Rapid transfers cause kernel panic

[Jeremy Seth Henry]

Our site has been experiencing problems with Samba causing kernel panics
under heavy use. We are running Mandrake 7.0 and Samba 2.0.6. We are no
longer using the distro included with Mandrake - we actually downloaded
the 2.0.6 release and recompiled. Downloading from Samba shares causes no
problems - and downloads are frequent and heavy - but uploading similar
amounts of data leaves a catatonic server at best and a kernel panic at
worst. Dropping several > 1Mb files on a share will usually do it. This
has been the case using a dual processor Celeron 433 and a single
processor AMD K6-2. Both systems have plenty of RAM and disk space and run
fine with NFS. We first noticed the problem on a AMD K5-133 system, and
initially attributed the problem to the NIC, but later tests proved the
hardware is fine. I ran a "samba-less" linux install on it for several
months with no unusual behavior.

This is highly repeatable - to the point that I have turned write
permissions off on all shares. I have taken to uploading with FTP using
LeechFTP to keep from trashing my partitions.

Since I didn't build the dual Celeron system, I will give the specs on my
AMD K6-2 system.

AMD K6-2 350 (not overclocked), 128Mb RAM
the NIC is a LiteON 10/100Mbps board (Linksys) This board sits on a
100Mbps switched ethernet and operates in full duplex. (it's a quick
network)

Although I can't imagine the disks being a problem, here's the disk
layout:

Primary (hda1) is at 25% usage and is a 1.6Gb WD PIO Mode4 drive (/)
Secondary (hdb1) is at 4% usage and is a 1Gb WD PIO Mode3 drive (/home)
Tertiary (hdc1) is at 46% usage and is a 1.2Gb WD PIO Mode4 drive (/usr)

There aren't any shares on those drives though (other than the
"home" directories - here is the SCSI subsystem:

Softraid /dev/md0 consists of 4 identical Seagate 2Gb UW disks (/export1)
Volume is at 49% usage
/dev/sda1 is a 4Gb Quantum UW disk (/export2)
Volume is at 40% usage

All of the shares on the above two filesystems.

The clients are a mix of NT, linux, Win98, Win95 and Win2K. I have been
able to reproduce this problem on all client OS' so far - so that doesn't
appear to be a factor. Also, the problem appears limited to copying files
- opening/saving does't seem to cause a problem - even with Photoshop and
its huge PSD files.

This system normally runs X, so I can't see the kernel panic
screens. Given the repeatable nature of the problem, I can drop out of X
and cause the crash. If any other info is needed - send me an email. We
are trying to find better workarounds or solutions, as the current
situation is kind of ridiculous.

Thanks,
Seth Henry
jshenry at net-noise.com