Skip to contentSkip to conversation

btrfs: zoned: reduce lock contention in zoned extent allocator

PATCH In review Johannes Thumshirn 3 participants 6 messages Jun 24, 2025 – Jun 25, 2025 Btrfs file system

lore
Share & tools
lore
#0 lore raw

v2 cover btrfs: zoned: reduce lock contention in zoned extent allocator

From: Johannes Thumshirn <[email protected]>

While investigating performance issues on zoned btrfs, I've analyzed lockstat output while running fio workloads.

Three locks that have always been on top are the space_info's ->lock, as well as fs_info's ->treelog_bg_lock and ->relocation_bg_lock.

The first one is not as trivial to get rid and most likely legit in most cases (see patch #3). The other two are only protecting a single value in fs_info and can be replaced by REAC_ONCE() and WRITE_ONCE().

The series has been run through fstests on tcmu-runner emulated SMR drives as well as Damien's fio benchmarks on real ZNS hardware and no regressions have been observed.

Changes in v2:
- Rewording of the commit message (Damien)
- No more local variables caching the values (Damien)

Johannes Thumshirn (3):
btrfs: zoned: get rid of relocation_bg_lock
btrfs: zoned: get rid of treelog_bg_lock
btrfs: zoned: don't hold space_info lock on zoned allocation

 fs/btrfs/disk-io.c     |  2 -
 fs/btrfs/extent-tree.c | 86 +++++++++++-------------------------------
 fs/btrfs/fs.h          |  7 +---
 fs/btrfs/zoned.c       | 12 +++---
 fs/btrfs/zoned.h       |  6 +--
 5 files changed, 31 insertions(+), 82 deletions(-)
Signature · 1 line
2.49.0
#1 lore raw

PATCH v2 1/3 btrfs: zoned: get rid of relocation_bg_lock

From: Johannes Thumshirn <[email protected]>

Lockstat analysis of benchmark workloads shows a very high contention on relocation_bg_lock. But the lock only protects a single field in 'struct btrfs_fs_info', namely 'u64 data_reloc_bg'.

Use READ_ONCE()/WRITE_ONCE() to access 'btrfs_fs_info::data_reloc_bg'.

---
 fs/btrfs/disk-io.c     |  1 -
 fs/btrfs/extent-tree.c | 34 ++++++++++++----------------------
 fs/btrfs/fs.h          |  6 +-----
 fs/btrfs/zoned.c       | 10 ++++------
 4 files changed, 17 insertions(+), 34 deletions(-)
fs/btrfs/disk-io.c +0 -1
index ee3cdd7035cc..24896322376d 100644
@@ -2791,7 +2791,6 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	spin_lock_init(&fs_info->unused_bgs_lock);
 	spin_lock_init(&fs_info->treelog_bg_lock);
 	spin_lock_init(&fs_info->zone_active_bgs_lock);
-	spin_lock_init(&fs_info->relocation_bg_lock);
 	rwlock_init(&fs_info->tree_mod_log_lock);
 	rwlock_init(&fs_info->global_root_lock);
 	mutex_init(&fs_info->unused_bg_unpin_mutex);
fs/btrfs/extent-tree.c +12 -22
index 10f50c725313..80ceca89a9ef 100644
@@ -3842,7 +3842,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	u64 avail;
 	u64 bytenr = block_group->start;
 	u64 log_bytenr;
-	u64 data_reloc_bytenr;
 	int ret = 0;
 	bool skip = false;
 
@@ -3865,14 +3864,9 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	 * Do not allow non-relocation blocks in the dedicated relocation block
 	 * group, and vice versa.
 	 */
-	spin_lock(&fs_info->relocation_bg_lock);
-	data_reloc_bytenr = fs_info->data_reloc_bg;
-	if (data_reloc_bytenr &&
-	    ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
-	     (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
-		skip = true;
-	spin_unlock(&fs_info->relocation_bg_lock);
-	if (skip)
+	if (READ_ONCE(fs_info->data_reloc_bg) &&
+	    ((ffe_ctl->for_data_reloc && bytenr != READ_ONCE(fs_info->data_reloc_bg)) ||
+	     (!ffe_ctl->for_data_reloc && bytenr == READ_ONCE(fs_info->data_reloc_bg))))
 		return 1;
 
 	/* Check RO and no space case before trying to activate it */
@@ -3899,7 +3893,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	spin_lock(&space_info->lock);
 	spin_lock(&block_group->lock);
 	spin_lock(&fs_info->treelog_bg_lock);
-	spin_lock(&fs_info->relocation_bg_lock);
 
 	if (ret)
 		goto out;
@@ -3908,8 +3901,8 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	       block_group->start == fs_info->treelog_bg ||
 	       fs_info->treelog_bg == 0);
 	ASSERT(!ffe_ctl->for_data_reloc ||
-	       block_group->start == fs_info->data_reloc_bg ||
-	       fs_info->data_reloc_bg == 0);
+	       block_group->start == READ_ONCE(fs_info->data_reloc_bg) ||
+	       READ_ONCE(fs_info->data_reloc_bg) == 0);
 
 	if (block_group->ro ||
 	    (!ffe_ctl->for_data_reloc &&
@@ -3932,7 +3925,7 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	 * Do not allow currently used block group to be the data relocation
 	 * dedicated block group.
 	 */
-	if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
+	if (ffe_ctl->for_data_reloc && READ_ONCE(fs_info->data_reloc_bg) == 0 &&
 	    (block_group->used || block_group->reserved)) {
 		ret = 1;
 		goto out;
@@ -3957,8 +3950,8 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 		fs_info->treelog_bg = block_group->start;
 
 	if (ffe_ctl->for_data_reloc) {
-		if (!fs_info->data_reloc_bg)
-			fs_info->data_reloc_bg = block_group->start;
+		if (READ_ONCE(fs_info->data_reloc_bg) == 0)
+			WRITE_ONCE(fs_info->data_reloc_bg, block_group->start);
 		/*
 		 * Do not allow allocations from this block group, unless it is
 		 * for data relocation. Compared to increasing the ->ro, setting
@@ -3994,8 +3987,7 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	if (ret && ffe_ctl->for_treelog)
 		fs_info->treelog_bg = 0;
 	if (ret && ffe_ctl->for_data_reloc)
-		fs_info->data_reloc_bg = 0;
-	spin_unlock(&fs_info->relocation_bg_lock);
+		WRITE_ONCE(fs_info->data_reloc_bg, 0);
 	spin_unlock(&fs_info->treelog_bg_lock);
 	spin_unlock(&block_group->lock);
 	spin_unlock(&space_info->lock);
@@ -4303,11 +4295,9 @@ static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info,
 		if (fs_info->treelog_bg)
 			ffe_ctl->hint_byte = fs_info->treelog_bg;
 		spin_unlock(&fs_info->treelog_bg_lock);
-	} else if (ffe_ctl->for_data_reloc) {
-		spin_lock(&fs_info->relocation_bg_lock);
-		if (fs_info->data_reloc_bg)
-			ffe_ctl->hint_byte = fs_info->data_reloc_bg;
-		spin_unlock(&fs_info->relocation_bg_lock);
+	} else if (ffe_ctl->for_data_reloc &&
+		   READ_ONCE(fs_info->data_reloc_bg)) {
+			ffe_ctl->hint_byte = READ_ONCE(fs_info->data_reloc_bg);
 	} else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
 		struct btrfs_block_group *block_group;
 
fs/btrfs/fs.h +1 -5
index b239e4b8421c..570f4b85096c 100644
@@ -849,11 +849,7 @@ struct btrfs_fs_info {
 	spinlock_t treelog_bg_lock;
 	u64 treelog_bg;
 
-	/*
-	 * Start of the dedicated data relocation block group, protected by
-	 * relocation_bg_lock.
-	 */
-	spinlock_t relocation_bg_lock;
+	/* Start of the dedicated data relocation block group */
 	u64 data_reloc_bg;
 	struct mutex zoned_data_reloc_io_lock;
 
fs/btrfs/zoned.c +4 -6
index bd987c90a05c..421afdb6eb39 100644
@@ -2496,10 +2496,8 @@ void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
 {
 	struct btrfs_fs_info *fs_info = bg->fs_info;
 
-	spin_lock(&fs_info->relocation_bg_lock);
-	if (fs_info->data_reloc_bg == bg->start)
-		fs_info->data_reloc_bg = 0;
-	spin_unlock(&fs_info->relocation_bg_lock);
+	if (READ_ONCE(fs_info->data_reloc_bg) == bg->start)
+		WRITE_ONCE(fs_info->data_reloc_bg, 0);
 }
 
 void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info)
@@ -2518,7 +2516,7 @@ void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info)
 	if (!btrfs_is_zoned(fs_info))
 		return;
 
-	if (fs_info->data_reloc_bg)
+	if (READ_ONCE(fs_info->data_reloc_bg))
 		return;
 
 	if (sb_rdonly(fs_info->sb))
@@ -2539,7 +2537,7 @@ void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info)
 			continue;
 		}
 
-		fs_info->data_reloc_bg = bg->start;
+		WRITE_ONCE(fs_info->data_reloc_bg, bg->start);
 		set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &bg->runtime_flags);
 		btrfs_zone_activate(bg);
 
Signature · 1 line
2.49.0
#4 lore raw

PATCH v2 2/3 btrfs: zoned: get rid of treelog_bg_lock

In reply to Johannes Thumshirn

From: Johannes Thumshirn <[email protected]>

Lockstat analysis of benchmark workloads shows a very high contention on treelog_bg_lock. But the lock only protects a single field in 'struct btrfs_fs_info', namely 'u64 treelog_bg'.

Use READ_ONCE()/WRITE_ONCE() to access 'btrfs_fs_info::treelog_bg'.

---
 fs/btrfs/disk-io.c     |  1 -
 fs/btrfs/extent-tree.c | 49 ++++++++++--------------------------------
 fs/btrfs/fs.h          |  1 -
 fs/btrfs/zoned.c       |  2 +-
 fs/btrfs/zoned.h       |  6 ++----
 5 files changed, 14 insertions(+), 45 deletions(-)
fs/btrfs/disk-io.c +0 -1
index 24896322376d..a54218717cb4 100644
@@ -2789,7 +2789,6 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	spin_lock_init(&fs_info->defrag_inodes_lock);
 	spin_lock_init(&fs_info->super_lock);
 	spin_lock_init(&fs_info->unused_bgs_lock);
-	spin_lock_init(&fs_info->treelog_bg_lock);
 	spin_lock_init(&fs_info->zone_active_bgs_lock);
 	rwlock_init(&fs_info->tree_mod_log_lock);
 	rwlock_init(&fs_info->global_root_lock);
fs/btrfs/extent-tree.c +11 -38

105 lines not on this page · raw message

fs/btrfs/fs.h +0 -1
index 570f4b85096c..a388af40a251 100644
@@ -846,7 +846,6 @@ struct btrfs_fs_info {
 	u64 max_zone_append_size;
 
 	struct mutex zoned_meta_io_lock;
-	spinlock_t treelog_bg_lock;
 	u64 treelog_bg;
 
 	/* Start of the dedicated data relocation block group */
fs/btrfs/zoned.c +1 -1
index 421afdb6eb39..98f483def77d 100644
@@ -1948,7 +1948,7 @@ static bool check_bg_is_active(struct btrfs_eb_write_context *ctx,
 	if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags))
 		return true;
 
-	if (fs_info->treelog_bg == block_group->start) {
+	if (READ_ONCE(fs_info->treelog_bg) == block_group->start) {
 		if (!btrfs_zone_activate(block_group)) {
 			int ret_fin = btrfs_zone_finish_one_bg(fs_info);
 
fs/btrfs/zoned.h +2 -4
index 6e11533b8e14..f71026b6c100 100644
@@ -387,10 +387,8 @@ static inline void btrfs_clear_treelog_bg(struct btrfs_block_group *bg)
 	if (!btrfs_is_zoned(fs_info))
 		return;
 
-	spin_lock(&fs_info->treelog_bg_lock);
-	if (fs_info->treelog_bg == bg->start)
-		fs_info->treelog_bg = 0;
-	spin_unlock(&fs_info->treelog_bg_lock);
+	if (READ_ONCE(fs_info->treelog_bg) == bg->start)
+		WRITE_ONCE(fs_info->treelog_bg, 0);
 }
 
 static inline void btrfs_zoned_data_reloc_lock(struct btrfs_inode *inode)
Signature · 1 line
2.49.0
#5 lore raw

PATCH v2 3/3 btrfs: zoned: don't hold space_info lock on zoned allocation

In reply to Johannes Thumshirn

From: Johannes Thumshirn <[email protected]>

The zoned extent allocator holds 'struct btrfs_space_info::lock' nearly over the entirety of the allocation process, but nothing in do_allocation_zoned() is actually accessing fields of 'struct btrfs_space_info'.

Furthermore taking lock_stat snapshots in performance testing, always shows the space_info::lock as the most contented lock in the entire system.

Remove locking the space_info lock during do_allocation_zoned() to reduce lock contention.

---
 fs/btrfs/extent-tree.c | 3 ---
 1 file changed, 3 deletions(-)
fs/btrfs/extent-tree.c +0 -3
index 39d3984153a2..e8d607877fb4 100644
@@ -3819,7 +3819,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 			       struct btrfs_block_group **bg_ret)
 {
 	struct btrfs_fs_info *fs_info = block_group->fs_info;
-	struct btrfs_space_info *space_info = block_group->space_info;
 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
 	u64 start = block_group->start;
 	u64 num_bytes = ffe_ctl->num_bytes;
@@ -3868,7 +3867,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 		 */
 	}
 
-	spin_lock(&space_info->lock);
 	spin_lock(&block_group->lock);
 
 	if (ret)
@@ -3966,7 +3964,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	if (ret && ffe_ctl->for_data_reloc)
 		WRITE_ONCE(fs_info->data_reloc_bg, 0);
 	spin_unlock(&block_group->lock);
-	spin_unlock(&space_info->lock);
 	return ret;
 }
 
Signature · 1 line
2.49.0
#2 lore raw

Re: [PATCH v2 1/3] btrfs: zoned: get rid of relocation_bg_lock

In reply to Johannes Thumshirn
Quoted 87 lines · Johannes Thumshirn

From: Johannes Thumshirn <[email protected]>

Lockstat analysis of benchmark workloads shows a very high contention on relocation_bg_lock. But the lock only protects a single field in 'struct btrfs_fs_info', namely 'u64 data_reloc_bg'.

Use READ_ONCE()/WRITE_ONCE() to access 'btrfs_fs_info::data_reloc_bg'.

---
 fs/btrfs/disk-io.c     |  1 -
 fs/btrfs/extent-tree.c | 34 ++++++++++++----------------------
 fs/btrfs/fs.h          |  6 +-----
 fs/btrfs/zoned.c       | 10 ++++------
 4 files changed, 17 insertions(+), 34 deletions(-)
fs/btrfs/disk-io.c +0 -1
index ee3cdd7035cc..24896322376d 100644
@@ -2791,7 +2791,6 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	spin_lock_init(&fs_info->unused_bgs_lock);
 	spin_lock_init(&fs_info->treelog_bg_lock);
 	spin_lock_init(&fs_info->zone_active_bgs_lock);
-	spin_lock_init(&fs_info->relocation_bg_lock);
 	rwlock_init(&fs_info->tree_mod_log_lock);
 	rwlock_init(&fs_info->global_root_lock);
 	mutex_init(&fs_info->unused_bg_unpin_mutex);
fs/btrfs/extent-tree.c +6 -13
index 10f50c725313..80ceca89a9ef 100644
@@ -3842,7 +3842,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	u64 avail;
 	u64 bytenr = block_group->start;
 	u64 log_bytenr;
-	u64 data_reloc_bytenr;
 	int ret = 0;
 	bool skip = false;
 
@@ -3865,14 +3864,9 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	 * Do not allow non-relocation blocks in the dedicated relocation block
 	 * group, and vice versa.
 	 */
-	spin_lock(&fs_info->relocation_bg_lock);
-	data_reloc_bytenr = fs_info->data_reloc_bg;
-	if (data_reloc_bytenr &&
-	    ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
-	     (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
-		skip = true;
-	spin_unlock(&fs_info->relocation_bg_lock);
-	if (skip)
+	if (READ_ONCE(fs_info->data_reloc_bg) &&
+	    ((ffe_ctl->for_data_reloc && bytenr != READ_ONCE(fs_info->data_reloc_bg)) ||
+	     (!ffe_ctl->for_data_reloc && bytenr == READ_ONCE(fs_info->data_reloc_bg))))
 		return 1;
 
 	/* Check RO and no space case before trying to activate it */
@@ -3899,7 +3893,6 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	spin_lock(&space_info->lock);
 	spin_lock(&block_group->lock);
 	spin_lock(&fs_info->treelog_bg_lock);
-	spin_lock(&fs_info->relocation_bg_lock);
 
 	if (ret)
 		goto out;
@@ -3908,8 +3901,8 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	       block_group->start == fs_info->treelog_bg ||
 	       fs_info->treelog_bg == 0);
 	ASSERT(!ffe_ctl->for_data_reloc ||
-	       block_group->start == fs_info->data_reloc_bg ||
-	       fs_info->data_reloc_bg == 0);
+	       block_group->start == READ_ONCE(fs_info->data_reloc_bg) ||
+	       READ_ONCE(fs_info->data_reloc_bg) == 0);
 
 	if (block_group->ro ||
 	    (!ffe_ctl->for_data_reloc &&
@@ -3932,7 +3925,7 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	 * Do not allow currently used block group to be the data relocation
 	 * dedicated block group.
 	 */
-	if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
+	if (ffe_ctl->for_data_reloc && READ_ONCE(fs_info->data_reloc_bg) == 0 &&
 	    (block_group->used || block_group->reserved)) {
 		ret = 1;
 		goto out;

fs/btrfs/extent-tree.c static int do_allocation_zoned(struct btrfs_block_group *block_group, · line 3954

Quoted 7 lines
fs/btrfs/extent-tree.c +2 -2
@@ -3957,8 +3950,8 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 		fs_info->treelog_bg = block_group->start;
 
 	if (ffe_ctl->for_data_reloc) {
-		if (!fs_info->data_reloc_bg)
-			fs_info->data_reloc_bg = block_group->start;
+		if (READ_ONCE(fs_info->data_reloc_bg) == 0)
+			WRITE_ONCE(fs_info->data_reloc_bg, block_group->start);

What happens two threads reach here at the same time and each has a different block group locked? One of them will get an unexpected state.

Apparently, it does not happen because we have the space_info->lock around, but patch 3 will remove it...

Quoted 80 lines
fs/btrfs/extent-tree.c +4 -7
 		/*
 		 * Do not allow allocations from this block group, unless it is
 		 * for data relocation. Compared to increasing the ->ro, setting
@@ -3994,8 +3987,7 @@ static int do_allocation_zoned(struct btrfs_block_group *block_group,
 	if (ret && ffe_ctl->for_treelog)
 		fs_info->treelog_bg = 0;
 	if (ret && ffe_ctl->for_data_reloc)
-		fs_info->data_reloc_bg = 0;
-	spin_unlock(&fs_info->relocation_bg_lock);
+		WRITE_ONCE(fs_info->data_reloc_bg, 0);
 	spin_unlock(&fs_info->treelog_bg_lock);
 	spin_unlock(&block_group->lock);
 	spin_unlock(&space_info->lock);
@@ -4303,11 +4295,9 @@ static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info,
 		if (fs_info->treelog_bg)
 			ffe_ctl->hint_byte = fs_info->treelog_bg;
 		spin_unlock(&fs_info->treelog_bg_lock);
-	} else if (ffe_ctl->for_data_reloc) {
-		spin_lock(&fs_info->relocation_bg_lock);
-		if (fs_info->data_reloc_bg)
-			ffe_ctl->hint_byte = fs_info->data_reloc_bg;
-		spin_unlock(&fs_info->relocation_bg_lock);
+	} else if (ffe_ctl->for_data_reloc &&
+		   READ_ONCE(fs_info->data_reloc_bg)) {
+			ffe_ctl->hint_byte = READ_ONCE(fs_info->data_reloc_bg);
 	} else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
 		struct btrfs_block_group *block_group;
 
fs/btrfs/fs.h +1 -5
index b239e4b8421c..570f4b85096c 100644
@@ -849,11 +849,7 @@ struct btrfs_fs_info {
 	spinlock_t treelog_bg_lock;
 	u64 treelog_bg;
 
-	/*
-	 * Start of the dedicated data relocation block group, protected by
-	 * relocation_bg_lock.
-	 */
-	spinlock_t relocation_bg_lock;
+	/* Start of the dedicated data relocation block group */
 	u64 data_reloc_bg;
 	struct mutex zoned_data_reloc_io_lock;
 
fs/btrfs/zoned.c +4 -6
index bd987c90a05c..421afdb6eb39 100644
@@ -2496,10 +2496,8 @@ void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
 {
 	struct btrfs_fs_info *fs_info = bg->fs_info;
 
-	spin_lock(&fs_info->relocation_bg_lock);
-	if (fs_info->data_reloc_bg == bg->start)
-		fs_info->data_reloc_bg = 0;
-	spin_unlock(&fs_info->relocation_bg_lock);
+	if (READ_ONCE(fs_info->data_reloc_bg) == bg->start)
+		WRITE_ONCE(fs_info->data_reloc_bg, 0);
 }
 
 void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info)
@@ -2518,7 +2516,7 @@ void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info)
 	if (!btrfs_is_zoned(fs_info))
 		return;
 
-	if (fs_info->data_reloc_bg)
+	if (READ_ONCE(fs_info->data_reloc_bg))
 		return;
 
 	if (sb_rdonly(fs_info->sb))
@@ -2539,7 +2537,7 @@ void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info)
 			continue;
 		}
 
-		fs_info->data_reloc_bg = bg->start;
+		WRITE_ONCE(fs_info->data_reloc_bg, bg->start);
 		set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &bg->runtime_flags);
 		btrfs_zone_activate(bg);
 
#3 lore raw

Re: [PATCH v2 1/3] btrfs: zoned: get rid of relocation_bg_lock

Quoted 5 lines · Naohiro Aota

What happens two threads reach here at the same time and each has a different block group locked? One of them will get an unexpected state.

Apparently, it does not happen because we have the space_info->lock around, but patch 3 will remove it...

But can this really happen, now that we have dedicated space_infos for relocation and treelog?