[Bugme-janitors] [Bug 24012] Commit g18733b0 hangs my LVM+raid10 system

bugzilla-daemon at bugzilla.kernel.org bugzilla-daemon at bugzilla.kernel.org
Mon Nov 29 15:40:05 PST 2010


https://bugzilla.kernel.org/show_bug.cgi?id=24012





--- Comment #5 from Neil Brown <neilb at suse.de>  2010-11-29 23:40:03 ---
Very strange...
This is what I see in the 'Show State' output:

Two processes - jbd2/dm-0-8 and fsck.ext4 - are stuck in md_flush_request.
This means they are trying to perform REQ_FLUSH requests, but md only allows
one at a time, and there is one currently in progress.

One process - flush-253:0 - is stuck in md_write_start. waiting for a
super-block update to complete.   This process is handling a regular write
request, not a REQ_FLUSH request (if it were, the stack would be different). 
So it isn't blocking the REQ_FLUSH requests.

One process - md0_raid10 - is writing out the superblock, on behalf of the
md_write_start request.  It has submitted the IO (which would be REQ_FLUSH, and
would have been even before the offending commit) and is waiting for it to
complete.  But for some reason it isn't completing.

The only way this would be different before the identified commit is that the
first two processes would not be stuck in md_flush_request.  They would have
progressed further and would be stuck in md_write_start instead.  And there is
nothing they could do between where md_flush_request is called and where
md_write_start is called that could possibly unblock the other threads.

The number of ata errors about  'flush' are certainly worrying.  I would
certainly be more comfortable if I could see a stack trace without any ATA
error - possible it would be different and the differences might be
illuminating.

So: could you try to produce another sysrq-T output, and if you can get one
with no errors reported, so much the better.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.


More information about the Bugme-janitors mailing list