bdi-default hung waiting for kthread_stop to finish
I was testing some cifs patches by running the fsstress program from LTP on a cifs mount on a very recent kernel. After running ~70 minutes or so, I got this in the ring buffer:
INFO: task bdi-default:26 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
bdi-default D 0000000000000001 3256 26 2 0x00000000
ffff88003e789c60 0000000000000046 ffffffff00000000 00000000001d5180 00000000001d5180 ffff88003e782450 ffff88003e789fd8 00000000001d5180 00000000001d5180 00000000001d5180 00000000001d5180 ffff88003e789fd8
Call Trace: [<ffffffff814994e1>] schedule_timeout+0x39/0xfe [<ffffffff8107d99a>] ? lock_acquired+0x1fd/0x20c [<ffffffff8107fc2a>] ? lock_release+0x19a/0x1a6 [<ffffffff81080158>] ? trace_hardirqs_on+0xd/0xf [<ffffffff814992bb>] wait_for_common+0xb0/0x10a [<ffffffff81049fc1>] ? default_wake_function+0x0/0x14 [<ffffffff814993cd>] wait_for_completion+0x1d/0x1f [<ffffffff8106c497>] kthread_stop+0x73/0xc2 [<ffffffff810f95f3>] bdi_forker_thread+0x38c/0x42a [<ffffffff810f9267>] ? bdi_forker_thread+0x0/0x42a [<ffffffff8106c41c>] kthread+0x9d/0xa5 [<ffffffff8100ab24>] kernel_thread_helper+0x4/0x10 [<ffffffff8149bc10>] ? restore_args+0x0/0x30 [<ffffffff8106c37f>] ? kthread+0x0/0xa5 [<ffffffff8100ab20>] ? kernel_thread_helper+0x0/0x10
I'm assuming that it's waiting on flush-* to exit. The problem is that all of the ones that I see on the box just seem to be sitting there idle. Here's the one for cifs, but the others seem to be sitting in exactly the same spot:
flush-cifs-2 S 0000000000000003 6512 8398 2 0x00000000
ffff880007f41e20 0000000000000046 ffff880007f41d90 00000000001d5180 00000000001d5180 ffff88003bcec8a0 ffff880007f41fd8 00000000001d5180 00000000001d5180 00000000001d5180 00000000001d5180 ffff880007f41fd8
Call Trace: [<ffffffff8114987f>] bdi_writeback_thread+0x14f/0x211 [<ffffffff81149730>] ? bdi_writeback_thread+0x0/0x211 [<ffffffff8106c41c>] kthread+0x9d/0xa5 [<ffffffff8100ab24>] kernel_thread_helper+0x4/0x10 [<ffffffff8149bc10>] ? restore_args+0x0/0x30 [<ffffffff8106c37f>] ? kthread+0x0/0xa5 [<ffffffff8100ab20>] ? kernel_thread_helper+0x0/0x10
Stuff like statfs() and readdir() to the cifs mount seems to be fine, but writeback seems to be frozen.
2.6.36-0.14.rc3.git0.fc15.x86_64
The machine is a KVM guest. The fsstress command is:
# fsstress -d /mnt/cifs/fsstress -l0 -n 1000 -p 8
Let me know if other info would be helpful in tracking this down.
Signature · 1 line
Jeff Layton <[email protected]>