Re: [Doh] WG Review: DNS Over HTTPS (doh)

["Paul Hoffman" <[e-mail]>]

On 15 Sep 2017, at 9:44, Ted Hardie wrote:

=>> This working group will standardize encodings for DNS queries and 
responses
>> that are suitable for use in HTTPS. This will enable the domain name 
>> system
>> to function over certain paths where existing DNS methods (UDP, TLS, 
>> and
>> DTLS)
>> experience problems.  The working group will re-use HTTPS methods, 
>> error
>> codes, and other semantics to the greatest extent possible.  The use 
>> of
>> HTTPS
>> provides integrity and confidentiality, and it also allows the 
>> transport to
>> interoperate with common HTTPS infrastructure and policy.
>>
>>
> I appreciate the charter's use of "HTTPS" as a signal that these are
> intended to be TLS-protected HTTP sessions.  I note, however, that 
> there is
> considerable ambiguity still present.  HTTPS can mean HTTP 1.1 over 
> TLS,
> HTTP/2 over TLS, and it may mean HTTP over QUIC at some point soon (in 
> some
> deployments it already means that).  The document named as input 
> specifies
> HTTP/2  over TLS.  While the working group may, of course, change that 
> to
> support HTTP 1.1 and/or QUIC, it might be useful for the charter to
> indicate which of these is potentially in scope.

Yes, please. The charter should say "HTTP/2 over TLS". There is no 
reason for current browsers adding this feature to add it using an 
obsolete protocol. If someone wants to create a diff from the eventual 
protocol for HTTP 1.1, they can do that without forcing the document to 
add a comparison of the two transports to what is supposed to be a 
short, concise document.

> If the community is sure
> now that HTTP over QUIC is in scope, for example, having that noted in 
> the
> charter by adding the QUIC working group to list of working groups to
> consult would be useful.

The deadline for this WG is well ahead of when HTTP-over-QUIC will be 
finalized. Instead, when HTTP-over-QUIC is finalized, an update to this 
document should be pretty easy to produce.

> I will confess a bias here: while I think it is useful to match the
> capability set of HTTP over QUIC to that of HTTP over other transports 
> as
> much as possible, I think making that a key part of this work is not a 
> good
> initial direction.  For one thing, there are some aspects of the QUIC
> transport that are still in discussion, and I think it will slow this 
> work
> a bit to track those.  More importantly, though, I think this would be 
> the
> wrong way to do DNS over QUIC (draft-huitema-quic-dnsoquic-00 shows a
> different approach).  Having QUIC be an early focus here may solidify 
> an
> approach that is simple, but not nearly as complete as we could 
> deliver
> with a DNS over QUIC.
>
> (This is in part because of the choice to use the udp wireformat as 
> the
> baseline HTTP response here, rather than specifying DNS responses over 
> a
> transport construct like a QUIC stream.  The working group could, of
> course, change that, but it would seriously shift the direction of its
> input document to do so).

Fully agree.

>> Specification of how the DNS data may be used for new use cases, and
>> the discovery of the DOH servers, are out of scope for the working 
>> group.
>>
>>
> While it is useful to know that discovery is out of scope here, I 
> think
> having a quick community discussion now of where it might be in scope 
> is
> useful.  There are some potentially interesting questions buried in 
> that,
> especially in how we expect interworking with existing systems to go.
> Note that even if the service discovery method provides an HTTPS URI 
> for
> the name server, the questions above related to HTTP version or 
> transport
> may bite you.  For server discovery based on address and port, the
> situation is much the same unless the port is clearly marked as TCP or
> UDP.  And if the server discovery includes no port at all, then a 
> happy
> eyeballs type method may be needed.
>
> This may all fall into a combo of DHCP and DNSOP work, but giving 
> somewhat
> clean lines for it now seems like it will make the later work go 
> faster.

If you want to propose a new WG for that, please do so; there are a lot 
of people interested in that topic. It definitely goes across multiple 
areas of interest, such as "discovery" and "addressing" and even 
"trust". Personally, I don't think it applies to a transport document.

>> Milestones:
>>
>>   Apr 2018 - Submit specification for performing DNS queries over 
>> HTTPS to
>>   the IESG for publication as PS
>>
>>
> I admire the optimism in this.

It was not optimistic with the charter that was originally sent to the 
IESG; see <https://datatracker.ietf.org/doc/charter-ietf-doh/00-00/>. As 
more issues are added to the charter, it makes sense to have to extend 
the milestone deliverable.

--Paul Hoffman