# Which Problems Are Solved
#12313 lets clients register themselves as OIDC applications (RFC 7591),
but a dynamically registered client has no way to manage itself
afterwards:
- The registration response returns no `registration_access_token` and
no `registration_client_uri`, which RFC 7591 and 7592 clients (including
the MCP SDKs) expect in order to manage their own registration.
- There is no way to read back, update or delete a registration without
an operator acting on the client's behalf through the Console or the
Management API. A client that needs to rotate a redirect URI has to ask
a human.
# How the Problems Are Solved
Implement the [OAuth 2.0 Dynamic Client Registration Management Protocol
(RFC 7592)](https://datatracker.ietf.org/doc/html/rfc7592): `GET`, `PUT`
and `DELETE` on `/oauth/v2/register/{client_id}`, wired through the same
`op.WithSetRouter` hook as the registration endpoint.
## Control plane: the same security settings, no new switch
The management endpoints are gated on the
`DynamicClientRegistrationSettings.enabled` introduced in #12313. No new
setting, no feature flag: a client that may register may manage what it
registered. When registration is disabled the endpoints answer `404`,
exactly as `POST /oauth/v2/register` does.
`allow_unauthenticated` selects the registration mode and does not apply
here, because the registration access token authorizes these endpoints
in both modes.
## Authorization: the registration access token, and nothing else
RFC 7592 §3 defines the credential as the registration access token
issued at registration and bound to a single client, not a user
credential. So a client manages only itself, proving it with the token
it was handed:
| | Registration (`POST`) | Management (`GET` / `PUT` / `DELETE`) |
| --- | --- | --- |
| Credential | Access token, or none in open mode | Registration access
token |
| Permission | `project.app.register_dynamic` in token mode | None |
Neither `project.app.register_dynamic`, which governs who may *create*
clients, nor `project.app.write` applies to the management endpoints.
One authorization path per endpoint is easier to reason about than two,
and there is no user identity to check against in the first place.
Operators keep managing dynamically registered clients through the
Management API and the Console, like any other application; they cannot
obtain a client's registration access token, by design.
Status codes: a missing, malformed or non matching token is `401`; a
client that no longer exists is `404`. The token's `client_id` is
checked against the path before any lookup, so a token bound to another
client is rejected without touching the database.
## The token itself
- **Stored like a client secret.** Only its passwap hash is persisted,
through a new
`project.application.config.oidc.registration_token.changed` event; the
plain token leaves the server exactly once, in the response that issues
it.
- **Bound to its client and organization** through authenticated
encryption with the instance key, the same mechanism as refresh tokens.
The opaque token handed to the client encrypts
`client_id:org_id:secret`, so the endpoint recovers the binding before
checking the secret against the stored hash.
- **Verified in O(1)** against the hash on the projection, folded into
the client lookup the handler already does, with a strongly consistent
fallback to the eventstore so a token that was just rotated is accepted
before it is projected.
- **Rotated on update** (RFC 7592 §4.4) and invalidated implicitly when
the client is deleted. A read does not rotate it. Every update rotates,
including one that changes no metadata, so a client always gets a usable
token back.
## Reusing the existing persistence
A small refactor extracts the shared OIDC application creation and
update logic (`pushOIDCApplication` gains a variadic `extraEvents`, and
`oidcApplicationChangeEvent` is lifted out of `UpdateOIDCApplication`),
so the dynamic commands reuse it without the permission check. The
registration token event is pushed atomically with the application, so a
registered client is never left unmanageable.
# Additional Changes
- Setup step **74** adds the `registration_token` column to the
`apps7_oidc_configs` projection, with a reducer for the new event
modelled on the client secret. Applications are not part of the
relational storage projection, so there is no second write path to keep
in sync.
- The registration response (`POST`) now also carries
`registration_access_token` and `registration_client_uri`.
- Docs: a "Manage a registration" section in the dynamic client
registration guide covering the three operations, the authorization
model, token rotation and the new limitations, plus the management
endpoints on the OIDC endpoints page.
- Unit tests for the command layer (issue, verify, rotate, update,
delete), the projection reducer, and the token binding and response
mapping. An integration test covers the full lifecycle (register, read,
update with rotation, delete, read again with `404`), a token bound to
another client, and the endpoints not being served when registration is
disabled. None of the integration calls carries a user access token,
which pins down that the management endpoints require no permission of
their own.
# Additional Context
- Part of the MCP authorization effort in #9810. Follow-up to #12313
(RFC 7591), which is now merged; this branch has been rebased on `main`
and carries the settings and authorization model agreed there.
- **Open question for the review**: the authorization model above
(registration access token only, no operator path through these
endpoints) is the one I proposed when #12313 was reviewed, but it was
never explicitly confirmed. If you would rather have `project.app.write`
also grant access to a client in the operator's own organization, that
is a contained change in `authorizeClientManagement`.
- Rotation is eventually consistent: right after an update the previous
token may keep working for a brief moment until the rotation is
projected, while the new token works immediately through the eventstore
fallback. This matches how ZITADEL validates access tokens, where
revocation is likewise eventually consistent. It is documented in the
guide and asserted as such in the integration test.
- A database migration is introduced for the new projection column. No
proto change and no new feature flag.
- Verified locally: full build, `go vet` including `-tags integration`,
`golangci-lint` with 0 issues, and the unit suites for every touched
package. As before, I cannot run the integration suite here, so CI
remains the check on it.
# Which Problems Are Solved
Memberships did not have a fields table fill migration.
# How the Problems Are Solved
Add filling of membership fields to the repeatable steps.
# Additional Changes
- Use the same repeatable step for multiple fill fields handlers.
- Fix an error for PostgreSQL 15 where a subquery in a `FROM` clause
needs an alias ing the `permitted_orgs` function.
# Additional Context
- Part of https://github.com/zitadel/zitadel/issues/9188
- Introduced in https://github.com/zitadel/zitadel/pull/9152
# Which Problems Are Solved
Do not push secret succeeded and failed events for API and OIDC clients
on the introspection endpoint.
On instances where introspection was fequently called, the pushed events
created issues on duplicate primary keys, due to collisions on the
`sequence` column in the eventstore. As the event pusher retries on this
collision and we pushed above mentioned events async, it would create a
backpressure of concurrent pushers and effectively cripple an instance.
We considered that pushing these events have little value with regards
to the audit trail, as we do not push similar events when client
assertion is used. Also, before #7657 the events were defined, but not
pushed.
# How the Problems Are Solved
- Removed API secret check succeeded and faild event definitions
- Removed OIDC secret check succeeded and faild event definitions
- Push only Hash Updated event when needed
# Additional Changes
- None
# Additional Context
- Fixes https://github.com/zitadel/zitadel/issues/8479
- Closes https://github.com/zitadel/zitadel/issues/8430
- Intoduced in https://github.com/zitadel/zitadel/pull/7657
Even though this is a feature it's released as fix so that we can back port to earlier revisions.
As reported by multiple users startup of ZITADEL after leaded to downtime and worst case rollbacks to the previously deployed version.
The problem starts rising when there are too many events to process after the start of ZITADEL. The root cause are changes on projections (database tables) which must be recomputed. This PR solves this problem by adding a new step to the setup phase which prefills the projections. The step can be enabled by adding the `--init-projections`-flag to `setup`, `start-from-init` and `start-from-setup`. Setting this flag results in potentially longer duration of the setup phase but reduces the risk of the problems mentioned in the paragraph above.
* docs: update cockroachdb version to 22.2
* feat(adminAPI): ListEventTypes returns the list of event types ZITADEL implements
* feat(adminAPI): ListAggregateTypes returns the list of aggregate types ZITADEL implements
* feat(adminAPI): ListEvents allows `IAM_OWNERS` to search for events
jmaretteandGitHub
Tim MöhlmannandGitHub
SilvanandGitHub
Stefan BenzandGitHub
