RE: [dnsext] Use of DNS SRV records for host selection

["Greg Daley" <[e-mail]>]

Hi Colm, 

>> The SRV record contains only a Target field, which is required to
>> have a host lookup (such as A or AAAA), not allowed to be an alias
>> (CNAME or DNAME).
> 
> I merely meant if this pattern was valid;
> 
> question;      _foo._host.example.com IN SRV
> answer:        _foo._host.example.com IN CNAME _bar._host.example.com
> 
> ...
> 
> _bar._host.example.com IN SRV
> 
> I think it should be, with regular CNAME/DNAME behaviour applying ...
> but it's why I wondered if "negative response" should imply "zero
> answers". It was just illustrative.

I believe for the initial response it would be possible to use
CNAME/DNAME, but would there then be a (restarted) SRV lookup on
the canonical name?

S5.1 of DNAME-bis specification identifies that the DNAME substitution
occurs before specific (MX,SRV,PTR) records:

http://tools.ietf.org/html/draft-ietf-dnsext-rfc2672bis-dname-18

For CNAME RFC2181 S10.3 is more explicit about this, where it discusses
MX and NS  records.  CNAME is included in the response with no additional
section processing, with the query potentially being restarted (with the
same QTYPE, and the CNAME substituted).

I don't think that these issues have been widely examined because for service
(not host) SRV records your would not typically have a (non-SRV) CNAME response.

I would guess that this CNAME behaviour is valid as well, although in
absence of guidance (other than that for the Target field of RFC2782),
people have been assuming that no CNAME record can be used in response to SRV
queries:

https://issues.asterisk.org/view.php?id=6314


>> Thanks, as you point out, the description for \X states  that "\." 
>> can be placed in a label.
>>
>> I was using the preferred syntax from S2.3.1 which indicates
>> alphanumeric and hyphens compose labels:
>>
>> <domain> ::= <subdomain> | " "
>> <subdomain> ::= <label> | <subdomain> "." <label>
>> <label> ::= <letter> [ [ <ldh-str> ] <let-dig> ]
>> <ldh-str> ::= <let-dig-hyp> | <let-dig-hyp> <ldh-str>
>> <let-dig-hyp> ::= <let-dig> | "-"
>>
>> I will try to stick to hostname/domainname in descriptions.
>>
>> There would seem to be a divergence then from having short 
>> unstructured Service labels, as are currently defined for SRV, and the 
>> hostname (label) element which could contain embedded "."
>>
>> This only matters where the host SRV record is required to take a
>> particular structure with underscores for the _Protocol element.
> 
> I just think it will be easier for implementors if when passed;
> 
>    "foo\.bar.example.com"
> 
> that the the lookup is for;
> 
>   _foo\.bar._host.example.com
> 
> rather than
> 
>   _foo._host.bar.example.com
> 
> it's one boundary case with the existing language. DNS resolvers and
> clients can typically avoid text processing, but as-defined, the
> language of the I-D would require string manipulation.

I will make sure that the text is updated.

Greg