Re: Some observations on ssh/sshd

[Svante Signell]

On Wed, 2015-09-09 at 09:42 +0200, Samuel Thibault wrote:
> Uh.
> 
> I just never noticed those.  On which box do you see that?  Can you
> reproduce this on a freshly installed box?

I see this on several qemu boxes. Doing it on a freshly installed one
will take some time.

> Svante Signell, le Wed 09 Sep 2015 08:09:11 +0200, a écrit :
> > 1) In more than 50% of the times of removing a socket with unink or rm
> > ssh/sshd hangs.
> 
> When you say "removing a socket", you mean a named socket which was just
> created for whatever use?

Yes, removing a named socket, AF_UNIX and either SOCK_DGRAM or
SOCK_STREAM

> > 2) Starting sshd on the client normally, ssh connections are not
> > responding until 
> 
> When you mean "Starting sshd on the client", you mean running
> /etc/init.d/sshd restart?

I meant on the qemu client, should have written qemu guest.
No, e.g. ssh -p 5557 localhost

> > task (/bin/sh(576)) increasing a bogus port 104 by 1, most probably a
> > bug.
> > task (/bin/sh(576)) deallocating a bogus port 104 by 1, most probably a
> > bug.
> 
> I've never seen those messages for /bin/sh.

Old qemu installations upgraded to latest released Debian Hurd versions.

> > 3) When ssh connections are locked, see 2) above, logging in on the
> > console and
> > /etc/init.d/ssh stop
> > /etc/init.d/ssh start
> > usually hangs the sshd start. Breaking the start with ^C/^Z often
> > results in a segfault. After that sshd starts fine.
> 
> Never seen that either.