RedHat/CentOS 4.8 Bug Fix

Problems:

• I/O and/or machine hangs when userspace monitoring software uses a SCSI ioctl driver.
• Sudden high load or IO wait, system unresponsive

Environment:

• Red Hat Enterprise Linux 4.8
• Note that Red Hat Enterprise Linux 5 (and later) is specifically not affected.
• Qlogic HBA

Solutions:

• A first workaround is to avoid using any monitoring/management software which uses SCSI IOCTLs.
• A second workaround is to move the ioctl module (qioctlmod.ko) to a location where it cannot be loaded.
• Fixed in http://rhn.redhat.com/errata/RHBA-2011-0968.html
  * kernel-2.6.9-101.EL
   

More details:

You can follow the steps below to confirm this bug.

1. Look for the presence of qioctlmod in lsmod output:

# lsmod | grep ioctl
qioctlmod 79553 0
qla2xxx 204065 242 qioctlmod,qla2400

2. Look in /var/log/messages for messages containing qim_cmd_wait with "status=1":

# grep qim_cmd_wait /var/log/messages

Mar 11 14:15:12 somehost kernel: qim_cmd_wait(8): entered
Mar 11 14:15:12 somehost kernel: qim_cmd_wait(8): exiting - ind=405, status=1.

3. Data from "sar" confirms high I/O wait and high load around the time of the "qim_cmd_wait" messages.