[Samba] Strange crashes and disconnection from PDC?

[Andrew Bartlett]

Nir Soffer wrote:
> 
> > Can you set a 'panic action', (panic action = /bin/sleep 9000 works
> > well) and attach a debugger?
> >
> > We need a 'bt full' to see what's going on.  Even better if you can
> > compile with -g (--enable-debug configure switch for that).
> >
> >
> >> I'll be glad if anyone has any ideas... Is this is a known issue in
> >> 2.2.4? Has it been resolved? Isn't the SMB Magic supposed to be in
> the
> >> beggining of the packet and not in the middle? Could this be some
> >> buffer going ballistic and screwing up the alignment of the packet
> >> with something else, consequently causing a SIGSEGV?
> 
> > Interesting theory.  That certainly is the SMB signiture (0ff SMB).
> 
> > See what you can get out of the debugger - and the last statements
> from
> > a high level debug could help.
> 
> Grumble grumble. Getting Outlook quoting to work like pine/elm won't
> be easy.
> 
> Oh well - to the task at hand...

:-)

> I'm afraid that this has happened only once - and I have no idea what
> triggered it, so I can't really try to bt it. If I could, I would. :).
> 
> Am I right in what I thought? The SMB signature should be around
> offset zero, and not the middle of the packet?

Indeed - assuming that NT correctly captured the request.  Could it be
the whole IP packet?  With the IP header etc?

> If you have any idea what could trigger such a behaviour, I'll be glad
> to attempt to reproduce this... Otherwise, I'm afraid I'm at a dead
> end.
> 
> To elaborate a bit more on the configuration - this is a cluster
> environment, where two nodes (the ones running smbd) experienced
> the same symptoms simultenously. This happened on another cluster,
> and IIRC on both nodes too, albeit two hours later.
> 
> This has me confused. The PDC could've have been spewing some
> garbage or something, but that would've crashed several other
> servers. Yet it didn't. The fact that it happened in two nodes of the
> same cluster suggests it was some sort of a PDC screw up wrt to the
> specific NetBIOS name of that cluster... We've been getting several
> log entries before this crash that had something to do with the fact
> that it lost the credentials on the PDC.
> 
> Being of the Windows world, the error code was naturally 0, so all I
> can tell you that the operation failed because of SUCCESS... :)

Can you get me the log entries you do have?

> To make a long and rambling post short -
> 
> Happened only once, don't know how to reproduce it, will surely do so
> if/when it happens. Will be glad for some clues as to how to reproduce
> it.

I'll see if the log file can give me some clues - but I'm at a loss
too...

Andrew Bartlett

-- 
Andrew Bartlett                                 abartlet at pcug.org.au
Manager, Authentication Subsystems, Samba Team  abartlet at samba.org
Student Network Administrator, Hawker College   abartlet at hawkerc.net
http://samba.org     http://build.samba.org     http://hawkerc.net