Users LOCKED OUT (was :Re: printing, locking, etc in 2.2.0)

[Andrew Bartlett]

Sorry about the incomplete e-mail, my mouse slipped...

Andrew Bartlett wrote:
> 
> Jeremy Allison wrote:
> >
> > Andrew Bartlett wrote:
> > >
> > > My installation of last nights CVS, plus my pam patch (irrelevent for
> > > this particular code path) also shows:
> > >
> > > May 23 09:44:42 jessie smbd[15437]: art14 (192.168.29.10) closed
> > > connection to service webbc
> > > May 23 09:44:42 jessie smbd[15417]: too many connections - rejected
> > >
> > > My theroy is that process 15437 is locking at least part of
> > > connections.tdb, and that this code in connections.c:claim_connection()
> > > is then failing:
> > >
> > >         if (tdb_store(tdb, kbuf, dbuf, TDB_REPLACE) != 0)
> > >                 ret = False;
> > >
> > > Can this happen?  Should it be fixed?  Note that we have never had
> > > connection limits at our site.  In any case, it has only happend once
> > > (that I have seen), and is certainly not critical.
> >
> > Hmmm. That's a concern. I've checked over the code, and
> > the tdb_store should block until it gets a lock, even if
> > it's in competition with another smbd.
> >
> > I've added some tdb_errstr() calls to smbd/connection.c
> > so we can see what's going on if any tdb calls fail.
> >
> 
> Unfortunetly we seem to have hit a bigger bug in this area, while
> previously I had only seen one line in the logfile to indicate the
> problem, this time I herd it by mobile...  Only about 8 studnets + staff
> were able to log into the server, with the remainder locked out with the
'too many connections - rejected' message.  
> 
> I've cvs updated and installed, causing the problem to 'go away'
because it also restarted samba.  I'll see what the error is if we ever
get it again.

In the meantime, i have the connections.tdb, all 8MB of it.  I wonder if
this is some internal limit?  Note that we have NO error handling for
the delete case, a failure doesn't rate the slightest mention.  Could
the be an issue?

Anyway, is there somebody who would like to analyze the tdb?  

Andrew Bartlett

-- 
Andrew Bartlett
abartlet at pcug.org.au