FYI, failures in tst-fdopendir for 2.44

[Florian Weimer]

* Andreas K. Huettel:

>> > FYI, I've seen this fail to flaky O_NOATIME detection on tmpfs with relatime default:
>> > https://sourceware.org/glibc/wiki/Testing/Tests/dirent/tst-fdopendir
>> >
>> > The right answer is to rewrite the test using the FUSE framework to look for
>> > fdopendir reopening the file descriptor with different flags.
>> 
>> We've tracked this down to the EDR (Endpoint Detection & Response) tool
>> used at Red Hat:
>> 
>>   osquery does not open files with O_NOATIME during scanning/hashing
>>   <https://github.com/osquery/osquery/issues/8965>
>
> Just to make sure I understand this correctly- 
>
> osquery interposes the function call even during running the glibc
> testsuite, that introduces a subtle change in behaviour and as a
> result the test sometimes fails?
>
> That indeed sounds more like a problem with osquery.

I don't think it interposes anything.  Apparently it uses a combination
of inotify and the kernel auditing subsystem.  It is a problem with
osquery.

(Maybe GNU make could use a jobserver with POSIX semaphores.  It would
avoid such problems.  But GNU make shouldn't have to change for this.)

Thanks,
Florian