Re: [OAUTH-WG] [JAR] scope parameter outside request object of OIDC request

[Takahiko Kawasaki <[e-mail]>]

"there seems to be a good consensus that servers must not require the
duplicates."

from FAPI Issue 315: PAR certification question - must servers allow
requests where scope/response_type only passed to PAR endpoint
https://bitbucket.org/openid/fapi/issues/315/par-certification-question-must-servers

A consensus has been reached which is enough at least for the conformance
suite. It is the same as my suggestion. Thank you for discussion.

Taka


On Wed, Sep 30, 2020 at 2:21 PM Takahiko Kawasaki <[e-mail]> wrote:

> So, my suggestion is "When JAR compatible behaviors are employed, AS
> implementations should not require that an OIDC request come with `scope`
> query/form parameter when the request uses a request object." (NOTE: "an
> OIDC request" implies that the request object contains `scope` including
> `openid`.)
>
> Any thoughts?
>
> Taka
>
> On Thu, Sep 24, 2020 at 11:07 PM Takahiko Kawasaki <[e-mail]>
> wrote:
>
>> Hi Vladimir,
>>
>> Just FYI. To be exact, FAPI (version 1) Part 1 (Read-Only) does not
>> require all request parameters be put duplicately in a request object. It
>> is FAPI (version 1) Part 2 (Read-Write) (Section 5.2.2
>> <https://openid.net/specs/openid-financial-api-part-2-ID2.html#authorization-server>
>> Clause 10) that has the requirement. In the context of FAPI Part 1, a
>> request object does not have to be used. One more note is that parameters
>> inside a request object and parameters outside a request object are merged
>> (as they are in OIDC Core 1.0) when the authorization request is being made
>> for FAPI Read-Only APIs (not for FAPI Read-Write APIs).
>>
>> Taka
>>
>>
>> On Thu, Sep 24, 2020 at 7:14 PM Vladimir Dzhuvinov <
>> [e-mail]> wrote:
>>
>>> Hi Taka,
>>>
>>> Speaking of the OIDC Core 1.0 conformance tests, IMO those should not
>>> change with the publication of JAR.
>>>
>>> Speaking of the FAPI 1.0 tests, those already require all request
>>> parameters to be JWT-secured, which makes the requests also JAR compliant:
>>> all parameters are found in the JWT, with scope (as complete scope or
>>> minimally required scope=openid), response_type, client_id and redirect_uri
>>> also having a copy outside the JWT, as query parameters. Thus the request
>>> is OIDC as well as JAR compliant.
>>>
>>> If I had an RP I would always construct OIDC auth requests like that, to
>>> make sure they comply with OIDC as well as the new JAR spec (and will not
>>> have issues with servers which implement both specs but are not able to
>>> "switch" behavior for some reason).
>>>
>>> Vladimir
>>> On 23/09/2020 14:58, Takahiko Kawasaki wrote:
>>>
>>> Hi Vladimir,
>>>
>>> Thank you for your reply. It sounds that your opinion is "`scope`
>>> request parameter must exist outside the request object even if JAR applies
>>> if the authorization request is an OIDC request". I'm on the fence on this
>>> topic and just wondered whether those who had wanted to remove
>>> `response_type` outside the request object (although doing it was a
>>> breaking change) would want to remove `scope` outside the request object
>>> too with the same motivation (although I don't remember well what was the
>>> motivation). JAR dares to drop `response_type`, so it would not be
>>> surprising to see that JAR dares to drop `scope` (including `openid`) too.
>>>
>>> OIDC Core 1.0 requires `response_type`, but JAR allows omission of the
>>> parameter if the parameter is included in the request object.
>>>
>>> If we applied the same logic, we would be able to state:
>>>
>>> OIDC Core 1.0 requires `scope` (including `openid`), but JAR allows
>>> omission of the parameter if the parameter is included in the request
>>> object.
>>>
>>> In terms of `response_type`, practically speaking, JAR has modified OIDC
>>> Core 1.0. Because JAR has already been allowed to go so far as that point,
>>> I would say it is difficult to find a convincing reason not to allow
>>> omission of `scope`.
>>>
>>> AFAIK, in the context of OIDC Core 1.0, parameters that are required to
>>> exist outside a request object even if they are included in the request
>>> object are `client_id`, `response_type` and `scope`. Because `client_id` is
>>> mandatory in JAR (it has become mandatory after long discussion),
>>> discussion for the parameter is not needed. Because the community has
>>> already reached consensus that `response_type` can be omitted, discussion
>>> for the parameter is not needed, either. What I've brought here is
>>> discussion for `scope`, hopefully the last parameter that is affected by
>>> JAR.
>>>
>>> Again, I'm on the fence on this topic. However, because logical
>>> conclusion (at least of mine) is that JAR should allow omission of `scope`
>>> (it also should be noted that JAR's basic rule prohibits referring to
>>> request parameters outside a request object), I want to see explicit
>>> consensus if `scope` (including `openid`) outside a request object is still
>>> required even after JAR is enabled.
>>>
>>> In short, my question is "Should `scope` be omitted?" I guess that the
>>> conclusion will affect the official conformance suite.
>>>
>>> Best Regards,
>>> Takahiko Kawasaki
>>> Authlete, Inc.
>>>
>>>
>>>
>>> On Tue, Sep 22, 2020 at 5:59 AM Vladimir Dzhuvinov <
>>> [e-mail]> wrote:
>>>
>>>> Hi Taka,
>>>> On 21/09/2020 20:12, Takahiko Kawasaki wrote:
>>>>
>>>> If we allow JAR (JWT Secured Authorization Request) to relax the
>>>> requirement of `response_type` request parameter (outside a request object)
>>>> from mandatory to optional, should we relax the following requirement of
>>>> `scope` request parameter stated in OIDC Core 1.0 Section 6.1, too?
>>>>
>>>> ----------
>>>> Even if a scope parameter is present in the Request Object value, a
>>>> scope parameter MUST always be passed using the OAuth 2.0 request syntax
>>>> containing the openid scope value to indicate to the underlying OAuth 2.0
>>>> logic that this is an OpenID Connect request.
>>>> ----------
>>>>
>>>> Otherwise, an authorization request like
>>>> "client_id=...&request(_uri)=..." fails if the request object represents an
>>>> OIDC request. An authorization request has to look like
>>>> "client_id=...&request(_uri)=...&scope=openid" (`scope` including `openid`
>>>> has to be given) even if the authorization server conforms to JAR and
>>>> allows omission of `response_type` request parameter.
>>>>
>>>> The bottom of section 5 has normative text which allows a JAR compliant
>>>> server to also comply with the OIDC spec with its own style of request /
>>>> request_uri parameter handling insofar as to not reject other query params
>>>> (such as scope, etc). The difference is that according to JAR their values
>>>> cannot be used or merged (as in OIDC). But what can be reasonably done is
>>>> to detect scope=openid as you say and then switch to OIDC style request
>>>> object behavior.
>>>>
>>>> https://tools.ietf.org/html/draft-ietf-oauth-jwsreq-30#section-5
>>>>
>>>>    The client MAY send the parameters included in the request object
>>>>    duplicated in the query parameters as well for the backward
>>>>    compatibility etc.  However, the authorization server supporting this
>>>>    specification MUST only use the parameters included in the request
>>>>    object.
>>>>
>>>> The confusion between the two specs clears when it's seen that the
>>>> request objects in OIDC and JAR have different objectives.
>>>>
>>>> In OIDC the objective is to enable securing of selected parameters.
>>>>
>>>> In JAR the objective is to secure the entire authz request.
>>>>
>>>>
>>>>
>>>> I think that implementers want to know consensus on this because it
>>>> affects implementations. Has this been discussed yet?
>>>>
>>>> Best Regards,
>>>> Takahiko Kawasaki
>>>> Authlete, Inc.
>>>>
>>>>
>>>> Vladimir
>>>>
>>>
>>> _______________________________________________
>>> OAuth mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/oauth
>>>
>>