samba - Mar 2009 - Samba logs fill up disk with inotify errors, smbd 100% CPU

Hello,
One of the server I maintain has Ubuntu 8.04 LTS server with samba 
3.0.28a. Kernel version is 2.6.24-16-server. The samba setup for this 
server is a workgroup environment, with user level security.
At least once a week, sometimes more often, the server becomes 
unavailable because the system disk is full. It appears that the samba 
logs are the problem. Although the samba configuration file indicates 
that logs should be limited to 1000K, one of the machine specific logs 
(/var/log/samba/log.machine_xyz.old) grows to the point of entirely 
filling the disk (140 GB of log). The log file contains endless 
repetitions of the same message:
smbd/notify_inotify.c:inotify_handler(240) No data on inotify fd?!
At the same time, the smbd process for the user related to the faulty 
machine log uses 100% CPU, and must be stopped with kill -9. Note that 
it not always the same machine that creates the problem.
Because some users are never affecting the system, we tried to identify 
if the origin of the problem was linked to a specific data manipulation 
by some of the 'bad' users, but we are still clueless.
I found some info in this list with similarities to my problem where the 
suggestion is to include "notify:inotify = false" in [global] to see
if
it helps (http://www.mail-archive.com/samba%40lists.samba.org/msg98074.html)
I've just done this, so I will have to wait and see.
Otherwise, would you have any idea or suggestion ?
Any help would be greatly appreciated.

Am Dienstag, 17. M?rz 2009 schrieb Stephane Carre:
> Hello,
> One of the server I maintain has Ubuntu 8.04 LTS server with samba 
> 3.0.28a. Kernel version is 2.6.24-16-server. The samba setup for this 
> server is a workgroup environment, with user level security.
> At least once a week, sometimes more often, the server becomes 
> unavailable because the system disk is full. It appears that the samba 
> logs are the problem. Although the samba configuration file indicates 
> that logs should be limited to 1000K, one of the machine specific logs 
> (/var/log/samba/log.machine_xyz.old) grows to the point of entirely 
> filling the disk (140 GB of log). The log file contains endless 
> repetitions of the same message:
> smbd/notify_inotify.c:inotify_handler(240) No data on inotify fd?!
> At the same time, the smbd process for the user related to the faulty 
> machine log uses 100% CPU, and must be stopped with kill -9. Note that 
> it not always the same machine that creates the problem.
> Because some users are never affecting the system, we tried to identify 
> if the origin of the problem was linked to a specific data manipulation 
> by some of the 'bad' users, but we are still clueless.
> I found some info in this list with similarities to my problem where the 
> suggestion is to include "notify:inotify = false" in [global] to
see if
> it helps
(http://www.mail-archive.com/samba%40lists.samba.org/msg98074.html)
> I've just done this, so I will have to wait and see.
> Otherwise, would you have any idea or suggestion ?
> Any help would be greatly appreciated.
> 
looks like a buggy kernel. There have been kernel issues with inotify:
  http://ubuntuforums.org/showthread.php?t=655504
  https://bugzilla.novell.com/show_bug.cgi?id=463372
Google will return more results...

Cheers, G?nter