- From: Roberto Peon <grmocg@gmail.com>
- Date: Fri, 3 Apr 2015 12:06:36 -0700
- To: Nicholas Hurley <hurley@mozilla.com>
- Cc: HTTP Working Group <ietf-http-wg@w3.org>
- Message-ID: <CAP+FsNcGAJjRXpQPKOs9rLk-5=JYjj24=DxNHCAv+Mib5v+2GA@mail.gmail.com>
Does anyone recall why 6066 has no SNI for IP literals? (It could be an empty SNI field or the SNI could indicate the IP literal)? -=R On Fri, Apr 3, 2015 at 11:37 AM, Nicholas Hurley <hurley@mozilla.com> wrote: > All, > > While looking at https://github.com/molnarg/node-http2/issues/69 I came > to the realization that it appears we have (unintentionally) made it > impossible to speak h2 when connecting directly to an IP address (as in, IP > address typed into URL bar as opposed to hostname typed into URL bar) and > remain compliant with both the h2 spec and RFC 6066. 6066 specifies that > SNI is not to be sent for an IP literal, while h2 requires SNI. You can see > the conflict. > > In node-http2, we have decided to relax the SNI requirement, and still > speak h2 to clients that don't give us any SNI, under the assumption that > this (IP in URL bar, or equivalent) is the case we are hitting. I had also > filed a bug against Firefox to stop advertising h2 in the cases where we > won't send SNI, but am rethinking that idea, as it was pointed out (rightly > so) that a lot of test servers never have a hostname associated with them, > and not being able to talk h2 to test servers seems like a Bad Idea :) > > FWIW, I checked Safari, Chrome, IE (11 on Windows 7), and Firefox. Both > Safari and Chrome send SNI regardless of IP or hostname, so they will not > run into this problem. IE and Firefox both send SNI only for hostnames (at > least in the configurations I tested), so they will hit this problem. > (Obvious caveat: non-Firefox browsers may have changed their behavior in > later versions than I have access to, so of course my testing may not hold > true in the future.) > > I talked briefly to Martin offline, and he says we may be able to get a > clarification on this point in during AUTH48 to (my words, now, not his) > perhaps relax this restriction, or at least make it clear that you probably > don't need to require SNI in a testing situation, in order to avoid this > problem. > > Thoughts? >
Received on Friday, 3 April 2015 19:07:03 UTC