Re: [TLS] MTI extensions?

[Jeffrey Walton <[e-mail]>]

> This is a significant consideration for SMTP with DANE, because
> clients use the TLSA base domain (typically MX host name) in the
> SNI message, but will also accept other names (typically the
> recipient domain for which the server is an MX host).  The server
> may have just one "UCC" certificate for the recipient domain, and
> MUST NOT abort the connection just because the client seems to want

I think cases like this should be considered further. There's too many
open questions surrounding use cases like this.

Jeff

On Sun, Mar 15, 2015 at 5:28 PM, Viktor Dukhovni <[e-mail]> wrote:
> On Sun, Mar 15, 2015 at 02:13:14PM -0700, Eric Rescorla wrote:
>
>> SNI might be plausible as a MUST implement but there are plenty of cases
>> where it's not sensible to be MUST send, as in where there is no server
>> hostname. E.g., WebRTC.
>
> Furthermore, even when SNI is sent, a server should generally NOT
> abort the handshake when it does not find a match for the requested
> name.
>
> Rather, it is better for the server to choose *some* other (often
> default) certificate, and leave it up to the client to decide
> whether that one is also acceptable.
>
> This is a signficant consideration for SMTP with DANE, because
> clients use the TLSA base domain (typically MX host name) in the
> SNI message, but will also accept other names (typically the
> recipient domain for which the server is an MX host).  The server
> may have just one "UCC" certificate for the recipient domain, and
> MUST NOT abort the connection just because the client seems to want
> something else.