Joe Abley wrote:
> On 11 Nov 2026, at 09:09, Ralf Weber <[email protected]> wrote:
> 
> > Looks like I’m not the first one to have thought about that got pointed to 
> > by Joe:
> >    
> > https://datatracker.ietf.org/doc/draft-jabley-dnsop-no-longer-support-any/
> > Which I think is a good starting point for us to work on
> 
> The QTYPE ANY appreciation society always welcomes new members!
> 
> That draft in its current form continues a conservative approach to 
> eliminating ANY from the Internet. The advice could be made stronger. I think 
> the core point you made the other day in Vancouver that ANY is so unreliable 
> at this point that it's not really useful for anything was a good one. Maybe 
> it's time to give it the hard MUST NOT treatment. 
> 
> Another thing we talked about briefly was whether advice for QTYPE=ANY could 
> be combined with advice about QTYPE=RRSIG which is a current topic of 
> conversation, and perhaps also QTYPE=NSEC and QTYPE=NSEC3 fall in a similar 
> bucket. 
> 
> In my mind ANY has a different history and a different set of practical 
> considerations from the others and it's useful to treat it separately, but it 
> would be good to hear what others think. 

Hi,

I read the draft and the slides from Ralf's presentation [0] but I don't
see a clear rationale for why an RCODE=4 (NOTIMP) response would be
preferable to, say, an RFC 8482 §4.2 (HINFO) response. The closest seems
to be the anecdote in the slide deck that speculates that a broken RFC
8482 implementation might have been responsible for events where broken
responses were observed from root nameservers?

[0]: 
https://indico.dns-oarc.net/event/58/contributions/1286/attachments/1227/2631/StopANY.pdf

According to RFC 8906, RCODE=4 (NOTIMP) should not be used for an
unimplemented QTYPE (specifically, an unimplemented non- "Meta-RR"
QTYPE):

    For unimplemented opcodes, NOTIMP is the expected response code. [...]
    For unimplemented type codes, and in the absence of other errors,
    the only valid response is NOERROR if the qname exists and NXDOMAIN
    otherwise. For Meta-RRs, NOTIMP may be returned instead.

According to RFC 6895, QTYPE=ANY is not a Meta-TYPE:

    There are currently three Meta-TYPEs assigned: OPT [RFC6891], TSIG
    [RFC2845], and TKEY [RFC2930].  There are currently five QTYPEs
    assigned: * (ALL/ANY), MAILA, MAILB, AXFR, and IXFR.

This makes sense, since there is no resource record with RRTYPE 255 that
appears on the wire, like there is for OPT, TSIG, and TKEY.

(Side note, this text in 6895 about there being only five QTYPEs seems
a little weird, vs. the definition in RFC 1035 §3.2.3 about QTYPEs
being a superset of TYPEs, and the use of "QTYPE" to refer to both
the type field in the question section of a DNS message, as well as a
classification of RRTYPE code points.)

The IANA Resource Record (RR) TYPEs registry doesn't seem to record
whether an individual code point in the 128-255 QTYPE and Meta-Type
range is a QTYPE or a Meta-Type, and it doesn't seem that QTYPE=ANY
has been reclassified by any other documents. The closest seems to
be this text in the Terminology section of RFC 8482 about ANY being
a "meta-query":

    In this document, "ANY query" refers to a DNS meta-query with
    QTYPE=ANY.  An "ANY response" is a response to such a query. 

But that seems to stop short of an explicit specification update. If
8482 did re-classify ANY as a Meta-TYPE, I'd expect it to say so in
section 7 or 10.

So, given that ANY is a QTYPE and not a Meta-TYPE, does returning
RCODE=4 (NOTIMP) result in problematic behavior from resolvers
that don't implement this draft, such as retrying the query against
other nameservers for the zone? If so, that seems like a significant
disadvantage compared to RFC 8482 responses that ought to be called out
in the draft, and maybe warrants a phased approach? Such as updating
resolvers before nameservers.

Similarly, are there issues with recursive resolvers returning RCODE=4
to stub clients, like causing the stub clients to retry the query
against the other resolver addresses configured into the stub resolver?

Side note, in the draft I found the use of the unqualified terms "DNS
client" and "DNS server" a little unclear. Recursive resolvers function
as clients when sending queries to nameservers and as servers when
sending responses to stub resolvers, so it seems that both the "DNS
client" and "DNS server" guidance applies to recursive resolvers.
Maybe it would be better to issue separate guidance for stub resolvers,
full-service resolvers, authoritative servers, etc., especially if a
phased approach is used.

-- 
Robert Edmonds
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to