Merge branch 'integration' into for-linus

fs/btrfs/btrfs_inode.h

@@ -34,6 +34,9 @@ struct btrfs_inode {
 	 */
 	struct btrfs_key location;
 
+	/* Lock for counters */
+	spinlock_t lock;
+
 	/* the extent_tree has caches of all the extent mappings to disk */
 	struct extent_map_tree extent_tree;
 
@@ -134,8 +137,8 @@ struct btrfs_inode {
 	 * items we think we'll end up using, and reserved_extents is the number
 	 * of extent items we've reserved metadata for.
 	 */
-	atomic_t outstanding_extents;
-	atomic_t reserved_extents;
+	unsigned outstanding_extents;
+	unsigned reserved_extents;
 
 	/*
 	 * ordered_data_close is set by truncate when a file that used
@@ -184,4 +187,13 @@ static inline void btrfs_i_size_write(struct inode *inode, u64 size)
 	BTRFS_I(inode)->disk_i_size = size;
 }
 
+static inline bool btrfs_is_free_space_inode(struct btrfs_root *root,
+				       struct inode *inode)
+{
+	if (root == root->fs_info->tree_root ||
+	    BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
+		return true;
+	return false;
+}
+
 #endif

fs/btrfs/ctree.c

@@ -54,8 +54,13 @@ noinline void btrfs_set_path_blocking(struct btrfs_path *p)
 {
 	int i;
 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
-		if (p->nodes[i] && p->locks[i])
-			btrfs_set_lock_blocking(p->nodes[i]);
+		if (!p->nodes[i] || !p->locks[i])
+			continue;
+		btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
+		if (p->locks[i] == BTRFS_READ_LOCK)
+			p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
+		else if (p->locks[i] == BTRFS_WRITE_LOCK)
+			p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
 	}
 }
 
@@ -68,7 +73,7 @@ noinline void btrfs_set_path_blocking(struct btrfs_path *p)
  * for held
  */
 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
-					struct extent_buffer *held)
+					struct extent_buffer *held, int held_rw)
 {
 	int i;
 
@@ -79,19 +84,29 @@ noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
 	 * really sure by forcing the path to blocking before we clear
 	 * the path blocking.
 	 */
-	if (held)
-		btrfs_set_lock_blocking(held);
+	if (held) {
+		btrfs_set_lock_blocking_rw(held, held_rw);
+		if (held_rw == BTRFS_WRITE_LOCK)
+			held_rw = BTRFS_WRITE_LOCK_BLOCKING;
+		else if (held_rw == BTRFS_READ_LOCK)
+			held_rw = BTRFS_READ_LOCK_BLOCKING;
+	}
 	btrfs_set_path_blocking(p);
 #endif
 
 	for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
-		if (p->nodes[i] && p->locks[i])
-			btrfs_clear_lock_blocking(p->nodes[i]);
+		if (p->nodes[i] && p->locks[i]) {
+			btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
+			if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
+				p->locks[i] = BTRFS_WRITE_LOCK;
+			else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
+				p->locks[i] = BTRFS_READ_LOCK;
+		}
 	}
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 	if (held)
-		btrfs_clear_lock_blocking(held);
+		btrfs_clear_lock_blocking_rw(held, held_rw);
 #endif
 }
 
@@ -119,7 +134,7 @@ noinline void btrfs_release_path(struct btrfs_path *p)
 		if (!p->nodes[i])
 			continue;
 		if (p->locks[i]) {
-			btrfs_tree_unlock(p->nodes[i]);
+			btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
 			p->locks[i] = 0;
 		}
 		free_extent_buffer(p->nodes[i]);
@@ -167,6 +182,25 @@ struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
 	return eb;
 }
 
+/* loop around taking references on and locking the root node of the
+ * tree until you end up with a lock on the root.  A locked buffer
+ * is returned, with a reference held.
+ */
+struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+
+	while (1) {
+		eb = btrfs_root_node(root);
+		btrfs_tree_read_lock(eb);
+		if (eb == root->node)
+			break;
+		btrfs_tree_read_unlock(eb);
+		free_extent_buffer(eb);
+	}
+	return eb;
+}
+
 /* cowonly root (everything not a reference counted cow subvolume), just get
  * put onto a simple dirty list.  transaction.c walks this to make sure they
  * get properly updated on disk.
@@ -626,14 +660,6 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 	for (i = start_slot; i < end_slot; i++) {
 		int close = 1;
 
-		if (!parent->map_token) {
-			map_extent_buffer(parent,
-					btrfs_node_key_ptr_offset(i),
-					sizeof(struct btrfs_key_ptr),
-					&parent->map_token, &parent->kaddr,
-					&parent->map_start, &parent->map_len,
-					KM_USER1);
-		}
 		btrfs_node_key(parent, &disk_key, i);
 		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
 			continue;
@@ -656,11 +682,6 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 			last_block = blocknr;
 			continue;
 		}
-		if (parent->map_token) {
-			unmap_extent_buffer(parent, parent->map_token,
-					    KM_USER1);
-			parent->map_token = NULL;
-		}
 
 		cur = btrfs_find_tree_block(root, blocknr, blocksize);
 		if (cur)
@@ -701,11 +722,6 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 		btrfs_tree_unlock(cur);
 		free_extent_buffer(cur);
 	}
-	if (parent->map_token) {
-		unmap_extent_buffer(parent, parent->map_token,
-				    KM_USER1);
-		parent->map_token = NULL;
-	}
 	return err;
 }
 
@@ -746,7 +762,6 @@ static noinline int generic_bin_search(struct extent_buffer *eb,
 	struct btrfs_disk_key *tmp = NULL;
 	struct btrfs_disk_key unaligned;
 	unsigned long offset;
-	char *map_token = NULL;
 	char *kaddr = NULL;
 	unsigned long map_start = 0;
 	unsigned long map_len = 0;
@@ -756,18 +771,13 @@ static noinline int generic_bin_search(struct extent_buffer *eb,
 		mid = (low + high) / 2;
 		offset = p + mid * item_size;
 
-		if (!map_token || offset < map_start ||
+		if (!kaddr || offset < map_start ||
 		    (offset + sizeof(struct btrfs_disk_key)) >
 		    map_start + map_len) {
-			if (map_token) {
-				unmap_extent_buffer(eb, map_token, KM_USER0);
-				map_token = NULL;
-			}
 
 			err = map_private_extent_buffer(eb, offset,
 						sizeof(struct btrfs_disk_key),
-						&map_token, &kaddr,
-						&map_start, &map_len, KM_USER0);
+						&kaddr, &map_start, &map_len);
 
 			if (!err) {
 				tmp = (struct btrfs_disk_key *)(kaddr + offset -
@@ -790,14 +800,10 @@ static noinline int generic_bin_search(struct extent_buffer *eb,
 			high = mid;
 		else {
 			*slot = mid;
-			if (map_token)
-				unmap_extent_buffer(eb, map_token, KM_USER0);
 			return 0;
 		}
 	}
 	*slot = low;
-	if (map_token)
-		unmap_extent_buffer(eb, map_token, KM_USER0);
 	return 1;
 }
 
@@ -890,7 +896,8 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 
 	mid = path->nodes[level];
 
-	WARN_ON(!path->locks[level]);
+	WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
+		path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
 
 	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
@@ -1228,7 +1235,6 @@ static void reada_for_search(struct btrfs_root *root,
 	u32 nr;
 	u32 blocksize;
 	u32 nscan = 0;
-	bool map = true;
 
 	if (level != 1)
 		return;
@@ -1250,19 +1256,8 @@ static void reada_for_search(struct btrfs_root *root,
 
 	nritems = btrfs_header_nritems(node);
 	nr = slot;
-	if (node->map_token || path->skip_locking)
-		map = false;
 
 	while (1) {
-		if (map && !node->map_token) {
-			unsigned long offset = btrfs_node_key_ptr_offset(nr);
-			map_private_extent_buffer(node, offset,
-						  sizeof(struct btrfs_key_ptr),
-						  &node->map_token,
-						  &node->kaddr,
-						  &node->map_start,
-						  &node->map_len, KM_USER1);
-		}
 		if (direction < 0) {
 			if (nr == 0)
 				break;
@@ -1281,11 +1276,6 @@ static void reada_for_search(struct btrfs_root *root,
 		if ((search <= target && target - search <= 65536) ||
 		    (search > target && search - target <= 65536)) {
 			gen = btrfs_node_ptr_generation(node, nr);
-			if (map && node->map_token) {
-				unmap_extent_buffer(node, node->map_token,
-						    KM_USER1);
-				node->map_token = NULL;
-			}
 			readahead_tree_block(root, search, blocksize, gen);
 			nread += blocksize;
 		}
@@ -1293,10 +1283,6 @@ static void reada_for_search(struct btrfs_root *root,
 		if ((nread > 65536 || nscan > 32))
 			break;
 	}
-	if (map && node->map_token) {
-		unmap_extent_buffer(node, node->map_token, KM_USER1);
-		node->map_token = NULL;
-	}
 }
 
 /*
@@ -1409,7 +1395,7 @@ static noinline void unlock_up(struct btrfs_path *path, int level,
 
 		t = path->nodes[i];
 		if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
-			btrfs_tree_unlock(t);
+			btrfs_tree_unlock_rw(t, path->locks[i]);
 			path->locks[i] = 0;
 		}
 	}
@@ -1436,7 +1422,7 @@ noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
 			continue;
 		if (!path->locks[i])
 			continue;
-		btrfs_tree_unlock(path->nodes[i]);
+		btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
 		path->locks[i] = 0;
 	}
 }
@@ -1485,6 +1471,8 @@ read_block_for_search(struct btrfs_trans_handle *trans,
 			 * we can trust our generation number
 			 */
 			free_extent_buffer(tmp);
+			btrfs_set_path_blocking(p);
+
 			tmp = read_tree_block(root, blocknr, blocksize, gen);
 			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
 				*eb_ret = tmp;
@@ -1540,20 +1528,27 @@ read_block_for_search(struct btrfs_trans_handle *trans,
 static int
 setup_nodes_for_search(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, struct btrfs_path *p,
-		       struct extent_buffer *b, int level, int ins_len)
+		       struct extent_buffer *b, int level, int ins_len,
+		       int *write_lock_level)
 {
 	int ret;
 	if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
 	    BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
 		int sret;
 
+		if (*write_lock_level < level + 1) {
+			*write_lock_level = level + 1;
+			btrfs_release_path(p);
+			goto again;
+		}
+
 		sret = reada_for_balance(root, p, level);
 		if (sret)
 			goto again;
 
 		btrfs_set_path_blocking(p);
 		sret = split_node(trans, root, p, level);
-		btrfs_clear_path_blocking(p, NULL);
+		btrfs_clear_path_blocking(p, NULL, 0);
 
 		BUG_ON(sret > 0);
 		if (sret) {
@@ -1565,13 +1560,19 @@ setup_nodes_for_search(struct btrfs_trans_handle *trans,
 		   BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
 		int sret;
 
+		if (*write_lock_level < level + 1) {
+			*write_lock_level = level + 1;
+			btrfs_release_path(p);
+			goto again;
+		}
+
 		sret = reada_for_balance(root, p, level);
 		if (sret)
 			goto again;
 
 		btrfs_set_path_blocking(p);
 		sret = balance_level(trans, root, p, level);
-		btrfs_clear_path_blocking(p, NULL);
+		btrfs_clear_path_blocking(p, NULL, 0);
 
 		if (sret) {
 			ret = sret;
@@ -1615,27 +1616,78 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 	int err;
 	int level;
 	int lowest_unlock = 1;
+	int root_lock;
+	/* everything at write_lock_level or lower must be write locked */
+	int write_lock_level = 0;
 	u8 lowest_level = 0;
 
 	lowest_level = p->lowest_level;
 	WARN_ON(lowest_level && ins_len > 0);
 	WARN_ON(p->nodes[0] != NULL);
 
-	if (ins_len < 0)
+	if (ins_len < 0) {
 		lowest_unlock = 2;
 
+		/* when we are removing items, we might have to go up to level
+		 * two as we update tree pointers  Make sure we keep write
+		 * for those levels as well
+		 */
+		write_lock_level = 2;
+	} else if (ins_len > 0) {
+		/*
+		 * for inserting items, make sure we have a write lock on
+		 * level 1 so we can update keys
+		 */
+		write_lock_level = 1;
+	}
+
+	if (!cow)
+		write_lock_level = -1;
+
+	if (cow && (p->keep_locks || p->lowest_level))
+		write_lock_level = BTRFS_MAX_LEVEL;
+
 again:
+	/*
+	 * we try very hard to do read locks on the root
+	 */
+	root_lock = BTRFS_READ_LOCK;
+	level = 0;
 	if (p->search_commit_root) {
+		/*
+		 * the commit roots are read only
+		 * so we always do read locks
+		 */
 		b = root->commit_root;
 		extent_buffer_get(b);
+		level = btrfs_header_level(b);
 		if (!p->skip_locking)
-			btrfs_tree_lock(b);
+			btrfs_tree_read_lock(b);
 	} else {
-		if (p->skip_locking)
+		if (p->skip_locking) {
 			b = btrfs_root_node(root);
-		else
-			b = btrfs_lock_root_node(root);
+			level = btrfs_header_level(b);
+		} else {
+			/* we don't know the level of the root node
+			 * until we actually have it read locked
+			 */
+			b = btrfs_read_lock_root_node(root);
+			level = btrfs_header_level(b);
+			if (level <= write_lock_level) {
+				/* whoops, must trade for write lock */
+				btrfs_tree_read_unlock(b);
+				free_extent_buffer(b);
+				b = btrfs_lock_root_node(root);
+				root_lock = BTRFS_WRITE_LOCK;
+
+				/* the level might have changed, check again */
+				level = btrfs_header_level(b);
+			}
+		}
 	}
+	p->nodes[level] = b;
+	if (!p->skip_locking)
+		p->locks[level] = root_lock;
 
 	while (b) {
 		level = btrfs_header_level(b);
@@ -1644,10 +1696,6 @@ again:
 		 * setup the path here so we can release it under lock
 		 * contention with the cow code
 		 */
-		p->nodes[level] = b;
-		if (!p->skip_locking)
-			p->locks[level] = 1;
-
 		if (cow) {
 			/*
 			 * if we don't really need to cow this block
@@ -1659,6 +1707,16 @@ again:
 
 			btrfs_set_path_blocking(p);
 
+			/*
+			 * must have write locks on this node and the
+			 * parent
+			 */
+			if (level + 1 > write_lock_level) {
+				write_lock_level = level + 1;
+				btrfs_release_path(p);
+				goto again;
+			}
+
 			err = btrfs_cow_block(trans, root, b,
 					      p->nodes[level + 1],
 					      p->slots[level + 1], &b);
@@ -1671,10 +1729,7 @@ cow_done:
 		BUG_ON(!cow && ins_len);
 
 		p->nodes[level] = b;
-		if (!p->skip_locking)
-			p->locks[level] = 1;
-
-		btrfs_clear_path_blocking(p, NULL);
+		btrfs_clear_path_blocking(p, NULL, 0);
 
 		/*
 		 * we have a lock on b and as long as we aren't changing
@@ -1700,7 +1755,7 @@ cow_done:
 			}
 			p->slots[level] = slot;
 			err = setup_nodes_for_search(trans, root, p, b, level,
-						     ins_len);
+					     ins_len, &write_lock_level);
 			if (err == -EAGAIN)
 				goto again;
 			if (err) {
@@ -1710,6 +1765,19 @@ cow_done:
 			b = p->nodes[level];
 			slot = p->slots[level];
 
+			/*
+			 * slot 0 is special, if we change the key
+			 * we have to update the parent pointer
+			 * which means we must have a write lock
+			 * on the parent
+			 */
+			if (slot == 0 && cow &&
+			    write_lock_level < level + 1) {
+				write_lock_level = level + 1;
+				btrfs_release_path(p);
+				goto again;
+			}
+
 			unlock_up(p, level, lowest_unlock);
 
 			if (level == lowest_level) {
@@ -1728,23 +1796,42 @@ cow_done:
 			}
 
 			if (!p->skip_locking) {
-				btrfs_clear_path_blocking(p, NULL);
-				err = btrfs_try_spin_lock(b);
-
-				if (!err) {
-					btrfs_set_path_blocking(p);
-					btrfs_tree_lock(b);
-					btrfs_clear_path_blocking(p, b);
+				level = btrfs_header_level(b);
+				if (level <= write_lock_level) {
+					err = btrfs_try_tree_write_lock(b);
+					if (!err) {
+						btrfs_set_path_blocking(p);
+						btrfs_tree_lock(b);
+						btrfs_clear_path_blocking(p, b,
+								  BTRFS_WRITE_LOCK);
+					}
+					p->locks[level] = BTRFS_WRITE_LOCK;
+				} else {
+					err = btrfs_try_tree_read_lock(b);
+					if (!err) {
+						btrfs_set_path_blocking(p);
+						btrfs_tree_read_lock(b);
+						btrfs_clear_path_blocking(p, b,
+								  BTRFS_READ_LOCK);
+					}
+					p->locks[level] = BTRFS_READ_LOCK;
 				}
+				p->nodes[level] = b;
 			}
 		} else {
 			p->slots[level] = slot;
 			if (ins_len > 0 &&
 			    btrfs_leaf_free_space(root, b) < ins_len) {
+				if (write_lock_level < 1) {
+					write_lock_level = 1;
+					btrfs_release_path(p);
+					goto again;
+				}
+
 				btrfs_set_path_blocking(p);
 				err = split_leaf(trans, root, key,
 						 p, ins_len, ret == 0);
-				btrfs_clear_path_blocking(p, NULL);
+				btrfs_clear_path_blocking(p, NULL, 0);
 
 				BUG_ON(err > 0);
 				if (err) {
@@ -2025,7 +2112,7 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 	add_root_to_dirty_list(root);
 	extent_buffer_get(c);
 	path->nodes[level] = c;
-	path->locks[level] = 1;
+	path->locks[level] = BTRFS_WRITE_LOCK;
 	path->slots[level] = 0;
 	return 0;
 }
@@ -2253,14 +2340,6 @@ static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
 		if (path->slots[0] == i)
 			push_space += data_size;
 
-		if (!left->map_token) {
-			map_extent_buffer(left, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&left->map_token, &left->kaddr,
-					&left->map_start, &left->map_len,
-					KM_USER1);
-		}
-
 		this_item_size = btrfs_item_size(left, item);
 		if (this_item_size + sizeof(*item) + push_space > free_space)
 			break;
@@ -2271,10 +2350,6 @@ static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
 			break;
 		i--;
 	}
-	if (left->map_token) {
-		unmap_extent_buffer(left, left->map_token, KM_USER1);
-		left->map_token = NULL;
-	}
 
 	if (push_items == 0)
 		goto out_unlock;
@@ -2316,21 +2391,10 @@ static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
 	push_space = BTRFS_LEAF_DATA_SIZE(root);
 	for (i = 0; i < right_nritems; i++) {
 		item = btrfs_item_nr(right, i);
-		if (!right->map_token) {
-			map_extent_buffer(right, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&right->map_token, &right->kaddr,
-					&right->map_start, &right->map_len,
-					KM_USER1);
-		}
 		push_space -= btrfs_item_size(right, item);
 		btrfs_set_item_offset(right, item, push_space);
 	}
 
-	if (right->map_token) {
-		unmap_extent_buffer(right, right->map_token, KM_USER1);
-		right->map_token = NULL;
-	}
 	left_nritems -= push_items;
 	btrfs_set_header_nritems(left, left_nritems);
 
@@ -2467,13 +2531,6 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 
 	for (i = 0; i < nr; i++) {
 		item = btrfs_item_nr(right, i);
-		if (!right->map_token) {
-			map_extent_buffer(right, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&right->map_token, &right->kaddr,
-					&right->map_start, &right->map_len,
-					KM_USER1);
-		}
 
 		if (!empty && push_items > 0) {
 			if (path->slots[0] < i)
@@ -2496,11 +2553,6 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 		push_space += this_item_size + sizeof(*item);
 	}
 
-	if (right->map_token) {
-		unmap_extent_buffer(right, right->map_token, KM_USER1);
-		right->map_token = NULL;
-	}
-
 	if (push_items == 0) {
 		ret = 1;
 		goto out;
@@ -2530,23 +2582,12 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 		u32 ioff;
 
 		item = btrfs_item_nr(left, i);
-		if (!left->map_token) {
-			map_extent_buffer(left, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&left->map_token, &left->kaddr,
-					&left->map_start, &left->map_len,
-					KM_USER1);
-		}
 
 		ioff = btrfs_item_offset(left, item);
 		btrfs_set_item_offset(left, item,
 		      ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
 	}
 	btrfs_set_header_nritems(left, old_left_nritems + push_items);
-	if (left->map_token) {
-		unmap_extent_buffer(left, left->map_token, KM_USER1);
-		left->map_token = NULL;
-	}
 
 	/* fixup right node */
 	if (push_items > right_nritems) {
@@ -2574,21 +2615,9 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 	for (i = 0; i < right_nritems; i++) {
 		item = btrfs_item_nr(right, i);
 
-		if (!right->map_token) {
-			map_extent_buffer(right, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&right->map_token, &right->kaddr,
-					&right->map_start, &right->map_len,
-					KM_USER1);
-		}
-
 		push_space = push_space - btrfs_item_size(right, item);
 		btrfs_set_item_offset(right, item, push_space);
 	}
-	if (right->map_token) {
-		unmap_extent_buffer(right, right->map_token, KM_USER1);
-		right->map_token = NULL;
-	}
 
 	btrfs_mark_buffer_dirty(left);
 	if (right_nritems)
@@ -2729,23 +2758,10 @@ static noinline int copy_for_split(struct btrfs_trans_handle *trans,
 		struct btrfs_item *item = btrfs_item_nr(right, i);
 		u32 ioff;
 
-		if (!right->map_token) {
-			map_extent_buffer(right, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&right->map_token, &right->kaddr,
-					&right->map_start, &right->map_len,
-					KM_USER1);
-		}
-
 		ioff = btrfs_item_offset(right, item);
 		btrfs_set_item_offset(right, item, ioff + rt_data_off);
 	}
 
-	if (right->map_token) {
-		unmap_extent_buffer(right, right->map_token, KM_USER1);
-		right->map_token = NULL;
-	}
-
 	btrfs_set_header_nritems(l, mid);
 	ret = 0;
 	btrfs_item_key(right, &disk_key, 0);
@@ -3264,23 +3280,10 @@ int btrfs_truncate_item(struct btrfs_trans_handle *trans,
 		u32 ioff;
 		item = btrfs_item_nr(leaf, i);
 
-		if (!leaf->map_token) {
-			map_extent_buffer(leaf, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&leaf->map_token, &leaf->kaddr,
-					&leaf->map_start, &leaf->map_len,
-					KM_USER1);
-		}
-
 		ioff = btrfs_item_offset(leaf, item);
 		btrfs_set_item_offset(leaf, item, ioff + size_diff);
 	}
 
-	if (leaf->map_token) {
-		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-		leaf->map_token = NULL;
-	}
-
 	/* shift the data */
 	if (from_end) {
 		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
@@ -3377,22 +3380,10 @@ int btrfs_extend_item(struct btrfs_trans_handle *trans,
 		u32 ioff;
 		item = btrfs_item_nr(leaf, i);
 
-		if (!leaf->map_token) {
-			map_extent_buffer(leaf, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&leaf->map_token, &leaf->kaddr,
-					&leaf->map_start, &leaf->map_len,
-					KM_USER1);
-		}
 		ioff = btrfs_item_offset(leaf, item);
 		btrfs_set_item_offset(leaf, item, ioff - data_size);
 	}
 
-	if (leaf->map_token) {
-		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-		leaf->map_token = NULL;
-	}
-
 	/* shift the data */
 	memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
 		      data_end - data_size, btrfs_leaf_data(leaf) +
@@ -3494,27 +3485,13 @@ int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
 		 */
 		/* first correct the data pointers */
-		WARN_ON(leaf->map_token);
 		for (i = slot; i < nritems; i++) {
 			u32 ioff;
 
 			item = btrfs_item_nr(leaf, i);
-			if (!leaf->map_token) {
-				map_extent_buffer(leaf, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&leaf->map_token, &leaf->kaddr,
-					&leaf->map_start, &leaf->map_len,
-					KM_USER1);
-			}
-
 			ioff = btrfs_item_offset(leaf, item);
 			btrfs_set_item_offset(leaf, item, ioff - total_data);
 		}
-		if (leaf->map_token) {
-			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-			leaf->map_token = NULL;
-		}
-
 		/* shift the items */
 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
 			      btrfs_item_nr_offset(slot),
@@ -3608,27 +3585,13 @@ int setup_items_for_insert(struct btrfs_trans_handle *trans,
 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
 		 */
 		/* first correct the data pointers */
-		WARN_ON(leaf->map_token);
 		for (i = slot; i < nritems; i++) {
 			u32 ioff;
 
 			item = btrfs_item_nr(leaf, i);
-			if (!leaf->map_token) {
-				map_extent_buffer(leaf, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&leaf->map_token, &leaf->kaddr,
-					&leaf->map_start, &leaf->map_len,
-					KM_USER1);
-			}
-
 			ioff = btrfs_item_offset(leaf, item);
 			btrfs_set_item_offset(leaf, item, ioff - total_data);
 		}
-		if (leaf->map_token) {
-			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-			leaf->map_token = NULL;
-		}
-
 		/* shift the items */
 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
 			      btrfs_item_nr_offset(slot),
@@ -3840,22 +3803,10 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 			u32 ioff;
 
 			item = btrfs_item_nr(leaf, i);
-			if (!leaf->map_token) {
-				map_extent_buffer(leaf, (unsigned long)item,
-					sizeof(struct btrfs_item),
-					&leaf->map_token, &leaf->kaddr,
-					&leaf->map_start, &leaf->map_len,
-					KM_USER1);
-			}
 			ioff = btrfs_item_offset(leaf, item);
 			btrfs_set_item_offset(leaf, item, ioff + dsize);
 		}
 
-		if (leaf->map_token) {
-			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-			leaf->map_token = NULL;
-		}
-
 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
 			      btrfs_item_nr_offset(slot + nr),
 			      sizeof(struct btrfs_item) *
@@ -4004,11 +3955,11 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
 
 	WARN_ON(!path->keep_locks);
 again:
-	cur = btrfs_lock_root_node(root);
+	cur = btrfs_read_lock_root_node(root);
 	level = btrfs_header_level(cur);
 	WARN_ON(path->nodes[level]);
 	path->nodes[level] = cur;
-	path->locks[level] = 1;
+	path->locks[level] = BTRFS_READ_LOCK;
 
 	if (btrfs_header_generation(cur) < min_trans) {
 		ret = 1;
@@ -4098,12 +4049,12 @@ find_next_key:
 		cur = read_node_slot(root, cur, slot);
 		BUG_ON(!cur);
 
-		btrfs_tree_lock(cur);
+		btrfs_tree_read_lock(cur);
 
-		path->locks[level - 1] = 1;
+		path->locks[level - 1] = BTRFS_READ_LOCK;
 		path->nodes[level - 1] = cur;
 		unlock_up(path, level, 1);
-		btrfs_clear_path_blocking(path, NULL);
+		btrfs_clear_path_blocking(path, NULL, 0);
 	}
 out:
 	if (ret == 0)
@@ -4218,30 +4169,21 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
 	u32 nritems;
 	int ret;
 	int old_spinning = path->leave_spinning;
-	int force_blocking = 0;
+	int next_rw_lock = 0;
 
 	nritems = btrfs_header_nritems(path->nodes[0]);
 	if (nritems == 0)
 		return 1;
 
-	/*
-	 * we take the blocks in an order that upsets lockdep.  Using
-	 * blocking mode is the only way around it.
-	 */
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-	force_blocking = 1;
-#endif
-
 	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
 again:
 	level = 1;
 	next = NULL;
+	next_rw_lock = 0;
 	btrfs_release_path(path);
 
 	path->keep_locks = 1;
-
-	if (!force_blocking)
-		path->leave_spinning = 1;
+	path->leave_spinning = 1;
 
 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 	path->keep_locks = 0;
@@ -4281,11 +4223,12 @@ again:
 		}
 
 		if (next) {
-			btrfs_tree_unlock(next);
+			btrfs_tree_unlock_rw(next, next_rw_lock);
 			free_extent_buffer(next);
 		}
 
 		next = c;
+		next_rw_lock = path->locks[level];
 		ret = read_block_for_search(NULL, root, path, &next, level,
 					    slot, &key);
 		if (ret == -EAGAIN)
@@ -4297,15 +4240,14 @@ again:
 		}
 
 		if (!path->skip_locking) {
-			ret = btrfs_try_spin_lock(next);
+			ret = btrfs_try_tree_read_lock(next);
 			if (!ret) {
 				btrfs_set_path_blocking(path);
-				btrfs_tree_lock(next);
-				if (!force_blocking)
-					btrfs_clear_path_blocking(path, next);
+				btrfs_tree_read_lock(next);
+				btrfs_clear_path_blocking(path, next,
+							  BTRFS_READ_LOCK);
 			}
-			if (force_blocking)
-				btrfs_set_lock_blocking(next);
+			next_rw_lock = BTRFS_READ_LOCK;
 		}
 		break;
 	}
@@ -4314,14 +4256,13 @@ again:
 		level--;
 		c = path->nodes[level];
 		if (path->locks[level])
-			btrfs_tree_unlock(c);
+			btrfs_tree_unlock_rw(c, path->locks[level]);
 
 		free_extent_buffer(c);
 		path->nodes[level] = next;
 		path->slots[level] = 0;
 		if (!path->skip_locking)
-			path->locks[level] = 1;
-
+			path->locks[level] = next_rw_lock;
 		if (!level)
 			break;
 
@@ -4336,16 +4277,14 @@ again:
 		}
 
 		if (!path->skip_locking) {
-			btrfs_assert_tree_locked(path->nodes[level]);
-			ret = btrfs_try_spin_lock(next);
+			ret = btrfs_try_tree_read_lock(next);
 			if (!ret) {
 				btrfs_set_path_blocking(path);
-				btrfs_tree_lock(next);
-				if (!force_blocking)
-					btrfs_clear_path_blocking(path, next);
+				btrfs_tree_read_lock(next);
+				btrfs_clear_path_blocking(path, next,
+							  BTRFS_READ_LOCK);
 			}
-			if (force_blocking)
-				btrfs_set_lock_blocking(next);
+			next_rw_lock = BTRFS_READ_LOCK;
 		}
 	}
 	ret = 0;

fs/btrfs/ctree.h

@@ -755,6 +755,8 @@ struct btrfs_space_info {
 				   chunks for this space */
 	unsigned int chunk_alloc:1;	/* set if we are allocating a chunk */
 
+	unsigned int flush:1;		/* set if we are trying to make space */
+
 	unsigned int force_alloc;	/* set if we need to force a chunk
 					   alloc for this space */
 
@@ -764,7 +766,7 @@ struct btrfs_space_info {
 	struct list_head block_groups[BTRFS_NR_RAID_TYPES];
 	spinlock_t lock;
 	struct rw_semaphore groups_sem;
-	atomic_t caching_threads;
+	wait_queue_head_t wait;
 };
 
 struct btrfs_block_rsv {
@@ -824,6 +826,7 @@ struct btrfs_caching_control {
 	struct list_head list;
 	struct mutex mutex;
 	wait_queue_head_t wait;
+	struct btrfs_work work;
 	struct btrfs_block_group_cache *block_group;
 	u64 progress;
 	atomic_t count;
@@ -1032,6 +1035,8 @@ struct btrfs_fs_info {
 	struct btrfs_workers endio_write_workers;
 	struct btrfs_workers endio_freespace_worker;
 	struct btrfs_workers submit_workers;
+	struct btrfs_workers caching_workers;
+
 	/*
 	 * fixup workers take dirty pages that didn't properly go through
 	 * the cow mechanism and make them safe to write.  It happens
@@ -2128,7 +2133,7 @@ static inline bool btrfs_mixed_space_info(struct btrfs_space_info *space_info)
 
 /* extent-tree.c */
 static inline u64 btrfs_calc_trans_metadata_size(struct btrfs_root *root,
-						 int num_items)
+						 unsigned num_items)
 {
 	return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
 		3 * num_items;
@@ -2222,9 +2227,6 @@ void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *ionde);
 void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
 int btrfs_check_data_free_space(struct inode *inode, u64 bytes);
 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes);
-int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root,
-				int num_items);
 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root);
 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
@@ -2330,7 +2332,7 @@ struct btrfs_path *btrfs_alloc_path(void);
 void btrfs_free_path(struct btrfs_path *p);
 void btrfs_set_path_blocking(struct btrfs_path *p);
 void btrfs_clear_path_blocking(struct btrfs_path *p,
-			       struct extent_buffer *held);
+			       struct extent_buffer *held, int held_rw);
 void btrfs_unlock_up_safe(struct btrfs_path *p, int level);
 
 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,

fs/btrfs/delayed-inode.c

@@ -735,7 +735,7 @@ static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
 	}
 
 	/* reset all the locked nodes in the patch to spinning locks. */
-	btrfs_clear_path_blocking(path, NULL);
+	btrfs_clear_path_blocking(path, NULL, 0);
 
 	/* insert the keys of the items */
 	ret = setup_items_for_insert(trans, root, path, keys, data_size,

fs/btrfs/dir-item.c

@@ -89,13 +89,8 @@ int btrfs_insert_xattr_item(struct btrfs_trans_handle *trans,
 	data_size = sizeof(*dir_item) + name_len + data_len;
 	dir_item = insert_with_overflow(trans, root, path, &key, data_size,
 					name, name_len);
-	/*
-	 * FIXME: at some point we should handle xattr's that are larger than
-	 * what we can fit in our leaf.  We set location to NULL b/c we arent
-	 * pointing at anything else, that will change if we store the xattr
-	 * data in a separate inode.
-	 */
-	BUG_ON(IS_ERR(dir_item));
+	if (IS_ERR(dir_item))
+		return PTR_ERR(dir_item);
 	memset(&location, 0, sizeof(location));
 
 	leaf = path->nodes[0];

fs/btrfs/disk-io.c

@@ -100,38 +100,83 @@ struct async_submit_bio {
 	struct btrfs_work work;
 };
 
-/* These are used to set the lockdep class on the extent buffer locks.
- * The class is set by the readpage_end_io_hook after the buffer has
- * passed csum validation but before the pages are unlocked.
+/*
+ * Lockdep class keys for extent_buffer->lock's in this root.  For a given
+ * eb, the lockdep key is determined by the btrfs_root it belongs to and
+ * the level the eb occupies in the tree.
+ *
+ * Different roots are used for different purposes and may nest inside each
+ * other and they require separate keysets.  As lockdep keys should be
+ * static, assign keysets according to the purpose of the root as indicated
+ * by btrfs_root->objectid.  This ensures that all special purpose roots
+ * have separate keysets.
  *
- * The lockdep class is also set by btrfs_init_new_buffer on freshly
- * allocated blocks.
+ * Lock-nesting across peer nodes is always done with the immediate parent
+ * node locked thus preventing deadlock.  As lockdep doesn't know this, use
+ * subclass to avoid triggering lockdep warning in such cases.
  *
- * The class is based on the level in the tree block, which allows lockdep
- * to know that lower nodes nest inside the locks of higher nodes.
+ * The key is set by the readpage_end_io_hook after the buffer has passed
+ * csum validation but before the pages are unlocked.  It is also set by
+ * btrfs_init_new_buffer on freshly allocated blocks.
  *
- * We also add a check to make sure the highest level of the tree is
- * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
- * code needs update as well.
+ * We also add a check to make sure the highest level of the tree is the
+ * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
+ * needs update as well.
  */
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 # if BTRFS_MAX_LEVEL != 8
 #  error
 # endif
-static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
-static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
-	/* leaf */
-	"btrfs-extent-00",
-	"btrfs-extent-01",
-	"btrfs-extent-02",
-	"btrfs-extent-03",
-	"btrfs-extent-04",
-	"btrfs-extent-05",
-	"btrfs-extent-06",
-	"btrfs-extent-07",
-	/* highest possible level */
-	"btrfs-extent-08",
+
+static struct btrfs_lockdep_keyset {
+	u64			id;		/* root objectid */
+	const char		*name_stem;	/* lock name stem */
+	char			names[BTRFS_MAX_LEVEL + 1][20];
+	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
+} btrfs_lockdep_keysets[] = {
+	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
+	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
+	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
+	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
+	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
+	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
+	{ .id = BTRFS_ORPHAN_OBJECTID,		.name_stem = "orphan"	},
+	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
+	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
+	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
+	{ .id = 0,				.name_stem = "tree"	},
 };
+
+void __init btrfs_init_lockdep(void)
+{
+	int i, j;
+
+	/* initialize lockdep class names */
+	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
+		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
+
+		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
+			snprintf(ks->names[j], sizeof(ks->names[j]),
+				 "btrfs-%s-%02d", ks->name_stem, j);
+	}
+}
+
+void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
+				    int level)
+{
+	struct btrfs_lockdep_keyset *ks;
+
+	BUG_ON(level >= ARRAY_SIZE(ks->keys));
+
+	/* find the matching keyset, id 0 is the default entry */
+	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
+		if (ks->id == objectid)
+			break;
+
+	lockdep_set_class_and_name(&eb->lock,
+				   &ks->keys[level], ks->names[level]);
+}
+
 #endif
 
 /*
@@ -217,7 +262,6 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 	unsigned long len;
 	unsigned long cur_len;
 	unsigned long offset = BTRFS_CSUM_SIZE;
-	char *map_token = NULL;
 	char *kaddr;
 	unsigned long map_start;
 	unsigned long map_len;
@@ -228,8 +272,7 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 	len = buf->len - offset;
 	while (len > 0) {
 		err = map_private_extent_buffer(buf, offset, 32,
-					&map_token, &kaddr,
-					&map_start, &map_len, KM_USER0);
+					&kaddr, &map_start, &map_len);
 		if (err)
 			return 1;
 		cur_len = min(len, map_len - (offset - map_start));
@@ -237,7 +280,6 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 				      crc, cur_len);
 		len -= cur_len;
 		offset += cur_len;
-		unmap_extent_buffer(buf, map_token, KM_USER0);
 	}
 	if (csum_size > sizeof(inline_result)) {
 		result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
@@ -494,15 +536,6 @@ static noinline int check_leaf(struct btrfs_root *root,
 	return 0;
 }
 
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
-{
-	lockdep_set_class_and_name(&eb->lock,
-			   &btrfs_eb_class[level],
-			   btrfs_eb_name[level]);
-}
-#endif
-
 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 			       struct extent_state *state)
 {
@@ -553,7 +586,8 @@ static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 	}
 	found_level = btrfs_header_level(eb);
 
-	btrfs_set_buffer_lockdep_class(eb, found_level);
+	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
+				       eb, found_level);
 
 	ret = csum_tree_block(root, eb, 1);
 	if (ret) {
@@ -1603,7 +1637,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 		goto fail_bdi;
 	}
 
-	fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
+	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 
 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
 	INIT_LIST_HEAD(&fs_info->trans_list);
@@ -1807,6 +1841,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 			   fs_info->thread_pool_size),
 			   &fs_info->generic_worker);
 
+	btrfs_init_workers(&fs_info->caching_workers, "cache",
+			   2, &fs_info->generic_worker);
+
 	/* a higher idle thresh on the submit workers makes it much more
 	 * likely that bios will be send down in a sane order to the
 	 * devices
@@ -1860,6 +1897,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	btrfs_start_workers(&fs_info->endio_write_workers, 1);
 	btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
 	btrfs_start_workers(&fs_info->delayed_workers, 1);
+	btrfs_start_workers(&fs_info->caching_workers, 1);
 
 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -2117,6 +2155,7 @@ fail_sb_buffer:
 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
 	btrfs_stop_workers(&fs_info->submit_workers);
 	btrfs_stop_workers(&fs_info->delayed_workers);
+	btrfs_stop_workers(&fs_info->caching_workers);
 fail_alloc:
 	kfree(fs_info->delayed_root);
 fail_iput:
@@ -2584,6 +2623,7 @@ int close_ctree(struct btrfs_root *root)
 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
 	btrfs_stop_workers(&fs_info->submit_workers);
 	btrfs_stop_workers(&fs_info->delayed_workers);
+	btrfs_stop_workers(&fs_info->caching_workers);
 
 	btrfs_close_devices(fs_info->fs_devices);
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);

fs/btrfs/disk-io.h

@@ -87,10 +87,14 @@ int btree_lock_page_hook(struct page *page);
 
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
-void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level);
+void btrfs_init_lockdep(void);
+void btrfs_set_buffer_lockdep_class(u64 objectid,
+			            struct extent_buffer *eb, int level);
 #else
-static inline void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb,
-						 int level)
+static inline void btrfs_init_lockdep(void)
+{ }
+static inline void btrfs_set_buffer_lockdep_class(u64 objectid,
+					struct extent_buffer *eb, int level)
 {
 }
 #endif

fs/btrfs/extent-tree.c

@@ -320,12 +320,12 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 	return total_added;
 }
 
-static int caching_kthread(void *data)
+static noinline void caching_thread(struct btrfs_work *work)
 {
-	struct btrfs_block_group_cache *block_group = data;
-	struct btrfs_fs_info *fs_info = block_group->fs_info;
-	struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
-	struct btrfs_root *extent_root = fs_info->extent_root;
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_caching_control *caching_ctl;
+	struct btrfs_root *extent_root;
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
 	struct btrfs_key key;
@@ -334,9 +334,14 @@ static int caching_kthread(void *data)
 	u32 nritems;
 	int ret = 0;
 
+	caching_ctl = container_of(work, struct btrfs_caching_control, work);
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+	extent_root = fs_info->extent_root;
+
 	path = btrfs_alloc_path();
 	if (!path)
-		return -ENOMEM;
+		goto out;
 
 	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 
@@ -433,13 +438,11 @@ err:
 	free_excluded_extents(extent_root, block_group);
 
 	mutex_unlock(&caching_ctl->mutex);
+out:
 	wake_up(&caching_ctl->wait);
 
 	put_caching_control(caching_ctl);
-	atomic_dec(&block_group->space_info->caching_threads);
 	btrfs_put_block_group(block_group);
-
-	return 0;
 }
 
 static int cache_block_group(struct btrfs_block_group_cache *cache,
@@ -449,7 +452,6 @@ static int cache_block_group(struct btrfs_block_group_cache *cache,
 {
 	struct btrfs_fs_info *fs_info = cache->fs_info;
 	struct btrfs_caching_control *caching_ctl;
-	struct task_struct *tsk;
 	int ret = 0;
 
 	smp_mb();
@@ -501,6 +503,7 @@ static int cache_block_group(struct btrfs_block_group_cache *cache,
 	caching_ctl->progress = cache->key.objectid;
 	/* one for caching kthread, one for caching block group list */
 	atomic_set(&caching_ctl->count, 2);
+	caching_ctl->work.func = caching_thread;
 
 	spin_lock(&cache->lock);
 	if (cache->cached != BTRFS_CACHE_NO) {
@@ -516,16 +519,9 @@ static int cache_block_group(struct btrfs_block_group_cache *cache,
 	list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 	up_write(&fs_info->extent_commit_sem);
 
-	atomic_inc(&cache->space_info->caching_threads);
 	btrfs_get_block_group(cache);
 
-	tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
-			  cache->key.objectid);
-	if (IS_ERR(tsk)) {
-		ret = PTR_ERR(tsk);
-		printk(KERN_ERR "error running thread %d\n", ret);
-		BUG();
-	}
+	btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 
 	return ret;
 }
@@ -2932,9 +2928,10 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 	found->full = 0;
 	found->force_alloc = CHUNK_ALLOC_NO_FORCE;
 	found->chunk_alloc = 0;
+	found->flush = 0;
+	init_waitqueue_head(&found->wait);
 	*space_info = found;
 	list_add_rcu(&found->list, &info->space_info);
-	atomic_set(&found->caching_threads, 0);
 	return 0;
 }
 
@@ -3314,6 +3311,14 @@ static int shrink_delalloc(struct btrfs_trans_handle *trans,
 	if (reserved == 0)
 		return 0;
 
+	smp_mb();
+	if (root->fs_info->delalloc_bytes == 0) {
+		if (trans)
+			return 0;
+		btrfs_wait_ordered_extents(root, 0, 0);
+		return 0;
+	}
+
 	max_reclaim = min(reserved, to_reclaim);
 
 	while (loops < 1024) {
@@ -3356,6 +3361,8 @@ static int shrink_delalloc(struct btrfs_trans_handle *trans,
 		}
 
 	}
+	if (reclaimed >= to_reclaim && !trans)
+		btrfs_wait_ordered_extents(root, 0, 0);
 	return reclaimed >= to_reclaim;
 }
 
@@ -3380,15 +3387,36 @@ static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
 	u64 num_bytes = orig_bytes;
 	int retries = 0;
 	int ret = 0;
-	bool reserved = false;
 	bool committed = false;
+	bool flushing = false;
 
 again:
-	ret = -ENOSPC;
-	if (reserved)
-		num_bytes = 0;
-
+	ret = 0;
 	spin_lock(&space_info->lock);
+	/*
+	 * We only want to wait if somebody other than us is flushing and we are
+	 * actually alloed to flush.
+	 */
+	while (flush && !flushing && space_info->flush) {
+		spin_unlock(&space_info->lock);
+		/*
+		 * If we have a trans handle we can't wait because the flusher
+		 * may have to commit the transaction, which would mean we would
+		 * deadlock since we are waiting for the flusher to finish, but
+		 * hold the current transaction open.
+		 */
+		if (trans)
+			return -EAGAIN;
+		ret = wait_event_interruptible(space_info->wait,
+					       !space_info->flush);
+		/* Must have been interrupted, return */
+		if (ret)
+			return -EINTR;
+
+		spin_lock(&space_info->lock);
+	}
+
+	ret = -ENOSPC;
 	unused = space_info->bytes_used + space_info->bytes_reserved +
 		 space_info->bytes_pinned + space_info->bytes_readonly +
 		 space_info->bytes_may_use;
@@ -3403,8 +3431,7 @@ again:
 	if (unused <= space_info->total_bytes) {
 		unused = space_info->total_bytes - unused;
 		if (unused >= num_bytes) {
-			if (!reserved)
-				space_info->bytes_reserved += orig_bytes;
+			space_info->bytes_reserved += orig_bytes;
 			ret = 0;
 		} else {
 			/*
@@ -3429,17 +3456,14 @@ again:
 	 * to reclaim space we can actually use it instead of somebody else
 	 * stealing it from us.
 	 */
-	if (ret && !reserved) {
-		space_info->bytes_reserved += orig_bytes;
-		reserved = true;
+	if (ret && flush) {
+		flushing = true;
+		space_info->flush = 1;
 	}
 
 	spin_unlock(&space_info->lock);
 
-	if (!ret)
-		return 0;
-
-	if (!flush)
+	if (!ret || !flush)
 		goto out;
 
 	/*
@@ -3447,11 +3471,11 @@ again:
 	 * metadata until after the IO is completed.
 	 */
 	ret = shrink_delalloc(trans, root, num_bytes, 1);
-	if (ret > 0)
-		return 0;
-	else if (ret < 0)
+	if (ret < 0)
 		goto out;
 
+	ret = 0;
+
 	/*
 	 * So if we were overcommitted it's possible that somebody else flushed
 	 * out enough space and we simply didn't have enough space to reclaim,
@@ -3462,11 +3486,11 @@ again:
 		goto again;
 	}
 
-	spin_lock(&space_info->lock);
 	/*
 	 * Not enough space to be reclaimed, don't bother committing the
 	 * transaction.
 	 */
+	spin_lock(&space_info->lock);
 	if (space_info->bytes_pinned < orig_bytes)
 		ret = -ENOSPC;
 	spin_unlock(&space_info->lock);
@@ -3474,10 +3498,13 @@ again:
 		goto out;
 
 	ret = -EAGAIN;
-	if (trans || committed)
+	if (trans)
 		goto out;
 
 	ret = -ENOSPC;
+	if (committed)
+		goto out;
+
 	trans = btrfs_join_transaction(root);
 	if (IS_ERR(trans))
 		goto out;
@@ -3489,12 +3516,12 @@ again:
 	}
 
 out:
-	if (reserved) {
+	if (flushing) {
 		spin_lock(&space_info->lock);
-		space_info->bytes_reserved -= orig_bytes;
+		space_info->flush = 0;
+		wake_up_all(&space_info->wait);
 		spin_unlock(&space_info->lock);
 	}
-
 	return ret;
 }
 
@@ -3704,7 +3731,6 @@ int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
 	if (commit_trans) {
 		if (trans)
 			return -EAGAIN;
-
 		trans = btrfs_join_transaction(root);
 		BUG_ON(IS_ERR(trans));
 		ret = btrfs_commit_transaction(trans, root);
@@ -3874,26 +3900,6 @@ int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
-int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
-				 struct btrfs_root *root,
-				 int num_items)
-{
-	u64 num_bytes;
-	int ret;
-
-	if (num_items == 0 || root->fs_info->chunk_root == root)
-		return 0;
-
-	num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
-	ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
-				  num_bytes);
-	if (!ret) {
-		trans->bytes_reserved += num_bytes;
-		trans->block_rsv = &root->fs_info->trans_block_rsv;
-	}
-	return ret;
-}
-
 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
 				  struct btrfs_root *root)
 {
@@ -3944,6 +3950,30 @@ int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
 	return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
 }
 
+static unsigned drop_outstanding_extent(struct inode *inode)
+{
+	unsigned dropped_extents = 0;
+
+	spin_lock(&BTRFS_I(inode)->lock);
+	BUG_ON(!BTRFS_I(inode)->outstanding_extents);
+	BTRFS_I(inode)->outstanding_extents--;
+
+	/*
+	 * If we have more or the same amount of outsanding extents than we have
+	 * reserved then we need to leave the reserved extents count alone.
+	 */
+	if (BTRFS_I(inode)->outstanding_extents >=
+	    BTRFS_I(inode)->reserved_extents)
+		goto out;
+
+	dropped_extents = BTRFS_I(inode)->reserved_extents -
+		BTRFS_I(inode)->outstanding_extents;
+	BTRFS_I(inode)->reserved_extents -= dropped_extents;
+out:
+	spin_unlock(&BTRFS_I(inode)->lock);
+	return dropped_extents;
+}
+
 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
 {
 	return num_bytes >>= 3;
@@ -3953,9 +3983,8 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
-	u64 to_reserve;
-	int nr_extents;
-	int reserved_extents;
+	u64 to_reserve = 0;
+	unsigned nr_extents = 0;
 	int ret;
 
 	if (btrfs_transaction_in_commit(root->fs_info))
@@ -3963,66 +3992,49 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
 
 	num_bytes = ALIGN(num_bytes, root->sectorsize);
 
-	nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
-	reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
+	spin_lock(&BTRFS_I(inode)->lock);
+	BTRFS_I(inode)->outstanding_extents++;
+
+	if (BTRFS_I(inode)->outstanding_extents >
+	    BTRFS_I(inode)->reserved_extents) {
+		nr_extents = BTRFS_I(inode)->outstanding_extents -
+			BTRFS_I(inode)->reserved_extents;
+		BTRFS_I(inode)->reserved_extents += nr_extents;
 
-	if (nr_extents > reserved_extents) {
-		nr_extents -= reserved_extents;
 		to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
-	} else {
-		nr_extents = 0;
-		to_reserve = 0;
 	}
+	spin_unlock(&BTRFS_I(inode)->lock);
 
 	to_reserve += calc_csum_metadata_size(inode, num_bytes);
 	ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
-	if (ret)
+	if (ret) {
+		unsigned dropped;
+		/*
+		 * We don't need the return value since our reservation failed,
+		 * we just need to clean up our counter.
+		 */
+		dropped = drop_outstanding_extent(inode);
+		WARN_ON(dropped > 1);
 		return ret;
-
-	atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
-	atomic_inc(&BTRFS_I(inode)->outstanding_extents);
+	}
 
 	block_rsv_add_bytes(block_rsv, to_reserve, 1);
 
-	if (block_rsv->size > 512 * 1024 * 1024)
-		shrink_delalloc(NULL, root, to_reserve, 0);
-
 	return 0;
 }
 
 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
-	u64 to_free;
-	int nr_extents;
-	int reserved_extents;
+	u64 to_free = 0;
+	unsigned dropped;
 
 	num_bytes = ALIGN(num_bytes, root->sectorsize);
-	atomic_dec(&BTRFS_I(inode)->outstanding_extents);
-	WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
-
-	reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
-	do {
-		int old, new;
-
-		nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
-		if (nr_extents >= reserved_extents) {
-			nr_extents = 0;
-			break;
-		}
-		old = reserved_extents;
-		nr_extents = reserved_extents - nr_extents;
-		new = reserved_extents - nr_extents;
-		old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
-				     reserved_extents, new);
-		if (likely(old == reserved_extents))
-			break;
-		reserved_extents = old;
-	} while (1);
+	dropped = drop_outstanding_extent(inode);
 
 	to_free = calc_csum_metadata_size(inode, num_bytes);
-	if (nr_extents > 0)
-		to_free += btrfs_calc_trans_metadata_size(root, nr_extents);
+	if (dropped > 0)
+		to_free += btrfs_calc_trans_metadata_size(root, dropped);
 
 	btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
 				to_free);
@@ -4990,14 +5002,10 @@ have_block_group:
 			}
 
 			/*
-			 * We only want to start kthread caching if we are at
-			 * the point where we will wait for caching to make
-			 * progress, or if our ideal search is over and we've
-			 * found somebody to start caching.
+			 * The caching workers are limited to 2 threads, so we
+			 * can queue as much work as we care to.
 			 */
-			if (loop > LOOP_CACHING_NOWAIT ||
-			    (loop > LOOP_FIND_IDEAL &&
-			     atomic_read(&space_info->caching_threads) < 2)) {
+			if (loop > LOOP_FIND_IDEAL) {
 				ret = cache_block_group(block_group, trans,
 							orig_root, 0);
 				BUG_ON(ret);
@@ -5219,8 +5227,7 @@ loop:
 		if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
 			found_uncached_bg = false;
 			loop++;
-			if (!ideal_cache_percent &&
-			    atomic_read(&space_info->caching_threads))
+			if (!ideal_cache_percent)
 				goto search;
 
 			/*
@@ -5623,7 +5630,7 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
 	if (!buf)
 		return ERR_PTR(-ENOMEM);
 	btrfs_set_header_generation(buf, trans->transid);
-	btrfs_set_buffer_lockdep_class(buf, level);
+	btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
 	btrfs_tree_lock(buf);
 	clean_tree_block(trans, root, buf);
 
@@ -5910,7 +5917,7 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
 			return 1;
 
 		if (path->locks[level] && !wc->keep_locks) {
-			btrfs_tree_unlock(eb);
+			btrfs_tree_unlock_rw(eb, path->locks[level]);
 			path->locks[level] = 0;
 		}
 		return 0;
@@ -5934,7 +5941,7 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
 	 * keep the tree lock
 	 */
 	if (path->locks[level] && level > 0) {
-		btrfs_tree_unlock(eb);
+		btrfs_tree_unlock_rw(eb, path->locks[level]);
 		path->locks[level] = 0;
 	}
 	return 0;
@@ -6047,7 +6054,7 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
 	BUG_ON(level != btrfs_header_level(next));
 	path->nodes[level] = next;
 	path->slots[level] = 0;
-	path->locks[level] = 1;
+	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
 	wc->level = level;
 	if (wc->level == 1)
 		wc->reada_slot = 0;
@@ -6118,7 +6125,7 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 			BUG_ON(level == 0);
 			btrfs_tree_lock(eb);
 			btrfs_set_lock_blocking(eb);
-			path->locks[level] = 1;
+			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
 
 			ret = btrfs_lookup_extent_info(trans, root,
 						       eb->start, eb->len,
@@ -6127,8 +6134,7 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 			BUG_ON(ret);
 			BUG_ON(wc->refs[level] == 0);
 			if (wc->refs[level] == 1) {
-				btrfs_tree_unlock(eb);
-				path->locks[level] = 0;
+				btrfs_tree_unlock_rw(eb, path->locks[level]);
 				return 1;
 			}
 		}
@@ -6150,7 +6156,7 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 		    btrfs_header_generation(eb) == trans->transid) {
 			btrfs_tree_lock(eb);
 			btrfs_set_lock_blocking(eb);
-			path->locks[level] = 1;
+			path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
 		}
 		clean_tree_block(trans, root, eb);
 	}
@@ -6229,7 +6235,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
 				return 0;
 
 			if (path->locks[level]) {
-				btrfs_tree_unlock(path->nodes[level]);
+				btrfs_tree_unlock_rw(path->nodes[level],
+						     path->locks[level]);
 				path->locks[level] = 0;
 			}
 			free_extent_buffer(path->nodes[level]);
@@ -6281,7 +6288,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
 		path->nodes[level] = btrfs_lock_root_node(root);
 		btrfs_set_lock_blocking(path->nodes[level]);
 		path->slots[level] = 0;
-		path->locks[level] = 1;
+		path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
 		memset(&wc->update_progress, 0,
 		       sizeof(wc->update_progress));
 	} else {
@@ -6449,7 +6456,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 	level = btrfs_header_level(node);
 	path->nodes[level] = node;
 	path->slots[level] = 0;
-	path->locks[level] = 1;
+	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
 
 	wc->refs[parent_level] = 1;
 	wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
@@ -6524,15 +6531,28 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
 	return flags;
 }
 
-static int set_block_group_ro(struct btrfs_block_group_cache *cache)
+static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
 {
 	struct btrfs_space_info *sinfo = cache->space_info;
 	u64 num_bytes;
+	u64 min_allocable_bytes;
 	int ret = -ENOSPC;
 
 	if (cache->ro)
 		return 0;
 
+	/*
+	 * We need some metadata space and system metadata space for
+	 * allocating chunks in some corner cases until we force to set
+	 * it to be readonly.
+	 */
+	if ((sinfo->flags &
+	     (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
+	    !force)
+		min_allocable_bytes = 1 * 1024 * 1024;
+	else
+		min_allocable_bytes = 0;
+
 	spin_lock(&sinfo->lock);
 	spin_lock(&cache->lock);
 	num_bytes = cache->key.offset - cache->reserved - cache->pinned -
@@ -6540,7 +6560,8 @@ static int set_block_group_ro(struct btrfs_block_group_cache *cache)
 
 	if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
 	    sinfo->bytes_may_use + sinfo->bytes_readonly +
-	    cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
+	    cache->reserved_pinned + num_bytes + min_allocable_bytes <=
+	    sinfo->total_bytes) {
 		sinfo->bytes_readonly += num_bytes;
 		sinfo->bytes_reserved += cache->reserved_pinned;
 		cache->reserved_pinned = 0;
@@ -6571,7 +6592,7 @@ int btrfs_set_block_group_ro(struct btrfs_root *root,
 		do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
 			       CHUNK_ALLOC_FORCE);
 
-	ret = set_block_group_ro(cache);
+	ret = set_block_group_ro(cache, 0);
 	if (!ret)
 		goto out;
 	alloc_flags = get_alloc_profile(root, cache->space_info->flags);
@@ -6579,7 +6600,7 @@ int btrfs_set_block_group_ro(struct btrfs_root *root,
 			     CHUNK_ALLOC_FORCE);
 	if (ret < 0)
 		goto out;
-	ret = set_block_group_ro(cache);
+	ret = set_block_group_ro(cache, 0);
 out:
 	btrfs_end_transaction(trans, root);
 	return ret;
@@ -7016,7 +7037,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 
 		set_avail_alloc_bits(root->fs_info, cache->flags);
 		if (btrfs_chunk_readonly(root, cache->key.objectid))
-			set_block_group_ro(cache);
+			set_block_group_ro(cache, 1);
 	}
 
 	list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
@@ -7030,9 +7051,9 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		 * mirrored block groups.
 		 */
 		list_for_each_entry(cache, &space_info->block_groups[3], list)
-			set_block_group_ro(cache);
+			set_block_group_ro(cache, 1);
 		list_for_each_entry(cache, &space_info->block_groups[4], list)
-			set_block_group_ro(cache);
+			set_block_group_ro(cache, 1);
 	}
 
 	init_global_block_rsv(info);

fs/btrfs/extent_io.c

@@ -281,11 +281,10 @@ static int merge_state(struct extent_io_tree *tree,
 		if (other->start == state->end + 1 &&
 		    other->state == state->state) {
 			merge_cb(tree, state, other);
-			other->start = state->start;
-			state->tree = NULL;
-			rb_erase(&state->rb_node, &tree->state);
-			free_extent_state(state);
-			state = NULL;
+			state->end = other->end;
+			other->tree = NULL;
+			rb_erase(&other->rb_node, &tree->state);
+			free_extent_state(other);
 		}
 	}
 
@@ -351,7 +350,6 @@ static int insert_state(struct extent_io_tree *tree,
 		       "%llu %llu\n", (unsigned long long)found->start,
 		       (unsigned long long)found->end,
 		       (unsigned long long)start, (unsigned long long)end);
-		free_extent_state(state);
 		return -EEXIST;
 	}
 	state->tree = tree;
@@ -500,7 +498,8 @@ again:
 			cached_state = NULL;
 		}
 
-		if (cached && cached->tree && cached->start == start) {
+		if (cached && cached->tree && cached->start <= start &&
+		    cached->end > start) {
 			if (clear)
 				atomic_dec(&cached->refs);
 			state = cached;
@@ -742,7 +741,8 @@ again:
 	spin_lock(&tree->lock);
 	if (cached_state && *cached_state) {
 		state = *cached_state;
-		if (state->start == start && state->tree) {
+		if (state->start <= start && state->end > start &&
+		    state->tree) {
 			node = &state->rb_node;
 			goto hit_next;
 		}
@@ -783,13 +783,13 @@ hit_next:
 		if (err)
 			goto out;
 
-		next_node = rb_next(node);
 		cache_state(state, cached_state);
 		merge_state(tree, state);
 		if (last_end == (u64)-1)
 			goto out;
 
 		start = last_end + 1;
+		next_node = rb_next(&state->rb_node);
 		if (next_node && start < end && prealloc && !need_resched()) {
 			state = rb_entry(next_node, struct extent_state,
 					 rb_node);
@@ -862,7 +862,6 @@ hit_next:
 		 * Avoid to free 'prealloc' if it can be merged with
 		 * the later extent.
 		 */
-		atomic_inc(&prealloc->refs);
 		err = insert_state(tree, prealloc, start, this_end,
 				   &bits);
 		BUG_ON(err == -EEXIST);
@@ -872,7 +871,6 @@ hit_next:
 			goto out;
 		}
 		cache_state(prealloc, cached_state);
-		free_extent_state(prealloc);
 		prealloc = NULL;
 		start = this_end + 1;
 		goto search_again;
@@ -1564,7 +1562,8 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
 	int bitset = 0;
 
 	spin_lock(&tree->lock);
-	if (cached && cached->tree && cached->start == start)
+	if (cached && cached->tree && cached->start <= start &&
+	    cached->end > start)
 		node = &cached->rb_node;
 	else
 		node = tree_search(tree, start);
@@ -2432,6 +2431,7 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
 	pgoff_t index;
 	pgoff_t end;		/* Inclusive */
 	int scanned = 0;
+	int tag;
 
 	pagevec_init(&pvec, 0);
 	if (wbc->range_cyclic) {
@@ -2442,11 +2442,16 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
 		end = wbc->range_end >> PAGE_CACHE_SHIFT;
 		scanned = 1;
 	}
+	if (wbc->sync_mode == WB_SYNC_ALL)
+		tag = PAGECACHE_TAG_TOWRITE;
+	else
+		tag = PAGECACHE_TAG_DIRTY;
 retry:
+	if (wbc->sync_mode == WB_SYNC_ALL)
+		tag_pages_for_writeback(mapping, index, end);
 	while (!done && !nr_to_write_done && (index <= end) &&
-	       (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
-			      PAGECACHE_TAG_DIRTY, min(end - index,
-				  (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
+	       (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
+			min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
 		unsigned i;
 
 		scanned = 1;
@@ -3022,8 +3027,15 @@ static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
 		return NULL;
 	eb->start = start;
 	eb->len = len;
-	spin_lock_init(&eb->lock);
-	init_waitqueue_head(&eb->lock_wq);
+	rwlock_init(&eb->lock);
+	atomic_set(&eb->write_locks, 0);
+	atomic_set(&eb->read_locks, 0);
+	atomic_set(&eb->blocking_readers, 0);
+	atomic_set(&eb->blocking_writers, 0);
+	atomic_set(&eb->spinning_readers, 0);
+	atomic_set(&eb->spinning_writers, 0);
+	init_waitqueue_head(&eb->write_lock_wq);
+	init_waitqueue_head(&eb->read_lock_wq);
 
 #if LEAK_DEBUG
 	spin_lock_irqsave(&leak_lock, flags);
@@ -3119,7 +3131,7 @@ struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
 		i = 0;
 	}
 	for (; i < num_pages; i++, index++) {
-		p = find_or_create_page(mapping, index, GFP_NOFS | __GFP_HIGHMEM);
+		p = find_or_create_page(mapping, index, GFP_NOFS);
 		if (!p) {
 			WARN_ON(1);
 			goto free_eb;
@@ -3266,6 +3278,22 @@ int set_extent_buffer_dirty(struct extent_io_tree *tree,
 	return was_dirty;
 }
 
+static int __eb_straddles_pages(u64 start, u64 len)
+{
+	if (len < PAGE_CACHE_SIZE)
+		return 1;
+	if (start & (PAGE_CACHE_SIZE - 1))
+		return 1;
+	if ((start + len) & (PAGE_CACHE_SIZE - 1))
+		return 1;
+	return 0;
+}
+
+static int eb_straddles_pages(struct extent_buffer *eb)
+{
+	return __eb_straddles_pages(eb->start, eb->len);
+}
+
 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
 				struct extent_buffer *eb,
 				struct extent_state **cached_state)
@@ -3277,8 +3305,10 @@ int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
 	num_pages = num_extent_pages(eb->start, eb->len);
 	clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
 
-	clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
-			      cached_state, GFP_NOFS);
+	if (eb_straddles_pages(eb)) {
+		clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
+				      cached_state, GFP_NOFS);
+	}
 	for (i = 0; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
 		if (page)
@@ -3296,8 +3326,10 @@ int set_extent_buffer_uptodate(struct extent_io_tree *tree,
 
 	num_pages = num_extent_pages(eb->start, eb->len);
 
-	set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
-			    NULL, GFP_NOFS);
+	if (eb_straddles_pages(eb)) {
+		set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
+				    NULL, GFP_NOFS);
+	}
 	for (i = 0; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
 		if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
@@ -3320,9 +3352,12 @@ int extent_range_uptodate(struct extent_io_tree *tree,
 	int uptodate;
 	unsigned long index;
 
-	ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
-	if (ret)
-		return 1;
+	if (__eb_straddles_pages(start, end - start + 1)) {
+		ret = test_range_bit(tree, start, end,
+				     EXTENT_UPTODATE, 1, NULL);
+		if (ret)
+			return 1;
+	}
 	while (start <= end) {
 		index = start >> PAGE_CACHE_SHIFT;
 		page = find_get_page(tree->mapping, index);
@@ -3350,10 +3385,12 @@ int extent_buffer_uptodate(struct extent_io_tree *tree,
 	if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
 		return 1;
 
-	ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
-			   EXTENT_UPTODATE, 1, cached_state);
-	if (ret)
-		return ret;
+	if (eb_straddles_pages(eb)) {
+		ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
+				   EXTENT_UPTODATE, 1, cached_state);
+		if (ret)
+			return ret;
+	}
 
 	num_pages = num_extent_pages(eb->start, eb->len);
 	for (i = 0; i < num_pages; i++) {
@@ -3386,9 +3423,11 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 	if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
 		return 0;
 
-	if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
-			   EXTENT_UPTODATE, 1, NULL)) {
-		return 0;
+	if (eb_straddles_pages(eb)) {
+		if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
+				   EXTENT_UPTODATE, 1, NULL)) {
+			return 0;
+		}
 	}
 
 	if (start) {
@@ -3492,9 +3531,8 @@ void read_extent_buffer(struct extent_buffer *eb, void *dstv,
 		page = extent_buffer_page(eb, i);
 
 		cur = min(len, (PAGE_CACHE_SIZE - offset));
-		kaddr = kmap_atomic(page, KM_USER1);
+		kaddr = page_address(page);
 		memcpy(dst, kaddr + offset, cur);
-		kunmap_atomic(kaddr, KM_USER1);
 
 		dst += cur;
 		len -= cur;
@@ -3504,9 +3542,9 @@ void read_extent_buffer(struct extent_buffer *eb, void *dstv,
 }
 
 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
-			       unsigned long min_len, char **token, char **map,
+			       unsigned long min_len, char **map,
 			       unsigned long *map_start,
-			       unsigned long *map_len, int km)
+			       unsigned long *map_len)
 {
 	size_t offset = start & (PAGE_CACHE_SIZE - 1);
 	char *kaddr;
@@ -3536,42 +3574,12 @@ int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
 	}
 
 	p = extent_buffer_page(eb, i);
-	kaddr = kmap_atomic(p, km);
-	*token = kaddr;
+	kaddr = page_address(p);
 	*map = kaddr + offset;
 	*map_len = PAGE_CACHE_SIZE - offset;
 	return 0;
 }
 
-int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
-		      unsigned long min_len,
-		      char **token, char **map,
-		      unsigned long *map_start,
-		      unsigned long *map_len, int km)
-{
-	int err;
-	int save = 0;
-	if (eb->map_token) {
-		unmap_extent_buffer(eb, eb->map_token, km);
-		eb->map_token = NULL;
-		save = 1;
-	}
-	err = map_private_extent_buffer(eb, start, min_len, token, map,
-				       map_start, map_len, km);
-	if (!err && save) {
-		eb->map_token = *token;
-		eb->kaddr = *map;
-		eb->map_start = *map_start;
-		eb->map_len = *map_len;
-	}
-	return err;
-}
-
-void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
-{
-	kunmap_atomic(token, km);
-}
-
 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
 			  unsigned long start,
 			  unsigned long len)
@@ -3595,9 +3603,8 @@ int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
 
 		cur = min(len, (PAGE_CACHE_SIZE - offset));
 
-		kaddr = kmap_atomic(page, KM_USER0);
+		kaddr = page_address(page);
 		ret = memcmp(ptr, kaddr + offset, cur);
-		kunmap_atomic(kaddr, KM_USER0);
 		if (ret)
 			break;
 
@@ -3630,9 +3637,8 @@ void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
 		WARN_ON(!PageUptodate(page));
 
 		cur = min(len, PAGE_CACHE_SIZE - offset);
-		kaddr = kmap_atomic(page, KM_USER1);
+		kaddr = page_address(page);
 		memcpy(kaddr + offset, src, cur);
-		kunmap_atomic(kaddr, KM_USER1);
 
 		src += cur;
 		len -= cur;
@@ -3661,9 +3667,8 @@ void memset_extent_buffer(struct extent_buffer *eb, char c,
 		WARN_ON(!PageUptodate(page));
 
 		cur = min(len, PAGE_CACHE_SIZE - offset);
-		kaddr = kmap_atomic(page, KM_USER0);
+		kaddr = page_address(page);
 		memset(kaddr + offset, c, cur);
-		kunmap_atomic(kaddr, KM_USER0);
 
 		len -= cur;
 		offset = 0;
@@ -3694,9 +3699,8 @@ void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
 
 		cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
 
-		kaddr = kmap_atomic(page, KM_USER0);
+		kaddr = page_address(page);
 		read_extent_buffer(src, kaddr + offset, src_offset, cur);
-		kunmap_atomic(kaddr, KM_USER0);
 
 		src_offset += cur;
 		len -= cur;
@@ -3709,20 +3713,17 @@ static void move_pages(struct page *dst_page, struct page *src_page,
 		       unsigned long dst_off, unsigned long src_off,
 		       unsigned long len)
 {
-	char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
+	char *dst_kaddr = page_address(dst_page);
 	if (dst_page == src_page) {
 		memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
 	} else {
-		char *src_kaddr = kmap_atomic(src_page, KM_USER1);
+		char *src_kaddr = page_address(src_page);
 		char *p = dst_kaddr + dst_off + len;
 		char *s = src_kaddr + src_off + len;
 
 		while (len--)
 			*--p = *--s;
-
-		kunmap_atomic(src_kaddr, KM_USER1);
 	}
-	kunmap_atomic(dst_kaddr, KM_USER0);
 }
 
 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
@@ -3735,20 +3736,17 @@ static void copy_pages(struct page *dst_page, struct page *src_page,
 		       unsigned long dst_off, unsigned long src_off,
 		       unsigned long len)
 {
-	char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
+	char *dst_kaddr = page_address(dst_page);
 	char *src_kaddr;
 
 	if (dst_page != src_page) {
-		src_kaddr = kmap_atomic(src_page, KM_USER1);
+		src_kaddr = page_address(src_page);
 	} else {
 		src_kaddr = dst_kaddr;
 		BUG_ON(areas_overlap(src_off, dst_off, len));
 	}
 
 	memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
-	kunmap_atomic(dst_kaddr, KM_USER0);
-	if (dst_page != src_page)
-		kunmap_atomic(src_kaddr, KM_USER1);
 }
 
 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,

fs/btrfs/extent_io.h

@@ -120,8 +120,6 @@ struct extent_state {
 struct extent_buffer {
 	u64 start;
 	unsigned long len;
-	char *map_token;
-	char *kaddr;
 	unsigned long map_start;
 	unsigned long map_len;
 	struct page *first_page;
@@ -130,14 +128,26 @@ struct extent_buffer {
 	struct rcu_head rcu_head;
 	atomic_t refs;
 
-	/* the spinlock is used to protect most operations */
-	spinlock_t lock;
+	/* count of read lock holders on the extent buffer */
+	atomic_t write_locks;
+	atomic_t read_locks;
+	atomic_t blocking_writers;
+	atomic_t blocking_readers;
+	atomic_t spinning_readers;
+	atomic_t spinning_writers;
+
+	/* protects write locks */
+	rwlock_t lock;
 
-	/*
-	 * when we keep the lock held while blocking, waiters go onto
-	 * the wq
+	/* readers use lock_wq while they wait for the write
+	 * lock holders to unlock
 	 */
-	wait_queue_head_t lock_wq;
+	wait_queue_head_t write_lock_wq;
+
+	/* writers use read_lock_wq while they wait for readers
+	 * to unlock
+	 */
+	wait_queue_head_t read_lock_wq;
 };
 
 static inline void extent_set_compress_type(unsigned long *bio_flags,
@@ -279,15 +289,10 @@ int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
 int extent_buffer_uptodate(struct extent_io_tree *tree,
 			   struct extent_buffer *eb,
 			   struct extent_state *cached_state);
-int map_extent_buffer(struct extent_buffer *eb, unsigned long offset,
-		      unsigned long min_len, char **token, char **map,
-		      unsigned long *map_start,
-		      unsigned long *map_len, int km);
 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset,
-		      unsigned long min_len, char **token, char **map,
+		      unsigned long min_len, char **map,
 		      unsigned long *map_start,
-		      unsigned long *map_len, int km);
-void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km);
+		      unsigned long *map_len);
 int extent_range_uptodate(struct extent_io_tree *tree,
 			  u64 start, u64 end);
 int extent_clear_unlock_delalloc(struct inode *inode,

fs/btrfs/file-item.c

@@ -177,6 +177,15 @@ static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
 
 	WARN_ON(bio->bi_vcnt <= 0);
 
+	/*
+	 * the free space stuff is only read when it hasn't been
+	 * updated in the current transaction.  So, we can safely
+	 * read from the commit root and sidestep a nasty deadlock
+	 * between reading the free space cache and updating the csum tree.
+	 */
+	if (btrfs_is_free_space_inode(root, inode))
+		path->search_commit_root = 1;
+
 	disk_bytenr = (u64)bio->bi_sector << 9;
 	if (dio)
 		offset = logical_offset;
@@ -664,10 +673,6 @@ int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 	struct btrfs_sector_sum *sector_sum;
 	u32 nritems;
 	u32 ins_size;
-	char *eb_map;
-	char *eb_token;
-	unsigned long map_len;
-	unsigned long map_start;
 	u16 csum_size =
 		btrfs_super_csum_size(&root->fs_info->super_copy);
 
@@ -814,30 +819,9 @@ found:
 	item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
 				      btrfs_item_size_nr(leaf, path->slots[0]));
-	eb_token = NULL;
 next_sector:
 
-	if (!eb_token ||
-	   (unsigned long)item + csum_size >= map_start + map_len) {
-		int err;
-
-		if (eb_token)
-			unmap_extent_buffer(leaf, eb_token, KM_USER1);
-		eb_token = NULL;
-		err = map_private_extent_buffer(leaf, (unsigned long)item,
-						csum_size,
-						&eb_token, &eb_map,
-						&map_start, &map_len, KM_USER1);
-		if (err)
-			eb_token = NULL;
-	}
-	if (eb_token) {
-		memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
-		       &sector_sum->sum, csum_size);
-	} else {
-		write_extent_buffer(leaf, &sector_sum->sum,
-				    (unsigned long)item, csum_size);
-	}
+	write_extent_buffer(leaf, &sector_sum->sum, (unsigned long)item, csum_size);
 
 	total_bytes += root->sectorsize;
 	sector_sum++;
@@ -850,10 +834,7 @@ next_sector:
 			goto next_sector;
 		}
 	}
-	if (eb_token) {
-		unmap_extent_buffer(leaf, eb_token, KM_USER1);
-		eb_token = NULL;
-	}
+
 	btrfs_mark_buffer_dirty(path->nodes[0]);
 	if (total_bytes < sums->len) {
 		btrfs_release_path(path);

fs/btrfs/file.c

@@ -1081,7 +1081,8 @@ static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 
 again:
 	for (i = 0; i < num_pages; i++) {
-		pages[i] = grab_cache_page(inode->i_mapping, index + i);
+		pages[i] = find_or_create_page(inode->i_mapping, index + i,
+					       GFP_NOFS);
 		if (!pages[i]) {
 			faili = i - 1;
 			err = -ENOMEM;
@@ -1238,9 +1239,11 @@ static noinline ssize_t __btrfs_buffered_write(struct file *file,
 		 * managed to copy.
 		 */
 		if (num_pages > dirty_pages) {
-			if (copied > 0)
-				atomic_inc(
-					&BTRFS_I(inode)->outstanding_extents);
+			if (copied > 0) {
+				spin_lock(&BTRFS_I(inode)->lock);
+				BTRFS_I(inode)->outstanding_extents++;
+				spin_unlock(&BTRFS_I(inode)->lock);
+			}
 			btrfs_delalloc_release_space(inode,
 					(num_pages - dirty_pages) <<
 					PAGE_CACHE_SHIFT);

fs/btrfs/free-space-cache.c

@@ -98,6 +98,12 @@ struct inode *lookup_free_space_inode(struct btrfs_root *root,
 		return inode;
 
 	spin_lock(&block_group->lock);
+	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) {
+		printk(KERN_INFO "Old style space inode found, converting.\n");
+		BTRFS_I(inode)->flags &= ~BTRFS_INODE_NODATASUM;
+		block_group->disk_cache_state = BTRFS_DC_CLEAR;
+	}
+
 	if (!btrfs_fs_closing(root->fs_info)) {
 		block_group->inode = igrab(inode);
 		block_group->iref = 1;
@@ -135,7 +141,7 @@ int __create_free_space_inode(struct btrfs_root *root,
 	btrfs_set_inode_gid(leaf, inode_item, 0);
 	btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
 	btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS |
-			      BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM);
+			      BTRFS_INODE_PREALLOC);
 	btrfs_set_inode_nlink(leaf, inode_item, 1);
 	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
 	btrfs_set_inode_block_group(leaf, inode_item, offset);
@@ -239,17 +245,12 @@ int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 	struct btrfs_free_space_header *header;
 	struct extent_buffer *leaf;
 	struct page *page;
-	u32 *checksums = NULL, *crc;
-	char *disk_crcs = NULL;
 	struct btrfs_key key;
 	struct list_head bitmaps;
 	u64 num_entries;
 	u64 num_bitmaps;
 	u64 generation;
-	u32 cur_crc = ~(u32)0;
 	pgoff_t index = 0;
-	unsigned long first_page_offset;
-	int num_checksums;
 	int ret = 0;
 
 	INIT_LIST_HEAD(&bitmaps);
@@ -292,16 +293,6 @@ int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 	if (!num_entries)
 		goto out;
 
-	/* Setup everything for doing checksumming */
-	num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
-	checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
-	if (!checksums)
-		goto out;
-	first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
-	disk_crcs = kzalloc(first_page_offset, GFP_NOFS);
-	if (!disk_crcs)
-		goto out;
-
 	ret = readahead_cache(inode);
 	if (ret)
 		goto out;
@@ -311,18 +302,12 @@ int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 		struct btrfs_free_space *e;
 		void *addr;
 		unsigned long offset = 0;
-		unsigned long start_offset = 0;
 		int need_loop = 0;
 
 		if (!num_entries && !num_bitmaps)
 			break;
 
-		if (index == 0) {
-			start_offset = first_page_offset;
-			offset = start_offset;
-		}
-
-		page = grab_cache_page(inode->i_mapping, index);
+		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
 		if (!page)
 			goto free_cache;
 
@@ -342,8 +327,15 @@ int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 		if (index == 0) {
 			u64 *gen;
 
-			memcpy(disk_crcs, addr, first_page_offset);
-			gen = addr + (sizeof(u32) * num_checksums);
+			/*
+			 * We put a bogus crc in the front of the first page in
+			 * case old kernels try to mount a fs with the new
+			 * format to make sure they discard the cache.
+			 */
+			addr += sizeof(u64);
+			offset += sizeof(u64);
+
+			gen = addr;
 			if (*gen != BTRFS_I(inode)->generation) {
 				printk(KERN_ERR "btrfs: space cache generation"
 				       " (%llu) does not match inode (%llu)\n",
@@ -355,24 +347,10 @@ int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
 				page_cache_release(page);
 				goto free_cache;
 			}
-			crc = (u32 *)disk_crcs;
-		}
-		entry = addr + start_offset;
-
-		/* First lets check our crc before we do anything fun */
-		cur_crc = ~(u32)0;
-		cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc,
-					  PAGE_CACHE_SIZE - start_offset);
-		btrfs_csum_final(cur_crc, (char *)&cur_crc);
-		if (cur_crc != *crc) {
-			printk(KERN_ERR "btrfs: crc mismatch for page %lu\n",
-			       index);
-			kunmap(page);
-			unlock_page(page);
-			page_cache_release(page);
-			goto free_cache;
+			addr += sizeof(u64);
+			offset += sizeof(u64);
 		}
-		crc++;
+		entry = addr;
 
 		while (1) {
 			if (!num_entries)
@@ -470,8 +448,6 @@ next:
 
 	ret = 1;
 out:
-	kfree(checksums);
-	kfree(disk_crcs);
 	return ret;
 free_cache:
 	__btrfs_remove_free_space_cache(ctl);
@@ -569,8 +545,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 	struct btrfs_key key;
 	u64 start, end, len;
 	u64 bytes = 0;
-	u32 *crc, *checksums;
-	unsigned long first_page_offset;
+	u32 crc = ~(u32)0;
 	int index = 0, num_pages = 0;
 	int entries = 0;
 	int bitmaps = 0;
@@ -590,34 +565,13 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 	num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
 		PAGE_CACHE_SHIFT;
 
-	/* Since the first page has all of our checksums and our generation we
-	 * need to calculate the offset into the page that we can start writing
-	 * our entries.
-	 */
-	first_page_offset = (sizeof(u32) * num_pages) + sizeof(u64);
-
 	filemap_write_and_wait(inode->i_mapping);
 	btrfs_wait_ordered_range(inode, inode->i_size &
 				 ~(root->sectorsize - 1), (u64)-1);
 
-	/* make sure we don't overflow that first page */
-	if (first_page_offset + sizeof(struct btrfs_free_space_entry) >= PAGE_CACHE_SIZE) {
-		/* this is really the same as running out of space, where we also return 0 */
-		printk(KERN_CRIT "Btrfs: free space cache was too big for the crc page\n");
-		ret = 0;
-		goto out_update;
-	}
-
-	/* We need a checksum per page. */
-	crc = checksums = kzalloc(sizeof(u32) * num_pages, GFP_NOFS);
-	if (!crc)
-		return -1;
-
 	pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS);
-	if (!pages) {
-		kfree(crc);
+	if (!pages)
 		return -1;
-	}
 
 	/* Get the cluster for this block_group if it exists */
 	if (block_group && !list_empty(&block_group->cluster_list))
@@ -640,7 +594,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 	 * know and don't freak out.
 	 */
 	while (index < num_pages) {
-		page = grab_cache_page(inode->i_mapping, index);
+		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
 		if (!page) {
 			int i;
 
@@ -648,7 +602,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 				unlock_page(pages[i]);
 				page_cache_release(pages[i]);
 			}
-			goto out_free;
+			goto out;
 		}
 		pages[index] = page;
 		index++;
@@ -668,17 +622,11 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 	/* Write out the extent entries */
 	do {
 		struct btrfs_free_space_entry *entry;
-		void *addr;
+		void *addr, *orig;
 		unsigned long offset = 0;
-		unsigned long start_offset = 0;
 
 		next_page = false;
 
-		if (index == 0) {
-			start_offset = first_page_offset;
-			offset = start_offset;
-		}
-
 		if (index >= num_pages) {
 			out_of_space = true;
 			break;
@@ -686,10 +634,26 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 
 		page = pages[index];
 
-		addr = kmap(page);
-		entry = addr + start_offset;
+		orig = addr = kmap(page);
+		if (index == 0) {
+			u64 *gen;
 
-		memset(addr, 0, PAGE_CACHE_SIZE);
+			/*
+			 * We're going to put in a bogus crc for this page to
+			 * make sure that old kernels who aren't aware of this
+			 * format will be sure to discard the cache.
+			 */
+			addr += sizeof(u64);
+			offset += sizeof(u64);
+
+			gen = addr;
+			*gen = trans->transid;
+			addr += sizeof(u64);
+			offset += sizeof(u64);
+		}
+		entry = addr;
+
+		memset(addr, 0, PAGE_CACHE_SIZE - offset);
 		while (node && !next_page) {
 			struct btrfs_free_space *e;
 
@@ -752,13 +716,19 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 				next_page = true;
 			entry++;
 		}
-		*crc = ~(u32)0;
-		*crc = btrfs_csum_data(root, addr + start_offset, *crc,
-				       PAGE_CACHE_SIZE - start_offset);
-		kunmap(page);
 
-		btrfs_csum_final(*crc, (char *)crc);
-		crc++;
+		/* Generate bogus crc value */
+		if (index == 0) {
+			u32 *tmp;
+			crc = btrfs_csum_data(root, orig + sizeof(u64), crc,
+					      PAGE_CACHE_SIZE - sizeof(u64));
+			btrfs_csum_final(crc, (char *)&crc);
+			crc++;
+			tmp = orig;
+			*tmp = crc;
+		}
+
+		kunmap(page);
 
 		bytes += PAGE_CACHE_SIZE;
 
@@ -779,11 +749,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 
 		addr = kmap(page);
 		memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE);
-		*crc = ~(u32)0;
-		*crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE);
 		kunmap(page);
-		btrfs_csum_final(*crc, (char *)crc);
-		crc++;
 		bytes += PAGE_CACHE_SIZE;
 
 		list_del_init(&entry->list);
@@ -796,7 +762,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 				     i_size_read(inode) - 1, &cached_state,
 				     GFP_NOFS);
 		ret = 0;
-		goto out_free;
+		goto out;
 	}
 
 	/* Zero out the rest of the pages just to make sure */
@@ -811,20 +777,6 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 		index++;
 	}
 
-	/* Write the checksums and trans id to the first page */
-	{
-		void *addr;
-		u64 *gen;
-
-		page = pages[0];
-
-		addr = kmap(page);
-		memcpy(addr, checksums, sizeof(u32) * num_pages);
-		gen = addr + (sizeof(u32) * num_pages);
-		*gen = trans->transid;
-		kunmap(page);
-	}
-
 	ret = btrfs_dirty_pages(root, inode, pages, num_pages, 0,
 					    bytes, &cached_state);
 	btrfs_drop_pages(pages, num_pages);
@@ -833,7 +785,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 
 	if (ret) {
 		ret = 0;
-		goto out_free;
+		goto out;
 	}
 
 	BTRFS_I(inode)->generation = trans->transid;
@@ -850,7 +802,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
 				 EXTENT_DIRTY | EXTENT_DELALLOC |
 				 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
-		goto out_free;
+		goto out;
 	}
 	leaf = path->nodes[0];
 	if (ret > 0) {
@@ -866,7 +818,7 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 					 EXTENT_DO_ACCOUNTING, 0, 0, NULL,
 					 GFP_NOFS);
 			btrfs_release_path(path);
-			goto out_free;
+			goto out;
 		}
 	}
 	header = btrfs_item_ptr(leaf, path->slots[0],
@@ -879,11 +831,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 
 	ret = 1;
 
-out_free:
-	kfree(checksums);
+out:
 	kfree(pages);
-
-out_update:
 	if (ret != 1) {
 		invalidate_inode_pages2_range(inode->i_mapping, 0, index);
 		BTRFS_I(inode)->generation = 0;

fs/btrfs/inode.c

@@ -750,15 +750,6 @@ static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 	return alloc_hint;
 }
 
-static inline bool is_free_space_inode(struct btrfs_root *root,
-				       struct inode *inode)
-{
-	if (root == root->fs_info->tree_root ||
-	    BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
-		return true;
-	return false;
-}
-
 /*
  * when extent_io.c finds a delayed allocation range in the file,
  * the call backs end up in this code.  The basic idea is to
@@ -791,7 +782,7 @@ static noinline int cow_file_range(struct inode *inode,
 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 	int ret = 0;
 
-	BUG_ON(is_free_space_inode(root, inode));
+	BUG_ON(btrfs_is_free_space_inode(root, inode));
 	trans = btrfs_join_transaction(root);
 	BUG_ON(IS_ERR(trans));
 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
@@ -1072,7 +1063,7 @@ static noinline int run_delalloc_nocow(struct inode *inode,
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 
-	nolock = is_free_space_inode(root, inode);
+	nolock = btrfs_is_free_space_inode(root, inode);
 
 	if (nolock)
 		trans = btrfs_join_transaction_nolock(root);
@@ -1298,7 +1289,9 @@ static int btrfs_split_extent_hook(struct inode *inode,
 	if (!(orig->state & EXTENT_DELALLOC))
 		return 0;
 
-	atomic_inc(&BTRFS_I(inode)->outstanding_extents);
+	spin_lock(&BTRFS_I(inode)->lock);
+	BTRFS_I(inode)->outstanding_extents++;
+	spin_unlock(&BTRFS_I(inode)->lock);
 	return 0;
 }
 
@@ -1316,7 +1309,9 @@ static int btrfs_merge_extent_hook(struct inode *inode,
 	if (!(other->state & EXTENT_DELALLOC))
 		return 0;
 
-	atomic_dec(&BTRFS_I(inode)->outstanding_extents);
+	spin_lock(&BTRFS_I(inode)->lock);
+	BTRFS_I(inode)->outstanding_extents--;
+	spin_unlock(&BTRFS_I(inode)->lock);
 	return 0;
 }
 
@@ -1337,12 +1332,15 @@ static int btrfs_set_bit_hook(struct inode *inode,
 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
 		struct btrfs_root *root = BTRFS_I(inode)->root;
 		u64 len = state->end + 1 - state->start;
-		bool do_list = !is_free_space_inode(root, inode);
+		bool do_list = !btrfs_is_free_space_inode(root, inode);
 
-		if (*bits & EXTENT_FIRST_DELALLOC)
+		if (*bits & EXTENT_FIRST_DELALLOC) {
 			*bits &= ~EXTENT_FIRST_DELALLOC;
-		else
-			atomic_inc(&BTRFS_I(inode)->outstanding_extents);
+		} else {
+			spin_lock(&BTRFS_I(inode)->lock);
+			BTRFS_I(inode)->outstanding_extents++;
+			spin_unlock(&BTRFS_I(inode)->lock);
+		}
 
 		spin_lock(&root->fs_info->delalloc_lock);
 		BTRFS_I(inode)->delalloc_bytes += len;
@@ -1370,12 +1368,15 @@ static int btrfs_clear_bit_hook(struct inode *inode,
 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
 		struct btrfs_root *root = BTRFS_I(inode)->root;
 		u64 len = state->end + 1 - state->start;
-		bool do_list = !is_free_space_inode(root, inode);
+		bool do_list = !btrfs_is_free_space_inode(root, inode);
 
-		if (*bits & EXTENT_FIRST_DELALLOC)
+		if (*bits & EXTENT_FIRST_DELALLOC) {
 			*bits &= ~EXTENT_FIRST_DELALLOC;
-		else if (!(*bits & EXTENT_DO_ACCOUNTING))
-			atomic_dec(&BTRFS_I(inode)->outstanding_extents);
+		} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
+			spin_lock(&BTRFS_I(inode)->lock);
+			BTRFS_I(inode)->outstanding_extents--;
+			spin_unlock(&BTRFS_I(inode)->lock);
+		}
 
 		if (*bits & EXTENT_DO_ACCOUNTING)
 			btrfs_delalloc_release_metadata(inode, len);
@@ -1477,7 +1478,7 @@ static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 
 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
 
-	if (is_free_space_inode(root, inode))
+	if (btrfs_is_free_space_inode(root, inode))
 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
 	else
 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
@@ -1726,7 +1727,7 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 		return 0;
 	BUG_ON(!ordered_extent);
 
-	nolock = is_free_space_inode(root, inode);
+	nolock = btrfs_is_free_space_inode(root, inode);
 
 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
 		BUG_ON(!list_empty(&ordered_extent->list));
@@ -2531,13 +2532,6 @@ static void btrfs_read_locked_inode(struct inode *inode)
 
 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
 				    struct btrfs_inode_item);
-	if (!leaf->map_token)
-		map_private_extent_buffer(leaf, (unsigned long)inode_item,
-					  sizeof(struct btrfs_inode_item),
-					  &leaf->map_token, &leaf->kaddr,
-					  &leaf->map_start, &leaf->map_len,
-					  KM_USER1);
-
 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 	inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 	inode->i_uid = btrfs_inode_uid(leaf, inode_item);
@@ -2575,11 +2569,6 @@ cache_acl:
 	if (!maybe_acls)
 		cache_no_acl(inode);
 
-	if (leaf->map_token) {
-		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-		leaf->map_token = NULL;
-	}
-
 	btrfs_free_path(path);
 
 	switch (inode->i_mode & S_IFMT) {
@@ -2624,13 +2613,6 @@ static void fill_inode_item(struct btrfs_trans_handle *trans,
 			    struct btrfs_inode_item *item,
 			    struct inode *inode)
 {
-	if (!leaf->map_token)
-		map_private_extent_buffer(leaf, (unsigned long)item,
-					  sizeof(struct btrfs_inode_item),
-					  &leaf->map_token, &leaf->kaddr,
-					  &leaf->map_start, &leaf->map_len,
-					  KM_USER1);
-
 	btrfs_set_inode_uid(leaf, item, inode->i_uid);
 	btrfs_set_inode_gid(leaf, item, inode->i_gid);
 	btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
@@ -2659,11 +2641,6 @@ static void fill_inode_item(struct btrfs_trans_handle *trans,
 	btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 	btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 	btrfs_set_inode_block_group(leaf, item, 0);
-
-	if (leaf->map_token) {
-		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
-		leaf->map_token = NULL;
-	}
 }
 
 /*
@@ -2684,7 +2661,7 @@ noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
 	 * The data relocation inode should also be directly updated
 	 * without delay
 	 */
-	if (!is_free_space_inode(root, inode)
+	if (!btrfs_is_free_space_inode(root, inode)
 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
 		ret = btrfs_delayed_update_inode(trans, root, inode);
 		if (!ret)
@@ -3398,7 +3375,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 
 	ret = -ENOMEM;
 again:
-	page = grab_cache_page(mapping, index);
+	page = find_or_create_page(mapping, index, GFP_NOFS);
 	if (!page) {
 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 		goto out;
@@ -3634,7 +3611,7 @@ void btrfs_evict_inode(struct inode *inode)
 
 	truncate_inode_pages(&inode->i_data, 0);
 	if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
-			       is_free_space_inode(root, inode)))
+			       btrfs_is_free_space_inode(root, inode)))
 		goto no_delete;
 
 	if (is_bad_inode(inode)) {
@@ -4277,7 +4254,7 @@ int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 	if (BTRFS_I(inode)->dummy_inode)
 		return 0;
 
-	if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
+	if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
 		nolock = true;
 
 	if (wbc->sync_mode == WB_SYNC_ALL) {
@@ -6735,8 +6712,9 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
 	ei->index_cnt = (u64)-1;
 	ei->last_unlink_trans = 0;
 
-	atomic_set(&ei->outstanding_extents, 0);
-	atomic_set(&ei->reserved_extents, 0);
+	spin_lock_init(&ei->lock);
+	ei->outstanding_extents = 0;
+	ei->reserved_extents = 0;
 
 	ei->ordered_data_close = 0;
 	ei->orphan_meta_reserved = 0;
@@ -6774,8 +6752,8 @@ void btrfs_destroy_inode(struct inode *inode)
 
 	WARN_ON(!list_empty(&inode->i_dentry));
 	WARN_ON(inode->i_data.nrpages);
-	WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
-	WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
+	WARN_ON(BTRFS_I(inode)->outstanding_extents);
+	WARN_ON(BTRFS_I(inode)->reserved_extents);
 
 	/*
 	 * This can happen where we create an inode, but somebody else also
@@ -6830,7 +6808,7 @@ int btrfs_drop_inode(struct inode *inode)
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 
 	if (btrfs_root_refs(&root->root_item) == 0 &&
-	    !is_free_space_inode(root, inode))
+	    !btrfs_is_free_space_inode(root, inode))
 		return 1;
 	else
 		return generic_drop_inode(inode);

fs/btrfs/ioctl.c

@@ -867,8 +867,8 @@ again:
 	/* step one, lock all the pages */
 	for (i = 0; i < num_pages; i++) {
 		struct page *page;
-		page = grab_cache_page(inode->i_mapping,
-					    start_index + i);
+		page = find_or_create_page(inode->i_mapping,
+					    start_index + i, GFP_NOFS);
 		if (!page)
 			break;
 
@@ -938,7 +938,9 @@ again:
 			  GFP_NOFS);
 
 	if (i_done != num_pages) {
-		atomic_inc(&BTRFS_I(inode)->outstanding_extents);
+		spin_lock(&BTRFS_I(inode)->lock);
+		BTRFS_I(inode)->outstanding_extents++;
+		spin_unlock(&BTRFS_I(inode)->lock);
 		btrfs_delalloc_release_space(inode,
 				     (num_pages - i_done) << PAGE_CACHE_SHIFT);
 	}

fs/btrfs/locking.c

@@ -24,185 +24,197 @@
 #include "extent_io.h"
 #include "locking.h"
 
-static inline void spin_nested(struct extent_buffer *eb)
-{
-	spin_lock(&eb->lock);
-}
+void btrfs_assert_tree_read_locked(struct extent_buffer *eb);
 
 /*
- * Setting a lock to blocking will drop the spinlock and set the
- * flag that forces other procs who want the lock to wait.  After
- * this you can safely schedule with the lock held.
+ * if we currently have a spinning reader or writer lock
+ * (indicated by the rw flag) this will bump the count
+ * of blocking holders and drop the spinlock.
  */
-void btrfs_set_lock_blocking(struct extent_buffer *eb)
+void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw)
 {
-	if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) {
-		set_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags);
-		spin_unlock(&eb->lock);
+	if (rw == BTRFS_WRITE_LOCK) {
+		if (atomic_read(&eb->blocking_writers) == 0) {
+			WARN_ON(atomic_read(&eb->spinning_writers) != 1);
+			atomic_dec(&eb->spinning_writers);
+			btrfs_assert_tree_locked(eb);
+			atomic_inc(&eb->blocking_writers);
+			write_unlock(&eb->lock);
+		}
+	} else if (rw == BTRFS_READ_LOCK) {
+		btrfs_assert_tree_read_locked(eb);
+		atomic_inc(&eb->blocking_readers);
+		WARN_ON(atomic_read(&eb->spinning_readers) == 0);
+		atomic_dec(&eb->spinning_readers);
+		read_unlock(&eb->lock);
 	}
-	/* exit with the spin lock released and the bit set */
+	return;
 }
 
 /*
- * clearing the blocking flag will take the spinlock again.
- * After this you can't safely schedule
+ * if we currently have a blocking lock, take the spinlock
+ * and drop our blocking count
  */
-void btrfs_clear_lock_blocking(struct extent_buffer *eb)
+void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
 {
-	if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) {
-		spin_nested(eb);
-		clear_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags);
-		smp_mb__after_clear_bit();
+	if (rw == BTRFS_WRITE_LOCK_BLOCKING) {
+		BUG_ON(atomic_read(&eb->blocking_writers) != 1);
+		write_lock(&eb->lock);
+		WARN_ON(atomic_read(&eb->spinning_writers));
+		atomic_inc(&eb->spinning_writers);
+		if (atomic_dec_and_test(&eb->blocking_writers))
+			wake_up(&eb->write_lock_wq);
+	} else if (rw == BTRFS_READ_LOCK_BLOCKING) {
+		BUG_ON(atomic_read(&eb->blocking_readers) == 0);
+		read_lock(&eb->lock);
+		atomic_inc(&eb->spinning_readers);
+		if (atomic_dec_and_test(&eb->blocking_readers))
+			wake_up(&eb->read_lock_wq);
 	}
-	/* exit with the spin lock held */
+	return;
 }
 
 /*
- * unfortunately, many of the places that currently set a lock to blocking
- * don't end up blocking for very long, and often they don't block
- * at all.  For a dbench 50 run, if we don't spin on the blocking bit
- * at all, the context switch rate can jump up to 400,000/sec or more.
- *
- * So, we're still stuck with this crummy spin on the blocking bit,
- * at least until the most common causes of the short blocks
- * can be dealt with.
+ * take a spinning read lock.  This will wait for any blocking
+ * writers
  */
-static int btrfs_spin_on_block(struct extent_buffer *eb)
+void btrfs_tree_read_lock(struct extent_buffer *eb)
 {
-	int i;
-
-	for (i = 0; i < 512; i++) {
-		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-			return 1;
-		if (need_resched())
-			break;
-		cpu_relax();
+again:
+	wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
+	read_lock(&eb->lock);
+	if (atomic_read(&eb->blocking_writers)) {
+		read_unlock(&eb->lock);
+		wait_event(eb->write_lock_wq,
+			   atomic_read(&eb->blocking_writers) == 0);
+		goto again;
 	}
-	return 0;
+	atomic_inc(&eb->read_locks);
+	atomic_inc(&eb->spinning_readers);
 }
 
 /*
- * This is somewhat different from trylock.  It will take the
- * spinlock but if it finds the lock is set to blocking, it will
- * return without the lock held.
- *
- * returns 1 if it was able to take the lock and zero otherwise
- *
- * After this call, scheduling is not safe without first calling
- * btrfs_set_lock_blocking()
+ * returns 1 if we get the read lock and 0 if we don't
+ * this won't wait for blocking writers
  */
-int btrfs_try_spin_lock(struct extent_buffer *eb)
+int btrfs_try_tree_read_lock(struct extent_buffer *eb)
 {
-	int i;
+	if (atomic_read(&eb->blocking_writers))
+		return 0;
 
-	if (btrfs_spin_on_block(eb)) {
-		spin_nested(eb);
-		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-			return 1;
-		spin_unlock(&eb->lock);
+	read_lock(&eb->lock);
+	if (atomic_read(&eb->blocking_writers)) {
+		read_unlock(&eb->lock);
+		return 0;
 	}
-	/* spin for a bit on the BLOCKING flag */
-	for (i = 0; i < 2; i++) {
-		cpu_relax();
-		if (!btrfs_spin_on_block(eb))
-			break;
-
-		spin_nested(eb);
-		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-			return 1;
-		spin_unlock(&eb->lock);
-	}
-	return 0;
+	atomic_inc(&eb->read_locks);
+	atomic_inc(&eb->spinning_readers);
+	return 1;
 }
 
 /*
- * the autoremove wake function will return 0 if it tried to wake up
- * a process that was already awake, which means that process won't
- * count as an exclusive wakeup.  The waitq code will continue waking
- * procs until it finds one that was actually sleeping.
- *
- * For btrfs, this isn't quite what we want.  We want a single proc
- * to be notified that the lock is ready for taking.  If that proc
- * already happen to be awake, great, it will loop around and try for
- * the lock.
- *
- * So, btrfs_wake_function always returns 1, even when the proc that we
- * tried to wake up was already awake.
+ * returns 1 if we get the read lock and 0 if we don't
+ * this won't wait for blocking writers or readers
  */
-static int btrfs_wake_function(wait_queue_t *wait, unsigned mode,
-			       int sync, void *key)
+int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 {
-	autoremove_wake_function(wait, mode, sync, key);
+	if (atomic_read(&eb->blocking_writers) ||
+	    atomic_read(&eb->blocking_readers))
+		return 0;
+	write_lock(&eb->lock);
+	if (atomic_read(&eb->blocking_writers) ||
+	    atomic_read(&eb->blocking_readers)) {
+		write_unlock(&eb->lock);
+		return 0;
+	}
+	atomic_inc(&eb->write_locks);
+	atomic_inc(&eb->spinning_writers);
 	return 1;
 }
 
 /*
- * returns with the extent buffer spinlocked.
- *
- * This will spin and/or wait as required to take the lock, and then
- * return with the spinlock held.
- *
- * After this call, scheduling is not safe without first calling
- * btrfs_set_lock_blocking()
+ * drop a spinning read lock
+ */
+void btrfs_tree_read_unlock(struct extent_buffer *eb)
+{
+	btrfs_assert_tree_read_locked(eb);
+	WARN_ON(atomic_read(&eb->spinning_readers) == 0);
+	atomic_dec(&eb->spinning_readers);
+	atomic_dec(&eb->read_locks);
+	read_unlock(&eb->lock);
+}
+
+/*
+ * drop a blocking read lock
+ */
+void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
+{
+	btrfs_assert_tree_read_locked(eb);
+	WARN_ON(atomic_read(&eb->blocking_readers) == 0);
+	if (atomic_dec_and_test(&eb->blocking_readers))
+		wake_up(&eb->read_lock_wq);
+	atomic_dec(&eb->read_locks);
+}
+
+/*
+ * take a spinning write lock.  This will wait for both
+ * blocking readers or writers
  */
 int btrfs_tree_lock(struct extent_buffer *eb)
 {
-	DEFINE_WAIT(wait);
-	wait.func = btrfs_wake_function;
-
-	if (!btrfs_spin_on_block(eb))
-		goto sleep;
-
-	while(1) {
-		spin_nested(eb);
-
-		/* nobody is blocking, exit with the spinlock held */
-		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-			return 0;
-
-		/*
-		 * we have the spinlock, but the real owner is blocking.
-		 * wait for them
-		 */
-		spin_unlock(&eb->lock);
-
-		/*
-		 * spin for a bit, and if the blocking flag goes away,
-		 * loop around
-		 */
-		cpu_relax();
-		if (btrfs_spin_on_block(eb))
-			continue;
-sleep:
-		prepare_to_wait_exclusive(&eb->lock_wq, &wait,
-					  TASK_UNINTERRUPTIBLE);
-
-		if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-			schedule();
-
-		finish_wait(&eb->lock_wq, &wait);
+again:
+	wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0);
+	wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
+	write_lock(&eb->lock);
+	if (atomic_read(&eb->blocking_readers)) {
+		write_unlock(&eb->lock);
+		wait_event(eb->read_lock_wq,
+			   atomic_read(&eb->blocking_readers) == 0);
+		goto again;
 	}
+	if (atomic_read(&eb->blocking_writers)) {
+		write_unlock(&eb->lock);
+		wait_event(eb->write_lock_wq,
+			   atomic_read(&eb->blocking_writers) == 0);
+		goto again;
+	}
+	WARN_ON(atomic_read(&eb->spinning_writers));
+	atomic_inc(&eb->spinning_writers);
+	atomic_inc(&eb->write_locks);
 	return 0;
 }
 
+/*
+ * drop a spinning or a blocking write lock.
+ */
 int btrfs_tree_unlock(struct extent_buffer *eb)
 {
-	/*
-	 * if we were a blocking owner, we don't have the spinlock held
-	 * just clear the bit and look for waiters
-	 */
-	if (test_and_clear_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-		smp_mb__after_clear_bit();
-	else
-		spin_unlock(&eb->lock);
-
-	if (waitqueue_active(&eb->lock_wq))
-		wake_up(&eb->lock_wq);
+	int blockers = atomic_read(&eb->blocking_writers);
+
+	BUG_ON(blockers > 1);
+
+	btrfs_assert_tree_locked(eb);
+	atomic_dec(&eb->write_locks);
+
+	if (blockers) {
+		WARN_ON(atomic_read(&eb->spinning_writers));
+		atomic_dec(&eb->blocking_writers);
+		smp_wmb();
+		wake_up(&eb->write_lock_wq);
+	} else {
+		WARN_ON(atomic_read(&eb->spinning_writers) != 1);
+		atomic_dec(&eb->spinning_writers);
+		write_unlock(&eb->lock);
+	}
 	return 0;
 }
 
 void btrfs_assert_tree_locked(struct extent_buffer *eb)
 {
-	if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-		assert_spin_locked(&eb->lock);
+	BUG_ON(!atomic_read(&eb->write_locks));
+}
+
+void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
+{
+	BUG_ON(!atomic_read(&eb->read_locks));
 }

fs/btrfs/locking.h

@@ -19,11 +19,43 @@
 #ifndef __BTRFS_LOCKING_
 #define __BTRFS_LOCKING_
 
+#define BTRFS_WRITE_LOCK 1
+#define BTRFS_READ_LOCK 2
+#define BTRFS_WRITE_LOCK_BLOCKING 3
+#define BTRFS_READ_LOCK_BLOCKING 4
+
 int btrfs_tree_lock(struct extent_buffer *eb);
 int btrfs_tree_unlock(struct extent_buffer *eb);
 int btrfs_try_spin_lock(struct extent_buffer *eb);
 
-void btrfs_set_lock_blocking(struct extent_buffer *eb);
-void btrfs_clear_lock_blocking(struct extent_buffer *eb);
+void btrfs_tree_read_lock(struct extent_buffer *eb);
+void btrfs_tree_read_unlock(struct extent_buffer *eb);
+void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb);
+void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw);
+void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw);
 void btrfs_assert_tree_locked(struct extent_buffer *eb);
+int btrfs_try_tree_read_lock(struct extent_buffer *eb);
+int btrfs_try_tree_write_lock(struct extent_buffer *eb);
+
+static inline void btrfs_tree_unlock_rw(struct extent_buffer *eb, int rw)
+{
+	if (rw == BTRFS_WRITE_LOCK || rw == BTRFS_WRITE_LOCK_BLOCKING)
+		btrfs_tree_unlock(eb);
+	else if (rw == BTRFS_READ_LOCK_BLOCKING)
+		btrfs_tree_read_unlock_blocking(eb);
+	else if (rw == BTRFS_READ_LOCK)
+		btrfs_tree_read_unlock(eb);
+	else
+		BUG();
+}
+
+static inline void btrfs_set_lock_blocking(struct extent_buffer *eb)
+{
+	btrfs_set_lock_blocking_rw(eb, BTRFS_WRITE_LOCK);
+}
+
+static inline void btrfs_clear_lock_blocking(struct extent_buffer *eb)
+{
+	btrfs_clear_lock_blocking_rw(eb, BTRFS_WRITE_LOCK_BLOCKING);
+}
 #endif

fs/btrfs/relocation.c

@@ -2955,7 +2955,8 @@ static int relocate_file_extent_cluster(struct inode *inode,
 			page_cache_sync_readahead(inode->i_mapping,
 						  ra, NULL, index,
 						  last_index + 1 - index);
-			page = grab_cache_page(inode->i_mapping, index);
+			page = find_or_create_page(inode->i_mapping, index,
+						   GFP_NOFS);
 			if (!page) {
 				btrfs_delalloc_release_metadata(inode,
 							PAGE_CACHE_SIZE);

fs/btrfs/struct-funcs.c

@@ -50,36 +50,22 @@ u##bits btrfs_##name(struct extent_buffer *eb,				\
 	unsigned long part_offset = (unsigned long)s;			\
 	unsigned long offset = part_offset + offsetof(type, member);	\
 	type *p;							\
-	/* ugly, but we want the fast path here */			\
-	if (eb->map_token && offset >= eb->map_start &&			\
-	    offset + sizeof(((type *)0)->member) <= eb->map_start +	\
-	    eb->map_len) {						\
-		p = (type *)(eb->kaddr + part_offset - eb->map_start);	\
-		return le##bits##_to_cpu(p->member);			\
-	}								\
-	{								\
-		int err;						\
-		char *map_token;					\
-		char *kaddr;						\
-		int unmap_on_exit = (eb->map_token == NULL);		\
-		unsigned long map_start;				\
-		unsigned long map_len;					\
-		u##bits res;						\
-		err = map_extent_buffer(eb, offset,			\
-				sizeof(((type *)0)->member),		\
-				&map_token, &kaddr,			\
-				&map_start, &map_len, KM_USER1);	\
-		if (err) {						\
-			__le##bits leres;				\
-			read_eb_member(eb, s, type, member, &leres);	\
-			return le##bits##_to_cpu(leres);		\
-		}							\
-		p = (type *)(kaddr + part_offset - map_start);		\
-		res = le##bits##_to_cpu(p->member);			\
-		if (unmap_on_exit)					\
-			unmap_extent_buffer(eb, map_token, KM_USER1);	\
-		return res;						\
-	}								\
+	int err;						\
+	char *kaddr;						\
+	unsigned long map_start;				\
+	unsigned long map_len;					\
+	u##bits res;						\
+	err = map_private_extent_buffer(eb, offset,		\
+			sizeof(((type *)0)->member),		\
+			&kaddr, &map_start, &map_len);		\
+	if (err) {						\
+		__le##bits leres;				\
+		read_eb_member(eb, s, type, member, &leres);	\
+		return le##bits##_to_cpu(leres);		\
+	}							\
+	p = (type *)(kaddr + part_offset - map_start);		\
+	res = le##bits##_to_cpu(p->member);			\
+	return res;						\
 }									\
 void btrfs_set_##name(struct extent_buffer *eb,				\
 				    type *s, u##bits val)		\
@@ -87,36 +73,21 @@ void btrfs_set_##name(struct extent_buffer *eb,				\
 	unsigned long part_offset = (unsigned long)s;			\
 	unsigned long offset = part_offset + offsetof(type, member);	\
 	type *p;							\
-	/* ugly, but we want the fast path here */			\
-	if (eb->map_token && offset >= eb->map_start &&			\
-	    offset + sizeof(((type *)0)->member) <= eb->map_start +	\
-	    eb->map_len) {						\
-		p = (type *)(eb->kaddr + part_offset - eb->map_start);	\
-		p->member = cpu_to_le##bits(val);			\
-		return;							\
-	}								\
-	{								\
-		int err;						\
-		char *map_token;					\
-		char *kaddr;						\
-		int unmap_on_exit = (eb->map_token == NULL);		\
-		unsigned long map_start;				\
-		unsigned long map_len;					\
-		err = map_extent_buffer(eb, offset,			\
-				sizeof(((type *)0)->member),		\
-				&map_token, &kaddr,			\
-				&map_start, &map_len, KM_USER1);	\
-		if (err) {						\
-			__le##bits val2;				\
-			val2 = cpu_to_le##bits(val);			\
-			write_eb_member(eb, s, type, member, &val2);	\
-			return;						\
-		}							\
-		p = (type *)(kaddr + part_offset - map_start);		\
-		p->member = cpu_to_le##bits(val);			\
-		if (unmap_on_exit)					\
-			unmap_extent_buffer(eb, map_token, KM_USER1);	\
-	}								\
+	int err;						\
+	char *kaddr;						\
+	unsigned long map_start;				\
+	unsigned long map_len;					\
+	err = map_private_extent_buffer(eb, offset,		\
+			sizeof(((type *)0)->member),		\
+			&kaddr, &map_start, &map_len);		\
+	if (err) {						\
+		__le##bits val2;				\
+		val2 = cpu_to_le##bits(val);			\
+		write_eb_member(eb, s, type, member, &val2);	\
+		return;						\
+	}							\
+	p = (type *)(kaddr + part_offset - map_start);		\
+	p->member = cpu_to_le##bits(val);			\
 }
 
 #include "ctree.h"
@@ -125,15 +96,6 @@ void btrfs_node_key(struct extent_buffer *eb,
 		    struct btrfs_disk_key *disk_key, int nr)
 {
 	unsigned long ptr = btrfs_node_key_ptr_offset(nr);
-	if (eb->map_token && ptr >= eb->map_start &&
-	    ptr + sizeof(*disk_key) <= eb->map_start + eb->map_len) {
-		memcpy(disk_key, eb->kaddr + ptr - eb->map_start,
-			sizeof(*disk_key));
-		return;
-	} else if (eb->map_token) {
-		unmap_extent_buffer(eb, eb->map_token, KM_USER1);
-		eb->map_token = NULL;
-	}
 	read_eb_member(eb, (struct btrfs_key_ptr *)ptr,
 		       struct btrfs_key_ptr, key, disk_key);
 }

fs/btrfs/transaction.c

@@ -260,7 +260,7 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 {
 	struct btrfs_trans_handle *h;
 	struct btrfs_transaction *cur_trans;
-	int retries = 0;
+	u64 num_bytes = 0;
 	int ret;
 
 	if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
@@ -274,6 +274,19 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 		h->block_rsv = NULL;
 		goto got_it;
 	}
+
+	/*
+	 * Do the reservation before we join the transaction so we can do all
+	 * the appropriate flushing if need be.
+	 */
+	if (num_items > 0 && root != root->fs_info->chunk_root) {
+		num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
+		ret = btrfs_block_rsv_add(NULL, root,
+					  &root->fs_info->trans_block_rsv,
+					  num_bytes);
+		if (ret)
+			return ERR_PTR(ret);
+	}
 again:
 	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 	if (!h)
@@ -310,24 +323,9 @@ again:
 		goto again;
 	}
 
-	if (num_items > 0) {
-		ret = btrfs_trans_reserve_metadata(h, root, num_items);
-		if (ret == -EAGAIN && !retries) {
-			retries++;
-			btrfs_commit_transaction(h, root);
-			goto again;
-		} else if (ret == -EAGAIN) {
-			/*
-			 * We have already retried and got EAGAIN, so really we
-			 * don't have space, so set ret to -ENOSPC.
-			 */
-			ret = -ENOSPC;
-		}
-
-		if (ret < 0) {
-			btrfs_end_transaction(h, root);
-			return ERR_PTR(ret);
-		}
+	if (num_bytes) {
+		h->block_rsv = &root->fs_info->trans_block_rsv;
+		h->bytes_reserved = num_bytes;
 	}
 
 got_it:
@@ -499,10 +497,17 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 	}
 
 	if (lock && cur_trans->blocked && !cur_trans->in_commit) {
-		if (throttle)
+		if (throttle) {
+			/*
+			 * We may race with somebody else here so end up having
+			 * to call end_transaction on ourselves again, so inc
+			 * our use_count.
+			 */
+			trans->use_count++;
 			return btrfs_commit_transaction(trans, root);
-		else
+		} else {
 			wake_up_process(info->transaction_kthread);
+		}
 	}
 
 	WARN_ON(cur_trans != info->running_transaction);

fs/btrfs/tree-log.c

@@ -1730,8 +1730,8 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
 				btrfs_read_buffer(next, ptr_gen);
 
 				btrfs_tree_lock(next);
-				clean_tree_block(trans, root, next);
 				btrfs_set_lock_blocking(next);
+				clean_tree_block(trans, root, next);
 				btrfs_wait_tree_block_writeback(next);
 				btrfs_tree_unlock(next);
 
@@ -1796,8 +1796,8 @@ static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
 				next = path->nodes[*level];
 
 				btrfs_tree_lock(next);
-				clean_tree_block(trans, root, next);
 				btrfs_set_lock_blocking(next);
+				clean_tree_block(trans, root, next);
 				btrfs_wait_tree_block_writeback(next);
 				btrfs_tree_unlock(next);
 
@@ -1864,8 +1864,8 @@ static int walk_log_tree(struct btrfs_trans_handle *trans,
 			next = path->nodes[orig_level];
 
 			btrfs_tree_lock(next);
-			clean_tree_block(trans, log, next);
 			btrfs_set_lock_blocking(next);
+			clean_tree_block(trans, log, next);
 			btrfs_wait_tree_block_writeback(next);
 			btrfs_tree_unlock(next);
 

fs/btrfs/volumes.c

@@ -3595,7 +3595,7 @@ int btrfs_read_sys_array(struct btrfs_root *root)
 	if (!sb)
 		return -ENOMEM;
 	btrfs_set_buffer_uptodate(sb);
-	btrfs_set_buffer_lockdep_class(sb, 0);
+	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
 
 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
 	array_size = btrfs_super_sys_array_size(super_copy);

fs/btrfs/xattr.c

@@ -102,43 +102,57 @@ static int do_setxattr(struct btrfs_trans_handle *trans,
 	if (!path)
 		return -ENOMEM;
 
-	/* first lets see if we already have this xattr */
-	di = btrfs_lookup_xattr(trans, root, path, btrfs_ino(inode), name,
-				strlen(name), -1);
-	if (IS_ERR(di)) {
-		ret = PTR_ERR(di);
-		goto out;
-	}
-
-	/* ok we already have this xattr, lets remove it */
-	if (di) {
-		/* if we want create only exit */
-		if (flags & XATTR_CREATE) {
-			ret = -EEXIST;
+	if (flags & XATTR_REPLACE) {
+		di = btrfs_lookup_xattr(trans, root, path, btrfs_ino(inode), name,
+					name_len, -1);
+		if (IS_ERR(di)) {
+			ret = PTR_ERR(di);
+			goto out;
+		} else if (!di) {
+			ret = -ENODATA;
 			goto out;
 		}
-
 		ret = btrfs_delete_one_dir_name(trans, root, path, di);
-		BUG_ON(ret);
+		if (ret)
+			goto out;
 		btrfs_release_path(path);
+	}
 
-		/* if we don't have a value then we are removing the xattr */
-		if (!value)
+again:
+	ret = btrfs_insert_xattr_item(trans, root, path, btrfs_ino(inode),
+				      name, name_len, value, size);
+	if (ret == -EEXIST) {
+		if (flags & XATTR_CREATE)
 			goto out;
-	} else {
+		/*
+		 * We can't use the path we already have since we won't have the
+		 * proper locking for a delete, so release the path and
+		 * re-lookup to delete the thing.
+		 */
 		btrfs_release_path(path);
+		di = btrfs_lookup_xattr(trans, root, path, btrfs_ino(inode),
+					name, name_len, -1);
+		if (IS_ERR(di)) {
+			ret = PTR_ERR(di);
+			goto out;
+		} else if (!di) {
+			/* Shouldn't happen but just in case... */
+			btrfs_release_path(path);
+			goto again;
+		}
 
-		if (flags & XATTR_REPLACE) {
-			/* we couldn't find the attr to replace */
-			ret = -ENODATA;
+		ret = btrfs_delete_one_dir_name(trans, root, path, di);
+		if (ret)
 			goto out;
+
+		/*
+		 * We have a value to set, so go back and try to insert it now.
+		 */
+		if (value) {
+			btrfs_release_path(path);
+			goto again;
 		}
 	}
-
-	/* ok we have to create a completely new xattr */
-	ret = btrfs_insert_xattr_item(trans, root, path, btrfs_ino(inode),
-				      name, name_len, value, size);
-	BUG_ON(ret);
 out:
 	btrfs_free_path(path);
 	return ret;