draidcfg -p 2 -d 3 -s 1 -n 6 locks on small disks (less than 1200mb)

[gmelikov]

System information

Type	Version/Name
Distribution Name	Centos
Distribution Version	7
Linux Kernel	3.10.0-514.26.1.el7.x86_64
Architecture	x64
ZFS Version	6e39308
SPL Version	0.7.0-rc4_5_g7a35f2b

Describe the problem you're observing

All zfs related commands freezes (and we cannot reboot system by reboot command) then create 6 disks draid2 pool on disks less than 1200mb.

If we try it on 1300mb disks then it works as supposed.

Describe how to reproduce the problem

rm -rf /var/tmp/test
mkdir /var/tmp/test
cd /var/tmp/test
for fn in {1..6}; do dd if=/dev/zero of=/var/tmp/test/$fn.img bs=1M count=1200; done
modprobe zfs
draidcfg -p 2 -d 3 -s 1 -n 6 /var/tmp/test/draid.nvl
zpool create -f testpool draid2 cfg=/var/tmp/test/draid.nvl /var/tmp/test/{1..6}.img
#command freezes, see logs below

Include any warning/errors/backtraces from the system logs

[  200.799473] spl: loading out-of-tree module taints kernel.
[  200.799564] spl: module verification failed: signature and/or required key missing - tainting kernel
[  200.802464] SPL: Loaded module v0.7.0-rc3_8_g481762f (DEBUG mode)
[  200.803106] znvpair: module license 'CDDL' taints kernel.
[  200.803108] Disabling lock debugging due to kernel taint
[  202.396449] ZFS: Loaded module v0.7.0-rc3, ZFS pool version 5000, ZFS filesystem version 5
[  203.562942]  sdb: sdb1 sdb9
[  203.565623]  sdb: sdb1 sdb9
[  203.567864]  sdb: sdb1 sdb9
[  203.762573]  sdc: sdc1 sdc9
[  203.770463]  sdc: sdc1 sdc9
[  203.772247]  sdc: sdc1 sdc9
[  203.975979]  sdd: sdd1 sdd9
[  203.983113]  sdd: sdd1 sdd9
[  203.985945]  sdd: sdd1 sdd9
[  204.195823]  sde: sde1 sde9
[  204.203310]  sde: sde1 sde9
[  204.207913]  sde: sde1 sde9
[  204.399645]  sdf: sdf1 sdf9
[  204.408676]  sdf: sdf1 sdf9
[  204.571065]  sdg: sdg1 sdg9
[  204.574878]  sdg: sdg1 sdg9
[  204.579216]  sdg: sdg1 sdg9
[  204.803557] SPL: using hostid 0x00000000
[  204.858356] WARNING: Pool 'mgsmdt-001' has encountered an uncorrectable I/O failure and has been suspended.

[  360.437443] INFO: task zpool:3293 blocked for more than 120 seconds.
[  360.437485] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  360.437510] zpool           D ffff880076f00a58     0  3293   3250 0x00000080
[  360.437514]  ffff8800780c7c78 0000000000000086 ffff880078053ec0 ffff8800780c7fd8
[  360.437516]  ffff8800780c7fd8 ffff8800780c7fd8 ffff880078053ec0 ffff880076f00ae0
[  360.437518]  ffff880076f00a30 ffff880076f00ae8 0000000000000000 ffff880076f00a58
[  360.437520] Call Trace:
[  360.437533]  [<ffffffff8168c7f9>] schedule+0x29/0x70
[  360.437545]  [<ffffffffa065160d>] cv_wait_common+0x14d/0x2a0 [spl]
[  360.437550]  [<ffffffff810b1b20>] ? wake_up_atomic_t+0x30/0x30
[  360.437554]  [<ffffffffa0651775>] __cv_wait+0x15/0x20 [spl]
[  360.437594]  [<ffffffffa07a3f7f>] txg_wait_synced+0xdf/0x120 [zfs]
[  360.437617]  [<ffffffffa079706d>] spa_create+0x6ad/0x9e0 [zfs]
[  360.437641]  [<ffffffffa07d9442>] zfs_ioc_pool_create+0x152/0x270 [zfs]
[  360.437644]  [<ffffffff811dd875>] ? __kmalloc+0x55/0x240
[  360.437668]  [<ffffffffa07d6ac6>] zfsdev_ioctl+0x506/0x550 [zfs]
[  360.437671]  [<ffffffff812127a5>] do_vfs_ioctl+0x2d5/0x4b0
[  360.437674]  [<ffffffff81692ce1>] ? __do_page_fault+0x171/0x450
[  360.437675]  [<ffffffff81212a21>] SyS_ioctl+0xa1/0xc0
[  360.437677]  [<ffffffff81697809>] system_call_fastpath+0x16/0x1b
[  360.437685] INFO: task txg_sync:3484 blocked for more than 120 seconds.
[  360.437707] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  360.437731] txg_sync        D ffff880076b983a8     0  3484      2 0x00000080
[  360.437733]  ffff88007736bae0 0000000000000046 ffff880076b60000 ffff88007736bfd8
[  360.437735]  ffff88007736bfd8 ffff88007736bfd8 ffff880076b60000 ffff88007aa56c40
[  360.437736]  0000000000000000 7fffffffffffffff 0000000000000001 ffff880076b983a8
[  360.437737] Call Trace:
[  360.437740]  [<ffffffff8168c7f9>] schedule+0x29/0x70
[  360.437742]  [<ffffffff8168a239>] schedule_timeout+0x239/0x2c0
[  360.437766]  [<ffffffffa07f626f>] ? zio_taskq_dispatch+0x8f/0xa0 [zfs]
[  360.437789]  [<ffffffffa07f62b2>] ? zio_issue_async+0x12/0x20 [zfs]
[  360.437812]  [<ffffffffa07fa97c>] ? zio_nowait+0xbc/0x150 [zfs]
[  360.437814]  [<ffffffff8168bd9e>] io_schedule_timeout+0xae/0x130
[  360.437816]  [<ffffffff810b1816>] ? prepare_to_wait_exclusive+0x56/0x90
[  360.437818]  [<ffffffff8168be38>] io_schedule+0x18/0x20
[  360.437822]  [<ffffffffa06515ab>] cv_wait_common+0xeb/0x2a0 [spl]
[  360.437824]  [<ffffffff810b1b20>] ? wake_up_atomic_t+0x30/0x30
[  360.437828]  [<ffffffffa06517b8>] __cv_wait_io+0x18/0x20 [spl]
[  360.437851]  [<ffffffffa07fa29b>] zio_wait+0xfb/0x190 [zfs]
[  360.437871]  [<ffffffffa07757ef>] dsl_pool_sync+0xbf/0x440 [zfs]
[  360.437893]  [<ffffffffa0791af7>] spa_sync+0x417/0xda0 [zfs]
[  360.437896]  [<ffffffff810c54f2>] ? default_wake_function+0x12/0x20
[  360.437933]  [<ffffffffa07a4ea8>] txg_sync_thread+0x2d8/0x490 [zfs]
[  360.438037]  [<ffffffffa07a4bd0>] ? txg_init+0x280/0x280 [zfs]
[  360.438042]  [<ffffffffa064a73a>] thread_generic_wrapper+0x7a/0xc0 [spl]
[  360.438045]  [<ffffffffa064a6c0>] ? __thread_exit+0x20/0x20 [spl]
[  360.438048]  [<ffffffff810b0a4f>] kthread+0xcf/0xe0
[  360.438050]  [<ffffffff810b0980>] ? kthread_create_on_node+0x140/0x140
[  360.438051]  [<ffffffff81697758>] ret_from_fork+0x58/0x90
[  360.438053]  [<ffffffff810b0980>] ? kthread_create_on_node+0x140/0x140

stracedraid6.log.zip

[thegreatgazoo]

This is likely an issue we already fixed in the internal dev repo. The current github version hard codes every 1 out of 20 metaslabs to be a mirrored metaslab (for small blocks). When the drive is too small, there might be a deadlock at pool creation because there's less than 20 metaslabs.

With openzfs#5182 managing the mirrored metaslabs, it's more dynamic and smarter as well.