On 30 June 2016 at 16:25, Marek Posolda <mposolda@redhat.com> wrote:

I am adding some OIDC specs JIRAs with possibility how to fix them. I am including those, which will be easy to fix and I can look into them later today or tomorrow before PTO :

https://issues.jboss.org/browse/KEYCLOAK-3189 - Add 'typ' to JWT header

https://issues.jboss.org/browse/KEYCLOAK-3190 - Add 'kid' to JWT header

https://issues.jboss.org/browse/KEYCLOAK-3217 - UserInfo endpoint not accessible by POST request secured with Bearer header

https://issues.jboss.org/browse/KEYCLOAK-3147 - OpenID Connect auth request redirect_uri behaviour not according to spec

+1 To just require redirect_uri in either case

We need separate JIRAs for scope=openid and another one for removal of ID token when scope=openid isn't used

https://issues.jboss.org/browse/KEYCLOAK-3222 - WellKnown endpoint doesn't return supported types of client authentication

https://issues.jboss.org/browse/KEYCLOAK-3219 - WellKnown endpoint doesn't support claims_supported

Maybe this one isn't so straightforward due to protocol mappers?

All of those are quite straightforward and easy to fix IMO.

Besides that, there are those 2, which I first rather want to confirm what exactly to do:

- https://issues.jboss.org/browse/KEYCLOAK-3221 Tokens not invalidated if an attempt to reuse code is made

We have just single-use code (which is good), however OAuth2 specs recommends to invalidate existing tokens if an attempt to reuse code is done. And one OIDC test is in WARNING state because of it (it tries to access UserInfo endpoint with the accessToken issued with the reused code).

I can see 2 possibilities to fix:
a) Invalidate just single clientSession where "code" attempt reuse was made
b) Logout whole userSession

It looks to me that (a) is sufficient solution. The potential issue with (b) is, that if attacker can steal code, it gives him the possibility to trigger global logout of user from all apps. WDYT?

I think option (a) is the correct approach. I interpret the spec to state that tokens associated with the specific code should be invalidated. Option (b) seems a bit over the top and not necessary I also wonder if there's non-malicious situations where an application could try to exchange a code twice, for example after a timeout or an automated connection retry?

- https://issues.jboss.org/browse/KEYCLOAK-3218 Support for "max_age" in AuthorizationEndpoint and "auth_time" claim on IDToken

The possibility to implement is :
- Add new note AUTH_TIME to UserSessionModel. It will be time when authentication of user was fully finished (including requiredActions). Session note is used just so we don't need to change the model ;)
- If "max_age" parameter was requested, the "auth_time" will be added to IDToken (or I will re-check specs if we should rather always add it to IDToken)
- I am also thinking about adding hook to CookieAuthenticator, so that if max_age parameter was used and userSession authTime is too "old", the CookieAuthenticator will be ignored and user will need to re-authenticate with other authenticators (username/password form etc). Then authTime will be updated on userSession once authentication is finished.

+1 Sounds good

WDYT?

That will leave us with bigger things for OIDC Basic certification ( scope parameter support, possibly 'claims' param support and 'acr' support).

Marek

_______________________________________________
keycloak-dev mailing list
keycloak-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/keycloak-dev