Re: [OAUTH-WG] Downgrade attacks on PKCE

[Vladimir Dzhuvinov <[e-mail]>]

Thanks for laying out the solutions so neatly.

We would prefer #2 the "dynamic" solution because it wouldn't require us
to do any changes. I've had the impression that the unexpected
code_verifier case was somehow covered as an error in RFC 7636 but
checked the spec now and apparently it isn't.

Vladimir

On 30/05/2020 10:58, Daniel Fett wrote:
> Hi all,
>
> Aaron, Dick, Torsten and I today discussed the downgrade attacks on
> PKCE [1] and how to mitigate them in OAuth 2.1 and 2.0. We came to the
> conclusion that we have two options:
>
> *1. "Static" Solution*
>
> For every client_id that is registered with an AS, the AS MUST either
> always enforce the use of PKCE or always enforce the use of nonce.
> Whether PKCE or nonce is enforced can be part of the client
> registration or configured in other ways.
>
> In other words: A single client is not allowed to switch between using
> PKCE and using nonce.
>
> Note that the client is allowed to use the respective other mechanism
> on top of the enforced one.
>
> Properties:
>
>   * Easy to understand mitigation.
>   * Implementation is mainly a new data field and a check in the
>     authorization request.
>   * Not compatible to deployments where clients sometimes use nonce
>     and sometimes use PKCE with the same client_id. *
>     ***
>
> **
>
> *2. "Dynamic" Solution*
>
> Each AS that supports PKCE MUST check whether a code challenge is
> contained in the authorization request. This information MUST be bound
> to the code that is issued.
>
> When a code arrives at the token endpoint, the AS MUST do the
> following check:
>
>  1. If there was a code_challenge in the authorization request for
>     which this code was issued, there must be a code_verifier in the
>     token request and it must be verified according to RFC7636. (This
>     is no change from the current behavior in RFC7636.)
>  2. If there was no code_challenge in the authorization request, any
>     request to the token endpoint containing a code_verifier MUST be
>     rejected.
>
> Properties:
>
>   * No change in behavior needed for properly implemented clients.
>     Backwards compatible for all existing deployments.
>   * Implementation is mainly some logic for the authorization
>     endpoint, token endpoint, and a new data field in the
>     authorization session maintained by the AS.
>   * Slightly more complex to implement for the AS, maybe.
>
> We would like to hear the feedback from the working group on these two
> solutions before proceeding to propose wording for the affected documents.
>
> [1] https://danielfett.de/2020/05/16/pkce-vs-nonce-equivalent-or-not/
>
> -Daniel