Re: [OAUTH-WG] [UNVERIFIED SENDER] Re: I-D Action: draft-ietf-oauth-access-token-jwt-03.txt

["Richard Backman, Annabelle" <[e-mail]>]

> That's a pretty strong statement :)

One I should’ve clarified. 😃 I don’t mean that the one-RS-per-AT model is not used at all, just that it is not universal and comes with real, practical tradeoffs that may not be appropriate for all use cases. Consequently, we should not design fundamental specs that mandate its adoption.


> …knowledge of that level isn't necessary.

Either the RS and AS have a shared understanding of that level, or the RS is trusting the AS to decide what “AuthenticatedClient” means, and set it accordingly. The latter requires that the RS only supports ASes that have a shared (or substantially similar) understanding of what that level is, which is unlikely outside of a closed system. In that case, there isn’t a lot of value in providing a standard claim, as the closed system could just as easily define a proprietary one.


> Could you expand on a concrete scenario that would lead to an antipattern because of the presence of those claims in the profile?

If I could, then I wouldn’t be concerned about defining this piece of the puzzle without the rest of it. 😃

In your experience, are the existing systems that use these claims reading them from ID Tokens, or from JWT access tokens? Are these typically clients that are accessing their own RSes, using tokens vended by an IdP? (e.g., a SaaS product that gets tokens from a hosted IDaaS product) I ask because OIDC clients can specifiy max_age and acr_values, and if client and RS are owned by the same entity then they can do whatever proprietary thing they want to indicate the reason for an authorization failure.

–
Annabelle Richard Backman
AWS Identity


From: Vittorio Bertocci <[e-mail]>
Date: Tuesday, December 17, 2019 at 12:04 PM
To: "Richard Backman, Annabelle" <[e-mail]>
Cc: Vittorio Bertocci <Vittorio=[e-mail]>, IETF oauth WG <[e-mail]>, "[e-mail]" <[e-mail]>
Subject: Re: [UNVERIFIED SENDER] Re: [OAUTH-WG] I-D Action: draft-ietf-oauth-access-token-jwt-03.txt

The one-RS-per-AT model is an ideal that simply doesn’t match reality.
That's a pretty strong statement :) I have worked with a very large number of developers, on a very large number of applications. I don't dispute that your experience might be different, but the one AT per RS is daily reality for hundreds of thousands of apps in production today. That's real by any measure of reality. The fact that AD offers RTs that behave like TGTs makes it possible.
RAR is an interesting development, but it's in the inception phase. The aud and scopes claims are already in nearly all the proprietary JWT implementations in production today, as shown in the research I presented when the draft was adopted by the WG, hence it seems natural to include them in an interop profile.
Nothing prevents us from developing it further as new approaches gain traction on the market, but for the time being I think the value of this would be to bring some order in existing implementations and contain abuses such as id_tokens used in lieu of access tokens today.

I’m skeptical of the idea that there is just one level of client identity proofing that RSes might care about.
In the scenario that led me to include this item in the discussion, knowledge of that level isn't necessary. Again, I agree it would be useful to define those levels, but I don't think it would be in scope here. If inclusion of that info in the JWT AT would be conditional to do all that work, I'd much rather drop that aspect in favor of making faster progress.

Without taking the whole picture into account, there is significant risk that we will define something that turns out to be woefully insufficient (or worse, establishes an anti-pattern). That said, I agree that it’s probably not appropriate for this draft, and recommend dropping these claims. They can always be defined in a separate draft, along with the other elements necessary to communicate step-up requirements.
Could you expand on a concrete scenario that would lead to an antipattern because of the presence of those claims in the profile?
There is already logic out there predicated on those claim values, precisely because ID_token defines them and vendors latch onto them. I am concerned that if we don't include those claims in the profile, people will simply keep using ID_tokens for calling APIs.


On Tue, Dec 17, 2019 at 11:39 AM Richard Backman, Annabelle <[e-mail]<mailto:[e-mail]>> wrote:
> In any case the intent was always to only allow a singe resource per AT…

My confusion actually comes from the last paragraph of §3, where it says “If a scope parameter is present in the request, the authorization server SHOULD use it to infer the value of the default resource indicator to be used in the aud claim.” I interpreted that to leave room for an AS inferring the same “generic” resource indicator for resources served by different RSes. The one-RS-per-AT model is an ideal that simply doesn’t match reality. Clients don’t want to have to juggle a bunch of different tokens, nor should they need to. I’m coming to think that we shouldn’t actually use `aud` and `scope` in JWT ATs at all, and instead adopt the structure RAR defines used for the `authorization_details` parameter (I think some iteration is required, but it seems natural for these two to describe authorizations in the same or compatible ways).


> …the resource wants to know whether this particular token was obtained proving the identity of the client…

I’m skeptical of the idea that there is just one level of client identity proofing that RSes might care about. There might be some clear lines that can be drawn, but we should be explicit about what those lines represent, e.g., “client authenticated using pre-established credential”, “client is running on end user-controlled device”, etc.


> I am not trying to create a complete framework for handling step up and the like, I am trying to create an interoperable representation of those values no matter how they were obtained.

Without taking the whole picture into account, there is significant risk that we will define something that turns out to be woefully insufficient (or worse, establishes an anti-pattern). That said, I agree that it’s probably not appropriate for this draft, and recommend dropping these claims. They can always be defined in a separate draft, along with the other elements necessary to communicate step-up requirements.

–
Annabelle Richard Backman
AWS Identity


From: Vittorio Bertocci <Vittorio=[e-mail]<mailto:[e-mail]>>
Date: Monday, December 16, 2019 at 9:31 PM
To: "Richard Backman, Annabelle" <[e-mail]<mailto:[e-mail]>>
Cc: IETF oauth WG <[e-mail]<mailto:[e-mail]>>, Vittorio Bertocci <[e-mail]<mailto:[e-mail]>>, "[e-mail]<mailto:[e-mail]>" <[e-mail]<mailto:[e-mail]>>
Subject: [UNVERIFIED SENDER] Re: [OAUTH-WG] I-D Action: draft-ietf-oauth-access-token-jwt-03.txt

Re: aliases, I see where the confusion is coming from!
I updated the request section, but the session 2.2 data structure still mentions the aliases. That should be cleaned up as well.
In any case the intent was always to only allow a singe resource per AT, the alias list was only for helping in cases where an AS identifies the same resource thru multiple IDs and the actual aud value depends on what ID the client requested. However we discussed this with Brian and he convinced me that it was just too ambiguous- your remark reinforces that impression. I’ll clean up 2.2 and eliminate references to aliases from there as well.
Thanks!


On Mon, Dec 16, 2019 at 20:19 Vittorio Bertocci <[e-mail]<mailto:[e-mail]>> wrote:
Thanks Annabelle.

Does a mobile app that uses Dynamic Client Registration to establish a client secret count as an “authenticated client”?
I think it should count, tho I am not sure what the RS would do with it. Dynamic client registration would likely not have much use for this.
In the cases I have seen, the idea is that a client (whose clientID is already known to the RS) can obtain tokens thru different grants, some of them confidential and others public; and the resource wants to know whether this particular token was obtained proving the identity of the client so that it can decide whether to grant extra privileges for that particular client in the current call. This scenario does not require the extra details about authentication method/strength you mention that, I agree, would be pretty hard to reach consensus on.
TL;DR: there is at least one scenario that is satisfied by the simple bool, as the previous knowledge of the client solves the issues you mention out of band.

authentication session properties:
 Let me try another angle. Say that I perform an authz code grant asking for AT, ID_T and RT- obtaining AT', ID_T' and RT'.
The values of auth_time, acr and amr in AT' will be the same as the corresponding claims in ID_T'. When the client uses RT' to obtain AT`N, AT'N+1 etc etc, the values of those claims will remain the same for every AT'n obtained by RT'.
Now, imagine that something happens (ignore what for the time being) that causes the client to perform a step up auth, which requires the user to perform interactive auth hence results in a new authz grant. The client will obtain a new tuple  AT", ID_T" and RT". The exact same rules described for the ' tuple apply, with the new values determined by the new authentication: AT" auth_time/acr/amr will be the same as ID_T", and those values will remain unchanged for every AT"n derived by RT".
If we want this to apply to the implicit flow as well, you can substitute the RT with the session artifact.
Does that help clarifying the intent? If yes, do you feel that the current language does not describe this?

use case for putting these in the access token
I am not trying to create a complete framework for handling step up and the like, I am trying to create an interoperable representation of those values no matter how they were obtained.
Consider the case in which an API allows certain operations only if a given amr value is present in the token. Whether that amr is required upfront on first authentication, as step up or any other reason is out of scope for this profile IMO: the AS might have all sorts of heuristics that determine that (user membership to a given group or role; geofencing; risk scoring; etc) that have nothing to do with scopes requested, and are not statically determined by RS-AS communications.
Now, I agree that formalizing how a RS communicates to the AS via the client the need to step up and in what terms would be super useful; however I think that's well beyond the scope of this profile.. Various vendors already have their own mechanisms for handling step up signaling from RS to AS (I am very familiar with the Microsoft one) and all I am looking for is a way of standardizing how to represent the outcomes, so that API gateways and SDK providers can provide tools that work across vendors; and possibly reuse some of the reasoning that was already done for id_tokens.
TL;DR: I think we are doing something useful in formalizing how to embed that info an ATs even if we don't force vendors to give up their current mechanisms to populate those values.

Regarding access tokens that are used to access different resource servers
The intent is to explicitly disallow the use of an AT with multiple resource servers. If a client wants to talk with 2 different RSes, the client should ask for 2 different ATs.
The spec is unambiguous on this IMO, here's the language I use in 03.


If it receives a request for an access token containing more than one

   resource parameter, an authorization server issuing JWT access tokens

   MUST reject the request and fail with "invalid_request" as described

   in section 4.1.2.1 of [RFC6749]<https://datatracker.ietf.org/doc/html/rfc6749#section-4.1.2.1> or with "invalid_target" as defined

   in section 2 of [ResourceIndicators<https://datatracker.ietf.org/doc/html/draft-ietf-oauth-access-token-jwt-03#ref-ResourceIndicators>].  See Section 2.2<https://datatracker.ietf.org/doc/html/draft-ietf-oauth-access-token-jwt-03#section-2.2> and Section 5<https://datatracker.ietf.org/doc/html/draft-ietf-oauth-access-token-jwt-03#section-5>

   for more details on how this measure ensures there's no confusion on

   to what resource the access token granted scopes apply.
  Is it possible you are referring to an older draft?


On Mon, Dec 16, 2019 at 5:28 PM Richard Backman, Annabelle <richanna=[e-mail]<mailto:[e-mail]>> wrote:
1. Regarding AuthenticatedClient:
Does a mobile app that uses Dynamic Client Registration to establish a client secret count as an “authenticated client”? What if that registration included an assertion signed by a trusted entity (e.g., device management software) using a certificate that had been pushed to the device? Without some kind of NIST LoA-style definition of “authenticated”, a single Boolean “AuthenticatedClient” value is too ambiguous to be useful. However, I’m skeptical that we could find consensus on what that definition should be, and it may be better to define client analogs to `acr` and `amr` that provide standard ways of expressing client authentication details without getting prescriptive about what kind of authentication is “good enough.”

2. Regarding authentication session properties:
I’m confused by the definitions given for `auth_time`, `acr`, and `amr`. For `auth_time`, it says:

“…its value will either remain the same for all the JWT access tokens issued within that session or be updated to the time of latest authentication if reauthentication occurred mid-session…”

But the “For example” at the end of that definition implies that `auth_time` will not be updated. The definitions for `acr` and `amr` say the same, but also say that the “…same considerations presented for auth_time apply…” What’s the intention here: are they fixed, updated, or is it up to the deployment to decide?

I’d like to better understand the use case for putting these in the access token. If the RS is making authorization decisions based on these claims, that implies that the RS cannot rely on the AS to determine (e.g., from the requested scope) the required authentication method(s), freshness, etc. If the AS could be relied upon for this, then presumably it would not issue RTs and ATs for a scope without requiring the end user to meet the appropriate authentication requirements. If this is the case, then the RS must have a way to signal to the client (and then the AS) that a step-up authentication is required. Is there an existing mechanism through which they could do this? All I can come up with is cramming the information into the scope attribute of a WWW-Authenticate challenge, but that has the same scope opacity violation problem as described in this draft’s Security Considerations. Without a clear way to express the step-up requirements, this feels incomplete..

3. Regarding access tokens that are used to access different resource servers:
Reading between the lines, I *think* the intention is that a JWT access token that is intended to be used to access two different resources at two different RSes should look something like:

{
"aud": "https://generic-resource-indicator.example.com/",
"scope": "resource_foo resource_bar",
...
}

And the expectation is that each RS should recognize that generic audience. Is this correct? If so it seems to go against the spirit of resource indicators as indicating the target or location of a resource. It’s stretching that definition, at the very least.

But as I said, I’m reading between the lines here. If this is the intention, it should be clearly stated. Alternatively, remove (or change to a SHOULD) the requirement that multi-value `aud` claims must only contain aliases for the same resource indicator.

–
Annabelle Richard Backman
AWS Identity


From: OAuth <[e-mail]<mailto:[e-mail]>> on behalf of Vittorio Bertocci <Vittorio=[e-mail]<mailto:[e-mail]>>
Date: Monday, December 16, 2019 at 2:51 PM
To: IETF oauth WG <[e-mail]<mailto:[e-mail]>>
Cc: "[e-mail]<mailto:[e-mail]>" <[e-mail]<mailto:[e-mail]>>
Subject: Re: [OAUTH-WG] I-D Action: draft-ietf-oauth-access-token-jwt-03.txt

Dear all,
I finally found the time to push a new draft of the JWT profile for ATs. This version incorporates the feedback kindly provided by Brian and Aaron at IETF105.
Unfortunately I didn't have a chance to participate to IETF106, hence we didn't do much progress since then.
There are still two areas we didn't manage to reach consensus on:

1. Mechanisms for signaling whether the AT was obtained by a user or an application

This point was the subject of intense debate on the list leading to IETF105..
One insight that emerged from the IETF105 discussion was that the use case here is more about whether the AT has been obtained via a grant that authenticated the client (regardless of what specific grant was used) vs a public client grant- that information can be relevant to a resource as it determines whether the identity of the client can be used in authorization decisions (in the former case) or not (in the public client case). If that's the scenario we are truly after, a simple, possibly optional boolean claim (say AuthenticatedClient) would suffice.
Note for the people who didn't read the spec: I removed any reference to this already in draft-ietf-oauth-access-token-jwt-01. I think this is an important and useful info and standardizing it would be beneficial for middleware and SDK creators, but ultimately this is an interop profile hence if there's no strong desire to reach an agreement here, I am also OK with dropping the matter and not address this function in the spec.
To summarize, I have two questions here:
A. Do we believe it's worth it to formalize in the profile a mechanism to indicate whether the client th obtained the AT authenticated itself with the AS?
B. If yes, are you OK wiht an optional bool claim here?

2. Claims indicating session properties

This is one area where I am far more passionate about, as it represents a fundamental aspect that production systems (and in particular 1st party solutions) already rely on in the current proprietary JTW ATs.
The profile currently includes the claims auth_time, acr and amr - capturing the values of those properties at the time of the latest authentication the user performed in the session used to issue the current AT: at session creation, at auth step up time and so on.
Those are values that API need in order to make authorization decisions; in the case of 1st party APIs, the AT is the only artifact they ever receive hence there is no other way for an API to determine whether the caller performed say MFA in order to obtain the current AT.
Since the first draft, some people signaled concerns about the current mechanisms thru which those claims get their value. For example, it has been proposed to maintain the original values of auth_time, acr & amr and introduce additional claims to indicate changes (like stepup).
I am not clear on what cold go wrong with the current approach, hence I don't have an immediate grasp of how changes like the above would help.
If the people uncomfortable with the current proposal could detail their concern, we can brainstorm ways to make that info available to API in a safer way. To summarize:
A. Do you have concerns with the current auth_time, arm, acr proposal? Can you spell them out? (please read the relevant parts of the spec before chiming in :))
B. If yes, what can we do to make that info available to RSs in a way that makes you comfortable?


Thanks

V.

On Mon, Dec 16, 2019 at 2:49 PM <[e-mail]<mailto:[e-mail]>> wrote:

A New Internet-Draft is available from the on-line Internet-Drafts directories.
This draft is a work item of the Web Authorization Protocol WG of the IETF.

        Title           : JSON Web Token (JWT) Profile for OAuth 2.0 Access Tokens
        Author          : Vittorio Bertocci
        Filename        : draft-ietf-oauth-access-token-jwt-03.txt
        Pages           : 16
        Date            : 2019-12-16

Abstract:
   This specification defines a profile for issuing OAuth 2.0 access
   tokens in JSON web token (JWT) format.  Authorization servers and
   resource servers from different vendors can leverage this profile to
   issue and consume access tokens in interoperable manner.


The IETF datatracker status page for this draft is:
https://datatracker.ietf.org/doc/draft-ietf-oauth-access-token-jwt/

There are also htmlized versions available at:
https://tools.ietf.org/html/draft-ietf-oauth-access-token-jwt-03
https://datatracker.ietf.org/doc/html/draft-ietf-oauth-access-token-jwt-03

A diff from the previous version is available at:
https://www.ietf.org/rfcdiff?url2=draft-ietf-oauth-access-token-jwt-03


Please note that it may take a couple of minutes from the time of submission
until the htmlized version and diff are available at tools.ietf.org<http://tools.ietf.org>.

Internet-Drafts are also available by anonymous FTP at:
ftp://ftp..ietf.org/internet-drafts/<ftp://ftp.ietf.org/internet-drafts/>

_______________________________________________
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