Re: [Bugme-new] [Bug 32832] New: shutdown(2) does not fully shut down socket any more

On Wed, Apr 13, 2011 at 04:55:27AM +0200, Eric Dumazet wrote:

Le mardi 12 avril 2011 à 16:17 -0700, David Miller a écrit :

quoted

From: Andrew Morton <akpm@linux-foundation.org>
Date: Tue, 12 Apr 2011 16:15:56 -0700

quoted

(switched to email.  Please respond via emailed reply-to-all, not via the
bugzilla web interface).

Stephen Hemminger forwarded this to the list last week, and Eric
Dumazet is actively working on a fix.

I worked on it this week end to discover FreeBSD 8.1 would not allow
several CLOSE sockets bound to same port even with REUSEADDR.

So haproxy claim is a bit wrong (its trick doesnt work on FreeBSD), and
used an undocumented linux feature.

I feel this case is a call for SO_REUSEPORT, eventually.

http://www.unixguide.net/network/socketfaq/4.11.shtml

  SO_REUSEADDR allows your server to bind to an address which is in a
  TIME_WAIT state.  It does not allow more than one server to bind to
  the same address.  It was mentioned that use of this flag can create a
  security risk because another server can bind to a the same port, by
  binding to a specific address as opposed to INADDR_ANY.  The
  SO_REUSEPORT flag allows multiple processes to bind to the same
  address provided all of them use the SO_REUSEPORT option.


Since SO_REUSEPORT is not a 'stable fix', I suggest we revert the patch,
and eventually work on SO_REUSEPORT on net-next-2.6

What do you think ?

Not entirely related, but FWIW I think that SO_REUSEPORT would
be rather useful for haproxy.

I've been working on allowing haproxy to be reconfigured without dropping
or reusing connections. I have done this by re-using open sockets. But it
would have been rather a lot easier to achieve using SO_REUSEPORT -
assuming that I understand SO_REUSEPORT correctly.