ZFS Deadlock: 'Directory of Death'

[Allan Jude]

On 2016-07-10 13:54, Ori Bernstein wrote:
> I've got a fun issue: There's a directory of death on my desktop running ZFS.
> When installing ports, I Any program that tries to touch a directory in
> /usr/ports hangs, deadlocked in the VFS cache lookup:
> 
> 	$ sudo procstat -kk 94688
> 	PID    TID COMM             TDNAME           KSTACK                       
> 	94688 102697 fuser            -                mi_switch+0xe1 sleepq_wait+0x3a sleeplk+0x15d __lockmgr_args+0x91a vop_stdlock+0x3c VOP_LOCK1_APV+0xab _vn_lock+0x43 vget+0x73 cache_lookup+0x5d5 vfs_cache_lookup+0xac VOP_LOOKUP_APV+0xa1 lookup+0x5a1 namei+0x4d4 kern_statat_vnhook+0xae sys_stat+0x2d amd64_syscall+0x40f Xfast_syscall+0xfb 
> 
> 
> The process that seems to have triggered this was a port install of NPM, which seems to have
> deadlocked while renaming some files:
> 
> 	$ ps a 93931
> 	93931 15- D      0:11.87 npm (node)
> 
> 	$ sudo procstat -kk 93931
> 	PID    TID COMM             TDNAME           KSTACK                       
> 	93931 101449 node             -                mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d umtxq_sleep+0x125 do_sem_wait+0x4db __umtx_op_sem_wait+0x73 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 101977 node             -                mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d kern_kevent_fp+0x399 kern_kevent+0x9f sys_kevent+0x12a amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102252 node             V8 WorkerThread  mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d umtxq_sleep+0x125 do_sem_wait+0x4db __umtx_op_sem_wait+0x73 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102258 node             V8 WorkerThread  mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d umtxq_sleep+0x125 do_sem_wait+0x4db __umtx_op_sem_wait+0x73 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102262 node             V8 WorkerThread  mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d umtxq_sleep+0x125 do_sem_wait+0x4db __umtx_op_sem_wait+0x73 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102266 node             V8 WorkerThread  mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d umtxq_sleep+0x125 do_sem_wait+0x4db __umtx_op_sem_wait+0x73 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102279 node             -                mi_switch+0xe1 sleepq_catch_signals+0xab sleepq_wait_sig+0xf _sleep+0x27d umtxq_sleep+0x125 do_wait+0x387 __umtx_op_wait_uint_private+0x83 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102285 node             -                mi_switch+0xe1 sleepq_wait+0x3a sleeplk+0x15d __lockmgr_args+0xca0 vop_stdlock+0x3c VOP_LOCK1_APV+0xab _vn_lock+0x43 vputx+0x21f zfs_rename_unlock+0x3e zfs_freebsd_rename+0xe39 VOP_RENAME_APV+0xab kern_renameat+0x4a6 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102287 node             -                mi_switch+0xe1 sleepq_wait+0x3a sleeplk+0x15d __lockmgr_args+0x91a vop_stdlock+0x3c VOP_LOCK1_APV+0xab _vn_lock+0x43 zfs_lookup+0x45d zfs_freebsd_lookup+0x6d VOP_CACHEDLOOKUP_APV+0xa1 vfs_cache_lookup+0xd6 VOP_LOOKUP_APV+0xa1 lookup+0x5a1 namei+0x4d4 kern_renameat+0x1b7 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 	93931 102294 node             -                mi_switch+0xe1 sleepq_wait+0x3a sleeplk+0x15d __lockmgr_args+0x91a vop_stdlock+0x3c VOP_LOCK1_APV+0xab _vn_lock+0x43 vget+0x73 cache_lookup+0x5d5 vfs_cache_lookup+0xac VOP_LOOKUP_APV+0xa1 lookup+0x5a1 namei+0x4d4 kern_renameat+0x1b7 amd64_syscall+0x40f Xfast_syscall+0xfb 
> 
> The specific directory that is causing this issue is:
> 
> 	/usr/ports/www/npm/work/stage/usr/local/lib/node_modules/npm/node_modules/npmlog/node_modules/gauge
> 
> Right now I'm running on a stock FreeBSD 10.3 release.
> 
> 	FreeBSD oneeye 10.3-RELEASE-p4 FreeBSD 10.3-RELEASE-p4 #0: Sat May 28 12:23:44 UTC 2016     root at amd64-builder.daemonology.net:/usr/obj/usr/src/sys/GENERIC  amd64
> 
> I'm going to guess that this deadlock will go away if I reboot this box, but
> I haven't rebooted yet. Since this isn't really impacting anything important
> that I'm doing, I can keep it up for a bit if someone wants me to gather some
> extra information to track down the root cause.
> 

It looks like your problem is similar to:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=209158

If you read that PR and find your problem is similar, please add your
information to that post.


-- 
Allan Jude