Run it · 4 of 4
Security
What Cairn OBS can and can't do on your network, how agents and the server trust each other, what's sealed and what's audited, and how to report a vulnerability.
A monitor sees everything, so it's a tempting target. Cairn OBS is built so that taking it over gets an attacker as little as possible.
It looks; it doesn't touch
- No remote shell, no "run this command", no "restart this service".
- Every account it's given elsewhere is read-only: Kubernetes
get,listandwatch; PVEAuditor on Proxmox; Read-only on vCenter; Prism Viewer on Nutanix; Read-Only on Pure; hmcviewer on the HMC; a profile with no special authorities on IBM i; a Vault policy that can't read a secret. - The two exceptions, both for certificates, are off until each host or cluster allows them in its own config, and each is a button an admin presses (Certificates). Because the host's own config decides, a stolen obs-beacond can't turn them on.
A stolen Cairn OBS session shows someone your dashboards. It doesn't hand them your servers.
Agents and the server
- Agents only connect out. Hosts need no open port.
- The agent port is TLS 1.3 with client certificates from a CA obs-beacond makes for itself, used for nothing else. Agents pin that CA by its fingerprint, learned from the join token.
- Each agent has one live certificate, for 90 days, renewed after 60. Renewing, joining again or removing a host revokes the one before at once.
- A host's identity comes from its certificate, never from what it says about itself, so one host can't report as another.
- Join tokens work once, within an hour, and only their SHA-256 is kept.
- The agent runs as an unprivileged user. Its root helper answers only a short list of read requests on a local socket, and the one thing it can change, a certificate reload, it checks against the host's own config, not the agent's word.
- On Windows the agent runs as its own virtual account with no administrator rights.
What's kept, and how
- Secrets are sealed in the database with the state key (AES-256-GCM): sign-in providers' tokens, channel secrets entered in the web interface, the agent CA's key. Session cookies are stored only as their SHA-256. The database alone signs nobody in.
- In files, secrets are always files of their own, never inline in the config, and they never appear in logs or errors.
- Internet clients' addresses never leave the host. The maps count public connections; they don't record who made them.
- No private key is ever read for the certificate inventory.
- No analytics, no telemetry. Cairn OBS doesn't phone home. It talks to what you point it at, and the web interface loads nothing from outside.
Sign-in
- Single sign-on over OpenID Connect, with roles from groups or verified emails, rechecked every 15 minutes. Nobody who matches no rule gets in.
- Passkeys, or a password and an authenticator code, never a password alone. Five failures in 15 minutes block that address or account for the rest of that time.
- Setup links work once, and the secret part never reaches a server log.
- Viewers don't see container log lines, Configure or the audit log.
The audit log
Every change is in the audit log, with who made it and from where: sign-ins, settings, users, hosts joined and removed, certificates issued to agents, accepted DNS changes, test notices, every reload or renew asked for and what came of it, and every emergency sign-in. Operators and admins can read it, under Configure › Audit log.
The server's unit
The packaged obs-beacond.service runs as the obs-beacon user with no
capabilities but binding ports below 1024, a read-only view of the system
(ProtectSystem=strict), no access to home folders, and no new privileges.
The container image runs as a non-root user with a read-only root and no
capabilities.
Releases
Each release's files are listed with their SHA-256 in SHA256SUMS, signed
with the Cairn OBS release key using minisign
(Download and check). install.sh
checks the agent package against SHA256SUMS before installing it.
Reporting a vulnerability
Email security@coffeylabs.org. Please don't open a public issue for a security problem. We'll answer, agree a date for a fix and a disclosure with you, and credit you if you'd like.