Re: [OAUTH-WG] MTLS and in-browser clients using the token endpoint

[Neil Madden <[e-mail]>]

On the assumption that this is likely to be a requirement from customers, I’d be in favour of a new server metadata field. My favourite bikeshed colour would be:

tls_client_auth_token_endpoint

On another metadata-related note, given that the additional security of certificate-bound access tokens vanishes if the resource server doesn’t understand and enforce the certificate-binding associated with the access token, is there a general need for a client to be able to discover if any given RS does actually support this? Otherwise the whole scheme “fails open” in that the access token silently degrades to a normal bearer token. Or do we consider it unlikely that an RS is going to accept a TLS client certificate without supporting this?

— Neil

> On 17 Dec 2018, at 20:26, Brian Campbell <bcampbell=[e-mail]> wrote:
> 
> While there's been some disagreement about the specific wording etc., there does seem to be general consensus coming out of this WG to, in one form or another, recommend against the use of the implicit grant in favor of authorization code. In order to follow that recommendation, in-browser JavaScript clients will need to use XHR/fetch (and likely CORS) to make requests directly to the token endpoint.
> 
> Meanwhile there is the MTLS document utilizes TLS client certificates at the token endpoint for client authentication and/or certificate bound access tokens. The security BCP draft even recommends sender/key constrained access tokens and MTLS is close to the only viable way to do that at this time. 
> 
> Unfortunately, however, these two things don't play very nice together. When a browser makes a TLS connection where a client cert is requested by the server in the handshake, even when client certificates are optional and even when it's fetch/XHR, most/many/all browsers will throw up some kind of certificate selection interface to the user.  Which is typically a very very bad user experience. From a practical standpoint, this means that a single deployment cannot really support the MTLS draft and have in-browser JavaScript clients using authorization code at the same time. 
> 
> In order to address the conflict here, I'd propose that the MTLS draft introduce a new optional AS metadata parameter that is an MTLS enabled token endpoint alias. Clients that are doing MTLS client authentication and/or certificate bound access tokens would/should/must use the alternative token endpoint when present in the AS's metadata. While all other clients continue to use the standard token endpoint as they always have. This would allow for an AS to deploy an alternative token endpoint alias on a distinct host or port where it will request client certs in the TLS handshake for OAuth clients that use it while keeping the regular token endpoint as it normally is for other clients, especially in-browser JavaScript clients. 
> 
> Thoughts, objections, agreements, etc., on this proposal?
> 
> PS Bikeshedding on a name for the metadata parameter is also welcome. Some ideas to start:
> token_endpoint_mtls_alias
> token_endpoint_mtls
> mtls_token_endpoint_alias
> mtls_token_endpoint
> alt_token_endpoint_mtls
> mtls_token_endpoint_alt
> a_token_endpoint_that_a_client_wanting_to_do_mtls_stuff_a_la_RFC_[TBD]_should_use
> equally_poor_idea_here
> 
> 
> 
> 
> 
> 
> 
> 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._______________________________________________
> OAuth mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/oauth