- Add custom RateLimitMiddleware using governor crate for per-IP rate limiting
- Two-tier rate limiting: destructive (20 req/min, burst 10) and read (120 req/min, burst 30)
- Health endpoints (/health, /api/v1/system/info) exempt from rate limiting
- Add max_queue_depth to JobManager (default: 100, configurable via config.yaml)
- Return 429 Too Many Requests with Retry-After header when queue is full
- Add RateLimitConfig to config.yaml with all rate limit settings
- Add 10 tests covering rate limiting, queue depth, and configuration defaults
CI/CD Pipeline / Build Debian Package (push) Failing after 4s
CI/CD Pipeline / Build Arch Package (push) Successful in 2m24s
CI/CD Pipeline / Build RPM Package (push) Successful in 2m15s
CI/CD Pipeline / Build Alpine Package (push) Successful in 3m16s
* test: add CRL unit tests and CrlAwareVerifier construction tests (PR 6 of 6)
* fix(ci): rename fmt job to match required status check context
---------
Co-authored-by: git-echo <git-echo@moon-dragon.us>
Implements agent-side CRL consumption for mTLS certificate revocation
checking, as specified in issue #20.
Changes:
- NEW: src/auth/crl.rs - CRL loading, parsing, signature verification,
in-memory revoked serial index (HashSet), 24h background refresh task
- MODIFY: src/auth/mtls.rs - CrlAwareVerifier wrapping WebPkiClientVerifier
with post-chain CRL serial lookup; fails closed on invalid signature,
degrades gracefully when CRL is missing
- MODIFY: src/auth/mod.rs - Register crl module, re-export CrlState/CrlStatus
- MODIFY: src/config/loader.rs - Add crl_path field to TlsConfig
- MODIFY: src/main.rs - Load CRL on startup, spawn refresh task, wire
SharedCrlState into server and health endpoint
- MODIFY: src/api/handlers/system.rs - Add crl_status and crl_age_seconds
to health check response
- MODIFY: Cargo.toml - Add arc-swap, base64 deps; enable x509-parser
verify feature for CRL signature verification
Design decisions:
- ArcSwap for lock-free atomic CRL state swaps on the hot path
- O(1) serial lookup via HashSet<String> of hex-encoded serials
- Stale CRL = continue serving + warn + health reports degraded
- Invalid CRL signature = refuse to start (fail-closed)
- Missing CRL = fall back to WebPKI-only (backward compatible)
Companion to PR #26 in linux-patch-manager (manager-side CRL generation)
Refs: #20
- Auto-enrollment on startup when certs are missing/invalid and enrollment.manager_url configured
- Certificate validation (existence, parse, expiry, key match, CA trust)
- --enroll exits after completion (no port conflict with systemd service)
- --renew-certs flag for manual cert renewal
- SO_REUSEADDR on TcpListener::bind (prevents Address already in use)
- Polling token persistence for enrollment resume after restart
- Exit code strategy (0=clean, 1=error, 2=enrollment in progress)
- HTTP 409 (host already exists) handling during enrollment
- Move 'Listening on' log after actual bind
- Increase RestartSec to 10s and add StartLimitBurst=5
- Postinst checks for certs and enrollment URL, prints guidance
- EnrollmentConfig.manager_url changed to Option<String>
- cert_renewal_threshold_days and polling_token config fields
- Updated SPEC.md and DEPLOYMENT_GUIDE.md with new workflow
- RCA document for crash loop root cause analysis
- Version bumped to 1.2.0
- New src/packages/cache.rs module with PackageCacheState, stale detection,
state persistence, 404 retry logic
- Add refresh_package_cache() and last_cache_update() to PackageManagerBackend
trait, implemented on all 5 backends (APT, DNF, YUM, APK, Pacman)
- Health check now reports last_cache_update and cache_status fields,
triggers cache refresh if stale (>4h), returns degraded on failure
- Patch apply jobs now force cache refresh before applying patches,
with 404/fetch error retry (1 retry after cache refresh)
- Cache state persists to /var/lib/linux_patch_api/state/cache.json
- Version bump to 1.1.17
- Update ARCHITECTURE.md and REQUIREMENTS.md (FR-007)
Closes: #2
Root cause: .apk-install is not a valid abuild suffix (lines 247-257 of abuild).
abuild expects SEPARATE files: pkgname.pre-install, .post-install, .pre-deinstall, .post-deinstall.
The old single .apk-install file caused abuild to die with "unknown install script suffix",
but CI used || true which masked the failure, so APK was built WITHOUT install scripts.
Verified on actual Alpine runner: install script suffixes now pass abuild validation.
- configs/linux-patch-api.pre-install: create dirs, set permissions (matches Debian preinst)
- configs/linux-patch-api.post-install: copy example configs, enable service (matches Debian postinst)
- configs/linux-patch-api.pre-deinstall: stop and disable service (matches Debian prerm)
- configs/linux-patch-api.post-deinstall: clean up empty dirs (matches Debian postrm)
- Removed configs/linux-patch-api.apk-install (invalid format)
- Updated build-alpine.sh: copy 4 install scripts to workspace, updated install= line in APKBUILD
- Insert 'Clean previous build artifacts' step (cargo clean + rm old .deb)
before Build Debian package in both build-deb and build-deb-u2204 jobs.
- Bump version to 1.1.7 to ensure a clean build from scratch.
- Update debian/changelog with 1.1.7-1 entry.
- Add rustls::crypto::aws_lc_rs::default_provider().install_default()
in main() before any TLS operations to prevent startup panic
- Bump version from 1.1.5 to 1.1.6
- Update debian/changelog with 1.1.6-1 entry
Requiring default-features=false on reqwest prevents native-tls/openssl-sys
from being pulled in as transitive dependencies, which broke static linking
on Alpine musl target. Also reverts invalid openssl-static package from CI.
- Cargo.toml: add default-features = false to reqwest dependency
- ci.yml: revert non-existent openssl-static package
- Add ServiceStatus struct with name, display_name, active_state, sub_state,
load_state, enabled_state, main_pid, healthy fields
- Add get_service_status() to PackageManagerBackend trait
- Implement get_service_status() in AptBackend with systemd and OpenRC support
- Add get_service_status HTTP handler in system.rs
- Add /system/services/{name} route
- Add E2E test for service status endpoint
- Bump version to 0.3.6
- Remove CapabilityBoundingSet and AmbientCapabilities (apt needs full root capabilities)
- Remove ReadWritePaths (unnecessary without ProtectSystem=strict)
- Fix E2E test: properly FAIL on status=failed package operations
- Fix E2E test: require status=completed for install/update/remove lifecycle
- Update dpkg packaging service file to match configs/
- Bump version to 0.3.5
- Remove sudo from apt command execution (service runs as root)
- Remove RestrictSUIDSGID from systemd service (blocks setuid for apt/dpkg)
- Remove NoNewPrivileges from systemd service (blocks sudo PERM_SUDOERS)
- Bump version to 0.3.2
- Fix reboot_system() to use shutdown -r +N for delayed reboots
- Fix patches handler to call reboot_system() instead of just logging
- Add CAP_SYS_BOOT capability to systemd service for LXC reboot support
- Remove unused warn import from packages/mod.rs
- Bump version to 0.3.1
- Fix ServerConfig::builder() to builder_with_provider() for TLS 1.3 enforcement
- Add aws_lc_rs feature to rustls in Cargo.toml
- Fix clippy unnecessary_map_or -> is_some_and in packages/mod.rs
