- From: Alan Karp <alanhkarp@gmail.com>
- Date: Tue, 1 Jun 2021 20:51:45 -0700
- To: W3C CCG Chairs <w3c.ccg@gmail.com>
- Cc: "W3C Credentials CG (Public List)" <public-credentials@w3.org>
- Message-ID: <CANpA1Z3mW33bAr8tTzLdnZubgskxQ_6zresMP2VXOGyMMQFwUg@mail.gmail.com>
I just read through the minutes and looked at the link to the Solid use cases. I hope I interpreted the conversation correctly. As far as the work plan goes, OAuth has some known problems, GNAP isn't ready, and zcap-ld isn't widely used. Fortunately, I think you can make progress without committing to any particular approach up front since all capability systems have a lot in common. Any capability system must be able to support attenuated redelegation, e.g., Alice with read/append/write permission delegates read/append to Bob who delegates read to Carol, and Carol can read even though the resource server never heard of her. Including revocation is a bonus. I don't think you should base an authorization design on the Solid use cases; they are reasonable but too simple. They mostly involve one party making requests specifying a single resource. Almost any permission scheme can be made to work in such cases. You need at least two parties making requests on multiple resources to stress the design. Norm Hardy's confused deputy, https://crypto.stanford.edu/cs155old/cs155-spring09/papers/ConfusedDeputy.html, is such a use case. Although the example seems contrived today, both Cross Site Request Forgery and Server Side Request Forgery are confused deputy vulnerabilities. "ACLs Don't", https://agoric.com/assets/pdf/papers/acls-dont.pdf, goes into more detail. I'm particularly fond of the transitive access problem use case, https://www.hpl.hp.com/techreports/2008/HPL-2008-204R1.html because no other access control approach works for it. -------------- Alan Karp On Tue, Jun 1, 2021 at 4:54 PM W3C CCG Chairs <w3c.ccg@gmail.com> wrote: > Thanks to aaron_coburn for scribing this week! The minutes > for this week's Verifiable Credentials HTTP API telecon are now available: > > https://w3c-ccg.github.io/meetings/2021-06-01-vchttpapi > > Full text of the discussion follows for W3C archival purposes. > Audio from the meeting is available as well (link provided below). > > ---------------------------------------------------------------- > Verifiable Credentials HTTP API Telecon Minutes for 2021-06-01 > > Agenda: > > https://lists.w3.org/Archives/Public/public-credentials/2021Jun/0000.html > Topics: > 1. Use Cases Update > 2. Authorization Part Deux > Organizer: > Manu Sporny > Scribe: > aaron_coburn > Present: > Manu Sporny, Mike Varley, Markus Sabadello, Eric Schuh, Aaron > Coburn, Henry Story, Adrian Gropper, Andreas Freund, Sanuja, > Diwala, Brent Shambauh, Juan Caballero, Ted Thibodeau, Anil John, > Orie Steele, David Ward > Audio: > https://w3c-ccg.github.io/meetings/2021-06-01/audio.ogg > > aaron_coburn is scribing. > Manu Sporny: Welcome to the weekly HTTP API call. This is the > new time for the call > ... today's agenda: continuation of authZ discussion > ... possibly some preliminary descisions if there is consensus > <tallted> confirming -- delete repeating Thursday 3pm ET, > effective this week? > ... are there any updates to the agenda? > > Topic: Use Cases Update > > Eric Schuh: Update on use cases document > ... the new date is june 14, two weeks from today > Eric Schuh: Deadline for new use cases: June 15 > <tallted> s/june 14/june 15/ > ... more use cases have been added > ... many are single-sentence, so we are looking forward to > expanding those > Juan Caballero: We have been leaving comments on some longer use > cases > ... with clarifying questions > Manu Sporny: Please add more if you have interesting use cases > ... any other status updates? > Juan Caballero: One of Mike's use case: he already answered a > clarifying question; it might be good to walk through one of his > use cases at the end (Trust Agent) > ... whether the trust agent is an api detail > Manu Sporny: Any other use cases for the agenda? > > Topic: Authorization Part Deux > > ... if not we will get started with Authorization > Mike Varley: +1 To tackle the Trust Agent use case if there is > time and determine if it is in scope. > ... to summarize last call: what is in scope / out of scope > ... don't put future authZ schemes out of scope (GNAP/ZCAP/etc) > ... one suggestion to put such things as GNAP out of scope > initially > ... one suggestion to focus on OAuth2 > ... today: see if we can get to a proposal > ... is this summary accurate? > Adrian Gropper: OAuth2 is about authorization but not delegation > ... how do we determine how delegation is in scope/out of scope > ... esp. w/r/t self-soverign > ... do we need to include UMA? > Mike Varley: Not about delegation > ... clarify: is this for system-to-system (e.g. > client_credentials) > ... or both? > ... or personal interaction w.r.t consent? > Manu Sporny: Several layers to this. Most focus has been on > client_credentials > ... for general access control (simple case) OAuth2 makes sense > ... for those endpoints that could be abused w/o an authZ layer > ... or: for those endpoints (such as presentation) one doesn't > need authorization > ... those very much in scope > ... delegation is trickier: what are you delegating? > ... we need a concrete use case > .... how does one doe that w/ OAuth2 > ... how does one do that with other mechanisms > Andreas: pull vs. push use case > Manu Sporny: Is it assumed that authZ is required to do that? > Andreas: assumption is that authZ is needed > ... e.g. I have a presentation to make > ... here is domain/endpoint/challenge > Manu Sporny: There is a "start workflow" use case like this > ... is there authZ required on this endpoint, or do you just > start > Andreas: is there something out of band in the communication? > Henry Story: I was wondering how the Authorization fits with the > Solid Authorization use cases we put together > > https://solid.github.io/authorization-panel/authorization-ucr/#uc-privacy > <andreas_freund> @aaron ... there is the assumption of out of > band communication to establish a trust relationship between > requester and target systems > Manu Sporny: Delegation -- what use case affects the HTTP API? > <andreas_freund> that is what i meant > Thanks @andreas > <bblfish> ok, changed browser > Adrian Gropper: Receive direct from the issuer or if the message > needs to be encrypted so the holder cannot read the data > ... streaming of credentials -- could be discarded by the > holder > ... cases where revocation mechanisms are not available > Henry Story: Solid authorization use cases -- these could be > helpful > ... possibly the terminology is different > Manu Sporny: Could you send your message to the CCG mailing list > ... could you provide a summary now? > Henry Story: Read access / write access to a resource > ... use cases for credentials (over a certain age) > ... how a client would know what credentials it has to give > ... server doesn't want to leak too much information > Manu Sporny: Some folks see the HTTP API is an ecosystem API > when in fact it is more focused on issuing VCs > ... the UCs provided by Henry might be more appropriate for > confidential wallets > ... looking at the endpoints themselves, we need to ask whether > certain endpoints need authZ or delegation > ... seems that @adrian's UCs relate to the content of the data > Adrian Gropper: The VC is a resource > ... what is manu asking? There is just the endpoint that will > accept/produce the VC. > Manu Sporny: We need to be more specific about what these > endpoints do if we want to talk about authZ > ... otherwise we'll keep going around in circles > ... one possiblity: we talk about each endpoint and discuss > what the endpoitn needs > Andreas: need to distinguish b/t authZ of the endpoint and authZ > of the resource > ... can the target system be called by the requestor system? > ... I have SAP (A) on one side and Oracle (A) > ... if oracle (B) doesn't know that SAP (B) isn't delegated by > SAP (A), it would reject that request > ... but we would want a delegation model that would allow these > sorts of interactions > Adrian Gropper: We have HTTP in the specification name > ... one brings a token that allows one to call that resource > ... what is the issue about multiple endpoints? HTTP says it > all > Andreas: the different endpoints do different things: some are > public some are not > Adrian Gropper: Does the resource owner have ability to delegate > to another entity > ... difference b/t oauth2 and uma > Andreas: e.g. SAP (A) delegates to SAP (B) > Andreas: system a requests a resource (a presentation) > Manu Sporny: I think the kind of delegation adrian is talking > about is out of scope > ... by "resource", the language adrian is using could be > interpreted in several ways > ... if you're reasoning about the VC, that is out of scope > Adrian Gropper: Agreed > Manu Sporny: Can you hit the endpoint or not > ... if so, you can issue anything > ... unless we have a very fine-grained authZ framework > ... two levels: can we do the simple stuff (OAuth2) > ... e.g. direct system-to-system calls > ... after that, what kind of delegation do we want to support > ... keep the complex authZ mechanisms in mind > ... but keep the simple use cases front and center > Adrian Gropper: OAuth uses string to determine scope > ... in confidential storage WG: does the subject have the right > to delegate access > ... not whether the subject has the ability to read/write > ... sees this as a human rights issue > Manu Sporny: I will try to pose the issue differently > ... I don't understand the human rights issue > ... can we talk about this as a priorities thing? > ... start with the simple tasks > ... move the complex items to later (while keeping them in > mind) > ... would there be any objections to prioritizing the simple > stuff > <adrian_gropper> Yes - I would object. > Adrian Gropper: Object on the grounds of 10 years of experience > of the disaster caused by oauth in the health care industry > ... colleagues have all moved from OAuth to GNAP > ... reduced to login with Google/FB > ... OAuth needs to go away > Orie Steele: -1 To moving from an establshed standard to a DRAFT > that is WIP. > ... serves B2B use cases > ... doesn't help self-soveign protocols > Manu Sporny: Anyone to speak to OAuth2 use cases > Ted Thibodeau: OAuth is not reduced to those two providers > ... we have an auth layer that allows dozens of providers to be > used > ... argument is spurious > ... many have chosen to leave OAuth, but the reasoning is > flawed > Orie Steele: As a standard, OAuth has been adopted by a large > number of providers > ... it's not a draft or work item > ... it has flaws > <bblfish> On the whole I am on Adrian's side, I don't think it > OAuth for Solid is really the right tool for example. > ... many of the design decisions have caused trouble for OAuth > ... but it has also stood up over time > ... there isn't a better solution today for OAuth > ... there are social issues ("nascar problem") but that doesn't > mean that OAuth isn't a foundation worth building on > ... starting with OAuth could, in fact, help this effort to > take off > ... it has been very difficult to replace OAuth > Manu Sporny: I think we're conflating things > <orie> sure OIDC is built on OAuth.... > ... when we say OAuth2 means "login with Google/FB", that's > OIDC > ... I have the same issue with that > ... that is not what we're talking about > <tallted> OIDC is also not trapped in a G/FB coin-flip > <orie> ^ bingo > ... when we mean applying auth to these endpoints > ... we are not talking about integrating OIDC into these > endpoints > ... if people are objecting to SSO, that is not what we are > talking about > ... we are talking about the underlying OAuth protocol > Orie Steele: Take your objects to ODIC to the OIDF :) I hear they > have their own way of doing SIOP / SSI :) > Adrian Gropper: The problem as manu stated, is not with OIDC. > The problem is with registration > <tallted> false. > <orie> false > Manu Sporny: How is it incompatible? > https://github.com/go-oauth2/redis > Adrian Gropper: The service provider needs a process by which > they issue a client id and that process involves some sort of > identity verification > Ted Thibodeau: Proof of falsity: http://youid.openlinksw.com/ > ... e.g. if you have a valid credit card, we will issue X to > you > <tallted> q > <orie> there is a difference between "authenticating an > application" and a "human being".... > ... the reason OAuth only works in B2B case is because of the > client registration requirement > Manu Sporny: I do not agree > ... one can set up your own server without needing to use any > of the stuff described > ... to get an oauth client_id and secret, all you need is a > webpage > ... I do agree about the use of OIDC to force the use of large > providers > Ted Thibodeau: The UID link above allows one to set up with > sufficient crypto to publish the public portion of your sign-in > material > ... there is no reason to have to use the big 2 or 3 providers > Adrian Gropper: To Orie's point: the reason for my objection > (and the reason to do GNAP in parallel): safety is not the issue. > The issue is about privacy > ... the barrier to client registration is a barrier to adoption > ... when client registration is introduced, you create a system > that is unfair to the subjects > <juan_caballero_(dif/spruce)> but what if all the subjects of the > VCs in question are inanimate objects and batches of steel or > coal? > ... the argument is based on privacy not security > <juan_caballero_(dif/spruce)> OAuth is being used to authenticate > servers which are passing between them VCs about rocks > Ted Thibodeau: You rejected my argument because of the last > sentence. > Adrian Gropper: I provided an example that oauth is successful > in B2B b/c of client registration > Ted Thibodeau: Dozens of instances where a verifiable identity > does not require a B2B registration > Manu Sporny: Make sure no one gets too passionate and that there > are no personal attacks > ... would like to collect some data (poll not a decision) > ... just a +1/-1 poll > ... start with adrian's proposal > ... to prioritize working with OAuth2 and GNAP simulaneously > ... just getting feedback > ... there will be several pos > Manu Sporny: POLL: Prioritize working on OAuth2 and GNAP > simultaneously when adding authorization to VC HTTP API > endpoints. > ... polls > Adrian Gropper: +1 > Orie Steele: -1 > Manu Sporny: -1 > <tallted> -.9 > <markus_sabadello> +0.5 > <mike_varley> -1; I do not feel GNAP is ready to be worked with > yet. But I hope it gets there soon > <eric_schuh> 0 > <bblfish> ISn't GNAP a 100 page proposal that is only just > started being worked on at the IETF? > <david_ward> -0.1 > Anil John: -1 As GNAP is still too immature > <orie> my reason for -1 is that both GNAP is not stable enough > to use in production today, and its additional complexity which > is orthogonal to our mission > Manu Sporny: GNAP is in early days at IETF > ... counter proposal: to prioritize working with OAuth2 > Anil John: XACML :-) > <markus_sabadello> XDI link contracts! > Manu Sporny: POLL: Prioritize working on OAuth2 for authorization > to VC HTTP API endpoints as a priority 1 item and enable the use > of other authorization mechanisms like GNAP, ZCAPs, etc. as a > priority 2 item. > <adrian_gropper> - 1 > Orie Steele: -1 To getting tangled in lower priority work > <tallted> wording bites me... > Mike Varley: +1 To OAuth 2.0 client credentials, +0.1 for user > authorization > <markus_sabadello> 0 > <eric_schuh> 0 > <orie> "lower priority" means a license to distract and waste > call time, but no commitment/ > Juan Caballero: What do the priority numbers mean? > Ted Thibodeau: +1 Presuming *enabling* other authz potential in > future is not blocked, is actively worked toward > <orie> at least to me > Manu Sporny: Very hand-wavy > Juan Caballero: Recalls concerns about how OAuth2 becomes > possible but infeasable at scale > <orie> I would be supportive of "not blocking future solutions" > and "not spending time on them other than when they are at risk > or having the door shut on them". > Manu Sporny: What we are seeing is that there is a preference > for OAuth2 > ... we are at the top of the hour > <tallted> *nods* yes, Orie > ... we will come back to this next week > ... adrian can you come up with a proposal that would get > consensus? > <orie> I suspect that HTTP headers are not going away anytime > soon. > ... we can start with one endpoint > ... e.g. verify endpoint -- what authZ should it support? > <david_ward> Can the technology used be kicked down the road a > bit? Is it actually important at this time until how the end > points fit the use cases are worked out? > ... we will put aside time for use cases next wee > ... thank you everyone for your engagement > <bblfish> ciao > Ted Thibodeau: +1 David_Ward > > >
Received on Wednesday, 2 June 2021 03:53:29 UTC