[PATCH v2] btrfs: test block group size class loading logic

[PATCH v2] btrfs: test block group size class loading logic

[Boris Burkov]

Add a new test which checks that size classes in freshly loaded block
groups after a cycle mount match size classes before going down

Depends on the kernel patch:
btrfs: add size class stats to sysfs

Signed-off-by: Boris Burkov <boris@bur.io>
---
v2: drop the fixed_by_kernel_commit since the fix is not out past the
btrfs development tree, so the fix is getting rolled in to the original
broken commit. Modified the commit message to note the dependency on the
new sysfs counters.

 tests/btrfs/283     | 49 +++++++++++++++++++++++++++++++++++++++++++++
 tests/btrfs/283.out |  2 ++
 2 files changed, 51 insertions(+)
 create mode 100755 tests/btrfs/283
 create mode 100644 tests/btrfs/283.out

diff --git a/tests/btrfs/283 b/tests/btrfs/283
new file mode 100755
index 00000000..b3614786
--- /dev/null
+++ b/tests/btrfs/283
@@ -0,0 +1,49 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2023 Meta Platforms, Inc.  All Rights Reserved.
+#
+# FS QA Test 283
+#
+# Test that mounting a btrfs filesystem properly loads block group size classes.
+#
+. ./common/preamble
+_begin_fstest auto quick mount
+
+sysfs_size_classes() {
+	local uuid="$(findmnt -n -o UUID "$SCRATCH_MNT")"
+	cat "/sys/fs/btrfs/$uuid/allocation/data/size_classes"
+}
+
+_supported_fs btrfs
+_require_scratch
+_require_btrfs_fs_sysfs
+
+f="$SCRATCH_MNT/f"
+small=$((16 * 1024))
+medium=$((1024 * 1024))
+large=$((16 * 1024 * 1024))
+
+_scratch_mkfs >/dev/null
+_scratch_mount
+# Write files with extents in each size class
+$XFS_IO_PROG -fc "pwrite -q 0 $small" $f.small
+$XFS_IO_PROG -fc "pwrite -q 0 $medium" $f.medium
+$XFS_IO_PROG -fc "pwrite -q 0 $large" $f.large
+# Sync to force the extent allocation
+sync
+pre=$(sysfs_size_classes)
+
+# cycle mount to drop the block group cache
+_scratch_cycle_mount
+
+# Another write causes us to actually load the block groups
+$XFS_IO_PROG -fc "pwrite -q 0 $large" $f.large.2
+sync
+
+post=$(sysfs_size_classes)
+diff <(echo $pre) <(echo $post)
+
+echo "Silence is golden"
+# success, all done
+status=0
+exit
diff --git a/tests/btrfs/283.out b/tests/btrfs/283.out
new file mode 100644
index 00000000..efb2c583
--- /dev/null
+++ b/tests/btrfs/283.out
@@ -0,0 +1,2 @@
+QA output created by 283
+Silence is golden
-- 
2.39.1

Re: [PATCH v2] btrfs: test block group size class loading logic

[Anand Jain]

On 26/01/2023 06:43, Boris Burkov wrote:
> Add a new test which checks that size classes in freshly loaded block
> groups after a cycle mount match size classes before going down
> 
> Depends on the kernel patch:
> btrfs: add size class stats to sysfs
> 
> Signed-off-by: Boris Burkov <boris@bur.io>
> ---
> v2: drop the fixed_by_kernel_commit since the fix is not out past the
> btrfs development tree, so the fix is getting rolled in to the original
> broken commit. Modified the commit message to note the dependency on the
> new sysfs counters.

You can add it as below, it will be helpful to some extent.
Depending on the status of the kernel patch at a later point,
we can update the fixed_by_kernel_commit.

fixed_by_kernel_commit xxxxxxxx "btrfs: add size class stats to sysfs".


> 
>   tests/btrfs/283     | 49 +++++++++++++++++++++++++++++++++++++++++++++
>   tests/btrfs/283.out |  2 ++
>   2 files changed, 51 insertions(+)
>   create mode 100755 tests/btrfs/283
>   create mode 100644 tests/btrfs/283.out
> 
> diff --git a/tests/btrfs/283 b/tests/btrfs/283
> new file mode 100755
> index 00000000..b3614786
> --- /dev/null
> +++ b/tests/btrfs/283
> @@ -0,0 +1,49 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2023 Meta Platforms, Inc.  All Rights Reserved.
> +#
> +# FS QA Test 283
> +#
> +# Test that mounting a btrfs filesystem properly loads block group size classes.
> +#
> +. ./common/preamble
> +_begin_fstest auto quick mount
> +
> +sysfs_size_classes() {
> +	local uuid="$(findmnt -n -o UUID "$SCRATCH_MNT")"
> +	cat "/sys/fs/btrfs/$uuid/allocation/data/size_classes"

The kernels without the sysfs patch must fail with not_run.
  _require_fs_sysfs() will help here.

Thx.