Bug#393704: interactive client dies on ill-timed SIGWINCH
On Tue, Oct 17, 2006 at 02:30:15PM +0100, Ian Jackson wrote:
> I often ask my window manager to bring up `xterm -e ssh somehost'. I
> have the wm configured to allow me to interactively place resize the
> window. Sometimes, when I do this, the xterm and ssh vanish. Process
> accounting shows ssh dying with exit status 255.
>
> I've spoken to other users and they report that this happens to them
> sometimes too.
>
> I suspect that the SIGWINCH arrives at a time when ssh is buggily
> unable to cope with EINTR.
This would not especially surprise me.
> I realise that this report isn't particularly informative. Please let
> me know what additional tests/diagnostics you'd like me to do.
> Unfortunately I can't reproduce the symptoms on demand; it happens to
> me spontaneously about once every few weeks. Any instrumentation
> you want me to do would have to be reasonable to do for all of my ssh
> sessions until the bug recurs, and would have to not affect the
> timings significantly.
Could you redirect ssh's stderr to a file (if it isn't already going to
~/.xsession-errors or similar) and find out if it emits any error
messages before exiting? The general style of OpenSSH code is such that
I would expect there to be an error message to identify the failing
system call.
Thanks,
--
Colin Watson [cjwatson@debian.org]
Reply to: