[PATCH] fstests: btrfs: add a test case for failed compressed inline attempt

[PATCH] fstests: btrfs: add a test case for failed compressed inline attempt

[Qu Wenruo]

[BUG]
There is a long existing behavior that if a compressed inline attempt
failed, the whole inode is marked incompressible.

Furthermore, the default threshold for compressed inline implies a
pretty high compression ratio (2K max inline vs 4K block size, thus 50%
compression ratio).

This makes btrfs to mark inodes incompressible prematurely.

[TEST]
The new test case is based on the previous btrfs/343 one, which will do:

- Create and mount the fs with compress,max_inline=4
  4 bytes will be too small for any header, thus all compressed inline
  writes should fail.

- Do the following operations:
  * Buffered write [0, blocksize / 2)
    Which can be inlined if using the default max_inline mount option
    and 4K fs block.
    But the new max_inline=4 will reject such compressed inline attempt.

    And such one block write should not trigger compression thus no
    compression ratio based check.

  * Sync
    For unpatched kernels, this will set the inode with NOCOMPRESS flag
    and reject all future compression on that inode.

  * Buffered write [1M, 2M)
    For unpatched kernels, the range will not be compressed due to the
    NOCOMPRESS flag.

- Unmount the fs

- Maker sure that:
  * The inode has no NOCOMPRESS flag
  * File extent at file offset 1M is being compressed

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 tests/btrfs/344     | 53 +++++++++++++++++++++++++++++++++++++++++++++
 tests/btrfs/344.out |  2 ++
 2 files changed, 55 insertions(+)
 create mode 100755 tests/btrfs/344
 create mode 100644 tests/btrfs/344.out

diff --git a/tests/btrfs/344 b/tests/btrfs/344
new file mode 100755
index 00000000..4695104f
--- /dev/null
+++ b/tests/btrfs/344
@@ -0,0 +1,53 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2026 SUSE S.A.  All Rights Reserved.
+#
+# FS QA Test 344
+#
+# Check if a failed inline attempt for compression write will mark
+# the whole inode as incompressible
+#
+. ./common/preamble
+_begin_fstest auto quick compress
+
+_require_scratch
+_require_btrfs_command inspect-internal dump-tree
+
+_fixed_by_kernel_commit xxxxxxxxxxxx \
+	"btrfs: do not mark inode incompressible after inline attempt failed"
+
+_scratch_mkfs >>$seqres.full 2>&1
+
+# This inline limit is too small for any header.
+# Thus all compressed inline attempt should fail.
+_scratch_mount "-o compress,max_inline=4"
+
+blocksize=$(_get_file_block_size $SCRATCH_MNT)
+
+# The initial half block write can be compressed, but the compressed
+# size will not meet the 4 bytes limits, thus it will not be inlined.
+#
+# Normally such single block write itself should not mark the inode
+# incompressible, no matter if there is an inline attempt or not.
+#
+# The 1M write should be compressed.
+$XFS_IO_PROG -f -c "pwrite 0 $(( $blocksize / 2 ))" -c sync \
+		-c "pwrite 1m 1m" $SCRATCH_MNT/foobar >> $seqres.full
+ino=$(stat -c "%i" $SCRATCH_MNT/foobar)
+_scratch_unmount
+
+# Dump the fs tree into seqres.full for debug.
+$BTRFS_UTIL_PROG inspect-internal dump-tree -t 5 $SCRATCH_DEV >> $seqres.full
+
+# Check the NOCOMPRESS flag of the inode.
+$BTRFS_UTIL_PROG inspect-internal dump-tree -t 5 $SCRATCH_DEV |\
+grep -A 4 -e "item .* key ($ino INODE_ITEM 0)" | grep -q NOCOMPRESS
+[ $? -eq 0 ] && echo "inode $ino has NOCOMPRESS flag"
+
+# Check the file extent at file offset 1m.
+$BTRFS_UTIL_PROG inspect-internal dump-tree -t 5 $SCRATCH_DEV |\
+grep -A 4 -e "item .* key ($ino EXTENT_DATA 1048576)" | grep -q "compression 0"
+[ $? -eq 0 ] && echo "inode $ino file offset 1M is not compressed"
+
+echo "Silence is golden"
+_exit 0
diff --git a/tests/btrfs/344.out b/tests/btrfs/344.out
new file mode 100644
index 00000000..b950d5ae
--- /dev/null
+++ b/tests/btrfs/344.out
@@ -0,0 +1,2 @@
+QA output created by 344
+Silence is golden
-- 
2.51.2

Re: [PATCH] fstests: btrfs: add a test case for failed compressed inline attempt

[Filipe Manana]

On Mon, Feb 16, 2026 at 3:40 AM Qu Wenruo <wqu@suse.com> wrote:
>
> [BUG]
> There is a long existing behavior that if a compressed inline attempt
> failed, the whole inode is marked incompressible.
>
> Furthermore, the default threshold for compressed inline implies a
> pretty high compression ratio (2K max inline vs 4K block size, thus 50%
> compression ratio).
>
> This makes btrfs to mark inodes incompressible prematurely.
>
> [TEST]
> The new test case is based on the previous btrfs/343 one, which will do:
>
> - Create and mount the fs with compress,max_inline=4
>   4 bytes will be too small for any header, thus all compressed inline
>   writes should fail.
>
> - Do the following operations:
>   * Buffered write [0, blocksize / 2)
>     Which can be inlined if using the default max_inline mount option
>     and 4K fs block.
>     But the new max_inline=4 will reject such compressed inline attempt.
>
>     And such one block write should not trigger compression thus no
>     compression ratio based check.
>
>   * Sync
>     For unpatched kernels, this will set the inode with NOCOMPRESS flag
>     and reject all future compression on that inode.
>
>   * Buffered write [1M, 2M)
>     For unpatched kernels, the range will not be compressed due to the
>     NOCOMPRESS flag.
>
> - Unmount the fs
>
> - Maker sure that:
>   * The inode has no NOCOMPRESS flag
>   * File extent at file offset 1M is being compressed
>
> Signed-off-by: Qu Wenruo <wqu@suse.com>
> ---
>  tests/btrfs/344     | 53 +++++++++++++++++++++++++++++++++++++++++++++
>  tests/btrfs/344.out |  2 ++
>  2 files changed, 55 insertions(+)
>  create mode 100755 tests/btrfs/344
>  create mode 100644 tests/btrfs/344.out
>
> diff --git a/tests/btrfs/344 b/tests/btrfs/344
> new file mode 100755
> index 00000000..4695104f
> --- /dev/null
> +++ b/tests/btrfs/344
> @@ -0,0 +1,53 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2026 SUSE S.A.  All Rights Reserved.
> +#
> +# FS QA Test 344
> +#
> +# Check if a failed inline attempt for compression write will mark
> +# the whole inode as incompressible
> +#
> +. ./common/preamble
> +_begin_fstest auto quick compress
> +
> +_require_scratch
> +_require_btrfs_command inspect-internal dump-tree
> +
> +_fixed_by_kernel_commit xxxxxxxxxxxx \
> +       "btrfs: do not mark inode incompressible after inline attempt failed"
> +
> +_scratch_mkfs >>$seqres.full 2>&1
> +
> +# This inline limit is too small for any header.
> +# Thus all compressed inline attempt should fail.

attempt -> attempts

> +_scratch_mount "-o compress,max_inline=4"
> +
> +blocksize=$(_get_file_block_size $SCRATCH_MNT)
> +
> +# The initial half block write can be compressed, but the compressed
> +# size will not meet the 4 bytes limits, thus it will not be inlined.

limits -> limit

Otherwise, it looks good, thanks.

Reviewed-by: Filipe Manana <fdmanana@suse.com>


> +#
> +# Normally such single block write itself should not mark the inode
> +# incompressible, no matter if there is an inline attempt or not.
> +#
> +# The 1M write should be compressed.
> +$XFS_IO_PROG -f -c "pwrite 0 $(( $blocksize / 2 ))" -c sync \
> +               -c "pwrite 1m 1m" $SCRATCH_MNT/foobar >> $seqres.full
> +ino=$(stat -c "%i" $SCRATCH_MNT/foobar)
> +_scratch_unmount
> +
> +# Dump the fs tree into seqres.full for debug.
> +$BTRFS_UTIL_PROG inspect-internal dump-tree -t 5 $SCRATCH_DEV >> $seqres.full
> +
> +# Check the NOCOMPRESS flag of the inode.
> +$BTRFS_UTIL_PROG inspect-internal dump-tree -t 5 $SCRATCH_DEV |\
> +grep -A 4 -e "item .* key ($ino INODE_ITEM 0)" | grep -q NOCOMPRESS
> +[ $? -eq 0 ] && echo "inode $ino has NOCOMPRESS flag"
> +
> +# Check the file extent at file offset 1m.
> +$BTRFS_UTIL_PROG inspect-internal dump-tree -t 5 $SCRATCH_DEV |\
> +grep -A 4 -e "item .* key ($ino EXTENT_DATA 1048576)" | grep -q "compression 0"
> +[ $? -eq 0 ] && echo "inode $ino file offset 1M is not compressed"
> +
> +echo "Silence is golden"
> +_exit 0
> diff --git a/tests/btrfs/344.out b/tests/btrfs/344.out
> new file mode 100644
> index 00000000..b950d5ae
> --- /dev/null
> +++ b/tests/btrfs/344.out
> @@ -0,0 +1,2 @@
> +QA output created by 344
> +Silence is golden
> --
> 2.51.2
>
>