Sorry for taking so long in providing you feedback, I moved from one company to another and there is another engineer dealing with this case on the previous company (and I lost emails from this thread, only noticed it because debian bug merged this into its bug and I got my name highlighted).
Nevertheless, yes, I discovered - through the dump and execution logic - final user was enforcing it by configuration file without any particular reason for it.
I didn't have the chance to ask final user to remove that and try again (which I think subsequent communication here might indicate as next steps), but I'm almost sure that this is indeed the issue.
Tks for providing feedback so fast to all my investigation!! The only thing here as a *TODO* for upstream work, if I might say that, would be to understand better why fcntl is the cause for the race (not guaranteeing atomicity).
Likely a sync and/or I/O barrier issue ? Exactly how you described in your tests!
Hello Eric,
Sorry for taking so long in providing you feedback, I moved from one company to another and there is another engineer dealing with this case on the previous company (and I lost emails from this thread, only noticed it because debian bug merged this into its bug and I got my name highlighted).
Nevertheless, yes, I discovered - through the dump and execution logic - final user was enforcing it by configuration file without any particular reason for it.
I didn't have the chance to ask final user to remove that and try again (which I think subsequent communication here might indicate as next steps), but I'm almost sure that this is indeed the issue.
Tks for providing feedback so fast to all my investigation!! The only thing here as a *TODO* for upstream work, if I might say that, would be to understand better why fcntl is the cause for the race (not guaranteeing atomicity).
Likely a sync and/or I/O barrier issue ? Exactly how you described in your tests!