[Samba] Hangs trying to lock locking.tdb ?

[Dave Gilbert (Home)]

Hi,
   server Kernel: Linux 2.4.21
   server Samba: 2.2.3a-12.3 from Debian package
   Client      : Win2k (& XP?)

We have a problem where when someone browses a directory on a share they 
see a LONG (~1 minute?) pause.  The server is very lightly loaded.  This 
doesn't seem to happen if they are logged on as nobody/guest, only if 
they are logged on as a user; it happens for multiple people from 
multiple clients.  The data is stored on a local ext3 filesystem to the 
Samba server.

On examining the server it looks like there are a lot of smbd's running 
for each user who observes the problem; and in the logs there are 
multiple 'connect to service ... as user ...' with differnt PIDs
and the gaps between them may well correspond to about the length of the 
pauses; there is no close between the multiple 'connect' lines.  The 
older process still exists.

So I'm assuming something times out and tries for another connection.

An ethereal trace shows a NT CREATE packet and then very little else 
from the server for the gap and then it looks like it relogs on.

So I took strace to the older of the processes and found it is stuck in:

  fcntl64(13, F_SETLKW64, {type=F_WRLCK, whence=SEEK_SET, start=284, 
len=1} <unfinished ...>

13 corresponds to locking.tdb; this file is also on a local (ext3) 
filesystem.

smbstatus shows many open services for the afflicted users but no locked 
files.


  Is this a known problem known to be fixed in a later version? I can't 
see any comments along those lines in the Debian changelogs for the 
Woody version of Samba; I'd be happy to update to something else if 
there was a reasonable confidence it was a known fixed issue.

  Suggestions on how to debug this further would be welcome.

Thanks in advance,

Dave
P.S. I'd appreciate a cc on the reply but I'll poll the list.