Re: [OAUTH-WG] draft-ietf-oauth-pop-key-distribution-01 and Open Issues

[John Bradley <[e-mail]>]

For signing authentication requests you include the keyid in the JWT, and the AS looks in the JWKS to find the correct key if there is more than one.

I don't think that is a problem

What we probably need to do is pass a keyid in the request if there is more than one signing key registered for the client.

John B.

> On Mar 5, 2015, at 12:34 PM, Hannes Tschofenig <[e-mail]> wrote:
> 
> Hi John,
> 
> 
> On 03/05/2015 10:27 AM, John Bradley wrote:
>> inline
>>> On Mar 5, 2015, at 9:59 AM, Hannes Tschofenig <[e-mail]> wrote:
>>> 
>>> Hi all,
>>> 
>>> I refreshed the PoP key distribution document. No changes to the
>>> content of the document.
>>> 
>>> The document contains two questions, namely
>>> 
>>> QUESTION: A benefit of asymmetric cryptography is to allow clients to
>>>  request a PoP token for use with multiple resource servers.  The
>>>  downside of that approach is linkability since different resource
>>>  servers will be able to link individual requests to the same client.
>>>  (The same is true if the a single public key is linked with PoP
>>>  tokens used with different resource servers.)  Nevertheless, to
>>>  support the functionality the audience parameter could carry an array
>>>  of values.  Is this desirable?
>>> 
>>> 
>>> Hannes: My view is that we do not want to introduce likability into
>>> OAuth via the use of these keys. As such, different keys for different
>>> origins.
>>> 
>>> 
>> John:  Having an array increases complexity and decreases privacy by allowing RS to link.
>> 
>> Audiance should be a single value.  That requires separate keys for symmetric, or asymmetric keys provisioned by the AS.
>> 
>> For asymmetric keys provisioned by the client it would be up to the client to decide if using the same key for multiple RS makes sense.  
>> 
>> It might be that there is a single key provisioned by a MSM in the tpm that they want to use for all the connections as that is the most secure, and are not concerned with correlation as all the RS are internal to a single enterprise.
>> 
> 
> OK.
> 
>> 
>> 
>>> QUESTION: Should we register the token_type and alg parameters for use
>>> with the dynamic client registration protocol?
>>> 
>>> Hannes: I believe we should register these two parameters into the
>>> dynamic client registration protocol since that allows us to configure
>>> the values for the client rather than exchanging them with every message.
>>> 
>> John:  Yes I had assumed that that was a no brainer.
>> The other question on registering is if we should allow the client to preregister a public key as well.
>> In some situations the client may want to always use the same key and making them push it each time is a waste of bandwidth.
> The dynamic client registration already supports the feature of
> uploading a public key to the authorization server and hence the missing
> feature to support that case is to allow the client to refer to that key
> somehow. I assume that the public key uploaded by the dynamic client
> registration protocol is referenced using a URI. The question is just
> what that URI would be.
> 
> Unfortunately, the Dynamic Client Registration spec is silent about how
> to reference the public key uploaded via the jwks parameter. Did we just
> found a bug in the spec?
> 
> Ciao
> Hannes
> 
> 
>> 
>> John B.
>>> Feedback appreciated before the submission deadline.
>>> 
>>> Ciao
>>> Hannes
>>> 
>>> _______________________________________________
>>> OAuth mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/oauth