[mcc0nnell]

Summary

Apply MS_RDONLY together with the other per-mount restrictions when the new mount API clones a recursive bind subtree.

open_tree(..., AT_RECURSIVE) clones nested submounts, but doBindMountAt() previously stripped MS_RDONLY before the recursive mount_setattr() call. The later bind remount only affects the top mount. With a writable jail root, nested submounts could therefore remain writable.

This passes the full requested flag set to applyMountFlags(), so MOUNT_ATTR_RDONLY is applied recursively before move_mount().

Fixes #369.

Validation

• Project Dockerfile builds successfully with the patch.
• New mount API smoke test exits successfully on Linux 6.8.
• Focused kernel-level reproduction using a recursive open_tree() clone with a nested tmpfs:
  • before: top bind read-only, nested tmpfs writable
  • after recursive MOUNT_ATTR_RDONLY: top bind read-only, nested tmpfs read-only
• git diff --check passes.

This intentionally does not change the separate mount-attribute failure handling tracked in #312.

[mcc0nnell]

Superseded by #371, which includes the same recursive-readonly correction plus fail-closed handling and a privilege-gated regression for nested tmpfs, writable recursive binds, and read-only file binds. Consolidating review on #371 to avoid duplicate PRs. Thanks!