Call for testing: OpenSSH 6.8

[Corinna Vinschen]

Hi Damien,

On Feb 24 04:26, Damien Miller wrote:
> On Mon, 23 Feb 2015, Corinna Vinschen wrote:
> 
> > leaves the ports 3301/3302 in TIME_WAIT state (as is 4242 from some
> > earlier test).  Here are the relevant excerpts from ps -e and (Windows)
> > netstat output.  The first group is the output prior to the above test:
> 
> [snip]
> 
> > This may well be a problem local to Windows.  Btw., the large number of
> > AF_INET sockets is a result of the way how Cygwin implements AF_LOCAL
> > sockets:  They are emulated by local AF_INET sockets since WIndows
> > doesn't know the concept of AF_LOCAL sockets.
> 
> Does CYGWIN implement setsockopt(s, SOL_SOCKET, SO_REUSEADDR, ...)?
> We set this for (AFAIK) all forwarding listeners to prevent TIME_WAIT
> collisions.

Yes, it does.  But given how screwed up this part of the Windows sockets
implementation is, I'm not at all sure this works 100% reliable.

> > Note that there are still sleep processes running.  So on a hunch I just
> > added a `sleep 30' between the two tests and, lo and behold, the
> > forwarding.sh test completes successfully every time:
> 
> IMO it's probably a simple race condition rather than a TCP thing, and
> the test before the "transfer over chained unix domain socket" one does
> look like it fails to wait for the backgrounded ssh to finish (the wait
> doesn't wait for the background ssh, but the one following it).
> 
> Does the following help?

It does!  Thanks, that looks much neater than a 30 seconds sleep :)


Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.mindrot.org/pipermail/openssh-unix-dev/attachments/20150223/1a6905b2/attachment.bin>