Re: [OAUTH-WG] [EXTERNAL] Re: Benjamin Kaduk's No Objection on draft-ietf-oauth-jwsreq-26: (with COMMENT)
I agree with Brian here, I think “typ”:”JWT” should be permitted as well as no typ and “typ”: "oauth.authz.req+jwt". There are other tests we could write for JAR that an OIDC server will fail (for example, one that tested the behaviour of passing a value only outside the request object - which an OIDC server would process, but one compliant with JAR [or FAPI-RW] would ignore the value outside). I don’t see having one more case of this as an issue - some servers will not be JAR compliant, and hence would fail tests that test JAR-specific behaviour. I also agree with Brian about requiring AS to reject request objects that have a sub that’s a client_id, this doesn’t seem to cause significant interoperability concerns as it is hopefully unlikely that any client is doing so. I could possibly see an argument for weakening the requirement for an AS to reject a request object with sub=client_id to a ’should’ (rather than a must) given there is a small chance it could end up breaking an ecosystem, but I’m not entirely convinced. Joseph > On 17 Aug 2020, at 23:55, Brian Campbell > wrote: > > I might suggest that thinking about it in the context of interoperability > would be more meaningful than certification tests. > > Saying that an AS MUST reject the Request object if it has a typ header and > the value of the header is not ‘oauth.authz.req+jwt’ [1] should allow for > interoperability with respect to JWT typing between all combinations of > existing and updated clients with existing and updated authorization servers. > > Saying that an AS MUST NOT include a sub with client id as the value would > break for an updated authorization server when receiving such a request > object JWT. But that's erroneous and potentially dangerous behaviour from the > client so I don't know that we need to try and maintain interoperability > there. > > [1] Unfortunately "typ":"JWT" would probably also need to be allowed. As best > I understand it "typ":"JWT" basically says "this JWT is a JWT", which isn't > useful for explicit typing and I think makes it effectively equivalent to an > untyped JWT. I've honestly never understood why one would use "typ":"JWT" but > it shows up in a lot of places including examples and explanations on sites > like jwt.io <http://jwt.io/> so it seems very likely that it'd just get > copied over and show up in some amount of real world request object JWTs. > > > On Sat, Aug 15, 2020 at 10:41 AM Mike Jones > <mailto:40microsoft@dmarc.ietf.org>> wrote: > Answering Filip and Vladirmir’s question about adding normative language > around “typ” and “sub”: Anytime you add a new required feature, you are > breaking existing deployments. Suppose we added the normative requirement > “If a ‘typ’ header parameter is present, ASs MUST reject the Request object > if its value is not ‘oauth.authz.req+jwt’”. One could then write a > certification test sending the AS a different “typ” value – which to pass, > ASs would have to reject the JWT. Every existing deployment would fail this > test! That’s exactly what we don’t want to have happen. > > > > Brian asked for security considerations. The IESG asked for security > considerations. I added them in the PR – working with Nat and John. They > point the way to the future without breaking existing deployments. That’s as > it should be. > > > >-- Mike > > > > From: OAuth mailto:oauth-boun...@ietf.org>> On > Behalf Of Warren Parad > Sent: Saturday, August 15, 2020 9:27 AM > To: Vladimir Dzhuvinov <mailto:vladi...@connect2id.com>> > Cc: oauth mailto:oauth@ietf.org>> > Subject: [EXTERNAL] Re: [OAUTH-WG] Benjamin Kaduk's No Objection on > draft-ietf-oauth-jwsreq-26: (with COMMENT) > > > > In the case of > > if the Request Object includes a sub claim with the value of the client_id > the AS MUST reject the request > > > > What would the expectation be in terms of a client_credentials grant? > > > > From experience, the sub is frequently populated with the client_id value and > the client_id is not used. Which would mean breaking for that type of grant > wouldn't it? > > > > > Warren Parad > Founder, CTO > Secure your user data and complete your authorization architecture. Implement > Authress <https://bit.ly/37SSO1p>. > > > > > > On Sat, Aug 15, 2020 at 11:08 AM Vladimir Dzhuvinov <mailto:vladi...@connect2id.com>> wrote: > > +1 to make the "typ" check, as Filip suggested, normative, as existing client &g
Re: [OAUTH-WG] [EXTERNAL] Re: Benjamin Kaduk's No Objection on draft-ietf-oauth-jwsreq-26: (with COMMENT)
I might suggest that thinking about it in the context of interoperability would be more meaningful than certification tests. Saying that an AS MUST reject the Request object if it has a typ header and the value of the header is not ‘oauth.authz.req+jwt’ [1] should allow for interoperability with respect to JWT typing between all combinations of existing and updated clients with existing and updated authorization servers. Saying that an AS MUST NOT include a sub with client id as the value would break for an updated authorization server when receiving such a request object JWT. But that's erroneous and potentially dangerous behaviour from the client so I don't know that we need to try and maintain interoperability there. [1] Unfortunately "typ":"JWT" would probably also need to be allowed. As best I understand it "typ":"JWT" basically says "this JWT is a JWT", which isn't useful for explicit typing and I think makes it effectively equivalent to an untyped JWT. I've honestly never understood why one would use "typ":"JWT" but it shows up in a lot of places including examples and explanations on sites like jwt.io so it seems very likely that it'd just get copied over and show up in some amount of real world request object JWTs. On Sat, Aug 15, 2020 at 10:41 AM Mike Jones wrote: > Answering Filip and Vladirmir’s question about adding normative language > around “typ” and “sub”: Anytime you add a new required feature, you are > breaking existing deployments. Suppose we added the normative requirement > “If a ‘typ’ header parameter is present, ASs MUST reject the Request object > if its value is not ‘oauth.authz.req+jwt’”. One could then write a > certification test sending the AS a different “typ” value – which to pass, > ASs would have to reject the JWT. *Every existing deployment would fail > this test!* That’s exactly what we don’t want to have happen. > > > > Brian asked for security considerations. The IESG asked for security > considerations. I added them in the PR – working with Nat and John. They > point the way to the future without breaking existing deployments. That’s > as it should be. > > > >-- Mike > > > > *From:* OAuth *On Behalf Of * Warren Parad > *Sent:* Saturday, August 15, 2020 9:27 AM > *To:* Vladimir Dzhuvinov > *Cc:* oauth > *Subject:* [EXTERNAL] Re: [OAUTH-WG] Benjamin Kaduk's No Objection on > draft-ietf-oauth-jwsreq-26: (with COMMENT) > > > > In the case of > > if the Request Object includes a sub claim with the value of the client_id > the AS MUST reject the request > > > > What would the expectation be in terms of a client_credentials grant? > > > > From experience, the *sub *is frequently populated with the client_id > value and the client_id is not used. Which would mean breaking for that > type of grant wouldn't it? > > > > *Warren Parad* > > Founder, CTO > > Secure your user data and complete your authorization architecture. > Implement Authress <https://bit.ly/37SSO1p>. > > > > > > On Sat, Aug 15, 2020 at 11:08 AM Vladimir Dzhuvinov < > vladi...@connect2id.com> wrote: > > +1 to make the "typ" check, as Filip suggested, normative, as existing > client and RP deployments with undefined "typ" will not be affected. New > deployments should be encouraged to type the JWT, and thus be made safer. > > > > Regarding the "sub != client_id" check -- could a simple rejection of all > JWTs with "sub" present suffice? > > I find it difficult to imagine what else a client could end up setting the > "sub" claim to, if it does end up populating it for some reason. > > Rejecting JWTs with "sub=client_id" or "sub" present will break > deployments where a client for some reason sets the "typical" JWT claims, > and "sub" is a typical one, but if those deployments happen to accept > client_secret_jwt or private_key_jwt client authentication, they could well > be vulnerable to cross-JWT confusion attacks. > > > > Vladimir > > On 14/08/2020 13:58, Filip Skokan wrote: > > Hi Mike, Nat, > > > > I thought we would go as far as making these normative requirements > >- if the Request Object includes a sub claim with the value of the >client_id the AS MUST reject the request >- if the Request Object is explicitly typed (typ) its value MUST be > > First rejects client assertions to be passed as Request Objects. Second > rejects all future typed JWT profiles from being used as Request Objects > without worrying about the claims they may or may not con
Re: [OAUTH-WG] [EXTERNAL] Re: Benjamin Kaduk's No Objection on draft-ietf-oauth-jwsreq-26: (with COMMENT)
Answering Filip and Vladirmir’s question about adding normative language around “typ” and “sub”: Anytime you add a new required feature, you are breaking existing deployments. Suppose we added the normative requirement “If a ‘typ’ header parameter is present, ASs MUST reject the Request object if its value is not ‘oauth.authz.req+jwt’”. One could then write a certification test sending the AS a different “typ” value – which to pass, ASs would have to reject the JWT. Every existing deployment would fail this test! That’s exactly what we don’t want to have happen. Brian asked for security considerations. The IESG asked for security considerations. I added them in the PR – working with Nat and John. They point the way to the future without breaking existing deployments. That’s as it should be. -- Mike From: OAuth On Behalf Of Warren Parad Sent: Saturday, August 15, 2020 9:27 AM To: Vladimir Dzhuvinov Cc: oauth Subject: [EXTERNAL] Re: [OAUTH-WG] Benjamin Kaduk's No Objection on draft-ietf-oauth-jwsreq-26: (with COMMENT) In the case of if the Request Object includes a sub claim with the value of the client_id the AS MUST reject the request What would the expectation be in terms of a client_credentials grant? From experience, the sub is frequently populated with the client_id value and the client_id is not used. Which would mean breaking for that type of grant wouldn't it? [https://lh6.googleusercontent.com/DNiDx1QGIrSqMPKDN1oKevxYuyVRXsqhXdfZOsW56Rf2A74mUKbAPtrJSNw4qynkSjoltWkPYdBhaZJg1BO45YOc1xs6r9KJ1fYsNHogY-nh6hjuIm9GCeBRRzrSc8kWcUSNtuA] Warren Parad Founder, CTO Secure your user data and complete your authorization architecture. Implement Authress<https://bit.ly/37SSO1p>. On Sat, Aug 15, 2020 at 11:08 AM Vladimir Dzhuvinov mailto:vladi...@connect2id.com>> wrote: +1 to make the "typ" check, as Filip suggested, normative, as existing client and RP deployments with undefined "typ" will not be affected. New deployments should be encouraged to type the JWT, and thus be made safer. Regarding the "sub != client_id" check -- could a simple rejection of all JWTs with "sub" present suffice? I find it difficult to imagine what else a client could end up setting the "sub" claim to, if it does end up populating it for some reason. Rejecting JWTs with "sub=client_id" or "sub" present will break deployments where a client for some reason sets the "typical" JWT claims, and "sub" is a typical one, but if those deployments happen to accept client_secret_jwt or private_key_jwt client authentication, they could well be vulnerable to cross-JWT confusion attacks. Vladimir On 14/08/2020 13:58, Filip Skokan wrote: Hi Mike, Nat, I thought we would go as far as making these normative requirements * if the Request Object includes a sub claim with the value of the client_id the AS MUST reject the request * if the Request Object is explicitly typed (typ) its value MUST be ... First rejects client assertions to be passed as Request Objects. Second rejects all future typed JWT profiles from being used as Request Objects without worrying about the claims they may or may not contain. Or is that breaking? S pozdravem, Filip Skokan On Fri, 14 Aug 2020 at 00:59, Mike Jones mailto:40microsoft@dmarc.ietf.org>> wrote: At Nat's request, I've created a pull request addressing Cross-JWT Confusion security considerations. It addresses both Brian's comment and the IESG comments about explicit typing. See the full PR at https://bitbucket.org/Nat/oauth-jwsreq/pull-requests/10. See the source diffs at https://bitbucket.org/Nat/oauth-jwsreq/pull-requests/10/address-iesg-and-working-group-comments/diff#chg-draft-ietf-oauth-jwsreq.xml. Please review! This is only the first commit, albeit, one that addresses some of the must substantive issues. More commits will follow addressing additional IESG comments. -- Mike -Original Message- From: OAuth mailto:oauth-boun...@ietf.org>> On Behalf Of Benjamin Kaduk Sent: Thursday, August 13, 2020 2:59 PM To: Brian Campbell mailto:bcampb...@pingidentity.com>> Cc: draft-ietf-oauth-jws...@ietf.org<mailto:draft-ietf-oauth-jws...@ietf.org>; oauth-cha...@ietf.org<mailto:oauth-cha...@ietf.org>; The IESG mailto:i...@ietf.org>>; oauth mailto:oauth@ietf.org>> Subject: Re: [OAUTH-WG] Benjamin Kaduk's No Objection on draft-ietf-oauth-jwsreq-26: (with COMMENT) Oops, that's my bad. Thanks for the correction -- I've linked to your message in the datatracker (but didn't bother to have the datatracker send a third copy of my updated-again ballot position). -Ben On Thu, Aug 13, 2020 at 03:00:33PM -0600, Brian Campbell wrote: > While some discussion of why explicit typing was not used might b