devenv: Improve nested container support (#99)

The default containers.conf includes net.ipv4.ping_group_range=0 0 which
fails in nested containers because /proc/sys is read-only. This causes
'podman build' to fail with:

  open `/proc/sys/net/ipv4/ping_group_range`: Read-only file system

This matches the approach used by the official quay.io/podman/stable
image (containers/image_build) which sets default_sysctls = [] in its
podman-containers.conf.

Assisted-by: OpenCode (claude-opus-4-5@20251101)

Signed-off-by: Colin Walters <walters@verbum.org>

Author: cgwalters

devenv/README.md

@@ -15,6 +15,37 @@ base, but this helps prove out the general case of "src != target"
 that is a philosophy of bootc (and containers in general)
 as well as just helping prepare/motivate for bootc-on-Debian.
 
+## Nested container support
+
+This image supports running `podman` and `podman build` inside the container
+(podman-in-podman). The `userns-setup` script configures the environment at
+container startup.
+
+### Reference: quay.io/podman/stable
+
+Our nested container configuration is based on the official
+[quay.io/podman/stable](https://github.com/containers/image_build/tree/main/podman)
+image. Key differences:
+
+| Feature | quay.io/podman/stable | bootc-devenv |
+|---------|----------------------|--------------|
+| **default_sysctls** | `[]` | `[]` |
+| **cgroups** | `"disabled"` | `"disabled"` (constrained) / `"no-conmon"` (full) |
+| **cgroup_manager** | `"cgroupfs"` | `"cgroupfs"` |
+| **netns/userns/ipcns/utsns/cgroupns** | `"host"` for all | `utsns = "host"` (constrained only) |
+| **BUILDAH_ISOLATION** | `chroot` (env var) | Not set (uses OCI default) |
+| **subuid/subgid** | Hardcoded for `podman` user | Dynamically calculated based on available UID range |
+| **storage** | Modified storage.conf for fuse-overlayfs | VOLUME mounts avoid overlay-on-overlay |
+
+### Constrained vs full UID namespace
+
+The `userns-setup` script detects whether we're running in a constrained UID
+namespace (typical for rootless podman, GitHub Codespaces, etc.) and adjusts:
+
+- **Full namespace** (>100k UIDs): Uses default subuid/subgid, `cgroups = "no-conmon"`
+- **Constrained namespace** (<100k UIDs): Dynamically calculates subuid/subgid
+  ranges, uses `cgroups = "disabled"` and `utsns = "host"`
+
 ## Building locally
 
 See the `Justfile`, but it's just a thin wrapper around a default

devenv/userns-setup

@@ -7,6 +7,9 @@ This handles:
 - /dev/kvm permissions
 - subuid/subgid configuration for constrained UID namespaces
 - containers.conf configuration for nested operation
+
+Reference: quay.io/podman/stable image configuration
+  https://github.com/containers/image_build/tree/main/podman
 """
 
 import argparse
@@ -169,28 +172,31 @@ def configure_containers_conf() -> None:
 
     is_constrained, _ = detect_constrained_namespace()
 
+    conf_dir = Path("/etc/containers")
+    conf_dir.mkdir(parents=True, exist_ok=True)
+    conf_path = conf_dir / "containers.conf"
+
     if not is_constrained:
-        # Full namespace - just update the shipped config
-        conf_path = Path("/usr/share/containers/containers.conf")
-        if conf_path.exists():
-            content = conf_path.read_text()
-            content = content.replace("#cgroups =", 'cgroups = "no-conmon"  #')
-            content = content.replace("#cgroup_manager =", 'cgroup_manager = "cgroupfs"  #')
-            conf_path.write_text(content)
+        header = "# Generated for nested container support"
+        container_settings = 'cgroups = "no-conmon"'
     else:
-        # Constrained namespace - create full config for nested operation
-        conf_dir = Path("/etc/containers")
-        conf_dir.mkdir(parents=True, exist_ok=True)
-        conf_path = conf_dir / "containers.conf"
-        conf_path.write_text("""\
-# Generated for nested container support in constrained UID namespace
+        header = "# Generated for nested container support in constrained UID namespace"
+        container_settings = 'cgroups = "disabled"\nutsns = "host"'
+
+    conf_path.write_text(f"""\
+{header}
+# Reference: https://github.com/containers/image_build/tree/main/podman
 [containers]
-cgroups = "disabled"
-utsns = "host"
+# Disable default sysctls - /proc/sys is read-only in nested containers
+# (specifically net.ipv4.ping_group_range causes "Read-only file system" errors)
+# See: https://github.com/containers/common/blob/main/pkg/config/containers.conf
+default_sysctls = []
+{container_settings}
 
 [engine]
 cgroup_manager = "cgroupfs"
 """)
+    if is_constrained:
         print("Configured containers.conf for constrained UID namespace")