From 3904661440d8fb5ca2e2274a2b4a38320ad83e2e Mon Sep 17 00:00:00 2001
From: Paul Meyer <katexochen0@gmail.com>
Date: Wed, 26 Aug 2026 11:33:34 +0200
Subject: [PATCH] test: make boot mode work in unprivileged containers

In a mount namespace owned by a non-initial user namespace, the kernel
only allows mounting a fresh proc or sysfs if the namespace already
contains a fully visible mount of the same filesystem type
(mount_too_revealing()). systemd-nspawn's pivoted root contains no such
mount, so booting the test container inside an unprivileged container
(e.g. a containerized CI runner) fails with EPERM when mounting /proc.

When running in a non-initial user namespace, satisfy the check by
binding a read-only instance of each into the container under /work.
Skipping the binds on regular privileged runs keeps the container
isolated from the host's proc and sysfs. Binding non-recursively keeps
foreign submounts (e.g. lxcfs overmounts) out, which would spoil full
visibility where they end up locked, and which read-only would not
cover.

Signed-off-by: Paul Meyer <katexochen0@gmail.com>
---
 .../integration-test-wrapper.py               | 30 ++++++++++++++++++-
 1 file changed, 29 insertions(+), 1 deletion(-)

diff --git a/test/integration-tests/integration-test-wrapper.py b/test/integration-tests/integration-test-wrapper.py
index ed8a8f2607..19dcf2bc28 100755
--- a/test/integration-tests/integration-test-wrapper.py
+++ b/test/integration-tests/integration-test-wrapper.py
@@ -365,6 +365,15 @@ def process_coverage(args: argparse.Namespace, summary: Summary, name: str, jour
             print(f'Wrote coverage report for {name} to {output}', file=sys.stderr)
 
 
+def in_userns() -> bool:
+    # The initial user namespace has a well-known inode number (USER_NS_INIT_INO, linux/nsfs.h).
+    # Without kernel support for user namespaces, ours can only be the initial one.
+    try:
+        return os.stat('/proc/self/ns/user').st_ino != 0xEFFFFFFD
+    except OSError:
+        return False
+
+
 def coco_host_type() -> Optional[str]:
     try:
         if Path('/sys/module/kvm_intel/parameters/tdx').read_text().strip() == 'Y':
@@ -706,7 +715,26 @@ def main() -> None:
         *(['--runtime-build-sources=no', '--register=no'] if not sys.stdin.isatty() else []),
         'vm' if vm else 'boot',
         *(
-            ['--', '--capability=CAP_BPF', f'--suppress-sync={"yes" if args.suppress_sync else "no"}', *vm_images_args, *coco_boot_args]
+            [
+                '--',
+                '--capability=CAP_BPF',
+                f'--suppress-sync={"yes" if args.suppress_sync else "no"}',
+                # In a non-initial user namespace the kernel only allows mounting a fresh proc or
+                # sysfs if the mount namespace already contains a fully visible mount of the same
+                # filesystem type (mount_too_revealing()), which nspawn's pivoted root doesn't.
+                # Satisfy the check with read-only binds, but only when actually confined, as
+                # they expose the outer proc and sysfs to the container. Bind non-recursively:
+                # carried-along submounts (e.g. lxcfs overmounts of container runtimes) would
+                # spoil full visibility wherever they end up locked, and --bind-ro wouldn't make
+                # them read-only either.
+                *(
+                    ['--bind-ro=/proc:/work/proc:norbind', '--bind-ro=/sys:/work/sys:norbind']
+                    if in_userns()
+                    else []
+                ),
+                *vm_images_args,
+                *coco_boot_args,
+            ]
             if not vm
             else []
         ),