btrfs-progs: add block-group-tree to the default mkfs/convert features
Subsystems (2)
v2 cover btrfs-progs: add block-group-tree to the default mkfs/convert features
[CHANGELOG]
v2:
- Automatically remove bgt feature when dependent feature is missing Instead of erroring out. This will allow us to run the existing no-holes/v1 free space cache test cases without any modification.
I was planning to do this during v6.12 but forgot it and now the next LTS kernel release is not that far away, it's finally time to make the switch.
The first patch is to change mkfs/btrfs-convert from rejecting "bgt,^no-holes" to disabling bgt when no-holes/fst feature is not selected.
This allows the existing "mkfs.btrfs ^no-holes" to be executed without extra modification.
The second patch is a large page-size specific fix, where on 64K page size systems misc/057 will fail due to subpage mounts always enable v2 free space cache, resulting later conversion failure to fst (as it's already fst).
The final patch is the one enabling new default block-group-tree feature for mkfs and convert.
Qu Wenruo (3):
btrfs-progs: disable block-group-tree feature if dependency is missing
btrfs-progs: misc-tests: check if free space tree is enabled after
mount
btrfs-progs: add block-group-tree to the default mkfs features
Documentation/mkfs.btrfs.rst | 7 ++++++- common/fsfeatures.c | 2 +- common/fsfeatures.h | 3 ++- convert/main.c | 4 ++-- mkfs/main.c | 10 +++++++--- tests/misc-tests/057-btrfstune-free-space-tree/test.sh | 8 ++++++++ 6 files changed, 26 insertions(+), 8 deletions(-)
--
2.52.0
PATCH v2 1/3 btrfs-progs: disable block-group-tree feature if dependency is missing
Currently mkfs.btrfs will fail with the following features:
# mkfs.btrfs -O block-group-tree,^no-holes $dev btrfs-progs v6.17.1 See https://btrfs.readthedocs.io for more information.
ERROR: block group tree requires no-holes and free-space-tree features
That's due to the artificial feature requirements from block-group-tree, which requires no-holes and free-space-tree.
But such mandatory rejection will block our migration to the new default block-group-tree tree features as a lot of no-holes test cases will cause false alerts (and we do not have plan to deprecate explicit holes yet, or do we?).
So to avoid the new default block-group-tree feature from causing tons of false alerts, automatically disable block-group-tree feature if the end user disables no-holes or free-space-tree features.
Now the above command will success, but with extra warnings and block-group-tree disabled.
# mkfs.btrfs -O block-group-tree,^no-holes $dev btrfs-progs v6.17.1 See https://btrfs.readthedocs.io for more information.
WARNING: disabling block-group-tree feature due to missing no-holes and free-space-tree features
Label: (null)
UUID: 3c7adb81-c0c3-4980-8ef3-485019e5d8c4
Node size: 16384
Sector size: 4096 (CPU page size: 4096)
Filesystem size: 1.00GiB
Block group profiles:
Data: single 8.00MiB
Metadata: DUP 51.19MiB
System: DUP 8.00MiB
SSD detected: no
Zoned device: no
Features: extref, skinny-metadata, free-space-tree
Checksum: crc32c...
--- Documentation/mkfs.btrfs.rst | 5 +++++ convert/main.c | 4 ++-- mkfs/main.c | 10 +++++++--- 3 files changed, 14 insertions(+), 5 deletions(-)
Documentation/mkfs.btrfs.rst +5 -0
@@ -436,6 +436,11 @@ block-group-tree enabled at *mkfs* time is possible, see :doc:`btrfstune`. Online conversion is not possible. + .. note:: + This feature requires ``no-holes`` and ``free-space-tree`` + features, if those dependency features are disabled, + ``block-group-tree`` feature will also be disabled automatically. + .. _mkfs-feature-raid-stripe-tree: raid-stripe-tree
convert/main.c +2 -2
@@ -1214,8 +1214,8 @@ static int do_convert(const char *devname, u32 convert_flags, u32 nodesize,if ((features->compat_ro_flags &BTRFS_FEATURE_COMPAT_RO_BLOCK_GROUP_TREE ) && (!(features->incompat_flags &BTRFS_FEATURE_INCOMPAT_NO_HOLES ) || !(features->compat_ro_flags &BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE ))) { -error ("block group tree requires no-holes and free-space-tree features" ); - goto fail; +warning ("disabling block-group-tree feature due to missing no-holes and free-space-tree features" ); + features->compat_ro_flags &= ~BTRFS_FEATURE_COMPAT_RO_BLOCK_GROUP_TREE ; } fd =open (devname,O_RDWR );if (fd <0 ) {
mkfs/main.c +7 -3
@@ -1849,13 +1849,17 @@ int BOX_MAIN(mkfs)(int argc, char **argv) } } -/* Block group tree feature requires no-holes and free-space-tree. */ +/* +* Block group tree feature requires no-holes and free-space-tree. +* And if those dependency is disabled, also disable block-group-tree feature. +*/ if (features.compat_ro_flags &BTRFS_FEATURE_COMPAT_RO_BLOCK_GROUP_TREE && (!(features.incompat_flags &BTRFS_FEATURE_INCOMPAT_NO_HOLES ) || !(features.compat_ro_flags &BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE ))) { -error ("block group tree requires no-holes and free-space-tree features" ); -exit (1 ); +warning ("disabling block-group-tree feature due to missing no-holes and free-space-tree features" ); + features.compat_ro_flags &= ~BTRFS_FEATURE_COMPAT_RO_BLOCK_GROUP_TREE ; } +if (opt_zoned) {const int blkid_version =blkid_get_library_version (NULL ,NULL );
Signature · 1 line
2.52.0
PATCH v2 2/3 btrfs-progs: misc-tests: check if free space tree is enabled after mount
In reply to Qu WenruoFor bs < ps cases, a mount will always enable free-space-tree due to the limitation of v1 space cache.
Test case misc/057 will lead to false failure if the page size is larger than the default 4K sector size.
Add an extra check on free-space-tree after fs population, and skip the test case if so.
--- tests/misc-tests/057-btrfstune-free-space-tree/test.sh | 8 ++++++++ 1 file changed, 8 insertions(+)
tests/misc-tests/057-btrfstune-free-space-tree/test.sh +8 -0
@@ -17,6 +17,14 @@ run_check_mount_test_dev populate_fs run_check_umount_test_dev +# Check if the fs has free space tree already. Currently bs < ps mount +# will always enable free-space-tree (no support for v1 free space cache) +if run_check_stdout "$TOP/btrfs" inspect-internal dump-super "$TEST_DEV" |\ + grep -q "FREE_SPACE_TREE"; then + _not_run "free-space-tree is always enabled for page size $(getconf PAGESIZE)" +fi + run_check $SUDO_HELPER "$TOP/btrfstune" --convert-to-free-space-tree "$TEST_DEV" + run_check "$TOP/btrfs" check "$TEST_DEV"
Signature · 1 line
2.52.0
PATCH v2 3/3 btrfs-progs: add block-group-tree to the default mkfs features
In reply to Qu WenruoThe block-group-tree feature is already considered safe since v6.6 kernels, and btrfs-progs have also improved its off-line conversion ability (btrfstune --convert-to|from-block-group-tree).
Now it's time to make bgt feature as the default mkfs features. This affects both mkfs and btrfs-convert.
The target merge window is v6.18.
--- Documentation/mkfs.btrfs.rst | 2 +- common/fsfeatures.c | 2 +- common/fsfeatures.h | 3 ++- 3 files changed, 4 insertions(+), 3 deletions(-)
Documentation/mkfs.btrfs.rst +1 -1
@@ -424,7 +424,7 @@ free-space-tree .. _mkfs-feature-block-group-tree: block-group-tree - (kernel support since 6.1) + (default since btrfs-progs v6.19, kernel support since 6.1) Enable a dedicated b-tree for block group items, this greatly reduces mount time for large filesystems due to better data locality that
common/fsfeatures.c +1 -1
@@ -219,7 +219,7 @@ static const struct btrfs_feature mkfs_features[] = { .sysfs_name = "block_group_tree",VERSION_TO_STRING2 (compat,6 ,1 ),VERSION_TO_STRING2 (safe,6 ,6 ), -VERSION_NULL (default), +VERSION_TO_STRING2 (default,6 ,19 ), .desc = "block group tree, more efficient block group tracking to reduce mount time" }, #if EXPERIMENTAL
common/fsfeatures.h +2 -1
@@ -46,7 +46,8 @@ struct btrfs_mkfs_features {static const struct btrfs_mkfs_features btrfs_mkfs_default_features = { .compat_ro_flags =BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE | -BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID , +BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID | +BTRFS_FEATURE_COMPAT_RO_BLOCK_GROUP_TREE , .incompat_flags =BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF |BTRFS_FEATURE_INCOMPAT_NO_HOLES |BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA ,
Signature · 1 line
2.52.0
Hi Qu,
Quoted 10 lines · Qu Wenruo
[CHANGELOG]
v2:
- Automatically remove bgt feature when dependent feature is missing Instead of erroring out. This will allow us to run the existing no-holes/v1 free space cache test cases without any modification.I was planning to do this during v6.12 but forgot it and now the next LTS kernel release is not that far away, it's finally time to make the switch.
Does this also mean that block-group-tree is considered recommended and production-ready for users of linux-6.12 LTS, or that it will only be recommended for the next LTS kernel series?
Also, are there any known disadvantages that result from having used btrfstune compared with a pristine FS created with mkfs.btrfs -O block-group-tree ?
Best,
Nicholas
Quoted 10 lines · Nicholas D Steeves
Hi Qu,
Quoted 6 lines · Qu Wenruo
[CHANGELOG]
v2:
- Automatically remove bgt feature when dependent feature is missing Instead of erroring out. This will allow us to run the existing no-holes/v1 free space cache test cases without any modification.
Quoted 8 lines
Quoted 4 lines
I was planning to do this during v6.12 but forgot it and now the next LTS kernel release is not that far away, it's finally time to make the switch.
Does this also mean that block-group-tree is considered recommended and production-ready for users of linux-6.12 LTS, or that it will only be recommended for the next LTS kernel series?
I think 6.12 is already safe enough (the current safe version is 6.6).
Quoted 4 lines
Also, are there any known disadvantages that result from having used btrfstune compared with a pristine FS created with mkfs.btrfs -O block-group-tree ?
I can only come up with the disadvantage of btrfstune itself, e.g:
- Older bugs in btrfstune that failed to resume interrupted conversion
- Certain ENOSPC bugs if the fs is really full
- Offline conversion
As long as the fs can be safely converted, there should be no difference.
Thanks,
Qu
Quoted 3 lines
Best,
Nicholas