On Dec 15, 2011 12:27 AM, "Jim Schaad" <[email protected]> wrote:
> Normally I would want to have a hard check on the name of the TLS server
and matching w/ the certificate.  However, just like one could use ABFAB
for the purpose of doing user authentication, I think that one can use
ABFAB for doing server authentication.  That is one could be able to
authenticate either an anonymous TLS server, or a TLS server with whom you
do not have a trust anchor.

The tls-server-end-point channel binding tour does not require that the
server cert be valid to any trust anchor, not does out require pre-sharing
of self-signed certs, not does it require any particular names our name
forms to appear in the server cert.  I.e., server certs can be anonymous
and pseudonymous if you like.

If you don't want to use server certs at all (self-signed our otherwise)
you can always use anon DHE cipher suites and use the tls-unique channel
binding type.

The tls-server-end-point CB type is easy to support almost universally,
whereas tls-unique support is a bit harder to come by (not all TLS stacks
support it, sadly).

> This would be similar to what you laid out for the use authentication
method.  That is one would setup the TLS channel, perform the ABFAB
authentication with the required channel binding from the TLS session and
then throw away the ABFAB work after the ABFAB has completed it
authentication process (i.e. the MIC is checked).  At this point one should
have mutual authentication, and the EAP channel binding should have
compared the name the client is looking from (from the TLS certificate or
the URL) with that provided by the ABFAB RP.
>
> Does this seem a reasonable approach in some circumstances?

Yes, it does.

>  Are there circumstances you can see where this would be unreasonable?

I can't think of any.

Nico
--
_______________________________________________
abfab mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/abfab

Reply via email to