If the smbd is exiting it should not bury the fact under a DEBUG(10, (...

[Jeremy Allison]

On Sun, May 12, 2013 at 03:00:45PM -0700, Richard Sharpe wrote:
> On Sun, May 12, 2013 at 2:56 PM, Andrew Bartlett <abartlet at samba.org> wrote:
> > On Sun, 2013-05-12 at 14:49 -0700, Richard Sharpe wrote:
> >> On Sun, May 12, 2013 at 2:57 AM, Stefan (metze) Metzmacher
> >> <metze at samba.org> wrote:
> >> > Hi Richard,
> >> >
> >> >> Here is a little patch to have any exits of the smbd log at level 0.
> >> >> We were seeing problems at a customer site and it wasn't until we got
> >> >> lucky with a debug 10 log that we figured out what was going on.
> >> >>
> >> >> Attached is a small patch to fix that. Please discuss and push if acceptable.
> >> >
> >> > This is also called with NT_STATUS_CONNECTION_DISCONNECTED or similar
> >> > things,
> >> > when the transport connection gets disconnected, we should not log this
> >> > at level 0.
> >>
> >> OK, you are correct. Can we change the signature of xxx to include
> >> whether or not we should report this at level 0 or a higher level?
> >>
> >> It is very important in the field to get warnings that software errors
> >> below us are causing an smbd to exit. It took us a week to discover
> >> that EINVAL being returned from the driver, for example, was causing
> >> problems at a customer site.
> >
> > Why are these problems then causing a normal termination, not the
> > abnormal termination path, with the stacktrace?
> 
> Because smbd_server_connection_terminate_ex calls exit_server_cleanly.

That's because client termination is a "normal" error. The question
we really need to solve is how to detect an "abnormal" termination
like your driver issue.