[prev in list] [next in list] [prev in thread] [next in thread] 

List:       hurd-bug
Subject:    Re: GNUstep - check for reuse address
From:       Samuel Thibault <samuel.thibault () gnu ! org>
Date:       2019-01-29 17:49:01
Message-ID: 20190129174901.woznelpxvmpkqjch () function
[Download RAW message or body]

Svante Signell, le mar. 29 janv. 2019 18:37:02 +0100, a ecrit:
> On Sat, 2016-01-09 at 00:37 +0100, Samuel Thibault wrote:
> > Svante Signell, on Fri 08 Jan 2016 21:59:56 +0100, wrote:
> > > > Yes. And SO_REUSEADDR won't help there :)
> > > 
> > > Samuel, this is exactly what the SO_REUSEADDR in pflocal should do:
> > 
> > Except no Unix makes it do that.
> 
> Well, it works for GNU/Linux.

No, it doesn't actually. The function returns 0, but GNU/Linux doesn't
actually do anything, one actually has to unlink the old socket by hand
to be able to reuse the path.

> > > Unlink the old socket and create a new one with the same name. (I
> > > wonder how GNU/Linux is implementing this?)
> > 
> > It doesn't.
> 
> How is it implemented then? I need to find a good testcase for this.

It just doesn't implement it. See for instance (in Python to make it
simpler):

#!/usr/bin/python3
import socket
s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM, 0)
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind("/tmp/foo")
s.close()

Try to run that twice on Linux, it'll fail.

Really, SO_REUSEADDR on local sockets doesn't make sense, it should just
not be called on local sockets.

Samuel

[prev in list] [next in list] [prev in thread] [next in thread]