5 Commits

Author SHA1 Message Date
Paul Payne
60f3ca4a3a Security hardening: input validation, secrets protection, NATS auth
Config injection prevention:
- Add FQDN validation for domain names (RFC 1123) in Register/Update —
  rejects newlines, spaces, shell metacharacters that could inject into
  HAProxy/dnsmasq configs
- Add backend address validation (valid host:port format, valid IP or
  hostname, port 1-65535). DNS-only backends allow bare IPs.
- Add header key/value validation — keys must be HTTP token chars only,
  values must not contain newlines or NULs
- Add WireGuard peer name validation (alphanumeric + hyphens + underscores)
- Add defense-in-depth domain validation in certbot Provision()

Secrets protection:
- Remove ?raw=true bypass on GET /api/v1/secrets — secrets are now always
  redacted in API responses regardless of query parameters
- Update test to verify redaction cannot be bypassed

NATS authentication:
- Generate random auth token on first startup, store in secrets.yaml
- Pass token to embedded NATS server via Authorization option
- Internal client connects with the same token
- External NATS clients (Wild Cloud) must now authenticate

Security headers:
- Add X-Content-Type-Options: nosniff
- Add X-Frame-Options: DENY
- Add Cache-Control: no-store
2026-07-14 12:38:17 +00:00
Paul Payne
428d47f876 Refactor architecture: extract reconciler, add interfaces, reduce complexity, improve test coverage
Architecture:
- Extract reconcileNetworking into internal/reconcile package with 7 consumer-side interfaces
- Add locked modifyState helper to fix state.yaml read-modify-write race condition
- Extract CrowdSec and VPN handler groups with interfaces documenting dependency surface
- Replace raw map[string]any YAML manipulation with typed AddDHCPStaticLease/RemoveDHCPStaticLease
- Extract Cloudflare API functions into cfClient struct, eliminating repeated auth boilerplate

Complexity reduction:
- haproxy.GenerateWithOpts: 50 → 6 (extracted 8 focused helpers)
- reconcile.Reconcile: 42 → 11 (extracted buildRoutes, buildDNSEntries, writeHAProxyConfig)
- AutheliaUpdateConfig: 25 → 10 (extracted enableAuthelia/disableAuthelia)

Test coverage improvements:
- reconcile: 4.5% → 56.8% (stub-based tests for route building, DNS entries, orchestration)
- dnsfilter: 10.7% → 45.6% (Manager.Compile, AddList, ToggleList, custom entries)
- config: 67.3% → 89.8% (DHCP static lease mutation tests)
2026-07-14 04:21:30 +00:00
Paul Payne
79c0c32b98 services -> domains 2026-07-10 20:46:22 +00:00
Paul Payne
53a640c34f fix: Auto-recover NATS JetStream from stale store on startup
If the NATS server fails to start (stale lock files, corrupt store
from a crash), automatically clean the data directory and retry once.
This prevents the frustrating "NATS server failed to start" error
after unclean shutdowns.

The NATS KV data (registered services) is also stored as YAML files
in the services directory, so cleaning the NATS store loses nothing
— services are re-registered on the next startup or API call.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-07-09 14:12:29 +00:00
Paul Payne
6b58ebb692 feat: Add embedded NATS JetStream coordination bus
Central now starts an embedded NATS JetStream server as its core
coordination bus. This provides:

- wild-services KV bucket: service registrations from Cloud/Works
- wild-presence KV bucket: node liveness with TTL-based keys
- wild-events stream: events from all sources (24h retention)

NATS is started in main.go before the API and shut down on SIGTERM.
Cloud and Works will connect to Central's NATS (port 4222) to register
services and maintain presence.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-07-08 23:36:40 +00:00