[Bug 219224] sync hanging in _cv_wait zil_commit zfs_sync sys_sync

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=219224

--- Comment #1 from Tino Reinhardt <t at sdf.org> ---
I seem to be able to reproduce this on a similar system,
having cfengine running and run a loop in a shell like
"while :; do sync; done":

tools-2-b# ps ax | grep sync
   27  -  DL       1:27.96 [syncer]
 3667  1  R+    1017:57.50 sync

tools-2-b# procstat -kk 3667
  PID    TID COMM             TDNAME           KSTACK
 3667 102488 sync             -                dbuf_read+0x155
dnode_hold_impl+0x188 dmu_bonus_hold+0x1d zfs_zget+0xc5 zfs_get_data+0x40
zil_commit+0x6a1 zfs_sync+0xad sys_sync+0x146 amd64_syscall+0x4ce
Xfast_syscall+0xfb 
tools-2-b# procstat -kk 3667
  PID    TID COMM             TDNAME           KSTACK
 3667 102488 sync             -                dnode_hold_impl+0x409
dmu_bonus_hold+0x1d zfs_zget+0xc5 zfs_get_data+0x40 zil_commit+0x6a1
zfs_sync+0xad sys_sync+0x146 amd64_syscall+0x4ce Xfast_syscall+0xfb 
tools-2-b# procstat -kk 3667
  PID    TID COMM             TDNAME           KSTACK
 3667 102488 sync             -                dmu_zfetch+0x7b dbuf_read+0x155
dnode_hold_impl+0x188 dmu_bonus_hold+0x1d zfs_zget+0xc5 zfs_get_data+0x40
zil_commit+0x6a1 zfs_sync+0xad sys_sync+0x146 amd64_syscall+0x4ce
Xfast_syscall+0xfb

tools-2-b# ps ax | awk '$3~"D"' | tail
   22  -  DL       0:00.00 [vmdaemon]
   23  -  DL       0:00.01 [pagezero]
   24  -  DL       0:02.13 [bufspacedaemon]
   25  -  DL       0:02.06 [bufdaemon]
   26  -  DL       0:02.31 [vnlru]
   27  -  DL       1:26.84 [syncer]
  643  -  DL       1:03.87 [pf purge]
 3650  -  DsJ      0:00.05 /var/cfengine/bin/cf-agent -f failsafe.cf
11495  -  DJ       0:00.01 /var/cfengine/bin/cf-agent -f failsafe.cf
96409  -  DL       0:15.85 [ftcleanup]

tools-2-b# procstat -k 3650
  PID    TID COMM             TDNAME           KSTACK                       
 3650 102929 cf-agent         -                mi_switch sleepq_wait _cv_wait
zil_commit zfs_freebsd_putpages VOP_PUTPAGES_APV vnode_pager_putpages
vm_pageout_flush vm_object_page_collect_flush vm_object_page_clean
vm_object_terminate zfs_freebsd_reclaim VOP_RECLAIM_APV vgonel vrecycle
zfs_freebsd_inactive VOP_INACTIVE_APV vinactive

The shell loop is running in a screen and this indicates
it took around 5 minutes for a sync(8) to stay hanging around:

tools-2-b# ps ax | grep sync
   27  -  DL       1:27.96 [syncer]
 3667  1  R+    1017:57.50 sync

tools-2-b# ps olstart 3667
STARTED
Thu May 11 19:50:28 2017

This is the hanging cf-agent process:

tools-2-b# ps olstart 3650
STARTED
Thu May 11 19:50:28 2017

>From this on, other calls to sync(2) are hanging:

tools-2-b# printf '#include<unistd.h>\n int main(){sync();}' | cc -x c -o sync
-
tools-2-b# ./sync &
[1] 67191
tools-2-b# procstat -kk 67191
  PID    TID COMM             TDNAME           KSTACK
67191 103195 sync             -                mi_switch+0xd2 sleepq_wait+0x3a
_cv_wait+0x194 zil_commit+0x85 zfs_sync+0xad sys_sync+0x146 amd64_syscall+0x4ce
Xfast_syscall+0xfb

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