Version: 1.0 (2026-02-05) Status: Canonical bypass catalog for the v1 contract.
This catalog records known bypass surfaces and their disposition. Each entry is classified as accepted, mitigated, or roadmap to keep claims defensible.
- Root compromise / kernel compromise
- Out of scope by threat model. Kernel modules or root can bypass policy.
- Non-LSM enforcement paths when BPF LSM is unavailable
- Tracepoint fallback is audit-only; syscall deny is not possible.
- Privileged container escape with host-level capabilities
- Treated as root-equivalent in scope definition.
- Symlink swaps
- Canonical path resolution + inode-based enforcement.
- Evidence:
docs/EDGE_CASE_COMPLIANCE_SUITE.md(symlink swap scenarios).
- Rename / hardlink path drift
- Inode-based deny persists across renames and hardlinks.
- Evidence:
docs/EDGE_CASE_COMPLIANCE_SUITE.md(rename + hardlink scenarios).
- Bind-mount aliases
- Enforcement is inode-driven; path telemetry can differ by namespace.
- Evidence:
docs/EDGE_CASE_COMPLIANCE_SUITE.md(bind‑mount alias scenarios).
- Outbound message sends (
sendmsg)- Covered by the same remote endpoint deny semantics as
connect()when the kernel exposessocket_sendmsg. - Evidence:
docs/NETWORK_LAYER_DESIGN.md,docs/POLICY_SEMANTICS.md.
- Covered by the same remote endpoint deny semantics as
- Pre-accept inbound policy coverage
- Add earlier inbound filtering or richer hook coverage before
accept()returns the socket to user space.
- Add earlier inbound filtering or richer hook coverage before
- Broader filesystem matrix
- Extend validation beyond ext4/xfs to additional FS types.
- Namespace-specific path views
- Improve operator tooling to reconcile path differences across namespaces.