docs(sso): document the machine surface, and stop restating it in nix

`docs/swarm/sso.md` described a person in a browser. The swarm's other
callers — the telemetry collector, the queue's auth-callout responder, each
hive's agents — hold no session and follow no redirect, and nothing operator-
facing said how they authenticate. Its relying-party table is forge and
matrix, both browser surfaces.

The new section carries what `swarm-authelia.nix` was holding in comments:
one client per hive because identity belongs to the directory, the audience
being that client id rather than a parallel naming scheme, and signed rather
than opaque tokens because the collector verifies offline against
`/jwks.json` while the queue introspects.

It also states the fail-closed rule once, in the place a reader looks before
touching a vhost: an error page answers 200, and `auth_request` reads any 2xx
as access granted. That shape has now appeared three times — this module's
`/api/` prefix and both of victorialogs' routes — which is what makes it
documentation rather than a comment.

The two comment blocks those replace shrink to the part that is genuinely
local: the submodule-typing reason these clients are a definition rather than
an append, and a loud warning against folding the machine prefix back into
`/`. The security warning stays at the site; only its consequence list moves.

Comments 495 -> 465 lines. Option `description` strings are untouched: they
are the source `pkgs.nixosOptionsDoc` renders into the operator's options
reference, so trimming one would delete published documentation rather than a
duplicate.
This commit is contained in:
atlas 2026-09-02 02:53:51 +02:00 committed by mara
commit 9b14014077
2 changed files with 73 additions and 41 deletions

View file

@ -177,6 +177,69 @@ Two consequences worth stating plainly:
directly and authenticates itself. The forward-auth vhosts protect
browser surfaces; this is not one of them.
## Machine clients
Everything above is a person in a browser. A swarm also has callers that
hold no session and follow no redirect: the telemetry collector, the
queue's auth-callout responder, and each hive's own agents.
**One client per hive, not one per service.** A hive's identity belongs to
the directory rather than to whichever service happens to consume it, so a
hive holds a single OIDC client — `<hiveClientPrefix><hive>` — and mints a
different token per service from it. The alternative, letting each
consuming subsystem declare its own list, collides on the same client id
the moment a second consumer appears.
**The audience is that client id.** A swarm service that has to tell hives
apart needs one name both sides already agree on, and the client id is
already that name. A parallel per-hive naming scheme would be a second
thing to keep in step, and it drifts silently — a mismatch presents as a
valid token refused at the target, which reads like a broken credential
rather than a broken name.
**Tokens are signed (`RS256`), not opaque**, because a resource server
that cannot call the provider back is a real case here: the telemetry
collector verifies offline against `/jwks.json`, and an opaque token gives
it nothing to verify. The queue's responder introspects instead — a
different question asked of the same token, and the reason both
`/api/oidc/introspection` and `/jwks.json` have to stay reachable.
### Machine callers must fail closed
⚠️ **An error page that answers `200` is a security bug, not a cosmetic
one.** The browser surface intercepts upstream errors and serves a
friendly "SSO is unavailable" page; that page is a file, so it returns
`200`. Any machine caller routed through it receives a success carrying
HTML instead of the failure that actually happened:
- `/api/authz/auth-request` — nginx `auth_request` treats **any 2xx as
success**, so a down provider means *access granted*
- `/api/oidc/introspection` — a token check that answers `200`
- `/api/oidc/token`, `/.well-known/openid-configuration` — a client
parsing an error page as its JSON document
So authelia's `/api/` and `/.well-known/` prefixes are routed **without**
error interception. The split is by *audience*, not by an enumerated path
list: a human gets the page, every machine caller gets the status.
Enumerating endpoints individually would leave the next one added
silently intercepted.
The same shape bites any machine route behind a browser-shaped gate: a
`302` to a login page is followed, the login page answers `200`, and the
caller reports success while nothing happened. Log ingest hit exactly this
and lost eleven hours of delivery in silence.
**Checking it, if you change this routing.** Point the vhost at a dead
upstream and compare three requests, not one:
1. through `/` — must still serve the friendly page
2. through `/api/` — must deny
3. a direct dial to authelia — must match what (2) did
All three matter. A change that silently deleted the browser page would
pass a deny-only check, and one that quietly stopped denying would pass a
page-only check. This was verified that way when the split was introduced.
## What this does not do
- **It does not disable local login.** Each service keeps its password