On 2014-06-11 17:45, Viktor Dukhovni wrote:
> On Wed, Jun 11, 2014 at 05:21:22PM +0200, Kim Alvefur wrote:
> 
>> On 2014-06-11 03:24, Peter Saint-Andre wrote:
>>> Matt Miller and I submitted a new version of draft-ietf-dane-srv today.
>>
>> Why does the text it Section 3 say to do queries in parallel when
>> section 3.2 says that one has to do A/AAAA lookups and validation (which
>> depend on SRV) before deciding to do TLSA lookups?
> 
> It says "where possible".  As I see it the only sensible opportunity
> for parallel lookup is to find the A or AAAA records of the subset
> of the SRV hosts with the highest (untried) priority, then among
> those that actually have an IP address, one can choose a randomized
> server based on the weight, at which point I would only retrieve
> TLSA records for that host.
> 
> Generally, all the SRV hosts are up and it is sufficient to try
> one connection, so there is no benefit in performing multiple TLSA
> lookups for clients that make long-lived infrequent connections.
> 
> In summary I think that while some clients may decide to parallelize
> some DNS lookups, this is an implementation decision that is perhaps
> out of scope for the DANE SRV specification.

That's what I was aiming for.

> 
>> And what does the security status of A/AAAA records matter if you have a
>> secure reference to the certificate you see?  Unless of course if you
>> get bogus, then aborting the connection to that target is sane.
> 
> See Section 2.2.2 of the SMTP draft, which provides the rationale:
> 
>     
> http://vdukhovni.github.io/ietf/draft-ietf-dane-smtp-with-dane-10.html#rfc.section.2.2.2

Perhaps a reference to that then?

>>>    SRV is secure:  The reference identifiers SHALL include both the
>>>       service domain and the SRV target server host name (e.g., include
>>>       both "im.example.com" and "xmpp23.hosting.example.net").  The
>>>       target server host name is the preferred name for TLS SNI or its
>>>       equivalent.
>>
>> In XMPP, sending to='xmpp123.hosting.example.net' instead of
>> to='example.com' does not seem sensible to me.  The server should know
>> which cert to present for each domain, or probably just shows the same
>> for all if it is a multi-tenant hosting service.
> 
> This suggestion breaks cross-domain hosting:
> 
>       _service.example.org.   IN SRV 0 0 12345 server.example.net.
> 
> The server.example.net hosting provider can obtain certs for its
> own name, managing certs from each hosted domain is a significant
> burden.

Then how is server.example.net supposed to know that you want to talk to
example.org?  "its equivalent" is said to be the stream addressing
earlier which makes this weird.

>> And I don't think one should look too hard at the name in a certificate
>> if you have a matching TLSA record.
> 
> You're not thinking about TA certificate usages PKIX-TA(0) and
> DANE-TA(2) which absolutely require name checks.  Even PKIX-EE(1)
> requires name checks as part of the PKIX validation.
> 

Right, of course.

--
Kim "Zash" Alvefur


Attachment: signature.asc
Description: OpenPGP digital signature

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

Reply via email to