45891 - IPv6: "Host" header is wrong in requests to IPv6 literal addresses

Status: RESOLVED FIXED

(Core :: Networking, defect, P3)

Description

[Jonathan Lennox]

(This bug report is largely thanks to David Madore.  I tracked down the specific
RFC grammar references.)

When one uses Mozilla to retrieve a URL with a literal IPv6 address
and port number, e.g.

        http://[::1]:8088/foo/

(this means "get relative /foo/ from HTTP server on port 8088 of the
loopback address (::1) using the syntax defined by RFC 2732), the
request it sends to the port is something like this:

GET /foo/ HTTP/1.1
Host: ::1:8088
User-Agent: Mozilla/5.0 (X11; U; Linux 2.2.17pre10-emerald i686; en-US; m17)
Gecko/20000709
Accept: */*
Accept-Language: en
Accept-Encoding: gzip,deflate,compress,identity
Keep-Alive: 300
Connection: keep-alive

Note in particular the "Host" header.  RFC 2617 (HTTP) defines the syntax of the
"Host" header as:

       Host = "Host" ":" host [ ":" port ] ; Section 3.2.2

Section 3.2.2 gives the grammar of the http URL.  Therefore, for IPv6 literal
addresses the Host header needs to follow the RFC 2732 grammar for "host", just
as with URLs:

      host          = hostname | IPv4address | IPv6reference
      IPv6reference = "[" IPv6address "]"

Without this, the server cannot tell whether what is meant is port 8088 of
host ::1 rather than (default) port 80 of host ::1:8088 (which is a
valid IPv6 address).

The correct way for the "Host" header to appear in this request is

Host: [::1]:8088

Experimentally, using the KAME IPv6 patch for Apache-1.3.12, it seems
that the latter syntax is recognized (and the former is not)

Comment 1

[John G. Myers]

Attachment: proposed fix

Comment 2

[John G. Myers]

Reassigning to module owner for review and checkin (I don't have CVS access).

Comment 3

[John G. Myers]

I have CVS access now, but I need review and approval.

Comment 4

[Wan-Teh Chang]

I reviewed jgmyers' patch and think it is good.
(I am not the module owner.)

Comment 5

[Gagan]

seems reasonable to me... r=gagan

Comment 6

[Jonathan Lennox]

Using strchr seems inefficient -- it means scanning every host string so that
the uncommon case of IPv6 literals is handled correctly -- but given that
nsStdURL::AppendString does the same thing, I suppose it's reasonable.

Comment 7

[John G. Myers]

Reassign to self, waiting for approval.

Comment 8

[John G. Myers]

Assign to waterson for approval.

Comment 9

[Chris Waterson]

a=waterson. jgmeyers: in the future, just send me email or cc me on the bug; you
don't have to assign the bug to me.

Comment 10

[John G. Myers]

Fix checked in.

Comment 11

[John G. Myers]

*** Bug 55815 has been marked as a duplicate of this bug. ***

Comment 12

[benc]

Relieving tever of IPv6 related QA.

Comment 13

[benc]

Jonathan: if this works for you, can you VERIFY this bug?

Comment 14

[benc]

-> pushing out networking qa of ipv6.