Skip to content
Log inDashboard

Security policy

Email [email protected], in English.

An encryption key (age and PGP) is not published yet. Until it is, send the description and impact by email and say that you hold exploit details; we reply with a way to send them privately. When the key exists, its fingerprint appears here and in security.txt (Encryption:), and the key itself at https://opdns.io/.well-known/pgp-key.txt.

Please include what is affected (component, version or commit, hostname), steps to reproduce, the impact as you understand it, and whether you want to be credited. Do not report vulnerabilities in public GitHub issues or discussions.

In scope:

Component Examples
Resolver (opdns-edge, Unbound configuration) malformed DNS, DoH, DoT, DoQ or DoH3 input; identification bypass (answering a profile’s queries unfiltered, or as another profile); cache poisoning through our layer; bypass of rate limits or response-size caps; EDE or policy confusion
Control plane API (api.opdns.io, /v1/*) sign-in through opauth (the OIDC client: state, nonce, PKCE, redirects, the staff checks of the operator console), sessions, step-up, API token scopes, cross-account access to profiles, logs or analytics, DDNS token misuse
Dashboard (app.opdns.io) XSS, CSRF, clickjacking on authenticated pages, token leakage
Link (link.opdns.net) and self-hosted node (opdns-node) node impersonation, reading another node’s logs, relayed-query injection, local page or local API exposure
List pipeline forging or replaying a signed list artifact, getting a domain into the fleet artifact without our signing key
Operator tooling anything that changes operator blocks or source blocks without the operator
This repository secrets in history, CI workflows exploitable from fork PRs

Out of scope:

  • The unfiltered public resolver path answering queries: that is intended. Amplification beyond our stated caps is in scope.
  • Denial of service by volume against our fleet, and load testing. Report the idea, do not run it.
  • Findings in third-party services we use (our hosting and network providers, the mail provider, the status page provider). Report those to the vendor. Tell us if the issue is in our configuration of them.
  • Blocklist contents, meaning a domain wrongly blocked or not blocked. Send that to [email protected] as list feedback.
  • Missing security headers or best-practice notes with no demonstrated impact, SPF/DMARC reports without a spoofing path, and self-XSS.
  • Social engineering of us or our users, and physical attacks.
  • Vulnerabilities in Unbound, Go or other upstream software itself. Report those upstream, and tell us if our use makes them worse.

If you act in good faith under this policy, we will not pursue or support legal action against you, and we consider your research authorised under applicable anti-hacking and anti-circumvention laws (in Canada, among others, the Criminal Code’s unauthorised-use-of-computer provisions). Good faith means:

  • testing only against your own account, profiles and nodes, or against a local build (the repository’s local development setup runs the whole system on your machine, and we prefer that you test there);
  • not accessing, keeping or modifying other users’ data beyond the minimum needed to show the issue. If you reach someone else’s data, stop and tell us;
  • no degradation of the service for others, no spam, and no social engineering;
  • giving us reasonable time to fix the issue before disclosure (below).

If a third party brings a claim against you over research that followed this policy, we will say publicly that your actions were authorised. We cannot bind other parties, such as our providers.

Step Target
Acknowledgement 3 business days
Triage (validity, severity, scope) 10 business days
Fix for critical and high (exploitable remotely, cross-account data, fleet-wide integrity) 30 days
Fix for medium and low 90 days
Coordinated public disclosure when fixed, or at 90 days, whichever is first, unless we agree otherwise

opdns is run by one person. If a target slips, we tell you why and give a new date. We keep you informed at each step. Severity uses CVSS 4.0 as a guide, adjusted for a DNS privacy product: anything that exposes query logs or identifies users counts as high at least.

For the self-hosted node, fixes ship as a new container image and binary. The release notes say whether an update is security-relevant. Nodes do not auto-update, so we also post the advisory on GitHub Security Advisories and request a CVE where one applies.

We credit reporters in the advisory and on a thanks page, unless you prefer not to be named. There is no bug bounty: no cash rewards and no swag during the beta, credit only. If we start one, it will be announced here and will not apply retroactively.

The hosted service is always the current version. For the self-hosted node, the latest release and the one before it receive security fixes during the beta.