[PATCH v2] xfs: add a test for number of open zones on conventional devices

[PATCH v2] xfs: add a test for number of open zones on conventional devices

[Christoph Hellwig]

Test that mounts of zoned file systems on conventional devices don't
create more open zones than allowed when the last blocks in one or more
zones have been invalidated.

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

 - delete odd files
 - fix the sysfs attr check to be on the scratch and not the test fs

 tests/xfs/4201     | 66 ++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/4201.out |  1 +
 2 files changed, 67 insertions(+)
 create mode 100755 tests/xfs/4201
 create mode 100644 tests/xfs/4201.out

diff --git a/tests/xfs/4201 b/tests/xfs/4201
new file mode 100755
index 000000000000..150978a9deda
--- /dev/null
+++ b/tests/xfs/4201
@@ -0,0 +1,66 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2026 Christoph Hellwig.
+#
+# FS QA Test No. 4201
+#
+# Test that mounts of zoned file systems on conventional devices don't create
+# more open zones than allowed when the last blocks in one or more zones have
+# been invalidated.
+#
+
+. ./common/preamble
+_begin_fstest auto quick zone
+
+_require_scratch
+_require_odirect
+_require_non_zoned_device $SCRATCH_DEV
+_require_non_zoned_device $SCRATCH_RTDEV
+
+open_zones=5
+blocks=65536
+bsize=4096
+
+zone_size_mib=$((blocks * bsize / 1024 / 1024))
+file_size_mib=$((zone_size_mib / 2))
+nr_files=$((open_zones * 3 - 1))
+
+echo "zone size: $zone_size_mib" >>$seqres.full
+echo "file size: $file_size_mib" >>$seqres.full
+echo "nr files: $nr_files" >>$seqres.full
+
+_scratch_mkfs -b size=${bsize} -r zoned=1,rgsize=${blocks}b \
+	> "$seqres.full" 2>&1
+_scratch_mount -o max_open_zones=${open_zones}
+
+_has_fs_sysfs_attr $SCRATCH_DEV zoned/nr_open_zones || \
+	_notrun "This test requires /sys/fs/${FSTYP}/$SCRATCH_DEV/zoned/nr_open_zones"
+
+for i in $(seq 1 $nr_files); do
+	dd if=/dev/zero of=$SCRATCH_MNT/$i bs=1M count=$file_size_mib \
+		 oflag=direct > /dev/null 2>&1
+done
+
+# delete odd numbered files as they should be at the end of the zones
+for ((i = 0; i < $nr_files; i = $((i + 2)))); do
+	rm $SCRATCH_MNT/$((i + 1))
+done
+
+_scratch_unmount
+_scratch_mount -o max_open_zones=${open_zones}
+
+max_open=$(_get_fs_sysfs_attr $SCRATCH_DEV zoned/max_open_zones)
+nr_open=$(_get_fs_sysfs_attr $SCRATCH_DEV zoned/nr_open_zones)
+
+if [ ${max_open} -ne $((open_zones - 1)) ]; then
+	_fail "Test needs adjustments for multiple open GC zones"
+fi
+
+if [ ${nr_open} -gt ${max_open} ]; then
+	_fail "Too many open zones: $nr_open/$max_open"
+fi
+
+_scratch_unmount
+
+status=0
+exit
diff --git a/tests/xfs/4201.out b/tests/xfs/4201.out
new file mode 100644
index 000000000000..32e395dd4d6d
--- /dev/null
+++ b/tests/xfs/4201.out
@@ -0,0 +1 @@
+QA output created by 4201
-- 
2.47.3

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Damien Le Moal]

On 4/1/26 00:23, Christoph Hellwig wrote:
> Test that mounts of zoned file systems on conventional devices don't
> create more open zones than allowed when the last blocks in one or more
> zones have been invalidated.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Looks good to me.

Reviewed-by: Damien Le Moal <dlemoal@kernel.org>

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Zorro Lang]

On Tue, Mar 31, 2026 at 05:23:20PM +0200, Christoph Hellwig wrote:
> Test that mounts of zoned file systems on conventional devices don't
> create more open zones than allowed when the last blocks in one or more
> zones have been invalidated.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
> 
>  - delete odd files
>  - fix the sysfs attr check to be on the scratch and not the test fs
> 
>  tests/xfs/4201     | 66 ++++++++++++++++++++++++++++++++++++++++++++++
>  tests/xfs/4201.out |  1 +
>  2 files changed, 67 insertions(+)
>  create mode 100755 tests/xfs/4201
>  create mode 100644 tests/xfs/4201.out
> 
> diff --git a/tests/xfs/4201 b/tests/xfs/4201
> new file mode 100755
> index 000000000000..150978a9deda
> --- /dev/null
> +++ b/tests/xfs/4201
> @@ -0,0 +1,66 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2026 Christoph Hellwig.
> +#
> +# FS QA Test No. 4201
> +#
> +# Test that mounts of zoned file systems on conventional devices don't create
> +# more open zones than allowed when the last blocks in one or more zones have
> +# been invalidated.
> +#
> +
> +. ./common/preamble
> +_begin_fstest auto quick zone
> +
> +_require_scratch
> +_require_odirect
> +_require_non_zoned_device $SCRATCH_DEV
> +_require_non_zoned_device $SCRATCH_RTDEV
> +
> +open_zones=5
> +blocks=65536
> +bsize=4096
> +
> +zone_size_mib=$((blocks * bsize / 1024 / 1024))
> +file_size_mib=$((zone_size_mib / 2))
> +nr_files=$((open_zones * 3 - 1))
> +
> +echo "zone size: $zone_size_mib" >>$seqres.full
> +echo "file size: $file_size_mib" >>$seqres.full
> +echo "nr files: $nr_files" >>$seqres.full
> +
> +_scratch_mkfs -b size=${bsize} -r zoned=1,rgsize=${blocks}b \
> +	> "$seqres.full" 2>&1

I usually tack on "|| _fail 'mkfs failed'" after _scratch_mount if it's
using extra options. That way, we'll know for sure if the mkfs actually
failed.

> +_scratch_mount -o max_open_zones=${open_zones}
> +
> +_has_fs_sysfs_attr $SCRATCH_DEV zoned/nr_open_zones || \
> +	_notrun "This test requires /sys/fs/${FSTYP}/$SCRATCH_DEV/zoned/nr_open_zones"

Can this be replaced with "_require_fs_sysfs_attr" ? _require_fs_sysfs_attr also can
help to get the real path to the sysfs directory for the fs on a device.

Thanks,
Zorro

> +
> +for i in $(seq 1 $nr_files); do
> +	dd if=/dev/zero of=$SCRATCH_MNT/$i bs=1M count=$file_size_mib \
> +		 oflag=direct > /dev/null 2>&1
> +done
> +
> +# delete odd numbered files as they should be at the end of the zones
> +for ((i = 0; i < $nr_files; i = $((i + 2)))); do
> +	rm $SCRATCH_MNT/$((i + 1))
> +done
> +
> +_scratch_unmount
> +_scratch_mount -o max_open_zones=${open_zones}
> +
> +max_open=$(_get_fs_sysfs_attr $SCRATCH_DEV zoned/max_open_zones)
> +nr_open=$(_get_fs_sysfs_attr $SCRATCH_DEV zoned/nr_open_zones)
> +
> +if [ ${max_open} -ne $((open_zones - 1)) ]; then
> +	_fail "Test needs adjustments for multiple open GC zones"
> +fi
> +
> +if [ ${nr_open} -gt ${max_open} ]; then
> +	_fail "Too many open zones: $nr_open/$max_open"
> +fi
> +
> +_scratch_unmount

Is this line part of the test?

Thanks,
Zorro

> +
> +status=0
> +exit
> diff --git a/tests/xfs/4201.out b/tests/xfs/4201.out
> new file mode 100644
> index 000000000000..32e395dd4d6d
> --- /dev/null
> +++ b/tests/xfs/4201.out
> @@ -0,0 +1 @@
> +QA output created by 4201
> -- 
> 2.47.3
> 
> 

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Christoph Hellwig]

On Wed, Apr 01, 2026 at 05:18:07AM +0800, Zorro Lang wrote:
> > +
> > +_scratch_mkfs -b size=${bsize} -r zoned=1,rgsize=${blocks}b \
> > +	> "$seqres.full" 2>&1
> 
> I usually tack on "|| _fail 'mkfs failed'" after _scratch_mount if it's
> using extra options. That way, we'll know for sure if the mkfs actually
> failed.

Yes, that is useful here, especially if we want to support older kernels
before the zoned device support or !CONFIG_XFS_RT kernels.

> > +_has_fs_sysfs_attr $SCRATCH_DEV zoned/nr_open_zones || \
> > +	_notrun "This test requires /sys/fs/${FSTYP}/$SCRATCH_DEV/zoned/nr_open_zones"
> 
> Can this be replaced with "_require_fs_sysfs_attr" ? _require_fs_sysfs_attr also can
> help to get the real path to the sysfs directory for the fs on a device.

No, this must happen on the scratch file system where we force the
zoned allocator, as the attribute only exists for those.

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Zorro Lang]

On Wed, Apr 01, 2026 at 08:18:28AM +0200, Christoph Hellwig wrote:
> On Wed, Apr 01, 2026 at 05:18:07AM +0800, Zorro Lang wrote:
> > > +
> > > +_scratch_mkfs -b size=${bsize} -r zoned=1,rgsize=${blocks}b \
> > > +	> "$seqres.full" 2>&1
> > 
> > I usually tack on "|| _fail 'mkfs failed'" after _scratch_mount if it's
> > using extra options. That way, we'll know for sure if the mkfs actually
> > failed.
> 
> Yes, that is useful here, especially if we want to support older kernels
> before the zoned device support or !CONFIG_XFS_RT kernels.

Oh, if so, we might need _require_realtime. But that's still not enough to
check mkfs.xfs and mount.xfs on old system (then notrun), we might need
more _require_* helpers for this test. What do you think?

Thanks,
Zorro

> 
> > > +_has_fs_sysfs_attr $SCRATCH_DEV zoned/nr_open_zones || \
> > > +	_notrun "This test requires /sys/fs/${FSTYP}/$SCRATCH_DEV/zoned/nr_open_zones"
> > 
> > Can this be replaced with "_require_fs_sysfs_attr" ? _require_fs_sysfs_attr also can
> > help to get the real path to the sysfs directory for the fs on a device.
> 
> No, this must happen on the scratch file system where we force the
> zoned allocator, as the attribute only exists for those.
> 

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Christoph Hellwig]

On Wed, Apr 01, 2026 at 03:59:52PM +0800, Zorro Lang wrote:
> > > > +_scratch_mkfs -b size=${bsize} -r zoned=1,rgsize=${blocks}b \
> > > > +	> "$seqres.full" 2>&1
> > > 
> > > I usually tack on "|| _fail 'mkfs failed'" after _scratch_mount if it's
> > > using extra options. That way, we'll know for sure if the mkfs actually
> > > failed.
> > 
> > Yes, that is useful here, especially if we want to support older kernels
> > before the zoned device support or !CONFIG_XFS_RT kernels.
> 
> Oh, if so, we might need _require_realtime. But that's still not enough to
> check mkfs.xfs and mount.xfs on old system (then notrun), we might need
> more _require_* helpers for this test. What do you think?


I think _notrun if mkfs or the initial mount fails might be easier and
better here.

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Zorro Lang]

On Wed, Apr 01, 2026 at 04:06:46PM +0200, Christoph Hellwig wrote:
> On Wed, Apr 01, 2026 at 03:59:52PM +0800, Zorro Lang wrote:
> > > > > +_scratch_mkfs -b size=${bsize} -r zoned=1,rgsize=${blocks}b \
> > > > > +	> "$seqres.full" 2>&1
> > > > 
> > > > I usually tack on "|| _fail 'mkfs failed'" after _scratch_mount if it's
> > > > using extra options. That way, we'll know for sure if the mkfs actually
> > > > failed.
> > > 
> > > Yes, that is useful here, especially if we want to support older kernels
> > > before the zoned device support or !CONFIG_XFS_RT kernels.
> > 
> > Oh, if so, we might need _require_realtime. But that's still not enough to
> > check mkfs.xfs and mount.xfs on old system (then notrun), we might need
> > more _require_* helpers for this test. What do you think?
> 
> 
> I think _notrun if mkfs or the initial mount fails might be easier and
> better here.

How about a _require_xfs_mkfs_zoned ?

And about `_scratch_mount -o max_open_zones=${open_zones}`, if max_open_zones
mount option isn't supported, _scratch_mount will fail directly. So how about

  _try_scratch_mount -o max_open_zones=${open_zones} || \
          _notrun "max_open_zones option is not supported"

Or we can have a helper for that, as xfs/642~644 and xfs/668 also need that.

Thanks,
Zorro

> 

Re: [PATCH v2] xfs: add a test for number of open zones on conventional devices

[Christoph Hellwig]

On Tue, Apr 07, 2026 at 01:05:28AM +0800, Zorro Lang wrote:
> > 
> > I think _notrun if mkfs or the initial mount fails might be easier and
> > better here.
> 
> How about a _require_xfs_mkfs_zoned ?
> 
> And about `_scratch_mount -o max_open_zones=${open_zones}`, if max_open_zones
> mount option isn't supported, _scratch_mount will fail directly. So how about
> 
>   _try_scratch_mount -o max_open_zones=${open_zones} || \
>           _notrun "max_open_zones option is not supported"
> 
> Or we can have a helper for that, as xfs/642~644 and xfs/668 also need that.

I'll take a look.  max_open_zones is always supported for zoned xfs
file systems - the notrun is intended to cover the case where the kernel
can't mount zoned file systems - either because it is too old, or because
XFS_RT is not enabled.