You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Justin Erenkrantz <ju...@erenkrantz.com> on 2004/04/03 19:29:00 UTC
Re: cvs commit: apr/network_io/win32 sockopt.c
--On Saturday, April 3, 2004 5:15 PM +0000 jerenkrantz@apache.org wrote:
> jerenkrantz 2004/04/03 09:15:52
>
> Modified: . CHANGES
> network_io/unix sockopt.c
> network_io/win32 sockopt.c
> Log:
> Even if we do not check the value of APR_IPV6_V6ONLY before setting, we
> still must set the option cache if we're setting APR_IPV6_V6ONLY.
> Otherwise, later apr_socket_opt_get calls will fail.
>
> This fixes Listen directives on IPv6 machines where IPV6_V6ONLY may be
> enabled (i.e. FreeBSD 5.x, OpenBSD, NetBSD) where it would not properly
> bind to the IPv4 socket.
This fixes an issue we saw on hermes.apache.org where the Listen directive
wasn't binding to the correct interfaces.
I was pretty sure that httpd was doing the right thing here, but it was APR
that lost the setting of APR_IPV6_V6ONLY - so httpd thought that IPV6_V6ONLY
was disabled when it really had already been enabled. Hence, it collapsed the
IPv4 address into the IPv6 socket - causing it not to bind to IPv4. -- justin
Re: cvs commit: apr/network_io/win32 sockopt.c
Posted by André Malo <nd...@perlig.de>.
* Justin Erenkrantz <ju...@erenkrantz.com> wrote:
> --On Sunday, April 4, 2004 12:08 AM +0200 André Malo <nd...@perlig.de> wrote:
>
> > It doesn't compile.
> >
> > apr_set_option(sock, APR_IPV6_V6ONLY, on);
>
> Would have saved me about 10 minutes of searching if you had said *which*
> branch and OS it didn't compile on. ;-)
ugh, sorry ;-) next time ...
thanks.
nd
Re: cvs commit: apr/network_io/win32 sockopt.c
Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Sunday, April 4, 2004 12:08 AM +0200 André Malo <nd...@perlig.de> wrote:
> It doesn't compile.
>
> apr_set_option(sock, APR_IPV6_V6ONLY, on);
Would have saved me about 10 minutes of searching if you had said *which*
branch and OS it didn't compile on. ;-)
Anyhoo, just corrected it on 0.9 branch for Unix. I caught that for Win32,
but apparently not for Unix... *sigh* -- justin
Re: cvs commit: apr/network_io/win32 sockopt.c
Posted by André Malo <nd...@perlig.de>.
* Justin Erenkrantz <ju...@erenkrantz.com> wrote:
> --On Saturday, April 3, 2004 5:15 PM +0000 jerenkrantz@apache.org wrote:
>
> > jerenkrantz 2004/04/03 09:15:52
> >
> > Modified: . CHANGES
> > network_io/unix sockopt.c
> > network_io/win32 sockopt.c
It doesn't compile.
apr_set_option(sock, APR_IPV6_V6ONLY, on);
should probably be
apr_set_option(&sock->netmask, APR_IPV6_V6ONLY, 1);
or something?
nd
Re: cvs commit: apr/network_io/win32 sockopt.c
Posted by Jess Holle <je...@ptc.com>.
While patching sockopt.c, I should point out the need for the following
case in the Win32 sockopt.c's apr_socket_opt_set routine:
case APR_SO_SNDBUF:
if (setsockopt(sock->socketdes, SOL_SOCKET, SO_SNDBUF, (void
*)&on, sizeof(int)) == -1) {
return apr_get_netos_error();
}
break;
Without it one cannot set the send buffer size on Windows, which leads
to a performance penalty on Windows.
--
Jess Holle
Justin Erenkrantz wrote:
> --On Saturday, April 3, 2004 5:15 PM +0000 jerenkrantz@apache.org wrote:
>
>> jerenkrantz 2004/04/03 09:15:52
>>
>> Modified: . CHANGES
>> network_io/unix sockopt.c
>> network_io/win32 sockopt.c
>> Log:
>> Even if we do not check the value of APR_IPV6_V6ONLY before
>> setting, we
>> still must set the option cache if we're setting APR_IPV6_V6ONLY.
>> Otherwise, later apr_socket_opt_get calls will fail.
>>
>> This fixes Listen directives on IPv6 machines where IPV6_V6ONLY may be
>> enabled (i.e. FreeBSD 5.x, OpenBSD, NetBSD) where it would not
>> properly
>> bind to the IPv4 socket.
>
>
> This fixes an issue we saw on hermes.apache.org where the Listen
> directive wasn't binding to the correct interfaces.
>
> I was pretty sure that httpd was doing the right thing here, but it
> was APR that lost the setting of APR_IPV6_V6ONLY - so httpd thought
> that IPV6_V6ONLY was disabled when it really had already been
> enabled. Hence, it collapsed the IPv4 address into the IPv6 socket -
> causing it not to bind to IPv4. -- justin
>
Re: cvs commit: apr/network_io/win32 sockopt.c
Posted by Jess Holle <je...@ptc.com>.
While patching sockopt.c, I should point out the need for the following
case in the Win32 sockopt.c's apr_socket_opt_set routine:
case APR_SO_SNDBUF:
if (setsockopt(sock->socketdes, SOL_SOCKET, SO_SNDBUF, (void
*)&on, sizeof(int)) == -1) {
return apr_get_netos_error();
}
break;
Without it one cannot set the send buffer size on Windows, which leads
to a performance penalty on Windows.
--
Jess Holle
Justin Erenkrantz wrote:
> --On Saturday, April 3, 2004 5:15 PM +0000 jerenkrantz@apache.org wrote:
>
>> jerenkrantz 2004/04/03 09:15:52
>>
>> Modified: . CHANGES
>> network_io/unix sockopt.c
>> network_io/win32 sockopt.c
>> Log:
>> Even if we do not check the value of APR_IPV6_V6ONLY before
>> setting, we
>> still must set the option cache if we're setting APR_IPV6_V6ONLY.
>> Otherwise, later apr_socket_opt_get calls will fail.
>>
>> This fixes Listen directives on IPv6 machines where IPV6_V6ONLY may be
>> enabled (i.e. FreeBSD 5.x, OpenBSD, NetBSD) where it would not
>> properly
>> bind to the IPv4 socket.
>
>
> This fixes an issue we saw on hermes.apache.org where the Listen
> directive wasn't binding to the correct interfaces.
>
> I was pretty sure that httpd was doing the right thing here, but it
> was APR that lost the setting of APR_IPV6_V6ONLY - so httpd thought
> that IPV6_V6ONLY was disabled when it really had already been
> enabled. Hence, it collapsed the IPv4 address into the IPv6 socket -
> causing it not to bind to IPv4. -- justin
>