On Tue, 08 Dec 2009 14:18:24 +0100
ml-raid@syscall.eu wrote:

Hello,
I have a 2.6.27.29-4-grsec (from http://kernelsec.cr0.org) Linux box, on
x86, running debian lenny. On sunday, during the monthly rebuild of my
standard RAID1 setup (/proc/mdstat):

Personalities : [raid1]
md127 : active raid1 sda1[0] sdb1[1]
      488383936 blocks [2/2] [UU]

unused devices: <none>

The kernel started outputting thousands of backtraces (see attached log)
The IO throughput was drastically reduced but the resync finally
completed after about 38 hours.

This is the first time I have been bitten by this bug. The server has
been running on this kernel for 74 days and the previous resync didn't
exhibit any problem.

This is not a RAID problem - it was a problem with the grsec patches.
They assume that any atomic variable is used as a counter and should not
overflow.  That is not the case with bd_disk->sync_io.  It is expected to
overflow.  So the "PAX: ..." report is a false positive.

NeilBrown