Re: Regression in svnserve: IPv6 causes IPv4 not to work [was: Branching for 1.3.]

Greg Hudson wrote:
> ...
>>Because on OS's that *do* allow both IPv4 and IPv6 connections on
>>the same listening socket, you *can't* bind two separate sockets on
>>the "any" address.
>
>
> Even if we set IPV6_V6ONLY on the v6 socket?

Yes, because on OpenBSD (and perhaps others of the *BSD lineage)
the IPV6_V6ONLY socket option is apparently read-only...

> (If the OS does not define
> IPV6_V6ONLY, then I think we can safely assume that it will allow v4 and
> v6 connections on the same socket.)

No, actually, I don't think that is the case. Before IPV6_V6ONLY,
whether an IPv6 socket would accept IPv4 connections was undefined
by the standards gods. Apparently (I'm going off of the *BSD list
archives here) when they added the IPV6_V6ONLY socket option, some
developers (mainly commercial UNIX vendors) had already allowed
IPv4 over IPv6, while other developers did not. They "compromised"
by saying that IPV6_V6ONLY may be read-only so that no one had to
change their implementations, just add the socket option so that
developers could automatically handle things...

In short, if setting IPV6_V6ONLY to 0 fails, or if IPV6_V6ONLY is
not available, then you need to have a second listen socket bound
to the IPv4 "any" address (assuming that you want that).

-- 
______________________________________________________________________
Michael Sweet, Easy Software Products           mike at easysw dot com
Internet Printing and Publishing Software        http://www.easysw.com
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org