Re: [OAUTH-WG] Fwd: New Version Notification for draft-richer-oauth-introspection-01.txt

Justin Richer <> Thu, 10 January 2013 14:18 UTC

Return-Path: <>
Received: from localhost (localhost []) by (Postfix) with ESMTP id 06FD421F862E for <>; Thu, 10 Jan 2013 06:18:36 -0800 (PST)
X-Virus-Scanned: amavisd-new at
X-Spam-Flag: NO
X-Spam-Score: -5.445
X-Spam-Status: No, score=-5.445 tagged_above=-999 required=5 tests=[AWL=-0.447, BAYES_00=-2.599, HTML_MESSAGE=0.001, J_BACKHAIR_22=1, J_CHICKENPOX_54=0.6, RCVD_IN_DNSWL_MED=-4]
Received: from ([]) by localhost ( []) (amavisd-new, port 10024) with ESMTP id 5l7+UZaH+Kqh for <>; Thu, 10 Jan 2013 06:18:34 -0800 (PST)
Received: from ( []) by (Postfix) with ESMTP id 1DF3B21F85EA for <>; Thu, 10 Jan 2013 06:18:33 -0800 (PST)
Received: from (localhost.localdomain []) by localhost (Postfix) with SMTP id BC88E1F2EED; Thu, 10 Jan 2013 09:18:32 -0500 (EST)
Received: from IMCCAS01.MITRE.ORG ( []) by (Postfix) with ESMTP id 834E51F2EBD; Thu, 10 Jan 2013 09:18:32 -0500 (EST)
Received: from [] ( by IMCCAS01.MITRE.ORG ( with Microsoft SMTP Server (TLS) id 14.2.318.4; Thu, 10 Jan 2013 09:18:32 -0500
Message-ID: <>
Date: Thu, 10 Jan 2013 09:18:28 -0500
From: Justin Richer <>
User-Agent: Mozilla/5.0 (X11; Linux i686; rv:17.0) Gecko/17.0 Thunderbird/17.0
MIME-Version: 1.0
To: George Fletcher <>
References: <> <> <> <> <> <B33BFB58CCC8BE4998958016839DE27E06873FA8@IMCMBX01.MITRE.ORG> <>
In-Reply-To: <>
Content-Type: multipart/alternative; boundary="------------060302090502040403040603"
X-Originating-IP: []
Cc: "" <>
Subject: Re: [OAUTH-WG] Fwd: New Version Notification for draft-richer-oauth-introspection-01.txt
X-Mailman-Version: 2.1.12
Precedence: list
List-Id: OAUTH WG <>
List-Unsubscribe: <>, <>
List-Archive: <>
List-Post: <>
List-Help: <>
List-Subscribe: <>, <>
X-List-Received-Date: Thu, 10 Jan 2013 14:18:36 -0000

In traditional OAuth, there really isn't a baked-in notion of 'audience' 
since the AS<->PR connection is completely out of scope. However, in 
practice, when you've got more than one PR per AS, you'll have some 
notion of 'audience'. It's definitely possible to handle this with 
'scope', especially if you want the client to have a say in the matter. 
But since you could have your scopes and audiences defined independently 
(one scope across several audiences, one audience with many scopes, and 
any other combination thereof) I think it makes sense to at least define 
a place for the AS to express this back to the PR. JWT has the exact 
same claim for the exact same reason.

As George points out below, this also really comes into play in the 
chaining case, where you've got one PR calling another PR and you need 
to keep things straight in a large backend.

So while I agree it'd be better if OAuth had an 'audience' concept all 
the way through, I don't think it should be precluded from the 
introspection response just because it doesn't.

  -- Justin

On 01/09/2013 04:47 PM, George Fletcher wrote:
> I had the same confusion about "what is 'audience' in OAuth?" today 
> working on a completely different project.
> I think for the default OAuth2 deployment, scopes take the place of 
> audience because the scopes identify the authorization grant(s) at the 
> resource servers affiliated with the Authorization Server. The client 
> can present the token to any resource server and if the necessary 
> authorization grant(s) are present, the protected resource is 
> returned. The client doesn't have to explicitly call out that it is 
> going to present the token to the 'mail service', it just needs to ask 
> for the 'readMail' scope.
> So, in regards to an AS implementation of the introspection endpoint, 
> what are the expectations for how the AS fills in the 'audience' 
> field. Should the AS not return the field if there is no audience? 
> Should the AS return "itself" as the audience? If a token has scopes 
> of 'readMail writeMail readBuddyList sendIM' then what is the correct 
> 'audience' of the token? Should it be an array of the resource servers 
> that depend on those scopes?
> I can see value in the chaining scenario of a client asking the AS for 
> a token that it will give to another party to present and storing that 
> intermediate party in the token. But for the default OAuth2 case, 
> should audience be omitted? or be the same value as 'client_id'?
> Thanks,
> George
> On 1/9/13 3:15 PM, Richer, Justin P. wrote:
>> On Jan 9, 2013, at 3:05 PM, Torsten Lodderstedt 
>> < <>> wrote:
>>> Hi Justin,
>>> Am 09.01.2013 um 20:35 schrieb Justin Richer < 
>>> <>>:
>>>> Thanks for the review, answers inline:
>>>>> why is there a need for both scope and audience? I would assume 
>>>>> the scope of the authorization request is typically turned into an 
>>>>> audience of an access token.
>>>> You can have an audience of a single server that has multiple 
>>>> scopes, or a single scope that's across multiple servers. Scope is 
>>>> an explicit construct in OAuth2, and while it is sometimes used for 
>>>> audience restriction purposes, they really are independent. Note 
>>>> that both of these are optional in the response -- if the AS has no 
>>>> notion of audience restriction in its stored token metadata, then 
>>>> it just doesn't return the "audience" field.
>>> You are making an interesting point here. To differentiate the 
>>> resource server and the permissions of a particular at this server 
>>> makes a lot of sense. BUT: the authorization request does not allow 
>>> the client to specify both in separate parameters. Instead both must 
>>> be folded into a single "scope" parameter. If I got your example 
>>> correctly, the scope of the request would be
>>> scope=myserver:read
>>> whereas the results of the introspection would be
>>> scope=read
>>> audience=myserver
>>> It's probably the different semantics of scope that confused me.
>> No, sorry if I was unclear: scope is scope, no different semantics. 
>> In this example case, you'd ask for scope=myserver:read and get back 
>> scope=myserver:read. I'm not suggesting that these be split up. Since 
>> the AS in this case knows that there's an audience, so it can return 
>> audience=myserver as well. The fact that it knows this through the 
>> scope mechanism is entirely system-dependent.
>> I agree that the lack of a method for specifying audience does make 
>> returning this field a little odd for simple OAuth deployments, but 
>> since audience restriction is a big part of clustered and enterprise 
>> deployments (in my personal experience), then it's something very 
>> useful to have the server return.
>>>>> Generally, wouldn't it be simpler (spec-wise) to just return a JWT 
>>>>> instead of inventing another set of JSON elements?
>>>> What would be the utility in returning a JWT? The RS/client making 
>>>> the call isn't going to take these results and present them 
>>>> elsewhere, so I don't want to give the impression that it's a 
>>>> token. (This, incidentally, is one of the main problems I have with 
>>>> the Ping introspection approach, which uses the Token Endpoint and 
>>>> invents a "token type" as its return value.) Also, the resource 
>>>> server would have to parse the JWT instead of raw JSON, the latter 
>>>> of which is easier and far more common. Besides, I'd have to invent 
>>>> new claims for things like "valid" and "scopes" and what not, so 
>>>> I'd be extending JWT anyway.
>>>> So while I think it's far preferable to use an actual JSON object, 
>>>> I'd be fine with re-using JWT claim names in the response if people 
>>>> prefer that. I tried to just use the expanded text since size 
>>>> constraints are not an issue outside of a JWT, so "issued_at" 
>>>> instead of "iat".
>>>> Finally, note that this is *not* the same as the old OIDC CheckId 
>>>> endpoint which merely parsed and unwrapped the data inside the 
>>>> token itself. This mechanism works just as well with an 
>>>> unstructured token as input since the AS can store all of the 
>>>> token's metadata, like expiration, separately and use the token's 
>>>> value as a lookup key.
>>> I probably didn't describe well what I meant. I would suggest to 
>>> return a JWT claim set from the introspection endpoint. That way one 
>>> could use the same claims (e.g. iat instead of issued_at) for 
>>> structured and handle-based tokens. So the logic operating on the 
>>> token data could be the same.
>> OK, I follow you now. I'd be fine with re-using the JWT claim names 
>> and extending the namespace with the OAuth-specific parameters, like 
>> scope, that make sense here.
>>  -- Justin
>>> regards,
>>> Torsten.
>>>>  -- Justin
>>>>> Am 09.01.2013 um 20:10 schrieb Justin Richer < 
>>>>> <>>:
>>>>>> Updated the introspection draft with feedback from the UMA WG, 
>>>>>> who have incorporated it into their latest revision of UMA.
>>>>>> I would like this document to become a working group item.
>>>>>>  -- Justin
>>>>>> -------- Original Message --------
>>>>>> Subject: 	New Version Notification for 
>>>>>> draft-richer-oauth-introspection-01.txt
>>>>>> Date: 	Tue, 8 Jan 2013 14:48:47 -0800
>>>>>> From: 	<>
>>>>>> To: 	<>
>>>>>> A new version of I-D, draft-richer-oauth-introspection-01.txt
>>>>>> has been successfully submitted by Justin Richer and posted to the
>>>>>> IETF repository.
>>>>>> Filename:	 draft-richer-oauth-introspection
>>>>>> Revision:	 01
>>>>>> Title:		 OAuth Token Introspection
>>>>>> Creation date:	 2013-01-08
>>>>>> WG ID:		 Individual Submission
>>>>>> Number of pages: 6
>>>>>> URL:
>>>>>> Status:
>>>>>> Htmlized:
>>>>>> Diff:
>>>>>> Abstract:
>>>>>>     This specification defines a method for a client or protected
>>>>>>     resource to query an OAuth authorization server to determine meta-
>>>>>>     information about an OAuth token.
>>>>>> The IETF Secretariat
>>>>>> _______________________________________________
>>>>>> OAuth mailing list
>>>>>> <>
>> _______________________________________________
>> OAuth mailing list