Add opt-in seccomp BPF allowlist behind --seccomp

A memory-corruption RCE in device emulation could otherwise pivot to
arbitrary host syscalls (open, execve, socket, unlink). The new filter
reduces blast radius to the syscalls every device worker, the vcpu
loop, and pthread internals actually need.

src/seccomp.c builds an arch-aware BPF allowlist for x86_64 and
aarch64. x86_64 explicitly rejects the x32 ABI: x32 shares
AUDIT_ARCH_X86_64 but tags syscall numbers with __X32_SYSCALL_BIT
(0x40000000), so a naive allowlist that copies an Internet example
without this guard would let an attacker pivot to x32 syscall numbers
that alias different kernel handlers. Default action is
SECCOMP_RET_KILL_PROCESS so a worker thread that takes a denied
syscall aborts the whole VMM rather than leaving the device in an
unrecoverable state.

Install uses seccomp(2) directly with SECCOMP_FILTER_FLAG_TSYNC. The
serial worker thread is already running by the time seccomp_apply()
is called (spawned in vm_arch_init_platform_device during vm_init);
plain prctl(PR_SET_SECCOMP) installs only on the calling thread,
leaving an attacker a path through the pre-existing worker. TSYNC's
return is three-way: 0 success, -1 errno error, positive TID for
partial-sync failure -- a naive < 0 check would silently treat
partial-sync as success and leave the process unfiltered, so any
non-zero return is reported as failure and surfaces the offending
TID.

The flag is opt-in via --seccomp so existing test and development
workflows are unaffected. CI gains a second "boot test (seccomp)"
step on host-x64 that reuses .ci/autorun.sh with
KVM_HOST_FLAGS=--seccomp; reaching the "Linux version " banner
exercises prctl(PR_SET_NO_NEW_PRIVS), seccomp(2)+TSYNC over the
already-running serial worker, and the early KVM_RUN dispatch under
the filter, so a regression that drops a steady-state syscall from
the allowlist surfaces here as a SIGSYS before the banner.

Boot-tested on x86_64 and aarch64: --seccomp boots Linux to the
busybox console prompt with virtio-blk mounting ext4 r/w (exercises
pread/pwrite/fdatasync) and virtio-net probed; the lazy
virtio-blk/virtio-net worker spawn paths inside vm_run rely on
clone/clone3 + set_robust_list + rseq + sigaltstack being
allowlisted.

Author: jserv

.github/workflows/main.yml

@@ -185,6 +185,19 @@ jobs:
       - name: boot test
         run: .ci/autorun.sh
         timeout-minutes: 5
+      # Smoke-test the --seccomp opt-in path on the same nested-KVM runner.
+      # The smoke test only waits for the "Linux version " banner before
+      # sending Ctrl-A x, but reaching the banner already exercises
+      # prctl(PR_SET_NO_NEW_PRIVS), seccomp(2)+TSYNC install over the
+      # already-running serial worker, and the early KVM_RUN dispatch
+      # under the filter. A regression that drops a steady-state syscall
+      # from src/seccomp.c's allowlist surfaces here as a SIGSYS before
+      # the banner appears.
+      - name: boot test (seccomp)
+        run: .ci/autorun.sh
+        env:
+          KVM_HOST_FLAGS: --seccomp
+        timeout-minutes: 5
 
   # arm64 host build: configs/linux-arm64.config has no prebuilt path and
   # the GitHub-hosted ubuntu-24.04-arm runners (Cobalt 100) do NOT expose

Makefile

@@ -32,6 +32,7 @@ OBJS := \
 	virtio-blk.o \
 	virtio-net.o \
 	diskimg.o \
+	seccomp.o \
 	main.o
 
 ifeq ($(ARCH), x86_64)
@@ -66,9 +67,15 @@ $(OUT)/ext4.img:
 	$(Q)dd if=/dev/zero of=$@ bs=4k count=600
 	$(Q)mkfs.ext4 -F $@
 
+# KVM_HOST_FLAGS forwards extra flags to the binary so CI and developers
+# can opt into --seccomp without duplicating the recipe. Empty by default
+# to keep `make check` matching the documented invocation.
+KVM_HOST_FLAGS ?=
+
 check: $(BIN) $(LINUX_IMG) $(ROOTFS_IMG) $(OUT)/ext4.img
 	$(VECHO) "\nOnce the message 'Kernel panic' appears, press Ctrl-C to exit\n\n"
-	$(Q)sudo $(BIN) -k $(LINUX_IMG) -i $(ROOTFS_IMG) -d $(OUT)/ext4.img
+	$(Q)sudo $(BIN) -k $(LINUX_IMG) -i $(ROOTFS_IMG) -d $(OUT)/ext4.img \
+	    $(KVM_HOST_FLAGS)
 
 clean:
 	$(VECHO) "Cleaning...\n"

README.md

@@ -43,14 +43,31 @@ make check
 ### Start Emulator
 
 ```
-$ build/kvm-host -k bzImage [-i initrd] [-d disk-image]
+$ build/kvm-host -k bzImage [-i initrd] [-d disk-image] [--seccomp]
 ```
 
 `bzImage` is the path to linux kernel bzImage. The bzImage file is in a specific format,
 containing concatenated `bootsect.o + setup.o + misc.o + piggy.o`. `initrd` is the path to
 initial RAM disk image, which is an optional argument.
 `disk-image` is the path to disk image which can be mounted as a block device via virtio. For the reference Linux guest, ext4 filesystem is used for disk image.
 
+`--seccomp` is an opt-in defense-in-depth flag that installs a seccomp BPF
+allowlist over the steady-state KVM_RUN loop. Once active, only the
+syscalls that the vcpu, virtio-blk, virtio-net, and serial workers need
+are permitted; anything else (including a memory-corruption RCE in
+device emulation pivoting to `execve`, `open`, or `socket`) terminates
+the process with `SIGSYS`. The filter is applied via `seccomp(2)` with
+`SECCOMP_FILTER_FLAG_TSYNC` so already-running worker threads inherit
+it. The flag is off by default so existing test and development
+workflows are unaffected. CI exercises both paths
+(`.github/workflows/main.yml`).
+
+To run `make check` with the filter enabled:
+
+```shell
+$ make KVM_HOST_FLAGS=--seccomp check
+```
+
 ### Exit Emulator
 
 To exit kvm-host, press "Ctrl-A", release both keys, and then press "x".

src/main.c

@@ -4,9 +4,18 @@
 #include <unistd.h>
 
 #include "err.h"
+#include "seccomp.h"
 #include "vm.h"
 
 static char *kernel_file = NULL, *initrd_file = NULL, *diskimg_file = NULL;
+static int enable_seccomp = 0;
+
+/* Long-only option ids start above the ASCII range so they can never collide
+ * with a short-option char in the getopt_long return.
+ */
+enum {
+    OPT_SECCOMP = 256,
+};
 
 #define print_option(args, help_msg) printf("  %-30s%s", args, help_msg)
 
@@ -19,6 +28,8 @@ static void usage(const char *execpath)
     print_option("-i, --initrd initrd", "Initial RAM disk image\n");
     print_option("-d, --disk disk-image",
                  "Disk image for virtio-blk devices\n");
+    print_option("--seccomp",
+                 "Install a seccomp BPF allowlist before vm_run.\n");
 }
 
 static struct termios saved_attributes;
@@ -50,9 +61,8 @@ int main(int argc, char *argv[])
 {
     int option_index = 0;
     struct option opts[] = {
-        {"kernel", 1, NULL, 'k'},
-        {"initrd", 1, NULL, 'i'},
-        {"disk", 1, NULL, 'd'},
+        {"kernel", 1, NULL, 'k'}, {"initrd", 1, NULL, 'i'},
+        {"disk", 1, NULL, 'd'},   {"seccomp", 0, NULL, OPT_SECCOMP},
         {"help", 0, NULL, 'h'},
     };
 
@@ -69,6 +79,9 @@ int main(int argc, char *argv[])
         case 'd':
             diskimg_file = optarg;
             break;
+        case OPT_SECCOMP:
+            enable_seccomp = 1;
+            break;
         case 'h':
             usage(argv[0]);
             exit(0);
@@ -97,8 +110,15 @@ int main(int argc, char *argv[])
     if (vm_late_init(&vm) < 0)
         return -1;
 
-    /* Switch the terminal to raw mode only once setup has succeeded so that
-     * any error from the load/init paths above is rendered on a normal tty.
+    /* Lock down the syscall surface before raw-mode and vm_run, so a
+     * memory-corruption RCE in device emulation cannot escape to arbitrary host
+     * syscalls. Off by default — opt in via --seccomp.
+     */
+    if (enable_seccomp && seccomp_apply() < 0)
+        return -1;
+
+    /* Switch the terminal to raw mode only once setup has succeeded so that any
+     * error from the load/init paths above is rendered on a normal tty.
      */
     set_input_mode();