Re: More 'D' state processes [was: Re: Weird problem: mdadm blocks]

On Mon, 06 Jan 2014 17:00:35 +0100 hans@hanswkraus.com wrote:

 

Hi Neil, 

the output of 'uname -r' is: 3.2.0-4-amd64. A standard Debian Wheezy.

I rebooted the system in the meantime. The re-sync started anew but
finished otherwise without flaws. 

I guess there must be some other bug then. I still expect that it is related
to md_flush_request().  Probably related to the fact that you have an md
device on top of other md devices....

I had a quick look and cannot find the cause of the problem.  Maybe it will
have to wait until someone else hit it :-(

Thanks for the report,
NeilBrown

Regards, Hans 

Am 05.01.2014 23:00, schrieb NeilBrown: 

quoted

On Sun, 22 Dec 2013 17:43:43 +0100 Hans Kraus [off-list ref] wrote:

quoted

Hi Neil,

Hi Hans,
sorry for the delay - Christmas/New Year vacation...
[40800.777037] xfsaild/dm-0 D ffff88003754c300 0 20798 2 0x00000000 [40800.777042] ffff88003754c300 0000000000000046 0000000000000000 ffff88005d7c51a0 [40800.777047] 0000000000013780 ffff88001b01ffd8 ffff88001b01ffd8 ffff88003754c300 [40800.777052] ffff880071fede00 ffffffff81070fc1 0000000000000046 ffff88003734a400 [40800.777057] Call Trace: [40800.777061] [<ffffffff81070fc1>] ? arch_local_irq_save+0x11/0x17 [40800.777071] [<ffffffffa0170466>] ? md_flush_request+0x96/0x111 [md_mod] [40800.777076] [<ffffffff8103f6c4>] ? try_to_wake_up+0x197/0x197 [40800.777082] [<ffffffffa0f11711>] ? make_request+0x25/0x37a [raid456] [40800.777091] [<ffffffffa0185873>] ?

This looks like the most likely root of the problem - something wrong in
md_flush_request.

There was a bug here fixed in 2.6.37

commit a035fc3e2531703b539f23bec4ca7943cfc69349
Author: NeilBrown [off-list ref]
Date: Thu Dec 9 16:17:51 2010 +1100

 md: fix possible deadlock in handling flush requests.

You didn't say which kernel you were running. Could it be earlier than
2.6.37???

NeilBrown