324 views
# OAuth WG Interim Meeting ## Date November 3rd, 2021, 12:00pm EDT ## Topic - Token Chaining Presenter: Kelley Burgin Draft: [Token Chaining - Single ICAM](https://www.mitre.org/publications/technical-papers/token-and-identity-chaining-between-protected-resources-in-a-single-icam-ecosystem-using-oauth-token-exchange) Draft: [Token Chaining - Multiple ICAMs](https://www.mitre.org/publications/technical-papers/token-and-identity-chaining-between-protected-resources-in-a-multiple-icam-ecosystem-using-oauth-token-exchange) Slides: [Token Chaining](https://datatracker.ietf.org/meeting/interim-2021-oauth-15/materials/slides-interim-2021-oauth-15-sessa-token-chaining-overview-00) # Notes Note taker: Justin Richer ## Presentation: Token and Identity Chaining - MITRE published two profiles for identity briding for OAuth 2.0 & OIDC - "Enterprise Mission Tailored Profiles" - Leverage existing environments and previous work Token Chaining: - Problem: client calls PR1, PR1 needs to call PR2, etc. - two cases: all PRs trust one AS, PRs trust different AS's Solution in single ICAM: - PR1 acts as client to get token for PR2 using token exchange - tokens can be chained through multiple PRs by repeating process - claims inside token: - "aud" is target PR - "act" is chain of token history before we got to the current token Multiple ICAM system: - More complicated - user could authenticate in any way, doesn't need to be PKI, just needs to auth to first AS Multiple ICAM solution options: - looking for feedback on different options 1. PR1 gets a token from its own AS via token exchange - Pros: PR1 doesn't need a relationship w/another AS domain - requirements that everyone's registered ahead of time - Cons: PR2 needs to trust tokens from AS1 - has to do token introspection (cross-domain) but through its own AS2 Justin: this isn't off-label use of introspection Brian: but it's not common to use Kelley: PR2 calls AS2, not AS1 Justin: that is actually off-label, then Jeff: is there an assumption of the identifier space being shared? Kelley: we're assuming there are relationships bbetween parties Aaron: on introspection, one AS would use introspection of other AS, is that the case? Kelley: Idea is yes, AS2 validates tokens for PR2, how that happens is outside of scope 2. PR1 gets a token from PR2's AS via token exchange - pros: PR2 only needs to validate tokens in its own domain - nothing special for PR2 - cons: PR's need to register across domains in order to get cross-domain tokens - registration across all other possible domains where it might need to talk - AS2 has to trust tokens from AS1 Brian: in general, token exchange draft defines token exchange URI as particular to the AS you're talking about; something else would need to be used to denote tokens from another AS; profile should add it 3a. PR1 gets a JWT from its own AS that PR2 sends to PR2's AS as an assertion grant - token exchange can be used to get assertion token (JWT) - pros: PR2 only sees access tokens from its own AS - JWTs are meant to be used across domains - Cons: AS needs to trust assertions from AS1 - AS's need to agree on content/details of assertions, need to communicate & trust - additional steps; could add latency? 3b. PR1 gets 1nd access token from its own AS using a JWT assertion - PR1 does token exchange but gets back token, not assertion - AS1 knows how to make "tokens for outside organizations", contacts/coordinates with AS2 - Pros: PR1 doesn't need to be registered across domain, - PR2 only gets tokens ussed by AS2 - cons: - as2 has to trust assertions from AS1 - AS's need to agree on content/details of assertions - additional steps & latency ## Questions / Comments / Discussion Brian: like model (3b), but it never seems to work out as well as it looks on paper ... all these tokens are bound to the cert of the client making the call, right? Kelley: yes, we require MTLS with tokens Brian: that gets problematic b/c you need to issue a token for a client/RS but issued to the AS; binding certificates correctly gets really strange. I'm not sure 3B is even viable w/o a lot more information about certificates. Kelley: this is the kind of feedback we're looking for, if it's not feasible Brian: speaking to the logistics of who holds the key; in 3B it doesn't make sense b/c caller isn't entity holding certificate. less about current implementations, more about logistics of the flow. ... in step 4 of diagram for 3b, have one service calling another, but result is token used in (6) but token is bound to certificate in PR1 Kelley: we assume all servers have NPE certs Brian: even if everybody does have a cert, certs are established btw two entities but it's bound to a different certificate Kelley: hadn't thought of that Warren: similar question but to other cases; don't understand the use cases, how is RS at the end verifying identity that's coming in? ... is expectation that data w/subs is going to verify each layer? if no, is it only trusting last AS? ... what's the purpose of stacking? you're just collecting them, could be a list Kelley: we want to make sure we know the client and intended recipient, then transferred along in chain Warren: nothing stops AS3 from lying about what it did to get there Kelley: last AS signs the token it generates Beth: we designed this for a very specific environment, there's an attribute sharing infrastructure underlying the AS's ... v. strong trust infrastructure, does not work in every environment ## WG Call Dmitry: question on other token binding mechanisms; have mentioned MTLS, might any other mechanisms (Dpop) be relevant here? Kelley: we're waiting to see how DPoP turns out ... we like the server nonce, that fixes a lot of issues ... there's another draft out, asymmetric keypairs Rifaat: is there interest in solving this problem? (no +1's in the chat) Justin: concerned about scope and unknown assumptions underlying the proposals ... historical chaining draft: https://tools.ietf.org/id/draft-richer-oauth-chain-00.html Warren: if we treat these as a single AS, then this feels similar to DPoP but only in a way we could stack DPoP on top of each other ... as long as there's no language about crossing security domains beth: two different proposals, single and mutli security system ... might have blurred together ... v. interested in making things as standardized as possible ... don't want specialized solutions if we can avoid it, even thought we have specialized environment Rifaat: don't see much support at this stage, but maybe you got good feedback and can try again in the future # Attendees * Rifaat Shekh-Yusef/Auth0 (chair) * Hannes Tschofenig/Arm (chair) * Kelley Burgin/MITRE (Presenter) * Justin Richer * Peter Yee, AKAYLA * Jeff Craig, Google * Brian Campbell, Ping * Aaron Parecki, Okta * Dmitry Telegin, Backbase / Keycloak * Marie-Helene Bouchard (Government of Canada) * Beth Abramowitz, MITRE * Roman Danyliw, CMU SEI * Warren Parad, Rhosys * Marius Ciocan * Vittorio Bertocci (Auth0 | Okta) * Filip Skokan (Auth0) * Roberto Polli * Michael Peck * Casey Yourman, Ford Credit # Recording https://ietf.webex.com/recordingservice/sites/ietf/recording/7599b82b1eed103abb7e0050568ced19/playback