Re: SETTINGS_MIXED_SCHEME_PERMITTED | Re: I-D Action: draft-ietf-httpbis-http2-encryption-07.txt

[Martin Thomson <[e-mail]>]

And now that I read this thread, I find that the point about origins
over connections is pretty convincing.  I should read all before
committing to mistakes :)

However, perhaps there is some simplification to be salvaged.  I think
that Mike's observation suggests that we can remove "tls-ports".  Once
the TLS-enabled port acknowledges that it understand that it can
receive requests for http://<foo> then maybe that's enough (in
addition to it having a valid certificate, that is).

And, while I'm on the topic, "lifetime" is a bit jarring now that we
don't have a commitment.  To that end, a simpler formulation suggests
itself:

[ "http://example.com", "http://example.com:5602" ]

That should make Mark happy about not having to reconcile "lifetime"
with the cache freshness lifetime.

On 5 October 2016 at 19:53, Patrick McManus <[e-mail]> wrote:
> I think I agree with Kari - this is per origin and not per connection..
> having implemented the .wk thing I found it pretty workable and the server
> side folks liked that this could be done with simply configuration to
> reflect origin policy via headers and .wk rather than a protocol extension.
>
> On Wed, Oct 5, 2016 at 6:51 AM, Kari Hurtta <[e-mail]>
> wrote:
>>
>> Mike Bishop <[e-mail]>: (Tue Oct  4 20:38:45 2016)
>>
>> > Taking a step back, what is the list of ports actually buying us now?
>> > The port can be obtained by the client from the Alt-Svc header.  The fact
>> > that the port is legitimate and not hijacked is verified by finding that it
>> > has a certificate.  What we're actually confirming is that the origin
>> > supports mixed schemes.  The lifetime is already present in the Alt-Svc
>> > advertisement, and I haven't heard a compelling reason to have a separate
>> > lifetime.  Should we just define SETTINGS_MIXED_SCHEME_PERMITTED and call it
>> > a day?
>>
>> Hmm.
>>
>> SETTINGS_MIXED_SCHEME_PERMITTED is per connection. I assume that HTTP/2
>> server sends it on SETTINGS frame to HTTP/2 client (similar than what
>> I contemplated for SETTINGS_WEBSOCKET_CAPABLE at
>> https://lists.w3.org/Archives/Public/ietf-http-wg/2016OctDec/0033.html )
>>
>> http-opportunistic response tells here that given port for that
>> origin handles http -scheme when sent via TLS.
>>
>> connection apply probably for several origins. TLS connection
>> may be terminated by reverse proxy. And different origins
>> are served by different processes or servers behind of
>> reverse proxy.
>>
>> I guess that SETTINGS_MIXED_SCHEME_PERMITTED is too wide.
>>
>> "tls-ports"  should perhaps now be "mixed-scheme-listeners"
>> giving [ "alternative-server:port" ].
>>
>> / Kari Hurtta
>>
>>
>