One of the patches for bug 591483 introduces a page refcount leak. Machines that have CIFS filesystems under heavy I/O (particularly for reads) are susceptible to OOM kills. I've proposed a patch upstream to fix this which has been acked by Nick Piggin.
Once I confirm that the CIFS maintainer will take in the patch, I'll post it for RHEL6. I think we want this in the next beta if at all possible.
Patch(es) available on kernel-2.6.32-32.el6
I run fsstress on cifs filesystem for a few hours and no OOM kills error occurs on.
# uname -a
Linux dell-pem605-01.rhts.eng.bos.redhat.com 2.6.32-54.el6.x86_64 #1 SMP Wed Jul 28 00:06:54 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
# rpm -qa|grep samba
# rpm -qa|grep nfs-utils
# cat /proc/mounts
//localhost/public/ /cifs cifs rw,mand,relatime,unc=\\localhost\public,username=root,uid=0,noforceuid,gid=0,noforcegid,addr=0000:0000:0000:0000:0000:0000:0000:0001,posixpaths,acl,rsize=16384,wsize=57344 0 0
# nohup ./fsstress -d /cifs -l 0 -n 1000 -p 4 -r &
Red Hat Enterprise Linux 6.0 is now available and should resolve
the problem described in this bug report. This report is therefore being closed
with a resolution of CURRENTRELEASE. You may reopen this bug report if the
solution does not work for you.