[PATCH v2 0/2] add test for zone_gc_low_space attribute writes

[PATCH v2 0/2] add test for zone_gc_low_space attribute writes

[Hans Holmberg]

Adds a test for zone_gc_low_space attribute writes and some helpers to
go along with that.

The new test will fail currently, but i've submitted a fix:
https://lore.kernel.org/linux-xfs/20260325124312.26349-1-hans.holmberg@wdc.com/

v2:
- factored out useful helpers to common/xfs
- style fixes suggested by Christoph
- added a check that the scratch rt device is zoned before reading zone
  capacity

v1:
https://lore.kernel.org/fstests/20260325125013.26631-1-hans.holmberg@wdc.com/

Hans Holmberg (2):
  xfs/647: turn rt dev block device lookup into a common helper
  xfs: test that zone_gc_low_space writes start gc for rw fses

 common/xfs        | 28 ++++++++++++++++++
 tests/xfs/647     | 10 +------
 tests/xfs/999     | 72 +++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/999.out |  2 ++
 4 files changed, 103 insertions(+), 9 deletions(-)
 create mode 100755 tests/xfs/999
 create mode 100644 tests/xfs/999.out

-- 
2.34.1

[PATCH v2 1/2] xfs/647: turn rt dev block device lookup into a common helper

[Hans Holmberg]

Turn the logic figuring out on which block device the rt data is stored on
into a common helper.

Signed-off-by: Hans Holmberg <hans.holmberg@wdc.com>
---
 common/xfs    | 11 +++++++++++
 tests/xfs/647 | 10 +---------
 2 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/common/xfs b/common/xfs
index 7fa0db2e26b4..c81f939e68d2 100644
--- a/common/xfs
+++ b/common/xfs
@@ -737,6 +737,17 @@ _xfs_metadump_supports_rt()
 	$XFS_METADUMP_PROG --help 2>&1 | grep -q -- '-r rtdev'
 }
 
+# What block device is rt data located on for scratch?
+_xfs_get_scratch_rtdev_bdev()
+{
+	# Is the rt dev internal or not?
+	if [ -z "$SCRATCH_RTDEV" ]; then
+		echo "$SCRATCH_DEV"
+	else
+		echo "$SCRATCH_RTDEV"
+	fi
+}
+
 # Snapshot the metadata on the scratch device
 _scratch_xfs_metadump()
 {
diff --git a/tests/xfs/647 b/tests/xfs/647
index 76959b28c779..0ef340a14b07 100755
--- a/tests/xfs/647
+++ b/tests/xfs/647
@@ -17,15 +17,7 @@ _begin_fstest auto quick zone
 _require_scratch
 _require_realtime
 
-#
-# Figure out if the rt section is internal or not
-#
-if [ -z "$SCRATCH_RTDEV" ]; then
-	zdev=$SCRATCH_DEV
-else
-	zdev=$SCRATCH_RTDEV
-fi
-
+zdev="$(_xfs_get_scratch_rtdev_bdev)"
 _require_zoned_device $zdev
 _require_command "$BLKZONE_PROG" blkzone
 
-- 
2.34.1

Re: [PATCH v2 1/2] xfs/647: turn rt dev block device lookup into a common helper

[Darrick J. Wong]

On Fri, Mar 27, 2026 at 01:34:28PM +0100, Hans Holmberg wrote:
> Turn the logic figuring out on which block device the rt data is stored on
> into a common helper.
> 
> Signed-off-by: Hans Holmberg <hans.holmberg@wdc.com>

Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  common/xfs    | 11 +++++++++++
>  tests/xfs/647 | 10 +---------
>  2 files changed, 12 insertions(+), 9 deletions(-)
> 
> diff --git a/common/xfs b/common/xfs
> index 7fa0db2e26b4..c81f939e68d2 100644
> --- a/common/xfs
> +++ b/common/xfs
> @@ -737,6 +737,17 @@ _xfs_metadump_supports_rt()
>  	$XFS_METADUMP_PROG --help 2>&1 | grep -q -- '-r rtdev'
>  }
>  
> +# What block device is rt data located on for scratch?
> +_xfs_get_scratch_rtdev_bdev()
> +{
> +	# Is the rt dev internal or not?
> +	if [ -z "$SCRATCH_RTDEV" ]; then
> +		echo "$SCRATCH_DEV"
> +	else
> +		echo "$SCRATCH_RTDEV"
> +	fi
> +}
> +
>  # Snapshot the metadata on the scratch device
>  _scratch_xfs_metadump()
>  {
> diff --git a/tests/xfs/647 b/tests/xfs/647
> index 76959b28c779..0ef340a14b07 100755
> --- a/tests/xfs/647
> +++ b/tests/xfs/647
> @@ -17,15 +17,7 @@ _begin_fstest auto quick zone
>  _require_scratch
>  _require_realtime
>  
> -#
> -# Figure out if the rt section is internal or not
> -#
> -if [ -z "$SCRATCH_RTDEV" ]; then
> -	zdev=$SCRATCH_DEV
> -else
> -	zdev=$SCRATCH_RTDEV
> -fi
> -
> +zdev="$(_xfs_get_scratch_rtdev_bdev)"
>  _require_zoned_device $zdev
>  _require_command "$BLKZONE_PROG" blkzone
>  
> -- 
> 2.34.1
> 
> 

Re: [PATCH v2 1/2] xfs/647: turn rt dev block device lookup into a common helper

[Christoph Hellwig]

On Fri, Mar 27, 2026 at 01:34:28PM +0100, Hans Holmberg wrote:
> Turn the logic figuring out on which block device the rt data is stored on
> into a common helper.

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v2 2/2] xfs: test that zone_gc_low_space writes start gc for rw fses

[Hans Holmberg]

Test that writes to the sysfs attribute zone_gc_low_space triggers
garbage collection for rw (but not ro) file systems.

Signed-off-by: Hans Holmberg <hans.holmberg@wdc.com>
---
 common/xfs        | 17 +++++++++++
 tests/xfs/999     | 72 +++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/999.out |  2 ++
 3 files changed, 91 insertions(+)
 create mode 100755 tests/xfs/999
 create mode 100644 tests/xfs/999.out

diff --git a/common/xfs b/common/xfs
index c81f939e68d2..45d7f75d2064 100644
--- a/common/xfs
+++ b/common/xfs
@@ -748,6 +748,23 @@ _xfs_get_scratch_rtdev_bdev()
 	fi
 }
 
+# Get a stat from /proc/self/mountstats for a specific mount point
+_xfs_get_mountstat() {
+	awk -v mount="$1" -v stat="$2" \
+		'$0~"mounted on "mount" with fstype xfs"{f=1} f&&index($0,stat){print $NF;exit}' \
+		/proc/self/mountstats
+}
+
+# Get the user available blocks for a rt mount
+_xfs_get_user_available_rt_blocks() {
+	_xfs_get_mountstat "$1" "user available RT blocks:"
+}
+
+# Check if gc is required for a rt mount
+_xfs_get_rt_gc_required () {
+	_xfs_get_mountstat "$1" "RT GC required:"
+}
+
 # Snapshot the metadata on the scratch device
 _scratch_xfs_metadump()
 {
diff --git a/tests/xfs/999 b/tests/xfs/999
new file mode 100755
index 000000000000..2c5b54368b6d
--- /dev/null
+++ b/tests/xfs/999
@@ -0,0 +1,72 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2026 Western Digital Corporation
+#
+# FS QA Test No. 999
+#
+# Tests that writes to zonegc_low_space will trigger start of garbage
+# collection for rw (but not ro) file systems
+#
+. ./common/preamble
+_begin_fstest auto rw zone quick
+
+. ./common/zoned
+
+_require_scratch
+_require_odirect
+
+zdev="$(_xfs_get_scratch_rtdev_bdev)"
+_require_zoned_device "$zdev"
+capacity=$(_zone_capacity 0 $zdev)
+_scratch_mkfs_sized $(($capacity * 6)) > $seqres.full 2>&1
+
+# limit max open zones to avoid gc to trigger before we want it to
+export MOUNT_OPTIONS="$MOUNT_OPTIONS -o max_open_zones=2"
+_try_scratch_mount || _notrun "mount option not supported"
+_require_xfs_scratch_zoned
+
+# create three zone-sized files and invalidate 95% of the data
+for i in $(seq 1 3); do
+	filename=$SCRATCH_MNT/data_$i
+	dd if=/dev/zero of=$filename bs=1M oflag=direct\
+		count=$(($capacity / (1024 * 1024))) >> $seqres.full 2>&1
+	truncate -s $(($capacity / 20)) $filename
+done
+
+# remount read-only and set the gc limit to free up unused space
+_scratch_remount ro
+rt_available_pre="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
+_set_fs_sysfs_attr $SCRATCH_DEV "zoned/zonegc_low_space" 100
+
+# reclaiming a zone with 5% used blocks should take way less than 1 second
+sleep 1
+
+# gc should not run for read only file systems
+rt_available_post="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
+[ "$rt_available_post" -gt "$rt_available_pre" ] && \
+	_fail "gc should not run while read only"
+
+_set_fs_sysfs_attr $SCRATCH_DEV "zoned/zonegc_low_space" 0
+
+# remount rw and check that gc starts, frees up blocks and stops after
+# the threshold is set to reclaim the unused blocks
+_scratch_remount rw
+
+rt_available_pre="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
+_set_fs_sysfs_attr $SCRATCH_DEV "zoned/zonegc_low_space" 100
+
+i=0
+while [ $(_xfs_get_rt_gc_required "$SCRATCH_MNT") -eq "1" ]; do
+	sleep 1
+	let i=$i+1
+	[ $i -gt 3 ] && _fail "gc did not complete within a reasonable time"
+done
+
+rt_available_post="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
+[ "$rt_available_pre" -ge "$rt_available_post" ] && \
+	_fail "gc did not free up space"
+
+echo "Silence is golden"
+status=0
+
+exit
diff --git a/tests/xfs/999.out b/tests/xfs/999.out
new file mode 100644
index 000000000000..3b276ca804fc
--- /dev/null
+++ b/tests/xfs/999.out
@@ -0,0 +1,2 @@
+QA output created by 999
+Silence is golden
-- 
2.34.1

Re: [PATCH v2 2/2] xfs: test that zone_gc_low_space writes start gc for rw fses

[Darrick J. Wong]

On Fri, Mar 27, 2026 at 01:34:29PM +0100, Hans Holmberg wrote:
> Test that writes to the sysfs attribute zone_gc_low_space triggers
> garbage collection for rw (but not ro) file systems.
> 
> Signed-off-by: Hans Holmberg <hans.holmberg@wdc.com>

Looks good to me, given what I saw of the kernel patch.  It probably
needs a _fixed_by_kernel_commit tag once the patch goes upstream.

Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  common/xfs        | 17 +++++++++++
>  tests/xfs/999     | 72 +++++++++++++++++++++++++++++++++++++++++++++++
>  tests/xfs/999.out |  2 ++
>  3 files changed, 91 insertions(+)
>  create mode 100755 tests/xfs/999
>  create mode 100644 tests/xfs/999.out
> 
> diff --git a/common/xfs b/common/xfs
> index c81f939e68d2..45d7f75d2064 100644
> --- a/common/xfs
> +++ b/common/xfs
> @@ -748,6 +748,23 @@ _xfs_get_scratch_rtdev_bdev()
>  	fi
>  }
>  
> +# Get a stat from /proc/self/mountstats for a specific mount point
> +_xfs_get_mountstat() {
> +	awk -v mount="$1" -v stat="$2" \
> +		'$0~"mounted on "mount" with fstype xfs"{f=1} f&&index($0,stat){print $NF;exit}' \
> +		/proc/self/mountstats
> +}
> +
> +# Get the user available blocks for a rt mount
> +_xfs_get_user_available_rt_blocks() {
> +	_xfs_get_mountstat "$1" "user available RT blocks:"
> +}
> +
> +# Check if gc is required for a rt mount
> +_xfs_get_rt_gc_required () {
> +	_xfs_get_mountstat "$1" "RT GC required:"
> +}
> +
>  # Snapshot the metadata on the scratch device
>  _scratch_xfs_metadump()
>  {
> diff --git a/tests/xfs/999 b/tests/xfs/999
> new file mode 100755
> index 000000000000..2c5b54368b6d
> --- /dev/null
> +++ b/tests/xfs/999
> @@ -0,0 +1,72 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2026 Western Digital Corporation
> +#
> +# FS QA Test No. 999
> +#
> +# Tests that writes to zonegc_low_space will trigger start of garbage
> +# collection for rw (but not ro) file systems
> +#
> +. ./common/preamble
> +_begin_fstest auto rw zone quick
> +
> +. ./common/zoned
> +
> +_require_scratch
> +_require_odirect
> +
> +zdev="$(_xfs_get_scratch_rtdev_bdev)"
> +_require_zoned_device "$zdev"
> +capacity=$(_zone_capacity 0 $zdev)
> +_scratch_mkfs_sized $(($capacity * 6)) > $seqres.full 2>&1
> +
> +# limit max open zones to avoid gc to trigger before we want it to
> +export MOUNT_OPTIONS="$MOUNT_OPTIONS -o max_open_zones=2"
> +_try_scratch_mount || _notrun "mount option not supported"
> +_require_xfs_scratch_zoned
> +
> +# create three zone-sized files and invalidate 95% of the data
> +for i in $(seq 1 3); do
> +	filename=$SCRATCH_MNT/data_$i
> +	dd if=/dev/zero of=$filename bs=1M oflag=direct\
> +		count=$(($capacity / (1024 * 1024))) >> $seqres.full 2>&1
> +	truncate -s $(($capacity / 20)) $filename
> +done
> +
> +# remount read-only and set the gc limit to free up unused space
> +_scratch_remount ro
> +rt_available_pre="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
> +_set_fs_sysfs_attr $SCRATCH_DEV "zoned/zonegc_low_space" 100
> +
> +# reclaiming a zone with 5% used blocks should take way less than 1 second
> +sleep 1
> +
> +# gc should not run for read only file systems
> +rt_available_post="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
> +[ "$rt_available_post" -gt "$rt_available_pre" ] && \
> +	_fail "gc should not run while read only"
> +
> +_set_fs_sysfs_attr $SCRATCH_DEV "zoned/zonegc_low_space" 0
> +
> +# remount rw and check that gc starts, frees up blocks and stops after
> +# the threshold is set to reclaim the unused blocks
> +_scratch_remount rw
> +
> +rt_available_pre="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
> +_set_fs_sysfs_attr $SCRATCH_DEV "zoned/zonegc_low_space" 100
> +
> +i=0
> +while [ $(_xfs_get_rt_gc_required "$SCRATCH_MNT") -eq "1" ]; do
> +	sleep 1
> +	let i=$i+1
> +	[ $i -gt 3 ] && _fail "gc did not complete within a reasonable time"
> +done
> +
> +rt_available_post="$(_xfs_get_user_available_rt_blocks "$SCRATCH_MNT")"
> +[ "$rt_available_pre" -ge "$rt_available_post" ] && \
> +	_fail "gc did not free up space"
> +
> +echo "Silence is golden"
> +status=0
> +
> +exit
> diff --git a/tests/xfs/999.out b/tests/xfs/999.out
> new file mode 100644
> index 000000000000..3b276ca804fc
> --- /dev/null
> +++ b/tests/xfs/999.out
> @@ -0,0 +1,2 @@
> +QA output created by 999
> +Silence is golden
> -- 
> 2.34.1
> 
> 

Re: [PATCH v2 2/2] xfs: test that zone_gc_low_space writes start gc for rw fses

[Christoph Hellwig]

On Fri, Mar 27, 2026 at 01:34:29PM +0100, Hans Holmberg wrote:
> Test that writes to the sysfs attribute zone_gc_low_space triggers
> garbage collection for rw (but not ro) file systems.

.. and add common helpers for that.  I tend to split the helpers out
into separate commit, but I'm not sure if Zorro cares all that much
about it.

Otherwise looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>