Re: [jose] [OAUTH-WG] preventing confusion of one kind of JWT for another in JWT BCP

[Brian Campbell <[e-mail]>]

Yeah, I've implemented it in pretty much in the way Jim describes.

In my implementation of JWT validation the application sets things up by
pushing down all the info needed for validation (things like expected
audience & issuer, keys, allowed algorithms, etc.). So I'm probably biased
by that but also telling it that 'p' is okay as a critical header doesn't
feel like any more of a layering violation than anything else.

This example
<https://bitbucket.org/b_c/jose4j/wiki/JWT%20Examples#markdown-header-producing-and-consuming-a-signed-jwt>
shows how to use the library for regular JWT validation. This method
<http://static.javadoc.io/org.bitbucket.b_c/jose4j/0.5.6/org/jose4j/jwt/consumer/JwtConsumerBuilder.html#setJwsCustomizer%28org.jose4j.jwt.consumer.JwsCustomizer%29>
would be used to tell the JWT validator/consumer that the 'p' as a critical
header should be accepted. And then the JWT validator/consumer could be
told what value of 'p' is acceptable or required by adding a validator for
it
<http://static.javadoc.io/org.bitbucket.b_c/jose4j/0.5.6/org/jose4j/jwt/consumer/JwtConsumerBuilder.html#registerValidator(org.jose4j.jwt.consumer.Validator)>
or the application code could just look at the headers after and check it
that way. That's how it could be done now with the existing general API. If
the BCP was to formalize something, I'd probably enhance the API with
something specific for it to make it even easier.




On Thu, Jul 27, 2017 at 4:56 PM, Jim Schaad <[e-mail]> wrote:

> One simple way to implement it would be to have an call which says “I will
> deal with the following items if they exist”.  This means that all the
> application needs to do is to say that it will process p and not what
> values are acceptable.
>
>
>
> Jim
>
>
>
>
>
> *From:* jose [mailto:[e-mail]] *On Behalf Of *John Bradley
> *Sent:* Thursday, July 27, 2017 3:24 PM
> *To:* Brian Campbell <[e-mail]>
> *Cc:* Nathaniel McCallum <[e-mail]>; oauth <[e-mail]>;
> [e-mail]; Phil Hunt <[e-mail]>
> *Subject:* Re: [jose] [OAUTH-WG] preventing confusion of one kind of JWT
> for another in JWT BCP
>
>
>
> My only concern with a crit header is that it is potentially another
> layering violation.
>
>
>
> It really should be the application and not JWS/JWE that determines if the
> content of the JWT is correct at a application layer.
>
>
>
> To get your idea to work the application would need to push down to the
> JWT layer some value/s of p that it was willing to accept.   Without crit
> the application could just pass along typ or p to the application layer for
> matching.
>
>
>
> I don’t hate the idea but it seems like it is likely to work against
> having generic JOSE libs separate from JWT. Perhaps that is the reality
> anyway.
>
>
>
> How would you implement it?
>
>
>
> John B.
>
>
>
> On Jul 27, 2017, at 6:10 PM, Brian Campbell <[e-mail]>
> wrote:
>
>
>
> I wouldn't disagree that the issue with Connect is overstated sometimes.
> And it's a non-issue with sec event due to the "nonce" claim (as you point
> out) as well as the omission of the "exp" claim. Assuming correct
> validation anyway.
>
> The BCP draft suggests explicit typing to prevent Cross-JWT Confusion more
> generally (not just sec events and id tokens). I was wondering if some use
> of "crit" might accomplish the same thing but also cover the case of
> existing JWT implementations being confused with new JWT profiles.
>
>
>
> On Thu, Jul 27, 2017 at 3:40 PM, John Bradley <[e-mail]> wrote:
>
> Not that I am against the general desire for JWT not to be confused, but
> the attack surface for someone to take a sec event from one context and
> replay it as a id_token to login is very small and if the client is
> correctly configured would not work as it is now.
>
>
>
> The only way you could replay the sec event JWT is via the implicit flow.
>   The other flows would require a malicious AS and at that point you have
> bigger problems with a compromised token endpoint.
>
>
>
> If you were to create a implicit response with the sec event token that
> has the correct issuer and aud, connect requires that nonce value from the
> request be present in the id_token.   This would require that the attacker
> somehow get the AS to put not only the claim but the correct value from the
> request into the token.
>
>
>
> All in all I think that it is more likely (not 100% I admit) that a client
> will do that rather than checking the typ or a new crit JWT header
> parameter.
>
>
>
> I am all for a general solution for this but I think the issue with
> Connect is overstated sometimes.
>
>
>
> John B.
>
>
>
> On Jul 27, 2017, at 5:19 PM, Phil Hunt <[e-mail]> wrote:
>
>
>
> We have the use case in SECEVENTS where a logout event (e.g. OIDF
> backchannel logout) is extremely close to an ID_TOKEN.
>
>
>
> Older relying parties who are do not yet support logout could be tricked
> into accepting a logout assertion as an ID_TOKEN since they are too
> similar, and  because a valid ID_TOKEN parser is in theory allowed to
> ignore claims it does not understand (e.g. “events”) - leading to a
> possible erroneous acceptance of the logout event AS and ID_TOKEN.
>
>
>
> Because of this the issue of distinguishing classes or types of JWTs
> emerged.
>
>
>
> We discussed a number of differentiators like “aud”, “typ”, “crit", etc
> that would help.  But that really lead to the idea there there should be
> some best practices in the JWT BCP covering the issue(s).
>
>
>
> Phil
>
>
>
> Oracle Corporation, Identity Cloud Services Architect & Standards
>
> @independentid
>
> www.independentid.com
>
> [e-mail]
>
>
>
> On Jul 27, 2017, at 2:00 PM, Nathaniel McCallum <[e-mail]>
> wrote:
>
>
>
> Even after reading the whole section, I still don't understand the
> problem. Yes, a class of attack could exist where an attacker
> substitutes a valid JWT from one security context into another
> context. But isn't this resolved by audience validation?
>
> On Thu, Jul 27, 2017 at 3:34 PM, Brian Campbell
> <[e-mail]> wrote:
>
> The draft describes it in
> https://urldefense.proofpoint.com/v2/url?u=https-3A__tools.
> ietf.org_html_draft-2Dsheffer-2Doauth-2Djwt-2Dbcp-2D01-
> 23section-2D2.7&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PQcxBKCX5YTpkKY057SbK10&r=
> JBm5biRrKugCH0FkITSeGJxPEivzjWwlNKe4C_lLIGk&m=
> 02yWnzzFGNlefgVregxSncXu0b9sbh5pTrtBfQsC52A&s=
> 9GmPCouCMXE4enC48gfq0l0yc7ZlIxvEBAnQCVja_kY&e=
>
> On Thu, Jul 27, 2017 at 1:30 PM, Nathaniel McCallum <[e-mail]
> >
> wrote:
>
>
> What class of attacks is this trying to prevent? I frankly don't see a
> problem with confusing different types of JWT. But I may just be
> ignorant.
>
> On Thu, Jul 27, 2017 at 2:49 PM, Brian Campbell
> <[e-mail]> wrote:
>
> During the first WG meeting last week I asked if use of the JOSE "crit"
> (Critical) Header Parameter had been considered as a recommendation for
> preventing confusion of one kind of JWT for another. Time was running
> short
> in the meeting so there wasn't much discussion and it was requested that
> I
> take the question to the list. And so here on the list is that.
>
> Section 3.9 of the JWT BCP draft now recommends explicit typing using
> the
> "typ" JWS/JWE header parameter but does concede that 'the use of
> explicit
> typing may not achieve disambiguation from existing kinds of JWTs, as
> the
> validation rules for existing kinds JWTs often do not use the "typ"
> header
> parameter value.'  And the recommendations for how to use the Type
> Header
> Parameter in JWT strongly suggest that it's not currently being used for
> any
> validation.
>
> Alternatively using the JWS/JWE "crit" (Critical) Header Parameter to
> signal
> the type/intent/profile/application of a JWT could achieve
> disambiguation
> even in validation of existing kinds of JWTs. The critical header lists
> other headers which must be understood and processed by the receiver and
> that the JWS/JWE is invalid if those listed aren't understood. So a new
> type/profile of JWT that uses the "crit" header would produce JWTs that
> would be rejected even by existing applications of JWT validation (that
> actually implement "crit" properly anyway).
>
> The JWT BCP could suggest the use of "crit" in conjunction with a
> profile/application/type specific header. Or it could provide a bit more
> of
> a framework like defining a registering a new JOSE header "p" (strawman
> of p
> as a very short name for profile) and create a registry for its values.
> A
> JWT header using that approach might look like the following where the
> value
> 1 is registered as some cool new JWT profile/application. The consumer
> of
> such a JWT would have to understand and process the "p" header, which
> would
> mean checking that it had the value expected.
>
>     {
>      "alg":"ES256",
>      "crit":["p"],
>      "p":1
>     }
>
> A JOSE compliant JWT validator would reject such a JWT even for an OAuth
> access token or OIDC id_token because the "p" header isn't known or
> understood but is marked as critical.
>
> To me, that seems like an approach to preventing confusion that has more
> teeth than the "typ" header. Which is why I asked about it last week and
> am
> now bringing it to the list.
>
>
>
>
>
>
>
>
>
>
> CONFIDENTIALITY NOTICE: This email may contain confidential and
> privileged
> material for the sole use of the intended recipient(s). Any review, use,
> distribution or disclosure by others is strictly prohibited.  If you
> have
> received this communication in error, please notify the sender
> immediately
> by e-mail and delete the message and any file attachments from your
> computer. Thank you.
> _______________________________________________
> jose mailing list
> [e-mail]
> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.
> ietf.org_mailman_listinfo_jose&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PQcxBKCX5Y
> TpkKY057SbK10&r=JBm5biRrKugCH0FkITSeGJxPEivzjWwlNKe4C_lLIGk&m=
> 02yWnzzFGNlefgVregxSncXu0b9sbh5pTrtBfQsC52A&s=eJ8QZBwe5AO-
> RDl_E16U1q_KpobaeSRUP4Cp2W-_jJU&e=
>
>
>
>
> CONFIDENTIALITY NOTICE: This email may contain confidential and privileged
> material for the sole use of the intended recipient(s). Any review, use,
> distribution or disclosure by others is strictly prohibited.  If you have
> received this communication in error, please notify the sender immediately
> by e-mail and delete the message and any file attachments from your
> computer. Thank you.
>
>
> _______________________________________________
> jose mailing list
> [e-mail]
> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.
> ietf.org_mailman_listinfo_jose&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PQcxBKCX5Y
> TpkKY057SbK10&r=JBm5biRrKugCH0FkITSeGJxPEivzjWwlNKe4C_lLIGk&m=
> 02yWnzzFGNlefgVregxSncXu0b9sbh5pTrtBfQsC52A&s=eJ8QZBwe5AO-
> RDl_E16U1q_KpobaeSRUP4Cp2W-_jJU&e=
>
>
>
> _______________________________________________
> OAuth mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/oauth
>
>
>
>
> _______________________________________________
> OAuth mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/oauth
>
>
>
>
> *CONFIDENTIALITY NOTICE: This email may contain confidential and
> privileged material for the sole use of the intended recipient(s). Any
> review, use, distribution or disclosure by others is strictly prohibited.
> If you have received this communication in error, please notify the sender
> immediately by e-mail and delete the message and any file attachments from
> your computer. Thank you.*
>
>
>

-- 
*CONFIDENTIALITY NOTICE: This email may contain confidential and privileged 
material for the sole use of the intended recipient(s). Any review, use, 
distribution or disclosure by others is strictly prohibited.  If you have 
received this communication in error, please notify the sender immediately 
by e-mail and delete the message and any file attachments from your 
computer. Thank you.*