Skip to content

Security: LLCForge/Notary

Security

SECURITY.md

Security Policy

Notary runs close to network infrastructure, so we take security reports seriously and appreciate responsible disclosure.

Supported versions

Version Supported
Latest release Yes
main branch Yes (best effort)
Older releases No, please upgrade

Reporting a vulnerability

Do not open a public issue, discussion or pull request for security problems.

Report privately through GitHub Security Advisories:

  1. Go to https://github.com/LLCForge/Notary/security/advisories/new.
  2. Describe the issue, affected version or commit, and steps to reproduce.
  3. Include a proof of concept if you have one. Strip any real customer data, IP addresses or credentials from captures and logs.

What to expect:

  • Acknowledgement within 3 business days.
  • An initial assessment and severity rating within 10 business days.
  • A fix or mitigation plan, coordinated disclosure date, and credit in the advisory (unless you prefer to stay anonymous).

In scope

  • Incorrect validation results (a route reported Valid that RFC 6811 says is Invalid, or a wrong ASPA verdict).
  • Remote crash or memory exhaustion through crafted RTR PDUs (including ASPA), JSON exports, SLURM files, or responses from RIPEstat, Pathology, bgp.tools, the RIPE DB or whois servers.
  • TLS certificate or SSH host key verification bypass on tls:// / ssh:// RTR transports.
  • Query API exposure beyond its read-only contract: bearer-token bypass on /v1/*, the batch endpoint, history or ASPA queries, or the admin routes (/v1/silences, POST /v1/reload).
  • Script injection or data exfiltration through the embedded web UI.
  • Secrets (tokens, passwords, SSH passphrases, webhook URLs) leaking into logs, metrics or API responses.
  • Dependency vulnerabilities that are reachable from Notary's code paths.

Out of scope

  • Vulnerabilities in third-party services or protocols themselves (report those upstream).
  • Issues that require an attacker to already control the host or the config file.
  • Denial of service by volumetric flooding of the network the tool runs on.

Hardening recommendations

  • Run as an unprivileged user; grant only the capabilities the tool needs.
  • Bind management and API listeners to localhost or a management network.
  • Keep secrets in environment variables or your secret manager, never in the config file or in version control.
  • Use tls:// or ssh:// for RTR across any network you do not fully control; plain TCP RTR is unauthenticated. Always set ssh_host_key from a fingerprint you verified out of band.
  • Set [api] token_env when the API is reachable by anyone but you. The API has no per-connection header timeout; put a reverse proxy in front if it must face untrusted clients.
  • Set separate, long (>= 16 bytes) tokens for [silences] api_token_env and [health] admin_token_env; they allow muting alerts and reloading the configuration.
  • The data directory holds the alert queue (alert contents), silences and VRP history; keep it readable only by the service account (the shipped systemd unit, Windows installer and container already do).

Attack surface added in 0.2

Surface Mitigation
POST /v1/validity batch 4 MiB body cap, max_batch routes (default 10000), per-entry errors, optional bearer token
/v1/history, /v1/aspa Bounded parameters (prefix 64 bytes, path 4096 bytes, limit <= 1000), file reads off the async runtime, optional bearer token
Web UI (/, /ui/*) Static files compiled into the binary, CSP script-src 'self', no inline script, textContent only, frame-ancestors 'none'
RTR over TLS / SSH rustls with verified server certificates; SSH host key pinned, RSA disabled (RUSTSEC-2023-0071), credentials from env
RTR v2 ASPA PDUs Strict length and provider-list validation, 64 KiB PDU cap
Outbound calls (Pathology, bgp.tools, RIPE DB, whois) Fixed hosts or configured URLs only, response size caps, timeouts, never on the alert path

There aren't any published security advisories