570283 - Stop sending Proxy-Connection

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect)

Description

[u7736]

User-Agent:       Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10_6_3; en-au) AppleWebKit/531.22.7 (KHTML, like Gecko) Version/4.0.5 Safari/531.22.7
Build Identifier: 

Mozilla should stop sending the Proxy-Connection header when a HTTP proxy is configured.

This header is non-standard, doesn't work well (because proxies will forward it), and isn't implemented by modern proxies.

Furthermore, removing it will have the worst effect of not negotiating persistent connections with very old proxies; it will not cause interoperability issues.

Reproducible: Always

Comment 1

[u7736]

A bit more.

The worst case scenario here is if a HTTP/1.0 proxy that:
 - does not understand persistent connections to servers (i.e., only close delimitation), AND
 - doesn't understand and therefore forwards Connection

will hang and wait for close, because it unknowingly sends Connection: keep-alive to a HTTP/1.0 server.

However, I note that Safari already sends

  Connection: keep-alive
  Proxy-Connection: keep-alive

in their requests when a proxy is configured, so it appears that this very unlikely configuration isn't seen in practice today.

Comment 2

[error]

Tracing a connection with Firebug does not show the presence of Proxy-Connection in the request headers being sent from my copy of Firefox 3.6.13 when it's set to use a proxy server. Nor is the header seen in Wireshark, which I think is pretty conclusive. If Firefox is sending this header in any circumstances, they aren't normal circumstances that I've been able to reproduce.

Comment 3

[Boris Zbarsky [:bzbarsky]]

Quite odd.  nsHttpHandler::AddStandardRequestHeaders clearly adds a Proxy-Connection header (with comments as to why it's doing it, note), when useProxy is true.

Comment 4

[Boris Zbarsky [:bzbarsky]]

And note that so does nsHttpConnection::SetupSSLProxyConnect.

Comment 5

[Mark Nottingham]

What proxy were you connecting to? Specifically, is it HTTP/1.0 or 1.1?

Comment 6

[Patrick McManus [:mcmanus]]

Attachment: patch 0

mark is right.

Comment 7

[Patrick McManus [:mcmanus]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/d28039477b6d

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/d28039477b6d

Comment 9

[Patrick McManus [:mcmanus]]

in bug 828236 we've got a case of failed NTLM auth on a CONNECT method against a squid/2.6.STABLE9.. there are other reports of 2.7 failure too.

this would probably be a problem with end host NTLM too when using the proxy, but that's going to be pretty rare.

the "worst case" of losing the persistent connection breaks the damn stateful ntlm.