illumos gate

According the RFC 7530, section 9.4. Blocking Locks, the NFSv4 server should maintain an ordered list of pending blocking locks. This is not our case, and thus the locking could be seen as unfair from NFSv4 clients.

To reproduce the issue you could use the attached lock.c test:

# gcc -o lock lock.c 
# ./lock /mnt/file
child: locked
child: locked
child: locked
child: locked
child: locked
child: locked
child: locked
child: locked
child: locked
^C
#

Note: /mnt is mounted from other illumos box using NFSv4 (-o vers=4).

When the same test is run either locally, or over NFSv3, the result is following (as one would expect):

# ./lock /tmp/file
child: locked
parent: locked
child: locked
parent: locked
child: locked
parent: locked
child: locked
parent: locked
^C
#