On Oct 21, 2021, at 7:07 AM, Sara Dickinson <[email protected]> wrote:
> 
> 
> 
>> On 21 Oct 2021, at 00:16, Paul Hoffman <[email protected]> wrote:
>> 
>> After reading the -06 somewhat carefully, I have only one question: why 
>> should the DNS message ID be set to 0? If there's a good reason, is should 
>> be listed in Section 5.2.1, but if there isn't, the requirement should be 
>> removed. As the text indicates, this requirement makes forwarding more 
>> difficult. It also makes a special case for a client implementation that can 
>> do both DoQ and DoT.
> 
> Hi Paul,
> 
> The stream mapping of DoQ allows for unambiguous correlation or queries and 
> responses so the the message ID isn’t needed. Using 0 means the number of 
> outstanding queries on a DoQ connection isn’t limited to the Message ID space 
> which would be an artificial limit inherited from DNS-over-UPD/TCP. It seems 
> wrong to impose that on a transport specifically designed to multiplex, just 
> to make proxying simpler.  Given that RFC8484 says "DoH clients ... SHOULD 
> use a DNS ID of 0 in every DNS request.” that proxy problem already exists 
> and isn’t unique to DoQ. 

Arrrgh. I vaguely remembered that discussion. I forgot that we ended up with 
that in DoH, so it's OK if we end up with it in DoQ.

> It’s presence would also change the error condition from ‘what if the message 
> ID isn’t 0’ to 'what happens if the response ID on a stream doesn’t match the 
> query ID?’. It’s marginal but the former feels slightly cleaner.

We disagree on which feels cleaner, but that's moot. My argument against 
always-0 (it differs from DoT) is trumped by it being the same as DoH. This is 
a hassle that implementers have already seen.

--Paul Hoffman

Attachment: smime.p7s
Description: S/MIME cryptographic signature

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

Reply via email to