Milter-greylist

> I suspect you got the message in question even before you tried to fix
> the problem, and you get "cannot set IPV6_V6ONLY:" message as well,
> right? 

Yes, I got it once.

> If so, your system uses IPv4-mapped IPv6 address and we cannot disable
> use of it.  In that case, your system listen IPv4 by IPv6 socket.  So,
> you don't need to listen on IPv4 separately.
> 
> attila> Probably with SO_REUSEPORT that message would go away.
> 
> If is is your case, the error is expected behavior, and use of
> SO_REUSEPORT is not good way, IMHO.  Just disabling the message is
> enough.
> 
> Index: sync.c
> diff -u -p sync.c.orig sync.c
> --- sync.c.orig	2009-05-15 16:55:20.000000000 +0900
> +++ sync.c	2009-05-15 18:42:14.249372861 +0900
> @@ -961,15 +961,19 @@ sync_listen(addr, port, sms)
>  		optval = 1;
>  		if ((setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
>  		    &optval, sizeof(optval))) != 0) {
> +#if 0
>  			mg_log(LOG_ERR, "cannot set IPV6_V6ONLY: %s",
>  			    strerror(errno));
> +#endif
>  		}
>  	}
>  #endif
>  
>  	if (bind(s, SA(&laddr), laddrlen) != 0) {
> +#if 0
>  		mg_log(LOG_ERR, "cannot start MX sync, bind failed: %s",
>  		    strerror(errno));
> +#endif
>  		sms->runs = 0;
>  		close(s);
>  		return;

I still feel it is an overhead to try
to open a new socket for every new recipient,
even if there is no error message about the
bind failure because we hide it.

What about the following:
When we once get the message
"cannot set IPV6_V6ONLY"
consider it as serious
(actually no real support for IPv6 in the kernel)
so close the socket do not start
the IPv6 sync_master thread
but set the sync_master6.runs equals to -1
without associated thread running.
The mg will not try to start it again later.

That way the IPv4 thread sync_master thread will start
happily no need the SO_REUSEPORT.

Please see the attached patch for an implementation.
Also based on the vanilla 4.3.2.

Bests,
Attila