samba daemon deadlocks

Bug #250655 reported by Jaap Versteegh
This bug report is a duplicate of:  Bug #117730: cannot use sqlite3 on cifs mount. Edit Remove
4
Affects Status Importance Assigned to Milestone
samba (Ubuntu)
Incomplete
Undecided
Unassigned

Bug Description

Binary package hint: samba

I have firefox's (3.0) profile stored on a samba share. Ocasionally (about twice a day) when entering data into a webcontrol, firefox freezes. The cause seems to be the samba daemon on the server that deadlocks. A new samba daemon is spawned after a few seconds, but the dead daemon remains in memory and also keeps locking the firefox profile directory. It can only be killed with kill -9.
Connecting to the dead daemon with gdb produces the following callstack:
#0 0x00007faf9272023e in __lll_lock_wait_private () from /lib/libc.so.6
#1 0x00007faf926b4b92 in _L_lock_15195 () from /lib/libc.so.6
#2 0x00007faf926b3c11 in free () from /lib/libc.so.6
#3 0x00007faf926cb0ce in tzset_internal () from /lib/libc.so.6
#4 0x00007faf926cb177 in __tz_convert () from /lib/libc.so.6
#5 0x0000000000605a16 in current_timestring ()
#6 0x00000000006001fc in dbghdr ()
#7 0x0000000000601544 in ?? ()
#8 <signal handler called>
#9 0x00007faf926b05d0 in _int_malloc () from /lib/libc.so.6
#10 0x00007faf926b1b88 in _int_realloc () from /lib/libc.so.6
#11 0x00007faf926b3ea0 in realloc () from /lib/libc.so.6
#12 0x0000000000614d5a in Realloc ()
#13 0x00000000005c4c42 in brl_lock ()
#14 0x00000000005c1d7c in do_lock ()
#15 0x00000000004a469b in ?? ()
#16 0x00000000004a9618 in handle_trans2 ()
#17 0x00000000004afc3a in reply_trans2 ()
#18 0x00000000004c879e in ?? ()
#19 0x00000000004c9b92 in smbd_process ()
#20 0x00000000006c5f5d in main ()

Ubuntu is a completely up-to-date 8.04 (on both desktop and server machines).
Samba installed on the server is:
Architecture: amd64
Version: 3.0.28a-1ubuntu4.4

Smb.conf is added as an attachment.

Files in /var/log/samba have no entries at the time of the deadlock occurence.

Revision history for this message
Jaap Versteegh (j-r-versteegh) wrote :
Revision history for this message
Mathias Gug (mathiaz) wrote : Re: [Bug 250655] [NEW] samba daemon deadlocks

On Mon, Jul 21, 2008 at 11:46:39PM -0000, Jaap Versteegh wrote:
> I have firefox's (3.0) profile stored on a samba share. Ocasionally (about twice a day) when entering data into a webcontrol, firefox freezes. The cause seems to be the samba daemon on the server that deadlocks. A new samba daemon is spawned after a few seconds, but the dead daemon remains in memory and also keeps locking the firefox profile directory. It can only be killed with kill -9.

How do you mount your home directory on your client ?

  status incomplete

--
Mathias Gug
Ubuntu Developer http://www.ubuntu.com

Mathias Gug (mathiaz)
Changed in samba:
status: New → Incomplete
Revision history for this message
Toby Collett (thjc) wrote :

This may muddy the waters a little, but I am also seeing similar issues with samba mounts, which seem to be related to firefox 3.0, or at least made worse by it...

We are using pam_mount to mount user home drives off a samba server. The server is running Gutsy (samba 3.0.26a-1ubuntu2.5) and the client has recently been upgraded to Hardy (samba 3.0.28a-1ubuntu4.4).

The behaviour seems to vary a little, I am working on extracting more detail, but have limited time to focus on this particular issue, currently I have moved the firefox profile onto a local partition instead.

The common case was a disconnection occured with the server for an undetermined reason, which triggered a reconnection to the server, as posted above this lead to many files being locked out as the previous connection had not been cleaned up correctly.

Once the system has these lock issues pretty much all of the applications running start to lock up (due to the server home drive mount).

I have also seen some kernel oops messages from the cifs driver, but cant track any of these down in the logs at the moment, so will post if they reoccur. These seem to be independant of the lock up issues (or perhaps caused by rather than triggering).

The home drive mount : cat /proc/mounts
//infosphere/tcollett /home/tcollett cifs rw,mand,relatime,unc=\\infosphere\tcollett,username=tcollett,uid=9101,gid=513,posixpaths,rsize=16384,wsize=57344 0 0

Also there seems to have been some possibly related issues running firefox 3 on Gutsy, but all the affected users had downgraded to firefox 2 by the time I upgraded to hardy, so I am unable to confirm this.

This issue is the remaining one to resolve before the office switches to Hardy, so hopefully it can be resolved, I will do my best to provide additional information as requested or as it becomes available

Revision history for this message
Mathias Gug (mathiaz) wrote :

There are known issues related to cifs (namely byte range locking not working correctly with sqlite3, which is used by firefox3).

See bug 117730.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Bug attachments

Remote bug watches

Bug watches keep track of this bug in other bug trackers.