file system deadlock in RELENG_11

[Shane Ambler]

On 25/08/2017 05:40, Mike Tancsa wrote:
> On 8/24/2017 4:01 PM, Mike Tancsa wrote:
>> OK, this is fairly easy to repeat. If I start a sync of a snapshot via
>> zrep, it hangs the box.  CTRL+T shows
>>
>>
>> DEBUG: overiding stale lock on zroot/chyves from pid 19378
>> sending zroot/chyves at zrep_000010 to 10.151.9.2:zroot/chyves
>> cannot receive new filesystem stream: destination
>> 'zroot/chyves/guests/resi/disk1' exists
>> must specify -F to overwrite it

Are you sending a snapshot from the host to the guest?
or is the guest sending to the host?

>> load: 0.48  cmd: zfs 29690 [tx->tx_sync_done_cv] 358.94r 0.00u 0.00s 0%
>> 3476k
>> load: 0.48  cmd: zfs 29690 [tx->tx_sync_done_cv] 360.42r 0.00u 0.00s 0%
> 
> I was able to get ps to run and not sure if its helpful or not, but
> these are the two unkillable zfs processes
> 
> 
> root         5   0.0  0.0       0    128  -  DL   10:09      0:05.58
> [zfskern]
> root     29683   0.0  0.0    7752   3824  5  DE+  15:53      0:00.32 zfs
> send -R -I zroot/chyves at zrep_00000d zroot/chyves at zrep_000010
> root     29690   0.0  0.0    7752   3492  5  D+   15:53      0:00.00 zfs
> rename zroot/chyves at zrep_000010 zroot/chyves at zrep_000010_unsent
> 

Should be unrelated but several years ago there was a deadlock issue
when renaming a zvol, unless the fix got undone recently this may just
be a distraction or a hint of a solution...

https://svnweb.freebsd.org/base?view=revision&revision=272474


-- 
FreeBSD - the place to B...Storing Data

Shane Ambler