fix(vsock-host): poison interrupted frame writes

[seven332]

Summary

• centralize host request-frame writes behind a guarded helper
• poison and close the vsock connection if an active frame write fails or is cancelled
• add regression coverage for active-write cancellation and pre-lock cancellation

Tests

• cargo test --manifest-path crates/Cargo.toml -p vsock-host
• cargo clippy --manifest-path crates/Cargo.toml -p vsock-host --all-targets -- -D warnings
• pre-commit: cargo-fmt, cargo-doc, cargo-clippy

Fixes #12205

[codecov]

Codecov Report

❌ Patch coverage is 95.74468% with 8 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
crates/vsock-host/src/lib.rs	95.74%	8 Missing ⚠️

📢 Thoughts on this report? Let us know!

[seven332]

Code Review: PR #12247

Summary

Reviewed the vsock-host cancellation fix and tests. No P0/P1 issues found. The guarded write helper preserves the queued-writer cancellation behavior while poisoning the connection on active frame-write cancellation or failure.

Key Findings

Critical Issues (P0)

None.

High Priority (P1)

None.

Testing Review

Coverage

• fix: active request-frame write cancellation corrupting the shared connection -> covered by test_cancel_during_frame_write_closes_connection.
• fix: cancellation while waiting for writer lock should not close the connection -> covered by test_cancel_while_waiting_for_writer_lock_does_not_close_connection.
• fix: bounded exec active write cancellation must clean pending response and stream registrations -> covered by test_cancel_during_bounded_exec_frame_write_cleans_up_registrations.

Convention Compliance

No violations found. The Rust inline tests use real UnixStream pairs, avoid artificial sleeps/fake timers, and synchronize via Notify or explicit polling. The test-only setsockopt unsafe block is narrowly scoped and documents its safety condition.

Testing Verdict: Adequate

Bad Smell Analysis

No bad-smell blockers found. No new public API surface, no fallback/error swallowing pattern, no artificial production delay, and no resource ownership issue found. The raw fd remains borrowed; socket ownership stays with the split stream halves.

Recommendations

No code changes requested.

Verdict

LGTM

---

Full review details: codereviews/20260509/
Testing standards: docs/testing.md