feat: add support for read-only bind mounts in the linux sandbox (#9112)

### Motivation

- Landlock alone cannot prevent writes to sensitive in-repo files like
`.git/` when the repo root is writable, so explicit mount restrictions
are required for those paths.
- The sandbox must set up any mounts before calling Landlock so Landlock
can still be applied afterwards and the two mechanisms compose
correctly.

### Description

- Add a new `linux-sandbox` helper `apply_read_only_mounts` in
`linux-sandbox/src/mounts.rs` that: unshares namespaces, maps uids/gids
when required, makes mounts private, bind-mounts targets, and remounts
them read-only.
- Wire the mount step into the sandbox flow by calling
`apply_read_only_mounts(...)` before network/seccomp and before applying
Landlock rules in `linux-sandbox/src/landlock.rs`.
This commit is contained in:
viyatb-oai
2026-01-14 08:30:46 -08:00
committed by GitHub
parent bcd7858ced
commit e1447c3009
8 changed files with 676 additions and 13 deletions

View File

@@ -6,3 +6,52 @@ This crate is responsible for producing:
- a lib crate that exposes the business logic of the executable as `run_main()` so that
- the `codex-exec` CLI can check if its arg0 is `codex-linux-sandbox` and, if so, execute as if it were `codex-linux-sandbox`
- this should also be true of the `codex` multitool CLI
## Git safety mounts (Linux)
When the sandbox policy allows workspace writes, the Linux sandbox uses a user
namespace plus a mount namespace to bind-mount sensitive subpaths read-only
before applying Landlock rules. This keeps Git and Codex metadata immutable
while still allowing writes to other workspace files, including worktree setups
where `.git` is a pointer file.
Protected subpaths under each writable root include:
- `.git` (directory or pointer file)
- the resolved `gitdir:` target when `.git` is a pointer file
- `.codex` when present
### How this plays with Landlock
Mount permissions and Landlock intersect: if a bind mount is read-only, writes
are denied even if Landlock would allow them. For that reason, the sandbox sets
up the read-only mounts *before* calling `landlock_restrict_self()` and then
applies Landlock rules on top.
### Quick manual test
Run the sandbox directly with a workspace-write policy (from a Git repository
root):
```bash
codex-linux-sandbox \
--sandbox-policy-cwd "$PWD" \
--sandbox-policy '{"type":"workspace-write"}' \
-- bash -lc '
set -euo pipefail
echo "should fail" > .git/config && exit 1 || true
echo "should fail" > .git/hooks/pre-commit && exit 1 || true
echo "should fail" > .git/index.lock && exit 1 || true
echo "should fail" > .codex/config.toml && exit 1 || true
echo "ok" > sandbox-write-test.txt
'
```
Expected behavior:
- Writes to `.git/config` fail with `Read-only file system`.
- Creating or modifying files under `.git/hooks/` fails.
- Writing `.git/index.lock` fails (since `.git` is read-only).
- Writes under `.codex/` fail when the directory exists.
- Writing a normal repo file succeeds.