btrfs: stop reserving excessive space for block group item updates

Space for block group item updates, necessary after allocating or
deallocating an extent from a block group, is reserved in the delayed
refs block reserve. Currently we do this by incrementing the transaction
handle's delayed_ref_updates counter and then calling
btrfs_update_delayed_refs_rsv(), which will increase the size of the
delayed refs block reserve by an amount that corresponds to the same
amount we use for delayed refs, given by btrfs_calc_delayed_ref_bytes().

That is an excessive amount because it corresponds to the amount of space
needed to insert one item in a btree (btrfs_calc_insert_metadata_size())
times 2 when the free space tree feature is enabled. All we need is an
amount as given by btrfs_calc_metadata_size(), since we only need to
update an existing block group item in the extent tree (or block group
tree if this feature is enabled). By using btrfs_calc_metadata_size() we
will need to reserve 4 times less space when using the free space tree
and 2 times less space when not using it, putting less pressure on space
reservation.

So use helpers to reserve and release space for block group item updates
that use btrfs_calc_metadata_size() for calculation of the space.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/block-group.c

@@ -1286,7 +1286,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 	/* Once for the lookup reference */
 	btrfs_put_block_group(block_group);
 	if (remove_rsv)
-		btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
+		btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
 	btrfs_free_path(path);
 	return ret;
 }
@@ -3369,7 +3369,7 @@ int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
 		if (should_put)
 			btrfs_put_block_group(cache);
 		if (drop_reserve)
-			btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
+			btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
 		/*
 		 * Avoid blocking other tasks for too long. It might even save
 		 * us from writing caches for block groups that are going to be
@@ -3516,7 +3516,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
 		/* If its not on the io list, we need to put the block group */
 		if (should_put)
 			btrfs_put_block_group(cache);
-		btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
+		btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
 		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
 	spin_unlock(&cur_trans->dirty_bgs_lock);
@@ -3545,6 +3545,7 @@ int btrfs_update_block_group(struct btrfs_trans_handle *trans,
 	struct btrfs_block_group *cache;
 	u64 old_val;
 	bool reclaim = false;
+	bool bg_already_dirty = true;
 	int factor;
 
 	/* Block accounting for super block */
@@ -3613,7 +3614,7 @@ int btrfs_update_block_group(struct btrfs_trans_handle *trans,
 	spin_lock(&trans->transaction->dirty_bgs_lock);
 	if (list_empty(&cache->dirty_list)) {
 		list_add_tail(&cache->dirty_list, &trans->transaction->dirty_bgs);
-		trans->delayed_ref_updates++;
+		bg_already_dirty = false;
 		btrfs_get_block_group(cache);
 	}
 	spin_unlock(&trans->transaction->dirty_bgs_lock);
@@ -3633,7 +3634,8 @@ int btrfs_update_block_group(struct btrfs_trans_handle *trans,
 	btrfs_put_block_group(cache);
 
 	/* Modified block groups are accounted for in the delayed_refs_rsv. */
-	btrfs_update_delayed_refs_rsv(trans);
+	if (!bg_already_dirty)
+		btrfs_inc_delayed_refs_rsv_bg_updates(info);
 
 	return 0;
 }

fs/btrfs/delayed-ref.c

@@ -125,6 +125,41 @@ void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
 	trans->delayed_ref_csum_deletions = 0;
 }
 
+/*
+ * Adjust the size of the delayed refs block reserve for 1 block group item
+ * update.
+ */
+void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
+
+	spin_lock(&delayed_rsv->lock);
+	/*
+	 * Updating a block group item does not result in new nodes/leaves and
+	 * does not require changing the free space tree, only the extent tree
+	 * or the block group tree, so this is all we need.
+	 */
+	delayed_rsv->size += btrfs_calc_metadata_size(fs_info, 1);
+	delayed_rsv->full = false;
+	spin_unlock(&delayed_rsv->lock);
+}
+
+/*
+ * Adjust the size of the delayed refs block reserve to release space for 1
+ * block group item update.
+ */
+void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
+	const u64 num_bytes = btrfs_calc_metadata_size(fs_info, 1);
+	u64 released;
+
+	released = btrfs_block_rsv_release(fs_info, delayed_rsv, num_bytes, NULL);
+	if (released > 0)
+		trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
+					      0, released, 0);
+}
+
 /*
  * Transfer bytes to our delayed refs rsv.
  *

fs/btrfs/delayed-ref.h

@@ -436,6 +436,8 @@ int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
 
 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
+void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
+void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
 				  enum btrfs_reserve_flush_enum flush);
 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,

fs/btrfs/disk-io.c

@@ -4774,7 +4774,7 @@ void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
 
 		spin_unlock(&cur_trans->dirty_bgs_lock);
 		btrfs_put_block_group(cache);
-		btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
+		btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
 		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
 	spin_unlock(&cur_trans->dirty_bgs_lock);