Re: raid6 grow "hangs" on 2.6.27 with mounted FS

[roma1390]

Neil Brown wrote:
> On Saturday March 21, roma1390@xxxxxxxxx wrote:
> I can almost reproduce this.
> However when I run it, the "mdadm --grow" prints     mdadm: Need to 
backup 384K of critical section..
> (as expected) and then hangs.
>
> If I interrupt it and proceed, then the umount hangs.
>
> Is this the case for you?

Yes, mdadm hangs by itself, If i start mdadm in background and wait some 
time for reconstrucion, umount still hangs. I didn't touch mdadm, and 
mdadm likes to stop...

> The umount hangs because there is something important that mdadm needs
> to do which it didn't do because it was interrupted.  During the
> 'critical section', mdadm causes all writes to the start of the device
> to be blocked.  The umount tries to write the filesystem superblock
> and hangs.
> You can test if this is the problem by running the command
>
>  cat /sys/block/md2/md/suspend_hi > /sys/block/md2/md/suspend_lo

First try was:

  mdadm --grow /dev/md2 --raid-devices=5 &
  sleep 3
  cat /sys/block/md2/md/suspend_hi
  # output: 768
  cat /sys/block/md2/md/suspend_lo
  # output: 0
  cat /sys/block/md2/md/suspend_hi > /sys/block/md2/md/suspend_lo
  sleep 20
  umount /dev/md2

And this works! mdadm unhangs, and umount doesn't block any more.

> That should allow the 'umount' to complete.
>
> This will only happen on very small arrays that take less than a
> couple of seconds for the reshape to complete.  I have a patch for
> mdadm which makes it more robust in this situation.  It will be in
> future releases.
>
> Does this explain what is happening to you?

Yes, thanks. May be if I retest this situation with same kernel and 
limited recovery bandwith, then may be i can't hit same problem again?

> Thanks,
> NeilBrown

Thanks.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/