Re: [PATCH] block: report errno when flock fcntl fails

[David Edmondson]

On Wednesday, 2020-12-16 at 12:57:46 +01, Kevin Wolf wrote:

> Am 16.12.2020 um 12:38 hat David Edmondson geschrieben:
>> On Wednesday, 2020-12-16 at 12:29:40 +01, Kevin Wolf wrote:
>> 
>> > Am 15.12.2020 um 20:01 hat David Edmondson geschrieben:
>> >> When a call to fcntl(2) for the purpose of manipulating file locks
>> >> fails, report the error returned by fcntl.
>> >> 
>> >> Signed-off-by: David Edmondson <david.edmondson@oracle.com>
>> >
>> > Is appending "Resource temporarily unavailable" in the common case (a
>> > file locked by another process) actually an improvement for the message
>> > or more confusing? It would be good to mention the motivation for the
>> > change in the commit message.
>> 
>> Distinguishing between the common case and others is at least a start
>> when trying to figure out why it failed. We have potentially useful
>> information to assist in diagnosis, why throw it away?
>
> I agree in principle, just when I saw the result, it felt more confusing
> than helpful. Maybe the best option would be to include the errno only
> if it's different from EAGAIN? Then the qemu-iotests reference output
> would probably stay unchanged, too.

This is a pretty low-level error report even today - unless you
understand the implementation then it doesn't shout "the file is being
shared".

Given that, I don't see any value in eliding the EAGAIN message, as I
have to know that the lack of the errno string means that it was EAGAIN,
meaning that another process holds a lock.

>> In the case I encountered the failure appears to have been caused by
>> SELinux misconfiguration.
>
> Ah, so you got EPERM?

Yes.

> It can useful information in the future, let's include it in the commit
> message.

Okay.

dme.
-- 
My girl Friday, she no square.