[PATCH v5 0/3] btrfs: zoned: automatic BG reclaim

[PATCH v5 0/3] btrfs: zoned: automatic BG reclaim

[Johannes Thumshirn]

When a file gets deleted on a zoned file system, the space freed is not
returned back into the block group's free space, but is migrated to
zone_unusable.

As this zone_unusable space is behind the current write pointer it is not
possible to use it for new allocations. In the current implementation a
zone is reset once all of the block group's space is accounted as zone
unusable.

This behaviour can lead to premature ENOSPC errors on a busy file system.

Instead of only reclaiming the zone once it is completely unusable,
kick off a reclaim job once the amount of unusable bytes exceeds a user
configurable threshold between 51% and 100%. It can be set per mounted
filesystem via the sysfs tunable bg_reclaim_threshold which is set to 75%
per default.

Similar to reclaiming unused block groups, these dirty block groups are
added to a to_reclaim list and then on a transaction commit, the reclaim
process is triggered but after we deleted unused block groups, which will
free space for the relocation process.

Zones that are 100% full and zone unusable already get reclaimed atomatically
on transaction commit. Another improvement on the garbage collection side of
zoned btrfs would be no to reclaim block groups that have used, pinned and
reserved = 0 but zone_unusable > 0. This is not yet included as it needs
further reaserch and testing.

Changes to v4:
- Bail out on unmount (Josef)
- Fix delete extents comment (Filipe)
- Use constant for default threashold (David)
- document reclaim_bgs_mutex (David)

Changes to v3:
- Special case "discarding" after relocation (Filipe)

Changes to v2:
- Fix locking in multiple ways (Filipe)
- Offload reclaim into workqueue (Josef)
- Add patch discarding/zone-resetting after successfull relocation (Anand)

Changes to v1:
- Document sysfs parameter (David)
- Add info print for reclaim (Josef)
- Rename delete_unused_bgs_mutex to reclaim_bgs_lock (Filipe)
- Remove list_is_singular check (Filipe)
- Document of space_info->groups_sem use (Filipe)

Johannes Thumshirn (3):
  btrfs: zoned: reset zones of relocated block groups
  btrfs: rename delete_unused_bgs_mutex
  btrfs: zoned: automatically reclaim zones

 fs/btrfs/block-group.c       | 106 ++++++++++++++++++++++++++++++++++-
 fs/btrfs/block-group.h       |   3 +
 fs/btrfs/ctree.h             |   8 ++-
 fs/btrfs/disk-io.c           |  19 ++++++-
 fs/btrfs/free-space-cache.c  |   9 ++-
 fs/btrfs/sysfs.c             |  35 ++++++++++++
 fs/btrfs/volumes.c           |  63 +++++++++++++--------
 fs/btrfs/volumes.h           |   1 +
 fs/btrfs/zoned.h             |   2 +
 include/trace/events/btrfs.h |  12 ++++
 10 files changed, 226 insertions(+), 32 deletions(-)

-- 
2.30.0

[PATCH v5 1/3] btrfs: zoned: reset zones of relocated block groups

[Johannes Thumshirn]

When relocating a block group the freed up space is not discarded in one
big block, but each extent is discarded on it's own with -odisard=sync.

For a zoned filesystem we need to discard the whole block group at once,
so btrfs_discard_extent() will translate the discard into a
REQ_OP_ZONE_RESET operation, which then resets the device's zone.

Link: https://lore.kernel.org/linux-btrfs/459e2932c48e12e883dcfd3dda828d9da251d5b5.1617962110.git.johannes.thumshirn@wdc.com
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
---
 fs/btrfs/volumes.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 6d9b2369f17a..124c17ec53e9 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -3103,6 +3103,7 @@ static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
 	struct btrfs_root *root = fs_info->chunk_root;
 	struct btrfs_trans_handle *trans;
 	struct btrfs_block_group *block_group;
+	u64 length;
 	int ret;
 
 	/*
@@ -3130,8 +3131,22 @@ static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
 	if (!block_group)
 		return -ENOENT;
 	btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
+	length = block_group->length;
 	btrfs_put_block_group(block_group);
 
+	/*
+	 * On a zoned file system, discard the whole block group, this will
+	 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
+	 * resetting the zone fails, don't treat it as a fatal problem from the
+	 * filesystem's point of view.
+	 */
+	if (btrfs_is_zoned(fs_info)) {
+		ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
+		if (ret)
+			btrfs_info(fs_info, "failed to reset zone %llu",
+				   chunk_offset);
+	}
+
 	trans = btrfs_start_trans_remove_block_group(root->fs_info,
 						     chunk_offset);
 	if (IS_ERR(trans)) {
-- 
2.30.0

[PATCH v5 2/3] btrfs: rename delete_unused_bgs_mutex

[Johannes Thumshirn]

As a preparation for another user, rename the unused_bgs_mutex into
reclaim_bgs_lock.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/block-group.c |  6 +++---
 fs/btrfs/ctree.h       |  3 ++-
 fs/btrfs/disk-io.c     |  6 +++---
 fs/btrfs/volumes.c     | 46 +++++++++++++++++++++---------------------
 4 files changed, 31 insertions(+), 30 deletions(-)

diff --git a/fs/btrfs/block-group.c b/fs/btrfs/block-group.c
index 293f3169be80..bbb5a6e170c7 100644
--- a/fs/btrfs/block-group.c
+++ b/fs/btrfs/block-group.c
@@ -1289,7 +1289,7 @@ void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
 	 * Long running balances can keep us blocked here for eternity, so
 	 * simply skip deletion if we're unable to get the mutex.
 	 */
-	if (!mutex_trylock(&fs_info->delete_unused_bgs_mutex))
+	if (!mutex_trylock(&fs_info->reclaim_bgs_lock))
 		return;
 
 	spin_lock(&fs_info->unused_bgs_lock);
@@ -1462,12 +1462,12 @@ void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
 		spin_lock(&fs_info->unused_bgs_lock);
 	}
 	spin_unlock(&fs_info->unused_bgs_lock);
-	mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+	mutex_unlock(&fs_info->reclaim_bgs_lock);
 	return;
 
 flip_async:
 	btrfs_end_transaction(trans);
-	mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+	mutex_unlock(&fs_info->reclaim_bgs_lock);
 	btrfs_put_block_group(block_group);
 	btrfs_discard_punt_unused_bgs_list(fs_info);
 }
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 2c858d5349c8..68ee130b5a2a 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -957,7 +957,8 @@ struct btrfs_fs_info {
 	spinlock_t unused_bgs_lock;
 	struct list_head unused_bgs;
 	struct mutex unused_bg_unpin_mutex;
-	struct mutex delete_unused_bgs_mutex;
+	/* protects about to be deleted block groups */
+	struct mutex reclaim_bgs_lock;
 
 	/* Cached block sizes */
 	u32 nodesize;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 0a1182694f48..e52b89ad0a61 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1890,10 +1890,10 @@ static int cleaner_kthread(void *arg)
 		btrfs_run_defrag_inodes(fs_info);
 
 		/*
-		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
+		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
 		 * with relocation (btrfs_relocate_chunk) and relocation
 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
-		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
+		 * after acquiring fs_info->reclaim_bgs_lock. So we
 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
 		 * unused block groups.
 		 */
@@ -2876,7 +2876,7 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	spin_lock_init(&fs_info->treelog_bg_lock);
 	rwlock_init(&fs_info->tree_mod_log_lock);
 	mutex_init(&fs_info->unused_bg_unpin_mutex);
-	mutex_init(&fs_info->delete_unused_bgs_mutex);
+	mutex_init(&fs_info->reclaim_bgs_lock);
 	mutex_init(&fs_info->reloc_mutex);
 	mutex_init(&fs_info->delalloc_root_mutex);
 	mutex_init(&fs_info->zoned_meta_io_lock);
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 124c17ec53e9..c6568f61b2b0 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -3118,7 +3118,7 @@ static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
 	 * we release the path used to search the chunk/dev tree and before
 	 * the current task acquires this mutex and calls us.
 	 */
-	lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
+	lockdep_assert_held(&fs_info->reclaim_bgs_lock);
 
 	/* step one, relocate all the extents inside this chunk */
 	btrfs_scrub_pause(fs_info);
@@ -3187,10 +3187,10 @@ static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
 	key.type = BTRFS_CHUNK_ITEM_KEY;
 
 	while (1) {
-		mutex_lock(&fs_info->delete_unused_bgs_mutex);
+		mutex_lock(&fs_info->reclaim_bgs_lock);
 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
 		if (ret < 0) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			goto error;
 		}
 		BUG_ON(ret == 0); /* Corruption */
@@ -3198,7 +3198,7 @@ static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
 					  key.type);
 		if (ret)
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 		if (ret < 0)
 			goto error;
 		if (ret > 0)
@@ -3219,7 +3219,7 @@ static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
 			else
 				BUG_ON(ret);
 		}
-		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+		mutex_unlock(&fs_info->reclaim_bgs_lock);
 
 		if (found_key.offset == 0)
 			break;
@@ -3759,10 +3759,10 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 			goto error;
 		}
 
-		mutex_lock(&fs_info->delete_unused_bgs_mutex);
+		mutex_lock(&fs_info->reclaim_bgs_lock);
 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
 		if (ret < 0) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			goto error;
 		}
 
@@ -3776,7 +3776,7 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 		ret = btrfs_previous_item(chunk_root, path, 0,
 					  BTRFS_CHUNK_ITEM_KEY);
 		if (ret) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			ret = 0;
 			break;
 		}
@@ -3786,7 +3786,7 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
 
 		if (found_key.objectid != key.objectid) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			break;
 		}
 
@@ -3803,12 +3803,12 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 
 		btrfs_release_path(path);
 		if (!ret) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			goto loop;
 		}
 
 		if (counting) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			spin_lock(&fs_info->balance_lock);
 			bctl->stat.expected++;
 			spin_unlock(&fs_info->balance_lock);
@@ -3833,7 +3833,7 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 					count_meta < bctl->meta.limit_min)
 				|| ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
 					count_sys < bctl->sys.limit_min)) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			goto loop;
 		}
 
@@ -3847,7 +3847,7 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 			ret = btrfs_may_alloc_data_chunk(fs_info,
 							 found_key.offset);
 			if (ret < 0) {
-				mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+				mutex_unlock(&fs_info->reclaim_bgs_lock);
 				goto error;
 			} else if (ret == 1) {
 				chunk_reserved = 1;
@@ -3855,7 +3855,7 @@ static int __btrfs_balance(struct btrfs_fs_info *fs_info)
 		}
 
 		ret = btrfs_relocate_chunk(fs_info, found_key.offset);
-		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+		mutex_unlock(&fs_info->reclaim_bgs_lock);
 		if (ret == -ENOSPC) {
 			enospc_errors++;
 		} else if (ret == -ETXTBSY) {
@@ -4740,16 +4740,16 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 	key.type = BTRFS_DEV_EXTENT_KEY;
 
 	do {
-		mutex_lock(&fs_info->delete_unused_bgs_mutex);
+		mutex_lock(&fs_info->reclaim_bgs_lock);
 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 		if (ret < 0) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			goto done;
 		}
 
 		ret = btrfs_previous_item(root, path, 0, key.type);
 		if (ret) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			if (ret < 0)
 				goto done;
 			ret = 0;
@@ -4762,7 +4762,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 
 		if (key.objectid != device->devid) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			btrfs_release_path(path);
 			break;
 		}
@@ -4771,7 +4771,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 		length = btrfs_dev_extent_length(l, dev_extent);
 
 		if (key.offset + length <= new_size) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			btrfs_release_path(path);
 			break;
 		}
@@ -4787,12 +4787,12 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 		 */
 		ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
 		if (ret < 0) {
-			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+			mutex_unlock(&fs_info->reclaim_bgs_lock);
 			goto done;
 		}
 
 		ret = btrfs_relocate_chunk(fs_info, chunk_offset);
-		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+		mutex_unlock(&fs_info->reclaim_bgs_lock);
 		if (ret == -ENOSPC) {
 			failed++;
 		} else if (ret) {
@@ -8018,7 +8018,7 @@ static int relocating_repair_kthread(void *data)
 		return -EBUSY;
 	}
 
-	mutex_lock(&fs_info->delete_unused_bgs_mutex);
+	mutex_lock(&fs_info->reclaim_bgs_lock);
 
 	/* Ensure block group still exists */
 	cache = btrfs_lookup_block_group(fs_info, target);
@@ -8040,7 +8040,7 @@ static int relocating_repair_kthread(void *data)
 out:
 	if (cache)
 		btrfs_put_block_group(cache);
-	mutex_unlock(&fs_info->delete_unused_bgs_mutex);
+	mutex_unlock(&fs_info->reclaim_bgs_lock);
 	btrfs_exclop_finish(fs_info);
 
 	return ret;
-- 
2.30.0

[PATCH v5 3/3] btrfs: zoned: automatically reclaim zones

[Johannes Thumshirn]

When a file gets deleted on a zoned file system, the space freed is not
returned back into the block group's free space, but is migrated to
zone_unusable.

As this zone_unusable space is behind the current write pointer it is not
possible to use it for new allocations. In the current implementation a
zone is reset once all of the block group's space is accounted as zone
unusable.

This behaviour can lead to premature ENOSPC errors on a busy file system.

Instead of only reclaiming the zone once it is completely unusable,
kick off a reclaim job once the amount of unusable bytes exceeds a user
configurable threshold between 51% and 100%. It can be set per mounted
filesystem via the sysfs tunable bg_reclaim_threshold which is set to 75%
per default.

Similar to reclaiming unused block groups, these dirty block groups are
added to a to_reclaim list and then on a transaction commit, the reclaim
process is triggered but after we deleted unused block groups, which will
free space for the relocation process.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
---
 fs/btrfs/block-group.c       | 100 +++++++++++++++++++++++++++++++++++
 fs/btrfs/block-group.h       |   3 ++
 fs/btrfs/ctree.h             |   5 ++
 fs/btrfs/disk-io.c           |  13 +++++
 fs/btrfs/free-space-cache.c  |   9 +++-
 fs/btrfs/sysfs.c             |  35 ++++++++++++
 fs/btrfs/volumes.c           |   2 +-
 fs/btrfs/volumes.h           |   1 +
 fs/btrfs/zoned.h             |   2 +
 include/trace/events/btrfs.h |  12 +++++
 10 files changed, 180 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/block-group.c b/fs/btrfs/block-group.c
index bbb5a6e170c7..327f29739aeb 100644
--- a/fs/btrfs/block-group.c
+++ b/fs/btrfs/block-group.c
@@ -1485,6 +1485,96 @@ void btrfs_mark_bg_unused(struct btrfs_block_group *bg)
 	spin_unlock(&fs_info->unused_bgs_lock);
 }
 
+void btrfs_reclaim_bgs_work(struct work_struct *work)
+{
+	struct btrfs_fs_info *fs_info =
+		container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
+	struct btrfs_block_group *bg;
+	struct btrfs_space_info *space_info;
+	int ret;
+
+	if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
+		return;
+
+	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
+		return;
+
+	mutex_lock(&fs_info->reclaim_bgs_lock);
+	spin_lock(&fs_info->unused_bgs_lock);
+	while (!list_empty(&fs_info->reclaim_bgs)) {
+		bg = list_first_entry(&fs_info->reclaim_bgs,
+				      struct btrfs_block_group,
+				      bg_list);
+		list_del_init(&bg->bg_list);
+
+		space_info = bg->space_info;
+		spin_unlock(&fs_info->unused_bgs_lock);
+
+		/* Don't want to race with allocators so take the groups_sem */
+		down_write(&space_info->groups_sem);
+
+		spin_lock(&bg->lock);
+		if (bg->reserved || bg->pinned || bg->ro) {
+			/*
+			 * We want to bail if we made new allocations or have
+			 * outstanding allocations in this block group.  We do
+			 * the ro check in case balance is currently acting on
+			 * this block group.
+			 */
+			spin_unlock(&bg->lock);
+			up_write(&space_info->groups_sem);
+			goto next;
+		}
+		spin_unlock(&bg->lock);
+
+		/* Get out fast, in case we're unmounting the FS. */
+		if (btrfs_fs_closing(fs_info)) {
+			up_write(&space_info->groups_sem);
+			goto next;
+		}
+
+		ret = inc_block_group_ro(bg, 0);
+		up_write(&space_info->groups_sem);
+		if (ret < 0)
+			goto next;
+
+		btrfs_info(fs_info, "reclaiming chunk %llu", bg->start);
+		trace_btrfs_reclaim_block_group(bg);
+		ret = btrfs_relocate_chunk(fs_info, bg->start);
+		if (ret)
+			btrfs_err(fs_info, "error relocating chunk %llu",
+				  bg->start);
+
+next:
+		btrfs_put_block_group(bg);
+		spin_lock(&fs_info->unused_bgs_lock);
+	}
+	spin_unlock(&fs_info->unused_bgs_lock);
+	mutex_unlock(&fs_info->reclaim_bgs_lock);
+	btrfs_exclop_finish(fs_info);
+}
+
+void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info)
+{
+	spin_lock(&fs_info->unused_bgs_lock);
+	if (!list_empty(&fs_info->reclaim_bgs))
+		queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work);
+	spin_unlock(&fs_info->unused_bgs_lock);
+}
+
+void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg)
+{
+	struct btrfs_fs_info *fs_info = bg->fs_info;
+
+	spin_lock(&fs_info->unused_bgs_lock);
+	if (list_empty(&bg->bg_list)) {
+		btrfs_get_block_group(bg);
+		trace_btrfs_add_reclaim_block_group(bg);
+		list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs);
+	}
+	spin_unlock(&fs_info->unused_bgs_lock);
+}
+
 static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
 			   struct btrfs_path *path)
 {
@@ -3446,6 +3536,16 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
 	}
 	spin_unlock(&info->unused_bgs_lock);
 
+	spin_lock(&info->unused_bgs_lock);
+	while (!list_empty(&info->reclaim_bgs)) {
+		block_group = list_first_entry(&info->reclaim_bgs,
+					       struct btrfs_block_group,
+					       bg_list);
+		list_del_init(&block_group->bg_list);
+		btrfs_put_block_group(block_group);
+	}
+	spin_unlock(&info->unused_bgs_lock);
+
 	spin_lock(&info->block_group_cache_lock);
 	while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
 		block_group = rb_entry(n, struct btrfs_block_group,
diff --git a/fs/btrfs/block-group.h b/fs/btrfs/block-group.h
index 3ecc3372a5ce..7b927425dc71 100644
--- a/fs/btrfs/block-group.h
+++ b/fs/btrfs/block-group.h
@@ -264,6 +264,9 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 			     u64 group_start, struct extent_map *em);
 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info);
 void btrfs_mark_bg_unused(struct btrfs_block_group *bg);
+void btrfs_reclaim_bgs_work(struct work_struct *work);
+void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info);
+void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg);
 int btrfs_read_block_groups(struct btrfs_fs_info *info);
 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
 			   u64 type, u64 chunk_offset, u64 size);
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 68ee130b5a2a..7c72dcd71547 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -954,6 +954,11 @@ struct btrfs_fs_info {
 	struct work_struct async_data_reclaim_work;
 	struct work_struct preempt_reclaim_work;
 
+	/* Used to reclaim partially filled block groups in the background */
+	struct work_struct reclaim_bgs_work;
+	struct list_head reclaim_bgs;
+	int bg_reclaim_threshold;
+
 	spinlock_t unused_bgs_lock;
 	struct list_head unused_bgs;
 	struct mutex unused_bg_unpin_mutex;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index e52b89ad0a61..b7e81c17d3e2 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1898,6 +1898,13 @@ static int cleaner_kthread(void *arg)
 		 * unused block groups.
 		 */
 		btrfs_delete_unused_bgs(fs_info);
+
+		/*
+		 * Reclaim block groups in the reclaim_bgs list after we deleted
+		 * all unused block_groups. This possibly gives us some more free
+		 * space.
+		 */
+		btrfs_reclaim_bgs(fs_info);
 sleep:
 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
 		if (kthread_should_park())
@@ -2886,6 +2893,7 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	INIT_LIST_HEAD(&fs_info->space_info);
 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
 	INIT_LIST_HEAD(&fs_info->unused_bgs);
+	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
 #ifdef CONFIG_BTRFS_DEBUG
 	INIT_LIST_HEAD(&fs_info->allocated_roots);
 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
@@ -2974,6 +2982,9 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	fs_info->swapfile_pins = RB_ROOT;
 
 	fs_info->send_in_progress = 0;
+
+	fs_info->bg_reclaim_threshold = DEFAULT_RECLAIM_THRESH;
+	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
 }
 
 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
@@ -4332,6 +4343,8 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	cancel_work_sync(&fs_info->async_data_reclaim_work);
 	cancel_work_sync(&fs_info->preempt_reclaim_work);
 
+	cancel_work_sync(&fs_info->reclaim_bgs_work);
+
 	/* Cancel or finish ongoing discard work */
 	btrfs_discard_cleanup(fs_info);
 
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 9988decd5717..e54466fc101f 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -11,6 +11,7 @@
 #include <linux/ratelimit.h>
 #include <linux/error-injection.h>
 #include <linux/sched/mm.h>
+#include "misc.h"
 #include "ctree.h"
 #include "free-space-cache.h"
 #include "transaction.h"
@@ -2539,6 +2540,7 @@ int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
 					u64 bytenr, u64 size, bool used)
 {
+	struct btrfs_fs_info *fs_info = block_group->fs_info;
 	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
 	u64 offset = bytenr - block_group->start;
 	u64 to_free, to_unusable;
@@ -2569,8 +2571,13 @@ static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
 	}
 
 	/* All the region is now unusable. Mark it as unused and reclaim */
-	if (block_group->zone_unusable == block_group->length)
+	if (block_group->zone_unusable == block_group->length) {
 		btrfs_mark_bg_unused(block_group);
+	} else if (block_group->zone_unusable >=
+		   div_factor_fine(block_group->length,
+				   fs_info->bg_reclaim_threshold)) {
+		btrfs_mark_bg_to_reclaim(block_group);
+	}
 
 	return 0;
 }
diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
index a99d1f415a7f..436ac7b4b334 100644
--- a/fs/btrfs/sysfs.c
+++ b/fs/btrfs/sysfs.c
@@ -980,6 +980,40 @@ static ssize_t btrfs_read_policy_store(struct kobject *kobj,
 }
 BTRFS_ATTR_RW(, read_policy, btrfs_read_policy_show, btrfs_read_policy_store);
 
+static ssize_t btrfs_bg_reclaim_threshold_show(struct kobject *kobj,
+					       struct kobj_attribute *a,
+					       char *buf)
+{
+	struct btrfs_fs_info *fs_info = to_fs_info(kobj);
+	ssize_t ret;
+
+	ret = scnprintf(buf, PAGE_SIZE, "%d\n", fs_info->bg_reclaim_threshold);
+
+	return ret;
+}
+
+static ssize_t btrfs_bg_reclaim_threshold_store(struct kobject *kobj,
+						struct kobj_attribute *a,
+						const char *buf, size_t len)
+{
+	struct btrfs_fs_info *fs_info = to_fs_info(kobj);
+	int thresh;
+	int ret;
+
+	ret = kstrtoint(buf, 10, &thresh);
+	if (ret)
+		return ret;
+
+	if (thresh <= 50 || thresh > 100)
+		return -EINVAL;
+
+	fs_info->bg_reclaim_threshold = thresh;
+
+	return len;
+}
+BTRFS_ATTR_RW(, bg_reclaim_threshold, btrfs_bg_reclaim_threshold_show,
+	      btrfs_bg_reclaim_threshold_store);
+
 static const struct attribute *btrfs_attrs[] = {
 	BTRFS_ATTR_PTR(, label),
 	BTRFS_ATTR_PTR(, nodesize),
@@ -991,6 +1025,7 @@ static const struct attribute *btrfs_attrs[] = {
 	BTRFS_ATTR_PTR(, exclusive_operation),
 	BTRFS_ATTR_PTR(, generation),
 	BTRFS_ATTR_PTR(, read_policy),
+	BTRFS_ATTR_PTR(, bg_reclaim_threshold),
 	NULL,
 };
 
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index c6568f61b2b0..08541ef5fc39 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -3098,7 +3098,7 @@ int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
 	return ret;
 }
 
-static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
+int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
 {
 	struct btrfs_root *root = fs_info->chunk_root;
 	struct btrfs_trans_handle *trans;
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index d4c3e0dd32b8..9c0d84e5ec06 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -484,6 +484,7 @@ void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf);
 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info);
 int btrfs_recover_balance(struct btrfs_fs_info *fs_info);
 int btrfs_pause_balance(struct btrfs_fs_info *fs_info);
+int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset);
 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info);
 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info);
 int btrfs_uuid_scan_kthread(void *data);
diff --git a/fs/btrfs/zoned.h b/fs/btrfs/zoned.h
index 61e969652fe1..5d74b453b9b9 100644
--- a/fs/btrfs/zoned.h
+++ b/fs/btrfs/zoned.h
@@ -9,6 +9,8 @@
 #include "disk-io.h"
 #include "block-group.h"
 
+#define DEFAULT_RECLAIM_THRESH 75
+
 struct btrfs_zoned_device_info {
 	/*
 	 * Number of zones, zone size and types of zones if bdev is a
diff --git a/include/trace/events/btrfs.h b/include/trace/events/btrfs.h
index 0551ea65374f..a41dd8a0c730 100644
--- a/include/trace/events/btrfs.h
+++ b/include/trace/events/btrfs.h
@@ -1903,6 +1903,18 @@ DEFINE_EVENT(btrfs__block_group, btrfs_add_unused_block_group,
 	TP_ARGS(bg_cache)
 );
 
+DEFINE_EVENT(btrfs__block_group, btrfs_add_reclaim_block_group,
+	TP_PROTO(const struct btrfs_block_group *bg_cache),
+
+	TP_ARGS(bg_cache)
+);
+
+DEFINE_EVENT(btrfs__block_group, btrfs_reclaim_block_group,
+	TP_PROTO(const struct btrfs_block_group *bg_cache),
+
+	TP_ARGS(bg_cache)
+);
+
 DEFINE_EVENT(btrfs__block_group, btrfs_skip_unused_block_group,
 	TP_PROTO(const struct btrfs_block_group *bg_cache),
 
-- 
2.30.0

Re: [PATCH v5 1/3] btrfs: zoned: reset zones of relocated block groups

[David Sterba]

On Mon, Apr 19, 2021 at 04:41:00PM +0900, Johannes Thumshirn wrote:
> When relocating a block group the freed up space is not discarded in one
> big block, but each extent is discarded on it's own with -odisard=sync.
> 
> For a zoned filesystem we need to discard the whole block group at once,
> so btrfs_discard_extent() will translate the discard into a
> REQ_OP_ZONE_RESET operation, which then resets the device's zone.
> 
> Link: https://lore.kernel.org/linux-btrfs/459e2932c48e12e883dcfd3dda828d9da251d5b5.1617962110.git.johannes.thumshirn@wdc.com

The link points to the same patch in v3, is there something missing in
this patch that should be added to the changelog?

Re: [PATCH v5 3/3] btrfs: zoned: automatically reclaim zones

[David Sterba]

On Mon, Apr 19, 2021 at 04:41:02PM +0900, Johannes Thumshirn wrote:
> +void btrfs_reclaim_bgs_work(struct work_struct *work)
> +{
> +	struct btrfs_fs_info *fs_info =
> +		container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
> +	struct btrfs_block_group *bg;
> +	struct btrfs_space_info *space_info;
> +	int ret;
> +
> +	if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
> +		return;
> +
> +	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
> +		return;
> +
> +	mutex_lock(&fs_info->reclaim_bgs_lock);
> +	spin_lock(&fs_info->unused_bgs_lock);
> +	while (!list_empty(&fs_info->reclaim_bgs)) {
> +		bg = list_first_entry(&fs_info->reclaim_bgs,
> +				      struct btrfs_block_group,
> +				      bg_list);
> +		list_del_init(&bg->bg_list);
> +
> +		space_info = bg->space_info;
> +		spin_unlock(&fs_info->unused_bgs_lock);
> +
> +		/* Don't want to race with allocators so take the groups_sem */
> +		down_write(&space_info->groups_sem);
> +
> +		spin_lock(&bg->lock);
> +		if (bg->reserved || bg->pinned || bg->ro) {
> +			/*
> +			 * We want to bail if we made new allocations or have
> +			 * outstanding allocations in this block group.  We do
> +			 * the ro check in case balance is currently acting on
> +			 * this block group.
> +			 */
> +			spin_unlock(&bg->lock);
> +			up_write(&space_info->groups_sem);
> +			goto next;
> +		}
> +		spin_unlock(&bg->lock);
> +
> +		/* Get out fast, in case we're unmounting the FS. */
> +		if (btrfs_fs_closing(fs_info)) {
> +			up_write(&space_info->groups_sem);
> +			goto next;
> +		}
> +
> +		ret = inc_block_group_ro(bg, 0);
> +		up_write(&space_info->groups_sem);
> +		if (ret < 0)
> +			goto next;
> +
> +		btrfs_info(fs_info, "reclaiming chunk %llu", bg->start);

This could state the bg usage ratio, bg->used / bg->length .

> +		trace_btrfs_reclaim_block_group(bg);
> +		ret = btrfs_relocate_chunk(fs_info, bg->start);
> +		if (ret)
> +			btrfs_err(fs_info, "error relocating chunk %llu",
> +				  bg->start);
> +
> +next:
> +		btrfs_put_block_group(bg);
> +		spin_lock(&fs_info->unused_bgs_lock);
> +	}
> +	spin_unlock(&fs_info->unused_bgs_lock);
> +	mutex_unlock(&fs_info->reclaim_bgs_lock);
> +	btrfs_exclop_finish(fs_info);
> +}

Re: [PATCH v5 3/3] btrfs: zoned: automatically reclaim zones

[David Sterba]

On Mon, Apr 19, 2021 at 04:41:02PM +0900, Johannes Thumshirn wrote:
> --- a/fs/btrfs/zoned.h
> +++ b/fs/btrfs/zoned.h
> @@ -9,6 +9,8 @@
>  #include "disk-io.h"
>  #include "block-group.h"
>  
> +#define DEFAULT_RECLAIM_THRESH 75

This is not a .c local define so it needs a prefix and a comment what
the value means so something like

/*
 * Block groups filled more than this value (percents) will be scheduled
 * for background reclaim.
 */
#define BTRFS_DEFAULT_RECLAIM_THRESH		75

Re: [PATCH v5 1/3] btrfs: zoned: reset zones of relocated block groups

[Johannes Thumshirn]

On 19/04/2021 19:13, David Sterba wrote:
> On Mon, Apr 19, 2021 at 04:41:00PM +0900, Johannes Thumshirn wrote:
>> When relocating a block group the freed up space is not discarded in one
>> big block, but each extent is discarded on it's own with -odisard=sync.
>>
>> For a zoned filesystem we need to discard the whole block group at once,
>> so btrfs_discard_extent() will translate the discard into a
>> REQ_OP_ZONE_RESET operation, which then resets the device's zone.
>>
>> Link: https://lore.kernel.org/linux-btrfs/459e2932c48e12e883dcfd3dda828d9da251d5b5.1617962110.git.johannes.thumshirn@wdc.com
> 
> The link points to the same patch in v3, is there something missing in
> this patch that should be added to the changelog?
> 

I included this link so one can look up the discussion between Filipe and
me that lead to this decision. But I guess this is not really relevant. 

Re: [PATCH v5 3/3] btrfs: zoned: automatically reclaim zones

[Johannes Thumshirn]

On 19/04/2021 19:20, David Sterba wrote:
> On Mon, Apr 19, 2021 at 04:41:02PM +0900, Johannes Thumshirn wrote:
>> +void btrfs_reclaim_bgs_work(struct work_struct *work)
>> +{
>> +	struct btrfs_fs_info *fs_info =
>> +		container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
>> +	struct btrfs_block_group *bg;
>> +	struct btrfs_space_info *space_info;
>> +	int ret;
>> +
>> +	if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
>> +		return;
>> +
>> +	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
>> +		return;
>> +
>> +	mutex_lock(&fs_info->reclaim_bgs_lock);
>> +	spin_lock(&fs_info->unused_bgs_lock);
>> +	while (!list_empty(&fs_info->reclaim_bgs)) {
>> +		bg = list_first_entry(&fs_info->reclaim_bgs,
>> +				      struct btrfs_block_group,
>> +				      bg_list);
>> +		list_del_init(&bg->bg_list);
>> +
>> +		space_info = bg->space_info;
>> +		spin_unlock(&fs_info->unused_bgs_lock);
>> +
>> +		/* Don't want to race with allocators so take the groups_sem */
>> +		down_write(&space_info->groups_sem);
>> +
>> +		spin_lock(&bg->lock);
>> +		if (bg->reserved || bg->pinned || bg->ro) {
>> +			/*
>> +			 * We want to bail if we made new allocations or have
>> +			 * outstanding allocations in this block group.  We do
>> +			 * the ro check in case balance is currently acting on
>> +			 * this block group.
>> +			 */
>> +			spin_unlock(&bg->lock);
>> +			up_write(&space_info->groups_sem);
>> +			goto next;
>> +		}
>> +		spin_unlock(&bg->lock);
>> +
>> +		/* Get out fast, in case we're unmounting the FS. */
>> +		if (btrfs_fs_closing(fs_info)) {
>> +			up_write(&space_info->groups_sem);
>> +			goto next;
>> +		}
>> +
>> +		ret = inc_block_group_ro(bg, 0);
>> +		up_write(&space_info->groups_sem);
>> +		if (ret < 0)
>> +			goto next;
>> +
>> +		btrfs_info(fs_info, "reclaiming chunk %llu", bg->start);
> 
> This could state the bg usage ratio, bg->used / bg->length .

OK, will do.

Re: [PATCH v5 3/3] btrfs: zoned: automatically reclaim zones

[Johannes Thumshirn]

On 19/04/2021 19:29, David Sterba wrote:
> On Mon, Apr 19, 2021 at 04:41:02PM +0900, Johannes Thumshirn wrote:
>> --- a/fs/btrfs/zoned.h
>> +++ b/fs/btrfs/zoned.h
>> @@ -9,6 +9,8 @@
>>  #include "disk-io.h"
>>  #include "block-group.h"
>>  
>> +#define DEFAULT_RECLAIM_THRESH 75
> 
> This is not a .c local define so it needs a prefix and a comment what
> the value means so something like
> 
> /*
>  * Block groups filled more than this value (percents) will be scheduled
>  * for background reclaim.
>  */
> #define BTRFS_DEFAULT_RECLAIM_THRESH		75
> 

Of cause, sorry, will do ASAP.

Re: [PATCH v5 1/3] btrfs: zoned: reset zones of relocated block groups

[David Sterba]

On Tue, Apr 20, 2021 at 07:40:37AM +0000, Johannes Thumshirn wrote:
> On 19/04/2021 19:13, David Sterba wrote:
> > On Mon, Apr 19, 2021 at 04:41:00PM +0900, Johannes Thumshirn wrote:
> >> When relocating a block group the freed up space is not discarded in one
> >> big block, but each extent is discarded on it's own with -odisard=sync.
> >>
> >> For a zoned filesystem we need to discard the whole block group at once,
> >> so btrfs_discard_extent() will translate the discard into a
> >> REQ_OP_ZONE_RESET operation, which then resets the device's zone.
> >>
> >> Link: https://lore.kernel.org/linux-btrfs/459e2932c48e12e883dcfd3dda828d9da251d5b5.1617962110.git.johannes.thumshirn@wdc.com
> > 
> > The link points to the same patch in v3, is there something missing in
> > this patch that should be added to the changelog?
> > 
> 
> I included this link so one can look up the discussion between Filipe and
> me that lead to this decision. But I guess this is not really relevant. 

So what'd recommend and expect is to put a summary of the discussion to
the changelog, because the discussions can take many mails and the
contents is scattered, and link the first relevant mail in case the full
story could be interesting. Eventually the link could be in text with
some hint what to look for in the mail.