hello,

Since a short while (I know it worked with 3.0-rc1) loopback-mounting an 
image on a samba server mounted with cifs oopses the kernel.

Szenario:
- Client system: Thinkpad T60, 32bit, Debian Unstable
- Server system: Lacie edmini NAS device, exporting 500GB disk with 
  samba, 450GB crypto image file on it
- mount -t cifs -o username=backups,uid=1000,gid=0,file_mode=0664,dir_mode=0775,rw //192.168.10.98/backups /mnt/real-backups
- losetup /dev/loop7 /mnt/real-backups/loop.img
- cryptsetup luksOpen /dev/loop7 cifsloop
- mount -t ext4 -o rw,noatime,delalloc,barrier=0 /dev/mapper/cifsloop /mnt/backups

This worked since ages, but since some time after 2.6.39 (last test 
with 3.0-rc5+git ba466c74d99b91b2) this throws a kernel oops at the 
"losetup" stage.

As there were no logs written, full screenshot of the oops here:
https://www.sdinet.de/tmp/IMG_20110704_002831.jpg


Hand-written extract:

BUG: unable to handle kernel NULL pointer dereference at (null)
IP: bdi_queue_work+0x3b/0x70
Process losetup
Call trace:
 writeback_inode_sb_nr+0x50
 writeback_inode_sb+0x14
 __sync_filesystem+0x45
 sync_one_sb+0x17
 iterate_supers+0x74
 ? __sync_filesystems+0x80
 sys_sync+0x21
 systenter_do_call+0x12


Please CC: me on replies.

c'ya
sven-haegar

-- 
Three may keep a secret, if two of them are dead.
- Ben F.