Re: [PATCH] file-posix: detect the lock using the real file

[Li Feng]

Kevin Wolf <kwolf@redhat.com> 于2020年12月10日周四 上午1:43写道：
>
> Am 09.12.2020 um 10:33 hat Daniel P. Berrangé geschrieben:
> > On Tue, Dec 08, 2020 at 03:38:22PM +0100, Kevin Wolf wrote:
> > > Am 08.12.2020 um 13:59 hat Li Feng geschrieben:
> > > > This patch addresses this issue:
> > > > When accessing a volume on an NFS filesystem without supporting the 
> > > > file lock,
> > > > tools, like qemu-img, will complain "Failed to lock byte 100".
> > > >
> > > > In the original code, the qemu_has_ofd_lock will test the lock on the
> > > > "/dev/null" pseudo-file. Actually, the file.locking is per-drive 
> > > > property,
> > > > which depends on the underlay filesystem.
> > > >
> > > > In this patch, make the 'qemu_has_ofd_lock' with a filename be more 
> > > > generic
> > > > and reasonable.
> > > >
> > > > Signed-off-by: Li Feng <fengli@smartx.com>
> > >
> > > Do you know any way how I could configure either the NFS server or the
> > > NFS client such that locking would fail? For any patch related to this,
> > > it would be good if I could even test the scenario.
> >
> > One could write a qtest that uses an LD_PRELOAD to replace the standard
> > glibc fcntl() function with one that returns an error for locking commands.
>
> Sounds a bit ugly, but for regression testing it could make sense.
>
> However, part of the testing would be to verify that we our checks
> actually match the kernel code, which this approach couldn't solve.
>
Hi, Kevin and Daniel.

How about this patch? I think it's very straightforward.
Except we need a mock syscall test case.

> Kevin
>