Re: [OAUTH-WG] Concerning OAuth introspection

[Anthony Nadalin <[e-mail]>]

I would not say that this is incompatible with OAUTH at all , as OAUTH has a physical and logical endpoints. We had to live through the Web Service endpoint nightmare were we had to have separate services, nuts

From: Justin Richer [mailto:[e-mail]]
Sent: Wednesday, January 30, 2013 7:20 AM
To: Shiu Fun Poon
Cc: Anthony Nadalin; Nat Sakimura; [e-mail]
Subject: Re: [OAUTH-WG] Concerning OAuth introspection



Hi.. Tony.. You are able to present this better than I do.
Justin,
Currently as it is, the spec is unflexible.   So when I send a request to an endpoint, the endpoint will need to have information like /revoke, /introspection, and others.  I can imagine the documented usage of this will a pages long just to describe all the endpoint that is needed to support an application.  (so far, /authz, /token, /introspection, /revoke, /.....), and this will be painful in the multi-tenant environment.

So why not allow the flexibility of using one endpoint /token (example only), and make the caller to tell you want the caller wants ?  It can be an optional field, and a hint to the authorization/token endpoint on what you want to do.
This is incompatible with OAuth as it stands today, which defines the auth endpoint and token endpoint as logically separate, and therefore there is not a parameter defined that would allow for such functionality switching.

That said, an *installation* could implement it this way if they wanted to, they'd just have to document the endpoints like this:

token endpoint: /oauth?op=token
auth endpoint: /oauth?op=auth
introspection endpoint: /oauth?op=introspect
revocation endpoint: /oauth?op=revoke

etc. The key difference here is that the "op=" parameter is *system specific* and is *not* part of the spec itself. And I think that's a good design to continue to follow.

Right now, the registration endpoint is the only one that defines an "operation" parameter, and I was positing the question of defining it in terms of three different endpoints instead. Most of the time, these endpoints will have different URLs, as outlined below, but specific instances could use some kind of "operation" parameter if they wanted to.

Defining it as one endpoint with a switch parameter actually decreses flexibility quite a lot, especially if you're talking about dispatching to different kinds of functionality all together, which is the use case you brought up. There are some places where that could make sense, and the definition of different endpoints allows you to do that in specific instances of a system without breaking the assumptions of clients.

What Tony was talking about was allowing something to be either three different URLs *or* using a spec-defined "operation" parameter. That suggestion is completely nuts, in my opinion.


And it does not violate the rest/json guideline.
Yes it absolutely does. The REST principle is that a URL represents one entity and the HTTP verbs represent different actions on that entity. Using a query parameter to switch is very much directly against REST guidelines. Not to say that OAuth is RESTful -- it's not, by a long shot. But it does follow many rest-like principles, the endpoint definitions being one of them.



Even with oauth specification, it provides a hint on what is to come, e.g. grant_type refresh_token, indicates you want to exchange a valid refresh_token to an access_token.  There is something in the payload which tell you what you need to do.  In this case, there is nothing in the payload which indicate what is expected.
These are functionality switches on the same kind of action, not dispatch to different actions. you still do authentication/authorization at the auth endpoint, you still get tokens from the token endpoint, etc.



If you standard that now (on the optional field), there is a chance that companies can implement this according to what will work best for them, and we actually have a chance to get this working between different products.
It's too late to standardize that field in the core, which is really where it would belong. But as it stands today, an OAuth client is going to need to be able to handle separate URLs for each defined endpoint already, so it can already handle the case where it happens to be the same base URL with different operations.

For what it's worth, what I was trying to get discussion on was whether it made sense to make Dynamic Registration look like the rest of OAuth with separate endpoints, or not.

 -- Justin




On Wed, Jan 23, 2013 at 1:05 PM, Justin Richer <[e-mail]<mailto:[e-mail]>> wrote:
I am very confused, and I need someone to explain to me what I am missing. Why won't it work to just pick one? What are people "already stuck with" that this would affect? It's not like we're trying to unseat a well-established protocol with a wide installation base here.

How will giving people the choice between:
/oauth/register?operation=client_register
/oauth/register?operation=client_update
/oauth/register?operation=rotate_secret

and:
/oauth/client_register
/oauth/client_update
/oauth/rotate_secret

help multitenancy? How does it even affect that use case? Consider that the base URL for all of these is completely up to the host environment (nothing is bound to the root URL). Consider that clients still have to know what the URL (or URLs) are, in either case. Consider that clients still need to know how to manage all the parameters and responses.

If anything, keeping it the way that it is with a single URL could be argued to help multitenancy because setting up routing to multiple URL endpoints can sometimes be problematic in hosted environments. However, OAuth already defines a bunch of endpoints, and we have to define at least one more with this extension, so I'm not convinced that having three with specific functions is really any different from having one with three functions from a development, deployment, and implementation perspective. I can tell you from experience that in our own server code, the difference is trivial. (And from OAuth1 experience, you can always have a query parameter as part of your endpoint URL if you need to. You might hate yourself for doing it that way, but nothing says your base URL can't already have parameters on it. A client just needs to know how to appropriately tack its parameters onto an existing URL, and any HTTP client worth its salt will know how to augment a query parameter set with new items.)

The *real* difference between the two approaches is a philosophical design one. The former overloads one URL with multiple functions switched by a flag, the latter uses the URL itself as an implicit flag. Under the hood, these could (and in many cases will) be all served by the same chunks of code. The only difference is how this switch in functionality is presented.


With that said, can somebody please explain to me how allowing *both* of these as options simultaneously (what I understand Tony to be suggesting) is a good idea, or how multitenancy even comes into play? Because I am completely not seeing how these are related.

Thanks,
 -- Justin



On 01/23/2013 12:46 PM, Anthony Nadalin wrote:

It will not work the way you have it, as people do multi-tendency different and they are already stuck with the method that they have chosen, so they need the flexability, to restrict this is nuts as people won't use it.



-----Original Message-----

From: Justin Richer [mailto:[e-mail]]

Sent: Wednesday, January 23, 2013 9:27 AM

To: Anthony Nadalin

Cc: Nat Sakimura; Shiu Fun Poon; [e-mail]<mailto:[e-mail]>

Subject: Re: [OAUTH-WG] Concerning OAuth introspection



I completely disagree with this assessment. Multi-tenancy will work just fine (or even better) if everyone uses the same pattern. Telling someone "it might be three different urls or it might be all one url with a parameter" is just asking for a complete disaster. What does the flexibility of allowing two approaches actually accomplish?



You can argue about the merits of either approach, but having both as unspecified options for registration, which is meant to help things get going in a cold-boot environment, is just plain nuts.





-- Justin







On 01/23/2013 12:21 PM, Anthony Nadalin wrote:

Registration has to work in a multi-tenant environment  so flexibility

is needed



-----Original Message-----

From: Justin Richer [mailto:[e-mail]]

Sent: Wednesday, January 23, 2013 9:18 AM

To: Anthony Nadalin

Cc: Nat Sakimura; Shiu Fun Poon; [e-mail]<mailto:[e-mail]>

Subject: Re: [OAUTH-WG] Concerning OAuth introspection



Because then nobody would know how to actually use the thing.



In my opinion, this is a key place where this kind of flexibility is a very bad thing. Registration needs to work one fairly predictable way.



-- Justin



On 01/23/2013 12:14 PM, Anthony Nadalin wrote:

Why not just have a physical and logical endpoint options



-----Original Message-----

From: [e-mail]<mailto:[e-mail]> [mailto:[e-mail]] On

Behalf Of Justin Richer

Sent: Wednesday, January 23, 2013 7:47 AM

To: Nat Sakimura

Cc: Shiu Fun Poon; [e-mail]<mailto:[e-mail]>

Subject: Re: [OAUTH-WG] Concerning OAuth introspection



Which brings up an interesting question for the Registration doc: right now, it's set up as a single endpoint with three operations. We could instead define three endpoints for the different operations.



I've not been keen to make that deep of a cutting change to it, but it would certainly be cleaner and more RESTful API design. What do others think?



-- Justin





On 01/22/2013 08:05 PM, Nat Sakimura wrote:

"Action" goes against REST principle.

I do not think it is a good idea.



=nat via iPhone



Jan 23, 2013 4:00、Justin Richer <[e-mail]><mailto:[e-mail]> のメッセージ:



(CC'ing the working group)



I'm not sure what the "action/operation" flag would accomplish. The idea behind having different endpoints in OAuth is that they each do different kinds of things. The only "action/operation" that I had envisioned for the introspection endpoint is introspection itself: "I have a token, what does it mean?"



Note that client_id and client_secret *can* already be used at this endpoint if the server supports that as part of their client credentials setup. The examples use HTTP Basic with client id and secret right now. Basically, the client can authenticate however it wants, including any of the methods that OAuth2 allows on the token endpoint. It could also authenticate with an access token. At least, that's the intent of the introspection draft -- if that's unclear, I'd be happy to accept suggested changes to clarify this text.



 -- Justin



On 01/22/2013 01:00 PM, Shiu Fun Poon wrote:

Justin,



This spec is looking good..



One thing I would like to recommend is to add "action"/"operation"

to the request.  (and potentially add client_id and client_secret)



So the request will be like :

token                                             REQUIRED

operation (wording to be determine)  OPTIONAL inquire (default) | revoke ...

resource_id                                    OPTIONAL

client_id                                         OPTIONAL

client_secret                                   OPTIONAL



And for the OAuth client information, it should be an optional parameter (in case it is a public client or client is authenticated with SSL mutual authentication).



Please consider.



ShiuFun

_______________________________________________

OAuth mailing list

[e-mail]<mailto:[e-mail]>

https://www.ietf.org/mailman/listinfo/oauth

_______________________________________________

OAuth mailing list

[e-mail]<mailto:[e-mail]>

https://www.ietf.org/mailman/listinfo/oauth