smb_panic and pdb

[Andrew Bartlett]

On Mon, 2018-10-29 at 10:21 -0700, Jeremy Allison via samba-technical
wrote:
> On Mon, Oct 29, 2018 at 07:38:55AM +0100, Andreas Schneider wrote:
> > On Friday, 26 October 2018 21:29:23 CET Jeremy Allison wrote:
> > > On Wed, Oct 24, 2018 at 05:13:04PM +0200, Andreas Schneider via samba-
> > 
> > technical wrote:
> > > > Hello,
> > > > 
> > > > we have a retrace report which starts to glow. winbind is create a core
> > > > file during a name lookup because it fails to load the pdb methods and
> > > > calls smb_panic().
> > > > 
> > > > This is probably a configuration issue.
> > > > 
> > > > However how should we deal with that in a more graceful way and write what
> > > > is wrong the the log file instead of core dumping?
> > > > 
> > > > This has been reported over 6000 times in the meantime. The backtrace is
> > > > here:
> > > > 
> > > > https://retrace.fedoraproject.org/faf/reports/1577174/
> > > 
> > > The backtrace shows it printing a name of the backend
> > > method it's trying and failing to find:
> > > 
> > >  224                 char *msg = NULL;
> > >  225                 if (asprintf(&msg, "pdb_get_methods: "
> > >  226                              "failed to get pdb methods for backend
> > > %s\n", 227                              lp_passdb_backend()) > 0) {
> > >  228                         smb_panic(msg);
> > > 
> > > Do you have a log file showing what the
> > > name is ?
> > > 
> > > It'll be what the user configured in:
> > > 
> > > passdb backend = XXXXXX
> > > 
> > > so I'm guessing a miss-spelling here will
> > > cause the smb_panic.
> > > 
> > > We should probably print the error message
> > > here and exit more gracefully.
> > 
> > How do we do that, call exit_server_cleanly() instead or should we correctly 
> > return error codes in pdb functions?
> 
> Looking at return error codes from pdb functions means changing
> more code, but is probably a better long term solution. For
> the immediate fedora bug it's probably better to exit. Check
> what happens in winbindd to exit_server_cleanly(), I think it
> just calls exit(0) via the shim.

My suggestion is to have winbindd (and a few other key utilities) check
this at startup when we prepare messaging etc by calling
pdb_get_methods_reload() directly.

That could then print a clear, concise message to the user.

BTW: I had considered this might be an AD DC bug, where the
pdb_samba_dsdb modules might be in different and not installed package,
but we wouldn't get to this codepath on the AD DC (we go via ncacn_np
pipes for LSA).  

However could there be some other situation where winbindd is installed
without all the passdb plugins being present?

Thanks,

Andrew Bartlett

-- 
Andrew Bartlett                       http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba