Re: [OAUTH-WG] some implementation feedback with the PKI method of OAuth MTLS client authentication

[John Bradley <[e-mail]>]

Having discussed it with Brian, I agree that removing “tls_client_auth_root” is the way to go.  
It would be hard to implement in some cases, and it is up to the AS to configure the roots it trusts for client authentication.

In reality every TLS client auth deployment is likely to have custom rules about trust.  I think this parameter adds confusion rather than reducing it.

John B.


On Aug 28, 2017, at 10:05 AM, Brian Campbell <[e-mail]> wrote:
> 
> Some feedback was received recently off-list that pointed out difficulties with implementation around the "tls_client_auth_root_dn" constraint in the PKI method of OAuth MTLS client authentication from draft-ietf-oauth-mtls-03. Basically the feedback was that popular web servers such as Nginx and Apache don't expose sufficient information (easily or in some cases at all) from the client certificate chain to the application to enable proper checking of "tls_client_auth_root_dn". 
> 
> Following from that and some additional reasoning below, I'm proposing that "tls_client_auth_root_dn" be dropped from the draft-ietf-oauth-mtls draft. 
> 
> The original idea behind the "tls_client_auth_root_dn" client metadata parameter came from an MTLS client authentication feature we added to our AS product years ago. The feature provided a way to allow the AS to more tightly constrain trust in a particular context (from an otherwise global list of trust anchors). It was fine as a specific product feature but should have stayed at that. When I added metadata to the OAuth MTLS draft, I added the "tls_client_auth_root_dn" parameter with that AS product feature in mind as something an AS *might* want to be able to do (without thinking thorough it all sufficiently). But having it as a client metadata parameter has wider implications including shifting trust control to the client and requiring ASs to support it. So, after thinking about it some more and also seeing the potential implementation difficulties, I don't believe it's appropriate to have in the specification. The AS should be at liberty to do chain validation with the PKI method however is most appropriate for it. And not be required to support one specific way of doing things implied by "tls_client_auth_root_dn" (which is even infeasible to implement in some environments).  
> 
> 
> 
> 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