Belatedly... Sorry if I sound like a broken record on this, but: Most of UMA's
use involve letting a user introduce their various hosts (UMA-flavored resource
servers) to their single chosen "authorization manager" (UMA-flavored
authorization server), by treating the former as a dynamically introduced OAuth
client of the latter. This sounds an awful lot like the question originally
posed in this thread. We have exactly the problem of figuring out how the host
can tell the AM what resources the AM should be protecting and what can be done
to them, which we've begun to solve with what we're calling a "resource
registration API" (RRAPI).
(BTW, we're also working on an I-D to submit here that proposes a solution for
discovery/dynamic registration that meets our needs. Hoping it can feed into
Eran's discovery work.)
Eve
On 28 Jul 2010, at 1:23 AM, Torsten Lodderstedt wrote:
> thanks for sharing your thoughts.
>
> Differentiating resource servers by using different end-user authorization
> endpoint URLs is an option. I dont't know how this will work in conjunction
> with discovery, especially since this differentiation might by required on
> other endpoints, too. For example, if a client wants to obtain an access
> token for the end-user's credentials, it has to identify the resource server
> also on the tokens endpoint. There might be additional endpoint for other
> flows with similar requirements, e.g. the device flow.
>
> Based on your proposal, a derived solution could be to define a standard
> parameter "resourceserver" and to state how clients should use this parameter
> on the different endpoints. On the coding level, there would be no difference
> to your proposal :-) But the client would be able to construct such a URL on
> its own.
>
> Authorizing access for different resource servers is indeed an issue for me.
> How would you propose to add the namespace? Shall the scope obtained from the
> resource server already contain such a namespace are shall there be a rule to
> construct such namespaced-ed scopes in the spec?
>
> regards,
> Torsten.
>
> Am 25.07.2010 19:11, schrieb Andrew Arnott:
>>
>> It seems to me that if one auth server can create tokens for a diverse set
>> of resource servers, then why not have different user authorization endpoint
>> URLs for each type of resource server, as an added differentiator for the
>> scope (a namespace, if you will)?
>>
>> So suppose one auth server serves two different photo services, both using
>> overlapping scopes such that a client requesting access of some scope is
>> otherwise ambiguous between which resource server the client needs access
>> to. The auth server that serves both resource servers could have two end
>> user authorization endpoints:
>> http://auth.server.org/authorize?resourceserver=1
>> http://auth.server.org/authorize?resourceserver=2
>>
>> And that way the auth server knows exactly what the client needs.
>>
>> The only scenario this doesn't allow for is for a user to authorize a
>> client's access to both resource servers in one redirect. If this were an
>> issue, perhaps you can apply a namespace-like prefix to each scope substring:
>>
>> rs1:photo:read rs2:photo:read
>>
>> Which would serve a similar purpose.
>>
>> --
>> Andrew Arnott
>> "I [may] not agree with what you have to say, but I'll defend to the death
>> your right to say it." - S. G. Tallentyre
>>
>>
>> On Thu, Jul 22, 2010 at 3:07 PM, Torsten Lodderstedt
>> <tors...@lodderstedt.net> wrote:
>> no one else in the group having an opinion on this topic?
>>
>>
>>
>> Am 15.07.2010 20:14, schrieb Marius Scurtescu:
>> On Thu, Jul 15, 2010 at 10:03 AM, Torsten Lodderstedt
>> <tors...@lodderstedt.net> wrote:
>> As I have written in my reply to Marius's posting. I'm fine with including
>> server ids in scopes. But this requires a definition of the scope's syntax
>> and semantics in the spec. Otherwise, scope interpretation (and server
>> identification) will be deployment specific.
>> Sure, it is deployment specific, but why is that an issue?
>>
>> In your case, the authz server and all the resource servers are
>> managed by the same organization, right?
>>
>> Do clients need to be aware of the actual resource server?
>>
>> You can probably create a separate spec that defines scope syntax for
>> this purpose, if really needed. Does it have to be in core?
>>
>> Marius
>>
>> Solving the challenge I described in a deployment specific way is not an
>> issue. But the consequence is that authz server, resource servers and
>> clients are tight together.
>>
>> Let me ask you one question: Why are we working together towards a standard
>> protocol? I can tell you my expectations: I hope there will be broad support
>> not only by libraries, but also by ready-to-use services and clients, so we
>> could integrate such services into our deployment easily. Moreover, I would
>> like to see OAuth to be included in application/service protocols like
>> PortableContacts, SIP, WebDAV, IMAP, ...
>>
>> So what if I would like to use standard clients to access our services?
>> Using scopes for specifying resource server id's in this case is also simple
>> - if you take an isolated view. But since scopes may be used to specifiy a
>> lot of other things, like resources, permissions, and durations, handling
>> w/o a more detailed spec will in practice be impossible.
>>
>> Suppose a WebDAV service for media data access. Any WebDAV client knows the
>> WebDAV protocol (== interface), e.g. the supported methods (GET, PUT, POST,
>> DELETE, COPY, MOVE) and how to traverse directories. So it is sufficient to
>> configure the client with the URL of my personal web storage. To start with
>> let's assume, scopes are used to designate resource servers only. So the
>> server's scope could be "webstorage".
>>
>> WWW-Authenticate OAuth realm='webstorage' scope="webstorage"
>>
>> The client could just pass this parameter to the authz server and everything
>> is fine.
>>
>> On the next level, let's assume the (future) WebDAV standard with
>> OAuth-support uses one permission per method type. So the full scope could
>> be as follows:
>>
>> WWW-Authenticate OAuth realm='webstorage' scope="webstorage:GET
>> webstorage:PUT webstorage:POST webstorage:DELETE webstorage:COPY
>> webstorage:MOVE"
>>
>> Passing this scope w/o any unmodified to the authz server is not an issue.
>> But this implies the client asks for full access to the users media storage.
>> Since our client is a gallery application, it requires the "GET" permission
>> only. How does the client know which of the scope values to pick for the
>> end-user authorization process? It must somehow select "webstorage:GET".
>>
>> But how?
>>
>> In my personal opinion, clients should be enabled to interpret, combine and
>> even create scopes. And yes, this should go to the core of the spec.
>>
>> regards,
>> Torsten.
>>
>>
>>
>>
>> _______________________________________________
>> OAuth mailing list
>> OAuth@ietf.org
>> https://www.ietf.org/mailman/listinfo/oauth
>>
>> _______________________________________________
>> OAuth mailing list
>> OAuth@ietf.org
>> https://www.ietf.org/mailman/listinfo/oauth
>>
> _______________________________________________
> OAuth mailing list
> OAuth@ietf.org
> https://www.ietf.org/mailman/listinfo/oauth
Eve Maler
http://www.xmlgrrl.com/blog
http://www.twitter.com/xmlgrrl
http://www.linkedin.com/in/evemaler
_______________________________________________
OAuth mailing list
OAuth@ietf.org
https://www.ietf.org/mailman/listinfo/oauth
