Re: [PATCH 5/5] overlay: test mount error cases with exclusive directories

[Eryu Guan <eguan@redhat.com>]

On Wed, Jul 12, 2017 at 02:08:56PM +0300, Amir Goldstein wrote:
> On Wed, Jul 12, 2017 at 1:47 PM, Eryu Guan <eguan@redhat.com> wrote:
> > On Tue, Jul 11, 2017 at 10:52:06PM +0300, Amir Goldstein wrote:
> >> Expect EBUSY when trying to mount overlay when:
> >> - Upper dir is in-use by another overlay mount
> >> - Work dir is in-use by another overlay mount
> >>
> >> Signed-off-by: Amir Goldstein <amir73il@gmail.com>
> >
> > From patch 2/5, seems this behavior requires inode index feature to be
> > enabled, and now test fails on old kernels for me. Is this expected?
> >
> 
> From my cover letter:
> "The new test overlay/036 checks for new overlay mount behaviors related
> to exclusivity of upper and work dirs among overlay mounts. This test
> is expected to fail on upstream kernel, but for a good reason, because
> the mount configurations that it tests (sharing upper dir among two
> overlay mounts) are not wise at all."
> 
> So the answer is yes, this is expected and also does not depend
> on index feature.

I usually read the cover letter (I saved it in my 'review' mailbox along
with patches to review), but somehow I forgot about it this time..

> 
> I know this is a bit confusing, but inode index enforces that 2
> *subsequent* overlay
> mount don't reuse the same upper/work dir with different lower/upper.
> The test that checks this behavior is overlay/037, still on my
> overlayfs-devel branch.
> 
> overlay/036 verifies that 2 *concurrent* overlay mount don't reuse the same
> upper/work dir. The is enforced by overlayfs-next regardless on index feature.
> A setup like this will cause troubles in old kernel as well, but mount
> will not detect it.

Thanks for the explanation! The commit log and test description don't
provide enough background information. If you could resend with detailed
explanation on the test, that'd be great!

Thanks,
Eryu