Eve Maler
Cell +1 425.345.6756 | Skype: xmlgrrl | Twitter: @xmlgrrl

On Thu, Jun 8, 2017 at 10:35 AM, Pedro Igor Silva <psilva@redhat.com> wrote:

On Thu, Jun 8, 2017 at 2:05 PM, Eve Maler <eve@xmlgrrl.com> wrote:

Welcome back, Pedro! A few notes:

Thanks, Eve.

Aren't many other grant types applicable to multiple client types? And if someone is interested to restrict the UMA grant to apply only to confidential clients by writing a formal profile, we have the example of the extension Assertion Framework, which gets specialized by the SAML bearer assertion profile (not that the latter actually does this particular kind of specialization...).

Yes. You just nailed down the points I've brought to this discussion. Although not mentioned explicitly in UMA 2.0, we can authenticate clients when doing UMA Grant by pushing a previously issued access token. Just wanted to make it clear and confirm we could follow this path with a bless from the expert group. As I did not find anything explicitly in the specs, I found better to ask first before coding.

It would be good to learn more about the contexts in which you're using OIDC. From time to time, we've talked about the potential benefits of a sort of "OIDC or at least UserInfo endpoint that is UMA-protected", because then the resource owner could share standardized identity data with autonomous third parties, gain positive control over that data, not have to be present when the data is requested, etc. In fact, a person in a requesting party role could potentially use this mechanism to protect the claims sought by a resource owner's AS for authorization assessment.

For sure there are a lot of benefits for resource owners in protecting UserInfo with UMA. And I don't think this is something hard to achieve if you have UserInfo and its data/claims properly represented as resources and scopes in the AS.

I know you'll get angry with me :) But right now we don't fully support UMA. Our plan is now focus on updating our stuff to UMA 2.0 and implement the missing pieces. We have very limited resources to work on this and our requirements are really driven by demands from our community and customers. But we already had people asking for more features from UMA, specially those related with user managed access/resource sharing (the honeypot).

We do not yet have an interop or testing conformance program, but the number of serious requests is increasing, and we put some effort into this in the past. Time to pick up where we left off; I will make a note to follow up.

Great, thanks for the update.

Eve Maler
Cell +1 425.345.6756 | Skype: xmlgrrl | Twitter: @xmlgrrl

On Thu, Jun 8, 2017 at 5:57 AM, Justin Richer <jricher@mit.edu> wrote:

You're conflating the grant type (and its associated redirect) with public clients. They're two separate aspects of the protocol and are not tightly tied together, though you do often find one with the other. It's not "public clients" that obtain tokens from the authorization endpoint, it's "implicit grant" clients. You can use a public client with non-implicit flows if you want to -- in fact, a public client with the authorization code flow (and PKCE) is the recommended method for native applications these days.

You don't use the implicit grant with UMA, you use the UMA grant. The UMA grant doesn't use the authorization endpoint at all. There is no "implicit UMA grant". If you want to use the UMA grant with an in-browser client, you can do so. The in-browser client just calls the token endpoint like any other client would.

I don't understand your statement about a bearer token instead of the implicit grant -- if you're doing the implicit grant today in just about every implementation, you're getting a bearer token. And to do UMA you need to do the UMA grant anyway, so the implicit grant doesn't apply here. And there aren't standardized non-bearer tokens defined yet, so you're likely doing bearer tokens whether you're doing UMA or plain OAuth anyway.

OpenID Connect is built on OAuth, not UMA. There's no definition for doing OpenID Connect on top of UMA, and I would argue that it doesn't really make sense: OpenID Connect is an identity protocol and the whole point is to delegate access to the current user's identity API, which is what OAuth does. UMA would give you access to some other party's identity API, which doesn't tell you who's logged in right now. While they're often used together, they solve very different use cases. Think of it like having a hammer, a saw, and a screwdriver in the toolbox. They're all there for different reasons but you can use them all to build your cabinet or whatever.

Hopefully this helps untangle some things.

-- Justin

On 6/8/2017 7:29 AM, Pedro Igor Silva wrote:

OK. Thanks, Justin.

OAuth2 spec does cover public clients, which are usually related with usage of a implicit grant. But in UMA 2.0 it seems we have now a specific grant type, which flow and examples are really specific to confidential clients. And more important, UMA grant type does not seem to support redirection which is something essential if you want to support public clients. Based on OAuth2 specs, public clients are able to obtain tokens from the authorization endpoint (not token endpoint) based on their client_id and redirection_uri when using implicit grant.

Like I previously mentioned, to avoid using implicit grant we would just support a bearer token when using UMA grant type (and token endpoint). Some of our use cases are related with public clients authenticating with Keycloak based on OpenID Connect where an ID token and AT is already available to them.

One last question, this one about UMA compliance. We would like to include Keycloak in the list of UMA implementations after reviewing what we have in order to conform with latest version of the spec. Is there a TCK or some other process we need to follow to achieve this ?

On Wed, Jun 7, 2017 at 10:27 PM, Justin Richer <jricher@mit.edu> wrote:

There’s nothing in the spec about public clients because it’s all covered in OAuth 2 (RFC6749 specifically).

— Justin

On Jun 7, 2017, at 12:57 PM, Pedro Igor Silva <psilva@redhat.com> wrote:

On Wed, Jun 7, 2017 at 12:21 PM, Justin Richer <jricher@mit.edu> wrote:

That is not correct. You can have a public client in UMA 2.0, and in fact you inherit all of the various OAuth client types and authentication methods since you’re talking directly to the token endpoint to get the RPT instead of a special thing. But remember, you needed to talk to the AS to get an AAT previously anyway, so you’ve already needed to have those bits in place in your client.

Really, do whatever you’d have done previously to get the AAT (which required an OAuth transaction with the token endpoint) and use it with the token endpoint to get the RPT now. So you always needed a client ID before anyway, that’s not new, and now you can use your secret or key or nothing just like you used to, but in a different part of the process.

Yes, use the AAT with the token endpoint is what I have in mind. If that is fine and still in sync with the spec for me is fine. But, is there anything in spec about public clients ?

Is it recommended that you use a confidential client where possible? Definitely, as the increased security is well worth the minimal effort that’s usually required. There are a lot of cases where it doesn’t make sense though, like a browser or native app that’s not using dynamic registration. In any event, this is all now directly inherited from OAuth instead of being reinvented from OAuth, as it was in 1.0.

That is exactly one of the use cases we need to support.

— Justin

On Jun 7, 2017, at 10:58 AM, Pedro Igor Silva <psilva@redhat.com> wrote:

Hello,

It has been a long time since my last message to this group. Since then we have implemented quite a few things in Keycloak [1] in order to support fine-grained permissions and UMA 1.0.

We are not full compliant yet and some pieces are still missing in our implementation. But we already have the backbone to start implementing now the rest of the spec.

One of my main tasks is now update our implementation to UMA 2.0 and it seems that one of the main changes is related with the removal of AAT in favor of a specific grant type for UMA.

In UMA 1.0 clients were not really forced to authenticate using client credentials when interacting with the RPT endpoint (Authorization API), but just use a AAT as a bearer token. Now, with UMA 2.0, it seems that clients really need to be confidential and use its credentials (e.g.: id/secret or jwt) to authenticate with the token endpoint when using UMA Grant.

Is that correct ?

[1] https://keycloak.gitbooks.io/documentation/authorization_services/index.html

Thanks.
Pedro Igor

_______________________________________________
WG-UMA mailing list
WG-UMA@kantarainitiative.org
http://kantarainitiative.org/mailman/listinfo/wg-uma

_______________________________________________
WG-UMA mailing list
WG-UMA@kantarainitiative.org
http://kantarainitiative.org/mailman/listinfo/wg-uma