[Bug 229958] ctld / zfs deadlock with 'zfs rename ...'

Fri Jan 11 19:43:19 UTC 2019

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=229958

--- Comment #10 from Dave Baukus <daveb at spectralogic.com> ---
Here's another twist on the same theme.
The theme being that serializing zvol.c by overloading the 
spa_namespace_lock is specious at best.

This time, deadlkres() panicked the system because we have
zpool import thread holding the spa_namespace_lock while sleeping
in g_access() waiting for (gp->flags & G_GEOM_IN_ACCESS) != 0.

Meanwhile, another g_access() thread  set 
gp->flags |= G_GEOM_IN_ACCESS and executed gp->access(). 
In this case gp->access is zvol_geom_access().
zvol_geom_access() executes zvol_close() where it tries to acquire
the spa_namespace_lock - Deadlock.

Details below:

Thread 1666 (Thread 101829):
(struct thread *)0xfffff801630eb580, tid 101829
   zpool :: (struct proc *)0xfffff802e7bfca50, pid 39115
   args: zpool import

[Switching to thread 1666 (Thread 101829)]
STACK:
#0  sched_switch (td=0xfffff801630eb580, newtd=0xfffff80003ae1580,
flags=<optimized out>) at sys/kern/sched_ule.c:2112
#1  mi_switch (flags=260, newtd=0x0) at sys/kern/kern_synch.c:439
#2  sleepq_switch (wchan=0xfffff8040afc2700, pri=0) at
sys/kern/subr_sleepqueue.c:613
#3  sleepq_wait (wchan=0xfffff8040afc2700, pri=0) at
sys/kern/subr_sleepqueue.c:692
#4  _sleep (ident=0xfffff8040afc2700, lock=<optimized out>, priority=0,
wmesg=0xffffffff809042a5 "gtopol", sbt=<optimized out>, pr=0, flags=256) at
sys/kern/kern_synch.c:217
#5  g_access (cp=0xfffff805da3f3a80, dcr=<optimized out>, dcw=0, dce=1) at
sys/geom/geom_subr.c:932
#6  vdev_geom_attach (pp=0xfffff801ffe33800, vd=0x0, sanity=<optimized out>) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c:278
#7  vdev_attach_ok (vd=0xfffff8054ca2e000, pp=0xfffff801ffe33800) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c:650
#8  vdev_geom_attach_by_guids (vd=<optimized out>) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c:722
#9  vdev_geom_open_by_guids (vd=<optimized out>) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c:755
#10 vdev_geom_open (vd=0xfffff8054ca2e000, psize=0xfffffe01330aa168,
max_psize=0xfffffe01330aa170, logical_ashift=0xfffffe01330aa158,
physical_ashift=0xfffffe01330aa160) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c:860
#11 vdev_open (vd=0xfffff8054ca2e000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c:1550
#12 vdev_open_children (vd=0xfffff8054cbad000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c:1478
#13 vdev_mirror_open (vd=0xfffff8054cbad000, asize=0xfffffe01330aa228,
max_asize=0xfffffe01330aa230, logical_ashift=0xfffffe01330aa218,
physical_ashift=0xfffffe01330aa220) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_mirror.c:305
#14 vdev_open (vd=0xfffff8054cbad000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c:1550
#15 vdev_open_children (vd=0xfffff8054c996000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c:1478
#16 vdev_root_open (vd=0xfffff8054c996000, asize=0xfffffe01330aa308,
max_asize=0xfffffe01330aa310, logical_ashift=0xfffffe01330aa2f8,
physical_ashift=0xfffffe01330aa300) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_root.c:96
#17 vdev_open (vd=0xfffff8054c996000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c:1550
#18 spa_ld_open_vdevs (spa=0xfffffe013f864000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c:2564
#19 spa_ld_mos_init (spa=0xfffffe013f864000, type=SPA_IMPORT_EXISTING) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c:3568
#20 spa_ld_mos_with_trusted_config (spa=0xfffffe013f864000,
type=SPA_IMPORT_EXISTING, update_config_cache=0xfffffe01330aa514) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c:3711
#21 spa_load_impl (spa=0xfffffe013f864000, type=SPA_IMPORT_EXISTING,
ereport=<optimized out>) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c:3765
#22 spa_load (spa=0xfffffe013f864000, state=<optimized out>,
type=SPA_IMPORT_EXISTING) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c:2348
#23 spa_tryimport (tryconfig=<optimized out>) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c:5638
#24 zfs_ioc_pool_tryimport (zc=0xfffffe013c3d9000) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c:1756

--------------------------------------------

Thread 1328 (Thread 100162):
(struct thread *)0xfffff80003d23000, tid 100162
   geom :: (struct proc *)0xfffff80003da8a50, pid 13

#0  sched_switch (td=0xfffff80003d23000, newtd=0xfffff80003ad0000,
flags=<optimized out>) at sys/kern/sched_ule.c:2112
2112                    cpuid = PCPU_GET(cpuid);

STACK:
#0  sched_switch (td=0xfffff80003d23000, newtd=0xfffff80003ad0000,
flags=<optimized out>) at sys/kern/sched_ule.c:2112
#1  mi_switch (flags=260, newtd=0x0) at sys/kern/kern_synch.c:439
#2  sleepq_switch (wchan=0xffffffff81cb39a8 <spa_namespace_lock>, pri=0) at
sys/kern/subr_sleepqueue.c:613
#3  sleepq_wait (wchan=0xffffffff81cb39a8 <spa_namespace_lock>, pri=0) at
sys/kern/subr_sleepqueue.c:692
#4  _sx_xlock_hard (sx=0xffffffff81cb39a8 <spa_namespace_lock>, x=<optimized
out>, opts=<optimized out>, file=<optimized out>, line=<optimized out>) at
sys/kern/kern_sx.c:857
#5  _sx_xlock (sx=0xffffffff81cb39a8 <spa_namespace_lock>, opts=0,
file=0xffffffff81c09a5f
"sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c", line=1276) at
sys/kern/kern_sx.c:328
#6  zvol_close (flag=0, pp=<optimized out>, count=<optimized out>) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c:1276
#7  zvol_geom_access (pp=0xfffff801ffe33800, acr=<optimized out>,
acw=<optimized out>, ace=0) at
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c:2739
#8  g_access (cp=0xfffff8052a7a3c00, dcr=<optimized out>, dcw=0, dce=0) at
sys/geom/geom_subr.c:977
#9  g_label_taste (mp=0xffffffff80a7dfe0 <g_label_class>,
pp=0xfffff801ffe33800, flags=<optimized out>) at sys/geom/label/g_label.c:392
#10 g_new_provider_event (arg=0xfffff801ffe33800, flag=<optimized out>) at
sys/geom/geom_subr.c:557
#11 one_event () at sys/geom/geom_event.c:268
#12 g_run_events () at sys/geom/geom_event.c:291
#13 fork_exit (callout=0xffffffff80481010 <g_event_procbody>, arg=0x0,
frame=0xfffffe010e5a0ac0) at sys/kern/kern_fork.c:1057
#14 <signal handler called>

PENDING ON LOCK:
spa_namespace_lock, (struct sx *)0xffffffff81cb39a8
$7 = {
  lock_object = {
    lo_name = 0xffffffff81c26ae4,
    lo_flags = 0x22710000,
    lo_data = 0x0,
    lo_witness = 0x0
  },
  sx_lock = 0xfffff801630eb584
}
LOCK HOLDER:
(struct thread *)0xfffff801630eb580, tid 101829
   zpool :: (struct proc *)0xfffff802e7bfca50, pid 39115
   args: zpool import

-- 
You are receiving this mail because:
You are the assignee for the bug.