Re: [OAUTH-WG] OAuth Discovery spec pared down to its essence

Step 0 is  something we need to discuss. 

An OAuth protected resource can be hacked if a client uses an evil discovery and it returns a proxied resource server.  Just as a token server can be proxied, so can a resource server.  So the OAuth server may be issuing tokens in a secure way, but then it gives it to a client that uses them at the wrong endpoint.

I believe the resource and the As must be bound together in a security relationship. The security descriptor needs to reflect this.

I am ok with separate app data discovery.  But from the oauth security perspective, the oauth server needs to indicate that it is the correct server to use for Rx.

There are probably multiple ways to do this, but I think the discovery draft has only addressed 2 of 3 endpoints.  We need all of them bound together.

@independentid
www.independentid.com <http://www.independentid.com/>phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>

> On Feb 18, 2016, at 11:30 PM, Nat Sakimura <n-sakimura@nri.co.jp> wrote:
> 
> 
> 
> 2016年2月19日(金) 14:58 Phil Hunt (IDM) <phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>>:
> Inline...
> 
> Phil
> 
> On Feb 18, 2016, at 22:19, Nat Sakimura <n-sakimura@nri.co.jp <mailto:n-sakimura@nri.co.jp>> wrote:
> 
>> Thanks for the explanation. Let me re-formulate.  <>
>>  
>> 
>> Assumption
>> 
> 
> 
>> 1.     There are resource server – authorization server pairs: R1A1 … RnAn. 
>> 
>> 2.     There are clients C1 … Cm. 
>> 
>> 3.     These instances can be hosted on a multi-tenancy environment. 
>> 
>>  
>> 
>> Flow
>> 
> 
> Step 0. The client discovers the resource server endpoint and its configs. 
> 
> Sure, but that's out of scope, is it not? If I am not mistaken, we are talking about OAuth server configuration discovery. I do agree that resource discovery in itself is a very interesting and important work which I would love to work on, but that is not what we are talking about here, I think.  
>  
> Question: should that include oauth?? John seems to imply yes. 
> 
> I do not think so. I think he was referring to the configuration of Ay that corresponds to Ry. John?  
>> 1.     Client Cx goes to a resource server Ry, but he was denied of the access and was told to get an access token. Now Cx needs to know where to go. 
>> 
>> 2.     Cx uses << Discovery>> to find the OAuth endpoints and the associated metadata on Ay that corresponds to Ry. 
>> 
> 
> Where is the discovery for Ay?
> I was just trying to understand your use case. I have not put any solution into it. That's why I marked discovery as <<Discovery>>.  
> 
> The default one that the draft suggest is the https://authority <https://authority/>(uri(Ay))/.well-known/openid-configuration. 
> What John is suggesting is that we could also have https://authority(uri(Ay))/.well-known/oauth-config-for-Ry <https://authority(uri(Ay))/.well-known/oauth-config-for-Ry>, if I read his message correctly. 
> 
> 
>> 3.     Cx goes and fetches the Discovery file. 
>> 
>> 4.     Cx goes to Ay to get authorized using the config info in the Discovery file and the rest is normal RFC6749. 
>> 
> 
> Referring to the risk that a client discovered from a bad source (eg to insert a fake token endpoint), how does Ay know what discovery the client used was correct?  This might not be a problem in reality but it needs to work. 
> 
> Right. We are currently asking the clients to be sane enough to get the configuration for the server directly over https at the authorization server, more or less. When Ay gets an authorization request, it has no way of knowing that the client got the config from the right endpoint. Are you suggesting something like have the client send the signature of the config file that it is using now? 
>>  
>> 
>> Is this correct? 
>> 
>>  
>> 
>> Nat
>> 
>>  
>> 
>> --
>> 
>> PLEASE READ :This e-mail is confidential and intended for the
>> 
>> named recipient only. If you are not an intended recipient,
>> 
>> please notify the sender  and delete this e-mail.
>> 
>>  
>> 
>> From: Phil Hunt (IDM) [mailto:phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>] 
>> Sent: Friday, February 19, 2016 1:58 PM
>> To: Nat Sakimura
>> Cc: Mike Jones; oauth@ietf.org <mailto:oauth@ietf.org>
>> Subject: Re: [OAUTH-WG] OAuth Discovery spec pared down to its essence
>> 
>>  
>> 
>> No. Much simpler. 
>> 
>>  
>> 
>> A service provider has decided to have a separate oauth server for each web 'property'. This could be because they were acquired separately and run different infrastructures. Or the business structure keeps each BU completely separate. 
>> 
>>  
>> 
>> The client can't really depend on previously known or hard coded endpoints because there are 1000s of instances deployed (eg as in tenancies). 
>> 
>>  
>> 
>> This dynamic discovery is going to be particularly true of open source software that customers choose to host on PaaS cloud providers of their choosing. 
>> 
>> 
>> Phil
>> 
>> 
>> On Feb 18, 2016, at 19:04, Nat Sakimura <n-sakimura@nri.co.jp <mailto:n-sakimura@nri.co.jp>> wrote:
>> 
>> Hi Phil, 
>> 
>>  
>> 
>> You wrote: 
>> 
>> > If example.com <http://example.com/> had separate oauth servers for services xyz and abc, 
>> 
>> > how would discovery work from a single /.well-known endpoint?
>> 
>>  
>> 
>> I am trying to understand your use case, but I am not sure if I do. 
>> 
>>  
>> 
>> The use case seems to be such that: 
>> 
>>  
>> 
>> -       There is a client C1. It could be a CRM or any kind of application that uses RFC6749 and RFC6750 to access other resources a resource server R1. C1 and R1 has a pre-configured relationship. 
>> 
>> -       The resource server R1 supports RFC6750, and can have multiple OAuth RFC6749 endpoints that it supports, which are A1, …, An. 
>> 
>> -       Ax supports multiple resource services, Rx. 
>> 
>> -       There is a user U1 that wants to access C1, which in turn access R1. U1 gets authenticated somehow at C1. It could be either through a password system at C1, or through a federated login protocol supported at Ax, such as OpenID Connect. 
>> 
>>  
>> 
>> Another possibility is a case where Cx = Rx, which makes things a bit simpler. 
>> 
>>  
>> 
>> Is this what you have in mind? Please let me know. If it is not, please correct me.
>> 
>>  
>> 
>> Cheers, 
>> 
>>  
>> 
>> Nat
>> 
>> --
>> 
>> PLEASE READ :This e-mail is confidential and intended for the
>> 
>> named recipient only. If you are not an intended recipient,
>> 
>> please notify the sender  and delete this e-mail.
>> 
>>  
>> 
>> From: OAuth [mailto:oauth-bounces@ietf.org <mailto:oauth-bounces@ietf.org>] On Behalf Of Phil Hunt (IDM)
>> Sent: Friday, February 19, 2016 2:09 AM
>> To: Mike Jones
>> Cc: oauth@ietf.org <mailto:oauth@ietf.org>
>> Subject: Re: [OAUTH-WG] OAuth Discovery spec pared down to its essence
>> 
>>  
>> 
>> How does the client request the oauth configuration assigned to xyz?
>> 
>>  
>> 
>> The example you give appears to presume a single oauth infrastructure for all apps. 
>> 
>>  
>> 
>> The only way right now to have apps specific oauth is to infer the relation by the domain "xyz.example.com <http://xyz.example.com/>".  
>> 
>>  
>> 
>> That makes discovery more complex because there arw many more discovery locations and many more configurations to maintain. 
>> 
>>  
>> 
>> If example.com <http://example.com/> had separate oauth servers for services xyz and abc, how would discovery work from a single /.well-know endpoint?
>> 
>> 
>> Phil
>> 
>> 
>> On Feb 18, 2016, at 09:41, Mike Jones <Michael.Jones@microsoft.com <mailto:Michael.Jones@microsoft.com>> wrote:
>> 
>> Let me second John’s point that OAuth configuration information and application configuration information need not be interspersed.  For instance, if the service is at https://example.com <https://example.com/> and the XYZ application is being used, then these configuration metadata documents could both be used:
>> 
>> ·       https://example.com/.well-known/openid-configuration <https://example.com/.well-known/openid-configuration> - OAuth configuration metadata
>> 
>> ·       https://example.com/.well-known/xyz-configuration <https://example.com/.well-known/xyz-configuration> - XYZ configuration metadata
>> 
>>  
>> 
>> There’s not much point in defining a new /.well-known/oauth2.0 value, since there is no such thing as generic OAuth 2.0.  By definition, it must always be used in an application context that profiles OAuth 2.0 to enable interoperability.  The existing /.well-known/openid-configuration value works fine for this purpose.  Yes, the optics of having a different value might seem better but it comes at the cost of interoperability problems.  In my view, interop trumps optics.
>> 
>>  
>> 
>> To a point that George Fletcher made, WebFinger could still be used to learn the locations of these configuration metadata documents if that makes sense in the application context.  The editors took WebFinger out of the OAuth Discovery document since it isn’t always applicable.
>> 
>>  
>> 
>>                                                           Cheers,
>> 
>>                                                           -- Mike
>> 
>>  
>> 
>> From: John Bradley [mailto:ve7jtb@ve7jtb.com <mailto:ve7jtb@ve7jtb.com>] 
>> Sent: Thursday, February 18, 2016 7:41 AM
>> To: Phil Hunt <phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>>
>> Cc: Mike Jones <Michael.Jones@microsoft.com <mailto:Michael.Jones@microsoft.com>>; oauth@ietf.org <mailto:oauth@ietf.org>
>> Subject: Re: [OAUTH-WG] OAuth Discovery spec pared down to its essence
>> 
>>  
>> 
>> I suspect that the configuration well-knowns are going to be on the root domain.   You could try and get a user to put in crm.example.com <http://crm.example.com/>, but I suspect that is not going to work.
>> 
>>  
>> 
>> If the app doesn’t have a specific protocol identifier then it would use the default.  
>> 
>>  
>> 
>> I don’t know if you can get around having some sort of app/protocol identifier configured in the app.
>> 
>>  
>> 
>> John B.
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>> On Feb 18, 2016, at 9:49 AM, Phil Hunt <phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>> wrote:
>> 
>>  
>> 
>> resource service X could be any http accessible service:
>> 
>>  
>> 
>> * CRM
>> 
>> * Finance
>> 
>> * Payroll
>> 
>> * ERP
>> 
>> * any application on the web.
>> 
>>  
>> 
>> The spec seems to suggest that we use /.well-known/crm to discover OAuth config for crm.  But that may cause conflict if crm has its own discovery. Which leads us down the path of doing something like “crm-oauth”.
>> 
>>  
>> 
>> Then there is confusion about what host the discovery is done on.
>> 
>>  
>> 
>> For example, hypothetically do I do:
>> 
>>  
>> 
>> GET /.well-known/crm
>> 
>> Host: example.com <http://example.com/>
>>  
>> 
>> But what about the CRM’s configuration information. Is this stomping on it?
>> 
>>  
>> 
>> Or, what If we put the oauth configuration at the host for the crm service:
>> 
>> GET /.well-known/openid-configuration
>> 
>> Host: crm.example.com <http://crm.example.com/>
>>  
>> 
>> I think the point is that there is a relationship between a protected resource and its designated OAuth service. 
>> 
>>  
>> 
>> The client needs to discover:
>> 
>> * Where is its designated resource service and what security does it use
>> 
>> * If it is OAuth, where is the intended OAuth configuration for that resource service instance?
>> 
>>  
>> 
>> Phil
>> 
>>  
>> 
>> @independentid
>> 
>> www.independentid.com <http://www.independentid.com/>
>> phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>> On Feb 18, 2016, at 7:19 AM, John Bradley <ve7jtb@ve7jtb.com <mailto:ve7jtb@ve7jtb.com>> wrote:
>> 
>>  
>> 
>> Can you clarify what you mean by “resource service x”?
>> 
>>  
>> 
>> Is that the RS base URI for the resource,  a specific URI that the client is requesting?
>> 
>>  
>> 
>> That is getting UMA ish. 
>> 
>>  
>> 
>> The concept of a base RS URI is a rat hole that I prefer not to go down, as it is something everyone thinks exists but like SCIM if it exists it is protocol or deployment specific.
>> 
>>  
>> 
>> The notion that you would send the URI you are planning on requesting to a Webfinger server to find the OAuth server, is probably going to have privacy issues.
>> 
>>  
>> 
>> I suspect that you need to hand back a error from the resource to say where the AS is, or have a .well-known for the RS.
>> 
>>  
>> 
>> RS discovery probably wants to be separate from AS discovery.  (Yes I do think we need something,  UMA rpt or something like it might be a way to go)
>> 
>>  
>> 
>> John B.
>> 
>>  
>> 
>> On Feb 18, 2016, at 9:06 AM, Phil Hunt <phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>> wrote:
>> 
>>  
>> 
>> Maybe SCIM was a bad example.  It functions as a RESTful resource in the context of OAuth.
>> 
>>  
>> 
>> I find the use of OIDC to be confusing as an example (and the default) because it is both an OAuth resource and a security service.  It is a modification of OAuth.
>> 
>>  
>> 
>> Start thinking about every application ever written that uses OAuth. Are we expecting 100s of thousands of these to each register?
>> 
>>  
>> 
>> To me, this specification is a fine specification for OIDC and it should be published there because the specification defines how to discovery OAuth and OpenID information.
>> 
>>  
>> 
>> Likewise you suggest it is ok for SCIM to do the same. 
>> 
>>  
>> 
>> How do we expect normal applications to set up and do discovery?
>> 
>>  
>> 
>> It seems to me that an “OAUTH” discovery spec should have a parameter to ask, I want to discover OAuth configuration for resource service X.
>> 
>>  
>> 
>> That still allows me to have a separate discovery service that says, tell me about resource service X itself.
>> 
>>  
>> 
>> BTW. I think we are FAR from Last Call on this topic.
>> 
>>  
>> 
>> Phil
>> 
>>  
>> 
>> @independentid
>> 
>> www.independentid.com <http://www.independentid.com/>
>> phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>> On Feb 18, 2016, at 6:55 AM, John Bradley <ve7jtb@ve7jtb.com <mailto:ve7jtb@ve7jtb.com>> wrote:
>> 
>>  
>> 
>> Diffrent protocols like Connect and SCIM may have different configurations, endpoints , keys , authentication methods, scopes etc.
>> 
>>  
>> 
>> It should be posable to have them as one document, but forcing them to use one document is going to cause a explosion of claim registration for discovery.
>> 
>>  
>> 
>> I think it is better for SCIM to register one well known than to have to register 20 claims with scim prefixes or something silly like that.
>> 
>>  
>> 
>> Name-spacing the claims by allowing them to be in different well known files is not unreasonable.
>> 
>>  
>> 
>> Remember some of these protocols may be hosted on SaaS so there is no guarantee that all protocols will have the same OAuth Config.
>> 
>>  
>> 
>> Nothing stops a protocol from doing what it likes with webfinger if it wants to use that for discovery.
>> 
>>  
>> 
>> In principal I like the idea of having another protocol as an example.
>> 
>>  
>> 
>> My only concern is that I haven’t seen any discussion of your SCIM discovery document in the SCIM WG.  
>> 
>> I personally think sorting out discovery for SCIM is a good idea,  but OAUTh is but one of several authentication methods for SCIM, and there are probably other non OAuth things that want to be described.
>> 
>>  
>> 
>> I would feel better about using it as an example if it were adopted by the WG and some general interest shown.
>> 
>>  
>> 
>> I encourage you to do that so we can use it as a example.
>> 
>>  
>> 
>> John B.
>> 
>>  
>> 
>> On Feb 18, 2016, at 8:35 AM, Phil Hunt <phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>> wrote:
>> 
>>  
>> 
>> I still find the following text objectionable and confusing…
>> 
>>    By default, for historical reasons, unless an application-specific
>>    well-known URI path suffix is registered and used for an application,
>>    the client for that application SHOULD use the well-known URI path
>>    suffix "openid-configuration" and publish the metadata document at
>>    the path formed by concatenating "/.well-known/openid-configuration"
>>    to the authorization server's issuer identifier.  As described in
>>    Section 5 <http://tools.ietf.org/html/draft-ietf-oauth-discovery-01#section-5>, despite the identifier
>>    "/.well-known/openid-configuration", appearing to be OpenID-specific,
>>    its usage in this specification is actually referring to a general
>>    OAuth 2.0 feature that is not specific to OpenID Connect.
>>  
>> 
>> Further, as a default “openid-configuration” as the default further gives people the impression that a plain OAuth server *is* an authentication server and that the normal access token received is evidence of a successful authentication.
>> 
>>  
>> 
>> It would be better to point out that application may include oauth discovery in their discovery URI and that OAuth is an example of this. It might be good to include two examples.  E.g. OIDC and SCIM (as another referenceable example).
>> 
>>  
>> 
>>  GET /.well-known/openid-configuration
>> and
>> 
>>  GET /.well-known/scim
>> Retrieve the OAuth configuration for the application openid and scim respectively.
>> 
>>  
>> 
>> The use of:
>> 
>>  GET /.well-known/oauth2/
>> Should be the default used when there is no known application based well-known application based URI discovery.
>> 
>>  
>> 
>> Of course, the concern I raised earlier is that this approach of application specific URIs ends up requiring every application to make an IANA registration if they don’t want to use the default of “oauth2” (or “openid-configuration”).  Is that what the authors expect?
>> 
>>  
>> 
>> It seemed better to me to use the webfinger syntax to allow the client to say “I want the designated OAuth configuration for the resource service X” would be a better design that avoids extensive IANA registration.
>> 
>>  
>> 
>> Phil
>> 
>>  
>> 
>> @independentid
>> 
>> www.independentid.com <http://www.independentid.com/>
>> phil.hunt@oracle.com <mailto:phil.hunt@oracle.com>
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>> On Feb 17, 2016, at 11:48 PM, Mike Jones <Michael.Jones@microsoft.com <mailto:Michael.Jones@microsoft.com>> wrote:
>> 
>>  
>> 
>> In response to working group input, this version of the OAuth Discovery specification has been pared down to its essence – leaving only the features that are already widely deployed. Specifically, all that remains is the definition of the authorization server discovery metadata document and the metadata values used in it. The WebFinger discovery logic has been removed. The relationship between the issuer identifier URL and the well-known URI path relative to it at which the discovery metadata document is located has also been clarified.
>> 
>>  
>> 
>> Given that this now describes only features that are in widespread deployment, the editors believe that this version is ready for working group last call.
>> 
>>  
>> 
>> The specification is available at:
>> 
>> ·http://tools.ietf.org/html/draft-ietf-oauth-discovery-01 <http://tools.ietf.org/html/draft-ietf-oauth-discovery-01>
>>  
>> 
>> An HTML-formatted version is also available at:
>> 
>> ·http://self-issued.info/docs/draft-ietf-oauth-discovery-01.html <http://self-issued.info/docs/draft-ietf-oauth-discovery-01.html>
>>  
>> 
>>   -- Mike & Nat & John
>> 
>>  
>> 
>> P.S. This notice was also posted athttp://self-issued.info/?p=1544 <http://self-issued.info/?p=1544>and as@selfissued <https://twitter.com/selfissued>.
>> 
>> _______________________________________________
>> OAuth mailing list
>> OAuth@ietf.org <mailto:OAuth@ietf.org>
>> https://www.ietf.org/mailman/listinfo/oauth <https://www.ietf.org/mailman/listinfo/oauth>
>>  
>> 
>> _______________________________________________
>> OAuth mailing list
>> OAuth@ietf.org <mailto:OAuth@ietf.org>
>> https://www.ietf.org/mailman/listinfo/oauth <https://www.ietf.org/mailman/listinfo/oauth>
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
>>  
>> 
> _______________________________________________
> OAuth mailing list
> OAuth@ietf.org <mailto:OAuth@ietf.org>
> https://www.ietf.org/mailman/listinfo/oauth <https://www.ietf.org/mailman/listinfo/oauth>