[Triumf-linux-managers] nfs problem in SL5 latest kernels

[Kelvin Raywood]

This post refers to SL-5 installations only.  If you are running Sl-4 or 
some other Linux distribution, then you will not have the issue 
described here.

RedHat recently introduced some changes that into the kernel that 
seriously degrades the performance of nfs v3 clients and causes them to 
really hammer an nfs server.

ScientificLinux and CentOS rebuild the RedHat kernels, so this problem 
is also present in the latest kernels of those distros; kernels 
2.6.18-53.1.6 and 2.6.18-53.1.13.

See
http://listserv.fnal.gov/scripts/wa.exe?A2=ind0801&L=scientific-linux-devel&T=0&P=5427
https://bugzilla.redhat.com/show_bug.cgi?id=431092
http://lists.centos.org/pipermail/centos/2008-January/093336.html

Kernel packages based on 2.6.18-53.1.13 but with a patch to fix the nfs 
problem were built by one of the main CentOS developers.  The patch came 
from a RH engineer who backported it from their RHEL-5.2 test kernel. 
These kernels are compatible with SL.

We normally encourage people to stick with the standard kernels but this 
case warrants a depature from that position. The standard TRIUMF 
kickstart-install sets up several nfs mounts including /triumfcs/linux, 
/triumfcs/mirror and /triumfcs/trshare.  Thus, SL-5 workstations running 
one of the problem kernels can inadvertently seriously impact the 
performance of the central servers.  Also, many TRIUMF groups do nfs 
mounting between their workstations; especially those that are part of 
NIS clusters.

Therefore,  I have added the patched kernel packages to the TRIUMF rpm 
repository so that people who have installed SL-5 from my kickstarts 
will get them with their next "yum update".

So if you are running SL-5, please check your kernel version with
"uname -r".

2.6.18-53.1.4 or earlier: you do not have the nfs problem, but those 
kernels do have the vmsplice privilege-escalation bug.

2.6.18-53.1.6:  This kernel has the nfs and vmsplice bugs

2.6.18-53.1.13: vmsplice-bug fixed but still has nfs problem.

2.6.18-53.1.13.el5.bz321111:  This is that patched kernel that has both 
problems fixed.

If you are running a kernel with the nfs problem then **please update 
and reboot as soon as possible**.  If you are running a kernel with only 
the vmsplice bug, then your decision to update and reboot should be 
based upon the risk of one of your users escalating their priviliges to 
root.

Note also, in a standard SL installation, including those performed with 
the TRIUMF kickstarts, the kernel is not upgraded by the automatic 
nightly update procedure.  It requires you to manually run "yum update" 
and the reboot.

--
Kel Raywood
TRIUMF Network & Computing Services