samba - Mar 2004 - System lockups with CIFS

I am attempting to troubleshoot a rather perplexing problem. Whenever I 
attempt to transfer a non-trivial amount of amount to a share mounted 
via the new CIFS kernel mod, my system will experience a hard lock 
requiring reboot.

The setup

linux client, running 2.6.3 (gentoo-dev-sources) with the preemptible 
kernel option enabled. Samba version as reported by smdb --version is 
3.0.1. Using mount.cifs directly and mounting via /etc/fstab with same 
options.

Windows server, running 2003 Server. Share in question is a full 
partition formatted with FAT32 filesystem. SMB signing is enabled.

The Problem

I can mount the share fine from the linux box as mentioned using 
mount.cifs. lsmod shows that the cifs mod is loaded in the kernel and I 
can browse the share, create and remove files and directories. Numerous 
attempts to transfer a number of small data files (mp3's) have all 
resulted in system lockups, but only above a certain amount of 
data(~30mb?) Transferring the same batch of files will result in locks 
at different places, and I have encountered the problem with other 
groups of files. A check on the windows machine shows the session and 
the file being transferred when it hangs as being held open. Killing the 
session does not fix the linux machine. The locked system will not 
accept any terminal input (mouse or kb) and will respond to pings, but 
not allow ssh.

When checking the logs I encounter the following progression:

a large number of these (~30-50)

CIFS VFS: CIFS: caught signal
CIFS VFS: Send error in write = -4

followed by a number of these (~20-25)

CIFS VFS: Need to reconnect after session died to server
CIFS VFS: Send error in write = -5

and finally (varies)
CIFS VFS: Error 0xfffffffb or (-5 decimal) on cifs_get_inode_info in lookup

sometimes there is a large block of

CIFS VFS: CIFS: caught signal

during the session. After that, the system needs to be rebooted. 
Exporting the same partition via samba on the linux box and initiating 
the same transfer from the windows system works fine, so I don't think 
it is a disk or network issue. Before the upgrade to 2003, I had been 
mounting the same disk on the same server fine using smbfs, but the SMB 
  signing prompted me to attempt cifs.

Does anybody have any ideas why this would be happening? I am aware I 
can still use smbfs if I make some registry changes on the 2k3 system, 
but I would really like to know the cause of the lockups.

--brendan

I am having *exactly* the same problem, as soon as I find time I will strace a
cp command from a linux client. I already contacted the maintainer (copy is on
this list). 
Since it occurs in Windows, too, it rather seems to be a bug in the kernel
code or Samba. (I suspected client first)
What type of hard disk(s) do you have in that server?
Which version of CIFS VFS and mount.cifs?

I've got 2.63 stock and mount.cifs 1.1, upgraded the server kernel's
CIFS to
cifs-1.0.4-2.6 but not tested yet. (tho I understodd this only affects client
side anyway, Samba should be resonsible for the server side handling)

As CIFS is twice as fast as smbfs and handles special chars properly (as
opposed to smbfs which drove me nuts) I'd like to keep it.

On Mon, 08 Mar 2004 23:39:31 -0800
Brendan Melia <brendan@northwest.com> wrote:
> I am attempting to troubleshoot a rather perplexing problem. Whenever I 
> attempt to transfer a non-trivial amount of amount to a share mounted 
> via the new CIFS kernel mod, my system will experience a hard lock 
> requiring reboot.