asynchronous zvol minor operations cause deadlock in buildbot

[ryao]

The deadlock was introduced by a0bd735, which itself was a deadlock fix. It caused a buildbot failure on #6058:

http://build.zfsonlinux.org/builders/CentOS%206.7%20x86_64%20%28TEST%29/builds/5360/steps/shell_10/logs/console

We had a zvol attempt to create minors during pool export.

1. spa_export_common() took the namespace lock in the zpool thread.
2. The zpool thread blocked on the taskq being destroyed.
3. A taskq thread that was executing minor creation blocked on the namespace lock.
4. Deadlock

z_zvol        D 0000000000000001     0 12229      2 0x00000080
 ffff8801b6f6bc30 0000000000000046 ffff8801b6f6bbe0 ffffffffa06310f7
 ffff8801b6f6bbf0 0000000000000282 ffff8801ba692738 ffff8801ba692730
 ffff8801b6f6bbc0 ffffffff812998e9 ffff880131e0dad8 ffff8801b6f6bfd8
Call Trace:
 [<ffffffffa06310f7>] ? zap_lookup_norm+0x97/0xb0 [zfs]
 [<ffffffff812998e9>] ? cpumask_next_and+0x29/0x50
 [<ffffffff8154c086>] __mutex_lock_slowpath+0x96/0x210
 [<ffffffff8154bbab>] mutex_lock+0x2b/0x50
 [<ffffffffa05fc002>] spa_open_common+0x72/0x490 [zfs]
 [<ffffffff8100969d>] ? __switch_to+0x7d/0x340
 [<ffffffffa0688a60>] ? zvol_create_minors_cb+0x0/0x170 [zfs]
 [<ffffffffa05fc433>] spa_open+0x13/0x20 [zfs]
 [<ffffffffa0599eb9>] dmu_objset_find+0x39/0x80 [zfs]
 [<ffffffffa0347579>] ? spl_kmem_alloc+0xc9/0x1e0 [spl]
 [<ffffffffa068ae64>] zvol_task_cb+0x4a4/0x530 [zfs]
 [<ffffffffa034c844>] taskq_thread+0x294/0x590 [spl]
 [<ffffffff8106c480>] ? default_wake_function+0x0/0x20
 [<ffffffffa034c5b0>] ? taskq_thread+0x0/0x590 [spl]
 [<ffffffff810a63ae>] kthread+0x9e/0xc0
 [<ffffffff8100c28a>] child_rip+0xa/0x20
 [<ffffffff810a6310>] ? kthread+0x0/0xc0
 [<ffffffff8100c280>] ? child_rip+0x0/0x20
INFO: task zpool:20233 blocked for more than 120 seconds.
      Tainted: P           -- ------------    2.6.32-696.1.1.el6.x86_64 #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
zpool         D 0000000000000000     0 20233  20229 0x00000080
 ffff8801ea35fc48 0000000000000086 0000000000000000 ffff8801efc701c0
 000000000000003c ffff8801efec29f0 000006df85aa4a1d ffffffff81181554
 ffff8801efc212c0 00000001006ebb68 ffff880100dba5f8 ffff8801ea35ffd8
Call Trace:
 [<ffffffff81181554>] ? free_block+0x154/0x170
 [<ffffffff8106c480>] ? default_wake_function+0x0/0x20
 [<ffffffffa034b84d>] taskq_wait+0x5d/0x90 [spl]
 [<ffffffff810a6840>] ? autoremove_wake_function+0x0/0x40
 [<ffffffffa034b9a7>] taskq_destroy+0x57/0x3f0 [spl]
 [<ffffffffa05f7ad7>] spa_deactivate+0x77/0x6c0 [zfs]
 [<ffffffffa05f8f97>] spa_export_common+0x257/0x4d0 [zfs]
 [<ffffffffa03472bb>] ? spl_kmem_free+0x2b/0x40 [spl]
 [<ffffffffa05f924b>] spa_export+0x1b/0x20 [zfs]
 [<ffffffffa06458bf>] zfs_ioc_pool_export+0x3f/0x50 [zfs]
 [<ffffffffa0647f94>] zfsdev_ioctl+0x4d4/0x540 [zfs]
 [<ffffffff810521a4>] ? __do_page_fault+0x1f4/0x500
 [<ffffffff811af632>] vfs_ioctl+0x22/0xa0
 [<ffffffff811af7d4>] do_vfs_ioctl+0x84/0x580
 [<ffffffff811afd51>] sys_ioctl+0x81/0xa0
 [<ffffffff8100b0d2>] system_call_fastpath+0x16/0x1b

[loli10K]

Isn't this just a duplicate of #5195?

[ryao]

@loli10K It is. Thanks.