Security

Report a security problem

JailOps runs a root helper on your hosts and holds the keys to your backups, so a security problem in it matters. If you think you found one, write to security@jailops.net rather than opening a public issue.

How to report

One email is enough.

Send it to security@jailops.net, in English. Please include:

  • The version: what jailops version prints, and the FreeBSD release and architecture.
  • The part: the daemon (jailopsd), the root helper (jailops-priv), the command line, a web console, the backup node, the packages or the app catalog.
  • What an attacker needs: network access to a port, an account and its role, access to the host, a jail.
  • The steps that show the problem, and what you saw.
  • What it allows in your view, and whether anyone else knows of it.

The same address is in /.well-known/security.txt. For anything that is not a security problem, use the issue tracker or hello@jailops.net.

What happens next

  • Reports are read and handled in good faith, by the people who write JailOps.
  • The aim is to confirm the problem, fix it, and release the fix before the details are public.
  • Each release lists its security fixes in the changelog, with the upgrade steps they need.

Independent reviews

What they found, and what changed.

JailOps has had independent reviews of the API, the web console, the command line and the appliance release, and an external security audit of 2 November 2026 by @samarason, which reported twelve findings. Each finding was checked against the code. The ones that held are fixed; the ones that were already closed in the code are recorded as such in the changelog, with tests that keep them closed. What was found and fixed, by kind:

  • The boundary of the root helper. Paths the daemon could have used to reach the jails' setuid programs or another jail's files: storage roots now stay root-only and cannot be mounted over, the helper's file operations stay inside one jail's tree, and its socket and the pid files moved out of directories the daemon can write.
  • Sign-in and sessions. Login attempts are counted and limited before the password check, an IPv6 source counts as its /64, revoked identities lose their event streams and console sessions, and only a global administrator sees other users' sessions.
  • The API. Request bodies refuse unknown or duplicate members, a request without an identity learns only the API version and how to sign in, and every answer is marked not to be cached.
  • What the daemon may send where. Log forwarding may read its credentials only from one root-owned directory and may not send to loopback, link-local or broadcast addresses or to the daemon's own ports; WireGuard routes are checked so a tunnel cannot take the host's traffic.
  • Output and audit. Text from the server is cleaned before it reaches a terminal, and audit entries are escaped and limited in size.
  • Host management and setup. A restore puts back only what a module owns, a setup token stops working once an administrator exists, and the last global administrator cannot be removed.
  • The backup node and encrypted volumes. A repository binds to the first host's key, a wrong escrow file or chunk can be removed by the node's administrator, and the root helper limits wrong passphrases itself.

The full list, item by item, is in the changelog.

Thanks to @samarason for the external security audit of 2 November 2026. The audit covered the code as it was then; it is not a review of a later release.

The security model

  • The daemon runs as an unprivileged user. A small root helper does the privileged work from a fixed list of typed operations and checks every request against allow-lists that root writes; it does not trust the daemon.
  • The API is on a local socket until you turn on the network listener or the console, which speak TLS 1.3 only.
  • Jails share the host's kernel. A project separates names, quotas, networks and firewall policy, not the kernel: tenants that must be protected from each other need a boundary chosen for that, such as hosts of their own.

Read docs/security-model.md before you give anyone the admin role or edit priv.conf.