On Mon, Aug 30, 2021 at 12:08 PM Nikolay Borisov <nborisov@xxxxxxxx> wrote:
>
>
>
> On 30.08.21 г. 13:56, Filipe Manana wrote:
> > On Mon, Aug 30, 2021 at 8:18 AM Nikolay Borisov <nborisov@xxxxxxxx> wrote:
> >>
> >>
> >>
> >> <snip>
> >>> Finally, this would also be a good opportunity to test regular renames
> >>> with subvolumes too, as we had bugs and regressions in the past like
> >>> in commit 4871c1588f92c6c13f4713a7009f25f217055807 ("Btrfs: use right
> >>> root when checking for hash collision
> >>> "), and never got any test cases for them.
> >>
> >> What specific tests do you have in mind? Ordinary renames of files
> >> within a subvolume are already tested by merit of generic geneirc/02[345].
> >
> > So besides the case mentioned in that patch's changelog (renaming a
> > subvolume), checking that we can't rename an inode across subvolumes.
> > Something like checking that:
> >
> > rename /mnt/subvol1/file /mnt/subvol2/file
> >
> > fails with -EXDEV.
>
> But this is already checked by merit of using this line:
>
> _rename_tests_source_dest $SCRATCH_MNT/subvol1/src
> $SCRATCH_MNT/subvol2/dst "cross-subvol" "-x"
>
>
> it tests multiple combinations of regular/symbolic/directory/dev/null
> pairs. I.e :
>
> cross-subvol regu/regu -> Invalid cross-device link
>
>
>
> So this is already covered I'd say. Or you mean to test those
> combinations even without rename exchange, which would boil down to
> calling _rename_tests_source_dest without the -x flag.
Yes, without "-x" (that's what I meant by "regular renames").
>
> >
> > Thanks.
> >
> >
> >>
> >> The test in the patch you cited is basically renaming a subvolume within
> >> the same subvolume, that's easy enough.
> >>
> >> <snip>
> >
> >
> >
--
Filipe David Manana,
“Whether you think you can, or you think you can't — you're right.”
