Re: [OAUTH-WG] Registration: Scope Values

[Justin Richer <[e-mail]>]

Thing is, there's nothing normative about the enforcing statement that I 
made below, so I don't think it's any more restrictive than RFC 6749 
which lets the AS replace a client's requested scopes at the time of 
token issuance with whatever it pleases. But that said, I'd be just as 
happy to leave it like this with no further restrictions:

Space-separated list of scope values (as described in OAuth 2.0 
Section 3.3 [RFC6749] <http://tools.ietf.org/html/rfc6749#section-3.3>) 
that the Client can use when requesting access tokens from the 
Authorization Server.

And call it a day. This parallels the text for grant_types ("Array of 
OAuth 2.0 grant types that the Client may use [when accessing the Token 
Endpoint].") and response_types ("Array of the OAuth 2.0 response types 
that the Client may use [when accessing the Authorization Endpoint]."), 
and I think this is the correct approach. I only started to add the 
restrictive text because I thought that people were making the argument 
that it was necessary -- I'd rather not have it.

Plus, it's an optional field in the metadata, so if you've got 
structured scopes where this doesn't make sense, don't use it. If you 
don't do a per-client scope restriction, don't use it.

The interoperability is defined in light of the interoperability of 
scopes themselves: this is a field to request/grant a bag of strings 
that only make sense in light of a given API. For that to make real 
sense, I think that we need to differentiate an OAuth client as a 
generic *library* from an OAuth client as a generic accessing 
*application*. It's very useful for a generic *library* that handles the 
authorization layer for an application to have a slot for registering 
scopes and finding out what scopes the app's been registered for. It's 
up to the app (not the library) to figure out how to translate those 
into scopes to ask for at authorization time. Sometimes that means just 
passing the string, and sometimes it means the construction of a 
structured value like 
"urn:example:channel=HBO&urn:example:rating=G,PG-13". The library 
doesn't care, the application does, and we should focus on 
interoperability from the library's perspective. Similarly, on the 
server side, the libraries will tell you the bag of scopes that the 
client was registered for, and what bag of scopes the client asked for 
during authorization. It's up to the server *application* to harmonize 
those two in a way that makes sense for the API that it's protecting. 
Just like it's up to the protected resource *application* to figure out 
what a scope means in a given context.

So let's just leave it unrestricted but keep the slot for communicating 
this piece of information with the same semantics that the communication 
between the client and server take on for every other field: client asks 
for a thing, server says that client actually gets a thing, and it's 
implicitly up to the server to do the right thing and enforce things in 
a way that makes sense for the application no matter what the client does.

To take Tony's example, client requests no scopes at registration, 
server grants scope "A" at registration. Client then requests scope "X" 
at authorization, server is free to deny the request (invalid_scope 
error), allow authorization because it knows how "A" and "X" are 
related, require user intervention, or really, whatever it likes. The 
libraries, where I argue the interoperability cries really matter, don't 
care, and shouldn't care.

  -- Justin

On 04/18/2013 01:04 PM, Mike Jones wrote:
>
> Saying anything normative about "enforcing restrictions" is going 
> beyond RFC 6749 semantics.  Indeed, you'd said that "I agree that we 
> shouldn't try to "solve" scope in registration.", but talking about 
> restrictions is going down the slippery "solving it" path.
>
> At most we can say that the two parties are making declarations to one 
> another about scopes that they may choose to use, but we can't assume 
> that this is an exclusive list and that other scope values such as 
> "urn:example:channel=HBO&urn:example:rating=G,PG-13" might not be 
> used, even if the client registers saying that it intends to use the 
> "OATC" scope value.  We could maybe even say that some services may 
> use a static set of scopes and might choose to limit the scopes that a 
> client may use to those that it declared to the server or to those 
> that the server returned to the client.  That's a HINT that some 
> services might do this.  But we can't say anything normative about 
> such possible behaviors, because it goes beyond RFC 6749.
>
> -- Mike
>
> *From:*Richer, Justin P. [mailto:[e-mail]]
> *Sent:* Thursday, April 18, 2013 9:26 AM
> *To:* Anthony Nadalin
> *Cc:* Tim Bray; Mike Jones; [e-mail]
> *Subject:* Re: [OAUTH-WG] Registration: Scope Values
>
> This doesn't actually break the semantics because the client MUST 
> accept what the server tells it over anything that it asks for in the 
> first place. The server has the final say. So in this case, if your 
> client asks for nothing, the server says "A B C", the client now knows 
> it can ask for "A B C" scopes.
>
> I'm still in favor of not putting the restricting language in the 
> scope definition at all, instead have it say something like:
>
>     "Space-separated list of scope values (as described in OAuth 2.0
>     Section 3.3 [RFC6749]
>     <http://tools.ietf.org/html/rfc6749#section-3.3>) that the Client
>     can use when requesting access tokens from the Authorization
>     Server. As scope values are service specific, the means of the
>     Authorization Server enforcing this restriction are outside the
>     scope of this specification."
>
> Couple this with the overall paragraph that says that the client is 
> requesting values that the server is potentially overriding with its 
> declarations, and I think that addresses everything without getting 
> into confusing language that doesn't add to interoperability.
>
>  -- Justin
>
> On Apr 18, 2013, at 12:13 PM, Anthony Nadalin <[e-mail] 
> <mailto:[e-mail]>> wrote:
>
>
>
> If I don't specify a scope, then the server can allocate a default (or 
> default set), thus that breaks the semantics you describe
>
> *From:*[e-mail] <mailto:[e-mail]> 
> [mailto:[e-mail] <mailto:[e-mail]>]*On Behalf 
> Of*Tim Bray
> *Sent:*Thursday, April 18, 2013 9:04 AM
> *To:*Mike Jones
> *Cc:*[e-mail] <mailto:[e-mail]>
> *Subject:*Re: [OAUTH-WG] Registration: Scope Values
>
> I'm unconvinced, Mike.  Obviously you're right about the looseness of 
> OAuth2 scope specification, but this is a very specific semantic of 
> what happens when you register, and I don't think we're bound by 
> history here.   If we can't safely say anything about what the list of 
> scopes means, then I'm with you let's take them out.  But the most 
> obvious intended semantic is (from the client) "I promise to ask only 
> for these" and from the server "These are the only ones I'll give you 
> tokens for."  Or does someone have use-cases for an alternative semantic?
>
> To make this concrete, I propose the following:
>
> "Space-separated list of scope values (as described in OAuth 2.0 
> Section 3.3 [RFC6749] 
> <http://tools.ietf.org/html/rfc6749#section-3.3>) that the client is 
> declaring to the server that it will restrict itself to when 
> requesting access tokens, and that the server is declaring to the 
> client that it is registered to use when requesting access tokens. 
> Clients SHOULD assume that servers will refuse to grant access tokens 
> for scopes not in the list provided by the server."
>
> On Thu, Apr 18, 2013 at 8:55 AM, Mike Jones 
> <[e-mail] <mailto:[e-mail]>> wrote:
>
>     I don't think it's possible to define what it means in an
>     interoperable way because OAuth didn't specify scopes in an
>     interoperable way.  No, I don't like that either, but I think
>     that's where things are.  That's why I was advocating deleting
>     this registration feature entirely.
>
>     But understanding it might be useful in some contexts, I'm OK
>     keeping it, provided we be clear that the semantics of "registered
>     to use" are service-specific.
>
>     -- Mike
>
>     *From:*Tim Bray [mailto:[e-mail] <mailto:[e-mail]>]
>     *Sent:*Thursday, April 18, 2013 8:36 AM
>     *To:*Mike Jones
>     *Cc:*Justin Richer;[e-mail] <mailto:[e-mail]>
>
>
>     *Subject:*Re: [OAUTH-WG] Registration: Scope Values
>
>     On the server-to-client side, what does "registered to use" mean?
>      Does it mean that the client should assume that any scopes not on
>     the list WILL not be granted, MAY not be granted.... or what?  Is
>     this already covered elsewhere? -T
>
>     On Thu, Apr 18, 2013 at 8:28 AM, Mike Jones
>     <[e-mail] <mailto:[e-mail]>>
>     wrote:
>
>     Thanks, Justin.  I agree with the need for the generic two-sided
>     language.  I'd still keep this language for scope, because we want
>     to capture the "declaring" aspect in this case:
>
>     "Space separated list of scope values (as described in OAuth
>     2.0Section 3.3 [RFC6749]
>     <http://tools.ietf.org/html/rfc6749#section-3.3>) that the client
>     is declaring to the server that it may use when requesting access
>     tokens and that the server is declaring to the client that it is
>     registered to use when requesting access tokens.".
>
>     You should probably also reinforce that scope values are
>     service-specific and may not consist only of a static set of
>     string values, and that therefore, in some cases, an exhaustive
>     list of registered scope values is not possible.
>
>     -- Mike
>
>     *From:*Justin Richer [mailto:[e-mail]
>     <mailto:[e-mail]>]
>     *Sent:*Monday, April 15, 2013 12:29 PM
>
>
>     *To:*Mike Jones
>     *Cc:*Tim Bray;[e-mail] <mailto:[e-mail]>
>     *Subject:*Re: [OAUTH-WG] Registration: Scope Values
>
>     I think that because the "declaration" issue affects all
>     parameters in the list, not just scope, we should adopt this in a
>     higher level paragraph and leave it out of the individual
>     parameter descriptions. Thus, something like this inserted as the
>     second paragraph in section 2:
>
>     The client metadata values serve two parallel purposes in the
>     overall OAuth Dynamic Registration protocol:
>
>      - the Client requesting its desired values for each parameter to
>     the Authorization Server in a [register] or [update] request,
>      - the Authorization Server informing the Client of the current
>     values of each parameter that the Client has been registered to
>     use through a [client information response].
>
>     An Authorization Server MAY override any value that a Client
>     requests during the registration process (including any omitted
>     values) and replace the requested value with a default. The
>     normative indications in the following list apply to the Client's
>     declaration of its desired values.
>
>     The Authorization Server SHOULD provide documentation for any
>     fields that it requires to be filled in by the client or to have
>     particular values or formats. Extensions and profiles...
>
>
>     And then remove the sidedness-language from the scope parameter
>     and any other parameters where it might have crept in inadvertently.
>
>      -- Justin
>
>     On 04/15/2013 01:29 PM, Mike Jones wrote:
>
>         We could fix the one-sided language by changing
>
>         "Space separated list of scope values (as described in OAuth
>         2.0Section 3.3 [RFC6749]
>         <http://tools.ietf.org/html/rfc6749#section-3.3>) that the
>         client is declaring that it may use when requesting access
>         tokens."
>
>         to
>
>         "Space separated list of scope values (as described in OAuth
>         2.0Section 3.3 [RFC6749]
>         <http://tools.ietf.org/html/rfc6749#section-3.3>) that the
>         client is declaring to the server that it may use when
>         requesting access tokens and that the server is declaring to
>         the client that it is registered to use when requesting access
>         tokens.".
>
>         Again, I chose the "registered to use" language carefully --
>         because in the general case it's not a restriction on the
>         values that the client can use -- just a statement by the
>         server to the client that it is registered to use those
>         particular values.  In both cases, the parties are making
>         declarations to one another.
>
>         If you adopt that language (or keep the original language),
>         then yes, I'd consider this closed.
>
>         -- Mike
>
>         *From:*Justin Richer [mailto:[e-mail]]
>         *Sent:*Monday, April 15, 2013 9:57 AM
>         *To:*Mike Jones
>         *Cc:*Tim Bray;[e-mail] <mailto:[e-mail]>
>         *Subject:*Re: [OAUTH-WG] Registration: Scope Values
>
>         I absolutely do not want to delete this feature, as (having
>         implemented it) I think it's very useful. This is a very
>         established pattern in manual registration: I know of many,
>         many OAuth2 servers and clients that are set up where the
>         client must pre-register a set of scopes.
>
>         I don't like the language of "the client is declaring" because
>         it's too one-sided. The client might not have declared
>         anything, and it might be the server that's declaring
>         something to the client. Deleting the "is declaring" bit
>         removes that unintended restriction of the language while
>         keeping the original meaning intact. I actually thought that I
>         had fixed that before the last draft went in but apparently I
>         missed this one.
>
>         I will work on clarifying the intent of the whole metadata set
>         in its introductory paragraph(s) so that it's clear that all
>         of these fields are used in both of these situations:
>
>          1) The client declaring to the server its desire to use a
>         particular value
>          2) The server declaring to the client that it has been
>         registered with a particular value
>
>         This should hopefully clear up the issue in the editor's note
>         that I currently have at the top of that section right now, too.
>
>         Mike, since you were the one who originally brought up the
>         issue, and you're fine with the existing text, can I take this
>         as closed now? Assuming that you agree with deleting "is
>         declaring" for reasons stated above, I'm fine with leaving
>         everything else as is and staying quiet on what the server has
>         to do with the scopes.
>
>          -- Justin
>
>         On 04/15/2013 12:44 PM, Mike Jones wrote:
>
>             I think that the existing wording is superior to the
>             proposed changed wording.  The existing wording is:
>
>                scope
>
>                   OPTIONAL.  Space separated list of scope values (as
>             described in
>
>                   OAuth 2.0Section 3.3 [RFC6749]
>             <http://tools.ietf.org/html/rfc6749#section-3.3>) that the
>             client is declaring that
>
>                   it may use when requesting access tokens.  If
>             omitted, an
>
>                   Authorization Server MAY register a Client with a
>             default set of
>
>                   scopes.
>
>             For instance, the current "client is declaring" wording
>             will always be correct, whereas as the change to "client
>             can use" wording implies a restriction on client behavior
>             that is not always applicable.  The "client is declaring"
>             wording was specific and purposefully chosen, and I think
>             should be retained.  In particular, we can't do anything
>             that implies that only the registered scopes values can be
>             used.  At the OAuth spec level, this is a hint as to
>             possible future client behavior -- not a restriction on
>             future client behavior.
>
>             Also, for the reasons that Tim stated, I'm strongly
>             against any "matching" or "regex" language in the spec
>             pertaining to scopes -- as it's not actionable.
>
>             So I'd propose that we leave the existing scope wording in
>             place.  Alternatively, I'd also be fine with deleting this
>             feature entirely, as I don't think it's useful in the
>             general case.
>
>             -- Mike
>
>             *From:*[e-mail]
>             <mailto:[e-mail]>[mailto:[e-mail]]*On
>             Behalf Of*Justin Richer
>             *Sent:*Monday, April 15, 2013 8:05 AM
>             *To:*Tim Bray;[e-mail] <mailto:[e-mail]>
>             *Subject:*Re: [OAUTH-WG] Registration: Scope Values
>
>             On 04/15/2013 10:52 AM, Tim Bray wrote:
>
>
>             I'd use the existing wording; it's perfectly clear.
>              Failing that, if there's strong demand for registration
>             of structured scopes, bless the use of regexes, either
>             PCREs or some careful subset.
>
>
>             Thanks for the feedback -- Of these two choices, I'd
>             rather leave it as-is.
>
>
>
>             However, I'd subtract the sentence "If omitted, an
>             Authorization Server MAY register a Client with a default
>             set of  scopes."  It adds no value; if the client doesn't
>             declare scopes, the client doesn't declare scopes, that's
>             all.  -T
>
>
>             Remember, all of these fields aren't just for the client
>             *request*, they're also for the server's *response* to
>             either a POST, PUT, or GET request. (I didn't realize it,
>             but perhaps the wording as stated right now doesn't make
>             that clear -- I need to fix that.) The value that it adds
>             is if the client doesn't ask for any particular scopes,
>             the server can still assign it scopes and the client can
>             do something smart with that. Dumb clients are allowed to
>             ignore it if it doesn't mean anything to them.
>
>             This is how our server implementation actually works right
>             now. If the client doesn't ask for anything specific at
>             registration, the server hands it a bag of "default"
>             scopes. Same thing happens at auth time -- if the client
>             doesn't ask for any particular scopes, the server hands it
>             all of its registered scopes as a default. Granted, on our
>             server, scopes are just simple strings right now, so they
>             get compared at the auth endpoint with an exact
>             string-match metric and set-based logic.
>
>              -- Justin
>
>
>
>             On Mon, Apr 15, 2013 at 7:35 AM, Justin Richer
>             <[e-mail] <mailto:[e-mail]>> wrote:
>
>             What would you suggest for wording here, then? Keeping in
>             mind that we cannot (and don't want to) prohibit
>             expression-based scopes.
>
>              -- Justin
>
>             On 04/15/2013 10:33 AM, Tim Bray wrote:
>
>                 No, I mean it's not interoperable at the
>                 software-developer level.  I can't register scopes at
>                 authorization time with any predictable effect that I
>                 can write code to support, either client or server
>                 side, without out-of-line non-interoperable knowledge
>                 about the behavior of the server.
>
>                 I guess I'm just not used to OAuth's culture of having
>                 no expectation that things will be specified tightly
>                 enough that I can write code to implement as
>                 specified.  -T
>
>                 On Mon, Apr 15, 2013 at 7:15 AM, Justin Richer
>                 <[e-mail] <mailto:[e-mail]>> wrote:
>
>                 Scopes aren't meant to be interoperable between
>                 services since they're necessarily API-specific. The
>                 only interoperable bit is that there's *some* place to
>                 put the values and that it's expressed as a bag of
>                 space-separated strings. How those strings get
>                 interpreted and enforced (which is really what's at
>                 stake here) is up to the AS and PR (or a higher-level
>                 protocol like UMA).
>
>                  -- Justin
>
>                 On 04/15/2013 10:13 AM, Tim Bray wrote:
>
>                     This, as written, has zero interoperability. I
>                     think this feature can really only be made useful
>                     in the case where scopes are fixed strings.
>
>                     -T
>
>                     On Apr 15, 2013 6:54 AM, "Justin Richer"
>                     <[e-mail] <mailto:[e-mail]>> wrote:
>
>                     You are correct that the idea behind the "scope"
>                     parameter at registration is a constraint on
>                     authorization-time scopes that are made available.
>                     It's both a means for the client to request a set
>                     of valid scopes and for the server to provision
>                     (and echo back to the client) a set of valid scopes.
>
>                     I *really* don't want to try to define a matching
>                     language for scope expressions. For that to work,
>                     all servers would need to be able to process the
>                     regular expressions for all clients, even if the
>                     servers themselves only support simple-string
>                     scope values. Any regular expression syntax we
>                     pick here is guaranteed to be incompatible with
>                     something, and I think the complexity doesn't buy
>                     much. Also, I think you suddenly have a potential
>                     security issue if you have a bad regex in place on
>                     either end.
>
>                     As it stands today, the server can interpret the
>                     incoming registration scopes and enforce them
>                     however it wants to. The real trick comes not from
>                     assigning the values to a particular client but to
>                     enforcing them, and I think that's always going to
>                     be service-specific. We're just not as clear on
>                     that as we could be.
>
>                     After looking over everyone's comments so far, I'd
>                     like to propose the following text for that section:
>
>
>                         scope
>
>                            OPTIONAL.  Space separated list of scope values (as described in
>
>                            OAuth 2.0Section 3.3 [RFC6749]  <http://tools.ietf.org/html/rfc6749#section-3.3>) that the client can use when
>
>                            requesting access tokens.  As scope values are service-specific,
>
>                            the Authorization Server MAY define its own matching rules when
>
>                            determining if a scope value used during an authorization request
>
>                            is valid according to the scope values assigned during
>
>                            registration. Possible matching rules include wildcard patterns,
>
>                            regular expressions, or exactly matching the string. If omitted,
>
>                            an Authorization Server MAY register a Client with a default
>
>                            set of scopes.
>
>
>                     Comments? Improvements?
>
>                      -- Justin
>
>
>                     On 04/14/2013 08:23 PM, Manger, James H wrote:
>
>                         Presumably at app registration time any scope specification is really a constraint on the scope values that can be requested in an authorization flow.
>
>                           
>
>                         So ideally registration should accept rules for matching scopes, as opposed to actual scope values.
>
>                           
>
>                         You can try to use scope values as their own matching rules. That is fine for a small set of "static" scopes. It starts to fail when there are a large number of scopes, or scopes that can include parameters (resource paths? email addresses?). You can try to patch those failures by allowing services to define service-specific special "wildcard" scope values that can only be used during registration (eg "read:*").
>
>                           
>
>                         Alternatively, replace 'scope' in registration with 'scope_regex' that holds a regular expression that all scope values in an authorization flow must match.
>
>                           
>
>                         --
>
>                         James Manger
>
>                         _______________________________________________
>
>                         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
>
> _______________________________________________
> OAuth mailing list
> [e-mail] <mailto:[e-mail]>
> https://www.ietf.org/mailman/listinfo/oauth
>