8001-Add-APFS-driver.patch

From c968d4f93d7f9fe17d170511147958bdcac1c69b Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Tue, 17 Mar 2026 16:24:16 +0000
Subject: [PATCH] Add APFS driver

---
 fs/apfs/Makefile                  |   28 +
 fs/apfs/apfs.h                    | 1348 ++++++++++++
 fs/apfs/apfs_raw.h                | 1570 ++++++++++++++
 fs/apfs/btree.c                   | 1174 +++++++++++
 fs/apfs/compress.c                |  478 +++++
 fs/apfs/dir.c                     | 1505 ++++++++++++++
 fs/apfs/extents.c                 | 2374 ++++++++++++++++++++++
 fs/apfs/file.c                    |  224 ++
 fs/apfs/inode.c                   | 2598 ++++++++++++++++++++++++
 fs/apfs/key.c                     |  337 +++
 fs/apfs/libzbitmap.c              |  442 ++++
 fs/apfs/libzbitmap.h              |   31 +
 fs/apfs/lzfse/lzfse.h             |   67 +
 fs/apfs/lzfse/lzfse_decode.c      |   80 +
 fs/apfs/lzfse/lzfse_decode_base.c |  743 +++++++
 fs/apfs/lzfse/lzfse_fse.c         |  126 ++
 fs/apfs/lzfse/lzfse_fse.h         |  573 ++++++
 fs/apfs/lzfse/lzfse_internal.h    |  482 +++++
 fs/apfs/lzfse/lzvn_decode_base.c  |  487 +++++
 fs/apfs/lzfse/lzvn_decode_base.h  |   73 +
 fs/apfs/message.c                 |   29 +
 fs/apfs/namei.c                   |  146 ++
 fs/apfs/node.c                    | 2069 +++++++++++++++++++
 fs/apfs/object.c                  |  315 +++
 fs/apfs/snapshot.c                |  687 +++++++
 fs/apfs/spaceman.c                | 1479 ++++++++++++++
 fs/apfs/super.c                   | 2282 +++++++++++++++++++++
 fs/apfs/symlink.c                 |   80 +
 fs/apfs/transaction.c             | 1034 ++++++++++
 fs/apfs/unicode.c                 | 3156 +++++++++++++++++++++++++++++
 fs/apfs/unicode.h                 |   27 +
 fs/apfs/version.h                 |    1 +
 fs/apfs/xattr.c                   |  914 +++++++++
 fs/apfs/xfield.c                  |  171 ++
 34 files changed, 27130 insertions(+)
 create mode 100644 fs/apfs/Makefile
 create mode 100644 fs/apfs/apfs.h
 create mode 100644 fs/apfs/apfs_raw.h
 create mode 100644 fs/apfs/btree.c
 create mode 100644 fs/apfs/compress.c
 create mode 100644 fs/apfs/dir.c
 create mode 100644 fs/apfs/extents.c
 create mode 100644 fs/apfs/file.c
 create mode 100644 fs/apfs/inode.c
 create mode 100644 fs/apfs/key.c
 create mode 100644 fs/apfs/libzbitmap.c
 create mode 100644 fs/apfs/libzbitmap.h
 create mode 100644 fs/apfs/lzfse/lzfse.h
 create mode 100644 fs/apfs/lzfse/lzfse_decode.c
 create mode 100644 fs/apfs/lzfse/lzfse_decode_base.c
 create mode 100644 fs/apfs/lzfse/lzfse_fse.c
 create mode 100644 fs/apfs/lzfse/lzfse_fse.h
 create mode 100644 fs/apfs/lzfse/lzfse_internal.h
 create mode 100644 fs/apfs/lzfse/lzvn_decode_base.c
 create mode 100644 fs/apfs/lzfse/lzvn_decode_base.h
 create mode 100644 fs/apfs/message.c
 create mode 100644 fs/apfs/namei.c
 create mode 100644 fs/apfs/node.c
 create mode 100644 fs/apfs/object.c
 create mode 100644 fs/apfs/snapshot.c
 create mode 100644 fs/apfs/spaceman.c
 create mode 100644 fs/apfs/super.c
 create mode 100644 fs/apfs/symlink.c
 create mode 100644 fs/apfs/transaction.c
 create mode 100644 fs/apfs/unicode.c
 create mode 100644 fs/apfs/unicode.h
 create mode 100644 fs/apfs/version.h
 create mode 100644 fs/apfs/xattr.c
 create mode 100644 fs/apfs/xfield.c

diff --git a/fs/apfs/Makefile b/fs/apfs/Makefile
new file mode 100644
index 000000000..2bafc31f1
--- /dev/null
+++ b/fs/apfs/Makefile
@@ -0,0 +1,28 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Makefile for the out-of-tree Linux APFS module.
+#
+
+KERNELRELEASE ?= $(shell uname -r)
+KERNEL_DIR    ?= /lib/modules/$(KERNELRELEASE)/build
+PWD           := $(shell pwd)
+
+obj-m = apfs.o
+apfs-y := btree.o compress.o dir.o extents.o file.o inode.o key.o libzbitmap.o \
+	  lzfse/lzfse_decode.o lzfse/lzfse_decode_base.o lzfse/lzfse_fse.o \
+	  lzfse/lzvn_decode_base.o message.o namei.o node.o object.o snapshot.o \
+	  spaceman.o super.o symlink.o transaction.o unicode.o xattr.o xfield.o
+
+# If you want mounts to be writable by default, run the build as:
+#   make APFS_CONFIG=-DCONFIG_APFS_RW_ALWAYS
+# This is risky and not generally recommended.
+ccflags-y += $(APFS_CONFIG)
+
+default:
+	./genver.sh
+	make -C $(KERNEL_DIR) M=$(PWD)
+install:
+	make -C $(KERNEL_DIR) M=$(PWD) modules_install
+clean:
+	rm -f version.h
+	make -C $(KERNEL_DIR) M=$(PWD) clean
diff --git a/fs/apfs/apfs.h b/fs/apfs/apfs.h
new file mode 100644
index 000000000..27fb215ed
--- /dev/null
+++ b/fs/apfs/apfs.h
@@ -0,0 +1,1348 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#ifndef _APFS_H
+#define _APFS_H
+
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/list.h>
+#include <linux/types.h>
+#include <linux/version.h>
+
+/*
+ * Redhat kernels regularly break the api, so they need their own version
+ * checks. Make sure they always fail for non-rhel kernels.
+ *
+ * By the way, if rhel keeps picking up breaking patches for old majors after
+ * a new one is out, this check could break down and things could get more
+ * complicated. I don't think they do that though...
+ */
+#ifdef RHEL_RELEASE
+#define RHEL_VERSION_GE(a, b)	(RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(a, b))
+#else
+#define RHEL_VERSION_GE(a, b)	0
+#endif
+
+#include "apfs_raw.h"
+
+#define EFSBADCRC	EBADMSG		/* Bad CRC detected */
+#define EFSCORRUPTED	EUCLEAN		/* Filesystem is corrupted */
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 14, 0) /* SB_RDONLY came in 4.14 */
+#define SB_RDONLY	MS_RDONLY
+#define SB_SILENT	MS_SILENT
+#define SB_NOSEC	MS_NOSEC
+#define SB_ACTIVE	MS_ACTIVE
+static inline bool sb_rdonly(const struct super_block *sb) { return sb->s_flags & SB_RDONLY; }
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 3, 0)
+#define lockdep_assert_held_write(l)	((void)(l))
+#endif
+
+/* Compatibility wrapper around submit_bh() */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 0, 0) || RHEL_VERSION_GE(9, 2)
+#define apfs_submit_bh(op, op_flags, bh) submit_bh(op | op_flags, bh)
+#else
+#define apfs_submit_bh(op, op_flags, bh) submit_bh(op, op_flags, bh)
+#endif
+
+/*
+ * Parameter for the snapshot creation ioctl
+ */
+struct apfs_ioctl_snap_name {
+	char name[APFS_SNAP_MAX_NAMELEN + 1];
+};
+
+#define APFS_IOC_SET_DFLT_PFK	_IOW('@', 0x80, struct apfs_wrapped_crypto_state)
+#define APFS_IOC_SET_DIR_CLASS	_IOW('@', 0x81, u32)
+#define APFS_IOC_SET_PFK	_IOW('@', 0x82, struct apfs_wrapped_crypto_state)
+#define APFS_IOC_GET_CLASS	_IOR('@', 0x83, u32)
+#define APFS_IOC_GET_PFK	_IOR('@', 0x84, struct apfs_wrapped_crypto_state)
+#define APFS_IOC_TAKE_SNAPSHOT	_IOW('@', 0x85, struct apfs_ioctl_snap_name)
+
+/*
+ * In-memory representation of an APFS object
+ */
+struct apfs_object {
+	struct super_block *sb;
+	u64 block_nr;
+	u64 oid;
+
+	/*
+	 * Buffer head containing the one block of the object, may be NULL if
+	 * the object is only in memory. TODO: support objects with more than
+	 * one block.
+	 */
+	struct buffer_head *o_bh;
+	char *data; /* The raw object */
+	bool ephemeral; /* Is this an ephemeral object? */
+};
+
+/* Constants used in managing the size of a node's table of contents */
+#define APFS_BTREE_TOC_ENTRY_INCREMENT	8
+#define APFS_BTREE_TOC_ENTRY_MAX_UNUSED	(2 * BTREE_TOC_ENTRY_INCREMENT)
+
+/*
+ * In-memory representation of an APFS node
+ */
+struct apfs_node {
+	u32 tree_type;		/* Tree type (subtype of the node object) */
+	u16 flags;		/* Node flags */
+	u32 records;		/* Number of records in the node */
+
+	int key;		/* Offset of the key area in the block */
+	int free;		/* Offset of the free area in the block */
+	int val;		/* Offset of the value area in the block */
+
+	int key_free_list_len;	/* Length of the fragmented free key space */
+	int val_free_list_len;	/* Length of the fragmented free value space */
+
+	struct apfs_object object; /* Object holding the node */
+};
+
+/**
+ * apfs_node_is_leaf - Check if a b-tree node is a leaf
+ * @node: the node to check
+ */
+static inline bool apfs_node_is_leaf(struct apfs_node *node)
+{
+	return (node->flags & APFS_BTNODE_LEAF) != 0;
+}
+
+/**
+ * apfs_node_is_root - Check if a b-tree node is the root
+ * @node: the node to check
+ */
+static inline bool apfs_node_is_root(struct apfs_node *node)
+{
+	return (node->flags & APFS_BTNODE_ROOT) != 0;
+}
+
+/**
+ * apfs_node_has_fixed_kv_size - Check if a b-tree node has fixed key/value
+ * sizes
+ * @node: the node to check
+ */
+static inline bool apfs_node_has_fixed_kv_size(struct apfs_node *node)
+{
+	return (node->flags & APFS_BTNODE_FIXED_KV_SIZE) != 0;
+}
+
+struct apfs_ip_bitmap_block_info {
+	bool	dirty;		/* Do we need to commit this block to disk? */
+	void	*block;		/* In-memory address of the bitmap block */
+};
+
+/*
+ * Space manager data in memory.
+ */
+struct apfs_spaceman {
+	struct apfs_spaceman_phys *sm_raw; /* On-disk spaceman structure */
+	struct apfs_nxsb_info	  *sm_nxi; /* Container superblock */
+	u32			  sm_size; /* Size of @sm_raw in bytes */
+
+	u32 sm_blocks_per_chunk;	/* Blocks covered by a bitmap block */
+	u32 sm_chunks_per_cib;		/* Chunk count in a chunk-info block */
+	u64 sm_block_count;		/* Block count for the container */
+	u64 sm_chunk_count;		/* Number of bitmap blocks */
+	u32 sm_cib_count;		/* Number of chunk-info blocks */
+	u64 sm_free_count;		/* Number of free blocks */
+	u32 sm_addr_offset;		/* Offset of cib addresses in @sm_raw */
+	u64 sm_main_fq_nodes;		/* Number of nodes in the main fq */
+
+	/*
+	 * A range of freed blocks not yet put in the free queue. Extend this as
+	 * much as possible before creating an actual record.
+	 */
+	u64 sm_free_cache_base;
+	u64 sm_free_cache_blkcnt;
+
+	/* Shift to match an ip block with its bitmap in the array */
+	int sm_ip_bmaps_shift;
+	/* Mask to find an ip block's offset inside its ip bitmap */
+	u32 sm_ip_bmaps_mask;
+	/* Number of ip bitmaps */
+	u32 sm_ip_bmaps_count;
+	/* List of ip bitmaps, in order */
+	struct apfs_ip_bitmap_block_info sm_ip_bmaps[];
+};
+
+#define APFS_TRANS_MAIN_QUEUE_MAX	10000
+#define APFS_TRANS_BUFFERS_MAX		65536
+#define APFS_TRANS_STARTS_MAX		65536
+
+/* Possible states for the container transaction structure */
+#define APFS_NX_TRANS_FORCE_COMMIT	1	/* Commit guaranteed */
+#define APFS_NX_TRANS_DEFER_COMMIT	2	/* Commit banned right now */
+#define APFS_NX_TRANS_COMMITTING	4	/* Commit ongoing */
+#define APFS_NX_TRANS_INCOMPLETE_BLOCK	8	/* A data block is not written in full */
+
+/*
+ * ENOSPC checks before transactions are more or less coarse depending on this.
+ */
+enum apfs_trans_kind {
+	APFS_TRANS_REG,		/* Most transactions */
+	APFS_TRANS_DEL,		/* Transactions that usually free space */
+	APFS_TRANS_SYNC,	/* Sync (or unmount) transaction */
+};
+
+/* The space requirements for each kind of transaction, in blocks */
+#define APFS_REG_ROOM	128
+#define APFS_DEL_ROOM	20
+#define APFS_SYNC_ROOM	6
+
+/*
+ * Structure that keeps track of a container transaction.
+ */
+struct apfs_nx_transaction {
+	unsigned int t_state;
+
+	struct delayed_work t_work;	/* Work task for transaction commits */
+	struct super_block *t_work_sb;	/* sb that last queued a commit */
+
+	struct list_head t_inodes;	/* List of inodes in the transaction */
+	struct list_head t_buffers;	/* List of buffers in the transaction */
+	size_t t_buffers_count;		/* Count of items on the list */
+	int t_starts_count;		/* Count of starts for transaction */
+};
+
+/* State bits for buffer heads in a transaction */
+#define BH_TRANS	BH_PrivateStart		/* Attached to a transaction */
+#define BH_CSUM		(BH_PrivateStart + 1)	/* Requires checksum update */
+BUFFER_FNS(TRANS, trans);
+BUFFER_FNS(CSUM, csum);
+
+/*
+ * Additional information for a buffer in a transaction.
+ */
+struct apfs_bh_info {
+	struct buffer_head	*bh;	/* The buffer head */
+	struct list_head	list;	/* List of buffers in the transaction */
+};
+
+/*
+ * List entry for an in-memory ephemeral object
+ */
+struct apfs_ephemeral_object_info {
+	u64	oid;		/* Ephemeral object id */
+	u32	size;		/* Size of the object in bytes */
+	void	*object;	/* In-memory address of the object */
+};
+
+/*
+ * We allocate a fixed space for the list of ephemeral objects. I don't
+ * actually know how big this should be allowed to get, but all the objects
+ * must be written down with each transaction commit, so probably not too big.
+ */
+#define APFS_EPHEMERAL_LIST_SIZE	32768
+#define APFS_EPHEMERAL_LIST_LIMIT	(APFS_EPHEMERAL_LIST_SIZE / sizeof(struct apfs_ephemeral_object_info))
+
+/* Mount option flags for a container */
+#define APFS_CHECK_NODES	1
+#define APFS_READWRITE		2
+/*
+ * Mount options for a container are decided on its first mount. This flag lets
+ * future mounts know that has happened already.
+ */
+#define APFS_FLAGS_SET		4
+
+/*
+ * Wrapper around block devices for portability.
+ */
+struct apfs_blkdev_info {
+	struct block_device *blki_bdev;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 9, 0) || RHEL_VERSION_GE(9, 5)
+	struct file *blki_bdev_file;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+	struct bdev_handle *blki_bdev_handle;
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 5, 0) && !RHEL_VERSION_GE(9, 4)
+	fmode_t blki_mode;
+#endif
+
+	/*
+	 * For tier 2, we need to remember the mount path for show_options().
+	 * We don't need this for the main device of course, but better to
+	 * keep things consistent.
+	 */
+	char *blki_path;
+};
+
+/*
+ * Container superblock data in memory
+ */
+struct apfs_nxsb_info {
+	/* Block device info for the container */
+	struct apfs_blkdev_info *nx_blkdev_info;
+	struct apfs_blkdev_info *nx_tier2_info;
+	u64 nx_tier2_bno; /* Offset for tier 2 block numbers */
+
+	struct apfs_nx_superblock *nx_raw; /* On-disk main sb */
+	u64 nx_bno; /* Current block number for the checkpoint superblock */
+	u64 nx_xid; /* Latest transaction id */
+
+	/* List of ephemeral objects in memory (except the superblock) */
+	struct apfs_ephemeral_object_info *nx_eph_list;
+	int nx_eph_count;
+
+	struct list_head vol_list;	/* List of mounted volumes in container */
+
+	unsigned int nx_flags;	/* Mount options shared by all volumes */
+	unsigned int nx_refcnt; /* Number of mounted volumes in container */
+
+	/* TODO: handle block sizes above the maximum of PAGE_SIZE? */
+	unsigned long nx_blocksize;
+	unsigned char nx_blocksize_bits;
+
+	struct apfs_spaceman *nx_spaceman;
+	struct apfs_nx_transaction nx_transaction;
+	int nx_trans_buffers_max;
+
+	/* For now, a single semaphore for every operation */
+	struct rw_semaphore nx_big_sem;
+
+	/* List of currently mounted containers */
+	struct list_head nx_list;
+};
+
+extern struct mutex nxs_mutex;
+
+/*
+ * Omap mapping in memory.
+ * TODO: could this and apfs_omap_rec be the same struct?
+ */
+struct apfs_omap_map {
+	u64 xid;
+	u64 bno;
+	u32 flags;
+};
+
+/*
+ * Omap record data in memory
+ */
+struct apfs_omap_rec {
+	u64 oid;
+	u64 bno;
+};
+
+#define APFS_OMAP_CACHE_SLOTS		128
+#define APFS_OMAP_CACHE_SLOT_MASK	(APFS_OMAP_CACHE_SLOTS - 1)
+
+/**
+ * Cache of omap records
+ */
+struct apfs_omap_cache {
+	struct apfs_omap_rec recs[APFS_OMAP_CACHE_SLOTS];
+	bool disabled;
+	spinlock_t lock;
+};
+
+/*
+ * Omap structure shared by all snapshots for the same volume.
+ */
+struct apfs_omap {
+	struct apfs_node *omap_root;
+	struct apfs_omap_cache omap_cache;
+
+	/* Transaction id for most recent snapshot */
+	u64 omap_latest_snap;
+
+	/* Number of snapshots sharing this omap */
+	unsigned int omap_refcnt;
+};
+
+/*
+ * Volume superblock data in memory
+ */
+struct apfs_sb_info {
+	struct apfs_nxsb_info *s_nxi; /* In-memory container sb for volume */
+	struct list_head list;		/* List of mounted volumes in container */
+	struct apfs_superblock *s_vsb_raw; /* On-disk volume sb */
+
+	dev_t s_anon_dev; /* Anonymous device for this volume-snapshot */
+
+	char *s_tier2_path; /* Path to the tier 2 device */
+
+	char *s_snap_name; /* Label for the mounted snapshot */
+	u64 s_snap_xid; /* Transaction id for mounted snapshot */
+
+	struct apfs_node *s_cat_root;	/* Root of the catalog tree */
+	struct apfs_omap *s_omap;	/* The object map */
+
+	struct apfs_object s_vobject;	/* Volume superblock object */
+
+	/* Mount options */
+	unsigned int s_vol_nr;		/* Index of the volume in the sb list */
+	kuid_t s_uid;			/* uid to override on-disk uid */
+	kgid_t s_gid;			/* gid to override on-disk gid */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	unsigned int s_mount_opt;
+#endif
+
+	struct apfs_crypto_state_val *s_dflt_pfk; /* default per-file key */
+
+	struct inode *s_private_dir;	/* Inode for the private directory */
+	struct work_struct s_orphan_cleanup_work;
+	atomic_t s_orphan_cleanup_err;	/* Error from last orphan cleanup */
+};
+
+static inline struct apfs_sb_info *APFS_SB(struct super_block *sb)
+{
+	return sb->s_fs_info;
+}
+
+static inline bool apfs_is_sealed(struct super_block *sb)
+{
+	u64 flags = le64_to_cpu(APFS_SB(sb)->s_vsb_raw->apfs_incompatible_features);
+
+	return flags & APFS_INCOMPAT_SEALED_VOLUME;
+}
+
+/**
+ * apfs_vol_is_encrypted - Check if a volume is encrypting files
+ * @sb: superblock
+ */
+static inline bool apfs_vol_is_encrypted(struct super_block *sb)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+
+	return (vsb_raw->apfs_fs_flags & cpu_to_le64(APFS_FS_UNENCRYPTED)) == 0;
+}
+
+/**
+ * APFS_NXI - Get the shared container info for a volume's superblock
+ * @sb: superblock structure
+ */
+static inline struct apfs_nxsb_info *APFS_NXI(struct super_block *sb)
+{
+	return APFS_SB(sb)->s_nxi;
+}
+
+/**
+ * APFS_SM - Get the shared spaceman struct for a volume's superblock
+ * @sb: superblock structure
+ */
+static inline struct apfs_spaceman *APFS_SM(struct super_block *sb)
+{
+	return APFS_NXI(sb)->nx_spaceman;
+}
+
+static inline bool apfs_is_case_insensitive(struct super_block *sb)
+{
+	return (APFS_SB(sb)->s_vsb_raw->apfs_incompatible_features &
+	       cpu_to_le64(APFS_INCOMPAT_CASE_INSENSITIVE)) != 0;
+}
+
+static inline bool apfs_is_normalization_insensitive(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	u64 flags = le64_to_cpu(sbi->s_vsb_raw->apfs_incompatible_features);
+
+	if (apfs_is_case_insensitive(sb))
+		return true;
+	if (flags & APFS_INCOMPAT_NORMALIZATION_INSENSITIVE)
+		return true;
+	return false;
+}
+
+/**
+ * apfs_max_maps_per_block - Find the maximum map count for a mapping block
+ * @sb: superblock structure
+ */
+static inline int apfs_max_maps_per_block(struct super_block *sb)
+{
+	unsigned long maps_size;
+
+	maps_size = (sb->s_blocksize - sizeof(struct apfs_checkpoint_map_phys));
+	return maps_size / sizeof(struct apfs_checkpoint_mapping);
+}
+
+/*
+ * In-memory representation of a key, as relevant for a b-tree query.
+ */
+struct apfs_key {
+	u64		id;
+	u64		number;	/* Extent offset, name hash or transaction id */
+	const char	*name;	/* On-disk name string */
+	u8		type;	/* Record type (0 for the omap) */
+};
+
+/**
+ * apfs_init_free_queue_key - Initialize an in-memory key for a free queue query
+ * @xid:	transaction id
+ * @paddr:	block number
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_free_queue_key(u64 xid, u64 paddr,
+					    struct apfs_key *key)
+{
+	key->id = xid;
+	key->type = 0;
+	key->number = paddr;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_omap_key - Initialize an in-memory key for an omap query
+ * @oid:	object id
+ * @xid:	latest transaction id
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_omap_key(u64 oid, u64 xid, struct apfs_key *key)
+{
+	key->id = oid;
+	key->type = 0;
+	key->number = xid;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_extent_key - Initialize an in-memory key for an extentref query
+ * @bno:	physical block number for the start of the extent
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_extent_key(u64 bno, struct apfs_key *key)
+{
+	key->id = bno;
+	key->type = APFS_TYPE_EXTENT;
+	key->number = 0;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_inode_key - Initialize an in-memory key for an inode query
+ * @ino:	inode number
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_inode_key(u64 ino, struct apfs_key *key)
+{
+	key->id = ino;
+	key->type = APFS_TYPE_INODE;
+	key->number = 0;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_file_extent_key - Initialize an in-memory key for an extent query
+ * @id:		extent id
+ * @offset:	logical address (0 for a multiple query)
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_file_extent_key(u64 id, u64 offset,
+					     struct apfs_key *key)
+{
+	key->id = id;
+	key->type = APFS_TYPE_FILE_EXTENT;
+	key->number = offset;
+	key->name = NULL;
+}
+
+static inline void apfs_init_fext_key(u64 id, u64 offset, struct apfs_key *key)
+{
+	key->id = id;
+	key->type = 0;
+	key->number = offset;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_dstream_id_key - Initialize an in-memory key for a dstream query
+ * @id:		data stream id
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_dstream_id_key(u64 id, struct apfs_key *key)
+{
+	key->id = id;
+	key->type = APFS_TYPE_DSTREAM_ID;
+	key->number = 0;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_crypto_state_key - Initialize an in-memory key for a crypto query
+ * @id:		crypto state id
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_crypto_state_key(u64 id, struct apfs_key *key)
+{
+	key->id = id;
+	key->type = APFS_TYPE_CRYPTO_STATE;
+	key->number = 0;
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_sibling_link_key - Initialize an in-memory key for a sibling query
+ * @ino:	inode number
+ * @id:		sibling id
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_sibling_link_key(u64 ino, u64 id,
+					      struct apfs_key *key)
+{
+	key->id = ino;
+	key->type = APFS_TYPE_SIBLING_LINK;
+	key->number = id; /* Only guessing (TODO) */
+	key->name = NULL;
+}
+
+/**
+ * apfs_init_sibling_map_key - Initialize in-memory key for a sibling map query
+ * @id:		sibling id
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_sibling_map_key(u64 id, struct apfs_key *key)
+{
+	key->id = id;
+	key->type = APFS_TYPE_SIBLING_MAP;
+	key->number = 0;
+	key->name = NULL;
+}
+
+extern void apfs_init_drec_key(struct super_block *sb, u64 ino, const char *name,
+			       unsigned int name_len, struct apfs_key *key);
+
+/**
+ * apfs_init_xattr_key - Initialize an in-memory key for a xattr query
+ * @ino:	inode number of the parent file
+ * @name:	xattr name (NULL for a multiple query)
+ * @key:	apfs_key structure to initialize
+ */
+static inline void apfs_init_xattr_key(u64 ino, const char *name,
+				       struct apfs_key *key)
+{
+	key->id = ino;
+	key->type = APFS_TYPE_XATTR;
+	key->number = 0;
+	key->name = name;
+}
+
+static inline void apfs_init_snap_metadata_key(u64 xid, struct apfs_key *key)
+{
+	key->id = xid;
+	key->type = APFS_TYPE_SNAP_METADATA;
+	key->number = 0;
+	key->name = NULL;
+}
+
+static inline void apfs_init_snap_name_key(const char *name, struct apfs_key *key)
+{
+	key->id = APFS_SNAP_NAME_OBJ_ID;
+	key->type = APFS_TYPE_SNAP_NAME;
+	key->number = 0;
+	key->name = name;
+}
+
+static inline void apfs_init_omap_snap_key(u64 xid, struct apfs_key *key)
+{
+	key->id = xid;
+	key->type = 0;
+	key->number = 0;
+	key->name = NULL;
+}
+
+/**
+ * apfs_key_set_hdr - Set the header for a raw catalog key
+ * @type:	record type
+ * @id:		record id
+ * @key:	the key to initialize
+ */
+static inline void apfs_key_set_hdr(u64 type, u64 id, void *key)
+{
+	struct apfs_key_header *hdr = key;
+
+	hdr->obj_id_and_type = cpu_to_le64(id | type << APFS_OBJ_TYPE_SHIFT);
+}
+
+/**
+ * apfs_cat_type - Read the record type of a catalog key
+ * @key: the raw catalog key
+ */
+static inline int apfs_cat_type(struct apfs_key_header *key)
+{
+	return (le64_to_cpu(key->obj_id_and_type) & APFS_OBJ_TYPE_MASK) >> APFS_OBJ_TYPE_SHIFT;
+}
+
+/**
+ * apfs_cat_cnid - Read the cnid value on a catalog key
+ * @key: the raw catalog key
+ *
+ * TODO: rename this function, since it's not just for the catalog anymore
+ */
+static inline u64 apfs_cat_cnid(struct apfs_key_header *key)
+{
+	return le64_to_cpu(key->obj_id_and_type) & APFS_OBJ_ID_MASK;
+}
+
+/* Flags for the query structure */
+#define APFS_QUERY_TREE_MASK	000177	/* Which b-tree we query */
+#define APFS_QUERY_OMAP		000001	/* This is a b-tree object map query */
+#define APFS_QUERY_CAT		000002	/* This is a catalog tree query */
+#define APFS_QUERY_FREE_QUEUE	000004	/* This is a free queue query */
+#define APFS_QUERY_EXTENTREF	000010	/* This is an extent reference query */
+#define APFS_QUERY_FEXT		000020	/* This is a fext tree query */
+#define APFS_QUERY_SNAP_META	000040	/* This is a snapshot meta query */
+#define APFS_QUERY_OMAP_SNAP	000100	/* This is an omap snapshots query */
+#define APFS_QUERY_NEXT		000200	/* Find next of multiple matches */
+#define APFS_QUERY_EXACT	000400	/* Search for an exact match */
+#define APFS_QUERY_DONE		001000	/* The search at this level is over */
+#define APFS_QUERY_ANY_NAME	002000	/* Multiple search for any name */
+#define APFS_QUERY_ANY_NUMBER	004000	/* Multiple search for any number */
+#define APFS_QUERY_MULTIPLE	(APFS_QUERY_ANY_NAME | APFS_QUERY_ANY_NUMBER)
+#define APFS_QUERY_PREV		010000	/* Find previous record */
+
+/*
+ * Structure used to retrieve data from an APFS B-Tree.
+ */
+struct apfs_query {
+	struct apfs_node *node;		/* Node being searched */
+	struct apfs_key key;		/* What the query is looking for */
+
+	struct apfs_query *parent;	/* Query for parent node */
+	unsigned int flags;
+
+	/* Set by the query on success */
+	int index;			/* Index of the entry in the node */
+	int key_off;			/* Offset of the key in the node */
+	int key_len;			/* Length of the key */
+	int off;			/* Offset of the value in the node */
+	int len;			/* Length of the value */
+
+	int depth;			/* Put a limit on recursion */
+};
+
+/**
+ * apfs_query_storage - Get the storage type for a query's btree
+ * @query: the query structure
+ */
+static inline u32 apfs_query_storage(struct apfs_query *query)
+{
+	if (query->flags & APFS_QUERY_OMAP)
+		return APFS_OBJ_PHYSICAL;
+	if (query->flags & APFS_QUERY_CAT)
+		return APFS_OBJ_VIRTUAL;
+	if (query->flags & APFS_QUERY_FEXT)
+		return APFS_OBJ_PHYSICAL;
+	if (query->flags & APFS_QUERY_FREE_QUEUE)
+		return APFS_OBJ_EPHEMERAL;
+	if (query->flags & APFS_QUERY_EXTENTREF)
+		return APFS_OBJ_PHYSICAL;
+	if (query->flags & APFS_QUERY_SNAP_META)
+		return APFS_OBJ_PHYSICAL;
+	if (query->flags & APFS_QUERY_OMAP_SNAP)
+		return APFS_OBJ_PHYSICAL;
+
+	/* Absurd, but don't panic: let the callers fail and report it */
+	return -1;
+}
+
+/*
+ * Extent record data in memory
+ */
+struct apfs_file_extent {
+	u64 logical_addr;
+	u64 phys_block_num;
+	u64 len;
+	u64 crypto_id;
+};
+
+/*
+ * Physical extent record data in memory
+ */
+struct apfs_phys_extent {
+	u64 bno;
+	u64 blkcount;
+	u64 len;	/* In bytes */
+	u32 refcnt;
+	u8 kind;
+};
+
+/*
+ * Data stream info in memory
+ */
+struct apfs_dstream_info {
+	struct super_block	*ds_sb;		/* Filesystem superblock */
+	struct inode		*ds_inode;	/* NULL for xattr dstreams */
+	u64			ds_id;		/* ID of the extent records */
+	u64			ds_size;	/* Length of the stream */
+	u64			ds_sparse_bytes;/* Hole byte count in stream */
+	struct apfs_file_extent	ds_cached_ext;	/* Latest extent record */
+	bool			ds_ext_dirty;	/* Is ds_cached_ext dirty? */
+	spinlock_t		ds_ext_lock;	/* Protects ds_cached_ext */
+	bool			ds_shared;	/* Has multiple references? */
+};
+
+/**
+ * apfs_alloced_size - Return the alloced size for a data stream
+ * @dstream: data stream info
+ *
+ * TODO: is this always correct? Or could the extents have an unused tail?
+ */
+static inline u64 apfs_alloced_size(struct apfs_dstream_info *dstream)
+{
+	struct super_block *sb = dstream->ds_sb;
+	u64 blks = (dstream->ds_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+
+	return blks << sb->s_blocksize_bits;
+}
+
+/*
+ * APFS inode data in memory
+ */
+struct apfs_inode_info {
+	u64			i_ino64;	 /* 32-bit-safe inode number */
+	u64			i_parent_id;	 /* ID of primary parent */
+	struct timespec64	i_crtime;	 /* Time of creation */
+	u32			i_nchildren;	 /* Child count for directory */
+	uid_t			i_saved_uid;	 /* User ID on disk */
+	gid_t			i_saved_gid;	 /* Group ID on disk */
+	u32			i_key_class;	 /* Security class for directory */
+	u64			i_int_flags;	 /* Internal flags */
+	u32			i_bsd_flags;	 /* BSD flags */
+	struct list_head	i_list;		 /* List of inodes in transaction */
+
+	bool			 i_has_dstream;	 /* Is there a dstream record? */
+	struct apfs_dstream_info i_dstream;	 /* Dstream data, if any */
+
+	bool			i_cleaned;	 /* Orphan data already deleted */
+
+	struct inode vfs_inode;
+};
+
+static inline struct apfs_inode_info *APFS_I(const struct inode *inode)
+{
+	return container_of(inode, struct apfs_inode_info, vfs_inode);
+}
+
+/**
+ * apfs_ino - Get the 64-bit id of an inode
+ * @inode: the vfs inode
+ *
+ * Returns all 64 bits of @inode's id, even on 32-bit architectures.
+ */
+static inline u64 apfs_ino(const struct inode *inode)
+{
+	return APFS_I(inode)->i_ino64;
+}
+
+/**
+ * apfs_set_ino - Set a 64-bit id on an inode
+ * @inode: the vfs inode
+ * @id:	   id to set
+ *
+ * Sets both the vfs inode number and the actual 32-bit-safe id.
+ */
+static inline void apfs_set_ino(struct inode *inode, u64 id)
+{
+	inode->i_ino = id; /* Higher bits may be lost, but it doesn't matter */
+	APFS_I(inode)->i_ino64 = id;
+}
+
+/* Make the compiler complain if we ever access i_ino directly by mistake */
+#define i_ino	DONT_USE_I_INO
+
+/*
+ * Directory entry record in memory
+ */
+struct apfs_drec {
+	u8 *name;
+	u64 ino;
+	u64 sibling_id; /* The sibling id; 0 if none */
+	int name_len;
+	unsigned int type;
+};
+
+/*
+ * Xattr record data in memory
+ */
+struct apfs_xattr {
+	u8 *name;
+	u8 *xdata;
+	int name_len;
+	int xdata_len;
+	bool has_dstream;
+};
+
+struct apfs_compressed_data {
+	bool has_dstream;
+	u64 size;
+	union {
+		struct apfs_dstream_info *dstream;
+		void *data;
+	};
+};
+
+/*
+ * Report function name and line number for the message types that are likely
+ * to signal a bug, to make things easier for reporters. Don't do this for the
+ * common messages, there is no point and it makes the console look too busy.
+ */
+#define apfs_emerg(sb, fmt, ...) apfs_msg(sb, KERN_EMERG, __func__, __LINE__, fmt, ##__VA_ARGS__)
+#define apfs_alert(sb, fmt, ...) apfs_msg(sb, KERN_ALERT, __func__, __LINE__, fmt, ##__VA_ARGS__)
+#define apfs_crit(sb, fmt, ...) apfs_msg(sb, KERN_CRIT, __func__, __LINE__, fmt, ##__VA_ARGS__)
+#define apfs_err(sb, fmt, ...) apfs_msg(sb, KERN_ERR, __func__, __LINE__, fmt, ##__VA_ARGS__)
+#define apfs_warn(sb, fmt, ...) apfs_msg(sb, KERN_WARNING, NULL, 0, fmt, ##__VA_ARGS__)
+#define apfs_notice(sb, fmt, ...) apfs_msg(sb, KERN_NOTICE, NULL, 0, fmt, ##__VA_ARGS__)
+#define apfs_info(sb, fmt, ...) apfs_msg(sb, KERN_INFO, NULL, 0, fmt, ##__VA_ARGS__)
+
+#ifdef CONFIG_APFS_DEBUG
+#define ASSERT(expr)	WARN_ON(!(expr))
+#define apfs_debug(sb, fmt, ...) apfs_msg(sb, KERN_DEBUG, __func__, __LINE__, fmt, ##__VA_ARGS__)
+#else
+#define ASSERT(expr)	((void)0)
+#define apfs_debug(sb, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
+#endif /* CONFIG_APFS_DEBUG */
+
+/**
+ * apfs_assert_in_transaction - Assert that the object is in current transaction
+ * @sb:		superblock structure
+ * @obj:	on-disk object to check
+ */
+#define apfs_assert_in_transaction(sb, obj)				\
+do {									\
+	(void)sb;							\
+	(void)obj;							\
+	ASSERT(le64_to_cpu((obj)->o_xid) == APFS_NXI(sb)->nx_xid);	\
+} while (0)
+
+/* btree.c */
+extern struct apfs_node *apfs_query_root(const struct apfs_query *query);
+extern struct apfs_query *apfs_alloc_query(struct apfs_node *node,
+					   struct apfs_query *parent);
+extern void apfs_free_query(struct apfs_query *query);
+extern int apfs_btree_query(struct super_block *sb, struct apfs_query **query);
+extern int apfs_omap_lookup_block(struct super_block *sb, struct apfs_omap *omap, u64 id, u64 *block, bool write);
+extern int apfs_omap_lookup_newest_block(struct super_block *sb, struct apfs_omap *omap, u64 id, u64 *block, bool write);
+extern int apfs_create_omap_rec(struct super_block *sb, u64 oid, u64 bno);
+extern int apfs_delete_omap_rec(struct super_block *sb, u64 oid);
+extern int apfs_query_join_transaction(struct apfs_query *query);
+extern int __apfs_btree_insert(struct apfs_query *query, void *key, int key_len, void *val, int val_len);
+extern int apfs_btree_insert(struct apfs_query *query, void *key, int key_len,
+			     void *val, int val_len);
+extern int apfs_btree_remove(struct apfs_query *query);
+extern void apfs_btree_change_node_count(struct apfs_query *query, int change);
+extern int apfs_btree_replace(struct apfs_query *query, void *key, int key_len,
+			      void *val, int val_len);
+extern void apfs_query_direct_forward(struct apfs_query *query);
+
+/* compress.c */
+extern int apfs_compress_get_size(struct inode *inode, loff_t *size);
+
+/* dir.c */
+extern int apfs_inode_by_name(struct inode *dir, const struct qstr *child,
+			      u64 *ino);
+extern int apfs_mkany(struct inode *dir, struct dentry *dentry,
+		      umode_t mode, dev_t rdev, const char *symname);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+extern int apfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
+		      dev_t rdev);
+extern int apfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
+extern int apfs_rename(struct inode *old_dir, struct dentry *old_dentry,
+		       struct inode *new_dir, struct dentry *new_dentry,
+		       unsigned int flags);
+extern int apfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
+		       bool excl);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+extern int apfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
+		      struct dentry *dentry, umode_t mode, dev_t rdev);
+extern int apfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
+		      struct dentry *dentry, umode_t mode);
+extern int apfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
+		       struct dentry *old_dentry, struct inode *new_dir,
+		       struct dentry *new_dentry, unsigned int flags);
+extern int apfs_create(struct user_namespace *mnt_userns, struct inode *dir,
+		       struct dentry *dentry, umode_t mode, bool excl);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 15, 0)
+extern int apfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
+		      struct dentry *dentry, umode_t mode, dev_t rdev);
+extern int apfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
+		      struct dentry *dentry, umode_t mode);
+extern int apfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
+		       struct dentry *old_dentry, struct inode *new_dir,
+		       struct dentry *new_dentry, unsigned int flags);
+extern int apfs_create(struct mnt_idmap *idmap, struct inode *dir,
+		       struct dentry *dentry, umode_t mode, bool excl);
+#else
+extern int apfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
+		      struct dentry *dentry, umode_t mode, dev_t rdev);
+extern struct dentry *apfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, struct dentry *dentry, umode_t mode);
+extern int apfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
+		       struct dentry *old_dentry, struct inode *new_dir,
+		       struct dentry *new_dentry, unsigned int flags);
+extern int apfs_create(struct mnt_idmap *idmap, struct inode *dir,
+		       struct dentry *dentry, umode_t mode, bool excl);
+#endif
+
+extern int apfs_link(struct dentry *old_dentry, struct inode *dir,
+		     struct dentry *dentry);
+extern int apfs_unlink(struct inode *dir, struct dentry *dentry);
+extern int apfs_rmdir(struct inode *dir, struct dentry *dentry);
+extern int apfs_delete_orphan_link(struct inode *inode);
+extern u64 apfs_any_orphan_ino(struct super_block *sb, u64 *ino_p);
+
+/* extents.c */
+extern int apfs_extent_from_query(struct apfs_query *query,
+				  struct apfs_file_extent *extent);
+extern int apfs_logic_to_phys_bno(struct apfs_dstream_info *dstream, sector_t dsblock, u64 *bno);
+extern int __apfs_get_block(struct apfs_dstream_info *dstream, sector_t iblock,
+			    struct buffer_head *bh_result, int create);
+extern int apfs_get_block(struct inode *inode, sector_t iblock,
+			  struct buffer_head *bh_result, int create);
+extern int apfs_flush_extent_cache(struct apfs_dstream_info *dstream);
+extern int apfs_dstream_get_new_bno(struct apfs_dstream_info *dstream, u64 dsblock, u64 *bno);
+extern int apfs_get_new_block(struct inode *inode, sector_t iblock,
+			      struct buffer_head *bh_result, int create);
+extern int apfs_truncate(struct apfs_dstream_info *dstream, loff_t new_size);
+extern int apfs_inode_delete_front(struct inode *inode);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 20, 0)
+extern loff_t apfs_remap_file_range(struct file *src_file, loff_t off, struct file *dst_file, loff_t destoff, loff_t len, unsigned int remap_flags);
+#else
+extern int apfs_clone_file_range(struct file *src_file, loff_t off, struct file *dst_file, loff_t destoff, u64 len);
+#endif
+extern int apfs_clone_extents(struct apfs_dstream_info *dstream, u64 new_id);
+extern int apfs_nonsparse_dstream_read(struct apfs_dstream_info *dstream, void *buf, size_t count, u64 offset);
+extern void apfs_nonsparse_dstream_preread(struct apfs_dstream_info *dstream);
+
+/* file.c */
+extern int apfs_file_mmap(struct file *file, struct vm_area_struct *vma);
+extern int apfs_fsync(struct file *file, loff_t start, loff_t end, int datasync);
+
+/* inode.c */
+extern struct inode *apfs_iget(struct super_block *sb, u64 cnid);
+extern int apfs_update_inode(struct inode *inode, char *new_name);
+extern void apfs_schedule_orphan_cleanup(struct super_block *sb);
+extern void apfs_orphan_cleanup_work(struct work_struct *work);
+extern void apfs_evict_inode(struct inode *inode);
+extern struct inode *apfs_new_inode(struct inode *dir, umode_t mode,
+				    dev_t rdev);
+extern int apfs_create_inode_rec(struct super_block *sb, struct inode *inode,
+				 struct dentry *dentry);
+extern int apfs_inode_create_exclusive_dstream(struct inode *inode);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+extern int __apfs_write_begin(const struct kiocb *iocb, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int flags, struct page **pagep, void **fsdata);
+extern int __apfs_write_end(const struct kiocb *iocb, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int copied, struct page *page, void *fsdata);
+#else
+extern int __apfs_write_begin(struct file *file, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int flags, struct page **pagep, void **fsdata);
+extern int __apfs_write_end(struct file *file, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int copied, struct page *page, void *fsdata);
+#endif
+extern int apfs_dstream_adj_refcnt(struct apfs_dstream_info *dstream, u32 delta);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+extern int apfs_setattr(struct dentry *dentry, struct iattr *iattr);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+extern int apfs_setattr(struct user_namespace *mnt_userns,
+			struct dentry *dentry, struct iattr *iattr);
+#else
+extern int apfs_setattr(struct mnt_idmap *idmap,
+			struct dentry *dentry, struct iattr *iattr);
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+extern int apfs_update_time(struct inode *inode, enum fs_update_time time, unsigned int flags);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+extern int apfs_update_time(struct inode *inode, struct timespec64 *time, int flags);
+#else
+extern int apfs_update_time(struct inode *inode, int flags);
+#endif
+long apfs_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
+long apfs_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0) /* No statx yet... */
+extern int apfs_getattr(struct vfsmount *mnt, struct dentry *dentry,
+			struct kstat *stat);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+extern int apfs_getattr(const struct path *path, struct kstat *stat,
+			u32 request_mask, unsigned int query_flags);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+extern int apfs_getattr(struct user_namespace *mnt_userns,
+		const struct path *path, struct kstat *stat, u32 request_mask,
+		unsigned int query_flags);
+#else
+extern int apfs_getattr(struct mnt_idmap *idmap,
+		const struct path *path, struct kstat *stat, u32 request_mask,
+		unsigned int query_flags);
+#endif
+
+extern int apfs_crypto_adj_refcnt(struct super_block *sb, u64 crypto_id, int delta);
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+#define fileattr file_kattr
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 3, 0) || RHEL_VERSION_GE(9, 6)
+extern int apfs_fileattr_get(struct dentry *dentry, struct fileattr *fa);
+extern int apfs_fileattr_set(struct mnt_idmap *idmap, struct dentry *dentry, struct fileattr *fa);
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(5, 13, 0)
+extern int apfs_fileattr_get(struct dentry *dentry, struct fileattr *fa);
+extern int apfs_fileattr_set(struct user_namespace *mnt_userns, struct dentry *dentry, struct fileattr *fa);
+#endif
+
+/* key.c */
+extern int apfs_filename_cmp(struct super_block *sb, const char *name1, unsigned int len1, const char *name2, unsigned int len2);
+extern int apfs_keycmp(struct apfs_key *k1, struct apfs_key *k2);
+extern int apfs_read_cat_key(void *raw, int size, struct apfs_key *key, bool hashed);
+extern int apfs_read_fext_key(void *raw, int size, struct apfs_key *key);
+extern int apfs_read_free_queue_key(void *raw, int size, struct apfs_key *key);
+extern int apfs_read_omap_key(void *raw, int size, struct apfs_key *key);
+extern int apfs_read_extentref_key(void *raw, int size, struct apfs_key *key);
+extern int apfs_read_snap_meta_key(void *raw, int size, struct apfs_key *key);
+extern int apfs_read_omap_snap_key(void *raw, int size, struct apfs_key *key);
+
+/* message.c */
+extern __printf(5, 6) void apfs_msg(struct super_block *sb, const char *prefix, const char *func, int line, const char *fmt, ...);
+
+/* node.c */
+extern struct apfs_node *apfs_read_node(struct super_block *sb, u64 oid,
+					u32 storage, bool write);
+extern void apfs_update_node(struct apfs_node *node);
+extern int apfs_delete_node(struct apfs_node *node, int type);
+extern int apfs_node_query(struct super_block *sb, struct apfs_query *query);
+extern void apfs_node_query_first(struct apfs_query *query);
+extern int apfs_omap_map_from_query(struct apfs_query *query, struct apfs_omap_map *map);
+extern int apfs_node_split(struct apfs_query *query);
+extern int apfs_node_locate_key(struct apfs_node *node, int index, int *off);
+extern void apfs_node_free(struct apfs_node *node);
+extern void apfs_node_free_range(struct apfs_node *node, u16 off, u16 len);
+extern bool apfs_node_has_room(struct apfs_node *node, int length, bool replace);
+extern int apfs_node_replace(struct apfs_query *query, void *key, int key_len, void *val, int val_len);
+extern int apfs_node_insert(struct apfs_query *query, void *key, int key_len, void *val, int val_len);
+extern int apfs_create_single_rec_node(struct apfs_query *query, void *key, int key_len, void *val, int val_len);
+extern int apfs_make_empty_btree_root(struct super_block *sb, u32 subtype, u64 *oid);
+
+/* object.c */
+extern int apfs_obj_verify_csum(struct super_block *sb, struct buffer_head *bh);
+extern void apfs_obj_set_csum(struct super_block *sb, struct apfs_obj_phys *obj);
+extern int apfs_multiblock_verify_csum(char *object, u32 size);
+extern void apfs_multiblock_set_csum(char *object, u32 size);
+extern int apfs_create_cpm_block(struct super_block *sb, u64 bno, struct buffer_head **bh_p);
+extern int apfs_create_cpoint_map(struct super_block *sb, struct apfs_checkpoint_map_phys *cpm, struct apfs_obj_phys *obj, u64 bno, u32 size);
+extern struct apfs_ephemeral_object_info *apfs_ephemeral_object_lookup(struct super_block *sb, u64 oid);
+extern struct buffer_head *apfs_read_object_block(struct super_block *sb, u64 bno, bool write, bool preserve);
+extern u32 apfs_index_in_data_area(struct super_block *sb, u64 bno);
+extern u64 apfs_data_index_to_bno(struct super_block *sb, u32 index);
+
+/* snapshot.c */
+extern int apfs_ioc_take_snapshot(struct file *file, void __user *user_arg);
+extern int apfs_switch_to_snapshot(struct super_block *sb);
+
+/* spaceman.c */
+extern int apfs_read_spaceman(struct super_block *sb);
+extern int apfs_free_queue_insert_nocache(struct super_block *sb, u64 bno, u64 count);
+extern int apfs_free_queue_insert(struct super_block *sb, u64 bno, u64 count);
+extern int apfs_spaceman_allocate_block(struct super_block *sb, u64 *bno, bool backwards);
+extern int apfs_write_ip_bitmaps(struct super_block *sb);
+extern int apfs_spaceman_get_free_blkcnt(struct super_block *sb, u64 *blkcnt);
+
+/* super.c */
+extern int apfs_map_volume_super_bno(struct super_block *sb, u64 bno, bool check);
+extern int apfs_map_volume_super(struct super_block *sb, bool write);
+extern void apfs_unmap_volume_super(struct super_block *sb);
+extern int apfs_read_omap(struct super_block *sb, bool write);
+extern int apfs_read_catalog(struct super_block *sb, bool write);
+extern int apfs_sync_fs(struct super_block *sb, int wait);
+
+/* transaction.c */
+extern int apfs_cpoint_data_free(struct super_block *sb, u64 bno);
+extern void apfs_transaction_init(struct apfs_nx_transaction *trans);
+extern int apfs_transaction_start(struct super_block *sb, enum apfs_trans_kind kind);
+extern int apfs_transaction_commit(struct super_block *sb);
+extern void apfs_inode_join_transaction(struct super_block *sb, struct inode *inode);
+extern int apfs_transaction_join(struct super_block *sb,
+				 struct buffer_head *bh);
+void apfs_transaction_abort(struct super_block *sb);
+extern int apfs_transaction_flush_all_inodes(struct super_block *sb);
+extern int apfs_read_ephemeral_objects(struct super_block *sb);
+
+/* xattr.c */
+extern int ____apfs_xattr_get(struct inode *inode, const char *name, void *buffer,
+			      size_t size, bool only_whole);
+extern int __apfs_xattr_get(struct inode *inode, const char *name, void *buffer,
+			    size_t size);
+extern int apfs_delete_all_xattrs(struct inode *inode);
+extern int apfs_xattr_set(struct inode *inode, const char *name, const void *value,
+			  size_t size, int flags);
+extern ssize_t apfs_listxattr(struct dentry *dentry, char *buffer, size_t size);
+extern int apfs_xattr_get_compressed_data(struct inode *inode, const char *name, struct apfs_compressed_data *cdata);
+extern void apfs_release_compressed_data(struct apfs_compressed_data *cdata);
+extern int apfs_compressed_data_read(struct apfs_compressed_data *cdata, void *buf, size_t count, u64 offset);
+
+/* xfield.c */
+extern int apfs_find_xfield(u8 *xfields, int len, u8 xtype, char **xval);
+extern int apfs_init_xfields(u8 *buffer, int buflen);
+extern int apfs_insert_xfield(u8 *buffer, int buflen,
+			      const struct apfs_x_field *xkey,
+			      const void *xval);
+
+/*
+ * Inode and file operations
+ */
+
+/* compress.c */
+extern const struct address_space_operations apfs_compress_aops;
+extern const struct file_operations apfs_compress_file_operations;
+
+/* dir.c */
+extern const struct file_operations apfs_dir_operations;
+
+/* file.c */
+extern const struct file_operations apfs_file_operations;
+extern const struct inode_operations apfs_file_inode_operations;
+
+/* namei.c */
+extern const struct inode_operations apfs_dir_inode_operations;
+extern const struct inode_operations apfs_special_inode_operations;
+extern const struct dentry_operations apfs_dentry_operations;
+
+/* symlink.c */
+extern const struct inode_operations apfs_symlink_inode_operations;
+
+/* xattr.c */
+extern const struct xattr_handler *apfs_xattr_handlers[];
+
+/**
+ * apfs_assert_query_is_valid - Assert that all of a query's ancestors are set
+ * @query: the query to check
+ *
+ * A query may lose some of its ancestors during a node split, but nothing
+ * should be done to such a query until it gets refreshed.
+ */
+static inline void apfs_assert_query_is_valid(const struct apfs_query *query)
+{
+	ASSERT(apfs_node_is_root(apfs_query_root(query)));
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 15, 0)
+#define page_has_buffers(page)	folio_buffers(page_folio(page))
+#endif
+
+static inline int apfs_inode_state_read_once(struct inode *inode)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 19, 0)
+	return inode_state_read_once(inode);
+#else
+	return inode->i_state;
+#endif
+}
+
+static inline void apfs_inode_state_set_raw(struct inode *inode, int flags)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 19, 0)
+	inode_state_set_raw(inode, flags);
+#else
+	inode->i_state |= flags;
+#endif
+}
+
+static inline void apfs_inode_state_clear_raw(struct inode *inode, int flags)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 19, 0)
+	inode_state_clear_raw(inode, flags);
+#else
+	inode->i_state &= ~flags;
+#endif
+}
+
+/*
+ * TODO: the following are modified variants of buffer head functions that will
+ * work with the shared block device for the container. The correct approach
+ * here would be to avoid buffer heads and use bios, but for now this will do.
+ */
+
+static inline void
+apfs_map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
+{
+	struct apfs_nxsb_info *nxi = NULL;
+	struct apfs_blkdev_info *info = NULL;
+
+	nxi = APFS_NXI(sb);
+	info = nxi->nx_blkdev_info;
+
+	/*
+	 * Block numbers above s_tier2_bno are for the tier 2 device of a
+	 * fusion drive. Don't bother returning an error code if a regular
+	 * drive gets corrupted and reports a tier 2 block number: just treat
+	 * it the same as any other out-of-range block.
+	 */
+	if (block >= nxi->nx_tier2_bno) {
+		if (nxi->nx_tier2_info) {
+			info = nxi->nx_tier2_info;
+			block -= nxi->nx_tier2_bno;
+		} else {
+			apfs_err(sb, "block number in tier 2 range (0x%llx)", (unsigned long long)block);
+		}
+	}
+
+	set_buffer_mapped(bh);
+	bh->b_bdev = info->blki_bdev;
+	bh->b_blocknr = block;
+	bh->b_size = sb->s_blocksize;
+}
+
+static inline struct buffer_head *
+apfs_sb_bread(struct super_block *sb, sector_t block)
+{
+	struct apfs_nxsb_info *nxi = NULL;
+	struct apfs_blkdev_info *info = NULL;
+
+	nxi = APFS_NXI(sb);
+	info = nxi->nx_blkdev_info;
+
+	if (block >= nxi->nx_tier2_bno) {
+		if (!nxi->nx_tier2_info) {
+			/* Not really a fusion drive, so it's corrupted */
+			apfs_err(sb, "block number in tier 2 range (0x%llx)", (unsigned long long)block);
+			return NULL;
+		}
+		info = nxi->nx_tier2_info;
+		block -= nxi->nx_tier2_bno;
+	}
+
+	return __bread_gfp(info->blki_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
+}
+
+/* Like apfs_getblk(), but doesn't mark the buffer uptodate */
+static inline struct buffer_head *
+__apfs_getblk(struct super_block *sb, sector_t block)
+{
+	struct apfs_nxsb_info *nxi = NULL;
+	struct apfs_blkdev_info *info = NULL;
+
+	nxi = APFS_NXI(sb);
+	info = nxi->nx_blkdev_info;
+
+	if (block >= nxi->nx_tier2_bno) {
+		if (!nxi->nx_tier2_info) {
+			/* Not really a fusion drive, so it's corrupted */
+			apfs_err(sb, "block number in tier 2 range (0x%llx)", (unsigned long long)block);
+			return NULL;
+		}
+		info = nxi->nx_tier2_info;
+		block -= nxi->nx_tier2_bno;
+	}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	return __getblk_gfp(info->blki_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
+#else
+	return bdev_getblk(info->blki_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
+#endif
+}
+
+/* Use instead of apfs_sb_bread() for blocks that will just be overwritten */
+static inline struct buffer_head *
+apfs_getblk(struct super_block *sb, sector_t block)
+{
+	struct buffer_head *bh = NULL;
+
+	bh = __apfs_getblk(sb, block);
+	if (bh)
+		set_buffer_uptodate(bh);
+	return bh;
+}
+
+#endif	/* _APFS_H */
diff --git a/fs/apfs/apfs_raw.h b/fs/apfs/apfs_raw.h
new file mode 100644
index 000000000..415e6b713
--- /dev/null
+++ b/fs/apfs/apfs_raw.h
@@ -0,0 +1,1570 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2019 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ *
+ * Definitions for all on-disk data structures.
+ */
+
+#ifndef _APFS_RAW_H
+#define _APFS_RAW_H
+
+#include <linux/types.h>
+#include <linux/uuid.h>
+
+/* Object identifiers constants */
+#define APFS_OID_NX_SUPERBLOCK			1
+#define APFS_OID_INVALID			0ULL
+#define APFS_OID_RESERVED_COUNT			1024
+
+/* Object type masks */
+#define APFS_OBJECT_TYPE_MASK			0x0000ffff
+#define APFS_OBJECT_TYPE_FLAGS_MASK		0xffff0000
+#define APFS_OBJ_STORAGETYPE_MASK		0xc0000000
+#define APFS_OBJECT_TYPE_FLAGS_DEFINED_MASK	0xf8000000
+
+/* Object types */
+#define APFS_OBJECT_TYPE_NX_SUPERBLOCK		0x00000001
+#define APFS_OBJECT_TYPE_BTREE			0x00000002
+#define APFS_OBJECT_TYPE_BTREE_NODE		0x00000003
+#define APFS_OBJECT_TYPE_SPACEMAN		0x00000005
+#define APFS_OBJECT_TYPE_SPACEMAN_CAB		0x00000006
+#define APFS_OBJECT_TYPE_SPACEMAN_CIB		0x00000007
+#define APFS_OBJECT_TYPE_SPACEMAN_BITMAP	0x00000008
+#define APFS_OBJECT_TYPE_SPACEMAN_FREE_QUEUE	0x00000009
+#define APFS_OBJECT_TYPE_EXTENT_LIST_TREE	0x0000000a
+#define APFS_OBJECT_TYPE_OMAP			0x0000000b
+#define APFS_OBJECT_TYPE_CHECKPOINT_MAP		0x0000000c
+#define APFS_OBJECT_TYPE_FS			0x0000000d
+#define APFS_OBJECT_TYPE_FSTREE			0x0000000e
+#define APFS_OBJECT_TYPE_BLOCKREFTREE		0x0000000f
+#define APFS_OBJECT_TYPE_SNAPMETATREE		0x00000010
+#define APFS_OBJECT_TYPE_NX_REAPER		0x00000011
+#define APFS_OBJECT_TYPE_NX_REAP_LIST		0x00000012
+#define APFS_OBJECT_TYPE_OMAP_SNAPSHOT		0x00000013
+#define APFS_OBJECT_TYPE_EFI_JUMPSTART		0x00000014
+#define APFS_OBJECT_TYPE_FUSION_MIDDLE_TREE	0x00000015
+#define APFS_OBJECT_TYPE_NX_FUSION_WBC		0x00000016
+#define APFS_OBJECT_TYPE_NX_FUSION_WBC_LIST	0x00000017
+#define APFS_OBJECT_TYPE_ER_STATE		0x00000018
+#define APFS_OBJECT_TYPE_GBITMAP		0x00000019
+#define APFS_OBJECT_TYPE_GBITMAP_TREE		0x0000001a
+#define APFS_OBJECT_TYPE_GBITMAP_BLOCK		0x0000001b
+#define APFS_OBJECT_TYPE_ER_RECOVERY_BLOCK	0x0000001c
+#define APFS_OBJECT_TYPE_SNAP_META_EXT		0x0000001d
+#define APFS_OBJECT_TYPE_INTEGRITY_META		0x0000001e
+#define APFS_OBJECT_TYPE_FEXT_TREE		0x0000001f
+#define APFS_OBJECT_TYPE_RESERVED_20		0x00000020
+#define APFS_OBJECT_TYPE_INVALID		0x00000000
+#define APFS_OBJECT_TYPE_TEST			0x000000ff
+
+/* Object type flags */
+#define APFS_OBJ_VIRTUAL			0x00000000
+#define APFS_OBJ_EPHEMERAL			0x80000000
+#define APFS_OBJ_PHYSICAL			0x40000000
+#define APFS_OBJ_NOHEADER			0x20000000
+#define APFS_OBJ_ENCRYPTED			0x10000000
+#define APFS_OBJ_NONPERSISTENT			0x08000000
+
+#define APFS_MAX_CKSUM_SIZE 8
+
+/*
+ * On-disk representation of an APFS object
+ */
+struct apfs_obj_phys {
+/*00*/	__le64 o_cksum;		/* Fletcher checksum */
+	__le64 o_oid;		/* Object-id */
+/*10*/	__le64 o_xid;		/* Transaction ID */
+	__le32 o_type;		/* Object type */
+	__le32 o_subtype;	/* Object subtype */
+} __packed;
+
+/* Flags for the object map structure */
+#define APFS_OMAP_MANUALLY_MANAGED	0x00000001
+#define APFS_OMAP_ENCRYPTING		0x00000002
+#define APFS_OMAP_DECRYPTING		0x00000004
+#define APFS_OMAP_KEYROLLING		0x00000008
+#define APFS_OMAP_CRYPTO_GENERATION	0x00000010
+#define APFS_OMAP_FLAGS_VALID_MASK	(APFS_OMAP_MANUALLY_MANAGED \
+					| APFS_OMAP_ENCRYPTING \
+					| APFS_OMAP_DECRYPTING \
+					| APFS_OMAP_KEYROLLING \
+					| APFS_OMAP_CRYPTO_GENERATION)
+
+/*
+ * On-disk representation of an object map
+ */
+struct apfs_omap_phys {
+/*00*/	struct apfs_obj_phys om_o;
+/*20*/	__le32 om_flags;
+	__le32 om_snap_count;
+	__le32 om_tree_type;
+	__le32 om_snapshot_tree_type;
+/*30*/	__le64 om_tree_oid;
+	__le64 om_snapshot_tree_oid;
+/*40*/	__le64 om_most_recent_snap;
+	__le64 om_pending_revert_min;
+	__le64 om_pending_revert_max;
+} __packed;
+
+/* Object map value flags */
+#define APFS_OMAP_VAL_DELETED		0x00000001
+#define APFS_OMAP_VAL_SAVED		0x00000002
+#define APFS_OMAP_VAL_ENCRYPTED		0x00000004
+#define APFS_OMAP_VAL_NOHEADER		0x00000008
+#define APFS_OMAP_VAL_CRYPTO_GENERATION	0x00000010
+#define APFS_OMAP_VAL_FLAGS_VALID_MASK	(APFS_OMAP_VAL_DELETED \
+					| APFS_OMAP_VAL_SAVED \
+					| APFS_OMAP_VAL_ENCRYPTED \
+					| APFS_OMAP_VAL_NOHEADER \
+					| APFS_OMAP_VAL_CRYPTO_GENERATION)
+
+/*
+ * Structure of a value in an object map B-tree
+ */
+struct apfs_omap_val {
+	__le32 ov_flags;
+	__le32 ov_size;
+	__le64 ov_paddr;
+} __packed;
+
+/*
+ * Structure of a value in an omap's snapshot tree
+ */
+struct apfs_omap_snapshot {
+	__le32 oms_flags;
+	__le32 oms_pad;
+	__le64 oms_oid;
+} __packed;
+
+/* B-tree node flags */
+#define APFS_BTNODE_ROOT		0x0001
+#define APFS_BTNODE_LEAF		0x0002
+#define APFS_BTNODE_FIXED_KV_SIZE	0x0004
+#define APFS_BTNODE_CHECK_KOFF_INVAL	0x8000
+#define APFS_BTNODE_MASK		0x0007	/* Valid on-disk flags */
+
+/* B-tree location constants */
+#define APFS_BTOFF_INVALID		0xffff
+
+/*
+ * Structure storing a location inside a B-tree node
+ */
+struct apfs_nloc {
+	__le16 off;
+	__le16 len;
+} __packed;
+
+/*
+ * Structure storing the location of a key/value pair within a B-tree node
+ */
+struct apfs_kvloc {
+	struct apfs_nloc k;
+	struct apfs_nloc v;
+} __packed;
+
+/*
+ * Structure storing the location of a key/value pair within a B-tree node
+ * having fixed-size key and value (flag APFS_BTNODE_FIXED_KV_SIZE is present)
+ */
+struct apfs_kvoff {
+	__le16 k;
+	__le16 v;
+} __packed;
+
+/*
+ * On-disk representation of a B-tree node
+ */
+struct apfs_btree_node_phys {
+/*00*/	struct apfs_obj_phys btn_o;
+/*20*/	__le16 btn_flags;
+	__le16 btn_level;
+	__le32 btn_nkeys;
+/*28*/	struct apfs_nloc btn_table_space;
+	struct apfs_nloc btn_free_space;
+	struct apfs_nloc btn_key_free_list;
+	struct apfs_nloc btn_val_free_list;
+/*38*/	__le64 btn_data[];
+} __packed;
+
+/* B-tree info flags */
+#define APFS_BTREE_UINT64_KEYS		0x00000001
+#define APFS_BTREE_SEQUENTIAL_INSERT	0x00000002
+#define APFS_BTREE_ALLOW_GHOSTS		0x00000004
+#define APFS_BTREE_EPHEMERAL		0x00000008
+#define APFS_BTREE_PHYSICAL		0x00000010
+#define APFS_BTREE_NONPERSISTENT	0x00000020
+#define APFS_BTREE_KV_NONALIGNED	0x00000040
+#define APFS_BTREE_FLAGS_VALID_MASK	(APFS_BTREE_UINT64_KEYS \
+					| APFS_BTREE_SEQUENTIAL_INSERT \
+					| APFS_BTREE_ALLOW_GHOSTS \
+					| APFS_BTREE_EPHEMERAL \
+					| APFS_BTREE_PHYSICAL \
+					| APFS_BTREE_NONPERSISTENT \
+					| APFS_BTREE_KV_NONALIGNED)
+
+/*
+ * Structure used to store information about a B-tree that won't change
+ * over time
+ */
+struct apfs_btree_info_fixed {
+	__le32 bt_flags;
+	__le32 bt_node_size;
+	__le32 bt_key_size;
+	__le32 bt_val_size;
+} __packed;
+
+/*
+ * Structure used to store information about a B-tree (located at the end of
+ * a B-tree root node block)
+ */
+struct apfs_btree_info {
+	struct apfs_btree_info_fixed bt_fixed;
+	__le32 bt_longest_key;			/* Longest key ever stored */
+	__le32 bt_longest_val;			/* Longest value ever stored */
+	__le64 bt_key_count;
+	__le64 bt_node_count;
+} __packed;
+
+/*
+ * Structure of the value of a directory entry. This is the data in
+ * the catalog nodes for record type APFS_TYPE_DIR_REC.
+ */
+struct apfs_drec_val {
+	__le64 file_id;
+	__le64 date_added;
+	__le16 flags;
+	u8 xfields[];
+} __packed;
+
+/* Physical extent records */
+#define APFS_PEXT_LEN_MASK	0x0fffffffffffffffULL
+#define APFS_PEXT_KIND_MASK	0xf000000000000000ULL
+#define APFS_PEXT_KIND_SHIFT	60
+
+/* The kind of a physical extent record */
+enum {
+	APFS_KIND_ANY		= 0,
+	APFS_KIND_NEW		= 1,
+	APFS_KIND_UPDATE	= 2,
+	APFS_KIND_DEAD		= 3,
+	APFS_KIND_UPDATE_REFCNT	= 4,
+
+	APFS_KIND_INVALID	= 255 /* This is weird, won't fit in 4 bits */
+};
+
+#define APFS_OWNING_OBJ_ID_INVALID	(~0ULL)
+#define APFS_OWNING_OBJ_ID_UNKNOWN	(~1ULL)
+
+/*
+ * Structure of a physical extent record
+ */
+struct apfs_phys_ext_val {
+	__le64 len_and_kind;
+	__le64 owning_obj_id;
+	__le32 refcnt;
+} __packed;
+
+/* File extent records */
+#define APFS_FILE_EXTENT_LEN_MASK	0x00ffffffffffffffULL
+#define APFS_FILE_EXTENT_FLAG_MASK	0xff00000000000000ULL
+#define APFS_FILE_EXTENT_FLAG_SHIFT	56
+#define APFS_FILE_EXTENT_CRYPTO_FLAG	0x01	/* Made-up name */
+#define APFS_FILE_EXTENT_PREALLOCATED	0x02	/* Made-up name */
+#define APFS_VALID_FILE_EXTENT_FLAGS	(APFS_FILE_EXTENT_CRYPTO_FLAG \
+					| APFS_FILE_EXTENT_PREALLOCATED)
+
+/*
+ * Put a bound on maximum file size so that a growing truncation will always
+ * produce a single hole extent, even if 64k block sizes were in use. Larger
+ * file sizes could be supported with multiple extents of course, but it takes
+ * some work and I don't see the point.
+ */
+#define APFS_MAX_FILE_SIZE		0x00ffffffffff0000ULL
+
+/*
+ * Structure of a file extent record
+ */
+struct apfs_file_extent_val {
+	__le64 len_and_flags;
+	__le64 phys_block_num;
+	__le64 crypto_id;
+} __packed;
+
+/*
+ * Structure of a data stream record
+ */
+struct apfs_dstream_id_val {
+	__le32 refcnt;
+} __packed;
+
+#define APFS_CP_MAX_WRAPPEDKEYSIZE	128
+
+/*
+ * Structure used to store the encryption state for PFKs
+ */
+struct apfs_wrapped_crypto_state {
+	__le16 major_version;
+	__le16 minor_version;
+	__le32 cpflags;
+	__le32 persistent_class;
+	__le32 key_os_version;
+	__le16 key_revision;
+	__le16 key_len;
+	u8 persistent_key[];
+} __packed;
+
+/*
+ * Structure of a crypto state record
+ */
+struct apfs_crypto_state_val {
+	__le32 refcnt;
+	struct apfs_wrapped_crypto_state state;
+} __packed;
+
+/* Inode numbers for special inodes */
+#define APFS_INVALID_INO_NUM		0
+
+#define APFS_ROOT_DIR_PARENT		1	/* Root directory parent */
+#define APFS_ROOT_DIR_INO_NUM		2	/* Root directory */
+#define APFS_PRIV_DIR_INO_NUM		3	/* Private directory */
+#define APFS_SNAP_DIR_INO_NUM		6	/* Snapshots metadata */
+#define APFS_PURGEABLE_DIR_INO_NUM	7	/* Parent of purgeable files */
+
+/* Smallest inode number available for user content */
+#define APFS_MIN_USER_INO_NUM		16
+
+#define APFS_UNIFIED_ID_SPACE_MARK	0x0800000000000000
+
+/* Inode internal flags */
+#define APFS_INODE_IS_APFS_PRIVATE		0x00000001
+#define APFS_INODE_MAINTAIN_DIR_STATS		0x00000002
+#define APFS_INODE_DIR_STATS_ORIGIN		0x00000004
+#define APFS_INODE_PROT_CLASS_EXPLICIT		0x00000008
+#define APFS_INODE_WAS_CLONED			0x00000010
+#define APFS_INODE_FLAG_UNUSED			0x00000020
+#define APFS_INODE_HAS_SECURITY_EA		0x00000040
+#define APFS_INODE_BEING_TRUNCATED		0x00000080
+#define APFS_INODE_HAS_FINDER_INFO		0x00000100
+#define APFS_INODE_IS_SPARSE			0x00000200
+#define APFS_INODE_WAS_EVER_CLONED		0x00000400
+#define APFS_INODE_ACTIVE_FILE_TRIMMED		0x00000800
+#define APFS_INODE_PINNED_TO_MAIN		0x00001000
+#define APFS_INODE_PINNED_TO_TIER2		0x00002000
+#define APFS_INODE_HAS_RSRC_FORK		0x00004000
+#define APFS_INODE_NO_RSRC_FORK			0x00008000
+#define APFS_INODE_ALLOCATION_SPILLEDOVER	0x00010000
+#define APFS_INODE_FAST_PROMOTE			0x00020000
+#define APFS_INODE_HAS_UNCOMPRESSED_SIZE	0x00040000
+#define APFS_INODE_IS_PURGEABLE			0x00080000
+#define APFS_INODE_WANTS_TO_BE_PURGEABLE	0x00100000
+#define APFS_INODE_IS_SYNC_ROOT			0x00200000
+#define APFS_INODE_SNAPSHOT_COW_EXEMPTION	0x00400000
+/* This flag is not documented */
+#define APFS_INODE_HAS_PURGEABLE_FLAGS		0x02000000
+
+/* Masks for internal flags */
+#define APFS_VALID_INTERNAL_INODE_FLAGS	(APFS_INODE_IS_APFS_PRIVATE \
+					| APFS_INODE_MAINTAIN_DIR_STATS \
+					| APFS_INODE_DIR_STATS_ORIGIN \
+					| APFS_INODE_PROT_CLASS_EXPLICIT \
+					| APFS_INODE_WAS_CLONED \
+					| APFS_INODE_HAS_SECURITY_EA \
+					| APFS_INODE_BEING_TRUNCATED \
+					| APFS_INODE_HAS_FINDER_INFO \
+					| APFS_INODE_IS_SPARSE \
+					| APFS_INODE_WAS_EVER_CLONED \
+					| APFS_INODE_ACTIVE_FILE_TRIMMED \
+					| APFS_INODE_PINNED_TO_MAIN \
+					| APFS_INODE_PINNED_TO_TIER2 \
+					| APFS_INODE_HAS_RSRC_FORK \
+					| APFS_INODE_NO_RSRC_FORK \
+					| APFS_INODE_ALLOCATION_SPILLEDOVER \
+					| APFS_INODE_FAST_PROMOTE \
+					| APFS_INODE_HAS_UNCOMPRESSED_SIZE \
+					| APFS_INODE_IS_PURGEABLE \
+					| APFS_INODE_WANTS_TO_BE_PURGEABLE \
+					| APFS_INODE_IS_SYNC_ROOT \
+					| APFS_INODE_SNAPSHOT_COW_EXEMPTION \
+					| APFS_INODE_HAS_PURGEABLE_FLAGS)
+#define APFS_INODE_INHERITED_INTERNAL_FLAGS	(APFS_INODE_MAINTAIN_DIR_STATS)
+#define APFS_INDOE_CLONED_INTERNAL_FLAGS	(APFS_INODE_HAS_RSRC_FORK \
+						| APFS_INODE_NO_RSRC_FORK \
+						| APFS_INODE_HAS_FINDER_INFO)
+#define APFS_INODE_PINNED_MASK			(APFS_INODE_PINNED_TO_MAIN \
+						| APFS_INODE_PINNED_TO_TIER2)
+
+/* BSD flags */
+#define APFS_INOBSD_NODUMP			0x00000001
+#define APFS_INOBSD_IMMUTABLE			0x00000002
+#define APFS_INOBSD_APPEND			0x00000004
+#define APFS_INOBSD_COMPRESSED			0x00000020
+
+/*
+ * Structure of an inode as stored as a B-tree value
+ */
+struct apfs_inode_val {
+/*00*/	__le64 parent_id;
+	__le64 private_id;
+/*10*/	__le64 create_time;
+	__le64 mod_time;
+	__le64 change_time;
+	__le64 access_time;
+/*30*/	__le64 internal_flags;
+	union {
+		__le32 nchildren;
+		__le32 nlink;
+	};
+	__le32 default_protection_class;
+/*40*/	__le32 write_generation_counter;
+	__le32 bsd_flags;
+	__le32 owner;
+	__le32 group;
+/*50*/	__le16 mode;
+	__le16 pad1;
+	__le64 uncompressed_size;
+/*5C*/	u8 xfields[];
+} __packed;
+
+/* Extended field types for dentries */
+#define APFS_DREC_EXT_TYPE_SIBLING_ID 1
+
+/* Extended field types for inodes */
+#define APFS_INO_EXT_TYPE_SNAP_XID 1
+#define APFS_INO_EXT_TYPE_DELTA_TREE_OID 2
+#define APFS_INO_EXT_TYPE_DOCUMENT_ID 3
+#define APFS_INO_EXT_TYPE_NAME 4
+#define APFS_INO_EXT_TYPE_PREV_FSIZE 5
+#define APFS_INO_EXT_TYPE_RESERVED_6 6
+#define APFS_INO_EXT_TYPE_FINDER_INFO 7
+#define APFS_INO_EXT_TYPE_DSTREAM 8
+#define APFS_INO_EXT_TYPE_RESERVED_9 9
+#define APFS_INO_EXT_TYPE_DIR_STATS_KEY 10
+#define APFS_INO_EXT_TYPE_FS_UUID 11
+#define APFS_INO_EXT_TYPE_RESERVED_12 12
+#define APFS_INO_EXT_TYPE_SPARSE_BYTES 13
+#define APFS_INO_EXT_TYPE_RDEV 14
+#define APFS_INO_EXT_TYPE_PURGEABLE_FLAGS 15
+#define APFS_INO_EXT_TYPE_ORIG_SYNC_ROOT_ID 16
+
+/* Extended field flags */
+#define APFS_XF_DATA_DEPENDENT		0x01
+#define APFS_XF_DO_NOT_COPY		0x02
+#define APFS_XF_RESERVED_4		0x04
+#define APFS_XF_CHILDREN_INHERIT	0x08
+#define APFS_XF_USER_FIELD		0x10
+#define APFS_XF_SYSTEM_FIELD		0x20
+#define APFS_XF_RESERVED_40		0x40
+#define APFS_XF_RESERVED_80		0x80
+
+/* Constants for extended fields */
+#define APFS_MIN_DOC_ID 3	/* Smallest not reserved document id */
+
+/*
+ * Structure used to store the number and size of an xfield collection.  The
+ * official reference seems to be wrong about @xf_used_data: it's the size of
+ * the xfield values alone, without the metadata.
+ */
+struct apfs_xf_blob {
+	__le16 xf_num_exts;
+	__le16 xf_used_data;
+	u8 xf_data[];
+} __packed;
+
+/*
+ * Structure used to describe an extended field
+ */
+struct apfs_x_field {
+	u8 x_type;
+	u8 x_flags;
+	__le16 x_size;
+} __packed;
+
+/*
+ * Structure used to store information about a data stream
+ */
+struct apfs_dstream {
+	__le64 size;
+	__le64 alloced_size;
+	__le64 default_crypto_id;
+	__le64 total_bytes_written;
+	__le64 total_bytes_read;
+} __packed;
+
+/*
+ * Structure used to store directory information
+ */
+struct apfs_dir_stats_val {
+	__le64 num_children;
+	__le64 total_size;
+	__le64 chained_key;
+	__le64 gen_count;
+} __packed;
+
+/*
+ * Structure of the value for a sibling link record.  These are used to
+ * list the hard links for a given inode.
+ */
+struct apfs_sibling_val {
+	__le64 parent_id;
+	__le16 name_len;
+	u8 name[];
+} __packed;
+
+/*
+ * Structure of the value for a sibling map record.  No idea what these are for.
+ */
+struct apfs_sibling_map_val {
+	__le64 file_id;
+} __packed;
+
+/*
+ * Structure of a key in an object map B-tree
+ */
+struct apfs_omap_key {
+	__le64 ok_oid;
+	__le64 ok_xid;
+} __packed;
+
+/*
+ * Structure of a key in a free-space queue b-tree
+ */
+struct apfs_spaceman_free_queue_key {
+	__le64 sfqk_xid;
+	__le64 sfqk_paddr;
+} __packed;
+
+/* Catalog records types */
+enum {
+	APFS_TYPE_ANY			= 0,
+	APFS_TYPE_SNAP_METADATA		= 1,
+	APFS_TYPE_EXTENT		= 2,
+	APFS_TYPE_INODE			= 3,
+	APFS_TYPE_XATTR			= 4,
+	APFS_TYPE_SIBLING_LINK		= 5,
+	APFS_TYPE_DSTREAM_ID		= 6,
+	APFS_TYPE_CRYPTO_STATE		= 7,
+	APFS_TYPE_FILE_EXTENT		= 8,
+	APFS_TYPE_DIR_REC		= 9,
+	APFS_TYPE_DIR_STATS		= 10,
+	APFS_TYPE_SNAP_NAME		= 11,
+	APFS_TYPE_SIBLING_MAP		= 12,
+	APFS_TYPE_MAX_VALID		= 12,
+	APFS_TYPE_MAX			= 15,
+	APFS_TYPE_INVALID		= 15,
+};
+
+/* Bit masks for the 'obj_id_and_type' field of a key header */
+#define APFS_OBJ_ID_MASK		0x0fffffffffffffffULL
+#define APFS_OBJ_TYPE_MASK		0xf000000000000000ULL
+#define APFS_OBJ_TYPE_SHIFT		60
+
+/* Key header for filesystem-object keys */
+struct apfs_key_header {
+	__le64 obj_id_and_type;
+} __packed;
+
+/*
+ * Structure of the key for a physical extent record
+ */
+struct apfs_phys_ext_key {
+	struct apfs_key_header hdr;
+} __packed;
+
+/*
+ * Structure of the key for an inode record
+ */
+struct apfs_inode_key {
+	struct apfs_key_header hdr;
+} __packed;
+
+/*
+ * Structure of the key for a file extent record
+ */
+struct apfs_file_extent_key {
+	struct apfs_key_header hdr;
+	__le64 logical_addr;
+} __packed;
+
+/*
+ * Structure of the key for a data stream record
+ */
+struct apfs_dstream_id_key {
+	struct apfs_key_header hdr;
+} __packed;
+
+/*
+ * Structure of the key for a crypto state record
+ */
+struct apfs_crypto_state_key {
+	struct apfs_key_header hdr;
+} __packed;
+
+/* Bit masks for the 'name_len_and_hash' field of a directory entry */
+#define APFS_DREC_LEN_MASK	0x000003ff
+#define APFS_DREC_HASH_MASK	0xfffffc00
+#define APFS_DREC_HASH_SHIFT	10
+
+/* We could support bigger filenames, but I don't know if Linux allows it */
+#define APFS_NAME_LEN		NAME_MAX
+
+/* Bit masks for the 'type' field of a directory entry */
+enum {
+	APFS_DREC_TYPE_MASK	= 0x000f,
+	APFS_DREC_RESERVED_10	= 0x0010,
+
+	/* These flags are not documented */
+	APFS_DREC_PURGEABLE_2	= 0x0200,
+	APFS_DREC_PURGEABLE_8	= 0x0800,
+};
+
+#define APFS_DREC_PURGEABLE	(APFS_DREC_PURGEABLE_2 | APFS_DREC_PURGEABLE_8)
+
+/*
+ * Structure of the key for a directory entry - no hash, used on normalization
+ * sensitive volumes
+ */
+struct apfs_drec_key {
+	struct apfs_key_header hdr;
+	__le16 name_len;
+	u8 name[];
+} __packed;
+
+/*
+ * Structure of the key for a directory entry, including a precomputed
+ * hash of its name
+ */
+struct apfs_drec_hashed_key {
+	struct apfs_key_header hdr;
+	__le32 name_len_and_hash;
+	u8 name[];
+} __packed;
+
+/*
+ * Structure of the key for an extended attributes record
+ */
+struct apfs_xattr_key {
+	struct apfs_key_header hdr;
+	__le16 name_len;
+	u8 name[];
+} __packed;
+
+/*
+ * Structure of the key for a snapshot metadata record
+ */
+struct apfs_snap_metadata_key {
+	struct apfs_key_header hdr;
+} __packed;
+
+/* The snapshot name headers always have this placeholder object id */
+#define APFS_SNAP_NAME_OBJ_ID	(~0ULL & APFS_OBJ_ID_MASK)
+
+/*
+ * Structure of the key for a snapshot name record
+ */
+struct apfs_snap_name_key {
+	struct apfs_key_header hdr;
+	__le16 name_len;
+	u8 name[];
+} __packed;
+
+/*
+ * Structure of the key for a sibling link record
+ */
+struct apfs_sibling_link_key {
+	struct apfs_key_header hdr;
+	__le64 sibling_id;
+} __packed;
+
+/*
+ * Structure of the key for a siblink map record
+ */
+struct apfs_sibling_map_key {
+	struct apfs_key_header hdr;
+} __packed;
+
+/*
+ * On-disk allocation info for a chunk of blocks
+ */
+struct apfs_chunk_info {
+	__le64	ci_xid;
+	__le64	ci_addr;
+	__le32	ci_block_count;
+	__le32	ci_free_count;
+	__le64	ci_bitmap_addr;
+} __packed;
+
+/* Constants for the chunk info block */
+#define	APFS_CI_COUNT_MASK		0x000FFFFF
+#define	APFS_CI_COUNT_RESERVED_MASK	0xFFF00000
+
+/*
+ * Structure of a block with an array of chunk allocation info structures
+ */
+struct apfs_chunk_info_block {
+	struct apfs_obj_phys	cib_o;
+	__le32			cib_index;
+	__le32			cib_chunk_info_count;
+	struct apfs_chunk_info	cib_chunk_info[];
+} __packed;
+
+/*
+ * Structure of a block with an array of addresses to chunk information blocks
+ */
+struct apfs_cib_addr_block {
+	struct apfs_obj_phys	cab_o;
+	__le32			cab_index;
+	__le32			cab_cib_count;
+	__le64			cab_cib_addr[];
+} __packed;
+
+/*
+ * On-disk structure for a free queue
+ */
+struct apfs_spaceman_free_queue {
+	__le64	sfq_count;
+	__le64	sfq_tree_oid;
+	__le64	sfq_oldest_xid;
+	__le16	sfq_tree_node_limit;
+	__le16	sfq_pad16;
+	__le32	sfq_pad32;
+	__le64	sfq_reserved;
+} __packed;
+
+/* Indexes for a free queue array */
+enum {
+	APFS_SFQ_IP	= 0,
+	APFS_SFQ_MAIN	= 1,
+	APFS_SFQ_TIER2	= 2,
+	APFS_SFQ_COUNT	= 3
+};
+
+/*
+ * On-disk structure for device allocation information
+ */
+struct apfs_spaceman_device {
+	__le64	sm_block_count;
+	__le64	sm_chunk_count;
+	__le32	sm_cib_count;
+	__le32	sm_cab_count;
+	__le64	sm_free_count;
+	__le32	sm_addr_offset;
+	__le32	sm_reserved;
+	__le64	sm_reserved2;
+} __packed;
+
+/* Indexes for a device array */
+enum {
+	APFS_SD_MAIN	= 0,
+	APFS_SD_TIER2	= 1,
+	APFS_SD_COUNT	= 2
+};
+
+/*
+ * On-disk structure to describe allocation zone boundaries
+ */
+struct apfs_spaceman_allocation_zone_boundaries {
+	__le64	saz_zone_start;
+	__le64	saz_zone_end;
+} __packed;
+
+/* Allocation zone constants */
+#define	APFS_SM_ALLOCZONE_INVALID_END_BOUNDARY		0
+#define	APFS_SM_ALLOCZONE_NUM_PREVIOUS_BOUNDARIES	7
+
+struct apfs_spaceman_allocation_zone_info_phys {
+	struct apfs_spaceman_allocation_zone_boundaries	saz_current_boundaries;
+	struct apfs_spaceman_allocation_zone_boundaries
+	     saz_previous_boundaries[APFS_SM_ALLOCZONE_NUM_PREVIOUS_BOUNDARIES];
+
+	__le16	saz_zone_id;
+	__le16	saz_previous_boundary_index;
+	__le32	saz_reserved;
+} __packed;
+
+/* Datazone constants */
+#define	APFS_SM_DATAZONE_ALLOCZONE_COUNT	8
+
+struct apfs_spaceman_datazone_info_phys {
+	struct apfs_spaceman_allocation_zone_info_phys
+	  sdz_allocation_zones[APFS_SD_COUNT][APFS_SM_DATAZONE_ALLOCZONE_COUNT];
+} __packed;
+
+/* Internal-pool bitmap constants */
+#define	APFS_SPACEMAN_IP_BM_TX_MULTIPLIER	16
+#define	APFS_SPACEMAN_IP_BM_INDEX_INVALID	0xFFFF
+#define	APFS_SPACEMAN_IP_BM_BLOCK_COUNT_MAX	0xFFFE
+
+/* Space manager flags */
+#define	APFS_SM_FLAG_VERSIONED		0x00000001
+#define	APFS_SM_FLAGS_VALID_MASK	APFS_SM_FLAG_VERSIONED
+
+/*
+ * On-disk structure for the space manager
+ */
+struct apfs_spaceman_phys {
+	struct apfs_obj_phys			sm_o;
+	__le32					sm_block_size;
+	__le32					sm_blocks_per_chunk;
+	__le32					sm_chunks_per_cib;
+	__le32					sm_cibs_per_cab;
+	struct apfs_spaceman_device		sm_dev[APFS_SD_COUNT];
+	__le32					sm_flags;
+	__le32					sm_ip_bm_tx_multiplier;
+	__le64					sm_ip_block_count;
+	__le32					sm_ip_bm_size_in_blocks;
+	__le32					sm_ip_bm_block_count;
+	__le64					sm_ip_bm_base;
+	__le64					sm_ip_base;
+	__le64					sm_fs_reserve_block_count;
+	__le64					sm_fs_reserve_alloc_count;
+	struct apfs_spaceman_free_queue		sm_fq[APFS_SFQ_COUNT];
+	__le16					sm_ip_bm_free_head;
+	__le16					sm_ip_bm_free_tail;
+	__le32					sm_ip_bm_xid_offset;
+	__le32					sm_ip_bitmap_offset;
+	__le32					sm_ip_bm_free_next_offset;
+	__le32					sm_version;
+	__le32					sm_struct_size;
+	struct apfs_spaceman_datazone_info_phys	sm_datazone;
+} __packed;
+
+/*
+ * Structure used to store a range of physical blocks
+ */
+struct apfs_prange {
+	__le64 pr_start_paddr;
+	__le64 pr_block_count;
+} __packed;
+
+/* Reaper flags */
+#define	APFS_NR_BHM_FLAG		0x00000001
+#define	APFS_NR_CONTINUE		0x00000002
+#define APFS_NR_FLAGS_VALID_MASK	(APFS_NR_BHM_FLAG | APFS_NR_CONTINUE)
+
+/*
+ * On-disk reaper structure
+ */
+struct apfs_nx_reaper_phys {
+	struct apfs_obj_phys	nr_o;
+	__le64			nr_next_reap_id;
+	__le64			nr_completed_id;
+	__le64			nr_head;
+	__le64			nr_tail;
+	__le32			nr_flags;
+	__le32			nr_rlcount;
+	__le32			nr_type;
+	__le32			nr_size;
+	__le64			nr_fs_oid;
+	__le64			nr_oid;
+	__le64			nr_xid;
+	__le32			nr_nrle_flags;
+	__le32			nr_state_buffer_size;
+	u8			nr_state_buffer[];
+} __packed;
+
+struct apfs_nx_reap_list_entry {
+	__le32	nrle_next;
+	__le32	nrle_flags;
+	__le32	nrle_type;
+	__le32	nrle_size;
+	__le64	nrle_fs_oid;
+	__le64	nrle_oid;
+	__le64	nrle_xid;
+} __packed;
+
+struct apfs_nx_reap_list_phys {
+	struct apfs_obj_phys		nrl_o;
+	__le64				nrl_next;
+	__le32				nrl_flags;
+	__le32				nrl_max;
+	__le32				nrl_count;
+	__le32				nrl_first;
+	__le32				nrl_last;
+	__le32				nrl_free;
+	struct apfs_nx_reap_list_entry	nrl_entries[];
+} __packed;
+
+/* EFI constants */
+#define APFS_NX_EFI_JUMPSTART_MAGIC	0x5244534A
+#define APFS_NX_EFI_JUMPSTART_VERSION	1
+
+/*
+ * Information about the embedded EFI driver
+ */
+struct apfs_nx_efi_jumpstart {
+	struct apfs_obj_phys	nej_o;
+	__le32			nej_magic;
+	__le32			nej_version;
+	__le32			nej_efi_file_len;
+	__le32			nej_num_extents;
+	__le64			nej_reserved[16];
+	struct apfs_prange	nej_rec_extents[];
+} __packed;
+
+/* Main container */
+
+/* Container constants */
+#define APFS_NX_MAGIC				0x4253584E
+#define APFS_NX_BLOCK_NUM			0
+#define APFS_NX_MAX_FILE_SYSTEMS		100
+
+#define APFS_NX_EPH_INFO_COUNT			4
+#define APFS_NX_EPH_MIN_BLOCK_COUNT		8
+#define APFS_NX_MAX_FILE_SYSTEM_EPH_STRUCTS	4
+#define APFS_NX_TX_MIN_CHECKPOINT_COUNT		4
+#define APFS_NX_EPH_INFO_VERSION_1		1
+
+/* Container flags */
+#define APFS_NX_RESERVED_1			0x00000001LL
+#define APFS_NX_RESERVED_2			0x00000002LL
+#define APFS_NX_CRYPTO_SW			0x00000004LL
+#define APFS_NX_FLAGS_VALID_MASK		(APFS_NX_RESERVED_1 \
+						| APFS_NX_RESERVED_2 \
+						| APFS_NX_CRYPTO_SW)
+
+/* Optional container feature flags */
+#define APFS_NX_FEATURE_DEFRAG			0x0000000000000001ULL
+#define APFS_NX_FEATURE_LCFD			0x0000000000000002ULL
+#define APFS_NX_SUPPORTED_FEATURES_MASK		(APFS_NX_FEATURE_DEFRAG | \
+						APFS_NX_FEATURE_LCFD)
+
+/* Read-only compatible container feature flags */
+#define APFS_NX_SUPPORTED_ROCOMPAT_MASK		(0x0ULL)
+
+/* Incompatible container feature flags */
+#define APFS_NX_INCOMPAT_VERSION1		0x0000000000000001ULL
+#define APFS_NX_INCOMPAT_VERSION2		0x0000000000000002ULL
+#define APFS_NX_INCOMPAT_FUSION			0x0000000000000100ULL
+#define APFS_NX_SUPPORTED_INCOMPAT_MASK		(APFS_NX_INCOMPAT_VERSION2 \
+						| APFS_NX_INCOMPAT_FUSION)
+
+/* Block and container sizes */
+#define APFS_NX_MINIMUM_BLOCK_SIZE		4096
+#define APFS_NX_DEFAULT_BLOCK_SIZE		4096
+#define APFS_NX_MAXIMUM_BLOCK_SIZE		65536
+#define APFS_NX_MINIMUM_CONTAINER_SIZE		1048576
+
+/* Indexes into a container superblock's array of counters */
+enum {
+	APFS_NX_CNTR_OBJ_CKSUM_SET	= 0,
+	APFS_NX_CNTR_OBJ_CKSUM_FAIL	= 1,
+
+	APFS_NX_NUM_COUNTERS		= 32
+};
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 13, 0)
+#define UUID_SIZE 16
+#endif
+
+/*
+ * On-disk representation of the container superblock
+ */
+struct apfs_nx_superblock {
+/*00*/	struct apfs_obj_phys nx_o;
+/*20*/	__le32 nx_magic;
+	__le32 nx_block_size;
+	__le64 nx_block_count;
+
+/*30*/	__le64 nx_features;
+	__le64 nx_readonly_compatible_features;
+	__le64 nx_incompatible_features;
+
+/*48*/	char nx_uuid[UUID_SIZE];
+
+/*58*/	__le64 nx_next_oid;
+	__le64 nx_next_xid;
+
+/*68*/	__le32 nx_xp_desc_blocks;
+	__le32 nx_xp_data_blocks;
+/*70*/	__le64 nx_xp_desc_base;
+	__le64 nx_xp_data_base;
+	__le32 nx_xp_desc_next;
+	__le32 nx_xp_data_next;
+/*88*/	__le32 nx_xp_desc_index;
+	__le32 nx_xp_desc_len;
+	__le32 nx_xp_data_index;
+	__le32 nx_xp_data_len;
+
+/*98*/	__le64 nx_spaceman_oid;
+	__le64 nx_omap_oid;
+	__le64 nx_reaper_oid;
+
+/*B0*/	__le32 nx_test_type;
+
+	__le32 nx_max_file_systems;
+/*B8*/	__le64 nx_fs_oid[APFS_NX_MAX_FILE_SYSTEMS];
+/*3D8*/	__le64 nx_counters[APFS_NX_NUM_COUNTERS];
+/*4D8*/	struct apfs_prange nx_blocked_out_prange;
+	__le64 nx_evict_mapping_tree_oid;
+/*4F0*/	__le64 nx_flags;
+	__le64 nx_efi_jumpstart;
+/*500*/	char nx_fusion_uuid[UUID_SIZE];
+	struct apfs_prange nx_keylocker;
+/*520*/	__le64 nx_ephemeral_info[APFS_NX_EPH_INFO_COUNT];
+
+/*540*/	__le64 nx_test_oid;
+
+	__le64 nx_fusion_mt_oid;
+/*550*/	__le64 nx_fusion_wbc_oid;
+	struct apfs_prange nx_fusion_wbc;
+
+	__le64 nx_newest_mounted_version;
+
+/*570*/	struct apfs_prange nx_mkb_locker;
+} __packed;
+
+/*
+ * A mapping from an ephemeral object id to its physical address
+ */
+struct apfs_checkpoint_mapping {
+	__le32 cpm_type;
+	__le32 cpm_subtype;
+	__le32 cpm_size;
+	__le32 cpm_pad;
+	__le64 cpm_fs_oid;
+	__le64 cpm_oid;
+	__le64 cpm_paddr;
+} __packed;
+
+/* Checkpoint flags */
+#define	APFS_CHECKPOINT_MAP_LAST	0x00000001
+
+/*
+ * A checkpoint-mapping block
+ */
+struct apfs_checkpoint_map_phys {
+	struct apfs_obj_phys		cpm_o;
+	__le32				cpm_flags;
+	__le32				cpm_count;
+	struct apfs_checkpoint_mapping	cpm_map[];
+} __packed;
+
+/* Volume */
+
+/* Volume constants */
+#define APFS_MAGIC				0x42535041
+
+#define APFS_MAX_HIST				8
+#define APFS_VOLNAME_LEN			256
+
+/* Volume flags */
+#define APFS_FS_UNENCRYPTED			0x00000001LL
+#define APFS_FS_EFFACEABLE			0x00000002LL
+#define APFS_FS_RESERVED_4			0x00000004LL
+#define APFS_FS_ONEKEY				0x00000008LL
+#define APFS_FS_SPILLEDOVER			0x00000010LL
+#define APFS_FS_RUN_SPILLOVER_CLEANER		0x00000020LL
+#define APFS_FS_ALWAYS_CHECK_EXTENTREF		0x00000040LL
+#define APFS_FS_PREVIOUSLY_SEALED		0x00000080LL /* Made-up name */
+#define APFS_FS_PFK				0x00000100LL /* Made-up name */
+#define APFS_FS_UNKNOWN_200			0x00000200LL
+#define APFS_FS_FLAGS_VALID_MASK		(APFS_FS_UNENCRYPTED \
+						| APFS_FS_EFFACEABLE \
+						| APFS_FS_RESERVED_4 \
+						| APFS_FS_ONEKEY \
+						| APFS_FS_SPILLEDOVER \
+						| APFS_FS_RUN_SPILLOVER_CLEANER \
+						| APFS_FS_ALWAYS_CHECK_EXTENTREF \
+						| APFS_FS_PREVIOUSLY_SEALED \
+						| APFS_FS_PFK \
+						| APFS_FS_UNKNOWN_200)
+
+#define APFS_FS_CRYPTOFLAGS			(APFS_FS_UNENCRYPTED \
+						| APFS_FS_EFFACEABLE \
+						| APFS_FS_ONEKEY)
+
+/* Volume roles */
+#define APFS_VOLUME_ENUM_SHIFT		6
+#define APFS_VOL_ROLE_NONE		0x0000
+#define APFS_VOL_ROLE_SYSTEM		0x0001
+#define APFS_VOL_ROLE_USER		0x0002
+#define APFS_VOL_ROLE_RECOVERY		0x0004
+#define APFS_VOL_ROLE_VM		0x0008
+#define APFS_VOL_ROLE_PREBOOT		0x0010
+#define APFS_VOL_ROLE_INSTALLER		0x0020
+#define APFS_VOL_ROLE_DATA		(1 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_BASEBAND		(2 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_UPDATE		(3 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_XART		(4 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_HARDWARE		(5 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_BACKUP		(6 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_RESERVED_7	(7 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_RESERVED_8	(8 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_ENTERPRISE	(9 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_RESERVED_10	(10 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLE_PRELOGIN		(11 << APFS_VOLUME_ENUM_SHIFT)
+#define APFS_VOL_ROLES_VALID_MASK	(APFS_VOL_ROLE_SYSTEM \
+					| APFS_VOL_ROLE_USER \
+					| APFS_VOL_ROLE_RECOVERY \
+					| APFS_VOL_ROLE_VM \
+					| APFS_VOL_ROLE_PREBOOT \
+					| APFS_VOL_ROLE_INSTALLER \
+					| APFS_VOL_ROLE_DATA \
+					| APFS_VOL_ROLE_BASEBAND \
+					| APFS_VOL_ROLE_UPDATE \
+					| APFS_VOL_ROLE_XART \
+					| APFS_VOL_ROLE_HARDWARE \
+					| APFS_VOL_ROLE_BACKUP \
+					| APFS_VOL_ROLE_RESERVED_7 \
+					| APFS_VOL_ROLE_RESERVED_8 \
+					| APFS_VOL_ROLE_ENTERPRISE \
+					| APFS_VOL_ROLE_RESERVED_10 \
+					| APFS_VOL_ROLE_PRELOGIN)
+
+/* Optional volume feature flags */
+#define APFS_FEATURE_DEFRAG_PRERELEASE		0x00000001LL
+#define APFS_FEATURE_HARDLINK_MAP_RECORDS	0x00000002LL
+#define APFS_FEATURE_DEFRAG			0x00000004LL
+#define APFS_FEATURE_STRICTATIME		0x00000008LL
+#define APFS_FEATURE_VOLGRP_SYSTEM_INO_SPACE	0x00000010LL
+
+#define APFS_SUPPORTED_FEATURES_MASK	(APFS_FEATURE_DEFRAG \
+					| APFS_FEATURE_DEFRAG_PRERELEASE \
+					| APFS_FEATURE_HARDLINK_MAP_RECORDS \
+					| APFS_FEATURE_STRICTATIME \
+					| APFS_FEATURE_VOLGRP_SYSTEM_INO_SPACE)
+
+/* Read-only compatible volume feature flags */
+#define APFS_SUPPORTED_ROCOMPAT_MASK		(0x0ULL)
+
+/* Incompatible volume feature flags */
+#define APFS_INCOMPAT_CASE_INSENSITIVE		0x00000001LL
+#define APFS_INCOMPAT_DATALESS_SNAPS		0x00000002LL
+#define APFS_INCOMPAT_ENC_ROLLED		0x00000004LL
+#define APFS_INCOMPAT_NORMALIZATION_INSENSITIVE	0x00000008LL
+#define APFS_INCOMPAT_INCOMPLETE_RESTORE	0x00000010LL
+#define APFS_INCOMPAT_SEALED_VOLUME		0x00000020LL
+#define APFS_INCOMPAT_PFK			0x00000040LL /* Made-up name */
+#define APFS_INCOMPAT_EXTENT_PREALLOC_FLAG	0x00000080LL /* Made-up name */
+#define APFS_INCOMPAT_SECONDARY_FSROOT		0x00000100LL /* Made-up name */
+
+#define APFS_SUPPORTED_INCOMPAT_MASK (APFS_INCOMPAT_CASE_INSENSITIVE \
+				     | APFS_INCOMPAT_DATALESS_SNAPS \
+				     | APFS_INCOMPAT_ENC_ROLLED \
+				     | APFS_INCOMPAT_NORMALIZATION_INSENSITIVE \
+				     | APFS_INCOMPAT_INCOMPLETE_RESTORE \
+				     | APFS_INCOMPAT_SEALED_VOLUME \
+				     | APFS_INCOMPAT_PFK \
+				     | APFS_INCOMPAT_EXTENT_PREALLOC_FLAG \
+				     | APFS_INCOMPAT_SECONDARY_FSROOT)
+
+#define APFS_MODIFIED_NAMELEN	      32
+
+/*
+ * Structure containing information about a program that modified the volume
+ */
+struct apfs_modified_by {
+	u8 id[APFS_MODIFIED_NAMELEN];
+	__le64 timestamp;
+	__le64 last_xid;
+} __packed;
+
+/* Version constants for wrapped meta crypto state */
+#define APFS_WMCS_MAJOR_VERSION		5
+#define APFS_WMCS_MINOR_VERSION		0
+
+/* Protection classes */
+#define APFS_PROTECTION_CLASS_DIR_NONE	0 /* Inherits the directory's default */
+#define APFS_PROTECTION_CLASS_A		1
+#define APFS_PROTECTION_CLASS_B		2
+#define APFS_PROTECTION_CLASS_C		3
+#define APFS_PROTECTION_CLASS_D		4 /* No protection */
+#define APFS_PROTECTION_CLASS_F		6 /* No protection, nonpersistent key */
+
+/* Encryption identifiers */
+#define APFS_CRYPTO_SW_ID		4
+#define APFS_CRYPTO_RESERVED_5		5
+#define APFS_UNASSIGNED_CRYPTO_ID	(~0ULL)
+
+/* Doc id index flags. I'm making up the names for now. */
+#define APFS_DOC_ID_HAS_PREV_TREE	0x00000001
+#define APFS_DOC_ID_UNKNOWN_02		0x00000002
+#define APFS_DOC_ID_UNKNOWN_04		0x00000004
+#define APFS_DOC_ID_UNKNOWN_08		0x00000008
+#define APFS_DOC_ID_UNKNOWN_10		0x00000010
+#define APFS_DOC_ID_VALID_FLAGS		(APFS_DOC_ID_HAS_PREV_TREE \
+					| APFS_DOC_ID_UNKNOWN_02 \
+					| APFS_DOC_ID_UNKNOWN_04 \
+					| APFS_DOC_ID_UNKNOWN_08 \
+					| APFS_DOC_ID_UNKNOWN_10)
+
+/*
+ * Structure used to store the encryption state
+ */
+struct apfs_wrapped_meta_crypto_state {
+	__le16 major_version;
+	__le16 minor_version;
+	__le32 cpflags;
+	__le32 persistent_class;
+	__le32 key_os_version;
+	__le16 key_revision;
+	__le16 unused;
+} __packed;
+
+/*
+ * On-disk representation of a volume superblock
+ */
+struct apfs_superblock {
+/*00*/	struct apfs_obj_phys apfs_o;
+
+/*20*/	__le32 apfs_magic;
+	__le32 apfs_fs_index;
+
+/*28*/	__le64 apfs_features;
+	__le64 apfs_readonly_compatible_features;
+	__le64 apfs_incompatible_features;
+
+/*40*/	__le64 apfs_unmount_time;
+
+	__le64 apfs_fs_reserve_block_count;
+	__le64 apfs_fs_quota_block_count;
+	__le64 apfs_fs_alloc_count;
+
+/*60*/	struct apfs_wrapped_meta_crypto_state apfs_meta_crypto;
+
+/*74*/	__le32 apfs_root_tree_type;
+	__le32 apfs_extentref_tree_type;
+	__le32 apfs_snap_meta_tree_type;
+
+/*80*/	__le64 apfs_omap_oid;
+	__le64 apfs_root_tree_oid;
+	__le64 apfs_extentref_tree_oid;
+	__le64 apfs_snap_meta_tree_oid;
+
+/*A0*/	__le64 apfs_revert_to_xid;
+	__le64 apfs_revert_to_sblock_oid;
+
+/*B0*/	__le64 apfs_next_obj_id;
+
+/*B8*/	__le64 apfs_num_files;
+	__le64 apfs_num_directories;
+	__le64 apfs_num_symlinks;
+	__le64 apfs_num_other_fsobjects;
+	__le64 apfs_num_snapshots;
+
+/*E0*/	__le64 apfs_total_blocks_alloced;
+	__le64 apfs_total_blocks_freed;
+
+/*F0*/	char apfs_vol_uuid[UUID_SIZE];
+/*100*/	__le64 apfs_last_mod_time;
+
+	__le64 apfs_fs_flags;
+
+/*110*/	struct apfs_modified_by apfs_formatted_by;
+/*140*/	struct apfs_modified_by apfs_modified_by[APFS_MAX_HIST];
+
+/*2C0*/	u8 apfs_volname[APFS_VOLNAME_LEN];
+/*3C0*/	__le32 apfs_next_doc_id;
+
+	__le16 apfs_role;
+	__le16 reserved;
+
+/*3C8*/	__le64 apfs_root_to_xid;
+	__le64 apfs_er_state_oid;
+
+	__le64 apfs_cloneinfo_id_epoch;
+	__le64 apfs_cloneinfo_xid;
+
+	__le64 apfs_snap_meta_ext_oid;
+
+/*3F0*/	char apfs_volume_group_id[UUID_SIZE];
+
+/*400*/	__le64 apfs_integrity_meta_oid;
+
+	__le64 apfs_fext_tree_oid;
+/*410*/	__le32 apfs_fext_tree_type;
+
+	__le32 reserved_type;
+	__le64 reserved_oid;
+
+/*420*/	__le64 apfs_doc_id_index_xid;
+	__le32 apfs_doc_id_index_flags;
+	__le32 apfs_doc_id_tree_type;
+/*430*/	__le64 apfs_doc_id_tree_oid; /* Made-up name */
+	__le64 apfs_prev_doc_id_tree_oid;
+	__le64 apfs_doc_id_fixup_cursor;
+	__le64 apfs_sec_root_tree_oid;
+/*450*/	__le32 apfs_sec_root_tree_type;
+} __packed;
+
+/* Extended attributes constants */
+#define APFS_XATTR_MAX_EMBEDDED_SIZE	3804
+
+/* Extended attributes names */
+#define APFS_XATTR_NAME_SYMLINK		"com.apple.fs.symlink"
+#define APFS_XATTR_NAME_COMPRESSED	"com.apple.decmpfs"
+#define APFS_XATTR_NAME_RSRC_FORK	"com.apple.ResourceFork"
+#define APFS_XATTR_NAME_SECURITY	"com.apple.system.Security"
+#define APFS_XATTR_NAME_FINDER_INFO	"com.apple.FinderInfo"
+
+/* Extended attributes flags */
+enum {
+	APFS_XATTR_DATA_STREAM		= 0x00000001,
+	APFS_XATTR_DATA_EMBEDDED	= 0x00000002,
+	APFS_XATTR_FILE_SYSTEM_OWNED	= 0x00000004,
+	APFS_XATTR_RESERVED_8		= 0x00000008,
+	APFS_XATTR_UNKNOWN_10		= 0x00000010,
+};
+
+#define APFS_XATTR_VALID_FLAGS 0x0000001f
+
+/*
+ * Structure of the value of an extended attributes record
+ */
+struct apfs_xattr_val {
+	__le16 flags;
+	__le16 xdata_len;
+	u8 xdata[];
+} __packed;
+
+/*
+ * Structure used to store the data of an extended attributes record
+ */
+struct apfs_xattr_dstream {
+	__le64 xattr_obj_id;
+	struct apfs_dstream dstream;
+} __packed;
+
+/*
+ * Integrity metadata for a sealed volume
+ */
+struct apfs_integrity_meta_phys {
+	struct apfs_obj_phys im_o;
+
+	__le32	im_version;
+	__le32	im_flags;
+	__le32	im_hash_type;
+	__le32	im_root_hash_offset;
+	__le64	im_broken_xid;
+	__le64	im_reserved[9];
+} __packed;
+
+/*
+ * Version numbers for the integrity metadata structure
+ */
+enum {
+	APFS_INTEGRITY_META_VERSION_INVALID	= 0,
+	APFS_INTEGRITY_META_VERSION_1		= 1,
+	APFS_INTEGRITY_META_VERSION_2		= 2,
+	APFS_INTEGRITY_META_VERSION_HIGHEST	= APFS_INTEGRITY_META_VERSION_2
+};
+
+/* Flags used by integrity metadata */
+#define APFS_SEAL_BROKEN (1U << 0)
+
+/*
+ * Constants used to identify hash algorithms
+ */
+enum {
+	APFS_HASH_INVALID	= 0,
+	APFS_HASH_SHA256	= 0x1,
+	APFS_HASH_SHA512_256	= 0x2,
+	APFS_HASH_SHA384	= 0x3,
+	APFS_HASH_SHA512	= 0x4,
+
+	APFS_HASH_MIN		= APFS_HASH_SHA256,
+	APFS_HASH_MAX		= APFS_HASH_SHA512,
+	APFS_HASH_DEFAULT	= APFS_HASH_SHA256,
+};
+
+#define APFS_HASH_CCSHA256_SIZE		32
+#define APFS_HASH_CCSHA512_256_SIZE	32
+#define APFS_HASH_CCSHA384_SIZE		48
+#define APFS_HASH_CCSHA512_SIZE		64
+
+#define APFS_HASH_MAX_SIZE		64
+
+/*
+ * Structure of a key in a fext tree
+ */
+struct apfs_fext_tree_key {
+	__le64 private_id;
+	__le64 logical_addr;
+} __packed;
+
+/*
+ * Structure of a value in a fext tree
+ */
+struct apfs_fext_tree_val {
+	__le64 len_and_flags;
+	__le64 phys_block_num;
+} __packed;
+
+/*
+ * Structure of the key for a file info record
+ */
+struct apfs_file_info_key {
+	struct apfs_key_header hdr;
+	__le64 info_and_lba;
+} __packed;
+
+#define APFS_FILE_INFO_LBA_MASK		0x00ffffffffffffffULL
+#define APFS_FILE_INFO_TYPE_MASK	0xff00000000000000ULL
+#define APFS_FILE_INFO_TYPE_SHIFT	56
+
+/*
+ * A hash of file data
+ */
+struct apfs_file_data_hash_val {
+	__le16	hashed_len;
+	u8	hash_size;
+	u8	hash[];
+} __packed;
+
+#define APFS_FILE_INFO_DATA_HASH	1
+
+/*
+ * Structure of the value for a file info record
+ */
+struct apfs_file_info_val {
+	union {
+		struct apfs_file_data_hash_val dhash;
+	};
+} __packed;
+
+#define APFS_BTREE_NODE_HASH_SIZE_MAX	64
+
+/*
+ * Structure of the value of an index record for a hashed catalog tree
+ */
+struct apfs_btn_index_node_val {
+	__le64	binv_child_oid;
+	/*
+	 * The reference seems to be wrong about the hash size, at least for
+	 * SHA-256. TODO: check what happens with other hash functions.
+	 */
+	u8	binv_child_hash[APFS_HASH_CCSHA256_SIZE];
+} __packed;
+
+/*
+ * Compressed file header
+ */
+struct apfs_compress_hdr {
+	__le32 signature;
+	__le32 algo;
+	__le64 size;
+} __packed;
+
+#define APFS_COMPRESS_ZLIB_ATTR		3
+#define APFS_COMPRESS_ZLIB_RSRC		4
+#define APFS_COMPRESS_LZVN_ATTR		7
+#define APFS_COMPRESS_LZVN_RSRC		8
+#define APFS_COMPRESS_PLAIN_ATTR	9
+#define APFS_COMPRESS_PLAIN_RSRC	10
+#define APFS_COMPRESS_LZFSE_ATTR	11
+#define APFS_COMPRESS_LZFSE_RSRC	12
+#define APFS_COMPRESS_LZBITMAP_ATTR	13
+#define APFS_COMPRESS_LZBITMAP_RSRC	14
+
+struct apfs_compress_rsrc_hdr {
+	__be32 data_offs;
+	__be32 mgmt_offs;
+	__be32 data_size;
+	__be32 mgmt_size;
+} __packed;
+
+#define APFS_COMPRESS_BLOCK		0x10000
+
+struct apfs_compress_rsrc_data {
+	__le32 unknown;
+	__le32 num;
+	struct apfs_compress_rsrc_block {
+		__le32 offs;
+		__le32 size;
+	} __packed block[];
+} __packed;
+
+#define APFS_FUSION_TIER2_DEVICE_BYTE_ADDR	0x4000000000000000ULL
+
+/*
+ * TODO: this is a placeholder, create some snapshots with the official
+ * implementation to find the actual limit.
+ */
+#define APFS_SNAP_MAX_NAMELEN		255
+
+/*
+ * Structure of the value of a snapshot metadata record
+ */
+struct apfs_snap_metadata_val {
+	__le64 extentref_tree_oid;
+	__le64 sblock_oid;
+	__le64 create_time;
+	__le64 change_time;
+	__le64 inum;
+	__le32 extentref_tree_type;
+	__le32 flags;
+	__le16 name_len;
+	u8 name[];
+} __packed;
+
+/* Snapshot metadata record flags */
+enum {
+	APFS_SNAP_META_PENDING_DATALESS	= 0x00000001,
+	APFS_SNAP_MERGE_IN_PROGRESS	= 0x00000002,
+};
+
+/*
+ * Structure of the value of a snapshot name record
+ */
+struct apfs_snap_name_val {
+	__le64 snap_xid;
+} __packed;
+
+/*
+ * Structure of the extended snapshot metadata
+ */
+struct apfs_snap_meta_ext {
+	struct apfs_obj_phys sme_o;
+
+	__le32	sme_version;
+	__le32	sme_flags;
+	__le64	sme_snap_xid;
+	char	sme_uuid[UUID_SIZE];
+	__le64	sme_token;
+} __packed;
+
+#define APFS_OBJECT_TYPE_KEYBAG	0x6b657973 /* Spells 'syek' */
+
+#define	APFS_VOL_KEYBAG_ENTRY_MAX_SIZE		512
+#define APFS_FV_PERSONAL_RECOVERY_KEY_UUID	"EBC6C064-0000-11AA-AA11-00306543ECAC"
+
+/* Keybag entry types */
+enum {
+	KB_TAG_UNKNOWN			= 0,
+	KB_TAG_RESERVED_1		= 1,
+
+	KB_TAG_VOLUME_KEY		= 2,
+	KB_TAG_VOLUME_UNLOCK_RECORDS	= 3,
+	KB_TAG_VOLUME_PASSPHRASE_HINT	= 4,
+
+	KB_TAG_WRAPPING_M_KEY		= 5,
+	KB_TAG_VOLUME_M_KEY		= 6,
+
+	KB_TAG_RESERVED_F8		= 0xF8
+};
+
+/*
+ * Structure of a single entry in the keybag
+ */
+struct apfs_keybag_entry {
+	char	ke_uuid[UUID_SIZE];
+	__le16	ke_tag;
+	__le16	ke_keylen;
+	__le32	padding;
+	u8	ke_keydata[];
+} __packed;
+
+#define APFS_KEYBAG_VERSION	2
+
+/*
+ * Structure of the locker in the keybag
+ */
+struct apfs_kb_locker {
+	__le16				kl_version;
+	__le16				kl_nkeys;
+	__le32				kl_nbytes;
+	__le64				padding;
+	struct apfs_keybag_entry	kl_entries[];
+} __packed;
+
+#endif	/* _APFS_RAW_H */
diff --git a/fs/apfs/btree.c b/fs/apfs/btree.c
new file mode 100644
index 000000000..e0b4c2998
--- /dev/null
+++ b/fs/apfs/btree.c
@@ -0,0 +1,1174 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/buffer_head.h>
+#include <linux/slab.h>
+#include "apfs.h"
+
+struct apfs_node *apfs_query_root(const struct apfs_query *query)
+{
+	while (query->parent)
+		query = query->parent;
+	ASSERT(apfs_node_is_root(query->node));
+	return query->node;
+}
+
+static u64 apfs_catalog_base_oid(struct apfs_query *query)
+{
+	struct apfs_query *root_query = NULL;
+
+	root_query = query;
+	while (root_query->parent)
+		root_query = root_query->parent;
+
+	return root_query->node->object.oid;
+}
+
+/**
+ * apfs_child_from_query - Read the child id found by a successful nonleaf query
+ * @query:	the query that found the record
+ * @child:	Return parameter.  The child id found.
+ *
+ * Reads the child id in the nonleaf node record into @child and performs a
+ * basic sanity check as a protection against crafted filesystems.  Returns 0
+ * on success or -EFSCORRUPTED otherwise.
+ */
+static int apfs_child_from_query(struct apfs_query *query, u64 *child)
+{
+	struct super_block *sb = query->node->object.sb;
+	char *raw = query->node->object.data;
+
+	if (query->flags & APFS_QUERY_CAT && apfs_is_sealed(sb)) {
+		struct apfs_btn_index_node_val *index_val = NULL;
+
+		if (query->len != sizeof(*index_val)) {
+			apfs_err(sb, "bad sealed index value length (%d)", query->len);
+			return -EFSCORRUPTED;
+		}
+		index_val = (struct apfs_btn_index_node_val *)(raw + query->off);
+		*child = le64_to_cpu(index_val->binv_child_oid) + apfs_catalog_base_oid(query);
+	} else {
+		if (query->len != 8) { /* The value on a nonleaf node is the child id */
+			apfs_err(sb, "bad index value length (%d)", query->len);
+			return -EFSCORRUPTED;
+		}
+		*child = le64_to_cpup((__le64 *)(raw + query->off));
+	}
+	return 0;
+}
+
+/**
+ * apfs_omap_cache_lookup - Look for an oid in an omap's cache
+ * @omap:	the object map
+ * @oid:	object id to look up
+ * @bno:	on return, the block number for the oid
+ *
+ * Returns 0 on success, or -1 if this mapping is not cached.
+ */
+static int apfs_omap_cache_lookup(struct apfs_omap *omap, u64 oid, u64 *bno)
+{
+	struct apfs_omap_cache *cache = &omap->omap_cache;
+	struct apfs_omap_rec *record = NULL;
+	int slot;
+	int ret = -1;
+
+	if (cache->disabled)
+		return -1;
+
+	/* Uninitialized cache records use OID 0, so check this just in case */
+	if (!oid)
+		return -1;
+
+	slot = oid & APFS_OMAP_CACHE_SLOT_MASK;
+	record = &cache->recs[slot];
+
+	spin_lock(&cache->lock);
+	if (record->oid == oid) {
+		*bno = record->bno;
+		ret = 0;
+	}
+	spin_unlock(&cache->lock);
+
+	return ret;
+}
+
+/**
+ * apfs_omap_cache_save - Save a record in an omap's cache
+ * @omap:	the object map
+ * @oid:	object id of the record
+ * @bno:	block number for the oid
+ */
+static void apfs_omap_cache_save(struct apfs_omap *omap, u64 oid, u64 bno)
+{
+	struct apfs_omap_cache *cache = &omap->omap_cache;
+	struct apfs_omap_rec *record = NULL;
+	int slot;
+
+	if (cache->disabled)
+		return;
+
+	slot = oid & APFS_OMAP_CACHE_SLOT_MASK;
+	record = &cache->recs[slot];
+
+	spin_lock(&cache->lock);
+	record->oid = oid;
+	record->bno = bno;
+	spin_unlock(&cache->lock);
+}
+
+/**
+ * apfs_omap_cache_delete - Try to delete a record from an omap's cache
+ * @omap:	the object map
+ * @oid:	object id of the record
+ */
+static void apfs_omap_cache_delete(struct apfs_omap *omap, u64 oid)
+{
+	struct apfs_omap_cache *cache = &omap->omap_cache;
+	struct apfs_omap_rec *record = NULL;
+	int slot;
+
+	if (cache->disabled)
+		return;
+
+	slot = oid & APFS_OMAP_CACHE_SLOT_MASK;
+	record = &cache->recs[slot];
+
+	spin_lock(&cache->lock);
+	if (record->oid == oid) {
+		record->oid = 0;
+		record->bno = 0;
+	}
+	spin_unlock(&cache->lock);
+}
+
+/**
+ * apfs_mounted_xid - Returns the mounted xid for this superblock
+ * @sb:	superblock structure
+ *
+ * This function is needed instead of APFS_NXI(@sb)->nx_xid in situations where
+ * we might be working with a snapshot. Snapshots are read-only and should
+ * mostly ignore xids, so this only appears to matter for omap lookups.
+ */
+static inline u64 apfs_mounted_xid(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+
+	return sbi->s_snap_xid ? sbi->s_snap_xid : nxi->nx_xid;
+}
+
+/**
+ * apfs_xid_in_snapshot - Check if an xid is part of a snapshot
+ * @omap:	the object map
+ * @xid:	the xid to check
+ */
+static inline bool apfs_xid_in_snapshot(struct apfs_omap *omap, u64 xid)
+{
+	return xid <= omap->omap_latest_snap;
+}
+
+/**
+ * apfs_omap_lookup_block_with_xid - Find bno of a virtual object from oid/xid
+ * @sb:		filesystem superblock
+ * @omap:	object map to be searched
+ * @id:		id of the node
+ * @xid:	transaction id
+ * @block:	on return, the found block number
+ * @write:	get write access to the object?
+ *
+ * Searches @omap for the most recent matching object with a transaction id
+ * below @xid. Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_omap_lookup_block_with_xid(struct super_block *sb, struct apfs_omap *omap, u64 id, u64 xid, u64 *block, bool write)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_query *query;
+	struct apfs_omap_map map = {0};
+	int ret = 0;
+
+	if (!write) {
+		if (!apfs_omap_cache_lookup(omap, id, block))
+			return 0;
+	}
+
+	query = apfs_alloc_query(omap->omap_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_omap_key(id, xid, &query->key);
+	query->flags |= APFS_QUERY_OMAP;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		if (ret != -ENODATA)
+			apfs_err(sb, "query failed for oid 0x%llx, xid 0x%llx", id, xid);
+		goto fail;
+	}
+
+	ret = apfs_omap_map_from_query(query, &map);
+	if (ret) {
+		apfs_alert(sb, "bad object map leaf block: 0x%llx",
+			   query->node->object.block_nr);
+		goto fail;
+	}
+	*block = map.bno;
+
+	if (write) {
+		struct apfs_omap_key key;
+		struct apfs_omap_val val;
+		struct buffer_head *new_bh;
+		bool preserve;
+
+		preserve = apfs_xid_in_snapshot(omap, map.xid);
+
+		new_bh = apfs_read_object_block(sb, *block, write, preserve);
+		if (IS_ERR(new_bh)) {
+			apfs_err(sb, "CoW failed for oid 0x%llx, xid 0x%llx", id, xid);
+			ret = PTR_ERR(new_bh);
+			goto fail;
+		}
+
+		key.ok_oid = cpu_to_le64(id);
+		key.ok_xid = cpu_to_le64(nxi->nx_xid);
+		val.ov_flags = cpu_to_le32(map.flags);
+		val.ov_size = cpu_to_le32(sb->s_blocksize);
+		val.ov_paddr = cpu_to_le64(new_bh->b_blocknr);
+
+		if (preserve)
+			ret = apfs_btree_insert(query, &key, sizeof(key), &val, sizeof(val));
+		else
+			ret = apfs_btree_replace(query, &key, sizeof(key), &val, sizeof(val));
+		if (ret)
+			apfs_err(sb, "CoW omap update failed (oid 0x%llx, xid 0x%llx)", id, xid);
+
+		*block = new_bh->b_blocknr;
+		brelse(new_bh);
+	}
+
+	apfs_omap_cache_save(omap, id, *block);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_omap_lookup_block - Find the block number of a b-tree node from its id
+ * @sb:		filesystem superblock
+ * @omap:	object map to be searched
+ * @id:		id of the node
+ * @block:	on return, the found block number
+ * @write:	get write access to the object?
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_omap_lookup_block(struct super_block *sb, struct apfs_omap *omap, u64 id, u64 *block, bool write)
+{
+	return apfs_omap_lookup_block_with_xid(sb, omap, id, apfs_mounted_xid(sb), block, write);
+}
+
+/**
+ * apfs_omap_lookup_newest_block - Find newest bno for a virtual object's oid
+ * @sb:		filesystem superblock
+ * @omap:	object map to be searched
+ * @id:		id of the object
+ * @block:	on return, the found block number
+ * @write:	get write access to the object?
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_omap_lookup_newest_block(struct super_block *sb, struct apfs_omap *omap, u64 id, u64 *block, bool write)
+{
+	return apfs_omap_lookup_block_with_xid(sb, omap, id, -1, block, write);
+}
+
+/**
+ * apfs_create_omap_rec - Create a record in the volume's omap tree
+ * @sb:		filesystem superblock
+ * @oid:	object id
+ * @bno:	block number
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_create_omap_rec(struct super_block *sb, u64 oid, u64 bno)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_omap *omap = sbi->s_omap;
+	struct apfs_query *query;
+	struct apfs_omap_key raw_key;
+	struct apfs_omap_val raw_val;
+	int ret;
+
+	query = apfs_alloc_query(omap->omap_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_omap_key(oid, nxi->nx_xid, &query->key);
+	query->flags |= APFS_QUERY_OMAP;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for oid 0x%llx, bno 0x%llx", oid, bno);
+		goto fail;
+	}
+
+	raw_key.ok_oid = cpu_to_le64(oid);
+	raw_key.ok_xid = cpu_to_le64(nxi->nx_xid);
+	raw_val.ov_flags = 0;
+	raw_val.ov_size = cpu_to_le32(sb->s_blocksize);
+	raw_val.ov_paddr = cpu_to_le64(bno);
+
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key),
+				&raw_val, sizeof(raw_val));
+	if (ret) {
+		apfs_err(sb, "insertion failed for oid 0x%llx, bno 0x%llx", oid, bno);
+		goto fail;
+	}
+
+	apfs_omap_cache_save(omap, oid, bno);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_delete_omap_rec - Delete an existing record from the volume's omap tree
+ * @sb:		filesystem superblock
+ * @oid:	object id for the record
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_delete_omap_rec(struct super_block *sb, u64 oid)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_omap *omap = sbi->s_omap;
+	struct apfs_query *query;
+	int ret;
+
+	query = apfs_alloc_query(omap->omap_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_omap_key(oid, nxi->nx_xid, &query->key);
+	query->flags |= APFS_QUERY_OMAP;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret == -ENODATA) {
+		apfs_err(sb, "nonexistent record (oid 0x%llx)", oid);
+		ret = -EFSCORRUPTED;
+		goto fail;
+	}
+	if (ret) {
+		apfs_err(sb, "query failed (oid 0x%llx)", oid);
+		goto fail;
+	}
+	ret = apfs_btree_remove(query);
+	if (ret) {
+		apfs_err(sb, "removal failed (oid 0x%llx)", oid);
+		goto fail;
+	}
+	apfs_omap_cache_delete(omap, oid);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_alloc_query - Allocates a query structure
+ * @node:	node to be searched
+ * @parent:	query for the parent node
+ *
+ * Callers other than apfs_btree_query() should set @parent to NULL, and @node
+ * to the root of the b-tree. They should also initialize most of the query
+ * fields themselves; when @parent is not NULL the query will inherit them.
+ *
+ * Returns the allocated query, or NULL in case of failure.
+ */
+struct apfs_query *apfs_alloc_query(struct apfs_node *node,
+				    struct apfs_query *parent)
+{
+	struct apfs_query *query;
+
+	query = kzalloc(sizeof(*query), GFP_KERNEL);
+	if (!query)
+		return NULL;
+
+	/* To be released by free_query. */
+	query->node = node;
+
+	if (parent) {
+		query->key = parent->key;
+		query->flags = parent->flags & ~(APFS_QUERY_DONE | APFS_QUERY_NEXT);
+		query->parent = parent;
+		query->depth = parent->depth + 1;
+	}
+
+	/*
+	 * We start the search with the last record and go backwards, but
+	 * some queries later use the PREV flag later to list them in order.
+	 */
+	if (query->flags & APFS_QUERY_PREV)
+		query->index = -1;
+	else
+		query->index = node->records;
+
+	return query;
+}
+
+/**
+ * apfs_free_query - Free a query structure
+ * @query: query to free
+ *
+ * Also frees the ancestor queries, if they are kept.
+ */
+void apfs_free_query(struct apfs_query *query)
+{
+	while (query) {
+		struct apfs_query *parent = query->parent;
+
+		/* The caller decides whether to free the root node */
+		if (query->depth != 0)
+			apfs_node_free(query->node);
+		kfree(query);
+		query = parent;
+	}
+}
+
+/**
+ * apfs_query_set_before_first - Set the query to point before the first record
+ * @sb:		superblock structure
+ * @query:	the query to set
+ *
+ * Queries set in this way are used to insert a record before the first one.
+ * Only the leaf gets set to the -1 entry; queries for other levels must be set
+ * to 0, since the first entry in each index node will need to be modified.
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_query_set_before_first(struct super_block *sb, struct apfs_query **query)
+{
+	struct apfs_node *node = NULL;
+	struct apfs_query *parent = NULL;
+	u64 child_id;
+	u32 storage = apfs_query_storage(*query);
+	int err;
+
+	while ((*query)->depth < 12) {
+		if (apfs_node_is_leaf((*query)->node)) {
+			(*query)->index = -1;
+			return 0;
+		}
+		apfs_node_query_first(*query);
+
+		err = apfs_child_from_query(*query, &child_id);
+		if (err) {
+			apfs_alert(sb, "bad index block: 0x%llx",
+				   (*query)->node->object.block_nr);
+			return err;
+		}
+
+		/* Now go a level deeper */
+		node = apfs_read_node(sb, child_id, storage, false /* write */);
+		if (IS_ERR(node)) {
+			apfs_err(sb, "failed to read child 0x%llx of node 0x%llx", child_id, (*query)->node->object.oid);
+			return PTR_ERR(node);
+		}
+
+		parent = *query;
+		*query = apfs_alloc_query(node, parent);
+		if (!*query) {
+			apfs_node_free(node);
+			*query = parent;
+			return -ENOMEM;
+		}
+		node = NULL;
+	}
+
+	apfs_err(sb, "btree is too high");
+	return -EFSCORRUPTED;
+}
+
+/**
+ * apfs_btree_query - Execute a query on a b-tree
+ * @sb:		filesystem superblock
+ * @query:	the query to execute
+ *
+ * Searches the b-tree starting at @query->index in @query->node, looking for
+ * the record corresponding to @query->key.
+ *
+ * Returns 0 in case of success and sets the @query->len, @query->off and
+ * @query->index fields to the results of the query. @query->node will now
+ * point to the leaf node holding the record.
+ *
+ * In case of failure returns an appropriate error code.
+ */
+int apfs_btree_query(struct super_block *sb, struct apfs_query **query)
+{
+	struct apfs_node *node;
+	struct apfs_query *parent;
+	u64 child_id;
+	u32 storage = apfs_query_storage(*query);
+	int err;
+
+next_node:
+	if ((*query)->depth >= 12) {
+		/*
+		 * We need a maximum depth for the tree so we can't loop
+		 * forever if the filesystem is damaged. 12 should be more
+		 * than enough to map every block.
+		 */
+		apfs_err(sb, "btree is too high");
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	err = apfs_node_query(sb, *query);
+	if (err == -ENODATA && !(*query)->parent && (*query)->index == -1) {
+		/*
+		 * We may be trying to insert a record before all others: don't
+		 * let the query give up at the root node.
+		 */
+		err = apfs_query_set_before_first(sb, query);
+		if (err) {
+			apfs_err(sb, "failed to set before the first record");
+			goto fail;
+		}
+		err = -ENODATA;
+		goto fail;
+	} else if (err == -EAGAIN) {
+		if (!(*query)->parent) {
+			/* We are at the root of the tree */
+			err = -ENODATA;
+			goto fail;
+		}
+
+		/* Move back up one level and continue the query */
+		parent = (*query)->parent;
+		(*query)->parent = NULL; /* Don't free the parent */
+		apfs_free_query(*query);
+		*query = parent;
+		goto next_node;
+	} else if (err) {
+		goto fail;
+	}
+	if (apfs_node_is_leaf((*query)->node)) /* All done */
+		return 0;
+
+	err = apfs_child_from_query(*query, &child_id);
+	if (err) {
+		apfs_alert(sb, "bad index block: 0x%llx",
+			   (*query)->node->object.block_nr);
+		goto fail;
+	}
+
+	/* Now go a level deeper and search the child */
+	node = apfs_read_node(sb, child_id, storage, false /* write */);
+	if (IS_ERR(node)) {
+		apfs_err(sb, "failed to read node 0x%llx", child_id);
+		err = PTR_ERR(node);
+		goto fail;
+	}
+
+	if (node->object.oid != child_id)
+		apfs_debug(sb, "corrupt b-tree");
+
+	/*
+	 * Remember the parent node and index in case the search needs
+	 * to be continued later.
+	 */
+	parent = *query;
+	*query = apfs_alloc_query(node, parent);
+	if (!*query) {
+		apfs_node_free(node);
+		*query = parent;
+		err = -ENOMEM;
+		goto fail;
+	}
+	node = NULL;
+	goto next_node;
+
+fail:
+	/* Don't leave stale record info here or some callers will use it */
+	(*query)->key_len = (*query)->len = 0;
+	return err;
+}
+
+static int __apfs_btree_replace(struct apfs_query *query, void *key, int key_len, void *val, int val_len);
+
+/**
+ * apfs_query_join_transaction - Add the found node to the current transaction
+ * @query: query that found the node
+ */
+int apfs_query_join_transaction(struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	u64 oid = node->object.oid;
+	u32 storage = apfs_query_storage(query);
+	struct apfs_obj_phys *raw = NULL;
+
+	/*
+	 * Ephemeral objects are checkpoint data, and all of their xids get
+	 * updated on commit. There is no real need to do it here as well, but
+	 * it's better for consistency with the other object types.
+	 */
+	if (storage == APFS_OBJ_EPHEMERAL) {
+		ASSERT(node->object.ephemeral);
+		raw = (void *)node->object.data;
+		raw->o_xid = cpu_to_le64(APFS_NXI(sb)->nx_xid);
+		return 0;
+	}
+
+	if (buffer_trans(node->object.o_bh)) /* Already in the transaction */
+		return 0;
+	/* Root nodes should join the transaction before the query is created */
+	ASSERT(!apfs_node_is_root(node));
+
+	node = apfs_read_node(sb, oid, storage, true /* write */);
+	if (IS_ERR(node)) {
+		apfs_err(sb, "Cow failed for node 0x%llx", oid);
+		return PTR_ERR(node);
+	}
+	apfs_node_free(query->node);
+	query->node = node;
+
+	if (storage == APFS_OBJ_PHYSICAL && query->parent) {
+		__le64 bno = cpu_to_le64(node->object.block_nr);
+
+		/* The parent node needs to report the new location */
+		return __apfs_btree_replace(query->parent, NULL /* key */, 0 /* key_len */, &bno, sizeof(bno));
+	}
+	return 0;
+}
+
+/**
+ * apfs_btree_change_rec_count - Update the b-tree info before a record change
+ * @query:	query used to insert/remove/replace the leaf record
+ * @change:	change in the record count
+ * @key_len:	length of the new leaf record key (0 if removed or unchanged)
+ * @val_len:	length of the new leaf record value (0 if removed or unchanged)
+ *
+ * Don't call this function if @query->parent was reset to NULL, or if the same
+ * is true of any of its ancestor queries.
+ */
+static void apfs_btree_change_rec_count(struct apfs_query *query, int change,
+					int key_len, int val_len)
+{
+	struct super_block *sb;
+	struct apfs_node *root;
+	struct apfs_btree_node_phys *root_raw;
+	struct apfs_btree_info *info;
+
+	if (change == -1)
+		ASSERT(!key_len && !val_len);
+	ASSERT(apfs_node_is_leaf(query->node));
+
+	root = apfs_query_root(query);
+	ASSERT(apfs_node_is_root(root));
+
+	sb = root->object.sb;
+	root_raw = (void *)root->object.data;
+	info = (void *)root_raw + sb->s_blocksize - sizeof(*info);
+
+	apfs_assert_in_transaction(sb, &root_raw->btn_o);
+	if (key_len > le32_to_cpu(info->bt_longest_key))
+		info->bt_longest_key = cpu_to_le32(key_len);
+	if (val_len > le32_to_cpu(info->bt_longest_val))
+		info->bt_longest_val = cpu_to_le32(val_len);
+	le64_add_cpu(&info->bt_key_count, change);
+}
+
+/**
+ * apfs_btree_change_node_count - Change the node count for a b-tree
+ * @query:	query used to remove/create the node
+ * @change:	change in the node count
+ *
+ * Also changes the node count in the volume superblock.  Don't call this
+ * function if @query->parent was reset to NULL, or if the same is true of
+ * any of its ancestor queries.
+ */
+void apfs_btree_change_node_count(struct apfs_query *query, int change)
+{
+	struct super_block *sb;
+	struct apfs_node *root;
+	struct apfs_btree_node_phys *root_raw;
+	struct apfs_btree_info *info;
+
+	root = apfs_query_root(query);
+	ASSERT(apfs_node_is_root(root));
+
+	sb = root->object.sb;
+	root_raw = (void *)root->object.data;
+	info = (void *)root_raw + sb->s_blocksize - sizeof(*info);
+
+	apfs_assert_in_transaction(sb, &root_raw->btn_o);
+	le64_add_cpu(&info->bt_node_count, change);
+}
+
+/**
+ * apfs_query_refresh - Recreate a catalog query invalidated by node splits
+ * @old_query:	query chain to refresh
+ * @root:	root node of the query chain
+ * @nodata:	is the query expected to find nothing?
+ *
+ * On success, @old_query is left pointing to the same leaf record, but with
+ * valid ancestor queries as well. Returns a negative error code in case of
+ * failure, or 0 on success.
+ */
+static int apfs_query_refresh(struct apfs_query *old_query, struct apfs_node *root, bool nodata)
+{
+	struct super_block *sb = NULL;
+	struct apfs_query *new_query = NULL;
+	int err = 0;
+
+	sb = root->object.sb;
+
+	if (!apfs_node_is_leaf(old_query->node)) {
+		apfs_alert(sb, "attempting refresh of non-leaf query");
+		return -EFSCORRUPTED;
+	}
+	if (apfs_node_is_root(old_query->node)) {
+		apfs_alert(sb, "attempting refresh of root query");
+		return -EFSCORRUPTED;
+	}
+
+	new_query = apfs_alloc_query(root, NULL /* parent */);
+	if (!new_query)
+		return -ENOMEM;
+	new_query->key = old_query->key;
+	new_query->flags = old_query->flags & ~(APFS_QUERY_DONE | APFS_QUERY_NEXT);
+
+	err = apfs_btree_query(sb, &new_query);
+	if (!nodata && err == -ENODATA) {
+		apfs_err(sb, "record should exist");
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+	if (err && err != -ENODATA) {
+		apfs_err(sb, "failed to rerun");
+		goto fail;
+	}
+	err = 0;
+
+	/* Replace the parent of the original query with the new valid one */
+	apfs_free_query(old_query->parent);
+	old_query->parent = new_query->parent;
+	new_query->parent = NULL;
+
+	/*
+	 * The records may have moved around so update this too. TODO: rework
+	 * the query struct so this stuff is not needed.
+	 */
+	ASSERT(old_query->node->object.oid == new_query->node->object.oid);
+	old_query->index = new_query->index;
+	old_query->key_off = new_query->key_off;
+	old_query->key_len = new_query->key_len;
+	old_query->off = new_query->off;
+	old_query->len = new_query->len;
+	old_query->depth = new_query->depth;
+
+fail:
+	apfs_free_query(new_query);
+	return err;
+}
+
+/**
+ * __apfs_btree_insert - Insert a new record into a b-tree (at any level)
+ * @query:	query run to search for the record
+ * @key:	on-disk record key
+ * @key_len:	length of @key
+ * @val:	on-disk record value (NULL for ghost records)
+ * @val_len:	length of @val (0 for ghost records)
+ *
+ * The new record is placed right after the one found by @query.  On success,
+ * returns 0 and sets @query to the new record; returns a negative error code
+ * in case of failure, which may be -EAGAIN if a split happened and the caller
+ * must retry.
+ */
+int __apfs_btree_insert(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw;
+	int needed_room;
+	int err;
+
+	apfs_assert_query_is_valid(query);
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+
+	node = query->node;
+	node_raw = (void *)node->object.data;
+	apfs_assert_in_transaction(node->object.sb, &node_raw->btn_o);
+
+	needed_room = key_len + val_len;
+	if (!apfs_node_has_room(node, needed_room, false /* replace */)) {
+		if (node->records == 1) {
+			/* The new record just won't fit in the node */
+			err = apfs_create_single_rec_node(query, key, key_len, val, val_len);
+			if (err && err != -EAGAIN)
+				apfs_err(sb, "failed to create single-record node");
+			return err;
+		}
+		err = apfs_node_split(query);
+		if (err && err != -EAGAIN) {
+			apfs_err(sb, "node split failed");
+			return err;
+		}
+		return -EAGAIN;
+	}
+
+	apfs_assert_query_is_valid(query);
+
+	if (query->parent && query->index == -1) {
+		/* We are about to insert a record before all others */
+		err = __apfs_btree_replace(query->parent, key, key_len, NULL /* val */, 0 /* val_len */);
+		if (err) {
+			if (err != -EAGAIN)
+				apfs_err(sb, "parent update failed");
+			return err;
+		}
+	}
+
+	apfs_assert_query_is_valid(query);
+
+	err = apfs_node_insert(query, key, key_len, val, val_len);
+	if (err) {
+		apfs_err(sb, "node record insertion failed");
+		return err;
+	}
+	return 0;
+}
+
+/**
+ * apfs_btree_insert - Insert a new record into a b-tree leaf
+ * @query:	query run to search for the record
+ * @key:	on-disk record key
+ * @key_len:	length of @key
+ * @val:	on-disk record value (NULL for ghost records)
+ * @val_len:	length of @val (0 for ghost records)
+ *
+ * The new record is placed right after the one found by @query.  On success,
+ * returns 0 and sets @query to the new record; returns a negative error code
+ * in case of failure.
+ */
+int apfs_btree_insert(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct super_block *sb = NULL;
+	struct apfs_node *root = NULL, *leaf = NULL;
+	int err;
+
+	root = apfs_query_root(query);
+	ASSERT(apfs_node_is_root(root));
+	leaf = query->node;
+	ASSERT(apfs_node_is_leaf(leaf));
+	sb = root->object.sb;
+
+	while (true) {
+		err = __apfs_btree_insert(query, key, key_len, val, val_len);
+		if (err != -EAGAIN) {
+			if (err)
+				return err;
+			break;
+		}
+		err = apfs_query_refresh(query, root, true /* nodata */);
+		if (err) {
+			apfs_err(sb, "query refresh failed");
+			return err;
+		}
+	}
+
+	apfs_assert_query_is_valid(query);
+	apfs_btree_change_rec_count(query, 1 /* change */, key_len, val_len);
+	return 0;
+}
+
+/**
+ * __apfs_btree_remove - Remove a record from a b-tree (at any level)
+ * @query:	exact query that found the record
+ *
+ * Returns 0 on success, or a negative error code in case of failure, which may
+ * be -EAGAIN if a split happened and the caller must retry.
+ */
+static int __apfs_btree_remove(struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw;
+	int later_entries = node->records - query->index - 1;
+	int err;
+
+	apfs_assert_query_is_valid(query);
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+
+	node = query->node;
+	node_raw = (void *)query->node->object.data;
+	apfs_assert_in_transaction(node->object.sb, &node_raw->btn_o);
+
+	if (query->parent && node->records == 1) {
+		/* Just get rid of the node */
+		err = __apfs_btree_remove(query->parent);
+		if (err == -EAGAIN)
+			return -EAGAIN;
+		if (err) {
+			apfs_err(sb, "parent index removal failed");
+			return err;
+		}
+		apfs_btree_change_node_count(query, -1 /* change */);
+		err = apfs_delete_node(node, query->flags & APFS_QUERY_TREE_MASK);
+		if (err) {
+			apfs_err(sb, "node deletion failed");
+			return err;
+		}
+		return 0;
+	}
+
+	apfs_assert_query_is_valid(query);
+
+	/* The first key in a node must match the parent record's */
+	if (query->parent && query->index == 0) {
+		int first_key_len, first_key_off;
+		void *key;
+
+		first_key_len = apfs_node_locate_key(node, 1, &first_key_off);
+		if (!first_key_len)
+			return -EFSCORRUPTED;
+		key = (void *)node_raw + first_key_off;
+
+		err = __apfs_btree_replace(query->parent, key, first_key_len, NULL /* val */, 0 /* val_len */);
+		if (err == -EAGAIN)
+			return -EAGAIN;
+		if (err) {
+			apfs_err(sb, "parent update failed");
+			return err;
+		}
+	}
+
+	apfs_assert_query_is_valid(query);
+
+	/* Remove the entry from the table of contents */
+	if (apfs_node_has_fixed_kv_size(node)) {
+		struct apfs_kvoff *toc_entry;
+
+		toc_entry = (struct apfs_kvoff *)node_raw->btn_data +
+								query->index;
+		memmove(toc_entry, toc_entry + 1,
+			later_entries * sizeof(*toc_entry));
+	} else {
+		struct apfs_kvloc *toc_entry;
+
+		toc_entry = (struct apfs_kvloc *)node_raw->btn_data +
+								query->index;
+		memmove(toc_entry, toc_entry + 1,
+			later_entries * sizeof(*toc_entry));
+	}
+
+	apfs_node_free_range(node, query->key_off, query->key_len);
+	apfs_node_free_range(node, query->off, query->len);
+
+	--node->records;
+	if (node->records == 0) {
+		/* All descendants are gone, root is the whole tree */
+		node_raw->btn_level = 0;
+		node->flags |= APFS_BTNODE_LEAF;
+	}
+	apfs_update_node(node);
+
+	--query->index;
+	return 0;
+}
+
+/**
+ * apfs_btree_remove - Remove a record from a b-tree leaf
+ * @query:	exact query that found the record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_btree_remove(struct apfs_query *query)
+{
+	struct super_block *sb = NULL;
+	struct apfs_node *root = NULL, *leaf = NULL;
+	int err;
+
+	root = apfs_query_root(query);
+	ASSERT(apfs_node_is_root(root));
+	leaf = query->node;
+	ASSERT(apfs_node_is_leaf(leaf));
+	sb = root->object.sb;
+
+	while (true) {
+		err = __apfs_btree_remove(query);
+		if (err != -EAGAIN) {
+			if (err)
+				return err;
+			break;
+		}
+		err = apfs_query_refresh(query, root, false /* nodata */);
+		if (err) {
+			apfs_err(sb, "query refresh failed");
+			return err;
+		}
+	}
+
+	apfs_assert_query_is_valid(query);
+	apfs_btree_change_rec_count(query, -1 /* change */, 0 /* key_len */, 0 /* val_len */);
+	return 0;
+}
+
+/**
+ * __apfs_btree_replace - Replace a record in a b-tree (at any level)
+ * @query:	exact query that found the record
+ * @key:	new on-disk record key (NULL if unchanged)
+ * @key_len:	length of @key
+ * @val:	new on-disk record value (NULL if unchanged)
+ * @val_len:	length of @val
+ *
+ * It's important that the order of the records is not changed by the new @key.
+ * This function is not needed to replace an old value with a new one of the
+ * same length: it can just be overwritten in place.
+ *
+ * Returns 0 on success, and @query is left pointing to the same record; returns
+ * a negative error code in case of failure, which may be -EAGAIN if a split
+ * happened and the caller must retry.
+ */
+static int __apfs_btree_replace(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw;
+	int needed_room;
+	int err;
+
+	ASSERT(key || val);
+	apfs_assert_query_is_valid(query);
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+
+	node = query->node;
+	node_raw = (void *)node->object.data;
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+
+	needed_room = key_len + val_len;
+	/* We can reuse the space of the replaced key/value */
+	if (key)
+		needed_room -= query->key_len;
+	if (val)
+		needed_room -= query->len;
+
+	if (!apfs_node_has_room(node, needed_room, true /* replace */)) {
+		if (node->records == 1) {
+			apfs_alert(sb, "no room in empty node?");
+			return -EFSCORRUPTED;
+		}
+		err = apfs_node_split(query);
+		if (err && err != -EAGAIN) {
+			apfs_err(sb, "node split failed");
+			return err;
+		}
+		return -EAGAIN;
+	}
+
+	apfs_assert_query_is_valid(query);
+
+	/* The first key in a node must match the parent record's */
+	if (key && query->parent && query->index == 0) {
+		err = __apfs_btree_replace(query->parent, key, key_len, NULL /* val */, 0 /* val_len */);
+		if (err) {
+			if (err != -EAGAIN)
+				apfs_err(sb, "parent update failed");
+			return err;
+		}
+	}
+
+	apfs_assert_query_is_valid(query);
+
+	err = apfs_node_replace(query, key, key_len, val, val_len);
+	if (err) {
+		apfs_err(sb, "node record replacement failed");
+		return err;
+	}
+	return 0;
+}
+
+/**
+ * apfs_btree_replace - Replace a record in a b-tree leaf
+ * @query:	exact query that found the record
+ * @key:	new on-disk record key (NULL if unchanged)
+ * @key_len:	length of @key
+ * @val:	new on-disk record value (NULL if unchanged)
+ * @val_len:	length of @val
+ *
+ * It's important that the order of the records is not changed by the new @key.
+ * This function is not needed to replace an old value with a new one of the
+ * same length: it can just be overwritten in place.
+ *
+ * Returns 0 on success, and @query is left pointing to the same record; returns
+ * a negative error code in case of failure.
+ */
+int apfs_btree_replace(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct super_block *sb = NULL;
+	struct apfs_node *root = NULL, *leaf = NULL;
+	int err;
+
+	root = apfs_query_root(query);
+	ASSERT(apfs_node_is_root(root));
+	leaf = query->node;
+	ASSERT(apfs_node_is_leaf(leaf));
+	sb = root->object.sb;
+
+	while (true) {
+		err = __apfs_btree_replace(query, key, key_len, val, val_len);
+		if (err != -EAGAIN) {
+			if (err)
+				return err;
+			break;
+		}
+		err = apfs_query_refresh(query, root, false /* nodata */);
+		if (err) {
+			apfs_err(sb, "query refresh failed");
+			return err;
+		}
+	}
+
+	apfs_assert_query_is_valid(query);
+	apfs_btree_change_rec_count(query, 0 /* change */, key_len, val_len);
+	return 0;
+}
+
+/**
+ * apfs_query_direct_forward - Set a query to start listing records forwards
+ * @query: a successfully executed query
+ *
+ * Multiple queries list records backwards, but queries marked with this
+ * function after execution will go in the opposite direction.
+ */
+void apfs_query_direct_forward(struct apfs_query *query)
+{
+	if (query->flags & APFS_QUERY_PREV)
+		return;
+
+	apfs_assert_query_is_valid(query);
+	ASSERT(apfs_node_is_leaf(query->node));
+
+	while (query) {
+		query->flags |= APFS_QUERY_PREV;
+		query = query->parent;
+	}
+}
diff --git a/fs/apfs/compress.c b/fs/apfs/compress.c
new file mode 100644
index 000000000..2e05e7a02
--- /dev/null
+++ b/fs/apfs/compress.c
@@ -0,0 +1,478 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020 Corellium LLC
+ */
+
+#include <linux/slab.h>
+#include <linux/zlib.h>
+#include <linux/mutex.h>
+
+#include "apfs.h"
+#include "libzbitmap.h"
+#include "lzfse/lzfse.h"
+#include "lzfse/lzvn_decode_base.h"
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 12, 0)
+
+#include <linux/vmalloc.h>
+
+static inline void *kvmalloc(size_t size, gfp_t flags)
+{
+	gfp_t kmalloc_flags = flags;
+	void *ret;
+
+	if ((flags & GFP_KERNEL) != GFP_KERNEL)
+		return kmalloc(size, flags);
+
+	if (size > PAGE_SIZE)
+		kmalloc_flags |= __GFP_NOWARN | __GFP_NORETRY;
+
+	ret = kmalloc(size, flags);
+	if (ret || size < PAGE_SIZE)
+		return ret;
+
+	return vmalloc(size);
+}
+
+#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(4, 12, 0) */
+
+struct apfs_compress_file_data {
+	struct apfs_compress_hdr hdr;
+	u8 *buf;
+	ssize_t bufblk;
+	size_t bufsize;
+	struct mutex mtx;
+	struct super_block *sb;
+	struct apfs_compressed_data cdata;
+};
+
+static inline int apfs_compress_is_rsrc(u32 algo)
+{
+	return (algo & 1) == 0;
+}
+
+static inline bool apfs_compress_is_supported(u32 algo)
+{
+	switch (algo) {
+	case APFS_COMPRESS_ZLIB_RSRC:
+	case APFS_COMPRESS_ZLIB_ATTR:
+	case APFS_COMPRESS_LZVN_RSRC:
+	case APFS_COMPRESS_LZVN_ATTR:
+	case APFS_COMPRESS_PLAIN_RSRC:
+	case APFS_COMPRESS_PLAIN_ATTR:
+	case APFS_COMPRESS_LZFSE_RSRC:
+	case APFS_COMPRESS_LZFSE_ATTR:
+	case APFS_COMPRESS_LZBITMAP_RSRC:
+	case APFS_COMPRESS_LZBITMAP_ATTR:
+		return true;
+	default:
+		/* Once will usually be enough, don't flood the console */
+		pr_err_once("APFS: unsupported compression algorithm (%u)\n", algo);
+		return false;
+	}
+}
+
+static int apfs_compress_file_open(struct inode *inode, struct file *filp)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_compress_file_data *fd;
+	ssize_t res;
+	bool is_rsrc;
+
+	/*
+	 * The official implementation seems to transparently decompress files
+	 * when you write to them. Doing that atomically inside the kernel is
+	 * probably a chore, so for now I'll just leave it to the user to make
+	 * an uncompressed copy themselves and replace the original. I might
+	 * fix this in the future, but only if people complain (TODO).
+	 */
+	if (filp->f_mode & FMODE_WRITE) {
+		apfs_warn(sb, "writes to compressed files are not supported");
+		apfs_warn(sb, "you can work with a copy of the file instead");
+		return -EOPNOTSUPP;
+	}
+
+	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
+		return -EOVERFLOW;
+
+	fd = kzalloc(sizeof(*fd), GFP_KERNEL);
+	if (!fd)
+		return -ENOMEM;
+	mutex_init(&fd->mtx);
+	fd->sb = sb;
+
+	down_read(&nxi->nx_big_sem);
+
+	res = ____apfs_xattr_get(inode, APFS_XATTR_NAME_COMPRESSED, &fd->hdr, sizeof(fd->hdr), 0);
+	if (res != sizeof(fd->hdr)) {
+		apfs_err(sb, "decmpfs header read failed");
+		goto fail;
+	}
+
+	if (!apfs_compress_is_supported(le32_to_cpu(fd->hdr.algo))) {
+		res = -EOPNOTSUPP;
+		goto fail;
+	}
+
+	fd->buf = kvmalloc(APFS_COMPRESS_BLOCK, GFP_KERNEL);
+	if (!fd->buf) {
+		res = -ENOMEM;
+		goto fail;
+	}
+	fd->bufblk = -1;
+
+	is_rsrc = apfs_compress_is_rsrc(le32_to_cpu(fd->hdr.algo));
+	res = apfs_xattr_get_compressed_data(inode, is_rsrc ? APFS_XATTR_NAME_RSRC_FORK : APFS_XATTR_NAME_COMPRESSED, &fd->cdata);
+	if (res) {
+		apfs_err(sb, "failed to get compressed data");
+		goto fail;
+	}
+
+	up_read(&nxi->nx_big_sem);
+
+	filp->private_data = fd;
+	return 0;
+
+fail:
+	apfs_release_compressed_data(&fd->cdata);
+	if (fd->buf)
+		kvfree(fd->buf);
+	up_read(&nxi->nx_big_sem);
+	kfree(fd);
+	if (res > 0)
+		res = -EINVAL;
+	return res;
+}
+
+static int apfs_compress_file_read_block(struct apfs_compress_file_data *fd, loff_t block)
+{
+	struct super_block *sb = fd->sb;
+	struct apfs_compressed_data *comp_data = &fd->cdata;
+	u8 *cdata = NULL;
+	u8 *tmp = fd->buf;
+	u32 doffs = 0, coffs;
+	size_t csize, bsize;
+	int res = 0;
+
+	if (apfs_compress_is_rsrc(le32_to_cpu(fd->hdr.algo)) &&
+	   le32_to_cpu(fd->hdr.algo) != APFS_COMPRESS_LZBITMAP_RSRC &&
+	   le32_to_cpu(fd->hdr.algo) != APFS_COMPRESS_LZVN_RSRC &&
+	   le32_to_cpu(fd->hdr.algo) != APFS_COMPRESS_LZFSE_RSRC) {
+		struct apfs_compress_rsrc_hdr hdr = {0};
+		struct apfs_compress_rsrc_data cd = {0};
+		struct apfs_compress_rsrc_block blk = {0};
+		u32 blk_off;
+
+		res = apfs_compressed_data_read(comp_data, &hdr, sizeof(hdr), 0 /* offset */);
+		if (res) {
+			apfs_err(sb, "failed to read resource header");
+			return res;
+		}
+
+		doffs = be32_to_cpu(hdr.data_offs);
+		res = apfs_compressed_data_read(comp_data, &cd, sizeof(cd), doffs);
+		if (res) {
+			apfs_err(sb, "failed to read resource data header");
+			return res;
+		}
+		if (block >= le32_to_cpu(cd.num))
+			return 0;
+
+		blk_off = doffs + sizeof(cd) + sizeof(blk) * block;
+		res = apfs_compressed_data_read(comp_data, &blk, sizeof(blk), blk_off);
+		if (res) {
+			apfs_err(sb, "failed to read resource block metadata");
+			return res;
+		}
+
+		bsize = le64_to_cpu(fd->hdr.size) - block * APFS_COMPRESS_BLOCK;
+		if (bsize > APFS_COMPRESS_BLOCK)
+			bsize = APFS_COMPRESS_BLOCK;
+
+		csize = le32_to_cpu(blk.size);
+		coffs = le32_to_cpu(blk.offs) + 4;
+	} else if (apfs_compress_is_rsrc(le32_to_cpu(fd->hdr.algo))) {
+		__le32 blks[2];
+		u32 blk_off;
+
+		blk_off = doffs + sizeof(__le32) * block;
+		res = apfs_compressed_data_read(comp_data, blks, sizeof(blks), blk_off);
+		if (res) {
+			apfs_err(sb, "failed to read resource block metadata");
+			return res;
+		}
+
+		bsize = le64_to_cpu(fd->hdr.size) - block * APFS_COMPRESS_BLOCK;
+		if (bsize > APFS_COMPRESS_BLOCK)
+			bsize = APFS_COMPRESS_BLOCK;
+
+		coffs = le32_to_cpu(blks[0]);
+		csize = le32_to_cpu(blks[1]) - coffs;
+	} else {
+		/*
+		 * I think attr compression is only for single-block files, in
+		 * fact none of these files ever seem to decompress to more than
+		 * 2048 bytes.
+		 */
+		bsize = le64_to_cpu(fd->hdr.size);
+		if (block != 0 || bsize > APFS_COMPRESS_BLOCK) {
+			apfs_err(sb, "file too big for inline compression");
+			return -EFSCORRUPTED;
+		}
+
+		/* The first few bytes are the decmpfs header */
+		coffs = sizeof(struct apfs_compress_hdr);
+		csize = comp_data->size - sizeof(struct apfs_compress_hdr);
+	}
+
+	cdata = kvmalloc(csize, GFP_KERNEL);
+	if (!cdata)
+		return -ENOMEM;
+	res = apfs_compressed_data_read(comp_data, cdata, csize, doffs + coffs);
+	if (res) {
+		apfs_err(sb, "failed to read compressed block");
+		goto fail;
+	}
+
+	switch (le32_to_cpu(fd->hdr.algo)) {
+	case APFS_COMPRESS_ZLIB_RSRC:
+	case APFS_COMPRESS_ZLIB_ATTR:
+		if (cdata[0] == 0x78 && csize >= 2) {
+			res = zlib_inflate_blob(tmp, bsize, cdata + 2, csize - 2);
+			if (res <= 0) {
+				apfs_err(sb, "zlib decompression failed");
+				goto fail;
+			}
+			bsize = res;
+			res = 0;
+		} else if ((cdata[0] & 0x0F) == 0x0F) {
+			memcpy(tmp, &cdata[1], csize - 1);
+			bsize = csize - 1;
+		} else {
+			apfs_err(sb, "zlib decompression failed");
+			res = -EINVAL;
+			goto fail;
+		}
+		break;
+	case APFS_COMPRESS_LZVN_RSRC:
+	case APFS_COMPRESS_LZVN_ATTR:
+		if (cdata[0] == 0x06) {
+			memcpy(tmp, &cdata[1], csize - 1);
+			bsize = csize - 1;
+		} else {
+			lzvn_decoder_state dstate = {0};
+
+			dstate.src = cdata;
+			dstate.src_end = dstate.src + csize;
+			dstate.dst = dstate.dst_begin = tmp;
+			dstate.dst_end = dstate.dst + bsize;
+			lzvn_decode(&dstate);
+			bsize = dstate.dst - tmp;
+		}
+		break;
+	case APFS_COMPRESS_LZBITMAP_RSRC:
+	case APFS_COMPRESS_LZBITMAP_ATTR:
+		if (cdata[0] == 0x5a) {
+			res = zbm_decompress(tmp, bsize, cdata, csize, &bsize);
+			if (res < 0) {
+				apfs_err(sb, "lzbitmap decompression failed");
+				goto fail;
+			}
+			res = 0;
+		} else if ((cdata[0] & 0x0F) == 0x0F) {
+			memcpy(tmp, &cdata[1], csize - 1);
+			bsize = csize - 1;
+		} else {
+			apfs_err(sb, "lzbitmap decompression failed");
+			res = -EINVAL;
+			goto fail;
+		}
+		break;
+	case APFS_COMPRESS_LZFSE_RSRC:
+	case APFS_COMPRESS_LZFSE_ATTR:
+		if (cdata[0] == 0x62 && csize >= 2) {
+			res = lzfse_decode_buffer(tmp, bsize, cdata, csize, NULL);
+			if (res == 0) {
+				apfs_err(sb, "lzfse decompression failed");
+				/* Could be ENOMEM too... */
+				res = -EINVAL;
+				goto fail;
+			}
+			bsize = res;
+			res = 0;
+		} else {
+			/* cdata[0] == 0xff, apparently */
+			memcpy(tmp, &cdata[1], csize - 1);
+			bsize = csize - 1;
+		}
+		break;
+	case APFS_COMPRESS_PLAIN_RSRC:
+	case APFS_COMPRESS_PLAIN_ATTR:
+		memcpy(tmp, &cdata[1], csize - 1);
+		bsize = csize - 1;
+		break;
+	default:
+		res = -EINVAL;
+		goto fail;
+	}
+	fd->bufblk = block;
+	fd->bufsize = bsize;
+fail:
+	kvfree(cdata);
+	return res;
+}
+
+static int apfs_compress_file_release(struct inode *inode, struct file *filp)
+{
+	struct apfs_compress_file_data *fd = filp->private_data;
+
+	apfs_release_compressed_data(&fd->cdata);
+	if (fd->buf)
+		kvfree(fd->buf);
+	kfree(fd);
+	return 0;
+}
+
+static ssize_t apfs_compress_file_read_from_block(struct apfs_compress_file_data *fd, char *buf, size_t size, loff_t off)
+{
+	struct super_block *sb = fd->sb;
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_compressed_data cdata = fd->cdata;
+	loff_t block;
+	size_t bsize;
+	ssize_t res;
+
+	/*
+	 * Request reads of all blocks before actually working with any of them.
+	 * The compressed data is typically small enough that this is effective.
+	 * It would be much better to make an inode for the xattr dstream and
+	 * work with readahead as usual, but I'm not confident I can get that
+	 * right (TODO).
+	 */
+	if (cdata.has_dstream && off == 0) {
+		down_read(&nxi->nx_big_sem);
+		apfs_nonsparse_dstream_preread(cdata.dstream);
+		up_read(&nxi->nx_big_sem);
+	}
+
+	if (off >= le64_to_cpu(fd->hdr.size))
+		return 0;
+	if (size > le64_to_cpu(fd->hdr.size) - off)
+		size = le64_to_cpu(fd->hdr.size) - off;
+
+	block = off / APFS_COMPRESS_BLOCK;
+	off -= block * APFS_COMPRESS_BLOCK;
+	if (block != fd->bufblk) {
+		down_read(&nxi->nx_big_sem);
+		res = apfs_compress_file_read_block(fd, block);
+		up_read(&nxi->nx_big_sem);
+		if (res) {
+			apfs_err(sb, "failed to read block into buffer");
+			return res;
+		}
+	}
+	bsize = fd->bufsize;
+
+	if (bsize < off)
+		return 0;
+	bsize -= off;
+	if (size > bsize)
+		size = bsize;
+	memcpy(buf, fd->buf + off, size);
+	return size;
+}
+
+static ssize_t apfs_compress_file_read_page(struct file *filp, char *buf, loff_t off)
+{
+	struct apfs_compress_file_data *fd = filp->private_data;
+	loff_t step;
+	ssize_t block, res;
+	size_t size = PAGE_SIZE;
+
+	step = 0;
+	while (step < size) {
+		block = APFS_COMPRESS_BLOCK - ((off + step) & (APFS_COMPRESS_BLOCK - 1));
+		if (block > size - step)
+			block = size - step;
+		mutex_lock(&fd->mtx);
+		res = apfs_compress_file_read_from_block(fd, buf + step, block, off + step);
+		mutex_unlock(&fd->mtx);
+		if (res < block) {
+			if (res < 0 && !step)
+				return res;
+			step += res > 0 ? res : 0;
+			break;
+		}
+		step += block;
+	}
+	return step;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+static int apfs_compress_read_folio(struct file *filp, struct folio *folio)
+{
+	struct page *page = &folio->page;
+#else
+static int apfs_compress_readpage(struct file *filp, struct page *page)
+{
+#endif
+	char *addr = NULL;
+	ssize_t ret;
+	loff_t off;
+
+	/* Mostly copied from ext4_read_inline_page() */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 16, 0)
+	off = page_folio(page)->index << PAGE_SHIFT;
+#else
+	off = page->index << PAGE_SHIFT;
+#endif
+	addr = kmap(page);
+	ret = apfs_compress_file_read_page(filp, addr, off);
+	flush_dcache_page(page);
+	kunmap(page);
+	if (ret >= 0) {
+		zero_user_segment(page, ret, PAGE_SIZE);
+		SetPageUptodate(page);
+		ret = 0;
+	}
+
+	unlock_page(page);
+	return ret;
+}
+
+const struct address_space_operations apfs_compress_aops = {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+	.read_folio	= apfs_compress_read_folio,
+#else
+	.readpage	= apfs_compress_readpage,
+#endif
+};
+
+/* TODO: these operations are all happening without proper locks */
+const struct file_operations apfs_compress_file_operations = {
+	.open		= apfs_compress_file_open,
+	.llseek		= generic_file_llseek,
+	.read_iter	= generic_file_read_iter,
+	.release	= apfs_compress_file_release,
+	.mmap		= apfs_file_mmap,
+};
+
+int apfs_compress_get_size(struct inode *inode, loff_t *size)
+{
+	struct apfs_compress_hdr hdr;
+	int res = ____apfs_xattr_get(inode, APFS_XATTR_NAME_COMPRESSED, &hdr, sizeof(hdr), 0);
+
+	if (res < 0)
+		return res;
+	if (res != sizeof(hdr)) {
+		apfs_err(inode->i_sb, "decmpfs header read failed");
+		return 1;
+	}
+
+	if (!apfs_compress_is_supported(le32_to_cpu(hdr.algo)))
+		return 1;
+
+	*size = le64_to_cpu(hdr.size);
+	return 0;
+}
diff --git a/fs/apfs/dir.c b/fs/apfs/dir.c
new file mode 100644
index 000000000..01f409b45
--- /dev/null
+++ b/fs/apfs/dir.c
@@ -0,0 +1,1505 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/slab.h>
+#include <linux/buffer_head.h>
+#include "apfs.h"
+
+/**
+ * apfs_drec_from_query - Read the directory record found by a successful query
+ * @query:	the query that found the record
+ * @drec:	Return parameter.  The directory record found.
+ * @hashed:	is this record hashed?
+ *
+ * Reads the directory record into @drec and performs some basic sanity checks
+ * as a protection against crafted filesystems.  Returns 0 on success or
+ * -EFSCORRUPTED otherwise.
+ *
+ * The caller must not free @query while @drec is in use, because @drec->name
+ * points to data on disk.
+ */
+static int apfs_drec_from_query(struct apfs_query *query, struct apfs_drec *drec, bool hashed)
+{
+	struct super_block *sb = query->node->object.sb;
+	char *raw = query->node->object.data;
+	struct apfs_drec_hashed_key *de_hkey = NULL;
+	struct apfs_drec_key *de_ukey = NULL;
+	struct apfs_drec_val *de;
+	int namelen, xlen;
+	char *xval = NULL, *name;
+
+	namelen = query->key_len - (hashed ? sizeof(*de_hkey) : sizeof(*de_ukey));
+	if (namelen < 1) {
+		apfs_err(sb, "key is too small (%d)", query->key_len);
+		return -EFSCORRUPTED;
+	}
+	if (query->len < sizeof(*de)) {
+		apfs_err(sb, "value is too small (%d)", query->len);
+		return -EFSCORRUPTED;
+	}
+
+	de = (struct apfs_drec_val *)(raw + query->off);
+	if (hashed) {
+		de_hkey = (struct apfs_drec_hashed_key *)(raw + query->key_off);
+		if (namelen != (le32_to_cpu(de_hkey->name_len_and_hash) & APFS_DREC_LEN_MASK)) {
+			apfs_err(sb, "inconsistent name length");
+			return -EFSCORRUPTED;
+		}
+		name = de_hkey->name;
+	} else {
+		de_ukey = (struct apfs_drec_key *)(raw + query->key_off);
+		if (namelen != le16_to_cpu(de_ukey->name_len)) {
+			apfs_err(sb, "inconsistent name length");
+			return -EFSCORRUPTED;
+		}
+		name = de_ukey->name;
+	}
+
+	/* Filename must be NULL-terminated */
+	if (name[namelen - 1] != 0) {
+		apfs_err(sb, "null termination missing");
+		return -EFSCORRUPTED;
+	}
+
+	/* The dentry may have at most one xfield: the sibling id */
+	drec->sibling_id = 0;
+	xlen = apfs_find_xfield(de->xfields, query->len - sizeof(*de),
+				APFS_DREC_EXT_TYPE_SIBLING_ID, &xval);
+	if (xlen >= sizeof(__le64)) {
+		__le64 *sib_id = (__le64 *)xval;
+
+		drec->sibling_id = le64_to_cpup(sib_id);
+	}
+
+	drec->name = name;
+	drec->name_len = namelen - 1; /* Don't count the NULL termination */
+	drec->ino = le64_to_cpu(de->file_id);
+
+	drec->type = le16_to_cpu(de->flags) & APFS_DREC_TYPE_MASK;
+	if (drec->type != DT_FIFO && drec->type & 1) /* Invalid file type */
+		drec->type = DT_UNKNOWN;
+	return 0;
+}
+
+/**
+ * apfs_dentry_lookup - Lookup a dentry record in the catalog b-tree
+ * @dir:	parent directory
+ * @child:	filename
+ * @drec:	on return, the directory record found
+ *
+ * Runs a catalog query for @name in the @dir directory.  On success, sets
+ * @drec and returns a pointer to the query structure.  On failure, returns
+ * an appropriate error pointer.
+ */
+static struct apfs_query *apfs_dentry_lookup(struct inode *dir,
+					     const struct qstr *child,
+					     struct apfs_drec *drec)
+{
+	struct super_block *sb = dir->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	u64 cnid = apfs_ino(dir);
+	bool hashed = apfs_is_normalization_insensitive(sb);
+	int err;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return ERR_PTR(-ENOMEM);
+	apfs_init_drec_key(sb, cnid, child->name, child->len, &query->key);
+
+	/*
+	 * Distinct filenames in the same directory may (rarely) share the same
+	 * hash. The query code cannot handle that because their order in the
+	 * b-tree would	depend on their unnormalized original names. Just get
+	 * all the candidates and check them one by one.
+	 *
+	 * This is very wasteful for normalization-sensitive filesystems: there
+	 * are no hashes so we just check every single file in the directory for
+	 * no reason. This would be easy to avoid but does it matter? (TODO)
+	 */
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_ANY_NAME | APFS_QUERY_EXACT;
+	do {
+		err = apfs_btree_query(sb, &query);
+		if (err)
+			goto fail;
+		err = apfs_drec_from_query(query, drec, hashed);
+		if (err)
+			goto fail;
+	} while (unlikely(apfs_filename_cmp(sb, child->name, child->len, drec->name, drec->name_len)));
+
+	/*
+	 * We may need to refresh the query later, but the refresh code doesn't
+	 * know how to deal with hash collisions. Instead set the key to the
+	 * unnormalized name and pretend that this was never a multiple query
+	 * in the first place.
+	 */
+	query->key.name = drec->name;
+	query->flags &= ~(APFS_QUERY_MULTIPLE | APFS_QUERY_DONE | APFS_QUERY_NEXT);
+	return query;
+
+fail:
+	if (err != -ENODATA)
+		apfs_err(sb, "query failed in dir 0x%llx", cnid);
+	apfs_free_query(query);
+	return ERR_PTR(err);
+}
+
+/**
+ * apfs_inode_by_name - Find the cnid for a given filename
+ * @dir:	parent directory
+ * @child:	filename
+ * @ino:	on return, the inode number found
+ *
+ * Returns 0 and the inode number (which is the cnid of the file
+ * record); otherwise, return the appropriate error code.
+ */
+int apfs_inode_by_name(struct inode *dir, const struct qstr *child, u64 *ino)
+{
+	struct super_block *sb = dir->i_sb;
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_query *query;
+	struct apfs_drec drec;
+	int err = 0;
+
+	down_read(&nxi->nx_big_sem);
+	query = apfs_dentry_lookup(dir, child, &drec);
+	if (IS_ERR(query)) {
+		err = PTR_ERR(query);
+		goto out;
+	}
+	*ino = drec.ino;
+	apfs_free_query(query);
+out:
+	up_read(&nxi->nx_big_sem);
+	return err;
+}
+
+static int apfs_readdir(struct file *file, struct dir_context *ctx)
+{
+	struct inode *inode = file_inode(file);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_query *query;
+	u64 cnid = apfs_ino(inode);
+	loff_t pos;
+	bool hashed = apfs_is_normalization_insensitive(sb);
+	int err = 0;
+
+	down_read(&nxi->nx_big_sem);
+
+	/* Inode numbers might overflow here; follow btrfs in ignoring that */
+	if (!dir_emit_dots(file, ctx))
+		goto out;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	/* We want all the children for the cnid, regardless of the name */
+	apfs_init_drec_key(sb, cnid, NULL /* name */, 0 /* name_len */, &query->key);
+	query->flags = APFS_QUERY_CAT | APFS_QUERY_MULTIPLE | APFS_QUERY_EXACT;
+
+	pos = ctx->pos - 2;
+	while (1) {
+		struct apfs_drec drec;
+		/*
+		 * We query for the matching records, one by one. After we
+		 * pass ctx->pos we begin to emit them.
+		 *
+		 * TODO: Faster approach for large directories?
+		 */
+
+		err = apfs_btree_query(sb, &query);
+		if (err == -ENODATA) { /* Got all the records */
+			err = 0;
+			break;
+		}
+		if (err)
+			break;
+
+		err = apfs_drec_from_query(query, &drec, hashed);
+		if (err) {
+			apfs_alert(sb, "bad dentry record in directory 0x%llx",
+				   cnid);
+			break;
+		}
+
+		err = 0;
+		if (pos <= 0) {
+			if (!dir_emit(ctx, drec.name, drec.name_len,
+				      drec.ino, drec.type))
+				break;
+			++ctx->pos;
+		}
+		pos--;
+	}
+	apfs_free_query(query);
+
+out:
+	up_read(&nxi->nx_big_sem);
+	return err;
+}
+
+const struct file_operations apfs_dir_operations = {
+	.llseek		= generic_file_llseek,
+	.read		= generic_read_dir,
+	.iterate_shared	= apfs_readdir,
+	.fsync		= apfs_fsync,
+	.unlocked_ioctl	= apfs_dir_ioctl,
+};
+
+/**
+ * apfs_build_dentry_unhashed_key - Allocate and initialize the key for an unhashed dentry record
+ * @qname:	filename
+ * @parent_id:	inode number for the parent of the dentry
+ * @key_p:	on return, a pointer to the new on-disk key structure
+ *
+ * Returns the length of the key, or a negative error code in case of failure.
+ */
+static int apfs_build_dentry_unhashed_key(const struct qstr *qname, u64 parent_id,
+					  struct apfs_drec_key **key_p)
+{
+	struct apfs_drec_key *key;
+	u16 namelen = qname->len + 1; /* We count the null-termination */
+	int key_len;
+
+	key_len = sizeof(*key) + namelen;
+	key = kmalloc(key_len, GFP_KERNEL);
+	if (!key)
+		return -ENOMEM;
+
+	apfs_key_set_hdr(APFS_TYPE_DIR_REC, parent_id, key);
+	key->name_len = cpu_to_le16(namelen);
+	strscpy(key->name, qname->name, namelen);
+
+	*key_p = key;
+	return key_len;
+}
+
+/**
+ * apfs_build_dentry_hashed_key - Allocate and initialize the key for a hashed dentry record
+ * @qname:	filename
+ * @hash:	filename hash
+ * @parent_id:	inode number for the parent of the dentry
+ * @key_p:	on return, a pointer to the new on-disk key structure
+ *
+ * Returns the length of the key, or a negative error code in case of failure.
+ */
+static int apfs_build_dentry_hashed_key(const struct qstr *qname, u64 hash, u64 parent_id,
+					struct apfs_drec_hashed_key **key_p)
+{
+	struct apfs_drec_hashed_key *key;
+	u16 namelen = qname->len + 1; /* We count the null-termination */
+	int key_len;
+
+	key_len = sizeof(*key) + namelen;
+	key = kmalloc(key_len, GFP_KERNEL);
+	if (!key)
+		return -ENOMEM;
+
+	apfs_key_set_hdr(APFS_TYPE_DIR_REC, parent_id, key);
+	key->name_len_and_hash = cpu_to_le32(namelen | hash);
+	strscpy(key->name, qname->name, namelen);
+
+	*key_p = key;
+	return key_len;
+}
+
+/**
+ * apfs_build_dentry_val - Allocate and initialize the value for a dentry record
+ * @inode:	vfs inode for the dentry
+ * @sibling_id:	sibling id for this hardlink (0 for none)
+ * @val_p:	on return, a pointer to the new on-disk value structure
+ *
+ * Returns the length of the value, or a negative error code in case of failure.
+ */
+static int apfs_build_dentry_val(struct inode *inode, u64 sibling_id,
+				 struct apfs_drec_val **val_p)
+{
+	struct apfs_drec_val *val;
+	struct apfs_x_field xkey;
+	int total_xlen = 0, val_len;
+	__le64 raw_sibling_id = cpu_to_le64(sibling_id);
+	struct timespec64 now = current_time(inode);
+
+	/* The dentry record may have one xfield: the sibling id */
+	if (sibling_id)
+		total_xlen += sizeof(struct apfs_xf_blob) +
+			      sizeof(xkey) + sizeof(raw_sibling_id);
+
+	val_len = sizeof(*val) + total_xlen;
+	val = kmalloc(val_len, GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+	*val_p = val;
+
+	val->file_id = cpu_to_le64(apfs_ino(inode));
+	val->date_added = cpu_to_le64(timespec64_to_ns(&now));
+	val->flags = cpu_to_le16((inode->i_mode >> 12) & 15); /* File type */
+
+	if (!sibling_id)
+		return val_len;
+
+	/* The buffer was just allocated: none of these functions should fail */
+	apfs_init_xfields(val->xfields, val_len - sizeof(*val));
+	xkey.x_type = APFS_DREC_EXT_TYPE_SIBLING_ID;
+	xkey.x_flags = 0; /* TODO: proper flags here? */
+	xkey.x_size = cpu_to_le16(sizeof(raw_sibling_id));
+	apfs_insert_xfield(val->xfields, total_xlen, &xkey, &raw_sibling_id);
+	return val_len;
+}
+
+/**
+ * apfs_create_dentry_rec - Create a dentry record in the catalog b-tree
+ * @inode:	vfs inode for the dentry
+ * @qname:	filename
+ * @parent_id:	inode number for the parent of the dentry
+ * @sibling_id:	sibling id for this hardlink (0 for none)
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_dentry_rec(struct inode *inode, const struct qstr *qname,
+				  u64 parent_id, u64 sibling_id)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	void *raw_key = NULL;
+	struct apfs_drec_val *raw_val = NULL;
+	int key_len, val_len;
+	bool hashed = apfs_is_normalization_insensitive(sb);
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_drec_key(sb, parent_id, qname->name, qname->len, &query->key);
+	query->flags |= APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed in dir 0x%llx (hash 0x%llx)", parent_id, query->key.number);
+		goto fail;
+	}
+
+	if (hashed)
+		key_len = apfs_build_dentry_hashed_key(qname, query->key.number, parent_id,
+						       (struct apfs_drec_hashed_key **)&raw_key);
+	else
+		key_len = apfs_build_dentry_unhashed_key(qname, parent_id,
+							 (struct apfs_drec_key **)&raw_key);
+	if (key_len < 0) {
+		ret = key_len;
+		goto fail;
+	}
+
+	val_len = apfs_build_dentry_val(inode, sibling_id, &raw_val);
+	if (val_len < 0) {
+		ret = val_len;
+		goto fail;
+	}
+	/* TODO: deal with hash collisions */
+	ret = apfs_btree_insert(query, raw_key, key_len, raw_val, val_len);
+	if (ret)
+		apfs_err(sb, "insertion failed in dir 0x%llx (hash 0x%llx)", parent_id, query->key.number);
+
+fail:
+	kfree(raw_val);
+	kfree(raw_key);
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_build_sibling_val - Allocate and initialize a sibling link's value
+ * @dentry:	in-memory dentry for this hardlink
+ * @val_p:	on return, a pointer to the new on-disk value structure
+ *
+ * Returns the length of the value, or a negative error code in case of failure.
+ */
+static int apfs_build_sibling_val(struct dentry *dentry,
+				  struct apfs_sibling_val **val_p)
+{
+	struct apfs_sibling_val *val;
+	const struct qstr *qname = &dentry->d_name;
+	u16 namelen = qname->len + 1; /* We count the null-termination */
+	struct inode *parent = d_inode(dentry->d_parent);
+	int val_len;
+
+	val_len = sizeof(*val) + namelen;
+	val = kmalloc(val_len, GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	val->parent_id = cpu_to_le64(apfs_ino(parent));
+	val->name_len = cpu_to_le16(namelen);
+	strscpy(val->name, qname->name, namelen);
+
+	*val_p = val;
+	return val_len;
+}
+
+/**
+ * apfs_create_sibling_link_rec - Create a sibling link record for a dentry
+ * @dentry:	the in-memory dentry
+ * @inode:	vfs inode for the dentry
+ * @sibling_id:	sibling id for this hardlink
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_sibling_link_rec(struct dentry *dentry,
+					struct inode *inode, u64 sibling_id)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	struct apfs_sibling_link_key raw_key;
+	struct apfs_sibling_val *raw_val;
+	int val_len;
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_sibling_link_key(apfs_ino(inode), sibling_id, &query->key);
+	query->flags |= APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for ino 0x%llx, sibling 0x%llx", apfs_ino(inode), sibling_id);
+		goto fail;
+	}
+
+	apfs_key_set_hdr(APFS_TYPE_SIBLING_LINK, apfs_ino(inode), &raw_key);
+	raw_key.sibling_id = cpu_to_le64(sibling_id);
+	val_len = apfs_build_sibling_val(dentry, &raw_val);
+	if (val_len < 0)
+		goto fail;
+
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), raw_val, val_len);
+	if (ret)
+		apfs_err(sb, "insertion failed for ino 0x%llx, sibling 0x%llx", apfs_ino(inode), sibling_id);
+	kfree(raw_val);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_create_sibling_map_rec - Create a sibling map record for a dentry
+ * @dentry:	the in-memory dentry
+ * @inode:	vfs inode for the dentry
+ * @sibling_id:	sibling id for this hardlink
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_sibling_map_rec(struct dentry *dentry,
+				       struct inode *inode, u64 sibling_id)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	struct apfs_sibling_map_key raw_key;
+	struct apfs_sibling_map_val raw_val;
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_sibling_map_key(sibling_id, &query->key);
+	query->flags |= APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for sibling 0x%llx", sibling_id);
+		goto fail;
+	}
+
+	apfs_key_set_hdr(APFS_TYPE_SIBLING_MAP, sibling_id, &raw_key);
+	raw_val.file_id = cpu_to_le64(apfs_ino(inode));
+
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+	if (ret)
+		apfs_err(sb, "insertion failed for sibling 0x%llx", sibling_id);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_create_sibling_recs - Create sibling link and map records for a dentry
+ * @dentry:	the in-memory dentry
+ * @inode:	vfs inode for the dentry
+ * @sibling_id:	on return, the sibling id for this hardlink
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_sibling_recs(struct dentry *dentry,
+				    struct inode *inode, u64 *sibling_id)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	u64 cnid;
+	int ret;
+
+	/* Sibling ids come from the same pool as the inode numbers */
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	cnid = le64_to_cpu(vsb_raw->apfs_next_obj_id);
+	le64_add_cpu(&vsb_raw->apfs_next_obj_id, 1);
+
+	ret = apfs_create_sibling_link_rec(dentry, inode, cnid);
+	if (ret)
+		return ret;
+	ret = apfs_create_sibling_map_rec(dentry, inode, cnid);
+	if (ret)
+		return ret;
+
+	*sibling_id = cnid;
+	return 0;
+}
+
+/**
+ * apfs_create_dentry - Create all records for a new dentry
+ * @dentry:	the in-memory dentry
+ * @inode:	vfs inode for the dentry
+ *
+ * Creates the dentry record itself, as well as the sibling records if needed;
+ * also updates the child count for the parent inode.  Returns 0 on success or
+ * a negative error code in case of failure.
+ */
+static int apfs_create_dentry(struct dentry *dentry, struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct inode *parent = d_inode(dentry->d_parent);
+	u64 sibling_id = 0;
+	int err;
+
+	if (inode->i_nlink > 1) {
+		/* This is optional for a single link, so don't waste space */
+		err = apfs_create_sibling_recs(dentry, inode, &sibling_id);
+		if (err) {
+			apfs_err(sb, "failed to create sibling recs for ino 0x%llx", apfs_ino(inode));
+			return err;
+		}
+	}
+
+	err = apfs_create_dentry_rec(inode, &dentry->d_name, apfs_ino(parent), sibling_id);
+	if (err) {
+		apfs_err(sb, "failed to create drec for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	/* Now update the parent inode */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	parent->i_mtime = parent->i_ctime = current_time(inode);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	parent->i_mtime = inode_set_ctime_current(parent);
+#else
+	inode_set_mtime_to_ts(parent, inode_set_ctime_current(parent));
+#endif
+	++APFS_I(parent)->i_nchildren;
+	apfs_inode_join_transaction(parent->i_sb, parent);
+	return 0;
+}
+
+/**
+ * apfs_undo_create_dentry - Clean up apfs_create_dentry()
+ * @dentry: the in-memory dentry
+ */
+static void apfs_undo_create_dentry(struct dentry *dentry)
+{
+	struct inode *parent = d_inode(dentry->d_parent);
+
+	--APFS_I(parent)->i_nchildren;
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 19, 0)
+
+static inline void discard_new_inode(struct inode *inode)
+{
+	unlock_new_inode(inode);
+	iput(inode);
+}
+
+#endif
+
+int apfs_mkany(struct inode *dir, struct dentry *dentry, umode_t mode,
+	       dev_t rdev, const char *symname)
+{
+	struct super_block *sb = dir->i_sb;
+	struct inode *inode;
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	inode = apfs_new_inode(dir, mode, rdev);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		goto out_abort;
+	}
+
+	err = apfs_create_inode_rec(sb, inode, dentry);
+	if (err) {
+		apfs_err(sb, "failed to create inode rec for ino 0x%llx", apfs_ino(inode));
+		goto out_discard_inode;
+	}
+
+	err = apfs_create_dentry(dentry, inode);
+	if (err) {
+		apfs_err(sb, "failed to create dentry recs for ino 0x%llx", apfs_ino(inode));
+		goto out_discard_inode;
+	}
+
+	if (symname) {
+		err = apfs_xattr_set(inode, APFS_XATTR_NAME_SYMLINK, symname,
+				     strlen(symname) + 1, 0 /* flags */);
+		if (err == -ERANGE) {
+			err = -ENAMETOOLONG;
+			goto out_undo_create;
+		}
+		if (err) {
+			apfs_err(sb, "failed to set symlink xattr for ino 0x%llx", apfs_ino(inode));
+			goto out_undo_create;
+		}
+	}
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto out_undo_create;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 17, 0)
+	/* Apparently there is a lockdep bug here, but it doesn't matter */
+	unlock_new_inode(inode);
+	d_instantiate(dentry, inode);
+#else
+	d_instantiate_new(dentry, inode);
+#endif
+	return 0;
+
+out_undo_create:
+	apfs_undo_create_dentry(dentry);
+out_discard_inode:
+	/* Don't reset nlink: on-disk cleanup is unneeded and would deadlock */
+	discard_new_inode(inode);
+out_abort:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+int apfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
+	       dev_t rdev)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+int apfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
+	       struct dentry *dentry, umode_t mode, dev_t rdev)
+#else
+int apfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
+	       struct dentry *dentry, umode_t mode, dev_t rdev)
+#endif
+{
+	return apfs_mkany(dir, dentry, mode, rdev, NULL /* symname */);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+
+int apfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
+{
+	return apfs_mknod(dir, dentry, mode | S_IFDIR, 0 /* rdev */);
+}
+
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+
+int apfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
+	       struct dentry *dentry, umode_t mode)
+{
+	return apfs_mknod(mnt_userns, dir, dentry, mode | S_IFDIR, 0 /* rdev */);
+}
+
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 15, 0)
+
+int apfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
+	       struct dentry *dentry, umode_t mode)
+{
+	return apfs_mknod(idmap, dir, dentry, mode | S_IFDIR, 0 /* rdev */);
+}
+
+#else
+
+struct dentry *apfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, struct dentry *dentry, umode_t mode)
+{
+	return ERR_PTR(apfs_mknod(idmap, dir, dentry, mode | S_IFDIR, 0 /* rdev */));
+}
+
+#endif
+
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+
+int apfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
+		bool excl)
+{
+	return apfs_mknod(dir, dentry, mode, 0 /* rdev */);
+}
+
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+
+int apfs_create(struct user_namespace *mnt_userns, struct inode *dir,
+		struct dentry *dentry, umode_t mode, bool excl)
+{
+	return apfs_mknod(mnt_userns, dir, dentry, mode, 0 /* rdev */);
+}
+
+#else
+
+int apfs_create(struct mnt_idmap *idmap, struct inode *dir,
+		struct dentry *dentry, umode_t mode, bool excl)
+{
+	return apfs_mknod(idmap, dir, dentry, mode, 0 /* rdev */);
+}
+
+#endif
+
+
+/**
+ * apfs_prepare_dentry_for_link - Assign a sibling id and records to a dentry
+ * @dentry: the in-memory dentry (should be for a primary link)
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_prepare_dentry_for_link(struct dentry *dentry)
+{
+	struct inode *parent = d_inode(dentry->d_parent);
+	struct super_block *sb = parent->i_sb;
+	struct apfs_query *query;
+	struct apfs_drec drec;
+	u64 sibling_id;
+	int ret;
+
+	query = apfs_dentry_lookup(parent, &dentry->d_name, &drec);
+	if (IS_ERR(query)) {
+		apfs_err(sb, "lookup failed in dir 0x%llx", apfs_ino(parent));
+		return PTR_ERR(query);
+	}
+	if (drec.sibling_id) {
+		/* This dentry already has a sibling id xfield */
+		apfs_free_query(query);
+		return 0;
+	}
+
+	/* Don't modify the dentry record, just delete it to make a new one */
+	ret = apfs_btree_remove(query);
+	apfs_free_query(query);
+	if (ret) {
+		apfs_err(sb, "removal failed in dir 0x%llx", apfs_ino(parent));
+		return ret;
+	}
+
+	ret = apfs_create_sibling_recs(dentry, d_inode(dentry), &sibling_id);
+	if (ret) {
+		apfs_err(sb, "failed to create sibling recs in dir 0x%llx", apfs_ino(parent));
+		return ret;
+	}
+	return apfs_create_dentry_rec(d_inode(dentry), &dentry->d_name,
+				      apfs_ino(parent), sibling_id);
+}
+
+/**
+ * __apfs_undo_link - Clean up __apfs_link()
+ * @dentry: the in-memory dentry
+ * @inode:  target inode
+ */
+static void __apfs_undo_link(struct dentry *dentry, struct inode *inode)
+{
+	apfs_undo_create_dentry(dentry);
+	drop_nlink(inode);
+}
+
+/**
+ * __apfs_link - Link a dentry
+ * @old_dentry: dentry for the old link
+ * @dentry:	new dentry to link
+ *
+ * Does the same as apfs_link(), but without starting a transaction, taking a
+ * new reference to @old_dentry->d_inode, or instantiating @dentry.
+ */
+static int __apfs_link(struct dentry *old_dentry, struct dentry *dentry)
+{
+	struct inode *inode = d_inode(old_dentry);
+	struct super_block *sb = inode->i_sb;
+	int err;
+
+	/* First update the inode's link count */
+	inc_nlink(inode);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode->i_ctime = current_time(inode);
+#else
+	inode_set_ctime_current(inode);
+#endif
+	apfs_inode_join_transaction(inode->i_sb, inode);
+
+	if (inode->i_nlink == 2) {
+		/* A single link may lack sibling records, so create them now */
+		err = apfs_prepare_dentry_for_link(old_dentry);
+		if (err) {
+			apfs_err(sb, "failed to prepare original dentry");
+			goto fail;
+		}
+	}
+
+	err = apfs_create_dentry(dentry, inode);
+	if (err) {
+		apfs_err(sb, "failed to create new dentry");
+		goto fail;
+	}
+	return 0;
+
+fail:
+	drop_nlink(inode);
+	return err;
+}
+
+int apfs_link(struct dentry *old_dentry, struct inode *dir,
+	      struct dentry *dentry)
+{
+	struct super_block *sb = dir->i_sb;
+	struct inode *inode = d_inode(old_dentry);
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	err = __apfs_link(old_dentry, dentry);
+	if (err)
+		goto out_abort;
+	ihold(inode);
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto out_undo_link;
+
+	d_instantiate(dentry, inode);
+	return 0;
+
+out_undo_link:
+	iput(inode);
+	__apfs_undo_link(dentry, inode);
+out_abort:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_delete_sibling_link_rec - Delete the sibling link record for a dentry
+ * @dentry:	the in-memory dentry
+ * @sibling_id:	sibling id for this hardlink
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_delete_sibling_link_rec(struct dentry *dentry, u64 sibling_id)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct inode *inode = d_inode(dentry);
+	struct apfs_query *query = NULL;
+	int ret;
+
+	ASSERT(sibling_id);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_sibling_link_key(apfs_ino(inode), sibling_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret == -ENODATA) {
+		/* A dentry with a sibling id must have sibling records */
+		ret = -EFSCORRUPTED;
+	}
+	if (ret) {
+		apfs_err(sb, "query failed for ino 0x%llx, sibling 0x%llx", apfs_ino(inode), sibling_id);
+		goto fail;
+	}
+	ret = apfs_btree_remove(query);
+	if (ret)
+		apfs_err(sb, "removal failed for ino 0x%llx, sibling 0x%llx", apfs_ino(inode), sibling_id);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_delete_sibling_map_rec - Delete the sibling map record for a dentry
+ * @dentry:	the in-memory dentry
+ * @sibling_id:	sibling id for this hardlink
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_delete_sibling_map_rec(struct dentry *dentry, u64 sibling_id)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	int ret;
+
+	ASSERT(sibling_id);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_sibling_map_key(sibling_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret == -ENODATA) {
+		/* A dentry with a sibling id must have sibling records */
+		ret = -EFSCORRUPTED;
+	}
+	if (ret) {
+		apfs_err(sb, "query failed for sibling 0x%llx", sibling_id);
+		goto fail;
+	}
+	ret = apfs_btree_remove(query);
+	if (ret)
+		apfs_err(sb, "removal failed for sibling 0x%llx", sibling_id);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_delete_dentry - Delete all records for a dentry
+ * @dentry: the in-memory dentry
+ *
+ * Deletes the dentry record itself, as well as the sibling records if they
+ * exist; also updates the child count for the parent inode.  Returns 0 on
+ * success or a negative error code in case of failure.
+ */
+static int apfs_delete_dentry(struct dentry *dentry)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct inode *parent = d_inode(dentry->d_parent);
+	struct apfs_query *query;
+	struct apfs_drec drec;
+	int err;
+
+	query = apfs_dentry_lookup(parent, &dentry->d_name, &drec);
+	if (IS_ERR(query))
+		return PTR_ERR(query);
+	err = apfs_btree_remove(query);
+	apfs_free_query(query);
+	if (err) {
+		apfs_err(sb, "drec removal failed");
+		return err;
+	}
+
+	if (drec.sibling_id) {
+		err = apfs_delete_sibling_link_rec(dentry, drec.sibling_id);
+		if (err) {
+			apfs_err(sb, "sibling link removal failed");
+			return err;
+		}
+		err = apfs_delete_sibling_map_rec(dentry, drec.sibling_id);
+		if (err) {
+			apfs_err(sb, "sibling map removal failed");
+			return err;
+		}
+	}
+
+	/* Now update the parent inode */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	parent->i_mtime = parent->i_ctime = current_time(parent);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	parent->i_mtime = inode_set_ctime_current(parent);
+#else
+	inode_set_mtime_to_ts(parent, inode_set_ctime_current(parent));
+#endif
+	--APFS_I(parent)->i_nchildren;
+	apfs_inode_join_transaction(sb, parent);
+	return err;
+}
+
+/**
+ * apfs_undo_delete_dentry - Clean up apfs_delete_dentry()
+ * @dentry: the in-memory dentry
+ */
+static inline void apfs_undo_delete_dentry(struct dentry *dentry)
+{
+	struct inode *parent = d_inode(dentry->d_parent);
+
+	/* Cleanup for the on-disk changes will happen on transaction abort */
+	++APFS_I(parent)->i_nchildren;
+}
+
+/**
+ * apfs_sibling_link_from_query - Read the sibling link record found by a query
+ * @query:	the query that found the record
+ * @name:	on return, the name of link
+ * @parent:	on return, the inode number for the link's parent
+ *
+ * Reads the sibling link information into @parent and @name, and performs some
+ * basic sanity checks as a protection against crafted filesystems.  The caller
+ * must free @name after use.  Returns 0 on success or a negative error code in
+ * case of failure.
+ */
+static int apfs_sibling_link_from_query(struct apfs_query *query,
+					char **name, u64 *parent)
+{
+	struct super_block *sb = query->node->object.sb;
+	char *raw = query->node->object.data;
+	struct apfs_sibling_val *siblink;
+	int namelen = query->len - sizeof(*siblink);
+
+	if (namelen < 1) {
+		apfs_err(sb, "value is too small (%d)", query->len);
+		return -EFSCORRUPTED;
+	}
+	siblink = (struct apfs_sibling_val *)(raw + query->off);
+
+	if (namelen != le16_to_cpu(siblink->name_len)) {
+		apfs_err(sb, "inconsistent name length");
+		return -EFSCORRUPTED;
+	}
+	/* Filename must be NULL-terminated */
+	if (siblink->name[namelen - 1] != 0) {
+		apfs_err(sb, "null termination missing");
+		return -EFSCORRUPTED;
+	}
+
+	*name = kmalloc(namelen, GFP_KERNEL);
+	if (!*name)
+		return -ENOMEM;
+	strscpy(*name, siblink->name, namelen);
+	*parent = le64_to_cpu(siblink->parent_id);
+	return 0;
+}
+
+/**
+ * apfs_find_primary_link - Find the primary link for an inode
+ * @inode:	the vfs inode
+ * @name:	on return, the name of the primary link
+ * @parent:	on return, the inode number for the primary parent
+ *
+ * On success, returns 0 and sets @parent and @name; the second must be freed
+ * by the caller after use.  Returns a negative error code in case of failure.
+ */
+static int apfs_find_primary_link(struct inode *inode, char **name, u64 *parent)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	int err;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_sibling_link_key(apfs_ino(inode), 0 /* sibling_id */, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_ANY_NUMBER | APFS_QUERY_EXACT;
+
+	/* The primary link is the one with the lowest sibling id */
+	*name = NULL;
+	while (1) {
+		err = apfs_btree_query(sb, &query);
+		if (err == -ENODATA) /* No more link records */
+			break;
+		kfree(*name);
+		if (err) {
+			apfs_err(sb, "query failed for ino 0x%llx", apfs_ino(inode));
+			goto fail;
+		}
+
+		err = apfs_sibling_link_from_query(query, name, parent);
+		if (err) {
+			apfs_err(sb, "bad sibling link record for ino 0x%llx", apfs_ino(inode));
+			goto fail;
+		}
+	}
+	err = *name ? 0 : -EFSCORRUPTED; /* Sibling records must exist */
+	if (err)
+		apfs_err(sb, "query failed for ino 0x%llx", apfs_ino(inode));
+
+fail:
+	apfs_free_query(query);
+	return err;
+}
+
+/**
+ * apfs_orphan_name - Get the name for an orphan inode's invisible link
+ * @ino:	the inode number
+ * @qname:	on return, the name assigned to the link
+ *
+ * Returns 0 on success; the caller must remember to free @qname->name after
+ * use.  Returns a negative error code in case of failure.
+ */
+static int apfs_orphan_name(u64 ino, struct qstr *qname)
+{
+	int max_len;
+	char *name;
+
+	/* The name is the inode number in hex, with '-dead' suffix */
+	max_len = 2 + 16 + 5 + 1;
+	name = kmalloc(max_len, GFP_KERNEL);
+	if (!name)
+		return -ENOMEM;
+	qname->len = snprintf(name, max_len, "0x%llx-dead", ino);
+	qname->name = name;
+	return 0;
+}
+
+/**
+ * apfs_create_orphan_link - Create a link for an orphan inode under private-dir
+ * @inode:	the vfs inode
+ *
+ * On success, returns 0. Returns a negative error code in case of failure.
+ */
+static int apfs_create_orphan_link(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct inode *priv_dir = sbi->s_private_dir;
+	struct qstr qname;
+	int err = 0;
+
+	err = apfs_orphan_name(apfs_ino(inode), &qname);
+	if (err)
+		return err;
+	err = apfs_create_dentry_rec(inode, &qname, apfs_ino(priv_dir), 0 /* sibling_id */);
+	if (err) {
+		apfs_err(sb, "failed to create drec for ino 0x%llx", apfs_ino(inode));
+		goto fail;
+	}
+
+	/* Now update the child count for private-dir */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	priv_dir->i_mtime = priv_dir->i_ctime = current_time(priv_dir);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	priv_dir->i_mtime = inode_set_ctime_current(priv_dir);
+#else
+	inode_set_mtime_to_ts(priv_dir, inode_set_ctime_current(priv_dir));
+#endif
+	++APFS_I(priv_dir)->i_nchildren;
+	apfs_inode_join_transaction(sb, priv_dir);
+
+fail:
+	kfree(qname.name);
+	return err;
+}
+
+/**
+ * apfs_delete_orphan_link - Delete the link for an orphan inode
+ * @inode: the vfs inode
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_delete_orphan_link(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct inode *priv_dir = sbi->s_private_dir;
+	struct apfs_query *query;
+	struct qstr qname;
+	struct apfs_drec drec;
+	int err;
+
+	err = apfs_orphan_name(apfs_ino(inode), &qname);
+	if (err)
+		return err;
+
+	query = apfs_dentry_lookup(priv_dir, &qname, &drec);
+	if (IS_ERR(query)) {
+		apfs_err(sb, "dentry lookup failed");
+		err = PTR_ERR(query);
+		query = NULL;
+		goto fail;
+	}
+	err = apfs_btree_remove(query);
+	if (err) {
+		apfs_err(sb, "dentry removal failed");
+		goto fail;
+	}
+
+	/* Now update the child count for private-dir */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	priv_dir->i_mtime = priv_dir->i_ctime = current_time(priv_dir);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	priv_dir->i_mtime = inode_set_ctime_current(priv_dir);
+#else
+	inode_set_mtime_to_ts(priv_dir, inode_set_ctime_current(priv_dir));
+#endif
+	--APFS_I(priv_dir)->i_nchildren;
+	apfs_inode_join_transaction(sb, priv_dir);
+
+fail:
+	apfs_free_query(query);
+	kfree(qname.name);
+	return err;
+}
+
+/**
+ * __apfs_undo_unlink - Clean up __apfs_unlink()
+ * @dentry: dentry to unlink
+ */
+static void __apfs_undo_unlink(struct dentry *dentry)
+{
+	struct inode *inode = d_inode(dentry);
+
+	/* Silence warning about nlink 0->1 */
+	apfs_inode_state_set_raw(inode, I_LINKABLE);
+	inc_nlink(inode);
+	apfs_inode_state_clear_raw(inode, I_LINKABLE);
+
+	apfs_undo_delete_dentry(dentry);
+}
+
+/**
+ * apfs_vol_filecnt_dec - Update the volume file count after a new orphaning
+ * @orphan: the new orphan
+ */
+static void apfs_vol_filecnt_dec(struct inode *orphan)
+{
+	struct super_block *sb = orphan->i_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+
+	switch (orphan->i_mode & S_IFMT) {
+	case S_IFREG:
+		le64_add_cpu(&vsb_raw->apfs_num_files, -1);
+		break;
+	case S_IFDIR:
+		le64_add_cpu(&vsb_raw->apfs_num_directories, -1);
+		break;
+	case S_IFLNK:
+		le64_add_cpu(&vsb_raw->apfs_num_symlinks, -1);
+		break;
+	default:
+		le64_add_cpu(&vsb_raw->apfs_num_other_fsobjects, -1);
+		break;
+	}
+}
+
+/**
+ * __apfs_unlink - Unlink a dentry
+ * @dir:    parent directory
+ * @dentry: dentry to unlink
+ *
+ * Does the same as apfs_unlink(), but without starting a transaction.
+ */
+static int __apfs_unlink(struct inode *dir, struct dentry *dentry)
+{
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_inode_info *ai = APFS_I(inode);
+	char *primary_name = NULL;
+	int err;
+
+	err = apfs_delete_dentry(dentry);
+	if (err) {
+		apfs_err(sb, "failed to delete dentry recs for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	drop_nlink(inode);
+	if (!inode->i_nlink) {
+		/* Orphaned inodes continue to report their old location */
+		err = apfs_create_orphan_link(inode);
+		/* Orphans are not included in the volume file counts */
+		apfs_vol_filecnt_dec(inode);
+	} else {
+		/* We may have deleted the primary link, so get the new one */
+		err = apfs_find_primary_link(inode, &primary_name,
+					     &ai->i_parent_id);
+	}
+	if (err)
+		goto fail;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode->i_ctime = dir->i_ctime;
+#else
+	inode_set_ctime_to_ts(inode, inode_get_ctime(dir));
+#endif
+	/* TODO: defer write of the primary name? */
+	err = apfs_update_inode(inode, primary_name);
+	if (err)
+		apfs_err(sb, "inode update failed for 0x%llx", apfs_ino(inode));
+
+fail:
+	kfree(primary_name);
+	primary_name = NULL;
+	if (err)
+		__apfs_undo_unlink(dentry);
+	return err;
+}
+
+int apfs_unlink(struct inode *dir, struct dentry *dentry)
+{
+	struct super_block *sb = dir->i_sb;
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_DEL);
+	if (err)
+		return err;
+
+	err = __apfs_unlink(dir, dentry);
+	if (err)
+		goto out_abort;
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto out_undo_unlink;
+	return 0;
+
+out_undo_unlink:
+	__apfs_undo_unlink(dentry);
+out_abort:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+int apfs_rmdir(struct inode *dir, struct dentry *dentry)
+{
+	struct inode *inode = d_inode(dentry);
+
+	if (APFS_I(inode)->i_nchildren)
+		return -ENOTEMPTY;
+	return apfs_unlink(dir, dentry);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+int apfs_rename(struct inode *old_dir, struct dentry *old_dentry,
+		struct inode *new_dir, struct dentry *new_dentry,
+		unsigned int flags)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+int apfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
+		struct dentry *old_dentry, struct inode *new_dir,
+		struct dentry *new_dentry, unsigned int flags)
+#else
+int apfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
+		struct dentry *old_dentry, struct inode *new_dir,
+		struct dentry *new_dentry, unsigned int flags)
+#endif
+{
+	struct super_block *sb = old_dir->i_sb;
+	struct inode *old_inode = d_inode(old_dentry);
+	struct inode *new_inode = d_inode(new_dentry);
+	int err;
+
+	if (new_inode && APFS_I(new_inode)->i_nchildren)
+		return -ENOTEMPTY;
+
+	if (flags & ~RENAME_NOREPLACE) /* TODO: support RENAME_EXCHANGE */
+		return -EINVAL;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	if (new_inode) {
+		err = __apfs_unlink(new_dir, new_dentry);
+		if (err) {
+			apfs_err(sb, "unlink failed for replaced dentry");
+			goto out_abort;
+		}
+	}
+
+	err = __apfs_link(old_dentry, new_dentry);
+	if (err) {
+		apfs_err(sb, "link failed for new dentry");
+		goto out_undo_unlink_new;
+	}
+
+	err = __apfs_unlink(old_dir, old_dentry);
+	if (err) {
+		apfs_err(sb, "unlink failed for old dentry");
+		goto out_undo_link;
+	}
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto out_undo_unlink_old;
+	return 0;
+
+out_undo_unlink_old:
+	__apfs_undo_unlink(old_dentry);
+out_undo_link:
+	__apfs_undo_link(new_dentry, old_inode);
+out_undo_unlink_new:
+	if (new_inode)
+		__apfs_undo_unlink(new_dentry);
+out_abort:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_any_orphan_ino - Find the inode number for any orphaned regular file
+ * @sb:		filesytem superblock
+ * @ino_p:	on return, the inode number found
+ *
+ * Returns 0 on success, or a negative error code in case of failure, which may
+ * be -ENODATA if there are no orphan files.
+ */
+u64 apfs_any_orphan_ino(struct super_block *sb, u64 *ino_p)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	struct apfs_drec drec = {0};
+	struct qstr qname = {0};
+	bool hashed = apfs_is_normalization_insensitive(sb);
+	bool found = false;
+	int err;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_drec_key(sb, APFS_PRIV_DIR_INO_NUM, NULL /* name */, 0 /* name_len */, &query->key);
+	query->flags = APFS_QUERY_CAT | APFS_QUERY_MULTIPLE | APFS_QUERY_EXACT;
+
+	while (!found) {
+		err = apfs_btree_query(sb, &query);
+		if (err) {
+			if (err == -ENODATA)
+				goto out;
+			apfs_err(sb, "drec query failed for private dir");
+			goto out;
+		}
+		err = apfs_drec_from_query(query, &drec, hashed);
+		if (err) {
+			apfs_alert(sb, "bad dentry record in private dir");
+			goto out;
+		}
+
+		/* These files are deleted immediately by ->evict_inode() */
+		if (drec.type != DT_REG)
+			continue;
+
+		/*
+		 * Confirm that this is an orphan file, because the official
+		 * reference allows other uses for the private directory.
+		 */
+		err = apfs_orphan_name(drec.ino, &qname);
+		if (err)
+			goto out;
+		found = strcmp(drec.name, qname.name) == 0;
+		kfree(qname.name);
+		qname.name = NULL;
+	}
+	*ino_p = drec.ino;
+
+out:
+	apfs_free_query(query);
+	query = NULL;
+	return err;
+}
diff --git a/fs/apfs/extents.c b/fs/apfs/extents.c
new file mode 100644
index 000000000..fabb2d230
--- /dev/null
+++ b/fs/apfs/extents.c
@@ -0,0 +1,2374 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/buffer_head.h>
+#include <linux/slab.h>
+#include <linux/blk_types.h>
+#include "apfs.h"
+
+/**
+ * apfs_ext_is_hole - Does this extent represent a hole in a sparse file?
+ * @extent: the extent to check
+ */
+static inline bool apfs_ext_is_hole(struct apfs_file_extent *extent)
+{
+	return extent->phys_block_num == 0;
+}
+
+/**
+ * apfs_size_to_blocks - Return the block count for a given size, rounded up
+ * @sb:		filesystem superblock
+ * @size:	size in bytes
+ *
+ * TODO: reuse for inode.c
+ */
+static inline u64 apfs_size_to_blocks(struct super_block *sb, u64 size)
+{
+	return (size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+}
+
+/**
+ * apfs_extent_from_query - Read the extent found by a successful query
+ * @query:	the query that found the record
+ * @extent:	Return parameter.  The extent found.
+ *
+ * Reads the extent record into @extent and performs some basic sanity checks
+ * as a protection against crafted filesystems.  Returns 0 on success or
+ * -EFSCORRUPTED otherwise.
+ */
+int apfs_extent_from_query(struct apfs_query *query,
+			   struct apfs_file_extent *extent)
+{
+	struct super_block *sb = query->node->object.sb;
+	char *raw = query->node->object.data;
+	u64 ext_len;
+
+	if (!apfs_is_sealed(sb)) {
+		struct apfs_file_extent_val *ext = NULL;
+		struct apfs_file_extent_key *ext_key = NULL;
+
+		if (query->len != sizeof(*ext) || query->key_len != sizeof(*ext_key)) {
+			apfs_err(sb, "bad length of key (%d) or value (%d)", query->key_len, query->len);
+			return -EFSCORRUPTED;
+		}
+
+		ext = (struct apfs_file_extent_val *)(raw + query->off);
+		ext_key = (struct apfs_file_extent_key *)(raw + query->key_off);
+		ext_len = le64_to_cpu(ext->len_and_flags) & APFS_FILE_EXTENT_LEN_MASK;
+
+		extent->logical_addr = le64_to_cpu(ext_key->logical_addr);
+		extent->phys_block_num = le64_to_cpu(ext->phys_block_num);
+		extent->crypto_id = le64_to_cpu(ext->crypto_id);
+	} else {
+		struct apfs_fext_tree_val *fext_val = NULL;
+		struct apfs_fext_tree_key *fext_key = NULL;
+
+		if (query->len != sizeof(*fext_val) || query->key_len != sizeof(*fext_key)) {
+			apfs_err(sb, "bad length of sealed key (%d) or value (%d)", query->key_len, query->len);
+			return -EFSCORRUPTED;
+		}
+
+		fext_val = (struct apfs_fext_tree_val *)(raw + query->off);
+		fext_key = (struct apfs_fext_tree_key *)(raw + query->key_off);
+		ext_len = le64_to_cpu(fext_val->len_and_flags) & APFS_FILE_EXTENT_LEN_MASK;
+
+		extent->logical_addr = le64_to_cpu(fext_key->logical_addr);
+		extent->phys_block_num = le64_to_cpu(fext_val->phys_block_num);
+		extent->crypto_id = 0;
+	}
+
+	/* Extent length must be a multiple of the block size */
+	if (ext_len & (sb->s_blocksize - 1)) {
+		apfs_err(sb, "invalid length (0x%llx)", ext_len);
+		return -EFSCORRUPTED;
+	}
+	extent->len = ext_len;
+	return 0;
+}
+
+/**
+ * apfs_extent_read - Read the extent record that covers a block
+ * @dstream:	data stream info
+ * @dsblock:	logical number of the wanted block (must be in range)
+ * @extent:	Return parameter.  The extent found.
+ *
+ * Finds and caches the extent record.  On success, returns a pointer to the
+ * cache record; on failure, returns an error code.
+ */
+static int apfs_extent_read(struct apfs_dstream_info *dstream, sector_t dsblock,
+			    struct apfs_file_extent *extent)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_key key;
+	struct apfs_query *query;
+	struct apfs_file_extent *cache = &dstream->ds_cached_ext;
+	u64 iaddr = dsblock << sb->s_blocksize_bits;
+	struct apfs_node *root = NULL;
+	int ret = 0;
+
+	spin_lock(&dstream->ds_ext_lock);
+	if (iaddr >= cache->logical_addr &&
+	    iaddr < cache->logical_addr + cache->len) {
+		*extent = *cache;
+		spin_unlock(&dstream->ds_ext_lock);
+		return 0;
+	}
+	spin_unlock(&dstream->ds_ext_lock);
+
+	/* We will search for the extent that covers iblock */
+	if (!apfs_is_sealed(sb)) {
+		apfs_init_file_extent_key(dstream->ds_id, iaddr, &key);
+		root = sbi->s_cat_root;
+	} else {
+		apfs_init_fext_key(dstream->ds_id, iaddr, &key);
+		root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_fext_tree_oid), APFS_OBJ_PHYSICAL, false /* write */);
+		if (IS_ERR(root)) {
+			apfs_err(sb, "failed to read fext root 0x%llx", le64_to_cpu(vsb_raw->apfs_fext_tree_oid));
+			return PTR_ERR(root);
+		}
+	}
+
+	query = apfs_alloc_query(root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto done;
+	}
+	query->key = key;
+	query->flags = apfs_is_sealed(sb) ? APFS_QUERY_FEXT : APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		apfs_err(sb, "query failed for id 0x%llx, addr 0x%llx", dstream->ds_id, iaddr);
+		if (ret == -ENODATA)
+			ret = -EFSCORRUPTED;
+		goto done;
+	}
+
+	ret = apfs_extent_from_query(query, extent);
+	if (ret) {
+		apfs_err(sb, "bad extent record for dstream 0x%llx", dstream->ds_id);
+		goto done;
+	}
+	if (iaddr < extent->logical_addr || iaddr >= extent->logical_addr + extent->len) {
+		apfs_err(sb, "no extent for addr 0x%llx in dstream 0x%llx", iaddr, dstream->ds_id);
+		ret = -EFSCORRUPTED;
+		goto done;
+	}
+
+	/*
+	 * For now prioritize the deferral of writes.
+	 * i_extent_dirty is protected by the read semaphore.
+	 */
+	if (!dstream->ds_ext_dirty) {
+		spin_lock(&dstream->ds_ext_lock);
+		*cache = *extent;
+		spin_unlock(&dstream->ds_ext_lock);
+	}
+
+done:
+	apfs_free_query(query);
+	if (apfs_is_sealed(sb))
+		apfs_node_free(root);
+	return ret;
+}
+
+/**
+ * apfs_logic_to_phys_bno - Find the physical block number for a dstream block
+ * @dstream:	data stream info
+ * @dsblock:	logical number of the wanted block
+ * @bno:	on return, the physical block number (or zero for holes)
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_logic_to_phys_bno(struct apfs_dstream_info *dstream, sector_t dsblock, u64 *bno)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_file_extent ext;
+	u64 blk_off;
+	int ret;
+
+	ret = apfs_extent_read(dstream, dsblock, &ext);
+	if (ret)
+		return ret;
+
+	if (apfs_ext_is_hole(&ext)) {
+		*bno = 0;
+		return 0;
+	}
+
+	/* Find the block offset of iblock within the extent */
+	blk_off = dsblock - (ext.logical_addr >> sb->s_blocksize_bits);
+	*bno = ext.phys_block_num + blk_off;
+	return 0;
+}
+
+/* This does the same as apfs_get_block(), but without taking any locks */
+int __apfs_get_block(struct apfs_dstream_info *dstream, sector_t dsblock,
+		     struct buffer_head *bh_result, int create)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_file_extent ext;
+	u64 blk_off, bno, map_len;
+	int ret;
+
+	ASSERT(!create);
+
+	if (dsblock >= apfs_size_to_blocks(sb, dstream->ds_size))
+		return 0;
+
+	ret = apfs_extent_read(dstream, dsblock, &ext);
+	if (ret) {
+		apfs_err(sb, "extent read failed");
+		return ret;
+	}
+
+	/* Find the block offset of iblock within the extent */
+	blk_off = dsblock - (ext.logical_addr >> sb->s_blocksize_bits);
+
+	/* Make sure we don't read past the extent boundaries */
+	map_len = ext.len - (blk_off << sb->s_blocksize_bits);
+	if (bh_result->b_size > map_len)
+		bh_result->b_size = map_len;
+
+	/*
+	 * Save the requested mapping length as apfs_map_bh() replaces it with
+	 * the filesystem block size
+	 */
+	map_len = bh_result->b_size;
+	/* Extents representing holes have block number 0 */
+	if (!apfs_ext_is_hole(&ext)) {
+		/* Find the block number of iblock within the disk */
+		bno = ext.phys_block_num + blk_off;
+		apfs_map_bh(bh_result, sb, bno);
+	}
+	bh_result->b_size = map_len;
+	return 0;
+}
+
+int apfs_get_block(struct inode *inode, sector_t iblock,
+		   struct buffer_head *bh_result, int create)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(inode->i_sb);
+	struct apfs_inode_info *ai = APFS_I(inode);
+	int ret;
+
+	down_read(&nxi->nx_big_sem);
+	ret = __apfs_get_block(&ai->i_dstream, iblock, bh_result, create);
+	up_read(&nxi->nx_big_sem);
+	return ret;
+}
+
+/**
+ * apfs_set_extent_length - Set a new length in an extent record's value
+ * @ext: the extent record's value
+ * @len: the new length
+ *
+ * Preserves the flags, though none are defined yet and I don't know if that
+ * will ever be important.
+ */
+static inline void apfs_set_extent_length(struct apfs_file_extent_val *ext, u64 len)
+{
+	u64 len_and_flags = le64_to_cpu(ext->len_and_flags);
+	u64 flags = len_and_flags & APFS_FILE_EXTENT_FLAG_MASK;
+
+	ext->len_and_flags = cpu_to_le64(flags | len);
+}
+
+static int apfs_range_put_reference(struct super_block *sb, u64 paddr, u64 length);
+
+/**
+ * apfs_shrink_extent_head - Shrink an extent record in its head
+ * @query:	the query that found the record
+ * @dstream:	data stream info
+ * @start:	new logical start for the extent
+ *
+ * Also deletes the physical extent records for the head. Returns 0 on success
+ * or a negative error code in case of failure.
+ */
+static int apfs_shrink_extent_head(struct apfs_query *query, struct apfs_dstream_info *dstream, u64 start)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_file_extent_key key;
+	struct apfs_file_extent_val val;
+	struct apfs_file_extent extent;
+	u64 new_len, head_len;
+	void *raw = NULL;
+	int err = 0;
+
+	err = apfs_extent_from_query(query, &extent);
+	if (err) {
+		apfs_err(sb, "bad extent record for dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+	raw = query->node->object.data;
+	key = *(struct apfs_file_extent_key *)(raw + query->key_off);
+	val = *(struct apfs_file_extent_val *)(raw + query->off);
+
+	new_len = extent.logical_addr + extent.len - start;
+	head_len = extent.len - new_len;
+
+	/* Delete the physical records for the blocks lost in the shrinkage */
+	if (!apfs_ext_is_hole(&extent)) {
+		err = apfs_range_put_reference(sb, extent.phys_block_num, head_len);
+		if (err) {
+			apfs_err(sb, "failed to put range 0x%llx-0x%llx", extent.phys_block_num, head_len);
+			return err;
+		}
+	} else {
+		dstream->ds_sparse_bytes -= head_len;
+	}
+
+	/* This is the actual shrinkage of the logical extent */
+	key.logical_addr = cpu_to_le64(start);
+	apfs_set_extent_length(&val, new_len);
+	if (!apfs_ext_is_hole(&extent))
+		le64_add_cpu(&val.phys_block_num, head_len >> sb->s_blocksize_bits);
+	return apfs_btree_replace(query, &key, sizeof(key), &val, sizeof(val));
+}
+
+/**
+ * apfs_shrink_extent_tail - Shrink an extent record in its tail
+ * @query:	the query that found the record
+ * @dstream:	data stream info
+ * @end:	new logical end for the extent
+ *
+ * Also puts the physical extent records for the tail. Returns 0 on success or
+ * a negative error code in case of failure.
+ */
+static int apfs_shrink_extent_tail(struct apfs_query *query, struct apfs_dstream_info *dstream, u64 end)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_file_extent_val *val;
+	struct apfs_file_extent extent;
+	u64 new_len, new_blkcount, tail_len;
+	void *raw;
+	int err = 0;
+
+	ASSERT((end & (sb->s_blocksize - 1)) == 0);
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+
+	err = apfs_extent_from_query(query, &extent);
+	if (err) {
+		apfs_err(sb, "bad extent record for dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+	val = raw + query->off;
+
+	new_len = end - extent.logical_addr;
+	new_blkcount = new_len >> sb->s_blocksize_bits;
+	tail_len = extent.len - new_len;
+
+	/* Delete the physical records for the blocks lost in the shrinkage */
+	if (!apfs_ext_is_hole(&extent)) {
+		err = apfs_range_put_reference(sb, extent.phys_block_num + new_blkcount, tail_len);
+		if (err) {
+			apfs_err(sb, "failed to put range 0x%llx-0x%llx", extent.phys_block_num + new_blkcount, tail_len);
+			return err;
+		}
+	} else {
+		dstream->ds_sparse_bytes -= tail_len;
+	}
+
+	/* This is the actual shrinkage of the logical extent */
+	apfs_set_extent_length(val, new_len);
+	return err;
+}
+
+/**
+ * apfs_query_found_extent - Did this query find an extent with the right id?
+ * @query: the (successful) query that found the record
+ */
+static inline bool apfs_query_found_extent(struct apfs_query *query)
+{
+	void *raw = query->node->object.data;
+	struct apfs_key_header *hdr;
+
+	if (query->key_len < sizeof(*hdr))
+		return false;
+	hdr = raw + query->key_off;
+
+	if (apfs_cat_type(hdr) != APFS_TYPE_FILE_EXTENT)
+		return false;
+	if (apfs_cat_cnid(hdr) != query->key.id)
+		return false;
+	return true;
+}
+
+/**
+ * apfs_update_tail_extent - Grow the tail extent for a data stream
+ * @dstream:	data stream info
+ * @extent:	new in-memory extent
+ *
+ * Also takes care of any needed changes to the physical extent records. Returns
+ * 0 on success or a negative error code in case of failure.
+ */
+static int apfs_update_tail_extent(struct apfs_dstream_info *dstream, const struct apfs_file_extent *extent)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_file_extent_key raw_key;
+	struct apfs_file_extent_val raw_val;
+	u64 extent_id = dstream->ds_id;
+	int ret;
+	u64 new_crypto;
+
+	apfs_key_set_hdr(APFS_TYPE_FILE_EXTENT, extent_id, &raw_key);
+	raw_key.logical_addr = cpu_to_le64(extent->logical_addr);
+	raw_val.len_and_flags = cpu_to_le64(extent->len);
+	raw_val.phys_block_num = cpu_to_le64(extent->phys_block_num);
+	if (apfs_vol_is_encrypted(sb))
+		new_crypto = extent_id;
+	else
+		new_crypto = 0;
+	raw_val.crypto_id = cpu_to_le64(new_crypto);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	/* We want the last extent record */
+	apfs_init_file_extent_key(extent_id, -1, &query->key);
+	query->flags = APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for last extent of id 0x%llx", extent_id);
+		goto out;
+	}
+
+	if (ret == -ENODATA || !apfs_query_found_extent(query)) {
+		/* We are creating the first extent for the file */
+		ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+		if (ret) {
+			apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+			goto out;
+		}
+	} else {
+		struct apfs_file_extent tail;
+
+		ret = apfs_extent_from_query(query, &tail);
+		if (ret) {
+			apfs_err(sb, "bad extent record for dstream 0x%llx", dstream->ds_id);
+			goto out;
+		}
+
+		if (tail.logical_addr > extent->logical_addr) {
+			apfs_alert(sb, "extent is not tail - bug!");
+			ret = -EOPNOTSUPP;
+			goto out;
+		} else if (tail.logical_addr == extent->logical_addr) {
+			ret = apfs_btree_replace(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+			if (ret) {
+				apfs_err(sb, "update failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+				goto out;
+			}
+			if (apfs_ext_is_hole(&tail)) {
+				dstream->ds_sparse_bytes -= tail.len;
+			} else if (tail.phys_block_num != extent->phys_block_num) {
+				ret = apfs_range_put_reference(sb, tail.phys_block_num, tail.len);
+				if (ret) {
+					apfs_err(sb, "failed to put range 0x%llx-0x%llx", tail.phys_block_num, tail.len);
+					goto out;
+				}
+			}
+			if (new_crypto == tail.crypto_id)
+				goto out;
+			ret = apfs_crypto_adj_refcnt(sb, tail.crypto_id, -1);
+			if (ret) {
+				apfs_err(sb, "failed to put crypto id 0x%llx", tail.crypto_id);
+				goto out;
+			}
+		} else {
+			/*
+			 * TODO: we could actually also continue the tail extent
+			 * if it's right next to the new one (both logically and
+			 * physically), even if they don't overlap. Or maybe we
+			 * should always make sure that the tail extent is in
+			 * the cache before a write...
+			 */
+			if (extent->logical_addr < tail.logical_addr + tail.len) {
+				ret = apfs_shrink_extent_tail(query, dstream, extent->logical_addr);
+				if (ret) {
+					apfs_err(sb, "failed to shrink tail of dstream 0x%llx", extent_id);
+					goto out;
+				}
+			}
+			ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+			if (ret) {
+				apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+				goto out;
+			}
+		}
+	}
+
+	ret = apfs_crypto_adj_refcnt(sb, new_crypto, 1);
+	if (ret)
+		apfs_err(sb, "failed to take crypto id 0x%llx", new_crypto);
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_split_extent - Break an extent in two
+ * @query:	query pointing to the extent
+ * @div:	logical address for the division
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_split_extent(struct apfs_query *query, u64 div)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_file_extent_val *val1;
+	struct apfs_file_extent_key key2;
+	struct apfs_file_extent_val val2;
+	struct apfs_file_extent extent;
+	u64 len1, len2, blkcount1;
+	void *raw;
+	int err = 0;
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+
+	err = apfs_extent_from_query(query, &extent);
+	if (err) {
+		apfs_err(sb, "bad extent record");
+		return err;
+	}
+	val1 = raw + query->off;
+	val2 = *(struct apfs_file_extent_val *)(raw + query->off);
+	key2 = *(struct apfs_file_extent_key *)(raw + query->key_off);
+
+	len1 = div - extent.logical_addr;
+	blkcount1 = len1 >> sb->s_blocksize_bits;
+	len2 = extent.len - len1;
+
+	/* Modify the current extent in place to become the first half */
+	apfs_set_extent_length(val1, len1);
+
+	/* Insert the second half right after the first */
+	key2.logical_addr = cpu_to_le64(div);
+	if (!apfs_ext_is_hole(&extent))
+		val2.phys_block_num = cpu_to_le64(extent.phys_block_num + blkcount1);
+	apfs_set_extent_length(&val2, len2);
+	err = apfs_btree_insert(query, &key2, sizeof(key2), &val2, sizeof(val2));
+	if (err) {
+		apfs_err(sb, "insertion failed in division 0x%llx", div);
+		return err;
+	}
+
+	return apfs_crypto_adj_refcnt(sb, extent.crypto_id, 1);
+}
+
+/**
+ * apfs_update_mid_extent - Create or update a non-tail extent for a dstream
+ * @dstream:	data stream info
+ * @extent:	new in-memory extent
+ *
+ * Also takes care of any needed changes to the physical extent records. Returns
+ * 0 on success or a negative error code in case of failure.
+ */
+static int apfs_update_mid_extent(struct apfs_dstream_info *dstream, const struct apfs_file_extent *extent)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_key key;
+	struct apfs_query *query;
+	struct apfs_file_extent_key raw_key;
+	struct apfs_file_extent_val raw_val;
+	struct apfs_file_extent prev_ext;
+	u64 extent_id = dstream->ds_id;
+	u64 prev_crypto, new_crypto;
+	u64 prev_start, prev_end;
+	bool second_run = false;
+	int ret;
+
+	apfs_key_set_hdr(APFS_TYPE_FILE_EXTENT, extent_id, &raw_key);
+	raw_key.logical_addr = cpu_to_le64(extent->logical_addr);
+	raw_val.len_and_flags = cpu_to_le64(extent->len);
+	raw_val.phys_block_num = cpu_to_le64(extent->phys_block_num);
+	if (apfs_vol_is_encrypted(sb))
+		new_crypto = extent_id;
+	else
+		new_crypto = 0;
+	raw_val.crypto_id = cpu_to_le64(new_crypto);
+
+	apfs_init_file_extent_key(extent_id, extent->logical_addr, &key);
+
+search_and_insert:
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	query->key = key;
+	query->flags = APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+		goto out;
+	}
+
+	if (ret == -ENODATA || !apfs_query_found_extent(query)) {
+		/*
+		 * The new extent goes in a hole we just made, right at the
+		 * beginning of the file.
+		 */
+		if (!second_run) {
+			apfs_err(sb, "missing extent on dstream 0x%llx", extent_id);
+			ret = -EFSCORRUPTED;
+		} else {
+			ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+			if (ret)
+				apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+		}
+		goto out;
+	}
+
+	if (apfs_extent_from_query(query, &prev_ext)) {
+		apfs_err(sb, "bad mid extent record on dstream 0x%llx", extent_id);
+		ret = -EFSCORRUPTED;
+		goto out;
+	}
+	prev_crypto = prev_ext.crypto_id;
+	prev_start = prev_ext.logical_addr;
+	prev_end = prev_ext.logical_addr + prev_ext.len;
+
+	if (prev_end == extent->logical_addr && second_run) {
+		/* The new extent goes in the hole we just made */
+		ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+		if (ret) {
+			apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+			goto out;
+		}
+	} else if (prev_start == extent->logical_addr && prev_ext.len == extent->len) {
+		/* The old and new extents are the same logical block */
+		ret = apfs_btree_replace(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+		if (ret) {
+			apfs_err(sb, "update failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+			goto out;
+		}
+		if (apfs_ext_is_hole(&prev_ext)) {
+			dstream->ds_sparse_bytes -= prev_ext.len;
+		} else if (prev_ext.phys_block_num != extent->phys_block_num) {
+			ret = apfs_range_put_reference(sb, prev_ext.phys_block_num, prev_ext.len);
+			if (ret) {
+				apfs_err(sb, "failed to put range 0x%llx-0x%llx", prev_ext.phys_block_num, prev_ext.len);
+				goto out;
+			}
+		}
+		ret = apfs_crypto_adj_refcnt(sb, prev_crypto, -1);
+		if (ret) {
+			apfs_err(sb, "failed to put crypto id 0x%llx", prev_crypto);
+			goto out;
+		}
+	} else if (prev_start == extent->logical_addr) {
+		/* The new extent is the first logical block of the old one */
+		if (second_run) {
+			/* I don't know if this is possible, but be safe */
+			apfs_alert(sb, "recursion shrinking extent head for dstream 0x%llx", extent_id);
+			ret = -EFSCORRUPTED;
+			goto out;
+		}
+		ret = apfs_shrink_extent_head(query, dstream, extent->logical_addr + extent->len);
+		if (ret) {
+			apfs_err(sb, "failed to shrink extent in dstream 0x%llx", extent_id);
+			goto out;
+		}
+		/* The query should point to the previous record, start again */
+		apfs_free_query(query);
+		second_run = true;
+		goto search_and_insert;
+	} else if (prev_end == extent->logical_addr + extent->len) {
+		/* The new extent is the last logical block of the old one */
+		ret = apfs_shrink_extent_tail(query, dstream, extent->logical_addr);
+		if (ret) {
+			apfs_err(sb, "failed to shrink extent in dstream 0x%llx", extent_id);
+			goto out;
+		}
+		ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+		if (ret) {
+			apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", extent_id, extent->logical_addr);
+			goto out;
+		}
+	} else if (prev_start < extent->logical_addr && prev_end > extent->logical_addr + extent->len) {
+		/* The new extent is logically in the middle of the old one */
+		if (second_run) {
+			/* I don't know if this is possible, but be safe */
+			apfs_alert(sb, "recursion when splitting extents for dstream 0x%llx", extent_id);
+			ret = -EFSCORRUPTED;
+			goto out;
+		}
+		ret = apfs_split_extent(query, extent->logical_addr + extent->len);
+		if (ret) {
+			apfs_err(sb, "failed to split extent in dstream 0x%llx", extent_id);
+			goto out;
+		}
+		/* The split may make the query invalid */
+		apfs_free_query(query);
+		second_run = true;
+		goto search_and_insert;
+	} else {
+		/* I don't know what this is, be safe */
+		apfs_alert(sb, "strange extents for dstream 0x%llx", extent_id);
+		ret = -EFSCORRUPTED;
+		goto out;
+	}
+
+	ret = apfs_crypto_adj_refcnt(sb, new_crypto, 1);
+	if (ret)
+		apfs_err(sb, "failed to take crypto id 0x%llx", new_crypto);
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_update_extent - Create or update the extent record for an extent
+ * @dstream:	data stream info
+ * @extent:	new in-memory file extent
+ *
+ * The @extent must either be a new tail for the dstream, or a single block.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_update_extent(struct apfs_dstream_info *dstream, const struct apfs_file_extent *extent)
+{
+	struct super_block *sb = dstream->ds_sb;
+
+	if (extent->logical_addr + extent->len >= dstream->ds_size)
+		return apfs_update_tail_extent(dstream, extent);
+	if (extent->len > sb->s_blocksize) {
+		apfs_err(sb, "can't create mid extents of length 0x%llx", extent->len);
+		return -EOPNOTSUPP;
+	}
+	return apfs_update_mid_extent(dstream, extent);
+}
+
+static int apfs_extend_phys_extent(struct apfs_query *query, u64 bno, u64 blkcnt, u64 dstream_id)
+{
+	struct apfs_phys_ext_key raw_key;
+	struct apfs_phys_ext_val raw_val;
+	u64 kind = (u64)APFS_KIND_NEW << APFS_PEXT_KIND_SHIFT;
+
+	apfs_key_set_hdr(APFS_TYPE_EXTENT, bno, &raw_key);
+	raw_val.len_and_kind = cpu_to_le64(kind | blkcnt);
+	raw_val.owning_obj_id = cpu_to_le64(dstream_id);
+	raw_val.refcnt = cpu_to_le32(1);
+	return apfs_btree_replace(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+}
+
+static int apfs_insert_new_phys_extent(struct apfs_query *query, u64 bno, u64 blkcnt, u64 dstream_id)
+{
+	struct apfs_phys_ext_key raw_key;
+	struct apfs_phys_ext_val raw_val;
+	u64 kind = (u64)APFS_KIND_NEW << APFS_PEXT_KIND_SHIFT;
+
+	apfs_key_set_hdr(APFS_TYPE_EXTENT, bno, &raw_key);
+	raw_val.len_and_kind = cpu_to_le64(kind | blkcnt);
+	raw_val.owning_obj_id = cpu_to_le64(dstream_id);
+	raw_val.refcnt = cpu_to_le32(1);
+	return apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+}
+
+static int apfs_phys_ext_from_query(struct apfs_query *query, struct apfs_phys_extent *pext);
+
+/**
+ * apfs_insert_phys_extent - Create or grow the physical record for an extent
+ * @dstream:	data stream info for the extent
+ * @extent:	new in-memory file extent
+ *
+ * Only works for appending to extents, for now. TODO: reference counting.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_insert_phys_extent(struct apfs_dstream_info *dstream, const struct apfs_file_extent *extent)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_node *extref_root;
+	struct apfs_query *query = NULL;
+	struct apfs_phys_extent pext;
+	u64 blkcnt = extent->len >> sb->s_blocksize_bits;
+	u64 last_bno, new_base, new_blkcnt;
+	int ret;
+
+	extref_root = apfs_read_node(sb,
+				le64_to_cpu(vsb_raw->apfs_extentref_tree_oid),
+				APFS_OBJ_PHYSICAL, true /* write */);
+	if (IS_ERR(extref_root)) {
+		apfs_err(sb, "failed to read extref root 0x%llx", le64_to_cpu(vsb_raw->apfs_extentref_tree_oid));
+		return PTR_ERR(extref_root);
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	vsb_raw->apfs_extentref_tree_oid = cpu_to_le64(extref_root->object.oid);
+
+	query = apfs_alloc_query(extref_root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	/*
+	 * The cached logical extent may have been split into multiple physical
+	 * extents because of clones. If that happens, we want to grow the last
+	 * one.
+	 */
+	last_bno = extent->phys_block_num + blkcnt - 1;
+	apfs_init_extent_key(last_bno, &query->key);
+	query->flags = APFS_QUERY_EXTENTREF;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for paddr 0x%llx", last_bno);
+		goto out;
+	}
+
+	if (ret == -ENODATA) {
+		/* This is a fresh new physical extent */
+		ret = apfs_insert_new_phys_extent(query, extent->phys_block_num, blkcnt, dstream->ds_id);
+		if (ret)
+			apfs_err(sb, "insertion failed for paddr 0x%llx", extent->phys_block_num);
+		goto out;
+	}
+
+	ret = apfs_phys_ext_from_query(query, &pext);
+	if (ret) {
+		apfs_err(sb, "bad pext record for bno 0x%llx", last_bno);
+		goto out;
+	}
+	if (pext.bno + pext.blkcount <= extent->phys_block_num) {
+		/* Also a fresh new physical extent */
+		ret = apfs_insert_new_phys_extent(query, extent->phys_block_num, blkcnt, dstream->ds_id);
+		if (ret)
+			apfs_err(sb, "insertion failed for paddr 0x%llx", extent->phys_block_num);
+		goto out;
+	}
+
+	/*
+	 * There is an existing physical extent that overlaps the new one. The
+	 * cache was dirty, so the existing extent can't cover the whole tail.
+	 */
+	if (pext.bno + pext.blkcount >= extent->phys_block_num + blkcnt) {
+		apfs_err(sb, "dirty cache tail covered by existing physical extent 0x%llx-0x%llx", pext.bno, pext.blkcount);
+		ret = -EFSCORRUPTED;
+		goto out;
+	}
+	if (pext.refcnt == 1) {
+		new_base = pext.bno;
+		new_blkcnt = extent->phys_block_num + blkcnt - new_base;
+		ret = apfs_extend_phys_extent(query, new_base, new_blkcnt, dstream->ds_id);
+		if (ret)
+			apfs_err(sb, "update failed for paddr 0x%llx", new_base);
+	} else {
+		/*
+		 * We can't extend this one, because it would extend the other
+		 * references as well.
+		 */
+		new_base = pext.bno + pext.blkcount;
+		new_blkcnt = extent->phys_block_num + blkcnt - new_base;
+		ret = apfs_insert_new_phys_extent(query, new_base, new_blkcnt, dstream->ds_id);
+		if (ret)
+			apfs_err(sb, "insertion failed for paddr 0x%llx", new_base);
+	}
+
+out:
+	apfs_free_query(query);
+	apfs_node_free(extref_root);
+	return ret;
+}
+
+/**
+ * apfs_phys_ext_from_query - Read the physical extent record found by a query
+ * @query:	the (successful) query that found the record
+ * @pext:	on return, the physical extent read
+ *
+ * Reads the physical extent record into @pext and performs some basic sanity
+ * checks as a protection against crafted filesystems. Returns 0 on success or
+ * -EFSCORRUPTED otherwise.
+ */
+static int apfs_phys_ext_from_query(struct apfs_query *query, struct apfs_phys_extent *pext)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_phys_ext_key *key;
+	struct apfs_phys_ext_val *val;
+	char *raw = query->node->object.data;
+
+	if (query->len != sizeof(*val) || query->key_len != sizeof(*key)) {
+		apfs_err(sb, "bad length of key (%d) or value (%d)", query->key_len, query->len);
+		return -EFSCORRUPTED;
+	}
+
+	key = (struct apfs_phys_ext_key *)(raw + query->key_off);
+	val = (struct apfs_phys_ext_val *)(raw + query->off);
+
+	pext->bno = apfs_cat_cnid(&key->hdr);
+	pext->blkcount = le64_to_cpu(val->len_and_kind) & APFS_PEXT_LEN_MASK;
+	pext->len = pext->blkcount << sb->s_blocksize_bits;
+	pext->refcnt = le32_to_cpu(val->refcnt);
+	pext->kind = le64_to_cpu(val->len_and_kind) >> APFS_PEXT_KIND_SHIFT;
+	return 0;
+}
+
+/**
+ * apfs_free_phys_ext - Add all blocks in a physical extent to the free queue
+ * @sb:		superblock structure
+ * @pext:	physical range to free
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_free_phys_ext(struct super_block *sb, struct apfs_phys_extent *pext)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, -pext->blkcount);
+	le64_add_cpu(&vsb_raw->apfs_total_blocks_freed, pext->blkcount);
+
+	return apfs_free_queue_insert(sb, pext->bno, pext->blkcount);
+}
+
+/**
+ * apfs_put_phys_extent - Reduce the reference count for a physical extent
+ * @pext:	physical extent data, already read
+ * @query:	query that found the extent
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_put_phys_extent(struct apfs_phys_extent *pext, struct apfs_query *query)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_phys_ext_val *val;
+	void *raw;
+	int err;
+
+	if (--pext->refcnt == 0) {
+		err = apfs_btree_remove(query);
+		if (err) {
+			apfs_err(sb, "removal failed for paddr 0x%llx", pext->bno);
+			return err;
+		}
+		return pext->kind == APFS_KIND_NEW ? apfs_free_phys_ext(sb, pext) : 0;
+	}
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+	val = raw + query->off;
+	val->refcnt = cpu_to_le32(pext->refcnt);
+	return 0;
+}
+
+/**
+ * apfs_take_phys_extent - Increase the reference count for a physical extent
+ * @pext:	physical extent data, already read
+ * @query:	query that found the extent
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_take_phys_extent(struct apfs_phys_extent *pext, struct apfs_query *query)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_phys_ext_val *val;
+	void *raw;
+	int err;
+
+	/* An update extent may be dropped when a reference is taken */
+	if (++pext->refcnt == 0)
+		return apfs_btree_remove(query);
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+	val = raw + query->off;
+	val->refcnt = cpu_to_le32(pext->refcnt);
+	return 0;
+}
+
+/**
+ * apfs_set_phys_ext_length - Set new length in a physical extent record's value
+ * @pext:	the physical extent record's value
+ * @len:	the new length (in blocks)
+ *
+ * Preserves the kind, though I doubt that's the right thing to do in general.
+ */
+static inline void apfs_set_phys_ext_length(struct apfs_phys_ext_val *pext, u64 len)
+{
+	u64 len_and_kind = le64_to_cpu(pext->len_and_kind);
+	u64 kind = len_and_kind & APFS_PEXT_KIND_MASK;
+
+	pext->len_and_kind = cpu_to_le64(kind | len);
+}
+
+/**
+ * apfs_split_phys_ext - Break a physical extent in two
+ * @query:	query pointing to the extent
+ * @div:	first physical block number to come after the division
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_split_phys_ext(struct apfs_query *query, u64 div)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_phys_ext_val *val1;
+	struct apfs_phys_ext_key key2;
+	struct apfs_phys_ext_val val2;
+	struct apfs_phys_extent pextent;
+	u64 blkcount1, blkcount2;
+	void *raw;
+	int err = 0;
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+
+	err = apfs_phys_ext_from_query(query, &pextent);
+	if (err) {
+		apfs_err(sb, "bad pext record over div 0x%llx", div);
+		return err;
+	}
+	val1 = raw + query->off;
+	val2 = *(struct apfs_phys_ext_val *)(raw + query->off);
+	key2 = *(struct apfs_phys_ext_key *)(raw + query->key_off);
+
+	blkcount1 = div - pextent.bno;
+	blkcount2 = pextent.blkcount - blkcount1;
+
+	/* Modify the current extent in place to become the first half */
+	apfs_set_phys_ext_length(val1, blkcount1);
+
+	/* Insert the second half right after the first */
+	apfs_key_set_hdr(APFS_TYPE_EXTENT, div, &key2);
+	apfs_set_phys_ext_length(&val2, blkcount2);
+	return apfs_btree_insert(query, &key2, sizeof(key2), &val2, sizeof(val2));
+}
+
+/**
+ * apfs_create_update_pext - Create a reference update physical extent record
+ * @query:	query that searched for the physical extent
+ * @extent:	range of physical blocks to update
+ * @diff:	reference count change
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_update_pext(struct apfs_query *query, const struct apfs_file_extent *extent, u32 diff)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_phys_ext_key key = {0};
+	struct apfs_phys_ext_val val = {0};
+
+	apfs_key_set_hdr(APFS_TYPE_EXTENT, extent->phys_block_num, &key);
+	val.len_and_kind = cpu_to_le64((u64)APFS_KIND_UPDATE << APFS_PEXT_KIND_SHIFT | extent->len >> sb->s_blocksize_bits);
+	val.owning_obj_id = cpu_to_le64(APFS_OWNING_OBJ_ID_INVALID);
+	val.refcnt = cpu_to_le32(diff);
+	return apfs_btree_insert(query, &key, sizeof(key), &val, sizeof(val));
+}
+
+/**
+ * apfs_dstream_cache_is_tail - Is the tail of this dstream in its extent cache?
+ * @dstream: dstream to check
+ */
+static inline bool apfs_dstream_cache_is_tail(struct apfs_dstream_info *dstream)
+{
+	struct apfs_file_extent *cache = &dstream->ds_cached_ext;
+
+	/* nx_big_sem provides the locking for the cache here */
+	return cache->len && (dstream->ds_size <= cache->logical_addr + cache->len);
+}
+
+/**
+ * apfs_flush_extent_cache - Write the cached extent to the catalog, if dirty
+ * @dstream: data stream to flush
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_flush_extent_cache(struct apfs_dstream_info *dstream)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_file_extent *ext = &dstream->ds_cached_ext;
+	int err;
+
+	if (!dstream->ds_ext_dirty)
+		return 0;
+	ASSERT(ext->len > 0);
+
+	err = apfs_update_extent(dstream, ext);
+	if (err) {
+		apfs_err(sb, "extent update failed");
+		return err;
+	}
+	err = apfs_insert_phys_extent(dstream, ext);
+	if (err) {
+		apfs_err(sb, "pext insertion failed");
+		return err;
+	}
+
+	/*
+	 * TODO: keep track of the byte and block count through the use of
+	 * inode_add_bytes() and inode_set_bytes(). This hasn't been done with
+	 * care in the rest of the module and it doesn't seem to matter beyond
+	 * stat(), so I'm ignoring it for now.
+	 */
+
+	dstream->ds_ext_dirty = false;
+	return 0;
+}
+
+/**
+ * apfs_create_hole - Create and insert a hole extent for the dstream
+ * @dstream:	data stream info
+ * @start:	first logical block number for the hole
+ * @end:	first logical block number right after the hole
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ * TODO: what happens to the crypto refcount?
+ */
+static int apfs_create_hole(struct apfs_dstream_info *dstream, u64 start, u64 end)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_file_extent_key raw_key;
+	struct apfs_file_extent_val raw_val;
+	u64 extent_id = dstream->ds_id;
+	int ret;
+
+	if (start == end)
+		return 0;
+
+	/* File extent records use addresses, not block numbers */
+	start <<= sb->s_blocksize_bits;
+	end <<= sb->s_blocksize_bits;
+
+	apfs_key_set_hdr(APFS_TYPE_FILE_EXTENT, extent_id, &raw_key);
+	raw_key.logical_addr = cpu_to_le64(start);
+	raw_val.len_and_flags = cpu_to_le64(end - start);
+	raw_val.phys_block_num = cpu_to_le64(0); /* It's a hole... */
+	raw_val.crypto_id = cpu_to_le64(apfs_vol_is_encrypted(sb) ? extent_id : 0);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_file_extent_key(extent_id, start, &query->key);
+	query->flags = APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for id 0x%llx, addr 0x%llx", extent_id, start);
+		goto out;
+	}
+
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+	if (ret)
+		apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", extent_id, start);
+	dstream->ds_sparse_bytes += end - start;
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_zero_dstream_tail - Zero out stale bytes in a data stream's last block
+ * @dstream: data stream info
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_zero_dstream_tail(struct apfs_dstream_info *dstream)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct inode *inode = NULL;
+	struct page *page = NULL;
+	void *fsdata = NULL;
+	int valid_length;
+	int err;
+
+	/* No stale bytes if no actual content */
+	if (dstream->ds_size <= dstream->ds_sparse_bytes)
+		return 0;
+
+	/* No stale tail if the last block is fully used */
+	valid_length = dstream->ds_size & (sb->s_blocksize - 1);
+	if (valid_length == 0)
+		return 0;
+
+	inode = dstream->ds_inode;
+	if (!inode) {
+		/* This should never happen, but be safe */
+		apfs_alert(sb, "attempt to zero the tail of xattr dstream 0x%llx", dstream->ds_id);
+		return -EFSCORRUPTED;
+	}
+
+	/* This will take care of the CoW and zeroing */
+	err = __apfs_write_begin(NULL, inode->i_mapping, inode->i_size, 0, 0, &page, &fsdata);
+	if (err)
+		return err;
+	return __apfs_write_end(NULL, inode->i_mapping, inode->i_size, 0, 0, page, fsdata);
+}
+
+/**
+ * apfs_zero_bh_tail - Zero out stale bytes in a buffer head
+ * @sb:		filesystem superblock
+ * @bh:		buffer head to zero
+ * @length:	length of valid bytes to be left alone
+ */
+static void apfs_zero_bh_tail(struct super_block *sb, struct buffer_head *bh, u64 length)
+{
+	ASSERT(buffer_trans(bh));
+	if (length < sb->s_blocksize)
+		memset(bh->b_data + length, 0, sb->s_blocksize - length);
+}
+
+/**
+ * apfs_range_in_snap - Check if a given block range overlaps a snapshot
+ * @sb:		filesystem superblock
+ * @bno:	first block in the range
+ * @blkcnt:	block count for the range
+ * @in_snap:	on return, the result
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_range_in_snap(struct super_block *sb, u64 bno, u64 blkcnt, bool *in_snap)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_node *extref_root = NULL;
+	struct apfs_query *query = NULL;
+	struct apfs_phys_extent pext = {0};
+	int ret;
+
+	/* Avoid the tree queries when we don't even have snapshots */
+	if (vsb_raw->apfs_num_snapshots == 0) {
+		*in_snap = false;
+		return 0;
+	}
+
+	/*
+	 * Now check if the current physical extent tree has an entry for
+	 * these blocks
+	 */
+	extref_root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_extentref_tree_oid), APFS_OBJ_PHYSICAL, false /* write */);
+	if (IS_ERR(extref_root)) {
+		apfs_err(sb, "failed to read extref root 0x%llx", le64_to_cpu(vsb_raw->apfs_extentref_tree_oid));
+		return PTR_ERR(extref_root);
+	}
+
+	query = apfs_alloc_query(extref_root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	apfs_init_extent_key(bno, &query->key);
+	query->flags = APFS_QUERY_EXTENTREF;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for paddr 0x%llx", bno);
+		goto out;
+	}
+	if (ret == -ENODATA) {
+		*in_snap = false;
+		ret = 0;
+		goto out;
+	}
+
+	ret = apfs_phys_ext_from_query(query, &pext);
+	if (ret) {
+		apfs_err(sb, "bad pext record for paddr 0x%llx", bno);
+		goto out;
+	}
+
+	if (pext.bno <= bno && pext.bno + pext.blkcount >= bno + blkcnt) {
+		if (pext.kind == APFS_KIND_NEW) {
+			*in_snap = false;
+			goto out;
+		}
+	}
+
+	/*
+	 * I think the file extent could still be covered by two different
+	 * physical extents from the current tree, but it's easier to just
+	 * assume the worst here.
+	 */
+	*in_snap = true;
+
+out:
+	apfs_free_query(query);
+	apfs_node_free(extref_root);
+	return ret;
+}
+
+/**
+ * apfs_dstream_cache_in_snap - Check if the cached extent overlaps a snapshot
+ * @dstream:	the data stream to check
+ * @in_snap:	on return, the result
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_dstream_cache_in_snap(struct apfs_dstream_info *dstream, bool *in_snap)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_file_extent *cache = NULL;
+
+	/* All changes to extents get flushed when a snaphot is created */
+	if (dstream->ds_ext_dirty) {
+		*in_snap = false;
+		return 0;
+	}
+
+	cache = &dstream->ds_cached_ext;
+	return apfs_range_in_snap(sb, cache->phys_block_num, cache->len >> sb->s_blocksize_bits, in_snap);
+}
+
+/**
+ * apfs_dstream_get_new_block - Like the get_block_t function, but for dstreams
+ * @dstream:	data stream info
+ * @dsblock:	logical dstream block to map
+ * @bh_result:	buffer head to map (NULL if none)
+ * @bno:	if not NULL, the new block number is returned here
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_dstream_get_new_block(struct apfs_dstream_info *dstream, u64 dsblock, struct buffer_head *bh_result, u64 *bno)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_file_extent *cache = NULL;
+	u64 phys_bno, logical_addr, cache_blks, dstream_blks;
+	bool in_snap = true;
+	int err;
+
+	/* TODO: preallocate tail blocks */
+	logical_addr = dsblock << sb->s_blocksize_bits;
+
+	err = apfs_spaceman_allocate_block(sb, &phys_bno, false /* backwards */);
+	if (err) {
+		apfs_err(sb, "block allocation failed");
+		return err;
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, 1);
+	le64_add_cpu(&vsb_raw->apfs_total_blocks_alloced, 1);
+	if (bno)
+		*bno = phys_bno;
+
+	if (bh_result) {
+		apfs_map_bh(bh_result, sb, phys_bno);
+		err = apfs_transaction_join(sb, bh_result);
+		if (err)
+			return err;
+
+		if (!buffer_uptodate(bh_result)) {
+			/*
+			 * Truly new buffers need to be marked as such, to get
+			 * zeroed; this also takes care of holes in sparse files
+			 */
+			set_buffer_new(bh_result);
+		} else if (dstream->ds_size > logical_addr) {
+			/*
+			 * The last block may have stale data left from a
+			 * truncation
+			 */
+			apfs_zero_bh_tail(sb, bh_result, dstream->ds_size - logical_addr);
+		}
+	}
+
+	dstream_blks = apfs_size_to_blocks(sb, dstream->ds_size);
+	if (dstream_blks < dsblock) {
+		/*
+		 * This recurses into apfs_dstream_get_new_block() and dirties
+		 * the extent cache, so it must happen before flushing it.
+		 */
+		err = apfs_zero_dstream_tail(dstream);
+		if (err) {
+			apfs_err(sb, "failed to zero tail for dstream 0x%llx", dstream->ds_id);
+			return err;
+		}
+	}
+
+	err = apfs_dstream_cache_in_snap(dstream, &in_snap);
+	if (err)
+		return err;
+
+	cache = &dstream->ds_cached_ext;
+	cache_blks = apfs_size_to_blocks(sb, cache->len);
+
+	/* TODO: allow dirty caches of several blocks in the middle of a file */
+	if (!in_snap && apfs_dstream_cache_is_tail(dstream) &&
+	    logical_addr == cache->logical_addr + cache->len &&
+	    phys_bno == cache->phys_block_num + cache_blks) {
+		cache->len += sb->s_blocksize;
+		dstream->ds_ext_dirty = true;
+		return 0;
+	}
+
+	err = apfs_flush_extent_cache(dstream);
+	if (err) {
+		apfs_err(sb, "extent cache flush failed for dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+
+	if (dstream_blks < dsblock) {
+		/*
+		 * This puts new extents after the reported end of the file, so
+		 * it must happen after the flush to avoid conflict with those
+		 * extent operations.
+		 */
+		err = apfs_create_hole(dstream, dstream_blks, dsblock);
+		if (err) {
+			apfs_err(sb, "hole creation failed for dstream 0x%llx", dstream->ds_id);
+			return err;
+		}
+	}
+
+	cache->logical_addr = logical_addr;
+	cache->phys_block_num = phys_bno;
+	cache->len = sb->s_blocksize;
+	dstream->ds_ext_dirty = true;
+	return 0;
+}
+
+/**
+ * apfs_dstream_get_new_bno - Allocate a new block inside a dstream
+ * @dstream:	data stream info
+ * @dsblock:	logical dstream block to allocate
+ * @bno:	on return, the new block number
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_dstream_get_new_bno(struct apfs_dstream_info *dstream, u64 dsblock, u64 *bno)
+{
+	return apfs_dstream_get_new_block(dstream, dsblock, NULL /* bh_result */, bno);
+}
+
+int apfs_get_new_block(struct inode *inode, sector_t iblock,
+		       struct buffer_head *bh_result, int create)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+
+	ASSERT(create);
+	return apfs_dstream_get_new_block(&ai->i_dstream, iblock, bh_result, NULL /* bno */);
+}
+
+/**
+ * apfs_shrink_dstream_last_extent - Shrink last extent of dstream being resized
+ * @dstream:	data stream info
+ * @new_size:	new size for the whole data stream
+ *
+ * Deletes, shrinks or zeroes the last extent, as needed for the truncation of
+ * the data stream.
+ *
+ * Only works with the last extent, so it needs to be called repeatedly to
+ * complete the truncation. Returns -EAGAIN in that case, or 0 when the process
+ * is complete. Returns other negative error codes in case of failure.
+ */
+static int apfs_shrink_dstream_last_extent(struct apfs_dstream_info *dstream, loff_t new_size)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_file_extent tail;
+	u64 extent_id = dstream->ds_id;
+	int ret = 0;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_file_extent_key(extent_id, -1, &query->key);
+	query->flags = APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for last extent of id 0x%llx", extent_id);
+		goto out;
+	}
+
+	if (!apfs_query_found_extent(query)) {
+		/* No more extents, we deleted the whole file already? */
+		if (new_size) {
+			apfs_err(sb, "missing extent for dstream 0x%llx", extent_id);
+			ret = -EFSCORRUPTED;
+		} else {
+			ret = 0;
+		}
+		goto out;
+	}
+
+	ret = apfs_extent_from_query(query, &tail);
+	if (ret) {
+		apfs_err(sb, "bad tail extent record on dstream 0x%llx", extent_id);
+		goto out;
+	}
+
+	if (tail.logical_addr + tail.len < new_size) {
+		apfs_err(sb, "missing extent for dstream 0x%llx", extent_id);
+		ret = -EFSCORRUPTED; /* Tail extent missing */
+	} else if (tail.logical_addr + tail.len == new_size) {
+		ret = 0; /* Nothing more to be done */
+	} else if (tail.logical_addr >= new_size) {
+		/* This whole extent needs to go */
+		ret = apfs_btree_remove(query);
+		if (ret) {
+			apfs_err(sb, "removal failed for id 0x%llx, addr 0x%llx", dstream->ds_id, tail.logical_addr);
+			goto out;
+		}
+		if (apfs_ext_is_hole(&tail)) {
+			dstream->ds_sparse_bytes -= tail.len;
+		} else {
+			ret = apfs_range_put_reference(sb, tail.phys_block_num, tail.len);
+			if (ret) {
+				apfs_err(sb, "failed to put range 0x%llx-0x%llx", tail.phys_block_num, tail.len);
+				goto out;
+			}
+		}
+		ret = apfs_crypto_adj_refcnt(sb, tail.crypto_id, -1);
+		if (ret) {
+			apfs_err(sb, "failed to take crypto id 0x%llx", tail.crypto_id);
+			goto out;
+		}
+		ret = tail.logical_addr == new_size ? 0 : -EAGAIN;
+	} else {
+		/*
+		 * The file is being truncated in the middle of this extent.
+		 * TODO: preserve the physical tail to be overwritten later.
+		 */
+		new_size = apfs_size_to_blocks(sb, new_size) << sb->s_blocksize_bits;
+		ret = apfs_shrink_extent_tail(query, dstream, new_size);
+		if (ret)
+			apfs_err(sb, "failed to shrink tail of dstream 0x%llx", extent_id);
+	}
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_shrink_dstream - Shrink a data stream's extents to a new length
+ * @dstream:	data stream info
+ * @new_size:	the new size
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_shrink_dstream(struct apfs_dstream_info *dstream, loff_t new_size)
+{
+	int ret;
+
+	do {
+		ret = apfs_shrink_dstream_last_extent(dstream, new_size);
+	} while (ret == -EAGAIN);
+
+	return ret;
+}
+
+/**
+ * apfs_truncate - Truncate a data stream's content
+ * @dstream:	data stream info
+ * @new_size:	the new size
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_truncate(struct apfs_dstream_info *dstream, loff_t new_size)
+{
+	struct super_block *sb = dstream->ds_sb;
+	u64 old_blks, new_blks;
+	struct apfs_file_extent *cache = &dstream->ds_cached_ext;
+	int err;
+
+	/* TODO: don't write the cached extent if it will be deleted */
+	err = apfs_flush_extent_cache(dstream);
+	if (err) {
+		apfs_err(sb, "extent cache flush failed for dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+	dstream->ds_ext_dirty = false;
+
+	/* TODO: keep the cache valid on truncation */
+	cache->len = 0;
+
+	/* "<=", because a partial write may have left extents beyond the end */
+	if (new_size <= dstream->ds_size)
+		return apfs_shrink_dstream(dstream, new_size);
+
+	err = apfs_zero_dstream_tail(dstream);
+	if (err) {
+		apfs_err(sb, "failed to zero tail for dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+	new_blks = apfs_size_to_blocks(sb, new_size);
+	old_blks = apfs_size_to_blocks(sb, dstream->ds_size);
+	return apfs_create_hole(dstream, old_blks, new_blks);
+}
+
+/**
+ * apfs_dstream_delete_front - Deletes as many leading extents as possible
+ * @sb:		filesystem superblock
+ * @ds_id:	id for the dstream to delete
+ *
+ * Returns 0 on success, or a negative error code in case of failure, which may
+ * be -ENODATA if there are no more extents, or -EAGAIN if the free queue is
+ * getting too full.
+ */
+static int apfs_dstream_delete_front(struct super_block *sb, u64 ds_id)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_spaceman_free_queue *fq = NULL;
+	struct apfs_query *query = NULL;
+	struct apfs_file_extent head;
+	bool first_match = true;
+	int ret;
+
+	fq = &sm_raw->sm_fq[APFS_SFQ_MAIN];
+	if (le64_to_cpu(fq->sfq_count) > APFS_TRANS_MAIN_QUEUE_MAX)
+		return -EAGAIN;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_file_extent_key(ds_id, 0, &query->key);
+	query->flags = APFS_QUERY_CAT;
+
+next_extent:
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for first extent of id 0x%llx", ds_id);
+		goto out;
+	}
+	apfs_query_direct_forward(query);
+	if (!apfs_query_found_extent(query)) {
+		/*
+		 * After the original lookup, the query may not be set to the
+		 * first extent, but instead to the record that comes right
+		 * before.
+		 */
+		if (first_match) {
+			first_match = false;
+			goto next_extent;
+		}
+		ret = -ENODATA;
+		goto out;
+	}
+	first_match = false;
+
+	ret = apfs_extent_from_query(query, &head);
+	if (ret) {
+		apfs_err(sb, "bad head extent record on dstream 0x%llx", ds_id);
+		goto out;
+	}
+	ret = apfs_btree_remove(query);
+	if (ret) {
+		apfs_err(sb, "removal failed for id 0x%llx, addr 0x%llx", ds_id, head.logical_addr);
+		goto out;
+	}
+
+	/*
+	 * The official fsck doesn't complain about wrong sparse byte counts
+	 * for orphans, so I guess we don't need to update them here
+	 */
+	if (!apfs_ext_is_hole(&head)) {
+		ret = apfs_range_put_reference(sb, head.phys_block_num, head.len);
+		if (ret) {
+			apfs_err(sb, "failed to put range 0x%llx-0x%llx", head.phys_block_num, head.len);
+			goto out;
+		}
+		ret = apfs_crypto_adj_refcnt(sb, head.crypto_id, -1);
+		if (ret) {
+			apfs_err(sb, "failed to take crypto id 0x%llx", head.crypto_id);
+			goto out;
+		}
+	}
+
+	if (le64_to_cpu(fq->sfq_count) <= APFS_TRANS_MAIN_QUEUE_MAX)
+		goto next_extent;
+	ret = -EAGAIN;
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_inode_delete_front - Deletes as many leading extents as possible
+ * @inode:	inode to delete
+ *
+ * Tries to delete all extents for @inode, in which case it returns 0. If the
+ * free queue is getting too full, deletes as much as is reasonable and returns
+ * -EAGAIN. May return other negative error codes as well.
+ */
+int apfs_inode_delete_front(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_dstream_info *dstream = NULL;
+	struct apfs_inode_info *ai = APFS_I(inode);
+	int ret;
+
+	if (!ai->i_has_dstream)
+		return 0;
+
+	dstream = &ai->i_dstream;
+	ret = apfs_flush_extent_cache(dstream);
+	if (ret) {
+		apfs_err(sb, "extent cache flush failed for dstream 0x%llx", dstream->ds_id);
+		return ret;
+	}
+
+	ret = apfs_dstream_delete_front(sb, dstream->ds_id);
+	if (ret == -ENODATA)
+		return 0;
+	return ret;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 20, 0)
+loff_t apfs_remap_file_range(struct file *src_file, loff_t off, struct file *dst_file, loff_t destoff, loff_t len, unsigned int remap_flags)
+#else
+int apfs_clone_file_range(struct file *src_file, loff_t off, struct file *dst_file, loff_t destoff, u64 len)
+#endif
+{
+	struct inode *src_inode = file_inode(src_file);
+	struct inode *dst_inode = file_inode(dst_file);
+	struct apfs_inode_info *src_ai = APFS_I(src_inode);
+	struct apfs_inode_info *dst_ai = APFS_I(dst_inode);
+	struct apfs_dstream_info *src_ds = &src_ai->i_dstream;
+	struct apfs_dstream_info *dst_ds = &dst_ai->i_dstream;
+	struct super_block *sb = src_inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	const u64 xfield_flags = APFS_INODE_MAINTAIN_DIR_STATS | APFS_INODE_IS_SPARSE | APFS_INODE_HAS_PURGEABLE_FLAGS;
+	int err;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 20, 0)
+	if (remap_flags & ~(REMAP_FILE_ADVISORY))
+		return -EINVAL;
+#endif
+	if (src_inode == dst_inode)
+		return -EINVAL;
+
+	/* We only want to clone whole files, like in the official driver */
+	if (off != 0 || destoff != 0 || len != 0)
+		return -EINVAL;
+
+	/*
+	 * Clones here work in two steps: first the user creates an empty target
+	 * file, and then the user calls the ioctl, which replaces the file with
+	 * a clone. This is not atomic, of course.
+	 */
+	if (dst_ai->i_has_dstream || dst_ai->i_bsd_flags & APFS_INOBSD_COMPRESSED) {
+		apfs_warn(sb, "clones can only replace freshly created files");
+		return -EOPNOTSUPP;
+	}
+	if (dst_ai->i_int_flags & xfield_flags) {
+		apfs_warn(sb, "clone can't replace a file that has xfields");
+		return -EOPNOTSUPP;
+	}
+
+	if (!src_ai->i_has_dstream) {
+		apfs_warn(sb, "can't clone a file with no dstream");
+		return -EOPNOTSUPP;
+	}
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+	apfs_inode_join_transaction(sb, src_inode);
+	apfs_inode_join_transaction(sb, dst_inode);
+
+	err = apfs_flush_extent_cache(src_ds);
+	if (err) {
+		apfs_err(sb, "extent cache flush failed for dstream 0x%llx", src_ds->ds_id);
+		goto fail;
+	}
+	err = apfs_dstream_adj_refcnt(src_ds, +1);
+	if (err) {
+		apfs_err(sb, "failed to take dstream id 0x%llx", src_ds->ds_id);
+		goto fail;
+	}
+	src_ds->ds_shared = true;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	dst_inode->i_mtime = dst_inode->i_ctime = current_time(dst_inode);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	dst_inode->i_mtime = inode_set_ctime_current(dst_inode);
+#else
+	inode_set_mtime_to_ts(dst_inode, inode_set_ctime_current(dst_inode));
+#endif
+	dst_inode->i_size = src_inode->i_size;
+	dst_ai->i_key_class = src_ai->i_key_class;
+	dst_ai->i_int_flags = src_ai->i_int_flags;
+	dst_ai->i_bsd_flags = src_ai->i_bsd_flags;
+	dst_ai->i_has_dstream = true;
+
+	dst_ds->ds_sb = src_ds->ds_sb;
+	dst_ds->ds_inode = dst_inode;
+	dst_ds->ds_id = src_ds->ds_id;
+	dst_ds->ds_size = src_ds->ds_size;
+	dst_ds->ds_sparse_bytes = src_ds->ds_sparse_bytes;
+	dst_ds->ds_cached_ext = src_ds->ds_cached_ext;
+	dst_ds->ds_ext_dirty = false;
+	dst_ds->ds_shared = true;
+
+	dst_ai->i_int_flags |= APFS_INODE_WAS_EVER_CLONED | APFS_INODE_WAS_CLONED;
+	src_ai->i_int_flags |= APFS_INODE_WAS_EVER_CLONED;
+
+	/*
+	 * The sparse flag is the important one here: if we need it, it will get
+	 * set later by apfs_update_inode(), after the xfield gets created.
+	 */
+	dst_ai->i_int_flags &= ~xfield_flags;
+
+	/*
+	 * Commit the transaction to make sure all buffers in the source inode
+	 * go through copy-on-write. This is a bit excessive, but I don't expect
+	 * clones to be created often enough for it to matter.
+	 */
+	sbi->s_nxi->nx_transaction.t_state |= APFS_NX_TRANS_FORCE_COMMIT;
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return dst_ds->ds_size;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_extent_create_record - Create a logical extent record for a dstream id
+ * @sb:		filesystem superblock
+ * @dstream_id:	the dstream id
+ * @extent:	extent info for the record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_extent_create_record(struct super_block *sb, u64 dstream_id, struct apfs_file_extent *extent)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	struct apfs_file_extent_val raw_val;
+	struct apfs_file_extent_key raw_key;
+	int ret = 0;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_file_extent_key(dstream_id, extent->logical_addr, &query->key);
+	query->flags = APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for id 0x%llx, addr 0x%llx", dstream_id, extent->logical_addr);
+		goto out;
+	}
+
+	apfs_key_set_hdr(APFS_TYPE_FILE_EXTENT, dstream_id, &raw_key);
+	raw_key.logical_addr = cpu_to_le64(extent->logical_addr);
+	raw_val.len_and_flags = cpu_to_le64(extent->len);
+	raw_val.phys_block_num = cpu_to_le64(extent->phys_block_num);
+	raw_val.crypto_id = cpu_to_le64(apfs_vol_is_encrypted(sb) ? dstream_id : 0); /* TODO */
+
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+	if (ret)
+		apfs_err(sb, "insertion failed for id 0x%llx, addr 0x%llx", dstream_id, extent->logical_addr);
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_put_single_extent - Put a reference to a single extent
+ * @sb:		filesystem superblock
+ * @paddr_end:	first block after the extent to put
+ * @paddr_min:	don't put references before this block
+ *
+ * Puts a reference to the physical extent range that ends in @paddr_end. Sets
+ * @paddr_end to the beginning of the extent, so that the caller can continue
+ * with the previous one. Returns 0 on success, or a negative error code in
+ * case of failure.
+ *
+ * TODO: unify this with apfs_take_single_extent(), they are almost the same.
+ */
+static int apfs_put_single_extent(struct super_block *sb, u64 *paddr_end, u64 paddr_min)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_node *extref_root = NULL;
+	struct apfs_key key;
+	struct apfs_query *query = NULL;
+	struct apfs_phys_extent prev_ext;
+	u64 prev_start, prev_end;
+	bool cropped_head = false, cropped_tail = false;
+	struct apfs_file_extent tmp = {0}; /* TODO: clean up all the fake extent interfaces? */
+	int ret;
+
+	extref_root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_extentref_tree_oid), APFS_OBJ_PHYSICAL, true /* write */);
+	if (IS_ERR(extref_root)) {
+		apfs_err(sb, "failed to read extref root 0x%llx", le64_to_cpu(vsb_raw->apfs_extentref_tree_oid));
+		return PTR_ERR(extref_root);
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	vsb_raw->apfs_extentref_tree_oid = cpu_to_le64(extref_root->object.oid);
+
+	apfs_init_extent_key(*paddr_end - 1, &key);
+
+restart:
+	query = apfs_alloc_query(extref_root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	query->key = key;
+	query->flags = APFS_QUERY_EXTENTREF;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for paddr 0x%llx", *paddr_end - 1);
+		goto out;
+	}
+
+	if (ret == -ENODATA) {
+		/* The whole range to put is part of a snapshot */
+		tmp.phys_block_num = paddr_min;
+		tmp.len = (*paddr_end - paddr_min) << sb->s_blocksize_bits;
+		ret = apfs_create_update_pext(query, &tmp, -1);
+		*paddr_end = paddr_min;
+		goto out;
+	}
+
+	ret = apfs_phys_ext_from_query(query, &prev_ext);
+	if (ret) {
+		apfs_err(sb, "bad pext record over paddr 0x%llx", *paddr_end - 1);
+		goto out;
+	}
+	prev_start = prev_ext.bno;
+	prev_end = prev_ext.bno + prev_ext.blkcount;
+	if (prev_end < *paddr_end) {
+		/* The extent to put is part of a snapshot */
+		tmp.phys_block_num = max(prev_end, paddr_min);
+		tmp.len = (*paddr_end - tmp.phys_block_num) << sb->s_blocksize_bits;
+		ret = apfs_create_update_pext(query, &tmp, -1);
+		*paddr_end = tmp.phys_block_num;
+		goto out;
+	}
+
+	if ((cropped_tail && prev_end > *paddr_end) || (cropped_head && prev_start < paddr_min)) {
+		/* This should never happen, but be safe */
+		apfs_alert(sb, "recursion cropping physical extent 0x%llx-0x%llx", prev_start, prev_end);
+		ret = -EFSCORRUPTED;
+		goto out;
+	}
+
+	if (prev_end > *paddr_end) {
+		ret = apfs_split_phys_ext(query, *paddr_end);
+		if (ret) {
+			apfs_err(sb, "failed to split pext at 0x%llx", *paddr_end);
+			goto out;
+		}
+		/* The split may make the query invalid */
+		apfs_free_query(query);
+		cropped_tail = true;
+		goto restart;
+	}
+
+	if (prev_start < paddr_min) {
+		ret = apfs_split_phys_ext(query, paddr_min);
+		if (ret) {
+			apfs_err(sb, "failed to split pext at 0x%llx", paddr_min);
+			goto out;
+		}
+		/* The split may make the query invalid */
+		apfs_free_query(query);
+		cropped_head = true;
+		goto restart;
+	}
+
+	/* The extent to put already exists */
+	ret = apfs_put_phys_extent(&prev_ext, query);
+	if (ret)
+		apfs_err(sb, "failed to put pext at 0x%llx", prev_ext.bno);
+	*paddr_end = prev_start;
+
+out:
+	apfs_free_query(query);
+	apfs_node_free(extref_root);
+	return ret;
+}
+
+/**
+ * apfs_range_put_reference - Put a reference to a physical range
+ * @sb:		filesystem superblock
+ * @paddr:	first block of the range
+ * @length:	length of the range (in bytes)
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_range_put_reference(struct super_block *sb, u64 paddr, u64 length)
+{
+	u64 extent_end;
+	int err;
+
+	ASSERT(paddr);
+
+	extent_end = paddr + (length >> sb->s_blocksize_bits);
+	while (extent_end > paddr) {
+		err = apfs_put_single_extent(sb, &extent_end, paddr);
+		if (err)
+			return err;
+	}
+	return 0;
+}
+
+/**
+ * apfs_take_single_extent - Take a reference to a single extent
+ * @sb:		filesystem superblock
+ * @paddr_end:	first block after the extent to take
+ * @paddr_min:	don't take references before this block
+ *
+ * Takes a reference to the physical extent range that ends in @paddr_end. Sets
+ * @paddr_end to the beginning of the extent, so that the caller can continue
+ * with the previous one. Returns 0 on success, or a negative error code in
+ * case of failure.
+ */
+static int apfs_take_single_extent(struct super_block *sb, u64 *paddr_end, u64 paddr_min)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_node *extref_root = NULL;
+	struct apfs_key key;
+	struct apfs_query *query = NULL;
+	struct apfs_phys_extent prev_ext;
+	u64 prev_start, prev_end;
+	bool cropped_head = false, cropped_tail = false;
+	struct apfs_file_extent tmp = {0}; /* TODO: clean up all the fake extent interfaces? */
+	int ret;
+
+	extref_root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_extentref_tree_oid), APFS_OBJ_PHYSICAL, true /* write */);
+	if (IS_ERR(extref_root)) {
+		apfs_err(sb, "failed to read extref root 0x%llx", le64_to_cpu(vsb_raw->apfs_extentref_tree_oid));
+		return PTR_ERR(extref_root);
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	vsb_raw->apfs_extentref_tree_oid = cpu_to_le64(extref_root->object.oid);
+
+	apfs_init_extent_key(*paddr_end - 1, &key);
+
+restart:
+	query = apfs_alloc_query(extref_root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	query->key = key;
+	query->flags = APFS_QUERY_EXTENTREF;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for paddr 0x%llx", *paddr_end - 1);
+		goto out;
+	}
+
+	if (ret == -ENODATA) {
+		/* The whole range to take is part of a snapshot */
+		tmp.phys_block_num = paddr_min;
+		tmp.len = (*paddr_end - paddr_min) << sb->s_blocksize_bits;
+		ret = apfs_create_update_pext(query, &tmp, +1);
+		*paddr_end = paddr_min;
+		goto out;
+	}
+
+	ret = apfs_phys_ext_from_query(query, &prev_ext);
+	if (ret) {
+		apfs_err(sb, "bad pext record over paddr 0x%llx", *paddr_end - 1);
+		goto out;
+	}
+	prev_start = prev_ext.bno;
+	prev_end = prev_ext.bno + prev_ext.blkcount;
+	if (prev_end < *paddr_end) {
+		/* The extent to take is part of a snapshot */
+		tmp.phys_block_num = max(prev_end, paddr_min);
+		tmp.len = (*paddr_end - tmp.phys_block_num) << sb->s_blocksize_bits;
+		ret = apfs_create_update_pext(query, &tmp, +1);
+		*paddr_end = tmp.phys_block_num;
+		goto out;
+	}
+
+	if ((cropped_tail && prev_end > *paddr_end) || (cropped_head && prev_start < paddr_min)) {
+		/* This should never happen, but be safe */
+		apfs_alert(sb, "recursion cropping physical extent 0x%llx-0x%llx", prev_start, prev_end);
+		ret = -EFSCORRUPTED;
+		goto out;
+	}
+
+	if (prev_end > *paddr_end) {
+		ret = apfs_split_phys_ext(query, *paddr_end);
+		if (ret) {
+			apfs_err(sb, "failed to split pext at 0x%llx", *paddr_end);
+			goto out;
+		}
+		/* The split may make the query invalid */
+		apfs_free_query(query);
+		cropped_tail = true;
+		goto restart;
+	}
+
+	if (prev_start < paddr_min) {
+		ret = apfs_split_phys_ext(query, paddr_min);
+		if (ret) {
+			apfs_err(sb, "failed to split pext at 0x%llx", paddr_min);
+			goto out;
+		}
+		/* The split may make the query invalid */
+		apfs_free_query(query);
+		cropped_head = true;
+		goto restart;
+	}
+
+	/* The extent to take already exists */
+	ret = apfs_take_phys_extent(&prev_ext, query);
+	if (ret)
+		apfs_err(sb, "failed to take pext at 0x%llx", prev_ext.bno);
+	*paddr_end = prev_start;
+
+out:
+	apfs_free_query(query);
+	apfs_node_free(extref_root);
+	return ret;
+}
+
+/**
+ * apfs_range_take_reference - Take a reference to a physical range
+ * @sb:		filesystem superblock
+ * @paddr:	first block of the range
+ * @length:	length of the range (in bytes)
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_range_take_reference(struct super_block *sb, u64 paddr, u64 length)
+{
+	u64 extent_end;
+	int err;
+
+	ASSERT(paddr);
+
+	extent_end = paddr + (length >> sb->s_blocksize_bits);
+	while (extent_end > paddr) {
+		err = apfs_take_single_extent(sb, &extent_end, paddr);
+		if (err)
+			return err;
+	}
+	return 0;
+}
+
+/**
+ * apfs_clone_single_extent - Make a copy of an extent in a dstream to a new one
+ * @dstream:	old dstream
+ * @new_id:	id of the new dstream
+ * @log_addr:	logical address for the extent
+ *
+ * Duplicates the logical extent, and updates the references to the physical
+ * extents as required. Sets @addr to the end of the extent, so that the caller
+ * can continue in the same place. Returns 0 on success, or a negative error
+ * code in case of failure.
+ */
+static int apfs_clone_single_extent(struct apfs_dstream_info *dstream, u64 new_id, u64 *log_addr)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_file_extent extent;
+	int err;
+
+	err = apfs_extent_read(dstream, *log_addr >> sb->s_blocksize_bits, &extent);
+	if (err) {
+		apfs_err(sb, "failed to read an extent to clone for dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+	err = apfs_extent_create_record(sb, new_id, &extent);
+	if (err) {
+		apfs_err(sb, "failed to create extent record for clone of dstream 0x%llx", dstream->ds_id);
+		return err;
+	}
+
+	if (!apfs_ext_is_hole(&extent)) {
+		err = apfs_range_take_reference(sb, extent.phys_block_num, extent.len);
+		if (err) {
+			apfs_err(sb, "failed to take a reference to physical range 0x%llx-0x%llx", extent.phys_block_num, extent.len);
+			return err;
+		}
+	}
+
+	*log_addr += extent.len;
+	return 0;
+}
+
+/**
+ * apfs_clone_extents - Make a copy of all extents in a dstream to a new one
+ * @dstream:	old dstream
+ * @new_id:	id for the new dstream
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_clone_extents(struct apfs_dstream_info *dstream, u64 new_id)
+{
+	u64 next = 0;
+	int err;
+
+	while (next < dstream->ds_size) {
+		err = apfs_clone_single_extent(dstream, new_id, &next);
+		if (err)
+			return err;
+	}
+	return 0;
+}
+
+/**
+ * apfs_nonsparse_dstream_read - Read from a dstream without holes
+ * @dstream:	dstream to read
+ * @buf:	destination buffer
+ * @count:	exact number of bytes to read
+ * @offset:	dstream offset to read from
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_nonsparse_dstream_read(struct apfs_dstream_info *dstream, void *buf, size_t count, u64 offset)
+{
+	struct super_block *sb = dstream->ds_sb;
+	u64 logical_start_block, logical_end_block, log_bno, blkcnt, idx;
+	struct buffer_head **bhs = NULL;
+	int ret = 0;
+
+	/* Save myself from thinking about overflow here */
+	if (count >= APFS_MAX_FILE_SIZE || offset >= APFS_MAX_FILE_SIZE) {
+		apfs_err(sb, "dstream read overflow (0x%llx-0x%llx)", offset, (unsigned long long)count);
+		return -EFBIG;
+	}
+
+	if (offset + count > dstream->ds_size) {
+		apfs_err(sb, "reading past the end (0x%llx-0x%llx)", offset, (unsigned long long)count);
+		/* No caller is expected to legitimately read out-of-bounds */
+		return -EFSCORRUPTED;
+	}
+
+	logical_start_block = offset >> sb->s_blocksize_bits;
+	logical_end_block = (offset + count + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+	blkcnt = logical_end_block - logical_start_block;
+	bhs = kcalloc(blkcnt, sizeof(*bhs), GFP_KERNEL);
+	if (!bhs)
+		return -ENOMEM;
+
+	for (log_bno = logical_start_block; log_bno < logical_end_block; log_bno++) {
+		struct buffer_head *bh = NULL;
+		u64 bno = 0;
+
+		idx = log_bno - logical_start_block;
+
+		ret = apfs_logic_to_phys_bno(dstream, log_bno, &bno);
+		if (ret)
+			goto out;
+		if (bno == 0) {
+			apfs_err(sb, "nonsparse dstream has a hole");
+			ret = -EFSCORRUPTED;
+			goto out;
+		}
+
+		bhs[idx] = __apfs_getblk(sb, bno);
+		if (!bhs[idx]) {
+			apfs_err(sb, "failed to map block 0x%llx", bno);
+			ret = -EIO;
+			goto out;
+		}
+
+		bh = bhs[idx];
+		if (!buffer_uptodate(bh)) {
+			get_bh(bh);
+			lock_buffer(bh);
+			bh->b_end_io = end_buffer_read_sync;
+			apfs_submit_bh(REQ_OP_READ, 0, bh);
+		}
+	}
+
+	for (log_bno = logical_start_block; log_bno < logical_end_block; log_bno++) {
+		int off_in_block, left_in_block;
+
+		idx = log_bno - logical_start_block;
+		wait_on_buffer(bhs[idx]);
+		if (!buffer_uptodate(bhs[idx])) {
+			apfs_err(sb, "failed to read a block");
+			ret = -EIO;
+			goto out;
+		}
+
+		if (log_bno == logical_start_block)
+			off_in_block = offset & (sb->s_blocksize - 1);
+		else
+			off_in_block = 0;
+
+		if (log_bno == logical_end_block - 1)
+			left_in_block = count + offset - (log_bno << sb->s_blocksize_bits) - off_in_block;
+		else
+			left_in_block = sb->s_blocksize - off_in_block;
+
+		memcpy(buf, bhs[idx]->b_data + off_in_block, left_in_block);
+		buf += left_in_block;
+	}
+
+out:
+	if (bhs) {
+		for (idx = 0; idx < blkcnt; idx++)
+			brelse(bhs[idx]);
+		kfree(bhs);
+	}
+	return ret;
+}
+
+/**
+ * apfs_nonsparse_dstream_preread - Attempt to preread a dstream without holes
+ * @dstream:	dstream to preread
+ *
+ * Requests reads for all blocks of a dstream, but doesn't wait for the result.
+ */
+void apfs_nonsparse_dstream_preread(struct apfs_dstream_info *dstream)
+{
+	struct super_block *sb = dstream->ds_sb;
+	u64 logical_end_block, log_bno;
+
+	logical_end_block = (dstream->ds_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+
+	for (log_bno = 0; log_bno < logical_end_block; log_bno++) {
+		struct buffer_head *bh = NULL;
+		u64 bno = 0;
+		int ret;
+
+		ret = apfs_logic_to_phys_bno(dstream, log_bno, &bno);
+		if (ret || bno == 0)
+			return;
+
+		bh = __apfs_getblk(sb, bno);
+		if (!bh)
+			return;
+		if (!buffer_uptodate(bh)) {
+			get_bh(bh);
+			lock_buffer(bh);
+			bh->b_end_io = end_buffer_read_sync;
+			apfs_submit_bh(REQ_OP_READ, 0, bh);
+		}
+		brelse(bh);
+		bh = NULL;
+	}
+}
diff --git a/fs/apfs/file.c b/fs/apfs/file.c
new file mode 100644
index 000000000..3e8a55897
--- /dev/null
+++ b/fs/apfs/file.c
@@ -0,0 +1,224 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include "apfs.h"
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+#include <linux/splice.h>
+#endif
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 17, 0)
+typedef int vm_fault_t;
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 15, 0)
+static void wait_for_stable_page(struct page *page)
+{
+	return folio_wait_stable(page_folio(page));
+}
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0)
+static vm_fault_t apfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+#else
+static vm_fault_t apfs_page_mkwrite(struct vm_fault *vmf)
+{
+	struct vm_area_struct *vma = vmf->vma;
+#endif
+	struct page *page = vmf->page;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 7, 0)
+	struct folio *folio;
+#endif
+	struct inode *inode = file_inode(vma->vm_file);
+	struct super_block *sb = inode->i_sb;
+	struct buffer_head *bh, *head;
+	vm_fault_t ret = VM_FAULT_LOCKED;
+	unsigned int blocksize, block_start, len;
+	u64 size;
+	int err = 0;
+
+	sb_start_pagefault(inode->i_sb);
+	file_update_time(vma->vm_file);
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		goto out;
+	apfs_inode_join_transaction(sb, inode);
+
+	err = apfs_inode_create_exclusive_dstream(inode);
+	if (err) {
+		apfs_err(sb, "dstream creation failed for ino 0x%llx", apfs_ino(inode));
+		goto out_abort;
+	}
+
+	lock_page(page);
+	wait_for_stable_page(page);
+	if (page->mapping != inode->i_mapping) {
+		ret = VM_FAULT_NOPAGE;
+		goto out_unlock;
+	}
+
+	if (!page_has_buffers(page)) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+		create_empty_buffers(page, sb->s_blocksize, 0);
+#else
+		folio = page_folio(page);
+		bh = folio_buffers(folio);
+		if (!bh)
+			bh = create_empty_buffers(folio, sb->s_blocksize, 0);
+#endif
+	}
+
+	size = i_size_read(inode);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 16, 0)
+	if (page_folio(page)->index == size >> PAGE_SHIFT)
+#else
+	if (page->index == size >> PAGE_SHIFT)
+#endif
+		len = size & ~PAGE_MASK;
+	else
+		len = PAGE_SIZE;
+
+	/* The blocks were read on the fault, mark them as unmapped for CoW */
+	head = page_buffers(page);
+	blocksize = head->b_size;
+	for (bh = head, block_start = 0; bh != head || !block_start;
+	     block_start += blocksize, bh = bh->b_this_page) {
+		if (len > block_start) {
+			/* If it's not a hole, the fault read it already */
+			ASSERT(!buffer_mapped(bh) || buffer_uptodate(bh));
+			if (buffer_trans(bh))
+				continue;
+			clear_buffer_mapped(bh);
+		}
+	}
+	unlock_page(page); /* XXX: race? */
+
+	err = block_page_mkwrite(vma, vmf, apfs_get_new_block);
+	if (err) {
+		apfs_err(sb, "mkwrite failed for ino 0x%llx", apfs_ino(inode));
+		goto out_abort;
+	}
+	set_page_dirty(page);
+
+	/* An immediate commit would leave the page unlocked */
+	APFS_SB(sb)->s_nxi->nx_transaction.t_state |= APFS_NX_TRANS_DEFER_COMMIT;
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto out_unlock;
+	goto out;
+
+out_unlock:
+	unlock_page(page);
+out_abort:
+	apfs_transaction_abort(sb);
+out:
+	if (err)
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0) && !RHEL_VERSION_GE(9, 5)
+		ret = block_page_mkwrite_return(err);
+#else
+		ret = vmf_fs_error(err);
+#endif
+	sb_end_pagefault(inode->i_sb);
+	return ret;
+}
+
+static const struct vm_operations_struct apfs_file_vm_ops = {
+	.fault		= filemap_fault,
+	.map_pages	= filemap_map_pages,
+	.page_mkwrite	= apfs_page_mkwrite,
+};
+
+int apfs_file_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	struct address_space *mapping = file->f_mapping;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+	if (!mapping->a_ops->read_folio)
+#else
+	if (!mapping->a_ops->readpage)
+#endif
+		return -ENOEXEC;
+	file_accessed(file);
+	vma->vm_ops = &apfs_file_vm_ops;
+	return 0;
+}
+
+/*
+ * Just flush the whole transaction for now (TODO), since that's technically
+ * correct and easy to implement.
+ */
+int apfs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
+{
+	struct inode *inode = file->f_mapping->host;
+	struct super_block *sb = inode->i_sb;
+
+	return apfs_sync_fs(sb, true /* wait */);
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+static ssize_t apfs_copy_file_range(struct file *src_file, loff_t src_off,
+				    struct file *dst_file, loff_t dst_off,
+				    size_t len, unsigned int flags)
+{
+	return (splice_copy_file_range(src_file, src_off,
+		dst_file, dst_off, len));
+}
+#endif
+
+const struct file_operations apfs_file_operations = {
+	.llseek			= generic_file_llseek,
+	.read_iter		= generic_file_read_iter,
+	.write_iter		= generic_file_write_iter,
+	.mmap			= apfs_file_mmap,
+	.open			= generic_file_open,
+	.fsync			= apfs_fsync,
+	.unlocked_ioctl		= apfs_file_ioctl,
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+	.copy_file_range	= apfs_copy_file_range,
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(5, 3, 0)
+	.copy_file_range	= generic_copy_file_range,
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 20, 0)
+	.remap_file_range	= apfs_remap_file_range,
+#else
+	.clone_file_range	= apfs_clone_file_range,
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 5, 0)
+	.splice_read		= generic_file_splice_read,
+#else
+	.splice_read		= filemap_splice_read,
+#endif
+	.splice_write		= iter_file_splice_write,
+};
+
+#if LINUX_VERSION_CODE == KERNEL_VERSION(5, 3, 0)
+/*
+ * This is needed mainly to test clones with xfstests, so we only support the
+ * kernel version I use during testing. TODO: support all kernel versions.
+ */
+int apfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, u64 start, u64 len)
+{
+	return generic_block_fiemap(inode, fieinfo, start, len, apfs_get_block);
+}
+#endif
+
+const struct inode_operations apfs_file_inode_operations = {
+	.getattr	= apfs_getattr,
+	.listxattr	= apfs_listxattr,
+	.setattr	= apfs_setattr,
+	.update_time	= apfs_update_time,
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 13, 0)
+	.fileattr_get	= apfs_fileattr_get,
+	.fileattr_set	= apfs_fileattr_set,
+#endif
+#if LINUX_VERSION_CODE == KERNEL_VERSION(5, 3, 0)
+	.fiemap		= apfs_fiemap,
+#endif
+};
diff --git a/fs/apfs/inode.c b/fs/apfs/inode.c
new file mode 100644
index 000000000..7eb88bdb8
--- /dev/null
+++ b/fs/apfs/inode.c
@@ -0,0 +1,2598 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/slab.h>
+#include <linux/buffer_head.h>
+#include <linux/mount.h>
+#include <linux/mpage.h>
+#include <linux/blk_types.h>
+#include "apfs.h"
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+#include <linux/sched/mm.h>
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 13, 0)
+#include <linux/fileattr.h>
+#endif
+
+#define MAX_PFK_LEN	512
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 15, 0)
+static struct page *grab_cache_page_write_begin(struct address_space *mapping, pgoff_t index)
+{
+	return pagecache_get_page(mapping, index, FGP_WRITEBEGIN, mapping_gfp_mask(mapping));
+}
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+
+static int apfs_read_folio(struct file *file, struct folio *folio)
+{
+	return mpage_read_folio(folio, apfs_get_block);
+}
+
+#else
+
+static int apfs_readpage(struct file *file, struct page *page)
+{
+	return mpage_readpage(page, apfs_get_block);
+}
+
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0) /* Misses mpage_readpages() */
+
+static void apfs_readahead(struct readahead_control *rac)
+{
+	mpage_readahead(rac, apfs_get_block);
+}
+
+#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0) */
+
+static int apfs_readpages(struct file *file, struct address_space *mapping,
+			  struct list_head *pages, unsigned int nr_pages)
+{
+	return mpage_readpages(mapping, pages, nr_pages, apfs_get_block);
+}
+
+#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0) */
+
+/**
+ * apfs_create_dstream_rec - Create a data stream record
+ * @dstream: data stream info
+ *
+ * Does nothing if the record already exists.  TODO: support cloned files.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_dstream_rec(struct apfs_dstream_info *dstream)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_dstream_id_key raw_key;
+	struct apfs_dstream_id_val raw_val;
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_dstream_id_key(dstream->ds_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret != -ENODATA) /* Either an error, or the record already exists */
+		goto out;
+
+	apfs_key_set_hdr(APFS_TYPE_DSTREAM_ID, dstream->ds_id, &raw_key);
+	raw_val.refcnt = cpu_to_le32(1);
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+	if (ret) {
+		apfs_err(sb, "insertion failed for id 0x%llx", dstream->ds_id);
+		goto out;
+	}
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+static int apfs_check_dstream_refcnt(struct inode *inode);
+static int apfs_put_dstream_rec(struct apfs_dstream_info *dstream);
+
+/**
+ * apfs_inode_create_exclusive_dstream - Make an inode's dstream not shared
+ * @inode: the vfs inode
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_inode_create_exclusive_dstream(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct apfs_dstream_info *dstream = &ai->i_dstream;
+	u64 new_id;
+	int err;
+
+	if (!ai->i_has_dstream || !dstream->ds_shared)
+		return 0;
+
+	/*
+	 * The ds_shared field is not updated when the other user of the
+	 * dstream puts it, so it could be a false positive. Check it again
+	 * before actually putting the dstream. The double query is wasteful,
+	 * but I don't know if it makes sense to optimize this (TODO).
+	 */
+	err = apfs_check_dstream_refcnt(inode);
+	if (err) {
+		apfs_err(sb, "failed to check refcnt for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+	if (!dstream->ds_shared)
+		return 0;
+	err = apfs_put_dstream_rec(dstream);
+	if (err) {
+		apfs_err(sb, "failed to put dstream for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	new_id = le64_to_cpu(vsb_raw->apfs_next_obj_id);
+	le64_add_cpu(&vsb_raw->apfs_next_obj_id, 1);
+
+	err = apfs_clone_extents(dstream, new_id);
+	if (err) {
+		apfs_err(sb, "failed clone extents for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	dstream->ds_id = new_id;
+	err = apfs_create_dstream_rec(dstream);
+	if (err) {
+		apfs_err(sb, "failed to create dstream for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	dstream->ds_shared = false;
+	return 0;
+}
+
+/**
+ * apfs_inode_create_dstream_rec - Create the data stream record for an inode
+ * @inode: the vfs inode
+ *
+ * Does nothing if the record already exists.  TODO: support cloned files.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_inode_create_dstream_rec(struct inode *inode)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+	int err;
+
+	if (ai->i_has_dstream)
+		return apfs_inode_create_exclusive_dstream(inode);
+
+	err = apfs_create_dstream_rec(&ai->i_dstream);
+	if (err)
+		return err;
+
+	ai->i_has_dstream = true;
+	return 0;
+}
+
+/**
+ * apfs_dstream_adj_refcnt - Adjust dstream record refcount
+ * @dstream:	data stream info
+ * @delta:	desired change in reference count
+ *
+ * Deletes the record if the reference count goes to zero. Returns 0 on success
+ * or a negative error code in case of failure.
+ */
+int apfs_dstream_adj_refcnt(struct apfs_dstream_info *dstream, u32 delta)
+{
+	struct super_block *sb = dstream->ds_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_dstream_id_val raw_val;
+	void *raw = NULL;
+	u32 refcnt;
+	int ret;
+
+	ASSERT(APFS_I(dstream->ds_inode)->i_has_dstream);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_dstream_id_key(dstream->ds_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		apfs_err(sb, "query failed for id 0x%llx", dstream->ds_id);
+		if (ret == -ENODATA)
+			ret = -EFSCORRUPTED;
+		goto out;
+	}
+
+	if (query->len != sizeof(raw_val)) {
+		apfs_err(sb, "bad value length (%d)", query->len);
+		ret = -EFSCORRUPTED;
+		goto out;
+	}
+	raw = query->node->object.data;
+	raw_val = *(struct apfs_dstream_id_val *)(raw + query->off);
+	refcnt = le32_to_cpu(raw_val.refcnt);
+
+	refcnt += delta;
+	if (refcnt == 0) {
+		ret = apfs_btree_remove(query);
+		if (ret)
+			apfs_err(sb, "removal failed for id 0x%llx", dstream->ds_id);
+		goto out;
+	}
+
+	raw_val.refcnt = cpu_to_le32(refcnt);
+	ret = apfs_btree_replace(query, NULL /* key */, 0 /* key_len */, &raw_val, sizeof(raw_val));
+	if (ret)
+		apfs_err(sb, "update failed for id 0x%llx", dstream->ds_id);
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_put_dstream_rec - Put a reference for a data stream record
+ * @dstream: data stream info
+ *
+ * Deletes the record if the reference count goes to zero. Returns 0 on success
+ * or a negative error code in case of failure.
+ */
+static int apfs_put_dstream_rec(struct apfs_dstream_info *dstream)
+{
+	struct apfs_inode_info *ai = APFS_I(dstream->ds_inode);
+
+	if (!ai->i_has_dstream)
+		return 0;
+	return apfs_dstream_adj_refcnt(dstream, -1);
+}
+
+/**
+ * apfs_create_crypto_rec - Create the crypto state record for an inode
+ * @inode: the vfs inode
+ *
+ * Does nothing if the record already exists.  TODO: support cloned files.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_crypto_rec(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	struct apfs_query *query;
+	struct apfs_crypto_state_key raw_key;
+	int ret;
+
+	if (inode->i_size || inode->i_blocks) /* Already has a dstream */
+		return 0;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_crypto_state_key(dstream->ds_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret != -ENODATA) /* Either an error, or the record already exists */
+		goto out;
+
+	apfs_key_set_hdr(APFS_TYPE_CRYPTO_STATE, dstream->ds_id, &raw_key);
+	if (sbi->s_dflt_pfk) {
+		struct apfs_crypto_state_val *raw_val = sbi->s_dflt_pfk;
+		unsigned int key_len = le16_to_cpu(raw_val->state.key_len);
+
+		ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), raw_val, sizeof(*raw_val) + key_len);
+		if (ret)
+			apfs_err(sb, "insertion failed for id 0x%llx", dstream->ds_id);
+	} else {
+		struct apfs_crypto_state_val raw_val;
+
+		raw_val.refcnt = cpu_to_le32(1);
+		raw_val.state.major_version = cpu_to_le16(APFS_WMCS_MAJOR_VERSION);
+		raw_val.state.minor_version = cpu_to_le16(APFS_WMCS_MINOR_VERSION);
+		raw_val.state.cpflags = 0;
+		raw_val.state.persistent_class = cpu_to_le32(APFS_PROTECTION_CLASS_F);
+		raw_val.state.key_os_version = 0;
+		raw_val.state.key_revision = cpu_to_le16(1);
+		raw_val.state.key_len = cpu_to_le16(0);
+		ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+		if (ret)
+			apfs_err(sb, "insertion failed for id 0x%llx", dstream->ds_id);
+	}
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_dflt_key_class - Returns default key class for files in volume
+ * @sb: volume superblock
+ */
+static unsigned int apfs_dflt_key_class(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+
+	if (!sbi->s_dflt_pfk)
+		return APFS_PROTECTION_CLASS_F;
+
+	return le32_to_cpu(sbi->s_dflt_pfk->state.persistent_class);
+}
+
+/**
+ * apfs_create_crypto_rec - Adjust crypto state record refcount
+ * @sb: volume superblock
+ * @crypto_id: crypto_id to adjust
+ * @delta: desired change in reference count
+ *
+ * This function is used when adding or removing extents, as each extent holds
+ * a reference to the crypto ID. It should also be used when removing inodes,
+ * and in that case it should also remove the crypto record (TODO).
+ */
+int apfs_crypto_adj_refcnt(struct super_block *sb, u64 crypto_id, int delta)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_crypto_state_val *raw_val;
+	char *raw;
+	int ret;
+
+	if (!crypto_id)
+		return 0;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_crypto_state_key(crypto_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		apfs_err(sb, "query failed for id 0x%llx", crypto_id);
+		goto out;
+	}
+
+	ret = apfs_query_join_transaction(query);
+	if (ret) {
+		apfs_err(sb, "query join failed");
+		return ret;
+	}
+	raw = query->node->object.data;
+	raw_val = (void *)raw + query->off;
+
+	le32_add_cpu(&raw_val->refcnt, delta);
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_crypto_set_key - Modify content of crypto state record
+ * @sb: volume superblock
+ * @crypto_id: crypto_id to modify
+ * @new_val: new crypto state data; new_val->refcnt is overridden
+ *
+ * This function does not alter the inode's default protection class field.
+ * It needs to be done separately if the class changes.
+ */
+static int apfs_crypto_set_key(struct super_block *sb, u64 crypto_id, struct apfs_crypto_state_val *new_val)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_crypto_state_val *raw_val;
+	char *raw;
+	int ret;
+	unsigned int pfk_len;
+
+	if (!crypto_id)
+		return 0;
+
+	pfk_len = le16_to_cpu(new_val->state.key_len);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_crypto_state_key(crypto_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		apfs_err(sb, "query failed for id 0x%llx", crypto_id);
+		goto out;
+	}
+	raw = query->node->object.data;
+	raw_val = (void *)raw + query->off;
+
+	new_val->refcnt = raw_val->refcnt;
+
+	ret = apfs_btree_replace(query, NULL /* key */, 0 /* key_len */, new_val, sizeof(*new_val) + pfk_len);
+	if (ret)
+		apfs_err(sb, "update failed for id 0x%llx", crypto_id);
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_crypto_get_key - Retrieve content of crypto state record
+ * @sb: volume superblock
+ * @crypto_id: crypto_id to modify
+ * @val: result crypto state data
+ * @max_len: maximum allowed value of val->state.key_len
+ */
+static int apfs_crypto_get_key(struct super_block *sb, u64 crypto_id, struct apfs_crypto_state_val *val,
+			       unsigned int max_len)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_crypto_state_val *raw_val;
+	char *raw;
+	int ret;
+	unsigned int pfk_len;
+
+	if (!crypto_id)
+		return -ENOENT;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_crypto_state_key(crypto_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret)
+		goto out;
+	raw = query->node->object.data;
+	raw_val = (void *)raw + query->off;
+
+	pfk_len = le16_to_cpu(raw_val->state.key_len);
+	if (pfk_len > max_len) {
+		ret = -ENOSPC;
+		goto out;
+	}
+
+	memcpy(val, raw_val, sizeof(*val) + pfk_len);
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+int __apfs_write_begin(const struct kiocb *file, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int flags, struct page **pagep, void **fsdata)
+#else
+int __apfs_write_begin(struct file *file, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int flags, struct page **pagep, void **fsdata)
+#endif
+{
+	struct inode *inode = mapping->host;
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	struct super_block *sb = inode->i_sb;
+	struct page *page;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 7, 0)
+	struct folio *folio;
+#endif
+	struct buffer_head *bh, *head;
+	unsigned int blocksize, block_start, block_end, from, to;
+	pgoff_t index = pos >> PAGE_SHIFT;
+	sector_t iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
+	loff_t i_blks_end;
+	int err;
+
+	apfs_inode_join_transaction(sb, inode);
+
+	err = apfs_inode_create_dstream_rec(inode);
+	if (err) {
+		apfs_err(sb, "failed to create dstream for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	if (apfs_vol_is_encrypted(sb)) {
+		err = apfs_create_crypto_rec(inode);
+		if (err) {
+			apfs_err(sb, "crypto creation failed for ino 0x%llx", apfs_ino(inode));
+			return err;
+		}
+	}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+	flags = memalloc_nofs_save();
+	page = grab_cache_page_write_begin(mapping, index);
+	memalloc_nofs_restore(flags);
+#else
+	page = grab_cache_page_write_begin(mapping, index, flags | AOP_FLAG_NOFS);
+#endif
+	if (!page)
+		return -ENOMEM;
+	if (!page_has_buffers(page)) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+		create_empty_buffers(page, sb->s_blocksize, 0);
+#else
+		folio = page_folio(page);
+		bh = folio_buffers(folio);
+		if (!bh)
+			bh = create_empty_buffers(folio, sb->s_blocksize, 0);
+#endif
+	}
+
+	/* CoW moves existing blocks, so read them but mark them as unmapped */
+	head = page_buffers(page);
+	blocksize = head->b_size;
+	i_blks_end = (inode->i_size + sb->s_blocksize - 1) >> inode->i_blkbits;
+	i_blks_end <<= inode->i_blkbits;
+	if (i_blks_end >= pos) {
+		from = pos & (PAGE_SIZE - 1);
+		to = from + min(i_blks_end - pos, (loff_t)len);
+	} else {
+		/* TODO: deal with preallocated tail blocks */
+		from = UINT_MAX;
+		to = 0;
+	}
+	for (bh = head, block_start = 0; bh != head || !block_start;
+	     block_start = block_end, bh = bh->b_this_page, ++iblock) {
+		block_end = block_start + blocksize;
+		if (to > block_start && from < block_end) {
+			if (buffer_trans(bh))
+				continue;
+			if (!buffer_mapped(bh)) {
+				err = __apfs_get_block(dstream, iblock, bh,
+						       false /* create */);
+				if (err) {
+					apfs_err(sb, "failed to map block for ino 0x%llx", apfs_ino(inode));
+					goto out_put_page;
+				}
+			}
+			if (buffer_mapped(bh) && !buffer_uptodate(bh)) {
+				get_bh(bh);
+				lock_buffer(bh);
+				bh->b_end_io = end_buffer_read_sync;
+				apfs_submit_bh(REQ_OP_READ, 0, bh);
+				wait_on_buffer(bh);
+				if (!buffer_uptodate(bh)) {
+					apfs_err(sb, "failed to read block for ino 0x%llx", apfs_ino(inode));
+					err = -EIO;
+					goto out_put_page;
+				}
+			}
+			clear_buffer_mapped(bh);
+		}
+	}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+	err = __block_write_begin(page_folio(page), pos, len, apfs_get_new_block);
+#else
+	err = __block_write_begin(page, pos, len, apfs_get_new_block);
+#endif
+	if (err) {
+		apfs_err(sb, "CoW failed in inode 0x%llx", apfs_ino(inode));
+		goto out_put_page;
+	}
+
+	*pagep = page;
+	return 0;
+
+out_put_page:
+	unlock_page(page);
+	put_page(page);
+	return err;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+static int apfs_write_begin(const struct kiocb *file, struct address_space *mapping,
+			    loff_t pos, unsigned int len,
+			    struct folio **foliop, void **fsdata)
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+static int apfs_write_begin(struct file *file, struct address_space *mapping,
+			    loff_t pos, unsigned int len,
+			    struct folio **foliop, void **fsdata)
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+static int apfs_write_begin(struct file *file, struct address_space *mapping,
+			    loff_t pos, unsigned int len,
+			    struct page **pagep, void **fsdata)
+#else
+static int apfs_write_begin(struct file *file, struct address_space *mapping,
+			    loff_t pos, unsigned int len, unsigned int flags,
+			    struct page **pagep, void **fsdata)
+#endif
+{
+	struct inode *inode = mapping->host;
+	struct super_block *sb = inode->i_sb;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+	struct page *page = NULL;
+	struct page **pagep = &page;
+#endif
+	int err;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+	unsigned int flags = 0;
+#endif
+
+	if (unlikely(pos >= APFS_MAX_FILE_SIZE))
+		return -EFBIG;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	err = __apfs_write_begin(file, mapping, pos, len, flags, pagep, fsdata);
+	if (err)
+		goto fail;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+	*foliop = page_folio(page);
+#endif
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+int __apfs_write_end(const struct kiocb *file, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int copied, struct page *page, void *fsdata)
+#else
+int __apfs_write_end(struct file *file, struct address_space *mapping, loff_t pos, unsigned int len, unsigned int copied, struct page *page, void *fsdata)
+#endif
+{
+	struct inode *inode = mapping->host;
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	int ret, err;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+	ret = generic_write_end(file, mapping, pos, len, copied, page_folio(page), fsdata);
+#else
+	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
+#endif
+	dstream->ds_size = i_size_read(inode);
+	if (ret < len && pos + len > inode->i_size) {
+		truncate_pagecache(inode, inode->i_size);
+		err = apfs_truncate(dstream, inode->i_size);
+		if (err) {
+			apfs_err(inode->i_sb, "truncation failed for ino 0x%llx", apfs_ino(inode));
+			return err;
+		}
+	}
+	return ret;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+static int apfs_write_end(const struct kiocb *file, struct address_space *mapping,
+			  loff_t pos, unsigned int len, unsigned int copied,
+			  struct folio *folio, void *fsdata)
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+static int apfs_write_end(struct file *file, struct address_space *mapping,
+			  loff_t pos, unsigned int len, unsigned int copied,
+			  struct folio *folio, void *fsdata)
+#else
+static int apfs_write_end(struct file *file, struct address_space *mapping,
+			  loff_t pos, unsigned int len, unsigned int copied,
+			  struct page *page, void *fsdata)
+#endif
+{
+	struct inode *inode = mapping->host;
+	struct super_block *sb = inode->i_sb;
+	struct apfs_nx_transaction *trans = &APFS_NXI(sb)->nx_transaction;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 12, 0)
+	struct page *page = &folio->page;
+#endif
+	int ret, err;
+
+	ret = __apfs_write_end(file, mapping, pos, len, copied, page, fsdata);
+	if (ret < 0) {
+		err = ret;
+		goto fail;
+	}
+
+	if ((pos + ret) & (sb->s_blocksize - 1))
+		trans->t_state |= APFS_NX_TRANS_INCOMPLETE_BLOCK;
+	else
+		trans->t_state &= ~APFS_NX_TRANS_INCOMPLETE_BLOCK;
+
+	err = apfs_transaction_commit(sb);
+	if (!err)
+		return ret;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 18, 0) || RHEL_VERSION_GE(9, 3)
+static void apfs_noop_invalidate_folio(struct folio *folio, size_t offset, size_t length)
+#else
+static void apfs_noop_invalidatepage(struct page *page, unsigned int offset, unsigned int length)
+#endif
+{
+}
+
+/* bmap is not implemented to avoid issues with CoW on swapfiles */
+static const struct address_space_operations apfs_aops = {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 18, 0) || RHEL_VERSION_GE(9, 2)
+	.dirty_folio	= block_dirty_folio,
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(5, 14, 0)
+	.set_page_dirty	= __set_page_dirty_buffers,
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+	.read_folio	= apfs_read_folio,
+#else
+	.readpage	= apfs_readpage,
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
+	.readahead	= apfs_readahead,
+#else
+	.readpages	= apfs_readpages,
+#endif
+
+	.write_begin	= apfs_write_begin,
+	.write_end	= apfs_write_end,
+
+	/* The intention is to keep bhs around until the transaction is over */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 18, 0) || RHEL_VERSION_GE(9, 3)
+	.invalidate_folio = apfs_noop_invalidate_folio,
+#else
+	.invalidatepage	= apfs_noop_invalidatepage,
+#endif
+};
+
+/**
+ * apfs_inode_set_ops - Set up an inode's operations
+ * @inode:	vfs inode to set up
+ * @rdev:	device id (0 if not a device file)
+ * @compressed:	is this a compressed inode?
+ *
+ * For device files, also sets the device id to @rdev.
+ */
+static void apfs_inode_set_ops(struct inode *inode, dev_t rdev, bool compressed)
+{
+	/* A lot of operations still missing, of course */
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFREG:
+		inode->i_op = &apfs_file_inode_operations;
+		if (compressed) {
+			inode->i_fop = &apfs_compress_file_operations;
+			inode->i_mapping->a_ops = &apfs_compress_aops;
+		} else {
+			inode->i_fop = &apfs_file_operations;
+			inode->i_mapping->a_ops = &apfs_aops;
+		}
+		break;
+	case S_IFDIR:
+		inode->i_op = &apfs_dir_inode_operations;
+		inode->i_fop = &apfs_dir_operations;
+		break;
+	case S_IFLNK:
+		inode->i_op = &apfs_symlink_inode_operations;
+		break;
+	default:
+		inode->i_op = &apfs_special_inode_operations;
+		init_special_inode(inode, inode->i_mode, rdev);
+		break;
+	}
+}
+
+/**
+ * apfs_inode_from_query - Read the inode found by a successful query
+ * @query:	the query that found the record
+ * @inode:	vfs inode to be filled with the read data
+ *
+ * Reads the inode record into @inode and performs some basic sanity checks,
+ * mostly as a protection against crafted filesystems.  Returns 0 on success
+ * or a negative error code otherwise.
+ */
+static int apfs_inode_from_query(struct apfs_query *query, struct inode *inode)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct apfs_dstream_info *dstream = &ai->i_dstream;
+	struct apfs_inode_val *inode_val;
+	char *raw = query->node->object.data;
+	char *xval = NULL;
+	int xlen;
+	u32 rdev = 0, bsd_flags;
+	bool compressed = false;
+
+	if (query->len < sizeof(*inode_val))
+		goto corrupted;
+
+	inode_val = (struct apfs_inode_val *)(raw + query->off);
+
+	ai->i_parent_id = le64_to_cpu(inode_val->parent_id);
+	dstream->ds_id = le64_to_cpu(inode_val->private_id);
+	inode->i_mode = le16_to_cpu(inode_val->mode);
+	ai->i_key_class = le32_to_cpu(inode_val->default_protection_class);
+	ai->i_int_flags = le64_to_cpu(inode_val->internal_flags);
+
+	ai->i_saved_uid = le32_to_cpu(inode_val->owner);
+	i_uid_write(inode, ai->i_saved_uid);
+	ai->i_saved_gid = le32_to_cpu(inode_val->group);
+	i_gid_write(inode, ai->i_saved_gid);
+
+	ai->i_bsd_flags = bsd_flags = le32_to_cpu(inode_val->bsd_flags);
+	if (bsd_flags & APFS_INOBSD_IMMUTABLE)
+		inode->i_flags |= S_IMMUTABLE;
+	if (bsd_flags & APFS_INOBSD_APPEND)
+		inode->i_flags |= S_APPEND;
+
+	if (!S_ISDIR(inode->i_mode)) {
+		/*
+		 * Directory inodes don't store their link count, so to provide
+		 * it we would have to actually count the subdirectories. The
+		 * HFS/HFS+ modules just leave it at 1, and so do we, for now.
+		 */
+		set_nlink(inode, le32_to_cpu(inode_val->nlink));
+	} else {
+		ai->i_nchildren = le32_to_cpu(inode_val->nchildren);
+	}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode->i_ctime = ns_to_timespec64(le64_to_cpu(inode_val->change_time));
+#else
+	inode_set_ctime_to_ts(inode, ns_to_timespec64(le64_to_cpu(inode_val->change_time)));
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	inode->i_atime = ns_to_timespec64(le64_to_cpu(inode_val->access_time));
+	inode->i_mtime = ns_to_timespec64(le64_to_cpu(inode_val->mod_time));
+#else
+	inode_set_atime_to_ts(inode, ns_to_timespec64(le64_to_cpu(inode_val->access_time)));
+	inode_set_mtime_to_ts(inode, ns_to_timespec64(le64_to_cpu(inode_val->mod_time)));
+#endif
+	ai->i_crtime = ns_to_timespec64(le64_to_cpu(inode_val->create_time));
+
+	dstream->ds_size = inode->i_size = inode->i_blocks = 0;
+	ai->i_has_dstream = false;
+	if ((bsd_flags & APFS_INOBSD_COMPRESSED) && !S_ISDIR(inode->i_mode)) {
+		if (!apfs_compress_get_size(inode, &inode->i_size)) {
+			inode->i_blocks = (inode->i_size + 511) >> 9;
+			compressed = true;
+		}
+	} else {
+		xlen = apfs_find_xfield(inode_val->xfields,
+					query->len - sizeof(*inode_val),
+					APFS_INO_EXT_TYPE_DSTREAM, &xval);
+		if (xlen >= sizeof(struct apfs_dstream)) {
+			struct apfs_dstream *dstream_raw = (struct apfs_dstream *)xval;
+
+			dstream->ds_size = inode->i_size = le64_to_cpu(dstream_raw->size);
+			inode->i_blocks = le64_to_cpu(dstream_raw->alloced_size) >> 9;
+			ai->i_has_dstream = true;
+		}
+	}
+	xval = NULL;
+
+	/* TODO: move each xfield read to its own function */
+	dstream->ds_sparse_bytes = 0;
+	xlen = apfs_find_xfield(inode_val->xfields, query->len - sizeof(*inode_val), APFS_INO_EXT_TYPE_SPARSE_BYTES, &xval);
+	if (xlen >= sizeof(__le64)) {
+		__le64 *sparse_bytes_p = (__le64 *)xval;
+
+		dstream->ds_sparse_bytes = le64_to_cpup(sparse_bytes_p);
+	}
+	xval = NULL;
+
+	rdev = 0;
+	xlen = apfs_find_xfield(inode_val->xfields,
+				query->len - sizeof(*inode_val),
+				APFS_INO_EXT_TYPE_RDEV, &xval);
+	if (xlen >= sizeof(__le32)) {
+		__le32 *rdev_p = (__le32 *)xval;
+
+		rdev = le32_to_cpup(rdev_p);
+	}
+
+	apfs_inode_set_ops(inode, rdev, compressed);
+	return 0;
+
+corrupted:
+	apfs_err(inode->i_sb, "bad inode record for inode 0x%llx", apfs_ino(inode));
+	return -EFSCORRUPTED;
+}
+
+/**
+ * apfs_inode_lookup - Lookup an inode record in the catalog b-tree
+ * @inode:	vfs inode to lookup
+ *
+ * Runs a catalog query for the apfs_ino(@inode) inode record; returns a pointer
+ * to the query structure on success, or an error pointer in case of failure.
+ */
+static struct apfs_query *apfs_inode_lookup(const struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return ERR_PTR(-ENOMEM);
+	apfs_init_inode_key(apfs_ino(inode), &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (!ret)
+		return query;
+
+	/* Don't complain if an orphan is already gone */
+	if (!current_work() || ret != -ENODATA)
+		apfs_err(sb, "query failed for id 0x%llx", apfs_ino(inode));
+	apfs_free_query(query);
+	return ERR_PTR(ret);
+}
+
+/**
+ * apfs_test_inode - Check if the inode matches a 64-bit inode number
+ * @inode:	inode to test
+ * @cnid:	pointer to the inode number
+ */
+static int apfs_test_inode(struct inode *inode, void *cnid)
+{
+	u64 *ino = cnid;
+
+	return apfs_ino(inode) == *ino;
+}
+
+/**
+ * apfs_set_inode - Set a 64-bit inode number on the given inode
+ * @inode:	inode to set
+ * @cnid:	pointer to the inode number
+ */
+static int apfs_set_inode(struct inode *inode, void *cnid)
+{
+	apfs_set_ino(inode, *(u64 *)cnid);
+	return 0;
+}
+
+/**
+ * apfs_iget_locked - Wrapper for iget5_locked()
+ * @sb:		filesystem superblock
+ * @cnid:	64-bit inode number
+ *
+ * Works the same as iget_locked(), but can handle 64-bit inode numbers on
+ * 32-bit architectures.
+ */
+static struct inode *apfs_iget_locked(struct super_block *sb, u64 cnid)
+{
+	return iget5_locked(sb, cnid, apfs_test_inode, apfs_set_inode, &cnid);
+}
+
+/**
+ * apfs_check_dstream_refcnt - Check if an inode's dstream is shared
+ * @inode:	the inode to check
+ *
+ * Sets the value of ds_shared for the inode's dstream. Returns 0 on success,
+ * or a negative error code in case of failure.
+ */
+static int apfs_check_dstream_refcnt(struct inode *inode)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct apfs_dstream_info *dstream = &ai->i_dstream;
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	struct apfs_dstream_id_val raw_val;
+	void *raw = NULL;
+	u32 refcnt;
+	int ret;
+
+	if (!ai->i_has_dstream) {
+		dstream->ds_shared = false;
+		return 0;
+	}
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_dstream_id_key(dstream->ds_id, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		apfs_err(sb, "query failed for id 0x%llx", dstream->ds_id);
+		if (ret == -ENODATA)
+			ret = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	if (query->len != sizeof(raw_val)) {
+		ret = -EFSCORRUPTED;
+		goto fail;
+	}
+	raw = query->node->object.data;
+	raw_val = *(struct apfs_dstream_id_val *)(raw + query->off);
+	refcnt = le32_to_cpu(raw_val.refcnt);
+
+	dstream->ds_shared = refcnt > 1;
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_iget - Populate inode structures with metadata from disk
+ * @sb:		filesystem superblock
+ * @cnid:	inode number
+ *
+ * Populates the vfs inode and the corresponding apfs_inode_info structure.
+ * Returns a pointer to the vfs inode in case of success, or an appropriate
+ * error pointer otherwise.
+ */
+struct inode *apfs_iget(struct super_block *sb, u64 cnid)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct inode *inode;
+	struct apfs_query *query;
+	int err;
+
+	inode = apfs_iget_locked(sb, cnid);
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+	if (!(apfs_inode_state_read_once(inode) & I_NEW))
+		return inode;
+
+	down_read(&nxi->nx_big_sem);
+	query = apfs_inode_lookup(inode);
+	if (IS_ERR(query)) {
+		err = PTR_ERR(query);
+		/* Don't complain if an orphan is already gone */
+		if (!current_work() || err != -ENODATA)
+			apfs_err(sb, "lookup failed for ino 0x%llx", cnid);
+		goto fail;
+	}
+	err = apfs_inode_from_query(query, inode);
+	apfs_free_query(query);
+	if (err)
+		goto fail;
+	err = apfs_check_dstream_refcnt(inode);
+	if (err) {
+		apfs_err(sb, "refcnt check failed for ino 0x%llx", cnid);
+		goto fail;
+	}
+	up_read(&nxi->nx_big_sem);
+
+	/* Allow the user to override the ownership */
+	if (uid_valid(sbi->s_uid))
+		inode->i_uid = sbi->s_uid;
+	if (gid_valid(sbi->s_gid))
+		inode->i_gid = sbi->s_gid;
+
+	/* Inode flags are not important for now, leave them at 0 */
+	unlock_new_inode(inode);
+	return inode;
+
+fail:
+	up_read(&nxi->nx_big_sem);
+	iget_failed(inode);
+	return ERR_PTR(err);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0) /* No statx yet... */
+
+int apfs_getattr(struct vfsmount *mnt, struct dentry *dentry,
+		 struct kstat *stat)
+{
+	struct inode *inode = d_inode(dentry);
+
+	generic_fillattr(inode, stat);
+	stat->dev = APFS_SB(inode->i_sb)->s_anon_dev;
+	stat->ino = apfs_ino(inode);
+	return 0;
+}
+
+#else /* LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0) */
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+int apfs_getattr(const struct path *path, struct kstat *stat,
+		 u32 request_mask, unsigned int query_flags)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+int apfs_getattr(struct user_namespace *mnt_userns,
+		 const struct path *path, struct kstat *stat, u32 request_mask,
+		 unsigned int query_flags)
+#else
+int apfs_getattr(struct mnt_idmap *idmap,
+		 const struct path *path, struct kstat *stat, u32 request_mask,
+		 unsigned int query_flags)
+#endif
+{
+	struct inode *inode = d_inode(path->dentry);
+	struct apfs_inode_info *ai = APFS_I(inode);
+
+	stat->result_mask |= STATX_BTIME;
+	stat->btime = ai->i_crtime;
+
+	if (ai->i_bsd_flags & APFS_INOBSD_APPEND)
+		stat->attributes |= STATX_ATTR_APPEND;
+	if (ai->i_bsd_flags & APFS_INOBSD_IMMUTABLE)
+		stat->attributes |= STATX_ATTR_IMMUTABLE;
+	if (ai->i_bsd_flags & APFS_INOBSD_NODUMP)
+		stat->attributes |= STATX_ATTR_NODUMP;
+	if (ai->i_bsd_flags & APFS_INOBSD_COMPRESSED)
+		stat->attributes |= STATX_ATTR_COMPRESSED;
+
+	stat->attributes_mask |= STATX_ATTR_APPEND;
+	stat->attributes_mask |= STATX_ATTR_IMMUTABLE;
+	stat->attributes_mask |= STATX_ATTR_NODUMP;
+	stat->attributes_mask |= STATX_ATTR_COMPRESSED;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+	generic_fillattr(inode, stat);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+	generic_fillattr(mnt_userns, inode, stat);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	generic_fillattr(idmap, inode, stat);
+#else
+	generic_fillattr(idmap, request_mask, inode, stat);
+#endif
+
+	stat->dev = APFS_SB(inode->i_sb)->s_anon_dev;
+	stat->ino = apfs_ino(inode);
+	return 0;
+}
+
+#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0) */
+
+/**
+ * apfs_build_inode_val - Allocate and initialize the value for an inode record
+ * @inode:	vfs inode to record
+ * @qname:	filename for primary link
+ * @val_p:	on return, a pointer to the new on-disk value structure
+ *
+ * Returns the length of the value, or a negative error code in case of failure.
+ */
+static int apfs_build_inode_val(struct inode *inode, const struct qstr *qname,
+				struct apfs_inode_val **val_p)
+{
+	struct apfs_inode_val *val;
+	struct apfs_x_field xkey;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 7, 0)
+	struct timespec64 ts;
+#endif
+	int total_xlen, val_len;
+	bool is_device = S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode);
+	__le32 rdev;
+
+	/* The only required xfield is the name, and the id if it's a device */
+	total_xlen = sizeof(struct apfs_xf_blob);
+	total_xlen += sizeof(xkey) + round_up(qname->len + 1, 8);
+	if (is_device)
+		total_xlen += sizeof(xkey) + round_up(sizeof(rdev), 8);
+
+	val_len = sizeof(*val) + total_xlen;
+	val = kzalloc(val_len, GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	val->parent_id = cpu_to_le64(APFS_I(inode)->i_parent_id);
+	val->private_id = cpu_to_le64(apfs_ino(inode));
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	val->mod_time = cpu_to_le64(timespec64_to_ns(&inode->i_mtime));
+#else
+	ts = inode_get_mtime(inode);
+	val->mod_time = cpu_to_le64(timespec64_to_ns(&ts));
+#endif
+	val->create_time = val->change_time = val->access_time = val->mod_time;
+
+	if (S_ISDIR(inode->i_mode))
+		val->nchildren = 0;
+	else
+		val->nlink = cpu_to_le32(1);
+
+	val->owner = cpu_to_le32(i_uid_read(inode));
+	val->group = cpu_to_le32(i_gid_read(inode));
+	val->mode = cpu_to_le16(inode->i_mode);
+
+	/* The buffer was just allocated: none of these functions should fail */
+	apfs_init_xfields(val->xfields, total_xlen);
+	xkey.x_type = APFS_INO_EXT_TYPE_NAME;
+	xkey.x_flags = APFS_XF_DO_NOT_COPY;
+	xkey.x_size = cpu_to_le16(qname->len + 1);
+	apfs_insert_xfield(val->xfields, total_xlen, &xkey, qname->name);
+	if (is_device) {
+		rdev = cpu_to_le32(inode->i_rdev);
+		xkey.x_type = APFS_INO_EXT_TYPE_RDEV;
+		xkey.x_flags = 0; /* TODO: proper flags here? */
+		xkey.x_size = cpu_to_le16(sizeof(rdev));
+		apfs_insert_xfield(val->xfields, total_xlen, &xkey, &rdev);
+	}
+
+	*val_p = val;
+	return val_len;
+}
+
+/*
+ * apfs_inode_rename - Update the primary name reported in an inode record
+ * @inode:	the in-memory inode
+ * @new_name:	name of the new primary link (NULL if unchanged)
+ * @query:	the query that found the inode record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_inode_rename(struct inode *inode, char *new_name,
+			     struct apfs_query *query)
+{
+	char *raw = query->node->object.data;
+	struct apfs_inode_val *new_val = NULL;
+	int buflen, namelen;
+	struct apfs_x_field xkey;
+	int xlen;
+	int err;
+
+	if (!new_name)
+		return 0;
+
+	namelen = strlen(new_name) + 1; /* Count the null-termination */
+	buflen = query->len;
+	buflen += sizeof(struct apfs_x_field) + round_up(namelen, 8);
+	new_val = kzalloc(buflen, GFP_KERNEL);
+	if (!new_val)
+		return -ENOMEM;
+	memcpy(new_val, raw + query->off, query->len);
+
+	/* TODO: can we assume that all inode records have an xfield blob? */
+	xkey.x_type = APFS_INO_EXT_TYPE_NAME;
+	xkey.x_flags = APFS_XF_DO_NOT_COPY;
+	xkey.x_size = cpu_to_le16(namelen);
+	xlen = apfs_insert_xfield(new_val->xfields, buflen - sizeof(*new_val),
+				  &xkey, new_name);
+	if (!xlen) {
+		/* Buffer has enough space, but the metadata claims otherwise */
+		apfs_err(inode->i_sb, "bad xfields on inode 0x%llx", apfs_ino(inode));
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	/* Just remove the old record and create a new one */
+	err = apfs_btree_replace(query, NULL /* key */, 0 /* key_len */, new_val, sizeof(*new_val) + xlen);
+	if (err)
+		apfs_err(inode->i_sb, "update failed for ino 0x%llx", apfs_ino(inode));
+
+fail:
+	kfree(new_val);
+	return err;
+}
+
+/**
+ * apfs_create_dstream_xfield - Create the inode xfield for a new data stream
+ * @inode:	the in-memory inode
+ * @query:	the query that found the inode record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_create_dstream_xfield(struct inode *inode,
+				      struct apfs_query *query)
+{
+	char *raw = query->node->object.data;
+	struct apfs_inode_val *new_val;
+	struct apfs_dstream dstream_raw = {0};
+	struct apfs_x_field xkey;
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	int xlen;
+	int buflen;
+	int err;
+
+	buflen = query->len;
+	buflen += sizeof(struct apfs_x_field) + sizeof(dstream_raw);
+	new_val = kzalloc(buflen, GFP_KERNEL);
+	if (!new_val)
+		return -ENOMEM;
+	memcpy(new_val, raw + query->off, query->len);
+
+	dstream_raw.size = cpu_to_le64(inode->i_size);
+	dstream_raw.alloced_size = cpu_to_le64(apfs_alloced_size(dstream));
+	if (apfs_vol_is_encrypted(inode->i_sb))
+		dstream_raw.default_crypto_id = cpu_to_le64(dstream->ds_id);
+
+	/* TODO: can we assume that all inode records have an xfield blob? */
+	xkey.x_type = APFS_INO_EXT_TYPE_DSTREAM;
+	xkey.x_flags = APFS_XF_SYSTEM_FIELD;
+	xkey.x_size = cpu_to_le16(sizeof(dstream_raw));
+	xlen = apfs_insert_xfield(new_val->xfields, buflen - sizeof(*new_val),
+				  &xkey, &dstream_raw);
+	if (!xlen) {
+		/* Buffer has enough space, but the metadata claims otherwise */
+		apfs_err(inode->i_sb, "bad xfields on inode 0x%llx", apfs_ino(inode));
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	/* Just remove the old record and create a new one */
+	err = apfs_btree_replace(query, NULL /* key */, 0 /* key_len */, new_val, sizeof(*new_val) + xlen);
+	if (err)
+		apfs_err(inode->i_sb, "update failed for ino 0x%llx", apfs_ino(inode));
+
+fail:
+	kfree(new_val);
+	return err;
+}
+
+/**
+ * apfs_inode_resize - Update the sizes reported in an inode record
+ * @inode:	the in-memory inode
+ * @query:	the query that found the inode record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_inode_resize(struct inode *inode, struct apfs_query *query)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+	char *raw;
+	struct apfs_inode_val *inode_raw;
+	char *xval;
+	int xlen;
+	int err;
+
+	/* All dstream records must have a matching xfield, even if empty */
+	if (!ai->i_has_dstream)
+		return 0;
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(inode->i_sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+	inode_raw = (void *)raw + query->off;
+
+	xlen = apfs_find_xfield(inode_raw->xfields,
+				query->len - sizeof(*inode_raw),
+				APFS_INO_EXT_TYPE_DSTREAM, &xval);
+
+	if (xlen) {
+		struct apfs_dstream *dstream;
+
+		if (xlen != sizeof(*dstream)) {
+			apfs_err(inode->i_sb, "bad xlen (%d) on inode 0x%llx", xlen, apfs_ino(inode));
+			return -EFSCORRUPTED;
+		}
+		dstream = (struct apfs_dstream *)xval;
+
+		/* TODO: count bytes read and written */
+		dstream->size = cpu_to_le64(inode->i_size);
+		dstream->alloced_size = cpu_to_le64(apfs_alloced_size(&ai->i_dstream));
+		return 0;
+	}
+	/* This inode has no dstream xfield, so we need to create it */
+	return apfs_create_dstream_xfield(inode, query);
+}
+
+/**
+ * apfs_create_sparse_xfield - Create an inode xfield to count sparse bytes
+ * @inode:	the in-memory inode
+ * @query:	the query that found the inode record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_create_sparse_xfield(struct inode *inode, struct apfs_query *query)
+{
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	char *raw = query->node->object.data;
+	struct apfs_inode_val *new_val;
+	__le64 sparse_bytes;
+	struct apfs_x_field xkey;
+	int xlen;
+	int buflen;
+	int err;
+
+	buflen = query->len;
+	buflen += sizeof(struct apfs_x_field) + sizeof(sparse_bytes);
+	new_val = kzalloc(buflen, GFP_KERNEL);
+	if (!new_val)
+		return -ENOMEM;
+	memcpy(new_val, raw + query->off, query->len);
+
+	sparse_bytes = cpu_to_le64(dstream->ds_sparse_bytes);
+
+	/* TODO: can we assume that all inode records have an xfield blob? */
+	xkey.x_type = APFS_INO_EXT_TYPE_SPARSE_BYTES;
+	xkey.x_flags = APFS_XF_SYSTEM_FIELD | APFS_XF_CHILDREN_INHERIT;
+	xkey.x_size = cpu_to_le16(sizeof(sparse_bytes));
+	xlen = apfs_insert_xfield(new_val->xfields, buflen - sizeof(*new_val), &xkey, &sparse_bytes);
+	if (!xlen) {
+		/* Buffer has enough space, but the metadata claims otherwise */
+		apfs_err(inode->i_sb, "bad xfields on inode 0x%llx", apfs_ino(inode));
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	/* Just remove the old record and create a new one */
+	err = apfs_btree_replace(query, NULL /* key */, 0 /* key_len */, new_val, sizeof(*new_val) + xlen);
+	if (err)
+		apfs_err(inode->i_sb, "update failed for ino 0x%llx", apfs_ino(inode));
+
+fail:
+	kfree(new_val);
+	return err;
+}
+
+/**
+ * apfs_inode_resize_sparse - Update sparse byte count reported in inode record
+ * @inode:	the in-memory inode
+ * @query:	the query that found the inode record
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ *
+ * TODO: should the xfield be removed if the count reaches 0? Should the inode
+ * flag change?
+ */
+static int apfs_inode_resize_sparse(struct inode *inode, struct apfs_query *query)
+{
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	char *raw;
+	struct apfs_inode_val *inode_raw;
+	char *xval;
+	int xlen;
+	int err;
+
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(inode->i_sb, "query join failed");
+		return err;
+	}
+	raw = query->node->object.data;
+	inode_raw = (void *)raw + query->off;
+
+	xlen = apfs_find_xfield(inode_raw->xfields,
+				query->len - sizeof(*inode_raw),
+				APFS_INO_EXT_TYPE_SPARSE_BYTES, &xval);
+	if (!xlen && !dstream->ds_sparse_bytes)
+		return 0;
+
+	if (xlen) {
+		__le64 *sparse_bytes_p;
+
+		if (xlen != sizeof(*sparse_bytes_p)) {
+			apfs_err(inode->i_sb, "bad xlen (%d) on inode 0x%llx", xlen, apfs_ino(inode));
+			return -EFSCORRUPTED;
+		}
+		sparse_bytes_p = (__le64 *)xval;
+
+		*sparse_bytes_p = cpu_to_le64(dstream->ds_sparse_bytes);
+		return 0;
+	}
+	return apfs_create_sparse_xfield(inode, query);
+}
+
+/**
+ * apfs_update_inode - Update an existing inode record
+ * @inode:	the modified in-memory inode
+ * @new_name:	name of the new primary link (NULL if unchanged)
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_update_inode(struct inode *inode, char *new_name)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct apfs_dstream_info *dstream = &ai->i_dstream;
+	struct apfs_query *query;
+	struct apfs_btree_node_phys *node_raw;
+	struct apfs_inode_val *inode_raw;
+	int err;
+
+	err = apfs_flush_extent_cache(dstream);
+	if (err) {
+		apfs_err(sb, "extent cache flush failed for inode 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	query = apfs_inode_lookup(inode);
+	if (IS_ERR(query)) {
+		apfs_err(sb, "lookup failed for ino 0x%llx", apfs_ino(inode));
+		return PTR_ERR(query);
+	}
+
+	/* TODO: copy the record to memory and make all xfield changes there */
+	err = apfs_inode_rename(inode, new_name, query);
+	if (err) {
+		apfs_err(sb, "rename failed for ino 0x%llx", apfs_ino(inode));
+		goto fail;
+	}
+
+	err = apfs_inode_resize(inode, query);
+	if (err) {
+		apfs_err(sb, "resize failed for ino 0x%llx", apfs_ino(inode));
+		goto fail;
+	}
+
+	err = apfs_inode_resize_sparse(inode, query);
+	if (err) {
+		apfs_err(sb, "sparse resize failed for ino 0x%llx", apfs_ino(inode));
+		goto fail;
+	}
+	if (dstream->ds_sparse_bytes)
+		ai->i_int_flags |= APFS_INODE_IS_SPARSE;
+
+	/* TODO: just use apfs_btree_replace()? */
+	err = apfs_query_join_transaction(query);
+	if (err) {
+		apfs_err(sb, "query join failed");
+		goto fail;
+	}
+	node_raw = (void *)query->node->object.data;
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+	inode_raw = (void *)node_raw + query->off;
+
+	inode_raw->parent_id = cpu_to_le64(ai->i_parent_id);
+	inode_raw->private_id = cpu_to_le64(dstream->ds_id);
+	inode_raw->mode = cpu_to_le16(inode->i_mode);
+	inode_raw->owner = cpu_to_le32(i_uid_read(inode));
+	inode_raw->group = cpu_to_le32(i_gid_read(inode));
+	inode_raw->default_protection_class = cpu_to_le32(ai->i_key_class);
+	inode_raw->internal_flags = cpu_to_le64(ai->i_int_flags);
+	inode_raw->bsd_flags = cpu_to_le32(ai->i_bsd_flags);
+
+	/* Don't persist the uid/gid provided by the user on mount */
+	if (uid_valid(sbi->s_uid))
+		inode_raw->owner = cpu_to_le32(ai->i_saved_uid);
+	if (gid_valid(sbi->s_gid))
+		inode_raw->group = cpu_to_le32(ai->i_saved_gid);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode_raw->change_time = cpu_to_le64(timespec64_to_ns(&inode->i_ctime));
+#else
+	struct timespec64 ictime = inode_get_ctime(inode);
+	inode_raw->change_time = cpu_to_le64(timespec64_to_ns(&ictime));
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	inode_raw->access_time = cpu_to_le64(timespec64_to_ns(&inode->i_atime));
+	inode_raw->mod_time = cpu_to_le64(timespec64_to_ns(&inode->i_mtime));
+#else
+	struct timespec64 ts = inode_get_mtime(inode);
+	inode_raw->mod_time = cpu_to_le64(timespec64_to_ns(&ts));
+	ts = inode_get_atime(inode);
+	inode_raw->access_time = cpu_to_le64(timespec64_to_ns(&ts));
+#endif
+	inode_raw->create_time = cpu_to_le64(timespec64_to_ns(&ai->i_crtime));
+
+	if (S_ISDIR(inode->i_mode)) {
+		inode_raw->nchildren = cpu_to_le32(ai->i_nchildren);
+	} else {
+		/* The remaining link for orphan inodes is not counted */
+		inode_raw->nlink = cpu_to_le32(inode->i_nlink);
+	}
+
+fail:
+	apfs_free_query(query);
+	return err;
+}
+
+/**
+ * apfs_delete_inode - Delete an inode record
+ * @inode: the vfs inode to delete
+ *
+ * Returns 0 on success or a negative error code in case of failure, which may
+ * be -EAGAIN if the inode was not deleted in full.
+ */
+static int apfs_delete_inode(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct apfs_dstream_info *dstream = NULL;
+	struct apfs_query *query;
+	u64 old_dstream_id;
+	int ret;
+
+	ret = apfs_delete_all_xattrs(inode);
+	if (ret) {
+		apfs_err(sb, "xattr deletion failed for ino 0x%llx", apfs_ino(inode));
+		return ret;
+	}
+
+	dstream = &ai->i_dstream;
+	old_dstream_id = dstream->ds_id;
+
+	/*
+	 * This is very wasteful since all the new extents and references will
+	 * get deleted right away, but it only affects clones, so I don't see a
+	 * big reason to improve it (TODO)
+	 */
+	ret = apfs_inode_create_exclusive_dstream(inode);
+	if (ret) {
+		apfs_err(sb, "dstream creation failed for ino 0x%llx", apfs_ino(inode));
+		return ret;
+	}
+
+	/* TODO: what about partial deletion of xattrs? Is that allowed? */
+	ret = apfs_inode_delete_front(inode);
+	if (ret) {
+		/*
+		 * If the inode had too many extents, only the first few get
+		 * deleted and the inode remains in the orphan list for now.
+		 * I don't know why the deletion starts at the front, but it
+		 * seems to be what the official driver does.
+		 */
+		if (ret != -EAGAIN) {
+			apfs_err(sb, "head deletion failed for ino 0x%llx", apfs_ino(inode));
+			return ret;
+		}
+		if (dstream->ds_id != old_dstream_id) {
+			ret = apfs_update_inode(inode, NULL /* new_name */);
+			if (ret) {
+				apfs_err(sb, "dstream id update failed for orphan 0x%llx", apfs_ino(inode));
+				return ret;
+			}
+		}
+		return -EAGAIN;
+	}
+
+	ret = apfs_put_dstream_rec(dstream);
+	if (ret) {
+		apfs_err(sb, "failed to put dstream for ino 0x%llx", apfs_ino(inode));
+		return ret;
+	}
+	dstream = NULL;
+	ai->i_has_dstream = false;
+
+	query = apfs_inode_lookup(inode);
+	if (IS_ERR(query)) {
+		apfs_err(sb, "lookup failed for ino 0x%llx", apfs_ino(inode));
+		return PTR_ERR(query);
+	}
+	ret = apfs_btree_remove(query);
+	apfs_free_query(query);
+	if (ret) {
+		apfs_err(sb, "removal failed for ino 0x%llx", apfs_ino(inode));
+		return ret;
+	}
+
+	ai->i_cleaned = true;
+	return ret;
+}
+
+/**
+ * apfs_clean_single_orphan - Clean the given orphan file
+ * @inode:	inode for the file to clean
+ *
+ * Returns 0 on success or a negative error code in case of failure, which may
+ * be -EAGAIN if the file could not be deleted in full.
+ */
+static int apfs_clean_single_orphan(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	u64 ino = apfs_ino(inode);
+	bool eagain = false;
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_DEL);
+	if (err)
+		return err;
+	err = apfs_delete_inode(inode);
+	if (err) {
+		if (err != -EAGAIN) {
+			apfs_err(sb, "failed to delete orphan 0x%llx", ino);
+			goto fail;
+		}
+		eagain = true;
+	} else {
+		err = apfs_delete_orphan_link(inode);
+		if (err) {
+			apfs_err(sb, "failed to unlink orphan 0x%llx", ino);
+			goto fail;
+		}
+	}
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return eagain ? -EAGAIN : 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_clean_any_orphan - Pick an orphan and delete as much as reasonable
+ * @sb:		filesystem superblock
+ *
+ * Returns 0 on success, or a negative error code in case of failure, which may
+ * be -ENODATA if there are no more orphan files or -EAGAIN if a file could not
+ * be deleted in full.
+ */
+static int apfs_clean_any_orphan(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct inode *inode = NULL;
+	int err;
+	u64 ino;
+
+	down_read(&nxi->nx_big_sem);
+	err = apfs_any_orphan_ino(sb, &ino);
+	up_read(&nxi->nx_big_sem);
+	if (err) {
+		if (err == -ENODATA)
+			return -ENODATA;
+		apfs_err(sb, "failed to find orphan inode numbers");
+		return err;
+	}
+
+	inode = apfs_iget(sb, ino);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		if (err != -ENODATA) {
+			apfs_err(sb, "iget failed for orphan 0x%llx", ino);
+			return err;
+		}
+		/*
+		 * This happens rarely for files with no extents, if we hit a
+		 * race with ->evict_inode(). Not a problem: the file is gone.
+		 */
+		apfs_notice(sb, "orphan 0x%llx not found", ino);
+		return 0;
+	}
+
+	if (atomic_read(&inode->i_count) > 1)
+		goto out;
+	err = apfs_clean_single_orphan(inode);
+	if (err && err != -EAGAIN) {
+		apfs_err(sb, "failed to clean orphan 0x%llx", ino);
+		goto out;
+	}
+out:
+	iput(inode);
+	return err;
+}
+
+/**
+ * apfs_schedule_orphan_cleanup - Schedule cleanup for orphan inodes
+ * @sb: filesystem superblock
+ */
+void apfs_schedule_orphan_cleanup(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+
+	/*
+	 * Don't schedule cleanups during unmount: completing all of it could
+	 * take a while so just leave future mounts to handle the orphans.
+	 */
+	if (atomic_read(&sb->s_active) == 0)
+		return;
+
+	/*
+	 * Don't keep retrying orphan cleanups nonstop when they run into an
+	 * unexpected error: it won't do any good and it will flood dmesg. We
+	 * will retry eventually for ENOSPC, but that's handled elsewhere.
+	 */
+	if (atomic_read(&sbi->s_orphan_cleanup_err))
+		return;
+
+	schedule_work(&sbi->s_orphan_cleanup_work);
+}
+
+/**
+ * apfs_clean_orphans - Delete as many orphan files as is reasonable
+ * @sb: filesystem superblock
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_clean_orphans(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	int ret, i;
+
+	for (i = 0; i < 100; ++i) {
+		ret = apfs_clean_any_orphan(sb);
+		if (ret == 0)
+			continue;
+		if (ret == -ENODATA)
+			return 0;
+		if (ret == -EAGAIN)
+			break;
+		apfs_err(sb, "failed to delete an orphan file");
+		atomic_set(&sbi->s_orphan_cleanup_err, ret);
+		return ret;
+	}
+
+	/*
+	 * If a file is too big, or if there are too many files, take a break
+	 * and continue later.
+	 */
+	apfs_schedule_orphan_cleanup(sb);
+	return 0;
+}
+
+void apfs_evict_inode(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_inode_info *ai = APFS_I(inode);
+	int err;
+
+	if (is_bad_inode(inode) || inode->i_nlink || ai->i_cleaned)
+		goto out;
+
+	if (!ai->i_has_dstream || ai->i_dstream.ds_size == 0) {
+		/* For files with no extents, scheduled cleanup wastes time */
+		err = apfs_clean_single_orphan(inode);
+		if (err) {
+			apfs_err(sb, "failed to clean orphan 0x%llx (err:%d)", apfs_ino(inode), err);
+			atomic_set(&APFS_SB(sb)->s_orphan_cleanup_err, err);
+		}
+		goto out;
+	}
+
+	/* If the inode still has extents then schedule cleanup for the rest */
+	apfs_schedule_orphan_cleanup(sb);
+out:
+	truncate_inode_pages_final(&inode->i_data);
+	clear_inode(inode);
+}
+
+void apfs_orphan_cleanup_work(struct work_struct *work)
+{
+	struct super_block *sb = NULL;
+	struct apfs_sb_info *sbi = NULL;
+	struct inode *priv = NULL;
+	int err;
+
+	sbi = container_of(work, struct apfs_sb_info, s_orphan_cleanup_work);
+	priv = sbi->s_private_dir;
+	sb = priv->i_sb;
+
+	if (sb->s_flags & SB_RDONLY) {
+		apfs_alert(sb, "attempt to flush orphans in read-only mount");
+		return;
+	}
+
+	err = apfs_clean_orphans(sb);
+	if (err)
+		apfs_err(sb, "orphan cleanup failed (err:%d)", err);
+}
+
+/**
+ * apfs_insert_inode_locked - Wrapper for insert_inode_locked4()
+ * @inode: vfs inode to insert in cache
+ *
+ * Works the same as insert_inode_locked(), but can handle 64-bit inode numbers
+ * on 32-bit architectures.
+ */
+static int apfs_insert_inode_locked(struct inode *inode)
+{
+	u64 cnid = apfs_ino(inode);
+
+	return insert_inode_locked4(inode, cnid, apfs_test_inode, &cnid);
+}
+
+/**
+ * apfs_new_inode - Create a new in-memory inode
+ * @dir:	parent inode
+ * @mode:	mode bits for the new inode
+ * @rdev:	device id (0 if not a device file)
+ *
+ * Returns a pointer to the new vfs inode on success, or an error pointer in
+ * case of failure.
+ */
+struct inode *apfs_new_inode(struct inode *dir, umode_t mode, dev_t rdev)
+{
+	struct super_block *sb = dir->i_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct inode *inode;
+	struct apfs_inode_info *ai;
+	struct apfs_dstream_info *dstream;
+	u64 cnid;
+	struct timespec64 now;
+
+	/* Updating on-disk structures here is odd, but it works for now */
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+
+	inode = new_inode(sb);
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+	ai = APFS_I(inode);
+	dstream = &ai->i_dstream;
+
+	cnid = le64_to_cpu(vsb_raw->apfs_next_obj_id);
+	le64_add_cpu(&vsb_raw->apfs_next_obj_id, 1);
+	apfs_set_ino(inode, cnid);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+	inode_init_owner(inode, dir, mode);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+	inode_init_owner(&init_user_ns, inode, dir, mode);
+#else
+	inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
+#endif
+
+	ai->i_saved_uid = i_uid_read(inode);
+	ai->i_saved_gid = i_gid_read(inode);
+	ai->i_parent_id = apfs_ino(dir);
+	set_nlink(inode, 1);
+	ai->i_nchildren = 0;
+	if (apfs_vol_is_encrypted(sb) && S_ISREG(mode))
+		ai->i_key_class = apfs_dflt_key_class(sb);
+	else
+		ai->i_key_class = 0;
+	ai->i_int_flags = APFS_INODE_NO_RSRC_FORK;
+	ai->i_bsd_flags = 0;
+
+	ai->i_has_dstream = false;
+	dstream->ds_id = cnid;
+	dstream->ds_size = 0;
+	dstream->ds_sparse_bytes = 0;
+	dstream->ds_shared = false;
+
+	now = current_time(inode);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode->i_atime = inode->i_mtime = inode->i_ctime = ai->i_crtime = now;
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	inode_set_ctime_to_ts(inode, now);
+	inode->i_atime = inode->i_mtime = ai->i_crtime = now;
+#else
+	ai->i_crtime = simple_inode_init_ts(inode);
+#endif
+	vsb_raw->apfs_last_mod_time = cpu_to_le64(timespec64_to_ns(&now));
+
+	if (S_ISREG(mode))
+		le64_add_cpu(&vsb_raw->apfs_num_files, 1);
+	else if (S_ISDIR(mode))
+		le64_add_cpu(&vsb_raw->apfs_num_directories, 1);
+	else if (S_ISLNK(mode))
+		le64_add_cpu(&vsb_raw->apfs_num_symlinks, 1);
+	else
+		le64_add_cpu(&vsb_raw->apfs_num_other_fsobjects, 1);
+
+	if (apfs_insert_inode_locked(inode)) {
+		/* The inode number should have been free, but wasn't */
+		apfs_err(sb, "next obj_id (0x%llx) not free", cnid);
+		make_bad_inode(inode);
+		iput(inode);
+		return ERR_PTR(-EFSCORRUPTED);
+	}
+
+	/* No need to dirty the inode, we'll write it to disk right away */
+	apfs_inode_set_ops(inode, rdev, false /* compressed */);
+	return inode;
+}
+
+/**
+ * apfs_create_inode_rec - Create an inode record in the catalog b-tree
+ * @sb:		filesystem superblock
+ * @inode:	vfs inode to record
+ * @dentry:	dentry for primary link
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_create_inode_rec(struct super_block *sb, struct inode *inode,
+			  struct dentry *dentry)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_inode_key raw_key;
+	struct apfs_inode_val *raw_val;
+	int val_len;
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_inode_key(apfs_ino(inode), &query->key);
+	query->flags |= APFS_QUERY_CAT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret && ret != -ENODATA) {
+		apfs_err(sb, "query failed for ino 0x%llx", apfs_ino(inode));
+		goto fail;
+	}
+
+	apfs_key_set_hdr(APFS_TYPE_INODE, apfs_ino(inode), &raw_key);
+
+	val_len = apfs_build_inode_val(inode, &dentry->d_name, &raw_val);
+	if (val_len < 0) {
+		ret = val_len;
+		goto fail;
+	}
+
+	ret = apfs_btree_insert(query, &raw_key, sizeof(raw_key), raw_val, val_len);
+	if (ret)
+		apfs_err(sb, "insertion failed for ino 0x%llx", apfs_ino(inode));
+	kfree(raw_val);
+
+fail:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_setsize - Change the size of a regular file
+ * @inode:	the vfs inode
+ * @new_size:	the new size
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_setsize(struct inode *inode, loff_t new_size)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	int err;
+
+	if (new_size == inode->i_size)
+		return 0;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode->i_mtime = inode->i_ctime = current_time(inode);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 7, 0)
+	inode->i_mtime = inode_set_ctime_current(inode);
+#else
+	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
+#endif
+
+	err = apfs_inode_create_dstream_rec(inode);
+	if (err) {
+		apfs_err(sb, "failed to create dstream for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	/* Must be called before i_size is changed */
+	err = apfs_truncate(dstream, new_size);
+	if (err) {
+		apfs_err(sb, "truncation failed for ino 0x%llx", apfs_ino(inode));
+		return err;
+	}
+
+	truncate_setsize(inode, new_size);
+	dstream->ds_size = i_size_read(inode);
+	return 0;
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+int apfs_setattr(struct dentry *dentry, struct iattr *iattr)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+int apfs_setattr(struct user_namespace *mnt_userns,
+		 struct dentry *dentry, struct iattr *iattr)
+#else
+int apfs_setattr(struct mnt_idmap *idmap,
+		 struct dentry *dentry, struct iattr *iattr)
+#endif
+{
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = inode->i_sb;
+	bool resizing = S_ISREG(inode->i_mode) && (iattr->ia_valid & ATTR_SIZE);
+	bool shrinking = false;
+	int err;
+
+	if (resizing && iattr->ia_size > APFS_MAX_FILE_SIZE)
+		return -EFBIG;
+	if (resizing && iattr->ia_size < inode->i_size)
+		shrinking = true;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+	err = setattr_prepare(dentry, iattr);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+	err = setattr_prepare(&init_user_ns, dentry, iattr);
+#else
+	err = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
+#endif
+	if (err)
+		return err;
+
+	/* TODO: figure out why ->write_inode() isn't firing */
+	err = apfs_transaction_start(sb, shrinking ? APFS_TRANS_DEL : APFS_TRANS_REG);
+	if (err)
+		return err;
+	apfs_inode_join_transaction(sb, inode);
+
+	if (resizing) {
+		err = apfs_setsize(inode, iattr->ia_size);
+		if (err) {
+			apfs_err(sb, "setsize failed for ino 0x%llx", apfs_ino(inode));
+			goto fail;
+		}
+	}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+	setattr_copy(inode, iattr);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+	setattr_copy(&init_user_ns, inode, iattr);
+#else
+	setattr_copy(&nop_mnt_idmap, inode, iattr);
+#endif
+
+	mark_inode_dirty(inode);
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/* TODO: this only seems to be necessary because ->write_inode() isn't firing */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+int apfs_update_time(struct inode *inode, enum fs_update_time time, unsigned int flags)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+int apfs_update_time(struct inode *inode, struct timespec64 *time, int flags)
+#else
+int apfs_update_time(struct inode *inode, int flags)
+#endif
+{
+	struct super_block *sb = inode->i_sb;
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+	apfs_inode_join_transaction(sb, inode);
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	generic_update_time(inode, time, flags);
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0) && !RHEL_VERSION_GE(9, 6)
+	generic_update_time(inode, time, flags);
+#else
+	generic_update_time(inode, flags);
+#endif
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+static int apfs_ioc_set_dflt_pfk(struct file *file, void __user *user_pfk)
+{
+	struct inode *inode = file_inode(file);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_wrapped_crypto_state pfk_hdr;
+	struct apfs_crypto_state_val *pfk;
+	unsigned int key_len;
+
+	if (__copy_from_user(&pfk_hdr, user_pfk, sizeof(pfk_hdr)))
+		return -EFAULT;
+	key_len = le16_to_cpu(pfk_hdr.key_len);
+	if (key_len > MAX_PFK_LEN)
+		return -EFBIG;
+	pfk = kmalloc(sizeof(*pfk) + key_len, GFP_KERNEL);
+	if (!pfk)
+		return -ENOMEM;
+	if (__copy_from_user(&pfk->state, user_pfk, sizeof(pfk_hdr) + key_len)) {
+		kfree(pfk);
+		return -EFAULT;
+	}
+	pfk->refcnt = cpu_to_le32(1);
+
+	down_write(&nxi->nx_big_sem);
+
+	if (sbi->s_dflt_pfk)
+		kfree(sbi->s_dflt_pfk);
+	sbi->s_dflt_pfk = pfk;
+
+	up_write(&nxi->nx_big_sem);
+
+	return 0;
+}
+
+static int apfs_ioc_set_dir_class(struct file *file, u32 __user *user_class)
+{
+	struct inode *inode = file_inode(file);
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct super_block *sb = inode->i_sb;
+	u32 class;
+	int err;
+
+	if (get_user(class, user_class))
+		return -EFAULT;
+
+	ai->i_key_class = class;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+	apfs_inode_join_transaction(sb, inode);
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+static int apfs_ioc_set_pfk(struct file *file, void __user *user_pfk)
+{
+	struct inode *inode = file_inode(file);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_wrapped_crypto_state pfk_hdr;
+	struct apfs_crypto_state_val *pfk;
+	struct apfs_inode_info *ai = APFS_I(inode);
+	struct apfs_dstream_info *dstream = &ai->i_dstream;
+	unsigned int key_len, key_class;
+	int err;
+
+	if (__copy_from_user(&pfk_hdr, user_pfk, sizeof(pfk_hdr)))
+		return -EFAULT;
+	key_len = le16_to_cpu(pfk_hdr.key_len);
+	if (key_len > MAX_PFK_LEN)
+		return -EFBIG;
+	pfk = kmalloc(sizeof(*pfk) + key_len, GFP_KERNEL);
+	if (!pfk)
+		return -ENOMEM;
+	if (__copy_from_user(&pfk->state, user_pfk, sizeof(pfk_hdr) + key_len)) {
+		kfree(pfk);
+		return -EFAULT;
+	}
+	pfk->refcnt = cpu_to_le32(1);
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err) {
+		kfree(pfk);
+		return err;
+	}
+
+	err = apfs_crypto_set_key(sb, dstream->ds_id, pfk);
+	if (err)
+		goto fail;
+
+	key_class = le32_to_cpu(pfk_hdr.persistent_class);
+	if (ai->i_key_class != key_class) {
+		ai->i_key_class = key_class;
+		apfs_inode_join_transaction(sb, inode);
+	}
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	kfree(pfk);
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	kfree(pfk);
+	return err;
+}
+
+static int apfs_ioc_get_class(struct file *file, u32 __user *user_class)
+{
+	struct inode *inode = file_inode(file);
+	struct apfs_inode_info *ai = APFS_I(inode);
+	u32 class;
+
+	class = ai->i_key_class;
+	if (put_user(class, user_class))
+		return -EFAULT;
+	return 0;
+}
+
+static int apfs_ioc_get_pfk(struct file *file, void __user *user_pfk)
+{
+	struct inode *inode = file_inode(file);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_wrapped_crypto_state pfk_hdr;
+	struct apfs_crypto_state_val *pfk;
+	unsigned int max_len, key_len;
+	struct apfs_dstream_info *dstream = &APFS_I(inode)->i_dstream;
+	int err;
+
+	if (__copy_from_user(&pfk_hdr, user_pfk, sizeof(pfk_hdr)))
+		return -EFAULT;
+	max_len = le16_to_cpu(pfk_hdr.key_len);
+	if (max_len > MAX_PFK_LEN)
+		return -EFBIG;
+	pfk = kmalloc(sizeof(*pfk) + max_len, GFP_KERNEL);
+	if (!pfk)
+		return -ENOMEM;
+
+	down_read(&nxi->nx_big_sem);
+
+	err = apfs_crypto_get_key(sb, dstream->ds_id, pfk, max_len);
+	if (err)
+		goto fail;
+
+	up_read(&nxi->nx_big_sem);
+
+	key_len = le16_to_cpu(pfk->state.key_len);
+	if (__copy_to_user(user_pfk, &pfk->state, sizeof(pfk_hdr) + key_len)) {
+		kfree(pfk);
+		return -EFAULT;
+	}
+
+	kfree(pfk);
+	return 0;
+
+fail:
+	up_read(&nxi->nx_big_sem);
+	kfree(pfk);
+	return err;
+}
+
+/*
+ * Older kernels have no vfs_ioc_setflags_prepare(), so don't implement the
+ * SETFLAGS/GETFLAGS ioctls there. It should be easy to fix, but it's not
+ * really needed at all. Be careful with this macro check, because it nests
+ * over a few others.
+ */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 3, 0)
+
+/**
+ * apfs_getflags - Read an inode's bsd flags in FS_IOC_GETFLAGS format
+ * @inode: the vfs inode
+ */
+static unsigned int apfs_getflags(struct inode *inode)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+	unsigned int flags = 0;
+
+	if (ai->i_bsd_flags & APFS_INOBSD_APPEND)
+		flags |= FS_APPEND_FL;
+	if (ai->i_bsd_flags & APFS_INOBSD_IMMUTABLE)
+		flags |= FS_IMMUTABLE_FL;
+	if (ai->i_bsd_flags & APFS_INOBSD_NODUMP)
+		flags |= FS_NODUMP_FL;
+	return flags;
+}
+
+/**
+ * apfs_setflags - Set an inode's bsd flags
+ * @inode: the vfs inode
+ * @flags: flags to set, in FS_IOC_SETFLAGS format
+ */
+static void apfs_setflags(struct inode *inode, unsigned int flags)
+{
+	struct apfs_inode_info *ai = APFS_I(inode);
+	unsigned int i_flags = 0;
+
+	if (flags & FS_APPEND_FL) {
+		ai->i_bsd_flags |= APFS_INOBSD_APPEND;
+		i_flags |= S_APPEND;
+	} else {
+		ai->i_bsd_flags &= ~APFS_INOBSD_APPEND;
+	}
+
+	if (flags & FS_IMMUTABLE_FL) {
+		ai->i_bsd_flags |= APFS_INOBSD_IMMUTABLE;
+		i_flags |= S_IMMUTABLE;
+	} else {
+		ai->i_bsd_flags &= ~APFS_INOBSD_IMMUTABLE;
+	}
+
+	if (flags & FS_NODUMP_FL)
+		ai->i_bsd_flags |= APFS_INOBSD_NODUMP;
+	else
+		ai->i_bsd_flags &= ~APFS_INOBSD_NODUMP;
+
+	inode_set_flags(inode, i_flags, S_IMMUTABLE | S_APPEND);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 13, 0)
+
+/**
+ * apfs_ioc_getflags - Ioctl handler for FS_IOC_GETFLAGS
+ * @file:	affected file
+ * @arg:	ioctl argument
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_ioc_getflags(struct file *file, int __user *arg)
+{
+	unsigned int flags = apfs_getflags(file_inode(file));
+
+	return put_user(flags, arg);
+}
+
+/**
+ * apfs_do_ioc_setflags - Actual work for apfs_ioc_setflags(), after preparation
+ * @inode:	affected vfs inode
+ * @newflags:	inode flags to set, in FS_IOC_SETFLAGS format
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_do_ioc_setflags(struct inode *inode, unsigned int newflags)
+{
+	struct super_block *sb = inode->i_sb;
+	unsigned int oldflags;
+	int err;
+
+	lockdep_assert_held_write(&inode->i_rwsem);
+
+	oldflags = apfs_getflags(inode);
+	err = vfs_ioc_setflags_prepare(inode, oldflags, newflags);
+	if (err)
+		return err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	apfs_inode_join_transaction(sb, inode);
+	apfs_setflags(inode, newflags);
+	inode->i_ctime = current_time(inode);
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_ioc_setflags - Ioctl handler for FS_IOC_SETFLAGS
+ * @file:	affected file
+ * @arg:	ioctl argument
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_ioc_setflags(struct file *file, int __user *arg)
+{
+	struct inode *inode = file_inode(file);
+	struct super_block *sb = inode->i_sb;
+	unsigned int newflags;
+	int err;
+
+	if (sb->s_flags & SB_RDONLY)
+		return -EROFS;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+	if (!inode_owner_or_capable(inode))
+#else
+	if (!inode_owner_or_capable(&init_user_ns, inode))
+#endif
+		return -EPERM;
+
+	if (get_user(newflags, arg))
+		return -EFAULT;
+
+	if (newflags & ~(FS_APPEND_FL | FS_IMMUTABLE_FL | FS_NODUMP_FL))
+		return -EOPNOTSUPP;
+
+	err = mnt_want_write_file(file);
+	if (err)
+		return err;
+
+	inode_lock(inode);
+	err = apfs_do_ioc_setflags(inode, newflags);
+	inode_unlock(inode);
+
+	mnt_drop_write_file(file);
+	return err;
+}
+
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+
+int apfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
+{
+	unsigned int flags = apfs_getflags(d_inode(dentry));
+
+	fileattr_fill_flags(fa, flags);
+	return 0;
+}
+
+int apfs_fileattr_set(struct user_namespace *mnt_userns, struct dentry *dentry, struct fileattr *fa)
+{
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = inode->i_sb;
+	int err;
+
+	if (sb->s_flags & SB_RDONLY)
+		return -EROFS;
+
+	if (fa->flags & ~(FS_APPEND_FL | FS_IMMUTABLE_FL | FS_NODUMP_FL))
+		return -EOPNOTSUPP;
+	if (fileattr_has_fsx(fa))
+		return -EOPNOTSUPP;
+
+	lockdep_assert_held_write(&inode->i_rwsem);
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	apfs_inode_join_transaction(sb, inode);
+	apfs_setflags(inode, fa->flags);
+	inode->i_ctime = current_time(inode);
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		apfs_transaction_abort(sb);
+	return err;
+}
+
+#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(6, 3, 0) */
+
+int apfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
+{
+	unsigned int flags = apfs_getflags(d_inode(dentry));
+
+	fileattr_fill_flags(fa, flags);
+	return 0;
+}
+
+int apfs_fileattr_set(struct mnt_idmap *idmap, struct dentry *dentry, struct fileattr *fa)
+{
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = inode->i_sb;
+	int err;
+
+	if (sb->s_flags & SB_RDONLY)
+		return -EROFS;
+
+	if (fa->flags & ~(FS_APPEND_FL | FS_IMMUTABLE_FL | FS_NODUMP_FL))
+		return -EOPNOTSUPP;
+	if (fileattr_has_fsx(fa))
+		return -EOPNOTSUPP;
+
+	lockdep_assert_held_write(&inode->i_rwsem);
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	apfs_inode_join_transaction(sb, inode);
+	apfs_setflags(inode, fa->flags);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 6, 0)
+	inode->i_ctime = current_time(inode);
+#else
+	inode_set_ctime_current(inode);
+#endif
+
+	err = apfs_transaction_commit(sb);
+	if (err)
+		apfs_transaction_abort(sb);
+	return err;
+}
+
+#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(5, 13, 0) */
+
+#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(5, 3, 0) */
+
+long apfs_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	void __user *argp = (void __user *)arg;
+
+	switch (cmd) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 13, 0) && LINUX_VERSION_CODE >= KERNEL_VERSION(5, 3, 0)
+	case FS_IOC_GETFLAGS:
+		return apfs_ioc_getflags(file, argp);
+	case FS_IOC_SETFLAGS:
+		return apfs_ioc_setflags(file, argp);
+#endif
+	case APFS_IOC_SET_DFLT_PFK:
+		return apfs_ioc_set_dflt_pfk(file, argp);
+	case APFS_IOC_SET_DIR_CLASS:
+		return apfs_ioc_set_dir_class(file, argp);
+	case APFS_IOC_GET_CLASS:
+		return apfs_ioc_get_class(file, argp);
+	case APFS_IOC_TAKE_SNAPSHOT:
+		return apfs_ioc_take_snapshot(file, argp);
+	default:
+		return -ENOTTY;
+	}
+}
+
+long apfs_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	void __user *argp = (void __user *)arg;
+
+	switch (cmd) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 13, 0) && LINUX_VERSION_CODE >= KERNEL_VERSION(5, 3, 0)
+	case FS_IOC_GETFLAGS:
+		return apfs_ioc_getflags(file, argp);
+	case FS_IOC_SETFLAGS:
+		return apfs_ioc_setflags(file, argp);
+#endif
+	case APFS_IOC_SET_PFK:
+		return apfs_ioc_set_pfk(file, argp);
+	case APFS_IOC_GET_CLASS:
+		return apfs_ioc_get_class(file, argp);
+	case APFS_IOC_GET_PFK:
+		return apfs_ioc_get_pfk(file, argp);
+	default:
+		return -ENOTTY;
+	}
+}
diff --git a/fs/apfs/key.c b/fs/apfs/key.c
new file mode 100644
index 000000000..96cfa4a6c
--- /dev/null
+++ b/fs/apfs/key.c
@@ -0,0 +1,337 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/crc32c.h>
+#include "apfs.h"
+#include "unicode.h"
+
+/**
+ * apfs_filename_cmp - Normalize and compare two APFS filenames
+ * @sb:		filesystem superblock
+ * @name1:	first name to compare
+ * @len1:	length of @name1
+ * @name2:	second name to compare
+ * @len2:	length of the @name2
+ *
+ * Returns 0 if @name1 and @name2 are equal, or non-zero otherwise.
+ */
+int apfs_filename_cmp(struct super_block *sb,
+		      const char *name1, unsigned int len1,
+		      const char *name2, unsigned int len2)
+{
+	struct apfs_unicursor cursor1, cursor2;
+	bool case_fold = apfs_is_case_insensitive(sb);
+
+	if (!apfs_is_normalization_insensitive(sb)) {
+		if (len1 != len2)
+			return -1;
+		return memcmp(name1, name2, len1);
+	}
+
+	apfs_init_unicursor(&cursor1, name1, len1);
+	apfs_init_unicursor(&cursor2, name2, len2);
+
+	while (1) {
+		unicode_t uni1, uni2;
+
+		uni1 = apfs_normalize_next(&cursor1, case_fold);
+		uni2 = apfs_normalize_next(&cursor2, case_fold);
+
+		if (uni1 != uni2)
+			return uni1 < uni2 ? -1 : 1;
+		if (!uni1)
+			return 0;
+	}
+}
+
+/**
+ * apfs_keycmp - Compare two keys
+ * @k1:	first key to compare
+ * @k2:	second key to compare
+ *
+ * returns   0 if @k1 and @k2 are equal
+ *	   < 0 if @k1 comes before @k2 in the btree
+ *	   > 0 if @k1 comes after @k2 in the btree
+ */
+int apfs_keycmp(struct apfs_key *k1, struct apfs_key *k2)
+{
+	if (k1->id != k2->id)
+		return k1->id < k2->id ? -1 : 1;
+	if (k1->type != k2->type)
+		return k1->type < k2->type ? -1 : 1;
+	if (k1->number != k2->number)
+		return k1->number < k2->number ? -1 : 1;
+	if (!k1->name || !k2->name)
+		return 0;
+
+	/* Normalization seems to be ignored here, even for directory records */
+	return strcmp(k1->name, k2->name);
+}
+
+/**
+ * apfs_read_cat_key - Parse an on-disk catalog key
+ * @raw:	pointer to the raw key
+ * @size:	size of the raw key
+ * @key:	apfs_key structure to store the result
+ * @hashed:	are the directory records hashed?
+ *
+ * Returns 0 on success, or a negative error code otherwise.
+ */
+int apfs_read_cat_key(void *raw, int size, struct apfs_key *key, bool hashed)
+{
+	if (size < sizeof(struct apfs_key_header)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	key->id = apfs_cat_cnid((struct apfs_key_header *)raw);
+	key->type = apfs_cat_type((struct apfs_key_header *)raw);
+
+	switch (key->type) {
+	case APFS_TYPE_DIR_REC:
+		if (hashed) {
+			if (size < sizeof(struct apfs_drec_hashed_key) + 1 ||
+			    *((char *)raw + size - 1) != 0) {
+				/* Filename must have NULL-termination */
+				apfs_err(NULL, "invalid drec key (%d)", size);
+				return -EFSCORRUPTED;
+			}
+			/* Name length is not used in key comparisons, only the hash */
+			key->number = le32_to_cpu(
+			      ((struct apfs_drec_hashed_key *)raw)->name_len_and_hash) &
+								    APFS_DREC_HASH_MASK;
+			key->name = ((struct apfs_drec_hashed_key *)raw)->name;
+		} else {
+			if (size < sizeof(struct apfs_drec_key) + 1 ||
+			    *((char *)raw + size - 1) != 0) {
+				/* Filename must have NULL-termination */
+				apfs_err(NULL, "invalid drec key (%d)", size);
+				return -EFSCORRUPTED;
+			}
+			/* There's no hash */
+			key->number = 0;
+			key->name = ((struct apfs_drec_key *)raw)->name;
+		}
+		break;
+	case APFS_TYPE_XATTR:
+		if (size < sizeof(struct apfs_xattr_key) + 1 ||
+		    *((char *)raw + size - 1) != 0) {
+			/* xattr name must have NULL-termination */
+			apfs_err(NULL, "invalid xattr key (%d)", size);
+			return -EFSCORRUPTED;
+		}
+		key->number = 0;
+		key->name = ((struct apfs_xattr_key *)raw)->name;
+		break;
+	case APFS_TYPE_FILE_EXTENT:
+		if (size != sizeof(struct apfs_file_extent_key)) {
+			apfs_err(NULL, "bad key length (%d)", size);
+			return -EFSCORRUPTED;
+		}
+		key->number = le64_to_cpu(
+			((struct apfs_file_extent_key *)raw)->logical_addr);
+		key->name = NULL;
+		break;
+	case APFS_TYPE_SIBLING_LINK:
+		if (size != sizeof(struct apfs_sibling_link_key)) {
+			apfs_err(NULL, "bad key length (%d)", size);
+			return -EFSCORRUPTED;
+		}
+		key->number = le64_to_cpu(
+			((struct apfs_sibling_link_key *)raw)->sibling_id);
+		key->name = NULL;
+		break;
+	default:
+		key->number = 0;
+		key->name = NULL;
+		break;
+	}
+
+	return 0;
+}
+
+int apfs_read_fext_key(void *raw, int size, struct apfs_key *key)
+{
+	struct apfs_fext_tree_key *raw_key;
+
+	if (size != sizeof(*raw_key)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	raw_key = raw;
+
+	key->id = le64_to_cpu(raw_key->private_id);
+	key->type = 0;
+	key->number = le64_to_cpu(raw_key->logical_addr);
+	key->name = NULL;
+	return 0;
+}
+
+/**
+ * apfs_read_free_queue_key - Parse an on-disk free queue key
+ * @raw:	pointer to the raw key
+ * @size:	size of the raw key
+ * @key:	apfs_key structure to store the result
+ *
+ * Returns 0 on success, or a negative error code otherwise.
+ */
+int apfs_read_free_queue_key(void *raw, int size, struct apfs_key *key)
+{
+	struct apfs_spaceman_free_queue_key *raw_key;
+
+	if (size != sizeof(*raw_key)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	raw_key = raw;
+
+	key->id = le64_to_cpu(raw_key->sfqk_xid);
+	key->type = 0;
+	key->number = le64_to_cpu(raw_key->sfqk_paddr);
+	key->name = NULL;
+
+	return 0;
+}
+
+/**
+ * apfs_read_omap_key - Parse an on-disk object map key
+ * @raw:	pointer to the raw key
+ * @size:	size of the raw key
+ * @key:	apfs_key structure to store the result
+ *
+ * Returns 0 on success, or a negative error code otherwise.
+ */
+int apfs_read_omap_key(void *raw, int size, struct apfs_key *key)
+{
+	struct apfs_omap_key *raw_key;
+
+	if (size != sizeof(*raw_key)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	raw_key = raw;
+
+	key->id = le64_to_cpu(raw_key->ok_oid);
+	key->type = 0;
+	key->number = le64_to_cpu(raw_key->ok_xid);
+	key->name = NULL;
+
+	return 0;
+}
+
+/**
+ * apfs_read_extentref_key - Parse an on-disk extent reference tree key
+ * @raw:	pointer to the raw key
+ * @size:	size of the raw key
+ * @key:	apfs_key structure to store the result
+ *
+ * Returns 0 on success, or a negative error code otherwise.
+ */
+int apfs_read_extentref_key(void *raw, int size, struct apfs_key *key)
+{
+	if (size != sizeof(struct apfs_phys_ext_key)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	key->id = apfs_cat_cnid((struct apfs_key_header *)raw);
+	key->type = apfs_cat_type((struct apfs_key_header *)raw);
+	key->number = 0;
+	key->name = NULL;
+	return 0;
+}
+
+int apfs_read_snap_meta_key(void *raw, int size, struct apfs_key *key)
+{
+	if (size < sizeof(struct apfs_key_header)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	key->id = apfs_cat_cnid((struct apfs_key_header *)raw);
+	key->type = apfs_cat_type((struct apfs_key_header *)raw);
+	key->number = 0;
+
+	switch (key->type) {
+	case APFS_TYPE_SNAP_METADATA:
+		if (size != sizeof(struct apfs_snap_metadata_key)) {
+			apfs_err(NULL, "bad key length (%d)", size);
+			return -EFSCORRUPTED;
+		}
+		key->name = NULL;
+		return 0;
+	case APFS_TYPE_SNAP_NAME:
+		if (size < sizeof(struct apfs_snap_name_key) + 1 || *((char *)raw + size - 1) != 0) {
+			/* snapshot name must have NULL-termination */
+			apfs_err(NULL, "invalid snap name key (%d)", size);
+			return -EFSCORRUPTED;
+		}
+		key->name = ((struct apfs_snap_name_key *)raw)->name;
+		return 0;
+	default:
+		return -EFSCORRUPTED;
+	}
+}
+
+int apfs_read_omap_snap_key(void *raw, int size, struct apfs_key *key)
+{
+	__le64 *xid = NULL;
+
+	if (size != sizeof(*xid)) {
+		apfs_err(NULL, "bad key length (%d)", size);
+		return -EFSCORRUPTED;
+	}
+	xid = raw;
+
+	key->id = le64_to_cpup(xid);
+	key->number = 0;
+	key->name = NULL;
+	key->type = 0;
+	return 0;
+}
+
+/**
+ * apfs_init_drec_key - Initialize an in-memory key for a dentry query
+ * @sb:		filesystem superblock
+ * @ino:	inode number of the parent directory
+ * @name:	filename (NULL for a multiple query)
+ * @name_len:	filename length (0 if NULL)
+ * @key:	apfs_key structure to initialize
+ */
+void apfs_init_drec_key(struct super_block *sb, u64 ino, const char *name,
+			unsigned int name_len, struct apfs_key *key)
+{
+	struct apfs_unicursor cursor;
+	bool case_fold = apfs_is_case_insensitive(sb);
+	u32 hash = 0xFFFFFFFF;
+
+	key->id = ino;
+	key->type = APFS_TYPE_DIR_REC;
+	if (!apfs_is_normalization_insensitive(sb)) {
+		key->name = name;
+		key->number = 0;
+		return;
+	}
+
+	/* To respect normalization, queries can only consider the hash */
+	key->name = NULL;
+
+	if (!name) {
+		key->number = 0;
+		return;
+	}
+
+	apfs_init_unicursor(&cursor, name, name_len);
+
+	while (1) {
+		unicode_t utf32;
+
+		utf32 = apfs_normalize_next(&cursor, case_fold);
+		if (!utf32)
+			break;
+
+		hash = crc32c(hash, &utf32, sizeof(utf32));
+	}
+
+	/* The filename length doesn't matter, so it's left as zero */
+	key->number = hash << APFS_DREC_HASH_SHIFT;
+}
diff --git a/fs/apfs/libzbitmap.c b/fs/apfs/libzbitmap.c
new file mode 100644
index 000000000..a46cfdbde
--- /dev/null
+++ b/fs/apfs/libzbitmap.c
@@ -0,0 +1,442 @@
+// SPDX-License-Identifier: GPL-2.0+ OR MIT
+/*
+ * Copyright (C) 2022 Corellium LLC
+ *
+ * Author: Ernesto A. Fernández <ernesto@corellium.com>
+ *
+ * Ported from libzbitmap (https://github.com/eafer/libzbitmap). Only the
+ * decompression code is included.
+ */
+
+#include <linux/errno.h>
+#include <linux/string.h>
+#include "libzbitmap.h"
+
+#define ZBM_MAGIC       "ZBM\x09"
+#define ZBM_MAGIC_SZ    4
+
+#define ZBM_MAX_DECMP_CHUNK_SIZE        0x8000
+#define ZBM_MAX_DECMP_CHUNK_SIZE_BITS   15
+
+struct uint24 {
+    uint8_t     low;
+    uint8_t     mid;
+    uint8_t     hig;
+};
+
+/* This header is shared by both compressed and decompressed chunks */
+struct zbm_chunk_hdr {
+    struct uint24   len;        /* Length of the chunk */
+    struct uint24   decmp_len;  /* Length of the chunk after decompression */
+};
+
+/* The full header for compressed chunks */
+struct zbm_cmp_chunk_hdr {
+    /* Shared with decompressed chunks */
+    struct zbm_chunk_hdr hdr;
+
+    /* Offset for each of the three metadata areas */
+    struct uint24   meta_off_1;
+    struct uint24   meta_off_2;
+    struct uint24   meta_off_3;
+};
+
+/* Pointer to a half-byte */
+struct nybl_ptr {
+    uint8_t         *addr;  /* Address of the byte */
+    int             nibble; /* Which of the two nibbles? */
+};
+
+/* 0-2 and 0xf are not real bitmap indexes */
+#define ZBM_BITMAP_COUNT        (16 - 1 - 3)
+#define ZBM_BITMAP_BASE         3
+#define ZBM_BITMAP_BYTECNT      17
+#define ZBM_MAX_PERIOD_BYTECNT  2
+
+struct zbm_bmap {
+    uint8_t     bitmap;         /* The bitmap */
+    uint8_t     period_bytecnt; /* Read this many bytes to get the new period */
+};
+
+struct zbm_state {
+    /* Updated during a chunk read */
+    uint8_t         *dest;      /* Write the next byte here */
+    size_t          dest_left;  /* Room left in destination buffer */
+    uint32_t        written;    /* Bytes written so far for current chunk */
+    uint16_t        period;     /* Repetition period for decompression, in bytes */
+
+    /* Updated right before a chunk read */
+    const uint8_t   *src_end;   /* End of current chunk */
+    uint32_t        len;        /* Length of the chunk */
+    uint32_t        decmp_len;  /* Expected chunk length after decompression */
+
+    /* Updated after a chunk read */
+    const uint8_t   *src;       /* Start of buffer, or current chunk if any */
+    size_t          src_left;   /* Room left in the source buffer */
+    size_t          prewritten; /* Bytes written for previous chunks */
+
+    /* Current position in data and metadata areas for this chunk */
+    const uint8_t   *data;
+    const uint8_t   *meta_1;
+    const uint8_t   *meta_2;
+    struct nybl_ptr meta_3;
+
+    /* Array of bitmaps for the current chunk */
+    struct zbm_bmap bitmaps[ZBM_BITMAP_COUNT];
+};
+
+static int zbm_check_magic(struct zbm_state *state)
+{
+    if(state->src_left < ZBM_MAGIC_SZ)
+        return -EINVAL;
+
+    if(memcmp(state->src, ZBM_MAGIC, ZBM_MAGIC_SZ))
+        return -EINVAL;
+
+    state->src += ZBM_MAGIC_SZ;
+    state->src_left -= ZBM_MAGIC_SZ;
+    return 0;
+}
+
+static uint32_t zbm_u24_to_u32(struct uint24 n)
+{
+    uint32_t res;
+
+    res = n.hig;
+    res <<= 8;
+    res += n.mid;
+    res <<= 8;
+    res += n.low;
+    return res;
+}
+
+/* Some chunks just have regular uncompressed data, but with a header */
+static int zbm_chunk_is_uncompressed(struct zbm_state *state)
+{
+    return state->len == state->decmp_len + sizeof(struct zbm_chunk_hdr);
+}
+
+static int zbm_handle_uncompressed_chunk(struct zbm_state *state)
+{
+    state->meta_1 = state->meta_2 = NULL;
+    state->meta_3.addr = NULL;
+    state->meta_3.nibble = 0;
+    state->data = state->src + sizeof(struct zbm_chunk_hdr);
+    memcpy(state->dest, state->data, state->decmp_len);
+
+    state->dest += state->decmp_len;
+    state->dest_left -= state->decmp_len;
+    state->written = state->decmp_len;
+    return 0;
+}
+
+static int zbm_read_nibble(struct nybl_ptr *nybl, const uint8_t *limit, uint8_t *result)
+{
+    if(nybl->addr >= limit)
+        return -EINVAL;
+
+    if(nybl->nibble == 0) {
+        *result = *nybl->addr & 0xf;
+        nybl->nibble = 1;
+    } else {
+        *result = (*nybl->addr >> 4) & 0xf;
+        nybl->nibble = 0;
+        ++nybl->addr;
+    }
+    return 0;
+}
+
+static void zbm_rewind_nibble(struct nybl_ptr *nybl)
+{
+    if(nybl->nibble == 0) {
+        nybl->nibble = 1;
+        --nybl->addr;
+    } else {
+        nybl->nibble = 0;
+    }
+}
+
+static int zbm_apply_bitmap(struct zbm_state *state, struct zbm_bmap *bitmap)
+{
+    int i;
+
+    /* The periods are stored in the first metadata area */
+    if(bitmap->period_bytecnt) {
+        state->period = 0;
+        for(i = 0; i < bitmap->period_bytecnt; ++i) {
+            if(state->meta_1 >= state->src_end)
+                return -EINVAL;
+            state->period |= *state->meta_1 << i * 8;
+            ++state->meta_1;
+        }
+    }
+    if(state->period == 0)
+        return -EINVAL;
+
+    for(i = 0; i < 8; ++i) {
+        if(state->written == state->decmp_len)
+            break;
+        if(bitmap->bitmap & 1 << i) {
+            if(state->data >= state->src_end)
+                return -EINVAL;
+            *state->dest = *state->data;
+            ++state->data;
+        } else {
+            if(state->prewritten + state->written < state->period)
+                return -EINVAL;
+            *state->dest = *(state->dest - state->period);
+        }
+        ++state->dest;
+        --state->dest_left;
+        ++state->written;
+    }
+
+    return 0;
+}
+
+static int zbm_apply_bitmap_number(struct zbm_state *state, uint8_t bmp_num)
+{
+    struct zbm_bmap next = {0};
+
+    /* Not a valid bitmap number (it signals a repetition) */
+    if(bmp_num == 0xf)
+        return -EINVAL;
+
+    /* An actual index in the bitmap array */
+    if(bmp_num > ZBM_MAX_PERIOD_BYTECNT)
+        return zbm_apply_bitmap(state, &state->bitmaps[bmp_num - ZBM_BITMAP_BASE]);
+
+    /* For < 2, use the next bitmap in the second metadata area */
+    if(state->meta_2 >= state->src_end)
+        return -EINVAL;
+    next.bitmap = *state->meta_2;
+    next.period_bytecnt = bmp_num;
+    ++state->meta_2;
+    return zbm_apply_bitmap(state, &next);
+}
+
+/* Find out how many times we need to repeat the current bitmap operation */
+static int zbm_read_repetition_count(struct zbm_state *state, uint16_t *repeat)
+{
+    uint8_t nibble;
+    uint16_t total;
+    int err;
+
+    /* Don't confuse the trailing bitmaps with a repetition count */
+    if(state->decmp_len - state->written <= 8) {
+        *repeat = 1;
+        return 0;
+    }
+
+    err = zbm_read_nibble(&state->meta_3, state->src_end, &nibble);
+    if(err)
+        return err;
+
+    if(nibble != 0xf) {
+        /* No repetition count: the previous bitmap number gets applied once */
+        zbm_rewind_nibble(&state->meta_3);
+        *repeat = 1;
+        return 0;
+    }
+
+    /*
+     * Under this scheme, repeating a bitmap number 3 times wouldn't save any
+     * space, so the repetition count starts from 4.
+     */
+    total = 4;
+    while(nibble == 0xf) {
+        err = zbm_read_nibble(&state->meta_3, state->src_end, &nibble);
+        if(err)
+            return err;
+        total += nibble;
+        if(total < nibble)
+            return -EINVAL;
+    }
+
+    *repeat = total;
+    return 0;
+}
+
+static int zbm_decompress_single_bitmap(struct zbm_state *state)
+{
+    uint8_t bmp_num;
+    uint16_t repeat;
+    int i;
+    int err;
+
+    /* The current nibble is the offset of the next bitmap to apply */
+    err = zbm_read_nibble(&state->meta_3, state->src_end, &bmp_num);
+    if(err)
+        return err;
+
+    err = zbm_read_repetition_count(state, &repeat);
+    if(err)
+        return err;
+
+    for(i = 0; i < repeat; ++i) {
+        err = zbm_apply_bitmap_number(state, bmp_num);
+        if(err)
+            return err;
+    }
+    return 0;
+}
+
+/* Pointer to a bit */
+struct bit_ptr {
+    uint8_t         *addr;  /* Address of the byte */
+    int             offset; /* Bit number */
+};
+
+/* This function does not perform boundary checks, the caller must do it */
+static int zbm_read_single_bit(struct bit_ptr *bit)
+{
+    int res = *bit->addr >> bit->offset & 1;
+
+    ++bit->offset;
+    if(bit->offset != 8)
+        return res;
+    bit->offset = 0;
+    ++bit->addr;
+    return res;
+}
+
+static int zbm_read_single_bitmap(struct bit_ptr *bit, const uint8_t *limit, struct zbm_bmap *result)
+{
+    int i;
+
+    result->bitmap = 0;
+    result->period_bytecnt = 0;
+
+    /* The bitmap itself */
+    for(i = 0; i < 8; ++i) {
+        if(bit->addr >= limit)
+            return -EINVAL;
+        result->bitmap |= zbm_read_single_bit(bit) << i;
+    }
+
+    /*
+     * The two trailing bits tell us how many bytes to read for the next
+     * repetition period
+     */
+    for(i = 0; i < 2; ++i) {
+        if(bit->addr >= limit)
+            return -EINVAL;
+        result->period_bytecnt |= zbm_read_single_bit(bit) << i;
+    }
+
+    return 0;
+}
+
+static int zbm_read_bitmaps(struct zbm_state *state)
+{
+    struct bit_ptr bmap = {0};
+    int err, i;
+
+    if(state->len < ZBM_BITMAP_BYTECNT)
+        return -EINVAL;
+
+    bmap.addr = (uint8_t *)state->src_end - ZBM_BITMAP_BYTECNT;
+    bmap.offset = 0;
+
+    for(i = 0; i < ZBM_BITMAP_COUNT; ++i) {
+        err = zbm_read_single_bitmap(&bmap, state->src_end, &state->bitmaps[i]);
+        if(err)
+            return err;
+        if(state->bitmaps[i].period_bytecnt > ZBM_MAX_PERIOD_BYTECNT)
+            return -EINVAL;
+    }
+    return 0;
+}
+
+static int zbm_handle_compressed_chunk(struct zbm_state *state)
+{
+    const struct zbm_cmp_chunk_hdr *hdr = NULL;
+    uint32_t meta_off_1, meta_off_2, meta_off_3;
+    int err;
+
+    state->written = 0;
+    state->period = 8;
+
+    if(state->len < sizeof(*hdr))
+        return -EINVAL;
+    hdr = (struct zbm_cmp_chunk_hdr *)state->src;
+    state->data = state->src + sizeof(*hdr);
+
+    meta_off_1 = zbm_u24_to_u32(hdr->meta_off_1);
+    meta_off_2 = zbm_u24_to_u32(hdr->meta_off_2);
+    meta_off_3 = zbm_u24_to_u32(hdr->meta_off_3);
+    if(meta_off_1 >= state->len || meta_off_2 >= state->len || meta_off_3 >= state->len)
+        return -EINVAL;
+    state->meta_1 = state->src + meta_off_1;
+    state->meta_2 = state->src + meta_off_2;
+    state->meta_3.addr = (uint8_t *)state->src + meta_off_3;
+    state->meta_3.nibble = 0;
+
+    err = zbm_read_bitmaps(state);
+    if(err)
+        return err;
+
+    while(state->written < state->decmp_len) {
+        err = zbm_decompress_single_bitmap(state);
+        if(err)
+            return err;
+    }
+
+    return 0;
+}
+
+static int zbm_handle_chunk(struct zbm_state *state)
+{
+    const struct zbm_chunk_hdr *decmp_hdr = NULL;
+
+    if(state->src_left < sizeof(*decmp_hdr))
+        return -EINVAL;
+    decmp_hdr = (struct zbm_chunk_hdr *)state->src;
+
+    state->len = zbm_u24_to_u32(decmp_hdr->len);
+    if(state->len > state->src_left)
+        return -EINVAL;
+    state->src_end = state->src + state->len;
+
+    state->decmp_len = zbm_u24_to_u32(decmp_hdr->decmp_len);
+    if(state->decmp_len > ZBM_MAX_DECMP_CHUNK_SIZE)
+        return -EINVAL;
+    if(!state->dest) /* We just wanted the length, so we are done */
+        return 0;
+    if(state->decmp_len > state->dest_left)
+        return -ERANGE;
+
+    if(zbm_chunk_is_uncompressed(state))
+        return zbm_handle_uncompressed_chunk(state);
+
+    return zbm_handle_compressed_chunk(state);
+}
+
+int zbm_decompress(void *dest, size_t dest_size, const void *src, size_t src_size, size_t *out_len)
+{
+    struct zbm_state state = {0};
+    int err;
+
+    state.src = src;
+    state.src_left = src_size;
+    state.dest = dest;
+    state.dest_left = dest_size;
+    state.prewritten = 0;
+
+    err = zbm_check_magic(&state);
+    if(err)
+        return err;
+
+    /* The final chunk has zero decompressed length */
+    do {
+        err = zbm_handle_chunk(&state);
+        if(err)
+            return err;
+        state.src += state.len;
+        state.src_left -= state.len;
+        state.prewritten += state.decmp_len;
+    } while(state.decmp_len != 0);
+
+    *out_len = state.prewritten;
+    return 0;
+}
diff --git a/fs/apfs/libzbitmap.h b/fs/apfs/libzbitmap.h
new file mode 100644
index 000000000..5188d00d3
--- /dev/null
+++ b/fs/apfs/libzbitmap.h
@@ -0,0 +1,31 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR MIT */
+/*
+ * Copyright (c) 2022 Corellium LLC
+ *
+ * Author: Ernesto A. Fernández <ernesto@corellium.com>
+ *
+ * Ported from libzbitmap (https://github.com/eafer/libzbitmap). Only the
+ * decompression code is included.
+ */
+
+#ifndef _LIBZBITMAP_H
+#define _LIBZBITMAP_H
+
+#include <linux/errno.h>
+#include <linux/types.h>
+
+/**
+ * zbm_decompress - Decompress an LZBITMAP buffer
+ * @dest:       destination buffer (may be NULL)
+ * @dest_size:  size of the destination buffer
+ * @src:        source buffer
+ * @src_size:   size of the source buffer
+ * @out_len:    on return, the length of the decompressed output
+ *
+ * May be called with a NULL destination buffer to retrieve the expected length
+ * of the decompressed data. Returns 0 on success, or a negative error code in
+ * case of failure.
+ */
+int zbm_decompress(void *dest, size_t dest_size, const void *src, size_t src_size, size_t *out_len);
+
+#endif /* _LIBZBITMAP_H */
diff --git a/fs/apfs/lzfse/lzfse.h b/fs/apfs/lzfse/lzfse.h
new file mode 100644
index 000000000..017e0055d
--- /dev/null
+++ b/fs/apfs/lzfse/lzfse.h
@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: BSD-3-Clause */
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef LZFSE_H
+#define LZFSE_H
+
+#include <linux/stddef.h>
+#include <linux/types.h>
+
+/*! @abstract Get the required scratch buffer size to decompress using LZFSE. */
+size_t lzfse_decode_scratch_size(void);
+
+/*! @abstract Decompress a buffer using LZFSE.
+ *
+ *  @param dst_buffer
+ *  Pointer to the first byte of the destination buffer.
+ *
+ *  @param dst_size
+ *  Size of the destination buffer in bytes.
+ *
+ *  @param src_buffer
+ *  Pointer to the first byte of the source buffer.
+ *
+ *  @param src_size
+ *  Size of the source buffer in bytes.
+ *
+ *  @param scratch_buffer
+ *  If non-NULL, a pointer to scratch space for the routine to use as workspace;
+ *  the routine may use up to lzfse_decode_scratch_size( ) bytes of workspace
+ *  during its operation, and will not perform any internal allocations. If
+ *  NULL, the routine may allocate its own memory to use during operation via
+ *  a single call to malloc( ), and will release it by calling free( ) prior
+ *  to returning. For most use, passing NULL is perfectly satisfactory, but if
+ *  you require strict control over allocation, you will want to pass an
+ *  explicit scratch buffer.
+ *
+ *  @return
+ *  The number of bytes written to the destination buffer if the input is
+ *  successfully decompressed. If there is not enough space in the destination
+ *  buffer to hold the entire expanded output, only the first dst_size bytes
+ *  will be written to the buffer and dst_size is returned. Note that this
+ *  behavior differs from that of lzfse_encode_buffer.
+ */
+size_t lzfse_decode_buffer(uint8_t *__restrict dst_buffer, size_t dst_size,
+			   const uint8_t *__restrict src_buffer, size_t src_size,
+			   void *__restrict scratch_buffer);
+
+#endif /* LZFSE_H */
diff --git a/fs/apfs/lzfse/lzfse_decode.c b/fs/apfs/lzfse/lzfse_decode.c
new file mode 100644
index 000000000..481e973b3
--- /dev/null
+++ b/fs/apfs/lzfse/lzfse_decode.c
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: BSD-3-Clause
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* LZFSE decode API */
+
+#include <linux/slab.h>
+#include "lzfse.h"
+#include "lzfse_internal.h"
+
+size_t lzfse_decode_scratch_size(void)
+{
+	return sizeof(lzfse_decoder_state);
+}
+
+static size_t lzfse_decode_buffer_with_scratch(uint8_t *__restrict dst_buffer, size_t dst_size,
+					       const uint8_t *__restrict src_buffer,
+					       size_t src_size, void *__restrict scratch_buffer)
+{
+	int status;
+	lzfse_decoder_state *s = (lzfse_decoder_state *)scratch_buffer;
+
+	memset(s, 0x00, sizeof(*s));
+
+	/* Initialize state */
+	s->src = src_buffer;
+	s->src_begin = src_buffer;
+	s->src_end = s->src + src_size;
+	s->dst = dst_buffer;
+	s->dst_begin = dst_buffer;
+	s->dst_end = dst_buffer + dst_size;
+
+	/* Decode */
+	status = lzfse_decode(s);
+	if (status == LZFSE_STATUS_DST_FULL)
+		return dst_size;
+	if (status != LZFSE_STATUS_OK)
+		return 0; /* failed */
+	return (size_t)(s->dst - dst_buffer); /* bytes written */
+}
+
+size_t lzfse_decode_buffer(uint8_t *__restrict dst_buffer, size_t dst_size,
+			   const uint8_t *__restrict src_buffer, size_t src_size,
+			   void *__restrict scratch_buffer)
+{
+	int has_malloc = 0;
+	size_t ret = 0;
+
+	/* Deal with the possible NULL pointer */
+	if (scratch_buffer == NULL) {
+		/* +1 in case scratch size could be zero */
+		scratch_buffer = kmalloc(lzfse_decode_scratch_size() + 1, GFP_KERNEL);
+		has_malloc = 1;
+	}
+	if (scratch_buffer == NULL)
+		return 0;
+	ret = lzfse_decode_buffer_with_scratch(dst_buffer, dst_size, src_buffer, src_size,
+					       scratch_buffer);
+	if (has_malloc)
+		kfree(scratch_buffer);
+	return ret;
+}
diff --git a/fs/apfs/lzfse/lzfse_decode_base.c b/fs/apfs/lzfse/lzfse_decode_base.c
new file mode 100644
index 000000000..8cab7a340
--- /dev/null
+++ b/fs/apfs/lzfse/lzfse_decode_base.c
@@ -0,0 +1,743 @@
+// SPDX-License-Identifier: BSD-3-Clause
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "lzfse_internal.h"
+#include "lzvn_decode_base.h"
+
+/*! @abstract Decode an entry value from next bits of stream.
+ *  Return \p value, and set \p *nbits to the number of bits to consume
+ *  (starting with LSB).
+ */
+static inline int lzfse_decode_v1_freq_value(uint32_t bits, int *nbits)
+{
+	static const int8_t lzfse_freq_nbits_table[32] = { 2, 3, 2, 5, 2,  3, 2, 8, 2, 3, 2,
+							   5, 2, 3, 2, 14, 2, 3, 2, 5, 2, 3,
+							   2, 8, 2, 3, 2,  5, 2, 3, 2, 14 };
+	static const int8_t lzfse_freq_value_table[32] = { 0, 2,  1, 4, 0,  3, 1, -1, 0, 2, 1,
+							   5, 0,  3, 1, -1, 0, 2, 1,  6, 0, 3,
+							   1, -1, 0, 2, 1,  7, 0, 3,  1, -1 };
+
+	uint32_t b = bits & 31; /* lower 5 bits */
+	int n = lzfse_freq_nbits_table[b];
+	*nbits = n;
+
+	/* Special cases for > 5 bits encoding */
+	if (n == 8)
+		return 8 + ((bits >> 4) & 0xf);
+	if (n == 14)
+		return 24 + ((bits >> 4) & 0x3ff);
+
+	// <= 5 bits encoding from table
+	return lzfse_freq_value_table[b];
+}
+
+/*! @abstract Extracts up to 32 bits from a 64-bit field beginning at
+ *  \p offset, and zero-extends them to a \p uint32_t.
+ *
+ *  If we number the bits of \p v from 0 (least significant) to 63 (most
+ *  significant), the result is bits \p offset to \p offset+nbits-1.
+ */
+static inline uint32_t get_field(uint64_t v, int offset, int nbits)
+{
+	if (nbits == 32)
+		return (uint32_t)(v >> offset);
+	return (uint32_t)((v >> offset) & ((1 << nbits) - 1));
+}
+
+/*! @abstract Return \c header_size field from a \c lzfse_compressed_block_header_v2. */
+static inline uint32_t lzfse_decode_v2_header_size(const lzfse_compressed_block_header_v2 *in)
+{
+	return get_field(in->packed_fields[2], 0, 32);
+}
+
+/*! @abstract Decode all fields from a \c lzfse_compressed_block_header_v2 to a
+ * \c lzfse_compressed_block_header_v1.
+ * @return 0 on success.
+ * @return -1 on failure.
+ */
+static inline int lzfse_decode_v1(lzfse_compressed_block_header_v1 *out,
+				  const lzfse_compressed_block_header_v2 *in)
+{
+	uint64_t v0;
+	uint64_t v1;
+	uint64_t v2;
+	uint16_t *dst = NULL;
+	const uint8_t *src = NULL;
+	const uint8_t *src_end = NULL;
+	uint32_t accum = 0;
+	int accum_nbits = 0;
+	int nbits = 0;
+	int i;
+
+	/* Clear all fields */
+	memset(out, 0x00, sizeof(lzfse_compressed_block_header_v1));
+
+	v0 = in->packed_fields[0];
+	v1 = in->packed_fields[1];
+	v2 = in->packed_fields[2];
+
+	out->magic = LZFSE_COMPRESSEDV1_BLOCK_MAGIC;
+	out->n_raw_bytes = in->n_raw_bytes;
+
+	/* Literal state */
+	out->n_literals = get_field(v0, 0, 20);
+	out->n_literal_payload_bytes = get_field(v0, 20, 20);
+	out->literal_bits = (int)get_field(v0, 60, 3) - 7;
+	out->literal_state[0] = get_field(v1, 0, 10);
+	out->literal_state[1] = get_field(v1, 10, 10);
+	out->literal_state[2] = get_field(v1, 20, 10);
+	out->literal_state[3] = get_field(v1, 30, 10);
+
+	/* L,M,D state */
+	out->n_matches = get_field(v0, 40, 20);
+	out->n_lmd_payload_bytes = get_field(v1, 40, 20);
+	out->lmd_bits = (int)get_field(v1, 60, 3) - 7;
+	out->l_state = get_field(v2, 32, 10);
+	out->m_state = get_field(v2, 42, 10);
+	out->d_state = get_field(v2, 52, 10);
+
+	/* Total payload size */
+	out->n_payload_bytes = out->n_literal_payload_bytes + out->n_lmd_payload_bytes;
+
+	/* Freq tables */
+	dst = &(out->l_freq[0]);
+	src = &(in->freq[0]);
+	src_end = (const uint8_t *)in + get_field(v2, 0, 32); /* first byte after header */
+	accum = 0;
+	accum_nbits = 0;
+
+	/* No freq tables? */
+	if (src_end == src)
+		return 0; /* OK, freq tables were omitted */
+
+	for (i = 0; i < LZFSE_ENCODE_L_SYMBOLS + LZFSE_ENCODE_M_SYMBOLS + LZFSE_ENCODE_D_SYMBOLS +
+				LZFSE_ENCODE_LITERAL_SYMBOLS;
+	     i++) {
+		/*
+		 * Refill accum, one byte at a time, until we reach end of
+		 * header, or accum is full
+		 */
+		while (src < src_end && accum_nbits + 8 <= 32) {
+			accum |= (uint32_t)(*src) << accum_nbits;
+			accum_nbits += 8;
+			src++;
+		}
+
+		/* Decode and store value */
+		nbits = 0;
+		dst[i] = lzfse_decode_v1_freq_value(accum, &nbits);
+
+		if (nbits > accum_nbits)
+			return -1; /* failed */
+
+		/* Consume nbits bits */
+		accum >>= nbits;
+		accum_nbits -= nbits;
+	}
+
+	if (accum_nbits >= 8 || src != src_end)
+		return -1; /* we need to end up exactly at the end of header, with less than 8 bits in accumulator */
+
+	return 0;
+}
+
+static inline void copy(uint8_t *dst, const uint8_t *src, size_t length)
+{
+	const uint8_t *dst_end = dst + length;
+
+	do {
+		copy8(dst, src);
+		dst += 8;
+		src += 8;
+	} while (dst < dst_end);
+}
+
+static int lzfse_decode_lmd(lzfse_decoder_state *s)
+{
+	lzfse_compressed_block_decoder_state *bs = &(s->compressed_lzfse_block_state);
+	fse_state l_state = bs->l_state;
+	fse_state m_state = bs->m_state;
+	fse_state d_state = bs->d_state;
+	fse_in_stream in = bs->lmd_in_stream;
+	const uint8_t *src_start = s->src_begin;
+	const uint8_t *src = s->src + bs->lmd_in_buf;
+	const uint8_t *lit = bs->current_literal;
+	uint8_t *dst = s->dst;
+	uint32_t symbols = bs->n_matches;
+	int32_t L = bs->l_value;
+	int32_t M = bs->m_value;
+	int32_t D = bs->d_value;
+	int32_t new_d;
+
+	/*
+	 * Number of bytes remaining in the destination buffer, minus 32 to
+	 * provide a margin of safety for using overlarge copies on the fast
+	 * path. This is a signed quantity, and may go negative when we are
+	 * close to the end of the buffer. That's OK; we're careful about how
+	 * we handle it in the slow-and-careful match execution path.
+	 */
+	ptrdiff_t remaining_bytes = s->dst_end - dst - 32;
+
+	/*
+	 * If L or M is non-zero, that means that we have already started
+	 * decoding this block, and that we needed to interrupt decoding to get
+	 * more space from the caller. There's a pending L, M, D triplet that
+	 * we weren't able to completely process. Jump ahead to finish executing
+	 * that symbol before decoding new values.
+	 */
+	if (L || M)
+		goto ExecuteMatch;
+
+	while (symbols > 0) {
+		int res;
+		/* Decode the next L, M, D symbol from the input stream. */
+		res = fse_in_flush(&in, &src, src_start);
+		if (res)
+			return LZFSE_STATUS_ERROR;
+		L = fse_value_decode(&l_state, bs->l_decoder, &in);
+		if ((lit + L) >= (bs->literals + LZFSE_LITERALS_PER_BLOCK + 64))
+			return LZFSE_STATUS_ERROR;
+		res = fse_in_flush2(&in, &src, src_start);
+		if (res)
+			return LZFSE_STATUS_ERROR;
+		M = fse_value_decode(&m_state, bs->m_decoder, &in);
+		res = fse_in_flush2(&in, &src, src_start);
+		if (res)
+			return LZFSE_STATUS_ERROR;
+		new_d = fse_value_decode(&d_state, bs->d_decoder, &in);
+		D = new_d ? new_d : D;
+		symbols--;
+
+ExecuteMatch:
+		/*
+		 * Error if D is out of range, so that we avoid passing through
+		 * uninitialized data or accesssing memory out of the
+		 * destination buffer.
+		 */
+		if ((uint32_t)D > dst + L - s->dst_begin)
+			return LZFSE_STATUS_ERROR;
+
+		if (L + M <= remaining_bytes) {
+			size_t i;
+			/*
+			 * If we have plenty of space remaining, we can copy the
+			 * literal and match with 16- and 32-byte operations,
+			 * without worrying about writing off the end of the
+			 * buffer.
+			 */
+			remaining_bytes -= L + M;
+			copy(dst, lit, L);
+			dst += L;
+			lit += L;
+			/*
+			 * For the match, we have two paths; a fast copy by
+			 * 16-bytes if the match distance is large enough to
+			 * allow it, and a more careful path that applies a
+			 * permutation to account for the possible overlap
+			 * between source and destination if the distance is
+			 * small.
+			 */
+			if (D >= 8 || D >= M)
+				copy(dst, dst - D, M);
+			else
+				for (i = 0; i < M; i++)
+					dst[i] = dst[i - D];
+			dst += M;
+		}
+
+		else {
+			/*
+			 * Otherwise, we are very close to the end of the
+			 * destination buffer, so we cannot use wide copies that
+			 * slop off the end of the region that we are copying
+			 * to. First, we restore the true length remaining,
+			 * rather than the sham value we've been using so far.
+			 */
+			remaining_bytes += 32;
+			/*
+			 * Now, we process the literal. Either there's space for
+			 * it or there isn't; if there is, we copy the whole
+			 * thing and update all the pointers and lengths to
+			 * reflect the copy.
+			 */
+			if (L <= remaining_bytes) {
+				size_t i;
+
+				for (i = 0; i < L; i++)
+					dst[i] = lit[i];
+				dst += L;
+				lit += L;
+				remaining_bytes -= L;
+				L = 0;
+			}
+			/*
+			 * There isn't enough space to fit the whole literal.
+			 * Copy as much of it as we can, update the pointers and
+			 * the value of L, and report that the destination
+			 * buffer is full. Note that we always write right up to
+			 * the end of the destination buffer.
+			 */
+			else {
+				size_t i;
+
+				for (i = 0; i < remaining_bytes; i++)
+					dst[i] = lit[i];
+				dst += remaining_bytes;
+				lit += remaining_bytes;
+				L -= remaining_bytes;
+				goto DestinationBufferIsFull;
+			}
+			/*
+			 * The match goes just like the literal does. We copy as
+			 * much as we can byte-by-byte, and if we reach the end
+			 * of the buffer before finishing, we return to the
+			 * caller indicating that the buffer is full.
+			 */
+			if (M <= remaining_bytes) {
+				size_t i;
+
+				for (i = 0; i < M; i++)
+					dst[i] = dst[i - D];
+				dst += M;
+				remaining_bytes -= M;
+				M = 0;
+				(void)M; /* no dead store warning */
+				/*
+				 * We don't need to update M = 0, because
+				 * there's no partial symbol to continue
+				 * executing. Either we're at the end of the
+				 * block, in which case we will never need to
+				 * resume with this state, or we're going to
+				 * decode another L, M, D set, which will
+				 * overwrite M anyway.
+				 *
+				 * But we still set M = 0, to maintain the
+				 * post-condition.
+				 */
+			} else {
+				size_t i;
+
+				for (i = 0; i < remaining_bytes; i++)
+					dst[i] = dst[i - D];
+				dst += remaining_bytes;
+				M -= remaining_bytes;
+DestinationBufferIsFull:
+				/*
+				 * Because we want to be able to resume decoding
+				 * where we've left off (even in the middle of a
+				 * literal or match), we need to update all of
+				 * the block state fields with the current
+				 * values so that we can resume execution from
+				 * this point once the caller has given us more
+				 * space to write into.
+				 */
+				bs->l_value = L;
+				bs->m_value = M;
+				bs->d_value = D;
+				bs->l_state = l_state;
+				bs->m_state = m_state;
+				bs->d_state = d_state;
+				bs->lmd_in_stream = in;
+				bs->n_matches = symbols;
+				bs->lmd_in_buf = (uint32_t)(src - s->src);
+				bs->current_literal = lit;
+				s->dst = dst;
+				return LZFSE_STATUS_DST_FULL;
+			}
+			/*
+			 * Restore the "sham" decremented value of
+			 * remaining_bytes and continue to the next L, M, D
+			 * triple. We'll just be back in the careful path again,
+			 * but this only happens at the very end of the buffer,
+			 * so a little minor inefficiency here is a good
+			 * tradeoff for simpler code.
+			 */
+			remaining_bytes -= 32;
+		}
+	}
+	/*
+	 * Because we've finished with the whole block, we don't need to update
+	 * any of the blockstate fields; they will not be used again. We just
+	 * update the destination pointer in the state object and return.
+	 */
+	s->dst = dst;
+	return LZFSE_STATUS_OK;
+}
+
+int lzfse_decode(lzfse_decoder_state *s)
+{
+	while (1) {
+		/* Are we inside a block? */
+		switch (s->block_magic) {
+		case LZFSE_NO_BLOCK_MAGIC: {
+			uint32_t magic;
+			/* We need at least 4 bytes of magic number to identify next block */
+			if (s->src + 4 > s->src_end)
+				return LZFSE_STATUS_SRC_EMPTY; /* SRC truncated */
+			magic = load4(s->src);
+
+			if (magic == LZFSE_ENDOFSTREAM_BLOCK_MAGIC) {
+				s->src += 4;
+				s->end_of_stream = 1;
+				return LZFSE_STATUS_OK; /* done */
+			}
+
+			if (magic == LZFSE_UNCOMPRESSED_BLOCK_MAGIC) {
+				uncompressed_block_decoder_state *bs = NULL;
+
+				if (s->src + sizeof(uncompressed_block_header) > s->src_end)
+					return LZFSE_STATUS_SRC_EMPTY; /* SRC truncated */
+				/* Setup state for uncompressed block */
+				bs = &(s->uncompressed_block_state);
+				bs->n_raw_bytes = load4(
+					s->src + offsetof(uncompressed_block_header, n_raw_bytes));
+				s->src += sizeof(uncompressed_block_header);
+				s->block_magic = magic;
+				break;
+			}
+
+			if (magic == LZFSE_COMPRESSEDLZVN_BLOCK_MAGIC) {
+				lzvn_compressed_block_decoder_state *bs = NULL;
+
+				if (s->src + sizeof(lzvn_compressed_block_header) > s->src_end)
+					return LZFSE_STATUS_SRC_EMPTY; /* SRC truncated */
+				/* Setup state for compressed LZVN block */
+				bs = &(s->compressed_lzvn_block_state);
+				bs->n_raw_bytes =
+					load4(s->src +
+					      offsetof(lzvn_compressed_block_header, n_raw_bytes));
+				bs->n_payload_bytes =
+					load4(s->src + offsetof(lzvn_compressed_block_header,
+								n_payload_bytes));
+				bs->d_prev = 0;
+				s->src += sizeof(lzvn_compressed_block_header);
+				s->block_magic = magic;
+				break;
+			}
+
+			if (magic == LZFSE_COMPRESSEDV1_BLOCK_MAGIC ||
+			    magic == LZFSE_COMPRESSEDV2_BLOCK_MAGIC) {
+				lzfse_compressed_block_header_v1 header1;
+				size_t header_size = 0;
+				lzfse_compressed_block_decoder_state *bs = NULL;
+
+				/* Decode compressed headers */
+				if (magic == LZFSE_COMPRESSEDV2_BLOCK_MAGIC) {
+					const lzfse_compressed_block_header_v2 *header2;
+					int decodeStatus;
+					/* Check we have the fixed part of the structure */
+					if (s->src + offsetof(lzfse_compressed_block_header_v2,
+							      freq) >
+					    s->src_end)
+						return LZFSE_STATUS_SRC_EMPTY; /* SRC truncated */
+
+					/* Get size, and check we have the entire structure */
+					header2 = (const lzfse_compressed_block_header_v2 *)
+							  s->src; /* not aligned, OK */
+					header_size = lzfse_decode_v2_header_size(header2);
+					if (s->src + header_size > s->src_end)
+						return LZFSE_STATUS_SRC_EMPTY; /* SRC truncated */
+					decodeStatus = lzfse_decode_v1(&header1, header2);
+					if (decodeStatus != 0)
+						return LZFSE_STATUS_ERROR; /* failed */
+				} else {
+					if (s->src + sizeof(lzfse_compressed_block_header_v1) >
+					    s->src_end)
+						return LZFSE_STATUS_SRC_EMPTY; /* SRC truncated */
+					memcpy(&header1, s->src,
+					       sizeof(lzfse_compressed_block_header_v1));
+					header_size = sizeof(lzfse_compressed_block_header_v1);
+				}
+
+				/*
+				 * We require the header + entire encoded block
+				 * to be present in SRC during the entire block
+				 * decoding. This can be relaxed somehow, if it
+				 * becomes a limiting factor, at the price of a
+				 * more complex state maintenance. For DST, we
+				 * can't easily require space for the entire
+				 * decoded block, because it may expand to
+				 * something very very large.
+				 */
+				if (s->src + header_size + header1.n_literal_payload_bytes +
+					    header1.n_lmd_payload_bytes >
+				    s->src_end)
+					return LZFSE_STATUS_SRC_EMPTY; /* need all encoded block */
+
+				/* Sanity checks */
+				if (lzfse_check_block_header_v1(&header1) != 0)
+					return LZFSE_STATUS_ERROR;
+
+				/* Skip header */
+				s->src += header_size;
+
+				/* Setup state for compressed V1 block from header */
+				bs = &(s->compressed_lzfse_block_state);
+				bs->n_lmd_payload_bytes = header1.n_lmd_payload_bytes;
+				bs->n_matches = header1.n_matches;
+				fse_init_decoder_table(LZFSE_ENCODE_LITERAL_STATES,
+						       LZFSE_ENCODE_LITERAL_SYMBOLS,
+						       header1.literal_freq, bs->literal_decoder);
+				fse_init_value_decoder_table(LZFSE_ENCODE_L_STATES,
+							     LZFSE_ENCODE_L_SYMBOLS, header1.l_freq,
+							     l_extra_bits, l_base_value,
+							     bs->l_decoder);
+				fse_init_value_decoder_table(LZFSE_ENCODE_M_STATES,
+							     LZFSE_ENCODE_M_SYMBOLS, header1.m_freq,
+							     m_extra_bits, m_base_value,
+							     bs->m_decoder);
+				fse_init_value_decoder_table(LZFSE_ENCODE_D_STATES,
+							     LZFSE_ENCODE_D_SYMBOLS, header1.d_freq,
+							     d_extra_bits, d_base_value,
+							     bs->d_decoder);
+
+				/* Decode literals */
+				{
+					fse_in_stream in;
+					const uint8_t *buf_start = s->src_begin;
+					const uint8_t *buf;
+					fse_state state0;
+					fse_state state1;
+					fse_state state2;
+					fse_state state3;
+					uint32_t i;
+
+					s->src +=
+						header1.n_literal_payload_bytes; /* skip literal payload */
+					buf = s->src; /* read bits backwards from the end */
+					if (fse_in_init(&in, header1.literal_bits, &buf,
+							buf_start) != 0)
+						return LZFSE_STATUS_ERROR;
+
+					state0 = header1.literal_state[0];
+					state1 = header1.literal_state[1];
+					state2 = header1.literal_state[2];
+					state3 = header1.literal_state[3];
+
+					for (i = 0; i < header1.n_literals;
+					     i += 4 /* n_literals is multiple of 4 */) {
+#if FSE_IOSTREAM_64
+						if (fse_in_flush(&in, &buf, buf_start) != 0)
+							return LZFSE_STATUS_ERROR; /* [57, 64] bits */
+						bs->literals[i + 0] =
+							fse_decode(&state0, bs->literal_decoder,
+								   &in); /* 10b max */
+						bs->literals[i + 1] =
+							fse_decode(&state1, bs->literal_decoder,
+								   &in); /* 10b max */
+						bs->literals[i + 2] =
+							fse_decode(&state2, bs->literal_decoder,
+								   &in); /* 10b max */
+						bs->literals[i + 3] =
+							fse_decode(&state3, bs->literal_decoder,
+								   &in); /* 10b max */
+#else
+						if (fse_in_flush(&in, &buf, buf_start) != 0)
+							return LZFSE_STATUS_ERROR; /* [25, 23] bits */
+						bs->literals[i + 0] =
+							fse_decode(&state0, bs->literal_decoder,
+								   &in); /* 10b max */
+						bs->literals[i + 1] =
+							fse_decode(&state1, bs->literal_decoder,
+								   &in); /* 10b max */
+						if (fse_in_flush(&in, &buf, buf_start) != 0)
+							return LZFSE_STATUS_ERROR; /* [25, 23] bits */
+						bs->literals[i + 2] =
+							fse_decode(&state2, bs->literal_decoder,
+								   &in); /* 10b max */
+						bs->literals[i + 3] =
+							fse_decode(&state3, bs->literal_decoder,
+								   &in); /* 10b max */
+#endif
+					}
+
+					bs->current_literal = bs->literals;
+				} /* literals */
+
+				/*
+				 * SRC is not incremented to skip the LMD
+				 * payload, since we need it during block
+				 * decode. We will increment SRC at the end of
+				 * the block only after this point.
+				 */
+
+				/*
+				 * Initialize the L,M,D decode stream, do not
+				 * start decoding matches yet, and store decoder
+				 * state
+				 */
+				{
+					fse_in_stream in;
+					/* read bits backwards from the end */
+					const uint8_t *buf = s->src + header1.n_lmd_payload_bytes;
+
+					if (fse_in_init(&in, header1.lmd_bits, &buf, s->src) != 0)
+						return LZFSE_STATUS_ERROR;
+
+					bs->l_state = header1.l_state;
+					bs->m_state = header1.m_state;
+					bs->d_state = header1.d_state;
+					bs->lmd_in_buf = (uint32_t)(buf - s->src);
+					bs->l_value = bs->m_value = 0;
+					/*
+					 * Initialize D to an illegal value so
+					 * we can't erroneously use an
+					 * uninitialized "previous" value.
+					 */
+					bs->d_value = -1;
+					bs->lmd_in_stream = in;
+				}
+
+				s->block_magic = magic;
+				break;
+			}
+
+			/* Here we have an invalid magic number */
+			return LZFSE_STATUS_ERROR;
+		} /* LZFSE_NO_BLOCK_MAGIC */
+
+		case LZFSE_UNCOMPRESSED_BLOCK_MAGIC: {
+			uncompressed_block_decoder_state *bs = &(s->uncompressed_block_state);
+
+			/*
+			 * Compute the size (in bytes) of the data that we will
+			 * actually copy. This size is
+			 * minimum(bs->n_raw_bytes, space in src, space in dst).
+			 */
+
+			uint32_t copy_size = bs->n_raw_bytes; /* bytes left to copy */
+			size_t src_space, dst_space;
+
+			if (copy_size == 0) {
+				s->block_magic = 0;
+				break;
+			} /* end of block */
+
+			if (s->src_end <= s->src)
+				return LZFSE_STATUS_SRC_EMPTY; /* need more SRC data */
+			src_space = s->src_end - s->src;
+			if (copy_size > src_space)
+				copy_size = (uint32_t)src_space; /* limit to SRC data (> 0) */
+
+			if (s->dst_end <= s->dst)
+				return LZFSE_STATUS_DST_FULL; /* need more DST capacity */
+			dst_space = s->dst_end - s->dst;
+			if (copy_size > dst_space)
+				copy_size = (uint32_t)dst_space; /* limit to DST capacity (> 0) */
+
+			/*
+			 * Now that we know that the copy size is bounded to the
+			 * source and dest buffers, go ahead and copy the data.
+			 * We always have copy_size > 0 here
+			 */
+			memcpy(s->dst, s->src, copy_size);
+			s->src += copy_size;
+			s->dst += copy_size;
+			bs->n_raw_bytes -= copy_size;
+
+			break;
+		} /* LZFSE_UNCOMPRESSED_BLOCK_MAGIC */
+
+		case LZFSE_COMPRESSEDV1_BLOCK_MAGIC:
+		case LZFSE_COMPRESSEDV2_BLOCK_MAGIC: {
+			int status;
+			lzfse_compressed_block_decoder_state *bs =
+				&(s->compressed_lzfse_block_state);
+			/* Require the entire LMD payload to be in SRC */
+			if (s->src_end <= s->src ||
+			    bs->n_lmd_payload_bytes > (size_t)(s->src_end - s->src))
+				return LZFSE_STATUS_SRC_EMPTY;
+
+			status = lzfse_decode_lmd(s);
+			if (status != LZFSE_STATUS_OK)
+				return status;
+
+			s->block_magic = LZFSE_NO_BLOCK_MAGIC;
+			s->src += bs->n_lmd_payload_bytes; // to next block
+			break;
+		} /* LZFSE_COMPRESSEDV1_BLOCK_MAGIC || LZFSE_COMPRESSEDV2_BLOCK_MAGIC */
+
+		case LZFSE_COMPRESSEDLZVN_BLOCK_MAGIC: {
+			lzvn_compressed_block_decoder_state *bs = &(s->compressed_lzvn_block_state);
+			lzvn_decoder_state dstate;
+			size_t src_used, dst_used;
+
+			if (bs->n_payload_bytes > 0 && s->src_end <= s->src)
+				return LZFSE_STATUS_SRC_EMPTY; /* need more SRC data */
+
+			/* Init LZVN decoder state */
+			memset(&dstate, 0x00, sizeof(dstate));
+			dstate.src = s->src;
+			dstate.src_end = s->src_end;
+			if (dstate.src_end - s->src > bs->n_payload_bytes)
+				dstate.src_end =
+					s->src + bs->n_payload_bytes; /* limit to payload bytes */
+			dstate.dst_begin = s->dst_begin;
+			dstate.dst = s->dst;
+			dstate.dst_end = s->dst_end;
+			if (dstate.dst_end - s->dst > bs->n_raw_bytes)
+				dstate.dst_end = s->dst + bs->n_raw_bytes; /* limit to raw bytes */
+			dstate.d_prev = bs->d_prev;
+			dstate.end_of_stream = 0;
+
+			/* Run LZVN decoder */
+			lzvn_decode(&dstate);
+
+			/* Update our state */
+			src_used = dstate.src - s->src;
+			dst_used = dstate.dst - s->dst;
+			if (src_used > bs->n_payload_bytes || dst_used > bs->n_raw_bytes)
+				return LZFSE_STATUS_ERROR; /* sanity check */
+			s->src = dstate.src;
+			s->dst = dstate.dst;
+			bs->n_payload_bytes -= (uint32_t)src_used;
+			bs->n_raw_bytes -= (uint32_t)dst_used;
+			bs->d_prev = (uint32_t)dstate.d_prev;
+
+			/* Test end of block */
+			if (bs->n_payload_bytes == 0 && bs->n_raw_bytes == 0 &&
+			    dstate.end_of_stream) {
+				s->block_magic = 0;
+				break;
+			} /* block done */
+
+			/* Check for invalid state */
+			if (bs->n_payload_bytes == 0 || bs->n_raw_bytes == 0 ||
+			    dstate.end_of_stream)
+				return LZFSE_STATUS_ERROR;
+
+			/*
+			 * Here, block is not done and state is valid, so we
+			 * need more space in dst.
+			 */
+			return LZFSE_STATUS_DST_FULL;
+		}
+
+		default:
+			return LZFSE_STATUS_ERROR; /* invalid magic */
+
+		} /* switch magic */
+
+	} /* block loop */
+
+	return LZFSE_STATUS_OK;
+}
diff --git a/fs/apfs/lzfse/lzfse_fse.c b/fs/apfs/lzfse/lzfse_fse.c
new file mode 100644
index 000000000..7005bbe72
--- /dev/null
+++ b/fs/apfs/lzfse/lzfse_fse.c
@@ -0,0 +1,126 @@
+// SPDX-License-Identifier: BSD-3-Clause
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "lzfse_internal.h"
+
+/*
+ * Initialize decoder table T[NSTATES].
+ * NSTATES = sum FREQ[i] is the number of states (a power of 2)
+ * NSYMBOLS is the number of symbols.
+ * FREQ[NSYMBOLS] is a normalized histogram of symbol frequencies, with FREQ[i]
+ * >= 0.
+ * Some symbols may have a 0 frequency. In that case, they should not be
+ * present in the data.
+ */
+int fse_init_decoder_table(int nstates, int nsymbols, const uint16_t *__restrict freq,
+			   int32_t *__restrict t)
+{
+	int n_clz = __builtin_clz(nstates);
+	int sum_of_freq = 0;
+	int i, j0, j;
+
+	for (i = 0; i < nsymbols; i++) {
+		int f = (int)freq[i];
+		int k;
+
+		if (f == 0)
+			continue; /* skip this symbol, no occurrences */
+
+		sum_of_freq += f;
+
+		if (sum_of_freq > nstates)
+			return -1;
+
+		k = __builtin_clz(f) - n_clz; /* shift needed to ensure N <= (F<<K) < 2*N */
+		j0 = ((2 * nstates) >> k) - f;
+
+		/* Initialize all states S reached by this symbol: OFFSET <= S < OFFSET + F */
+		for (j = 0; j < f; j++) {
+			fse_decoder_entry e;
+
+			e.symbol = (uint8_t)i;
+			if (j < j0) {
+				e.k = (int8_t)k;
+				e.delta = (int16_t)(((f + j) << k) - nstates);
+			} else {
+				e.k = (int8_t)(k - 1);
+				e.delta = (int16_t)((j - j0) << (k - 1));
+			}
+
+			memcpy(t, &e, sizeof(e));
+			t++;
+		}
+	}
+
+	return 0; /* OK */
+}
+
+/*
+ * Initialize value decoder table T[NSTATES].
+ * NSTATES = sum FREQ[i] is the number of states (a power of 2)
+ * NSYMBOLS is the number of symbols.
+ * FREQ[NSYMBOLS] is a normalized histogram of symbol frequencies, with FREQ[i]
+ * >= 0.
+ * SYMBOL_VBITS[NSYMBOLS] and SYMBOLS_VBASE[NSYMBOLS] are the number of value
+ * bits to read and the base value for each symbol.
+ * Some symbols may have a 0 frequency. In that case, they should not be
+ * present in the data.
+ */
+void fse_init_value_decoder_table(int nstates, int nsymbols, const uint16_t *__restrict freq,
+				  const uint8_t *__restrict symbol_vbits,
+				  const int32_t *__restrict symbol_vbase,
+				  struct fse_value_decoder_entry *__restrict t)
+{
+	int n_clz = __builtin_clz(nstates);
+	int i;
+
+	for (i = 0; i < nsymbols; i++) {
+		struct fse_value_decoder_entry ei = { 0 };
+		int f = (int)freq[i];
+		int k, j0, j;
+
+		if (f == 0)
+			continue; /* skip this symbol, no occurrences */
+
+		k = __builtin_clz(f) - n_clz; /* shift needed to ensure N <= (F<<K) < 2*N */
+		j0 = ((2 * nstates) >> k) - f;
+
+		ei.value_bits = symbol_vbits[i];
+		ei.vbase = symbol_vbase[i];
+
+		/* Initialize all states S reached by this symbol: OFFSET <= S < OFFSET + F */
+		for (j = 0; j < f; j++) {
+			struct fse_value_decoder_entry e = ei;
+
+			if (j < j0) {
+				e.total_bits = (uint8_t)k + e.value_bits;
+				e.delta = (int16_t)(((f + j) << k) - nstates);
+			} else {
+				e.total_bits = (uint8_t)(k - 1) + e.value_bits;
+				e.delta = (int16_t)((j - j0) << (k - 1));
+			}
+
+			memcpy(t, &e, 8);
+			t++;
+		}
+	}
+}
diff --git a/fs/apfs/lzfse/lzfse_fse.h b/fs/apfs/lzfse/lzfse_fse.h
new file mode 100644
index 000000000..35b46866d
--- /dev/null
+++ b/fs/apfs/lzfse/lzfse_fse.h
@@ -0,0 +1,573 @@
+/* SPDX-License-Identifier: BSD-3-Clause */
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Finite state entropy coding (FSE)
+ * This is an implementation of the tANS algorithm described by Jarek Duda,
+ * we use the more descriptive name "Finite State Entropy".
+ */
+
+#ifndef LZFSE_FSE_H
+#define LZFSE_FSE_H
+
+#include <linux/types.h>
+#include <linux/string.h>
+
+/*
+ * Select between 32/64-bit I/O streams for FSE. Note that the FSE stream
+ * size need not match the word size of the machine, but in practice you
+ * want to use 64b streams on 64b systems for better performance.
+ */
+#if defined(_M_AMD64) || defined(__x86_64__) || defined(__arm64__)
+#define FSE_IOSTREAM_64 1
+#else
+#define FSE_IOSTREAM_64 0
+#endif
+
+/* MARK: - Bit utils */
+
+/*! @abstract Signed type used to represent bit count. */
+typedef int32_t fse_bit_count;
+
+/*! @abstract Unsigned type used to represent FSE state. */
+typedef uint16_t fse_state;
+
+/* Mask the NBITS lsb of X. 0 <= NBITS < 64 */
+static inline uint64_t fse_mask_lsb64(uint64_t x, fse_bit_count nbits)
+{
+	static const uint64_t mtable[65] = {
+		0x0000000000000000LLU, 0x0000000000000001LLU, 0x0000000000000003LLU,
+		0x0000000000000007LLU, 0x000000000000000fLLU, 0x000000000000001fLLU,
+		0x000000000000003fLLU, 0x000000000000007fLLU, 0x00000000000000ffLLU,
+		0x00000000000001ffLLU, 0x00000000000003ffLLU, 0x00000000000007ffLLU,
+		0x0000000000000fffLLU, 0x0000000000001fffLLU, 0x0000000000003fffLLU,
+		0x0000000000007fffLLU, 0x000000000000ffffLLU, 0x000000000001ffffLLU,
+		0x000000000003ffffLLU, 0x000000000007ffffLLU, 0x00000000000fffffLLU,
+		0x00000000001fffffLLU, 0x00000000003fffffLLU, 0x00000000007fffffLLU,
+		0x0000000000ffffffLLU, 0x0000000001ffffffLLU, 0x0000000003ffffffLLU,
+		0x0000000007ffffffLLU, 0x000000000fffffffLLU, 0x000000001fffffffLLU,
+		0x000000003fffffffLLU, 0x000000007fffffffLLU, 0x00000000ffffffffLLU,
+		0x00000001ffffffffLLU, 0x00000003ffffffffLLU, 0x00000007ffffffffLLU,
+		0x0000000fffffffffLLU, 0x0000001fffffffffLLU, 0x0000003fffffffffLLU,
+		0x0000007fffffffffLLU, 0x000000ffffffffffLLU, 0x000001ffffffffffLLU,
+		0x000003ffffffffffLLU, 0x000007ffffffffffLLU, 0x00000fffffffffffLLU,
+		0x00001fffffffffffLLU, 0x00003fffffffffffLLU, 0x00007fffffffffffLLU,
+		0x0000ffffffffffffLLU, 0x0001ffffffffffffLLU, 0x0003ffffffffffffLLU,
+		0x0007ffffffffffffLLU, 0x000fffffffffffffLLU, 0x001fffffffffffffLLU,
+		0x003fffffffffffffLLU, 0x007fffffffffffffLLU, 0x00ffffffffffffffLLU,
+		0x01ffffffffffffffLLU, 0x03ffffffffffffffLLU, 0x07ffffffffffffffLLU,
+		0x0fffffffffffffffLLU, 0x1fffffffffffffffLLU, 0x3fffffffffffffffLLU,
+		0x7fffffffffffffffLLU, 0xffffffffffffffffLLU,
+	};
+	return x & mtable[nbits];
+}
+
+/* Mask the NBITS lsb of X. 0 <= NBITS < 32 */
+static inline uint32_t fse_mask_lsb32(uint32_t x, fse_bit_count nbits)
+{
+	static const uint32_t mtable[33] = {
+		0x0000000000000000U, 0x0000000000000001U, 0x0000000000000003U, 0x0000000000000007U,
+		0x000000000000000fU, 0x000000000000001fU, 0x000000000000003fU, 0x000000000000007fU,
+		0x00000000000000ffU, 0x00000000000001ffU, 0x00000000000003ffU, 0x00000000000007ffU,
+		0x0000000000000fffU, 0x0000000000001fffU, 0x0000000000003fffU, 0x0000000000007fffU,
+		0x000000000000ffffU, 0x000000000001ffffU, 0x000000000003ffffU, 0x000000000007ffffU,
+		0x00000000000fffffU, 0x00000000001fffffU, 0x00000000003fffffU, 0x00000000007fffffU,
+		0x0000000000ffffffU, 0x0000000001ffffffU, 0x0000000003ffffffU, 0x0000000007ffffffU,
+		0x000000000fffffffU, 0x000000001fffffffU, 0x000000003fffffffU, 0x000000007fffffffU,
+		0x00000000ffffffffU,
+	};
+	return x & mtable[nbits];
+}
+
+/*! @abstract Select \c nbits at index \c start from \c x.
+ *  0 <= start <= start+nbits <= 64
+ */
+static __always_inline uint64_t fse_extract_bits64(uint64_t x, fse_bit_count start,
+						   fse_bit_count nbits)
+{
+	/* If START and NBITS are constants, map to bit-field extraction instructions */
+	if (__builtin_constant_p(start) && __builtin_constant_p(nbits))
+		return (x >> start) & ((1LLU << nbits) - 1LLU);
+
+	/* Otherwise, shift and mask */
+	return fse_mask_lsb64(x >> start, nbits);
+}
+
+/*! @abstract Select \c nbits at index \c start from \c x.
+ *  0 <= start <= start+nbits <= 32
+ */
+static __always_inline uint32_t fse_extract_bits32(uint32_t x, fse_bit_count start,
+						   fse_bit_count nbits)
+{
+	/* If START and NBITS are constants, map to bit-field extraction instructions */
+	if (__builtin_constant_p(start) && __builtin_constant_p(nbits))
+		return (x >> start) & ((1U << nbits) - 1U);
+
+	/* Otherwise, shift and mask */
+	return fse_mask_lsb32(x >> start, nbits);
+}
+
+/* MARK: - Bit stream */
+
+/*
+ * I/O streams
+ * The streams can be shared between several FSE encoders/decoders, which is why
+ * they are not in the state struct
+ */
+
+/*! @abstract Output stream, 64-bit accum. */
+typedef struct {
+	uint64_t accum; /* Output bits */
+	fse_bit_count accum_nbits; /* Number of valid bits in ACCUM, other bits are 0 */
+} fse_out_stream64;
+
+/*! @abstract Output stream, 32-bit accum. */
+typedef struct {
+	uint32_t accum; /* Output bits */
+	fse_bit_count accum_nbits; /* Number of valid bits in ACCUM, other bits are 0 */
+} fse_out_stream32;
+
+/*! @abstract Object representing an input stream. */
+typedef struct {
+	uint64_t accum; /* Input bits */
+	fse_bit_count accum_nbits; /* Number of valid bits in ACCUM, other bits are 0 */
+} fse_in_stream64;
+
+/*! @abstract Object representing an input stream. */
+typedef struct {
+	uint32_t accum; /* Input bits */
+	fse_bit_count accum_nbits; /* Number of valid bits in ACCUM, other bits are 0 */
+} fse_in_stream32;
+
+/*! @abstract Initialize an output stream object. */
+static __always_inline void fse_out_init64(fse_out_stream64 *s)
+{
+	s->accum = 0;
+	s->accum_nbits = 0;
+}
+
+/*! @abstract Initialize an output stream object. */
+static __always_inline void fse_out_init32(fse_out_stream32 *s)
+{
+	s->accum = 0;
+	s->accum_nbits = 0;
+}
+
+/*! @abstract Write full bytes from the accumulator to output buffer, ensuring
+ * accum_nbits is in [0, 7].
+ * We assume we can write 8 bytes to the output buffer \c (*pbuf[0..7]) in all
+ * cases.
+ * @note *pbuf is incremented by the number of written bytes.
+ */
+static __always_inline void fse_out_flush64(fse_out_stream64 *s, uint8_t **pbuf)
+{
+	fse_bit_count nbits = s->accum_nbits & -8; /* number of bits written, multiple of 8 */
+
+	/* Write 8 bytes of current accumulator */
+	memcpy(*pbuf, &(s->accum), 8);
+	*pbuf += (nbits >> 3); /* bytes */
+
+	/* Update state */
+	s->accum >>= nbits; /* remove nbits */
+	s->accum_nbits -= nbits;
+}
+
+/*! @abstract Write full bytes from the accumulator to output buffer, ensuring
+ * accum_nbits is in [0, 7].
+ * We assume we can write 4 bytes to the output buffer \c (*pbuf[0..3]) in all
+ * cases.
+ * @note *pbuf is incremented by the number of written bytes.
+ */
+static __always_inline void fse_out_flush32(fse_out_stream32 *s, uint8_t **pbuf)
+{
+	fse_bit_count nbits = s->accum_nbits & -8; /* number of bits written, multiple of 8 */
+
+	/* Write 4 bytes of current accumulator */
+	memcpy(*pbuf, &(s->accum), 4);
+	*pbuf += (nbits >> 3); /* bytes */
+
+	/* Update state */
+	s->accum >>= nbits; /* remove nbits */
+	s->accum_nbits -= nbits;
+}
+
+/*! @abstract Write the last bytes from the accumulator to output buffer,
+ * ensuring accum_nbits is in [-7, 0]. Bits are padded with 0 if needed.
+ * We assume we can write 8 bytes to the output buffer \c (*pbuf[0..7]) in all
+ * cases.
+ * @note *pbuf is incremented by the number of written bytes.
+ */
+static __always_inline void fse_out_finish64(fse_out_stream64 *s, uint8_t **pbuf)
+{
+	fse_bit_count nbits = (s->accum_nbits + 7) & -8; /* number of bits written, multiple of 8 */
+
+	/* Write 8 bytes of current accumulator */
+	memcpy(*pbuf, &(s->accum), 8);
+	*pbuf += (nbits >> 3); /* bytes */
+
+	/* Update state */
+	s->accum = 0; /* remove nbits */
+	s->accum_nbits -= nbits;
+}
+
+/*! @abstract Write the last bytes from the accumulator to output buffer,
+ * ensuring accum_nbits is in [-7, 0]. Bits are padded with 0 if needed.
+ * We assume we can write 4 bytes to the output buffer \c (*pbuf[0..3]) in all
+ * cases.
+ * @note *pbuf is incremented by the number of written bytes.
+ */
+static __always_inline void fse_out_finish32(fse_out_stream32 *s, uint8_t **pbuf)
+{
+	fse_bit_count nbits = (s->accum_nbits + 7) & -8; /* number of bits written, multiple of 8 */
+
+	/* Write 8 bytes of current accumulator */
+	memcpy(*pbuf, &(s->accum), 4);
+	*pbuf += (nbits >> 3); /* bytes */
+
+	/* Update state */
+	s->accum = 0; /* remove nbits */
+	s->accum_nbits -= nbits;
+}
+
+/*! @abstract Accumulate \c n bits \c b to output stream \c s. We \b must have:
+ * 0 <= b < 2^n, and N + s->accum_nbits <= 64.
+ * @note The caller must ensure out_flush is called \b before the accumulator
+ * overflows to more than 64 bits.
+ */
+static __always_inline void fse_out_push64(fse_out_stream64 *s, fse_bit_count n, uint64_t b)
+{
+	s->accum |= b << s->accum_nbits;
+	s->accum_nbits += n;
+}
+
+/*! @abstract Accumulate \c n bits \c b to output stream \c s. We \b must have:
+ * 0 <= n < 2^n, and n + s->accum_nbits <= 32.
+ * @note The caller must ensure out_flush is called \b before the accumulator
+ * overflows to more than 32 bits.
+ */
+static __always_inline void fse_out_push32(fse_out_stream32 *s, fse_bit_count n, uint32_t b)
+{
+	s->accum |= b << s->accum_nbits;
+	s->accum_nbits += n;
+}
+
+#define DEBUG_CHECK_INPUT_STREAM_PARAMETERS
+
+/*! @abstract   Initialize the fse input stream so that accum holds between 56
+ *  and 63 bits. We never want to have 64 bits in the stream, because that allows
+ *  us to avoid a special case in the fse_in_pull function (eliminating an
+ *  unpredictable branch), while not requiring any additional fse_flush
+ *  operations. This is why we have the special case for n == 0 (in which case
+ *  we want to load only 7 bytes instead of 8).
+ */
+static __always_inline int fse_in_checked_init64(fse_in_stream64 *s, fse_bit_count n,
+						 const uint8_t **pbuf, const uint8_t *buf_start)
+{
+	if (n) {
+		if (*pbuf < buf_start + 8)
+			return -1; /* out of range */
+		*pbuf -= 8;
+		memcpy(&(s->accum), *pbuf, 8);
+		s->accum_nbits = n + 64;
+	} else {
+		if (*pbuf < buf_start + 7)
+			return -1; /* out of range */
+		*pbuf -= 7;
+		memcpy(&(s->accum), *pbuf, 7);
+		s->accum &= 0xffffffffffffff;
+		s->accum_nbits = n + 56;
+	}
+
+	if ((s->accum_nbits < 56 || s->accum_nbits >= 64) || ((s->accum >> s->accum_nbits) != 0)) {
+		/*
+		 * the incoming input is wrong (encoder should have zeroed the
+		 * upper bits)
+		 */
+		return -1;
+	}
+
+	return 0; /* OK */
+}
+
+/*! @abstract Identical to previous function, but for 32-bit operation
+ * (resulting bit count is between 24 and 31 bits).
+ */
+static __always_inline int fse_in_checked_init32(fse_in_stream32 *s, fse_bit_count n,
+						 const uint8_t **pbuf, const uint8_t *buf_start)
+{
+	if (n) {
+		if (*pbuf < buf_start + 4)
+			return -1; /* out of range */
+		*pbuf -= 4;
+		memcpy(&(s->accum), *pbuf, 4);
+		s->accum_nbits = n + 32;
+	} else {
+		if (*pbuf < buf_start + 3)
+			return -1; /* out of range */
+		*pbuf -= 3;
+		memcpy(&(s->accum), *pbuf, 3);
+		s->accum &= 0xffffff;
+		s->accum_nbits = n + 24;
+	}
+
+	if ((s->accum_nbits < 24 || s->accum_nbits >= 32) || ((s->accum >> s->accum_nbits) != 0)) {
+		/*
+		 * the incoming input is wrong (encoder should have zeroed the
+		 * upper bits)
+		 */
+		return -1;
+	}
+
+	return 0; /* OK */
+}
+
+/*! @abstract  Read in new bytes from buffer to ensure that we have a full
+ * complement of bits in the stream object (again, between 56 and 63 bits).
+ * checking the new value of \c *pbuf remains >= \c buf_start.
+ * @return 0 if OK.
+ * @return -1 on failure.
+ */
+static __always_inline int fse_in_checked_flush64(fse_in_stream64 *s, const uint8_t **pbuf,
+						  const uint8_t *buf_start)
+{
+	/* Get number of bits to add to bring us into the desired range. */
+	fse_bit_count nbits = (63 - s->accum_nbits) & -8;
+	/* Convert bits to bytes and decrement buffer address, then load new data. */
+	const uint8_t *buf = (*pbuf) - (nbits >> 3);
+	uint64_t incoming;
+
+	if (buf < buf_start)
+		return -1; /* out of range */
+	*pbuf = buf;
+	memcpy(&incoming, buf, 8);
+	/* Update the state object and verify its validity (in DEBUG). */
+	s->accum = (s->accum << nbits) | fse_mask_lsb64(incoming, nbits);
+	s->accum_nbits += nbits;
+	DEBUG_CHECK_INPUT_STREAM_PARAMETERS
+	return 0; // OK
+}
+
+/*! @abstract Identical to previous function (but again, we're only filling
+ * a 32-bit field with between 24 and 31 bits).
+ */
+static __always_inline int fse_in_checked_flush32(fse_in_stream32 *s, const uint8_t **pbuf,
+						  const uint8_t *buf_start)
+{
+	/* Get number of bits to add to bring us into the desired range. */
+	fse_bit_count nbits = (31 - s->accum_nbits) & -8;
+
+	if (nbits > 0) {
+		/* Convert bits to bytes and decrement buffer address, then load new data. */
+		const uint8_t *buf = (*pbuf) - (nbits >> 3);
+		uint32_t incoming;
+
+		if (buf < buf_start)
+			return -1; /* out of range */
+
+		*pbuf = buf;
+
+		incoming = *((uint32_t *)buf);
+
+		/* Update the state object and verify its validity (in DEBUG). */
+		s->accum = (s->accum << nbits) | fse_mask_lsb32(incoming, nbits);
+		s->accum_nbits += nbits;
+	}
+	DEBUG_CHECK_INPUT_STREAM_PARAMETERS
+	return 0; /* OK */
+}
+
+/*! @abstract Pull n bits out of the fse stream object. */
+static __always_inline uint64_t fse_in_pull64(fse_in_stream64 *s, fse_bit_count n)
+{
+	uint64_t result;
+
+	s->accum_nbits -= n;
+	result = s->accum >> s->accum_nbits;
+	s->accum = fse_mask_lsb64(s->accum, s->accum_nbits);
+	return result;
+}
+
+/*! @abstract Pull n bits out of the fse stream object. */
+static __always_inline uint32_t fse_in_pull32(fse_in_stream32 *s, fse_bit_count n)
+{
+	uint32_t result;
+
+	s->accum_nbits -= n;
+	result = s->accum >> s->accum_nbits;
+	s->accum = fse_mask_lsb32(s->accum, s->accum_nbits);
+	return result;
+}
+
+/* MARK: - Encode/Decode */
+
+/* Map to 32/64-bit implementations and types for I/O */
+#if FSE_IOSTREAM_64
+
+typedef uint64_t fse_bits;
+typedef fse_out_stream64 fse_out_stream;
+typedef fse_in_stream64 fse_in_stream;
+#define fse_mask_lsb fse_mask_lsb64
+#define fse_extract_bits fse_extract_bits64
+#define fse_out_init fse_out_init64
+#define fse_out_flush fse_out_flush64
+#define fse_out_finish fse_out_finish64
+#define fse_out_push fse_out_push64
+#define fse_in_init fse_in_checked_init64
+#define fse_in_checked_init fse_in_checked_init64
+#define fse_in_flush fse_in_checked_flush64
+#define fse_in_checked_flush fse_in_checked_flush64
+#define fse_in_flush2(_unused, _parameters, _unused2) 0 /* nothing */
+#define fse_in_checked_flush2(_unused, _parameters) /* nothing */
+#define fse_in_pull fse_in_pull64
+
+#else
+
+typedef uint32_t fse_bits;
+typedef fse_out_stream32 fse_out_stream;
+typedef fse_in_stream32 fse_in_stream;
+#define fse_mask_lsb fse_mask_lsb32
+#define fse_extract_bits fse_extract_bits32
+#define fse_out_init fse_out_init32
+#define fse_out_flush fse_out_flush32
+#define fse_out_finish fse_out_finish32
+#define fse_out_push fse_out_push32
+#define fse_in_init fse_in_checked_init32
+#define fse_in_checked_init fse_in_checked_init32
+#define fse_in_flush fse_in_checked_flush32
+#define fse_in_checked_flush fse_in_checked_flush32
+#define fse_in_flush2 fse_in_checked_flush32
+#define fse_in_checked_flush2 fse_in_checked_flush32
+#define fse_in_pull fse_in_pull32
+
+#endif
+
+/*! @abstract  Entry for one state in the decoder table (32b). */
+typedef struct { /* DO NOT REORDER THE FIELDS */
+	int8_t k; /* Number of bits to read */
+	uint8_t symbol; /* Emitted symbol */
+	int16_t delta; /* Signed increment used to compute next state (+bias) */
+} fse_decoder_entry;
+
+/*! @abstract  Entry for one state in the value decoder table (64b). */
+struct fse_value_decoder_entry { /* DO NOT REORDER THE FIELDS */
+	uint8_t total_bits; /* state bits + extra value bits = shift for next decode */
+	uint8_t value_bits; /* extra value bits */
+	int16_t delta; /* state base (delta) */
+	int32_t vbase; /* value base */
+};
+
+/*! @abstract Decode and return symbol using the decoder table, and update
+ *  \c *pstate, \c in.
+ *  @note The caller must ensure we have enough bits available in the input
+ *  stream accumulator.
+ */
+static __always_inline uint8_t fse_decode(fse_state *__restrict pstate,
+					  const int32_t *__restrict decoder_table,
+					  fse_in_stream *__restrict in)
+{
+	int32_t e = decoder_table[*pstate];
+
+	/* Update state from K bits of input + DELTA */
+	*pstate = (fse_state)(e >> 16) + (fse_state)fse_in_pull(in, e & 0xff);
+
+	/* Return the symbol for this state */
+	return fse_extract_bits(e, 8, 8); /* symbol */
+}
+
+/*! @abstract Decode and return value using the decoder table, and update \c
+ *  *pstate, \c in.
+ * \c value_decoder_table[nstates]
+ * @note The caller must ensure we have enough bits available in the input
+ * stream accumulator.
+ */
+static __always_inline int32_t fse_value_decode(fse_state *__restrict pstate,
+						const struct fse_value_decoder_entry *value_decoder_table,
+						fse_in_stream *__restrict in)
+{
+	struct fse_value_decoder_entry entry = value_decoder_table[*pstate];
+	uint32_t state_and_value_bits = (uint32_t)fse_in_pull(in, entry.total_bits);
+	*pstate = (fse_state)(entry.delta + (state_and_value_bits >> entry.value_bits));
+	return (int32_t)(entry.vbase + fse_mask_lsb(state_and_value_bits, entry.value_bits));
+}
+
+/* MARK: - Tables */
+
+/*
+ * IMPORTANT: To properly decode an FSE encoded stream, both encoder/decoder
+ * tables shall be initialized with the same parameters, including the
+ * FREQ[NSYMBOL] array.
+ */
+
+/*! @abstract Sanity check on frequency table, verify sum of \c freq
+ *  is <= \c number_of_states.
+ */
+static __always_inline int fse_check_freq(const uint16_t *freq_table, const size_t table_size,
+					  const size_t number_of_states)
+{
+	size_t sum_of_freq = 0;
+	int i;
+
+	for (i = 0; i < table_size; i++)
+		sum_of_freq += freq_table[i];
+	return (sum_of_freq > number_of_states) ? -1 : 0;
+}
+
+/*! @abstract Initialize decoder table \c t[nstates].
+ *
+ * @param nstates
+ * sum \c freq[i]; the number of states (a power of 2).
+ *
+ * @param nsymbols
+ * the number of symbols.
+ *
+ * @param feq[nsymbols]
+ * a normalized histogram of symbol frequencies, with \c freq[i] >= 0.
+ * Some symbols may have a 0 frequency. In that case they should not be
+ * present in the data.
+ *
+ * @return 0 if OK.
+ * @return -1 on failure.
+ */
+int fse_init_decoder_table(int nstates, int nsymbols, const uint16_t *__restrict freq,
+			   int32_t *__restrict t);
+
+/*! @abstract Initialize value decoder table \c t[nstates].
+ *
+ * @param nstates
+ * sum \cfreq[i]; the number of states (a power of 2).
+ *
+ * @param nsymbols
+ * the number of symbols.
+ *
+ * @param freq[nsymbols]
+ * a normalized histogram of symbol frequencies, with \c freq[i] >= 0.
+ * \c symbol_vbits[nsymbols] and \c symbol_vbase[nsymbols] are the number of
+ * value bits to read and the base value for each symbol.
+ * Some symbols may have a 0 frequency.  In that case they should not be
+ * present in the data.
+ */
+void fse_init_value_decoder_table(int nstates, int nsymbols, const uint16_t *__restrict freq,
+				  const uint8_t *__restrict symbol_vbits,
+				  const int32_t *__restrict symbol_vbase,
+				  struct fse_value_decoder_entry *__restrict t);
+
+#endif /* LZFSE_FSE_H */
diff --git a/fs/apfs/lzfse/lzfse_internal.h b/fs/apfs/lzfse/lzfse_internal.h
new file mode 100644
index 000000000..148748082
--- /dev/null
+++ b/fs/apfs/lzfse/lzfse_internal.h
@@ -0,0 +1,482 @@
+/* SPDX-License-Identifier: BSD-3-Clause */
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef LZFSE_INTERNAL_H
+#define LZFSE_INTERNAL_H
+
+#include "lzfse_fse.h"
+#include <linux/limits.h>
+#include <linux/stddef.h>
+
+/*
+ * Throughout LZFSE we refer to "L", "M" and "D"; these will always appear as
+ * a triplet, and represent a "usual" LZ-style literal and match pair. "L"
+ * is the number of literal bytes, "M" is the number of match bytes, and "D"
+ * is the match "distance"; the distance in bytes between the current pointer
+ * and the start of the match.
+ */
+#define LZFSE_ENCODE_HASH_VALUES (1 << LZFSE_ENCODE_HASH_BITS)
+#define LZFSE_ENCODE_L_SYMBOLS 20
+#define LZFSE_ENCODE_M_SYMBOLS 20
+#define LZFSE_ENCODE_D_SYMBOLS 64
+#define LZFSE_ENCODE_LITERAL_SYMBOLS 256
+#define LZFSE_ENCODE_L_STATES 64
+#define LZFSE_ENCODE_M_STATES 64
+#define LZFSE_ENCODE_D_STATES 256
+#define LZFSE_ENCODE_LITERAL_STATES 1024
+#define LZFSE_MATCHES_PER_BLOCK 10000
+#define LZFSE_LITERALS_PER_BLOCK (4 * LZFSE_MATCHES_PER_BLOCK)
+#define LZFSE_DECODE_LITERALS_PER_BLOCK (4 * LZFSE_DECODE_MATCHES_PER_BLOCK)
+
+/*
+ * LZFSE internal status. These values are used by internal LZFSE routines
+ * as return codes. There should not be any good reason to change their
+ * values; it is plausible that additional codes might be added in the
+ * future.
+ */
+#define LZFSE_STATUS_OK 0
+#define LZFSE_STATUS_SRC_EMPTY -1
+#define LZFSE_STATUS_DST_FULL -2
+#define LZFSE_STATUS_ERROR -3
+
+/*
+ * Type representing an offset between elements in a buffer. On 64-bit
+ * systems, this is stored in a 64-bit container to avoid extra sign-
+ * extension operations in addressing arithmetic, but the value is always
+ * representable as a 32-bit signed value in LZFSE's usage.
+ */
+#if defined(_M_AMD64) || defined(__x86_64__) || defined(__arm64__)
+typedef int64_t lzfse_offset;
+#else
+typedef int32_t lzfse_offset;
+#endif
+
+typedef uint64_t uintmax_t;
+
+/*! @abstract Decoder state object for lzfse compressed blocks. */
+typedef struct {
+	/* Number of matches remaining in the block. */
+	uint32_t n_matches;
+	/* Number of bytes used to encode L, M, D triplets for the block. */
+	uint32_t n_lmd_payload_bytes;
+	/* Pointer to the next literal to emit. */
+	const uint8_t *current_literal;
+	/*
+	 * L, M, D triplet for the match currently being emitted. This is used
+	 * only if we need to restart after reaching the end of the destination
+	 * buffer in the middle of a literal or match.
+	 */
+	int32_t l_value, m_value, d_value;
+	/* FSE stream object. */
+	fse_in_stream lmd_in_stream;
+	/*
+	 * Offset of L,M,D encoding in the input buffer. Because we read through
+	 * an FSE stream *backwards* while decoding, this is decremented as we
+	 * move through a block.
+	 */
+	uint32_t lmd_in_buf;
+	/* The current state of the L, M, and D FSE decoders. */
+	uint16_t l_state, m_state, d_state;
+	/*
+	 * Internal FSE decoder tables for the current block. These have
+	 * alignment forced to 8 bytes to guarantee that a single state's
+	 * entry cannot span two cachelines.
+	 */
+	struct fse_value_decoder_entry l_decoder[LZFSE_ENCODE_L_STATES] __aligned(8);
+	struct fse_value_decoder_entry m_decoder[LZFSE_ENCODE_M_STATES] __aligned(8);
+	struct fse_value_decoder_entry d_decoder[LZFSE_ENCODE_D_STATES] __aligned(8);
+	int32_t literal_decoder[LZFSE_ENCODE_LITERAL_STATES];
+	/*
+	 * The literal stream for the block, plus padding to allow for faster
+	 * copy operations.
+	 */
+	uint8_t literals[LZFSE_LITERALS_PER_BLOCK + 64];
+} lzfse_compressed_block_decoder_state;
+
+/* Decoder state object for uncompressed blocks. */
+typedef struct {
+	uint32_t n_raw_bytes;
+} uncompressed_block_decoder_state;
+
+/*! @abstract Decoder state object for lzvn-compressed blocks. */
+typedef struct {
+	uint32_t n_raw_bytes;
+	uint32_t n_payload_bytes;
+	uint32_t d_prev;
+} lzvn_compressed_block_decoder_state;
+
+/*! @abstract Decoder state object. */
+typedef struct {
+	/*
+	 * Pointer to next byte to read from source buffer (this is advanced as
+	 * we decode; src_begin describe the buffer and do not change).
+	 */
+	const uint8_t *src;
+	/* Pointer to first byte of source buffer. */
+	const uint8_t *src_begin;
+	/* Pointer to one byte past the end of the source buffer. */
+	const uint8_t *src_end;
+	/*
+	 * Pointer to the next byte to write to destination buffer (this is
+	 * advanced as we decode; dst_begin and dst_end describe the buffer and
+	 * do not change).
+	 */
+	uint8_t *dst;
+	/* Pointer to first byte of destination buffer. */
+	uint8_t *dst_begin;
+	/* Pointer to one byte past the end of the destination buffer. */
+	uint8_t *dst_end;
+	/* 1 if we have reached the end of the stream, 0 otherwise. */
+	int end_of_stream;
+	/*
+	 * magic number of the current block if we are within a block,
+	 * LZFSE_NO_BLOCK_MAGIC otherwise.
+	 */
+	uint32_t block_magic;
+	lzfse_compressed_block_decoder_state compressed_lzfse_block_state;
+	lzvn_compressed_block_decoder_state compressed_lzvn_block_state;
+	uncompressed_block_decoder_state uncompressed_block_state;
+} lzfse_decoder_state;
+
+/* MARK: - Block header objects */
+
+#define LZFSE_NO_BLOCK_MAGIC 0x00000000 /* 0    (invalid) */
+#define LZFSE_ENDOFSTREAM_BLOCK_MAGIC 0x24787662 /* bvx$ (end of stream) */
+#define LZFSE_UNCOMPRESSED_BLOCK_MAGIC 0x2d787662 /* bvx- (raw data) */
+#define LZFSE_COMPRESSEDV1_BLOCK_MAGIC 0x31787662 /* bvx1 (lzfse compressed, uncompressed tables) */
+#define LZFSE_COMPRESSEDV2_BLOCK_MAGIC 0x32787662 /* bvx2 (lzfse compressed, compressed tables) */
+#define LZFSE_COMPRESSEDLZVN_BLOCK_MAGIC 0x6e787662 /* bvxn (lzvn compressed) */
+
+/*! @abstract Uncompressed block header in encoder stream. */
+typedef struct {
+	/* Magic number, always LZFSE_UNCOMPRESSED_BLOCK_MAGIC. */
+	uint32_t magic;
+	/* Number of raw bytes in block. */
+	uint32_t n_raw_bytes;
+} uncompressed_block_header;
+
+/*! @abstract Compressed block header with uncompressed tables. */
+typedef struct {
+	/* Magic number, always LZFSE_COMPRESSEDV1_BLOCK_MAGIC. */
+	uint32_t magic;
+	/* Number of decoded (output) bytes in block. */
+	uint32_t n_raw_bytes;
+	/* Number of encoded (source) bytes in block. */
+	uint32_t n_payload_bytes;
+	/* Number of literal bytes output by block (*not* the number of literals). */
+	uint32_t n_literals;
+	/* Number of matches in block (which is also the number of literals). */
+	uint32_t n_matches;
+	/* Number of bytes used to encode literals. */
+	uint32_t n_literal_payload_bytes;
+	/* Number of bytes used to encode matches. */
+	uint32_t n_lmd_payload_bytes;
+
+	/*
+	 * Final encoder states for the block, which will be the initial states
+	 * for the decoder:
+	 */
+	/* Final accum_nbits for literals stream. */
+	int32_t literal_bits;
+	/*
+	 * There are four interleaved streams of literals, so there are four
+	 * final states.
+	 */
+	uint16_t literal_state[4];
+	/* accum_nbits for the l, m, d stream. */
+	int32_t lmd_bits;
+	/* Final L (literal length) state. */
+	uint16_t l_state;
+	/* Final M (match length) state. */
+	uint16_t m_state;
+	/* Final D (match distance) state. */
+	uint16_t d_state;
+
+	/*
+	 * Normalized frequency tables for each stream. Sum of values in each
+	 * array is the number of states.
+	 */
+	uint16_t l_freq[LZFSE_ENCODE_L_SYMBOLS];
+	uint16_t m_freq[LZFSE_ENCODE_M_SYMBOLS];
+	uint16_t d_freq[LZFSE_ENCODE_D_SYMBOLS];
+	uint16_t literal_freq[LZFSE_ENCODE_LITERAL_SYMBOLS];
+} lzfse_compressed_block_header_v1;
+
+/*! @abstract Compressed block header with compressed tables. Note that because
+ *  freq[] is compressed, the structure-as-stored-in-the-stream is *truncated*;
+ *  we only store the used bytes of freq[]. This means that some extra care must
+ *  be taken when reading one of these headers from the stream.
+ */
+typedef struct {
+	/* Magic number, always LZFSE_COMPRESSEDV2_BLOCK_MAGIC. */
+	uint32_t magic;
+	/* Number of decoded (output) bytes in block. */
+	uint32_t n_raw_bytes;
+	/*
+	 * The fields n_payload_bytes ... d_state from the
+	 * lzfse_compressed_block_header_v1 object are packed into three 64-bit
+	 * fields in the compressed header, as follows:
+	 *
+	 *    offset  bits  value
+	 *    0       20    n_literals
+	 *    20      20    n_literal_payload_bytes
+	 *    40      20    n_matches
+	 *    60      3     literal_bits
+	 *    63      1     --- unused ---
+	 *
+	 *    0       10    literal_state[0]
+	 *    10      10    literal_state[1]
+	 *    20      10    literal_state[2]
+	 *    30      10    literal_state[3]
+	 *    40      20    n_lmd_payload_bytes
+	 *    60      3     lmd_bits
+	 *    63      1     --- unused ---
+	 *
+	 *    0       32    header_size (total header size in bytes; this does not
+	 *                  correspond to a field in the uncompressed header version,
+	 *                  but is required; we wouldn't know the size of the
+	 *                  compresssed header otherwise.
+	 *    32      10    l_state
+	 *    42      10    m_state
+	 *    52      10    d_state
+	 *    62      2     --- unused ---
+	 */
+	uint64_t packed_fields[3];
+	/*
+	 * Variable size freq tables, using a Huffman-style fixed encoding.
+	 * Size allocated here is an upper bound (all values stored on 16 bits).
+	 */
+	uint8_t freq[2 * (LZFSE_ENCODE_L_SYMBOLS + LZFSE_ENCODE_M_SYMBOLS + LZFSE_ENCODE_D_SYMBOLS +
+			  LZFSE_ENCODE_LITERAL_SYMBOLS)];
+} __packed __aligned(1) lzfse_compressed_block_header_v2;
+
+/*! @abstract LZVN compressed block header. */
+typedef struct {
+	/* Magic number, always LZFSE_COMPRESSEDLZVN_BLOCK_MAGIC. */
+	uint32_t magic;
+	/* Number of decoded (output) bytes. */
+	uint32_t n_raw_bytes;
+	/* Number of encoded (source) bytes. */
+	uint32_t n_payload_bytes;
+} lzvn_compressed_block_header;
+
+int lzfse_decode(lzfse_decoder_state *s);
+size_t lzvn_decode_scratch_size(void);
+size_t lzvn_decode_buffer(void *__restrict dst, size_t dst_size, const void *__restrict src,
+			  size_t src_size);
+
+/*! @abstract Signed offset in buffers, stored on either 32 or 64 bits. */
+#if defined(_M_AMD64) || defined(__x86_64__) || defined(__arm64__)
+typedef int64_t lzvn_offset;
+#else
+typedef int32_t lzvn_offset;
+#endif
+
+/* MARK: - LZFSE utility functions */
+
+/*! @abstract Load bytes from memory location SRC. */
+static __always_inline uint16_t load2(const void *ptr)
+{
+	uint16_t data;
+
+	memcpy(&data, ptr, sizeof(data));
+	return data;
+}
+
+static __always_inline uint32_t load4(const void *ptr)
+{
+	uint32_t data;
+
+	memcpy(&data, ptr, sizeof(data));
+	return data;
+}
+
+static __always_inline uint64_t load8(const void *ptr)
+{
+	uint64_t data;
+
+	memcpy(&data, ptr, sizeof(data));
+	return data;
+}
+
+/*! @abstract Store bytes to memory location DST. */
+static __always_inline void store2(void *ptr, uint16_t data)
+{
+	memcpy(ptr, &data, sizeof(data));
+}
+
+static __always_inline void store4(void *ptr, uint32_t data)
+{
+	memcpy(ptr, &data, sizeof(data));
+}
+
+static __always_inline void store8(void *ptr, uint64_t data)
+{
+	memcpy(ptr, &data, sizeof(data));
+}
+
+/*! @abstract Load+store bytes from locations SRC to DST. Not intended for use
+ * with overlapping buffers. Note that for LZ-style compression, you need
+ * copies to behave like naive memcpy( ) implementations do, splatting the
+ * leading sequence if the buffers overlap. This copy does not do that, so
+ * should not be used with overlapping buffers.
+ */
+static __always_inline void copy8(void *dst, const void *src)
+{
+	store8(dst, load8(src));
+}
+static __always_inline void copy16(void *dst, const void *src)
+{
+	uint64_t m0 = load8(src);
+	uint64_t m1 = load8((const unsigned char *)src + 8);
+
+	store8(dst, m0);
+	store8((unsigned char *)dst + 8, m1);
+}
+
+/*
+ * ===============================================================
+ * Bitfield Operations
+ */
+
+/*! @abstract Extracts \p width bits from \p container, starting with \p lsb; if
+ * we view \p container as a bit array, we extract \c container[lsb:lsb+width].
+ */
+static __always_inline uintmax_t extract(uintmax_t container, unsigned int lsb, unsigned int width)
+{
+	static const size_t container_width = sizeof(container) * 8;
+
+	if (width == container_width)
+		return container;
+	return (container >> lsb) & (((uintmax_t)1 << width) - 1);
+}
+
+/*! @abstract Inserts \p width bits from \p data into \p container, starting with \p lsb.
+ *  Viewed as bit arrays, the operations is:
+ * @code
+ * container[:lsb] is unchanged
+ * container[lsb:lsb+width] <-- data[0:width]
+ * container[lsb+width:] is unchanged
+ * @endcode
+ */
+static __always_inline uintmax_t insert(uintmax_t container, uintmax_t data, unsigned int lsb,
+					unsigned int width)
+{
+	static const size_t container_width = sizeof(container) * 8;
+	uintmax_t mask;
+
+	if (width == container_width)
+		return container;
+	mask = ((uintmax_t)1 << width) - 1;
+	return (container & ~(mask << lsb)) | (data & mask) << lsb;
+}
+
+/*! @abstract Perform sanity checks on the values of lzfse_compressed_block_header_v1.
+ * Test that the field values are in the allowed limits, verify that the
+ * frequency tables sum to value less than total number of states.
+ * @return 0 if all tests passed.
+ * @return negative error code with 1 bit set for each failed test.
+ */
+static __always_inline int
+lzfse_check_block_header_v1(const lzfse_compressed_block_header_v1 *header)
+{
+	int tests_results = 0;
+	uint16_t literal_state[4];
+	int res;
+
+	tests_results = tests_results |
+			((header->magic == LZFSE_COMPRESSEDV1_BLOCK_MAGIC) ? 0 : (1 << 0));
+	tests_results = tests_results |
+			((header->n_literals <= LZFSE_LITERALS_PER_BLOCK) ? 0 : (1 << 1));
+	tests_results = tests_results |
+			((header->n_matches <= LZFSE_MATCHES_PER_BLOCK) ? 0 : (1 << 2));
+
+	memcpy(literal_state, header->literal_state, sizeof(uint16_t) * 4);
+
+	tests_results = tests_results |
+			((literal_state[0] < LZFSE_ENCODE_LITERAL_STATES) ? 0 : (1 << 3));
+	tests_results = tests_results |
+			((literal_state[1] < LZFSE_ENCODE_LITERAL_STATES) ? 0 : (1 << 4));
+	tests_results = tests_results |
+			((literal_state[2] < LZFSE_ENCODE_LITERAL_STATES) ? 0 : (1 << 5));
+	tests_results = tests_results |
+			((literal_state[3] < LZFSE_ENCODE_LITERAL_STATES) ? 0 : (1 << 6));
+
+	tests_results = tests_results | ((header->l_state < LZFSE_ENCODE_L_STATES) ? 0 : (1 << 7));
+	tests_results = tests_results | ((header->m_state < LZFSE_ENCODE_M_STATES) ? 0 : (1 << 8));
+	tests_results = tests_results | ((header->d_state < LZFSE_ENCODE_D_STATES) ? 0 : (1 << 9));
+
+	res = fse_check_freq(header->l_freq, LZFSE_ENCODE_L_SYMBOLS, LZFSE_ENCODE_L_STATES);
+	tests_results = tests_results | ((res == 0) ? 0 : (1 << 10));
+	res = fse_check_freq(header->m_freq, LZFSE_ENCODE_M_SYMBOLS, LZFSE_ENCODE_M_STATES);
+	tests_results = tests_results | ((res == 0) ? 0 : (1 << 11));
+	res = fse_check_freq(header->d_freq, LZFSE_ENCODE_D_SYMBOLS, LZFSE_ENCODE_D_STATES);
+	tests_results = tests_results | ((res == 0) ? 0 : (1 << 12));
+	res = fse_check_freq(header->literal_freq, LZFSE_ENCODE_LITERAL_SYMBOLS,
+			     LZFSE_ENCODE_LITERAL_STATES);
+	tests_results = tests_results | ((res == 0) ? 0 : (1 << 13));
+
+	if (tests_results) {
+		/* each 1 bit is a test that failed (except for the sign bit) */
+		return tests_results | 0x80000000;
+	}
+
+	return 0; /* OK */
+}
+
+/*! @abstract The L, M, D data streams are all encoded as a "base" value, which is
+ * FSE-encoded, and an "extra bits" value, which is the difference between
+ * value and base, and is simply represented as a raw bit value (because it
+ * is the low-order bits of a larger number, not much entropy can be
+ * extracted from these bits by more complex encoding schemes). The following
+ * tables represent the number of low-order bits to encode separately and the
+ * base values for each of L, M, and D.
+ *
+ * @note The inverse tables for mapping the other way are significantly larger.
+ * Those tables have been split out to lzfse_encode_tables.h in order to keep
+ * this file relatively small.
+ */
+static const uint8_t l_extra_bits[LZFSE_ENCODE_L_SYMBOLS] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+							      0, 0, 0, 0, 0, 0, 2, 3, 5, 8 };
+static const int32_t l_base_value[LZFSE_ENCODE_L_SYMBOLS] = {
+	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 20, 28, 60
+};
+static const uint8_t m_extra_bits[LZFSE_ENCODE_M_SYMBOLS] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+							      0, 0, 0, 0, 0, 0, 3, 5, 8, 11 };
+static const int32_t m_base_value[LZFSE_ENCODE_M_SYMBOLS] = { 0,  1,  2,  3,  4,  5,  6,
+							      7,  8,  9,  10, 11, 12, 13,
+							      14, 15, 16, 24, 56, 312 };
+static const uint8_t d_extra_bits[LZFSE_ENCODE_D_SYMBOLS] = {
+	0,  0,	0,  0,	1,  1,	1,  1,	2,  2,	2,  2,	3,  3,	3,  3,	4,  4,	4,  4,	5,  5,
+	5,  5,	6,  6,	6,  6,	7,  7,	7,  7,	8,  8,	8,  8,	9,  9,	9,  9,	10, 10, 10, 10,
+	11, 11, 11, 11, 12, 12, 12, 12, 13, 13, 13, 13, 14, 14, 14, 14, 15, 15, 15, 15
+};
+static const int32_t d_base_value[LZFSE_ENCODE_D_SYMBOLS] = {
+	0,     1,     2,     3,	    4,	    6,	    8,	    10,	    12,	   16,	  20,
+	24,    28,    36,    44,    52,	    60,	    76,	    92,	    108,   124,	  156,
+	188,   220,   252,   316,   380,    444,    508,    636,    764,   892,	  1020,
+	1276,  1532,  1788,  2044,  2556,   3068,   3580,   4092,   5116,  6140,  7164,
+	8188,  10236, 12284, 14332, 16380,  20476,  24572,  28668,  32764, 40956, 49148,
+	57340, 65532, 81916, 98300, 114684, 131068, 163836, 196604, 229372
+};
+
+#endif /* LZFSE_INTERNAL_H */
diff --git a/fs/apfs/lzfse/lzvn_decode_base.c b/fs/apfs/lzfse/lzvn_decode_base.c
new file mode 100644
index 000000000..f2cb2afce
--- /dev/null
+++ b/fs/apfs/lzfse/lzvn_decode_base.c
@@ -0,0 +1,487 @@
+// SPDX-License-Identifier: BSD-3-Clause
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* LZVN low-level decoder */
+
+#include <linux/compiler.h>
+#include <linux/version.h>
+#include "lzvn_decode_base.h"
+
+/* Older kernel versions will still get an objtool warning here */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 3, 0)
+#define __annotate_jump_table
+#endif
+
+/*
+ * Both the source and destination buffers are represented by a pointer and
+ * a length; they are *always* updated in concert using this macro; however
+ * many bytes the pointer is advanced, the length is decremented by the same
+ * amount. Thus, pointer + length always points to the byte one past the end
+ * of the buffer.
+ */
+#define PTR_LEN_INC(_pointer, _length, _increment) (_pointer += _increment, _length -= _increment)
+
+/*
+ * Update state with current positions and distance, corresponding to the
+ * beginning of an instruction in both streams
+ */
+#define UPDATE_GOOD (state->src = src_ptr, state->dst = dst_ptr, state->d_prev = D)
+
+void lzvn_decode(lzvn_decoder_state *state)
+{
+	/* Jump table for all instructions */
+	static const void *opc_tbl[256] __annotate_jump_table = {
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&eos,   &&lrg_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&nop,   &&lrg_d, &&sml_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&nop,   &&lrg_d, &&sml_d, &&sml_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&udef,  &&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d,
+		&&sml_d, &&sml_d, &&udef,  &&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d,
+		&&sml_d, &&udef,  &&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d,
+		&&udef,	 &&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&udef,
+		&&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&sml_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&sml_d, &&sml_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d,
+		&&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&udef,  &&udef,  &&udef,  &&udef,	&&udef,
+		&&udef,	 &&udef,  &&udef,  &&udef,  &&udef,  &&udef,  &&udef,  &&udef,	&&udef,
+		&&udef,	 &&udef,  &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d,
+		&&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&sml_d,
+		&&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d, &&lrg_d, &&med_d, &&med_d,
+		&&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d,
+		&&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d,
+		&&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d, &&med_d,
+		&&med_d, &&med_d, &&med_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d,
+		&&pre_d, &&lrg_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&sml_d, &&pre_d,
+		&&lrg_d, &&udef,  &&udef,  &&udef,  &&udef,  &&udef,  &&udef,  &&udef,	&&udef,
+		&&udef,	 &&udef,  &&udef,  &&udef,  &&udef,  &&udef,  &&udef,  &&udef,	&&lrg_l,
+		&&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l,
+		&&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l, &&sml_l, &&lrg_m, &&sml_m, &&sml_m,
+		&&sml_m, &&sml_m, &&sml_m, &&sml_m, &&sml_m, &&sml_m, &&sml_m, &&sml_m, &&sml_m,
+		&&sml_m, &&sml_m, &&sml_m, &&sml_m
+	};
+	size_t src_len = state->src_end - state->src;
+	size_t dst_len = state->dst_end - state->dst;
+	const unsigned char *src_ptr = state->src;
+	unsigned char *dst_ptr = state->dst;
+	size_t D = state->d_prev;
+	size_t M;
+	size_t L;
+	size_t opc_len;
+	unsigned char opc;
+	uint16_t opc23;
+
+	if (src_len == 0 || dst_len == 0)
+		return; /* empty buffer */
+
+	/* Do we have a partially expanded match saved in state? */
+	if (state->L != 0 || state->M != 0) {
+		L = state->L;
+		M = state->M;
+		D = state->D;
+		opc_len = 0; /* we already skipped the op */
+		state->L = state->M = state->D = 0;
+		if (M == 0)
+			goto copy_literal;
+		if (L == 0)
+			goto copy_match;
+		goto copy_literal_and_match;
+	}
+
+	opc = src_ptr[0];
+
+	goto *opc_tbl[opc];
+/*
+ * ===============================================================
+ * These four opcodes (sml_d, med_d, lrg_d, and pre_d) encode both a
+ * literal and a match. The bulk of their implementations are shared;
+ * each label here only does the work of setting the opcode length (not
+ * including any literal bytes), and extracting the literal length, match
+ * length, and match distance (except in pre_d). They then jump into the
+ * shared implementation to actually output the literal and match bytes.
+ *
+ * No error checking happens in the first stage, except for ensuring that
+ * the source has enough length to represent the full opcode before
+ * reading past the first byte.
+ */
+sml_d:
+	UPDATE_GOOD;
+	/*
+	 * "small distance": This opcode has the structure LLMMMDDD DDDDDDDD
+	 * LITERAL where the length of literal (0-3 bytes) is encoded by the
+	 * high 2 bits of the first byte. We first extract the literal length so
+	 * we know how long the opcode is, then check that the source can hold
+	 * both this opcode and at least one byte of the next (because any valid
+	 * input stream must be terminated with an eos token).
+	 */
+	opc_len = 2;
+	L = (size_t)extract(opc, 6, 2);
+	M = (size_t)extract(opc, 3, 3) + 3;
+	/*
+	 * We need to ensure that the source buffer is long enough that we can
+	 * safely read this entire opcode, the literal that follows, and the
+	 * first byte of the next opcode.  Once we satisfy this requirement, we
+	 * can safely unpack the match distance. A check similar to this one is
+	 * present in all the opcode implementations.
+	 */
+	if (src_len <= opc_len + L)
+		return; /* source truncated */
+	D = (size_t)extract(opc, 0, 3) << 8 | src_ptr[1];
+	goto copy_literal_and_match;
+
+med_d:
+	UPDATE_GOOD;
+	/*
+	 * "medium distance": This is a minor variant of the "small distance"
+	 * encoding, where we will now use two extra bytes instead of one to
+	 * encode the restof the match length and distance. This allows an extra
+	 * two bits for the match length, and an extra three bits for the match
+	 * distance. The full structure of the opcode is
+	 * 101LLMMM DDDDDDMM DDDDDDDD LITERAL.
+	 */
+	opc_len = 3;
+	L = (size_t)extract(opc, 3, 2);
+	if (src_len <= opc_len + L)
+		return; /* source truncated */
+	opc23 = load2(&src_ptr[1]);
+	M = (size_t)((extract(opc, 0, 3) << 2 | extract(opc23, 0, 2)) + 3);
+	D = (size_t)extract(opc23, 2, 14);
+	goto copy_literal_and_match;
+
+lrg_d:
+	UPDATE_GOOD;
+	/*
+	 * "large distance": This is another variant of the "small distance"
+	 * encoding, where we will now use two extra bytes to encode the match
+	 * distance, which allows distances up to 65535 to be represented. The
+	 * full structure of the opcode is LLMMM111 DDDDDDDD DDDDDDDD LITERAL.
+	 */
+	opc_len = 3;
+	L = (size_t)extract(opc, 6, 2);
+	M = (size_t)extract(opc, 3, 3) + 3;
+	if (src_len <= opc_len + L)
+		return; /* source truncated */
+	D = load2(&src_ptr[1]);
+	goto copy_literal_and_match;
+
+pre_d:
+	UPDATE_GOOD;
+	/*
+	 * "previous distance": This opcode has the structure LLMMM110, where
+	 * the length of the literal (0-3 bytes) is encoded by the high 2 bits
+	 * of the first byte. We first extract the literal length so we know how
+	 * long the opcode is, then check that the source can hold both this
+	 * opcode and at least one byte of the next (because any valid input
+	 * stream must be terminated with an eos token).
+	 */
+	opc_len = 1;
+	L = (size_t)extract(opc, 6, 2);
+	M = (size_t)extract(opc, 3, 3) + 3;
+	if (src_len <= opc_len + L)
+		return; /* source truncated */
+	goto copy_literal_and_match;
+
+copy_literal_and_match:
+	/*
+	 * Common implementation of writing data for opcodes that have both a
+	 * literal and a match. We begin by advancing the source pointer past
+	 * the opcode, so that it points at the first literal byte (if L
+	 * is non-zero; otherwise it points at the next opcode).
+	 */
+	PTR_LEN_INC(src_ptr, src_len, opc_len);
+	/* Now we copy the literal from the source pointer to the destination */
+	if (__builtin_expect(dst_len >= 4 && src_len >= 4, 1)) {
+		/*
+		 * The literal is 0-3 bytes; if we are not near the end of the
+		 * buffer, we can safely just do a 4 byte copy (which is
+		 * guaranteed to cover the complete literal, and may include
+		 * some other bytes as well).
+		 */
+		store4(dst_ptr, load4(src_ptr));
+	} else if (L <= dst_len) {
+		/*
+		 * We are too close to the end of either the input or output
+		 * stream to be able to safely use a four-byte copy, but we will
+		 * not exhaust either stream (we already know that the source
+		 * will not be exhausted from checks in the individual opcode
+		 * implementations, and we just tested that dst_len > L). Thus,
+		 * we need to do a byte-by-byte copy of the literal. This is
+		 * slow, but it can only ever happen near the very end of a
+		 * buffer, so it is not an important case to optimize.
+		 */
+		size_t i;
+
+		for (i = 0; i < L; ++i)
+			dst_ptr[i] = src_ptr[i];
+	} else {
+		/* Destination truncated: fill DST, and store partial match */
+
+		/* Copy partial literal */
+		size_t i;
+
+		for (i = 0; i < dst_len; ++i)
+			dst_ptr[i] = src_ptr[i];
+		/* Save state */
+		state->src = src_ptr + dst_len;
+		state->dst = dst_ptr + dst_len;
+		state->L = L - dst_len;
+		state->M = M;
+		state->D = D;
+		return; /* destination truncated */
+	}
+	/*
+	 * Having completed the copy of the literal, we advance both the source
+	 * and destination pointers by the number of literal bytes.
+	 */
+	PTR_LEN_INC(dst_ptr, dst_len, L);
+	PTR_LEN_INC(src_ptr, src_len, L);
+	/*
+	 * Check if the match distance is valid; matches must not reference
+	 * bytes that preceed the start of the output buffer, nor can the match
+	 * distance be zero.
+	 */
+	if (D > dst_ptr - state->dst_begin || D == 0)
+		goto invalid_match_distance;
+copy_match:
+	/*
+	 * Now we copy the match from dst_ptr - D to dst_ptr. It is important to
+	 * keep in mind that we may have D < M, in which case the source and
+	 * destination windows overlap in the copy. The semantics of the match
+	 * copy are *not* those of memmove( ); if the buffers overlap it needs
+	 * to behave as though we were copying byte-by-byte in increasing
+	 * address order. If, for example, D is 1, the copy operation is
+	 * equivalent to:
+	 *
+	 *      memset(dst_ptr, dst_ptr[-1], M);
+	 *
+	 * i.e. it splats the previous byte. This means that we need to be very
+	 * careful about using wide loads or stores to perform the copy
+	 * operation.
+	 */
+	if (__builtin_expect(dst_len >= M + 7 && D >= 8, 1)) {
+		/*
+		 * We are not near the end of the buffer, and the match distance
+		 * is at least eight. Thus, we can safely loop using eight byte
+		 * copies. The last of these may slop over the intended end of
+		 * the match, but this is OK because we know we have a safety
+		 * bound away from the end of the destination buffer.
+		 */
+		size_t i;
+
+		for (i = 0; i < M; i += 8)
+			store8(&dst_ptr[i], load8(&dst_ptr[i - D]));
+	} else if (M <= dst_len) {
+		/*
+		 * Either the match distance is too small, or we are too close
+		 * to the end of the buffer to safely use eight byte copies.
+		 * Fall back on a simple byte-by-byte implementation.
+		 */
+		size_t i;
+
+		for (i = 0; i < M; ++i)
+			dst_ptr[i] = dst_ptr[i - D];
+	} else {
+		/* Destination truncated: fill DST, and store partial match */
+
+		/* Copy partial match */
+		size_t i;
+
+		for (i = 0; i < dst_len; ++i)
+			dst_ptr[i] = dst_ptr[i - D];
+		/* Save state */
+		state->src = src_ptr;
+		state->dst = dst_ptr + dst_len;
+		state->L = 0;
+		state->M = M - dst_len;
+		state->D = D;
+		return; /* destination truncated */
+	}
+	/*
+	 * Update the destination pointer and length to account for the bytes
+	 * written by the match, then load the next opcode byte and branch to
+	 * the appropriate implementation.
+	 */
+	PTR_LEN_INC(dst_ptr, dst_len, M);
+	opc = src_ptr[0];
+	goto *opc_tbl[opc];
+
+/*
+ * ===============================================================
+ * Opcodes representing only a match (no literal).
+ *  These two opcodes (lrg_m and sml_m) encode only a match. The match
+ *  distance is carried over from the previous opcode, so all they need
+ *  to encode is the match length. We are able to reuse the match copy
+ *  sequence from the literal and match opcodes to perform the actual
+ *  copy implementation.
+ */
+sml_m:
+	UPDATE_GOOD;
+	/*
+	 * "small match": This opcode has no literal, and uses the previous
+	 * match distance (i.e. it encodes only the match length), in a single
+	 * byte as 1111MMMM.
+	 */
+	opc_len = 1;
+	if (src_len <= opc_len)
+		return; /* source truncated */
+	M = (size_t)extract(opc, 0, 4);
+	PTR_LEN_INC(src_ptr, src_len, opc_len);
+	goto copy_match;
+
+lrg_m:
+	UPDATE_GOOD;
+	/*
+	 * "large match": This opcode has no literal, and uses the previous
+	 * match distance (i.e. it encodes only the match length). It is encoded
+	 * in two bytes as 11110000 MMMMMMMM.  Because matches smaller than 16
+	 * bytes can be represented by sml_m, there is an implicit bias of 16 on
+	 * the match length; the representable values are [16,271].
+	 */
+	opc_len = 2;
+	if (src_len <= opc_len)
+		return; /* source truncated */
+	M = src_ptr[1] + 16;
+	PTR_LEN_INC(src_ptr, src_len, opc_len);
+	goto copy_match;
+
+/*
+ * ===============================================================
+ * Opcodes representing only a literal (no match).
+ *  These two opcodes (lrg_l and sml_l) encode only a literal. There is no
+ *  match length or match distance to worry about (but we need to *not*
+ *  touch D, as it must be preserved between opcodes).
+ */
+sml_l:
+	UPDATE_GOOD;
+	/*
+	 * "small literal": This opcode has no match, and encodes only a literal
+	 * of length up to 15 bytes. The format is 1110LLLL LITERAL.
+	 */
+	opc_len = 1;
+	L = (size_t)extract(opc, 0, 4);
+	goto copy_literal;
+
+lrg_l:
+	UPDATE_GOOD;
+	/*
+	 * "large literal": This opcode has no match, and uses the previous
+	 * match distance (i.e. it encodes only the match length). It is encoded
+	 * in two bytes as 11100000 LLLLLLLL LITERAL.  Because literals smaller
+	 * than 16 bytes can be represented by sml_l, there is an implicit bias
+	 * of 16 on the literal length; the representable values are [16,271].
+	 */
+	opc_len = 2;
+	if (src_len <= 2)
+		return; /* source truncated */
+	L = src_ptr[1] + 16;
+	goto copy_literal;
+
+copy_literal:
+	/*
+	 * Check that the source buffer is large enough to hold the complete
+	 * literal and at least the first byte of the next opcode. If so,
+	 * advance the source pointer to point to the first byte of the literal
+	 * and adjust the source length accordingly.
+	 */
+	if (src_len <= opc_len + L)
+		return; /* source truncated */
+	PTR_LEN_INC(src_ptr, src_len, opc_len);
+	/* Now we copy the literal from the source pointer to the destination */
+	if (dst_len >= L + 7 && src_len >= L + 7) {
+		/*
+		 * We are not near the end of the source or destination buffers;
+		 * thus we can safely copy the literal using wide copies,
+		 * without worrying about reading or writing past the end of
+		 * either buffer.
+		 */
+		size_t i;
+
+		for (i = 0; i < L; i += 8)
+			store8(&dst_ptr[i], load8(&src_ptr[i]));
+	} else if (L <= dst_len) {
+		/*
+		 * We are too close to the end of either the input or output
+		 * stream to be able to safely use an eight-byte copy. Instead
+		 * we copy the literal byte-by-byte.
+		 */
+		size_t i;
+
+		for (i = 0; i < L; ++i)
+			dst_ptr[i] = src_ptr[i];
+	} else {
+		/* Destination truncated: fill DST, and store partial match */
+
+		/* Copy partial literal */
+		size_t i;
+
+		for (i = 0; i < dst_len; ++i)
+			dst_ptr[i] = src_ptr[i];
+		/* Save state */
+		state->src = src_ptr + dst_len;
+		state->dst = dst_ptr + dst_len;
+		state->L = L - dst_len;
+		state->M = 0;
+		state->D = D;
+		return; /* destination truncated */
+	}
+	/*
+	 * Having completed the copy of the literal, we advance both the source
+	 * and destination pointers by the number of literal bytes.
+	 */
+	PTR_LEN_INC(dst_ptr, dst_len, L);
+	PTR_LEN_INC(src_ptr, src_len, L);
+	/* Load the first byte of the next opcode, and jump to its implementation */
+	opc = src_ptr[0];
+	goto *opc_tbl[opc];
+
+/*
+ * ===============================================================
+ * Other opcodes
+ */
+nop:
+	UPDATE_GOOD;
+	opc_len = 1;
+	if (src_len <= opc_len)
+		return; /* source truncated */
+	PTR_LEN_INC(src_ptr, src_len, opc_len);
+	opc = src_ptr[0];
+	goto *opc_tbl[opc];
+
+eos:
+	opc_len = 8;
+	if (src_len < opc_len)
+		return; /* source truncated (here we don't need an extra byte for next op code) */
+	PTR_LEN_INC(src_ptr, src_len, opc_len);
+	state->end_of_stream = 1;
+	UPDATE_GOOD;
+	return; /* end-of-stream */
+
+/*
+ * ===============================================================
+ * Return on error
+ */
+udef:
+invalid_match_distance:
+
+	return; /* we already updated state */
+}
diff --git a/fs/apfs/lzfse/lzvn_decode_base.h b/fs/apfs/lzfse/lzvn_decode_base.h
new file mode 100644
index 000000000..90826f4d3
--- /dev/null
+++ b/fs/apfs/lzfse/lzvn_decode_base.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause */
+/*
+ * Copyright (c) 2015-2016, Apple Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+ *
+ * 1.  Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+ *
+ * 2.  Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
+ *     in the documentation and/or other materials provided with the distribution.
+ *
+ * 3.  Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * LZVN low-level decoder (v2)
+ * Functions in the low-level API should switch to these at some point.
+ * Apr 2014
+ */
+
+#ifndef LZVN_DECODE_BASE_H
+#define LZVN_DECODE_BASE_H
+
+#include "lzfse_internal.h"
+
+/*! @abstract Base decoder state. */
+typedef struct {
+	/* Decoder I/O */
+
+	/* Next byte to read in source buffer */
+	const unsigned char *src;
+	/* Next byte after source buffer */
+	const unsigned char *src_end;
+
+	/* Next byte to write in destination buffer (by decoder) */
+	unsigned char *dst;
+	/* Valid range for destination buffer is [dst_begin, dst_end - 1] */
+	unsigned char *dst_begin;
+	unsigned char *dst_end;
+	/* Next byte to read in destination buffer (modified by caller) */
+	unsigned char *dst_current;
+
+	/* Decoder state */
+
+	/*
+	 * Partially expanded match, or 0,0,0.
+	 * In that case, src points to the next literal to copy, or the next
+	 * op-code if L==0.
+	 */
+	size_t L, M, D;
+
+	/* Distance for last emitted match, or 0 */
+	lzvn_offset d_prev;
+
+	/* Did we decode end-of-stream? */
+	int end_of_stream;
+
+} lzvn_decoder_state;
+
+/*! @abstract Decode source to destination.
+ *  Updates \p state (src,dst,d_prev).
+ */
+void lzvn_decode(lzvn_decoder_state *state);
+
+#endif /* LZVN_DECODE_BASE_H */
diff --git a/fs/apfs/message.c b/fs/apfs/message.c
new file mode 100644
index 000000000..fdd74ccb8
--- /dev/null
+++ b/fs/apfs/message.c
@@ -0,0 +1,29 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/fs.h>
+#include "apfs.h"
+
+void apfs_msg(struct super_block *sb, const char *prefix, const char *func, int line, const char *fmt, ...)
+{
+	char *sb_id = NULL;
+	struct va_format vaf;
+	va_list args;
+
+	va_start(args, fmt);
+
+	vaf.fmt = fmt;
+	vaf.va = &args;
+
+	/* The superblock is not available to all callers */
+	sb_id = sb ? sb->s_id : "?";
+
+	if (func)
+		printk("%sAPFS (%s): %pV (%s:%d)\n", prefix, sb_id, &vaf, func, line);
+	else
+		printk("%sAPFS (%s): %pV\n", prefix, sb_id, &vaf);
+
+	va_end(args);
+}
diff --git a/fs/apfs/namei.c b/fs/apfs/namei.c
new file mode 100644
index 000000000..626de9f6b
--- /dev/null
+++ b/fs/apfs/namei.c
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/namei.h>
+#include "apfs.h"
+#include "unicode.h"
+
+static struct dentry *apfs_lookup(struct inode *dir, struct dentry *dentry,
+				  unsigned int flags)
+{
+	struct inode *inode = NULL;
+	u64 ino = 0;
+	int err;
+
+	if (dentry->d_name.len > APFS_NAME_LEN)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	err = apfs_inode_by_name(dir, &dentry->d_name, &ino);
+	if (err && err != -ENODATA) {
+		apfs_err(dir->i_sb, "inode lookup by name failed");
+		return ERR_PTR(err);
+	}
+
+	if (!err) {
+		inode = apfs_iget(dir->i_sb, ino);
+		if (IS_ERR(inode)) {
+			apfs_err(dir->i_sb, "iget failed");
+			return ERR_CAST(inode);
+		}
+	}
+
+	return d_splice_alias(inode, dentry);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+static int apfs_symlink(struct inode *dir, struct dentry *dentry,
+			const char *symname)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+static int apfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
+			struct dentry *dentry, const char *symname)
+#else
+static int apfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
+			struct dentry *dentry, const char *symname)
+#endif
+{
+	/* Symlink permissions don't mean anything and their value is fixed */
+	return apfs_mkany(dir, dentry, S_IFLNK | 0x1ed, 0 /* rdev */, symname);
+}
+
+const struct inode_operations apfs_dir_inode_operations = {
+	.create		= apfs_create,
+	.lookup		= apfs_lookup,
+	.link		= apfs_link,
+	.unlink		= apfs_unlink,
+	.symlink	= apfs_symlink,
+	.mkdir		= apfs_mkdir,
+	.rmdir		= apfs_rmdir,
+	.mknod		= apfs_mknod,
+	.rename		= apfs_rename,
+	.getattr	= apfs_getattr,
+	.listxattr      = apfs_listxattr,
+	.setattr	= apfs_setattr,
+	.update_time	= apfs_update_time,
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 13, 0)
+	.fileattr_get	= apfs_fileattr_get,
+	.fileattr_set	= apfs_fileattr_set,
+#endif
+};
+
+const struct inode_operations apfs_special_inode_operations = {
+	.getattr	= apfs_getattr,
+	.listxattr      = apfs_listxattr,
+	.setattr	= apfs_setattr,
+	.update_time	= apfs_update_time,
+};
+
+static int apfs_dentry_hash(const struct dentry *dir, struct qstr *child)
+{
+	struct apfs_unicursor cursor;
+	unsigned long hash;
+	bool case_fold = apfs_is_case_insensitive(dir->d_sb);
+
+	if (!apfs_is_normalization_insensitive(dir->d_sb))
+		return 0;
+
+	apfs_init_unicursor(&cursor, child->name, child->len);
+	hash = init_name_hash(dir);
+
+	while (1) {
+		int i;
+		unicode_t utf32;
+
+		utf32 = apfs_normalize_next(&cursor, case_fold);
+		if (!utf32)
+			break;
+
+		/* Hash the unicode character one byte at a time */
+		for (i = 0; i < 4; ++i) {
+			hash = partial_name_hash((u8)utf32, hash);
+			utf32 = utf32 >> 8;
+		}
+	}
+	child->hash = end_name_hash(hash);
+
+	/* TODO: return error instead of truncating invalid UTF-8? */
+	return 0;
+}
+
+static int apfs_dentry_compare(const struct dentry *dentry, unsigned int len,
+			       const char *str, const struct qstr *name)
+{
+	return apfs_filename_cmp(dentry->d_sb, name->name, name->len, str, len);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 14, 0)
+static int apfs_dentry_revalidate(struct dentry *dentry, unsigned int flags)
+#else
+static int apfs_dentry_revalidate(struct inode *dir, const struct qstr *name, struct dentry *dentry, unsigned int flags)
+#endif
+{
+	struct super_block *sb = dentry->d_sb;
+
+	if (flags & LOOKUP_RCU)
+		return -ECHILD;
+
+	/*
+	 * If we want to create a link with a name that normalizes to the same
+	 * as an existing negative dentry, then we first need to invalidate the
+	 * dentry; otherwise it would keep the existing name.
+	 */
+	if (d_really_is_positive(dentry))
+		return 1;
+	if (!apfs_is_case_insensitive(sb) && !apfs_is_normalization_insensitive(sb))
+		return 1;
+	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
+		return 0;
+	return 1;
+}
+
+const struct dentry_operations apfs_dentry_operations = {
+	.d_revalidate	= apfs_dentry_revalidate,
+	.d_hash		= apfs_dentry_hash,
+	.d_compare	= apfs_dentry_compare,
+};
diff --git a/fs/apfs/node.c b/fs/apfs/node.c
new file mode 100644
index 000000000..cc332da43
--- /dev/null
+++ b/fs/apfs/node.c
@@ -0,0 +1,2069 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/slab.h>
+#include <linux/buffer_head.h>
+#include "apfs.h"
+
+/**
+ * apfs_node_is_valid - Check basic sanity of the node index
+ * @sb:		filesystem superblock
+ * @node:	node to check
+ *
+ * Verifies that the node index fits in a single block, and that the number
+ * of records fits in the index. Without this check a crafted filesystem could
+ * pretend to have too many records, and calls to apfs_node_locate_key() and
+ * apfs_node_locate_value() would read beyond the limits of the node.
+ */
+static bool apfs_node_is_valid(struct super_block *sb,
+			       struct apfs_node *node)
+{
+	u32 records = node->records;
+	int index_size = node->key - sizeof(struct apfs_btree_node_phys);
+	int entry_size;
+
+	if (node->key > sb->s_blocksize)
+		return false;
+
+	entry_size = (apfs_node_has_fixed_kv_size(node)) ?
+		sizeof(struct apfs_kvoff) : sizeof(struct apfs_kvloc);
+
+	/* Coarse bound to prevent multiplication overflow in final check */
+	if (records > 1 << 16)
+		return false;
+
+	return records * entry_size <= index_size;
+}
+
+void apfs_node_free(struct apfs_node *node)
+{
+	struct apfs_object *obj = NULL;
+
+	if (!node)
+		return;
+	obj = &node->object;
+
+	if (obj->o_bh) {
+		brelse(obj->o_bh);
+		obj->o_bh = NULL;
+	} else if (!obj->ephemeral) {
+		/* Ephemeral data always remains in memory */
+		kfree(obj->data);
+	}
+	obj->data = NULL;
+
+	kfree(node);
+}
+
+/**
+ * apfs_read_node - Read a node header from disk
+ * @sb:		filesystem superblock
+ * @oid:	object id for the node
+ * @storage:	storage type for the node object
+ * @write:	request write access?
+ *
+ * Returns ERR_PTR in case of failure, otherwise return a pointer to the
+ * resulting apfs_node structure with the initial reference taken.
+ *
+ * For now we assume the node has not been read before.
+ */
+struct apfs_node *apfs_read_node(struct super_block *sb, u64 oid, u32 storage,
+				 bool write)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct buffer_head *bh = NULL;
+	struct apfs_ephemeral_object_info *eph_info = NULL;
+	struct apfs_btree_node_phys *raw = NULL;
+	struct apfs_node *node = NULL;
+	struct apfs_nloc *free_head = NULL;
+	u64 bno;
+	int err;
+
+	switch (storage) {
+	case APFS_OBJ_VIRTUAL:
+		/* All virtual nodes are inside a volume, at least for now */
+		err = apfs_omap_lookup_block(sb, sbi->s_omap, oid, &bno, write);
+		if (err) {
+			apfs_err(sb, "omap lookup failed for oid 0x%llx", oid);
+			return ERR_PTR(err);
+		}
+		/* CoW has already been done, don't worry about snapshots */
+		bh = apfs_read_object_block(sb, bno, write, false /* preserve */);
+		if (IS_ERR(bh)) {
+			apfs_err(sb, "object read failed for bno 0x%llx", bno);
+			return (void *)bh;
+		}
+		bno = bh->b_blocknr;
+		raw = (struct apfs_btree_node_phys *)bh->b_data;
+		break;
+	case APFS_OBJ_PHYSICAL:
+		bh = apfs_read_object_block(sb, oid, write, false /* preserve */);
+		if (IS_ERR(bh)) {
+			apfs_err(sb, "object read failed for bno 0x%llx", oid);
+			return (void *)bh;
+		}
+		bno = oid = bh->b_blocknr;
+		raw = (struct apfs_btree_node_phys *)bh->b_data;
+		break;
+	case APFS_OBJ_EPHEMERAL:
+		/* Ephemeral objects are already in memory */
+		eph_info = apfs_ephemeral_object_lookup(sb, oid);
+		if (IS_ERR(eph_info)) {
+			apfs_err(sb, "no ephemeral node for oid 0x%llx", oid);
+			return (void *)eph_info;
+		}
+		if (eph_info->size != sb->s_blocksize) {
+			apfs_err(sb, "unsupported size for ephemeral node (%u)", eph_info->size);
+			return ERR_PTR(-EOPNOTSUPP);
+		}
+		bno = 0; /* In memory, so meaningless */
+		raw = eph_info->object;
+		/* Only for consistency, will happen again on commit */
+		if (write)
+			raw->btn_o.o_xid = cpu_to_le64(nxi->nx_xid);
+		break;
+	default:
+		apfs_alert(sb, "invalid storage type %u - bug!", storage);
+		return ERR_PTR(-EINVAL);
+	}
+
+	node = kmalloc(sizeof(*node), GFP_KERNEL);
+	if (!node) {
+		brelse(bh);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	node->tree_type = le32_to_cpu(raw->btn_o.o_subtype);
+	node->flags = le16_to_cpu(raw->btn_flags);
+	node->records = le32_to_cpu(raw->btn_nkeys);
+	node->key = sizeof(*raw) + le16_to_cpu(raw->btn_table_space.off)
+				+ le16_to_cpu(raw->btn_table_space.len);
+	node->free = node->key + le16_to_cpu(raw->btn_free_space.off);
+	node->val = node->free + le16_to_cpu(raw->btn_free_space.len);
+
+	free_head = &raw->btn_key_free_list;
+	node->key_free_list_len = le16_to_cpu(free_head->len);
+	free_head = &raw->btn_val_free_list;
+	node->val_free_list_len = le16_to_cpu(free_head->len);
+
+	node->object.sb = sb;
+	node->object.block_nr = bno;
+	node->object.oid = oid;
+	node->object.o_bh = bh;
+	node->object.data = (char *)raw;
+	node->object.ephemeral = !bh;
+
+	/* Ephemeral objects already got checked on mount */
+	if (!node->object.ephemeral && nxi->nx_flags & APFS_CHECK_NODES && !apfs_obj_verify_csum(sb, bh)) {
+		/* TODO: don't check this twice for virtual/physical objects */
+		apfs_err(sb, "bad checksum for node in block 0x%llx", (unsigned long long)bno);
+		apfs_node_free(node);
+		return ERR_PTR(-EFSBADCRC);
+	}
+	if (!apfs_node_is_valid(sb, node)) {
+		apfs_err(sb, "bad node in block 0x%llx", (unsigned long long)bno);
+		apfs_node_free(node);
+		return ERR_PTR(-EFSCORRUPTED);
+	}
+
+	return node;
+}
+
+/**
+ * apfs_node_min_table_size - Return the minimum size for a node's toc
+ * @sb:		superblock structure
+ * @type:	tree type for the node
+ * @flags:	flags for the node
+ */
+static int apfs_node_min_table_size(struct super_block *sb, u32 type, u16 flags)
+{
+	bool leaf = flags & APFS_BTNODE_LEAF;
+	int key_size, val_size, toc_size;
+	int space, count;
+
+	/* Preallocate the whole table for trees with fixed key/value sizes */
+	switch (type) {
+	case APFS_OBJECT_TYPE_OMAP:
+		key_size = sizeof(struct apfs_omap_key);
+		val_size = leaf ? sizeof(struct apfs_omap_val) : sizeof(__le64);
+		toc_size = sizeof(struct apfs_kvoff);
+		break;
+	case APFS_OBJECT_TYPE_SPACEMAN_FREE_QUEUE:
+		key_size = sizeof(struct apfs_spaceman_free_queue_key);
+		val_size = sizeof(__le64); /* We assume no ghosts here */
+		toc_size = sizeof(struct apfs_kvoff);
+		break;
+	case APFS_OBJECT_TYPE_OMAP_SNAPSHOT:
+		key_size = sizeof(__le64);
+		val_size = leaf ? sizeof(struct apfs_omap_snapshot) : sizeof(__le64);
+		toc_size = sizeof(struct apfs_kvoff);
+		break;
+	case APFS_OBJECT_TYPE_FEXT_TREE:
+		key_size = sizeof(struct apfs_fext_tree_key);
+		val_size = leaf ? sizeof(struct apfs_fext_tree_val) : sizeof(__le64);
+		toc_size = sizeof(struct apfs_kvoff);
+		break;
+	default:
+		/* Make room for one record at least */
+		toc_size = sizeof(struct apfs_kvloc);
+		return APFS_BTREE_TOC_ENTRY_INCREMENT * toc_size;
+	}
+
+	/* The footer of root nodes is ignored for some reason */
+	space = sb->s_blocksize - sizeof(struct apfs_btree_node_phys);
+	count = space / (key_size + val_size + toc_size);
+	return count * toc_size;
+}
+
+/**
+ * apfs_set_empty_btree_info - Set the info footer for an empty b-tree node
+ * @sb:		filesystem superblock
+ * @info:	pointer to the on-disk info footer
+ * @subtype:	subtype of the root node, i.e., tree type
+ *
+ * For now only supports the extent reference tree.
+ */
+static void apfs_set_empty_btree_info(struct super_block *sb, struct apfs_btree_info *info, u32 subtype)
+{
+	u32 flags;
+
+	ASSERT(subtype == APFS_OBJECT_TYPE_BLOCKREFTREE || subtype == APFS_OBJECT_TYPE_OMAP_SNAPSHOT);
+
+	memset(info, 0, sizeof(*info));
+
+	flags = APFS_BTREE_PHYSICAL;
+	if (subtype == APFS_OBJECT_TYPE_BLOCKREFTREE)
+		flags |= APFS_BTREE_KV_NONALIGNED;
+
+	info->bt_fixed.bt_flags = cpu_to_le32(flags);
+	info->bt_fixed.bt_node_size = cpu_to_le32(sb->s_blocksize);
+	info->bt_key_count = 0;
+	info->bt_node_count = cpu_to_le64(1); /* Only one node: the root */
+	if (subtype == APFS_OBJECT_TYPE_BLOCKREFTREE)
+		return;
+
+	info->bt_fixed.bt_key_size = cpu_to_le32(8);
+	info->bt_longest_key = info->bt_fixed.bt_key_size;
+	info->bt_fixed.bt_val_size = cpu_to_le32(sizeof(struct apfs_omap_snapshot));
+	info->bt_longest_val = info->bt_fixed.bt_val_size;
+}
+
+/**
+ * apfs_make_empty_btree_root - Make an empty root for a b-tree
+ * @sb:		filesystem superblock
+ * @subtype:	subtype of the root node, i.e., tree type
+ * @oid:	on return, the root's object id
+ *
+ * For now only supports the extent reference tree and an omap's snapshot tree.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_make_empty_btree_root(struct super_block *sb, u32 subtype, u64 *oid)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_btree_node_phys *root = NULL;
+	struct buffer_head *bh = NULL;
+	u64 bno;
+	u16 flags;
+	int toc_len, free_len, head_len, info_len;
+	int err;
+
+	ASSERT(subtype == APFS_OBJECT_TYPE_BLOCKREFTREE || subtype == APFS_OBJECT_TYPE_OMAP_SNAPSHOT);
+
+	err = apfs_spaceman_allocate_block(sb, &bno, true /* backwards */);
+	if (err) {
+		apfs_err(sb, "block allocation failed");
+		return err;
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, 1);
+	le64_add_cpu(&vsb_raw->apfs_total_blocks_alloced, 1);
+
+	bh = apfs_getblk(sb, bno);
+	if (!bh)
+		return -EIO;
+	root = (void *)bh->b_data;
+	err = apfs_transaction_join(sb, bh);
+	if (err)
+		goto fail;
+	set_buffer_csum(bh);
+
+	flags = APFS_BTNODE_ROOT | APFS_BTNODE_LEAF;
+	if (subtype == APFS_OBJECT_TYPE_OMAP_SNAPSHOT)
+		flags |= APFS_BTNODE_FIXED_KV_SIZE;
+	root->btn_flags = cpu_to_le16(flags);
+
+	toc_len = apfs_node_min_table_size(sb, subtype, flags);
+	head_len = sizeof(*root);
+	info_len = sizeof(struct apfs_btree_info);
+	free_len = sb->s_blocksize - head_len - toc_len - info_len;
+
+	root->btn_level = 0; /* Root */
+
+	/* No keys and no values, so this is straightforward */
+	root->btn_nkeys = 0;
+	root->btn_table_space.off = 0;
+	root->btn_table_space.len = cpu_to_le16(toc_len);
+	root->btn_free_space.off = 0;
+	root->btn_free_space.len = cpu_to_le16(free_len);
+
+	/* No fragmentation */
+	root->btn_key_free_list.off = cpu_to_le16(APFS_BTOFF_INVALID);
+	root->btn_key_free_list.len = 0;
+	root->btn_val_free_list.off = cpu_to_le16(APFS_BTOFF_INVALID);
+	root->btn_val_free_list.len = 0;
+
+	apfs_set_empty_btree_info(sb, (void *)root + sb->s_blocksize - info_len, subtype);
+
+	root->btn_o.o_oid = cpu_to_le64(bno);
+	root->btn_o.o_xid = cpu_to_le64(APFS_NXI(sb)->nx_xid);
+	root->btn_o.o_type = cpu_to_le32(APFS_OBJECT_TYPE_BTREE | APFS_OBJ_PHYSICAL);
+	root->btn_o.o_subtype = cpu_to_le32(subtype);
+
+	*oid = bno;
+	err = 0;
+fail:
+	root = NULL;
+	brelse(bh);
+	bh = NULL;
+	return err;
+}
+
+/**
+ * apfs_create_node - Allocates a new nonroot b-tree node on disk
+ * @sb:		filesystem superblock
+ * @storage:	storage type for the node object
+ *
+ * On success returns a pointer to the new in-memory node structure; the object
+ * header is initialized, and the node fields are given reasonable defaults.
+ * On failure, returns an error pointer.
+ */
+static struct apfs_node *apfs_create_node(struct super_block *sb, u32 storage)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *msb_raw = nxi->nx_raw;
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_ephemeral_object_info *eph_info = NULL;
+	struct apfs_node *node = NULL;
+	struct buffer_head *bh = NULL;
+	struct apfs_btree_node_phys *raw = NULL;
+	u64 bno, oid;
+	int err;
+
+	switch (storage) {
+	case APFS_OBJ_VIRTUAL:
+		err = apfs_spaceman_allocate_block(sb, &bno, true /* backwards */);
+		if (err) {
+			apfs_err(sb, "block allocation failed");
+			return ERR_PTR(err);
+		}
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, 1);
+		le64_add_cpu(&vsb_raw->apfs_total_blocks_alloced, 1);
+
+		oid = le64_to_cpu(msb_raw->nx_next_oid);
+		le64_add_cpu(&msb_raw->nx_next_oid, 1);
+		err = apfs_create_omap_rec(sb, oid, bno);
+		if (err) {
+			apfs_err(sb, "omap rec creation failed (0x%llx-0x%llx)", oid, bno);
+			return ERR_PTR(err);
+		}
+		break;
+	case APFS_OBJ_PHYSICAL:
+		err = apfs_spaceman_allocate_block(sb, &bno, true /* backwards */);
+		if (err) {
+			apfs_err(sb, "block allocation failed");
+			return ERR_PTR(err);
+		}
+		/* We don't write to the container's omap */
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, 1);
+		le64_add_cpu(&vsb_raw->apfs_total_blocks_alloced, 1);
+		oid = bno;
+		break;
+	case APFS_OBJ_EPHEMERAL:
+		if (nxi->nx_eph_count >= APFS_EPHEMERAL_LIST_LIMIT) {
+			apfs_err(sb, "creating too many ephemeral objects?");
+			return ERR_PTR(-EOPNOTSUPP);
+		}
+		eph_info = &nxi->nx_eph_list[nxi->nx_eph_count++];
+		eph_info->object = kzalloc(sb->s_blocksize, GFP_KERNEL);
+		if (!eph_info->object)
+			return ERR_PTR(-ENOMEM);
+		eph_info->size = sb->s_blocksize;
+		oid = eph_info->oid = le64_to_cpu(msb_raw->nx_next_oid);
+		le64_add_cpu(&msb_raw->nx_next_oid, 1);
+		break;
+	default:
+		apfs_alert(sb, "invalid storage type %u - bug!", storage);
+		return ERR_PTR(-EINVAL);
+	}
+
+	if (storage == APFS_OBJ_EPHEMERAL) {
+		bh = NULL;
+		bno = 0;
+		raw = eph_info->object;
+	} else {
+		bh = apfs_getblk(sb, bno);
+		if (!bh)
+			return ERR_PTR(-EIO);
+		bno = bh->b_blocknr;
+		raw = (void *)bh->b_data;
+		err = apfs_transaction_join(sb, bh);
+		if (err)
+			goto fail;
+		set_buffer_csum(bh);
+	}
+
+	/* Set most of the object header, but the subtype is up to the caller */
+	raw->btn_o.o_oid = cpu_to_le64(oid);
+	raw->btn_o.o_xid = cpu_to_le64(nxi->nx_xid);
+	raw->btn_o.o_type = cpu_to_le32(storage | APFS_OBJECT_TYPE_BTREE_NODE);
+	raw->btn_o.o_subtype = 0;
+
+	/* The caller is expected to change most node fields */
+	raw->btn_flags = 0;
+	raw->btn_level = 0;
+	raw->btn_nkeys = 0;
+	raw->btn_table_space.off = 0; /* Put the toc right after the header */
+	raw->btn_table_space.len = 0;
+	raw->btn_free_space.off = 0;
+	raw->btn_free_space.len = cpu_to_le16(sb->s_blocksize - sizeof(*raw));
+	raw->btn_key_free_list.off = cpu_to_le16(APFS_BTOFF_INVALID);
+	raw->btn_key_free_list.len = 0;
+	raw->btn_val_free_list.off = cpu_to_le16(APFS_BTOFF_INVALID);
+	raw->btn_val_free_list.len = 0;
+
+	node = kmalloc(sizeof(*node), GFP_KERNEL);
+	if (!node) {
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	node->object.sb = sb;
+	node->object.block_nr = bno;
+	node->object.oid = oid;
+	node->object.o_bh = bh;
+	node->object.data = (char *)raw;
+	node->object.ephemeral = !bh;
+	return node;
+
+fail:
+	if (storage == APFS_OBJ_EPHEMERAL)
+		kfree(raw);
+	else
+		brelse(bh);
+	raw = NULL;
+	bh = NULL;
+	return ERR_PTR(err);
+}
+
+/**
+ * apfs_delete_node - Deletes a nonroot node from disk
+ * @node: node to delete
+ * @type: tree type for the query that found the node
+ *
+ * Does nothing to the in-memory node structure.  Returns 0 on success, or a
+ * negative error code in case of failure.
+ */
+int apfs_delete_node(struct apfs_node *node, int type)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_superblock *vsb_raw;
+	u64 oid = node->object.oid;
+	u64 bno = node->object.block_nr;
+	struct apfs_ephemeral_object_info *eph_info = NULL, *eph_info_end = NULL;
+	int err;
+
+	switch (type) {
+	case APFS_QUERY_CAT:
+		err = apfs_free_queue_insert(sb, bno, 1);
+		if (err) {
+			apfs_err(sb, "free queue insertion failed for 0x%llx", bno);
+			return err;
+		}
+		err = apfs_delete_omap_rec(sb, oid);
+		if (err) {
+			apfs_err(sb, "omap rec deletion failed (0x%llx)", oid);
+			return err;
+		}
+		vsb_raw = APFS_SB(sb)->s_vsb_raw;
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, -1);
+		le64_add_cpu(&vsb_raw->apfs_total_blocks_freed, 1);
+		return 0;
+	case APFS_QUERY_OMAP:
+	case APFS_QUERY_EXTENTREF:
+	case APFS_QUERY_SNAP_META:
+		err = apfs_free_queue_insert(sb, bno, 1);
+		if (err) {
+			apfs_err(sb, "free queue insertion failed for 0x%llx", bno);
+			return err;
+		}
+		/* We don't write to the container's omap */
+		vsb_raw = APFS_SB(sb)->s_vsb_raw;
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, -1);
+		le64_add_cpu(&vsb_raw->apfs_total_blocks_freed, 1);
+		return 0;
+	case APFS_QUERY_FREE_QUEUE:
+		eph_info_end = &nxi->nx_eph_list[nxi->nx_eph_count];
+		eph_info = apfs_ephemeral_object_lookup(sb, node->object.oid);
+		if (IS_ERR(eph_info)) {
+			apfs_alert(sb, "can't find ephemeral object to delete");
+			return PTR_ERR(eph_info);
+		}
+		kfree(eph_info->object);
+		eph_info->object = NULL;
+		memmove(eph_info, eph_info + 1, (char *)eph_info_end - (char *)(eph_info + 1));
+		eph_info_end->object = NULL;
+		--nxi->nx_eph_count;
+		return 0;
+	default:
+		apfs_alert(sb, "new query type must implement node deletion (%d)", type);
+		return -EOPNOTSUPP;
+	}
+}
+
+/**
+ * apfs_update_node - Update an existing node header
+ * @node: the modified in-memory node
+ */
+void apfs_update_node(struct apfs_node *node)
+{
+	struct super_block *sb = node->object.sb;
+	struct buffer_head *bh = node->object.o_bh;
+	struct apfs_btree_node_phys *raw = (void *)node->object.data;
+	struct apfs_nloc *free_head;
+	u32 tflags, type;
+	int toc_off;
+
+	apfs_assert_in_transaction(sb, &raw->btn_o);
+
+	raw->btn_o.o_oid = cpu_to_le64(node->object.oid);
+
+	/* The node may no longer be a root, so update the object type */
+	tflags = le32_to_cpu(raw->btn_o.o_type) & APFS_OBJECT_TYPE_FLAGS_MASK;
+	type = (node->flags & APFS_BTNODE_ROOT) ? APFS_OBJECT_TYPE_BTREE :
+						  APFS_OBJECT_TYPE_BTREE_NODE;
+	raw->btn_o.o_type = cpu_to_le32(type | tflags);
+	raw->btn_o.o_subtype = cpu_to_le32(node->tree_type);
+
+	raw->btn_flags = cpu_to_le16(node->flags);
+	raw->btn_nkeys = cpu_to_le32(node->records);
+
+	toc_off = sizeof(*raw) + le16_to_cpu(raw->btn_table_space.off);
+	raw->btn_table_space.len = cpu_to_le16(node->key - toc_off);
+	raw->btn_free_space.off = cpu_to_le16(node->free - node->key);
+	raw->btn_free_space.len = cpu_to_le16(node->val - node->free);
+
+	/* Reset the lists on zero length, a defragmentation is taking place */
+	free_head = &raw->btn_key_free_list;
+	free_head->len = cpu_to_le16(node->key_free_list_len);
+	if (!free_head->len)
+		free_head->off = cpu_to_le16(APFS_BTOFF_INVALID);
+	free_head = &raw->btn_val_free_list;
+	free_head->len = cpu_to_le16(node->val_free_list_len);
+	if (!free_head->len)
+		free_head->off = cpu_to_le16(APFS_BTOFF_INVALID);
+
+	if (bh) {
+		ASSERT(buffer_trans(bh));
+		ASSERT(buffer_csum(bh));
+	}
+}
+
+/**
+ * apfs_node_locate_key - Locate the key of a node record
+ * @node:	node to be searched
+ * @index:	number of the entry to locate
+ * @off:	on return will hold the offset in the block
+ *
+ * Returns the length of the key, or 0 in case of failure. The function checks
+ * that this length fits within the block; callers must use it to make sure
+ * they never operate outside its bounds.
+ */
+int apfs_node_locate_key(struct apfs_node *node, int index, int *off)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *raw;
+	int len;
+
+	if (index >= node->records) {
+		apfs_err(sb, "index out of bounds (%d of %d)", index, node->records);
+		return 0;
+	}
+
+	raw = (struct apfs_btree_node_phys *)node->object.data;
+	if (apfs_node_has_fixed_kv_size(node)) {
+		struct apfs_kvoff *entry;
+
+		entry = (struct apfs_kvoff *)raw->btn_data + index;
+
+		/* TODO: it would be cleaner to read this stuff from disk */
+		if (node->tree_type == APFS_OBJECT_TYPE_OMAP_SNAPSHOT)
+			len = 8;
+		else
+			len = 16;
+
+		/* Translate offset in key area to offset in block */
+		*off = node->key + le16_to_cpu(entry->k);
+	} else {
+		/* These node types have variable length keys and values */
+		struct apfs_kvloc *entry;
+
+		entry = (struct apfs_kvloc *)raw->btn_data + index;
+		len = le16_to_cpu(entry->k.len);
+		/* Translate offset in key area to offset in block */
+		*off = node->key + le16_to_cpu(entry->k.off);
+	}
+
+	if (*off + len > sb->s_blocksize) {
+		apfs_err(sb, "key out of bounds (%d-%d)", *off, len);
+		return 0;
+	}
+	return len;
+}
+
+/**
+ * apfs_node_locate_value - Locate the value of a node record
+ * @node:	node to be searched
+ * @index:	number of the entry to locate
+ * @off:	on return will hold the offset in the block
+ *
+ * Returns the length of the value, which may be 0 in case of corruption or if
+ * the record is a ghost. The function checks that this length fits within the
+ * block; callers must use it to make sure they never operate outside its
+ * bounds.
+ */
+static int apfs_node_locate_value(struct apfs_node *node, int index, int *off)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *raw;
+	int len;
+
+	if (index >= node->records) {
+		apfs_err(sb, "index out of bounds (%d of %d)", index, node->records);
+		return 0;
+	}
+
+	raw = (struct apfs_btree_node_phys *)node->object.data;
+	if (apfs_node_has_fixed_kv_size(node)) {
+		/* These node types have fixed length keys and values */
+		struct apfs_kvoff *entry;
+
+		entry = (struct apfs_kvoff *)raw->btn_data + index;
+		if (node->tree_type == APFS_OBJECT_TYPE_SPACEMAN_FREE_QUEUE) {
+			/* A free-space queue record may have no value */
+			if (le16_to_cpu(entry->v) == APFS_BTOFF_INVALID) {
+				*off = 0;
+				return 0;
+			}
+			len = 8;
+		} else {
+			/* This is an omap or omap snapshots node */
+			len = apfs_node_is_leaf(node) ? 16 : 8;
+		}
+		/*
+		 * Value offsets are counted backwards from the end of the
+		 * block, or from the beginning of the footer when it exists
+		 */
+		if (apfs_node_is_root(node)) /* has footer */
+			*off = sb->s_blocksize - sizeof(struct apfs_btree_info)
+					- le16_to_cpu(entry->v);
+		else
+			*off = sb->s_blocksize - le16_to_cpu(entry->v);
+	} else {
+		/* These node types have variable length keys and values */
+		struct apfs_kvloc *entry;
+
+		entry = (struct apfs_kvloc *)raw->btn_data + index;
+		len = le16_to_cpu(entry->v.len);
+		/*
+		 * Value offsets are counted backwards from the end of the
+		 * block, or from the beginning of the footer when it exists
+		 */
+		if (apfs_node_is_root(node)) /* has footer */
+			*off = sb->s_blocksize - sizeof(struct apfs_btree_info)
+					- le16_to_cpu(entry->v.off);
+		else
+			*off = sb->s_blocksize - le16_to_cpu(entry->v.off);
+	}
+
+	if (*off < 0 || *off + len > sb->s_blocksize) {
+		apfs_err(sb, "value out of bounds (%d-%d)", *off, len);
+		return 0;
+	}
+	return len;
+}
+
+/**
+ * apfs_create_toc_entry - Create the table-of-contents entry for a record
+ * @query: query pointing to the record
+ *
+ * Creates a toc entry for the record at index @query->index and increases
+ * @node->records.  The caller must ensure enough space in the table.
+ */
+static void apfs_create_toc_entry(struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *raw = (void *)node->object.data;
+	int value_end;
+	int recs = node->records;
+	int index = query->index;
+
+	value_end = sb->s_blocksize;
+	if (apfs_node_is_root(node))
+		value_end -= sizeof(struct apfs_btree_info);
+
+	if (apfs_node_has_fixed_kv_size(node)) {
+		struct apfs_kvoff *kvoff;
+
+		kvoff = (struct apfs_kvoff *)raw->btn_data + query->index;
+		memmove(kvoff + 1, kvoff, (recs - index) * sizeof(*kvoff));
+
+		if (!query->len) /* Ghost record */
+			kvoff->v = cpu_to_le16(APFS_BTOFF_INVALID);
+		else
+			kvoff->v = cpu_to_le16(value_end - query->off);
+		kvoff->k = cpu_to_le16(query->key_off - node->key);
+	} else {
+		struct apfs_kvloc *kvloc;
+
+		kvloc = (struct apfs_kvloc *)raw->btn_data + query->index;
+		memmove(kvloc + 1, kvloc, (recs - index) * sizeof(*kvloc));
+
+		kvloc->v.off = cpu_to_le16(value_end - query->off);
+		kvloc->v.len = cpu_to_le16(query->len);
+		kvloc->k.off = cpu_to_le16(query->key_off - node->key);
+		kvloc->k.len = cpu_to_le16(query->key_len);
+	}
+	node->records++;
+}
+
+/**
+ * apfs_key_from_query - Read the current key from a query structure
+ * @query:	the query, with @query->key_off and @query->key_len already set
+ * @key:	return parameter for the key
+ *
+ * Reads the key into @key and performs some basic sanity checks as a
+ * protection against crafted filesystems.  Returns 0 on success or a
+ * negative error code otherwise.
+ */
+static int apfs_key_from_query(struct apfs_query *query, struct apfs_key *key)
+{
+	struct super_block *sb = query->node->object.sb;
+	char *raw = query->node->object.data;
+	void *raw_key = (void *)(raw + query->key_off);
+	bool hashed;
+	int err = 0;
+
+	switch (query->flags & APFS_QUERY_TREE_MASK) {
+	case APFS_QUERY_CAT:
+		hashed = apfs_is_normalization_insensitive(sb);
+		err = apfs_read_cat_key(raw_key, query->key_len, key, hashed);
+		break;
+	case APFS_QUERY_OMAP:
+		err = apfs_read_omap_key(raw_key, query->key_len, key);
+		break;
+	case APFS_QUERY_FREE_QUEUE:
+		err = apfs_read_free_queue_key(raw_key, query->key_len, key);
+		break;
+	case APFS_QUERY_EXTENTREF:
+		err = apfs_read_extentref_key(raw_key, query->key_len, key);
+		break;
+	case APFS_QUERY_FEXT:
+		err = apfs_read_fext_key(raw_key, query->key_len, key);
+		break;
+	case APFS_QUERY_SNAP_META:
+		err = apfs_read_snap_meta_key(raw_key, query->key_len, key);
+		break;
+	case APFS_QUERY_OMAP_SNAP:
+		err = apfs_read_omap_snap_key(raw_key, query->key_len, key);
+		break;
+	default:
+		apfs_alert(sb, "new query type must implement key reads (%d)", query->flags & APFS_QUERY_TREE_MASK);
+		err = -EOPNOTSUPP;
+		break;
+	}
+	if (err)
+		apfs_err(sb, "bad node key in block 0x%llx", query->node->object.block_nr);
+
+	/* A multiple query must ignore some of these fields */
+	if (query->flags & APFS_QUERY_ANY_NAME)
+		key->name = NULL;
+	if (query->flags & APFS_QUERY_ANY_NUMBER)
+		key->number = 0;
+
+	return err;
+}
+
+/**
+ * apfs_node_prev - Find the previous record in the current node
+ * @sb:		filesystem superblock
+ * @query:	query in execution
+ *
+ * Returns 0 on success, -EAGAIN if the previous record is in another node,
+ * -ENODATA if no more records exist, or another negative error code in case
+ * of failure.
+ *
+ * The meaning of "next" and "previous" is reverted here, because regular
+ * multiple always start with the final record, and then they go backwards.
+ * TODO: consider renaming this for clarity.
+ */
+static int apfs_node_prev(struct super_block *sb, struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+
+	if (query->index + 1 == node->records) {
+		/* The next record may be in another node */
+		return -EAGAIN;
+	}
+	++query->index;
+
+	query->key_len = apfs_node_locate_key(node, query->index, &query->key_off);
+	if (query->key_len == 0) {
+		apfs_err(sb, "bad key for index %d", query->index);
+		return -EFSCORRUPTED;
+	}
+	query->len = apfs_node_locate_value(node, query->index, &query->off);
+	if (query->len == 0) {
+		apfs_err(sb, "bad value for index %d", query->index);
+		return -EFSCORRUPTED;
+	}
+	return 0;
+}
+
+/**
+ * apfs_node_next - Find the next matching record in the current node
+ * @sb:		filesystem superblock
+ * @query:	multiple query in execution
+ *
+ * Returns 0 on success, -EAGAIN if the next record is in another node,
+ * -ENODATA if no more matching records exist, or another negative error
+ * code in case of failure.
+ */
+static int apfs_node_next(struct super_block *sb, struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+	struct apfs_key curr_key;
+	int cmp, err;
+
+	if (query->flags & APFS_QUERY_DONE)
+		/* Nothing left to search; the query failed */
+		return -ENODATA;
+
+	if (!query->index) /* The next record may be in another node */
+		return -EAGAIN;
+	--query->index;
+
+	query->key_len = apfs_node_locate_key(node, query->index,
+					      &query->key_off);
+	err = apfs_key_from_query(query, &curr_key);
+	if (err) {
+		apfs_err(sb, "bad key for index %d", query->index);
+		return err;
+	}
+
+	cmp = apfs_keycmp(&curr_key, &query->key);
+
+	if (cmp > 0) {
+		apfs_err(sb, "records are out of order");
+		return -EFSCORRUPTED;
+	}
+
+	if (cmp != 0 && apfs_node_is_leaf(node) &&
+	    query->flags & APFS_QUERY_EXACT)
+		return -ENODATA;
+
+	query->len = apfs_node_locate_value(node, query->index, &query->off);
+	if (query->len == 0) {
+		apfs_err(sb, "bad value for index %d", query->index);
+		return -EFSCORRUPTED;
+	}
+
+	if (cmp != 0) {
+		/*
+		 * This is the last entry that can be relevant in this node.
+		 * Keep searching the children, but don't return to this level.
+		 */
+		query->flags |= APFS_QUERY_DONE;
+	}
+
+	return 0;
+}
+
+/**
+ * apfs_node_query - Execute a query on a single node
+ * @sb:		filesystem superblock
+ * @query:	the query to execute
+ *
+ * The search will start at index @query->index, looking for the key that comes
+ * right before @query->key, according to the order given by apfs_keycmp().
+ *
+ * The @query->index will be updated to the last index checked. This is
+ * important when searching for multiple entries, since the query may need
+ * to remember where it was on this level. If we are done with this node, the
+ * query will be flagged as APFS_QUERY_DONE, and the search will end in failure
+ * as soon as we return to this level. The function may also return -EAGAIN,
+ * to signal that the search should go on in a different branch.
+ *
+ * On success returns 0; the offset of the value within the block will be saved
+ * in @query->off, and its length in @query->len. The function checks that this
+ * length fits within the block; callers must use the returned value to make
+ * sure they never operate outside its bounds.
+ *
+ * -ENODATA will be returned if no appropriate entry was found, -EFSCORRUPTED
+ * in case of corruption.
+ */
+int apfs_node_query(struct super_block *sb, struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+	int left, right;
+	int cmp;
+	int err;
+
+	if (query->flags & APFS_QUERY_PREV)
+		return apfs_node_prev(sb, query);
+	if (query->flags & APFS_QUERY_NEXT)
+		return apfs_node_next(sb, query);
+
+	/* Search by bisection */
+	cmp = 1;
+	left = 0;
+	do {
+		struct apfs_key curr_key;
+
+		if (cmp > 0) {
+			right = query->index - 1;
+			if (right < left) {
+				query->index = -1;
+				return -ENODATA;
+			}
+			query->index = (left + right) / 2;
+		} else {
+			left = query->index;
+			query->index = DIV_ROUND_UP(left + right, 2);
+		}
+
+		query->key_len = apfs_node_locate_key(node, query->index,
+						      &query->key_off);
+		err = apfs_key_from_query(query, &curr_key);
+		if (err) {
+			apfs_err(sb, "bad key for index %d", query->index);
+			return err;
+		}
+
+		cmp = apfs_keycmp(&curr_key, &query->key);
+		if (cmp == 0 && !(query->flags & APFS_QUERY_MULTIPLE))
+			break;
+	} while (left != right);
+
+	if (cmp > 0) {
+		query->index = -1;
+		return -ENODATA;
+	}
+
+	if (cmp != 0 && apfs_node_is_leaf(query->node) &&
+	    query->flags & APFS_QUERY_EXACT)
+		return -ENODATA;
+
+	if (query->flags & APFS_QUERY_MULTIPLE) {
+		if (cmp != 0) /* Last relevant entry in level */
+			query->flags |= APFS_QUERY_DONE;
+		query->flags |= APFS_QUERY_NEXT;
+	}
+
+	query->len = apfs_node_locate_value(node, query->index, &query->off);
+	return 0;
+}
+
+/**
+ * apfs_node_query_first - Find the first record in a node
+ * @query: on return this query points to the record
+ */
+void apfs_node_query_first(struct apfs_query *query)
+{
+	struct apfs_node *node = query->node;
+
+	query->index = 0;
+	query->key_len = apfs_node_locate_key(node, query->index, &query->key_off);
+	query->len = apfs_node_locate_value(node, query->index, &query->off);
+}
+
+/**
+ * apfs_omap_map_from_query - Read the mapping found by a successful omap query
+ * @query:	the query that found the record
+ * @map:	Return parameter.  The mapping found.
+ *
+ * Returns -EOPNOTSUPP if the object doesn't fit in one block, and -EFSCORRUPTED
+ * if the filesystem appears to be malicious.  Otherwise, reads the mapping info
+ * in the omap record into @map and returns 0.
+ */
+int apfs_omap_map_from_query(struct apfs_query *query, struct apfs_omap_map *map)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_omap_key *key = NULL;
+	struct apfs_omap_val *val = NULL;
+	char *raw = query->node->object.data;
+
+	if (query->len != sizeof(*val) || query->key_len != sizeof(*key)) {
+		apfs_err(sb, "bad length of key (%d) or value (%d)", query->key_len, query->len);
+		return -EFSCORRUPTED;
+	}
+	key = (struct apfs_omap_key *)(raw + query->key_off);
+	val = (struct apfs_omap_val *)(raw + query->off);
+
+	/* TODO: support objects with multiple blocks */
+	if (le32_to_cpu(val->ov_size) != sb->s_blocksize) {
+		apfs_err(sb, "object size doesn't match block size");
+		return -EOPNOTSUPP;
+	}
+
+	map->xid = le64_to_cpu(key->ok_xid);
+	map->bno = le64_to_cpu(val->ov_paddr);
+	map->flags = le32_to_cpu(val->ov_flags);
+	return 0;
+}
+
+/**
+ * apfs_btree_inc_height - Increase the height of a b-tree
+ * @query: query pointing to the root node
+ *
+ * On success returns 0, and @query is left pointing to the same record.
+ * Returns a negative error code in case of failure.
+ */
+static int apfs_btree_inc_height(struct apfs_query *query)
+{
+	struct apfs_query *root_query;
+	struct apfs_node *root = query->node;
+	struct apfs_node *new_node;
+	struct super_block *sb = root->object.sb;
+	struct apfs_btree_node_phys *root_raw;
+	struct apfs_btree_node_phys *new_raw;
+	struct apfs_btree_info *info;
+	__le64 *raw_oid;
+	u32 storage = apfs_query_storage(query);
+
+	root_raw = (void *)root->object.data;
+	apfs_assert_in_transaction(sb, &root_raw->btn_o);
+
+	if (query->parent || query->depth) {
+		apfs_err(sb, "invalid root query");
+		return -EFSCORRUPTED;
+	}
+
+	/* Create a new child node */
+	new_node = apfs_create_node(sb, storage);
+	if (IS_ERR(new_node)) {
+		apfs_err(sb, "node creation failed");
+		return PTR_ERR(new_node);
+	}
+	new_node->flags = root->flags & ~APFS_BTNODE_ROOT;
+	new_node->tree_type = root->tree_type;
+
+	/* Move all records into the child node; get rid of the info footer */
+	new_node->records = root->records;
+	new_node->key = root->key;
+	new_node->free = root->free;
+	new_node->val = root->val + sizeof(*info);
+	new_node->key_free_list_len = root->key_free_list_len;
+	new_node->val_free_list_len = root->val_free_list_len;
+	new_raw = (void *)new_node->object.data;
+	/* Don't copy the object header, already set by apfs_create_node() */
+	memcpy((void *)new_raw + sizeof(new_raw->btn_o),
+	       (void *)root_raw + sizeof(root_raw->btn_o),
+	       root->free - sizeof(new_raw->btn_o));
+	memcpy((void *)new_raw + new_node->val,
+	       (void *)root_raw + root->val,
+	       sb->s_blocksize - new_node->val);
+	query->off += sizeof(*info);
+	apfs_update_node(new_node);
+
+	/* Add a new level to the query chain */
+	root_query = query->parent = apfs_alloc_query(root, NULL /* parent */);
+	if (!query->parent) {
+		apfs_node_free(new_node);
+		return -ENOMEM;
+	}
+	root_query->key = query->key;
+	root_query->flags = query->flags;
+	query->node = new_node;
+	query->depth = 1;
+
+	/* Now assemble the new root with only the first key */
+	root_query->key_len = apfs_node_locate_key(root, 0 /* index */,
+						   &root_query->key_off);
+	if (!root_query->key_len) {
+		apfs_err(sb, "bad key for index %d", 0);
+		return -EFSCORRUPTED;
+	}
+	root->key = sizeof(*root_raw) +
+		    apfs_node_min_table_size(sb, root->tree_type, root->flags & ~APFS_BTNODE_LEAF);
+	memmove((void *)root_raw + root->key,
+		(void *)root_raw + root_query->key_off, root_query->key_len);
+	root_query->key_off = root->key;
+	root->free = root->key + root_query->key_len;
+
+	/* The new root is a nonleaf node; the record value is the child id */
+	root->flags &= ~APFS_BTNODE_LEAF;
+	root->val = sb->s_blocksize - sizeof(*info) - sizeof(*raw_oid);
+	raw_oid = (void *)root_raw + root->val;
+	*raw_oid = cpu_to_le64(new_node->object.oid);
+	root_query->off = root->val;
+	root_query->len = sizeof(*raw_oid);
+
+	/* With the key and value in place, set the table-of-contents */
+	root->records = 0;
+	root_query->index = 0;
+	apfs_create_toc_entry(root_query);
+
+	/* There is no internal fragmentation */
+	root->key_free_list_len = 0;
+	root->val_free_list_len = 0;
+
+	/* Finally, update the node count in the info footer */
+	apfs_btree_change_node_count(root_query, 1 /* change */);
+
+	le16_add_cpu(&root_raw->btn_level, 1); /* TODO: move to update_node() */
+	apfs_update_node(root);
+	return 0;
+}
+
+/**
+ * apfs_copy_record_range - Copy a range of records to an empty node
+ * @dest_node:	destination node
+ * @src_node:	source node
+ * @start:	index of first record in range
+ * @end:	index of first record after the range
+ *
+ * Doesn't modify the info footer of root nodes. Returns 0 on success or a
+ * negative error code in case of failure.
+ */
+static int apfs_copy_record_range(struct apfs_node *dest_node,
+				  struct apfs_node *src_node,
+				  int start, int end)
+{
+	struct super_block *sb = dest_node->object.sb;
+	struct apfs_btree_node_phys *dest_raw;
+	struct apfs_btree_node_phys *src_raw;
+	struct apfs_query *query = NULL;
+	int toc_size, toc_entry_size;
+	int err;
+	int i;
+
+	dest_raw = (void *)dest_node->object.data;
+	src_raw = (void *)src_node->object.data;
+
+	ASSERT(!dest_node->records);
+	apfs_assert_in_transaction(sb, &dest_raw->btn_o);
+
+	/* Resize the table of contents so that all the records fit */
+	if (apfs_node_has_fixed_kv_size(src_node))
+		toc_entry_size = sizeof(struct apfs_kvoff);
+	else
+		toc_entry_size = sizeof(struct apfs_kvloc);
+	toc_size = apfs_node_min_table_size(sb, src_node->tree_type, src_node->flags);
+	if (toc_size < toc_entry_size * (end - start))
+		toc_size = toc_entry_size * round_up(end - start, APFS_BTREE_TOC_ENTRY_INCREMENT);
+	dest_node->key = sizeof(*dest_raw) + toc_size;
+	dest_node->free = dest_node->key;
+	dest_node->val = sb->s_blocksize;
+	if (apfs_node_is_root(dest_node))
+		dest_node->val -= sizeof(struct apfs_btree_info);
+
+	/* We'll use a temporary query structure to move the records around */
+	query = apfs_alloc_query(dest_node, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	err = -EFSCORRUPTED;
+	for (i = start; i < end; ++i) {
+		int len, off;
+
+		len = apfs_node_locate_key(src_node, i, &off);
+		if (dest_node->free + len > sb->s_blocksize) {
+			apfs_err(sb, "key of length %d doesn't fit", len);
+			goto fail;
+		}
+		memcpy((char *)dest_raw + dest_node->free,
+		       (char *)src_raw + off, len);
+		query->key_off = dest_node->free;
+		query->key_len = len;
+		dest_node->free += len;
+
+		len = apfs_node_locate_value(src_node, i, &off);
+		dest_node->val -= len;
+		if (dest_node->val < 0) {
+			apfs_err(sb, "value of length %d doesn't fit", len);
+			goto fail;
+		}
+		memcpy((char *)dest_raw + dest_node->val,
+		       (char *)src_raw + off, len);
+		query->off = dest_node->val;
+		query->len = len;
+
+		query->index = i - start;
+		apfs_create_toc_entry(query);
+	}
+	err = 0;
+
+fail:
+	apfs_free_query(query);
+	return err;
+}
+
+/**
+ * apfs_attach_child - Attach a new node to its parent
+ * @query:	query pointing to the previous record in the parent
+ * @child:	the new child node to attach
+ *
+ * Returns 0 on success or a negative error code in case of failure (which may
+ * be -EAGAIN if a node split has happened and the caller must refresh and
+ * retry).
+ */
+static int apfs_attach_child(struct apfs_query *query, struct apfs_node *child)
+{
+	struct apfs_object *object = &child->object;
+	struct apfs_btree_node_phys *raw = (void *)object->data;
+	int key_len, key_off;
+	__le64 raw_oid = cpu_to_le64(object->oid);
+
+	key_len = apfs_node_locate_key(child, 0, &key_off);
+	if (!key_len) {
+		/* This should never happen: @child was made by us */
+		apfs_alert(object->sb, "bad key for index %d", 0);
+		return -EFSCORRUPTED;
+	}
+
+	return __apfs_btree_insert(query, (void *)raw + key_off, key_len, &raw_oid, sizeof(raw_oid));
+}
+
+/**
+ * apfs_node_temp_dup - Make an in-memory duplicate of a node
+ * @original:	node to duplicate
+ * @duplicate:	on success, the duplicate node
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_node_temp_dup(const struct apfs_node *original, struct apfs_node **duplicate)
+{
+	struct super_block *sb = original->object.sb;
+	struct apfs_node *dup = NULL;
+	char *buffer = NULL;
+
+	dup = kmalloc(sizeof(*dup), GFP_KERNEL);
+	if (!dup)
+		return -ENOMEM;
+	*dup = *original;
+	dup->object.o_bh = NULL;
+	dup->object.data = NULL;
+	dup->object.ephemeral = false;
+
+	buffer = kmalloc(sb->s_blocksize, GFP_KERNEL);
+	if (!buffer) {
+		kfree(dup);
+		return -ENOMEM;
+	}
+	memcpy(buffer, original->object.data, sb->s_blocksize);
+	dup->object.data = buffer;
+
+	*duplicate = dup;
+	return 0;
+}
+
+/**
+ * apfs_node_split - Split a b-tree node in two
+ * @query: query pointing to the node
+ *
+ * On success returns 0, and @query is left pointing to the same record on the
+ * tip; to simplify the implementation, @query->parent is set to NULL. Returns
+ * a negative error code in case of failure, which may be -EAGAIN if a node
+ * split has happened and the caller must refresh and retry.
+ */
+int apfs_node_split(struct apfs_query *query)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_node *old_node = NULL, *new_node = NULL, *tmp_node = NULL;
+	struct apfs_btree_node_phys *new_raw = NULL, *old_raw = NULL;
+	u32 storage = apfs_query_storage(query);
+	int record_count, new_rec_count, old_rec_count;
+	int err;
+
+	apfs_assert_query_is_valid(query);
+
+	if (apfs_node_is_root(query->node)) {
+		err = apfs_btree_inc_height(query);
+		if (err) {
+			apfs_err(sb, "failed to increase tree height");
+			return err;
+		}
+	} else if (!query->parent) {
+		apfs_err(sb, "nonroot node with no parent");
+		return -EFSCORRUPTED;
+	}
+	old_node = query->node;
+
+	old_raw = (void *)old_node->object.data;
+	apfs_assert_in_transaction(sb, &old_raw->btn_o);
+
+	/*
+	 * To defragment the original node, we put all records in a temporary
+	 * in-memory node before dealing them out.
+	 */
+	err = apfs_node_temp_dup(old_node, &tmp_node);
+	if (err)
+		return err;
+
+	record_count = old_node->records;
+	if (record_count == 1) {
+		apfs_alert(sb, "splitting node with a single record");
+		err = -EFSCORRUPTED;
+		goto out;
+	}
+	new_rec_count = record_count / 2;
+	old_rec_count = record_count - new_rec_count;
+
+	/*
+	 * The second half of the records go into a new node. This is done
+	 * before the first half to avoid committing to any actual changes
+	 * until we know for sure that no ancestor splits are expected.
+	 */
+
+	new_node = apfs_create_node(sb, storage);
+	if (IS_ERR(new_node)) {
+		apfs_err(sb, "node creation failed");
+		err = PTR_ERR(new_node);
+		new_node = NULL;
+		goto out;
+	}
+	new_node->tree_type = old_node->tree_type;
+	new_node->flags = old_node->flags;
+	new_node->records = 0;
+	new_node->key_free_list_len = 0;
+	new_node->val_free_list_len = 0;
+	err = apfs_copy_record_range(new_node, tmp_node, old_rec_count, record_count);
+	if (err) {
+		apfs_err(sb, "record copy failed");
+		goto out;
+	}
+	new_raw = (void *)new_node->object.data;
+	apfs_assert_in_transaction(sb, &new_raw->btn_o);
+	new_raw->btn_level = old_raw->btn_level;
+	apfs_update_node(new_node);
+
+	err = apfs_attach_child(query->parent, new_node);
+	if (err) {
+		if (err != -EAGAIN) {
+			apfs_err(sb, "child attachment failed");
+			goto out;
+		}
+		err = apfs_delete_node(new_node, query->flags & APFS_QUERY_TREE_MASK);
+		if (err) {
+			apfs_err(sb, "node cleanup failed for query retry");
+			goto out;
+		}
+		err = -EAGAIN;
+		goto out;
+	}
+	apfs_assert_query_is_valid(query->parent);
+	apfs_btree_change_node_count(query->parent, 1 /* change */);
+
+	/*
+	 * No more risk of ancestor splits, now actual changes can be made. The
+	 * first half of the records go into the original node.
+	 */
+
+	old_node->records = 0;
+	old_node->key_free_list_len = 0;
+	old_node->val_free_list_len = 0;
+	err = apfs_copy_record_range(old_node, tmp_node, 0, old_rec_count);
+	if (err) {
+		apfs_err(sb, "record copy failed");
+		goto out;
+	}
+	apfs_update_node(old_node);
+
+	/* Point the query back to the original record */
+	if (query->index >= old_rec_count) {
+		/* The record got moved to the new node */
+		apfs_node_free(query->node);
+		query->node = new_node;
+		new_node = NULL;
+		query->index -= old_rec_count;
+	}
+
+	/*
+	 * This could be avoided in most cases, and queries could get refreshed
+	 * only when really orphaned. But refreshing queries is probably not a
+	 * bottleneck, and trying to be clever with this stuff has caused me a
+	 * lot of trouble already.
+	 */
+	apfs_free_query(query->parent);
+	query->parent = NULL; /* The caller only gets the leaf */
+
+out:
+	apfs_node_free(new_node);
+	apfs_node_free(tmp_node);
+	return err;
+}
+
+/* TODO: the following 4 functions could be reused elsewhere */
+
+/**
+ * apfs_off_to_val_off - Translate offset in node to offset in value area
+ * @node:	the node
+ * @off:	offset in the node
+ */
+static u16 apfs_off_to_val_off(struct apfs_node *node, u16 off)
+{
+	struct super_block *sb = node->object.sb;
+	u16 val_end;
+
+	val_end = sb->s_blocksize;
+	if (apfs_node_is_root(node)) /* has footer */
+		val_end -= sizeof(struct apfs_btree_info);
+	return val_end - off;
+}
+
+/**
+ * apfs_val_off_to_off - Translate offset in value area to offset in node
+ * @node:	the node
+ * @off:	offset in the value area
+ */
+static u16 apfs_val_off_to_off(struct apfs_node *node, u16 off)
+{
+	return apfs_off_to_val_off(node, off);
+}
+
+/**
+ * apfs_off_to_key_off - Translate offset in node to offset in key area
+ * @node:	the node
+ * @off:	offset in the node
+ */
+static u16 apfs_off_to_key_off(struct apfs_node *node, u16 off)
+{
+	return off - node->key;
+}
+
+/**
+ * apfs_key_off_to_off - Translate offset in key area to offset in node
+ * @node:	the node
+ * @off:	offset in the key area
+ */
+static u16 apfs_key_off_to_off(struct apfs_node *node, u16 off)
+{
+	return off + node->key;
+}
+
+/* The type of the previous four functions, used for node offset calculations */
+typedef u16 (*offcalc)(struct apfs_node *, u16);
+
+/**
+ * apfs_node_free_list_add - Add a free node segment to the proper free list
+ * @node:	node for the segment
+ * @off:	offset of the segment to add
+ * @len:	length of the segment to add
+ *
+ * The caller must ensure that the freed segment fits in the node.
+ */
+static void apfs_node_free_list_add(struct apfs_node *node, u16 off, u16 len)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw = (void *)node->object.data;
+	struct apfs_nloc *head, *new;
+	offcalc off_to_rel;
+
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+
+	if (off >= node->val) { /* Value area */
+		off_to_rel = apfs_off_to_val_off;
+		head = &node_raw->btn_val_free_list;
+		node->val_free_list_len += len;
+	} else { /* Key area */
+		off_to_rel = apfs_off_to_key_off;
+		head = &node_raw->btn_key_free_list;
+		node->key_free_list_len += len;
+	}
+
+	/* Very small segments are leaked until defragmentation */
+	if (len < sizeof(*new))
+		return;
+
+	/* The free list doesn't seem to be kept in any particular order */
+	new = (void *)node_raw + off;
+	new->off = head->off;
+	new->len = cpu_to_le16(len);
+	head->off = cpu_to_le16(off_to_rel(node, off));
+}
+
+/**
+ * apfs_node_free_range - Free space from a node's key or value areas
+ * @node:	the node
+ * @off:	offset to free
+ * @len:	length to free
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+void apfs_node_free_range(struct apfs_node *node, u16 off, u16 len)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *raw = (void *)node->object.data;
+
+	apfs_assert_in_transaction(sb, &raw->btn_o);
+
+	if (off == node->val)
+		node->val += len;
+	else if (off + len == node->free)
+		node->free -= len;
+	else
+		apfs_node_free_list_add(node, off, len);
+}
+
+/**
+ * apfs_node_free_list_unlink - Unlink an entry from a node's free list
+ * @prev:	previous entry
+ * @curr:	entry to unlink
+ */
+static void apfs_node_free_list_unlink(struct apfs_nloc *prev, struct apfs_nloc *curr)
+{
+	prev->off = curr->off;
+}
+
+/**
+ * apfs_node_free_list_alloc - Allocate a free segment from a free list
+ * @node:	the node
+ * @len:	length to allocate
+ * @value:	true to allocate in the value area, false for the key area
+ *
+ * Returns the offset in the node on success, or a negative error code in case
+ * of failure, which may be -ENOSPC if the node seems full.
+ */
+static int apfs_node_free_list_alloc(struct apfs_node *node, u16 len, bool value)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw = (void *)node->object.data;
+	struct apfs_nloc *head, *curr, *prev;
+	offcalc rel_to_off;
+	int *list_len;
+	int bound = sb->s_blocksize;
+
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+
+	if (value) { /* Value area */
+		rel_to_off = apfs_val_off_to_off;
+		head = &node_raw->btn_val_free_list;
+		list_len = &node->val_free_list_len;
+	} else { /* Key area */
+		rel_to_off = apfs_key_off_to_off;
+		head = &node_raw->btn_key_free_list;
+		list_len = &node->key_free_list_len;
+	}
+
+	if (*list_len < len)
+		return -ENOSPC;
+
+	prev = head;
+	while (bound--) {
+		u16 curr_off = le16_to_cpu(prev->off);
+		u16 abs_off = rel_to_off(node, curr_off);
+		u16 curr_len;
+
+		if (curr_off == APFS_BTOFF_INVALID)
+			return -ENOSPC;
+		if (abs_off + sizeof(*curr) > sb->s_blocksize) {
+			apfs_err(sb, "nloc out of bounds (%d-%d)", abs_off, (int)sizeof(*curr));
+			return -EFSCORRUPTED;
+		}
+		curr = (void *)node_raw + abs_off;
+
+		curr_len = le16_to_cpu(curr->len);
+		if (curr_len >= len) {
+			if (abs_off + curr_len > sb->s_blocksize) {
+				apfs_err(sb, "entry out of bounds (%d-%d)", abs_off, curr_len);
+				return -EFSCORRUPTED;
+			}
+			*list_len -= curr_len;
+			apfs_node_free_list_unlink(prev, curr);
+			apfs_node_free_list_add(node, abs_off + len, curr_len - len);
+			return abs_off;
+		}
+
+		prev = curr;
+	}
+
+	/* Don't loop forever if the free list is corrupted and doesn't end */
+	apfs_err(sb, "free list never ends");
+	return -EFSCORRUPTED;
+}
+
+/**
+ * apfs_node_alloc_key - Allocated free space for a new key
+ * @node:	the node to search
+ * @len:	wanted key length
+ *
+ * Returns the offset in the node on success, or a negative error code in case
+ * of failure, which may be -ENOSPC if the node seems full.
+ */
+static int apfs_node_alloc_key(struct apfs_node *node, u16 len)
+{
+	int off;
+
+	if (node->free + len <= node->val) {
+		off = node->free;
+		node->free += len;
+		return off;
+	}
+	return apfs_node_free_list_alloc(node, len, false /* value */);
+}
+
+/**
+ * apfs_node_alloc_val - Allocated free space for a new value
+ * @node:	the node to search
+ * @len:	wanted value length
+ *
+ * Returns the offset in the node on success, or a negative error code in case
+ * of failure, which may be -ENOSPC if the node seems full.
+ */
+static int apfs_node_alloc_val(struct apfs_node *node, u16 len)
+{
+	int off;
+
+	if (node->free + len <= node->val) {
+		off = node->val - len;
+		node->val -= len;
+		return off;
+	}
+	return apfs_node_free_list_alloc(node, len, true /* value */);
+}
+
+/**
+ * apfs_node_total_room - Total free space in a node
+ * @node: the node
+ */
+static int apfs_node_total_room(struct apfs_node *node)
+{
+	return node->val - node->free + node->key_free_list_len + node->val_free_list_len;
+}
+
+/**
+ * apfs_node_has_room - Check if a node has room for insertion or replacement
+ * @node:	node to check
+ * @length:	length of the needed space (may be negative on replace)
+ * @replace:	are we replacing a record?
+ */
+bool apfs_node_has_room(struct apfs_node *node, int length, bool replace)
+{
+	struct apfs_btree_node_phys *node_raw = (void *)node->object.data;
+	int toc_entry_size, needed_room;
+
+	if (apfs_node_has_fixed_kv_size(node))
+		toc_entry_size = sizeof(struct apfs_kvoff);
+	else
+		toc_entry_size = sizeof(struct apfs_kvloc);
+
+	needed_room = length;
+	if (!replace) {
+		if (sizeof(*node_raw) + (node->records + 1) * toc_entry_size > node->key)
+			needed_room += APFS_BTREE_TOC_ENTRY_INCREMENT * toc_entry_size;
+	}
+
+	return apfs_node_total_room(node) >= needed_room;
+}
+
+/**
+ * apfs_defragment_node - Make all free space in a node contiguous
+ * @node: node to defragment
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_defragment_node(struct apfs_node *node)
+{
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw = (void *)node->object.data;
+	struct apfs_node *tmp_node = NULL;
+	int record_count, err;
+
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+
+	/* Put all records in a temporary in-memory node and deal them out */
+	err = apfs_node_temp_dup(node, &tmp_node);
+	if (err)
+		return err;
+	record_count = node->records;
+	node->records = 0;
+	node->key_free_list_len = 0;
+	node->val_free_list_len = 0;
+	err = apfs_copy_record_range(node, tmp_node, 0, record_count);
+	if (err) {
+		apfs_err(sb, "record copy failed");
+		goto fail;
+	}
+	apfs_update_node(node);
+fail:
+	apfs_node_free(tmp_node);
+	return err;
+}
+
+/**
+ * apfs_node_update_toc_entry - Update a table of contents entry in place
+ * @query: query pointing to the toc entry
+ *
+ * The toc entry gets updated with the length and offset for the key/value
+ * provided by @query. Don't call this function for nodes with fixed length
+ * key/values, those never need to update their toc entries.
+ */
+static void apfs_node_update_toc_entry(struct apfs_query *query)
+{
+	struct super_block *sb = NULL;
+	struct apfs_node *node = NULL;
+	struct apfs_btree_node_phys *node_raw = NULL;
+	struct apfs_kvloc *kvloc = NULL;
+	int value_end;
+
+	node = query->node;
+	ASSERT(!apfs_node_has_fixed_kv_size(node));
+	sb = node->object.sb;
+	node_raw = (void *)node->object.data;
+
+	value_end = sb->s_blocksize;
+	if (apfs_node_is_root(node))
+		value_end -= sizeof(struct apfs_btree_info);
+
+	kvloc = (struct apfs_kvloc *)node_raw->btn_data + query->index;
+	kvloc->v.off = cpu_to_le16(value_end - query->off);
+	kvloc->v.len = cpu_to_le16(query->len);
+	kvloc->k.off = cpu_to_le16(query->key_off - node->key);
+	kvloc->k.len = cpu_to_le16(query->key_len);
+}
+
+/**
+ * apfs_node_replace - Replace a record in a node that has enough room
+ * @query:	exact query that found the record
+ * @key:	new on-disk record key (NULL if unchanged)
+ * @key_len:	length of @key
+ * @val:	new on-disk record value (NULL if unchanged)
+ * @val_len:	length of @val
+ *
+ * Returns 0 on success, and @query is left pointing to the same record. Returns
+ * a negative error code in case of failure.
+ */
+int apfs_node_replace(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw = (void *)node->object.data;
+	int key_off = 0, val_off = 0, err = 0;
+	bool defragged = false;
+	int qtree = query->flags & APFS_QUERY_TREE_MASK;
+
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+
+	/*
+	 * Free queues are weird because their tables of contents don't report
+	 * record lengths, as if they were fixed, but some of the leaf values
+	 * are actually "ghosts", that is, zero-length. Supporting replace of
+	 * such records would require some changes, and so far I've had no need
+	 * for it.
+	 */
+	(void)qtree;
+	ASSERT(!(qtree == APFS_QUERY_FREE_QUEUE && apfs_node_is_leaf(node)));
+
+retry:
+	if (key) {
+		if (key_len <= query->key_len) {
+			u16 end = query->key_off + key_len;
+			u16 diff = query->key_len - key_len;
+
+			apfs_node_free_range(node, end, diff);
+			key_off = query->key_off;
+		} else {
+			apfs_node_free_range(node, query->key_off, query->key_len);
+			key_off = apfs_node_alloc_key(node, key_len);
+			if (key_off < 0) {
+				if (key_off == -ENOSPC)
+					goto defrag;
+				return key_off;
+			}
+		}
+	}
+
+	if (val) {
+		if (val_len <= query->len) {
+			u16 end = query->off + val_len;
+			u16 diff = query->len - val_len;
+
+			apfs_node_free_range(node, end, diff);
+			val_off = query->off;
+		} else {
+			apfs_node_free_range(node, query->off, query->len);
+			val_off = apfs_node_alloc_val(node, val_len);
+			if (val_off < 0) {
+				if (val_off == -ENOSPC)
+					goto defrag;
+				return val_off;
+			}
+		}
+	}
+
+	if (key) {
+		query->key_off = key_off;
+		query->key_len = key_len;
+		memcpy((void *)node_raw + key_off, key, key_len);
+	}
+	if (val) {
+		query->off = val_off;
+		query->len = val_len;
+		memcpy((void *)node_raw + val_off, val, val_len);
+	}
+
+	/* If the key or value were resized, update the table of contents */
+	if (!apfs_node_has_fixed_kv_size(node))
+		apfs_node_update_toc_entry(query);
+
+	apfs_update_node(node);
+	return 0;
+
+defrag:
+	if (defragged) {
+		apfs_alert(sb, "no room in defragged node");
+		return -EFSCORRUPTED;
+	}
+
+	/* Crush the replaced entry, so that defragmentation is complete */
+	if (apfs_node_has_fixed_kv_size(node)) {
+		apfs_alert(sb, "failed to replace a fixed size record");
+		return -EFSCORRUPTED;
+	}
+	if (key)
+		query->key_len = 0;
+	if (val)
+		query->len = 0;
+	apfs_node_update_toc_entry(query);
+
+	err = apfs_defragment_node(node);
+	if (err) {
+		apfs_err(sb, "failed to defragment node");
+		return err;
+	}
+	defragged = true;
+
+	/* The record to replace probably moved around */
+	query->len = apfs_node_locate_value(query->node, query->index, &query->off);
+	query->key_len = apfs_node_locate_key(query->node, query->index, &query->key_off);
+	goto retry;
+}
+
+/**
+ * apfs_node_insert - Insert a new record in a node that has enough room
+ * @query:	query run to search for the record
+ * @key:	on-disk record key
+ * @key_len:	length of @key
+ * @val:	on-disk record value (NULL for ghost records)
+ * @val_len:	length of @val (0 for ghost records)
+ *
+ * The new record is placed right after the one found by @query. On success,
+ * returns 0 and sets @query to the new record. In case of failure, returns a
+ * negative error code and leaves @query pointing to the same record.
+ */
+int apfs_node_insert(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct apfs_node *node = query->node;
+	struct super_block *sb = node->object.sb;
+	struct apfs_btree_node_phys *node_raw = (void *)node->object.data;
+	int toc_entry_size;
+	int key_off, val_off, err;
+	bool defragged = false;
+
+	apfs_assert_in_transaction(sb, &node_raw->btn_o);
+
+retry:
+	if (apfs_node_has_fixed_kv_size(node))
+		toc_entry_size = sizeof(struct apfs_kvoff);
+	else
+		toc_entry_size = sizeof(struct apfs_kvloc);
+
+	/* Expand the table of contents if necessary */
+	if (sizeof(*node_raw) + (node->records + 1) * toc_entry_size > node->key) {
+		int new_key_base = node->key;
+		int new_free_base = node->free;
+		int inc;
+
+		inc = APFS_BTREE_TOC_ENTRY_INCREMENT * toc_entry_size;
+
+		new_key_base += inc;
+		new_free_base += inc;
+		if (new_free_base > node->val)
+			goto defrag;
+		memmove((void *)node_raw + new_key_base,
+			(void *)node_raw + node->key, node->free - node->key);
+
+		node->key = new_key_base;
+		node->free = new_free_base;
+		query->key_off += inc;
+	}
+
+	key_off = apfs_node_alloc_key(node, key_len);
+	if (key_off < 0) {
+		if (key_off == -ENOSPC)
+			goto defrag;
+		return key_off;
+	}
+
+	if (val) {
+		val_off = apfs_node_alloc_val(node, val_len);
+		if (val_off < 0) {
+			if (val_off == -ENOSPC) {
+				/*
+				 * There is no need for an update of the on-disk
+				 * node before the defrag, since only in-memory
+				 * data should be used there...
+				 */
+				goto defrag;
+			}
+			return val_off;
+		}
+	}
+
+	query->key_len = key_len;
+	query->key_off = key_off;
+	memcpy((void *)node_raw + key_off, key, key_len);
+
+	query->len = val_len;
+	if (val) {
+		query->off = val_off;
+		memcpy((void *)node_raw + val_off, val, val_len);
+	} else {
+		query->off = 0;
+	}
+
+	query->index++; /* The query returned the record right before @key */
+
+	/* Add the new entry to the table of contents */
+	apfs_create_toc_entry(query);
+
+	apfs_update_node(node);
+	return 0;
+
+defrag:
+	if (defragged) {
+		apfs_err(sb, "node reports incorrect free space");
+		return -EFSCORRUPTED;
+	}
+	err = apfs_defragment_node(node);
+	if (err) {
+		apfs_err(sb, "failed to defragment node");
+		return err;
+	}
+	defragged = true;
+	goto retry;
+}
+
+/**
+ * apfs_create_single_rec_node - Creates a new node with a single record
+ * @query:	query run to search for the record
+ * @key:	on-disk record key
+ * @key_len:	length of @key
+ * @val:	on-disk record value
+ * @val_len:	length of @val
+ *
+ * The new node is placed right after the one found by @query, which must have
+ * a single record. On success, returns 0 and sets @query to the new record;
+ * returns a negative error code in case of failure, which may be -EAGAIN if a
+ * node split has happened and the caller must refresh and retry.
+ */
+int apfs_create_single_rec_node(struct apfs_query *query, void *key, int key_len, void *val, int val_len)
+{
+	struct super_block *sb = NULL;
+	struct apfs_node *new_node = NULL, *prev_node = NULL;
+	struct apfs_btree_node_phys *prev_raw = NULL;
+	struct apfs_btree_node_phys *new_raw = NULL;
+	int err;
+
+	prev_node = query->node;
+	sb = prev_node->object.sb;
+
+	ASSERT(query->parent);
+	ASSERT(prev_node->records == 1);
+	ASSERT(val && val_len);
+
+	/* This function should only be needed for huge catalog records */
+	if (prev_node->tree_type != APFS_OBJECT_TYPE_FSTREE) {
+		apfs_err(sb, "huge node records in the wrong tree");
+		return -EFSCORRUPTED;
+	}
+
+	/*
+	 * This will only be called for leaf nodes because it's the values that
+	 * can get huge, not the keys. It will also never be called for root,
+	 * because the catalog always has more than a single record.
+	 */
+	if (apfs_node_is_root(prev_node) || !apfs_node_is_leaf(prev_node)) {
+		apfs_err(sb, "huge record in index node");
+		return -EFSCORRUPTED;
+	}
+
+	new_node = apfs_create_node(sb, apfs_query_storage(query));
+	if (IS_ERR(new_node)) {
+		apfs_err(sb, "node creation failed");
+		return PTR_ERR(new_node);
+	}
+	new_node->tree_type = prev_node->tree_type;
+	new_node->flags = prev_node->flags;
+	new_node->records = 0;
+	new_node->key_free_list_len = 0;
+	new_node->val_free_list_len = 0;
+	new_node->key = new_node->free = sizeof(*new_raw);
+	new_node->val = sb->s_blocksize; /* Nonroot */
+
+	prev_raw = (void *)prev_node->object.data;
+	new_raw = (void *)new_node->object.data;
+	apfs_assert_in_transaction(sb, &new_raw->btn_o);
+	new_raw->btn_level = prev_raw->btn_level;
+	apfs_update_node(new_node);
+
+	query->node = new_node;
+	new_node = NULL;
+	query->index = -1;
+	err = apfs_node_insert(query, key, key_len, val, val_len);
+	if (err) {
+		apfs_err(sb, "node record insertion failed");
+		goto fail;
+	}
+
+	err = apfs_attach_child(query->parent, query->node);
+	if (err) {
+		if (err != -EAGAIN) {
+			apfs_err(sb, "child attachment failed");
+			goto fail;
+		}
+		err = apfs_delete_node(query->node, query->flags & APFS_QUERY_TREE_MASK);
+		if (err) {
+			apfs_err(sb, "node cleanup failed for query retry");
+			goto fail;
+		}
+
+		/*
+		 * The query must remain pointing to the original node for the
+		 * refresh to take place. The index will not matter though.
+		 */
+		new_node = query->node;
+		query->node = prev_node;
+		prev_node = NULL;
+		err = -EAGAIN;
+		goto fail;
+	}
+	apfs_btree_change_node_count(query->parent, 1 /* change */);
+
+fail:
+	apfs_node_free(prev_node);
+	apfs_node_free(new_node);
+	return err;
+}
diff --git a/fs/apfs/object.c b/fs/apfs/object.c
new file mode 100644
index 000000000..9636a9a29
--- /dev/null
+++ b/fs/apfs/object.c
@@ -0,0 +1,315 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Checksum routines for an APFS object
+ */
+
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include "apfs.h"
+
+/*
+ * Note that this is not a generic implementation of fletcher64, as it assumes
+ * a message length that doesn't overflow sum1 and sum2.  This constraint is ok
+ * for apfs, though, since the block size is limited to 2^16.  For a more
+ * generic optimized implementation, see Nakassis (1988).
+ */
+static u64 apfs_fletcher64(void *addr, size_t len)
+{
+	__le32 *buff = addr;
+	u64 sum1 = 0;
+	u64 sum2 = 0;
+	u64 c1, c2;
+	int i, count_32;
+
+	count_32 = len >> 2;
+	for (i = 0; i < count_32; i++) {
+		sum1 += le32_to_cpu(buff[i]);
+		sum2 += sum1;
+	}
+
+	c1 = sum1 + sum2;
+	c1 = 0xFFFFFFFF - do_div(c1, 0xFFFFFFFF);
+	c2 = sum1 + c1;
+	c2 = 0xFFFFFFFF - do_div(c2, 0xFFFFFFFF);
+
+	return (c2 << 32) | c1;
+}
+
+int apfs_obj_verify_csum(struct super_block *sb, struct buffer_head *bh)
+{
+	/* The checksum may be stale until the transaction is committed */
+	if (buffer_trans(bh))
+		return 1;
+	return apfs_multiblock_verify_csum(bh->b_data, sb->s_blocksize);
+}
+
+/**
+ * apfs_multiblock_verify_csum - Verify an object's checksum
+ * @object:	the object to verify
+ * @size:	size of the object in bytes (may be multiple blocks)
+ *
+ * Returns 1 on success, 0 on failure.
+ */
+int apfs_multiblock_verify_csum(char *object, u32 size)
+{
+	struct apfs_obj_phys *obj = (struct apfs_obj_phys *)object;
+	u64 actual_csum, header_csum;
+
+	header_csum = le64_to_cpu(obj->o_cksum);
+	actual_csum = apfs_fletcher64(object + APFS_MAX_CKSUM_SIZE, size - APFS_MAX_CKSUM_SIZE);
+	return header_csum == actual_csum;
+}
+
+/**
+ * apfs_obj_set_csum - Set the fletcher checksum in an object header
+ * @sb:		superblock structure
+ * @obj:	the object header
+ *
+ * The object must have a length of a single block.
+ */
+void apfs_obj_set_csum(struct super_block *sb, struct apfs_obj_phys *obj)
+{
+	apfs_multiblock_set_csum((char *)obj, sb->s_blocksize);
+}
+
+/**
+ * apfs_multiblock_set_csum - Set an object's checksum
+ * @object:	the object to checksum
+ * @size:	size of the object in bytes (may be multiple blocks)
+ */
+void apfs_multiblock_set_csum(char *object, u32 size)
+{
+	struct apfs_obj_phys *obj = (struct apfs_obj_phys *)object;
+	u64 cksum;
+
+	cksum = apfs_fletcher64(object + APFS_MAX_CKSUM_SIZE, size - APFS_MAX_CKSUM_SIZE);
+	obj->o_cksum = cpu_to_le64(cksum);
+}
+
+/**
+ * apfs_create_cpm_block - Create a new checkpoint-mapping block
+ * @sb:		filesystem superblock
+ * @bno:	block number to use
+ * @bh_p:	on return, the buffer head for the block
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_create_cpm_block(struct super_block *sb, u64 bno, struct buffer_head **bh_p)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_checkpoint_map_phys *cpm = NULL;
+	struct buffer_head *bh = NULL;
+	int err;
+
+	bh = apfs_getblk(sb, bno);
+	if (!bh) {
+		apfs_err(sb, "failed to map cpm block");
+		return -EIO;
+	}
+	err = apfs_transaction_join(sb, bh);
+	if (err) {
+		brelse(bh);
+		return err;
+	}
+	set_buffer_csum(bh);
+
+	cpm = (void *)bh->b_data;
+	memset(cpm, 0, sb->s_blocksize);
+	cpm->cpm_o.o_oid = cpu_to_le64(bno);
+	cpm->cpm_o.o_xid = cpu_to_le64(nxi->nx_xid);
+	cpm->cpm_o.o_type = cpu_to_le32(APFS_OBJ_PHYSICAL | APFS_OBJECT_TYPE_CHECKPOINT_MAP);
+	cpm->cpm_o.o_subtype = cpu_to_le32(APFS_OBJECT_TYPE_INVALID);
+
+	/* For now: the caller will have to update these fields */
+	cpm->cpm_flags = cpu_to_le32(APFS_CHECKPOINT_MAP_LAST);
+	cpm->cpm_count = 0;
+
+	*bh_p = bh;
+	return 0;
+}
+
+/**
+ * apfs_create_cpoint_map - Create a checkpoint mapping for an object
+ * @sb:		filesystem superblock
+ * @cpm:	checkpoint mapping block to use
+ * @obj:	header for the ephemeral object
+ * @bno:	block number for the ephemeral object
+ * @size:	size of the ephemeral object in bytes
+ *
+ * Returns 0 on success or a negative error code in case of failure, which may
+ * be -ENOSPC if @cpm is full.
+ */
+int apfs_create_cpoint_map(struct super_block *sb, struct apfs_checkpoint_map_phys *cpm, struct apfs_obj_phys *obj, u64 bno, u32 size)
+{
+	struct apfs_checkpoint_mapping *map = NULL;
+	u32 cpm_count;
+
+	apfs_assert_in_transaction(sb, &cpm->cpm_o);
+
+	cpm_count = le32_to_cpu(cpm->cpm_count);
+	if (cpm_count >= apfs_max_maps_per_block(sb))
+		return -ENOSPC;
+	map = &cpm->cpm_map[cpm_count];
+	le32_add_cpu(&cpm->cpm_count, 1);
+
+	map->cpm_type = obj->o_type;
+	map->cpm_subtype = obj->o_subtype;
+	map->cpm_size = cpu_to_le32(size);
+	map->cpm_pad = 0;
+	map->cpm_fs_oid = 0;
+	map->cpm_oid = obj->o_oid;
+	map->cpm_paddr = cpu_to_le64(bno);
+	return 0;
+}
+
+/**
+ * apfs_index_in_data_area - Get position of block in current checkpoint's data
+ * @sb:		superblock structure
+ * @bno:	block number
+ */
+u32 apfs_index_in_data_area(struct super_block *sb, u64 bno)
+{
+	struct apfs_nx_superblock *raw_sb = APFS_NXI(sb)->nx_raw;
+	u64 data_base = le64_to_cpu(raw_sb->nx_xp_data_base);
+	u32 data_index = le32_to_cpu(raw_sb->nx_xp_data_index);
+	u32 data_blks = le32_to_cpu(raw_sb->nx_xp_data_blocks);
+	u64 tmp;
+
+	tmp = bno - data_base + data_blks - data_index;
+	return do_div(tmp, data_blks);
+}
+
+/**
+ * apfs_data_index_to_bno - Convert index in data area to block number
+ * @sb:		superblock structure
+ * @index:	index of the block in the current checkpoint's data area
+ */
+u64 apfs_data_index_to_bno(struct super_block *sb, u32 index)
+{
+	struct apfs_nx_superblock *raw_sb = APFS_NXI(sb)->nx_raw;
+	u64 data_base = le64_to_cpu(raw_sb->nx_xp_data_base);
+	u32 data_index = le32_to_cpu(raw_sb->nx_xp_data_index);
+	u32 data_blks = le32_to_cpu(raw_sb->nx_xp_data_blocks);
+
+	return data_base + (index + data_index) % data_blks;
+}
+
+/**
+ * apfs_ephemeral_object_lookup - Find an ephemeral object info in memory
+ * @sb:		superblock structure
+ * @oid:	ephemeral object id
+ *
+ * Returns a pointer to the object info on success, or an error pointer in case
+ * of failure.
+ */
+struct apfs_ephemeral_object_info *apfs_ephemeral_object_lookup(struct super_block *sb, u64 oid)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_ephemeral_object_info *list = NULL;
+	int i;
+
+	list = nxi->nx_eph_list;
+	for (i = 0; i < nxi->nx_eph_count; ++i) {
+		if (list[i].oid == oid)
+			return &list[i];
+	}
+	apfs_err(sb, "no mapping for oid 0x%llx", oid);
+	return ERR_PTR(-EFSCORRUPTED);
+}
+
+/**
+ * apfs_read_object_block - Map a non-ephemeral object block
+ * @sb:		superblock structure
+ * @bno:	block number for the object
+ * @write:	request write access?
+ * @preserve:	preserve the old block?
+ *
+ * On success returns the mapped buffer head for the object, which may now be
+ * in a new location if write access was requested.  Returns an error pointer
+ * in case of failure.
+ */
+struct buffer_head *apfs_read_object_block(struct super_block *sb, u64 bno, bool write, bool preserve)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_superblock *vsb_raw = NULL;
+	struct buffer_head *bh, *new_bh;
+	struct apfs_obj_phys *obj;
+	u32 type;
+	u64 new_bno;
+	int err;
+
+	ASSERT(write || !preserve);
+
+	bh = apfs_sb_bread(sb, bno);
+	if (!bh) {
+		apfs_err(sb, "failed to read object block 0x%llx", bno);
+		return ERR_PTR(-EIO);
+	}
+
+	obj = (struct apfs_obj_phys *)bh->b_data;
+	type = le32_to_cpu(obj->o_type);
+	ASSERT(!(type & APFS_OBJ_EPHEMERAL));
+	if (nxi->nx_flags & APFS_CHECK_NODES && !apfs_obj_verify_csum(sb, bh)) {
+		apfs_err(sb, "bad checksum for object in block 0x%llx", bno);
+		err = -EFSBADCRC;
+		goto fail;
+	}
+
+	if (!write)
+		return bh;
+	ASSERT(!(sb->s_flags & SB_RDONLY));
+
+	/* Is the object already part of the current transaction? */
+	if (obj->o_xid == cpu_to_le64(nxi->nx_xid))
+		return bh;
+
+	err = apfs_spaceman_allocate_block(sb, &new_bno, true /* backwards */);
+	if (err) {
+		apfs_err(sb, "block allocation failed");
+		goto fail;
+	}
+	new_bh = apfs_getblk(sb, new_bno);
+	if (!new_bh) {
+		apfs_err(sb, "failed to map block for CoW (0x%llx)", new_bno);
+		err = -EIO;
+		goto fail;
+	}
+	memcpy(new_bh->b_data, bh->b_data, sb->s_blocksize);
+
+	/*
+	 * Don't free the old copy of the object if it's part of a snapshot.
+	 * Also increase the allocation count, except for the volume superblock
+	 * which is never counted there.
+	 */
+	if (!preserve) {
+		err = apfs_free_queue_insert(sb, bh->b_blocknr, 1);
+		if (err)
+			apfs_err(sb, "free queue insertion failed for 0x%llx", (unsigned long long)bh->b_blocknr);
+	} else if ((type & APFS_OBJECT_TYPE_MASK) != APFS_OBJECT_TYPE_FS) {
+		vsb_raw = APFS_SB(sb)->s_vsb_raw;
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		le64_add_cpu(&vsb_raw->apfs_fs_alloc_count, 1);
+		le64_add_cpu(&vsb_raw->apfs_total_blocks_alloced, 1);
+	}
+
+	brelse(bh);
+	bh = new_bh;
+	new_bh = NULL;
+	if (err)
+		goto fail;
+	obj = (struct apfs_obj_phys *)bh->b_data;
+
+	if (type & APFS_OBJ_PHYSICAL)
+		obj->o_oid = cpu_to_le64(new_bno);
+	obj->o_xid = cpu_to_le64(nxi->nx_xid);
+	err = apfs_transaction_join(sb, bh);
+	if (err)
+		goto fail;
+
+	set_buffer_csum(bh);
+	return bh;
+
+fail:
+	brelse(bh);
+	return ERR_PTR(err);
+}
diff --git a/fs/apfs/snapshot.c b/fs/apfs/snapshot.c
new file mode 100644
index 000000000..8459a8d2a
--- /dev/null
+++ b/fs/apfs/snapshot.c
@@ -0,0 +1,687 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2022 Ernesto A. Fernández <ernesto@corellium.com>
+ */
+
+#include <linux/mount.h>
+#include <linux/slab.h>
+#include "apfs.h"
+
+/**
+ * apfs_create_superblock_snapshot - Take a snapshot of the volume superblock
+ * @sb:		superblock structure
+ * @bno:	on return, the block number for the new superblock copy
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_create_superblock_snapshot(struct super_block *sb, u64 *bno)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct buffer_head *curr_bh = NULL;
+	struct buffer_head *snap_bh = NULL;
+	struct apfs_superblock *snap_raw = NULL;
+	int err;
+
+	err = apfs_spaceman_allocate_block(sb, bno, true /* backwards */);
+	if (err) {
+		apfs_err(sb, "block allocation failed");
+		goto fail;
+	}
+
+	snap_bh = apfs_getblk(sb, *bno);
+	if (!snap_bh) {
+		apfs_err(sb, "failed to map block for volume snap (0x%llx)", *bno);
+		err = -EIO;
+		goto fail;
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+
+	curr_bh = sbi->s_vobject.o_bh;
+	memcpy(snap_bh->b_data, curr_bh->b_data, sb->s_blocksize);
+	curr_bh = NULL;
+
+	err = apfs_transaction_join(sb, snap_bh);
+	if (err)
+		goto fail;
+	set_buffer_csum(snap_bh);
+
+	snap_raw = (struct apfs_superblock *)snap_bh->b_data;
+	/* Volume superblocks in snapshots are physical objects */
+	snap_raw->apfs_o.o_oid = cpu_to_le64p(bno);
+	snap_raw->apfs_o.o_type = cpu_to_le32(APFS_OBJ_PHYSICAL | APFS_OBJECT_TYPE_FS);
+	/* The omap is shared with the current volume */
+	snap_raw->apfs_omap_oid = 0;
+	/* The extent reference tree is given by the snapshot metadata */
+	snap_raw->apfs_extentref_tree_oid = 0;
+	/* No snapshots inside snapshots */
+	snap_raw->apfs_snap_meta_tree_oid = 0;
+
+	err = 0;
+fail:
+	snap_raw = NULL;
+	brelse(snap_bh);
+	snap_bh = NULL;
+	return err;
+}
+
+static int apfs_create_snap_metadata_rec(struct inode *mntpoint, struct apfs_node *snap_root, const char *name, int name_len, u64 sblock_oid)
+{
+	struct super_block *sb = mntpoint->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_query *query = NULL;
+	struct apfs_snap_metadata_key raw_key;
+	struct apfs_snap_metadata_val *raw_val = NULL;
+	int val_len;
+	struct timespec64 now;
+	u64 xid = APFS_NXI(sb)->nx_xid;
+	int err;
+
+	query = apfs_alloc_query(snap_root, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	apfs_init_snap_metadata_key(xid, &query->key);
+	query->flags |= APFS_QUERY_SNAP_META | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err == 0) {
+		apfs_err(sb, "record exists for xid 0x%llx", xid);
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+	if (err != -ENODATA) {
+		apfs_err(sb, "query failed for xid 0x%llx", xid);
+		goto fail;
+	}
+
+	apfs_key_set_hdr(APFS_TYPE_SNAP_METADATA, xid, &raw_key);
+
+	val_len = sizeof(*raw_val) + name_len + 1;
+	raw_val = kzalloc(val_len, GFP_KERNEL);
+	if (!raw_val) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	raw_val->extentref_tree_oid = vsb_raw->apfs_extentref_tree_oid;
+	raw_val->sblock_oid = cpu_to_le64(sblock_oid);
+	now = current_time(mntpoint);
+	raw_val->create_time = cpu_to_le64(timespec64_to_ns(&now));
+	raw_val->change_time = raw_val->create_time;
+	raw_val->extentref_tree_type = vsb_raw->apfs_extentref_tree_type;
+	raw_val->flags = 0;
+	raw_val->name_len = cpu_to_le16(name_len + 1); /* Count the null byte */
+	strscpy(raw_val->name, name, name_len + 1);
+
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	raw_val->inum = vsb_raw->apfs_next_obj_id;
+	le64_add_cpu(&vsb_raw->apfs_next_obj_id, 1);
+
+	err = apfs_btree_insert(query, &raw_key, sizeof(raw_key), raw_val, val_len);
+	if (err)
+		apfs_err(sb, "insertion failed for xid 0x%llx", xid);
+fail:
+	kfree(raw_val);
+	raw_val = NULL;
+	apfs_free_query(query);
+	query = NULL;
+	return err;
+}
+
+static int apfs_create_snap_name_rec(struct apfs_node *snap_root, const char *name, int name_len)
+{
+	struct super_block *sb = snap_root->object.sb;
+	struct apfs_query *query = NULL;
+	struct apfs_snap_name_key *raw_key = NULL;
+	struct apfs_snap_name_val raw_val;
+	int key_len;
+	int err;
+
+	query = apfs_alloc_query(snap_root, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	apfs_init_snap_name_key(name, &query->key);
+	query->flags |= APFS_QUERY_SNAP_META | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err == 0) {
+		/* This should never happen here, the caller has checked */
+		apfs_alert(sb, "snapshot name record already exists");
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+	if (err != -ENODATA) {
+		apfs_err(sb, "query failed (%s)", name);
+		goto fail;
+	}
+
+	key_len = sizeof(*raw_key) + name_len + 1;
+	raw_key = kzalloc(key_len, GFP_KERNEL);
+	if (!raw_key) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	apfs_key_set_hdr(APFS_TYPE_SNAP_NAME, APFS_SNAP_NAME_OBJ_ID, raw_key);
+	raw_key->name_len = cpu_to_le16(name_len + 1); /* Count the null byte */
+	strscpy(raw_key->name, name, name_len + 1);
+
+	raw_val.snap_xid = cpu_to_le64(APFS_NXI(sb)->nx_xid);
+
+	err = apfs_btree_insert(query, raw_key, key_len, &raw_val, sizeof(raw_val));
+	if (err)
+		apfs_err(sb, "insertion failed (%s)", name);
+fail:
+	kfree(raw_key);
+	raw_key = NULL;
+	apfs_free_query(query);
+	query = NULL;
+	return err;
+}
+
+static int apfs_create_snap_meta_records(struct inode *mntpoint, const char *name, int name_len, u64 sblock_oid)
+{
+	struct super_block *sb = mntpoint->i_sb;
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_node *snap_root = NULL;
+	int err;
+
+	snap_root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_snap_meta_tree_oid), APFS_OBJ_PHYSICAL, true /* write */);
+	if (IS_ERR(snap_root)) {
+		apfs_err(sb, "failed to read snap meta root 0x%llx", le64_to_cpu(vsb_raw->apfs_snap_meta_tree_oid));
+		return PTR_ERR(snap_root);
+	}
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	vsb_raw->apfs_snap_meta_tree_oid = cpu_to_le64(snap_root->object.oid);
+
+	err = apfs_create_snap_metadata_rec(mntpoint, snap_root, name, name_len, sblock_oid);
+	if (err) {
+		apfs_err(sb, "meta rec creation failed");
+		goto fail;
+	}
+	err = apfs_create_snap_name_rec(snap_root, name, name_len);
+	if (err)
+		apfs_err(sb, "name rec creation failed");
+
+fail:
+	apfs_node_free(snap_root);
+	return err;
+}
+
+static int apfs_create_new_extentref_tree(struct super_block *sb)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	u64 oid;
+	int err;
+
+	err = apfs_make_empty_btree_root(sb, APFS_OBJECT_TYPE_BLOCKREFTREE, &oid);
+	if (err) {
+		apfs_err(sb, "failed to make empty root");
+		return err;
+	}
+
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	vsb_raw->apfs_extentref_tree_oid = cpu_to_le64(oid);
+	return 0;
+}
+
+/**
+ * apfs_update_omap_snap_tree - Add the current xid to the omap's snapshot tree
+ * @sb:		filesystem superblock
+ * @oid_p:	pointer to the on-disk block number for the root node
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_update_omap_snap_tree(struct super_block *sb, __le64 *oid_p)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_node *root = NULL;
+	u64 oid = le64_to_cpup(oid_p);
+	struct apfs_query *query = NULL;
+	__le64 raw_key;
+	struct apfs_omap_snapshot raw_val = {0};
+	int err;
+
+	/* An empty snapshots tree may not even have a root yet */
+	if (!oid) {
+		err = apfs_make_empty_btree_root(sb, APFS_OBJECT_TYPE_OMAP_SNAPSHOT, &oid);
+		if (err) {
+			apfs_err(sb, "failed to make empty root");
+			return err;
+		}
+	}
+
+	root = apfs_read_node(sb, oid, APFS_OBJ_PHYSICAL, true /* write */);
+	if (IS_ERR(root)) {
+		apfs_err(sb, "failed to read omap snap root 0x%llx", oid);
+		return PTR_ERR(root);
+	}
+	oid = 0;
+
+	query = apfs_alloc_query(root, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	apfs_init_omap_snap_key(nxi->nx_xid, &query->key);
+	query->flags = APFS_QUERY_OMAP_SNAP | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err == 0) {
+		apfs_err(sb, "record exists for xid 0x%llx", nxi->nx_xid);
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+	if (err != -ENODATA) {
+		apfs_err(sb, "query failed for xid 0x%llx", nxi->nx_xid);
+		goto fail;
+	}
+
+	raw_key = cpu_to_le64(nxi->nx_xid);
+	err = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+	if (err)
+		apfs_err(sb, "insertion failed for xid 0x%llx", nxi->nx_xid);
+	*oid_p = cpu_to_le64(root->object.block_nr);
+
+fail:
+	apfs_free_query(query);
+	query = NULL;
+	apfs_node_free(root);
+	root = NULL;
+	return err;
+}
+
+/**
+ * apfs_update_omap_snapshots - Add the current xid to the omap's snapshots
+ * @sb:	filesystem superblock
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_update_omap_snapshots(struct super_block *sb)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct buffer_head *bh = NULL;
+	struct apfs_omap_phys *omap = NULL;
+	u64 omap_blk;
+	u64 xid;
+	int err;
+
+	xid = APFS_NXI(sb)->nx_xid;
+
+	omap_blk = le64_to_cpu(vsb_raw->apfs_omap_oid);
+	bh = apfs_read_object_block(sb, omap_blk, true /* write */, false /* preserve */);
+	if (IS_ERR(bh)) {
+		apfs_err(sb, "CoW failed for bno 0x%llx", omap_blk);
+		return PTR_ERR(bh);
+	}
+	omap = (struct apfs_omap_phys *)bh->b_data;
+
+	apfs_assert_in_transaction(sb, &omap->om_o);
+	le32_add_cpu(&omap->om_snap_count, 1);
+	omap->om_most_recent_snap = cpu_to_le64(xid);
+	err = apfs_update_omap_snap_tree(sb, &omap->om_snapshot_tree_oid);
+	if (err)
+		apfs_err(sb, "omap snap tree update failed");
+
+	omap = NULL;
+	brelse(bh);
+	bh = NULL;
+	return err;
+}
+
+/**
+ * apfs_snapshot_name_check - Check if a snapshot with the given name exists
+ * @sb:		filesystem superblock
+ * @name:	name to check
+ * @name_len:	length of @name
+ * @eexist:	on return, whether the name exists or not
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_snapshot_name_check(struct super_block *sb, const char *name, int name_len, bool *eexist)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_node *snap_root = NULL;
+	struct apfs_query *query = NULL;
+	int err;
+
+	snap_root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_snap_meta_tree_oid), APFS_OBJ_PHYSICAL, false /* write */);
+	if (IS_ERR(snap_root)) {
+		apfs_err(sb, "failed to read snap meta root 0x%llx", le64_to_cpu(vsb_raw->apfs_snap_meta_tree_oid));
+		return PTR_ERR(snap_root);
+	}
+	vsb_raw = NULL;
+
+	query = apfs_alloc_query(snap_root, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto out;
+	}
+	apfs_init_snap_name_key(name, &query->key);
+	query->flags |= APFS_QUERY_SNAP_META | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err == 0) {
+		*eexist = true;
+	} else if (err == -ENODATA) {
+		*eexist = false;
+		err = 0;
+	} else {
+		apfs_err(sb, "query failed (%s)", name);
+	}
+
+out:
+	apfs_free_query(query);
+	query = NULL;
+	apfs_node_free(snap_root);
+	snap_root = NULL;
+	return err;
+}
+
+/**
+ * apfs_do_ioc_takesnapshot - Actual work for apfs_ioc_take_snapshot()
+ * @mntpoint:	inode of the mount point to snapshot
+ * @name:	label for the snapshot
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_do_ioc_take_snapshot(struct inode *mntpoint, const char *name, int name_len)
+{
+	struct super_block *sb = mntpoint->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = NULL;
+	struct apfs_omap *omap = sbi->s_omap;
+	u64 sblock_oid;
+	bool eexist = false;
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+
+	/*
+	 * This check can fail in normal operation, so run it before making any
+	 * changes and exit the transaction cleanly if necessary. The name
+	 * lookup will have to be repeated later, but it's ok because I don't
+	 * expect snapshot creation to be a bottleneck for anyone.
+	 */
+	err = apfs_snapshot_name_check(sb, name, name_len, &eexist);
+	if (err) {
+		apfs_err(sb, "snapshot name existence check failed");
+		goto fail;
+	}
+	if (eexist) {
+		err = apfs_transaction_commit(sb);
+		if (err)
+			goto fail;
+		apfs_warn(sb, "snapshot name already in use (%s)", name);
+		return -EEXIST;
+	}
+
+	/*
+	 * Flush the extent caches to the extenref tree before it gets moved to
+	 * the snapshot. It seems safer in general to avoid big unpredictable
+	 * changes to the layout after the snapshot is set up.
+	 */
+	err = apfs_transaction_flush_all_inodes(sb);
+	if (err) {
+		apfs_err(sb, "failed to flush all inodes");
+		goto fail;
+	}
+
+	err = apfs_create_superblock_snapshot(sb, &sblock_oid);
+	if (err) {
+		apfs_err(sb, "failed to snapshot superblock");
+		goto fail;
+	}
+
+	err = apfs_create_snap_meta_records(mntpoint, name, name_len, sblock_oid);
+	if (err) {
+		apfs_err(sb, "failed to create snap meta records");
+		goto fail;
+	}
+
+	err = apfs_create_new_extentref_tree(sb);
+	if (err) {
+		apfs_err(sb, "failed to create new extref tree");
+		goto fail;
+	}
+
+	err = apfs_update_omap_snapshots(sb);
+	if (err) {
+		apfs_err(sb, "failed to update omap snapshots");
+		goto fail;
+	}
+
+	/*
+	 * The official reference allows old implementations to ignore extended
+	 * snapshot metadata, so I don't see any reason why we can't do the
+	 * same for now.
+	 */
+
+	vsb_raw = sbi->s_vsb_raw;
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	le64_add_cpu(&vsb_raw->apfs_num_snapshots, 1);
+
+	omap->omap_latest_snap = APFS_NXI(sb)->nx_xid;
+
+	sbi->s_nxi->nx_transaction.t_state |= APFS_NX_TRANS_FORCE_COMMIT;
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_ioc_take_snapshot - Ioctl handler for APFS_IOC_CREATE_SNAPSHOT
+ * @file:	affected file
+ * @arg:	ioctl argument
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_ioc_take_snapshot(struct file *file, void __user *user_arg)
+{
+	struct inode *inode = file_inode(file);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_ioctl_snap_name *arg = NULL;
+	size_t name_len;
+	int err;
+
+	if (apfs_ino(inode) != APFS_ROOT_DIR_INO_NUM) {
+		apfs_info(sb, "snapshot must be requested on mountpoint");
+		return -ENOTTY;
+	}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+	if (!inode_owner_or_capable(inode))
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+	if (!inode_owner_or_capable(&init_user_ns, inode))
+#else
+	if (!inode_owner_or_capable(&nop_mnt_idmap, inode))
+#endif
+		return -EPERM;
+
+	err = mnt_want_write_file(file);
+	if (err)
+		return err;
+
+	arg = kzalloc(sizeof(*arg), GFP_KERNEL);
+	if (!arg) {
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	if (copy_from_user(arg, user_arg, sizeof(*arg))) {
+		err = -EFAULT;
+		goto fail;
+	}
+
+	name_len = strnlen(arg->name, sizeof(arg->name));
+	if (name_len == sizeof(arg->name)) {
+		apfs_warn(sb, "snapshot name is too long (%d)", (int)name_len);
+		err = -EINVAL;
+		goto fail;
+	}
+
+	err = apfs_do_ioc_take_snapshot(inode, arg->name, name_len);
+fail:
+	kfree(arg);
+	arg = NULL;
+	mnt_drop_write_file(file);
+	return err;
+}
+
+static int apfs_snap_xid_from_query(struct apfs_query *query, u64 *xid)
+{
+	char *raw = query->node->object.data;
+	__le64 *val = NULL;
+
+	if (query->len != sizeof(*val)) {
+		apfs_err(query->node->object.sb, "bad value length (%d)", query->len);
+		return -EFSCORRUPTED;
+	}
+	val = (__le64 *)(raw + query->off);
+
+	*xid = le64_to_cpup(val);
+	return 0;
+}
+
+static int apfs_snapshot_name_to_xid(struct apfs_node *snap_root, const char *name, u64 *xid)
+{
+	struct super_block *sb = snap_root->object.sb;
+	struct apfs_query *query = NULL;
+	int err;
+
+	query = apfs_alloc_query(snap_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_snap_name_key(name, &query->key);
+	query->flags |= APFS_QUERY_SNAP_META | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err) {
+		if (err != -ENODATA)
+			apfs_err(sb, "query failed (%s)", name);
+		goto fail;
+	}
+
+	err = apfs_snap_xid_from_query(query, xid);
+	if (err)
+		apfs_err(sb, "bad snap name record (%s)", name);
+fail:
+	apfs_free_query(query);
+	query = NULL;
+	return err;
+}
+
+static int apfs_snap_sblock_from_query(struct apfs_query *query, u64 *sblock_oid)
+{
+	struct super_block *sb = query->node->object.sb;
+	char *raw = query->node->object.data;
+	struct apfs_snap_metadata_val *val = NULL;
+
+	if (query->len < sizeof(*val)) {
+		apfs_err(sb, "bad value length (%d)", query->len);
+		return -EFSCORRUPTED;
+	}
+	val = (struct apfs_snap_metadata_val *)(raw + query->off);
+
+	/* I don't know anything about these snapshots, can they be mounted? */
+	if (le32_to_cpu(val->flags) & APFS_SNAP_META_PENDING_DATALESS) {
+		apfs_warn(sb, "the requested snapshot is dataless");
+		return -EOPNOTSUPP;
+	}
+
+	*sblock_oid = le64_to_cpu(val->sblock_oid);
+	return 0;
+}
+
+static int apfs_snapshot_xid_to_sblock(struct apfs_node *snap_root, u64 xid, u64 *sblock_oid)
+{
+	struct super_block *sb = snap_root->object.sb;
+	struct apfs_query *query = NULL;
+	int err;
+
+	query = apfs_alloc_query(snap_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_snap_metadata_key(xid, &query->key);
+	query->flags |= APFS_QUERY_SNAP_META | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err) {
+		apfs_err(sb, "query failed for xid 0x%llx", xid);
+		goto fail;
+	}
+
+	err = apfs_snap_sblock_from_query(query, sblock_oid);
+	if (err)
+		apfs_err(sb, "bad snap meta record for xid 0x%llx", xid);
+fail:
+	apfs_free_query(query);
+	query = NULL;
+	return err;
+}
+
+/**
+ * apfs_switch_to_snapshot - Start working with the snapshot volume superblock
+ * @sb: superblock structure
+ *
+ * Maps the volume superblock from the snapshot specified in the mount options.
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_switch_to_snapshot(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_node *snap_root = NULL;
+	const char *name = NULL;
+	u64 sblock_oid = 0;
+	u64 xid = 0;
+	int err;
+
+	ASSERT(sb->s_flags & SB_RDONLY);
+
+	name = sbi->s_snap_name;
+	if (strlen(name) > APFS_SNAP_MAX_NAMELEN) {
+		apfs_warn(sb, "snapshot name is too long");
+		return -EINVAL;
+	}
+
+	snap_root = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_snap_meta_tree_oid), APFS_OBJ_PHYSICAL, false /* write */);
+	if (IS_ERR(snap_root)) {
+		apfs_err(sb, "failed to read snap meta root 0x%llx", le64_to_cpu(vsb_raw->apfs_snap_meta_tree_oid));
+		return PTR_ERR(snap_root);
+	}
+	vsb_raw = NULL;
+
+	err = apfs_snapshot_name_to_xid(snap_root, name, &xid);
+	if (err) {
+		if (err == -ENODATA)
+			apfs_info(sb, "no snapshot under that name (%s)", name);
+		goto fail;
+	}
+	sbi->s_snap_xid = xid;
+
+	err = apfs_snapshot_xid_to_sblock(snap_root, xid, &sblock_oid);
+	if (err)
+		goto fail;
+
+	apfs_unmap_volume_super(sb);
+	err = apfs_map_volume_super_bno(sb, sblock_oid, nxi->nx_flags & APFS_CHECK_NODES);
+	if (err)
+		apfs_err(sb, "failed to map volume block 0x%llx", sblock_oid);
+
+fail:
+	apfs_node_free(snap_root);
+	return err;
+}
diff --git a/fs/apfs/spaceman.c b/fs/apfs/spaceman.c
new file mode 100644
index 000000000..31bf439e9
--- /dev/null
+++ b/fs/apfs/spaceman.c
@@ -0,0 +1,1479 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include "apfs.h"
+
+/**
+ * apfs_spaceman_read_cib_addr - Get the address of a cib from the spaceman
+ * @sb:		superblock structure
+ * @index:	index of the chunk-info block
+ *
+ * Returns the block number for the chunk-info block.
+ *
+ * This is not described in the official documentation; credit for figuring it
+ * out should go to Joachim Metz: <https://github.com/libyal/libfsapfs>.
+ */
+static u64 apfs_spaceman_read_cib_addr(struct super_block *sb, int index)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	u32 offset;
+	__le64 *addr_p;
+
+	offset = sm->sm_addr_offset + index * sizeof(*addr_p);
+	addr_p = (void *)sm_raw + offset;
+	return le64_to_cpup(addr_p);
+}
+
+/**
+ * apfs_spaceman_write_cib_addr - Store the address of a cib in the spaceman
+ * @sb:		superblock structure
+ * @index:	index of the chunk-info block
+ * @addr:	address of the chunk-info block
+ */
+static void apfs_spaceman_write_cib_addr(struct super_block *sb,
+					 int index, u64 addr)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	u32 offset;
+	__le64 *addr_p;
+
+	apfs_assert_in_transaction(sb, &sm_raw->sm_o);
+
+	offset = sm->sm_addr_offset + index * sizeof(*addr_p);
+	addr_p = (void *)sm_raw + offset;
+	*addr_p = cpu_to_le64(addr);
+}
+
+/**
+ * apfs_max_chunks_per_cib - Find the maximum chunk count for a chunk-info block
+ * @sb: superblock structure
+ */
+static inline int apfs_max_chunks_per_cib(struct super_block *sb)
+{
+	return (sb->s_blocksize - sizeof(struct apfs_chunk_info_block)) /
+						sizeof(struct apfs_chunk_info);
+}
+
+/**
+ * apfs_read_spaceman_dev - Read a space manager device structure
+ * @sb:		superblock structure
+ * @dev:	on-disk device structure
+ *
+ * Initializes the in-memory spaceman fields related to the main device; fusion
+ * drives are not yet supported.  Returns 0 on success, or a negative error code
+ * in case of failure.
+ */
+static int apfs_read_spaceman_dev(struct super_block *sb,
+				  struct apfs_spaceman_device *dev)
+{
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+
+	if (dev->sm_cab_count) {
+		apfs_err(sb, "large devices are not supported");
+		return -EINVAL;
+	}
+
+	spaceman->sm_block_count = le64_to_cpu(dev->sm_block_count);
+	spaceman->sm_chunk_count = le64_to_cpu(dev->sm_chunk_count);
+	spaceman->sm_cib_count = le32_to_cpu(dev->sm_cib_count);
+	spaceman->sm_free_count = le64_to_cpu(dev->sm_free_count);
+	spaceman->sm_addr_offset = le32_to_cpu(dev->sm_addr_offset);
+
+	/* Check that all the cib addresses fit in the spaceman object */
+	if ((long long)spaceman->sm_addr_offset +
+	    (long long)spaceman->sm_cib_count * sizeof(u64) > spaceman->sm_size) {
+		apfs_err(sb, "too many cibs (%u)", spaceman->sm_cib_count);
+		return -EFSCORRUPTED;
+	}
+
+	return 0;
+}
+
+/**
+ * apfs_spaceman_get_16 - Get a 16-bit value from an offset in the spaceman
+ * @sb:		superblock structure
+ * @off:	offset for the value
+ *
+ * Returns a pointer to the value, or NULL if it doesn't fit.
+ */
+static __le16 *apfs_spaceman_get_16(struct super_block *sb, size_t off)
+{
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = spaceman->sm_raw;
+
+	if (off > spaceman->sm_size)
+		return NULL;
+	if (off + sizeof(__le16) > spaceman->sm_size)
+		return NULL;
+	return (void *)sm_raw + off;
+}
+
+/**
+ * apfs_spaceman_get_64 - Get a 64-bit value from an offset in the spaceman
+ * @sb:		superblock structure
+ * @off:	offset for the value
+ *
+ * Returns a pointer to the value, or NULL if it doesn't fit.
+ */
+static __le64 *apfs_spaceman_get_64(struct super_block *sb, size_t off)
+{
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = spaceman->sm_raw;
+
+	if (off > spaceman->sm_size)
+		return NULL;
+	if (off + sizeof(__le64) > spaceman->sm_size)
+		return NULL;
+	return (void *)sm_raw + off;
+}
+
+/**
+ * apfs_allocate_ip_bitmap - Allocate a free ip bitmap block
+ * @sb:		filesystem superblock
+ * @offset_p:	on return, the offset from sm_ip_bm_base of the allocated block
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_allocate_ip_bitmap(struct super_block *sb, u16 *offset_p)
+{
+	struct apfs_spaceman *spaceman = NULL;
+	struct apfs_spaceman_phys *sm_raw = NULL;
+	u32 free_next_offset, old_head_off;
+	u16 free_head, blkcnt;
+	__le16 *old_head_p = NULL;
+
+	spaceman = APFS_SM(sb);
+	sm_raw = spaceman->sm_raw;
+	free_next_offset = le32_to_cpu(sm_raw->sm_ip_bm_free_next_offset);
+	free_head = le16_to_cpu(sm_raw->sm_ip_bm_free_head);
+	blkcnt = (u16)le32_to_cpu(sm_raw->sm_ip_bm_block_count);
+
+	/*
+	 * The "free_next" array is a linked list of free blocks that starts
+	 * with the "free_head". Allocate this head then, and make the next
+	 * block into the new head.
+	 */
+	old_head_off = free_next_offset + free_head * sizeof(*old_head_p);
+	old_head_p = apfs_spaceman_get_16(sb, old_head_off);
+	if (!old_head_p) {
+		apfs_err(sb, "free_next head offset out of bounds (%u)", old_head_off);
+		return -EFSCORRUPTED;
+	}
+	*offset_p = free_head;
+	free_head = le16_to_cpup(old_head_p);
+	sm_raw->sm_ip_bm_free_head = *old_head_p;
+	/* No longer free, no longer part of the linked list */
+	*old_head_p = cpu_to_le16(APFS_SPACEMAN_IP_BM_INDEX_INVALID);
+
+	/* Just a little sanity check because I've messed this up before */
+	if (free_head >= blkcnt || *offset_p >= blkcnt) {
+		apfs_err(sb, "free next list seems empty or corrupt");
+		return -EFSCORRUPTED;
+	}
+
+	return 0;
+}
+
+/**
+ * apfs_free_ip_bitmap - Free a used ip bitmap block
+ * @sb:		filesystem superblock
+ * @offset:	the offset from sm_ip_bm_base of the block to free
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_free_ip_bitmap(struct super_block *sb, u16 offset)
+{
+	struct apfs_spaceman *spaceman = NULL;
+	struct apfs_spaceman_phys *sm_raw = NULL;
+	u32 free_next_offset, old_tail_off;
+	u16 free_tail;
+	__le16 *old_tail_p = NULL;
+
+	spaceman = APFS_SM(sb);
+	sm_raw = spaceman->sm_raw;
+	free_next_offset = le32_to_cpu(sm_raw->sm_ip_bm_free_next_offset);
+	free_tail = le16_to_cpu(sm_raw->sm_ip_bm_free_tail);
+
+	/*
+	 * The "free_next" array is a linked list of free blocks that ends
+	 * with the "free_tail". The block getting freed will become the new
+	 * tail of the list.
+	 */
+	old_tail_off = free_next_offset + free_tail * sizeof(*old_tail_p);
+	old_tail_p = apfs_spaceman_get_16(sb, old_tail_off);
+	if (!old_tail_p) {
+		apfs_err(sb, "free_next tail offset out of bounds (%u)", old_tail_off);
+		return -EFSCORRUPTED;
+	}
+	*old_tail_p = cpu_to_le16(offset);
+	sm_raw->sm_ip_bm_free_tail = cpu_to_le16(offset);
+	free_tail = offset;
+
+	return 0;
+}
+
+/**
+ * apfs_reallocate_ip_bitmap - Find a new block for an ip bitmap
+ * @sb:		filesystem superblock
+ * @offset_p:	the offset from sm_ip_bm_base of the block to free
+ *
+ * On success returns 0 and updates @offset_p to the new offset allocated for
+ * the ip bitmap. Since blocks are allocated at the head of the list and freed
+ * at the tail, there is no risk of reuse by future reallocations within the
+ * same transaction (under there is some serious corruption, of course).
+ *
+ * Returns a negative error code in case of failure.
+ */
+static int apfs_reallocate_ip_bitmap(struct super_block *sb, __le16 *offset_p)
+{
+	int err;
+	u16 offset;
+
+	offset = le16_to_cpup(offset_p);
+
+	err = apfs_free_ip_bitmap(sb, offset);
+	if (err) {
+		apfs_err(sb, "failed to free ip bitmap %u", offset);
+		return err;
+	}
+	err = apfs_allocate_ip_bitmap(sb, &offset);
+	if (err) {
+		apfs_err(sb, "failed to allocate a new ip bitmap block");
+		return err;
+	}
+
+	*offset_p = cpu_to_le16(offset);
+	return 0;
+}
+
+/**
+ * apfs_write_single_ip_bitmap - Write a single ip bitmap to disk
+ * @sb:		filesystem superblock
+ * @bitmap:	bitmap to write
+ * @idx:	index of the ip bitmap to write
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_write_single_ip_bitmap(struct super_block *sb, char *bitmap, u32 idx)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = spaceman->sm_raw;
+	struct buffer_head *bh = NULL;
+	u64 ip_bm_base, ip_bitmap_bno;
+	u32 xid_off, ip_bitmap_off;
+	__le64 *xid_p = NULL;
+	__le16 *ip_bitmap_p = NULL;
+	int err;
+
+	ip_bm_base = le64_to_cpu(sm_raw->sm_ip_bm_base);
+
+	/* First update the xid, which is kept in a separate array */
+	xid_off = le32_to_cpu(sm_raw->sm_ip_bm_xid_offset) + idx * sizeof(*xid_p);
+	xid_p = apfs_spaceman_get_64(sb, xid_off);
+	if (!xid_p) {
+		apfs_err(sb, "xid out of bounds (%u)", xid_off);
+		return -EFSCORRUPTED;
+	}
+	*xid_p = cpu_to_le64(nxi->nx_xid);
+
+	/* Now get find new location for the ip bitmap (and free the old one) */
+	ip_bitmap_off = le32_to_cpu(sm_raw->sm_ip_bitmap_offset) + idx * sizeof(*ip_bitmap_p);
+	ip_bitmap_p = apfs_spaceman_get_16(sb, ip_bitmap_off);
+	if (!ip_bitmap_p) {
+		apfs_err(sb, "bmap offset out of bounds (%u)", ip_bitmap_off);
+		return -EFSCORRUPTED;
+	}
+	err = apfs_reallocate_ip_bitmap(sb, ip_bitmap_p);
+	if (err) {
+		apfs_err(sb, "failed to reallocate ip bitmap %u", le16_to_cpup(ip_bitmap_p));
+		return err;
+	}
+
+	/* Finally, write the dirty bitmap to the new location */
+	ip_bitmap_bno = ip_bm_base + le16_to_cpup(ip_bitmap_p);
+	bh = apfs_getblk(sb, ip_bitmap_bno);
+	if (!bh) {
+		apfs_err(sb, "failed to map block for CoW (0x%llx)", ip_bitmap_bno);
+		return -EIO;
+	}
+	memcpy(bh->b_data, bitmap, sb->s_blocksize);
+	err = apfs_transaction_join(sb, bh);
+	if (err)
+		goto fail;
+	bh = NULL;
+
+	spaceman->sm_ip_bmaps[idx].dirty = false;
+	return 0;
+
+fail:
+	brelse(bh);
+	bh = NULL;
+	return err;
+}
+
+/**
+ * apfs_write_ip_bitmaps - Write all dirty ip bitmaps to disk
+ * @sb: superblock structure
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_write_ip_bitmaps(struct super_block *sb)
+{
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = spaceman->sm_raw;
+	struct apfs_ip_bitmap_block_info *info = NULL;
+	u32 bmaps_count = spaceman->sm_ip_bmaps_count;
+	int err;
+	u32 i;
+
+	apfs_assert_in_transaction(sb, &sm_raw->sm_o);
+
+	for (i = 0; i < bmaps_count; ++i) {
+		info = &spaceman->sm_ip_bmaps[i];
+		if (!info->dirty)
+			continue;
+		err = apfs_write_single_ip_bitmap(sb, info->block, i);
+		if (err) {
+			apfs_err(sb, "failed to rotate ip bitmap %u", i);
+			return err;
+		}
+	}
+	return 0;
+}
+
+/**
+* apfs_read_single_ip_bitmap - Read a single ip bitmap to memory
+* @sb:	filesystem superblock
+* @idx:	index of the ip bitmap to read
+*
+* Returns 0 on success or a negative error code in case of failure.
+*/
+static int apfs_read_single_ip_bitmap(struct super_block *sb, u32 idx)
+{
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = spaceman->sm_raw;
+	struct buffer_head *bh = NULL;
+	char *bitmap = NULL;
+	u64 ip_bm_base, ip_bitmap_bno;
+	u32 ip_bitmap_off;
+	__le16 *ip_bitmap_p = NULL;
+	int err;
+
+	ip_bm_base = le64_to_cpu(sm_raw->sm_ip_bm_base);
+
+	ip_bitmap_off = le32_to_cpu(sm_raw->sm_ip_bitmap_offset) + idx * sizeof(*ip_bitmap_p);
+	ip_bitmap_p = apfs_spaceman_get_16(sb, ip_bitmap_off);
+	if (!ip_bitmap_p) {
+		apfs_err(sb, "bmap offset out of bounds (%u)", ip_bitmap_off);
+		return -EFSCORRUPTED;
+	}
+
+	bitmap = kmalloc(sb->s_blocksize, GFP_KERNEL);
+	if (!bitmap)
+		return -ENOMEM;
+
+	ip_bitmap_bno = ip_bm_base + le16_to_cpup(ip_bitmap_p);
+	bh = apfs_sb_bread(sb, ip_bitmap_bno);
+	if (!bh) {
+		apfs_err(sb, "failed to read ip bitmap (0x%llx)", ip_bitmap_bno);
+		err = -EIO;
+		goto fail;
+	}
+	memcpy(bitmap, bh->b_data, sb->s_blocksize);
+	brelse(bh);
+	bh = NULL;
+
+	spaceman->sm_ip_bmaps[idx].dirty = false;
+	spaceman->sm_ip_bmaps[idx].block = bitmap;
+	bitmap = NULL;
+	return 0;
+
+fail:
+	kfree(bitmap);
+	bitmap = NULL;
+	return err;
+}
+
+/**
+ * apfs_read_ip_bitmaps - Read all the ip bitmaps to memory
+ * @sb: superblock structure
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_read_ip_bitmaps(struct super_block *sb)
+{
+	struct apfs_spaceman *spaceman = APFS_SM(sb);
+	u32 bmaps_count = spaceman->sm_ip_bmaps_count;
+	int err;
+	u32 i;
+
+	for (i = 0; i < bmaps_count; ++i) {
+		err = apfs_read_single_ip_bitmap(sb, i);
+		if (err) {
+			apfs_err(sb, "failed to read ip bitmap %u", i);
+			return err;
+		}
+	}
+	return 0;
+}
+
+/*
+ * Free queue record data
+ */
+struct apfs_fq_rec {
+	u64 xid;
+	u64 bno;
+	u64 len;
+};
+
+/**
+ * apfs_fq_rec_from_query - Read the free queue record found by a query
+ * @query:	the query that found the record
+ * @fqrec:	on return, the free queue record
+ *
+ * Reads the free queue record into @fqrec and performs some basic sanity
+ * checks as a protection against crafted filesystems. Returns 0 on success
+ * or -EFSCORRUPTED otherwise.
+ */
+static int apfs_fq_rec_from_query(struct apfs_query *query, struct apfs_fq_rec *fqrec)
+{
+	char *raw = query->node->object.data;
+	struct apfs_spaceman_free_queue_key *key;
+
+	if (query->key_len != sizeof(*key)) {
+		apfs_err(query->node->object.sb, "bad key length (%d)", query->key_len);
+		return -EFSCORRUPTED;
+	}
+	key = (struct apfs_spaceman_free_queue_key *)(raw + query->key_off);
+
+	fqrec->xid = le64_to_cpu(key->sfqk_xid);
+	fqrec->bno = le64_to_cpu(key->sfqk_paddr);
+
+	if (query->len == 0) {
+		fqrec->len = 1; /* Ghost record */
+		return 0;
+	} else if (query->len == sizeof(__le64)) {
+		fqrec->len = le64_to_cpup((__le64 *)(raw + query->off));
+		return 0;
+	}
+	apfs_err(query->node->object.sb, "bad value length (%d)", query->len);
+	return -EFSCORRUPTED;
+}
+
+/**
+ * apfs_block_in_ip - Does this block belong to the internal pool?
+ * @sm:		in-memory spaceman structure
+ * @bno:	block number to check
+ */
+static inline bool apfs_block_in_ip(struct apfs_spaceman *sm, u64 bno)
+{
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	u64 start = le64_to_cpu(sm_raw->sm_ip_base);
+	u64 end = start + le64_to_cpu(sm_raw->sm_ip_block_count);
+
+	return bno >= start && bno < end;
+}
+
+/**
+ * apfs_ip_mark_free - Mark a block in the internal pool as free
+ * @sb:		superblock structure
+ * @bno:	block number (must belong to the ip)
+ */
+static int apfs_ip_mark_free(struct super_block *sb, u64 bno)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_ip_bitmap_block_info *info = NULL;
+
+	bno -= le64_to_cpu(sm_raw->sm_ip_base);
+	info = &sm->sm_ip_bmaps[bno >> sm->sm_ip_bmaps_shift];
+	__clear_bit_le(bno & sm->sm_ip_bmaps_mask, info->block);
+	info->dirty = true;
+
+	return 0;
+}
+
+/*
+ * apfs_main_free - Mark a regular block as free
+ */
+static int apfs_main_free(struct super_block *sb, u64 bno);
+
+/**
+ * apfs_flush_fq_rec - Delete a single fq record and mark its blocks as free
+ * @root:	free queue root node
+ * @xid:	transaction to target
+ * @len:	on return, the number of freed blocks
+ *
+ * Returns 0 on success, or a negative error code in case of failure. -ENODATA
+ * in particular means that there are no matching records left.
+ */
+static int apfs_flush_fq_rec(struct apfs_node *root, u64 xid, u64 *len)
+{
+	struct super_block *sb = root->object.sb;
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_query *query = NULL;
+	struct apfs_fq_rec fqrec = {0};
+	u64 bno;
+	int err;
+
+	query = apfs_alloc_query(root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_free_queue_key(xid, 0 /* paddr */, &query->key);
+	query->flags |= APFS_QUERY_FREE_QUEUE | APFS_QUERY_ANY_NUMBER | APFS_QUERY_EXACT;
+
+	err = apfs_btree_query(sb, &query);
+	if (err) {
+		if (err != -ENODATA)
+			apfs_err(sb, "query failed for xid 0x%llx, paddr 0x%llx", xid, 0ULL);
+		goto fail;
+	}
+	err = apfs_fq_rec_from_query(query, &fqrec);
+	if (err) {
+		apfs_err(sb, "bad free queue rec for xid 0x%llx", xid);
+		goto fail;
+	}
+
+	for (bno = fqrec.bno; bno < fqrec.bno + fqrec.len; ++bno) {
+		if (apfs_block_in_ip(sm, bno))
+			err = apfs_ip_mark_free(sb, bno);
+		else
+			err = apfs_main_free(sb, bno);
+		if (err) {
+			apfs_err(sb, "freeing block 0x%llx failed (%d)", (unsigned long long)bno, err);
+			goto fail;
+		}
+	}
+	err = apfs_btree_remove(query);
+	if (err) {
+		apfs_err(sb, "removal failed for xid 0x%llx", xid);
+		goto fail;
+	}
+	*len = fqrec.len;
+
+fail:
+	apfs_free_query(query);
+	return err;
+}
+
+/**
+ * apfs_free_queue_oldest_xid - Find the oldest xid among the free queue records
+ * @root: free queue root node
+ */
+static u64 apfs_free_queue_oldest_xid(struct apfs_node *root)
+{
+	struct apfs_spaceman_free_queue_key *key;
+	char *raw = root->object.data;
+	int len, off;
+
+	if (root->records == 0)
+		return 0;
+	len = apfs_node_locate_key(root, 0, &off);
+	if (len != sizeof(*key)) {
+		/* TODO: abort transaction */
+		apfs_err(root->object.sb, "bad key length (%d)", len);
+		return 0;
+	}
+	key = (struct apfs_spaceman_free_queue_key *)(raw + off);
+	return le64_to_cpu(key->sfqk_xid);
+}
+
+/**
+ * apfs_flush_free_queue - Free ip blocks queued by old transactions
+ * @sb:		superblock structure
+ * @qid:	queue to be freed
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_flush_free_queue(struct super_block *sb, unsigned int qid)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_spaceman_free_queue *fq = &sm_raw->sm_fq[qid];
+	struct apfs_node *fq_root;
+	struct apfs_btree_info *fq_info = NULL;
+	u64 oldest = le64_to_cpu(fq->sfq_oldest_xid);
+	int err;
+
+	fq_root = apfs_read_node(sb, le64_to_cpu(fq->sfq_tree_oid),
+				 APFS_OBJ_EPHEMERAL, true /* write */);
+	if (IS_ERR(fq_root)) {
+		apfs_err(sb, "failed to read fq root 0x%llx", le64_to_cpu(fq->sfq_tree_oid));
+		return PTR_ERR(fq_root);
+	}
+
+	while (oldest) {
+		/*
+		 * Blocks freed in the current transaction can't be reused
+		 * safely until after the commit, but I don't think there is
+		 * any point in preserving old transacions. I'm guessing the
+		 * official driver keeps multiple transactions going at the
+		 * same time, that must be why they need a free queue.
+		 */
+		if (oldest == nxi->nx_xid)
+			break;
+
+		while (true) {
+			u64 count = 0;
+
+			/* Probably not very efficient... */
+			err = apfs_flush_fq_rec(fq_root, oldest, &count);
+			if (err == -ENODATA) {
+				err = 0;
+				break;
+			} else if (err) {
+				apfs_err(sb, "failed to flush fq");
+				goto fail;
+			} else {
+				le64_add_cpu(&fq->sfq_count, -count);
+			}
+		}
+		oldest = apfs_free_queue_oldest_xid(fq_root);
+		fq->sfq_oldest_xid = cpu_to_le64(oldest);
+	}
+
+	if (qid == APFS_SFQ_MAIN) {
+		fq_info = (void *)fq_root->object.data + sb->s_blocksize - sizeof(*fq_info);
+		sm->sm_main_fq_nodes = le64_to_cpu(fq_info->bt_node_count);
+		if (sm->sm_main_fq_nodes != 1) {
+			apfs_alert(sb, "main queue wasn't flushed in full - bug!");
+			err = -EFSCORRUPTED;
+			goto fail;
+		}
+	}
+
+fail:
+	apfs_node_free(fq_root);
+	return err;
+}
+
+/**
+ * apfs_allocate_spaceman - Allocate an in-memory spaceman struct, if needed
+ * @sb:		superblock structure
+ * @raw:	on-disk spaceman struct
+ * @size:	size of the on-disk spaceman
+ *
+ * Returns the spaceman and sets it in the superblock info. Also performs all
+ * initializations for the internal pool, including reading all the ip bitmaps.
+ * This is a bit out of place here, but it's convenient because it has to
+ * happen only once.
+ *
+ * On failure, returns an error pointer.
+ */
+static struct apfs_spaceman *apfs_allocate_spaceman(struct super_block *sb, struct apfs_spaceman_phys *raw, u32 size)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_spaceman *spaceman = NULL;
+	int blk_bitcnt = sb->s_blocksize * 8;
+	size_t sm_size;
+	u32 bmap_cnt;
+	int err;
+
+	if (nxi->nx_spaceman)
+		return nxi->nx_spaceman;
+
+	/* We don't expect filesystems this big, it would be like 260 TiB */
+	bmap_cnt = le32_to_cpu(raw->sm_ip_bm_size_in_blocks);
+	if (bmap_cnt > 200) {
+		apfs_err(sb, "too many ip bitmap blocks (%u)", bmap_cnt);
+		return ERR_PTR(-EFSCORRUPTED);
+	}
+	sm_size = sizeof(*spaceman) + bmap_cnt * sizeof(spaceman->sm_ip_bmaps[0]);
+
+	spaceman = nxi->nx_spaceman = kzalloc(sm_size, GFP_KERNEL);
+	if (!spaceman)
+		return ERR_PTR(-ENOMEM);
+	spaceman->sm_nxi = nxi;
+	/*
+	 * These two fields must be set before reading the ip bitmaps, since
+	 * that stuff involves several variable-length arrays inside the
+	 * spaceman object itself.
+	 */
+	spaceman->sm_raw = raw;
+	spaceman->sm_size = size;
+
+	spaceman->sm_ip_bmaps_count = bmap_cnt;
+	spaceman->sm_ip_bmaps_mask = blk_bitcnt - 1;
+	spaceman->sm_ip_bmaps_shift = order_base_2(blk_bitcnt);
+
+	/* This must happen only once, so it's easier to just leave it here */
+	err = apfs_read_ip_bitmaps(sb);
+	if (err) {
+		apfs_err(sb, "failed to read the ip bitmaps");
+		kfree(spaceman);
+		nxi->nx_spaceman = spaceman = NULL;
+		return ERR_PTR(err);
+	}
+
+	return nxi->nx_spaceman;
+}
+
+/**
+ * apfs_read_spaceman - Find and read the space manager
+ * @sb: superblock structure
+ *
+ * Reads the space manager structure from disk and initializes its in-memory
+ * counterpart; returns 0 on success, or a negative error code in case of
+ * failure.
+ */
+int apfs_read_spaceman(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *raw_sb = nxi->nx_raw;
+	struct apfs_spaceman *spaceman = NULL;
+	struct apfs_ephemeral_object_info *sm_eph_info = NULL;
+	struct apfs_spaceman_phys *sm_raw;
+	u32 sm_flags;
+	u64 oid = le64_to_cpu(raw_sb->nx_spaceman_oid);
+	int err;
+
+	if (sb->s_flags & SB_RDONLY) /* The space manager won't be needed */
+		return 0;
+
+	sm_eph_info = apfs_ephemeral_object_lookup(sb, oid);
+	if (IS_ERR(sm_eph_info)) {
+		apfs_err(sb, "no spaceman object for oid 0x%llx", oid);
+		return PTR_ERR(sm_eph_info);
+	}
+	sm_raw = (struct apfs_spaceman_phys *)sm_eph_info->object;
+	sm_raw->sm_o.o_xid = cpu_to_le64(nxi->nx_xid);
+
+	spaceman = apfs_allocate_spaceman(sb, sm_raw, sm_eph_info->size);
+	if (IS_ERR(spaceman)) {
+		apfs_err(sb, "failed to allocate spaceman");
+		err = PTR_ERR(spaceman);
+		goto fail;
+	}
+
+	spaceman->sm_free_cache_base = spaceman->sm_free_cache_blkcnt = 0;
+
+	sm_flags = le32_to_cpu(sm_raw->sm_flags);
+	/* Undocumented feature, but it's too common to refuse to mount */
+	if (sm_flags & APFS_SM_FLAG_VERSIONED)
+		pr_warn_once("APFS: space manager is versioned\n");
+
+	/* Only read the main device; fusion drives are not yet supported */
+	err = apfs_read_spaceman_dev(sb, &sm_raw->sm_dev[APFS_SD_MAIN]);
+	if (err) {
+		apfs_err(sb, "failed to read main device");
+		goto fail;
+	}
+
+	spaceman->sm_blocks_per_chunk =
+				le32_to_cpu(sm_raw->sm_blocks_per_chunk);
+	spaceman->sm_chunks_per_cib = le32_to_cpu(sm_raw->sm_chunks_per_cib);
+	if (spaceman->sm_chunks_per_cib > apfs_max_chunks_per_cib(sb)) {
+		apfs_err(sb, "too many chunks per cib (%u)", spaceman->sm_chunks_per_cib);
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	/*
+	 * We flush free queues whole when each transaction begins, to make it
+	 * harder for the btrees to become too unbalanced.
+	 */
+	err = apfs_flush_free_queue(sb, APFS_SFQ_IP);
+	if (err) {
+		apfs_err(sb, "failed to flush ip fq");
+		goto fail;
+	}
+	err = apfs_flush_free_queue(sb, APFS_SFQ_MAIN);
+	if (err) {
+		apfs_err(sb, "failed to flush main fq");
+		goto fail;
+	}
+	return 0;
+
+fail:
+	spaceman->sm_raw = NULL;
+	return err;
+}
+
+/**
+ * apfs_write_spaceman - Write the in-memory spaceman fields to the disk buffer
+ * @sm: in-memory spaceman structure
+ *
+ * Copies the updated in-memory fields of the space manager into the on-disk
+ * structure; the buffer is not dirtied.
+ */
+static void apfs_write_spaceman(struct apfs_spaceman *sm)
+{
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_spaceman_device *dev_raw = &sm_raw->sm_dev[APFS_SD_MAIN];
+	struct apfs_nxsb_info *nxi;
+
+	nxi = sm->sm_nxi;
+	ASSERT(le64_to_cpu(sm_raw->sm_o.o_xid) == nxi->nx_xid);
+
+	dev_raw->sm_free_count = cpu_to_le64(sm->sm_free_count);
+}
+
+/**
+ * apfs_ip_find_free - Find a free block inside the internal pool
+ * @sb:		superblock structure
+ *
+ * Returns the block number for a free block, or 0 in case of corruption.
+ */
+static u64 apfs_ip_find_free(struct super_block *sb)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	int blk_bitcnt = sb->s_blocksize * 8;
+	u64 full_bitcnt = le64_to_cpu(sm_raw->sm_ip_block_count);
+	u32 i;
+
+	for (i = 0; i < sm->sm_ip_bmaps_count; ++i) {
+		char *bitmap = sm->sm_ip_bmaps[i].block;
+		u64 off_in_bmap_blk, off_in_ip;
+
+		off_in_bmap_blk = find_next_zero_bit_le(bitmap, blk_bitcnt, 0 /* offset */);
+		if (off_in_bmap_blk >= blk_bitcnt) /* No space in this chunk */
+			continue;
+
+		/* We found something, confirm that it's not outside the ip */
+		off_in_ip = (i << sm->sm_ip_bmaps_shift) + off_in_bmap_blk;
+		if (off_in_ip >= full_bitcnt)
+			break;
+		return le64_to_cpu(sm_raw->sm_ip_base) + off_in_ip;
+	}
+	apfs_err(sb, "internal pool seems full");
+	return 0;
+}
+
+/**
+ * apfs_chunk_find_free - Find a free block inside a chunk
+ * @sb:		superblock structure
+ * @bitmap:	allocation bitmap for the chunk, which should have free blocks
+ * @addr:	number of the first block in the chunk
+ *
+ * Returns the block number for a free block, or 0 in case of corruption.
+ */
+static u64 apfs_chunk_find_free(struct super_block *sb, char *bitmap, u64 addr)
+{
+	int bitcount = sb->s_blocksize * 8;
+	u64 bno;
+
+	bno = find_next_zero_bit_le(bitmap, bitcount, 0 /* offset */);
+	if (bno >= bitcount)
+		return 0;
+	return addr + bno;
+}
+
+/**
+ * apfs_ip_mark_used - Mark a block in the internal pool as used
+ * @sb:		superblock strucuture
+ * @bno:	block number (must belong to the ip)
+ */
+static void apfs_ip_mark_used(struct super_block *sb, u64 bno)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_ip_bitmap_block_info *info = NULL;
+
+	bno -= le64_to_cpu(sm_raw->sm_ip_base);
+	info = &sm->sm_ip_bmaps[bno >> sm->sm_ip_bmaps_shift];
+	__set_bit_le(bno & sm->sm_ip_bmaps_mask, info->block);
+	info->dirty = true;
+}
+
+/**
+ * apfs_chunk_mark_used - Mark a block inside a chunk as used
+ * @sb:		superblock structure
+ * @bitmap:	allocation bitmap for the chunk
+ * @bno:	block number (must belong to the chunk)
+ */
+static inline void apfs_chunk_mark_used(struct super_block *sb, char *bitmap,
+					u64 bno)
+{
+	int bitcount = sb->s_blocksize * 8;
+
+	__set_bit_le(bno & (bitcount - 1), bitmap);
+}
+
+/**
+ * apfs_chunk_mark_free - Mark a block inside a chunk as free
+ * @sb:		superblock structure
+ * @bitmap:	allocation bitmap for the chunk
+ * @bno:	block number (must belong to the chunk)
+ */
+static inline int apfs_chunk_mark_free(struct super_block *sb, char *bitmap,
+					u64 bno)
+{
+	int bitcount = sb->s_blocksize * 8;
+
+	return __test_and_clear_bit_le(bno & (bitcount - 1), bitmap);
+}
+
+/**
+ * apfs_free_queue_try_insert - Try to add a block range to its free queue
+ * @sb:		superblock structure
+ * @bno:	first block number to free
+ * @count:	number of consecutive blocks to free
+ *
+ * Same as apfs_free_queue_insert_nocache(), except that this one can also fail
+ * with -ENOSPC if there is no room for the new record.
+ */
+static int apfs_free_queue_try_insert(struct super_block *sb, u64 bno, u64 count)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_spaceman_free_queue *fq;
+	struct apfs_node *fq_root = NULL;
+	struct apfs_btree_info *fq_info = NULL;
+	struct apfs_query *query = NULL;
+	struct apfs_spaceman_free_queue_key raw_key;
+	bool ghost = count == 1;
+	int needed_room;
+	__le64 raw_val;
+	u64 node_count;
+	u16 node_limit;
+	unsigned int qid;
+	int err;
+
+	qid = apfs_block_in_ip(sm, bno) ? APFS_SFQ_IP : APFS_SFQ_MAIN;
+	fq = &sm_raw->sm_fq[qid];
+
+	fq_root = apfs_read_node(sb, le64_to_cpu(fq->sfq_tree_oid),
+				 APFS_OBJ_EPHEMERAL, true /* write */);
+	if (IS_ERR(fq_root)) {
+		apfs_err(sb, "failed to read fq root 0x%llx", le64_to_cpu(fq->sfq_tree_oid));
+		return PTR_ERR(fq_root);
+	}
+
+	query = apfs_alloc_query(fq_root, NULL /* parent */);
+	if (!query) {
+		err = -ENOMEM;
+		goto fail;
+	}
+	apfs_init_free_queue_key(nxi->nx_xid, bno, &query->key);
+	query->flags |= APFS_QUERY_FREE_QUEUE;
+
+	err = apfs_btree_query(sb, &query);
+	if (err && err != -ENODATA) {
+		apfs_err(sb, "query failed for xid 0x%llx, paddr 0x%llx", nxi->nx_xid, bno);
+		goto fail;
+	}
+
+	fq_info = (void *)fq_root->object.data + sb->s_blocksize - sizeof(*fq_info);
+	node_count = le64_to_cpu(fq_info->bt_node_count);
+	node_limit = le16_to_cpu(fq->sfq_tree_node_limit);
+	if (node_count == node_limit) {
+		needed_room = sizeof(raw_key) + (ghost ? 0 : sizeof(raw_val));
+		if (!apfs_node_has_room(query->node, needed_room, false /* replace */)) {
+			err = -ENOSPC;
+			goto fail;
+		}
+	}
+
+	raw_key.sfqk_xid = cpu_to_le64(nxi->nx_xid);
+	raw_key.sfqk_paddr = cpu_to_le64(bno);
+	if (ghost) {
+		/* A lack of value (ghost record) means single-block extent */
+		err = apfs_btree_insert(query, &raw_key, sizeof(raw_key), NULL /* val */, 0 /* val_len */);
+	} else {
+		raw_val = cpu_to_le64(count);
+		err = apfs_btree_insert(query, &raw_key, sizeof(raw_key), &raw_val, sizeof(raw_val));
+	}
+	if (err) {
+		apfs_err(sb, "insertion failed for xid 0x%llx, paddr 0x%llx", nxi->nx_xid, bno);
+		goto fail;
+	}
+
+	if (!fq->sfq_oldest_xid)
+		fq->sfq_oldest_xid = cpu_to_le64(nxi->nx_xid);
+	le64_add_cpu(&fq->sfq_count, count);
+
+	if (qid == APFS_SFQ_MAIN)
+		sm->sm_main_fq_nodes = le64_to_cpu(fq_info->bt_node_count);
+
+fail:
+	apfs_free_query(query);
+	apfs_node_free(fq_root);
+	return err;
+}
+
+/**
+ * apfs_free_queue_insert_nocache - Add a block range to its free queue
+ * @sb:		superblock structure
+ * @bno:	first block number to free
+ * @count:	number of consecutive blocks to free
+ *
+ * Same as apfs_free_queue_insert(), but writes to the free queue directly,
+ * bypassing the cache of the latest freed block range.
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_free_queue_insert_nocache(struct super_block *sb, u64 bno, u64 count)
+{
+	unsigned int qid;
+	int err;
+
+	err = apfs_free_queue_try_insert(sb, bno, count);
+	if (err == -ENOSPC) {
+		qid = apfs_block_in_ip(APFS_SM(sb), bno) ? APFS_SFQ_IP : APFS_SFQ_MAIN;
+		apfs_alert(sb, "free queue (%u) seems full - bug!", qid);
+		err = -EFSCORRUPTED;
+	}
+	if (err) {
+		apfs_err(sb, "fq insert failed (0x%llx-0x%llx)", bno, count);
+		return err;
+	}
+	return 0;
+}
+
+/**
+ * apfs_free_queue_insert - Add a block range to its free queue
+ * @sb:		superblock structure
+ * @bno:	first block number to free
+ * @count:	number of consecutive blocks to free
+ *
+ * Uses a cache to delay the actual tree operations as much as possible.
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_free_queue_insert(struct super_block *sb, u64 bno, u64 count)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	int err;
+
+	if (sm->sm_free_cache_base == 0) {
+		/* Nothing yet cached */
+		sm->sm_free_cache_base = bno;
+		sm->sm_free_cache_blkcnt = count;
+		return 0;
+	}
+
+	/*
+	 * First attempt to extend the cache of freed blocks, but never cache
+	 * a range that doesn't belong to a single free queue.
+	 */
+	if (apfs_block_in_ip(sm, bno) == apfs_block_in_ip(sm, sm->sm_free_cache_base)) {
+		if (bno == sm->sm_free_cache_base + sm->sm_free_cache_blkcnt) {
+			sm->sm_free_cache_blkcnt += count;
+			return 0;
+		}
+		if (bno + count == sm->sm_free_cache_base) {
+			sm->sm_free_cache_base -= count;
+			sm->sm_free_cache_blkcnt += count;
+			return 0;
+		}
+	}
+
+	/* Failed to extend the cache, so flush it and replace it */
+	err = apfs_free_queue_insert_nocache(sb, sm->sm_free_cache_base, sm->sm_free_cache_blkcnt);
+	if (err) {
+		apfs_err(sb, "fq cache flush failed (0x%llx-0x%llx)", sm->sm_free_cache_base, sm->sm_free_cache_blkcnt);
+		return err;
+	}
+	sm->sm_free_cache_base = bno;
+	sm->sm_free_cache_blkcnt = count;
+	return 0;
+}
+
+/**
+ * apfs_chunk_alloc_free - Allocate or free block in given CIB and chunk
+ * @sb:		superblock structure
+ * @cib_bh:	buffer head for the chunk-info block
+ * @index:	index of this chunk's info structure inside @cib
+ * @bno:	block number
+ * @is_alloc:	true to allocate, false to free
+ */
+static int apfs_chunk_alloc_free(struct super_block *sb,
+				 struct buffer_head **cib_bh,
+				 int index, u64 *bno, bool is_alloc)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_chunk_info_block *cib;
+	struct apfs_chunk_info *ci;
+	struct buffer_head *bmap_bh = NULL;
+	char *bmap = NULL;
+	bool old_cib = false;
+	bool old_bmap = false;
+	int err = 0;
+
+	cib = (struct apfs_chunk_info_block *)(*cib_bh)->b_data;
+	ci = &cib->cib_chunk_info[index];
+
+	/* Cibs and bitmaps from old transactions can't be modified in place */
+	if (le64_to_cpu(cib->cib_o.o_xid) < nxi->nx_xid)
+		old_cib = true;
+	if (le64_to_cpu(ci->ci_xid) < nxi->nx_xid)
+		old_bmap = true;
+	if (is_alloc && le32_to_cpu(ci->ci_free_count) < 1)
+		return -ENOSPC;
+
+	/* Read the current bitmap, or allocate it if necessary */
+	if (!ci->ci_bitmap_addr) {
+		u64 bmap_bno;
+
+		if (!is_alloc) {
+			apfs_err(sb, "attempt to free block in all-free chunk");
+			return -EFSCORRUPTED;
+		}
+
+		/* All blocks in this chunk are free */
+		bmap_bno = apfs_ip_find_free(sb);
+		if (!bmap_bno) {
+			apfs_err(sb, "no free blocks in ip");
+			return -EFSCORRUPTED;
+		}
+		bmap_bh = apfs_sb_bread(sb, bmap_bno);
+	} else {
+		bmap_bh = apfs_sb_bread(sb, le64_to_cpu(ci->ci_bitmap_addr));
+	}
+	if (!bmap_bh) {
+		apfs_err(sb, "failed to read bitmap block");
+		return -EIO;
+	}
+	bmap = bmap_bh->b_data;
+	if (!ci->ci_bitmap_addr) {
+		memset(bmap, 0, sb->s_blocksize);
+		old_bmap = false;
+	}
+
+	/* Write the bitmap to its location for the next transaction */
+	if (old_bmap) {
+		struct buffer_head *new_bmap_bh;
+		u64 new_bmap_bno;
+
+		new_bmap_bno = apfs_ip_find_free(sb);
+		if (!new_bmap_bno) {
+			apfs_err(sb, "no free blocks in ip");
+			err = -EFSCORRUPTED;
+			goto fail;
+		}
+
+		new_bmap_bh = apfs_getblk(sb, new_bmap_bno);
+		if (!new_bmap_bh) {
+			apfs_err(sb, "failed to map new bmap block (0x%llx)", new_bmap_bno);
+			err = -EIO;
+			goto fail;
+		}
+		memcpy(new_bmap_bh->b_data, bmap, sb->s_blocksize);
+		err = apfs_free_queue_insert(sb, bmap_bh->b_blocknr, 1);
+		brelse(bmap_bh);
+		bmap_bh = new_bmap_bh;
+		if (err) {
+			apfs_err(sb, "free queue insertion failed");
+			goto fail;
+		}
+		bmap = bmap_bh->b_data;
+	}
+	apfs_ip_mark_used(sb, bmap_bh->b_blocknr);
+
+	/* Write the cib to its location for the next transaction */
+	if (old_cib) {
+		struct buffer_head *new_cib_bh;
+		u64 new_cib_bno;
+
+		new_cib_bno = apfs_ip_find_free(sb);
+		if (!new_cib_bno) {
+			apfs_err(sb, "no free blocks in ip");
+			err = -EFSCORRUPTED;
+			goto fail;
+		}
+
+		new_cib_bh = apfs_getblk(sb, new_cib_bno);
+		if (!new_cib_bh) {
+			apfs_err(sb, "failed to map new cib block (0x%llx)", new_cib_bno);
+			err = -EIO;
+			goto fail;
+		}
+		memcpy(new_cib_bh->b_data, (*cib_bh)->b_data, sb->s_blocksize);
+		err = apfs_free_queue_insert(sb, (*cib_bh)->b_blocknr, 1);
+		brelse(*cib_bh);
+		*cib_bh = new_cib_bh;
+		if (err) {
+			apfs_err(sb, "free queue insertion failed");
+			goto fail;
+		}
+
+		err = apfs_transaction_join(sb, *cib_bh);
+		if (err)
+			goto fail;
+
+		cib = (struct apfs_chunk_info_block *)(*cib_bh)->b_data;
+		ci = &cib->cib_chunk_info[index];
+		cib->cib_o.o_oid = cpu_to_le64(new_cib_bno);
+		cib->cib_o.o_xid = cpu_to_le64(nxi->nx_xid);
+
+		apfs_ip_mark_used(sb, new_cib_bno);
+	}
+
+	/* The chunk info can be updated now */
+	apfs_assert_in_transaction(sb, &cib->cib_o);
+	ci->ci_xid = cpu_to_le64(nxi->nx_xid);
+	le32_add_cpu(&ci->ci_free_count, is_alloc ? -1 : 1);
+	ci->ci_bitmap_addr = cpu_to_le64(bmap_bh->b_blocknr);
+	ASSERT(buffer_trans(*cib_bh));
+	set_buffer_csum(*cib_bh);
+
+	/* Finally, allocate / free the actual block that was requested */
+	if (is_alloc) {
+		*bno = apfs_chunk_find_free(sb, bmap, le64_to_cpu(ci->ci_addr));
+		if (!*bno) {
+			apfs_err(sb, "no free blocks in chunk");
+			err = -EFSCORRUPTED;
+			goto fail;
+		}
+		apfs_chunk_mark_used(sb, bmap, *bno);
+		sm->sm_free_count -= 1;
+	} else {
+		if (!apfs_chunk_mark_free(sb, bmap, *bno)) {
+			apfs_err(sb, "block already marked as free (0x%llx)", *bno);
+			le32_add_cpu(&ci->ci_free_count, -1);
+			set_buffer_csum(*cib_bh);
+			err = -EFSCORRUPTED;
+		} else
+			sm->sm_free_count += 1;
+	}
+	mark_buffer_dirty(bmap_bh);
+
+fail:
+	brelse(bmap_bh);
+	return err;
+}
+
+/**
+ * apfs_chunk_allocate_block - Allocate a single block from a chunk
+ * @sb:		superblock structure
+ * @cib_bh:	buffer head for the chunk-info block
+ * @index:	index of this chunk's info structure inside @cib
+ * @bno:	on return, the allocated block number
+ *
+ * Finds a free block in the chunk and marks it as used; the buffer at @cib_bh
+ * may be replaced if needed for copy-on-write.  Returns 0 on success, or a
+ * negative error code in case of failure.
+ */
+static int apfs_chunk_allocate_block(struct super_block *sb,
+				     struct buffer_head **cib_bh,
+				     int index, u64 *bno)
+{
+	return apfs_chunk_alloc_free(sb, cib_bh, index, bno, true);
+}
+
+/**
+ * apfs_cib_allocate_block - Allocate a single block from a cib
+ * @sb:		superblock structure
+ * @cib_bh:	buffer head for the chunk-info block
+ * @bno:	on return, the allocated block number
+ * @backwards:	start the search on the last chunk
+ *
+ * Finds a free block among all the chunks in the cib and marks it as used; the
+ * buffer at @cib_bh may be replaced if needed for copy-on-write.  Returns 0 on
+ * success, or a negative error code in case of failure.
+ */
+static int apfs_cib_allocate_block(struct super_block *sb,
+				   struct buffer_head **cib_bh, u64 *bno, bool backwards)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_chunk_info_block *cib;
+	u32 chunk_count;
+	int i;
+
+	cib = (struct apfs_chunk_info_block *)(*cib_bh)->b_data;
+	if (nxi->nx_flags & APFS_CHECK_NODES && !apfs_obj_verify_csum(sb, *cib_bh)) {
+		apfs_err(sb, "bad checksum for chunk-info block");
+		return -EFSBADCRC;
+	}
+
+	/* Avoid out-of-bounds operations on corrupted cibs */
+	chunk_count = le32_to_cpu(cib->cib_chunk_info_count);
+	if (chunk_count > sm->sm_chunks_per_cib) {
+		apfs_err(sb, "too many chunks in cib (%u)", chunk_count);
+		return -EFSCORRUPTED;
+	}
+
+	for (i = 0; i < chunk_count; ++i) {
+		int index;
+		int err;
+
+		index = backwards ? chunk_count - 1 - i : i;
+
+		err = apfs_chunk_allocate_block(sb, cib_bh, index, bno);
+		if (err == -ENOSPC) /* This chunk is full */
+			continue;
+		if (err)
+			apfs_err(sb, "error during allocation");
+		return err;
+	}
+	return -ENOSPC;
+}
+
+/**
+ * apfs_spaceman_allocate_block - Allocate a single on-disk block
+ * @sb:		superblock structure
+ * @bno:	on return, the allocated block number
+ * @backwards:	start the search on the last chunk
+ *
+ * Finds a free block among the spaceman bitmaps and marks it as used.  Returns
+ * 0 on success, or a negative error code in case of failure.
+ */
+int apfs_spaceman_allocate_block(struct super_block *sb, u64 *bno, bool backwards)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	int i;
+
+	for (i = 0; i < sm->sm_cib_count; ++i) {
+		struct buffer_head *cib_bh;
+		u64 cib_bno;
+		int index;
+		int err;
+
+		/* Keep extents and metadata separate to limit fragmentation */
+		index = backwards ? sm->sm_cib_count - 1 - i : i;
+
+		cib_bno = apfs_spaceman_read_cib_addr(sb, index);
+		cib_bh = apfs_sb_bread(sb, cib_bno);
+		if (!cib_bh) {
+			apfs_err(sb, "failed to read cib");
+			return -EIO;
+		}
+
+		err = apfs_cib_allocate_block(sb, &cib_bh, bno, backwards);
+		if (!err) {
+			/* The cib may have been moved */
+			apfs_spaceman_write_cib_addr(sb, index, cib_bh->b_blocknr);
+			/* The free block count has changed */
+			apfs_write_spaceman(sm);
+		}
+		brelse(cib_bh);
+		if (err == -ENOSPC) /* This cib is full */
+			continue;
+		if (err)
+			apfs_err(sb, "error during allocation");
+		return err;
+	}
+	/*
+	 * We checked the free space before starting the transaction, so this
+	 * isn't expected to happen.
+	 */
+	apfs_err(sb, "ran out of space during transaction");
+	return -ENOSPC;
+}
+
+/**
+ * apfs_chunk_free - Mark a regular block as free given CIB and chunk
+ * @sb:		superblock structure
+ * @cib_bh:	buffer head for the chunk-info block
+ * @index:	index of this chunk's info structure inside @cib
+ * @bno:	block number (must not belong to the ip)
+ */
+static int apfs_chunk_free(struct super_block *sb,
+				struct buffer_head **cib_bh,
+				int index, u64 bno)
+{
+	return apfs_chunk_alloc_free(sb, cib_bh, index, &bno, false);
+}
+
+/**
+ * apfs_main_free - Mark a regular block as free
+ * @sb:		superblock structure
+ * @bno:	block number (must not belong to the ip)
+ */
+static int apfs_main_free(struct super_block *sb, u64 bno)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+	struct apfs_sb_info *sbi = NULL;
+	u64 cib_idx, chunk_idx;
+	struct buffer_head *cib_bh;
+	u64 cib_bno;
+	int err, orphan_err;
+
+	if (!sm_raw->sm_blocks_per_chunk || !sm_raw->sm_chunks_per_cib) {
+		apfs_err(sb, "block or chunk count not set");
+		return -EINVAL;
+	}
+	/* TODO: use bitshifts instead of do_div() */
+	chunk_idx = bno;
+	do_div(chunk_idx, sm->sm_blocks_per_chunk);
+	cib_idx = chunk_idx;
+	chunk_idx = do_div(cib_idx, sm->sm_chunks_per_cib);
+
+	cib_bno = apfs_spaceman_read_cib_addr(sb, cib_idx);
+	cib_bh = apfs_sb_bread(sb, cib_bno);
+	if (!cib_bh) {
+		apfs_err(sb, "failed to read cib");
+		return -EIO;
+	}
+
+	err = apfs_chunk_free(sb, &cib_bh, chunk_idx, bno);
+	if (!err) {
+		/* The cib may have been moved */
+		apfs_spaceman_write_cib_addr(sb, cib_idx, cib_bh->b_blocknr);
+		/* The free block count has changed */
+		apfs_write_spaceman(sm);
+	}
+	brelse(cib_bh);
+	if (err) {
+		apfs_err(sb, "error during free");
+		return err;
+	}
+
+	/* It may be time to resume orphan cleanups, if we made enough room */
+	sbi = APFS_SB(sb);
+	orphan_err = atomic_read(&sbi->s_orphan_cleanup_err);
+	if (orphan_err == -ENOSPC && sm->sm_free_count >= 2 * APFS_DEL_ROOM) {
+		atomic_set(&sbi->s_orphan_cleanup_err, 0);
+		apfs_schedule_orphan_cleanup(sb);
+	}
+
+	return err;
+}
+
+/**
+ * apfs_spaceman_get_free_blkcnt - Calculate the total number of free blocks
+ * @sb:		filesystem superblock
+ * @blkcnt:	on return, the total number of free blocks for all devices
+ *
+ * Can be called even if the spaceman has not been read (for example, on a
+ * read-only mount). Returns 0 on success, or a negative error code in case of
+ * failure.
+ */
+int apfs_spaceman_get_free_blkcnt(struct super_block *sb, u64 *blkcnt)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *raw_sb = NULL;
+	struct apfs_spaceman_phys *sm_raw = NULL;
+	struct apfs_ephemeral_object_info *sm_eph_info = NULL;
+	struct apfs_spaceman_device *dev = NULL;
+	u64 oid;
+	int err;
+
+	if (!nxi->nx_eph_list) {
+		err = apfs_read_ephemeral_objects(sb);
+		if (err) {
+			apfs_err(sb, "failed to read the ephemeral objects");
+			return err;
+		}
+	}
+
+	raw_sb = nxi->nx_raw;
+	oid = le64_to_cpu(raw_sb->nx_spaceman_oid);
+	sm_eph_info = apfs_ephemeral_object_lookup(sb, oid);
+	if (IS_ERR(sm_eph_info)) {
+		apfs_err(sb, "no spaceman object for oid 0x%llx", oid);
+		return PTR_ERR(sm_eph_info);
+	}
+	sm_raw = (struct apfs_spaceman_phys *)sm_eph_info->object;
+
+	*blkcnt = 0;
+	dev = &sm_raw->sm_dev[APFS_SD_MAIN];
+	*blkcnt += le64_to_cpu(dev->sm_free_count);
+	dev = &sm_raw->sm_dev[APFS_SD_TIER2];
+	*blkcnt += le64_to_cpu(dev->sm_free_count);
+	return 0;
+}
diff --git a/fs/apfs/super.c b/fs/apfs/super.c
new file mode 100644
index 000000000..9cf63e85d
--- /dev/null
+++ b/fs/apfs/super.c
@@ -0,0 +1,2282 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/backing-dev.h>
+#include <linux/blkdev.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/magic.h>
+#include <linux/slab.h>
+#include <linux/parser.h>
+#include <linux/buffer_head.h>
+#include <linux/statfs.h>
+#include <linux/seq_file.h>
+#include "apfs.h"
+#include "version.h"
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+#include <linux/fs_context.h>
+#endif
+
+#define APFS_MODULE_ID_STRING	"linux-apfs by eafer (" GIT_COMMIT ")"
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 16, 0) /* iversion came in 4.16 */
+#include <linux/iversion.h>
+#endif
+
+/* Keep a list of mounted containers, so that their volumes can share them */
+static LIST_HEAD(nxs);
+/*
+ * The main purpose of this mutex is to protect the list of containers and
+ * their reference counts, but it also has other uses during mounts/unmounts:
+ *   - it prevents new mounts from starting while an unmount is updating the
+ *     backup superblock (apfs_attach_nxi vs apfs_make_super_copy)
+ *   - it prevents a new container superblock read from starting while another
+ *     is taking place, which could cause leaks and other issues if both
+ *     containers are the same (apfs_read_main_super vs itself)
+ *   - it protects the list of volumes for each container, and keeps it
+ *     consistent with the reference count
+ *   - it prevents two different snapshots for a single volume from trying to
+ *     do the first read of their shared omap at the same time
+ *     (apfs_first_read_omap vs itself)
+ *   - it protects the reference count for that shared omap, keeping it
+ *     consistent with the number of volumes that are set with that omap
+ *   - it protects the container mount flags, so that they can only be set by
+ *     the first volume mount to attempt it (apfs_set_nx_flags vs itself)
+ */
+DEFINE_MUTEX(nxs_mutex);
+
+/**
+ * apfs_nx_find_by_dev - Search for a device in the list of mounted containers
+ * @dev:	device number of block device for the wanted container
+ *
+ * Returns a pointer to the container structure in the list, or NULL if the
+ * container isn't currently mounted.
+ */
+static struct apfs_nxsb_info *apfs_nx_find_by_dev(dev_t dev)
+{
+	struct apfs_nxsb_info *curr;
+
+	lockdep_assert_held(&nxs_mutex);
+	list_for_each_entry(curr, &nxs, nx_list) {
+		struct block_device *curr_bdev = curr->nx_blkdev_info->blki_bdev;
+
+		if (curr_bdev->bd_dev == dev)
+			return curr;
+	}
+	return NULL;
+}
+
+/**
+ * apfs_blkdev_set_blocksize - Set the blocksize for a block device
+ * @info:	info struct for the block device
+ * @size:	size to set
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_blkdev_set_blocksize(struct apfs_blkdev_info *info, int size)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 10, 0)
+	return set_blocksize(info->blki_bdev_file, size);
+#else
+	return set_blocksize(info->blki_bdev, size);
+#endif
+}
+
+/**
+ * apfs_sb_set_blocksize - Set the block size for the container's device
+ * @sb:		superblock structure
+ * @size:	size to set
+ *
+ * This is like sb_set_blocksize(), but it uses the container's device instead
+ * of the nonexistent volume device.
+ */
+static int apfs_sb_set_blocksize(struct super_block *sb, int size)
+{
+	if (apfs_blkdev_set_blocksize(APFS_NXI(sb)->nx_blkdev_info, size))
+		return 0;
+	sb->s_blocksize = size;
+	sb->s_blocksize_bits = blksize_bits(size);
+	return sb->s_blocksize;
+}
+
+/**
+ * apfs_read_super_copy - Read the copy of the container superblock in block 0
+ * @sb: superblock structure
+ *
+ * Returns a pointer to the buffer head, or an error pointer in case of failure.
+ */
+static struct buffer_head *apfs_read_super_copy(struct super_block *sb)
+{
+	struct buffer_head *bh;
+	struct apfs_nx_superblock *msb_raw;
+	int blocksize;
+	int err = -EINVAL;
+
+	/*
+	 * For now assume a small blocksize, we only need it so that we can
+	 * read the actual blocksize from disk.
+	 */
+	if (!apfs_sb_set_blocksize(sb, APFS_NX_DEFAULT_BLOCK_SIZE)) {
+		apfs_err(sb, "unable to set blocksize");
+		return ERR_PTR(err);
+	}
+	bh = apfs_sb_bread(sb, APFS_NX_BLOCK_NUM);
+	if (!bh) {
+		apfs_err(sb, "unable to read superblock");
+		return ERR_PTR(err);
+	}
+	msb_raw = (struct apfs_nx_superblock *)bh->b_data;
+	blocksize = le32_to_cpu(msb_raw->nx_block_size);
+
+	sb->s_magic = le32_to_cpu(msb_raw->nx_magic);
+	if (sb->s_magic != APFS_NX_MAGIC) {
+		apfs_warn(sb, "not an apfs container - are you mounting the right partition?");
+		goto fail;
+	}
+
+	if (sb->s_blocksize != blocksize) {
+		brelse(bh);
+
+		if (!apfs_sb_set_blocksize(sb, blocksize)) {
+			apfs_err(sb, "bad blocksize %d", blocksize);
+			return ERR_PTR(err);
+		}
+		bh = apfs_sb_bread(sb, APFS_NX_BLOCK_NUM);
+		if (!bh) {
+			apfs_err(sb, "unable to read superblock 2nd time");
+			return ERR_PTR(err);
+		}
+		msb_raw = (struct apfs_nx_superblock *)bh->b_data;
+	}
+
+	if (!apfs_obj_verify_csum(sb, bh))
+		apfs_notice(sb, "backup superblock seems corrupted");
+	return bh;
+
+fail:
+	brelse(bh);
+	return ERR_PTR(err);
+}
+
+/**
+ * apfs_make_super_copy - Write a copy of the checkpoint superblock to block 0
+ * @sb:	superblock structure
+ */
+static void apfs_make_super_copy(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = sbi->s_nxi;
+	struct buffer_head *bh;
+
+	if (!(nxi->nx_flags & APFS_READWRITE))
+		return;
+
+	/*
+	 * Only update the backup when the last volume is getting unmounted.
+	 * Of course a new mounter could still come along before we actually
+	 * release the nxi.
+	 */
+	mutex_lock(&nxs_mutex);
+	if (nxi->nx_refcnt > 1)
+		goto out_unlock;
+
+	bh = apfs_sb_bread(sb, APFS_NX_BLOCK_NUM);
+	if (!bh) {
+		apfs_err(sb, "failed to write block zero");
+		goto out_unlock;
+	}
+	memcpy(bh->b_data, nxi->nx_raw, sb->s_blocksize);
+	mark_buffer_dirty(bh);
+	brelse(bh);
+out_unlock:
+	mutex_unlock(&nxs_mutex);
+}
+
+static int apfs_check_nx_features(struct super_block *sb);
+static void apfs_set_trans_buffer_limit(struct super_block *sb);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 7, 0)
+static inline void import_uuid(uuid_t *dst, const __u8 *src)
+{
+	memcpy(dst, src, sizeof(uuid_t));
+}
+#endif
+
+/**
+ * apfs_check_fusion_uuid - Verify that the main and tier 2 devices match
+ * @sb: filesystem superblock
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_check_fusion_uuid(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = NULL;
+	struct apfs_nx_superblock *main_raw = NULL, *tier2_raw = NULL;
+	uuid_t main_uuid, tier2_uuid;
+	struct buffer_head *bh = NULL;
+
+	nxi = APFS_NXI(sb);
+	main_raw = nxi->nx_raw;
+	if (!main_raw) {
+		apfs_alert(sb, "fusion uuid checks are misplaced");
+		return -EINVAL;
+	}
+	import_uuid(&main_uuid, main_raw->nx_fusion_uuid);
+	main_raw = NULL;
+
+	if (!nxi->nx_tier2_info) {
+		/* Not a fusion drive */
+		if (!uuid_is_null(&main_uuid)) {
+			apfs_err(sb, "fusion uuid on a regular drive");
+			return -EFSCORRUPTED;
+		}
+		return 0;
+	}
+	if (uuid_is_null(&main_uuid)) {
+		apfs_err(sb, "no fusion uuid on fusion drive");
+		return -EFSCORRUPTED;
+	}
+
+	/* Tier 2 also has a copy of the superblock in block zero */
+	bh = apfs_sb_bread(sb, nxi->nx_tier2_bno);
+	if (IS_ERR(bh))
+		return PTR_ERR(bh);
+	tier2_raw = (struct apfs_nx_superblock *)bh->b_data;
+	import_uuid(&tier2_uuid, tier2_raw->nx_fusion_uuid);
+	brelse(bh);
+	bh = NULL;
+	tier2_raw = NULL;
+
+	/*
+	 * The only difference between both superblocks (other than the
+	 * checksum) is this one bit here, so it can be used to tell which is
+	 * main and which is tier 2. By the way, the reference seems to have
+	 * this backwards.
+	 */
+	if (main_uuid.b[15] & 0x01) {
+		apfs_warn(sb, "bad bit on main device - are you mixing up main and tier 2?");
+		return -EINVAL;
+	}
+	if (!(tier2_uuid.b[15] & 0x01)) {
+		apfs_warn(sb, "bad bit on tier 2 device - are you mixing up main and tier 2?");
+		return -EINVAL;
+	}
+	tier2_uuid.b[15] &= ~0x01;
+	if (!uuid_equal(&main_uuid, &tier2_uuid)) {
+		apfs_warn(sb, "the devices are not part of the same fusion drive");
+		return -EINVAL;
+	}
+	return 0;
+}
+
+/**
+ * apfs_read_main_super - Find the container superblock and read it into memory
+ * @sb:	superblock structure
+ *
+ * Returns a negative error code in case of failure.  On success, returns 0
+ * and sets the nx_raw and nx_xid fields of APFS_NXI(@sb).
+ */
+static int apfs_read_main_super(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct buffer_head *bh = NULL;
+	struct buffer_head *desc_bh = NULL;
+	struct apfs_nx_superblock *msb_raw;
+	u64 xid, bno = APFS_NX_BLOCK_NUM;
+	u64 desc_base;
+	u32 desc_blocks;
+	int err = -EINVAL;
+	int i;
+
+	mutex_lock(&nxs_mutex);
+
+	if (nxi->nx_blocksize) {
+		/* It's already mapped */
+		sb->s_blocksize = nxi->nx_blocksize;
+		sb->s_blocksize_bits = nxi->nx_blocksize_bits;
+		sb->s_magic = le32_to_cpu(nxi->nx_raw->nx_magic);
+		err = 0;
+		goto out;
+	}
+
+	/*
+	 * We won't know the block size until we read the backup superblock,
+	 * so we can't set this up correctly yet. But we do know that the
+	 * backup superblock itself is always in the main device.
+	 */
+	nxi->nx_tier2_bno = APFS_NX_BLOCK_NUM + 1;
+
+	/* Read the superblock from the last clean unmount */
+	bh = apfs_read_super_copy(sb);
+	if (IS_ERR(bh)) {
+		err = PTR_ERR(bh);
+		bh = NULL;
+		goto out;
+	}
+	msb_raw = (struct apfs_nx_superblock *)bh->b_data;
+
+	/*
+	 * Now that we confirmed the block size, we can set this up for real.
+	 * It's important to do this early because I don't know which mount
+	 * objects could get moved to tier 2.
+	 */
+	nxi->nx_tier2_bno = APFS_FUSION_TIER2_DEVICE_BYTE_ADDR >> sb->s_blocksize_bits;
+
+	/* We want to mount the latest valid checkpoint among the descriptors */
+	desc_base = le64_to_cpu(msb_raw->nx_xp_desc_base);
+	if (desc_base >> 63 != 0) {
+		/* The highest bit is set when checkpoints are not contiguous */
+		apfs_err(sb, "checkpoint descriptor tree not yet supported");
+		goto out;
+	}
+	desc_blocks = le32_to_cpu(msb_raw->nx_xp_desc_blocks);
+	if (desc_blocks > 10000) { /* Arbitrary loop limit, is it enough? */
+		apfs_err(sb, "too many checkpoint descriptors?");
+		err = -EFSCORRUPTED;
+		goto out;
+	}
+
+	/* Now we go through the checkpoints one by one */
+	xid = le64_to_cpu(msb_raw->nx_o.o_xid);
+	for (i = 0; i < desc_blocks; ++i) {
+		struct apfs_nx_superblock *desc_raw;
+
+		brelse(desc_bh);
+		desc_bh = apfs_sb_bread(sb, desc_base + i);
+		if (!desc_bh) {
+			apfs_err(sb, "unable to read checkpoint descriptor");
+			goto out;
+		}
+		desc_raw = (struct apfs_nx_superblock *)desc_bh->b_data;
+
+		if (le32_to_cpu(desc_raw->nx_magic) != APFS_NX_MAGIC)
+			continue; /* Not a superblock */
+		if (le64_to_cpu(desc_raw->nx_o.o_xid) <= xid)
+			continue; /* Old */
+		if (!apfs_obj_verify_csum(sb, desc_bh))
+			continue; /* Corrupted */
+
+		xid = le64_to_cpu(desc_raw->nx_o.o_xid);
+		msb_raw = desc_raw;
+		bno = desc_base + i;
+		brelse(bh);
+		bh = desc_bh;
+		desc_bh = NULL;
+	}
+
+	nxi->nx_raw = kmalloc(sb->s_blocksize, GFP_KERNEL);
+	if (!nxi->nx_raw) {
+		err = -ENOMEM;
+		goto out;
+	}
+	memcpy(nxi->nx_raw, bh->b_data, sb->s_blocksize);
+	nxi->nx_bno = bno;
+	nxi->nx_xid = xid;
+
+	/* For now we only support blocksize < PAGE_SIZE */
+	nxi->nx_blocksize = sb->s_blocksize;
+	nxi->nx_blocksize_bits = sb->s_blocksize_bits;
+	apfs_set_trans_buffer_limit(sb);
+
+	err = apfs_check_nx_features(sb);
+	if (err)
+		goto out;
+
+	/*
+	 * This check is technically too late: if main and tier 2 are backwards
+	 * then we have already attempted (and failed) to read the checkpoint
+	 * from tier 2. This may lead to a confusing error message if tier 2
+	 * is absurdly tiny, not a big deal.
+	 */
+	err = apfs_check_fusion_uuid(sb);
+	if (err)
+		goto out;
+
+out:
+	brelse(bh);
+	mutex_unlock(&nxs_mutex);
+	return err;
+}
+
+/**
+ * apfs_update_software_info - Write the module info to a modified volume
+ * @sb: superblock structure
+ *
+ * Writes this module's information to index zero of the apfs_modified_by
+ * array, shifting the rest of the entries to the right.
+ */
+static void apfs_update_software_info(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *raw = sbi->s_vsb_raw;
+	struct apfs_modified_by *mod_by;
+
+	ASSERT(sbi->s_vsb_raw);
+	apfs_assert_in_transaction(sb, &raw->apfs_o);
+	ASSERT(strlen(APFS_MODULE_ID_STRING) < APFS_MODIFIED_NAMELEN);
+	mod_by = raw->apfs_modified_by;
+
+	memmove(mod_by + 1, mod_by, (APFS_MAX_HIST - 1) * sizeof(*mod_by));
+	memset(mod_by->id, 0, sizeof(mod_by->id));
+	strscpy(mod_by->id, APFS_MODULE_ID_STRING, sizeof(mod_by->id));
+	mod_by->timestamp = cpu_to_le64(ktime_get_real_ns());
+	mod_by->last_xid = cpu_to_le64(APFS_NXI(sb)->nx_xid);
+}
+
+static struct file_system_type apfs_fs_type;
+
+/**
+ * apfs_blkdev_cleanup - Clean up after a block device
+ * @info:	info struct to clean up
+ */
+static void apfs_blkdev_cleanup(struct apfs_blkdev_info *info)
+{
+	if (!info)
+		return;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 9, 0) || RHEL_VERSION_GE(9, 5)
+	fput(info->blki_bdev_file);
+	info->blki_bdev_file = NULL;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+	bdev_release(info->blki_bdev_handle);
+	info->blki_bdev_handle = NULL;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 5, 0) || RHEL_VERSION_GE(9, 4)
+	blkdev_put(info->blki_bdev, &apfs_fs_type);
+#else
+	blkdev_put(info->blki_bdev, info->blki_mode);
+#endif
+	info->blki_bdev = NULL;
+
+	kfree(info->blki_path);
+	info->blki_path = NULL;
+	kfree(info);
+}
+
+/**
+ * apfs_free_main_super - Clean up apfs_read_main_super()
+ * @sbi:	in-memory superblock info
+ *
+ * It also cleans up after apfs_attach_nxi(), so the name is no longer accurate.
+ */
+static inline void apfs_free_main_super(struct apfs_sb_info *sbi)
+{
+	struct apfs_nxsb_info *nxi = sbi->s_nxi;
+	struct apfs_ephemeral_object_info *eph_list = NULL;
+	struct apfs_spaceman *sm = NULL;
+	u32 bmap_idx;
+	int i;
+
+	if (!nxi)
+		return;
+
+	mutex_lock(&nxs_mutex);
+
+	list_del(&sbi->list);
+	if (--nxi->nx_refcnt)
+		goto out;
+
+	/* Clean up all the ephemeral objects in memory */
+	eph_list = nxi->nx_eph_list;
+	if (eph_list) {
+		for (i = 0; i < nxi->nx_eph_count; ++i) {
+			kfree(eph_list[i].object);
+			eph_list[i].object = NULL;
+		}
+		kfree(eph_list);
+		eph_list = nxi->nx_eph_list = NULL;
+		nxi->nx_eph_count = 0;
+	}
+
+	kfree(nxi->nx_raw);
+	nxi->nx_raw = NULL;
+
+	apfs_blkdev_cleanup(nxi->nx_blkdev_info);
+	nxi->nx_blkdev_info = NULL;
+	apfs_blkdev_cleanup(nxi->nx_tier2_info);
+	nxi->nx_tier2_info = NULL;
+
+	list_del(&nxi->nx_list);
+	sm = nxi->nx_spaceman;
+	if (sm) {
+		for (bmap_idx = 0; bmap_idx < sm->sm_ip_bmaps_count; ++bmap_idx) {
+			kfree(sm->sm_ip_bmaps[bmap_idx].block);
+			sm->sm_ip_bmaps[bmap_idx].block = NULL;
+		}
+		kfree(sm);
+		nxi->nx_spaceman = sm = NULL;
+	}
+	kfree(nxi);
+out:
+	sbi->s_nxi = NULL;
+	mutex_unlock(&nxs_mutex);
+}
+
+/**
+ * apfs_map_volume_super_bno - Map a block containing a volume superblock
+ * @sb:		superblock structure
+ * @bno:	block to map
+ * @check:	verify the checksum?
+ */
+int apfs_map_volume_super_bno(struct super_block *sb, u64 bno, bool check)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = NULL;
+	struct buffer_head *bh = NULL;
+	int err;
+
+	bh = apfs_sb_bread(sb, bno);
+	if (!bh) {
+		apfs_err(sb, "unable to read volume superblock");
+		return -EINVAL;
+	}
+
+	vsb_raw = (struct apfs_superblock *)bh->b_data;
+	if (le32_to_cpu(vsb_raw->apfs_magic) != APFS_MAGIC) {
+		apfs_err(sb, "wrong magic in volume superblock");
+		err = -EINVAL;
+		goto fail;
+	}
+
+	/*
+	 * XXX: apfs_omap_lookup_block() only runs this check when write
+	 * is true, but it should always do it.
+	 */
+	if (check && !apfs_obj_verify_csum(sb, bh)) {
+		apfs_err(sb, "inconsistent volume superblock");
+		err = -EFSBADCRC;
+		goto fail;
+	}
+
+	sbi->s_vsb_raw = vsb_raw;
+	sbi->s_vobject.sb = sb;
+	sbi->s_vobject.block_nr = bno;
+	sbi->s_vobject.oid = le64_to_cpu(vsb_raw->apfs_o.o_oid);
+	brelse(sbi->s_vobject.o_bh);
+	sbi->s_vobject.o_bh = bh;
+	sbi->s_vobject.data = bh->b_data;
+	return 0;
+
+fail:
+	brelse(bh);
+	return err;
+}
+
+/**
+ * apfs_alloc_omap - Allocate and initialize an object map struct
+ *
+ * Returns the struct, or NULL in case of allocation failure.
+ */
+static struct apfs_omap *apfs_alloc_omap(void)
+{
+	struct apfs_omap *omap = NULL;
+	struct apfs_omap_cache *cache = NULL;
+
+	omap = kzalloc(sizeof(*omap), GFP_KERNEL);
+	if (!omap)
+		return NULL;
+	cache = &omap->omap_cache;
+	spin_lock_init(&cache->lock);
+	return omap;
+}
+
+/**
+ * apfs_map_volume_super - Find the volume superblock and map it into memory
+ * @sb:		superblock structure
+ * @write:	request write access?
+ *
+ * Returns a negative error code in case of failure.  On success, returns 0
+ * and sets APFS_SB(@sb)->s_vsb_raw and APFS_SB(@sb)->s_vobject.
+ */
+int apfs_map_volume_super(struct super_block *sb, bool write)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *msb_raw = nxi->nx_raw;
+	struct apfs_omap_phys *msb_omap_raw;
+	struct apfs_omap *omap = NULL;
+	struct apfs_node *vnode;
+	struct buffer_head *bh;
+	u64 vol_id;
+	u64 vsb;
+	int err;
+
+	ASSERT(msb_raw);
+
+	/* Get the id for the requested volume number */
+	if (sbi->s_vol_nr >= APFS_NX_MAX_FILE_SYSTEMS) {
+		apfs_err(sb, "volume number out of range");
+		return -EINVAL;
+	}
+	vol_id = le64_to_cpu(msb_raw->nx_fs_oid[sbi->s_vol_nr]);
+	if (vol_id == 0) {
+		apfs_err(sb, "requested volume does not exist");
+		return -EINVAL;
+	}
+
+	/* Get the container's object map */
+	bh = apfs_read_object_block(sb, le64_to_cpu(msb_raw->nx_omap_oid),
+				    write, false /* preserve */);
+	if (IS_ERR(bh)) {
+		apfs_err(sb, "unable to read container object map");
+		return PTR_ERR(bh);
+	}
+	if (write) {
+		ASSERT(buffer_trans(bh));
+		msb_raw->nx_omap_oid = cpu_to_le64(bh->b_blocknr);
+	}
+	msb_omap_raw = (struct apfs_omap_phys *)bh->b_data;
+
+	/* Get the root node for the container's omap */
+	vnode = apfs_read_node(sb, le64_to_cpu(msb_omap_raw->om_tree_oid),
+			       APFS_OBJ_PHYSICAL, write);
+	if (IS_ERR(vnode)) {
+		apfs_err(sb, "unable to read volume block");
+		err = PTR_ERR(vnode);
+		goto fail;
+	}
+	if (write) {
+		ASSERT(buffer_trans(bh));
+		msb_omap_raw->om_tree_oid = cpu_to_le64(vnode->object.block_nr);
+	}
+	msb_omap_raw = NULL;
+	brelse(bh);
+	bh = NULL;
+
+	omap = apfs_alloc_omap();
+	if (!omap) {
+		apfs_node_free(vnode);
+		return -ENOMEM;
+	}
+	omap->omap_root = vnode;
+
+	err = apfs_omap_lookup_block(sb, omap, vol_id, &vsb, write);
+	apfs_node_free(vnode);
+	vnode = NULL;
+	kfree(omap);
+	omap = NULL;
+	if (err) {
+		apfs_err(sb, "volume not found, likely corruption");
+		return err;
+	}
+
+	/*
+	 * Snapshots could get mounted during a transaction, so the fletcher
+	 * checksum doesn't have to be valid.
+	 */
+	return apfs_map_volume_super_bno(sb, vsb, !write && !sbi->s_snap_name);
+
+fail:
+	brelse(bh);
+	return err;
+}
+
+/**
+ * apfs_unmap_volume_super - Clean up apfs_map_volume_super()
+ * @sb:	filesystem superblock
+ */
+void apfs_unmap_volume_super(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_object *obj = &sbi->s_vobject;
+
+	obj->data = NULL;
+	brelse(obj->o_bh);
+	obj->o_bh = NULL;
+}
+
+/**
+ * apfs_get_omap - Get a reference to the omap, if it's already read
+ * @sb:	filesystem superblock
+ *
+ * Returns the omap struct, or NULL on failure.
+ */
+static struct apfs_omap *apfs_get_omap(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_sb_info *curr = NULL;
+	struct apfs_omap *omap = NULL;
+	struct apfs_omap_cache *cache = NULL;
+
+	lockdep_assert_held(&nxs_mutex);
+
+	list_for_each_entry(curr, &nxi->vol_list, list) {
+		if (curr == sbi)
+			continue;
+		if (curr->s_vol_nr == sbi->s_vol_nr) {
+			omap = curr->s_omap;
+			if (!omap) {
+				/*
+				 * This volume has already gone through
+				 * apfs_attach_nxi(), but its omap is either
+				 * not yet read or already put.
+				 */
+				continue;
+			}
+			cache = &omap->omap_cache;
+			++omap->omap_refcnt;
+			/* Right now the cache can't be shared like this */
+			cache->disabled = true;
+			return omap;
+		}
+	}
+	return NULL;
+}
+
+/**
+ * apfs_read_omap - Find and read the omap root node
+ * @sb:		superblock structure
+ * @write:	request write access?
+ *
+ * On success, returns 0 and sets the fields of APFS_SB(@sb)->s_omap; on failure
+ * returns a negative error code.
+ */
+int apfs_read_omap(struct super_block *sb, bool write)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_omap_phys *omap_raw;
+	struct apfs_node *omap_root;
+	struct apfs_omap *omap = NULL;
+	struct buffer_head *bh;
+	u64 omap_blk;
+	int err;
+
+	ASSERT(sbi->s_vsb_raw);
+
+	ASSERT(sbi->s_omap);
+	omap = sbi->s_omap;
+
+	/* Get the block holding the volume omap information */
+	omap_blk = le64_to_cpu(vsb_raw->apfs_omap_oid);
+	bh = apfs_read_object_block(sb, omap_blk, write, false /* preserve */);
+	if (IS_ERR(bh)) {
+		apfs_err(sb, "unable to read the volume object map");
+		return PTR_ERR(bh);
+	}
+	if (write) {
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		vsb_raw->apfs_omap_oid = cpu_to_le64(bh->b_blocknr);
+	}
+	omap_raw = (struct apfs_omap_phys *)bh->b_data;
+
+	/* Get the volume's object map */
+	omap_root = apfs_read_node(sb, le64_to_cpu(omap_raw->om_tree_oid),
+				   APFS_OBJ_PHYSICAL, write);
+	if (IS_ERR(omap_root)) {
+		apfs_err(sb, "unable to read the omap root node");
+		err = PTR_ERR(omap_root);
+		goto fail;
+	}
+	if (write) {
+		apfs_assert_in_transaction(sb, &omap_raw->om_o);
+		ASSERT(buffer_trans(bh));
+		omap_raw->om_tree_oid = cpu_to_le64(omap_root->object.block_nr);
+	}
+	omap->omap_latest_snap = le64_to_cpu(omap_raw->om_most_recent_snap);
+	omap_raw = NULL;
+	brelse(bh);
+
+	if (omap->omap_root)
+		apfs_node_free(omap->omap_root);
+	omap->omap_root = omap_root;
+	return 0;
+
+fail:
+	brelse(bh);
+	return err;
+}
+
+/**
+ * apfs_first_read_omap - Find and read the omap root node during mount
+ * @sb:		superblock structure
+ *
+ * On success, returns 0 and sets APFS_SB(@sb)->s_omap; on failure returns a
+ * negative error code.
+ */
+static int apfs_first_read_omap(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = NULL;
+	struct apfs_omap *omap = NULL;
+	int err;
+
+	/*
+	 * For each volume, the first mount that gets here is responsible
+	 * for reading the omap. Other mounts (for other snapshots) just
+	 * go through the container's volume list to retrieve it. This results
+	 * in coarse locking as usual: with some thought it would be possible
+	 * to allow other volumes to read their own omaps at the same time,
+	 * but I don't see the point.
+	 */
+	mutex_lock(&nxs_mutex);
+
+	sbi = APFS_SB(sb);
+
+	/* The current transaction and all snapshots share a single omap */
+	omap = apfs_get_omap(sb);
+	if (omap) {
+		sbi->s_omap = omap;
+		err = 0;
+		goto out;
+	}
+
+	omap = apfs_alloc_omap();
+	if (!omap) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	sbi->s_omap = omap;
+	err = apfs_read_omap(sb, false /* write */);
+	if (err) {
+		kfree(omap);
+		sbi->s_omap = NULL;
+		goto out;
+	}
+
+	++omap->omap_refcnt;
+	err = 0;
+out:
+	mutex_unlock(&nxs_mutex);
+	return err;
+}
+
+/**
+ * apfs_unset_omap - Unset the object map in a superblock
+ * @sb: superblock structure
+ *
+ * Shrinks the omap reference, frees the omap if needed, and sets the field to
+ * NULL atomically in relation to apfs_first_read_omap(). So, no other mount
+ * can grab a new reference halfway through.
+ */
+static void apfs_unset_omap(struct super_block *sb)
+{
+	struct apfs_omap **omap_p = NULL;
+	struct apfs_omap *omap = NULL;
+
+	omap_p = &APFS_SB(sb)->s_omap;
+	omap = *omap_p;
+	if (!omap)
+		return;
+
+	mutex_lock(&nxs_mutex);
+
+	if (--omap->omap_refcnt != 0)
+		goto out;
+
+	apfs_node_free(omap->omap_root);
+	kfree(omap);
+out:
+	*omap_p = NULL;
+	mutex_unlock(&nxs_mutex);
+}
+
+/**
+ * apfs_read_catalog - Find and read the catalog root node
+ * @sb:		superblock structure
+ * @write:	request write access?
+ *
+ * On success, returns 0 and sets APFS_SB(@sb)->s_cat_root; on failure returns
+ * a negative error code.
+ */
+int apfs_read_catalog(struct super_block *sb, bool write)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_superblock *vsb_raw = sbi->s_vsb_raw;
+	struct apfs_node *root_node;
+
+	ASSERT(sbi->s_omap && sbi->s_omap->omap_root);
+
+	root_node = apfs_read_node(sb, le64_to_cpu(vsb_raw->apfs_root_tree_oid),
+				   APFS_OBJ_VIRTUAL, write);
+	if (IS_ERR(root_node)) {
+		apfs_err(sb, "unable to read catalog root node");
+		return PTR_ERR(root_node);
+	}
+
+	if (sbi->s_cat_root)
+		apfs_node_free(sbi->s_cat_root);
+	sbi->s_cat_root = root_node;
+	return 0;
+}
+
+static void apfs_put_super(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nx_transaction *trans = NULL;
+
+	/* Cleanups won't reschedule themselves during unmount */
+	flush_work(&sbi->s_orphan_cleanup_work);
+
+	/* We are about to commit anyway */
+	trans = &APFS_NXI(sb)->nx_transaction;
+	cancel_delayed_work_sync(&trans->t_work);
+
+	/* Stop flushing orphans and update the volume as needed */
+	if (!(sb->s_flags & SB_RDONLY)) {
+		struct apfs_superblock *vsb_raw;
+		struct buffer_head *vsb_bh;
+		int err;
+
+		err = apfs_transaction_start(sb, APFS_TRANS_SYNC);
+		if (err) {
+			apfs_err(sb, "unmount transaction start failed (err:%d)", err);
+			goto fail;
+		}
+		vsb_raw = sbi->s_vsb_raw;
+		vsb_bh = sbi->s_vobject.o_bh;
+
+		apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+		ASSERT(buffer_trans(vsb_bh));
+
+		apfs_update_software_info(sb);
+		vsb_raw->apfs_unmount_time = cpu_to_le64(ktime_get_real_ns());
+		set_buffer_csum(vsb_bh);
+
+		/* Guarantee commit */
+		sbi->s_nxi->nx_transaction.t_state |= APFS_NX_TRANS_FORCE_COMMIT;
+		err = apfs_transaction_commit(sb);
+		if (err) {
+			apfs_err(sb, "unmount transaction commit failed (err:%d)", err);
+			apfs_transaction_abort(sb);
+			goto fail;
+		}
+	}
+
+	/*
+	 * Even if this particular volume/snapshot was read-only, the container
+	 * may have changed and need an update here.
+	 */
+	apfs_make_super_copy(sb);
+
+fail:
+	/*
+	 * This is essentially the cleanup for apfs_fill_super(). It goes here
+	 * because generic_shutdown_super() only calls ->put_super() when the
+	 * root dentry has been set, that is, when apfs_fill_super() succeeded.
+	 * The rest of the mount cleanup is done directly by ->kill_sb().
+	 */
+	iput(sbi->s_private_dir);
+	sbi->s_private_dir = NULL;
+	apfs_node_free(sbi->s_cat_root);
+	sbi->s_cat_root = NULL;
+	apfs_unset_omap(sb);
+	apfs_unmap_volume_super(sb);
+}
+
+static struct kmem_cache *apfs_inode_cachep;
+
+static struct inode *apfs_alloc_inode(struct super_block *sb)
+{
+	struct apfs_inode_info *ai;
+	struct apfs_dstream_info *dstream;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 18, 0) || RHEL_VERSION_GE(9, 1)
+	ai = alloc_inode_sb(sb, apfs_inode_cachep, GFP_KERNEL);
+#else
+	ai = kmem_cache_alloc(apfs_inode_cachep, GFP_KERNEL);
+#endif
+	if (!ai)
+		return NULL;
+	dstream = &ai->i_dstream;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 16, 0) /* iversion came in 4.16 */
+	inode_set_iversion(&ai->vfs_inode, 1);
+#else
+	ai->vfs_inode.i_version = 1;
+#endif
+	dstream->ds_sb = sb;
+	dstream->ds_inode = &ai->vfs_inode;
+	dstream->ds_cached_ext.len = 0;
+	dstream->ds_ext_dirty = false;
+	ai->i_nchildren = 0;
+	INIT_LIST_HEAD(&ai->i_list);
+	ai->i_cleaned = false;
+	return &ai->vfs_inode;
+}
+
+static void apfs_i_callback(struct rcu_head *head)
+{
+	struct inode *inode = container_of(head, struct inode, i_rcu);
+
+	kmem_cache_free(apfs_inode_cachep, APFS_I(inode));
+}
+
+static void apfs_destroy_inode(struct inode *inode)
+{
+	call_rcu(&inode->i_rcu, apfs_i_callback);
+}
+
+static void init_once(void *p)
+{
+	struct apfs_inode_info *ai = (struct apfs_inode_info *)p;
+	struct apfs_dstream_info *dstream = &ai->i_dstream;
+
+	spin_lock_init(&dstream->ds_ext_lock);
+	inode_init_once(&ai->vfs_inode);
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 9, 0)
+#define SLAB_MEM_SPREAD	0
+#endif
+
+static int __init init_inodecache(void)
+{
+	apfs_inode_cachep = kmem_cache_create("apfs_inode_cache",
+					     sizeof(struct apfs_inode_info),
+					     0, (SLAB_RECLAIM_ACCOUNT|
+						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
+					     init_once);
+	if (apfs_inode_cachep == NULL)
+		return -ENOMEM;
+	return 0;
+}
+
+static int apfs_write_inode(struct inode *inode, struct writeback_control *wbc)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_nxsb_info *nxi = APFS_SB(sb)->s_nxi;
+	int err;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_REG);
+	if (err)
+		return err;
+	err = apfs_update_inode(inode, NULL /* new_name */);
+	if (err)
+		goto fail;
+	/* Don't commit yet, or the inode will get flushed again and lock up */
+	nxi->nx_transaction.t_state |= APFS_NX_TRANS_DEFER_COMMIT;
+	err = apfs_transaction_commit(sb);
+	if (err)
+		goto fail;
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+static void destroy_inodecache(void)
+{
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
+	kmem_cache_destroy(apfs_inode_cachep);
+}
+
+static int apfs_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *msb_raw;
+	struct apfs_superblock *vol;
+	u64 fsid, free_blocks;
+	int err;
+
+	down_read(&nxi->nx_big_sem);
+	msb_raw = nxi->nx_raw;
+	vol = sbi->s_vsb_raw;
+
+	buf->f_type = APFS_NX_MAGIC;
+	/* Nodes are assumed to fit in a page, for now */
+	buf->f_bsize = sb->s_blocksize;
+
+	/* Volumes share the whole disk space */
+	buf->f_blocks = le64_to_cpu(msb_raw->nx_block_count);
+	/*
+	 * It takes some work to retrieve the free block count because we
+	 * can't assume that the spaceman has been read yet. It would be
+	 * cleaner if we always did that on first mount (TODO).
+	 */
+	err = apfs_spaceman_get_free_blkcnt(sb, &free_blocks);
+	if (err)
+		goto fail;
+	buf->f_bfree = free_blocks;
+	buf->f_bavail = free_blocks;
+
+	/* The file count is only for the mounted volume */
+	buf->f_files = le64_to_cpu(vol->apfs_num_files) +
+		       le64_to_cpu(vol->apfs_num_directories) +
+		       le64_to_cpu(vol->apfs_num_symlinks) +
+		       le64_to_cpu(vol->apfs_num_other_fsobjects);
+
+	/*
+	 * buf->f_ffree is left undefined for now. Maybe it should report the
+	 * number of available cnids, like hfsplus attempts to do.
+	 */
+
+	buf->f_namelen = APFS_NAME_LEN;
+
+	/* There are no clear rules for the fsid, so we follow ext2 here */
+	fsid = le64_to_cpup((void *)vol->apfs_vol_uuid) ^
+	       le64_to_cpup((void *)vol->apfs_vol_uuid + sizeof(u64));
+	buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
+	buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
+
+fail:
+	up_read(&nxi->nx_big_sem);
+	return err;
+}
+
+static int apfs_show_options(struct seq_file *seq, struct dentry *root)
+{
+	struct apfs_sb_info *sbi = APFS_SB(root->d_sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(root->d_sb);
+
+	if (sbi->s_vol_nr != 0)
+		seq_printf(seq, ",vol=%u", sbi->s_vol_nr);
+	if (sbi->s_snap_name)
+		seq_printf(seq, ",snap=%s", sbi->s_snap_name);
+	if (uid_valid(sbi->s_uid))
+		seq_printf(seq, ",uid=%u", from_kuid(&init_user_ns,
+						     sbi->s_uid));
+	if (gid_valid(sbi->s_gid))
+		seq_printf(seq, ",gid=%u", from_kgid(&init_user_ns,
+						     sbi->s_gid));
+	if (nxi->nx_flags & APFS_CHECK_NODES)
+		seq_puts(seq, ",cknodes");
+	if (nxi->nx_tier2_info)
+		seq_printf(seq, ",tier2=%s", nxi->nx_tier2_info->blki_path);
+
+	return 0;
+}
+
+int apfs_sync_fs(struct super_block *sb, int wait)
+{
+	int err;
+
+	/* TODO: actually start the commit and return without waiting? */
+	if (wait == 0)
+		return 0;
+
+	err = apfs_transaction_start(sb, APFS_TRANS_SYNC);
+	if (err)
+		return err;
+	APFS_SB(sb)->s_nxi->nx_transaction.t_state |= APFS_NX_TRANS_FORCE_COMMIT;
+	err = apfs_transaction_commit(sb);
+	if (err)
+		apfs_transaction_abort(sb);
+	return err;
+}
+
+/* Only supports read-only remounts, everything else is silently ignored */
+static int apfs_remount(struct super_block *sb, int *flags, char *data)
+{
+	int err = 0;
+
+	err = sync_filesystem(sb);
+	if (err)
+		return err;
+
+	/* TODO: race? Could a new transaction have started already? */
+	if (*flags & SB_RDONLY)
+		sb->s_flags |= SB_RDONLY;
+
+	/*
+	 * TODO: readwrite remounts seem simple enough, but I worry about
+	 * remounting aborted transactions. I would probably also need a
+	 * dry-run version of parse_options().
+	 */
+	apfs_notice(sb, "all remounts can do is turn a volume read-only");
+	return 0;
+}
+
+static const struct super_operations apfs_sops = {
+	.alloc_inode	= apfs_alloc_inode,
+	.destroy_inode	= apfs_destroy_inode,
+	.write_inode	= apfs_write_inode,
+	.evict_inode	= apfs_evict_inode,
+	.put_super	= apfs_put_super,
+	.sync_fs	= apfs_sync_fs,
+	.statfs		= apfs_statfs,
+#if LINUX_VERSION_CODE < KERNEL_VERSION(7, 0, 0)
+	.remount_fs	= apfs_remount,
+#endif
+	.show_options	= apfs_show_options,
+};
+
+enum {
+	Opt_readwrite, Opt_cknodes, Opt_uid, Opt_gid, Opt_vol, Opt_snap, Opt_tier2, Opt_err,
+};
+
+static const match_table_t tokens = {
+	{Opt_readwrite, "readwrite"},
+	{Opt_cknodes, "cknodes"},
+	{Opt_uid, "uid=%u"},
+	{Opt_gid, "gid=%u"},
+	{Opt_vol, "vol=%u"},
+	{Opt_snap, "snap=%s"},
+	{Opt_tier2, "tier2=%s"},
+	{Opt_err, NULL}
+};
+
+/**
+ * apfs_set_nx_flags - Set the mount flags for the container, if allowed
+ * @sb:		superblock structure
+ * @flags:	flags to set
+ */
+static void apfs_set_nx_flags(struct super_block *sb, unsigned int flags)
+{
+	struct apfs_nxsb_info *nxi = APFS_SB(sb)->s_nxi;
+
+	mutex_lock(&nxs_mutex);
+
+	/* The first mount thet gets here decides the flags for its container */
+	flags |= APFS_FLAGS_SET;
+	if (!(nxi->nx_flags & APFS_FLAGS_SET))
+		nxi->nx_flags = flags;
+	else if (flags != nxi->nx_flags)
+		apfs_warn(sb, "ignoring mount flags - container already mounted");
+
+	mutex_unlock(&nxs_mutex);
+}
+
+static void parse_options_set_flags(struct super_block *sb, struct apfs_sb_info *sbi,
+		unsigned int nx_flags)
+{
+	struct apfs_nxsb_info *nxi = sbi->s_nxi;
+
+	apfs_set_nx_flags(sb, nx_flags);
+	if (!(sb->s_flags & SB_RDONLY)) {
+		if (nxi->nx_flags & APFS_READWRITE) {
+			apfs_notice(sb, "experimental write support is enabled");
+		} else {
+			apfs_warn(sb, "experimental writes disabled to avoid data loss");
+			apfs_warn(sb, "if you really want them, check the README");
+			sb->s_flags |= SB_RDONLY;
+		}
+	}
+}
+
+/*
+ * Many of the parse_options() functions in other file systems return 0
+ * on error. This one returns an error code, and 0 on success.
+ */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+static int parse_options(struct fs_context *fc, char *options)
+{
+    struct apfs_sb_info *sbi = fc->s_fs_info;
+#else
+static int parse_options(struct super_block *sb, char *options)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+#endif
+	char *p;
+	substring_t args[MAX_OPT_ARGS];
+	int option;
+	int err = 0;
+	unsigned int nx_flags;
+
+	/* Set default values before parsing */
+	nx_flags = 0;
+
+#ifdef CONFIG_APFS_RW_ALWAYS
+	/* Still risky, but some packagers want writable mounts by default */
+	nx_flags |= APFS_READWRITE;
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	sbi->s_uid = INVALID_UID;
+	sbi->s_gid = INVALID_GID;
+#endif
+
+	if (!options)
+		goto out;
+
+	while ((p = strsep(&options, ",")) != NULL) {
+		int token;
+
+		if (!*p)
+			continue;
+		token = match_token(p, tokens, args);
+		switch (token) {
+		case Opt_readwrite:
+			/*
+			 * Write support is not safe yet, so keep it disabled
+			 * unless the user requests it explicitly.
+			 */
+			nx_flags |= APFS_READWRITE;
+			break;
+		case Opt_cknodes:
+			/*
+			 * Right now, node checksums are too costly to enable
+			 * by default.  TODO: try to improve this.
+			 */
+			nx_flags |= APFS_CHECK_NODES;
+			break;
+		case Opt_uid:
+			err = match_int(&args[0], &option);
+			if (err)
+				return err;
+			sbi->s_uid = make_kuid(current_user_ns(), option);
+			if (!uid_valid(sbi->s_uid)) {
+				apfs_err(NULL, "invalid uid");
+				return -EINVAL;
+			}
+			break;
+		case Opt_gid:
+			err = match_int(&args[0], &option);
+			if (err)
+				return err;
+			sbi->s_gid = make_kgid(current_user_ns(), option);
+			if (!gid_valid(sbi->s_gid)) {
+				apfs_err(NULL, "invalid gid");
+				return -EINVAL;
+			}
+			break;
+		case Opt_vol:
+		case Opt_snap:
+		case Opt_tier2:
+			/* Already read early on mount */
+			break;
+		default:
+			/*
+			 * We should have already checked the mount options in
+			 * apfs_preparse_options(), so this is a bug.
+			 */
+			apfs_alert(NULL, "invalid mount option %s", p);
+			return -EINVAL;
+		}
+	}
+
+out:
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	sbi->s_mount_opt = nx_flags;
+#else
+	parse_options_set_flags(sb, sbi, nx_flags);
+#endif
+	return 0;
+}
+
+/**
+ * apfs_check_nx_features - Check for unsupported features in the container
+ * @sb: superblock structure
+ *
+ * Returns -EINVAL if unsupported incompatible features are found, otherwise
+ * returns 0.
+ */
+static int apfs_check_nx_features(struct super_block *sb)
+{
+	struct apfs_nx_superblock *msb_raw = NULL;
+	u64 features;
+	bool fusion;
+
+	msb_raw = APFS_NXI(sb)->nx_raw;
+	if (!msb_raw) {
+		apfs_alert(sb, "feature checks are misplaced");
+		return -EINVAL;
+	}
+
+	features = le64_to_cpu(msb_raw->nx_incompatible_features);
+	if (features & ~APFS_NX_SUPPORTED_INCOMPAT_MASK) {
+		apfs_warn(sb, "unknown incompatible container features (0x%llx)", features);
+		return -EINVAL;
+	}
+
+	fusion = features & APFS_NX_INCOMPAT_FUSION;
+	if (fusion && !APFS_NXI(sb)->nx_tier2_info) {
+		apfs_warn(sb, "fusion drive - please use the \"tier2\" mount option");
+		return -EINVAL;
+	}
+	if (!fusion && APFS_NXI(sb)->nx_tier2_info) {
+		apfs_warn(sb, "not a fusion drive - what's the second disk for?");
+		return -EINVAL;
+	}
+	if (fusion) {
+		if (!sb_rdonly(sb)) {
+			apfs_warn(sb, "writes to fusion drives not yet supported");
+			sb->s_flags |= SB_RDONLY;
+		}
+	}
+
+	features = le64_to_cpu(msb_raw->nx_readonly_compatible_features);
+	if (features & ~APFS_NX_SUPPORTED_ROCOMPAT_MASK) {
+		apfs_warn(sb, "unknown read-only compatible container features (0x%llx)", features);
+		if (!sb_rdonly(sb)) {
+			apfs_warn(sb, "container can't be mounted read-write");
+			return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+/**
+ * apfs_check_vol_features - Check for unsupported features in the volume
+ * @sb: superblock structure
+ *
+ * Returns -EINVAL if unsupported incompatible features are found, otherwise
+ * returns 0.
+ */
+static int apfs_check_vol_features(struct super_block *sb)
+{
+	struct apfs_superblock *vsb_raw = NULL;
+	u64 features;
+
+	vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	if (!vsb_raw) {
+		apfs_alert(sb, "feature checks are misplaced");
+		return -EINVAL;
+	}
+
+	features = le64_to_cpu(vsb_raw->apfs_incompatible_features);
+	if (features & ~APFS_SUPPORTED_INCOMPAT_MASK) {
+		apfs_warn(sb, "unknown incompatible volume features (0x%llx)", features);
+		return -EINVAL;
+	}
+	if (features & APFS_INCOMPAT_DATALESS_SNAPS) {
+		/*
+		 * I haven't encountered dataless snapshots myself yet (TODO).
+		 * I'm not even sure what they are, so be safe.
+		 */
+		if (!sb_rdonly(sb)) {
+			apfs_warn(sb, "writes to volumes with dataless snapshots not yet supported");
+			return -EINVAL;
+		}
+		apfs_warn(sb, "volume has dataless snapshots");
+	}
+	if (features & APFS_INCOMPAT_ENC_ROLLED) {
+		apfs_warn(sb, "encrypted volumes are not supported");
+		return -EINVAL;
+	}
+	if (features & APFS_INCOMPAT_INCOMPLETE_RESTORE) {
+		apfs_warn(sb, "incomplete restore is not supported");
+		return -EINVAL;
+	}
+	if (features & APFS_INCOMPAT_PFK) {
+		apfs_warn(sb, "PFK is not supported");
+		return -EINVAL;
+	}
+	if (features & APFS_INCOMPAT_SECONDARY_FSROOT) {
+		apfs_warn(sb, "secondary fsroot is not supported");
+		return -EINVAL;
+	}
+	if (features & APFS_INCOMPAT_SEALED_VOLUME) {
+		if (!sb_rdonly(sb)) {
+			apfs_warn(sb, "writes to sealed volumes are not yet supported");
+			return -EINVAL;
+		}
+		apfs_info(sb, "volume is sealed");
+	}
+	/*
+	 * As far as I can see, all this feature seems to do is define a new
+	 * flag (which I call APFS_FILE_EXTENT_PREALLOCATED) for extents that
+	 * are fully after the end of their file. I don't get why this change
+	 * is incompatible instead of read-only compatible, so I fear I might
+	 * be missing something. I will never be certain though, so for now
+	 * allow the mount and hope for the best.
+	 */
+	if (features & APFS_INCOMPAT_EXTENT_PREALLOC_FLAG)
+		apfs_warn(sb, "extent prealloc flag is set");
+
+	features = le64_to_cpu(vsb_raw->apfs_fs_flags);
+	/* Some encrypted volumes are readable anyway */
+	if (!(features & APFS_FS_UNENCRYPTED))
+		apfs_warn(sb, "volume is encrypted, may not be read correctly");
+
+	features = le64_to_cpu(vsb_raw->apfs_readonly_compatible_features);
+	if (features & ~APFS_SUPPORTED_ROCOMPAT_MASK) {
+		apfs_warn(sb, "unknown read-only compatible volume features (0x%llx)", features);
+		if (!sb_rdonly(sb)) {
+			apfs_warn(sb, "volume can't be mounted read-write");
+			return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
+
+/**
+ * apfs_setup_bdi - Set up the bdi for the superblock
+ * @sb: superblock structure
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_setup_bdi(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_blkdev_info *bd_info = NULL;
+	struct backing_dev_info *bdi_dev = NULL, *bdi_sb = NULL;
+	int err;
+
+	bd_info = nxi->nx_blkdev_info;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 15, 0) || RHEL_VERSION_GE(9, 0)
+	bdi_dev = bd_info->blki_bdev->bd_disk->bdi;
+#else
+	bdi_dev = bd_info->blki_bdev->bd_bdi;
+#endif
+
+	err = super_setup_bdi(sb);
+	if (err)
+		return err;
+	bdi_sb = sb->s_bdi;
+
+	bdi_sb->ra_pages = bdi_dev->ra_pages;
+	bdi_sb->io_pages = bdi_dev->io_pages;
+
+	bdi_sb->capabilities = bdi_dev->capabilities;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 10, 0)
+	bdi_sb->capabilities &= ~BDI_CAP_WRITEBACK;
+#else
+	bdi_sb->capabilities |= BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY;
+#endif
+
+	return 0;
+}
+
+#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0) */
+
+/* This is needed for readahead, so old kernels will be slower */
+static int apfs_setup_bdi(struct super_block *sb)
+{
+	return 0;
+}
+
+#endif
+
+static void apfs_set_trans_buffer_limit(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	unsigned long memsize_in_blocks;
+	struct sysinfo info = {0};
+
+	si_meminfo(&info);
+	memsize_in_blocks = info.totalram << (PAGE_SHIFT - sb->s_blocksize_bits);
+
+	/*
+	 * Buffer heads are not reclaimed while they are part of the current
+	 * transaction, so systems with little memory will crash if we don't
+	 * commit often enough. This hack should make that happen in general,
+	 * but I still need to get the reclaim to work eventually (TODO).
+	 */
+	if (memsize_in_blocks >= 16 * APFS_TRANS_BUFFERS_MAX)
+		nxi->nx_trans_buffers_max = APFS_TRANS_BUFFERS_MAX;
+	else
+		nxi->nx_trans_buffers_max = memsize_in_blocks / 16;
+}
+
+static int apfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct inode *root = NULL, *priv = NULL;
+	int err;
+
+	/*
+	 * This function doesn't write anything to disk, that happens later
+	 * when an actual transaction begins. So, it's not generally a problem
+	 * if other mounts for the same container fill their own supers at the
+	 * same time (the few critical sections will be protected by
+	 * nxs_mutex), nor is it a problem if other mounted volumes want to
+	 * make reads while the mount is taking place. But we definitely don't
+	 * want any writes, or else we could find ourselves reading stale
+	 * blocks after CoW, among other issues.
+	 */
+	down_read(&APFS_NXI(sb)->nx_big_sem);
+
+	err = apfs_setup_bdi(sb);
+	if (err)
+		goto failed_volume;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	parse_options_set_flags(sb, sbi, sbi->s_mount_opt);
+#else
+	sbi->s_uid = INVALID_UID;
+	sbi->s_gid = INVALID_GID;
+	err = parse_options(sb, data);
+	if (err)
+		goto failed_volume;
+#endif
+	err = apfs_map_volume_super(sb, false /* write */);
+	if (err)
+		goto failed_volume;
+
+	err = apfs_check_vol_features(sb);
+	if (err)
+		goto failed_omap;
+
+	/*
+	 * The omap needs to be set before the call to apfs_read_catalog().
+	 * It's also shared with all the snapshots, so it needs to be read
+	 * before we switch to the old superblock.
+	 */
+	err = apfs_first_read_omap(sb);
+	if (err)
+		goto failed_omap;
+
+	if (sbi->s_snap_name) {
+		err = apfs_switch_to_snapshot(sb);
+		if (err)
+			goto failed_cat;
+	}
+
+	err = apfs_read_catalog(sb, false /* write */);
+	if (err)
+		goto failed_cat;
+
+	sb->s_op = &apfs_sops;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 17, 0)
+	set_default_d_op(sb, &apfs_dentry_operations);
+#else
+	sb->s_d_op = &apfs_dentry_operations;
+#endif
+	sb->s_xattr = apfs_xattr_handlers;
+	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_time_gran = 1; /* Nanosecond granularity */
+
+	/*
+	 * At this point everything is already set up for the inode reads,
+	 * which take care of their own locking as always.
+	 */
+	up_read(&APFS_NXI(sb)->nx_big_sem);
+
+	sbi->s_private_dir = apfs_iget(sb, APFS_PRIV_DIR_INO_NUM);
+	if (IS_ERR(sbi->s_private_dir)) {
+		apfs_err(sb, "unable to get private-dir inode");
+		err = PTR_ERR(sbi->s_private_dir);
+		goto failed_private_dir;
+	}
+
+	root = apfs_iget(sb, APFS_ROOT_DIR_INO_NUM);
+	if (IS_ERR(root)) {
+		apfs_err(sb, "unable to get root inode");
+		err = PTR_ERR(root);
+		goto failed_mount;
+	}
+
+	sb->s_root = d_make_root(root);
+	if (!sb->s_root) {
+		apfs_err(sb, "unable to get root dentry");
+		err = -ENOMEM;
+		goto failed_mount;
+	}
+
+	INIT_WORK(&sbi->s_orphan_cleanup_work, apfs_orphan_cleanup_work);
+	if (!(sb->s_flags & SB_RDONLY)) {
+		priv = sbi->s_private_dir;
+		if (APFS_I(priv)->i_nchildren)
+			schedule_work(&sbi->s_orphan_cleanup_work);
+	}
+	return 0;
+
+failed_mount:
+	iput(sbi->s_private_dir);
+failed_private_dir:
+	sbi->s_private_dir = NULL;
+	down_read(&APFS_NXI(sb)->nx_big_sem);
+	apfs_node_free(sbi->s_cat_root);
+failed_cat:
+	apfs_unset_omap(sb);
+failed_omap:
+	apfs_unmap_volume_super(sb);
+failed_volume:
+	up_read(&APFS_NXI(sb)->nx_big_sem);
+	return err;
+}
+
+/**
+ * apfs_strings_are_equal - Compare two possible NULL strings
+ * @str1: the first string
+ * @str2: the second string
+ */
+static bool apfs_strings_are_equal(const char *str1, const char *str2)
+{
+	if (str1 == str2) /* Both are NULL */
+		return true;
+	if (!str1 || !str2) /* One is NULL */
+		return false;
+	return strcmp(str1, str2) == 0;
+}
+
+/**
+ * apfs_test_super - Check if two volume superblocks are for the same volume
+ * @sb:		superblock structure for a currently mounted volume
+ * @data:	superblock info for the volume being mounted
+ */
+static int apfs_test_super(struct super_block *sb, void *data)
+{
+	struct apfs_sb_info *sbi_1 = data;
+	struct apfs_sb_info *sbi_2 = APFS_SB(sb);
+
+	if (sbi_1->s_nxi != sbi_2->s_nxi)
+		return false;
+	if (sbi_1->s_vol_nr != sbi_2->s_vol_nr)
+		return false;
+	return apfs_strings_are_equal(sbi_1->s_snap_name, sbi_2->s_snap_name);
+}
+
+/**
+ * apfs_set_super - Assign the device and an info struct to a superblock
+ * @sb:		superblock structure to set
+ * @data:	superblock info for the volume being mounted
+ */
+static int apfs_set_super(struct super_block *sb, void *data)
+{
+	struct apfs_sb_info *sbi = data;
+	struct apfs_nxsb_info *nxi = sbi->s_nxi;
+	int err;
+
+	/*
+	 * This fake device number will be unique to this volume-snapshot
+	 * combination. It gets reported by stat(), so that userland tools can
+	 * use it to tell different mountpoints apart.
+	 */
+	err = get_anon_bdev(&sbi->s_anon_dev);
+	if (err)
+		return err;
+
+	/*
+	 * This is the actual device number, shared by all volumes and
+	 * snapshots. It gets reported by the mountinfo file, and it seems that
+	 * udisks uses it to decide if a device is mounted, so it must be set.
+	 *
+	 * TODO: does this work for fusion drives?
+	 */
+	sb->s_dev = nxi->nx_blkdev_info->blki_bdev->bd_dev;
+
+	sb->s_fs_info = sbi;
+	return 0;
+}
+
+/*
+ * Wrapper for lookup_bdev() that supports older kernels.
+ */
+static int apfs_lookup_bdev(const char *pathname, dev_t *dev)
+{
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 11, 0)
+	struct block_device *bdev;
+
+	bdev = lookup_bdev(pathname);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
+
+	*dev = bdev->bd_dev;
+	bdput(bdev);
+	return 0;
+#else
+	return lookup_bdev(pathname, dev);
+#endif
+}
+
+/**
+ * apfs_blkdev_setup - Open a block device and set its info struct
+ * @info_p:	info struct to set
+ * @dev_name:	path name for the block device to open
+ * @mode:	FMODE_* mask
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 5, 0)
+static int apfs_blkdev_setup(struct apfs_blkdev_info **info_p, const char *dev_name, blk_mode_t mode)
+#else
+static int apfs_blkdev_setup(struct apfs_blkdev_info **info_p, const char *dev_name, fmode_t mode)
+#endif
+{
+	struct apfs_blkdev_info *info = NULL;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 9, 0) || RHEL_VERSION_GE(9, 5)
+	struct file *file = NULL;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+	struct bdev_handle *handle = NULL;
+#endif
+	struct block_device *bdev = NULL;
+	int ret;
+
+	info = kzalloc(sizeof(*info), GFP_KERNEL);
+	if (!info)
+		return -ENOMEM;
+	info->blki_path = kstrdup(dev_name, GFP_KERNEL);
+	if (!info->blki_path) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 9, 0) || RHEL_VERSION_GE(9, 5)
+	file = bdev_file_open_by_path(dev_name, mode, &apfs_fs_type, NULL);
+	if (IS_ERR(file)) {
+		ret = PTR_ERR(file);
+		goto fail;
+	}
+	info->blki_bdev_file = file;
+	bdev = file_bdev(file);
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
+	handle = bdev_open_by_path(dev_name, mode, &apfs_fs_type, NULL);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		goto fail;
+	}
+	info->blki_bdev_handle = handle;
+	bdev = handle->bdev;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 5, 0) || RHEL_VERSION_GE(9, 4)
+	bdev = blkdev_get_by_path(dev_name, mode, &apfs_fs_type, NULL);
+#else
+	bdev = blkdev_get_by_path(dev_name, mode, &apfs_fs_type);
+#endif
+
+	if (IS_ERR(bdev)) {
+		ret = PTR_ERR(bdev);
+		goto fail;
+	}
+	info->blki_bdev = bdev;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 5, 0) && !RHEL_VERSION_GE(9, 4)
+	info->blki_mode = mode;
+#endif
+	*info_p = info;
+	return 0;
+
+fail:
+	kfree(info->blki_path);
+	info->blki_path = NULL;
+	kfree(info);
+	info = NULL;
+	return ret;
+}
+
+/**
+ * apfs_attach_nxi - Attach container sb info to a volume's sb info
+ * @sbi:	new superblock info structure for the volume to be mounted
+ * @dev_name:	path name for the container's block device
+ * @mode:	FMODE_* mask
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 5, 0)
+static int apfs_attach_nxi(struct apfs_sb_info *sbi, const char *dev_name, blk_mode_t mode)
+#else
+static int apfs_attach_nxi(struct apfs_sb_info *sbi, const char *dev_name, fmode_t mode)
+#endif
+{
+	struct apfs_nxsb_info *nxi = NULL;
+	dev_t dev = 0;
+	int ret;
+
+	mutex_lock(&nxs_mutex);
+
+	ret = apfs_lookup_bdev(dev_name, &dev);
+	if (ret)
+		goto out;
+
+	nxi = apfs_nx_find_by_dev(dev);
+	if (!nxi) {
+		nxi = kzalloc(sizeof(*nxi), GFP_KERNEL);
+		if (!nxi) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ret = apfs_blkdev_setup(&nxi->nx_blkdev_info, dev_name, mode);
+		if (ret)
+			goto out;
+
+		if (sbi->s_tier2_path) {
+			ret = apfs_blkdev_setup(&nxi->nx_tier2_info, sbi->s_tier2_path, mode);
+			if (ret) {
+				apfs_blkdev_cleanup(nxi->nx_blkdev_info);
+				nxi->nx_blkdev_info = NULL;
+				goto out;
+			}
+			/* We won't need this anymore, so why waste memory? */
+			kfree(sbi->s_tier2_path);
+			sbi->s_tier2_path = NULL;
+		}
+
+		init_rwsem(&nxi->nx_big_sem);
+		list_add(&nxi->nx_list, &nxs);
+		INIT_LIST_HEAD(&nxi->vol_list);
+		apfs_transaction_init(&nxi->nx_transaction);
+	}
+
+	list_add(&sbi->list, &nxi->vol_list);
+	sbi->s_nxi = nxi;
+	++nxi->nx_refcnt;
+	ret = 0;
+out:
+	if (ret) {
+		kfree(nxi);
+		nxi = NULL;
+	}
+	mutex_unlock(&nxs_mutex);
+	return ret;
+}
+
+/**
+ * apfs_preparse_options - Parse the options used to identify a superblock
+ * @sbi:	superblock info
+ * @options:	options string to parse
+ *
+ * Returns 0 on success, or a negative error code in case of failure. Even on
+ * failure, the caller is responsible for freeing all superblock fields.
+ */
+static int apfs_preparse_options(struct apfs_sb_info *sbi, char *options)
+{
+	char *tofree = NULL;
+	char *p;
+	substring_t args[MAX_OPT_ARGS];
+	int err = 0;
+
+	/* Set default values before parsing */
+	sbi->s_vol_nr = 0;
+	sbi->s_snap_name = NULL;
+	sbi->s_tier2_path = NULL;
+
+	if (!options)
+		return 0;
+
+	/* Later parse_options() will need the unmodified options string */
+	options = kstrdup(options, GFP_KERNEL);
+	if (!options)
+		return -ENOMEM;
+	tofree = options;
+
+	while ((p = strsep(&options, ",")) != NULL) {
+		int token;
+
+		if (!*p)
+			continue;
+		token = match_token(p, tokens, args);
+		switch (token) {
+		case Opt_vol:
+			err = match_int(&args[0], &sbi->s_vol_nr);
+			if (err)
+				goto out;
+			break;
+		case Opt_snap:
+			kfree(sbi->s_snap_name);
+			sbi->s_snap_name = match_strdup(&args[0]);
+			if (!sbi->s_snap_name) {
+				err = -ENOMEM;
+				goto out;
+			}
+			break;
+		case Opt_tier2:
+			kfree(sbi->s_tier2_path);
+			sbi->s_tier2_path = match_strdup(&args[0]);
+			if (!sbi->s_tier2_path) {
+				err = -ENOMEM;
+				goto out;
+			}
+			break;
+		case Opt_readwrite:
+		case Opt_cknodes:
+		case Opt_uid:
+		case Opt_gid:
+			/* Not needed for sget(), will be read later */
+			break;
+		default:
+			apfs_warn(NULL, "invalid mount option %s", p);
+			err = -EINVAL;
+			goto out;
+		}
+	}
+	err = 0;
+out:
+	kfree(tofree);
+	return err;
+}
+
+/*
+ * This function is a copy of mount_bdev() that allows multiple mounts.
+ */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+static int apfs_test_super_fc(struct super_block *sb, struct fs_context *fc);
+static int apfs_set_super_fc(struct super_block *sb, struct fs_context *fc);
+static int apfs_get_tree(struct fs_context *fc)
+#else
+static struct dentry *apfs_mount(struct file_system_type *fs_type, int flags,
+				 const char *dev_name, void *data)
+#endif
+{
+	struct super_block *sb;
+	struct apfs_sb_info *sbi;
+	struct apfs_blkdev_info *bd_info = NULL, *tier2_info = NULL;
+	int error = 0;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	blk_mode_t mode = sb_open_mode(fc->sb_flags);
+	void *data = fc->fs_private;
+	const char *dev_name = fc->source;
+
+	sbi = fc->s_fs_info;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(6, 5, 0) || RHEL_VERSION_GE(9, 4)
+	blk_mode_t mode = sb_open_mode(flags);
+#else
+	fmode_t mode = FMODE_READ | FMODE_EXCL;
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(7, 0, 0)
+	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
+	if (!sbi)
+		return ERR_PTR(-ENOMEM);
+
+	/* Set up the fields that sget() will need to id the superblock */
+	error = apfs_preparse_options(sbi, data);
+	if (error)
+		goto out_free_sbi;
+#endif
+
+	/* Make sure that snapshots are mounted read-only */
+	if (sbi->s_snap_name)
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+		fc->sb_flags |= SB_RDONLY;
+#else
+		flags |= SB_RDONLY;
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 5, 0) && !RHEL_VERSION_GE(9, 4)
+	if (!(flags & SB_RDONLY))
+		mode |= FMODE_WRITE;
+#endif
+
+	error = apfs_attach_nxi(sbi, dev_name, mode);
+	if (error)
+		goto out_free_sbi;
+
+	/* TODO: lockfs stuff? Btrfs doesn't seem to care */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	fc->sb_flags |= SB_NOSEC;
+	sb = sget_fc(fc, apfs_test_super_fc, apfs_set_super_fc);
+#else
+	sb = sget(fs_type, apfs_test_super, apfs_set_super, flags | SB_NOSEC, sbi);
+#endif
+	if (IS_ERR(sb)) {
+		error = PTR_ERR(sb);
+		goto out_unmap_super;
+	}
+
+	bd_info = APFS_NXI(sb)->nx_blkdev_info;
+	tier2_info = APFS_NXI(sb)->nx_tier2_info;
+
+	/*
+	 * I'm doing something hacky with s_dev inside ->kill_sb(), so I want
+	 * to find out as soon as possible if I messed it up.
+	 */
+	WARN_ON(sb->s_dev != bd_info->blki_bdev->bd_dev);
+
+	if (sb->s_root) {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+		if ((fc->sb_flags ^ sb->s_flags) & SB_RDONLY) {
+#else
+		if ((flags ^ sb->s_flags) & SB_RDONLY) {
+#endif
+			error = -EBUSY;
+			deactivate_locked_super(sb);
+			goto out_unmap_super;
+		}
+		/* Only one superblock per volume */
+		apfs_free_main_super(sbi);
+		kfree(sbi->s_snap_name);
+		sbi->s_snap_name = NULL;
+		kfree(sbi->s_tier2_path);
+		sbi->s_tier2_path = NULL;
+		kfree(sbi);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+		fc->s_fs_info = NULL;
+#endif
+		sbi = NULL;
+	} else {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+		fc->s_fs_info = NULL;
+#endif
+		if (!sbi->s_snap_name && !tier2_info)
+			snprintf(sb->s_id, sizeof(sb->s_id), "%pg:%u", bd_info->blki_bdev, sbi->s_vol_nr);
+		else if (!tier2_info)
+			snprintf(sb->s_id, sizeof(sb->s_id), "%pg:%u:%s", bd_info->blki_bdev, sbi->s_vol_nr, sbi->s_snap_name);
+		else if (!sbi->s_snap_name)
+			snprintf(sb->s_id, sizeof(sb->s_id), "%pg+%pg:%u", bd_info->blki_bdev, tier2_info->blki_bdev, sbi->s_vol_nr);
+		else
+			snprintf(sb->s_id, sizeof(sb->s_id), "%pg+%pg:%u:%s", bd_info->blki_bdev, tier2_info->blki_bdev, sbi->s_vol_nr, sbi->s_snap_name);
+		error = apfs_read_main_super(sb);
+		if (error) {
+			deactivate_locked_super(sb);
+			goto out_return_error;
+		}
+#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 5, 0) && !RHEL_VERSION_GE(9, 4)
+		sb->s_mode = mode;
+#endif
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+		error = apfs_fill_super(sb, data, fc->sb_flags & SB_SILENT ? 1 : 0);
+#else
+		error = apfs_fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
+#endif
+		if (error) {
+			deactivate_locked_super(sb);
+			goto out_return_error;
+		}
+		sb->s_flags |= SB_ACTIVE;
+	}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	fc->root = dget(sb->s_root);
+	return 0;
+#else
+	return dget(sb->s_root);
+#endif
+
+out_unmap_super:
+#if LINUX_VERSION_CODE < KERNEL_VERSION(7, 0, 0)
+	apfs_free_main_super(sbi);
+#endif
+out_free_sbi:
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	return error;
+#else
+	kfree(sbi->s_snap_name);
+	kfree(sbi->s_tier2_path);
+	kfree(sbi);
+#endif
+out_return_error:
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+	return error;
+#else
+	return ERR_PTR(error);
+#endif
+}
+
+/**
+ * apfs_free_sb_info - Free the sb info and release all remaining fields
+ * @sb: superblock structure
+ *
+ * This function does not include the cleanup for apfs_fill_super(), which
+ * already took place inside ->put_super() (or maybe inside apfs_fill_super()
+ * itself if the mount failed).
+ */
+static void apfs_free_sb_info(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = NULL;
+
+	sbi = APFS_SB(sb);
+	apfs_free_main_super(sbi);
+	sb->s_fs_info = NULL;
+
+	kfree(sbi->s_snap_name);
+	sbi->s_snap_name = NULL;
+	kfree(sbi->s_tier2_path);
+	sbi->s_tier2_path = NULL;
+	if (sbi->s_dflt_pfk)
+		kfree(sbi->s_dflt_pfk);
+	kfree(sbi);
+	sbi = NULL;
+}
+
+static void apfs_kill_sb(struct super_block *sb)
+{
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+
+	/*
+	 * We need to delist the superblock before freeing its info to avoid a
+	 * race with apfs_test_super(), but we can't call kill_super_notify()
+	 * from the driver. The available wrapper is kill_anon_super(), but our
+	 * s_dev is set to the actual device (that gets freed later along with
+	 * the container), not to the anon device that we keep on the sbi. So,
+	 * we change that before the call; this is safe because other mounters
+	 * won't revive this super, even if apfs_test_super() succeeds.
+	 */
+	sb->s_dev = sbi->s_anon_dev;
+	kill_anon_super(sb);
+
+	apfs_free_sb_info(sb);
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(7, 0, 0)
+static int apfs_test_super_fc(struct super_block *sb, struct fs_context *fc)
+{
+	return apfs_test_super(sb, fc->s_fs_info);
+}
+
+static int apfs_set_super_fc(struct super_block *sb, struct fs_context *fc)
+{
+	return apfs_set_super(sb, fc->s_fs_info);
+}
+
+static int apfs_reconfigure(struct fs_context *fc)
+{
+	struct super_block *sb = fc->root->d_sb;
+	int flags = fc->sb_flags;
+
+	return apfs_remount(sb, &flags, fc->fs_private);
+}
+
+static void apfs_free_fc(struct fs_context *fc)
+{
+	struct apfs_sb_info *sbi = fc->s_fs_info;
+
+	if (sbi) {
+		apfs_free_main_super(sbi);
+		kfree(sbi->s_snap_name);
+		kfree(sbi->s_tier2_path);
+		kfree(sbi);
+	}
+}
+
+static int apfs_parse_monolithic(struct fs_context *fc, void *data)
+{
+	struct apfs_sb_info *sbi = fc->s_fs_info;
+	int result = apfs_preparse_options(sbi, data);
+
+	if (result)
+		return result;
+	return parse_options(fc, data);
+}
+
+static const struct fs_context_operations apfs_context_ops = {
+	.get_tree    = apfs_get_tree,
+	.reconfigure = apfs_reconfigure,
+	.free        = apfs_free_fc,
+	.parse_monolithic = apfs_parse_monolithic,
+};
+
+static int apfs_init_fs_context(struct fs_context *fc)
+{
+	struct apfs_sb_info *sbi;
+
+	/* Setup happens here now instead of the top of apfs_mount */
+	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
+	if (!sbi)
+		return -ENOMEM;
+
+	fc->s_fs_info = sbi;
+	fc->ops = &apfs_context_ops;
+	return 0;
+}
+#endif /* LINUX_VERSION_CODE >= 7.0.0 */
+
+static struct file_system_type apfs_fs_type = {
+	.owner		= THIS_MODULE,
+	.name		= "apfs",
+#if LINUX_VERSION_CODE < KERNEL_VERSION(7, 0, 0)
+	.mount		= apfs_mount,
+#else
+	.init_fs_context = apfs_init_fs_context,
+#endif
+	.kill_sb	= apfs_kill_sb,
+	.fs_flags	= FS_REQUIRES_DEV,
+};
+MODULE_ALIAS_FS("apfs");
+
+static int __init init_apfs_fs(void)
+{
+	int err = 0;
+
+	err = init_inodecache();
+	if (err)
+		return err;
+	err = register_filesystem(&apfs_fs_type);
+	if (err)
+		destroy_inodecache();
+	return err;
+}
+
+static void __exit exit_apfs_fs(void)
+{
+	unregister_filesystem(&apfs_fs_type);
+	destroy_inodecache();
+}
+
+MODULE_AUTHOR("Ernesto A. Fernández");
+MODULE_DESCRIPTION("Apple File System");
+MODULE_VERSION(GIT_COMMIT);
+MODULE_LICENSE("GPL");
+module_init(init_apfs_fs)
+module_exit(exit_apfs_fs)
diff --git a/fs/apfs/symlink.c b/fs/apfs/symlink.c
new file mode 100644
index 000000000..be4f9df8f
--- /dev/null
+++ b/fs/apfs/symlink.c
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include "apfs.h"
+
+/**
+ * apfs_get_link - Follow a symbolic link
+ * @dentry:	dentry for the link
+ * @inode:	inode for the link
+ * @done:	delayed call to free the returned buffer after use
+ *
+ * Returns a pointer to a buffer containing the target path, or an appropriate
+ * error pointer in case of failure.
+ */
+static const char *apfs_get_link(struct dentry *dentry, struct inode *inode,
+				 struct delayed_call *done)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	char *target = NULL;
+	int err;
+	int size;
+
+	down_read(&nxi->nx_big_sem);
+
+	if (!dentry) {
+		err = -ECHILD;
+		goto fail;
+	}
+
+	size = __apfs_xattr_get(inode, APFS_XATTR_NAME_SYMLINK,
+				NULL /* buffer */, 0 /* size */);
+	if (size < 0) { /* TODO: return a better error code */
+		apfs_err(sb, "symlink size read failed");
+		err = size;
+		goto fail;
+	}
+
+	target = kmalloc(size, GFP_KERNEL);
+	if (!target) {
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	size = __apfs_xattr_get(inode, APFS_XATTR_NAME_SYMLINK, target, size);
+	if (size < 0) {
+		apfs_err(sb, "symlink read failed");
+		err = size;
+		goto fail;
+	}
+	if (size == 0 || *(target + size - 1) != 0) {
+		/* Target path must be NULL-terminated */
+		apfs_err(sb, "bad link target in inode 0x%llx", apfs_ino(inode));
+		err = -EFSCORRUPTED;
+		goto fail;
+	}
+
+	up_read(&nxi->nx_big_sem);
+	set_delayed_call(done, kfree_link, target);
+	return target;
+
+fail:
+	kfree(target);
+	up_read(&nxi->nx_big_sem);
+	return ERR_PTR(err);
+}
+
+const struct inode_operations apfs_symlink_inode_operations = {
+	.get_link	= apfs_get_link,
+	.getattr	= apfs_getattr,
+	.listxattr	= apfs_listxattr,
+	.update_time	= apfs_update_time,
+#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0) /* Now this is the default */
+	.readlink	= generic_readlink,
+#endif
+};
diff --git a/fs/apfs/transaction.c b/fs/apfs/transaction.c
new file mode 100644
index 000000000..e743ccb0b
--- /dev/null
+++ b/fs/apfs/transaction.c
@@ -0,0 +1,1034 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/blkdev.h>
+#include <linux/rmap.h>
+#include "apfs.h"
+
+/**
+ * apfs_checkpoint_end - End the new checkpoint
+ * @sb:	filesystem superblock
+ *
+ * Flushes all changes to disk, and commits the new checkpoint by setting the
+ * fletcher checksum on its superblock.  Returns 0 on success, or a negative
+ * error code in case of failure.
+ */
+static int apfs_checkpoint_end(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_obj_phys *obj = &nxi->nx_raw->nx_o;
+	struct buffer_head *bh = NULL;
+	struct apfs_blkdev_info *bd_info = nxi->nx_blkdev_info;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 10, 0)
+	struct address_space *bdev_map = bd_info->blki_bdev->bd_mapping;
+#else
+	struct inode *bdev_inode = bd_info->blki_bdev->bd_inode;
+	struct address_space *bdev_map = bdev_inode->i_mapping;
+#endif
+	int err;
+
+	ASSERT(!(sb->s_flags & SB_RDONLY));
+
+	bh = apfs_getblk(sb, nxi->nx_bno);
+	if (!bh) {
+		apfs_err(sb, "failed to map new checkpoint superblock");
+		return -EIO;
+	}
+	obj->o_xid = cpu_to_le64(nxi->nx_xid);
+	apfs_obj_set_csum(sb, obj);
+	memcpy(bh->b_data, obj, sb->s_blocksize);
+
+	err = filemap_write_and_wait(bdev_map);
+	if (err)
+		goto out;
+
+	mark_buffer_dirty(bh);
+	err = sync_dirty_buffer(bh);
+	if (err)
+		goto out;
+
+	err = filemap_write_and_wait(bdev_map);
+out:
+	brelse(bh);
+	bh = NULL;
+	return err;
+}
+
+/**
+ * apfs_transaction_has_room - Is there enough free space for this transaction?
+ * @sb:		superblock structure
+ * @kind:	transaction kind for space preallocation
+ */
+static bool apfs_transaction_has_room(struct super_block *sb, enum apfs_trans_kind kind)
+{
+	struct apfs_nx_transaction *trans = NULL;
+	struct apfs_spaceman *sm = NULL;
+	u64 max_blks;
+
+	trans = &APFS_NXI(sb)->nx_transaction;
+	sm = APFS_SM(sb);
+
+	/*
+	 * It's hard to keep track of the maximum number of blocks that each
+	 * operation could need because of the complexities of apfs btrees,
+	 * plus snapshots and clones. I try to keep things simple here.
+	 */
+	switch (kind) {
+	case APFS_TRANS_REG:
+		/*
+		 * For transactions that are expected to reduce free space, we
+		 * use a very coarse bound (512 KiB) that is certain to be much
+		 * more than enough, and will always leave room for deletions.
+		 */
+		max_blks = APFS_REG_ROOM;
+		break;
+	case APFS_TRANS_DEL:
+		/*
+		 * For transactions that are likely to increase free space, we
+		 * use a much tighter bound (80 KiB), so that users have the
+		 * opportunity to fix their ENOSPC situation.
+		 *
+		 * For huge filesystems with huge numbers of records, there is
+		 * a tiny chance that this might be too permissive, in which
+		 * case the transaction will later abort. I think that's
+		 * acceptable.
+		 */
+		max_blks = APFS_DEL_ROOM;
+		break;
+	case APFS_TRANS_SYNC:
+		/*
+		 * We try to allow sync as much as possible, for the user's
+		 * peace of mind and because flushing the free queues could
+		 * make some room.
+		 *
+		 * Even if we do nothing the transaction still allocates a new
+		 * volume superblock, and new roots for the omap and catalog,
+		 * which consumes 6 blocks in total. This could be avoided...
+		 */
+		if (trans->t_starts_count == 0)
+			max_blks = APFS_SYNC_ROOM;
+		else
+			max_blks = 0;
+		break;
+	default:
+		apfs_alert(sb, "invalid transaction kind %u - bug!", kind);
+		return false;
+	}
+
+	return max_blks <= sm->sm_free_count;
+}
+
+/**
+ * apfs_read_single_ephemeral_object - Read a single ephemeral object to memory
+ * @sb:		filesystem superblock
+ * @map:	checkpoint mapping for the object
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_read_single_ephemeral_object(struct super_block *sb, struct apfs_checkpoint_mapping *map)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *raw_sb = NULL;
+	struct apfs_ephemeral_object_info *list = NULL;
+	struct buffer_head *bh = NULL;
+	char *object = NULL;
+	int count;
+	u32 data_blks, size;
+	u64 data_base, bno, oid;
+	int err, i, data_idx;
+
+	raw_sb = nxi->nx_raw;
+	data_base = le64_to_cpu(raw_sb->nx_xp_data_base);
+	data_blks = le32_to_cpu(raw_sb->nx_xp_data_blocks);
+
+	list = nxi->nx_eph_list;
+	count = nxi->nx_eph_count;
+	if (count >= APFS_EPHEMERAL_LIST_LIMIT) {
+		apfs_err(sb, "too many ephemeral objects?");
+		return -EOPNOTSUPP;
+	}
+
+	bno = le64_to_cpu(map->cpm_paddr);
+	oid = le64_to_cpu(map->cpm_oid);
+	size = le32_to_cpu(map->cpm_size);
+	if (size > sb->s_blocksize << 1) {
+		/*
+		 * No reason not to support bigger objects, but there has to be
+		 * a limit somewhere and this is all I've seen so far.
+		 */
+		apfs_warn(sb, "ephemeral object has more than 2 blocks");
+		return -EOPNOTSUPP;
+	}
+	if (!size || (size & (sb->s_blocksize - 1))) {
+		apfs_err(sb, "invalid object size (0x%x)", size);
+		return -EFSCORRUPTED;
+	}
+	object = kmalloc(size, GFP_KERNEL);
+	if (!object)
+		return -ENOMEM;
+
+	data_idx = bno - data_base;
+	for (i = 0; i < size >> sb->s_blocksize_bits; ++i) {
+		bh = apfs_sb_bread(sb, data_base + data_idx);
+		if (!bh) {
+			apfs_err(sb, "failed to read ephemeral block");
+			err = -EIO;
+			goto fail;
+		}
+		memcpy(object + (i << sb->s_blocksize_bits), bh->b_data, sb->s_blocksize);
+		brelse(bh);
+		bh = NULL;
+		/* Somewhat surprisingly, objects can wrap around */
+		if (++data_idx == data_blks)
+			data_idx = 0;
+	}
+
+	/*
+	 * The official reference requires that we always verify ephemeral
+	 * checksums on mount, so do it even if the user didn't ask. We should
+	 * actually try to mount an older checkpoint when this fails (TODO),
+	 * which I guess means that the official driver writes all checkpoint
+	 * blocks at once, instead of leaving the superblock for last like we
+	 * do.
+	 */
+	if (!apfs_multiblock_verify_csum(object, size)) {
+		apfs_err(sb, "bad checksum for ephemeral object 0x%llx", oid);
+		err = -EFSBADCRC;
+		goto fail;
+	}
+
+	list[count].oid = oid;
+	list[count].size = size;
+	list[count].object = object;
+	object = NULL;
+	nxi->nx_eph_count = count + 1;
+	return 0;
+
+fail:
+	kfree(object);
+	object = NULL;
+	return err;
+}
+
+/**
+ * apfs_read_single_cpm_block - Read all ephemeral objects in a cpm block
+ * @sb:		filesystem superblock
+ * @cpm_bno:	block number for the cpm block
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_read_single_cpm_block(struct super_block *sb, u64 cpm_bno)
+{
+	struct buffer_head *bh = NULL;
+	struct apfs_checkpoint_map_phys *cpm = NULL;
+	u32 map_count;
+	int err, i;
+
+	bh = apfs_sb_bread(sb, cpm_bno);
+	if (!bh) {
+		apfs_err(sb, "failed to read cpm block");
+		return -EIO;
+	}
+	if (!apfs_obj_verify_csum(sb, bh)) {
+		/*
+		 * The reference seems to imply that we need to check these on
+		 * mount, and retry an older checkpoint on failure (TODO).
+		 */
+		apfs_err(sb, "bad checksum for cpm block at 0x%llx", cpm_bno);
+		err = -EFSBADCRC;
+		goto out;
+	}
+	cpm = (struct apfs_checkpoint_map_phys *)bh->b_data;
+
+	map_count = le32_to_cpu(cpm->cpm_count);
+	if (map_count > apfs_max_maps_per_block(sb)) {
+		apfs_err(sb, "block has too many maps (%d)", map_count);
+		err = -EFSCORRUPTED;
+		goto out;
+	}
+
+	for (i = 0; i < map_count; ++i) {
+		err = apfs_read_single_ephemeral_object(sb, &cpm->cpm_map[i]);
+		if (err) {
+			apfs_err(sb, "failed to read ephemeral object %u", i);
+			goto out;
+		}
+	}
+
+out:
+	brelse(bh);
+	cpm = NULL;
+	return err;
+}
+
+static void apfs_force_readonly(struct apfs_nxsb_info *nxi);
+
+/**
+ * apfs_read_ephemeral_objects - Read all ephemeral objects to memory
+ * @sb:	superblock structure
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_read_ephemeral_objects(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *raw_sb = nxi->nx_raw;
+	u64 desc_base;
+	u32 desc_index, desc_blks, desc_len, i;
+	int err;
+
+	if (nxi->nx_eph_list) {
+		apfs_alert(sb, "attempt to reread ephemeral object list");
+		return -EFSCORRUPTED;
+	}
+	nxi->nx_eph_list = kzalloc(APFS_EPHEMERAL_LIST_SIZE, GFP_KERNEL);
+	if (!nxi->nx_eph_list)
+		return -ENOMEM;
+	nxi->nx_eph_count = 0;
+
+	desc_base = le64_to_cpu(raw_sb->nx_xp_desc_base);
+	desc_index = le32_to_cpu(raw_sb->nx_xp_desc_index);
+	desc_blks = le32_to_cpu(raw_sb->nx_xp_desc_blocks);
+	desc_len = le32_to_cpu(raw_sb->nx_xp_desc_len);
+
+	/* Last block in the area is superblock; the rest are mapping blocks */
+	for (i = 0; i < desc_len - 1; ++i) {
+		u64 cpm_bno = desc_base + (desc_index + i) % desc_blks;
+
+		err = apfs_read_single_cpm_block(sb, cpm_bno);
+		if (err) {
+			apfs_err(sb, "failed to read cpm block %u", i);
+			/* No transaction to abort yet */
+			apfs_force_readonly(nxi);
+			return err;
+		}
+	}
+	return 0;
+}
+
+static void apfs_trans_commit_work(struct work_struct *work)
+{
+	struct super_block *sb = NULL;
+	struct apfs_nxsb_info *nxi = NULL;
+	struct apfs_nx_transaction *trans = NULL;
+	int err;
+
+	trans = container_of(to_delayed_work(work), struct apfs_nx_transaction, t_work);
+	nxi = container_of(trans, struct apfs_nxsb_info, nx_transaction);
+	sb = trans->t_work_sb;
+
+	/*
+	 * If sb is set then the transaction already started, there is no need
+	 * for apfs_transaction_start() here. It would be cleaner to call it
+	 * anyway (and check in there if sb is set), but maxops is a problem
+	 * because we don't need any space. I really need to rethink that stuff
+	 * (TODO).
+	 */
+	down_write(&nxi->nx_big_sem);
+	if (!sb || sb->s_flags & SB_RDONLY) {
+		/* The commit already took place, or there was an abort */
+		up_write(&nxi->nx_big_sem);
+		return;
+	}
+
+	trans->t_state |= APFS_NX_TRANS_FORCE_COMMIT;
+	err = apfs_transaction_commit(sb);
+	if (err) {
+		apfs_err(sb, "queued commit failed (err:%d)", err);
+		apfs_transaction_abort(sb);
+	}
+}
+
+/**
+ * apfs_transaction_init - Initialize the transaction struct for the container
+ * @trans: the transaction structure
+ */
+void apfs_transaction_init(struct apfs_nx_transaction *trans)
+{
+	trans->t_state = 0;
+	INIT_DELAYED_WORK(&trans->t_work, apfs_trans_commit_work);
+	INIT_LIST_HEAD(&trans->t_inodes);
+	INIT_LIST_HEAD(&trans->t_buffers);
+	trans->t_buffers_count = 0;
+	trans->t_starts_count = 0;
+}
+
+/**
+ * apfs_transaction_start - Begin a new transaction
+ * @sb:		superblock structure
+ * @kind:	transaction kind for space preallocation
+ *
+ * Also locks the filesystem for writing; returns 0 on success or a negative
+ * error code in case of failure.
+ */
+int apfs_transaction_start(struct super_block *sb, enum apfs_trans_kind kind)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+	int err;
+
+	down_write(&nxi->nx_big_sem);
+
+	if (sb->s_flags & SB_RDONLY) {
+		/* A previous transaction has failed; this should be rare */
+		up_write(&nxi->nx_big_sem);
+		return -EROFS;
+	}
+
+	/*
+	 * Ephemeral objects are read only once, kept in memory, and committed
+	 * to disk along with each transaction.
+	 */
+	if (!nxi->nx_eph_list) {
+		err = apfs_read_ephemeral_objects(sb);
+		if (err) {
+			up_write(&nxi->nx_big_sem);
+			apfs_err(sb, "failed to read the ephemeral objects");
+			return err;
+		}
+	}
+
+	if (nx_trans->t_starts_count == 0) {
+		++nxi->nx_xid;
+		nxi->nx_raw->nx_next_xid = cpu_to_le64(nxi->nx_xid + 1);
+
+		err = apfs_read_spaceman(sb);
+		if (err) {
+			apfs_err(sb, "failed to read the spaceman");
+			goto fail;
+		}
+	}
+
+	/* Don't start transactions unless we are sure they fit in disk */
+	if (!apfs_transaction_has_room(sb, kind)) {
+		/* Commit what we have so far to flush the queues */
+		nx_trans->t_state |= APFS_NX_TRANS_FORCE_COMMIT;
+		err = apfs_transaction_commit(sb);
+		if (err) {
+			apfs_err(sb, "commit failed");
+			goto fail;
+		}
+		return -ENOSPC;
+	}
+
+	if (nx_trans->t_starts_count == 0) {
+		err = apfs_map_volume_super(sb, true /* write */);
+		if (err) {
+			apfs_err(sb, "CoW failed for volume super");
+			goto fail;
+		}
+
+		/* TODO: don't copy these nodes for transactions that don't use them */
+		err = apfs_read_omap(sb, true /* write */);
+		if (err) {
+			apfs_err(sb, "CoW failed for omap");
+			goto fail;
+		}
+		err = apfs_read_catalog(sb, true /* write */);
+		if (err) {
+			apfs_err(sb, "Cow failed for catalog");
+			goto fail;
+		}
+	}
+
+	nx_trans->t_starts_count++;
+	return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+/**
+ * apfs_transaction_flush_all_inodes - Flush inode metadata to the buffer heads
+ * @sb: superblock structure
+ *
+ * This messes a lot with the disk layout, so it must be called ahead of time
+ * if we need it to be stable for the rest or the transaction (for example, if
+ * we are setting up a snapshot).
+ */
+int apfs_transaction_flush_all_inodes(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+	int err = 0, curr_err;
+
+	ASSERT(!(sb->s_flags & SB_RDONLY));
+
+	while (!list_empty(&nx_trans->t_inodes)) {
+		struct apfs_inode_info *ai = NULL;
+		struct inode *inode = NULL;
+
+		ai = list_first_entry(&nx_trans->t_inodes, struct apfs_inode_info, i_list);
+		inode = &ai->vfs_inode;
+
+		/* This is a bit wasteful if the inode will get deleted */
+		curr_err = apfs_update_inode(inode, NULL /* new_name */);
+		if (curr_err)
+			err = curr_err;
+		apfs_inode_state_clear_raw(inode, I_DIRTY_ALL);
+
+		/*
+		 * The same inode may get dirtied again as soon as we release
+		 * the lock, and we don't want to miss that.
+		 */
+		list_del_init(&ai->i_list);
+
+		nx_trans->t_state |= APFS_NX_TRANS_COMMITTING;
+		up_write(&nxi->nx_big_sem);
+
+		/* Unlocked, so it may call evict() and wait for writeback */
+		iput(inode);
+
+		down_write(&nxi->nx_big_sem);
+		nx_trans->t_state = 0;
+
+		/* Transaction aborted during writeback, error code is lost */
+		if (sb->s_flags & SB_RDONLY) {
+			apfs_err(sb, "abort during inode writeback");
+			return -EROFS;
+		}
+	}
+
+	return err;
+}
+
+/**
+ * apfs_write_single_ephemeral_object - Write a single ephemeral object to bh's
+ * @sb:		filesystem superblock
+ * @obj_raw:	contents of the object
+ * @map:	checkpoint mapping for the object, already updated
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_write_single_ephemeral_object(struct super_block *sb, struct apfs_obj_phys *obj_raw, const struct apfs_checkpoint_mapping *map)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *raw_sb = NULL;
+	struct buffer_head *bh = NULL;
+	u64 data_base, bno;
+	u32 data_blks, size;
+	int err, i, data_idx;
+
+	raw_sb = nxi->nx_raw;
+	data_base = le64_to_cpu(raw_sb->nx_xp_data_base);
+	data_blks = le32_to_cpu(raw_sb->nx_xp_data_blocks);
+
+	bno = le64_to_cpu(map->cpm_paddr);
+	size = le32_to_cpu(map->cpm_size);
+	obj_raw->o_xid = cpu_to_le64(nxi->nx_xid);
+	apfs_multiblock_set_csum((char *)obj_raw, size);
+
+	data_idx = bno - data_base;
+	for (i = 0; i < size >> sb->s_blocksize_bits; ++i) {
+		bh = apfs_getblk(sb, data_base + data_idx);
+		if (!bh) {
+			apfs_err(sb, "failed to map ephemeral block");
+			return -EIO;
+		}
+		err = apfs_transaction_join(sb, bh);
+		if (err) {
+			brelse(bh);
+			bh = NULL;
+			return err;
+		}
+		memcpy(bh->b_data, (char *)obj_raw + (i << sb->s_blocksize_bits), sb->s_blocksize);
+		brelse(bh);
+		bh = NULL;
+		/* Somewhat surprisingly, objects can wrap around */
+		if (++data_idx == data_blks)
+			data_idx = 0;
+	}
+	return 0;
+}
+
+/**
+ * apfs_write_ephemeral_objects - Write all ephemeral objects to bh's
+ * @sb: filesystem superblock
+ *
+ * Returns 0 on sucess, or a negative error code in case of failure.
+ */
+static int apfs_write_ephemeral_objects(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_superblock *raw_sb = nxi->nx_raw;
+	struct apfs_checkpoint_map_phys *cpm = NULL;
+	struct buffer_head *cpm_bh = NULL;
+	struct apfs_ephemeral_object_info *eph_info = NULL;
+	u64 cpm_bno;
+	u64 desc_base, data_base;
+	u32 desc_index, desc_blks, desc_len, desc_next;
+	u32 data_index, data_blks, data_len, data_next;
+	u32 desc_limit, data_limit;
+	u32 obj_blkcnt;
+	int err, i, cpm_start;
+
+	if (!nxi->nx_eph_list) {
+		apfs_alert(sb, "missing ephemeral object list");
+		return -EFSCORRUPTED;
+	}
+
+	desc_next = le32_to_cpu(raw_sb->nx_xp_desc_next);
+	desc_base = le64_to_cpu(raw_sb->nx_xp_desc_base);
+	desc_index = desc_next; /* New checkpoint */
+	desc_blks = le32_to_cpu(raw_sb->nx_xp_desc_blocks);
+	desc_len = 0; /* For now */
+
+	data_next = le32_to_cpu(raw_sb->nx_xp_data_next);
+	data_base = le64_to_cpu(raw_sb->nx_xp_data_base);
+	data_index = data_next; /* New checkpoint */
+	data_blks = le32_to_cpu(raw_sb->nx_xp_data_blocks);
+	data_len = 0; /* For now */
+
+	/*
+	 * The reference doesn't mention anything about this, but I need to
+	 * put some sort of a limit or else the rings could wrap around and
+	 * corrupt themselves.
+	 */
+	desc_limit = desc_blks >> 2;
+	data_limit = data_blks >> 2;
+
+	for (i = 0; i < nxi->nx_eph_count; ++i) {
+		if (data_len == data_limit) {
+			apfs_err(sb, "too many checkpoint data blocks");
+			return -EFSCORRUPTED;
+		}
+
+		if (!cpm) {
+			cpm_start = i;
+			if (desc_len == desc_limit) {
+				apfs_err(sb, "too many checkpoint descriptor blocks");
+				return -EFSCORRUPTED;
+			}
+			cpm_bno = desc_base + (desc_index + desc_len) % desc_blks;
+			err = apfs_create_cpm_block(sb, cpm_bno, &cpm_bh);
+			if (err) {
+				apfs_err(sb, "failed to create cpm block");
+				return err;
+			}
+			cpm = (void *)cpm_bh->b_data;
+			desc_len += 1;
+		}
+
+		eph_info = &nxi->nx_eph_list[i];
+		data_next = (data_index + data_len) % data_blks;
+		obj_blkcnt = eph_info->size >> sb->s_blocksize_bits;
+
+		err = apfs_create_cpoint_map(sb, cpm, eph_info->object, data_base + data_next, eph_info->size);
+		if (err) {
+			if (err == -ENOSPC)
+				cpm->cpm_flags = 0; /* No longer the last */
+			brelse(cpm_bh);
+			cpm = NULL;
+			cpm_bh = NULL;
+			if (err == -ENOSPC) {
+				--i;
+				continue;
+			}
+			apfs_err(sb, "failed to create cpm map %d", i);
+			return err;
+		}
+		err = apfs_write_single_ephemeral_object(sb, eph_info->object, &cpm->cpm_map[i - cpm_start]);
+		if (err) {
+			brelse(cpm_bh);
+			cpm = NULL;
+			cpm_bh = NULL;
+			apfs_err(sb, "failed to write ephemeral object %d", i);
+			return err;
+		}
+		data_len += obj_blkcnt;
+	}
+
+	/*
+	 * The checkpoint superblock can't be set until the very end of the
+	 * transaction commit, but allocate its block here already.
+	 */
+	nxi->nx_bno = desc_base + (desc_index + desc_len) % desc_blks;
+	desc_len += 1;
+
+	desc_next = (desc_index + desc_len) % desc_blks;
+	data_next = (data_index + data_len) % data_blks;
+
+	raw_sb->nx_xp_desc_next = cpu_to_le32(desc_next);
+	raw_sb->nx_xp_desc_index = cpu_to_le32(desc_index);
+	raw_sb->nx_xp_desc_len = cpu_to_le32(desc_len);
+
+	raw_sb->nx_xp_data_next = cpu_to_le32(data_next);
+	raw_sb->nx_xp_data_index = cpu_to_le32(data_index);
+	raw_sb->nx_xp_data_len = cpu_to_le32(data_len);
+
+	return 0;
+}
+
+/**
+ * apfs_transaction_commit_nx - Definitely commit the current transaction
+ * @sb: superblock structure
+ */
+static int apfs_transaction_commit_nx(struct super_block *sb)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+	struct apfs_bh_info *bhi, *tmp;
+	int err = 0;
+
+	ASSERT(!(sb->s_flags & SB_RDONLY));
+
+	/* Before committing the bhs, write all inode metadata to them */
+	err = apfs_transaction_flush_all_inodes(sb);
+	if (err) {
+		apfs_err(sb, "failed to flush all inodes");
+		return err;
+	}
+
+	/*
+	 * Now that nothing else will be freed, flush the last update to the
+	 * free queues so that it can be committed to disk along with all the
+	 * ephemeral objects.
+	 */
+	if (sm->sm_free_cache_base) {
+		err = apfs_free_queue_insert_nocache(sb, sm->sm_free_cache_base, sm->sm_free_cache_blkcnt);
+		if (err) {
+			apfs_err(sb, "fq cache flush failed (0x%llx-0x%llx)", sm->sm_free_cache_base, sm->sm_free_cache_blkcnt);
+			return err;
+		}
+		sm->sm_free_cache_base = sm->sm_free_cache_blkcnt = 0;
+	}
+	/*
+	 * Writing the ip bitmaps modifies the spaceman, so it must happen
+	 * before we commit the ephemeral objects. It must also happen after we
+	 * flush the free queue, in case the last freed range was in the ip.
+	 */
+	err = apfs_write_ip_bitmaps(sb);
+	if (err) {
+		apfs_err(sb, "failed to commit the ip bitmaps");
+		return err;
+	}
+	err = apfs_write_ephemeral_objects(sb);
+	if (err)
+		return err;
+
+	list_for_each_entry(bhi, &nx_trans->t_buffers, list) {
+		struct buffer_head *bh = bhi->bh;
+
+		ASSERT(buffer_trans(bh));
+
+		if (buffer_csum(bh))
+			apfs_obj_set_csum(sb, (void *)bh->b_data);
+
+		clear_buffer_dirty(bh);
+		get_bh(bh);
+		lock_buffer(bh);
+		bh->b_end_io = end_buffer_write_sync;
+		apfs_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
+	}
+	list_for_each_entry_safe(bhi, tmp, &nx_trans->t_buffers, list) {
+		struct buffer_head *bh = bhi->bh;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 11, 0)
+		struct folio *folio = NULL;
+#else
+		struct page *page = NULL;
+#endif
+		bool is_metadata;
+
+		ASSERT(buffer_trans(bh));
+
+		wait_on_buffer(bh);
+		if (!buffer_uptodate(bh)) {
+			apfs_err(sb, "failed to write some blocks");
+			return -EIO;
+		}
+
+		list_del(&bhi->list);
+		clear_buffer_trans(bh);
+		nx_trans->t_buffers_count--;
+
+		bh->b_private = NULL;
+		bhi->bh = NULL;
+		kfree(bhi);
+		bhi = NULL;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 11, 0)
+		folio = page_folio(bh->b_page);
+		folio_get(folio);
+#else
+		page = bh->b_page;
+		get_page(page);
+#endif
+
+		is_metadata = buffer_csum(bh);
+		clear_buffer_csum(bh);
+		put_bh(bh);
+		bh = NULL;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 11, 0)
+		folio_lock(folio);
+		folio_mkclean(folio);
+#else
+		/* Future writes to mmapped areas should fault for CoW */
+		lock_page(page);
+		page_mkclean(page);
+#endif
+		/* XXX: otherwise, the page cache fills up and crashes the machine */
+		if (!is_metadata) {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 11, 0)
+			try_to_free_buffers(folio);
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(5, 19, 0) || RHEL_VERSION_GE(9, 3)
+			try_to_free_buffers(page_folio(page));
+#else
+			try_to_free_buffers(page);
+#endif
+		}
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 11, 0)
+		folio_unlock(folio);
+		folio_put(folio);
+#else
+		unlock_page(page);
+		put_page(page);
+#endif
+	}
+	err = apfs_checkpoint_end(sb);
+	if (err) {
+		apfs_err(sb, "failed to end the checkpoint");
+		return err;
+	}
+
+	nx_trans->t_starts_count = 0;
+	nx_trans->t_buffers_count = 0;
+	return 0;
+}
+
+/**
+ * apfs_transaction_need_commit - Evaluate if a commit is required
+ * @sb: superblock structure
+ */
+static bool apfs_transaction_need_commit(struct super_block *sb)
+{
+	struct apfs_spaceman *sm = APFS_SM(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+
+	if (nx_trans->t_state & APFS_NX_TRANS_DEFER_COMMIT) {
+		nx_trans->t_state &= ~APFS_NX_TRANS_DEFER_COMMIT;
+		return false;
+	}
+
+	/* Avoid nested commits on inode writeback */
+	if (nx_trans->t_state & APFS_NX_TRANS_COMMITTING)
+		return false;
+
+	if (nx_trans->t_state & APFS_NX_TRANS_FORCE_COMMIT) {
+		nx_trans->t_state = 0;
+		return true;
+	}
+
+	if (sm) {
+		struct apfs_spaceman_phys *sm_raw = sm->sm_raw;
+		struct apfs_spaceman_free_queue *fq_ip = &sm_raw->sm_fq[APFS_SFQ_IP];
+		struct apfs_spaceman_free_queue *fq_main = &sm_raw->sm_fq[APFS_SFQ_MAIN];
+		int buffers_max = nxi->nx_trans_buffers_max;
+		int starts_max = APFS_TRANS_STARTS_MAX;
+		int mq_max = APFS_TRANS_MAIN_QUEUE_MAX;
+		int maxnodes;
+
+		/*
+		 * Try to avoid committing halfway through a data block write,
+		 * otherwise the block will be put through copy-on-write again,
+		 * causing unnecessary fragmentation.
+		 */
+		if (nx_trans->t_state & APFS_NX_TRANS_INCOMPLETE_BLOCK) {
+			buffers_max += 50;
+			starts_max += 50;
+			mq_max += 20;
+		}
+
+		if (nx_trans->t_buffers_count > buffers_max)
+			return true;
+		if (nx_trans->t_starts_count > starts_max)
+			return true;
+
+		/*
+		 * The internal pool has enough blocks to map the container
+		 * exactly 3 times. Don't allow large transactions if we can't
+		 * be sure the bitmap changes will all fit.
+		 */
+		if (le64_to_cpu(fq_ip->sfq_count) * 3 > le64_to_cpu(sm_raw->sm_ip_block_count))
+			return true;
+
+		/* Don't let the main queue get too full either */
+		if (le64_to_cpu(fq_main->sfq_count) > mq_max)
+			return true;
+
+		/*
+		 * The main free queue can become unbalanced enough to reach
+		 * the node limit while being mostly empty. For now, the only
+		 * way I have to rebalance it is to flush it entirely with a
+		 * new transaction. We could wait longer to do it, but I don't
+		 * see the point.
+		 */
+		maxnodes = le16_to_cpu(fq_main->sfq_tree_node_limit);
+		maxnodes = (maxnodes + 1) >> 1;
+		if (sm->sm_main_fq_nodes > 1 && sm->sm_main_fq_nodes >= maxnodes)
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * apfs_transaction_commit - Possibly commit the current transaction
+ * @sb: superblock structure
+ *
+ * On success returns 0 and releases the big filesystem lock. On failure,
+ * returns a negative error code, and the caller is responsibly for aborting
+ * the transaction.
+ */
+int apfs_transaction_commit(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *trans = NULL;
+	int err = 0;
+
+	trans = &nxi->nx_transaction;
+
+	if (apfs_transaction_need_commit(sb)) {
+		err = apfs_transaction_commit_nx(sb);
+		if (err) {
+			apfs_err(sb, "transaction commit failed");
+			return err;
+		}
+		trans->t_work_sb = NULL;
+		cancel_delayed_work(&trans->t_work);
+	} else {
+		trans->t_work_sb = sb;
+		mod_delayed_work(system_wq, &trans->t_work, msecs_to_jiffies(100));
+	}
+
+	up_write(&nxi->nx_big_sem);
+	return 0;
+}
+
+/**
+ * apfs_inode_join_transaction - Add an inode to the current transaction
+ * @sb:		superblock structure
+ * @inode:	vfs inode to add
+ */
+void apfs_inode_join_transaction(struct super_block *sb, struct inode *inode)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+	struct apfs_inode_info *ai = APFS_I(inode);
+
+	ASSERT(!(sb->s_flags & SB_RDONLY));
+	lockdep_assert_held_write(&nxi->nx_big_sem);
+
+	if (!list_empty(&ai->i_list)) /* Already in the transaction */
+		return;
+
+	ihold(inode);
+	list_add(&ai->i_list, &nx_trans->t_inodes);
+}
+
+/**
+ * apfs_transaction_join - Add a buffer head to the current transaction
+ * @sb:	superblock structure
+ * @bh:	the buffer head
+ */
+int apfs_transaction_join(struct super_block *sb, struct buffer_head *bh)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+	struct apfs_bh_info *bhi;
+
+	ASSERT(!(sb->s_flags & SB_RDONLY));
+	lockdep_assert_held_write(&nxi->nx_big_sem);
+
+	if (buffer_trans(bh)) /* Already part of the only transaction */
+		return 0;
+
+	/* TODO: use a slab cache */
+	bhi = kzalloc(sizeof(*bhi), GFP_NOFS);
+	if (!bhi)
+		return -ENOMEM;
+	get_bh(bh);
+	bhi->bh = bh;
+	list_add(&bhi->list, &nx_trans->t_buffers);
+	nx_trans->t_buffers_count++;
+
+	set_buffer_trans(bh);
+	bh->b_private = bhi;
+	return 0;
+}
+
+/**
+ * apfs_force_readonly - Set the whole container as read-only
+ * @nxi: container superblock info
+ */
+static void apfs_force_readonly(struct apfs_nxsb_info *nxi)
+{
+	struct apfs_sb_info *sbi = NULL;
+	struct super_block *sb = NULL;
+
+	list_for_each_entry(sbi, &nxi->vol_list, list) {
+		sb = sbi->s_vobject.sb;
+		sb->s_flags |= SB_RDONLY;
+	}
+	nxi->nx_flags &= ~APFS_READWRITE;
+}
+
+/**
+ * apfs_transaction_abort - Abort the current transaction
+ * @sb: superblock structure
+ *
+ * Releases the big filesystem lock and clears the in-memory transaction data;
+ * the on-disk changes are irrelevant because the superblock checksum hasn't
+ * been written yet. Leaves the filesystem in read-only state.
+ */
+void apfs_transaction_abort(struct super_block *sb)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_nx_transaction *nx_trans = &nxi->nx_transaction;
+	struct apfs_bh_info *bhi, *tmp;
+	struct apfs_inode_info *ai, *ai_tmp;
+
+	if (sb->s_flags & SB_RDONLY) {
+		/* Transaction already aborted, do nothing */
+		ASSERT(list_empty(&nx_trans->t_inodes));
+		ASSERT(list_empty(&nx_trans->t_buffers));
+		up_write(&nxi->nx_big_sem);
+		return;
+	}
+
+	nx_trans->t_state = 0;
+	apfs_err(sb, "aborting transaction");
+
+	--nxi->nx_xid;
+	list_for_each_entry_safe(bhi, tmp, &nx_trans->t_buffers, list) {
+		struct buffer_head *bh = bhi->bh;
+
+		bh->b_private = NULL;
+		clear_buffer_dirty(bh);
+		clear_buffer_trans(bh);
+		clear_buffer_csum(bh);
+		brelse(bh);
+		bhi->bh = NULL;
+
+		list_del(&bhi->list);
+		kfree(bhi);
+	}
+
+	/*
+	 * It's not possible to undo in-memory changes from old operations in
+	 * the aborted transaction. To avoid corruption, never write again.
+	 */
+	apfs_force_readonly(nxi);
+
+	up_write(&nxi->nx_big_sem);
+
+	list_for_each_entry_safe(ai, ai_tmp, &nx_trans->t_inodes, i_list) {
+		list_del_init(&ai->i_list);
+		iput(&ai->vfs_inode);
+	}
+}
diff --git a/fs/apfs/unicode.c b/fs/apfs/unicode.c
new file mode 100644
index 000000000..7baf4305b
--- /dev/null
+++ b/fs/apfs/unicode.c
@@ -0,0 +1,3156 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ *
+ * Routines and data for the normalization of unicode strings.
+ * Somewhat based on linux/fs/hfsplus/unicode.c
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/nls.h>
+#include <linux/ctype.h>
+#include <linux/version.h>
+#include "unicode.h"
+
+/* The arrays of unicode data are defined at the bottom of the file */
+/* TODO: would a single trie with all the data be more efficient? */
+static u16 apfs_nfd_trie[];
+static unicode_t apfs_nfd[];
+static u16 apfs_cf_trie[];
+static unicode_t apfs_cf[];
+static u8 apfs_ccc_trie[];
+
+#define TRIE_HEIGHT		5
+
+/* A trie node has one child for each possible nibble in the key */
+#define TRIE_CHILD_SHIFT	4
+#define TRIE_CHILD_MASK		((1 << TRIE_CHILD_SHIFT) - 1)
+
+/* A trie value length is stored in the last three bits of its position */
+#define TRIE_POS_SHIFT		3
+#define TRIE_SIZE_MASK		((1 << TRIE_POS_SHIFT) - 1)
+
+/**
+ * apfs_trie_find - Look up a trie value
+ * @trie:	trie to search
+ * @key:	search key (a unicode character)
+ * @result:	on return, this either holds the value (on a ccc lookup) or its
+ *		position in the value array (on a cf or nfd lookup).
+ * @is_ccc:	true if this a ccc (canonical combining class) lookup
+ *
+ * Returns the length of the value (0 if it doesn't exist).
+ */
+static int apfs_trie_find(void *trie, unicode_t key, void *result, bool is_ccc)
+{
+	int node = 0;
+	int h;
+
+	for (h = TRIE_HEIGHT - 1; h >= 0; --h) {
+		int child = (key >> (TRIE_CHILD_SHIFT * h)) & TRIE_CHILD_MASK;
+		int child_index = (node << TRIE_CHILD_SHIFT) + child;
+
+		if (is_ccc)
+			node = ((u8 *)trie)[child_index];
+		else
+			node = ((u16 *)trie)[child_index];
+
+		if (node == 0) {
+			*(u8 *)result = 0;
+			return 0;
+		}
+	}
+
+	if (is_ccc) {
+		/* ccc values fit in one byte, so no need for a value array */
+		*(u8 *)result = node;
+		return 1;
+	}
+
+	*(u16 *)result = node >> TRIE_POS_SHIFT;
+	return node & TRIE_SIZE_MASK;
+}
+
+/**
+ * apfs_init_unicursor - Initialize an apfs_unicursor structure
+ * @cursor:	cursor to initialize
+ * @utf8str:	string to normalize
+ * @total_len:	length of the string
+ */
+void apfs_init_unicursor(struct apfs_unicursor *cursor, const char *utf8str, unsigned int total_len)
+{
+	cursor->utf8curr = utf8str;
+	cursor->total_len = total_len;
+	cursor->length = -1;
+	cursor->last_pos = -1;
+	cursor->last_ccc = 0;
+}
+
+#define HANGUL_S_BASE	0xac00
+#define HANGUL_L_BASE	0x1100
+#define HANGUL_V_BASE	0x1161
+#define HANGUL_T_BASE	0x11a7
+#define HANGUL_L_COUNT	19
+#define HANGUL_V_COUNT	21
+#define HANGUL_T_COUNT	28
+#define HANGUL_N_COUNT	(HANGUL_V_COUNT * HANGUL_T_COUNT)
+#define HANGUL_S_COUNT	(HANGUL_L_COUNT * HANGUL_N_COUNT)
+
+/**
+ * apfs_is_precomposed_hangul - Check if a character is a Hangul syllable
+ * @utf32char:	character to check
+ *
+ * This function was adapted from sample code in section 3.12 of the
+ * Unicode Standard, version 9.0.
+ *
+ * Copyright (C) 1991-2018 Unicode, Inc.  All rights reserved.  Distributed
+ * under the Terms of Use in http://www.unicode.org/copyright.html.
+ */
+static bool apfs_is_precomposed_hangul(unicode_t utf32char)
+{
+	int index;
+
+	index = utf32char - HANGUL_S_BASE;
+	return (index >= 0 && index < HANGUL_S_COUNT);
+}
+
+/* Signals the end of the normalization for a single character */
+#define NORM_END	(unicode_t)(-1)
+
+/**
+ * apfs_decompose_hangul - Decompose a Hangul syllable
+ * @utf32char:	Hangul syllable to decompose
+ * @off:	offset of the wanted character from the decomposition
+ *
+ * Returns the single character at offset @off in the decomposition of
+ * @utf32char, or NORM_END if this offset is past the end.
+ *
+ * This function was adapted from sample code in section 3.12 of the
+ * Unicode Standard, version 9.0.
+ *
+ * Copyright (C) 1991-2018 Unicode, Inc.  All rights reserved.  Distributed
+ * under the Terms of Use in http://www.unicode.org/copyright.html.
+ */
+static unicode_t apfs_decompose_hangul(unicode_t utf32char, int off)
+{
+	int index;
+	int l, v, t;
+
+	index = utf32char - HANGUL_S_BASE;
+
+	l = HANGUL_L_BASE + index / HANGUL_N_COUNT;
+	if (off == 0)
+		return l;
+
+	v = HANGUL_V_BASE + (index % HANGUL_N_COUNT) / HANGUL_T_COUNT;
+	if (off == 1)
+		return v;
+
+	t = HANGUL_T_BASE + index % HANGUL_T_COUNT;
+	if (off == 2 && t != HANGUL_T_BASE)
+		return t;
+
+	return NORM_END;
+}
+
+/**
+ * apfs_normalize_char - Normalize a unicode character
+ * @utf32char:	character to normalize
+ * @off:	offset of the wanted character from the normalization
+ * @case_fold:	case fold the char?
+ *
+ * Returns the single character at offset @off in the normalization of
+ * @utf32char, or NORM_END if this offset is past the end.
+ */
+static unicode_t apfs_normalize_char(unicode_t utf32char, int off,
+				     bool case_fold)
+{
+	int nfd_len;
+	unicode_t *nfd, *cf;
+	u16 pos;
+	int ret;
+
+	if (apfs_is_precomposed_hangul(utf32char)) /* Hangul has no case */
+		return apfs_decompose_hangul(utf32char, off);
+
+	ret = apfs_trie_find(apfs_nfd_trie, utf32char,
+			     &pos, false /* is_ccc */);
+	if (!ret) {
+		/* The decomposition is just the same character */
+		nfd_len = 1;
+		nfd = &utf32char;
+	} else {
+		nfd_len = ret;
+		nfd = &apfs_nfd[pos];
+	}
+
+	if (!case_fold) {
+		if (off < nfd_len)
+			return nfd[off];
+		return NORM_END;
+	}
+
+	for (; nfd_len > 0; nfd++, nfd_len--) {
+		int cf_len;
+
+		ret = apfs_trie_find(apfs_cf_trie, *nfd,
+				     &pos, false /* is_ccc */);
+		if (!ret) {
+			/* The case folding is just the same character */
+			cf_len = 1;
+			cf = nfd;
+		} else {
+			cf_len = ret;
+			cf = &apfs_cf[pos];
+		}
+
+		if (off < cf_len)
+			return cf[off];
+		off -= cf_len;
+	}
+
+	return NORM_END;
+}
+
+/**
+ * apfs_get_normalization_length - Count the characters until the next starter
+ * @utf8str:	string to normalize, may begin with several starters
+ * @total_len:	length of the string to normalize
+ * @case_fold:	true if the count should consider case folding
+ *
+ * Returns the number of unicode characters in the normalization of the
+ * substring that begins at @utf8str and ends at the first nonconsecutive
+ * starter. Or 0 if the substring has invalid UTF-8.
+ */
+static int apfs_get_normalization_length(const char *utf8str, unsigned int total_len, bool case_fold)
+{
+	int utf8len, pos, norm_len = 0;
+	bool starters_over = false;
+	unicode_t utf32char;
+
+	while (1) {
+		if (!total_len || !*utf8str)
+			return norm_len;
+		utf8len = utf8_to_utf32(utf8str, min(total_len, 4U), &utf32char);
+		if (utf8len < 0) /* Invalid unicode; don't normalize anything */
+			return 0;
+
+		for (pos = 0;; pos++, norm_len++) {
+			unicode_t utf32norm;
+			u8 ccc;
+
+			utf32norm = apfs_normalize_char(utf32char, pos,
+							case_fold);
+			if (utf32norm == NORM_END)
+				break;
+
+			apfs_trie_find(apfs_ccc_trie, utf32norm, &ccc,
+				       true /* is_ccc */);
+
+			if (ccc != 0)
+				starters_over = true;
+			else if (starters_over) /* Reached the next starter */
+				return norm_len;
+		}
+		utf8str += utf8len;
+		total_len -= utf8len;
+	}
+}
+
+/**
+ * apfs_normalize_next - Return the next normalized character from a string
+ * @cursor:	unicode cursor for the string
+ * @case_fold:	case fold the string?
+ *
+ * Sets @cursor->length to the length of the normalized substring between
+ * @cursor->utf8curr and the first nonconsecutive starter. Returns a single
+ * normalized character, setting @cursor->last_ccc and @cursor->last_pos to
+ * its CCC and position in the substring. When the end of the substring is
+ * reached, updates @cursor->utf8curr to point to the beginning of the next
+ * one.
+ *
+ * Returns 0 if the substring has invalid UTF-8.
+ */
+unicode_t apfs_normalize_next(struct apfs_unicursor *cursor, bool case_fold)
+{
+	const char *utf8str = cursor->utf8curr;
+	unsigned int total_len = cursor->total_len;
+	int str_pos, min_pos = -1;
+	unicode_t utf32min = 0;
+	u8 min_ccc;
+
+new_starter:
+	if (likely(isascii(*utf8str))) {
+		if (!total_len)
+			return 0;
+		cursor->utf8curr = utf8str + 1;
+		cursor->total_len = total_len - 1;
+		if (case_fold)
+			return tolower(*utf8str);
+		return *utf8str;
+	}
+
+	if (cursor->length < 0) {
+		cursor->length = apfs_get_normalization_length(utf8str, total_len, case_fold);
+		if (cursor->length == 0)
+			return 0;
+	}
+
+	str_pos = 0;
+	min_ccc = 0xFF;	/* Above all possible ccc's */
+
+	while (1) {
+		unicode_t utf32char;
+		int utf8len, pos;
+
+		utf8len = utf8_to_utf32(utf8str, min(total_len, 4U), &utf32char);
+		for (pos = 0;; pos++, str_pos++) {
+			unicode_t utf32norm;
+			u8 ccc;
+
+			utf32norm = apfs_normalize_char(utf32char, pos,
+							case_fold);
+			if (utf32norm == NORM_END)
+				break;
+
+			apfs_trie_find(apfs_ccc_trie, utf32norm, &ccc,
+				       true /* is_ccc */);
+
+			if (ccc >= min_ccc || ccc < cursor->last_ccc)
+				continue;
+			if (ccc > cursor->last_ccc ||
+			    str_pos > cursor->last_pos) {
+				utf32min = utf32norm;
+				min_ccc = ccc;
+				min_pos = str_pos;
+			}
+		}
+
+		utf8str += utf8len;
+		total_len -= utf8len;
+		if (str_pos == cursor->length) {
+			/* Reached the following starter */
+			if (min_ccc != 0xFF) {
+				/* Not done with this substring yet */
+				cursor->last_ccc = min_ccc;
+				cursor->last_pos = min_pos;
+				return utf32min;
+			}
+			/* Continue from the next starter */
+			apfs_init_unicursor(cursor, utf8str, total_len);
+			goto new_starter;
+		}
+	}
+}
+
+/*
+ * The following arrays were built with data provided by the Unicode Standard,
+ * version 9.0.
+ *
+ * Copyright (C) 1991-2018 Unicode, Inc.  All rights reserved.  Distributed
+ * under the Terms of Use in http://www.unicode.org/copyright.html.
+ */
+
+static u16 apfs_nfd_trie[] = {
+	/* Node for range 0x_____ */
+	0x0001, 0x0002, 0x0003, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0____ */
+	0x0004, 0x0005, 0x0006, 0x0007, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0008,
+	/* Node for range 0x1____ */
+	0x0000, 0x0009, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x000a, 0x0000, 0x0000,
+	/* Node for range 0x2____ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x000b,
+	/* Node for range 0x00___ */
+	0x000c, 0x000d, 0x000e, 0x000f, 0x0010, 0x0000, 0x0011, 0x0000,
+	0x0000, 0x0012, 0x0013, 0x0014, 0x0015, 0x0016, 0x0000, 0x0017,
+	/* Node for range 0x01___ */
+	0x0018, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0019, 0x0000, 0x0000, 0x001a, 0x001b,
+	/* Node for range 0x02___ */
+	0x001c, 0x001d, 0x001e, 0x001f, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0020, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x03___ */
+	0x0021, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0f___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0022, 0x0023, 0x0024, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x11___ */
+	0x0025, 0x0026, 0x0000, 0x0027, 0x0028, 0x0029, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1d___ */
+	0x0000, 0x002a, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x2f___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x002b, 0x002c, 0x002d, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x000__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x002e, 0x002f, 0x0030, 0x0031,
+	/* Node for range 0x001__ */
+	0x0032, 0x0033, 0x0034, 0x0035, 0x0036, 0x0037, 0x0038, 0x0039,
+	0x0000, 0x0000, 0x003a, 0x003b, 0x003c, 0x003d, 0x003e, 0x003f,
+	/* Node for range 0x002__ */
+	0x0040, 0x0041, 0x0042, 0x0043, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x003__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0044, 0x0000, 0x0000, 0x0045,
+	0x0046, 0x0047, 0x0048, 0x0049, 0x004a, 0x004b, 0x0000, 0x0000,
+	/* Node for range 0x004__ */
+	0x004c, 0x004d, 0x0000, 0x004e, 0x0000, 0x004f, 0x0000, 0x0050,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0051, 0x0052, 0x0053, 0x0054,
+	/* Node for range 0x006__ */
+	0x0000, 0x0000, 0x0055, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0056, 0x0057, 0x0000, 0x0000,
+	/* Node for range 0x009__ */
+	0x0000, 0x0000, 0x0058, 0x0059, 0x0000, 0x005a, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x005b, 0x005c, 0x0000, 0x0000,
+	/* Node for range 0x00a__ */
+	0x0000, 0x0000, 0x0000, 0x005d, 0x0000, 0x005e, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00b__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x005f, 0x0060, 0x0000, 0x0000,
+	0x0000, 0x0061, 0x0000, 0x0000, 0x0062, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00c__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0063, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0064, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00d__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0065, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0066, 0x0000, 0x0000,
+	/* Node for range 0x00f__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0067, 0x0068, 0x0069, 0x006a,
+	0x006b, 0x006c, 0x006d, 0x006e, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x010__ */
+	0x0000, 0x0000, 0x006f, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01b__ */
+	0x0070, 0x0071, 0x0000, 0x0072, 0x0073, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01e__ */
+	0x0074, 0x0075, 0x0076, 0x0077, 0x0078, 0x0079, 0x007a, 0x007b,
+	0x007c, 0x007d, 0x007e, 0x007f, 0x0080, 0x0081, 0x0082, 0x0083,
+	/* Node for range 0x01f__ */
+	0x0084, 0x0085, 0x0086, 0x0087, 0x0088, 0x0089, 0x008a, 0x008b,
+	0x008c, 0x008d, 0x008e, 0x008f, 0x0090, 0x0091, 0x0092, 0x0093,
+	/* Node for range 0x020__ */
+	0x0094, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x021__ */
+	0x0000, 0x0000, 0x0095, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0096, 0x0097, 0x0000, 0x0098, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x022__ */
+	0x0099, 0x0000, 0x009a, 0x0000, 0x009b, 0x0000, 0x009c, 0x009d,
+	0x009e, 0x0000, 0x009f, 0x0000, 0x0000, 0x0000, 0x00a0, 0x0000,
+	/* Node for range 0x023__ */
+	0x0000, 0x0000, 0x00a1, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x02a__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x00a2, 0x0000, 0x0000,
+	/* Node for range 0x030__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x00a3, 0x00a4, 0x00a5, 0x00a6,
+	0x0000, 0x00a7, 0x00a8, 0x00a9, 0x00aa, 0x00ab, 0x0000, 0x00ac,
+	/* Node for range 0x0f9__ */
+	0x00ad, 0x00ae, 0x00af, 0x00b0, 0x00b1, 0x00b2, 0x00b3, 0x00b4,
+	0x00b5, 0x00b6, 0x00b7, 0x00b8, 0x00b9, 0x00ba, 0x00bb, 0x00bc,
+	/* Node for range 0x0fa__ */
+	0x00bd, 0x00be, 0x00bf, 0x00c0, 0x00c1, 0x00c2, 0x00c3, 0x00c4,
+	0x00c5, 0x00c6, 0x00c7, 0x00c8, 0x00c9, 0x00ca, 0x0000, 0x0000,
+	/* Node for range 0x0fb__ */
+	0x0000, 0x00cb, 0x00cc, 0x00cd, 0x00ce, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x110__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x00cf, 0x00d0, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x111__ */
+	0x0000, 0x0000, 0x00d1, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x113__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x00d2, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x114__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x00d3, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x115__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x00d4, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1d1__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x00d5, 0x00d6, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x00d7, 0x00d8, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x2f8__ */
+	0x00d9, 0x00da, 0x00db, 0x00dc, 0x00dd, 0x00de, 0x00df, 0x00e0,
+	0x00e1, 0x00e2, 0x00e3, 0x00e4, 0x00e5, 0x00e6, 0x00e7, 0x00e8,
+	/* Node for range 0x2f9__ */
+	0x00e9, 0x00ea, 0x00eb, 0x00ec, 0x00ed, 0x00ee, 0x00ef, 0x00f0,
+	0x00f1, 0x00f2, 0x00f3, 0x00f4, 0x00f5, 0x00f6, 0x00f7, 0x00f8,
+	/* Node for range 0x2fa__ */
+	0x00f9, 0x00fa, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x000c_ */
+	0x0002, 0x0012, 0x0022, 0x0032, 0x0042, 0x0052, 0x0000, 0x0062,
+	0x0072, 0x0082, 0x0092, 0x00a2, 0x00b2, 0x00c2, 0x00d2, 0x00e2,
+	/* Node for range 0x000d_ */
+	0x0000, 0x00f2, 0x0102, 0x0112, 0x0122, 0x0132, 0x0142, 0x0000,
+	0x0000, 0x0152, 0x0162, 0x0172, 0x0182, 0x0192, 0x0000, 0x0000,
+	/* Node for range 0x000e_ */
+	0x01a2, 0x01b2, 0x01c2, 0x01d2, 0x01e2, 0x01f2, 0x0000, 0x0202,
+	0x0212, 0x0222, 0x0232, 0x0242, 0x0252, 0x0262, 0x0272, 0x0282,
+	/* Node for range 0x000f_ */
+	0x0000, 0x0292, 0x02a2, 0x02b2, 0x02c2, 0x02d2, 0x02e2, 0x0000,
+	0x0000, 0x02f2, 0x0302, 0x0312, 0x0322, 0x0332, 0x0000, 0x0342,
+	/* Node for range 0x0010_ */
+	0x0352, 0x0362, 0x0372, 0x0382, 0x0392, 0x03a2, 0x03b2, 0x03c2,
+	0x03d2, 0x03e2, 0x03f2, 0x0402, 0x0412, 0x0422, 0x0432, 0x0442,
+	/* Node for range 0x0011_ */
+	0x0000, 0x0000, 0x0452, 0x0462, 0x0472, 0x0482, 0x0492, 0x04a2,
+	0x04b2, 0x04c2, 0x04d2, 0x04e2, 0x04f2, 0x0502, 0x0512, 0x0522,
+	/* Node for range 0x0012_ */
+	0x0532, 0x0542, 0x0552, 0x0562, 0x0572, 0x0582, 0x0000, 0x0000,
+	0x0592, 0x05a2, 0x05b2, 0x05c2, 0x05d2, 0x05e2, 0x05f2, 0x0602,
+	/* Node for range 0x0013_ */
+	0x0612, 0x0000, 0x0000, 0x0000, 0x0622, 0x0632, 0x0642, 0x0652,
+	0x0000, 0x0662, 0x0672, 0x0682, 0x0692, 0x06a2, 0x06b2, 0x0000,
+	/* Node for range 0x0014_ */
+	0x0000, 0x0000, 0x0000, 0x06c2, 0x06d2, 0x06e2, 0x06f2, 0x0702,
+	0x0712, 0x0000, 0x0000, 0x0000, 0x0722, 0x0732, 0x0742, 0x0752,
+	/* Node for range 0x0015_ */
+	0x0762, 0x0772, 0x0000, 0x0000, 0x0782, 0x0792, 0x07a2, 0x07b2,
+	0x07c2, 0x07d2, 0x07e2, 0x07f2, 0x0802, 0x0812, 0x0822, 0x0832,
+	/* Node for range 0x0016_ */
+	0x0842, 0x0852, 0x0862, 0x0872, 0x0882, 0x0892, 0x0000, 0x0000,
+	0x08a2, 0x08b2, 0x08c2, 0x08d2, 0x08e2, 0x08f2, 0x0902, 0x0912,
+	/* Node for range 0x0017_ */
+	0x0922, 0x0932, 0x0942, 0x0952, 0x0962, 0x0972, 0x0982, 0x0992,
+	0x09a2, 0x09b2, 0x09c2, 0x09d2, 0x09e2, 0x09f2, 0x0a02, 0x0000,
+	/* Node for range 0x001a_ */
+	0x0a12, 0x0a22, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0a32,
+	/* Node for range 0x001b_ */
+	0x0a42, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x001c_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0a52, 0x0a62, 0x0a72,
+	/* Node for range 0x001d_ */
+	0x0a82, 0x0a92, 0x0aa2, 0x0ab2, 0x0ac2, 0x0ad3, 0x0aeb, 0x0b03,
+	0x0b1b, 0x0b33, 0x0b4b, 0x0b63, 0x0b7b, 0x0000, 0x0b93, 0x0bab,
+	/* Node for range 0x001e_ */
+	0x0bc3, 0x0bdb, 0x0bf2, 0x0c02, 0x0000, 0x0000, 0x0c12, 0x0c22,
+	0x0c32, 0x0c42, 0x0c52, 0x0c62, 0x0c73, 0x0c8b, 0x0ca2, 0x0cb2,
+	/* Node for range 0x001f_ */
+	0x0cc2, 0x0000, 0x0000, 0x0000, 0x0cd2, 0x0ce2, 0x0000, 0x0000,
+	0x0cf2, 0x0d02, 0x0d13, 0x0d2b, 0x0d42, 0x0d52, 0x0d62, 0x0d72,
+	/* Node for range 0x0020_ */
+	0x0d82, 0x0d92, 0x0da2, 0x0db2, 0x0dc2, 0x0dd2, 0x0de2, 0x0df2,
+	0x0e02, 0x0e12, 0x0e22, 0x0e32, 0x0e42, 0x0e52, 0x0e62, 0x0e72,
+	/* Node for range 0x0021_ */
+	0x0e82, 0x0e92, 0x0ea2, 0x0eb2, 0x0ec2, 0x0ed2, 0x0ee2, 0x0ef2,
+	0x0f02, 0x0f12, 0x0f22, 0x0f32, 0x0000, 0x0000, 0x0f42, 0x0f52,
+	/* Node for range 0x0022_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0f62, 0x0f72,
+	0x0f82, 0x0f92, 0x0fa3, 0x0fbb, 0x0fd3, 0x0feb, 0x1002, 0x1012,
+	/* Node for range 0x0023_ */
+	0x1023, 0x103b, 0x1052, 0x1062, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0034_ */
+	0x1071, 0x1079, 0x0000, 0x1081, 0x108a, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0037_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x1099, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x10a1, 0x0000,
+	/* Node for range 0x0038_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x10aa, 0x10ba, 0x10c9,
+	0x10d2, 0x10e2, 0x10f2, 0x0000, 0x1102, 0x0000, 0x1112, 0x1122,
+	/* Node for range 0x0039_ */
+	0x1133, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x003a_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x114a, 0x115a, 0x116a, 0x117a, 0x118a, 0x119a,
+	/* Node for range 0x003b_ */
+	0x11ab, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x003c_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x11c2, 0x11d2, 0x11e2, 0x11f2, 0x1202, 0x0000,
+	/* Node for range 0x003d_ */
+	0x0000, 0x0000, 0x0000, 0x1212, 0x1222, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0040_ */
+	0x1232, 0x1242, 0x0000, 0x1252, 0x0000, 0x0000, 0x0000, 0x1262,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x1272, 0x1282, 0x1292, 0x0000,
+	/* Node for range 0x0041_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x12a2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0043_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x12b2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0045_ */
+	0x12c2, 0x12d2, 0x0000, 0x12e2, 0x0000, 0x0000, 0x0000, 0x12f2,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x1302, 0x1312, 0x1322, 0x0000,
+	/* Node for range 0x0047_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x1332, 0x1342,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x004c_ */
+	0x0000, 0x1352, 0x1362, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x004d_ */
+	0x1372, 0x1382, 0x1392, 0x13a2, 0x0000, 0x0000, 0x13b2, 0x13c2,
+	0x0000, 0x0000, 0x13d2, 0x13e2, 0x13f2, 0x1402, 0x1412, 0x1422,
+	/* Node for range 0x004e_ */
+	0x0000, 0x0000, 0x1432, 0x1442, 0x1452, 0x1462, 0x1472, 0x1482,
+	0x0000, 0x0000, 0x1492, 0x14a2, 0x14b2, 0x14c2, 0x14d2, 0x14e2,
+	/* Node for range 0x004f_ */
+	0x14f2, 0x1502, 0x1512, 0x1522, 0x1532, 0x1542, 0x0000, 0x0000,
+	0x1552, 0x1562, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0062_ */
+	0x0000, 0x0000, 0x1572, 0x1582, 0x1592, 0x15a2, 0x15b2, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x006c_ */
+	0x15c2, 0x0000, 0x15d2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x006d_ */
+	0x0000, 0x0000, 0x0000, 0x15e2, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0092_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x15f2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0093_ */
+	0x0000, 0x1602, 0x0000, 0x0000, 0x1612, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0095_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x1622, 0x1632, 0x1642, 0x1652, 0x1662, 0x1672, 0x1682, 0x1692,
+	/* Node for range 0x009c_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x16a2, 0x16b2, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x009d_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x16c2, 0x16d2, 0x0000, 0x16e2,
+	/* Node for range 0x00a3_ */
+	0x0000, 0x0000, 0x0000, 0x16f2, 0x0000, 0x0000, 0x1702, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00a5_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x1712, 0x1722, 0x1732, 0x0000, 0x0000, 0x1742, 0x0000,
+	/* Node for range 0x00b4_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x1752, 0x0000, 0x0000, 0x1762, 0x1772, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00b5_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x1782, 0x1792, 0x0000, 0x0000,
+	/* Node for range 0x00b9_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x17a2, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00bc_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x17b2, 0x17c2, 0x17d2, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00c4_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x17e2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00cc_ */
+	0x17f2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x1802,
+	0x1812, 0x0000, 0x1822, 0x1833, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00d4_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x184a, 0x185a, 0x186a, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00dd_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x187a, 0x0000, 0x188a, 0x189b, 0x18b2, 0x0000,
+	/* Node for range 0x00f4_ */
+	0x0000, 0x0000, 0x0000, 0x18c2, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x18d2, 0x0000, 0x0000,
+	/* Node for range 0x00f5_ */
+	0x0000, 0x0000, 0x18e2, 0x0000, 0x0000, 0x0000, 0x0000, 0x18f2,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x1902, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00f6_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x1912, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00f7_ */
+	0x0000, 0x0000, 0x0000, 0x1922, 0x0000, 0x1932, 0x1942, 0x0000,
+	0x1952, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00f8_ */
+	0x0000, 0x1962, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00f9_ */
+	0x0000, 0x0000, 0x0000, 0x1972, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x1982, 0x0000, 0x0000,
+	/* Node for range 0x00fa_ */
+	0x0000, 0x0000, 0x1992, 0x0000, 0x0000, 0x0000, 0x0000, 0x19a2,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x19b2, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x00fb_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x19c2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0102_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x19d2, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01b0_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x19e2, 0x0000,
+	0x19f2, 0x0000, 0x1a02, 0x0000, 0x1a12, 0x0000, 0x1a22, 0x0000,
+	/* Node for range 0x01b1_ */
+	0x0000, 0x0000, 0x1a32, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01b3_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x1a42, 0x0000, 0x1a52, 0x0000, 0x0000,
+	/* Node for range 0x01b4_ */
+	0x1a62, 0x1a72, 0x0000, 0x1a82, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01e0_ */
+	0x1a92, 0x1aa2, 0x1ab2, 0x1ac2, 0x1ad2, 0x1ae2, 0x1af2, 0x1b02,
+	0x1b13, 0x1b2b, 0x1b42, 0x1b52, 0x1b62, 0x1b72, 0x1b82, 0x1b92,
+	/* Node for range 0x01e1_ */
+	0x1ba2, 0x1bb2, 0x1bc2, 0x1bd2, 0x1be3, 0x1bfb, 0x1c13, 0x1c2b,
+	0x1c42, 0x1c52, 0x1c62, 0x1c72, 0x1c83, 0x1c9b, 0x1cb2, 0x1cc2,
+	/* Node for range 0x01e2_ */
+	0x1cd2, 0x1ce2, 0x1cf2, 0x1d02, 0x1d12, 0x1d22, 0x1d32, 0x1d42,
+	0x1d52, 0x1d62, 0x1d72, 0x1d82, 0x1d92, 0x1da2, 0x1db3, 0x1dcb,
+	/* Node for range 0x01e3_ */
+	0x1de2, 0x1df2, 0x1e02, 0x1e12, 0x1e22, 0x1e32, 0x1e42, 0x1e52,
+	0x1e63, 0x1e7b, 0x1e92, 0x1ea2, 0x1eb2, 0x1ec2, 0x1ed2, 0x1ee2,
+	/* Node for range 0x01e4_ */
+	0x1ef2, 0x1f02, 0x1f12, 0x1f22, 0x1f32, 0x1f42, 0x1f52, 0x1f62,
+	0x1f72, 0x1f82, 0x1f92, 0x1fa2, 0x1fb3, 0x1fcb, 0x1fe3, 0x1ffb,
+	/* Node for range 0x01e5_ */
+	0x2013, 0x202b, 0x2043, 0x205b, 0x2072, 0x2082, 0x2092, 0x20a2,
+	0x20b2, 0x20c2, 0x20d2, 0x20e2, 0x20f3, 0x210b, 0x2122, 0x2132,
+	/* Node for range 0x01e6_ */
+	0x2142, 0x2152, 0x2162, 0x2172, 0x2183, 0x219b, 0x21b3, 0x21cb,
+	0x21e3, 0x21fb, 0x2212, 0x2222, 0x2232, 0x2242, 0x2252, 0x2262,
+	/* Node for range 0x01e7_ */
+	0x2272, 0x2282, 0x2292, 0x22a2, 0x22b2, 0x22c2, 0x22d2, 0x22e2,
+	0x22f3, 0x230b, 0x2323, 0x233b, 0x2352, 0x2362, 0x2372, 0x2382,
+	/* Node for range 0x01e8_ */
+	0x2392, 0x23a2, 0x23b2, 0x23c2, 0x23d2, 0x23e2, 0x23f2, 0x2402,
+	0x2412, 0x2422, 0x2432, 0x2442, 0x2452, 0x2462, 0x2472, 0x2482,
+	/* Node for range 0x01e9_ */
+	0x2492, 0x24a2, 0x24b2, 0x24c2, 0x24d2, 0x24e2, 0x24f2, 0x2502,
+	0x2512, 0x2522, 0x0000, 0x2532, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01ea_ */
+	0x2542, 0x2552, 0x2562, 0x2572, 0x2583, 0x259b, 0x25b3, 0x25cb,
+	0x25e3, 0x25fb, 0x2613, 0x262b, 0x2643, 0x265b, 0x2673, 0x268b,
+	/* Node for range 0x01eb_ */
+	0x26a3, 0x26bb, 0x26d3, 0x26eb, 0x2703, 0x271b, 0x2733, 0x274b,
+	0x2762, 0x2772, 0x2782, 0x2792, 0x27a2, 0x27b2, 0x27c3, 0x27db,
+	/* Node for range 0x01ec_ */
+	0x27f3, 0x280b, 0x2823, 0x283b, 0x2853, 0x286b, 0x2883, 0x289b,
+	0x28b2, 0x28c2, 0x28d2, 0x28e2, 0x28f2, 0x2902, 0x2912, 0x2922,
+	/* Node for range 0x01ed_ */
+	0x2933, 0x294b, 0x2963, 0x297b, 0x2993, 0x29ab, 0x29c3, 0x29db,
+	0x29f3, 0x2a0b, 0x2a23, 0x2a3b, 0x2a53, 0x2a6b, 0x2a83, 0x2a9b,
+	/* Node for range 0x01ee_ */
+	0x2ab3, 0x2acb, 0x2ae3, 0x2afb, 0x2b12, 0x2b22, 0x2b32, 0x2b42,
+	0x2b53, 0x2b6b, 0x2b83, 0x2b9b, 0x2bb3, 0x2bcb, 0x2be3, 0x2bfb,
+	/* Node for range 0x01ef_ */
+	0x2c13, 0x2c2b, 0x2c42, 0x2c52, 0x2c62, 0x2c72, 0x2c82, 0x2c92,
+	0x2ca2, 0x2cb2, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01f0_ */
+	0x2cc2, 0x2cd2, 0x2ce3, 0x2cfb, 0x2d13, 0x2d2b, 0x2d43, 0x2d5b,
+	0x2d72, 0x2d82, 0x2d93, 0x2dab, 0x2dc3, 0x2ddb, 0x2df3, 0x2e0b,
+	/* Node for range 0x01f1_ */
+	0x2e22, 0x2e32, 0x2e43, 0x2e5b, 0x2e73, 0x2e8b, 0x0000, 0x0000,
+	0x2ea2, 0x2eb2, 0x2ec3, 0x2edb, 0x2ef3, 0x2f0b, 0x0000, 0x0000,
+	/* Node for range 0x01f2_ */
+	0x2f22, 0x2f32, 0x2f43, 0x2f5b, 0x2f73, 0x2f8b, 0x2fa3, 0x2fbb,
+	0x2fd2, 0x2fe2, 0x2ff3, 0x300b, 0x3023, 0x303b, 0x3053, 0x306b,
+	/* Node for range 0x01f3_ */
+	0x3082, 0x3092, 0x30a3, 0x30bb, 0x30d3, 0x30eb, 0x3103, 0x311b,
+	0x3132, 0x3142, 0x3153, 0x316b, 0x3183, 0x319b, 0x31b3, 0x31cb,
+	/* Node for range 0x01f4_ */
+	0x31e2, 0x31f2, 0x3203, 0x321b, 0x3233, 0x324b, 0x0000, 0x0000,
+	0x3262, 0x3272, 0x3283, 0x329b, 0x32b3, 0x32cb, 0x0000, 0x0000,
+	/* Node for range 0x01f5_ */
+	0x32e2, 0x32f2, 0x3303, 0x331b, 0x3333, 0x334b, 0x3363, 0x337b,
+	0x0000, 0x3392, 0x0000, 0x33a3, 0x0000, 0x33bb, 0x0000, 0x33d3,
+	/* Node for range 0x01f6_ */
+	0x33ea, 0x33fa, 0x340b, 0x3423, 0x343b, 0x3453, 0x346b, 0x3483,
+	0x349a, 0x34aa, 0x34bb, 0x34d3, 0x34eb, 0x3503, 0x351b, 0x3533,
+	/* Node for range 0x01f7_ */
+	0x354a, 0x355a, 0x356a, 0x357a, 0x358a, 0x359a, 0x35aa, 0x35ba,
+	0x35ca, 0x35da, 0x35ea, 0x35fa, 0x360a, 0x361a, 0x0000, 0x0000,
+	/* Node for range 0x01f8_ */
+	0x362b, 0x3643, 0x365c, 0x367c, 0x369c, 0x36bc, 0x36dc, 0x36fc,
+	0x371b, 0x3733, 0x374c, 0x376c, 0x378c, 0x37ac, 0x37cc, 0x37ec,
+	/* Node for range 0x01f9_ */
+	0x380b, 0x3823, 0x383c, 0x385c, 0x387c, 0x389c, 0x38bc, 0x38dc,
+	0x38fb, 0x3913, 0x392c, 0x394c, 0x396c, 0x398c, 0x39ac, 0x39cc,
+	/* Node for range 0x01fa_ */
+	0x39eb, 0x3a03, 0x3a1c, 0x3a3c, 0x3a5c, 0x3a7c, 0x3a9c, 0x3abc,
+	0x3adb, 0x3af3, 0x3b0c, 0x3b2c, 0x3b4c, 0x3b6c, 0x3b8c, 0x3bac,
+	/* Node for range 0x01fb_ */
+	0x3bca, 0x3bda, 0x3beb, 0x3c02, 0x3c13, 0x0000, 0x3c2a, 0x3c3b,
+	0x3c52, 0x3c62, 0x3c72, 0x3c82, 0x3c92, 0x0000, 0x3ca1, 0x0000,
+	/* Node for range 0x01fc_ */
+	0x0000, 0x3caa, 0x3cbb, 0x3cd2, 0x3ce3, 0x0000, 0x3cfa, 0x3d0b,
+	0x3d22, 0x3d32, 0x3d42, 0x3d52, 0x3d62, 0x3d72, 0x3d82, 0x3d92,
+	/* Node for range 0x01fd_ */
+	0x3da2, 0x3db2, 0x3dc3, 0x3ddb, 0x0000, 0x0000, 0x3df2, 0x3e03,
+	0x3e1a, 0x3e2a, 0x3e3a, 0x3e4a, 0x0000, 0x3e5a, 0x3e6a, 0x3e7a,
+	/* Node for range 0x01fe_ */
+	0x3e8a, 0x3e9a, 0x3eab, 0x3ec3, 0x3eda, 0x3eea, 0x3efa, 0x3f0b,
+	0x3f22, 0x3f32, 0x3f42, 0x3f52, 0x3f62, 0x3f72, 0x3f82, 0x3f91,
+	/* Node for range 0x01ff_ */
+	0x0000, 0x0000, 0x3f9b, 0x3fb2, 0x3fc3, 0x0000, 0x3fda, 0x3feb,
+	0x4002, 0x4012, 0x4022, 0x4032, 0x4042, 0x4051, 0x0000, 0x0000,
+	/* Node for range 0x0200_ */
+	0x4059, 0x4061, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0212_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x4069, 0x0000,
+	0x0000, 0x0000, 0x4071, 0x407a, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0219_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x408a, 0x409a, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x021a_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x40aa, 0x0000,
+	/* Node for range 0x021c_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x40ba, 0x40ca, 0x40da,
+	/* Node for range 0x0220_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x40ea, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x40fa, 0x0000, 0x0000, 0x410a, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0222_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x411a, 0x0000, 0x412a, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0224_ */
+	0x0000, 0x413a, 0x0000, 0x0000, 0x414a, 0x0000, 0x0000, 0x415a,
+	0x0000, 0x416a, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0226_ */
+	0x417a, 0x0000, 0x418a, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x419a, 0x41aa, 0x41ba,
+	/* Node for range 0x0227_ */
+	0x41ca, 0x41da, 0x0000, 0x0000, 0x41ea, 0x41fa, 0x0000, 0x0000,
+	0x420a, 0x421a, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0228_ */
+	0x422a, 0x423a, 0x0000, 0x0000, 0x424a, 0x425a, 0x0000, 0x0000,
+	0x426a, 0x427a, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x022a_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x428a, 0x429a, 0x42aa, 0x42ba,
+	/* Node for range 0x022e_ */
+	0x42ca, 0x42da, 0x42ea, 0x42fa, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x430a, 0x431a, 0x432a, 0x433a, 0x0000, 0x0000,
+	/* Node for range 0x0232_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x4349, 0x4351, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x02ad_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x435a, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0304_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x436a, 0x0000, 0x437a, 0x0000,
+	/* Node for range 0x0305_ */
+	0x438a, 0x0000, 0x439a, 0x0000, 0x43aa, 0x0000, 0x43ba, 0x0000,
+	0x43ca, 0x0000, 0x43da, 0x0000, 0x43ea, 0x0000, 0x43fa, 0x0000,
+	/* Node for range 0x0306_ */
+	0x440a, 0x0000, 0x441a, 0x0000, 0x0000, 0x442a, 0x0000, 0x443a,
+	0x0000, 0x444a, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0307_ */
+	0x445a, 0x446a, 0x0000, 0x447a, 0x448a, 0x0000, 0x449a, 0x44aa,
+	0x0000, 0x44ba, 0x44ca, 0x0000, 0x44da, 0x44ea, 0x0000, 0x0000,
+	/* Node for range 0x0309_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x44fa, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x450a, 0x0000,
+	/* Node for range 0x030a_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x451a, 0x0000, 0x452a, 0x0000,
+	/* Node for range 0x030b_ */
+	0x453a, 0x0000, 0x454a, 0x0000, 0x455a, 0x0000, 0x456a, 0x0000,
+	0x457a, 0x0000, 0x458a, 0x0000, 0x459a, 0x0000, 0x45aa, 0x0000,
+	/* Node for range 0x030c_ */
+	0x45ba, 0x0000, 0x45ca, 0x0000, 0x0000, 0x45da, 0x0000, 0x45ea,
+	0x0000, 0x45fa, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x030d_ */
+	0x460a, 0x461a, 0x0000, 0x462a, 0x463a, 0x0000, 0x464a, 0x465a,
+	0x0000, 0x466a, 0x467a, 0x0000, 0x468a, 0x469a, 0x0000, 0x0000,
+	/* Node for range 0x030f_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x46aa, 0x0000, 0x0000, 0x46ba,
+	0x46ca, 0x46da, 0x46ea, 0x0000, 0x0000, 0x0000, 0x46fa, 0x0000,
+	/* Node for range 0x0f90_ */
+	0x4709, 0x4711, 0x4719, 0x4721, 0x4729, 0x4731, 0x4739, 0x4741,
+	0x4749, 0x4751, 0x4759, 0x4761, 0x4769, 0x4771, 0x4779, 0x4781,
+	/* Node for range 0x0f91_ */
+	0x4789, 0x4791, 0x4799, 0x47a1, 0x47a9, 0x47b1, 0x47b9, 0x47c1,
+	0x47c9, 0x47d1, 0x47d9, 0x47e1, 0x47e9, 0x47f1, 0x47f9, 0x4801,
+	/* Node for range 0x0f92_ */
+	0x4809, 0x4811, 0x4819, 0x4821, 0x4829, 0x4831, 0x4839, 0x4841,
+	0x4849, 0x4851, 0x4859, 0x4861, 0x4869, 0x4871, 0x4879, 0x4881,
+	/* Node for range 0x0f93_ */
+	0x4889, 0x4891, 0x4899, 0x48a1, 0x48a9, 0x48b1, 0x48b9, 0x48c1,
+	0x48c9, 0x48d1, 0x48d9, 0x48e1, 0x48e9, 0x48f1, 0x48f9, 0x4901,
+	/* Node for range 0x0f94_ */
+	0x4909, 0x4911, 0x4919, 0x4921, 0x4929, 0x4931, 0x4939, 0x4941,
+	0x4949, 0x4951, 0x4959, 0x4961, 0x4969, 0x4971, 0x4979, 0x4981,
+	/* Node for range 0x0f95_ */
+	0x4989, 0x4991, 0x4999, 0x49a1, 0x49a9, 0x49b1, 0x49b9, 0x49c1,
+	0x49c9, 0x49d1, 0x49d9, 0x49e1, 0x49e9, 0x49f1, 0x49f9, 0x4a01,
+	/* Node for range 0x0f96_ */
+	0x4a09, 0x4a11, 0x4a19, 0x4a21, 0x4a29, 0x4a31, 0x4a39, 0x4a41,
+	0x4a49, 0x4a51, 0x4a59, 0x4a61, 0x4a69, 0x4a71, 0x4a79, 0x4a81,
+	/* Node for range 0x0f97_ */
+	0x4a89, 0x4a91, 0x4a99, 0x4aa1, 0x4aa9, 0x4ab1, 0x4ab9, 0x4ac1,
+	0x4ac9, 0x4ad1, 0x4ad9, 0x4ae1, 0x4ae9, 0x4af1, 0x4af9, 0x4b01,
+	/* Node for range 0x0f98_ */
+	0x4b09, 0x4b11, 0x4b19, 0x4b21, 0x4b29, 0x4b31, 0x4b39, 0x4b41,
+	0x4b49, 0x4b51, 0x4b59, 0x4b61, 0x4b69, 0x4b71, 0x4b79, 0x4b81,
+	/* Node for range 0x0f99_ */
+	0x4b89, 0x4b91, 0x4b99, 0x4ba1, 0x4ba9, 0x4bb1, 0x4bb9, 0x4bc1,
+	0x4bc9, 0x4bd1, 0x4bd9, 0x4be1, 0x4be9, 0x4bf1, 0x4bf9, 0x4c01,
+	/* Node for range 0x0f9a_ */
+	0x4c09, 0x4c11, 0x4c19, 0x4c21, 0x4c29, 0x4c31, 0x4c39, 0x4c41,
+	0x4c49, 0x4c51, 0x4c59, 0x4c61, 0x4c69, 0x4c71, 0x4c79, 0x4c81,
+	/* Node for range 0x0f9b_ */
+	0x4c89, 0x4c91, 0x4c99, 0x4ca1, 0x4ca9, 0x4cb1, 0x4cb9, 0x4cc1,
+	0x4cc9, 0x4cd1, 0x4cd9, 0x4ce1, 0x4ce9, 0x4cf1, 0x4cf9, 0x4d01,
+	/* Node for range 0x0f9c_ */
+	0x4d09, 0x4d11, 0x4d19, 0x4d21, 0x4d29, 0x4d31, 0x4d39, 0x4d41,
+	0x4d49, 0x4d51, 0x4d59, 0x4d61, 0x4d69, 0x4d71, 0x4d79, 0x4d81,
+	/* Node for range 0x0f9d_ */
+	0x4d89, 0x4d91, 0x4d99, 0x4da1, 0x4da9, 0x4db1, 0x4db9, 0x4dc1,
+	0x4dc9, 0x4dd1, 0x4dd9, 0x4de1, 0x4de9, 0x4df1, 0x4df9, 0x4e01,
+	/* Node for range 0x0f9e_ */
+	0x4e09, 0x4e11, 0x4e19, 0x4e21, 0x4e29, 0x4e31, 0x4e39, 0x4e41,
+	0x4e49, 0x4e51, 0x4e59, 0x4e61, 0x4e69, 0x4e71, 0x4e79, 0x4e81,
+	/* Node for range 0x0f9f_ */
+	0x4e89, 0x4e91, 0x4e99, 0x4ea1, 0x4ea9, 0x4eb1, 0x4eb9, 0x4ec1,
+	0x4ec9, 0x4ed1, 0x4ed9, 0x4ee1, 0x4ee9, 0x4ef1, 0x4ef9, 0x4f01,
+	/* Node for range 0x0fa0_ */
+	0x4f09, 0x4f11, 0x4f19, 0x4f21, 0x4f29, 0x4f31, 0x4f39, 0x4f41,
+	0x4f49, 0x4f51, 0x4f59, 0x4f61, 0x4f69, 0x4f71, 0x0000, 0x0000,
+	/* Node for range 0x0fa1_ */
+	0x4f79, 0x0000, 0x4f81, 0x0000, 0x0000, 0x4f89, 0x4f91, 0x4f99,
+	0x4fa1, 0x4fa9, 0x4fb1, 0x4fb9, 0x4fc1, 0x4fc9, 0x4fd1, 0x0000,
+	/* Node for range 0x0fa2_ */
+	0x4fd9, 0x0000, 0x4fe1, 0x0000, 0x0000, 0x4fe9, 0x4ff1, 0x0000,
+	0x0000, 0x0000, 0x4ff9, 0x5001, 0x5009, 0x5011, 0x5019, 0x5021,
+	/* Node for range 0x0fa3_ */
+	0x5029, 0x5031, 0x5039, 0x5041, 0x5049, 0x5051, 0x5059, 0x5061,
+	0x5069, 0x5071, 0x5079, 0x5081, 0x5089, 0x5091, 0x5099, 0x50a1,
+	/* Node for range 0x0fa4_ */
+	0x50a9, 0x50b1, 0x50b9, 0x50c1, 0x50c9, 0x50d1, 0x50d9, 0x50e1,
+	0x50e9, 0x50f1, 0x50f9, 0x5101, 0x5109, 0x5111, 0x5119, 0x5121,
+	/* Node for range 0x0fa5_ */
+	0x5129, 0x5131, 0x5139, 0x5141, 0x5149, 0x5151, 0x5159, 0x5161,
+	0x5169, 0x5171, 0x5179, 0x5181, 0x5189, 0x5191, 0x5199, 0x51a1,
+	/* Node for range 0x0fa6_ */
+	0x51a9, 0x51b1, 0x51b9, 0x51c1, 0x51c9, 0x51d1, 0x51d9, 0x51e1,
+	0x51e9, 0x51f1, 0x51f9, 0x5201, 0x5209, 0x5211, 0x0000, 0x0000,
+	/* Node for range 0x0fa7_ */
+	0x5219, 0x5221, 0x5229, 0x5231, 0x5239, 0x5241, 0x5249, 0x5251,
+	0x5259, 0x5261, 0x5269, 0x5271, 0x5279, 0x5281, 0x5289, 0x5291,
+	/* Node for range 0x0fa8_ */
+	0x5299, 0x52a1, 0x52a9, 0x52b1, 0x52b9, 0x52c1, 0x52c9, 0x52d1,
+	0x52d9, 0x52e1, 0x52e9, 0x52f1, 0x52f9, 0x5301, 0x5309, 0x5311,
+	/* Node for range 0x0fa9_ */
+	0x5319, 0x5321, 0x5329, 0x5331, 0x5339, 0x5341, 0x5349, 0x5351,
+	0x5359, 0x5361, 0x5369, 0x5371, 0x5379, 0x5381, 0x5389, 0x5391,
+	/* Node for range 0x0faa_ */
+	0x5399, 0x53a1, 0x53a9, 0x53b1, 0x53b9, 0x53c1, 0x53c9, 0x53d1,
+	0x53d9, 0x53e1, 0x53e9, 0x53f1, 0x53f9, 0x5401, 0x5409, 0x5411,
+	/* Node for range 0x0fab_ */
+	0x5419, 0x5421, 0x5429, 0x5431, 0x5439, 0x5441, 0x5449, 0x5451,
+	0x5459, 0x5461, 0x5469, 0x5471, 0x5479, 0x5481, 0x5489, 0x5491,
+	/* Node for range 0x0fac_ */
+	0x5499, 0x54a1, 0x54a9, 0x54b1, 0x54b9, 0x54c1, 0x54c9, 0x54d1,
+	0x54d9, 0x54e1, 0x54e9, 0x54f1, 0x54f9, 0x5501, 0x5509, 0x5511,
+	/* Node for range 0x0fad_ */
+	0x5519, 0x5521, 0x5529, 0x5531, 0x5539, 0x5541, 0x5549, 0x5551,
+	0x5559, 0x5561, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0fb1_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x556a, 0x0000, 0x557a,
+	/* Node for range 0x0fb2_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x558a, 0x559a, 0x55ab, 0x55c3, 0x55da, 0x55ea,
+	/* Node for range 0x0fb3_ */
+	0x55fa, 0x560a, 0x561a, 0x562a, 0x563a, 0x564a, 0x565a, 0x0000,
+	0x566a, 0x567a, 0x568a, 0x569a, 0x56aa, 0x0000, 0x56ba, 0x0000,
+	/* Node for range 0x0fb4_ */
+	0x56ca, 0x56da, 0x0000, 0x56ea, 0x56fa, 0x0000, 0x570a, 0x571a,
+	0x572a, 0x573a, 0x574a, 0x575a, 0x576a, 0x577a, 0x578a, 0x0000,
+	/* Node for range 0x1109_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x579a, 0x0000, 0x57aa, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x110a_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x57ba, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1112_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x57ca, 0x57da,
+	/* Node for range 0x1134_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x57ea, 0x57fa, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x114b_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x580a, 0x581a, 0x0000, 0x582a, 0x0000,
+	/* Node for range 0x115b_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x583a, 0x584a, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1d15_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x585a, 0x586a,
+	/* Node for range 0x1d16_ */
+	0x587b, 0x5893, 0x58ab, 0x58c3, 0x58db, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1d1b_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x58f2, 0x5902, 0x5913, 0x592b, 0x5943,
+	/* Node for range 0x1d1c_ */
+	0x595b, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x2f80_ */
+	0x5971, 0x5979, 0x5981, 0x5989, 0x5991, 0x5999, 0x59a1, 0x59a9,
+	0x59b1, 0x59b9, 0x59c1, 0x59c9, 0x59d1, 0x59d9, 0x59e1, 0x59e9,
+	/* Node for range 0x2f81_ */
+	0x59f1, 0x59f9, 0x5a01, 0x5a09, 0x5a11, 0x5a19, 0x5a21, 0x5a29,
+	0x5a31, 0x5a39, 0x5a41, 0x5a49, 0x5a51, 0x5a59, 0x5a61, 0x5a69,
+	/* Node for range 0x2f82_ */
+	0x5a71, 0x5a79, 0x5a81, 0x5a89, 0x5a91, 0x5a99, 0x5aa1, 0x5aa9,
+	0x5ab1, 0x5ab9, 0x5ac1, 0x5ac9, 0x5ad1, 0x5ad9, 0x5ae1, 0x5ae9,
+	/* Node for range 0x2f83_ */
+	0x5af1, 0x5af9, 0x5b01, 0x5b09, 0x5b11, 0x5b19, 0x5b21, 0x5b29,
+	0x5b31, 0x5b39, 0x5b41, 0x5b49, 0x5b51, 0x5b59, 0x5b61, 0x5b69,
+	/* Node for range 0x2f84_ */
+	0x5b71, 0x5b79, 0x5b81, 0x5b89, 0x5b91, 0x5b99, 0x5ba1, 0x5ba9,
+	0x5bb1, 0x5bb9, 0x5bc1, 0x5bc9, 0x5bd1, 0x5bd9, 0x5be1, 0x5be9,
+	/* Node for range 0x2f85_ */
+	0x5bf1, 0x5bf9, 0x5c01, 0x5c09, 0x5c11, 0x5c19, 0x5c21, 0x5c29,
+	0x5c31, 0x5c39, 0x5c41, 0x5c49, 0x5c51, 0x5c59, 0x5c61, 0x5c69,
+	/* Node for range 0x2f86_ */
+	0x5c71, 0x5c79, 0x5c81, 0x5c89, 0x5c91, 0x5c99, 0x5ca1, 0x5ca9,
+	0x5cb1, 0x5cb9, 0x5cc1, 0x5cc9, 0x5cd1, 0x5cd9, 0x5ce1, 0x5ce9,
+	/* Node for range 0x2f87_ */
+	0x5cf1, 0x5cf9, 0x5d01, 0x5d09, 0x5d11, 0x5d19, 0x5d21, 0x5d29,
+	0x5d31, 0x5d39, 0x5d41, 0x5d49, 0x5d51, 0x5d59, 0x5d61, 0x5d69,
+	/* Node for range 0x2f88_ */
+	0x5d71, 0x5d79, 0x5d81, 0x5d89, 0x5d91, 0x5d99, 0x5da1, 0x5da9,
+	0x5db1, 0x5db9, 0x5dc1, 0x5dc9, 0x5dd1, 0x5dd9, 0x5de1, 0x5de9,
+	/* Node for range 0x2f89_ */
+	0x5df1, 0x5df9, 0x5e01, 0x5e09, 0x5e11, 0x5e19, 0x5e21, 0x5e29,
+	0x5e31, 0x5e39, 0x5e41, 0x5e49, 0x5e51, 0x5e59, 0x5e61, 0x5e69,
+	/* Node for range 0x2f8a_ */
+	0x5e71, 0x5e79, 0x5e81, 0x5e89, 0x5e91, 0x5e99, 0x5ea1, 0x5ea9,
+	0x5eb1, 0x5eb9, 0x5ec1, 0x5ec9, 0x5ed1, 0x5ed9, 0x5ee1, 0x5ee9,
+	/* Node for range 0x2f8b_ */
+	0x5ef1, 0x5ef9, 0x5f01, 0x5f09, 0x5f11, 0x5f19, 0x5f21, 0x5f29,
+	0x5f31, 0x5f39, 0x5f41, 0x5f49, 0x5f51, 0x5f59, 0x5f61, 0x5f69,
+	/* Node for range 0x2f8c_ */
+	0x5f71, 0x5f79, 0x5f81, 0x5f89, 0x5f91, 0x5f99, 0x5fa1, 0x5fa9,
+	0x5fb1, 0x5fb9, 0x5fc1, 0x5fc9, 0x5fd1, 0x5fd9, 0x5fe1, 0x5fe9,
+	/* Node for range 0x2f8d_ */
+	0x5ff1, 0x5ff9, 0x6001, 0x6009, 0x6011, 0x6019, 0x6021, 0x6029,
+	0x6031, 0x6039, 0x6041, 0x6049, 0x6051, 0x6059, 0x6061, 0x6069,
+	/* Node for range 0x2f8e_ */
+	0x6071, 0x6079, 0x6081, 0x6089, 0x6091, 0x6099, 0x60a1, 0x60a9,
+	0x60b1, 0x60b9, 0x60c1, 0x60c9, 0x60d1, 0x60d9, 0x60e1, 0x60e9,
+	/* Node for range 0x2f8f_ */
+	0x60f1, 0x60f9, 0x6101, 0x6109, 0x6111, 0x6119, 0x6121, 0x6129,
+	0x6131, 0x6139, 0x6141, 0x6149, 0x6151, 0x6159, 0x6161, 0x6169,
+	/* Node for range 0x2f90_ */
+	0x6171, 0x6179, 0x6181, 0x6189, 0x6191, 0x6199, 0x61a1, 0x61a9,
+	0x61b1, 0x61b9, 0x61c1, 0x61c9, 0x61d1, 0x61d9, 0x61e1, 0x61e9,
+	/* Node for range 0x2f91_ */
+	0x61f1, 0x61f9, 0x6201, 0x6209, 0x6211, 0x6219, 0x6221, 0x6229,
+	0x6231, 0x6239, 0x6241, 0x6249, 0x6251, 0x6259, 0x6261, 0x6269,
+	/* Node for range 0x2f92_ */
+	0x6271, 0x6279, 0x6281, 0x6289, 0x6291, 0x6299, 0x62a1, 0x62a9,
+	0x62b1, 0x62b9, 0x62c1, 0x62c9, 0x62d1, 0x62d9, 0x62e1, 0x62e9,
+	/* Node for range 0x2f93_ */
+	0x62f1, 0x62f9, 0x6301, 0x6309, 0x6311, 0x6319, 0x6321, 0x6329,
+	0x6331, 0x6339, 0x6341, 0x6349, 0x6351, 0x6359, 0x6361, 0x6369,
+	/* Node for range 0x2f94_ */
+	0x6371, 0x6379, 0x6381, 0x6389, 0x6391, 0x6399, 0x63a1, 0x63a9,
+	0x63b1, 0x63b9, 0x63c1, 0x63c9, 0x63d1, 0x63d9, 0x63e1, 0x63e9,
+	/* Node for range 0x2f95_ */
+	0x63f1, 0x63f9, 0x6401, 0x6409, 0x6411, 0x6419, 0x6421, 0x6429,
+	0x6431, 0x6439, 0x6441, 0x6449, 0x6451, 0x6459, 0x6461, 0x6469,
+	/* Node for range 0x2f96_ */
+	0x6471, 0x6479, 0x6481, 0x6489, 0x6491, 0x6499, 0x64a1, 0x64a9,
+	0x64b1, 0x64b9, 0x64c1, 0x64c9, 0x64d1, 0x64d9, 0x64e1, 0x64e9,
+	/* Node for range 0x2f97_ */
+	0x64f1, 0x64f9, 0x6501, 0x6509, 0x6511, 0x6519, 0x6521, 0x6529,
+	0x6531, 0x6539, 0x6541, 0x6549, 0x6551, 0x6559, 0x6561, 0x6569,
+	/* Node for range 0x2f98_ */
+	0x6571, 0x6579, 0x6581, 0x6589, 0x6591, 0x6599, 0x65a1, 0x65a9,
+	0x65b1, 0x65b9, 0x65c1, 0x65c9, 0x65d1, 0x65d9, 0x65e1, 0x65e9,
+	/* Node for range 0x2f99_ */
+	0x65f1, 0x65f9, 0x6601, 0x6609, 0x6611, 0x6619, 0x6621, 0x6629,
+	0x6631, 0x6639, 0x6641, 0x6649, 0x6651, 0x6659, 0x6661, 0x6669,
+	/* Node for range 0x2f9a_ */
+	0x6671, 0x6679, 0x6681, 0x6689, 0x6691, 0x6699, 0x66a1, 0x66a9,
+	0x66b1, 0x66b9, 0x66c1, 0x66c9, 0x66d1, 0x66d9, 0x66e1, 0x66e9,
+	/* Node for range 0x2f9b_ */
+	0x66f1, 0x66f9, 0x6701, 0x6709, 0x6711, 0x6719, 0x6721, 0x6729,
+	0x6731, 0x6739, 0x6741, 0x6749, 0x6751, 0x6759, 0x6761, 0x6769,
+	/* Node for range 0x2f9c_ */
+	0x6771, 0x6779, 0x6781, 0x6789, 0x6791, 0x6799, 0x67a1, 0x67a9,
+	0x67b1, 0x67b9, 0x67c1, 0x67c9, 0x67d1, 0x67d9, 0x67e1, 0x67e9,
+	/* Node for range 0x2f9d_ */
+	0x67f1, 0x67f9, 0x6801, 0x6809, 0x6811, 0x6819, 0x6821, 0x6829,
+	0x6831, 0x6839, 0x6841, 0x6849, 0x6851, 0x6859, 0x6861, 0x6869,
+	/* Node for range 0x2f9e_ */
+	0x6871, 0x6879, 0x6881, 0x6889, 0x6891, 0x6899, 0x68a1, 0x68a9,
+	0x68b1, 0x68b9, 0x68c1, 0x68c9, 0x68d1, 0x68d9, 0x68e1, 0x68e9,
+	/* Node for range 0x2f9f_ */
+	0x68f1, 0x68f9, 0x6901, 0x6909, 0x6911, 0x6919, 0x6921, 0x6929,
+	0x6931, 0x6939, 0x6941, 0x6949, 0x6951, 0x6959, 0x6961, 0x6969,
+	/* Node for range 0x2fa0_ */
+	0x6971, 0x6979, 0x6981, 0x6989, 0x6991, 0x6999, 0x69a1, 0x69a9,
+	0x69b1, 0x69b9, 0x69c1, 0x69c9, 0x69d1, 0x69d9, 0x69e1, 0x69e9,
+	/* Node for range 0x2fa1_ */
+	0x69f1, 0x69f9, 0x6a01, 0x6a09, 0x6a11, 0x6a19, 0x6a21, 0x6a29,
+	0x6a31, 0x6a39, 0x6a41, 0x6a49, 0x6a51, 0x6a59, 0x0000, 0x0000,
+};
+
+static unicode_t apfs_nfd[] = {
+	0x000041, 0x000300, 0x000041, 0x000301, 0x000041, 0x000302,
+	0x000041, 0x000303, 0x000041, 0x000308, 0x000041, 0x00030a,
+	0x000043, 0x000327, 0x000045, 0x000300, 0x000045, 0x000301,
+	0x000045, 0x000302, 0x000045, 0x000308, 0x000049, 0x000300,
+	0x000049, 0x000301, 0x000049, 0x000302, 0x000049, 0x000308,
+	0x00004e, 0x000303, 0x00004f, 0x000300, 0x00004f, 0x000301,
+	0x00004f, 0x000302, 0x00004f, 0x000303, 0x00004f, 0x000308,
+	0x000055, 0x000300, 0x000055, 0x000301, 0x000055, 0x000302,
+	0x000055, 0x000308, 0x000059, 0x000301, 0x000061, 0x000300,
+	0x000061, 0x000301, 0x000061, 0x000302, 0x000061, 0x000303,
+	0x000061, 0x000308, 0x000061, 0x00030a, 0x000063, 0x000327,
+	0x000065, 0x000300, 0x000065, 0x000301, 0x000065, 0x000302,
+	0x000065, 0x000308, 0x000069, 0x000300, 0x000069, 0x000301,
+	0x000069, 0x000302, 0x000069, 0x000308, 0x00006e, 0x000303,
+	0x00006f, 0x000300, 0x00006f, 0x000301, 0x00006f, 0x000302,
+	0x00006f, 0x000303, 0x00006f, 0x000308, 0x000075, 0x000300,
+	0x000075, 0x000301, 0x000075, 0x000302, 0x000075, 0x000308,
+	0x000079, 0x000301, 0x000079, 0x000308, 0x000041, 0x000304,
+	0x000061, 0x000304, 0x000041, 0x000306, 0x000061, 0x000306,
+	0x000041, 0x000328, 0x000061, 0x000328, 0x000043, 0x000301,
+	0x000063, 0x000301, 0x000043, 0x000302, 0x000063, 0x000302,
+	0x000043, 0x000307, 0x000063, 0x000307, 0x000043, 0x00030c,
+	0x000063, 0x00030c, 0x000044, 0x00030c, 0x000064, 0x00030c,
+	0x000045, 0x000304, 0x000065, 0x000304, 0x000045, 0x000306,
+	0x000065, 0x000306, 0x000045, 0x000307, 0x000065, 0x000307,
+	0x000045, 0x000328, 0x000065, 0x000328, 0x000045, 0x00030c,
+	0x000065, 0x00030c, 0x000047, 0x000302, 0x000067, 0x000302,
+	0x000047, 0x000306, 0x000067, 0x000306, 0x000047, 0x000307,
+	0x000067, 0x000307, 0x000047, 0x000327, 0x000067, 0x000327,
+	0x000048, 0x000302, 0x000068, 0x000302, 0x000049, 0x000303,
+	0x000069, 0x000303, 0x000049, 0x000304, 0x000069, 0x000304,
+	0x000049, 0x000306, 0x000069, 0x000306, 0x000049, 0x000328,
+	0x000069, 0x000328, 0x000049, 0x000307, 0x00004a, 0x000302,
+	0x00006a, 0x000302, 0x00004b, 0x000327, 0x00006b, 0x000327,
+	0x00004c, 0x000301, 0x00006c, 0x000301, 0x00004c, 0x000327,
+	0x00006c, 0x000327, 0x00004c, 0x00030c, 0x00006c, 0x00030c,
+	0x00004e, 0x000301, 0x00006e, 0x000301, 0x00004e, 0x000327,
+	0x00006e, 0x000327, 0x00004e, 0x00030c, 0x00006e, 0x00030c,
+	0x00004f, 0x000304, 0x00006f, 0x000304, 0x00004f, 0x000306,
+	0x00006f, 0x000306, 0x00004f, 0x00030b, 0x00006f, 0x00030b,
+	0x000052, 0x000301, 0x000072, 0x000301, 0x000052, 0x000327,
+	0x000072, 0x000327, 0x000052, 0x00030c, 0x000072, 0x00030c,
+	0x000053, 0x000301, 0x000073, 0x000301, 0x000053, 0x000302,
+	0x000073, 0x000302, 0x000053, 0x000327, 0x000073, 0x000327,
+	0x000053, 0x00030c, 0x000073, 0x00030c, 0x000054, 0x000327,
+	0x000074, 0x000327, 0x000054, 0x00030c, 0x000074, 0x00030c,
+	0x000055, 0x000303, 0x000075, 0x000303, 0x000055, 0x000304,
+	0x000075, 0x000304, 0x000055, 0x000306, 0x000075, 0x000306,
+	0x000055, 0x00030a, 0x000075, 0x00030a, 0x000055, 0x00030b,
+	0x000075, 0x00030b, 0x000055, 0x000328, 0x000075, 0x000328,
+	0x000057, 0x000302, 0x000077, 0x000302, 0x000059, 0x000302,
+	0x000079, 0x000302, 0x000059, 0x000308, 0x00005a, 0x000301,
+	0x00007a, 0x000301, 0x00005a, 0x000307, 0x00007a, 0x000307,
+	0x00005a, 0x00030c, 0x00007a, 0x00030c, 0x00004f, 0x00031b,
+	0x00006f, 0x00031b, 0x000055, 0x00031b, 0x000075, 0x00031b,
+	0x000041, 0x00030c, 0x000061, 0x00030c, 0x000049, 0x00030c,
+	0x000069, 0x00030c, 0x00004f, 0x00030c, 0x00006f, 0x00030c,
+	0x000055, 0x00030c, 0x000075, 0x00030c, 0x000055, 0x000308,
+	0x000304, 0x000075, 0x000308, 0x000304, 0x000055, 0x000308,
+	0x000301, 0x000075, 0x000308, 0x000301, 0x000055, 0x000308,
+	0x00030c, 0x000075, 0x000308, 0x00030c, 0x000055, 0x000308,
+	0x000300, 0x000075, 0x000308, 0x000300, 0x000041, 0x000308,
+	0x000304, 0x000061, 0x000308, 0x000304, 0x000041, 0x000307,
+	0x000304, 0x000061, 0x000307, 0x000304, 0x0000c6, 0x000304,
+	0x0000e6, 0x000304, 0x000047, 0x00030c, 0x000067, 0x00030c,
+	0x00004b, 0x00030c, 0x00006b, 0x00030c, 0x00004f, 0x000328,
+	0x00006f, 0x000328, 0x00004f, 0x000328, 0x000304, 0x00006f,
+	0x000328, 0x000304, 0x0001b7, 0x00030c, 0x000292, 0x00030c,
+	0x00006a, 0x00030c, 0x000047, 0x000301, 0x000067, 0x000301,
+	0x00004e, 0x000300, 0x00006e, 0x000300, 0x000041, 0x00030a,
+	0x000301, 0x000061, 0x00030a, 0x000301, 0x0000c6, 0x000301,
+	0x0000e6, 0x000301, 0x0000d8, 0x000301, 0x0000f8, 0x000301,
+	0x000041, 0x00030f, 0x000061, 0x00030f, 0x000041, 0x000311,
+	0x000061, 0x000311, 0x000045, 0x00030f, 0x000065, 0x00030f,
+	0x000045, 0x000311, 0x000065, 0x000311, 0x000049, 0x00030f,
+	0x000069, 0x00030f, 0x000049, 0x000311, 0x000069, 0x000311,
+	0x00004f, 0x00030f, 0x00006f, 0x00030f, 0x00004f, 0x000311,
+	0x00006f, 0x000311, 0x000052, 0x00030f, 0x000072, 0x00030f,
+	0x000052, 0x000311, 0x000072, 0x000311, 0x000055, 0x00030f,
+	0x000075, 0x00030f, 0x000055, 0x000311, 0x000075, 0x000311,
+	0x000053, 0x000326, 0x000073, 0x000326, 0x000054, 0x000326,
+	0x000074, 0x000326, 0x000048, 0x00030c, 0x000068, 0x00030c,
+	0x000041, 0x000307, 0x000061, 0x000307, 0x000045, 0x000327,
+	0x000065, 0x000327, 0x00004f, 0x000308, 0x000304, 0x00006f,
+	0x000308, 0x000304, 0x00004f, 0x000303, 0x000304, 0x00006f,
+	0x000303, 0x000304, 0x00004f, 0x000307, 0x00006f, 0x000307,
+	0x00004f, 0x000307, 0x000304, 0x00006f, 0x000307, 0x000304,
+	0x000059, 0x000304, 0x000079, 0x000304, 0x000300, 0x000301,
+	0x000313, 0x000308, 0x000301, 0x0002b9, 0x00003b, 0x0000a8,
+	0x000301, 0x000391, 0x000301, 0x0000b7, 0x000395, 0x000301,
+	0x000397, 0x000301, 0x000399, 0x000301, 0x00039f, 0x000301,
+	0x0003a5, 0x000301, 0x0003a9, 0x000301, 0x0003b9, 0x000308,
+	0x000301, 0x000399, 0x000308, 0x0003a5, 0x000308, 0x0003b1,
+	0x000301, 0x0003b5, 0x000301, 0x0003b7, 0x000301, 0x0003b9,
+	0x000301, 0x0003c5, 0x000308, 0x000301, 0x0003b9, 0x000308,
+	0x0003c5, 0x000308, 0x0003bf, 0x000301, 0x0003c5, 0x000301,
+	0x0003c9, 0x000301, 0x0003d2, 0x000301, 0x0003d2, 0x000308,
+	0x000415, 0x000300, 0x000415, 0x000308, 0x000413, 0x000301,
+	0x000406, 0x000308, 0x00041a, 0x000301, 0x000418, 0x000300,
+	0x000423, 0x000306, 0x000418, 0x000306, 0x000438, 0x000306,
+	0x000435, 0x000300, 0x000435, 0x000308, 0x000433, 0x000301,
+	0x000456, 0x000308, 0x00043a, 0x000301, 0x000438, 0x000300,
+	0x000443, 0x000306, 0x000474, 0x00030f, 0x000475, 0x00030f,
+	0x000416, 0x000306, 0x000436, 0x000306, 0x000410, 0x000306,
+	0x000430, 0x000306, 0x000410, 0x000308, 0x000430, 0x000308,
+	0x000415, 0x000306, 0x000435, 0x000306, 0x0004d8, 0x000308,
+	0x0004d9, 0x000308, 0x000416, 0x000308, 0x000436, 0x000308,
+	0x000417, 0x000308, 0x000437, 0x000308, 0x000418, 0x000304,
+	0x000438, 0x000304, 0x000418, 0x000308, 0x000438, 0x000308,
+	0x00041e, 0x000308, 0x00043e, 0x000308, 0x0004e8, 0x000308,
+	0x0004e9, 0x000308, 0x00042d, 0x000308, 0x00044d, 0x000308,
+	0x000423, 0x000304, 0x000443, 0x000304, 0x000423, 0x000308,
+	0x000443, 0x000308, 0x000423, 0x00030b, 0x000443, 0x00030b,
+	0x000427, 0x000308, 0x000447, 0x000308, 0x00042b, 0x000308,
+	0x00044b, 0x000308, 0x000627, 0x000653, 0x000627, 0x000654,
+	0x000648, 0x000654, 0x000627, 0x000655, 0x00064a, 0x000654,
+	0x0006d5, 0x000654, 0x0006c1, 0x000654, 0x0006d2, 0x000654,
+	0x000928, 0x00093c, 0x000930, 0x00093c, 0x000933, 0x00093c,
+	0x000915, 0x00093c, 0x000916, 0x00093c, 0x000917, 0x00093c,
+	0x00091c, 0x00093c, 0x000921, 0x00093c, 0x000922, 0x00093c,
+	0x00092b, 0x00093c, 0x00092f, 0x00093c, 0x0009c7, 0x0009be,
+	0x0009c7, 0x0009d7, 0x0009a1, 0x0009bc, 0x0009a2, 0x0009bc,
+	0x0009af, 0x0009bc, 0x000a32, 0x000a3c, 0x000a38, 0x000a3c,
+	0x000a16, 0x000a3c, 0x000a17, 0x000a3c, 0x000a1c, 0x000a3c,
+	0x000a2b, 0x000a3c, 0x000b47, 0x000b56, 0x000b47, 0x000b3e,
+	0x000b47, 0x000b57, 0x000b21, 0x000b3c, 0x000b22, 0x000b3c,
+	0x000b92, 0x000bd7, 0x000bc6, 0x000bbe, 0x000bc7, 0x000bbe,
+	0x000bc6, 0x000bd7, 0x000c46, 0x000c56, 0x000cbf, 0x000cd5,
+	0x000cc6, 0x000cd5, 0x000cc6, 0x000cd6, 0x000cc6, 0x000cc2,
+	0x000cc6, 0x000cc2, 0x000cd5, 0x000d46, 0x000d3e, 0x000d47,
+	0x000d3e, 0x000d46, 0x000d57, 0x000dd9, 0x000dca, 0x000dd9,
+	0x000dcf, 0x000dd9, 0x000dcf, 0x000dca, 0x000dd9, 0x000ddf,
+	0x000f42, 0x000fb7, 0x000f4c, 0x000fb7, 0x000f51, 0x000fb7,
+	0x000f56, 0x000fb7, 0x000f5b, 0x000fb7, 0x000f40, 0x000fb5,
+	0x000f71, 0x000f72, 0x000f71, 0x000f74, 0x000fb2, 0x000f80,
+	0x000fb3, 0x000f80, 0x000f71, 0x000f80, 0x000f92, 0x000fb7,
+	0x000f9c, 0x000fb7, 0x000fa1, 0x000fb7, 0x000fa6, 0x000fb7,
+	0x000fab, 0x000fb7, 0x000f90, 0x000fb5, 0x001025, 0x00102e,
+	0x001b05, 0x001b35, 0x001b07, 0x001b35, 0x001b09, 0x001b35,
+	0x001b0b, 0x001b35, 0x001b0d, 0x001b35, 0x001b11, 0x001b35,
+	0x001b3a, 0x001b35, 0x001b3c, 0x001b35, 0x001b3e, 0x001b35,
+	0x001b3f, 0x001b35, 0x001b42, 0x001b35, 0x000041, 0x000325,
+	0x000061, 0x000325, 0x000042, 0x000307, 0x000062, 0x000307,
+	0x000042, 0x000323, 0x000062, 0x000323, 0x000042, 0x000331,
+	0x000062, 0x000331, 0x000043, 0x000327, 0x000301, 0x000063,
+	0x000327, 0x000301, 0x000044, 0x000307, 0x000064, 0x000307,
+	0x000044, 0x000323, 0x000064, 0x000323, 0x000044, 0x000331,
+	0x000064, 0x000331, 0x000044, 0x000327, 0x000064, 0x000327,
+	0x000044, 0x00032d, 0x000064, 0x00032d, 0x000045, 0x000304,
+	0x000300, 0x000065, 0x000304, 0x000300, 0x000045, 0x000304,
+	0x000301, 0x000065, 0x000304, 0x000301, 0x000045, 0x00032d,
+	0x000065, 0x00032d, 0x000045, 0x000330, 0x000065, 0x000330,
+	0x000045, 0x000327, 0x000306, 0x000065, 0x000327, 0x000306,
+	0x000046, 0x000307, 0x000066, 0x000307, 0x000047, 0x000304,
+	0x000067, 0x000304, 0x000048, 0x000307, 0x000068, 0x000307,
+	0x000048, 0x000323, 0x000068, 0x000323, 0x000048, 0x000308,
+	0x000068, 0x000308, 0x000048, 0x000327, 0x000068, 0x000327,
+	0x000048, 0x00032e, 0x000068, 0x00032e, 0x000049, 0x000330,
+	0x000069, 0x000330, 0x000049, 0x000308, 0x000301, 0x000069,
+	0x000308, 0x000301, 0x00004b, 0x000301, 0x00006b, 0x000301,
+	0x00004b, 0x000323, 0x00006b, 0x000323, 0x00004b, 0x000331,
+	0x00006b, 0x000331, 0x00004c, 0x000323, 0x00006c, 0x000323,
+	0x00004c, 0x000323, 0x000304, 0x00006c, 0x000323, 0x000304,
+	0x00004c, 0x000331, 0x00006c, 0x000331, 0x00004c, 0x00032d,
+	0x00006c, 0x00032d, 0x00004d, 0x000301, 0x00006d, 0x000301,
+	0x00004d, 0x000307, 0x00006d, 0x000307, 0x00004d, 0x000323,
+	0x00006d, 0x000323, 0x00004e, 0x000307, 0x00006e, 0x000307,
+	0x00004e, 0x000323, 0x00006e, 0x000323, 0x00004e, 0x000331,
+	0x00006e, 0x000331, 0x00004e, 0x00032d, 0x00006e, 0x00032d,
+	0x00004f, 0x000303, 0x000301, 0x00006f, 0x000303, 0x000301,
+	0x00004f, 0x000303, 0x000308, 0x00006f, 0x000303, 0x000308,
+	0x00004f, 0x000304, 0x000300, 0x00006f, 0x000304, 0x000300,
+	0x00004f, 0x000304, 0x000301, 0x00006f, 0x000304, 0x000301,
+	0x000050, 0x000301, 0x000070, 0x000301, 0x000050, 0x000307,
+	0x000070, 0x000307, 0x000052, 0x000307, 0x000072, 0x000307,
+	0x000052, 0x000323, 0x000072, 0x000323, 0x000052, 0x000323,
+	0x000304, 0x000072, 0x000323, 0x000304, 0x000052, 0x000331,
+	0x000072, 0x000331, 0x000053, 0x000307, 0x000073, 0x000307,
+	0x000053, 0x000323, 0x000073, 0x000323, 0x000053, 0x000301,
+	0x000307, 0x000073, 0x000301, 0x000307, 0x000053, 0x00030c,
+	0x000307, 0x000073, 0x00030c, 0x000307, 0x000053, 0x000323,
+	0x000307, 0x000073, 0x000323, 0x000307, 0x000054, 0x000307,
+	0x000074, 0x000307, 0x000054, 0x000323, 0x000074, 0x000323,
+	0x000054, 0x000331, 0x000074, 0x000331, 0x000054, 0x00032d,
+	0x000074, 0x00032d, 0x000055, 0x000324, 0x000075, 0x000324,
+	0x000055, 0x000330, 0x000075, 0x000330, 0x000055, 0x00032d,
+	0x000075, 0x00032d, 0x000055, 0x000303, 0x000301, 0x000075,
+	0x000303, 0x000301, 0x000055, 0x000304, 0x000308, 0x000075,
+	0x000304, 0x000308, 0x000056, 0x000303, 0x000076, 0x000303,
+	0x000056, 0x000323, 0x000076, 0x000323, 0x000057, 0x000300,
+	0x000077, 0x000300, 0x000057, 0x000301, 0x000077, 0x000301,
+	0x000057, 0x000308, 0x000077, 0x000308, 0x000057, 0x000307,
+	0x000077, 0x000307, 0x000057, 0x000323, 0x000077, 0x000323,
+	0x000058, 0x000307, 0x000078, 0x000307, 0x000058, 0x000308,
+	0x000078, 0x000308, 0x000059, 0x000307, 0x000079, 0x000307,
+	0x00005a, 0x000302, 0x00007a, 0x000302, 0x00005a, 0x000323,
+	0x00007a, 0x000323, 0x00005a, 0x000331, 0x00007a, 0x000331,
+	0x000068, 0x000331, 0x000074, 0x000308, 0x000077, 0x00030a,
+	0x000079, 0x00030a, 0x00017f, 0x000307, 0x000041, 0x000323,
+	0x000061, 0x000323, 0x000041, 0x000309, 0x000061, 0x000309,
+	0x000041, 0x000302, 0x000301, 0x000061, 0x000302, 0x000301,
+	0x000041, 0x000302, 0x000300, 0x000061, 0x000302, 0x000300,
+	0x000041, 0x000302, 0x000309, 0x000061, 0x000302, 0x000309,
+	0x000041, 0x000302, 0x000303, 0x000061, 0x000302, 0x000303,
+	0x000041, 0x000323, 0x000302, 0x000061, 0x000323, 0x000302,
+	0x000041, 0x000306, 0x000301, 0x000061, 0x000306, 0x000301,
+	0x000041, 0x000306, 0x000300, 0x000061, 0x000306, 0x000300,
+	0x000041, 0x000306, 0x000309, 0x000061, 0x000306, 0x000309,
+	0x000041, 0x000306, 0x000303, 0x000061, 0x000306, 0x000303,
+	0x000041, 0x000323, 0x000306, 0x000061, 0x000323, 0x000306,
+	0x000045, 0x000323, 0x000065, 0x000323, 0x000045, 0x000309,
+	0x000065, 0x000309, 0x000045, 0x000303, 0x000065, 0x000303,
+	0x000045, 0x000302, 0x000301, 0x000065, 0x000302, 0x000301,
+	0x000045, 0x000302, 0x000300, 0x000065, 0x000302, 0x000300,
+	0x000045, 0x000302, 0x000309, 0x000065, 0x000302, 0x000309,
+	0x000045, 0x000302, 0x000303, 0x000065, 0x000302, 0x000303,
+	0x000045, 0x000323, 0x000302, 0x000065, 0x000323, 0x000302,
+	0x000049, 0x000309, 0x000069, 0x000309, 0x000049, 0x000323,
+	0x000069, 0x000323, 0x00004f, 0x000323, 0x00006f, 0x000323,
+	0x00004f, 0x000309, 0x00006f, 0x000309, 0x00004f, 0x000302,
+	0x000301, 0x00006f, 0x000302, 0x000301, 0x00004f, 0x000302,
+	0x000300, 0x00006f, 0x000302, 0x000300, 0x00004f, 0x000302,
+	0x000309, 0x00006f, 0x000302, 0x000309, 0x00004f, 0x000302,
+	0x000303, 0x00006f, 0x000302, 0x000303, 0x00004f, 0x000323,
+	0x000302, 0x00006f, 0x000323, 0x000302, 0x00004f, 0x00031b,
+	0x000301, 0x00006f, 0x00031b, 0x000301, 0x00004f, 0x00031b,
+	0x000300, 0x00006f, 0x00031b, 0x000300, 0x00004f, 0x00031b,
+	0x000309, 0x00006f, 0x00031b, 0x000309, 0x00004f, 0x00031b,
+	0x000303, 0x00006f, 0x00031b, 0x000303, 0x00004f, 0x00031b,
+	0x000323, 0x00006f, 0x00031b, 0x000323, 0x000055, 0x000323,
+	0x000075, 0x000323, 0x000055, 0x000309, 0x000075, 0x000309,
+	0x000055, 0x00031b, 0x000301, 0x000075, 0x00031b, 0x000301,
+	0x000055, 0x00031b, 0x000300, 0x000075, 0x00031b, 0x000300,
+	0x000055, 0x00031b, 0x000309, 0x000075, 0x00031b, 0x000309,
+	0x000055, 0x00031b, 0x000303, 0x000075, 0x00031b, 0x000303,
+	0x000055, 0x00031b, 0x000323, 0x000075, 0x00031b, 0x000323,
+	0x000059, 0x000300, 0x000079, 0x000300, 0x000059, 0x000323,
+	0x000079, 0x000323, 0x000059, 0x000309, 0x000079, 0x000309,
+	0x000059, 0x000303, 0x000079, 0x000303, 0x0003b1, 0x000313,
+	0x0003b1, 0x000314, 0x0003b1, 0x000313, 0x000300, 0x0003b1,
+	0x000314, 0x000300, 0x0003b1, 0x000313, 0x000301, 0x0003b1,
+	0x000314, 0x000301, 0x0003b1, 0x000313, 0x000342, 0x0003b1,
+	0x000314, 0x000342, 0x000391, 0x000313, 0x000391, 0x000314,
+	0x000391, 0x000313, 0x000300, 0x000391, 0x000314, 0x000300,
+	0x000391, 0x000313, 0x000301, 0x000391, 0x000314, 0x000301,
+	0x000391, 0x000313, 0x000342, 0x000391, 0x000314, 0x000342,
+	0x0003b5, 0x000313, 0x0003b5, 0x000314, 0x0003b5, 0x000313,
+	0x000300, 0x0003b5, 0x000314, 0x000300, 0x0003b5, 0x000313,
+	0x000301, 0x0003b5, 0x000314, 0x000301, 0x000395, 0x000313,
+	0x000395, 0x000314, 0x000395, 0x000313, 0x000300, 0x000395,
+	0x000314, 0x000300, 0x000395, 0x000313, 0x000301, 0x000395,
+	0x000314, 0x000301, 0x0003b7, 0x000313, 0x0003b7, 0x000314,
+	0x0003b7, 0x000313, 0x000300, 0x0003b7, 0x000314, 0x000300,
+	0x0003b7, 0x000313, 0x000301, 0x0003b7, 0x000314, 0x000301,
+	0x0003b7, 0x000313, 0x000342, 0x0003b7, 0x000314, 0x000342,
+	0x000397, 0x000313, 0x000397, 0x000314, 0x000397, 0x000313,
+	0x000300, 0x000397, 0x000314, 0x000300, 0x000397, 0x000313,
+	0x000301, 0x000397, 0x000314, 0x000301, 0x000397, 0x000313,
+	0x000342, 0x000397, 0x000314, 0x000342, 0x0003b9, 0x000313,
+	0x0003b9, 0x000314, 0x0003b9, 0x000313, 0x000300, 0x0003b9,
+	0x000314, 0x000300, 0x0003b9, 0x000313, 0x000301, 0x0003b9,
+	0x000314, 0x000301, 0x0003b9, 0x000313, 0x000342, 0x0003b9,
+	0x000314, 0x000342, 0x000399, 0x000313, 0x000399, 0x000314,
+	0x000399, 0x000313, 0x000300, 0x000399, 0x000314, 0x000300,
+	0x000399, 0x000313, 0x000301, 0x000399, 0x000314, 0x000301,
+	0x000399, 0x000313, 0x000342, 0x000399, 0x000314, 0x000342,
+	0x0003bf, 0x000313, 0x0003bf, 0x000314, 0x0003bf, 0x000313,
+	0x000300, 0x0003bf, 0x000314, 0x000300, 0x0003bf, 0x000313,
+	0x000301, 0x0003bf, 0x000314, 0x000301, 0x00039f, 0x000313,
+	0x00039f, 0x000314, 0x00039f, 0x000313, 0x000300, 0x00039f,
+	0x000314, 0x000300, 0x00039f, 0x000313, 0x000301, 0x00039f,
+	0x000314, 0x000301, 0x0003c5, 0x000313, 0x0003c5, 0x000314,
+	0x0003c5, 0x000313, 0x000300, 0x0003c5, 0x000314, 0x000300,
+	0x0003c5, 0x000313, 0x000301, 0x0003c5, 0x000314, 0x000301,
+	0x0003c5, 0x000313, 0x000342, 0x0003c5, 0x000314, 0x000342,
+	0x0003a5, 0x000314, 0x0003a5, 0x000314, 0x000300, 0x0003a5,
+	0x000314, 0x000301, 0x0003a5, 0x000314, 0x000342, 0x0003c9,
+	0x000313, 0x0003c9, 0x000314, 0x0003c9, 0x000313, 0x000300,
+	0x0003c9, 0x000314, 0x000300, 0x0003c9, 0x000313, 0x000301,
+	0x0003c9, 0x000314, 0x000301, 0x0003c9, 0x000313, 0x000342,
+	0x0003c9, 0x000314, 0x000342, 0x0003a9, 0x000313, 0x0003a9,
+	0x000314, 0x0003a9, 0x000313, 0x000300, 0x0003a9, 0x000314,
+	0x000300, 0x0003a9, 0x000313, 0x000301, 0x0003a9, 0x000314,
+	0x000301, 0x0003a9, 0x000313, 0x000342, 0x0003a9, 0x000314,
+	0x000342, 0x0003b1, 0x000300, 0x0003b1, 0x000301, 0x0003b5,
+	0x000300, 0x0003b5, 0x000301, 0x0003b7, 0x000300, 0x0003b7,
+	0x000301, 0x0003b9, 0x000300, 0x0003b9, 0x000301, 0x0003bf,
+	0x000300, 0x0003bf, 0x000301, 0x0003c5, 0x000300, 0x0003c5,
+	0x000301, 0x0003c9, 0x000300, 0x0003c9, 0x000301, 0x0003b1,
+	0x000313, 0x000345, 0x0003b1, 0x000314, 0x000345, 0x0003b1,
+	0x000313, 0x000300, 0x000345, 0x0003b1, 0x000314, 0x000300,
+	0x000345, 0x0003b1, 0x000313, 0x000301, 0x000345, 0x0003b1,
+	0x000314, 0x000301, 0x000345, 0x0003b1, 0x000313, 0x000342,
+	0x000345, 0x0003b1, 0x000314, 0x000342, 0x000345, 0x000391,
+	0x000313, 0x000345, 0x000391, 0x000314, 0x000345, 0x000391,
+	0x000313, 0x000300, 0x000345, 0x000391, 0x000314, 0x000300,
+	0x000345, 0x000391, 0x000313, 0x000301, 0x000345, 0x000391,
+	0x000314, 0x000301, 0x000345, 0x000391, 0x000313, 0x000342,
+	0x000345, 0x000391, 0x000314, 0x000342, 0x000345, 0x0003b7,
+	0x000313, 0x000345, 0x0003b7, 0x000314, 0x000345, 0x0003b7,
+	0x000313, 0x000300, 0x000345, 0x0003b7, 0x000314, 0x000300,
+	0x000345, 0x0003b7, 0x000313, 0x000301, 0x000345, 0x0003b7,
+	0x000314, 0x000301, 0x000345, 0x0003b7, 0x000313, 0x000342,
+	0x000345, 0x0003b7, 0x000314, 0x000342, 0x000345, 0x000397,
+	0x000313, 0x000345, 0x000397, 0x000314, 0x000345, 0x000397,
+	0x000313, 0x000300, 0x000345, 0x000397, 0x000314, 0x000300,
+	0x000345, 0x000397, 0x000313, 0x000301, 0x000345, 0x000397,
+	0x000314, 0x000301, 0x000345, 0x000397, 0x000313, 0x000342,
+	0x000345, 0x000397, 0x000314, 0x000342, 0x000345, 0x0003c9,
+	0x000313, 0x000345, 0x0003c9, 0x000314, 0x000345, 0x0003c9,
+	0x000313, 0x000300, 0x000345, 0x0003c9, 0x000314, 0x000300,
+	0x000345, 0x0003c9, 0x000313, 0x000301, 0x000345, 0x0003c9,
+	0x000314, 0x000301, 0x000345, 0x0003c9, 0x000313, 0x000342,
+	0x000345, 0x0003c9, 0x000314, 0x000342, 0x000345, 0x0003a9,
+	0x000313, 0x000345, 0x0003a9, 0x000314, 0x000345, 0x0003a9,
+	0x000313, 0x000300, 0x000345, 0x0003a9, 0x000314, 0x000300,
+	0x000345, 0x0003a9, 0x000313, 0x000301, 0x000345, 0x0003a9,
+	0x000314, 0x000301, 0x000345, 0x0003a9, 0x000313, 0x000342,
+	0x000345, 0x0003a9, 0x000314, 0x000342, 0x000345, 0x0003b1,
+	0x000306, 0x0003b1, 0x000304, 0x0003b1, 0x000300, 0x000345,
+	0x0003b1, 0x000345, 0x0003b1, 0x000301, 0x000345, 0x0003b1,
+	0x000342, 0x0003b1, 0x000342, 0x000345, 0x000391, 0x000306,
+	0x000391, 0x000304, 0x000391, 0x000300, 0x000391, 0x000301,
+	0x000391, 0x000345, 0x0003b9, 0x0000a8, 0x000342, 0x0003b7,
+	0x000300, 0x000345, 0x0003b7, 0x000345, 0x0003b7, 0x000301,
+	0x000345, 0x0003b7, 0x000342, 0x0003b7, 0x000342, 0x000345,
+	0x000395, 0x000300, 0x000395, 0x000301, 0x000397, 0x000300,
+	0x000397, 0x000301, 0x000397, 0x000345, 0x001fbf, 0x000300,
+	0x001fbf, 0x000301, 0x001fbf, 0x000342, 0x0003b9, 0x000306,
+	0x0003b9, 0x000304, 0x0003b9, 0x000308, 0x000300, 0x0003b9,
+	0x000308, 0x000301, 0x0003b9, 0x000342, 0x0003b9, 0x000308,
+	0x000342, 0x000399, 0x000306, 0x000399, 0x000304, 0x000399,
+	0x000300, 0x000399, 0x000301, 0x001ffe, 0x000300, 0x001ffe,
+	0x000301, 0x001ffe, 0x000342, 0x0003c5, 0x000306, 0x0003c5,
+	0x000304, 0x0003c5, 0x000308, 0x000300, 0x0003c5, 0x000308,
+	0x000301, 0x0003c1, 0x000313, 0x0003c1, 0x000314, 0x0003c5,
+	0x000342, 0x0003c5, 0x000308, 0x000342, 0x0003a5, 0x000306,
+	0x0003a5, 0x000304, 0x0003a5, 0x000300, 0x0003a5, 0x000301,
+	0x0003a1, 0x000314, 0x0000a8, 0x000300, 0x0000a8, 0x000301,
+	0x000060, 0x0003c9, 0x000300, 0x000345, 0x0003c9, 0x000345,
+	0x0003c9, 0x000301, 0x000345, 0x0003c9, 0x000342, 0x0003c9,
+	0x000342, 0x000345, 0x00039f, 0x000300, 0x00039f, 0x000301,
+	0x0003a9, 0x000300, 0x0003a9, 0x000301, 0x0003a9, 0x000345,
+	0x0000b4, 0x002002, 0x002003, 0x0003a9, 0x00004b, 0x000041,
+	0x00030a, 0x002190, 0x000338, 0x002192, 0x000338, 0x002194,
+	0x000338, 0x0021d0, 0x000338, 0x0021d4, 0x000338, 0x0021d2,
+	0x000338, 0x002203, 0x000338, 0x002208, 0x000338, 0x00220b,
+	0x000338, 0x002223, 0x000338, 0x002225, 0x000338, 0x00223c,
+	0x000338, 0x002243, 0x000338, 0x002245, 0x000338, 0x002248,
+	0x000338, 0x00003d, 0x000338, 0x002261, 0x000338, 0x00224d,
+	0x000338, 0x00003c, 0x000338, 0x00003e, 0x000338, 0x002264,
+	0x000338, 0x002265, 0x000338, 0x002272, 0x000338, 0x002273,
+	0x000338, 0x002276, 0x000338, 0x002277, 0x000338, 0x00227a,
+	0x000338, 0x00227b, 0x000338, 0x002282, 0x000338, 0x002283,
+	0x000338, 0x002286, 0x000338, 0x002287, 0x000338, 0x0022a2,
+	0x000338, 0x0022a8, 0x000338, 0x0022a9, 0x000338, 0x0022ab,
+	0x000338, 0x00227c, 0x000338, 0x00227d, 0x000338, 0x002291,
+	0x000338, 0x002292, 0x000338, 0x0022b2, 0x000338, 0x0022b3,
+	0x000338, 0x0022b4, 0x000338, 0x0022b5, 0x000338, 0x003008,
+	0x003009, 0x002add, 0x000338, 0x00304b, 0x003099, 0x00304d,
+	0x003099, 0x00304f, 0x003099, 0x003051, 0x003099, 0x003053,
+	0x003099, 0x003055, 0x003099, 0x003057, 0x003099, 0x003059,
+	0x003099, 0x00305b, 0x003099, 0x00305d, 0x003099, 0x00305f,
+	0x003099, 0x003061, 0x003099, 0x003064, 0x003099, 0x003066,
+	0x003099, 0x003068, 0x003099, 0x00306f, 0x003099, 0x00306f,
+	0x00309a, 0x003072, 0x003099, 0x003072, 0x00309a, 0x003075,
+	0x003099, 0x003075, 0x00309a, 0x003078, 0x003099, 0x003078,
+	0x00309a, 0x00307b, 0x003099, 0x00307b, 0x00309a, 0x003046,
+	0x003099, 0x00309d, 0x003099, 0x0030ab, 0x003099, 0x0030ad,
+	0x003099, 0x0030af, 0x003099, 0x0030b1, 0x003099, 0x0030b3,
+	0x003099, 0x0030b5, 0x003099, 0x0030b7, 0x003099, 0x0030b9,
+	0x003099, 0x0030bb, 0x003099, 0x0030bd, 0x003099, 0x0030bf,
+	0x003099, 0x0030c1, 0x003099, 0x0030c4, 0x003099, 0x0030c6,
+	0x003099, 0x0030c8, 0x003099, 0x0030cf, 0x003099, 0x0030cf,
+	0x00309a, 0x0030d2, 0x003099, 0x0030d2, 0x00309a, 0x0030d5,
+	0x003099, 0x0030d5, 0x00309a, 0x0030d8, 0x003099, 0x0030d8,
+	0x00309a, 0x0030db, 0x003099, 0x0030db, 0x00309a, 0x0030a6,
+	0x003099, 0x0030ef, 0x003099, 0x0030f0, 0x003099, 0x0030f1,
+	0x003099, 0x0030f2, 0x003099, 0x0030fd, 0x003099, 0x008c48,
+	0x0066f4, 0x008eca, 0x008cc8, 0x006ed1, 0x004e32, 0x0053e5,
+	0x009f9c, 0x009f9c, 0x005951, 0x0091d1, 0x005587, 0x005948,
+	0x0061f6, 0x007669, 0x007f85, 0x00863f, 0x0087ba, 0x0088f8,
+	0x00908f, 0x006a02, 0x006d1b, 0x0070d9, 0x0073de, 0x00843d,
+	0x00916a, 0x0099f1, 0x004e82, 0x005375, 0x006b04, 0x00721b,
+	0x00862d, 0x009e1e, 0x005d50, 0x006feb, 0x0085cd, 0x008964,
+	0x0062c9, 0x0081d8, 0x00881f, 0x005eca, 0x006717, 0x006d6a,
+	0x0072fc, 0x0090ce, 0x004f86, 0x0051b7, 0x0052de, 0x0064c4,
+	0x006ad3, 0x007210, 0x0076e7, 0x008001, 0x008606, 0x00865c,
+	0x008def, 0x009732, 0x009b6f, 0x009dfa, 0x00788c, 0x00797f,
+	0x007da0, 0x0083c9, 0x009304, 0x009e7f, 0x008ad6, 0x0058df,
+	0x005f04, 0x007c60, 0x00807e, 0x007262, 0x0078ca, 0x008cc2,
+	0x0096f7, 0x0058d8, 0x005c62, 0x006a13, 0x006dda, 0x006f0f,
+	0x007d2f, 0x007e37, 0x00964b, 0x0052d2, 0x00808b, 0x0051dc,
+	0x0051cc, 0x007a1c, 0x007dbe, 0x0083f1, 0x009675, 0x008b80,
+	0x0062cf, 0x006a02, 0x008afe, 0x004e39, 0x005be7, 0x006012,
+	0x007387, 0x007570, 0x005317, 0x0078fb, 0x004fbf, 0x005fa9,
+	0x004e0d, 0x006ccc, 0x006578, 0x007d22, 0x0053c3, 0x00585e,
+	0x007701, 0x008449, 0x008aaa, 0x006bba, 0x008fb0, 0x006c88,
+	0x0062fe, 0x0082e5, 0x0063a0, 0x007565, 0x004eae, 0x005169,
+	0x0051c9, 0x006881, 0x007ce7, 0x00826f, 0x008ad2, 0x0091cf,
+	0x0052f5, 0x005442, 0x005973, 0x005eec, 0x0065c5, 0x006ffe,
+	0x00792a, 0x0095ad, 0x009a6a, 0x009e97, 0x009ece, 0x00529b,
+	0x0066c6, 0x006b77, 0x008f62, 0x005e74, 0x006190, 0x006200,
+	0x00649a, 0x006f23, 0x007149, 0x007489, 0x0079ca, 0x007df4,
+	0x00806f, 0x008f26, 0x0084ee, 0x009023, 0x00934a, 0x005217,
+	0x0052a3, 0x0054bd, 0x0070c8, 0x0088c2, 0x008aaa, 0x005ec9,
+	0x005ff5, 0x00637b, 0x006bae, 0x007c3e, 0x007375, 0x004ee4,
+	0x0056f9, 0x005be7, 0x005dba, 0x00601c, 0x0073b2, 0x007469,
+	0x007f9a, 0x008046, 0x009234, 0x0096f6, 0x009748, 0x009818,
+	0x004f8b, 0x0079ae, 0x0091b4, 0x0096b8, 0x0060e1, 0x004e86,
+	0x0050da, 0x005bee, 0x005c3f, 0x006599, 0x006a02, 0x0071ce,
+	0x007642, 0x0084fc, 0x00907c, 0x009f8d, 0x006688, 0x00962e,
+	0x005289, 0x00677b, 0x0067f3, 0x006d41, 0x006e9c, 0x007409,
+	0x007559, 0x00786b, 0x007d10, 0x00985e, 0x00516d, 0x00622e,
+	0x009678, 0x00502b, 0x005d19, 0x006dea, 0x008f2a, 0x005f8b,
+	0x006144, 0x006817, 0x007387, 0x009686, 0x005229, 0x00540f,
+	0x005c65, 0x006613, 0x00674e, 0x0068a8, 0x006ce5, 0x007406,
+	0x0075e2, 0x007f79, 0x0088cf, 0x0088e1, 0x0091cc, 0x0096e2,
+	0x00533f, 0x006eba, 0x00541d, 0x0071d0, 0x007498, 0x0085fa,
+	0x0096a3, 0x009c57, 0x009e9f, 0x006797, 0x006dcb, 0x0081e8,
+	0x007acb, 0x007b20, 0x007c92, 0x0072c0, 0x007099, 0x008b58,
+	0x004ec0, 0x008336, 0x00523a, 0x005207, 0x005ea6, 0x0062d3,
+	0x007cd6, 0x005b85, 0x006d1e, 0x0066b4, 0x008f3b, 0x00884c,
+	0x00964d, 0x00898b, 0x005ed3, 0x005140, 0x0055c0, 0x00585a,
+	0x006674, 0x0051de, 0x00732a, 0x0076ca, 0x00793c, 0x00795e,
+	0x007965, 0x00798f, 0x009756, 0x007cbe, 0x007fbd, 0x008612,
+	0x008af8, 0x009038, 0x0090fd, 0x0098ef, 0x0098fc, 0x009928,
+	0x009db4, 0x0090de, 0x0096b7, 0x004fae, 0x0050e7, 0x00514d,
+	0x0052c9, 0x0052e4, 0x005351, 0x00559d, 0x005606, 0x005668,
+	0x005840, 0x0058a8, 0x005c64, 0x005c6e, 0x006094, 0x006168,
+	0x00618e, 0x0061f2, 0x00654f, 0x0065e2, 0x006691, 0x006885,
+	0x006d77, 0x006e1a, 0x006f22, 0x00716e, 0x00722b, 0x007422,
+	0x007891, 0x00793e, 0x007949, 0x007948, 0x007950, 0x007956,
+	0x00795d, 0x00798d, 0x00798e, 0x007a40, 0x007a81, 0x007bc0,
+	0x007df4, 0x007e09, 0x007e41, 0x007f72, 0x008005, 0x0081ed,
+	0x008279, 0x008279, 0x008457, 0x008910, 0x008996, 0x008b01,
+	0x008b39, 0x008cd3, 0x008d08, 0x008fb6, 0x009038, 0x0096e3,
+	0x0097ff, 0x00983b, 0x006075, 0x0242ee, 0x008218, 0x004e26,
+	0x0051b5, 0x005168, 0x004f80, 0x005145, 0x005180, 0x0052c7,
+	0x0052fa, 0x00559d, 0x005555, 0x005599, 0x0055e2, 0x00585a,
+	0x0058b3, 0x005944, 0x005954, 0x005a62, 0x005b28, 0x005ed2,
+	0x005ed9, 0x005f69, 0x005fad, 0x0060d8, 0x00614e, 0x006108,
+	0x00618e, 0x006160, 0x0061f2, 0x006234, 0x0063c4, 0x00641c,
+	0x006452, 0x006556, 0x006674, 0x006717, 0x00671b, 0x006756,
+	0x006b79, 0x006bba, 0x006d41, 0x006edb, 0x006ecb, 0x006f22,
+	0x00701e, 0x00716e, 0x0077a7, 0x007235, 0x0072af, 0x00732a,
+	0x007471, 0x007506, 0x00753b, 0x00761d, 0x00761f, 0x0076ca,
+	0x0076db, 0x0076f4, 0x00774a, 0x007740, 0x0078cc, 0x007ab1,
+	0x007bc0, 0x007c7b, 0x007d5b, 0x007df4, 0x007f3e, 0x008005,
+	0x008352, 0x0083ef, 0x008779, 0x008941, 0x008986, 0x008996,
+	0x008abf, 0x008af8, 0x008acb, 0x008b01, 0x008afe, 0x008aed,
+	0x008b39, 0x008b8a, 0x008d08, 0x008f38, 0x009072, 0x009199,
+	0x009276, 0x00967c, 0x0096e3, 0x009756, 0x0097db, 0x0097ff,
+	0x00980b, 0x00983b, 0x009b12, 0x009f9c, 0x02284a, 0x022844,
+	0x0233d5, 0x003b9d, 0x004018, 0x004039, 0x025249, 0x025cd0,
+	0x027ed3, 0x009f43, 0x009f8e, 0x0005d9, 0x0005b4, 0x0005f2,
+	0x0005b7, 0x0005e9, 0x0005c1, 0x0005e9, 0x0005c2, 0x0005e9,
+	0x0005bc, 0x0005c1, 0x0005e9, 0x0005bc, 0x0005c2, 0x0005d0,
+	0x0005b7, 0x0005d0, 0x0005b8, 0x0005d0, 0x0005bc, 0x0005d1,
+	0x0005bc, 0x0005d2, 0x0005bc, 0x0005d3, 0x0005bc, 0x0005d4,
+	0x0005bc, 0x0005d5, 0x0005bc, 0x0005d6, 0x0005bc, 0x0005d8,
+	0x0005bc, 0x0005d9, 0x0005bc, 0x0005da, 0x0005bc, 0x0005db,
+	0x0005bc, 0x0005dc, 0x0005bc, 0x0005de, 0x0005bc, 0x0005e0,
+	0x0005bc, 0x0005e1, 0x0005bc, 0x0005e3, 0x0005bc, 0x0005e4,
+	0x0005bc, 0x0005e6, 0x0005bc, 0x0005e7, 0x0005bc, 0x0005e8,
+	0x0005bc, 0x0005e9, 0x0005bc, 0x0005ea, 0x0005bc, 0x0005d5,
+	0x0005b9, 0x0005d1, 0x0005bf, 0x0005db, 0x0005bf, 0x0005e4,
+	0x0005bf, 0x011099, 0x0110ba, 0x01109b, 0x0110ba, 0x0110a5,
+	0x0110ba, 0x011131, 0x011127, 0x011132, 0x011127, 0x011347,
+	0x01133e, 0x011347, 0x011357, 0x0114b9, 0x0114ba, 0x0114b9,
+	0x0114b0, 0x0114b9, 0x0114bd, 0x0115b8, 0x0115af, 0x0115b9,
+	0x0115af, 0x01d157, 0x01d165, 0x01d158, 0x01d165, 0x01d158,
+	0x01d165, 0x01d16e, 0x01d158, 0x01d165, 0x01d16f, 0x01d158,
+	0x01d165, 0x01d170, 0x01d158, 0x01d165, 0x01d171, 0x01d158,
+	0x01d165, 0x01d172, 0x01d1b9, 0x01d165, 0x01d1ba, 0x01d165,
+	0x01d1b9, 0x01d165, 0x01d16e, 0x01d1ba, 0x01d165, 0x01d16e,
+	0x01d1b9, 0x01d165, 0x01d16f, 0x01d1ba, 0x01d165, 0x01d16f,
+	0x004e3d, 0x004e38, 0x004e41, 0x020122, 0x004f60, 0x004fae,
+	0x004fbb, 0x005002, 0x00507a, 0x005099, 0x0050e7, 0x0050cf,
+	0x00349e, 0x02063a, 0x00514d, 0x005154, 0x005164, 0x005177,
+	0x02051c, 0x0034b9, 0x005167, 0x00518d, 0x02054b, 0x005197,
+	0x0051a4, 0x004ecc, 0x0051ac, 0x0051b5, 0x0291df, 0x0051f5,
+	0x005203, 0x0034df, 0x00523b, 0x005246, 0x005272, 0x005277,
+	0x003515, 0x0052c7, 0x0052c9, 0x0052e4, 0x0052fa, 0x005305,
+	0x005306, 0x005317, 0x005349, 0x005351, 0x00535a, 0x005373,
+	0x00537d, 0x00537f, 0x00537f, 0x00537f, 0x020a2c, 0x007070,
+	0x0053ca, 0x0053df, 0x020b63, 0x0053eb, 0x0053f1, 0x005406,
+	0x00549e, 0x005438, 0x005448, 0x005468, 0x0054a2, 0x0054f6,
+	0x005510, 0x005553, 0x005563, 0x005584, 0x005584, 0x005599,
+	0x0055ab, 0x0055b3, 0x0055c2, 0x005716, 0x005606, 0x005717,
+	0x005651, 0x005674, 0x005207, 0x0058ee, 0x0057ce, 0x0057f4,
+	0x00580d, 0x00578b, 0x005832, 0x005831, 0x0058ac, 0x0214e4,
+	0x0058f2, 0x0058f7, 0x005906, 0x00591a, 0x005922, 0x005962,
+	0x0216a8, 0x0216ea, 0x0059ec, 0x005a1b, 0x005a27, 0x0059d8,
+	0x005a66, 0x0036ee, 0x0036fc, 0x005b08, 0x005b3e, 0x005b3e,
+	0x0219c8, 0x005bc3, 0x005bd8, 0x005be7, 0x005bf3, 0x021b18,
+	0x005bff, 0x005c06, 0x005f53, 0x005c22, 0x003781, 0x005c60,
+	0x005c6e, 0x005cc0, 0x005c8d, 0x021de4, 0x005d43, 0x021de6,
+	0x005d6e, 0x005d6b, 0x005d7c, 0x005de1, 0x005de2, 0x00382f,
+	0x005dfd, 0x005e28, 0x005e3d, 0x005e69, 0x003862, 0x022183,
+	0x00387c, 0x005eb0, 0x005eb3, 0x005eb6, 0x005eca, 0x02a392,
+	0x005efe, 0x022331, 0x022331, 0x008201, 0x005f22, 0x005f22,
+	0x0038c7, 0x0232b8, 0x0261da, 0x005f62, 0x005f6b, 0x0038e3,
+	0x005f9a, 0x005fcd, 0x005fd7, 0x005ff9, 0x006081, 0x00393a,
+	0x00391c, 0x006094, 0x0226d4, 0x0060c7, 0x006148, 0x00614c,
+	0x00614e, 0x00614c, 0x00617a, 0x00618e, 0x0061b2, 0x0061a4,
+	0x0061af, 0x0061de, 0x0061f2, 0x0061f6, 0x006210, 0x00621b,
+	0x00625d, 0x0062b1, 0x0062d4, 0x006350, 0x022b0c, 0x00633d,
+	0x0062fc, 0x006368, 0x006383, 0x0063e4, 0x022bf1, 0x006422,
+	0x0063c5, 0x0063a9, 0x003a2e, 0x006469, 0x00647e, 0x00649d,
+	0x006477, 0x003a6c, 0x00654f, 0x00656c, 0x02300a, 0x0065e3,
+	0x0066f8, 0x006649, 0x003b19, 0x006691, 0x003b08, 0x003ae4,
+	0x005192, 0x005195, 0x006700, 0x00669c, 0x0080ad, 0x0043d9,
+	0x006717, 0x00671b, 0x006721, 0x00675e, 0x006753, 0x0233c3,
+	0x003b49, 0x0067fa, 0x006785, 0x006852, 0x006885, 0x02346d,
+	0x00688e, 0x00681f, 0x006914, 0x003b9d, 0x006942, 0x0069a3,
+	0x0069ea, 0x006aa8, 0x0236a3, 0x006adb, 0x003c18, 0x006b21,
+	0x0238a7, 0x006b54, 0x003c4e, 0x006b72, 0x006b9f, 0x006bba,
+	0x006bbb, 0x023a8d, 0x021d0b, 0x023afa, 0x006c4e, 0x023cbc,
+	0x006cbf, 0x006ccd, 0x006c67, 0x006d16, 0x006d3e, 0x006d77,
+	0x006d41, 0x006d69, 0x006d78, 0x006d85, 0x023d1e, 0x006d34,
+	0x006e2f, 0x006e6e, 0x003d33, 0x006ecb, 0x006ec7, 0x023ed1,
+	0x006df9, 0x006f6e, 0x023f5e, 0x023f8e, 0x006fc6, 0x007039,
+	0x00701e, 0x00701b, 0x003d96, 0x00704a, 0x00707d, 0x007077,
+	0x0070ad, 0x020525, 0x007145, 0x024263, 0x00719c, 0x0243ab,
+	0x007228, 0x007235, 0x007250, 0x024608, 0x007280, 0x007295,
+	0x024735, 0x024814, 0x00737a, 0x00738b, 0x003eac, 0x0073a5,
+	0x003eb8, 0x003eb8, 0x007447, 0x00745c, 0x007471, 0x007485,
+	0x0074ca, 0x003f1b, 0x007524, 0x024c36, 0x00753e, 0x024c92,
+	0x007570, 0x02219f, 0x007610, 0x024fa1, 0x024fb8, 0x025044,
+	0x003ffc, 0x004008, 0x0076f4, 0x0250f3, 0x0250f2, 0x025119,
+	0x025133, 0x00771e, 0x00771f, 0x00771f, 0x00774a, 0x004039,
+	0x00778b, 0x004046, 0x004096, 0x02541d, 0x00784e, 0x00788c,
+	0x0078cc, 0x0040e3, 0x025626, 0x007956, 0x02569a, 0x0256c5,
+	0x00798f, 0x0079eb, 0x00412f, 0x007a40, 0x007a4a, 0x007a4f,
+	0x02597c, 0x025aa7, 0x025aa7, 0x007aee, 0x004202, 0x025bab,
+	0x007bc6, 0x007bc9, 0x004227, 0x025c80, 0x007cd2, 0x0042a0,
+	0x007ce8, 0x007ce3, 0x007d00, 0x025f86, 0x007d63, 0x004301,
+	0x007dc7, 0x007e02, 0x007e45, 0x004334, 0x026228, 0x026247,
+	0x004359, 0x0262d9, 0x007f7a, 0x02633e, 0x007f95, 0x007ffa,
+	0x008005, 0x0264da, 0x026523, 0x008060, 0x0265a8, 0x008070,
+	0x02335f, 0x0043d5, 0x0080b2, 0x008103, 0x00440b, 0x00813e,
+	0x005ab5, 0x0267a7, 0x0267b5, 0x023393, 0x02339c, 0x008201,
+	0x008204, 0x008f9e, 0x00446b, 0x008291, 0x00828b, 0x00829d,
+	0x0052b3, 0x0082b1, 0x0082b3, 0x0082bd, 0x0082e6, 0x026b3c,
+	0x0082e5, 0x00831d, 0x008363, 0x0083ad, 0x008323, 0x0083bd,
+	0x0083e7, 0x008457, 0x008353, 0x0083ca, 0x0083cc, 0x0083dc,
+	0x026c36, 0x026d6b, 0x026cd5, 0x00452b, 0x0084f1, 0x0084f3,
+	0x008516, 0x0273ca, 0x008564, 0x026f2c, 0x00455d, 0x004561,
+	0x026fb1, 0x0270d2, 0x00456b, 0x008650, 0x00865c, 0x008667,
+	0x008669, 0x0086a9, 0x008688, 0x00870e, 0x0086e2, 0x008779,
+	0x008728, 0x00876b, 0x008786, 0x0045d7, 0x0087e1, 0x008801,
+	0x0045f9, 0x008860, 0x008863, 0x027667, 0x0088d7, 0x0088de,
+	0x004635, 0x0088fa, 0x0034bb, 0x0278ae, 0x027966, 0x0046be,
+	0x0046c7, 0x008aa0, 0x008aed, 0x008b8a, 0x008c55, 0x027ca8,
+	0x008cab, 0x008cc1, 0x008d1b, 0x008d77, 0x027f2f, 0x020804,
+	0x008dcb, 0x008dbc, 0x008df0, 0x0208de, 0x008ed4, 0x008f38,
+	0x0285d2, 0x0285ed, 0x009094, 0x0090f1, 0x009111, 0x02872e,
+	0x00911b, 0x009238, 0x0092d7, 0x0092d8, 0x00927c, 0x0093f9,
+	0x009415, 0x028bfa, 0x00958b, 0x004995, 0x0095b7, 0x028d77,
+	0x0049e6, 0x0096c3, 0x005db2, 0x009723, 0x029145, 0x02921a,
+	0x004a6e, 0x004a76, 0x0097e0, 0x02940a, 0x004ab2, 0x029496,
+	0x00980b, 0x00980b, 0x009829, 0x0295b6, 0x0098e2, 0x004b33,
+	0x009929, 0x0099a7, 0x0099c2, 0x0099fe, 0x004bce, 0x029b30,
+	0x009b12, 0x009c40, 0x009cfd, 0x004cce, 0x004ced, 0x009d67,
+	0x02a0ce, 0x004cf8, 0x02a105, 0x02a20e, 0x02a291, 0x009ebb,
+	0x004d56, 0x009ef9, 0x009efe, 0x009f05, 0x009f0f, 0x009f16,
+	0x009f3b, 0x02a600,
+};
+
+static u16 apfs_cf_trie[] = {
+	/* Node for range 0x_____ */
+	0x0001, 0x0002, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0____ */
+	0x0003, 0x0004, 0x0005, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0006, 0x0000, 0x0000, 0x0000, 0x0000, 0x0007,
+	/* Node for range 0x1____ */
+	0x0008, 0x0009, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x000a, 0x0000,
+	/* Node for range 0x00___ */
+	0x000b, 0x000c, 0x000d, 0x000e, 0x000f, 0x0010, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01___ */
+	0x0011, 0x0000, 0x0000, 0x0012, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0013, 0x0000, 0x0014, 0x0015,
+	/* Node for range 0x02___ */
+	0x0000, 0x0016, 0x0000, 0x0000, 0x0017, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0018, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0a___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0019, 0x001a,
+	0x0000, 0x0000, 0x0000, 0x001b, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0f___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x001c, 0x0000, 0x0000, 0x0000, 0x001d,
+	/* Node for range 0x10___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x001e, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x001f, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x11___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0020, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1e___ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0021, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x000__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0022, 0x0023, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0024, 0x0025, 0x0026, 0x0000, 0x0000,
+	/* Node for range 0x001__ */
+	0x0027, 0x0028, 0x0029, 0x002a, 0x002b, 0x002c, 0x002d, 0x002e,
+	0x002f, 0x0030, 0x0031, 0x0032, 0x0033, 0x0034, 0x0035, 0x0036,
+	/* Node for range 0x002__ */
+	0x0037, 0x0038, 0x0039, 0x003a, 0x003b, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x003__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x003c, 0x0000, 0x0000, 0x003d,
+	0x003e, 0x003f, 0x0040, 0x0041, 0x0042, 0x0043, 0x0044, 0x0045,
+	/* Node for range 0x004__ */
+	0x0046, 0x0047, 0x0048, 0x0000, 0x0000, 0x0000, 0x0049, 0x004a,
+	0x004b, 0x004c, 0x004d, 0x004e, 0x004f, 0x0050, 0x0051, 0x0052,
+	/* Node for range 0x005__ */
+	0x0053, 0x0054, 0x0055, 0x0056, 0x0057, 0x0058, 0x0000, 0x0000,
+	0x0059, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x010__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x005a, 0x005b, 0x005c, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x013__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x005d,
+	/* Node for range 0x01c__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x005e, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01e__ */
+	0x005f, 0x0060, 0x0061, 0x0062, 0x0063, 0x0064, 0x0065, 0x0066,
+	0x0067, 0x0068, 0x0069, 0x006a, 0x006b, 0x006c, 0x006d, 0x006e,
+	/* Node for range 0x01f__ */
+	0x006f, 0x0070, 0x0071, 0x0072, 0x0073, 0x0074, 0x0075, 0x0000,
+	0x0076, 0x0077, 0x0078, 0x0079, 0x007a, 0x007b, 0x007c, 0x007d,
+	/* Node for range 0x021__ */
+	0x0000, 0x0000, 0x007e, 0x007f, 0x0000, 0x0000, 0x0080, 0x0000,
+	0x0081, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x024__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0082, 0x0083, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x02c__ */
+	0x0084, 0x0085, 0x0086, 0x0000, 0x0000, 0x0000, 0x0087, 0x0088,
+	0x0089, 0x008a, 0x008b, 0x008c, 0x008d, 0x008e, 0x008f, 0x0090,
+	/* Node for range 0x0a6__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0091, 0x0092, 0x0093, 0x0000,
+	0x0094, 0x0095, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0a7__ */
+	0x0000, 0x0000, 0x0096, 0x0097, 0x0098, 0x0099, 0x009a, 0x009b,
+	0x009c, 0x009d, 0x009e, 0x009f, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0ab__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x00a0,
+	0x00a1, 0x00a2, 0x00a3, 0x00a4, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0fb__ */
+	0x00a5, 0x00a6, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0ff__ */
+	0x0000, 0x0000, 0x00a7, 0x00a8, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x104__ */
+	0x00a9, 0x00aa, 0x00ab, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x00ac, 0x00ad, 0x00ae, 0x0000, 0x0000,
+	/* Node for range 0x10c__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x00af, 0x00b0, 0x00b1, 0x00b2, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x118__ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x00b3, 0x00b4, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1e9__ */
+	0x00b5, 0x00b6, 0x00b7, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0004_ */
+	0x0000, 0x0001, 0x0009, 0x0011, 0x0019, 0x0021, 0x0029, 0x0031,
+	0x0039, 0x0041, 0x0049, 0x0051, 0x0059, 0x0061, 0x0069, 0x0071,
+	/* Node for range 0x0005_ */
+	0x0079, 0x0081, 0x0089, 0x0091, 0x0099, 0x00a1, 0x00a9, 0x00b1,
+	0x00b9, 0x00c1, 0x00c9, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x000b_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x00d1, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x000c_ */
+	0x00d9, 0x00e1, 0x00e9, 0x00f1, 0x00f9, 0x0101, 0x0109, 0x0111,
+	0x0119, 0x0121, 0x0129, 0x0131, 0x0139, 0x0141, 0x0149, 0x0151,
+	/* Node for range 0x000d_ */
+	0x0159, 0x0161, 0x0169, 0x0171, 0x0179, 0x0181, 0x0189, 0x0000,
+	0x0191, 0x0199, 0x01a1, 0x01a9, 0x01b1, 0x01b9, 0x01c1, 0x01ca,
+	/* Node for range 0x0010_ */
+	0x01d9, 0x0000, 0x01e1, 0x0000, 0x01e9, 0x0000, 0x01f1, 0x0000,
+	0x01f9, 0x0000, 0x0201, 0x0000, 0x0209, 0x0000, 0x0211, 0x0000,
+	/* Node for range 0x0011_ */
+	0x0219, 0x0000, 0x0221, 0x0000, 0x0229, 0x0000, 0x0231, 0x0000,
+	0x0239, 0x0000, 0x0241, 0x0000, 0x0249, 0x0000, 0x0251, 0x0000,
+	/* Node for range 0x0012_ */
+	0x0259, 0x0000, 0x0261, 0x0000, 0x0269, 0x0000, 0x0271, 0x0000,
+	0x0279, 0x0000, 0x0281, 0x0000, 0x0289, 0x0000, 0x0291, 0x0000,
+	/* Node for range 0x0013_ */
+	0x029a, 0x0000, 0x02a9, 0x0000, 0x02b1, 0x0000, 0x02b9, 0x0000,
+	0x0000, 0x02c1, 0x0000, 0x02c9, 0x0000, 0x02d1, 0x0000, 0x02d9,
+	/* Node for range 0x0014_ */
+	0x0000, 0x02e1, 0x0000, 0x02e9, 0x0000, 0x02f1, 0x0000, 0x02f9,
+	0x0000, 0x0302, 0x0311, 0x0000, 0x0319, 0x0000, 0x0321, 0x0000,
+	/* Node for range 0x0015_ */
+	0x0329, 0x0000, 0x0331, 0x0000, 0x0339, 0x0000, 0x0341, 0x0000,
+	0x0349, 0x0000, 0x0351, 0x0000, 0x0359, 0x0000, 0x0361, 0x0000,
+	/* Node for range 0x0016_ */
+	0x0369, 0x0000, 0x0371, 0x0000, 0x0379, 0x0000, 0x0381, 0x0000,
+	0x0389, 0x0000, 0x0391, 0x0000, 0x0399, 0x0000, 0x03a1, 0x0000,
+	/* Node for range 0x0017_ */
+	0x03a9, 0x0000, 0x03b1, 0x0000, 0x03b9, 0x0000, 0x03c1, 0x0000,
+	0x03c9, 0x03d1, 0x0000, 0x03d9, 0x0000, 0x03e1, 0x0000, 0x03e9,
+	/* Node for range 0x0018_ */
+	0x0000, 0x03f1, 0x03f9, 0x0000, 0x0401, 0x0000, 0x0409, 0x0411,
+	0x0000, 0x0419, 0x0421, 0x0429, 0x0000, 0x0000, 0x0431, 0x0439,
+	/* Node for range 0x0019_ */
+	0x0441, 0x0449, 0x0000, 0x0451, 0x0459, 0x0000, 0x0461, 0x0469,
+	0x0471, 0x0000, 0x0000, 0x0000, 0x0479, 0x0481, 0x0000, 0x0489,
+	/* Node for range 0x001a_ */
+	0x0491, 0x0000, 0x0499, 0x0000, 0x04a1, 0x0000, 0x04a9, 0x04b1,
+	0x0000, 0x04b9, 0x0000, 0x0000, 0x04c1, 0x0000, 0x04c9, 0x04d1,
+	/* Node for range 0x001b_ */
+	0x0000, 0x04d9, 0x04e1, 0x04e9, 0x0000, 0x04f1, 0x0000, 0x04f9,
+	0x0501, 0x0000, 0x0000, 0x0000, 0x0509, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x001c_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0511, 0x0519, 0x0000, 0x0521,
+	0x0529, 0x0000, 0x0531, 0x0539, 0x0000, 0x0541, 0x0000, 0x0549,
+	/* Node for range 0x001d_ */
+	0x0000, 0x0551, 0x0000, 0x0559, 0x0000, 0x0561, 0x0000, 0x0569,
+	0x0000, 0x0571, 0x0000, 0x0579, 0x0000, 0x0000, 0x0581, 0x0000,
+	/* Node for range 0x001e_ */
+	0x0589, 0x0000, 0x0591, 0x0000, 0x0599, 0x0000, 0x05a1, 0x0000,
+	0x05a9, 0x0000, 0x05b1, 0x0000, 0x05b9, 0x0000, 0x05c1, 0x0000,
+	/* Node for range 0x001f_ */
+	0x05ca, 0x05d9, 0x05e1, 0x0000, 0x05e9, 0x0000, 0x05f1, 0x05f9,
+	0x0601, 0x0000, 0x0609, 0x0000, 0x0611, 0x0000, 0x0619, 0x0000,
+	/* Node for range 0x0020_ */
+	0x0621, 0x0000, 0x0629, 0x0000, 0x0631, 0x0000, 0x0639, 0x0000,
+	0x0641, 0x0000, 0x0649, 0x0000, 0x0651, 0x0000, 0x0659, 0x0000,
+	/* Node for range 0x0021_ */
+	0x0661, 0x0000, 0x0669, 0x0000, 0x0671, 0x0000, 0x0679, 0x0000,
+	0x0681, 0x0000, 0x0689, 0x0000, 0x0691, 0x0000, 0x0699, 0x0000,
+	/* Node for range 0x0022_ */
+	0x06a1, 0x0000, 0x06a9, 0x0000, 0x06b1, 0x0000, 0x06b9, 0x0000,
+	0x06c1, 0x0000, 0x06c9, 0x0000, 0x06d1, 0x0000, 0x06d9, 0x0000,
+	/* Node for range 0x0023_ */
+	0x06e1, 0x0000, 0x06e9, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x06f1, 0x06f9, 0x0000, 0x0701, 0x0709, 0x0000,
+	/* Node for range 0x0024_ */
+	0x0000, 0x0711, 0x0000, 0x0719, 0x0721, 0x0729, 0x0731, 0x0000,
+	0x0739, 0x0000, 0x0741, 0x0000, 0x0749, 0x0000, 0x0751, 0x0000,
+	/* Node for range 0x0034_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0759, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0037_ */
+	0x0761, 0x0000, 0x0769, 0x0000, 0x0000, 0x0000, 0x0771, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0779,
+	/* Node for range 0x0038_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0781, 0x0000,
+	0x0789, 0x0791, 0x0799, 0x0000, 0x07a1, 0x0000, 0x07a9, 0x07b1,
+	/* Node for range 0x0039_ */
+	0x07bb, 0x07d1, 0x07d9, 0x07e1, 0x07e9, 0x07f1, 0x07f9, 0x0801,
+	0x0809, 0x0811, 0x0819, 0x0821, 0x0829, 0x0831, 0x0839, 0x0841,
+	/* Node for range 0x003a_ */
+	0x0849, 0x0851, 0x0000, 0x0859, 0x0861, 0x0869, 0x0871, 0x0879,
+	0x0881, 0x0889, 0x0891, 0x0899, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x003b_ */
+	0x08a3, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x003c_ */
+	0x0000, 0x0000, 0x08b9, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x08c1,
+	/* Node for range 0x003d_ */
+	0x08c9, 0x08d1, 0x0000, 0x0000, 0x0000, 0x08d9, 0x08e1, 0x0000,
+	0x08e9, 0x0000, 0x08f1, 0x0000, 0x08f9, 0x0000, 0x0901, 0x0000,
+	/* Node for range 0x003e_ */
+	0x0909, 0x0000, 0x0911, 0x0000, 0x0919, 0x0000, 0x0921, 0x0000,
+	0x0929, 0x0000, 0x0931, 0x0000, 0x0939, 0x0000, 0x0941, 0x0000,
+	/* Node for range 0x003f_ */
+	0x0949, 0x0951, 0x0000, 0x0000, 0x0959, 0x0961, 0x0000, 0x0969,
+	0x0000, 0x0971, 0x0979, 0x0000, 0x0000, 0x0981, 0x0989, 0x0991,
+	/* Node for range 0x0040_ */
+	0x0999, 0x09a1, 0x09a9, 0x09b1, 0x09b9, 0x09c1, 0x09c9, 0x09d1,
+	0x09d9, 0x09e1, 0x09e9, 0x09f1, 0x09f9, 0x0a01, 0x0a09, 0x0a11,
+	/* Node for range 0x0041_ */
+	0x0a19, 0x0a21, 0x0a29, 0x0a31, 0x0a39, 0x0a41, 0x0a49, 0x0a51,
+	0x0a59, 0x0a61, 0x0a69, 0x0a71, 0x0a79, 0x0a81, 0x0a89, 0x0a91,
+	/* Node for range 0x0042_ */
+	0x0a99, 0x0aa1, 0x0aa9, 0x0ab1, 0x0ab9, 0x0ac1, 0x0ac9, 0x0ad1,
+	0x0ad9, 0x0ae1, 0x0ae9, 0x0af1, 0x0af9, 0x0b01, 0x0b09, 0x0b11,
+	/* Node for range 0x0046_ */
+	0x0b19, 0x0000, 0x0b21, 0x0000, 0x0b29, 0x0000, 0x0b31, 0x0000,
+	0x0b39, 0x0000, 0x0b41, 0x0000, 0x0b49, 0x0000, 0x0b51, 0x0000,
+	/* Node for range 0x0047_ */
+	0x0b59, 0x0000, 0x0b61, 0x0000, 0x0b69, 0x0000, 0x0b71, 0x0000,
+	0x0b79, 0x0000, 0x0b81, 0x0000, 0x0b89, 0x0000, 0x0b91, 0x0000,
+	/* Node for range 0x0048_ */
+	0x0b99, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0ba1, 0x0000, 0x0ba9, 0x0000, 0x0bb1, 0x0000,
+	/* Node for range 0x0049_ */
+	0x0bb9, 0x0000, 0x0bc1, 0x0000, 0x0bc9, 0x0000, 0x0bd1, 0x0000,
+	0x0bd9, 0x0000, 0x0be1, 0x0000, 0x0be9, 0x0000, 0x0bf1, 0x0000,
+	/* Node for range 0x004a_ */
+	0x0bf9, 0x0000, 0x0c01, 0x0000, 0x0c09, 0x0000, 0x0c11, 0x0000,
+	0x0c19, 0x0000, 0x0c21, 0x0000, 0x0c29, 0x0000, 0x0c31, 0x0000,
+	/* Node for range 0x004b_ */
+	0x0c39, 0x0000, 0x0c41, 0x0000, 0x0c49, 0x0000, 0x0c51, 0x0000,
+	0x0c59, 0x0000, 0x0c61, 0x0000, 0x0c69, 0x0000, 0x0c71, 0x0000,
+	/* Node for range 0x004c_ */
+	0x0c79, 0x0c81, 0x0000, 0x0c89, 0x0000, 0x0c91, 0x0000, 0x0c99,
+	0x0000, 0x0ca1, 0x0000, 0x0ca9, 0x0000, 0x0cb1, 0x0000, 0x0000,
+	/* Node for range 0x004d_ */
+	0x0cb9, 0x0000, 0x0cc1, 0x0000, 0x0cc9, 0x0000, 0x0cd1, 0x0000,
+	0x0cd9, 0x0000, 0x0ce1, 0x0000, 0x0ce9, 0x0000, 0x0cf1, 0x0000,
+	/* Node for range 0x004e_ */
+	0x0cf9, 0x0000, 0x0d01, 0x0000, 0x0d09, 0x0000, 0x0d11, 0x0000,
+	0x0d19, 0x0000, 0x0d21, 0x0000, 0x0d29, 0x0000, 0x0d31, 0x0000,
+	/* Node for range 0x004f_ */
+	0x0d39, 0x0000, 0x0d41, 0x0000, 0x0d49, 0x0000, 0x0d51, 0x0000,
+	0x0d59, 0x0000, 0x0d61, 0x0000, 0x0d69, 0x0000, 0x0d71, 0x0000,
+	/* Node for range 0x0050_ */
+	0x0d79, 0x0000, 0x0d81, 0x0000, 0x0d89, 0x0000, 0x0d91, 0x0000,
+	0x0d99, 0x0000, 0x0da1, 0x0000, 0x0da9, 0x0000, 0x0db1, 0x0000,
+	/* Node for range 0x0051_ */
+	0x0db9, 0x0000, 0x0dc1, 0x0000, 0x0dc9, 0x0000, 0x0dd1, 0x0000,
+	0x0dd9, 0x0000, 0x0de1, 0x0000, 0x0de9, 0x0000, 0x0df1, 0x0000,
+	/* Node for range 0x0052_ */
+	0x0df9, 0x0000, 0x0e01, 0x0000, 0x0e09, 0x0000, 0x0e11, 0x0000,
+	0x0e19, 0x0000, 0x0e21, 0x0000, 0x0e29, 0x0000, 0x0e31, 0x0000,
+	/* Node for range 0x0053_ */
+	0x0000, 0x0e39, 0x0e41, 0x0e49, 0x0e51, 0x0e59, 0x0e61, 0x0e69,
+	0x0e71, 0x0e79, 0x0e81, 0x0e89, 0x0e91, 0x0e99, 0x0ea1, 0x0ea9,
+	/* Node for range 0x0054_ */
+	0x0eb1, 0x0eb9, 0x0ec1, 0x0ec9, 0x0ed1, 0x0ed9, 0x0ee1, 0x0ee9,
+	0x0ef1, 0x0ef9, 0x0f01, 0x0f09, 0x0f11, 0x0f19, 0x0f21, 0x0f29,
+	/* Node for range 0x0055_ */
+	0x0f31, 0x0f39, 0x0f41, 0x0f49, 0x0f51, 0x0f59, 0x0f61, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0058_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0f6a,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x010a_ */
+	0x0f79, 0x0f81, 0x0f89, 0x0f91, 0x0f99, 0x0fa1, 0x0fa9, 0x0fb1,
+	0x0fb9, 0x0fc1, 0x0fc9, 0x0fd1, 0x0fd9, 0x0fe1, 0x0fe9, 0x0ff1,
+	/* Node for range 0x010b_ */
+	0x0ff9, 0x1001, 0x1009, 0x1011, 0x1019, 0x1021, 0x1029, 0x1031,
+	0x1039, 0x1041, 0x1049, 0x1051, 0x1059, 0x1061, 0x1069, 0x1071,
+	/* Node for range 0x010c_ */
+	0x1079, 0x1081, 0x1089, 0x1091, 0x1099, 0x10a1, 0x0000, 0x10a9,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x10b1, 0x0000, 0x0000,
+	/* Node for range 0x013f_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x10b9, 0x10c1, 0x10c9, 0x10d1, 0x10d9, 0x10e1, 0x0000, 0x0000,
+	/* Node for range 0x01c8_ */
+	0x10e9, 0x10f1, 0x10f9, 0x1101, 0x1109, 0x1111, 0x1119, 0x1121,
+	0x1129, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01e0_ */
+	0x1131, 0x0000, 0x1139, 0x0000, 0x1141, 0x0000, 0x1149, 0x0000,
+	0x1151, 0x0000, 0x1159, 0x0000, 0x1161, 0x0000, 0x1169, 0x0000,
+	/* Node for range 0x01e1_ */
+	0x1171, 0x0000, 0x1179, 0x0000, 0x1181, 0x0000, 0x1189, 0x0000,
+	0x1191, 0x0000, 0x1199, 0x0000, 0x11a1, 0x0000, 0x11a9, 0x0000,
+	/* Node for range 0x01e2_ */
+	0x11b1, 0x0000, 0x11b9, 0x0000, 0x11c1, 0x0000, 0x11c9, 0x0000,
+	0x11d1, 0x0000, 0x11d9, 0x0000, 0x11e1, 0x0000, 0x11e9, 0x0000,
+	/* Node for range 0x01e3_ */
+	0x11f1, 0x0000, 0x11f9, 0x0000, 0x1201, 0x0000, 0x1209, 0x0000,
+	0x1211, 0x0000, 0x1219, 0x0000, 0x1221, 0x0000, 0x1229, 0x0000,
+	/* Node for range 0x01e4_ */
+	0x1231, 0x0000, 0x1239, 0x0000, 0x1241, 0x0000, 0x1249, 0x0000,
+	0x1251, 0x0000, 0x1259, 0x0000, 0x1261, 0x0000, 0x1269, 0x0000,
+	/* Node for range 0x01e5_ */
+	0x1271, 0x0000, 0x1279, 0x0000, 0x1281, 0x0000, 0x1289, 0x0000,
+	0x1291, 0x0000, 0x1299, 0x0000, 0x12a1, 0x0000, 0x12a9, 0x0000,
+	/* Node for range 0x01e6_ */
+	0x12b1, 0x0000, 0x12b9, 0x0000, 0x12c1, 0x0000, 0x12c9, 0x0000,
+	0x12d1, 0x0000, 0x12d9, 0x0000, 0x12e1, 0x0000, 0x12e9, 0x0000,
+	/* Node for range 0x01e7_ */
+	0x12f1, 0x0000, 0x12f9, 0x0000, 0x1301, 0x0000, 0x1309, 0x0000,
+	0x1311, 0x0000, 0x1319, 0x0000, 0x1321, 0x0000, 0x1329, 0x0000,
+	/* Node for range 0x01e8_ */
+	0x1331, 0x0000, 0x1339, 0x0000, 0x1341, 0x0000, 0x1349, 0x0000,
+	0x1351, 0x0000, 0x1359, 0x0000, 0x1361, 0x0000, 0x1369, 0x0000,
+	/* Node for range 0x01e9_ */
+	0x1371, 0x0000, 0x1379, 0x0000, 0x1381, 0x0000, 0x138a, 0x139a,
+	0x13aa, 0x13ba, 0x13ca, 0x13d9, 0x0000, 0x0000, 0x13e2, 0x0000,
+	/* Node for range 0x01ea_ */
+	0x13f1, 0x0000, 0x13f9, 0x0000, 0x1401, 0x0000, 0x1409, 0x0000,
+	0x1411, 0x0000, 0x1419, 0x0000, 0x1421, 0x0000, 0x1429, 0x0000,
+	/* Node for range 0x01eb_ */
+	0x1431, 0x0000, 0x1439, 0x0000, 0x1441, 0x0000, 0x1449, 0x0000,
+	0x1451, 0x0000, 0x1459, 0x0000, 0x1461, 0x0000, 0x1469, 0x0000,
+	/* Node for range 0x01ec_ */
+	0x1471, 0x0000, 0x1479, 0x0000, 0x1481, 0x0000, 0x1489, 0x0000,
+	0x1491, 0x0000, 0x1499, 0x0000, 0x14a1, 0x0000, 0x14a9, 0x0000,
+	/* Node for range 0x01ed_ */
+	0x14b1, 0x0000, 0x14b9, 0x0000, 0x14c1, 0x0000, 0x14c9, 0x0000,
+	0x14d1, 0x0000, 0x14d9, 0x0000, 0x14e1, 0x0000, 0x14e9, 0x0000,
+	/* Node for range 0x01ee_ */
+	0x14f1, 0x0000, 0x14f9, 0x0000, 0x1501, 0x0000, 0x1509, 0x0000,
+	0x1511, 0x0000, 0x1519, 0x0000, 0x1521, 0x0000, 0x1529, 0x0000,
+	/* Node for range 0x01ef_ */
+	0x1531, 0x0000, 0x1539, 0x0000, 0x1541, 0x0000, 0x1549, 0x0000,
+	0x1551, 0x0000, 0x1559, 0x0000, 0x1561, 0x0000, 0x1569, 0x0000,
+	/* Node for range 0x01f0_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x1571, 0x1579, 0x1581, 0x1589, 0x1591, 0x1599, 0x15a1, 0x15a9,
+	/* Node for range 0x01f1_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x15b1, 0x15b9, 0x15c1, 0x15c9, 0x15d1, 0x15d9, 0x0000, 0x0000,
+	/* Node for range 0x01f2_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x15e1, 0x15e9, 0x15f1, 0x15f9, 0x1601, 0x1609, 0x1611, 0x1619,
+	/* Node for range 0x01f3_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x1621, 0x1629, 0x1631, 0x1639, 0x1641, 0x1649, 0x1651, 0x1659,
+	/* Node for range 0x01f4_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x1661, 0x1669, 0x1671, 0x1679, 0x1681, 0x1689, 0x0000, 0x0000,
+	/* Node for range 0x01f5_ */
+	0x1692, 0x0000, 0x16a3, 0x0000, 0x16bb, 0x0000, 0x16d3, 0x0000,
+	0x0000, 0x16e9, 0x0000, 0x16f1, 0x0000, 0x16f9, 0x0000, 0x1701,
+	/* Node for range 0x01f6_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x1709, 0x1711, 0x1719, 0x1721, 0x1729, 0x1731, 0x1739, 0x1741,
+	/* Node for range 0x01f8_ */
+	0x174a, 0x175a, 0x176a, 0x177a, 0x178a, 0x179a, 0x17aa, 0x17ba,
+	0x17ca, 0x17da, 0x17ea, 0x17fa, 0x180a, 0x181a, 0x182a, 0x183a,
+	/* Node for range 0x01f9_ */
+	0x184a, 0x185a, 0x186a, 0x187a, 0x188a, 0x189a, 0x18aa, 0x18ba,
+	0x18ca, 0x18da, 0x18ea, 0x18fa, 0x190a, 0x191a, 0x192a, 0x193a,
+	/* Node for range 0x01fa_ */
+	0x194a, 0x195a, 0x196a, 0x197a, 0x198a, 0x199a, 0x19aa, 0x19ba,
+	0x19ca, 0x19da, 0x19ea, 0x19fa, 0x1a0a, 0x1a1a, 0x1a2a, 0x1a3a,
+	/* Node for range 0x01fb_ */
+	0x0000, 0x0000, 0x1a4a, 0x1a5a, 0x1a6a, 0x0000, 0x1a7a, 0x1a8b,
+	0x1aa1, 0x1aa9, 0x1ab1, 0x1ab9, 0x1ac2, 0x0000, 0x1ad1, 0x0000,
+	/* Node for range 0x01fc_ */
+	0x0000, 0x0000, 0x1ada, 0x1aea, 0x1afa, 0x0000, 0x1b0a, 0x1b1b,
+	0x1b31, 0x1b39, 0x1b41, 0x1b49, 0x1b52, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01fd_ */
+	0x0000, 0x0000, 0x1b63, 0x1b7b, 0x0000, 0x0000, 0x1b92, 0x1ba3,
+	0x1bb9, 0x1bc1, 0x1bc9, 0x1bd1, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01fe_ */
+	0x0000, 0x0000, 0x1bdb, 0x1bf3, 0x1c0a, 0x0000, 0x1c1a, 0x1c2b,
+	0x1c41, 0x1c49, 0x1c51, 0x1c59, 0x1c61, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x01ff_ */
+	0x0000, 0x0000, 0x1c6a, 0x1c7a, 0x1c8a, 0x0000, 0x1c9a, 0x1cab,
+	0x1cc1, 0x1cc9, 0x1cd1, 0x1cd9, 0x1ce2, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0212_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x1cf1, 0x0000,
+	0x0000, 0x0000, 0x1cf9, 0x1d01, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0213_ */
+	0x0000, 0x0000, 0x1d09, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0216_ */
+	0x1d11, 0x1d19, 0x1d21, 0x1d29, 0x1d31, 0x1d39, 0x1d41, 0x1d49,
+	0x1d51, 0x1d59, 0x1d61, 0x1d69, 0x1d71, 0x1d79, 0x1d81, 0x1d89,
+	/* Node for range 0x0218_ */
+	0x0000, 0x0000, 0x0000, 0x1d91, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x024b_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x1d99, 0x1da1,
+	0x1da9, 0x1db1, 0x1db9, 0x1dc1, 0x1dc9, 0x1dd1, 0x1dd9, 0x1de1,
+	/* Node for range 0x024c_ */
+	0x1de9, 0x1df1, 0x1df9, 0x1e01, 0x1e09, 0x1e11, 0x1e19, 0x1e21,
+	0x1e29, 0x1e31, 0x1e39, 0x1e41, 0x1e49, 0x1e51, 0x1e59, 0x1e61,
+	/* Node for range 0x02c0_ */
+	0x1e69, 0x1e71, 0x1e79, 0x1e81, 0x1e89, 0x1e91, 0x1e99, 0x1ea1,
+	0x1ea9, 0x1eb1, 0x1eb9, 0x1ec1, 0x1ec9, 0x1ed1, 0x1ed9, 0x1ee1,
+	/* Node for range 0x02c1_ */
+	0x1ee9, 0x1ef1, 0x1ef9, 0x1f01, 0x1f09, 0x1f11, 0x1f19, 0x1f21,
+	0x1f29, 0x1f31, 0x1f39, 0x1f41, 0x1f49, 0x1f51, 0x1f59, 0x1f61,
+	/* Node for range 0x02c2_ */
+	0x1f69, 0x1f71, 0x1f79, 0x1f81, 0x1f89, 0x1f91, 0x1f99, 0x1fa1,
+	0x1fa9, 0x1fb1, 0x1fb9, 0x1fc1, 0x1fc9, 0x1fd1, 0x1fd9, 0x0000,
+	/* Node for range 0x02c6_ */
+	0x1fe1, 0x0000, 0x1fe9, 0x1ff1, 0x1ff9, 0x0000, 0x0000, 0x2001,
+	0x0000, 0x2009, 0x0000, 0x2011, 0x0000, 0x2019, 0x2021, 0x2029,
+	/* Node for range 0x02c7_ */
+	0x2031, 0x0000, 0x2039, 0x0000, 0x0000, 0x2041, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x2049, 0x2051,
+	/* Node for range 0x02c8_ */
+	0x2059, 0x0000, 0x2061, 0x0000, 0x2069, 0x0000, 0x2071, 0x0000,
+	0x2079, 0x0000, 0x2081, 0x0000, 0x2089, 0x0000, 0x2091, 0x0000,
+	/* Node for range 0x02c9_ */
+	0x2099, 0x0000, 0x20a1, 0x0000, 0x20a9, 0x0000, 0x20b1, 0x0000,
+	0x20b9, 0x0000, 0x20c1, 0x0000, 0x20c9, 0x0000, 0x20d1, 0x0000,
+	/* Node for range 0x02ca_ */
+	0x20d9, 0x0000, 0x20e1, 0x0000, 0x20e9, 0x0000, 0x20f1, 0x0000,
+	0x20f9, 0x0000, 0x2101, 0x0000, 0x2109, 0x0000, 0x2111, 0x0000,
+	/* Node for range 0x02cb_ */
+	0x2119, 0x0000, 0x2121, 0x0000, 0x2129, 0x0000, 0x2131, 0x0000,
+	0x2139, 0x0000, 0x2141, 0x0000, 0x2149, 0x0000, 0x2151, 0x0000,
+	/* Node for range 0x02cc_ */
+	0x2159, 0x0000, 0x2161, 0x0000, 0x2169, 0x0000, 0x2171, 0x0000,
+	0x2179, 0x0000, 0x2181, 0x0000, 0x2189, 0x0000, 0x2191, 0x0000,
+	/* Node for range 0x02cd_ */
+	0x2199, 0x0000, 0x21a1, 0x0000, 0x21a9, 0x0000, 0x21b1, 0x0000,
+	0x21b9, 0x0000, 0x21c1, 0x0000, 0x21c9, 0x0000, 0x21d1, 0x0000,
+	/* Node for range 0x02ce_ */
+	0x21d9, 0x0000, 0x21e1, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x21e9, 0x0000, 0x21f1, 0x0000, 0x0000,
+	/* Node for range 0x02cf_ */
+	0x0000, 0x0000, 0x21f9, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0a64_ */
+	0x2201, 0x0000, 0x2209, 0x0000, 0x2211, 0x0000, 0x2219, 0x0000,
+	0x2221, 0x0000, 0x2229, 0x0000, 0x2231, 0x0000, 0x2239, 0x0000,
+	/* Node for range 0x0a65_ */
+	0x2241, 0x0000, 0x2249, 0x0000, 0x2251, 0x0000, 0x2259, 0x0000,
+	0x2261, 0x0000, 0x2269, 0x0000, 0x2271, 0x0000, 0x2279, 0x0000,
+	/* Node for range 0x0a66_ */
+	0x2281, 0x0000, 0x2289, 0x0000, 0x2291, 0x0000, 0x2299, 0x0000,
+	0x22a1, 0x0000, 0x22a9, 0x0000, 0x22b1, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0a68_ */
+	0x22b9, 0x0000, 0x22c1, 0x0000, 0x22c9, 0x0000, 0x22d1, 0x0000,
+	0x22d9, 0x0000, 0x22e1, 0x0000, 0x22e9, 0x0000, 0x22f1, 0x0000,
+	/* Node for range 0x0a69_ */
+	0x22f9, 0x0000, 0x2301, 0x0000, 0x2309, 0x0000, 0x2311, 0x0000,
+	0x2319, 0x0000, 0x2321, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0a72_ */
+	0x0000, 0x0000, 0x2329, 0x0000, 0x2331, 0x0000, 0x2339, 0x0000,
+	0x2341, 0x0000, 0x2349, 0x0000, 0x2351, 0x0000, 0x2359, 0x0000,
+	/* Node for range 0x0a73_ */
+	0x0000, 0x0000, 0x2361, 0x0000, 0x2369, 0x0000, 0x2371, 0x0000,
+	0x2379, 0x0000, 0x2381, 0x0000, 0x2389, 0x0000, 0x2391, 0x0000,
+	/* Node for range 0x0a74_ */
+	0x2399, 0x0000, 0x23a1, 0x0000, 0x23a9, 0x0000, 0x23b1, 0x0000,
+	0x23b9, 0x0000, 0x23c1, 0x0000, 0x23c9, 0x0000, 0x23d1, 0x0000,
+	/* Node for range 0x0a75_ */
+	0x23d9, 0x0000, 0x23e1, 0x0000, 0x23e9, 0x0000, 0x23f1, 0x0000,
+	0x23f9, 0x0000, 0x2401, 0x0000, 0x2409, 0x0000, 0x2411, 0x0000,
+	/* Node for range 0x0a76_ */
+	0x2419, 0x0000, 0x2421, 0x0000, 0x2429, 0x0000, 0x2431, 0x0000,
+	0x2439, 0x0000, 0x2441, 0x0000, 0x2449, 0x0000, 0x2451, 0x0000,
+	/* Node for range 0x0a77_ */
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x2459, 0x0000, 0x2461, 0x0000, 0x2469, 0x2471, 0x0000,
+	/* Node for range 0x0a78_ */
+	0x2479, 0x0000, 0x2481, 0x0000, 0x2489, 0x0000, 0x2491, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x2499, 0x0000, 0x24a1, 0x0000, 0x0000,
+	/* Node for range 0x0a79_ */
+	0x24a9, 0x0000, 0x24b1, 0x0000, 0x0000, 0x0000, 0x24b9, 0x0000,
+	0x24c1, 0x0000, 0x24c9, 0x0000, 0x24d1, 0x0000, 0x24d9, 0x0000,
+	/* Node for range 0x0a7a_ */
+	0x24e1, 0x0000, 0x24e9, 0x0000, 0x24f1, 0x0000, 0x24f9, 0x0000,
+	0x2501, 0x0000, 0x2509, 0x2511, 0x2519, 0x2521, 0x2529, 0x0000,
+	/* Node for range 0x0a7b_ */
+	0x2531, 0x2539, 0x2541, 0x2549, 0x2551, 0x0000, 0x2559, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0ab7_ */
+	0x2561, 0x2569, 0x2571, 0x2579, 0x2581, 0x2589, 0x2591, 0x2599,
+	0x25a1, 0x25a9, 0x25b1, 0x25b9, 0x25c1, 0x25c9, 0x25d1, 0x25d9,
+	/* Node for range 0x0ab8_ */
+	0x25e1, 0x25e9, 0x25f1, 0x25f9, 0x2601, 0x2609, 0x2611, 0x2619,
+	0x2621, 0x2629, 0x2631, 0x2639, 0x2641, 0x2649, 0x2651, 0x2659,
+	/* Node for range 0x0ab9_ */
+	0x2661, 0x2669, 0x2671, 0x2679, 0x2681, 0x2689, 0x2691, 0x2699,
+	0x26a1, 0x26a9, 0x26b1, 0x26b9, 0x26c1, 0x26c9, 0x26d1, 0x26d9,
+	/* Node for range 0x0aba_ */
+	0x26e1, 0x26e9, 0x26f1, 0x26f9, 0x2701, 0x2709, 0x2711, 0x2719,
+	0x2721, 0x2729, 0x2731, 0x2739, 0x2741, 0x2749, 0x2751, 0x2759,
+	/* Node for range 0x0abb_ */
+	0x2761, 0x2769, 0x2771, 0x2779, 0x2781, 0x2789, 0x2791, 0x2799,
+	0x27a1, 0x27a9, 0x27b1, 0x27b9, 0x27c1, 0x27c9, 0x27d1, 0x27d9,
+	/* Node for range 0x0fb0_ */
+	0x27e2, 0x27f2, 0x2802, 0x2813, 0x282b, 0x2842, 0x2852, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0fb1_ */
+	0x0000, 0x0000, 0x0000, 0x2862, 0x2872, 0x2882, 0x2892, 0x28a2,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x0ff2_ */
+	0x0000, 0x28b1, 0x28b9, 0x28c1, 0x28c9, 0x28d1, 0x28d9, 0x28e1,
+	0x28e9, 0x28f1, 0x28f9, 0x2901, 0x2909, 0x2911, 0x2919, 0x2921,
+	/* Node for range 0x0ff3_ */
+	0x2929, 0x2931, 0x2939, 0x2941, 0x2949, 0x2951, 0x2959, 0x2961,
+	0x2969, 0x2971, 0x2979, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x1040_ */
+	0x2981, 0x2989, 0x2991, 0x2999, 0x29a1, 0x29a9, 0x29b1, 0x29b9,
+	0x29c1, 0x29c9, 0x29d1, 0x29d9, 0x29e1, 0x29e9, 0x29f1, 0x29f9,
+	/* Node for range 0x1041_ */
+	0x2a01, 0x2a09, 0x2a11, 0x2a19, 0x2a21, 0x2a29, 0x2a31, 0x2a39,
+	0x2a41, 0x2a49, 0x2a51, 0x2a59, 0x2a61, 0x2a69, 0x2a71, 0x2a79,
+	/* Node for range 0x1042_ */
+	0x2a81, 0x2a89, 0x2a91, 0x2a99, 0x2aa1, 0x2aa9, 0x2ab1, 0x2ab9,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x104b_ */
+	0x2ac1, 0x2ac9, 0x2ad1, 0x2ad9, 0x2ae1, 0x2ae9, 0x2af1, 0x2af9,
+	0x2b01, 0x2b09, 0x2b11, 0x2b19, 0x2b21, 0x2b29, 0x2b31, 0x2b39,
+	/* Node for range 0x104c_ */
+	0x2b41, 0x2b49, 0x2b51, 0x2b59, 0x2b61, 0x2b69, 0x2b71, 0x2b79,
+	0x2b81, 0x2b89, 0x2b91, 0x2b99, 0x2ba1, 0x2ba9, 0x2bb1, 0x2bb9,
+	/* Node for range 0x104d_ */
+	0x2bc1, 0x2bc9, 0x2bd1, 0x2bd9, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x10c8_ */
+	0x2be1, 0x2be9, 0x2bf1, 0x2bf9, 0x2c01, 0x2c09, 0x2c11, 0x2c19,
+	0x2c21, 0x2c29, 0x2c31, 0x2c39, 0x2c41, 0x2c49, 0x2c51, 0x2c59,
+	/* Node for range 0x10c9_ */
+	0x2c61, 0x2c69, 0x2c71, 0x2c79, 0x2c81, 0x2c89, 0x2c91, 0x2c99,
+	0x2ca1, 0x2ca9, 0x2cb1, 0x2cb9, 0x2cc1, 0x2cc9, 0x2cd1, 0x2cd9,
+	/* Node for range 0x10ca_ */
+	0x2ce1, 0x2ce9, 0x2cf1, 0x2cf9, 0x2d01, 0x2d09, 0x2d11, 0x2d19,
+	0x2d21, 0x2d29, 0x2d31, 0x2d39, 0x2d41, 0x2d49, 0x2d51, 0x2d59,
+	/* Node for range 0x10cb_ */
+	0x2d61, 0x2d69, 0x2d71, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	/* Node for range 0x118a_ */
+	0x2d79, 0x2d81, 0x2d89, 0x2d91, 0x2d99, 0x2da1, 0x2da9, 0x2db1,
+	0x2db9, 0x2dc1, 0x2dc9, 0x2dd1, 0x2dd9, 0x2de1, 0x2de9, 0x2df1,
+	/* Node for range 0x118b_ */
+	0x2df9, 0x2e01, 0x2e09, 0x2e11, 0x2e19, 0x2e21, 0x2e29, 0x2e31,
+	0x2e39, 0x2e41, 0x2e49, 0x2e51, 0x2e59, 0x2e61, 0x2e69, 0x2e71,
+	/* Node for range 0x1e90_ */
+	0x2e79, 0x2e81, 0x2e89, 0x2e91, 0x2e99, 0x2ea1, 0x2ea9, 0x2eb1,
+	0x2eb9, 0x2ec1, 0x2ec9, 0x2ed1, 0x2ed9, 0x2ee1, 0x2ee9, 0x2ef1,
+	/* Node for range 0x1e91_ */
+	0x2ef9, 0x2f01, 0x2f09, 0x2f11, 0x2f19, 0x2f21, 0x2f29, 0x2f31,
+	0x2f39, 0x2f41, 0x2f49, 0x2f51, 0x2f59, 0x2f61, 0x2f69, 0x2f71,
+	/* Node for range 0x1e92_ */
+	0x2f79, 0x2f81, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+	0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+};
+
+static unicode_t apfs_cf[] = {
+	0x000061, 0x000062, 0x000063, 0x000064, 0x000065, 0x000066,
+	0x000067, 0x000068, 0x000069, 0x00006a, 0x00006b, 0x00006c,
+	0x00006d, 0x00006e, 0x00006f, 0x000070, 0x000071, 0x000072,
+	0x000073, 0x000074, 0x000075, 0x000076, 0x000077, 0x000078,
+	0x000079, 0x00007a, 0x0003bc, 0x0000e0, 0x0000e1, 0x0000e2,
+	0x0000e3, 0x0000e4, 0x0000e5, 0x0000e6, 0x0000e7, 0x0000e8,
+	0x0000e9, 0x0000ea, 0x0000eb, 0x0000ec, 0x0000ed, 0x0000ee,
+	0x0000ef, 0x0000f0, 0x0000f1, 0x0000f2, 0x0000f3, 0x0000f4,
+	0x0000f5, 0x0000f6, 0x0000f8, 0x0000f9, 0x0000fa, 0x0000fb,
+	0x0000fc, 0x0000fd, 0x0000fe, 0x000073, 0x000073, 0x000101,
+	0x000103, 0x000105, 0x000107, 0x000109, 0x00010b, 0x00010d,
+	0x00010f, 0x000111, 0x000113, 0x000115, 0x000117, 0x000119,
+	0x00011b, 0x00011d, 0x00011f, 0x000121, 0x000123, 0x000125,
+	0x000127, 0x000129, 0x00012b, 0x00012d, 0x00012f, 0x000069,
+	0x000307, 0x000133, 0x000135, 0x000137, 0x00013a, 0x00013c,
+	0x00013e, 0x000140, 0x000142, 0x000144, 0x000146, 0x000148,
+	0x0002bc, 0x00006e, 0x00014b, 0x00014d, 0x00014f, 0x000151,
+	0x000153, 0x000155, 0x000157, 0x000159, 0x00015b, 0x00015d,
+	0x00015f, 0x000161, 0x000163, 0x000165, 0x000167, 0x000169,
+	0x00016b, 0x00016d, 0x00016f, 0x000171, 0x000173, 0x000175,
+	0x000177, 0x0000ff, 0x00017a, 0x00017c, 0x00017e, 0x000073,
+	0x000253, 0x000183, 0x000185, 0x000254, 0x000188, 0x000256,
+	0x000257, 0x00018c, 0x0001dd, 0x000259, 0x00025b, 0x000192,
+	0x000260, 0x000263, 0x000269, 0x000268, 0x000199, 0x00026f,
+	0x000272, 0x000275, 0x0001a1, 0x0001a3, 0x0001a5, 0x000280,
+	0x0001a8, 0x000283, 0x0001ad, 0x000288, 0x0001b0, 0x00028a,
+	0x00028b, 0x0001b4, 0x0001b6, 0x000292, 0x0001b9, 0x0001bd,
+	0x0001c6, 0x0001c6, 0x0001c9, 0x0001c9, 0x0001cc, 0x0001cc,
+	0x0001ce, 0x0001d0, 0x0001d2, 0x0001d4, 0x0001d6, 0x0001d8,
+	0x0001da, 0x0001dc, 0x0001df, 0x0001e1, 0x0001e3, 0x0001e5,
+	0x0001e7, 0x0001e9, 0x0001eb, 0x0001ed, 0x0001ef, 0x00006a,
+	0x00030c, 0x0001f3, 0x0001f3, 0x0001f5, 0x000195, 0x0001bf,
+	0x0001f9, 0x0001fb, 0x0001fd, 0x0001ff, 0x000201, 0x000203,
+	0x000205, 0x000207, 0x000209, 0x00020b, 0x00020d, 0x00020f,
+	0x000211, 0x000213, 0x000215, 0x000217, 0x000219, 0x00021b,
+	0x00021d, 0x00021f, 0x00019e, 0x000223, 0x000225, 0x000227,
+	0x000229, 0x00022b, 0x00022d, 0x00022f, 0x000231, 0x000233,
+	0x002c65, 0x00023c, 0x00019a, 0x002c66, 0x000242, 0x000180,
+	0x000289, 0x00028c, 0x000247, 0x000249, 0x00024b, 0x00024d,
+	0x00024f, 0x0003b9, 0x000371, 0x000373, 0x000377, 0x0003f3,
+	0x0003ac, 0x0003ad, 0x0003ae, 0x0003af, 0x0003cc, 0x0003cd,
+	0x0003ce, 0x0003b9, 0x000308, 0x000301, 0x0003b1, 0x0003b2,
+	0x0003b3, 0x0003b4, 0x0003b5, 0x0003b6, 0x0003b7, 0x0003b8,
+	0x0003b9, 0x0003ba, 0x0003bb, 0x0003bc, 0x0003bd, 0x0003be,
+	0x0003bf, 0x0003c0, 0x0003c1, 0x0003c3, 0x0003c4, 0x0003c5,
+	0x0003c6, 0x0003c7, 0x0003c8, 0x0003c9, 0x0003ca, 0x0003cb,
+	0x0003c5, 0x000308, 0x000301, 0x0003c3, 0x0003d7, 0x0003b2,
+	0x0003b8, 0x0003c6, 0x0003c0, 0x0003d9, 0x0003db, 0x0003dd,
+	0x0003df, 0x0003e1, 0x0003e3, 0x0003e5, 0x0003e7, 0x0003e9,
+	0x0003eb, 0x0003ed, 0x0003ef, 0x0003ba, 0x0003c1, 0x0003b8,
+	0x0003b5, 0x0003f8, 0x0003f2, 0x0003fb, 0x00037b, 0x00037c,
+	0x00037d, 0x000450, 0x000451, 0x000452, 0x000453, 0x000454,
+	0x000455, 0x000456, 0x000457, 0x000458, 0x000459, 0x00045a,
+	0x00045b, 0x00045c, 0x00045d, 0x00045e, 0x00045f, 0x000430,
+	0x000431, 0x000432, 0x000433, 0x000434, 0x000435, 0x000436,
+	0x000437, 0x000438, 0x000439, 0x00043a, 0x00043b, 0x00043c,
+	0x00043d, 0x00043e, 0x00043f, 0x000440, 0x000441, 0x000442,
+	0x000443, 0x000444, 0x000445, 0x000446, 0x000447, 0x000448,
+	0x000449, 0x00044a, 0x00044b, 0x00044c, 0x00044d, 0x00044e,
+	0x00044f, 0x000461, 0x000463, 0x000465, 0x000467, 0x000469,
+	0x00046b, 0x00046d, 0x00046f, 0x000471, 0x000473, 0x000475,
+	0x000477, 0x000479, 0x00047b, 0x00047d, 0x00047f, 0x000481,
+	0x00048b, 0x00048d, 0x00048f, 0x000491, 0x000493, 0x000495,
+	0x000497, 0x000499, 0x00049b, 0x00049d, 0x00049f, 0x0004a1,
+	0x0004a3, 0x0004a5, 0x0004a7, 0x0004a9, 0x0004ab, 0x0004ad,
+	0x0004af, 0x0004b1, 0x0004b3, 0x0004b5, 0x0004b7, 0x0004b9,
+	0x0004bb, 0x0004bd, 0x0004bf, 0x0004cf, 0x0004c2, 0x0004c4,
+	0x0004c6, 0x0004c8, 0x0004ca, 0x0004cc, 0x0004ce, 0x0004d1,
+	0x0004d3, 0x0004d5, 0x0004d7, 0x0004d9, 0x0004db, 0x0004dd,
+	0x0004df, 0x0004e1, 0x0004e3, 0x0004e5, 0x0004e7, 0x0004e9,
+	0x0004eb, 0x0004ed, 0x0004ef, 0x0004f1, 0x0004f3, 0x0004f5,
+	0x0004f7, 0x0004f9, 0x0004fb, 0x0004fd, 0x0004ff, 0x000501,
+	0x000503, 0x000505, 0x000507, 0x000509, 0x00050b, 0x00050d,
+	0x00050f, 0x000511, 0x000513, 0x000515, 0x000517, 0x000519,
+	0x00051b, 0x00051d, 0x00051f, 0x000521, 0x000523, 0x000525,
+	0x000527, 0x000529, 0x00052b, 0x00052d, 0x00052f, 0x000561,
+	0x000562, 0x000563, 0x000564, 0x000565, 0x000566, 0x000567,
+	0x000568, 0x000569, 0x00056a, 0x00056b, 0x00056c, 0x00056d,
+	0x00056e, 0x00056f, 0x000570, 0x000571, 0x000572, 0x000573,
+	0x000574, 0x000575, 0x000576, 0x000577, 0x000578, 0x000579,
+	0x00057a, 0x00057b, 0x00057c, 0x00057d, 0x00057e, 0x00057f,
+	0x000580, 0x000581, 0x000582, 0x000583, 0x000584, 0x000585,
+	0x000586, 0x000565, 0x000582, 0x002d00, 0x002d01, 0x002d02,
+	0x002d03, 0x002d04, 0x002d05, 0x002d06, 0x002d07, 0x002d08,
+	0x002d09, 0x002d0a, 0x002d0b, 0x002d0c, 0x002d0d, 0x002d0e,
+	0x002d0f, 0x002d10, 0x002d11, 0x002d12, 0x002d13, 0x002d14,
+	0x002d15, 0x002d16, 0x002d17, 0x002d18, 0x002d19, 0x002d1a,
+	0x002d1b, 0x002d1c, 0x002d1d, 0x002d1e, 0x002d1f, 0x002d20,
+	0x002d21, 0x002d22, 0x002d23, 0x002d24, 0x002d25, 0x002d27,
+	0x002d2d, 0x0013f0, 0x0013f1, 0x0013f2, 0x0013f3, 0x0013f4,
+	0x0013f5, 0x000432, 0x000434, 0x00043e, 0x000441, 0x000442,
+	0x000442, 0x00044a, 0x000463, 0x00a64b, 0x001e01, 0x001e03,
+	0x001e05, 0x001e07, 0x001e09, 0x001e0b, 0x001e0d, 0x001e0f,
+	0x001e11, 0x001e13, 0x001e15, 0x001e17, 0x001e19, 0x001e1b,
+	0x001e1d, 0x001e1f, 0x001e21, 0x001e23, 0x001e25, 0x001e27,
+	0x001e29, 0x001e2b, 0x001e2d, 0x001e2f, 0x001e31, 0x001e33,
+	0x001e35, 0x001e37, 0x001e39, 0x001e3b, 0x001e3d, 0x001e3f,
+	0x001e41, 0x001e43, 0x001e45, 0x001e47, 0x001e49, 0x001e4b,
+	0x001e4d, 0x001e4f, 0x001e51, 0x001e53, 0x001e55, 0x001e57,
+	0x001e59, 0x001e5b, 0x001e5d, 0x001e5f, 0x001e61, 0x001e63,
+	0x001e65, 0x001e67, 0x001e69, 0x001e6b, 0x001e6d, 0x001e6f,
+	0x001e71, 0x001e73, 0x001e75, 0x001e77, 0x001e79, 0x001e7b,
+	0x001e7d, 0x001e7f, 0x001e81, 0x001e83, 0x001e85, 0x001e87,
+	0x001e89, 0x001e8b, 0x001e8d, 0x001e8f, 0x001e91, 0x001e93,
+	0x001e95, 0x000068, 0x000331, 0x000074, 0x000308, 0x000077,
+	0x00030a, 0x000079, 0x00030a, 0x000061, 0x0002be, 0x001e61,
+	0x000073, 0x000073, 0x001ea1, 0x001ea3, 0x001ea5, 0x001ea7,
+	0x001ea9, 0x001eab, 0x001ead, 0x001eaf, 0x001eb1, 0x001eb3,
+	0x001eb5, 0x001eb7, 0x001eb9, 0x001ebb, 0x001ebd, 0x001ebf,
+	0x001ec1, 0x001ec3, 0x001ec5, 0x001ec7, 0x001ec9, 0x001ecb,
+	0x001ecd, 0x001ecf, 0x001ed1, 0x001ed3, 0x001ed5, 0x001ed7,
+	0x001ed9, 0x001edb, 0x001edd, 0x001edf, 0x001ee1, 0x001ee3,
+	0x001ee5, 0x001ee7, 0x001ee9, 0x001eeb, 0x001eed, 0x001eef,
+	0x001ef1, 0x001ef3, 0x001ef5, 0x001ef7, 0x001ef9, 0x001efb,
+	0x001efd, 0x001eff, 0x001f00, 0x001f01, 0x001f02, 0x001f03,
+	0x001f04, 0x001f05, 0x001f06, 0x001f07, 0x001f10, 0x001f11,
+	0x001f12, 0x001f13, 0x001f14, 0x001f15, 0x001f20, 0x001f21,
+	0x001f22, 0x001f23, 0x001f24, 0x001f25, 0x001f26, 0x001f27,
+	0x001f30, 0x001f31, 0x001f32, 0x001f33, 0x001f34, 0x001f35,
+	0x001f36, 0x001f37, 0x001f40, 0x001f41, 0x001f42, 0x001f43,
+	0x001f44, 0x001f45, 0x0003c5, 0x000313, 0x0003c5, 0x000313,
+	0x000300, 0x0003c5, 0x000313, 0x000301, 0x0003c5, 0x000313,
+	0x000342, 0x001f51, 0x001f53, 0x001f55, 0x001f57, 0x001f60,
+	0x001f61, 0x001f62, 0x001f63, 0x001f64, 0x001f65, 0x001f66,
+	0x001f67, 0x001f00, 0x0003b9, 0x001f01, 0x0003b9, 0x001f02,
+	0x0003b9, 0x001f03, 0x0003b9, 0x001f04, 0x0003b9, 0x001f05,
+	0x0003b9, 0x001f06, 0x0003b9, 0x001f07, 0x0003b9, 0x001f00,
+	0x0003b9, 0x001f01, 0x0003b9, 0x001f02, 0x0003b9, 0x001f03,
+	0x0003b9, 0x001f04, 0x0003b9, 0x001f05, 0x0003b9, 0x001f06,
+	0x0003b9, 0x001f07, 0x0003b9, 0x001f20, 0x0003b9, 0x001f21,
+	0x0003b9, 0x001f22, 0x0003b9, 0x001f23, 0x0003b9, 0x001f24,
+	0x0003b9, 0x001f25, 0x0003b9, 0x001f26, 0x0003b9, 0x001f27,
+	0x0003b9, 0x001f20, 0x0003b9, 0x001f21, 0x0003b9, 0x001f22,
+	0x0003b9, 0x001f23, 0x0003b9, 0x001f24, 0x0003b9, 0x001f25,
+	0x0003b9, 0x001f26, 0x0003b9, 0x001f27, 0x0003b9, 0x001f60,
+	0x0003b9, 0x001f61, 0x0003b9, 0x001f62, 0x0003b9, 0x001f63,
+	0x0003b9, 0x001f64, 0x0003b9, 0x001f65, 0x0003b9, 0x001f66,
+	0x0003b9, 0x001f67, 0x0003b9, 0x001f60, 0x0003b9, 0x001f61,
+	0x0003b9, 0x001f62, 0x0003b9, 0x001f63, 0x0003b9, 0x001f64,
+	0x0003b9, 0x001f65, 0x0003b9, 0x001f66, 0x0003b9, 0x001f67,
+	0x0003b9, 0x001f70, 0x0003b9, 0x0003b1, 0x0003b9, 0x0003ac,
+	0x0003b9, 0x0003b1, 0x000342, 0x0003b1, 0x000342, 0x0003b9,
+	0x001fb0, 0x001fb1, 0x001f70, 0x001f71, 0x0003b1, 0x0003b9,
+	0x0003b9, 0x001f74, 0x0003b9, 0x0003b7, 0x0003b9, 0x0003ae,
+	0x0003b9, 0x0003b7, 0x000342, 0x0003b7, 0x000342, 0x0003b9,
+	0x001f72, 0x001f73, 0x001f74, 0x001f75, 0x0003b7, 0x0003b9,
+	0x0003b9, 0x000308, 0x000300, 0x0003b9, 0x000308, 0x000301,
+	0x0003b9, 0x000342, 0x0003b9, 0x000308, 0x000342, 0x001fd0,
+	0x001fd1, 0x001f76, 0x001f77, 0x0003c5, 0x000308, 0x000300,
+	0x0003c5, 0x000308, 0x000301, 0x0003c1, 0x000313, 0x0003c5,
+	0x000342, 0x0003c5, 0x000308, 0x000342, 0x001fe0, 0x001fe1,
+	0x001f7a, 0x001f7b, 0x001fe5, 0x001f7c, 0x0003b9, 0x0003c9,
+	0x0003b9, 0x0003ce, 0x0003b9, 0x0003c9, 0x000342, 0x0003c9,
+	0x000342, 0x0003b9, 0x001f78, 0x001f79, 0x001f7c, 0x001f7d,
+	0x0003c9, 0x0003b9, 0x0003c9, 0x00006b, 0x0000e5, 0x00214e,
+	0x002170, 0x002171, 0x002172, 0x002173, 0x002174, 0x002175,
+	0x002176, 0x002177, 0x002178, 0x002179, 0x00217a, 0x00217b,
+	0x00217c, 0x00217d, 0x00217e, 0x00217f, 0x002184, 0x0024d0,
+	0x0024d1, 0x0024d2, 0x0024d3, 0x0024d4, 0x0024d5, 0x0024d6,
+	0x0024d7, 0x0024d8, 0x0024d9, 0x0024da, 0x0024db, 0x0024dc,
+	0x0024dd, 0x0024de, 0x0024df, 0x0024e0, 0x0024e1, 0x0024e2,
+	0x0024e3, 0x0024e4, 0x0024e5, 0x0024e6, 0x0024e7, 0x0024e8,
+	0x0024e9, 0x002c30, 0x002c31, 0x002c32, 0x002c33, 0x002c34,
+	0x002c35, 0x002c36, 0x002c37, 0x002c38, 0x002c39, 0x002c3a,
+	0x002c3b, 0x002c3c, 0x002c3d, 0x002c3e, 0x002c3f, 0x002c40,
+	0x002c41, 0x002c42, 0x002c43, 0x002c44, 0x002c45, 0x002c46,
+	0x002c47, 0x002c48, 0x002c49, 0x002c4a, 0x002c4b, 0x002c4c,
+	0x002c4d, 0x002c4e, 0x002c4f, 0x002c50, 0x002c51, 0x002c52,
+	0x002c53, 0x002c54, 0x002c55, 0x002c56, 0x002c57, 0x002c58,
+	0x002c59, 0x002c5a, 0x002c5b, 0x002c5c, 0x002c5d, 0x002c5e,
+	0x002c61, 0x00026b, 0x001d7d, 0x00027d, 0x002c68, 0x002c6a,
+	0x002c6c, 0x000251, 0x000271, 0x000250, 0x000252, 0x002c73,
+	0x002c76, 0x00023f, 0x000240, 0x002c81, 0x002c83, 0x002c85,
+	0x002c87, 0x002c89, 0x002c8b, 0x002c8d, 0x002c8f, 0x002c91,
+	0x002c93, 0x002c95, 0x002c97, 0x002c99, 0x002c9b, 0x002c9d,
+	0x002c9f, 0x002ca1, 0x002ca3, 0x002ca5, 0x002ca7, 0x002ca9,
+	0x002cab, 0x002cad, 0x002caf, 0x002cb1, 0x002cb3, 0x002cb5,
+	0x002cb7, 0x002cb9, 0x002cbb, 0x002cbd, 0x002cbf, 0x002cc1,
+	0x002cc3, 0x002cc5, 0x002cc7, 0x002cc9, 0x002ccb, 0x002ccd,
+	0x002ccf, 0x002cd1, 0x002cd3, 0x002cd5, 0x002cd7, 0x002cd9,
+	0x002cdb, 0x002cdd, 0x002cdf, 0x002ce1, 0x002ce3, 0x002cec,
+	0x002cee, 0x002cf3, 0x00a641, 0x00a643, 0x00a645, 0x00a647,
+	0x00a649, 0x00a64b, 0x00a64d, 0x00a64f, 0x00a651, 0x00a653,
+	0x00a655, 0x00a657, 0x00a659, 0x00a65b, 0x00a65d, 0x00a65f,
+	0x00a661, 0x00a663, 0x00a665, 0x00a667, 0x00a669, 0x00a66b,
+	0x00a66d, 0x00a681, 0x00a683, 0x00a685, 0x00a687, 0x00a689,
+	0x00a68b, 0x00a68d, 0x00a68f, 0x00a691, 0x00a693, 0x00a695,
+	0x00a697, 0x00a699, 0x00a69b, 0x00a723, 0x00a725, 0x00a727,
+	0x00a729, 0x00a72b, 0x00a72d, 0x00a72f, 0x00a733, 0x00a735,
+	0x00a737, 0x00a739, 0x00a73b, 0x00a73d, 0x00a73f, 0x00a741,
+	0x00a743, 0x00a745, 0x00a747, 0x00a749, 0x00a74b, 0x00a74d,
+	0x00a74f, 0x00a751, 0x00a753, 0x00a755, 0x00a757, 0x00a759,
+	0x00a75b, 0x00a75d, 0x00a75f, 0x00a761, 0x00a763, 0x00a765,
+	0x00a767, 0x00a769, 0x00a76b, 0x00a76d, 0x00a76f, 0x00a77a,
+	0x00a77c, 0x001d79, 0x00a77f, 0x00a781, 0x00a783, 0x00a785,
+	0x00a787, 0x00a78c, 0x000265, 0x00a791, 0x00a793, 0x00a797,
+	0x00a799, 0x00a79b, 0x00a79d, 0x00a79f, 0x00a7a1, 0x00a7a3,
+	0x00a7a5, 0x00a7a7, 0x00a7a9, 0x000266, 0x00025c, 0x000261,
+	0x00026c, 0x00026a, 0x00029e, 0x000287, 0x00029d, 0x00ab53,
+	0x00a7b5, 0x00a7b7, 0x0013a0, 0x0013a1, 0x0013a2, 0x0013a3,
+	0x0013a4, 0x0013a5, 0x0013a6, 0x0013a7, 0x0013a8, 0x0013a9,
+	0x0013aa, 0x0013ab, 0x0013ac, 0x0013ad, 0x0013ae, 0x0013af,
+	0x0013b0, 0x0013b1, 0x0013b2, 0x0013b3, 0x0013b4, 0x0013b5,
+	0x0013b6, 0x0013b7, 0x0013b8, 0x0013b9, 0x0013ba, 0x0013bb,
+	0x0013bc, 0x0013bd, 0x0013be, 0x0013bf, 0x0013c0, 0x0013c1,
+	0x0013c2, 0x0013c3, 0x0013c4, 0x0013c5, 0x0013c6, 0x0013c7,
+	0x0013c8, 0x0013c9, 0x0013ca, 0x0013cb, 0x0013cc, 0x0013cd,
+	0x0013ce, 0x0013cf, 0x0013d0, 0x0013d1, 0x0013d2, 0x0013d3,
+	0x0013d4, 0x0013d5, 0x0013d6, 0x0013d7, 0x0013d8, 0x0013d9,
+	0x0013da, 0x0013db, 0x0013dc, 0x0013dd, 0x0013de, 0x0013df,
+	0x0013e0, 0x0013e1, 0x0013e2, 0x0013e3, 0x0013e4, 0x0013e5,
+	0x0013e6, 0x0013e7, 0x0013e8, 0x0013e9, 0x0013ea, 0x0013eb,
+	0x0013ec, 0x0013ed, 0x0013ee, 0x0013ef, 0x000066, 0x000066,
+	0x000066, 0x000069, 0x000066, 0x00006c, 0x000066, 0x000066,
+	0x000069, 0x000066, 0x000066, 0x00006c, 0x000073, 0x000074,
+	0x000073, 0x000074, 0x000574, 0x000576, 0x000574, 0x000565,
+	0x000574, 0x00056b, 0x00057e, 0x000576, 0x000574, 0x00056d,
+	0x00ff41, 0x00ff42, 0x00ff43, 0x00ff44, 0x00ff45, 0x00ff46,
+	0x00ff47, 0x00ff48, 0x00ff49, 0x00ff4a, 0x00ff4b, 0x00ff4c,
+	0x00ff4d, 0x00ff4e, 0x00ff4f, 0x00ff50, 0x00ff51, 0x00ff52,
+	0x00ff53, 0x00ff54, 0x00ff55, 0x00ff56, 0x00ff57, 0x00ff58,
+	0x00ff59, 0x00ff5a, 0x010428, 0x010429, 0x01042a, 0x01042b,
+	0x01042c, 0x01042d, 0x01042e, 0x01042f, 0x010430, 0x010431,
+	0x010432, 0x010433, 0x010434, 0x010435, 0x010436, 0x010437,
+	0x010438, 0x010439, 0x01043a, 0x01043b, 0x01043c, 0x01043d,
+	0x01043e, 0x01043f, 0x010440, 0x010441, 0x010442, 0x010443,
+	0x010444, 0x010445, 0x010446, 0x010447, 0x010448, 0x010449,
+	0x01044a, 0x01044b, 0x01044c, 0x01044d, 0x01044e, 0x01044f,
+	0x0104d8, 0x0104d9, 0x0104da, 0x0104db, 0x0104dc, 0x0104dd,
+	0x0104de, 0x0104df, 0x0104e0, 0x0104e1, 0x0104e2, 0x0104e3,
+	0x0104e4, 0x0104e5, 0x0104e6, 0x0104e7, 0x0104e8, 0x0104e9,
+	0x0104ea, 0x0104eb, 0x0104ec, 0x0104ed, 0x0104ee, 0x0104ef,
+	0x0104f0, 0x0104f1, 0x0104f2, 0x0104f3, 0x0104f4, 0x0104f5,
+	0x0104f6, 0x0104f7, 0x0104f8, 0x0104f9, 0x0104fa, 0x0104fb,
+	0x010cc0, 0x010cc1, 0x010cc2, 0x010cc3, 0x010cc4, 0x010cc5,
+	0x010cc6, 0x010cc7, 0x010cc8, 0x010cc9, 0x010cca, 0x010ccb,
+	0x010ccc, 0x010ccd, 0x010cce, 0x010ccf, 0x010cd0, 0x010cd1,
+	0x010cd2, 0x010cd3, 0x010cd4, 0x010cd5, 0x010cd6, 0x010cd7,
+	0x010cd8, 0x010cd9, 0x010cda, 0x010cdb, 0x010cdc, 0x010cdd,
+	0x010cde, 0x010cdf, 0x010ce0, 0x010ce1, 0x010ce2, 0x010ce3,
+	0x010ce4, 0x010ce5, 0x010ce6, 0x010ce7, 0x010ce8, 0x010ce9,
+	0x010cea, 0x010ceb, 0x010cec, 0x010ced, 0x010cee, 0x010cef,
+	0x010cf0, 0x010cf1, 0x010cf2, 0x0118c0, 0x0118c1, 0x0118c2,
+	0x0118c3, 0x0118c4, 0x0118c5, 0x0118c6, 0x0118c7, 0x0118c8,
+	0x0118c9, 0x0118ca, 0x0118cb, 0x0118cc, 0x0118cd, 0x0118ce,
+	0x0118cf, 0x0118d0, 0x0118d1, 0x0118d2, 0x0118d3, 0x0118d4,
+	0x0118d5, 0x0118d6, 0x0118d7, 0x0118d8, 0x0118d9, 0x0118da,
+	0x0118db, 0x0118dc, 0x0118dd, 0x0118de, 0x0118df, 0x01e922,
+	0x01e923, 0x01e924, 0x01e925, 0x01e926, 0x01e927, 0x01e928,
+	0x01e929, 0x01e92a, 0x01e92b, 0x01e92c, 0x01e92d, 0x01e92e,
+	0x01e92f, 0x01e930, 0x01e931, 0x01e932, 0x01e933, 0x01e934,
+	0x01e935, 0x01e936, 0x01e937, 0x01e938, 0x01e939, 0x01e93a,
+	0x01e93b, 0x01e93c, 0x01e93d, 0x01e93e, 0x01e93f, 0x01e940,
+	0x01e941, 0x01e942, 0x01e943,
+};
+
+static u8 apfs_ccc_trie[] = {
+	/* Node for range 0x_____ */
+	0x01, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0____ */
+	0x03, 0x04, 0x05, 0x06, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x00, 0x08,
+	/* Node for range 0x1____ */
+	0x09, 0x0a, 0x00, 0x00, 0x00, 0x00, 0x0b, 0x00,
+	0x00, 0x00, 0x00, 0x0c, 0x00, 0x0d, 0x0e, 0x00,
+	/* Node for range 0x00___ */
+	0x00, 0x00, 0x00, 0x0f, 0x10, 0x11, 0x12, 0x13,
+	0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b,
+	/* Node for range 0x01___ */
+	0x1c, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x1e,
+	0x1f, 0x20, 0x21, 0x22, 0x23, 0x24, 0x00, 0x00,
+	/* Node for range 0x02___ */
+	0x25, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x26, 0x27, 0x00, 0x00,
+	/* Node for range 0x03___ */
+	0x28, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a___ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x29, 0x00,
+	0x2a, 0x2b, 0x2c, 0x2d, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0f___ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x2f, 0x00,
+	/* Node for range 0x10___ */
+	0x00, 0x30, 0x31, 0x32, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x11___ */
+	0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x3b,
+	0x00, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00,
+	/* Node for range 0x16___ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x3d, 0x3e, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1b___ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x3f, 0x00, 0x00, 0x00,
+	/* Node for range 0x1d___ */
+	0x00, 0x40, 0x41, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1e___ */
+	0x42, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x43, 0x44, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x003__ */
+	0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x004__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x4c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x005__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x4d, 0x4e, 0x4f, 0x50, 0x00, 0x00, 0x00,
+	/* Node for range 0x006__ */
+	0x00, 0x51, 0x00, 0x00, 0x52, 0x53, 0x00, 0x54,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x55, 0x56, 0x00,
+	/* Node for range 0x007__ */
+	0x00, 0x57, 0x00, 0x58, 0x59, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x5a, 0x5b,
+	/* Node for range 0x008__ */
+	0x00, 0x5c, 0x5d, 0x00, 0x00, 0x5e, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x5f, 0x60, 0x61,
+	/* Node for range 0x009__ */
+	0x00, 0x00, 0x00, 0x62, 0x63, 0x64, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x65, 0x66, 0x00, 0x00, 0x00,
+	/* Node for range 0x00a__ */
+	0x00, 0x00, 0x00, 0x67, 0x68, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x69, 0x6a, 0x00, 0x00, 0x00,
+	/* Node for range 0x00b__ */
+	0x00, 0x00, 0x00, 0x6b, 0x6c, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x6d, 0x00, 0x00, 0x00,
+	/* Node for range 0x00c__ */
+	0x00, 0x00, 0x00, 0x00, 0x6e, 0x6f, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x70, 0x71, 0x00, 0x00, 0x00,
+	/* Node for range 0x00d__ */
+	0x00, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x73, 0x00, 0x00, 0x00,
+	/* Node for range 0x00e__ */
+	0x00, 0x00, 0x00, 0x74, 0x75, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x76, 0x77, 0x00, 0x00, 0x00,
+	/* Node for range 0x00f__ */
+	0x00, 0x78, 0x00, 0x79, 0x00, 0x00, 0x00, 0x7a,
+	0x7b, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x00, 0x00,
+	/* Node for range 0x010__ */
+	0x00, 0x00, 0x00, 0x7d, 0x00, 0x00, 0x00, 0x00,
+	0x7e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x013__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x7f, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x017__ */
+	0x00, 0x80, 0x00, 0x81, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x82, 0x00, 0x00,
+	/* Node for range 0x018__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x83, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x019__ */
+	0x00, 0x00, 0x00, 0x84, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01a__ */
+	0x00, 0x85, 0x00, 0x00, 0x00, 0x00, 0x86, 0x87,
+	0x00, 0x00, 0x00, 0x88, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01b__ */
+	0x00, 0x00, 0x00, 0x89, 0x8a, 0x00, 0x8b, 0x8c,
+	0x00, 0x00, 0x8d, 0x00, 0x00, 0x00, 0x8e, 0x8f,
+	/* Node for range 0x01c__ */
+	0x00, 0x00, 0x00, 0x90, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x91, 0x92, 0x93,
+	/* Node for range 0x01d__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x94, 0x95, 0x96, 0x97,
+	/* Node for range 0x020__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x98, 0x99, 0x9a,
+	/* Node for range 0x02c__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x9b, 0x9c,
+	/* Node for range 0x02d__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x9d,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x9e, 0x9f,
+	/* Node for range 0x030__ */
+	0x00, 0x00, 0xa0, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0xa1, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a6__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa2, 0xa3,
+	0x00, 0xa4, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa5,
+	/* Node for range 0x0a8__ */
+	0xa6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0xa7, 0x00, 0xa8, 0xa9,
+	/* Node for range 0x0a9__ */
+	0x00, 0x00, 0xaa, 0x00, 0x00, 0xab, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xac, 0xad, 0x00, 0x00, 0x00,
+	/* Node for range 0x0aa__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xae, 0xaf, 0x00, 0x00, 0xb0,
+	/* Node for range 0x0ab__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xb1, 0x00,
+	/* Node for range 0x0fb__ */
+	0x00, 0xb2, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0fe__ */
+	0x00, 0x00, 0xb3, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x101__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xb4,
+	/* Node for range 0x102__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xb5, 0x00,
+	/* Node for range 0x103__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xb6,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x10a__ */
+	0xb7, 0x00, 0x00, 0xb8, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xb9, 0x00,
+	/* Node for range 0x110__ */
+	0x00, 0x00, 0x00, 0x00, 0xba, 0x00, 0x00, 0xbb,
+	0x00, 0x00, 0x00, 0xbc, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x111__ */
+	0xbd, 0x00, 0x00, 0xbe, 0x00, 0x00, 0x00, 0xbf,
+	0x00, 0x00, 0x00, 0x00, 0xc0, 0x00, 0x00, 0x00,
+	/* Node for range 0x112__ */
+	0x00, 0x00, 0x00, 0xc1, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc2, 0x00,
+	/* Node for range 0x113__ */
+	0x00, 0x00, 0x00, 0xc3, 0xc4, 0x00, 0xc5, 0xc6,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x114__ */
+	0x00, 0x00, 0x00, 0x00, 0xc7, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0xc8, 0x00, 0x00, 0x00,
+	/* Node for range 0x115__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xc9, 0xca, 0x00, 0x00, 0x00,
+	/* Node for range 0x116__ */
+	0x00, 0x00, 0x00, 0xcb, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xcc, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x117__ */
+	0x00, 0x00, 0xcd, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x11c__ */
+	0x00, 0x00, 0x00, 0xce, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x16a__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xcf,
+	/* Node for range 0x16b__ */
+	0x00, 0x00, 0x00, 0xd0, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1bc__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0xd1, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1d1__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xd2, 0xd3,
+	0xd4, 0x00, 0xd5, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1d2__ */
+	0x00, 0x00, 0x00, 0x00, 0xd6, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1e0__ */
+	0xd7, 0xd8, 0xd9, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1e8__ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xda, 0x00, 0x00,
+	/* Node for range 0x1e9__ */
+	0x00, 0x00, 0x00, 0x00, 0xdb, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0030_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0031_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe8, 0xdc, 0xdc,
+	0xdc, 0xdc, 0xe8, 0xd8, 0xdc, 0xdc, 0xdc, 0xdc,
+	/* Node for range 0x0032_ */
+	0xdc, 0xca, 0xca, 0xdc, 0xdc, 0xdc, 0xdc, 0xca,
+	0xca, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc,
+	/* Node for range 0x0033_ */
+	0xdc, 0xdc, 0xdc, 0xdc, 0x01, 0x01, 0x01, 0x01,
+	0x01, 0xdc, 0xdc, 0xdc, 0xdc, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0034_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xf0, 0xe6, 0xdc,
+	0xdc, 0xdc, 0xe6, 0xe6, 0xe6, 0xdc, 0xdc, 0x00,
+	/* Node for range 0x0035_ */
+	0xe6, 0xe6, 0xe6, 0xdc, 0xdc, 0xdc, 0xdc, 0xe6,
+	0xe8, 0xdc, 0xdc, 0xe6, 0xe9, 0xea, 0xea, 0xe9,
+	/* Node for range 0x0036_ */
+	0xea, 0xea, 0xe9, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0048_ */
+	0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0059_ */
+	0x00, 0xdc, 0xe6, 0xe6, 0xe6, 0xe6, 0xdc, 0xe6,
+	0xe6, 0xe6, 0xde, 0xdc, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x005a_ */
+	0xe6, 0xe6, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc,
+	0xe6, 0xe6, 0xdc, 0xe6, 0xe6, 0xde, 0xe4, 0xe6,
+	/* Node for range 0x005b_ */
+	0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11,
+	0x12, 0x13, 0x13, 0x14, 0x15, 0x16, 0x00, 0x17,
+	/* Node for range 0x005c_ */
+	0x00, 0x18, 0x19, 0x00, 0xe6, 0xdc, 0x00, 0x12,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0061_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0x1e, 0x1f, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0064_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f,
+	/* Node for range 0x0065_ */
+	0x20, 0x21, 0x22, 0xe6, 0xe6, 0xdc, 0xdc, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xdc, 0xe6, 0xe6, 0xdc,
+	/* Node for range 0x0067_ */
+	0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x006d_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00, 0xe6,
+	/* Node for range 0x006e_ */
+	0xe6, 0xe6, 0xe6, 0xdc, 0xe6, 0x00, 0x00, 0xe6,
+	0xe6, 0x00, 0xdc, 0xe6, 0xe6, 0xdc, 0x00, 0x00,
+	/* Node for range 0x0071_ */
+	0x00, 0x24, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0073_ */
+	0xe6, 0xdc, 0xe6, 0xe6, 0xdc, 0xe6, 0xe6, 0xdc,
+	0xdc, 0xdc, 0xe6, 0xdc, 0xdc, 0xe6, 0xdc, 0xe6,
+	/* Node for range 0x0074_ */
+	0xe6, 0xe6, 0xdc, 0xe6, 0xdc, 0xe6, 0xdc, 0xe6,
+	0xdc, 0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x007e_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x007f_ */
+	0xe6, 0xe6, 0xdc, 0xe6, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0081_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6,
+	0xe6, 0xe6, 0x00, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0082_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0xe6, 0xe6, 0xe6,
+	0x00, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00,
+	/* Node for range 0x0085_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0xdc, 0xdc, 0xdc, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x008d_ */
+	0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x008e_ */
+	0xe6, 0xe6, 0x00, 0xdc, 0xe6, 0xe6, 0xdc, 0xe6,
+	0xe6, 0xdc, 0xe6, 0xe6, 0xe6, 0xdc, 0xdc, 0xdc,
+	/* Node for range 0x008f_ */
+	0x1b, 0x1c, 0x1d, 0xe6, 0xe6, 0xe6, 0xdc, 0xe6,
+	0xe6, 0xdc, 0xdc, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0093_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x0094_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x0095_ */
+	0x00, 0xe6, 0xdc, 0xe6, 0xe6, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x009b_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x009c_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00a3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x00a4_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00ab_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x00ac_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00b3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x00b4_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00bc_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00c4_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00c5_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x54, 0x5b, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00cb_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x00cc_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00d4_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x00dc_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00e3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x67, 0x67, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00e4_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x6b, 0x6b, 0x6b, 0x6b, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00eb_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x76, 0x76, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00ec_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x7a, 0x7a, 0x7a, 0x7a, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00f1_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0xdc, 0xdc, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00f3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xdc, 0x00, 0xdc,
+	0x00, 0xd8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00f7_ */
+	0x00, 0x81, 0x82, 0x00, 0x84, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x82, 0x82, 0x82, 0x82, 0x00, 0x00,
+	/* Node for range 0x00f8_ */
+	0x82, 0x00, 0xe6, 0xe6, 0x09, 0x00, 0xe6, 0xe6,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x00fc_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xdc, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0103_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07,
+	0x00, 0x09, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0108_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xdc, 0x00, 0x00,
+	/* Node for range 0x0135_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0171_ */
+	0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0173_ */
+	0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x017d_ */
+	0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0x00, 0x00,
+	/* Node for range 0x018a_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0xe4, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0193_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0xde, 0xe6, 0xdc, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01a1_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6,
+	0xdc, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01a6_ */
+	0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01a7_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00, 0xdc,
+	/* Node for range 0x01ab_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xdc, 0xdc, 0xdc,
+	0xdc, 0xdc, 0xdc, 0xe6, 0xe6, 0xdc, 0x00, 0x00,
+	/* Node for range 0x01b3_ */
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01b4_ */
+	0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01b6_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xe6, 0xdc, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x01b7_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01ba_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x09, 0x09, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01be_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01bf_ */
+	0x00, 0x00, 0x09, 0x09, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01c3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01cd_ */
+	0xe6, 0xe6, 0xe6, 0x00, 0x01, 0xdc, 0xdc, 0xdc,
+	0xdc, 0xdc, 0xe6, 0xe6, 0xdc, 0xdc, 0xdc, 0xdc,
+	/* Node for range 0x01ce_ */
+	0xe6, 0x00, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
+	0x01, 0x00, 0x00, 0x00, 0x00, 0xdc, 0x00, 0x00,
+	/* Node for range 0x01cf_ */
+	0x00, 0x00, 0x00, 0x00, 0xe6, 0x00, 0x00, 0x00,
+	0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x01dc_ */
+	0xe6, 0xe6, 0xdc, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xdc, 0xe6, 0xe6, 0xea, 0xd6, 0xdc,
+	/* Node for range 0x01dd_ */
+	0xca, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x01de_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x01df_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xe6, 0xe9, 0xdc, 0xe6, 0xdc,
+	/* Node for range 0x020d_ */
+	0xe6, 0xe6, 0x01, 0x01, 0xe6, 0xe6, 0xe6, 0xe6,
+	0x01, 0x01, 0x01, 0xe6, 0xe6, 0x00, 0x00, 0x00,
+	/* Node for range 0x020e_ */
+	0x00, 0xe6, 0x00, 0x00, 0x00, 0x01, 0x01, 0xe6,
+	0xdc, 0xe6, 0x01, 0x01, 0xdc, 0xdc, 0xdc, 0xdc,
+	/* Node for range 0x020f_ */
+	0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x02ce_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6,
+	/* Node for range 0x02cf_ */
+	0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x02d7_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09,
+	/* Node for range 0x02de_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x02df_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0302_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0xda, 0xe4, 0xe8, 0xde, 0xe0, 0xe0,
+	/* Node for range 0x0309_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x08, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a66_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6,
+	/* Node for range 0x0a67_ */
+	0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00,
+	/* Node for range 0x0a69_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6,
+	/* Node for range 0x0a6f_ */
+	0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a80_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a8c_ */
+	0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a8e_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x0a8f_ */
+	0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a92_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xdc, 0xdc, 0xdc, 0x00, 0x00,
+	/* Node for range 0x0a95_ */
+	0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a9b_ */
+	0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0a9c_ */
+	0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0aab_ */
+	0xe6, 0x00, 0xe6, 0xe6, 0xdc, 0x00, 0x00, 0xe6,
+	0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6,
+	/* Node for range 0x0aac_ */
+	0x00, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0aaf_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x0abe_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x0fb1_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1a, 0x00,
+	/* Node for range 0x0fe2_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xdc,
+	0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xe6, 0xe6,
+	/* Node for range 0x101f_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xdc, 0x00, 0x00,
+	/* Node for range 0x102e_ */
+	0xdc, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1037_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x10a0_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xdc, 0x00, 0xe6,
+	/* Node for range 0x10a3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0xe6, 0x01, 0xdc, 0x00, 0x00, 0x00, 0x00, 0x09,
+	/* Node for range 0x10ae_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xdc, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1104_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1107_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09,
+	/* Node for range 0x110b_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x09, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1110_ */
+	0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1113_ */
+	0x00, 0x00, 0x00, 0x09, 0x09, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1117_ */
+	0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x111c_ */
+	0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1123_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x07, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x112e_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x07, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1133_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+	/* Node for range 0x1134_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00,
+	/* Node for range 0x1136_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00,
+	/* Node for range 0x1137_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1144_ */
+	0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x07, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x114c_ */
+	0x00, 0x00, 0x09, 0x07, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x115b_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09,
+	/* Node for range 0x115c_ */
+	0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1163_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09,
+	/* Node for range 0x116b_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x07,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1172_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x11c3_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09,
+	/* Node for range 0x16af_ */
+	0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x16b3_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1bc9_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00,
+	/* Node for range 0x1d16_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0xd8, 0xd8, 0x01,
+	0x01, 0x01, 0x00, 0x00, 0x00, 0xe2, 0xd8, 0xd8,
+	/* Node for range 0x1d17_ */
+	0xd8, 0xd8, 0xd8, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc,
+	/* Node for range 0x1d18_ */
+	0xdc, 0xdc, 0xdc, 0x00, 0x00, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xdc, 0xdc, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1d1a_ */
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6, 0x00, 0x00,
+	/* Node for range 0x1d24_ */
+	0x00, 0x00, 0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1e00_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0x00,
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x1e01_ */
+	0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6, 0xe6,
+	/* Node for range 0x1e02_ */
+	0xe6, 0xe6, 0x00, 0xe6, 0xe6, 0x00, 0xe6, 0xe6,
+	0xe6, 0xe6, 0xe6, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1e8d_ */
+	0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0xdc, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	/* Node for range 0x1e94_ */
+	0x00, 0x00, 0x00, 0x00, 0xe6, 0xe6, 0xe6, 0xe6,
+	0xe6, 0xe6, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00,
+};
diff --git a/fs/apfs/unicode.h b/fs/apfs/unicode.h
new file mode 100644
index 000000000..e3b7edc51
--- /dev/null
+++ b/fs/apfs/unicode.h
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#ifndef _APFS_UNICODE_H
+#define _APFS_UNICODE_H
+
+#include <linux/nls.h>
+
+/*
+ * This structure helps apfs_normalize_next() to retrieve one normalized
+ * (and case-folded) UTF-32 character at a time from a UTF-8 string.
+ */
+struct apfs_unicursor {
+	const char *utf8curr;	/* Start of UTF-8 to decompose and reorder */
+	unsigned int total_len;	/* Length of the whole UTF-8 string */
+	int length;		/* Length of normalization until next starter */
+	int last_pos;           /* Offset in substring of last char returned */
+	u8 last_ccc;		/* CCC of the last character returned */
+};
+
+extern void apfs_init_unicursor(struct apfs_unicursor *cursor, const char *utf8str, unsigned int total_len);
+extern unicode_t apfs_normalize_next(struct apfs_unicursor *cursor,
+				     bool case_fold);
+
+#endif	/* _APFS_UNICODE_H */
diff --git a/fs/apfs/version.h b/fs/apfs/version.h
new file mode 100644
index 000000000..e16b23c10
--- /dev/null
+++ b/fs/apfs/version.h
@@ -0,0 +1 @@
+#define GIT_COMMIT	""
diff --git a/fs/apfs/xattr.c b/fs/apfs/xattr.c
new file mode 100644
index 000000000..c21b8e469
--- /dev/null
+++ b/fs/apfs/xattr.c
@@ -0,0 +1,914 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2018 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/buffer_head.h>
+#include <linux/xattr.h>
+#include <linux/blk_types.h>
+#include "apfs.h"
+
+/**
+ * apfs_xattr_from_query - Read the xattr record found by a successful query
+ * @query:	the query that found the record
+ * @xattr:	Return parameter.  The xattr record found.
+ *
+ * Reads the xattr record into @xattr and performs some basic sanity checks
+ * as a protection against crafted filesystems.  Returns 0 on success or
+ * -EFSCORRUPTED otherwise.
+ *
+ * The caller must not free @query while @xattr is in use, because @xattr->name
+ * and @xattr->xdata point to data on disk.
+ */
+static int apfs_xattr_from_query(struct apfs_query *query,
+				 struct apfs_xattr *xattr)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_xattr_val *xattr_val;
+	struct apfs_xattr_key *xattr_key;
+	char *raw = query->node->object.data;
+	int datalen = query->len - sizeof(*xattr_val);
+	int namelen = query->key_len - sizeof(*xattr_key);
+
+	if (namelen < 1 || datalen < 0) {
+		apfs_err(sb, "bad length of name (%d) or data (%d)", namelen, datalen);
+		return -EFSCORRUPTED;
+	}
+
+	xattr_val = (struct apfs_xattr_val *)(raw + query->off);
+	xattr_key = (struct apfs_xattr_key *)(raw + query->key_off);
+
+	if (namelen != le16_to_cpu(xattr_key->name_len)) {
+		apfs_err(sb, "inconsistent name length (%d vs %d)", namelen, le16_to_cpu(xattr_key->name_len));
+		return -EFSCORRUPTED;
+	}
+
+	/* The xattr name must be NULL-terminated */
+	if (xattr_key->name[namelen - 1] != 0) {
+		apfs_err(sb, "null termination missing");
+		return -EFSCORRUPTED;
+	}
+
+	xattr->has_dstream = le16_to_cpu(xattr_val->flags) &
+			     APFS_XATTR_DATA_STREAM;
+
+	if (xattr->has_dstream && datalen != sizeof(struct apfs_xattr_dstream)) {
+		apfs_err(sb, "bad data length (%d)", datalen);
+		return -EFSCORRUPTED;
+	}
+	if (!xattr->has_dstream && datalen != le16_to_cpu(xattr_val->xdata_len)) {
+		apfs_err(sb, "inconsistent data length (%d vs %d)", datalen, le16_to_cpu(xattr_val->xdata_len));
+		return -EFSCORRUPTED;
+	}
+
+	xattr->name = xattr_key->name;
+	xattr->name_len = namelen - 1; /* Don't count the NULL termination */
+	xattr->xdata = xattr_val->xdata;
+	xattr->xdata_len = datalen;
+	return 0;
+}
+
+/**
+ * apfs_dstream_from_xattr - Get the dstream info for a dstream xattr
+ * @sb:		filesystem superblock
+ * @xattr:	in-memory xattr record (already sanity-checked)
+ * @dstream:	on return, the data stream info
+ */
+static void apfs_dstream_from_xattr(struct super_block *sb, struct apfs_xattr *xattr, struct apfs_dstream_info *dstream)
+{
+	struct apfs_xattr_dstream *xdata = (void *)xattr->xdata;
+
+	dstream->ds_sb = sb;
+	dstream->ds_inode = NULL;
+	dstream->ds_id = le64_to_cpu(xdata->xattr_obj_id);
+	dstream->ds_size = le64_to_cpu(xdata->dstream.size);
+	dstream->ds_sparse_bytes = 0; /* Irrelevant for xattrs */
+
+	dstream->ds_cached_ext.len = 0;
+	dstream->ds_ext_dirty = false;
+	spin_lock_init(&dstream->ds_ext_lock);
+
+	/* Xattrs can't be cloned */
+	dstream->ds_shared = false;
+}
+
+/**
+ * apfs_xattr_extents_read - Read the value of a xattr from its extents
+ * @parent:	inode the attribute belongs to
+ * @xattr:	the xattr structure
+ * @buffer:	where to copy the attribute value
+ * @size:	size of @buffer
+ * @only_whole:	are partial reads banned?
+ *
+ * Copies the value of @xattr to @buffer, if provided. If @buffer is NULL, just
+ * computes the size of the buffer required.
+ *
+ * Returns the number of bytes used/required, or a negative error code in case
+ * of failure.
+ */
+static int apfs_xattr_extents_read(struct inode *parent,
+				   struct apfs_xattr *xattr,
+				   void *buffer, size_t size, bool only_whole)
+{
+	struct super_block *sb = parent->i_sb;
+	struct apfs_dstream_info *dstream;
+	int length, ret;
+
+	dstream = kzalloc(sizeof(*dstream), GFP_KERNEL);
+	if (!dstream)
+		return -ENOMEM;
+	apfs_dstream_from_xattr(sb, xattr, dstream);
+
+	length = dstream->ds_size;
+	if (length < 0 || length < dstream->ds_size) {
+		/* TODO: avoid overflow here for huge compressed files */
+		apfs_warn(sb, "xattr is too big to read on linux (0x%llx)", dstream->ds_size);
+		ret = -E2BIG;
+		goto out;
+	}
+
+	if (!buffer) {
+		/* All we want is the length */
+		ret = length;
+		goto out;
+	}
+
+	if (only_whole) {
+		if (length > size) {
+			/* xattr won't fit in the buffer */
+			ret = -ERANGE;
+			goto out;
+		}
+	} else {
+		if (length > size)
+			length = size;
+	}
+
+	ret = apfs_nonsparse_dstream_read(dstream, buffer, length, 0 /* offset */);
+	if (ret == 0)
+		ret = length;
+
+out:
+	kfree(dstream);
+	return ret;
+}
+
+/**
+ * apfs_xattr_inline_read - Read the value of an inline xattr
+ * @xattr:	the xattr structure
+ * @buffer:	where to copy the attribute value
+ * @size:	size of @buffer
+ * @only_whole:	are partial reads banned?
+ *
+ * Copies the inline value of @xattr to @buffer, if provided. If @buffer is
+ * NULL, just computes the size of the buffer required.
+ *
+ * Returns the number of bytes used/required, or a negative error code in case
+ * of failure.
+ */
+static int apfs_xattr_inline_read(struct apfs_xattr *xattr, void *buffer, size_t size, bool only_whole)
+{
+	int length = xattr->xdata_len;
+
+	if (!buffer) /* All we want is the length */
+		return length;
+	if (only_whole) {
+		if (length > size) /* xattr won't fit in the buffer */
+			return -ERANGE;
+	} else {
+		if (length > size)
+			length = size;
+	}
+	memcpy(buffer, xattr->xdata, length);
+	return length;
+}
+
+/**
+ * apfs_xattr_get_compressed_data - Get the compressed data in a named attribute
+ * @inode:	inode the attribute belongs to
+ * @name:	name of the attribute
+ * @cdata:	compressed data struct to set on return
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_xattr_get_compressed_data(struct inode *inode, const char *name, struct apfs_compressed_data *cdata)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_xattr xattr;
+	u64 cnid = apfs_ino(inode);
+	int ret;
+
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_xattr_key(cnid, name, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		apfs_err(sb, "query failed for id 0x%llx (%s)", cnid, name);
+		goto done;
+	}
+
+	ret = apfs_xattr_from_query(query, &xattr);
+	if (ret) {
+		apfs_err(sb, "bad xattr record in inode 0x%llx", cnid);
+		goto done;
+	}
+
+	cdata->has_dstream = xattr.has_dstream;
+	if (cdata->has_dstream) {
+		struct apfs_dstream_info *dstream = NULL;
+
+		dstream = kzalloc(sizeof(*dstream), GFP_KERNEL);
+		if (!dstream) {
+			ret = -ENOMEM;
+			goto done;
+		}
+		apfs_dstream_from_xattr(sb, &xattr, dstream);
+
+		cdata->dstream = dstream;
+		cdata->size = dstream->ds_size;
+	} else {
+		void *data = NULL;
+		int len;
+
+		len = xattr.xdata_len;
+		if (len > APFS_XATTR_MAX_EMBEDDED_SIZE) {
+			apfs_err(sb, "inline xattr too big");
+			ret = -EFSCORRUPTED;
+			goto done;
+		}
+		data = kzalloc(len, GFP_KERNEL);
+		if (!data) {
+			ret = -ENOMEM;
+			goto done;
+		}
+		memcpy(data, xattr.xdata, len);
+
+		cdata->data = data;
+		cdata->size = len;
+	}
+	ret = 0;
+
+done:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_release_compressed_data - Clean up a compressed data struct
+ * @cdata: struct to clean up (but not free)
+ */
+void apfs_release_compressed_data(struct apfs_compressed_data *cdata)
+{
+	if (!cdata)
+		return;
+
+	if (cdata->has_dstream) {
+		kfree(cdata->dstream);
+		cdata->dstream = NULL;
+	} else {
+		kfree(cdata->data);
+		cdata->data = NULL;
+	}
+}
+
+/**
+ * apfs_compressed_data_read - Read from a compressed data struct
+ * @cdata:	compressed data struct
+ * @buf:	destination buffer
+ * @count:	exact number of bytes to read
+ * @offset:	dstream offset to read from
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_compressed_data_read(struct apfs_compressed_data *cdata, void *buf, size_t count, u64 offset)
+{
+	if (cdata->has_dstream)
+		return apfs_nonsparse_dstream_read(cdata->dstream, buf, count, offset);
+
+	if (offset > cdata->size || count > cdata->size - offset) {
+		apfs_err(NULL, "reading past the end (0x%llx-0x%llx)", offset, (unsigned long long)count);
+		/* No caller is expected to legitimately read out-of-bounds */
+		return -EFSCORRUPTED;
+	}
+	memcpy(buf, cdata->data + offset, count);
+	return 0;
+}
+
+/**
+ * __apfs_xattr_get - Find and read a named attribute
+ * @inode:	inode the attribute belongs to
+ * @name:	name of the attribute
+ * @buffer:	where to copy the attribute value
+ * @size:	size of @buffer
+ *
+ * This does the same as apfs_xattr_get(), but without taking any locks.
+ */
+int __apfs_xattr_get(struct inode *inode, const char *name, void *buffer,
+		     size_t size)
+{
+	return ____apfs_xattr_get(inode, name, buffer, size, true /* only_whole */);
+}
+
+/**
+ * ____apfs_xattr_get - Find and read a named attribute, optionally header only
+ * @inode:	inode the attribute belongs to
+ * @name:	name of the attribute
+ * @buffer:	where to copy the attribute value
+ * @size:	size of @buffer
+ * @only_whole:	must read complete (no partial header read allowed)
+ */
+int ____apfs_xattr_get(struct inode *inode, const char *name, void *buffer,
+		       size_t size, bool only_whole)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	struct apfs_xattr xattr;
+	u64 cnid = apfs_ino(inode);
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_xattr_key(cnid, name, &query->key);
+	query->flags |= APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		if (ret != -ENODATA)
+			apfs_err(sb, "query failed for id 0x%llx (%s)", cnid, name);
+		goto done;
+	}
+
+	ret = apfs_xattr_from_query(query, &xattr);
+	if (ret) {
+		apfs_err(sb, "bad xattr record in inode 0x%llx", cnid);
+		goto done;
+	}
+
+	if (xattr.has_dstream)
+		ret = apfs_xattr_extents_read(inode, &xattr, buffer, size, only_whole);
+	else
+		ret = apfs_xattr_inline_read(&xattr, buffer, size, only_whole);
+
+done:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_xattr_get - Find and read a named attribute
+ * @inode:	inode the attribute belongs to
+ * @name:	name of the attribute
+ * @buffer:	where to copy the attribute value
+ * @size:	size of @buffer
+ *
+ * Finds an extended attribute and copies its value to @buffer, if provided. If
+ * @buffer is NULL, just computes the size of the buffer required.
+ *
+ * Returns the number of bytes used/required, or a negative error code in case
+ * of failure.
+ */
+static int apfs_xattr_get(struct inode *inode, const char *name, void *buffer, size_t size)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(inode->i_sb);
+	int ret;
+
+	down_read(&nxi->nx_big_sem);
+	ret = __apfs_xattr_get(inode, name, buffer, size);
+	up_read(&nxi->nx_big_sem);
+	if (ret > XATTR_SIZE_MAX) {
+		apfs_warn(inode->i_sb, "xattr is too big to read on linux (%d)", ret);
+		return -E2BIG;
+	}
+	return ret;
+}
+
+static int apfs_xattr_osx_get(const struct xattr_handler *handler,
+				struct dentry *unused, struct inode *inode,
+				const char *name, void *buffer, size_t size)
+{
+	/* Ignore the fake 'osx' prefix */
+	return apfs_xattr_get(inode, name, buffer, size);
+}
+
+/**
+ * apfs_delete_xattr - Delete an extended attribute
+ * @query:	successful query pointing to the xattr to delete
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+static int apfs_delete_xattr(struct apfs_query *query)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_xattr xattr;
+	struct apfs_dstream_info *dstream;
+	int err;
+
+	err = apfs_xattr_from_query(query, &xattr);
+	if (err) {
+		apfs_err(sb, "bad xattr record");
+		return err;
+	}
+
+	if (!xattr.has_dstream)
+		return apfs_btree_remove(query);
+
+	dstream = kzalloc(sizeof(*dstream), GFP_KERNEL);
+	if (!dstream)
+		return -ENOMEM;
+	apfs_dstream_from_xattr(sb, &xattr, dstream);
+
+	/*
+	 * Remove the xattr record before truncation, because truncation creates
+	 * new queries and makes ours invalid. This stuff is all too subtle, I
+	 * really need to add some assertions (TODO).
+	 */
+	err = apfs_btree_remove(query);
+	if (err) {
+		apfs_err(sb, "removal failed");
+		goto fail;
+	}
+	err = apfs_truncate(dstream, 0);
+	if (err)
+		apfs_err(sb, "truncation failed for dstream 0x%llx", dstream->ds_id);
+
+fail:
+	kfree(dstream);
+	return err;
+}
+
+/**
+ * apfs_delete_any_xattr - Delete any single xattr for a given inode
+ * @inode: the vfs inode
+ *
+ * Intended to be called repeatedly, to delete all the xattrs one by one.
+ * Returns -EAGAIN on success until the process is complete, then it returns
+ * 0. Returns other negative error codes in case of failure.
+ */
+static int apfs_delete_any_xattr(struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query;
+	int ret;
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query)
+		return -ENOMEM;
+	apfs_init_xattr_key(apfs_ino(inode), NULL /* name */, &query->key);
+	query->flags = APFS_QUERY_CAT | APFS_QUERY_ANY_NAME | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		if (ret == -ENODATA)
+			ret = 0; /* No more xattrs, we are done */
+		else
+			apfs_err(sb, "query failed for ino 0x%llx", apfs_ino(inode));
+		goto out;
+	}
+
+	ret = apfs_delete_xattr(query);
+	if (!ret)
+		ret = -EAGAIN;
+	else
+		apfs_err(sb, "xattr deletion failed");
+
+out:
+	apfs_free_query(query);
+	return ret;
+}
+
+/**
+ * apfs_delete_all_xattrs - Delete all xattrs for a given inode
+ * @inode: the vfs inode
+ *
+ * Returns 0 on success or a negative error code in case of failure.
+ */
+int apfs_delete_all_xattrs(struct inode *inode)
+{
+	struct apfs_nxsb_info *nxi = APFS_NXI(inode->i_sb);
+	int ret;
+
+	lockdep_assert_held_write(&nxi->nx_big_sem);
+
+	do {
+		ret = apfs_delete_any_xattr(inode);
+	} while (ret == -EAGAIN);
+
+	return ret;
+}
+
+/**
+ * apfs_build_xattr_key - Allocate and initialize the key for a xattr record
+ * @name:	xattr name
+ * @ino:	inode number for xattr's owner
+ * @key_p:	on return, a pointer to the new on-disk key structure
+ *
+ * Returns the length of the key, or a negative error code in case of failure.
+ */
+static int apfs_build_xattr_key(const char *name, u64 ino, struct apfs_xattr_key **key_p)
+{
+	struct apfs_xattr_key *key;
+	u16 namelen = strlen(name) + 1; /* We count the null-termination */
+	int key_len;
+
+	key_len = sizeof(*key) + namelen;
+	key = kmalloc(key_len, GFP_KERNEL);
+	if (!key)
+		return -ENOMEM;
+
+	apfs_key_set_hdr(APFS_TYPE_XATTR, ino, key);
+	key->name_len = cpu_to_le16(namelen);
+	strscpy(key->name, name, namelen);
+
+	*key_p = key;
+	return key_len;
+}
+
+/**
+ * apfs_build_dstream_xattr_val - Allocate and init value for a dstream xattr
+ * @dstream:	data stream info
+ * @val_p:	on return, a pointer to the new on-disk value structure
+ *
+ * Returns the length of the value, or a negative error code in case of failure.
+ */
+static int apfs_build_dstream_xattr_val(struct apfs_dstream_info *dstream, struct apfs_xattr_val **val_p)
+{
+	struct apfs_xattr_val *val;
+	struct apfs_xattr_dstream *dstream_raw;
+	int val_len;
+
+	val_len = sizeof(*val) + sizeof(*dstream_raw);
+	val = kzalloc(val_len, GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	val->flags = cpu_to_le16(APFS_XATTR_DATA_STREAM);
+	val->xdata_len = cpu_to_le16(sizeof(*dstream_raw));
+
+	dstream_raw = (void *)val->xdata;
+	dstream_raw->xattr_obj_id = cpu_to_le64(dstream->ds_id);
+	dstream_raw->dstream.size = cpu_to_le64(dstream->ds_size);
+	dstream_raw->dstream.alloced_size = cpu_to_le64(apfs_alloced_size(dstream));
+	if (apfs_vol_is_encrypted(dstream->ds_sb))
+		dstream_raw->dstream.default_crypto_id = cpu_to_le64(dstream->ds_id);
+
+	*val_p = val;
+	return val_len;
+}
+
+/**
+ * apfs_build_inline_xattr_val - Allocate and init value for an inline xattr
+ * @value:	content of the xattr
+ * @size:	size of @value
+ * @val_p:	on return, a pointer to the new on-disk value structure
+ *
+ * Returns the length of the value, or a negative error code in case of failure.
+ */
+static int apfs_build_inline_xattr_val(const void *value, size_t size, struct apfs_xattr_val **val_p)
+{
+	struct apfs_xattr_val *val;
+	int val_len;
+
+	val_len = sizeof(*val) + size;
+	val = kmalloc(val_len, GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	val->flags = cpu_to_le16(APFS_XATTR_DATA_EMBEDDED);
+	val->xdata_len = cpu_to_le16(size);
+	memcpy(val->xdata, value, size);
+
+	*val_p = val;
+	return val_len;
+}
+
+/**
+ * apfs_create_xattr_dstream - Create the extents for a dstream xattr
+ * @sb:		filesystem superblock
+ * @value:	value for the attribute
+ * @size:	sizeo of @value
+ *
+ * Returns the info for the created data stream, or an error pointer in case
+ * of failure.
+ */
+static struct apfs_dstream_info *apfs_create_xattr_dstream(struct super_block *sb, const void *value, size_t size)
+{
+	struct apfs_superblock *vsb_raw = APFS_SB(sb)->s_vsb_raw;
+	struct apfs_dstream_info *dstream;
+	int blkcnt, i;
+	int err;
+
+	dstream = kzalloc(sizeof(*dstream), GFP_KERNEL);
+	if (!dstream)
+		return ERR_PTR(-ENOMEM);
+	dstream->ds_sb = sb;
+	spin_lock_init(&dstream->ds_ext_lock);
+
+	apfs_assert_in_transaction(sb, &vsb_raw->apfs_o);
+	dstream->ds_id = le64_to_cpu(vsb_raw->apfs_next_obj_id);
+	le64_add_cpu(&vsb_raw->apfs_next_obj_id, 1);
+
+	blkcnt = (size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+	for (i = 0; i < blkcnt; i++) {
+		struct buffer_head *bh;
+		int off, tocopy;
+		u64 bno;
+
+		err = apfs_dstream_get_new_bno(dstream, i, &bno);
+		if (err) {
+			apfs_err(sb, "failed to get new block in dstream 0x%llx", dstream->ds_id);
+			goto fail;
+		}
+		bh = apfs_sb_bread(sb, bno);
+		if (!bh) {
+			apfs_err(sb, "failed to read new block");
+			err = -EIO;
+			goto fail;
+		}
+
+		err = apfs_transaction_join(sb, bh);
+		if (err) {
+			brelse(bh);
+			goto fail;
+		}
+
+		off = i << sb->s_blocksize_bits;
+		tocopy = min(sb->s_blocksize, (unsigned long)(size - off));
+		memcpy(bh->b_data, value + off, tocopy);
+		if (tocopy < sb->s_blocksize)
+			memset(bh->b_data + tocopy, 0, sb->s_blocksize - tocopy);
+		brelse(bh);
+
+		dstream->ds_size += tocopy;
+	}
+
+	err = apfs_flush_extent_cache(dstream);
+	if (err) {
+		apfs_err(sb, "extent cache flush failed for dstream 0x%llx", dstream->ds_id);
+		goto fail;
+	}
+	return dstream;
+
+fail:
+	kfree(dstream);
+	return ERR_PTR(err);
+}
+
+/**
+ * apfs_xattr_dstream_from_query - Extract the dstream from a xattr record
+ * @query:	the query that found the record
+ * @dstream_p:	on return, the newly allocated dstream info (or NULL if none)
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+static int apfs_xattr_dstream_from_query(struct apfs_query *query, struct apfs_dstream_info **dstream_p)
+{
+	struct super_block *sb = query->node->object.sb;
+	struct apfs_dstream_info *dstream = NULL;
+	struct apfs_xattr xattr;
+	int err;
+
+	err = apfs_xattr_from_query(query, &xattr);
+	if (err) {
+		apfs_err(sb, "bad xattr record");
+		return err;
+	}
+
+	if (xattr.has_dstream) {
+		dstream = kzalloc(sizeof(*dstream), GFP_KERNEL);
+		if (!dstream)
+			return -ENOMEM;
+		apfs_dstream_from_xattr(sb, &xattr, dstream);
+	}
+	*dstream_p = dstream;
+	return 0;
+}
+
+/**
+ * apfs_xattr_set - Write a named attribute
+ * @inode:	inode the attribute will belong to
+ * @name:	name for the attribute
+ * @value:	value for the attribute
+ * @size:	size of @value
+ * @flags:	XATTR_REPLACE and XATTR_CREATE
+ *
+ * Returns 0 on success, or a negative error code in case of failure.
+ */
+int apfs_xattr_set(struct inode *inode, const char *name, const void *value,
+		   size_t size, int flags)
+{
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_query *query = NULL;
+	u64 cnid = apfs_ino(inode);
+	int key_len, val_len;
+	struct apfs_xattr_key *raw_key = NULL;
+	struct apfs_xattr_val *raw_val = NULL;
+	struct apfs_dstream_info *dstream = NULL;
+	struct apfs_dstream_info *old_dstream = NULL;
+	int ret;
+
+	if (size > APFS_XATTR_MAX_EMBEDDED_SIZE) {
+		dstream = apfs_create_xattr_dstream(sb, value, size);
+		if (IS_ERR(dstream)) {
+			apfs_err(sb, "failed to set xattr dstream for ino 0x%llx", apfs_ino(inode));
+			return PTR_ERR(dstream);
+		}
+	}
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto done;
+	}
+	apfs_init_xattr_key(cnid, name, &query->key);
+	query->flags = APFS_QUERY_CAT | APFS_QUERY_EXACT;
+
+	ret = apfs_btree_query(sb, &query);
+	if (ret) {
+		if (ret != -ENODATA) {
+			apfs_err(sb, "query failed for id 0x%llx (%s)", cnid, name);
+			goto done;
+		} else if (flags & XATTR_REPLACE) {
+			goto done;
+		}
+	} else if (flags & XATTR_CREATE) {
+		ret = -EEXIST;
+		goto done;
+	} else if (!value) {
+		ret = apfs_delete_xattr(query);
+		if (ret)
+			apfs_err(sb, "xattr deletion failed");
+		goto done;
+	} else {
+		/* Remember the old dstream to clean it up later */
+		ret = apfs_xattr_dstream_from_query(query, &old_dstream);
+		if (ret) {
+			apfs_err(sb, "failed to get the old dstream");
+			goto done;
+		}
+	}
+
+	key_len = apfs_build_xattr_key(name, cnid, &raw_key);
+	if (key_len < 0) {
+		ret = key_len;
+		goto done;
+	}
+
+	if (dstream)
+		val_len = apfs_build_dstream_xattr_val(dstream, &raw_val);
+	else
+		val_len = apfs_build_inline_xattr_val(value, size, &raw_val);
+	if (val_len < 0) {
+		ret = val_len;
+		goto done;
+	}
+
+	/* For now this is the only system xattr we support */
+	if (strcmp(name, APFS_XATTR_NAME_SYMLINK) == 0)
+		raw_val->flags |= cpu_to_le16(APFS_XATTR_FILE_SYSTEM_OWNED);
+
+	if (ret)
+		ret = apfs_btree_insert(query, raw_key, key_len, raw_val, val_len);
+	else
+		ret = apfs_btree_replace(query, raw_key, key_len, raw_val, val_len);
+	if (ret) {
+		apfs_err(sb, "insertion/update failed for id 0x%llx (%s)", cnid, name);
+		goto done;
+	}
+
+	if (old_dstream) {
+		ret = apfs_truncate(old_dstream, 0);
+		if (ret)
+			apfs_err(sb, "truncation failed for dstream 0x%llx", old_dstream->ds_id);
+	}
+
+done:
+	kfree(dstream);
+	kfree(old_dstream);
+	kfree(raw_val);
+	kfree(raw_key);
+	apfs_free_query(query);
+	return ret;
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 12, 0)
+static int apfs_xattr_osx_set(const struct xattr_handler *handler,
+	      struct dentry *unused, struct inode *inode, const char *name,
+	      const void *value, size_t size, int flags)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(6, 3, 0) && !RHEL_VERSION_GE(9, 6)
+static int apfs_xattr_osx_set(const struct xattr_handler *handler,
+		  struct user_namespace *mnt_userns, struct dentry *unused,
+		  struct inode *inode, const char *name, const void *value,
+		  size_t size, int flags)
+#else
+static int apfs_xattr_osx_set(const struct xattr_handler *handler,
+		  struct mnt_idmap *idmap, struct dentry *unused,
+		  struct inode *inode, const char *name, const void *value,
+		  size_t size, int flags)
+#endif
+{
+	struct super_block *sb = inode->i_sb;
+	int err;
+
+	err = apfs_transaction_start(sb, value ? APFS_TRANS_REG : APFS_TRANS_DEL);
+	if (err)
+		return err;
+
+	/* Ignore the fake 'osx' prefix */
+	err = apfs_xattr_set(inode, name, value, size, flags);
+	if (err)
+		goto fail;
+
+	err = apfs_transaction_commit(sb);
+	if (!err)
+		return 0;
+
+fail:
+	apfs_transaction_abort(sb);
+	return err;
+}
+
+static const struct xattr_handler apfs_xattr_osx_handler = {
+	.prefix	= XATTR_MAC_OSX_PREFIX,
+	.get	= apfs_xattr_osx_get,
+	.set	= apfs_xattr_osx_set,
+};
+
+/* On-disk xattrs have no namespace; use a fake 'osx' prefix in the kernel */
+const struct xattr_handler *apfs_xattr_handlers[] = {
+	&apfs_xattr_osx_handler,
+	NULL
+};
+
+ssize_t apfs_listxattr(struct dentry *dentry, char *buffer, size_t size)
+{
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = inode->i_sb;
+	struct apfs_sb_info *sbi = APFS_SB(sb);
+	struct apfs_nxsb_info *nxi = APFS_NXI(sb);
+	struct apfs_query *query;
+	u64 cnid = apfs_ino(inode);
+	size_t free = size;
+	ssize_t ret;
+
+	down_read(&nxi->nx_big_sem);
+
+	query = apfs_alloc_query(sbi->s_cat_root, NULL /* parent */);
+	if (!query) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	/* We want all the xattrs for the cnid, regardless of the name */
+	apfs_init_xattr_key(cnid, NULL /* name */, &query->key);
+	query->flags = APFS_QUERY_CAT | APFS_QUERY_MULTIPLE | APFS_QUERY_EXACT;
+
+	while (1) {
+		struct apfs_xattr xattr;
+
+		ret = apfs_btree_query(sb, &query);
+		if (ret == -ENODATA) { /* Got all the xattrs */
+			ret = size - free;
+			break;
+		}
+		if (ret) {
+			apfs_err(sb, "query failed for id 0x%llx", cnid);
+			break;
+		}
+
+		ret = apfs_xattr_from_query(query, &xattr);
+		if (ret) {
+			apfs_err(sb, "bad xattr key in inode %llx", cnid);
+			break;
+		}
+
+		if (buffer) {
+			/* Prepend the fake 'osx' prefix before listing */
+			if (xattr.name_len + XATTR_MAC_OSX_PREFIX_LEN + 1 >
+									free) {
+				ret = -ERANGE;
+				break;
+			}
+			memcpy(buffer, XATTR_MAC_OSX_PREFIX,
+			       XATTR_MAC_OSX_PREFIX_LEN);
+			buffer += XATTR_MAC_OSX_PREFIX_LEN;
+			memcpy(buffer, xattr.name, xattr.name_len + 1);
+			buffer += xattr.name_len + 1;
+		}
+		free -= xattr.name_len + XATTR_MAC_OSX_PREFIX_LEN + 1;
+	}
+
+fail:
+	apfs_free_query(query);
+	up_read(&nxi->nx_big_sem);
+	return ret;
+}
diff --git a/fs/apfs/xfield.c b/fs/apfs/xfield.c
new file mode 100644
index 000000000..b8cbe17fd
--- /dev/null
+++ b/fs/apfs/xfield.c
@@ -0,0 +1,171 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
+ */
+
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include "apfs.h"
+
+/**
+ * apfs_find_xfield - Find an extended field value in an inode or dentry record
+ * @xfields:	pointer to the on-disk xfield collection for the record
+ * @len:	length of the collection
+ * @xtype:	type of the xfield to retrieve
+ * @xval:	on return, a pointer to the wanted on-disk xfield value
+ *
+ * Returns the length of @xval on success, or 0 if no matching xfield was found;
+ * the caller must check that the expected structures fit before casting @xval.
+ */
+int apfs_find_xfield(u8 *xfields, int len, u8 xtype, char **xval)
+{
+	struct apfs_xf_blob *blob;
+	struct apfs_x_field *xfield;
+	int count;
+	int rest = len;
+	int i;
+
+	if (!len)
+		return 0; /* No xfield data */
+
+	rest -= sizeof(*blob);
+	if (rest < 0)
+		return 0; /* Corruption */
+	blob = (struct apfs_xf_blob *)xfields;
+
+	count = le16_to_cpu(blob->xf_num_exts);
+	rest -= count * sizeof(*xfield);
+	if (rest < 0)
+		return 0; /* Corruption */
+	xfield = (struct apfs_x_field *)blob->xf_data;
+
+	for (i = 0; i < count; ++i) {
+		int xlen;
+
+		/* Attribute length is padded to a multiple of 8 */
+		xlen = round_up(le16_to_cpu(xfield[i].x_size), 8);
+		if (xlen > rest)
+			return 0; /* Corruption */
+
+		if (xfield[i].x_type == xtype) {
+			*xval = (char *)xfields + len - rest;
+			return xlen;
+		}
+		rest -= xlen;
+	}
+	return 0;
+}
+
+/**
+ * apfs_init_xfields - Set an empty collection of xfields in a buffer
+ * @buffer:	buffer to hold the xfields
+ * @buflen:	length of the buffer; should be enough to fit an xfield blob
+ *
+ * Returns 0 on success, or -1 if the buffer isn't long enough.
+ */
+int apfs_init_xfields(u8 *buffer, int buflen)
+{
+	struct apfs_xf_blob *blob;
+
+	if (buflen < sizeof(*blob))
+		return -1;
+	blob = (struct apfs_xf_blob *)buffer;
+
+	blob->xf_num_exts = 0;
+	blob->xf_used_data = 0;
+	return 0;
+}
+
+/**
+ * apfs_insert_xfield - Add a new xfield to an in-memory collection
+ * @buffer:	buffer holding the collection of xfields
+ * @buflen:	length of the buffer; should be enough to fit the new xfield
+ * @xkey:	metadata for the new xfield
+ * @xval:	value for the new xfield
+ *
+ * Returns the new length of the collection, or 0 if it the allocation would
+ * overflow @buffer.
+ */
+int apfs_insert_xfield(u8 *buffer, int buflen, const struct apfs_x_field *xkey,
+		       const void *xval)
+{
+	struct apfs_xf_blob *blob;
+	struct apfs_x_field *curr_xkey;
+	void *curr_xval;
+	int count;
+	int rest = buflen;
+	u16 used_data;
+	int xlen, padded_xlen;
+	int meta_len, total_len;
+	int i;
+
+	xlen = le16_to_cpu(xkey->x_size);
+	padded_xlen = round_up(xlen, 8);
+
+	if (!buflen)
+		return 0;
+
+	rest -= sizeof(*blob);
+	if (rest < 0)
+		return 0;
+	blob = (struct apfs_xf_blob *)buffer;
+	used_data = le16_to_cpu(blob->xf_used_data);
+
+	count = le16_to_cpu(blob->xf_num_exts);
+	rest -= count * sizeof(*curr_xkey);
+	if (rest < 0)
+		return 0;
+	meta_len = buflen - rest;
+	curr_xkey = (struct apfs_x_field *)blob->xf_data;
+
+	for (i = 0; i < count; ++i, ++curr_xkey) {
+		int curr_xlen;
+
+		/* Attribute length is padded to a multiple of 8 */
+		curr_xlen = round_up(le16_to_cpu(curr_xkey->x_size), 8);
+		if (curr_xlen > rest)
+			return 0;
+		if (curr_xkey->x_type != xkey->x_type) {
+			rest -= curr_xlen;
+			continue;
+		}
+
+		/* The xfield already exists, so just resize it and set it */
+		memcpy(curr_xkey, xkey, sizeof(*curr_xkey));
+		if (padded_xlen > rest)
+			return 0;
+		curr_xval = buffer + buflen - rest;
+		rest -= max(padded_xlen, curr_xlen);
+		memmove(curr_xval + padded_xlen, curr_xval + curr_xlen, rest);
+		memcpy(curr_xval, xval, xlen);
+		memset(curr_xval + xlen, 0, padded_xlen - xlen);
+		used_data += padded_xlen - curr_xlen;
+
+		goto done;
+	}
+
+	/* Create a metadata entry for the new xfield */
+	rest -= sizeof(*curr_xkey);
+	if (rest < 0)
+		return 0;
+	meta_len += sizeof(*curr_xkey);
+	memmove(curr_xkey + 1, curr_xkey, buflen - meta_len);
+	memcpy(curr_xkey, xkey, sizeof(*curr_xkey));
+	++count;
+
+	/* Now set the xfield value */
+	if (padded_xlen > rest)
+		return 0;
+	curr_xval = buffer + buflen - rest;
+	memcpy(curr_xval, xval, xlen);
+	memset(curr_xval + xlen, 0, padded_xlen - xlen);
+	used_data += padded_xlen;
+
+done:
+	total_len = used_data + meta_len;
+	if (total_len > buflen)
+		return 0;
+	blob->xf_num_exts = cpu_to_le16(count);
+	blob->xf_used_data = cpu_to_le16(used_data);
+	return total_len;
+}
-- 
2.53.0