On Wed, Mar 22, 2023 at 1:59 AM Qu Wenruo <wqu@xxxxxxxx> wrote:
>
> [FALSE ALERT]
> If test case btrfs/246 is executed with a 16K page sized system (like
> some aarch64 SoCs) using 4K sector size (would be the new default), the
> test case would fail with output mismatch:
>
> btrfs/246 1s ... - output mismatch (see ~/xfstests-dev/results//btrfs/246.out.bad)
> --- tests/btrfs/246.out 2022-11-24 19:53:53.158470844 +0800
> +++ ~/xfstests-dev/results//btrfs/246.out.bad 2023-03-22 13:27:34.975796048 +0800
> @@ -3,3 +3,5 @@
> 0ca3bfdeda1ef5036bfa5dad078a9f15724e79cf296bd4388cf786bfaf4195d0 SCRATCH_MNT/foobar
> sha256sum after mount cycle
> 0ca3bfdeda1ef5036bfa5dad078a9f15724e79cf296bd4388cf786bfaf4195d0 SCRATCH_MNT/foobar
> +no inline extent found
> +no compressed extent found
> ...
> (Run 'diff -u ~/xfstests-dev/tests/btrfs/246.out ~/xfstests-dev/results//btrfs/246.out.bad' to see the entire diff)
>
> [CAUSE]
> For current btrfs subpage support, there are still some limitations:
>
> - No compressed write if the range is not fully page aligned
> - No inline extents creation
> Reading inline extents is still supported
>
> Thus we won't create such inlined compressed extent at all.
>
> [FIX]
> Just skip the test case if we can not even create a regular inline
> extent.
>
> This is done by a new require helper,
> _require_btrfs_inline_extent_creation(), which would detect if btrfs can
> even create an uncompressed inlined extent.
>
> Reported-by: Hector Martin <marcan@xxxxxxxxx>
> Signed-off-by: Qu Wenruo <wqu@xxxxxxxx>
> ---
> common/btrfs | 22 ++++++++++++++++++++++
> tests/btrfs/246 | 3 +++
> 2 files changed, 25 insertions(+)
>
> diff --git a/common/btrfs b/common/btrfs
> index 7c32360376c2..344509ce300c 100644
> --- a/common/btrfs
> +++ b/common/btrfs
> @@ -499,6 +499,28 @@ _require_btrfs_support_sectorsize()
> _notrun "sectorsize $sectorsize is not supported"
> }
>
> +_require_btrfs_inline_extents_creation()
> +{
> + local ino
> +
> + _require_xfs_io_command fiemap
> + _require_scratch
> +
> + _scratch_mkfs &> /dev/null
> + _scratch_mount -o max_inline=2048,compress=none
> + _pwrite_byte 0x00 0 1024 $SCRATCH_MNT/inline &> /dev/null
> + sync
> + $XFS_IO_PROG -c "fiemap -v" $SCRATCH_MNT/inline | tail -n 1 > $tmp.fiemap
> + _scratch_unmount
> + # 0x200 means inlined, 0x100 means not block aligned, 0x1 means
> + # the last extent.
> + if ! grep -q "0x301" $tmp.fiemap; then
> + rm -f -- $tmp.fiemap
> + _notrun "No inline extent creation support, maybe subpage?"
> + fi
> + rm -f -- $tmp.fiemap
> +}
> +
> _btrfs_metadump()
> {
> local device="$1"
> diff --git a/tests/btrfs/246 b/tests/btrfs/246
> index 0dcc7c0d1a43..2fe54f959048 100755
> --- a/tests/btrfs/246
> +++ b/tests/btrfs/246
> @@ -23,6 +23,9 @@ _cleanup()
> _supported_fs btrfs
> _require_scratch
>
> +# If it's subpage case, we don't support inline extents creation for now.
> +_require_btrfs_inline_extents_creation
> +
> _scratch_mkfs > /dev/null
> _scratch_mount -o compress,max_inline=2048
>
> --
> 2.39.2
>
LGTM.
Reviewed-by: Neal Gompa <neal@xxxxxxxxx>
--
真実はいつも一つ！/ Always, there's only one truth!
