[Bug 236220] ZFS vnode deadlock

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236220

--- Comment #1 from ncrogers at gmail.com ---
avg at freebsd.org pointed out in the freebsd-stable list that one thread is stuck
on zfs_znode_alloc, and that is likely the root of the problem. Indeed all the
systems I've witnessed this happen to have had one thread stuck on vn_lock ->
zfs_znode_alloc. It often seems to be a different application, so I'm still not
sure what particular workload is causing this.

For example, the below lines from procstat -kka are from three different
systems stuck in a hung / zfs deadlock state.

2329 100515 sh                  -                   mi_switch+0xe1
sleepq_wait+0x2c sleeplk+0x1c5 lockmgr_xlock_hard+0x19c VOP_LOCK1_APV+0x7e
_vn_lock+0x40 zfs_znode_alloc+0x434 zfs_mknode+0xa9d zfs_freebsd_create+0x512
VOP_CREATE_APV+0x78 vn_open_cred+0x2c9 kern_openat+0x20c amd64_syscall+0x369
fast_syscall_common+0x101 

36704 101146 perl                -                   mi_switch+0xe1
sleepq_wait+0x2c sleeplk+0x1c5 lockmgr_xlock_hard+0x19c VOP_LOCK1_APV+0x7e
_vn_lock+0x40 zfs_znode_alloc+0x434 zfs_mknode+0xa9d zfs_freebsd_create+0x512
VOP_CREATE_APV+0x78 vn_open_cred+0x2c9 kern_openat+0x20c amd64_syscall+0x369
fast_syscall_common+0x101 

1663 100503 postgres            -                   mi_switch+0xe1
sleepq_wait+0x2c sleeplk+0x1c5 lockmgr_xlock_hard+0x19c VOP_LOCK1_APV+0x7e
_vn_lock+0x40 zfs_znode_alloc+0x434 zfs_mknode+0xa9d zfs_freebsd_create+0x512
VOP_CREATE_APV+0x78 vn_open_cred+0x2c9 kern_openat+0x20c amd64_syscall+0x369
fast_syscall_common+0x101

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