Discovery advertised the device_code grant in grant_types_supported, but dynamic
client registration only allowed authorization_code/refresh_token, so a client
listing the device grant in its RFC 7591 metadata was rejected with
invalid_client_metadata.
Introduce a single source of truth — OidcController::SUPPORTED_GRANT_TYPES —
used by discovery (grant_types_supported) and by registration validation, so
advertisement and registration can never drift. The token dispatcher already
handles exactly these grants for all clients (they are user-context grants gated
by consent + user_allowed?, so there is no per-client grant restriction), making
advertisement, registration, and enforcement consistent.
Tests: a client registering with the device_code grant now succeeds, plus an
assertion that registration's accepted set equals discovery's advertised set.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F7cwhwDJp3MJJDoNPVE6zq
More device-flow security-review fixes.
Replay-revocation on device_code redemption (was: destroy on use):
- Redemption now marks the code redeemed_at and links the issued access/refresh
tokens back to it (new oidc_device_code_id FKs, on_delete: :nullify) instead of
destroying the row. A replayed redeemed code is detected as reuse — revoking
every descended token and returning a distinguishable "already been used"
invalid_grant rather than the generic "Invalid device_code" — mirroring the
authorization-code reuse semantics (RFC 6749 §4.1.2). The device_code FK is
carried forward across refresh rotation so revocation reaches the whole chain.
Redeemed codes are reaped by the existing expiry cleanup sweep.
user_code CSPRNG (RFC 8628 §6.1):
- The 8-char user_code is a credential (type it + Approve mints tokens), so draw
it from SecureRandom instead of Ruby's global Mersenne Twister (Array#sample).
Rate-limit the /device verification endpoint (RFC 8628 §5.1):
- Add the app's standard 10/min limit on show + verify so a signed-in user can't
brute-force the short code space to deny or hijack a pending authorization.
Also carried in this commit (other agent's working-tree change):
- Give each OidcController rate_limit a distinct name: so frequent device polling
on the token endpoint no longer shares one counter with (and 429s) unrelated
token/refresh/revoke/introspect calls.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016Q4ATZHoMCWqvSpE2yYoie
Two more fixes from the device-flow security review:
slow_down interval grew without bound (OidcController):
- Each too-fast poll bumped the persisted interval by 5s with no ceiling, so a
client polling slightly fast — or an attacker spamming a known device_code —
could balloon it (5→10→15→…) past the 10-minute expiry and starve a legitimate
client of its token. Clamp the bump to OidcDeviceCode::MAX_INTERVAL (30s); a
client polling at the advertised interval is never throttled.
Expired device codes accumulated forever (OidcTokenCleanupJob):
- Anonymous callers can create device codes via /oauth/device_authorization, and
expired/denied/abandoned rows were never reaped. Sweep rows past expiry (with a
1-hour grace to stay clear of in-flight redemption); redeemed codes are already
destroyed at token issuance.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016Q4ATZHoMCWqvSpE2yYoie
Fixes from review of the device flow / introspection / DCR work:
Introspection over-disclosure (RFC 7662):
- Restrict introspection to authorized callers — the client a token was issued
to, or a resource server registered (resource_identifiers) to serve the token's
bound RFC 8707 audience. Unauthorized callers get {active:false}, disclosing
nothing, instead of any confidential client reading any token.
- Scope-gate disclosed claims: username only with the email scope, groups only
with the groups scope (mirrors userinfo). ADR 0005.
Tokens minted for revoked users:
- Re-check application.user_allowed?(user) at mint time in the device, refresh,
and authorization-code grants (covers app-active, user-active, group
membership). A user deactivated or removed from the allowed group between
approval and the token request is refused with access_denied. The refresh
check runs before rotation so a denied refresh has no side effects.
Device authorization hardening:
- Require confidential clients to authenticate, and require PKCE (code_challenge)
up front for clients that require it, so an intercepted device_code plus a
known public client_id cannot redeem tokens.
- Merge (not overwrite) the shared consent record on device approval.
Tests: new oidc_introspection_test and oidc_mint_authorization_test; existing
PKCE/claims tests updated to grant app access (they created ungrouped apps and
relied on the token endpoint not checking authorization). Full suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F7cwhwDJp3MJJDoNPVE6zq
Clients can name a target resource server via the `resource` parameter at the
authorization and device-authorization endpoints. Clinch binds it to the issued
token as its audience and reports it at introspection as `aud`, so a token minted
for one API cannot be replayed against another that also trusts clinch.
- `resource` is validated syntax-only (absolute URI, no fragment) — pass-through,
no resource registry; the resource server enforces the audience on introspection
- threaded from authorize / device_authorization onto the auth/device code, then
the access and refresh tokens, and carried across refresh rotation
- invalid values are rejected with error=invalid_target
- introspection `aud` is the bound resource, falling back to the client_id
- ADR 0004 records the pass-through-vs-registry decision
Completes the clinch-side OAuth surface for MCP connectors (with DCR +
introspection); Protected Resource Metadata (RFC 9728) lives on the resource server.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F7cwhwDJp3MJJDoNPVE6zq
Adds three OAuth surfaces to the OIDC provider so CLIs, terminal agents, and
MCP connectors can authenticate as real users instead of using static API keys.
Device Authorization Grant (RFC 8628):
- OidcDeviceCode model (HMAC device_code, short plaintext user_code, nullable
user until approval, slow_down polling state), mirroring OidcAuthorizationCode
- POST /oauth/device_authorization issues the code pair + verification URIs
- device_code grant on /oauth/token returns authorization_pending / slow_down /
access_denied / expired_token, then the standard token triple; single-use
- Authenticated /device approval page, gated by Application#user_allowed?
Token Introspection (RFC 7662):
- POST /oauth/introspect: confidential-caller-authenticated; returns active,
scope, and the user's groups so resource servers can authorize on membership
Dynamic Client Registration (RFC 7591):
- POST /oauth/register creates public/confidential clients (PKCE required)
- Runtime toggle via a new Setting store + admin switch on the Applications page;
off by default, env var CLINCH_DCR_ENABLED as bootstrap fallback
- New clients are default-deny (no allowed_groups) until an admin grants access
- RFC 8414 metadata alias at /.well-known/oauth-authorization-server;
registration_endpoint advertised only while the window is open
Discovery advertises all three grants/endpoints. Seeds add a clinch-cli public
client and a c2a2-introspection confidential client. ADRs in docs/decisions
record the opaque-vs-JWT, device-flow, and DCR-security decisions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F7cwhwDJp3MJJDoNPVE6zq
Two HIGH-severity findings from the security review:
- ForwardAuth: when no host header was present, /api/verify skipped the
application lookup and group check entirely, returning 200 with identity
headers (including all of the user's groups). This bypassed per-domain
access control. Now fails closed with 403, and the unreachable
DEFAULT_HEADERS fallback (the bypass path) is removed so headers are
always scoped to a resolved, active application.
- OIDC: the consent endpoint was in the verify_authenticity_token skip
list, so a forged cross-site POST could silently grant OAuth scopes.
Removed :consent from the skip list (the form already embeds the token).
Adds regression tests for both: fail-closed with no identity headers when
host is absent, and 422 on a tokenless consent POST.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The previous implementation iterated find_each(&:revoke!) on both the
access-token and refresh-token associations. OidcAccessToken#revoke!
also cascades to its refresh tokens, so a chain of N access tokens with
their refresh tokens produced ~3N UPDATEs (outer loop + cascade +
outer refresh loop double-writing) all while holding a pessimistic
lock on the auth_code row. Replace with scoped update_all on each
association -- 2 UPDATEs total, no behavior change.
Also hoist the repeated refresh_token_record.oidc_authorization_code
lookup in the rotation path to a named local and drop the duplicated
inline comment.
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
The replay handler previously used a created_at time-range filter to
target access tokens and called update_all(expires_at:), which left
revoked_at nil, skipped refresh tokens entirely, and could miss or
falsely catch tokens from concurrent flows. Add an oidc_authorization_code
FK on both token tables, carry it through refresh-token rotation, and
use the association to revoke every descendant via revoke! (which sets
revoked_at and cascades access -> refresh).
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
The authorize action opened with ~55 lines of parameter validation that
ran before any business logic. Move the two RFC 6749 §4.1.2.1 checks
(client_id lookup, redirect_uri registration) into set_application and
validate_redirect_uri before_actions. The action body now starts at the
point where errors may legitimately redirect back to the client.
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
The authorize action called validate_claims_against_scopes with
requested_scopes before that local was assigned (assignment was ~100
lines later), raising NameError whenever a client passed a claims=
parameter. Move the scope normalization above the claims validation.
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
Remove PKCE plain method support (S256 only), enforce openid scope requirement,
filter to supported scopes, strip reserved claims from custom claims as
defense-in-depth, sanitize SVG icons with Loofah, add global input padding,
switch session cookies to SameSite=Lax, use Session.active scope, and remove
unsafe-eval from CSP.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
