On 21/05/26 12:13PM, Zorro Lang wrote:
> On Tue, May 25, 2021 at 09:09:07AM -0700, Darrick J. Wong wrote:
> > On Tue, May 25, 2021 at 12:46:42PM +0800, Zorro Lang wrote:
> > > If a swapfile doesn't contain even a single page-aligned contiguous
> > > range of blocks, it's an invalid swapfile, and might cause kernel
> > > issue. This case covered commit 5808fecc5723 ("iomap: Fix negative
> > > assignment to unsigned sis->pages in iomap_swapfile_activate").
> > >
> > > Signed-off-by: Zorro Lang <zlang@xxxxxxxxxx>
> > > ---
> > > tests/generic/638 | 79 +++++++++++++++++++++++++++++++++++++++++++
> > > tests/generic/638.out | 2 ++
> > > tests/generic/group | 1 +
> > > 3 files changed, 82 insertions(+)
> > > create mode 100755 tests/generic/638
> > > create mode 100644 tests/generic/638.out
> > >
> > > diff --git a/tests/generic/638 b/tests/generic/638
> > > new file mode 100755
> > > index 00000000..74bc39cb
> > > --- /dev/null
> > > +++ b/tests/generic/638
> > > @@ -0,0 +1,79 @@
> > > +#! /bin/bash
> > > +# SPDX-License-Identifier: GPL-2.0
> > > +# Copyright (c) 2021 Red Hat Inc. All Rights Reserved.
> > > +#
> > > +# FS QA Test 638
> > > +#
> > > +# Test small swapfile which doesn't contain even a single page-aligned contiguous
> > > +# range of blocks. This case covered commit 5808fecc5723 ("iomap: Fix negative
> > > +# assignment to unsigned sis->pages in iomap_swapfile_activate").
> > > +#
> > > +seq=`basename $0`
> > > +seqres=$RESULT_DIR/$seq
> > > +echo "QA output created by $seq"
> > > +
> > > +here=`pwd`
> > > +tmp=/tmp/$$
> > > +status=1 # failure is the default!
> > > +trap "_cleanup; exit \$status" 0 1 2 3 15
> > > +
> > > +_cleanup()
> > > +{
> > > + cd /
> > > + rm -f $tmp.*
> > > +}
> > > +
> > > +# get standard environment, filters and checks
> > > +. ./common/rc
> > > +. ./common/filter
> > > +
> > > +# remove previous $seqres.full before test
> > > +rm -f $seqres.full
> > > +
> > > +# real QA test starts here
> > > +_supported_fs generic
> > > +_require_scratch
> > > +_require_scratch_swapfile
> > > +_require_test_program mkswap
> > > +_require_test_program swapon
> > > +
> > > +make_unaligned_swapfile()
> > > +{
> > > + local fname=$1
> > > + local n=$((psize / bsize - 1))
> > > +
> > > + # Make sure the swapfile doesn't contain even a single page-aligned
> > > + # contiguous range of blocks. This's necessary to cover the bug
> > > + $XFS_IO_PROG -f -t -c "pwrite 0 $(((psize + bsize) * n))" $fname >> $seqres.full 2>&1
> > > + for((i=1; i<=n; i++));do
> > > + $XFS_IO_PROG -c "fcollapse $(((psize - bsize) * i)) $bsize" $fname
> > > + done
> > > + chmod 0600 $fname
> > > + $CHATTR_PROG +C $fname > /dev/null 2>&1
> > > + $here/src/mkswap $fname
> > > +}
> > > +
> > > +_scratch_mkfs >> $seqres.full 2>&1
> > > +_scratch_mount
> > > +psize=`get_page_size`
> > > +bsize=`_get_block_size $SCRATCH_MNT`
> >
> > _get_file_block_size, since fcollapse requires arguments aligned to the
> > file's allocation unit, which isn't necessarily the same as the fs block
> > size (bigalloc, rt, etc.)...
>
> You're right.
>
> >
> > > +# Due to we need page-unaligned blocks, so blocksize < pagesize is necessary.
> > > +# If not, try to make a smaller enough block size
> > > +if [ $bsize -ge $psize ];then
> > > + _scratch_unmount
> > > + _scratch_mkfs_blocksized 1024 >> $seqres.full 2>&1
> >
> > ...which leads me to the next question, which is: Should the above
> > helper override mkfs options as necessary to enforce that the allocation
> > unit is 1024 bytes? Or is it time for a new helper?
>
> Hmm... the _scratch_mkfs_blocksized() does:
>
> xfs)
> _scratch_mkfs_xfs $MKFS_OPTIONS -b size=$blocksize
>
> _scratch_mkfs_xfs() already use $MKFS_OPTIONS internally, don't need to give
> $MKFS_OPTIONS to it again. Although _scratch_mkfs_xfs deals with $MKFS_OPTIONS,
> it just drops whole $MKFS_OPTIONS if there're conflicts.
>
> The _scratch_mkfs_geom() does overwrite, but _scratch_mkfs_sized doesn't (sized
> filesystem is rarely seen in local.config). We might let _scratch_mkfs_blocksized
> learn _scratch_mkfs_geom.
>
> >
> > > + if [ $? -ne 0 ];then
> > > + _notrun "Can't make filesystem block size < page size"
> > > + fi
> > > + _scratch_mount
> > > + bsize=1024
> >
> > You probably want to re-check _get_file_block_size to make sure that you
> > actually got blocksize < pagesize in the format attempt.
>
> Yes, you're right. _scratch_mkfs_blocksized can't make sure to get that blocksize.
>
Sorry, about delay in getting a test ready for this kernel bug. Also I only now
noticed this patch of yours, after I had sent a patch generic/639 to cover this
bug. (sheer coincidence).
I have tested [1] on both x86 and Power platform for all filesystems config.
(ext4-1k, ext4-4k. ext4-64k, ext2, xfs-1k, xfs-4k, xfs-64k, btrfs) (64k if for Power)
But I am fine if you would like to push your patch upstream.
[1]: https://patchwork.kernel.org/project/fstests/patch/e1f9798462ef60648db24b6291e1b149b114f2f2.1622105066.git.riteshh@xxxxxxxxxxxxx/
-ritesh
