1 files changed, 83 insertions, 32 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 089b46c4d97f..79ac1ebabaf7 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -1003,6 +1003,48 @@ static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+static struct extent_buffer *alloc_tree_block_no_bg_flush(
+					  struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  u64 parent_start,
+					  const struct btrfs_disk_key *disk_key,
+					  int level,
+					  u64 hint,
+					  u64 empty_size)
+{
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct extent_buffer *ret;
+
+	/*
+	 * If we are COWing a node/leaf from the extent, chunk, device or free
+	 * space trees, make sure that we do not finish block group creation of
+	 * pending block groups. We do this to avoid a deadlock.
+	 * COWing can result in allocation of a new chunk, and flushing pending
+	 * block groups (btrfs_create_pending_block_groups()) can be triggered
+	 * when finishing allocation of a new chunk. Creation of a pending block
+	 * group modifies the extent, chunk, device and free space trees,
+	 * therefore we could deadlock with ourselves since we are holding a
+	 * lock on an extent buffer that btrfs_create_pending_block_groups() may
+	 * try to COW later.
+	 * For similar reasons, we also need to delay flushing pending block
+	 * groups when splitting a leaf or node, from one of those trees, since
+	 * we are holding a write lock on it and its parent or when inserting a
+	 * new root node for one of those trees.
+	 */
+	if (root == fs_info->extent_root ||
+	    root == fs_info->chunk_root ||
+	    root == fs_info->dev_root ||
+	    root == fs_info->free_space_root)
+		trans->can_flush_pending_bgs = false;
+
+	ret = btrfs_alloc_tree_block(trans, root, parent_start,
+				     root->root_key.objectid, disk_key, level,
+				     hint, empty_size);
+	trans->can_flush_pending_bgs = true;
+
+	return ret;
+}
+
 /*
  * does the dirty work in cow of a single block.  The parent block (if
  * supplied) is updated to point to the new cow copy.  The new buffer is marked
@@ -1050,26 +1092,8 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
 		parent_start = parent->start;
 
-	/*
-	 * If we are COWing a node/leaf from the extent, chunk or device trees,
-	 * make sure that we do not finish block group creation of pending block
-	 * groups. We do this to avoid a deadlock.
-	 * COWing can result in allocation of a new chunk, and flushing pending
-	 * block groups (btrfs_create_pending_block_groups()) can be triggered
-	 * when finishing allocation of a new chunk. Creation of a pending block
-	 * group modifies the extent, chunk and device trees, therefore we could
-	 * deadlock with ourselves since we are holding a lock on an extent
-	 * buffer that btrfs_create_pending_block_groups() may try to COW later.
-	 */
-	if (root == fs_info->extent_root ||
-	    root == fs_info->chunk_root ||
-	    root == fs_info->dev_root)
-		trans->can_flush_pending_bgs = false;
-
-	cow = btrfs_alloc_tree_block(trans, root, parent_start,
-			root->root_key.objectid, &disk_key, level,
-			search_start, empty_size);
-	trans->can_flush_pending_bgs = true;
+	cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
+					   level, search_start, empty_size);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -2412,6 +2436,16 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
 	if (tmp) {
 		/* first we do an atomic uptodate check */
 		if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
+			/*
+			 * Do extra check for first_key, eb can be stale due to
+			 * being cached, read from scrub, or have multiple
+			 * parents (shared tree blocks).
+			 */
+			if (btrfs_verify_level_key(fs_info, tmp,
+					parent_level - 1, &first_key, gen)) {
+				free_extent_buffer(tmp);
+				return -EUCLEAN;
+			}
 			*eb_ret = tmp;
 			return 0;
 		}
@@ -2624,14 +2658,27 @@ static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
 	root_lock = BTRFS_READ_LOCK;
 
 	if (p->search_commit_root) {
-		/* The commit roots are read only so we always do read locks */
-		if (p->need_commit_sem)
+		/*
+		 * The commit roots are read only so we always do read locks,
+		 * and we always must hold the commit_root_sem when doing
+		 * searches on them, the only exception is send where we don't
+		 * want to block transaction commits for a long time, so
+		 * we need to clone the commit root in order to avoid races
+		 * with transaction commits that create a snapshot of one of
+		 * the roots used by a send operation.
+		 */
+		if (p->need_commit_sem) {
 			down_read(&fs_info->commit_root_sem);
-		b = root->commit_root;
-		extent_buffer_get(b);
-		level = btrfs_header_level(b);
-		if (p->need_commit_sem)
+			b = btrfs_clone_extent_buffer(root->commit_root);
 			up_read(&fs_info->commit_root_sem);
+			if (!b)
+				return ERR_PTR(-ENOMEM);
+
+		} else {
+			b = root->commit_root;
+			extent_buffer_get(b);
+		}
+		level = btrfs_header_level(b);
 		/*
 		 * Ensure that all callers have set skip_locking when
 		 * p->search_commit_root = 1.
@@ -2757,6 +2804,10 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 again:
 	prev_cmp = -1;
 	b = btrfs_search_slot_get_root(root, p, write_lock_level);
+	if (IS_ERR(b)) {
+		ret = PTR_ERR(b);
+		goto done;
+	}
 
 	while (b) {
 		level = btrfs_header_level(b);
@@ -3364,8 +3415,8 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 	else
 		btrfs_node_key(lower, &lower_key, 0);
 
-	c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
-				   &lower_key, level, root->node->start, 0);
+	c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
+					 root->node->start, 0);
 	if (IS_ERR(c))
 		return PTR_ERR(c);
 
@@ -3494,8 +3545,8 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 	mid = (c_nritems + 1) / 2;
 	btrfs_node_key(c, &disk_key, mid);
 
-	split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
-			&disk_key, level, c->start, 0);
+	split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
+					     c->start, 0);
 	if (IS_ERR(split))
 		return PTR_ERR(split);
 
@@ -4279,8 +4330,8 @@ again:
 	else
 		btrfs_item_key(l, &disk_key, mid);
 
-	right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
-			&disk_key, 0, l->start, 0);
+	right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
+					     l->start, 0);
 	if (IS_ERR(right))
 		return PTR_ERR(right);