W3C home > Mailing lists > Public > public-credentials@w3.org > June 2021

Re: VC HTTP Authorization Conversation

From: Justin Richer <jricher@mit.edu>
Date: Fri, 11 Jun 2021 15:11:28 -0400
Message-Id: <D831BF2D-A581-45A4-AB36-214BDF3BA602@mit.edu>
Cc: Adrian Gropper <agropper@healthurl.com>, Juan Caballero <juan.caballero@spherity.com>, Manu Sporny <msporny@digitalbazaar.com>, W3C Credentials Community Group <public-credentials@w3.org>
To: Mike Varley <mike.varley@securekey.com>
+1 to Mike’s general descriptions of the scoping. 

From my perspective, and what I tried to lay out on the call, there are three touch points between VC-HTTP-API and authorization/delegation systems like OAuth2, GNAP, and UMA. Understanding which, if any, of these you’re talking about at a given time is key to understanding how they can connect to each other. 

On the call this week, I tried to lay out three dimensions in which VC-HTTP-API and GNAP fit together, and for two of these OAuth 2 also is a consideration. I’m going to lay them out here with my recommendation as to what, in my view, the WG should do with that aspect in order to keep things both focused and usable.

1) A means to protect calls to the VC-HTTP-API.

Define access to the newly-defined API in terms of security tokens and the rights associated with them. This is where I think the most value for this WG lies. 

My recommendations:

a) Declare VC-HTTP-API SHOULD be protected by OAuth 2 Bearer tokens or GNAP key-bound access tokens. 
b) When (a) is true, define API access actions in terms of GNAP resource access [1] and OAuth 2 RAR [2] structures. These structures are compatible.
c) Describe general security protections and assumptions for API access.
d) Leave the details of how (a) works, including getting tokens and validating tokens, firmly out of scope.

This is, in my opinion, the best way forward. It’s a step past “just declare it’s an API and don’t worry about security” as it gives some true interoperable guidance, but it’s a far cry from fully defining an integration layer where a universal one isn’t there. It also would allow people to have their own weird access rules if they wanted to, but define the interoperable access (for test suite harnesses and the like) in terms of existing standards. So the test suite would have explicit hooks for doing OAuth2 or GNAP or both, and implementations hooking into it would need to tie to those hooks — but this level of detail is part of the definition of the test harness and not the standard itself.

And in my professional opinion, don’t try to define something with cookies or invent your own access header system. Please.

2) A means to authorize calls to an arbitrary token-protected API.

Define a way for a resource server to use the VC-HTTP-API to retrieve access rights associated with a presented token or VC in a form the resource server can process and act on. This would be like OAuth/GNAP token introspection, but with an added semantic-like data layer provided by VC’s. 

This is where a lot of the discussion about cross-domain authorization seems to be landing. I think it’s interesting work, and there’s room for interoperable definitions here, but it’s a different direction of concern from what this WG is really focusing on. It would be an interesting driving use case for this group, but this group is looking to define the API and not the applications that use it. This would fall into the latter category.

In this bucket I also would put any use of structured tokens to convey VC information to an RS, whether VC-HTTP-API or not. 

3) A means to gain or prove authorization to gain access to a protected API.

Define a way for an authorization server to call a VC-HTTP-API or present a VC-HTTP-API for use in the 

This is a place where GNAP goes past what OAuth 2 can do, as GNAP has a flexible “get users involved” process[3] that can be extended to make use of this API.

I would love to see this defined either as part of this group or as a collaboration with this and the GNAP WG, but it should not be a priority. This is a USE of the API, and not definition of the API. This is why I added it to the use case documents.

 — Justin

[1] https://www.ietf.org/archive/id/draft-ietf-gnap-core-protocol-05.html#name-resource-access-rights
[2] https://www.ietf.org/archive/id/draft-ietf-oauth-rar-05.html#name-authorization-data-elements
[3] https://www.ietf.org/archive/id/draft-ietf-gnap-core-protocol-05.html#name-determining-authorization-a

> On Jun 11, 2021, at 9:49 AM, Mike Varley <mike.varley@securekey.com> wrote:
> Hi Adrian,
> > So people are suggesting that we should do OAuth2 first and then maybe do UMA or GNAP later. The group is unclear whether "later" is in another group and what customers will even care to implement attenuated delegation in the SSI protocols context. W3C might be just as happy for IETF to worry about that.
> The problem for the VC-HTTP-API may be ‘smaller’ then what you are describing. Currently there is a gap in how service providers (issuers/holder/verifiers) consistently, in a standardized way, can generate (or “mint”), verify, or obtain/receive these credentials. The VC-HTTP-API is a (hope to be) standard API that services can integrate with to consistently achieve those functions.
> In other words, system builders have a problem considering the bigger questions you are asking, Adrian, because we haven’t settled on an API for reliably working with credential containers or moving credentials around; let alone worrying about how we got the credential or engaged with a person/holder to begin with.
> Let’s assume we solved for the perfect SSI privacy preserving, delegation authorization and trust model:
> As an issuer, what API do I call to generate a credential for this authorized session?
> As an issuer (with a credential to deliver to a holder), is there an API I can call, or an API the holder can call to transfer the credential?
> As a holder/presenter of credentials, what API do I need to support to collect a credential or present a credential presentation?
> As a verifier, with the holder’s permission, what API to I support to reveive a presentation?
> As a verifier, now that I have a credential presentation, how to I verify it for authenticity (i.e. properly signed by the involved parties/entities)
> ++ (see use cases document)  
> VC-HTTP-API may be a “bottom up” approach, where you (and other interested in the authZ conversation) are approaching more “top down”. I believe others in the group (and I will add my name to the list) want to fully support the discussion you are driving towards with authorization and delegation, but in a sequence of operations, by the time the auth/bootstrapping/delegation/permission/policy decisioning is completed (by issuer or holder or verifier) the VC-HTTP-API just has some “work to be done”.
> For SSI to work the way we all want it to, your questions need to be answered.
> For SSI to work at all (generally), we need a standard way for components to communicate (and VC-HTTP-API is one of these methods).
> Does that help?
> MV
> From: Adrian Gropper <agropper@healthurl.com>
> Date: Thursday, June 10, 2021 at 10:54 PM
> To: Juan Caballero <juan.caballero@spherity.com>
> Cc: Manu Sporny <msporny@digitalbazaar.com>, W3C Credentials Community Group <public-credentials@w3.org>
> Subject: Re: VC HTTP Authorization Conversation
> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
> I'm obviously missing something, and instead of explaining it to me, people are just shouting louder and louder at why I need to get out of the way of progress.  
> Here's what I think I understand:
> ·         The VC-HTTP endpoint needs to be protected by some kind of bearer token from a client or a cookie from a browser.
> ·         This token is subject to some "bootstrapping" protocol that may or may not be in scope for this group.
> ·         Simply stopping with the token is "kicking the can down the road" inadequate for the group.
> ·         So the group wants to do something with OAuth2 that might or might not put the "bootstrap" in scope.
> ·         In (Alice-to-Bob) delegation use-cases where the client presenting the token has not pre-registered with the VC-HTTP server, OAuth2 may not be sufficient.
> ·         In these delegation use cases, the VC-HTTP endpoint would need to be protected by UMA or GNAP. 
> ·         But nobody here is arguing for UMA and GNAP is seen as presenting a delay to progress.
> ·         So folks want Adrian to stop insisting that Bob can be delegated by Alice to access the VC-HTTP endpoint because all of their use cases are fine with Alice and Alice's client always being an intermediary holder for the VC.
> I'm OK if the group wants to keep "bootstrap", and OAuth2 along with it, out of scope but that does not seem to be acceptable to some.
> So people are suggesting that we should do OAuth2 first and then maybe do UMA or GNAP later. The group is unclear whether "later" is in another group and what customers will even care to implement attenuated delegation in the SSI protocols context. W3C might be just as happy for IETF to worry about that.
> How'm I doing?
> - Adrian
> On Thu, Jun 10, 2021 at 5:37 PM Juan Caballero <juan.caballero@spherity.com <mailto:juan.caballero@spherity.com>> wrote:
> Inline
> On 6/10/2021 4:52 PM, Manu Sporny wrote:
> On 6/10/21 10:04 AM, Orie Steele wrote:
> I see the scope of the VC-HTTP-API as limited to the following:
> 1. HTTP endpoints for producing and consuming the W3C VC Data Model Objects
> 2. HTTP endpoints for producing and consuming extensions to the VC Data Model,
> or data structures it relies on, like Revocation Status and Credential Schemas
> 3. Recommendations for securing these HTTP endpoints using HTTP Headers.
> 4. Recommendations for exposing these HTTP endpoints inside and across trust
> domains.
> The above feels like a good start to a scoping statement.
> +1 from the use-case team.  We are coming up fast on the deadline for "first drafts" of use cases that are far enough along to anchor PRs. If you have opinions about #3 and #4 (especially an opinion like, "I would hate to see mTLS discouraged as a valid option for #3/4"), a quick way to lobby for that opinion is to submit a first-draft use case.  Feel free to "fork," riff, or expand on other peoples' submissions that are halfway to what you want to see, but too brief, too specific in in describing a solution rather than a problem, etc.  
> Perhaps it would be better to start a CCG work item dedicated to GNAP and VCs?
> +1 to this, Adrian clearly wants to have this discussion, and polling of the
> VC HTTP API group is showing that there is not a lot of support for doing that
> in that group:
> https://w3c-ccg.github.io/meetings/2021-06-01-vchttpapi/#57 <https://w3c-ccg.github.io/meetings/2021-06-01-vchttpapi/#57>
> A more focused group on how one gets authorization tokens to use with the VC
> HTTP API using GNAP might result in 1) a more focused discussion on that
> topic, and 2) a tighter scope for the VC HTTP API.
> Agreed that a tighter scope for this group will get results faster from this group.
> Or a battle royal between OAuth2.0 the only "real" standard being discussed
> and the drafts for GNAP, OIDF-SIOP, ZCaps with HTTP Signatures and DIDComm
> hosted by the ccg?
> I expect this to turn into painful discussion for everyone involved. While I'm
> pretty sure Orie was joking about this option, -1 to the concept of comparing
> half-baked things against each other when we're trying to make some concrete
> decisions for the VC HTTP API.
> Agreed that this group needs uncontroversial, concrete results, not design constraints imported from discusssions about standards or technologies still in flux. The people volunteering to make this API testable and get it taken seriously by today's vendors of production-grade software are not volunteering to build for the future, they are volunteering to build for their paying clients and customers.
> I'd prefer we either recommend OAuth2.0 and leave the door open to other ways
> to get an authorization token... 
> +1 to this proposal.
> Everyone is unanimous in leaving the door open.  Parties asking for multiple recommendations have not volunteered to do the extra work; we have volunteers committed to doing the OAuth2 work. 
> or we label the vc-http-api agnostic to
> authorization, and warn users that they will need to pair it with other specs
> to address security issues associated with exposing http endpoints on the
> internet.
> -1 to this proposal, we have kicked the "authorization can" down the road long
> enough... 18+ months now... let's not keep doing it when there is a fairly
> clear solution that all implementers seem to feel is a viable path forward as
> ONE option for authorization.
> -1 as well, ONE viable option is a stronger proposition than 0 viable options, 1.5, or even 1.9.
> I would also love to just do Issue and PR review, and have less open
> discussion on calls, we seem to be discouraging asynchronous work, maybe we
> can limit open discussion to 50% of the call time?
> Yes, agreed. Perhaps we start out with PR review first -- because that rewards
> the thing we want to see, which is work being done.
> Then issue review, because it rewards the second thing we want to see -- which
> is asynchronous work.
> +1 discussing possible futures for this spec is taking up way too much time.  Given the timeline for use-cases (and my own commitment to be at all meetings until they are stabilized), I'd propose that Tuesday's agenda be:
> 1.) PRs
> 2.) Issues
> 3.) Use-cases 
> However, the Authorization discussion HAS been processing issues. Namely:
> https://github.com/w3c-ccg/vc-http-api/issues/108 <https://github.com/w3c-ccg/vc-http-api/issues/108>
> https://github.com/w3c-ccg/vc-http-api/issues/181 <https://github.com/w3c-ccg/vc-http-api/issues/181>
> https://github.com/w3c-ccg/vc-http-api/issues/187 <https://github.com/w3c-ccg/vc-http-api/issues/187>
> I try very hard not to do open discussion that's not driving toward some sort
> of PROPOSAL or POLL at the end. What I've been trying to do is to make sure
> that we get to at least one concrete proposal on each call. Sometimes you
> can't get there w/o discussion, which is the case with authorization... but to
> be clear, we have been processing registered issues, even though we haven't
> been calling them out by issue number.
> Thanks for that.  Manu, you've been doing great work moving us through the issues as efficiently as possible given the lack of consensus on scope. 
> -- manu
> ----------------- 
> Juan Caballero 
> Research Lead, 
> Spherity <https://www.spherity.com/> GmbH, Emil-Figge-Straße 80, 44227 Dortmund 
> Co-chair, DIF Healthcare Use-cases Discussion Group <https://www.eventbrite.com/o/dif-healthcare-special-interest-group-31363301995> and DIF Interoperability Working Group <https://github.com/decentralized-identity/interoperability/blob/master/agenda.md> 
> Berlin-based: +49 1573 5994525 Signal/whatsapp: +1 415-3101351
> This email and any attachments are for the sole use of the intended recipients and may be privileged, confidential or otherwise exempt from disclosure under law. Any distribution, printing or other use by anyone other than the intended recipient is prohibited. If you are not an intended recipient, please contact the sender immediately, and permanently delete this email and its attachments.

Received on Friday, 11 June 2021 19:12:47 UTC

This archive was generated by hypermail 2.4.0 : Thursday, 24 March 2022 20:25:16 UTC