Re: [dnsoverhttp] [Ext] DNS over HTTP: next steps?

[Shane Kerr <[e-mail]>]

Erik,

Apologies for the late reply!

At 2017-01-08 20:34:23 +0900
Erik Kline <[e-mail]> wrote:

> Can you clarify something for me about the wireformat draft, section 3.4?
> 
> I believe that a client proxy does not include the DNS-over-TCP 2 byte
> length toward the server, instead relying on the HTTP content-length.
> Is it fair (and would be more clear) to simply say that the
> DNS-over-TCP 2 byte length is *never* included when encapsulating over
> HTTP regardless of role and query vs. response status?

I think that's fair.

The basic idea is that you don't need any length indicator for the DNS
message because HTTP already includes that information.

However, I have re-written this section at least 3 times and every time
someone ends up finding it unclear in some way. :( I think that I might
not be able to write this in good way. Maybe someone can help?
 
> I'm also a tad confused about how things look if one was, for example,
> doing a zone transfer via DNS-over-HTTP and the server wanted to send
> things with Transfer-Encoding of type chunked?

I haven't actually thought about it. Generally server-to-server
communications fall outside of the scope of the draft. I would expect
those to use normal DNS over TCP.

But in principle there is nothing to prevent a user from asking for a
zone transfer (I do it all the time, actually), and no reason why this
shouldn't work over HTTP.

> Actually, re-reading RFC 7230 sections 3.3.1 and 3.3.2, I believe that
> if it's a requirement to include a Content-Length header then that
> means Transfer-Encoding MUST NOT be used.  Is this fair to say, and is
> it true of all DNS-over-HTTP data streams, regardless of role and
> direction (i.e. it doesn't just apply to client requests)?

Thinking about it, I don't think there is a strong requirement to
include a Content-Length header. Honestly all that we really need is
for the data to end up on the other end, and to know how much got
there. Any transfer encoding should work for this.

Probably I should go through the document and remove references to
this, and make it more general to indicate this. Do you think it makes
sense to include Content-Length as an example of a possible way to do
this or would that confuse things?

> My apologies if I'm focusing on minutiae too early in the process.

Not at all! I'm not actually sure where this draft is process-wise. :)

Cheers,

--
Shane