On Wed, Jul 22, 2020 at 12:52:16PM +0100, Filipe Manana wrote:
> On Wed, Jul 22, 2020 at 12:48 PM Zorro Lang <zlang@xxxxxxxxxx> wrote:
> >
> > On Wed, Jul 22, 2020 at 02:38:32PM +0800, Murphy Zhou wrote:
> > > Hard coded reflink offset/length numbers are not working well on 64K page
> > > size arch like ppc64le, reporting:
> > > +XFS_IOC_CLONE_RANGE: Invalid argument
> > >
> > > Because they are not aligned with the current PAGE_SIZE.
> > >
> > > Translate original numbers into multiples of PAGE_SIZE. Then we're writing
> > > different testfiles with different md5sums when PAGE_SIZE is different.
> > > So we need to link different output file based on the current PAGE_SIZE.
> > >
> > > This testcase is covering a btrfs regression which has been fixed in
> > > v4.18. I tried to reproduce it on 4.17 kernel but failed with and
> > > without this patch.
> > >
> > > Signed-off-by: Murphy Zhou <jencce.kernel@xxxxxxxxx>
> > > ---
> > > .gitignore | 1 +
> > > tests/generic/501 | 17 +++++++++++++----
> > > tests/generic/{501.out => 501.out.4K} | 0
> > > tests/generic/501.out.64K | 5 +++++
> > > 4 files changed, 19 insertions(+), 4 deletions(-)
> > > rename tests/generic/{501.out => 501.out.4K} (100%)
> > > create mode 100644 tests/generic/501.out.64K
> > >
> > > diff --git a/.gitignore b/.gitignore
> > > index 5f5c4a0f..39318615 100644
> > > --- a/.gitignore
> > > +++ b/.gitignore
> > > @@ -263,6 +263,7 @@
> > > /tests/xfs/033.out
> > > /tests/xfs/071.out
> > > /tests/xfs/096.out
> > > +/tests/generic/501.out
> > >
> > > # cscope files
> > > cscope.*
> > > diff --git a/tests/generic/501 b/tests/generic/501
> > > index 0d1f6ffe..4e589cda 100755
> > > --- a/tests/generic/501
> > > +++ b/tests/generic/501
> > > @@ -9,6 +9,7 @@
> > > # the file again and then power fail, after we mount again the filesystem, no
> > > # file data was lost or corrupted.
> > > #
> > > +seqfull=$0
> > > seq=`basename $0`
> > > seqres=$RESULT_DIR/$seq
> > > echo "QA output created by $seq"
> > > @@ -42,14 +43,22 @@ _require_metadata_journaling $SCRATCH_DEV
> > > _init_flakey
> > > _mount_flakey
> > >
> > > -$XFS_IO_PROG -f -c "pwrite -S 0x18 9000K 6908K" $SCRATCH_MNT/foo >>$seqres.full
> > > -$XFS_IO_PROG -f -c "pwrite -S 0x20 2572K 156K" $SCRATCH_MNT/bar >>$seqres.full
> > > +pagesz=$(getconf PAGE_SIZE)
> > > +
> > > +if [ $pagesz -eq 65536 ] ; then
> > > + ln -sf $seqfull.out.64K $seqfull.out
> > > +elif [ $pagesz -eq 4096 ] ; then
> > > + ln -sf $seqfull.out.4K $seqfull.out
> > > +fi
> >
> > You can't sure there're only 4k and 64k system.
> > This case trys to make sure the md5 of $SCRATCH_MNT/bar isn't changed before and
> > after _flakey_drop_and_remount. So if you turn to use different pagesize, you
> > can silence the golden image, and compare the md5 quietly. Something likes(just
> > example):
>
> This can be simpler. The test can be changed to use multiples of 64K,
> so it will work everywhere (like other reflink tests).
> I'll try it out and test on 4.17 to confirm it still reproduces the
> original btrfs bug.
Sure, if so, it's a simpler way to fix this failure.
Thanks,
Zorro
>
> Thanks.
>
> >
> > diff --git a/tests/generic/501 b/tests/generic/501
> > index 0d1f6ffe..7e7f9be5 100755
> > --- a/tests/generic/501
> > +++ b/tests/generic/501
> > @@ -53,18 +53,21 @@ $XFS_IO_PROG -c "fsync" \
> > -c "fsync" \
> > $SCRATCH_MNT/bar >>$seqres.full
> >
> > -echo "File bar digest before power failure:"
> > -md5sum $SCRATCH_MNT/bar | _filter_scratch
> > +# File bar digest before power failure
> > +md5sum $SCRATCH_MNT/bar > $tmp.md5.org
> >
> > # Simulate a power failure and mount the filesystem to check that no file data
> > # was lost or corrupted.
> > _flakey_drop_and_remount
> >
> > -echo "File bar digest after power failure:"
> > -md5sum $SCRATCH_MNT/bar | _filter_scratch
> > +# File bar digest after power failure
> > +md5sum $SCRATCH_MNT/bar > $tmp.md5.new
> > +diff -u $tmp.md5.org $tmp.md5.new
> >
> > _unmount_flakey
> > _cleanup_flakey
> >
> > +echo "Silence is golden"
> > +
> > status=0
> > exit
> > diff --git a/tests/generic/501.out b/tests/generic/501.out
> > index 5d7da017..00133b63 100644
> > --- a/tests/generic/501.out
> > +++ b/tests/generic/501.out
> > @@ -1,5 +1,2 @@
> > QA output created by 501
> > -File bar digest before power failure:
> > -95a95813a8c2abc9aa75a6c2914a077e SCRATCH_MNT/bar
> > -File bar digest after power failure:
> > -95a95813a8c2abc9aa75a6c2914a077e SCRATCH_MNT/bar
> > +Silence is golden
> >
> > Thanks,
> > Zorro
> >
> > > +
> > > +$XFS_IO_PROG -f -c "pwrite -S 0x18 $((2250*pagesz)) $((1727*pagesz))" $SCRATCH_MNT/foo >>$seqres.full
> > > +$XFS_IO_PROG -f -c "pwrite -S 0x20 $((643*pagesz)) $((39*pagesz))" $SCRATCH_MNT/bar >>$seqres.full
> > >
> > > # We clone from file foo into a range of file bar that overlaps the existing
> > > # extent at file bar. The destination offset of the reflink operation matches
> > > -# the eof position of file bar minus 4Kb.
> > > +# the eof position of file bar minus 1 PAGE_SIZE.
> > > $XFS_IO_PROG -c "fsync" \
> > > - -c "reflink ${SCRATCH_MNT}/foo 0 2724K 15908K" \
> > > + -c "reflink ${SCRATCH_MNT}/foo 0 $((681*pagesz)) $((3977*pagesz))" \
> > > -c "fsync" \
> > > $SCRATCH_MNT/bar >>$seqres.full
> > >
> > > diff --git a/tests/generic/501.out b/tests/generic/501.out.4K
> > > similarity index 100%
> > > rename from tests/generic/501.out
> > > rename to tests/generic/501.out.4K
> > > diff --git a/tests/generic/501.out.64K b/tests/generic/501.out.64K
> > > new file mode 100644
> > > index 00000000..7e1013b2
> > > --- /dev/null
> > > +++ b/tests/generic/501.out.64K
> > > @@ -0,0 +1,5 @@
> > > +QA output created by 501
> > > +File bar digest before power failure:
> > > +3058797b969076e91c518cb206b21163 SCRATCH_MNT/bar
> > > +File bar digest after power failure:
> > > +3058797b969076e91c518cb206b21163 SCRATCH_MNT/bar
> > > --
> > > 2.20.1
> > >
> >
>
>
> --
> Filipe David Manana,
>
> “Whether you think you can, or you think you can't — you're right.”
>
