Re: [Cluster-devel] 2.6.37 - GFS2 trouble

[Steven Whitehouse]

Hi,

On Sun, 2011-01-09 at 22:06 +0100, Nikola Ciprich wrote:
> Hello,
> I wanted to try 2.6.37 on my cluster, but all tasks trying to access GFS2-mounted
> partition got stuck. Kernel started spitting following messages:
> 
> Jan  5 22:16:46 vbox1 [ 3001.948125] INFO: task gfs2_quotad:12532 blocked for more than 120 seconds.
> Jan  5 22:16:46 vbox1 [ 3001.948137] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> Jan  5 22:16:46 vbox1 [ 3001.948141] gfs2_quotad   D ffffffff8140a4c0     0 12532      2 0x00000080
> Jan  5 22:16:46 vbox1 [ 3001.948149]  ffff88040465fc30 0000000000000046 ffff88040465fb20 00000000000116c0
> Jan  5 22:16:46 vbox1 [ 3001.948156]  ffff8804054b48d8 0000000000000007 ffff8804054b4530 ffff88042fd45c40
> Jan  5 22:16:46 vbox1 [ 3001.948163]  ffff88040465ffd8 0000000000000000 000000000465fb50 ffffffff8136cce3
> Jan  5 22:16:46 vbox1 [ 3001.948170] Call Trace:
> Jan  5 22:16:46 vbox1 [ 3001.948184]  [<ffffffff8136cce3>] ? _raw_spin_unlock+0x13/0x40
> Jan  5 22:16:46 vbox1 [ 3001.948199]  [<ffffffffa04c3d38>] ? dlm_put_lockspace+0x28/0x30 [dlm]
> Jan  5 22:16:46 vbox1 [ 3001.948208]  [<ffffffffa04c2066>] ? dlm_lock+0x86/0x180 [dlm]
> Jan  5 22:16:46 vbox1 [ 3001.948228]  [<ffffffffa05696c0>] ? gdlm_bast+0x0/0x50 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948233]  [<ffffffff81045862>] ? update_curr+0xb2/0x170
> Jan  5 22:16:46 vbox1 [ 3001.948239]  [<ffffffff8103ad61>] ? get_parent_ip+0x11/0x50
> Jan  5 22:16:46 vbox1 [ 3001.948243]  [<ffffffff8103c4bd>] ? sub_preempt_count+0x9d/0xd0
> Jan  5 22:16:46 vbox1 [ 3001.948249]  [<ffffffff8136cc9d>] ? _raw_spin_unlock_irqrestore+0x1d/0x50
> Jan  5 22:16:46 vbox1 [ 3001.948260]  [<ffffffffa054a149>] gfs2_glock_holder_wait+0x9/0x10 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948265]  [<ffffffff8136a9e5>] __wait_on_bit+0x55/0x80
> Jan  5 22:16:46 vbox1 [ 3001.948275]  [<ffffffffa054a140>] ? gfs2_glock_holder_wait+0x0/0x10 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948285]  [<ffffffffa054a140>] ? gfs2_glock_holder_wait+0x0/0x10 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948291]  [<ffffffff8136aa88>] out_of_line_wait_on_bit+0x78/0x90
> Jan  5 22:16:46 vbox1 [ 3001.948296]  [<ffffffff8106ae70>] ? wake_bit_function+0x0/0x30
> Jan  5 22:16:46 vbox1 [ 3001.948301]  [<ffffffff8103ad61>] ? get_parent_ip+0x11/0x50
> Jan  5 22:16:46 vbox1 [ 3001.948311]  [<ffffffffa054a192>] gfs2_glock_wait+0x42/0x50 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948323]  [<ffffffffa054bc8d>] gfs2_glock_nq+0x28d/0x3a0 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948336]  [<ffffffffa0565fe9>] gfs2_statfs_sync+0x59/0x1a0 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948350]  [<ffffffffa0565fe1>] ? gfs2_statfs_sync+0x51/0x1a0 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948354]  [<ffffffff8103c4bd>] ? sub_preempt_count+0x9d/0xd0
> Jan  5 22:16:46 vbox1 [ 3001.948368]  [<ffffffffa055ec97>] quotad_check_timeo+0x57/0x90 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948381]  [<ffffffffa05606d7>] gfs2_quotad+0x207/0x240 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948386]  [<ffffffff8106ae30>] ? autoremove_wake_function+0x0/0x40
> Jan  5 22:16:46 vbox1 [ 3001.948391]  [<ffffffff8136cc9d>] ? _raw_spin_unlock_irqrestore+0x1d/0x50
> Jan  5 22:16:46 vbox1 [ 3001.948404]  [<ffffffffa05604d0>] ? gfs2_quotad+0x0/0x240 [gfs2]
> Jan  5 22:16:46 vbox1 [ 3001.948409]  [<ffffffff8106a906>] kthread+0x96/0xa0
> Jan  5 22:16:46 vbox1 [ 3001.948416]  [<ffffffff810032d4>] kernel_thread_helper+0x4/0x10
> Jan  5 22:16:46 vbox1 [ 3001.948420]  [<ffffffff8106a870>] ? kthread+0x0/0xa0
> Jan  5 22:16:46 vbox1 [ 3001.948425]  [<ffffffff810032d0>] ? kernel_thread_helper+0x0/0x10
> 
> Could somebody with insight have a look at this? Is it some known problem? 
> If I could somehow help to debug, I'll be happy to do so.
> cheers
> nik
> 
> 

Quotad is often the first victim of any slow down or problem on GFS2
since it runs periodically (even if you are not using quotas, it also
looks after statfs too).

So I can't tell directly from that information what is wrong. I'd
suggest taking glock dumps (via debugfs) as a first step. Grabbing
backtraces of the glock_workqueue and other processes using gfs2 is
probably the next step, and if that fails to point the way, the gfs2
tracepoints are the next thing to look at,

Steve.


--
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/