1 files changed, 40 insertions, 0 deletions

diff --git a/fs/btrfs/free-space-tree.c b/fs/btrfs/free-space-tree.c
index 39c6b96a4c25..b65a20fd519b 100644
--- a/fs/btrfs/free-space-tree.c
+++ b/fs/btrfs/free-space-tree.c
@@ -1217,6 +1217,7 @@ static int clear_free_space_tree(struct btrfs_trans_handle *trans,
 {
 	BTRFS_PATH_AUTO_FREE(path);
 	struct btrfs_key key;
+	struct rb_node *node;
 	int nr;
 	int ret;
 
@@ -1245,6 +1246,16 @@ static int clear_free_space_tree(struct btrfs_trans_handle *trans,
 		btrfs_release_path(path);
 	}
 
+	node = rb_first_cached(&trans->fs_info->block_group_cache_tree);
+	while (node) {
+		struct btrfs_block_group *bg;
+
+		bg = rb_entry(node, struct btrfs_block_group, cache_node);
+		clear_bit(BLOCK_GROUP_FLAG_FREE_SPACE_ADDED, &bg->runtime_flags);
+		node = rb_next(node);
+		cond_resched();
+	}
+
 	return 0;
 }
 
@@ -1334,12 +1345,18 @@ int btrfs_rebuild_free_space_tree(struct btrfs_fs_info *fs_info)
 
 		block_group = rb_entry(node, struct btrfs_block_group,
 				       cache_node);
+
+		if (test_bit(BLOCK_GROUP_FLAG_FREE_SPACE_ADDED,
+			     &block_group->runtime_flags))
+			goto next;
+
 		ret = populate_free_space_tree(trans, block_group);
 		if (ret) {
 			btrfs_abort_transaction(trans, ret);
 			btrfs_end_transaction(trans);
 			return ret;
 		}
+next:
 		if (btrfs_should_end_transaction(trans)) {
 			btrfs_end_transaction(trans);
 			trans = btrfs_start_transaction(free_space_root, 1);
@@ -1366,6 +1383,29 @@ static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
 
 	clear_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags);
 
+	/*
+	 * While rebuilding the free space tree we may allocate new metadata
+	 * block groups while modifying the free space tree.
+	 *
+	 * Because during the rebuild (at btrfs_rebuild_free_space_tree()) we
+	 * can use multiple transactions, every time btrfs_end_transaction() is
+	 * called at btrfs_rebuild_free_space_tree() we finish the creation of
+	 * new block groups by calling btrfs_create_pending_block_groups(), and
+	 * that in turn calls us, through add_block_group_free_space(), to add
+	 * a free space info item and a free space extent item for the block
+	 * group.
+	 *
+	 * Then later btrfs_rebuild_free_space_tree() may find such new block
+	 * groups and processes them with populate_free_space_tree(), which can
+	 * fail with EEXIST since there are already items for the block group in
+	 * the free space tree. Notice that we say "may find" because a new
+	 * block group may be added to the block groups rbtree in a node before
+	 * or after the block group currently being processed by the rebuild
+	 * process. So signal the rebuild process to skip such new block groups
+	 * if it finds them.
+	 */
+	set_bit(BLOCK_GROUP_FLAG_FREE_SPACE_ADDED, &block_group->runtime_flags);
+
 	ret = add_new_free_space_info(trans, block_group, path);
 	if (ret)
 		return ret;