Description of problem: When a mandatory lock request arrives at the posix locks xlator, it should be sent success only when all the conflicting operations(like write) which were already STACK WOUND have been UNWOUND too. If not, then in the rarest of the cases it may happen that a write from the client which had acquired the lock wins the race against the write which was WOUND but not UNWOUND there by causing loss of data guarantees. How reproducible: Very rare, but it a bug nevertheless. Credits: Raghavendra Gowdappa explained the possibility of this bug.
Thanks for the report, and considering the rarity of the bug, and as we had not seen this in last 3+ years, marking it as DEFERRED. We will revisit and pick this if we get time resource/time later.