On Nov 9, 2007 7:14 AM, Justin Piszcz [off-list ref] wrote:

On Thu, 8 Nov 2007, Carlos Carvalho wrote:

quoted

Jeff Lessem (Jeff@Lessem.org) wrote on 6 November 2007 22:00:

quoted

Dan Williams wrote:

quoted

The following patch, also attached, cleans up cases where the code looks
at sh->ops.pending when it should be looking at the consistent
stack-based snapshot of the operations flags.

I tried this patch (against a stock 2.6.23), and it did not work for
me.  Not only did I/O to the effected RAID5 & XFS partition stop, but
also I/O to all other disks.  I was not able to capture any debugging
information, but I should be able to do that tomorrow when I can hook
a serial console to the machine.

I'm not sure if my problem is identical to these others, as mine only
seems to manifest with RAID5+XFS.  The RAID rebuilds with no problem,
and I've not had any problems with RAID5+ext3.

Us too! We're stuck trying to build a disk server with several disks
in a raid5 array, and the rsync from the old machine stops writing to
the new filesystem. It only happens under heavy IO. We can make it
lock without rsync, using 8 simultaneous dd's to the array. All IO
stops, including the resync after a newly created raid or after an
unclean reboot.

We could not trigger the problem with ext3 or reiser3; it only happens
with xfs.

In our case all process using md4, including md4_resync, stay in D state.
Call Trace:
  [<ffffffff803615ac>] __generic_unplug_device+0x13/0x24
  [<ffffffff803622cf>] generic_unplug_device+0x18/0x28
  [<ffffffff803f2cf7>] get_active_stripe+0x22b/0x472
...
see dmesg (sysrq t) attached.

We can reproduce this problem in two machines with the same configuration:
  - 2 x Dual-Core Opteron 2.8GHz
  - 8GB memory
  - 3ware 9000 with 10 x 750GB sata disks
  - Debian Etch x86_64
  - raid5 + xfs (/dev/md4)
in all these stock kernel's:
  - 2.6.22.11, 2.6.22.12, 2.6.23.1, 2.6.24-rc2
running:
  - for i in f{0..7}; do (dd bs=1M count=100000 if=/dev/zero of=$i &); done

If we increase /sys/block/md4/md/stripe_cache_size the device and process
back to work.