Re: [dane] namespace management, DANE Client Authentication draft updated

[Viktor Dukhovni <[e-mail]>]

On Wed, Jan 13, 2016 at 08:06:17PM -0000, John Levine wrote:

> >    _service._client.node.example. IN TLSA ...
> 
> Ah, we're getting closer.

Well, I can live with the above, *if* there's a good reason to
expect conflicts that block use of CNAMEs.

> >I still don't see any use for _tcp/_udp in there.
> 
> RFC 6698 has _tcp _udp and _sctp protocols as part of the names for
> TLSA.

For a good reason, because the full prefix is "_<portnumber>._<proto>".
Without "_proto", we get TLSA records applied to the wrong service
endpoint, because UDP ports with the same number don't reach have
the same service as the TCP ports.

> It seems rather odd to have the protocol name for the server
> certificate but not for the client.

Because the client prefix-label a service *name*, so so the port
collision issue goes away.  We should not cargo-cult designs,
the rationale has to carry over logically, and false analogies
need to be avoided.

-- 
	Viktor.