[ofa-general] SRP f/s data corruption

[Cameron Harr]

Hello,
I'm seeing data corruption on an SRP-exported device and I'm fishing for 
any suggestions. I've seen the corruption in several ways, but here's a 
really simple way to reproduce it:

I format /dev/md0 with ext3 on a host (medusa) and export md0 via SRP. I 
mount it on the initiator (harpie), copy over a large file and verify 
that it's md5sum is the same as the original. Then I unmount/remount and 
see that the md5sum is different.

[root at harpie ~]# mount /mnt/medusa/
[root at harpie ~]# cp /usr/src/OFED-1.3.1.tgz /mnt/medusa/
[root at harpie ~]# md5sum /usr/src/OFED-1.3.1.tgz
69fe510fc78a39b627713cfb49ad4ca3  /usr/src/OFED-1.3.1.tgz
[root at harpie ~]# md5sum  /mnt/medusa/OFED-1.3.1.tgz
69fe510fc78a39b627713cfb49ad4ca3  /mnt/medusa/OFED-1.3.1.tgz
[root at harpie ~]# umount /mnt/medusa/
[root at harpie ~]# mount /mnt/medusa/
[root at harpie ~]# md5sum  /mnt/medusa/OFED-1.3.1.tgz
5b761a931bf8fa7273cccc505ff13121  /mnt/medusa/OFED-1.3.1.tgz

As a side note, right after I copy over the file and see that it has the 
correct md5sum, I can mount the same device read only on the target 
server and see the file, but it has a different md5sum.

In searching, I saw this problem here and tried dropping scst_threads to 
1, to no avail:
http://osdir.com/ml/windows.devel.drivers.openib/2007-12/msg00050.html

Ideas?
Thanks,
Cameron