Re: [PATCH v2 5/7] overlay: test concurrent copy up of lower hardlinks

[Eryu Guan <eguan@redhat.com>]

On Wed, Jul 05, 2017 at 01:46:51PM +0300, Amir Goldstein wrote:
> On Wed, Jul 5, 2017 at 12:59 PM, Eryu Guan <eguan@redhat.com> wrote:
> > On Tue, Jul 04, 2017 at 02:40:32PM +0300, Amir Goldstein wrote:
> >> Two tasks make a modification concurrently on two hardlinks of a large
> >> lower inode.  The copy up should be triggered by one of the tasks and the
> >> other should be waiting for copy up to complete.  Both copy up targets
> >> should end up being upper hardlinks and both metadata changes should be
> >> visible in both hardlinks.
> >>
> >> With kernel <= v4.12, hardlinks are broken on copy up, meaning that copy up
> >
> > So this will be fixed in 4.13-rc1 kernel?
> 
> It is fixed on current overlayfs-next branch. Yes.
> 
> 
> >> +
> >> +_do_cmd()
> >
> > I'd rename it to do_cmd for a local helper function.
> >
> 
> sure
> 
> >> +{
> >> +     echo "`date +%T` $1..." >> $seqres.full
> >> +     eval "$1"
> >> +     echo "`date +%T` ...$1" >> $seqres.full
> >> +}
> >> +
> >> +# Perform one modification on each hardlink (size and owner)
> >> +_do_cmd "echo >> $SCRATCH_MNT/one" &
> >> +#
> >> +# When hardlinks are broken and overlayfs supports concurrent copy up,
> >> +# $seqres.full will show that file two copy up started ~2s after file one
> >> +# copy up started and ended ~2s after file one copy up ended.
> >> +# With synchronized copy up of lower inodes, $seqres.full will show that
> >> +# file two copy up ended at the same time as file one copy up.
> >> +#
> >> +sleep 2
> >
> > If the first copyup finished within 2 seconds, it's not a concurrent
> > modification on the two hardlinks. Does that break the test assumption?
> 
> It will not be a race, plain 2 consequent copy ups, so not testing anything
> special, but won't cause test to fail.
> 
> I am testing overlayfs with 2 underlying fs configurations:
> 1. xfs with clone support
> 2. ext4
> 
> With xfs clone up takes fraction of a second, the test is not really concurrent.
> With ext4, on my laptop with SSD, the 500M copy up takes ~3 seconds,
> which is quite close to the 2 seconds delay.
> I increased the copy up file size to 1G and now first copy up takes ~6 seconds,
> so there is more slack for the 2 sec delay, but still keeps the test 'quick'.

Then what about don't sleep at all? So it doesn't rely on the copyup
speed, the only drawback is the timestamps won't tell us more about the
copyup behavior, but that doesn't seem critical to this test.

But if sleep is still needed, some comments about it would be good.

Thanks,
Eryu

> 
> >
> >> +_do_cmd "chown 100 $SCRATCH_MNT/two" &
> >> +
> >> +wait
> >> +
> >> +# Expect all hardlinks to show both metadata modifications
> >> +for f in zero one two; do
> >> +     _ls_l -n $SCRATCH_MNT/$f | awk '{ print $2, $3, $5, $9 }' | _filter_scratch
> >
> > Better to comment on what are the 4 fields.
> >
> 
> They show the metadata that was changes by chown and write.
> Will add a comment.
> 
> Thanks,
> Amir.
> --
> To unsubscribe from this list: send the line "unsubscribe fstests" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html