Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable

* git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable: (27 commits)
  Btrfs: add more error checking to btrfs_dirty_inode
  Btrfs: allow unaligned DIO
  Btrfs: drop verbose enospc printk
  Btrfs: Fix block generation verification race
  Btrfs: fix preallocation and nodatacow checks in O_DIRECT
  Btrfs: avoid ENOSPC errors in btrfs_dirty_inode
  Btrfs: move O_DIRECT space reservation to btrfs_direct_IO
  Btrfs: rework O_DIRECT enospc handling
  Btrfs: use async helpers for DIO write checksumming
  Btrfs: don't walk around with task->state != TASK_RUNNING
  Btrfs: do aio_write instead of write
  Btrfs: add basic DIO read/write support
  direct-io: do not merge logically non-contiguous requests
  direct-io: add a hook for the fs to provide its own submit_bio function
  fs: allow short direct-io reads to be completed via buffered IO
  Btrfs: Metadata ENOSPC handling for balance
  Btrfs: Pre-allocate space for data relocation
  Btrfs: Metadata ENOSPC handling for tree log
  Btrfs: Metadata reservation for orphan inodes
  Btrfs: Introduce global metadata reservation
  ...

fs/btrfs/async-thread.c

@@ -377,6 +377,7 @@ again:
 				if (!list_empty(&worker->pending) ||
 				    !list_empty(&worker->prio_pending)) {
 					spin_unlock_irq(&worker->lock);
+					set_current_state(TASK_RUNNING);
 					goto again;
 				}
 

fs/btrfs/btrfs_inode.h

@@ -137,8 +137,8 @@ struct btrfs_inode {
 	 * of extent items we've reserved metadata for.
 	 */
 	spinlock_t accounting_lock;
+	atomic_t outstanding_extents;
 	int reserved_extents;
-	int outstanding_extents;
 
 	/*
 	 * ordered_data_close is set by truncate when a file that used
@@ -151,6 +151,7 @@ struct btrfs_inode {
 	 * of these.
 	 */
 	unsigned ordered_data_close:1;
+	unsigned orphan_meta_reserved:1;
 	unsigned dummy_inode:1;
 
 	/*

fs/btrfs/ctree.c

@@ -280,7 +280,8 @@ int btrfs_block_can_be_shared(struct btrfs_root *root,
 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 				       struct btrfs_root *root,
 				       struct extent_buffer *buf,
-				       struct extent_buffer *cow)
+				       struct extent_buffer *cow,
+				       int *last_ref)
 {
 	u64 refs;
 	u64 owner;
@@ -366,6 +367,7 @@ static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 			BUG_ON(ret);
 		}
 		clean_tree_block(trans, root, buf);
+		*last_ref = 1;
 	}
 	return 0;
 }
@@ -392,6 +394,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	struct btrfs_disk_key disk_key;
 	struct extent_buffer *cow;
 	int level;
+	int last_ref = 0;
 	int unlock_orig = 0;
 	u64 parent_start;
 
@@ -442,7 +445,10 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 			    (unsigned long)btrfs_header_fsid(cow),
 			    BTRFS_FSID_SIZE);
 
-	update_ref_for_cow(trans, root, buf, cow);
+	update_ref_for_cow(trans, root, buf, cow, &last_ref);
+
+	if (root->ref_cows)
+		btrfs_reloc_cow_block(trans, root, buf, cow);
 
 	if (buf == root->node) {
 		WARN_ON(parent && parent != buf);
@@ -457,8 +463,8 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 		extent_buffer_get(cow);
 		spin_unlock(&root->node_lock);
 
-		btrfs_free_tree_block(trans, root, buf->start, buf->len,
-				parent_start, root->root_key.objectid, level);
+		btrfs_free_tree_block(trans, root, buf, parent_start,
+				      last_ref);
 		free_extent_buffer(buf);
 		add_root_to_dirty_list(root);
 	} else {
@@ -473,8 +479,8 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 		btrfs_set_node_ptr_generation(parent, parent_slot,
 					      trans->transid);
 		btrfs_mark_buffer_dirty(parent);
-		btrfs_free_tree_block(trans, root, buf->start, buf->len,
-				parent_start, root->root_key.objectid, level);
+		btrfs_free_tree_block(trans, root, buf, parent_start,
+				      last_ref);
 	}
 	if (unlock_orig)
 		btrfs_tree_unlock(buf);
@@ -949,6 +955,22 @@ int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 	return bin_search(eb, key, level, slot);
 }
 
+static void root_add_used(struct btrfs_root *root, u32 size)
+{
+	spin_lock(&root->accounting_lock);
+	btrfs_set_root_used(&root->root_item,
+			    btrfs_root_used(&root->root_item) + size);
+	spin_unlock(&root->accounting_lock);
+}
+
+static void root_sub_used(struct btrfs_root *root, u32 size)
+{
+	spin_lock(&root->accounting_lock);
+	btrfs_set_root_used(&root->root_item,
+			    btrfs_root_used(&root->root_item) - size);
+	spin_unlock(&root->accounting_lock);
+}
+
 /* given a node and slot number, this reads the blocks it points to.  The
  * extent buffer is returned with a reference taken (but unlocked).
  * NULL is returned on error.
@@ -1019,7 +1041,11 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		btrfs_tree_lock(child);
 		btrfs_set_lock_blocking(child);
 		ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
-		BUG_ON(ret);
+		if (ret) {
+			btrfs_tree_unlock(child);
+			free_extent_buffer(child);
+			goto enospc;
+		}
 
 		spin_lock(&root->node_lock);
 		root->node = child;
@@ -1034,11 +1060,12 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		btrfs_tree_unlock(mid);
 		/* once for the path */
 		free_extent_buffer(mid);
-		ret = btrfs_free_tree_block(trans, root, mid->start, mid->len,
-					    0, root->root_key.objectid, level);
+
+		root_sub_used(root, mid->len);
+		btrfs_free_tree_block(trans, root, mid, 0, 1);
 		/* once for the root ptr */
 		free_extent_buffer(mid);
-		return ret;
+		return 0;
 	}
 	if (btrfs_header_nritems(mid) >
 	    BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
@@ -1088,23 +1115,16 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		if (wret < 0 && wret != -ENOSPC)
 			ret = wret;
 		if (btrfs_header_nritems(right) == 0) {
-			u64 bytenr = right->start;
-			u32 blocksize = right->len;
-
 			clean_tree_block(trans, root, right);
 			btrfs_tree_unlock(right);
-			free_extent_buffer(right);
-			right = NULL;
 			wret = del_ptr(trans, root, path, level + 1, pslot +
 				       1);
 			if (wret)
 				ret = wret;
-			wret = btrfs_free_tree_block(trans, root,
-						     bytenr, blocksize, 0,
-						     root->root_key.objectid,
-						     level);
-			if (wret)
-				ret = wret;
+			root_sub_used(root, right->len);
+			btrfs_free_tree_block(trans, root, right, 0, 1);
+			free_extent_buffer(right);
+			right = NULL;
 		} else {
 			struct btrfs_disk_key right_key;
 			btrfs_node_key(right, &right_key, 0);
@@ -1136,21 +1156,15 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		BUG_ON(wret == 1);
 	}
 	if (btrfs_header_nritems(mid) == 0) {
-		/* we've managed to empty the middle node, drop it */
-		u64 bytenr = mid->start;
-		u32 blocksize = mid->len;
-
 		clean_tree_block(trans, root, mid);
 		btrfs_tree_unlock(mid);
-		free_extent_buffer(mid);
-		mid = NULL;
 		wret = del_ptr(trans, root, path, level + 1, pslot);
 		if (wret)
 			ret = wret;
-		wret = btrfs_free_tree_block(trans, root, bytenr, blocksize,
-					 0, root->root_key.objectid, level);
-		if (wret)
-			ret = wret;
+		root_sub_used(root, mid->len);
+		btrfs_free_tree_block(trans, root, mid, 0, 1);
+		free_extent_buffer(mid);
+		mid = NULL;
 	} else {
 		/* update the parent key to reflect our changes */
 		struct btrfs_disk_key mid_key;
@@ -1590,7 +1604,7 @@ read_block_for_search(struct btrfs_trans_handle *trans,
 	btrfs_release_path(NULL, p);
 
 	ret = -EAGAIN;
-	tmp = read_tree_block(root, blocknr, blocksize, gen);
+	tmp = read_tree_block(root, blocknr, blocksize, 0);
 	if (tmp) {
 		/*
 		 * If the read above didn't mark this buffer up to date,
@@ -1740,7 +1754,6 @@ again:
 					      p->nodes[level + 1],
 					      p->slots[level + 1], &b);
 			if (err) {
-				free_extent_buffer(b);
 				ret = err;
 				goto done;
 			}
@@ -2076,6 +2089,8 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 	if (IS_ERR(c))
 		return PTR_ERR(c);
 
+	root_add_used(root, root->nodesize);
+
 	memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
 	btrfs_set_header_nritems(c, 1);
 	btrfs_set_header_level(c, level);
@@ -2134,6 +2149,7 @@ static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
 	int nritems;
 
 	BUG_ON(!path->nodes[level]);
+	btrfs_assert_tree_locked(path->nodes[level]);
 	lower = path->nodes[level];
 	nritems = btrfs_header_nritems(lower);
 	BUG_ON(slot > nritems);
@@ -2202,6 +2218,8 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 	if (IS_ERR(split))
 		return PTR_ERR(split);
 
+	root_add_used(root, root->nodesize);
+
 	memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
 	btrfs_set_header_level(split, btrfs_header_level(c));
 	btrfs_set_header_bytenr(split, split->start);
@@ -2415,6 +2433,9 @@ static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
 
 	if (left_nritems)
 		btrfs_mark_buffer_dirty(left);
+	else
+		clean_tree_block(trans, root, left);
+
 	btrfs_mark_buffer_dirty(right);
 
 	btrfs_item_key(right, &disk_key, 0);
@@ -2660,6 +2681,8 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(left);
 	if (right_nritems)
 		btrfs_mark_buffer_dirty(right);
+	else
+		clean_tree_block(trans, root, right);
 
 	btrfs_item_key(right, &disk_key, 0);
 	wret = fixup_low_keys(trans, root, path, &disk_key, 1);
@@ -2669,8 +2692,6 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 	/* then fixup the leaf pointer in the path */
 	if (path->slots[0] < push_items) {
 		path->slots[0] += old_left_nritems;
-		if (btrfs_header_nritems(path->nodes[0]) == 0)
-			clean_tree_block(trans, root, path->nodes[0]);
 		btrfs_tree_unlock(path->nodes[0]);
 		free_extent_buffer(path->nodes[0]);
 		path->nodes[0] = left;
@@ -2932,10 +2953,10 @@ again:
 	right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
 					root->root_key.objectid,
 					&disk_key, 0, l->start, 0);
-	if (IS_ERR(right)) {
-		BUG_ON(1);
+	if (IS_ERR(right))
 		return PTR_ERR(right);
-	}
+
+	root_add_used(root, root->leafsize);
 
 	memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
 	btrfs_set_header_bytenr(right, right->start);
@@ -3054,7 +3075,8 @@ static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
 
 	btrfs_set_path_blocking(path);
 	ret = split_leaf(trans, root, &key, path, ins_len, 1);
-	BUG_ON(ret);
+	if (ret)
+		goto err;
 
 	path->keep_locks = 0;
 	btrfs_unlock_up_safe(path, 1);
@@ -3796,9 +3818,10 @@ static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
 	 */
 	btrfs_unlock_up_safe(path, 0);
 
-	ret = btrfs_free_tree_block(trans, root, leaf->start, leaf->len,
-				    0, root->root_key.objectid, 0);
-	return ret;
+	root_sub_used(root, leaf->len);
+
+	btrfs_free_tree_block(trans, root, leaf, 0, 1);
+	return 0;
 }
 /*
  * delete the item at the leaf level in path.  If that empties
@@ -3865,6 +3888,8 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		if (leaf == root->node) {
 			btrfs_set_header_level(leaf, 0);
 		} else {
+			btrfs_set_path_blocking(path);
+			clean_tree_block(trans, root, leaf);
 			ret = btrfs_del_leaf(trans, root, path, leaf);
 			BUG_ON(ret);
 		}

fs/btrfs/ctree.h

@@ -34,6 +34,7 @@
 
 struct btrfs_trans_handle;
 struct btrfs_transaction;
+struct btrfs_pending_snapshot;
 extern struct kmem_cache *btrfs_trans_handle_cachep;
 extern struct kmem_cache *btrfs_transaction_cachep;
 extern struct kmem_cache *btrfs_bit_radix_cachep;
@@ -663,6 +664,7 @@ struct btrfs_csum_item {
 #define BTRFS_BLOCK_GROUP_RAID1    (1 << 4)
 #define BTRFS_BLOCK_GROUP_DUP	   (1 << 5)
 #define BTRFS_BLOCK_GROUP_RAID10   (1 << 6)
+#define BTRFS_NR_RAID_TYPES	   5
 
 struct btrfs_block_group_item {
 	__le64 used;
@@ -674,42 +676,46 @@ struct btrfs_space_info {
 	u64 flags;
 
 	u64 total_bytes;	/* total bytes in the space */
-	u64 bytes_used;		/* total bytes used on disk */
+	u64 bytes_used;		/* total bytes used,
+				   this does't take mirrors into account */
 	u64 bytes_pinned;	/* total bytes pinned, will be freed when the
 				   transaction finishes */
 	u64 bytes_reserved;	/* total bytes the allocator has reserved for
 				   current allocations */
 	u64 bytes_readonly;	/* total bytes that are read only */
-	u64 bytes_super;	/* total bytes reserved for the super blocks */
-	u64 bytes_root;		/* the number of bytes needed to commit a
-				   transaction */
+
 	u64 bytes_may_use;	/* number of bytes that may be used for
 				   delalloc/allocations */
-	u64 bytes_delalloc;	/* number of bytes currently reserved for
-				   delayed allocation */
+	u64 disk_used;		/* total bytes used on disk */
 
 	int full;		/* indicates that we cannot allocate any more
 				   chunks for this space */
 	int force_alloc;	/* set if we need to force a chunk alloc for
 				   this space */
-	int force_delalloc;	/* make people start doing filemap_flush until
-				   we're under a threshold */
 
 	struct list_head list;
 
-	/* for controlling how we free up space for allocations */
-	wait_queue_head_t allocate_wait;
-	wait_queue_head_t flush_wait;
-	int allocating_chunk;
-	int flushing;
-
 	/* for block groups in our same type */
-	struct list_head block_groups;
+	struct list_head block_groups[BTRFS_NR_RAID_TYPES];
 	spinlock_t lock;
 	struct rw_semaphore groups_sem;
 	atomic_t caching_threads;
 };
 
+struct btrfs_block_rsv {
+	u64 size;
+	u64 reserved;
+	u64 freed[2];
+	struct btrfs_space_info *space_info;
+	struct list_head list;
+	spinlock_t lock;
+	atomic_t usage;
+	unsigned int priority:8;
+	unsigned int durable:1;
+	unsigned int refill_used:1;
+	unsigned int full:1;
+};
+
 /*
  * free clusters are used to claim free space in relatively large chunks,
  * allowing us to do less seeky writes.  They are used for all metadata
@@ -760,6 +766,7 @@ struct btrfs_block_group_cache {
 	spinlock_t lock;
 	u64 pinned;
 	u64 reserved;
+	u64 reserved_pinned;
 	u64 bytes_super;
 	u64 flags;
 	u64 sectorsize;
@@ -825,6 +832,22 @@ struct btrfs_fs_info {
 	/* logical->physical extent mapping */
 	struct btrfs_mapping_tree mapping_tree;
 
+	/* block reservation for extent, checksum and root tree */
+	struct btrfs_block_rsv global_block_rsv;
+	/* block reservation for delay allocation */
+	struct btrfs_block_rsv delalloc_block_rsv;
+	/* block reservation for metadata operations */
+	struct btrfs_block_rsv trans_block_rsv;
+	/* block reservation for chunk tree */
+	struct btrfs_block_rsv chunk_block_rsv;
+
+	struct btrfs_block_rsv empty_block_rsv;
+
+	/* list of block reservations that cross multiple transactions */
+	struct list_head durable_block_rsv_list;
+
+	struct mutex durable_block_rsv_mutex;
+
 	u64 generation;
 	u64 last_trans_committed;
 
@@ -927,7 +950,6 @@ struct btrfs_fs_info {
 	struct btrfs_workers endio_meta_write_workers;
 	struct btrfs_workers endio_write_workers;
 	struct btrfs_workers submit_workers;
-	struct btrfs_workers enospc_workers;
 	/*
 	 * fixup workers take dirty pages that didn't properly go through
 	 * the cow mechanism and make them safe to write.  It happens
@@ -943,6 +965,7 @@ struct btrfs_fs_info {
 	int do_barriers;
 	int closing;
 	int log_root_recovering;
+	int enospc_unlink;
 
 	u64 total_pinned;
 
@@ -1012,6 +1035,9 @@ struct btrfs_root {
 	struct completion kobj_unregister;
 	struct mutex objectid_mutex;
 
+	spinlock_t accounting_lock;
+	struct btrfs_block_rsv *block_rsv;
+
 	struct mutex log_mutex;
 	wait_queue_head_t log_writer_wait;
 	wait_queue_head_t log_commit_wait[2];
@@ -1043,7 +1069,6 @@ struct btrfs_root {
 	int ref_cows;
 	int track_dirty;
 	int in_radix;
-	int clean_orphans;
 
 	u64 defrag_trans_start;
 	struct btrfs_key defrag_progress;
@@ -1057,8 +1082,11 @@ struct btrfs_root {
 
 	struct list_head root_list;
 
-	spinlock_t list_lock;
+	spinlock_t orphan_lock;
 	struct list_head orphan_list;
+	struct btrfs_block_rsv *orphan_block_rsv;
+	int orphan_item_inserted;
+	int orphan_cleanup_state;
 
 	spinlock_t inode_lock;
 	/* red-black tree that keeps track of in-memory inodes */
@@ -1965,6 +1993,9 @@ void btrfs_put_block_group(struct btrfs_block_group_cache *cache);
 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root, unsigned long count);
 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
+int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 bytenr,
+			     u64 num_bytes, u64 *refs, u64 *flags);
 int btrfs_pin_extent(struct btrfs_root *root,
 		     u64 bytenr, u64 num, int reserved);
 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
@@ -1984,10 +2015,10 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 					u64 parent, u64 root_objectid,
 					struct btrfs_disk_key *key, int level,
 					u64 hint, u64 empty_size);
-int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root,
-			  u64 bytenr, u32 blocksize,
-			  u64 parent, u64 root_objectid, int level);
+void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct extent_buffer *buf,
+			   u64 parent, int last_ref);
 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
 					    struct btrfs_root *root,
 					    u64 bytenr, u32 blocksize,
@@ -2041,27 +2072,49 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 			   u64 size);
 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root, u64 group_start);
-int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
-				struct btrfs_block_group_cache *group);
-
 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags);
 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_reserve_metadata_space(struct btrfs_root *root, int num_items);
-int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items);
-int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
-					  struct inode *inode, int num_items);
-int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
-					struct inode *inode, int num_items);
-int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
-				u64 bytes);
-void btrfs_free_reserved_data_space(struct btrfs_root *root,
-				    struct inode *inode, u64 bytes);
-void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
-				 u64 bytes);
-void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
-			      u64 bytes);
+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, int *retries);
+void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root);
+int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
+				  struct inode *inode);
+void btrfs_orphan_release_metadata(struct inode *inode);
+int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
+				struct btrfs_pending_snapshot *pending);
+int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes);
+void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes);
+int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes);
+void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes);
+void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv);
+struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root);
+void btrfs_free_block_rsv(struct btrfs_root *root,
+			  struct btrfs_block_rsv *rsv);
+void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
+				 struct btrfs_block_rsv *rsv);
+int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_block_rsv *block_rsv,
+			u64 num_bytes, int *retries);
+int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  struct btrfs_block_rsv *block_rsv,
+			  u64 min_reserved, int min_factor);
+int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
+			    struct btrfs_block_rsv *dst_rsv,
+			    u64 num_bytes);
+void btrfs_block_rsv_release(struct btrfs_root *root,
+			     struct btrfs_block_rsv *block_rsv,
+			     u64 num_bytes);
+int btrfs_set_block_group_ro(struct btrfs_root *root,
+			     struct btrfs_block_group_cache *cache);
+int btrfs_set_block_group_rw(struct btrfs_root *root,
+			     struct btrfs_block_group_cache *cache);
 /* ctree.c */
 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 		     int level, int *slot);
@@ -2152,7 +2205,8 @@ static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf);
-int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref);
+int btrfs_drop_snapshot(struct btrfs_root *root,
+			struct btrfs_block_rsv *block_rsv, int update_ref);
 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
 			struct extent_buffer *node,
@@ -2245,6 +2299,12 @@ int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   const char *name, int name_len,
 			   u64 inode_objectid, u64 ref_objectid, u64 *index);
+struct btrfs_inode_ref *
+btrfs_lookup_inode_ref(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			const char *name, int name_len,
+			u64 inode_objectid, u64 ref_objectid, int mod);
 int btrfs_insert_empty_inode(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root,
 			     struct btrfs_path *path, u64 objectid);
@@ -2257,6 +2317,8 @@ int btrfs_del_csums(struct btrfs_trans_handle *trans,
 		    struct btrfs_root *root, u64 bytenr, u64 len);
 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
 			  struct bio *bio, u32 *dst);
+int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
+			      struct bio *bio, u64 logical_offset, u32 *dst);
 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root,
 			     u64 objectid, u64 pos,
@@ -2311,6 +2373,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 			       u32 min_type);
 
 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput);
+int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput);
 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
 			      struct extent_state **cached_state);
 int btrfs_writepages(struct address_space *mapping,
@@ -2349,10 +2412,20 @@ int btrfs_update_inode(struct btrfs_trans_handle *trans,
 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode);
 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode);
 void btrfs_orphan_cleanup(struct btrfs_root *root);
+void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
+				struct btrfs_pending_snapshot *pending,
+				u64 *bytes_to_reserve);
+void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
+				struct btrfs_pending_snapshot *pending);
+void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root);
 int btrfs_cont_expand(struct inode *inode, loff_t size);
 int btrfs_invalidate_inodes(struct btrfs_root *root);
 void btrfs_add_delayed_iput(struct inode *inode);
 void btrfs_run_delayed_iputs(struct btrfs_root *root);
+int btrfs_prealloc_file_range(struct inode *inode, int mode,
+			      u64 start, u64 num_bytes, u64 min_size,
+			      loff_t actual_len, u64 *alloc_hint);
 extern const struct dentry_operations btrfs_dentry_operations;
 
 /* ioctl.c */
@@ -2409,4 +2482,12 @@ int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 			    struct btrfs_root *root);
 int btrfs_recover_relocation(struct btrfs_root *root);
 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len);
+void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, struct extent_buffer *buf,
+			   struct extent_buffer *cow);
+void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
+			      struct btrfs_pending_snapshot *pending,
+			      u64 *bytes_to_reserve);
+void btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
+			      struct btrfs_pending_snapshot *pending);
 #endif

fs/btrfs/delayed-ref.c

@@ -318,107 +318,6 @@ out:
 	return ret;
 }
 
-/*
- * helper function to lookup reference count and flags of extent.
- *
- * the head node for delayed ref is used to store the sum of all the
- * reference count modifications queued up in the rbtree. the head
- * node may also store the extent flags to set. This way you can check
- * to see what the reference count and extent flags would be if all of
- * the delayed refs are not processed.
- */
-int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
-			     struct btrfs_root *root, u64 bytenr,
-			     u64 num_bytes, u64 *refs, u64 *flags)
-{
-	struct btrfs_delayed_ref_node *ref;
-	struct btrfs_delayed_ref_head *head;
-	struct btrfs_delayed_ref_root *delayed_refs;
-	struct btrfs_path *path;
-	struct btrfs_extent_item *ei;
-	struct extent_buffer *leaf;
-	struct btrfs_key key;
-	u32 item_size;
-	u64 num_refs;
-	u64 extent_flags;
-	int ret;
-
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
-
-	key.objectid = bytenr;
-	key.type = BTRFS_EXTENT_ITEM_KEY;
-	key.offset = num_bytes;
-	delayed_refs = &trans->transaction->delayed_refs;
-again:
-	ret = btrfs_search_slot(trans, root->fs_info->extent_root,
-				&key, path, 0, 0);
-	if (ret < 0)
-		goto out;
-
-	if (ret == 0) {
-		leaf = path->nodes[0];
-		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
-		if (item_size >= sizeof(*ei)) {
-			ei = btrfs_item_ptr(leaf, path->slots[0],
-					    struct btrfs_extent_item);
-			num_refs = btrfs_extent_refs(leaf, ei);
-			extent_flags = btrfs_extent_flags(leaf, ei);
-		} else {
-#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
-			struct btrfs_extent_item_v0 *ei0;
-			BUG_ON(item_size != sizeof(*ei0));
-			ei0 = btrfs_item_ptr(leaf, path->slots[0],
-					     struct btrfs_extent_item_v0);
-			num_refs = btrfs_extent_refs_v0(leaf, ei0);
-			/* FIXME: this isn't correct for data */
-			extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
-#else
-			BUG();
-#endif
-		}
-		BUG_ON(num_refs == 0);
-	} else {
-		num_refs = 0;
-		extent_flags = 0;
-		ret = 0;
-	}
-
-	spin_lock(&delayed_refs->lock);
-	ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
-	if (ref) {
-		head = btrfs_delayed_node_to_head(ref);
-		if (!mutex_trylock(&head->mutex)) {
-			atomic_inc(&ref->refs);
-			spin_unlock(&delayed_refs->lock);
-
-			btrfs_release_path(root->fs_info->extent_root, path);
-
-			mutex_lock(&head->mutex);
-			mutex_unlock(&head->mutex);
-			btrfs_put_delayed_ref(ref);
-			goto again;
-		}
-		if (head->extent_op && head->extent_op->update_flags)
-			extent_flags |= head->extent_op->flags_to_set;
-		else
-			BUG_ON(num_refs == 0);
-
-		num_refs += ref->ref_mod;
-		mutex_unlock(&head->mutex);
-	}
-	WARN_ON(num_refs == 0);
-	if (refs)
-		*refs = num_refs;
-	if (flags)
-		*flags = extent_flags;
-out:
-	spin_unlock(&delayed_refs->lock);
-	btrfs_free_path(path);
-	return ret;
-}
-
 /*
  * helper function to update an extent delayed ref in the
  * rbtree.  existing and update must both have the same

fs/btrfs/delayed-ref.h

@@ -167,9 +167,6 @@ int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
 struct btrfs_delayed_ref_head *
 btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
 int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr);
-int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
-			     struct btrfs_root *root, u64 bytenr,
-			     u64 num_bytes, u64 *refs, u64 *flags);
 int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans,
 			  u64 bytenr, u64 num_bytes, u64 orig_parent,
 			  u64 parent, u64 orig_ref_root, u64 ref_root,

fs/btrfs/disk-io.c

@@ -74,6 +74,11 @@ struct async_submit_bio {
 	int rw;
 	int mirror_num;
 	unsigned long bio_flags;
+	/*
+	 * bio_offset is optional, can be used if the pages in the bio
+	 * can't tell us where in the file the bio should go
+	 */
+	u64 bio_offset;
 	struct btrfs_work work;
 };
 
@@ -534,7 +539,8 @@ static void run_one_async_start(struct btrfs_work *work)
 	async = container_of(work, struct  async_submit_bio, work);
 	fs_info = BTRFS_I(async->inode)->root->fs_info;
 	async->submit_bio_start(async->inode, async->rw, async->bio,
-			       async->mirror_num, async->bio_flags);
+			       async->mirror_num, async->bio_flags,
+			       async->bio_offset);
 }
 
 static void run_one_async_done(struct btrfs_work *work)
@@ -556,7 +562,8 @@ static void run_one_async_done(struct btrfs_work *work)
 		wake_up(&fs_info->async_submit_wait);
 
 	async->submit_bio_done(async->inode, async->rw, async->bio,
-			       async->mirror_num, async->bio_flags);
+			       async->mirror_num, async->bio_flags,
+			       async->bio_offset);
 }
 
 static void run_one_async_free(struct btrfs_work *work)
@@ -570,6 +577,7 @@ static void run_one_async_free(struct btrfs_work *work)
 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 			int rw, struct bio *bio, int mirror_num,
 			unsigned long bio_flags,
+			u64 bio_offset,
 			extent_submit_bio_hook_t *submit_bio_start,
 			extent_submit_bio_hook_t *submit_bio_done)
 {
@@ -592,6 +600,7 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 
 	async->work.flags = 0;
 	async->bio_flags = bio_flags;
+	async->bio_offset = bio_offset;
 
 	atomic_inc(&fs_info->nr_async_submits);
 
@@ -627,7 +636,8 @@ static int btree_csum_one_bio(struct bio *bio)
 
 static int __btree_submit_bio_start(struct inode *inode, int rw,
 				    struct bio *bio, int mirror_num,
-				    unsigned long bio_flags)
+				    unsigned long bio_flags,
+				    u64 bio_offset)
 {
 	/*
 	 * when we're called for a write, we're already in the async
@@ -638,7 +648,8 @@ static int __btree_submit_bio_start(struct inode *inode, int rw,
 }
 
 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
-				 int mirror_num, unsigned long bio_flags)
+				 int mirror_num, unsigned long bio_flags,
+				 u64 bio_offset)
 {
 	/*
 	 * when we're called for a write, we're already in the async
@@ -648,7 +659,8 @@ static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 }
 
 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
-				 int mirror_num, unsigned long bio_flags)
+				 int mirror_num, unsigned long bio_flags,
+				 u64 bio_offset)
 {
 	int ret;
 
@@ -671,6 +683,7 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 	 */
 	return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 				   inode, rw, bio, mirror_num, 0,
+				   bio_offset,
 				   __btree_submit_bio_start,
 				   __btree_submit_bio_done);
 }
@@ -894,7 +907,8 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 	root->ref_cows = 0;
 	root->track_dirty = 0;
 	root->in_radix = 0;
-	root->clean_orphans = 0;
+	root->orphan_item_inserted = 0;
+	root->orphan_cleanup_state = 0;
 
 	root->fs_info = fs_info;
 	root->objectid = objectid;
@@ -903,13 +917,16 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 	root->name = NULL;
 	root->in_sysfs = 0;
 	root->inode_tree = RB_ROOT;
+	root->block_rsv = NULL;
+	root->orphan_block_rsv = NULL;
 
 	INIT_LIST_HEAD(&root->dirty_list);
 	INIT_LIST_HEAD(&root->orphan_list);
 	INIT_LIST_HEAD(&root->root_list);
 	spin_lock_init(&root->node_lock);
-	spin_lock_init(&root->list_lock);
+	spin_lock_init(&root->orphan_lock);
 	spin_lock_init(&root->inode_lock);
+	spin_lock_init(&root->accounting_lock);
 	mutex_init(&root->objectid_mutex);
 	mutex_init(&root->log_mutex);
 	init_waitqueue_head(&root->log_writer_wait);
@@ -968,42 +985,6 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
 	return 0;
 }
 
-int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
-			     struct btrfs_fs_info *fs_info)
-{
-	struct extent_buffer *eb;
-	struct btrfs_root *log_root_tree = fs_info->log_root_tree;
-	u64 start = 0;
-	u64 end = 0;
-	int ret;
-
-	if (!log_root_tree)
-		return 0;
-
-	while (1) {
-		ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
-				0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
-		if (ret)
-			break;
-
-		clear_extent_bits(&log_root_tree->dirty_log_pages, start, end,
-				  EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
-	}
-	eb = fs_info->log_root_tree->node;
-
-	WARN_ON(btrfs_header_level(eb) != 0);
-	WARN_ON(btrfs_header_nritems(eb) != 0);
-
-	ret = btrfs_free_reserved_extent(fs_info->tree_root,
-				eb->start, eb->len);
-	BUG_ON(ret);
-
-	free_extent_buffer(eb);
-	kfree(fs_info->log_root_tree);
-	fs_info->log_root_tree = NULL;
-	return 0;
-}
-
 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 					 struct btrfs_fs_info *fs_info)
 {
@@ -1191,19 +1172,23 @@ again:
 	if (root)
 		return root;
 
-	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
-	if (ret == 0)
-		ret = -ENOENT;
-	if (ret < 0)
-		return ERR_PTR(ret);
-
 	root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
 	if (IS_ERR(root))
 		return root;
 
-	WARN_ON(btrfs_root_refs(&root->root_item) == 0);
 	set_anon_super(&root->anon_super, NULL);
 
+	if (btrfs_root_refs(&root->root_item) == 0) {
+		ret = -ENOENT;
+		goto fail;
+	}
+
+	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
+	if (ret < 0)
+		goto fail;
+	if (ret == 0)
+		root->orphan_item_inserted = 1;
+
 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 	if (ret)
 		goto fail;
@@ -1212,10 +1197,9 @@ again:
 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
 				(unsigned long)root->root_key.objectid,
 				root);
-	if (ret == 0) {
+	if (ret == 0)
 		root->in_radix = 1;
-		root->clean_orphans = 1;
-	}
+
 	spin_unlock(&fs_info->fs_roots_radix_lock);
 	radix_tree_preload_end();
 	if (ret) {
@@ -1461,10 +1445,6 @@ static int cleaner_kthread(void *arg)
 	struct btrfs_root *root = arg;
 
 	do {
-		smp_mb();
-		if (root->fs_info->closing)
-			break;
-
 		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
 
 		if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
@@ -1477,11 +1457,9 @@ static int cleaner_kthread(void *arg)
 		if (freezing(current)) {
 			refrigerator();
 		} else {
-			smp_mb();
-			if (root->fs_info->closing)
-				break;
 			set_current_state(TASK_INTERRUPTIBLE);
-			schedule();
+			if (!kthread_should_stop())
+				schedule();
 			__set_current_state(TASK_RUNNING);
 		}
 	} while (!kthread_should_stop());
@@ -1493,36 +1471,40 @@ static int transaction_kthread(void *arg)
 	struct btrfs_root *root = arg;
 	struct btrfs_trans_handle *trans;
 	struct btrfs_transaction *cur;
+	u64 transid;
 	unsigned long now;
 	unsigned long delay;
 	int ret;
 
 	do {
-		smp_mb();
-		if (root->fs_info->closing)
-			break;
-
 		delay = HZ * 30;
 		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
 
-		mutex_lock(&root->fs_info->trans_mutex);
+		spin_lock(&root->fs_info->new_trans_lock);
 		cur = root->fs_info->running_transaction;
 		if (!cur) {
-			mutex_unlock(&root->fs_info->trans_mutex);
+			spin_unlock(&root->fs_info->new_trans_lock);
 			goto sleep;
 		}
 
 		now = get_seconds();
-		if (now < cur->start_time || now - cur->start_time < 30) {
-			mutex_unlock(&root->fs_info->trans_mutex);
+		if (!cur->blocked &&
+		    (now < cur->start_time || now - cur->start_time < 30)) {
+			spin_unlock(&root->fs_info->new_trans_lock);
 			delay = HZ * 5;
 			goto sleep;
 		}
-		mutex_unlock(&root->fs_info->trans_mutex);
-		trans = btrfs_start_transaction(root, 1);
-		ret = btrfs_commit_transaction(trans, root);
+		transid = cur->transid;
+		spin_unlock(&root->fs_info->new_trans_lock);
 
+		trans = btrfs_join_transaction(root, 1);
+		if (transid == trans->transid) {
+			ret = btrfs_commit_transaction(trans, root);
+			BUG_ON(ret);
+		} else {
+			btrfs_end_transaction(trans, root);
+		}
 sleep:
 		wake_up_process(root->fs_info->cleaner_kthread);
 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
@@ -1530,10 +1512,10 @@ sleep:
 		if (freezing(current)) {
 			refrigerator();
 		} else {
-			if (root->fs_info->closing)
-				break;
 			set_current_state(TASK_INTERRUPTIBLE);
-			schedule_timeout(delay);
+			if (!kthread_should_stop() &&
+			    !btrfs_transaction_blocked(root->fs_info))
+				schedule_timeout(delay);
 			__set_current_state(TASK_RUNNING);
 		}
 	} while (!kthread_should_stop());
@@ -1620,6 +1602,13 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
 	INIT_LIST_HEAD(&fs_info->space_info);
 	btrfs_mapping_init(&fs_info->mapping_tree);
+	btrfs_init_block_rsv(&fs_info->global_block_rsv);
+	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
+	btrfs_init_block_rsv(&fs_info->trans_block_rsv);
+	btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
+	btrfs_init_block_rsv(&fs_info->empty_block_rsv);
+	INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
+	mutex_init(&fs_info->durable_block_rsv_mutex);
 	atomic_set(&fs_info->nr_async_submits, 0);
 	atomic_set(&fs_info->async_delalloc_pages, 0);
 	atomic_set(&fs_info->async_submit_draining, 0);
@@ -1759,9 +1748,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 			   min_t(u64, fs_devices->num_devices,
 			   fs_info->thread_pool_size),
 			   &fs_info->generic_worker);
-	btrfs_init_workers(&fs_info->enospc_workers, "enospc",
-			   fs_info->thread_pool_size,
-			   &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
@@ -1809,7 +1795,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	btrfs_start_workers(&fs_info->endio_meta_workers, 1);
 	btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
 	btrfs_start_workers(&fs_info->endio_write_workers, 1);
-	btrfs_start_workers(&fs_info->enospc_workers, 1);
 
 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -1912,17 +1897,18 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 
 	csum_root->track_dirty = 1;
 
+	fs_info->generation = generation;
+	fs_info->last_trans_committed = generation;
+	fs_info->data_alloc_profile = (u64)-1;
+	fs_info->metadata_alloc_profile = (u64)-1;
+	fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
+
 	ret = btrfs_read_block_groups(extent_root);
 	if (ret) {
 		printk(KERN_ERR "Failed to read block groups: %d\n", ret);
 		goto fail_block_groups;
 	}
 
-	fs_info->generation = generation;
-	fs_info->last_trans_committed = generation;
-	fs_info->data_alloc_profile = (u64)-1;
-	fs_info->metadata_alloc_profile = (u64)-1;
-	fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
 					       "btrfs-cleaner");
 	if (IS_ERR(fs_info->cleaner_kthread))
@@ -1977,6 +1963,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	BUG_ON(ret);
 
 	if (!(sb->s_flags & MS_RDONLY)) {
+		ret = btrfs_cleanup_fs_roots(fs_info);
+		BUG_ON(ret);
+
 		ret = btrfs_recover_relocation(tree_root);
 		if (ret < 0) {
 			printk(KERN_WARNING
@@ -2040,7 +2029,6 @@ fail_sb_buffer:
 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
 	btrfs_stop_workers(&fs_info->endio_write_workers);
 	btrfs_stop_workers(&fs_info->submit_workers);
-	btrfs_stop_workers(&fs_info->enospc_workers);
 fail_iput:
 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
 	iput(fs_info->btree_inode);
@@ -2405,11 +2393,11 @@ int btrfs_commit_super(struct btrfs_root *root)
 	down_write(&root->fs_info->cleanup_work_sem);
 	up_write(&root->fs_info->cleanup_work_sem);
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_join_transaction(root, 1);
 	ret = btrfs_commit_transaction(trans, root);
 	BUG_ON(ret);
 	/* run commit again to drop the original snapshot */
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_join_transaction(root, 1);
 	btrfs_commit_transaction(trans, root);
 	ret = btrfs_write_and_wait_transaction(NULL, root);
 	BUG_ON(ret);
@@ -2426,15 +2414,15 @@ int close_ctree(struct btrfs_root *root)
 	fs_info->closing = 1;
 	smp_mb();
 
-	kthread_stop(root->fs_info->transaction_kthread);
-	kthread_stop(root->fs_info->cleaner_kthread);
-
 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
 		ret =  btrfs_commit_super(root);
 		if (ret)
 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
 	}
 
+	kthread_stop(root->fs_info->transaction_kthread);
+	kthread_stop(root->fs_info->cleaner_kthread);
+
 	fs_info->closing = 2;
 	smp_mb();
 
@@ -2473,7 +2461,6 @@ int close_ctree(struct btrfs_root *root)
 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
 	btrfs_stop_workers(&fs_info->endio_write_workers);
 	btrfs_stop_workers(&fs_info->submit_workers);
-	btrfs_stop_workers(&fs_info->enospc_workers);
 
 	btrfs_close_devices(fs_info->fs_devices);
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);

fs/btrfs/disk-io.h

@@ -87,7 +87,7 @@ int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 			int metadata);
 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 			int rw, struct bio *bio, int mirror_num,
-			unsigned long bio_flags,
+			unsigned long bio_flags, u64 bio_offset,
 			extent_submit_bio_hook_t *submit_bio_start,
 			extent_submit_bio_hook_t *submit_bio_done);
 
@@ -95,8 +95,6 @@ int btrfs_congested_async(struct btrfs_fs_info *info, int iodone);
 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info);
 int btrfs_write_tree_block(struct extent_buffer *buf);
 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf);
-int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
-			     struct btrfs_fs_info *fs_info);
 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 			     struct btrfs_fs_info *fs_info);
 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,

fs/btrfs/extent-tree.c

@@ -35,10 +35,9 @@
 
 static int update_block_group(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
-			      u64 bytenr, u64 num_bytes, int alloc,
-			      int mark_free);
-static int update_reserved_extents(struct btrfs_block_group_cache *cache,
-				   u64 num_bytes, int reserve);
+			      u64 bytenr, u64 num_bytes, int alloc);
+static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
+				 u64 num_bytes, int reserve, int sinfo);
 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root,
 				u64 bytenr, u64 num_bytes, u64 parent,
@@ -61,12 +60,6 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *extent_root, u64 alloc_bytes,
 			  u64 flags, int force);
-static int pin_down_bytes(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root,
-			  struct btrfs_path *path,
-			  u64 bytenr, u64 num_bytes,
-			  int is_data, int reserved,
-			  struct extent_buffer **must_clean);
 static int find_next_key(struct btrfs_path *path, int level,
 			 struct btrfs_key *key);
 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
@@ -91,8 +84,12 @@ void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 
 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 {
-	if (atomic_dec_and_test(&cache->count))
+	if (atomic_dec_and_test(&cache->count)) {
+		WARN_ON(cache->pinned > 0);
+		WARN_ON(cache->reserved > 0);
+		WARN_ON(cache->reserved_pinned > 0);
 		kfree(cache);
+	}
 }
 
 /*
@@ -319,7 +316,7 @@ static int caching_kthread(void *data)
 
 	exclude_super_stripes(extent_root, block_group);
 	spin_lock(&block_group->space_info->lock);
-	block_group->space_info->bytes_super += block_group->bytes_super;
+	block_group->space_info->bytes_readonly += block_group->bytes_super;
 	spin_unlock(&block_group->space_info->lock);
 
 	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
@@ -507,6 +504,9 @@ static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 	struct list_head *head = &info->space_info;
 	struct btrfs_space_info *found;
 
+	flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
+		 BTRFS_BLOCK_GROUP_METADATA;
+
 	rcu_read_lock();
 	list_for_each_entry_rcu(found, head, list) {
 		if (found->flags == flags) {
@@ -609,6 +609,113 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 	return ret;
 }
 
+/*
+ * helper function to lookup reference count and flags of extent.
+ *
+ * the head node for delayed ref is used to store the sum of all the
+ * reference count modifications queued up in the rbtree. the head
+ * node may also store the extent flags to set. This way you can check
+ * to see what the reference count and extent flags would be if all of
+ * the delayed refs are not processed.
+ */
+int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 bytenr,
+			     u64 num_bytes, u64 *refs, u64 *flags)
+{
+	struct btrfs_delayed_ref_head *head;
+	struct btrfs_delayed_ref_root *delayed_refs;
+	struct btrfs_path *path;
+	struct btrfs_extent_item *ei;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	u32 item_size;
+	u64 num_refs;
+	u64 extent_flags;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = num_bytes;
+	if (!trans) {
+		path->skip_locking = 1;
+		path->search_commit_root = 1;
+	}
+again:
+	ret = btrfs_search_slot(trans, root->fs_info->extent_root,
+				&key, path, 0, 0);
+	if (ret < 0)
+		goto out_free;
+
+	if (ret == 0) {
+		leaf = path->nodes[0];
+		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+		if (item_size >= sizeof(*ei)) {
+			ei = btrfs_item_ptr(leaf, path->slots[0],
+					    struct btrfs_extent_item);
+			num_refs = btrfs_extent_refs(leaf, ei);
+			extent_flags = btrfs_extent_flags(leaf, ei);
+		} else {
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+			struct btrfs_extent_item_v0 *ei0;
+			BUG_ON(item_size != sizeof(*ei0));
+			ei0 = btrfs_item_ptr(leaf, path->slots[0],
+					     struct btrfs_extent_item_v0);
+			num_refs = btrfs_extent_refs_v0(leaf, ei0);
+			/* FIXME: this isn't correct for data */
+			extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
+#else
+			BUG();
+#endif
+		}
+		BUG_ON(num_refs == 0);
+	} else {
+		num_refs = 0;
+		extent_flags = 0;
+		ret = 0;
+	}
+
+	if (!trans)
+		goto out;
+
+	delayed_refs = &trans->transaction->delayed_refs;
+	spin_lock(&delayed_refs->lock);
+	head = btrfs_find_delayed_ref_head(trans, bytenr);
+	if (head) {
+		if (!mutex_trylock(&head->mutex)) {
+			atomic_inc(&head->node.refs);
+			spin_unlock(&delayed_refs->lock);
+
+			btrfs_release_path(root->fs_info->extent_root, path);
+
+			mutex_lock(&head->mutex);
+			mutex_unlock(&head->mutex);
+			btrfs_put_delayed_ref(&head->node);
+			goto again;
+		}
+		if (head->extent_op && head->extent_op->update_flags)
+			extent_flags |= head->extent_op->flags_to_set;
+		else
+			BUG_ON(num_refs == 0);
+
+		num_refs += head->node.ref_mod;
+		mutex_unlock(&head->mutex);
+	}
+	spin_unlock(&delayed_refs->lock);
+out:
+	WARN_ON(num_refs == 0);
+	if (refs)
+		*refs = num_refs;
+	if (flags)
+		*flags = extent_flags;
+out_free:
+	btrfs_free_path(path);
+	return ret;
+}
+
 /*
  * Back reference rules.  Back refs have three main goals:
  *
@@ -1871,7 +1978,6 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-
 /* helper function to actually process a single delayed ref entry */
 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root,
@@ -1891,32 +1997,14 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 		BUG_ON(extent_op);
 		head = btrfs_delayed_node_to_head(node);
 		if (insert_reserved) {
-			int mark_free = 0;
-			struct extent_buffer *must_clean = NULL;
-
-			ret = pin_down_bytes(trans, root, NULL,
-					     node->bytenr, node->num_bytes,
-					     head->is_data, 1, &must_clean);
-			if (ret > 0)
-				mark_free = 1;
-
-			if (must_clean) {
-				clean_tree_block(NULL, root, must_clean);
-				btrfs_tree_unlock(must_clean);
-				free_extent_buffer(must_clean);
-			}
+			btrfs_pin_extent(root, node->bytenr,
+					 node->num_bytes, 1);
 			if (head->is_data) {
 				ret = btrfs_del_csums(trans, root,
 						      node->bytenr,
 						      node->num_bytes);
 				BUG_ON(ret);
 			}
-			if (mark_free) {
-				ret = btrfs_free_reserved_extent(root,
-							node->bytenr,
-							node->num_bytes);
-				BUG_ON(ret);
-			}
 		}
 		mutex_unlock(&head->mutex);
 		return 0;
@@ -2347,6 +2435,8 @@ int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
 		ret = 0;
 out:
 	btrfs_free_path(path);
+	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
+		WARN_ON(ret > 0);
 	return ret;
 }
 
@@ -2660,12 +2750,21 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 			     struct btrfs_space_info **space_info)
 {
 	struct btrfs_space_info *found;
+	int i;
+	int factor;
+
+	if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
+		     BTRFS_BLOCK_GROUP_RAID10))
+		factor = 2;
+	else
+		factor = 1;
 
 	found = __find_space_info(info, flags);
 	if (found) {
 		spin_lock(&found->lock);
 		found->total_bytes += total_bytes;
 		found->bytes_used += bytes_used;
+		found->disk_used += bytes_used * factor;
 		found->full = 0;
 		spin_unlock(&found->lock);
 		*space_info = found;
@@ -2675,18 +2774,20 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 	if (!found)
 		return -ENOMEM;
 
-	INIT_LIST_HEAD(&found->block_groups);
+	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
+		INIT_LIST_HEAD(&found->block_groups[i]);
 	init_rwsem(&found->groups_sem);
-	init_waitqueue_head(&found->flush_wait);
-	init_waitqueue_head(&found->allocate_wait);
 	spin_lock_init(&found->lock);
-	found->flags = flags;
+	found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
+				BTRFS_BLOCK_GROUP_SYSTEM |
+				BTRFS_BLOCK_GROUP_METADATA);
 	found->total_bytes = total_bytes;
 	found->bytes_used = bytes_used;
+	found->disk_used = bytes_used * factor;
 	found->bytes_pinned = 0;
 	found->bytes_reserved = 0;
 	found->bytes_readonly = 0;
-	found->bytes_delalloc = 0;
+	found->bytes_may_use = 0;
 	found->full = 0;
 	found->force_alloc = 0;
 	*space_info = found;
@@ -2711,19 +2812,6 @@ static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
 	}
 }
 
-static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
-{
-	spin_lock(&cache->space_info->lock);
-	spin_lock(&cache->lock);
-	if (!cache->ro) {
-		cache->space_info->bytes_readonly += cache->key.offset -
-					btrfs_block_group_used(&cache->item);
-		cache->ro = 1;
-	}
-	spin_unlock(&cache->lock);
-	spin_unlock(&cache->space_info->lock);
-}
-
 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
 {
 	u64 num_devices = root->fs_info->fs_devices->rw_devices;
@@ -2752,722 +2840,946 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
 	return flags;
 }
 
-static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
-{
-	struct btrfs_fs_info *info = root->fs_info;
-	u64 alloc_profile;
-
-	if (data) {
-		alloc_profile = info->avail_data_alloc_bits &
-			info->data_alloc_profile;
-		data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
-	} else if (root == root->fs_info->chunk_root) {
-		alloc_profile = info->avail_system_alloc_bits &
-			info->system_alloc_profile;
-		data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
-	} else {
-		alloc_profile = info->avail_metadata_alloc_bits &
-			info->metadata_alloc_profile;
-		data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
-	}
-
-	return btrfs_reduce_alloc_profile(root, data);
-}
-
-void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
+static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
 {
-	u64 alloc_target;
-
-	alloc_target = btrfs_get_alloc_profile(root, 1);
-	BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
-						       alloc_target);
+	if (flags & BTRFS_BLOCK_GROUP_DATA)
+		flags |= root->fs_info->avail_data_alloc_bits &
+			 root->fs_info->data_alloc_profile;
+	else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
+		flags |= root->fs_info->avail_system_alloc_bits &
+			 root->fs_info->system_alloc_profile;
+	else if (flags & BTRFS_BLOCK_GROUP_METADATA)
+		flags |= root->fs_info->avail_metadata_alloc_bits &
+			 root->fs_info->metadata_alloc_profile;
+	return btrfs_reduce_alloc_profile(root, flags);
 }
 
-static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
+static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
 {
-	u64 num_bytes;
-	int level;
-
-	level = BTRFS_MAX_LEVEL - 2;
-	/*
-	 * NOTE: these calculations are absolutely the worst possible case.
-	 * This assumes that _every_ item we insert will require a new leaf, and
-	 * that the tree has grown to its maximum level size.
-	 */
+	u64 flags;
 
-	/*
-	 * for every item we insert we could insert both an extent item and a
-	 * extent ref item.  Then for ever item we insert, we will need to cow
-	 * both the original leaf, plus the leaf to the left and right of it.
-	 *
-	 * Unless we are talking about the extent root, then we just want the
-	 * number of items * 2, since we just need the extent item plus its ref.
-	 */
-	if (root == root->fs_info->extent_root)
-		num_bytes = num_items * 2;
+	if (data)
+		flags = BTRFS_BLOCK_GROUP_DATA;
+	else if (root == root->fs_info->chunk_root)
+		flags = BTRFS_BLOCK_GROUP_SYSTEM;
 	else
-		num_bytes = (num_items + (2 * num_items)) * 3;
+		flags = BTRFS_BLOCK_GROUP_METADATA;
 
-	/*
-	 * num_bytes is total number of leaves we could need times the leaf
-	 * size, and then for every leaf we could end up cow'ing 2 nodes per
-	 * level, down to the leaf level.
-	 */
-	num_bytes = (num_bytes * root->leafsize) +
-		(num_bytes * (level * 2)) * root->nodesize;
+	return get_alloc_profile(root, flags);
+}
 
-	return num_bytes;
+void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
+{
+	BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
+						       BTRFS_BLOCK_GROUP_DATA);
 }
 
 /*
- * Unreserve metadata space for delalloc.  If we have less reserved credits than
- * we have extents, this function does nothing.
+ * This will check the space that the inode allocates from to make sure we have
+ * enough space for bytes.
  */
-int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
-					  struct inode *inode, int num_items)
+int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
 {
-	struct btrfs_fs_info *info = root->fs_info;
-	struct btrfs_space_info *meta_sinfo;
-	u64 num_bytes;
-	u64 alloc_target;
-	bool bug = false;
+	struct btrfs_space_info *data_sinfo;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	u64 used;
+	int ret = 0, committed = 0;
 
-	/* get the space info for where the metadata will live */
-	alloc_target = btrfs_get_alloc_profile(root, 0);
-	meta_sinfo = __find_space_info(info, alloc_target);
+	/* make sure bytes are sectorsize aligned */
+	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 
-	num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
-					   num_items);
+	data_sinfo = BTRFS_I(inode)->space_info;
+	if (!data_sinfo)
+		goto alloc;
 
-	spin_lock(&meta_sinfo->lock);
-	spin_lock(&BTRFS_I(inode)->accounting_lock);
-	if (BTRFS_I(inode)->reserved_extents <=
-	    BTRFS_I(inode)->outstanding_extents) {
-		spin_unlock(&BTRFS_I(inode)->accounting_lock);
-		spin_unlock(&meta_sinfo->lock);
-		return 0;
-	}
-	spin_unlock(&BTRFS_I(inode)->accounting_lock);
+again:
+	/* make sure we have enough space to handle the data first */
+	spin_lock(&data_sinfo->lock);
+	used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
+		data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
+		data_sinfo->bytes_may_use;
+
+	if (used + bytes > data_sinfo->total_bytes) {
+		struct btrfs_trans_handle *trans;
 
-	BTRFS_I(inode)->reserved_extents -= num_items;
-	BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
+		/*
+		 * if we don't have enough free bytes in this space then we need
+		 * to alloc a new chunk.
+		 */
+		if (!data_sinfo->full) {
+			u64 alloc_target;
 
-	if (meta_sinfo->bytes_delalloc < num_bytes) {
-		bug = true;
-		meta_sinfo->bytes_delalloc = 0;
-	} else {
-		meta_sinfo->bytes_delalloc -= num_bytes;
-	}
-	spin_unlock(&meta_sinfo->lock);
+			data_sinfo->force_alloc = 1;
+			spin_unlock(&data_sinfo->lock);
+alloc:
+			alloc_target = btrfs_get_alloc_profile(root, 1);
+			trans = btrfs_join_transaction(root, 1);
+			if (IS_ERR(trans))
+				return PTR_ERR(trans);
 
-	BUG_ON(bug);
+			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+					     bytes + 2 * 1024 * 1024,
+					     alloc_target, 0);
+			btrfs_end_transaction(trans, root);
+			if (ret < 0)
+				return ret;
 
-	return 0;
-}
+			if (!data_sinfo) {
+				btrfs_set_inode_space_info(root, inode);
+				data_sinfo = BTRFS_I(inode)->space_info;
+			}
+			goto again;
+		}
+		spin_unlock(&data_sinfo->lock);
 
-static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
-{
-	u64 thresh;
+		/* commit the current transaction and try again */
+		if (!committed && !root->fs_info->open_ioctl_trans) {
+			committed = 1;
+			trans = btrfs_join_transaction(root, 1);
+			if (IS_ERR(trans))
+				return PTR_ERR(trans);
+			ret = btrfs_commit_transaction(trans, root);
+			if (ret)
+				return ret;
+			goto again;
+		}
 
-	thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
-		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
-		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
-		meta_sinfo->bytes_may_use;
+#if 0 /* I hope we never need this code again, just in case */
+		printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
+		       "%llu bytes_reserved, " "%llu bytes_pinned, "
+		       "%llu bytes_readonly, %llu may use %llu total\n",
+		       (unsigned long long)bytes,
+		       (unsigned long long)data_sinfo->bytes_used,
+		       (unsigned long long)data_sinfo->bytes_reserved,
+		       (unsigned long long)data_sinfo->bytes_pinned,
+		       (unsigned long long)data_sinfo->bytes_readonly,
+		       (unsigned long long)data_sinfo->bytes_may_use,
+		       (unsigned long long)data_sinfo->total_bytes);
+#endif
+		return -ENOSPC;
+	}
+	data_sinfo->bytes_may_use += bytes;
+	BTRFS_I(inode)->reserved_bytes += bytes;
+	spin_unlock(&data_sinfo->lock);
 
-	thresh = meta_sinfo->total_bytes - thresh;
-	thresh *= 80;
-	do_div(thresh, 100);
-	if (thresh <= meta_sinfo->bytes_delalloc)
-		meta_sinfo->force_delalloc = 1;
-	else
-		meta_sinfo->force_delalloc = 0;
+	return 0;
 }
 
-struct async_flush {
-	struct btrfs_root *root;
-	struct btrfs_space_info *info;
-	struct btrfs_work work;
-};
-
-static noinline void flush_delalloc_async(struct btrfs_work *work)
+/*
+ * called when we are clearing an delalloc extent from the
+ * inode's io_tree or there was an error for whatever reason
+ * after calling btrfs_check_data_free_space
+ */
+void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
 {
-	struct async_flush *async;
-	struct btrfs_root *root;
-	struct btrfs_space_info *info;
-
-	async = container_of(work, struct async_flush, work);
-	root = async->root;
-	info = async->info;
-
-	btrfs_start_delalloc_inodes(root, 0);
-	wake_up(&info->flush_wait);
-	btrfs_wait_ordered_extents(root, 0, 0);
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_space_info *data_sinfo;
 
-	spin_lock(&info->lock);
-	info->flushing = 0;
-	spin_unlock(&info->lock);
-	wake_up(&info->flush_wait);
+	/* make sure bytes are sectorsize aligned */
+	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 
-	kfree(async);
+	data_sinfo = BTRFS_I(inode)->space_info;
+	spin_lock(&data_sinfo->lock);
+	data_sinfo->bytes_may_use -= bytes;
+	BTRFS_I(inode)->reserved_bytes -= bytes;
+	spin_unlock(&data_sinfo->lock);
 }
 
-static void wait_on_flush(struct btrfs_space_info *info)
+static void force_metadata_allocation(struct btrfs_fs_info *info)
 {
-	DEFINE_WAIT(wait);
-	u64 used;
-
-	while (1) {
-		prepare_to_wait(&info->flush_wait, &wait,
-				TASK_UNINTERRUPTIBLE);
-		spin_lock(&info->lock);
-		if (!info->flushing) {
-			spin_unlock(&info->lock);
-			break;
-		}
+	struct list_head *head = &info->space_info;
+	struct btrfs_space_info *found;
 
-		used = info->bytes_used + info->bytes_reserved +
-			info->bytes_pinned + info->bytes_readonly +
-			info->bytes_super + info->bytes_root +
-			info->bytes_may_use + info->bytes_delalloc;
-		if (used < info->total_bytes) {
-			spin_unlock(&info->lock);
-			break;
-		}
-		spin_unlock(&info->lock);
-		schedule();
+	rcu_read_lock();
+	list_for_each_entry_rcu(found, head, list) {
+		if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
+			found->force_alloc = 1;
 	}
-	finish_wait(&info->flush_wait, &wait);
+	rcu_read_unlock();
 }
 
-static void flush_delalloc(struct btrfs_root *root,
-				 struct btrfs_space_info *info)
+static int should_alloc_chunk(struct btrfs_space_info *sinfo,
+			      u64 alloc_bytes)
 {
-	struct async_flush *async;
-	bool wait = false;
-
-	spin_lock(&info->lock);
-
-	if (!info->flushing)
-		info->flushing = 1;
-	else
-		wait = true;
-
-	spin_unlock(&info->lock);
-
-	if (wait) {
-		wait_on_flush(info);
-		return;
-	}
-
-	async = kzalloc(sizeof(*async), GFP_NOFS);
-	if (!async)
-		goto flush;
-
-	async->root = root;
-	async->info = info;
-	async->work.func = flush_delalloc_async;
+	u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
 
-	btrfs_queue_worker(&root->fs_info->enospc_workers,
-			   &async->work);
-	wait_on_flush(info);
-	return;
+	if (sinfo->bytes_used + sinfo->bytes_reserved +
+	    alloc_bytes + 256 * 1024 * 1024 < num_bytes)
+		return 0;
 
-flush:
-	btrfs_start_delalloc_inodes(root, 0);
-	btrfs_wait_ordered_extents(root, 0, 0);
+	if (sinfo->bytes_used + sinfo->bytes_reserved +
+	    alloc_bytes < div_factor(num_bytes, 8))
+		return 0;
 
-	spin_lock(&info->lock);
-	info->flushing = 0;
-	spin_unlock(&info->lock);
-	wake_up(&info->flush_wait);
+	return 1;
 }
 
-static int maybe_allocate_chunk(struct btrfs_root *root,
-				 struct btrfs_space_info *info)
+static int do_chunk_alloc(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *extent_root, u64 alloc_bytes,
+			  u64 flags, int force)
 {
-	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
-	struct btrfs_trans_handle *trans;
-	bool wait = false;
+	struct btrfs_space_info *space_info;
+	struct btrfs_fs_info *fs_info = extent_root->fs_info;
 	int ret = 0;
-	u64 min_metadata;
-	u64 free_space;
 
-	free_space = btrfs_super_total_bytes(disk_super);
-	/*
-	 * we allow the metadata to grow to a max of either 10gb or 5% of the
-	 * space in the volume.
-	 */
-	min_metadata = min((u64)10 * 1024 * 1024 * 1024,
-			     div64_u64(free_space * 5, 100));
-	if (info->total_bytes >= min_metadata) {
-		spin_unlock(&info->lock);
-		return 0;
-	}
+	mutex_lock(&fs_info->chunk_mutex);
 
-	if (info->full) {
-		spin_unlock(&info->lock);
-		return 0;
+	flags = btrfs_reduce_alloc_profile(extent_root, flags);
+
+	space_info = __find_space_info(extent_root->fs_info, flags);
+	if (!space_info) {
+		ret = update_space_info(extent_root->fs_info, flags,
+					0, 0, &space_info);
+		BUG_ON(ret);
 	}
+	BUG_ON(!space_info);
 
-	if (!info->allocating_chunk) {
-		info->force_alloc = 1;
-		info->allocating_chunk = 1;
-	} else {
-		wait = true;
+	spin_lock(&space_info->lock);
+	if (space_info->force_alloc)
+		force = 1;
+	if (space_info->full) {
+		spin_unlock(&space_info->lock);
+		goto out;
 	}
 
-	spin_unlock(&info->lock);
-
-	if (wait) {
-		wait_event(info->allocate_wait,
-			   !info->allocating_chunk);
-		return 1;
+	if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
+		spin_unlock(&space_info->lock);
+		goto out;
 	}
+	spin_unlock(&space_info->lock);
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans) {
-		ret = -ENOMEM;
-		goto out;
+	/*
+	 * if we're doing a data chunk, go ahead and make sure that
+	 * we keep a reasonable number of metadata chunks allocated in the
+	 * FS as well.
+	 */
+	if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
+		fs_info->data_chunk_allocations++;
+		if (!(fs_info->data_chunk_allocations %
+		      fs_info->metadata_ratio))
+			force_metadata_allocation(fs_info);
 	}
 
-	ret = do_chunk_alloc(trans, root->fs_info->extent_root,
-			     4096 + 2 * 1024 * 1024,
-			     info->flags, 0);
-	btrfs_end_transaction(trans, root);
+	ret = btrfs_alloc_chunk(trans, extent_root, flags);
+	spin_lock(&space_info->lock);
 	if (ret)
-		goto out;
+		space_info->full = 1;
+	else
+		ret = 1;
+	space_info->force_alloc = 0;
+	spin_unlock(&space_info->lock);
 out:
-	spin_lock(&info->lock);
-	info->allocating_chunk = 0;
-	spin_unlock(&info->lock);
-	wake_up(&info->allocate_wait);
+	mutex_unlock(&extent_root->fs_info->chunk_mutex);
+	return ret;
+}
 
-	if (ret)
+static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct btrfs_space_info *sinfo, u64 num_bytes)
+{
+	int ret;
+	int end_trans = 0;
+
+	if (sinfo->full)
 		return 0;
-	return 1;
+
+	spin_lock(&sinfo->lock);
+	ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
+	spin_unlock(&sinfo->lock);
+	if (!ret)
+		return 0;
+
+	if (!trans) {
+		trans = btrfs_join_transaction(root, 1);
+		BUG_ON(IS_ERR(trans));
+		end_trans = 1;
+	}
+
+	ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+			     num_bytes + 2 * 1024 * 1024,
+			     get_alloc_profile(root, sinfo->flags), 0);
+
+	if (end_trans)
+		btrfs_end_transaction(trans, root);
+
+	return ret == 1 ? 1 : 0;
 }
 
 /*
- * Reserve metadata space for delalloc.
+ * shrink metadata reservation for delalloc
  */
-int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
-					struct inode *inode, int num_items)
+static int shrink_delalloc(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, u64 to_reclaim)
+{
+	struct btrfs_block_rsv *block_rsv;
+	u64 reserved;
+	u64 max_reclaim;
+	u64 reclaimed = 0;
+	int pause = 1;
+	int ret;
+
+	block_rsv = &root->fs_info->delalloc_block_rsv;
+	spin_lock(&block_rsv->lock);
+	reserved = block_rsv->reserved;
+	spin_unlock(&block_rsv->lock);
+
+	if (reserved == 0)
+		return 0;
+
+	max_reclaim = min(reserved, to_reclaim);
+
+	while (1) {
+		ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
+		if (!ret) {
+			__set_current_state(TASK_INTERRUPTIBLE);
+			schedule_timeout(pause);
+			pause <<= 1;
+			if (pause > HZ / 10)
+				pause = HZ / 10;
+		} else {
+			pause = 1;
+		}
+
+		spin_lock(&block_rsv->lock);
+		if (reserved > block_rsv->reserved)
+			reclaimed = reserved - block_rsv->reserved;
+		reserved = block_rsv->reserved;
+		spin_unlock(&block_rsv->lock);
+
+		if (reserved == 0 || reclaimed >= max_reclaim)
+			break;
+
+		if (trans && trans->transaction->blocked)
+			return -EAGAIN;
+	}
+	return reclaimed >= to_reclaim;
+}
+
+static int should_retry_reserve(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct btrfs_block_rsv *block_rsv,
+				u64 num_bytes, int *retries)
 {
-	struct btrfs_fs_info *info = root->fs_info;
-	struct btrfs_space_info *meta_sinfo;
-	u64 num_bytes;
-	u64 used;
-	u64 alloc_target;
-	int flushed = 0;
-	int force_delalloc;
+	struct btrfs_space_info *space_info = block_rsv->space_info;
+	int ret;
 
-	/* get the space info for where the metadata will live */
-	alloc_target = btrfs_get_alloc_profile(root, 0);
-	meta_sinfo = __find_space_info(info, alloc_target);
+	if ((*retries) > 2)
+		return -ENOSPC;
 
-	num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
-					   num_items);
-again:
-	spin_lock(&meta_sinfo->lock);
+	ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
+	if (ret)
+		return 1;
 
-	force_delalloc = meta_sinfo->force_delalloc;
+	if (trans && trans->transaction->in_commit)
+		return -ENOSPC;
 
-	if (unlikely(!meta_sinfo->bytes_root))
-		meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+	ret = shrink_delalloc(trans, root, num_bytes);
+	if (ret)
+		return ret;
 
-	if (!flushed)
-		meta_sinfo->bytes_delalloc += num_bytes;
+	spin_lock(&space_info->lock);
+	if (space_info->bytes_pinned < num_bytes)
+		ret = 1;
+	spin_unlock(&space_info->lock);
+	if (ret)
+		return -ENOSPC;
 
-	used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
-		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
-		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
-		meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+	(*retries)++;
 
-	if (used > meta_sinfo->total_bytes) {
-		flushed++;
+	if (trans)
+		return -EAGAIN;
 
-		if (flushed == 1) {
-			if (maybe_allocate_chunk(root, meta_sinfo))
-				goto again;
-			flushed++;
+	trans = btrfs_join_transaction(root, 1);
+	BUG_ON(IS_ERR(trans));
+	ret = btrfs_commit_transaction(trans, root);
+	BUG_ON(ret);
+
+	return 1;
+}
+
+static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
+				  u64 num_bytes)
+{
+	struct btrfs_space_info *space_info = block_rsv->space_info;
+	u64 unused;
+	int ret = -ENOSPC;
+
+	spin_lock(&space_info->lock);
+	unused = space_info->bytes_used + space_info->bytes_reserved +
+		 space_info->bytes_pinned + space_info->bytes_readonly;
+
+	if (unused < space_info->total_bytes)
+		unused = space_info->total_bytes - unused;
+	else
+		unused = 0;
+
+	if (unused >= num_bytes) {
+		if (block_rsv->priority >= 10) {
+			space_info->bytes_reserved += num_bytes;
+			ret = 0;
 		} else {
-			spin_unlock(&meta_sinfo->lock);
+			if ((unused + block_rsv->reserved) *
+			    block_rsv->priority >=
+			    (num_bytes + block_rsv->reserved) * 10) {
+				space_info->bytes_reserved += num_bytes;
+				ret = 0;
+			}
 		}
+	}
+	spin_unlock(&space_info->lock);
 
-		if (flushed == 2) {
-			filemap_flush(inode->i_mapping);
-			goto again;
-		} else if (flushed == 3) {
-			flush_delalloc(root, meta_sinfo);
-			goto again;
+	return ret;
+}
+
+static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
+					     struct btrfs_root *root)
+{
+	struct btrfs_block_rsv *block_rsv;
+	if (root->ref_cows)
+		block_rsv = trans->block_rsv;
+	else
+		block_rsv = root->block_rsv;
+
+	if (!block_rsv)
+		block_rsv = &root->fs_info->empty_block_rsv;
+
+	return block_rsv;
+}
+
+static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
+			       u64 num_bytes)
+{
+	int ret = -ENOSPC;
+	spin_lock(&block_rsv->lock);
+	if (block_rsv->reserved >= num_bytes) {
+		block_rsv->reserved -= num_bytes;
+		if (block_rsv->reserved < block_rsv->size)
+			block_rsv->full = 0;
+		ret = 0;
+	}
+	spin_unlock(&block_rsv->lock);
+	return ret;
+}
+
+static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
+				u64 num_bytes, int update_size)
+{
+	spin_lock(&block_rsv->lock);
+	block_rsv->reserved += num_bytes;
+	if (update_size)
+		block_rsv->size += num_bytes;
+	else if (block_rsv->reserved >= block_rsv->size)
+		block_rsv->full = 1;
+	spin_unlock(&block_rsv->lock);
+}
+
+void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
+			     struct btrfs_block_rsv *dest, u64 num_bytes)
+{
+	struct btrfs_space_info *space_info = block_rsv->space_info;
+
+	spin_lock(&block_rsv->lock);
+	if (num_bytes == (u64)-1)
+		num_bytes = block_rsv->size;
+	block_rsv->size -= num_bytes;
+	if (block_rsv->reserved >= block_rsv->size) {
+		num_bytes = block_rsv->reserved - block_rsv->size;
+		block_rsv->reserved = block_rsv->size;
+		block_rsv->full = 1;
+	} else {
+		num_bytes = 0;
+	}
+	spin_unlock(&block_rsv->lock);
+
+	if (num_bytes > 0) {
+		if (dest) {
+			block_rsv_add_bytes(dest, num_bytes, 0);
+		} else {
+			spin_lock(&space_info->lock);
+			space_info->bytes_reserved -= num_bytes;
+			spin_unlock(&space_info->lock);
 		}
-		spin_lock(&meta_sinfo->lock);
-		meta_sinfo->bytes_delalloc -= num_bytes;
-		spin_unlock(&meta_sinfo->lock);
-		printk(KERN_ERR "enospc, has %d, reserved %d\n",
-		       BTRFS_I(inode)->outstanding_extents,
-		       BTRFS_I(inode)->reserved_extents);
-		dump_space_info(meta_sinfo, 0, 0);
-		return -ENOSPC;
 	}
+}
 
-	BTRFS_I(inode)->reserved_extents += num_items;
-	check_force_delalloc(meta_sinfo);
-	spin_unlock(&meta_sinfo->lock);
+static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
+				   struct btrfs_block_rsv *dst, u64 num_bytes)
+{
+	int ret;
 
-	if (!flushed && force_delalloc)
-		filemap_flush(inode->i_mapping);
+	ret = block_rsv_use_bytes(src, num_bytes);
+	if (ret)
+		return ret;
 
+	block_rsv_add_bytes(dst, num_bytes, 1);
 	return 0;
 }
 
-/*
- * unreserve num_items number of items worth of metadata space.  This needs to
- * be paired with btrfs_reserve_metadata_space.
- *
- * NOTE: if you have the option, run this _AFTER_ you do a
- * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
- * oprations which will result in more used metadata, so we want to make sure we
- * can do that without issue.
- */
-int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
+void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
 {
-	struct btrfs_fs_info *info = root->fs_info;
-	struct btrfs_space_info *meta_sinfo;
-	u64 num_bytes;
+	memset(rsv, 0, sizeof(*rsv));
+	spin_lock_init(&rsv->lock);
+	atomic_set(&rsv->usage, 1);
+	rsv->priority = 6;
+	INIT_LIST_HEAD(&rsv->list);
+}
+
+struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
+{
+	struct btrfs_block_rsv *block_rsv;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 	u64 alloc_target;
-	bool bug = false;
 
-	/* get the space info for where the metadata will live */
-	alloc_target = btrfs_get_alloc_profile(root, 0);
-	meta_sinfo = __find_space_info(info, alloc_target);
+	block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
+	if (!block_rsv)
+		return NULL;
 
-	num_bytes = calculate_bytes_needed(root, num_items);
+	btrfs_init_block_rsv(block_rsv);
 
-	spin_lock(&meta_sinfo->lock);
-	if (meta_sinfo->bytes_may_use < num_bytes) {
-		bug = true;
-		meta_sinfo->bytes_may_use = 0;
-	} else {
-		meta_sinfo->bytes_may_use -= num_bytes;
-	}
-	spin_unlock(&meta_sinfo->lock);
+	alloc_target = btrfs_get_alloc_profile(root, 0);
+	block_rsv->space_info = __find_space_info(fs_info,
+						  BTRFS_BLOCK_GROUP_METADATA);
 
-	BUG_ON(bug);
+	return block_rsv;
+}
 
-	return 0;
+void btrfs_free_block_rsv(struct btrfs_root *root,
+			  struct btrfs_block_rsv *rsv)
+{
+	if (rsv && atomic_dec_and_test(&rsv->usage)) {
+		btrfs_block_rsv_release(root, rsv, (u64)-1);
+		if (!rsv->durable)
+			kfree(rsv);
+	}
 }
 
 /*
- * Reserve some metadata space for use.  We'll calculate the worste case number
- * of bytes that would be needed to modify num_items number of items.  If we
- * have space, fantastic, if not, you get -ENOSPC.  Please call
- * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
- * items you reserved, since whatever metadata you needed should have already
- * been allocated.
- *
- * This will commit the transaction to make more space if we don't have enough
- * metadata space.  THe only time we don't do this is if we're reserving space
- * inside of a transaction, then we will just return -ENOSPC and it is the
- * callers responsibility to handle it properly.
+ * make the block_rsv struct be able to capture freed space.
+ * the captured space will re-add to the the block_rsv struct
+ * after transaction commit
  */
-int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
+void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
+				 struct btrfs_block_rsv *block_rsv)
 {
-	struct btrfs_fs_info *info = root->fs_info;
-	struct btrfs_space_info *meta_sinfo;
-	u64 num_bytes;
-	u64 used;
-	u64 alloc_target;
-	int retries = 0;
+	block_rsv->durable = 1;
+	mutex_lock(&fs_info->durable_block_rsv_mutex);
+	list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
+	mutex_unlock(&fs_info->durable_block_rsv_mutex);
+}
 
-	/* get the space info for where the metadata will live */
-	alloc_target = btrfs_get_alloc_profile(root, 0);
-	meta_sinfo = __find_space_info(info, alloc_target);
+int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_block_rsv *block_rsv,
+			u64 num_bytes, int *retries)
+{
+	int ret;
 
-	num_bytes = calculate_bytes_needed(root, num_items);
+	if (num_bytes == 0)
+		return 0;
 again:
-	spin_lock(&meta_sinfo->lock);
+	ret = reserve_metadata_bytes(block_rsv, num_bytes);
+	if (!ret) {
+		block_rsv_add_bytes(block_rsv, num_bytes, 1);
+		return 0;
+	}
 
-	if (unlikely(!meta_sinfo->bytes_root))
-		meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+	ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
+	if (ret > 0)
+		goto again;
+
+	return ret;
+}
 
-	if (!retries)
-		meta_sinfo->bytes_may_use += num_bytes;
+int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  struct btrfs_block_rsv *block_rsv,
+			  u64 min_reserved, int min_factor)
+{
+	u64 num_bytes = 0;
+	int commit_trans = 0;
+	int ret = -ENOSPC;
 
-	used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
-		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
-		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
-		meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+	if (!block_rsv)
+		return 0;
 
-	if (used > meta_sinfo->total_bytes) {
-		retries++;
-		if (retries == 1) {
-			if (maybe_allocate_chunk(root, meta_sinfo))
-				goto again;
-			retries++;
-		} else {
-			spin_unlock(&meta_sinfo->lock);
-		}
+	spin_lock(&block_rsv->lock);
+	if (min_factor > 0)
+		num_bytes = div_factor(block_rsv->size, min_factor);
+	if (min_reserved > num_bytes)
+		num_bytes = min_reserved;
 
-		if (retries == 2) {
-			flush_delalloc(root, meta_sinfo);
-			goto again;
+	if (block_rsv->reserved >= num_bytes) {
+		ret = 0;
+	} else {
+		num_bytes -= block_rsv->reserved;
+		if (block_rsv->durable &&
+		    block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
+			commit_trans = 1;
+	}
+	spin_unlock(&block_rsv->lock);
+	if (!ret)
+		return 0;
+
+	if (block_rsv->refill_used) {
+		ret = reserve_metadata_bytes(block_rsv, num_bytes);
+		if (!ret) {
+			block_rsv_add_bytes(block_rsv, num_bytes, 0);
+			return 0;
 		}
-		spin_lock(&meta_sinfo->lock);
-		meta_sinfo->bytes_may_use -= num_bytes;
-		spin_unlock(&meta_sinfo->lock);
+	}
 
-		dump_space_info(meta_sinfo, 0, 0);
-		return -ENOSPC;
+	if (commit_trans) {
+		if (trans)
+			return -EAGAIN;
+
+		trans = btrfs_join_transaction(root, 1);
+		BUG_ON(IS_ERR(trans));
+		ret = btrfs_commit_transaction(trans, root);
+		return 0;
 	}
 
-	check_force_delalloc(meta_sinfo);
-	spin_unlock(&meta_sinfo->lock);
+	WARN_ON(1);
+	printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
+		block_rsv->size, block_rsv->reserved,
+		block_rsv->freed[0], block_rsv->freed[1]);
 
-	return 0;
+	return -ENOSPC;
+}
+
+int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
+			    struct btrfs_block_rsv *dst_rsv,
+			    u64 num_bytes)
+{
+	return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
+}
+
+void btrfs_block_rsv_release(struct btrfs_root *root,
+			     struct btrfs_block_rsv *block_rsv,
+			     u64 num_bytes)
+{
+	struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
+	if (global_rsv->full || global_rsv == block_rsv ||
+	    block_rsv->space_info != global_rsv->space_info)
+		global_rsv = NULL;
+	block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
 }
 
 /*
- * This will check the space that the inode allocates from to make sure we have
- * enough space for bytes.
+ * helper to calculate size of global block reservation.
+ * the desired value is sum of space used by extent tree,
+ * checksum tree and root tree
  */
-int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
-				u64 bytes)
+static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
 {
-	struct btrfs_space_info *data_sinfo;
-	u64 used;
-	int ret = 0, committed = 0, flushed = 0;
+	struct btrfs_space_info *sinfo;
+	u64 num_bytes;
+	u64 meta_used;
+	u64 data_used;
+	int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
+#if 0
+	/*
+	 * per tree used space accounting can be inaccuracy, so we
+	 * can't rely on it.
+	 */
+	spin_lock(&fs_info->extent_root->accounting_lock);
+	num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
+	spin_unlock(&fs_info->extent_root->accounting_lock);
 
-	/* make sure bytes are sectorsize aligned */
-	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
+	spin_lock(&fs_info->csum_root->accounting_lock);
+	num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
+	spin_unlock(&fs_info->csum_root->accounting_lock);
 
-	data_sinfo = BTRFS_I(inode)->space_info;
-	if (!data_sinfo)
-		goto alloc;
+	spin_lock(&fs_info->tree_root->accounting_lock);
+	num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
+	spin_unlock(&fs_info->tree_root->accounting_lock);
+#endif
+	sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
+	spin_lock(&sinfo->lock);
+	data_used = sinfo->bytes_used;
+	spin_unlock(&sinfo->lock);
 
-again:
-	/* make sure we have enough space to handle the data first */
-	spin_lock(&data_sinfo->lock);
-	used = data_sinfo->bytes_used + data_sinfo->bytes_delalloc +
-		data_sinfo->bytes_reserved + data_sinfo->bytes_pinned +
-		data_sinfo->bytes_readonly + data_sinfo->bytes_may_use +
-		data_sinfo->bytes_super;
+	sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
+	spin_lock(&sinfo->lock);
+	meta_used = sinfo->bytes_used;
+	spin_unlock(&sinfo->lock);
 
-	if (used + bytes > data_sinfo->total_bytes) {
-		struct btrfs_trans_handle *trans;
+	num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
+		    csum_size * 2;
+	num_bytes += div64_u64(data_used + meta_used, 50);
 
-		if (!flushed) {
-			spin_unlock(&data_sinfo->lock);
-			flush_delalloc(root, data_sinfo);
-			flushed = 1;
-			goto again;
-		}
+	if (num_bytes * 3 > meta_used)
+		num_bytes = div64_u64(meta_used, 3);
 
-		/*
-		 * if we don't have enough free bytes in this space then we need
-		 * to alloc a new chunk.
-		 */
-		if (!data_sinfo->full) {
-			u64 alloc_target;
+	return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
+}
 
-			data_sinfo->force_alloc = 1;
-			spin_unlock(&data_sinfo->lock);
-alloc:
-			alloc_target = btrfs_get_alloc_profile(root, 1);
-			trans = btrfs_start_transaction(root, 1);
-			if (!trans)
-				return -ENOMEM;
+static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
+	struct btrfs_space_info *sinfo = block_rsv->space_info;
+	u64 num_bytes;
 
-			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
-					     bytes + 2 * 1024 * 1024,
-					     alloc_target, 0);
-			btrfs_end_transaction(trans, root);
-			if (ret)
-				return ret;
+	num_bytes = calc_global_metadata_size(fs_info);
 
-			if (!data_sinfo) {
-				btrfs_set_inode_space_info(root, inode);
-				data_sinfo = BTRFS_I(inode)->space_info;
-			}
-			goto again;
-		}
-		spin_unlock(&data_sinfo->lock);
+	spin_lock(&block_rsv->lock);
+	spin_lock(&sinfo->lock);
 
-		/* commit the current transaction and try again */
-		if (!committed && !root->fs_info->open_ioctl_trans) {
-			committed = 1;
-			trans = btrfs_join_transaction(root, 1);
-			if (!trans)
-				return -ENOMEM;
-			ret = btrfs_commit_transaction(trans, root);
-			if (ret)
-				return ret;
-			goto again;
-		}
+	block_rsv->size = num_bytes;
 
-		printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
-		       ", %llu bytes_used, %llu bytes_reserved, "
-		       "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
-		       "%llu total\n", (unsigned long long)bytes,
-		       (unsigned long long)data_sinfo->bytes_delalloc,
-		       (unsigned long long)data_sinfo->bytes_used,
-		       (unsigned long long)data_sinfo->bytes_reserved,
-		       (unsigned long long)data_sinfo->bytes_pinned,
-		       (unsigned long long)data_sinfo->bytes_readonly,
-		       (unsigned long long)data_sinfo->bytes_may_use,
-		       (unsigned long long)data_sinfo->total_bytes);
-		return -ENOSPC;
+	num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
+		    sinfo->bytes_reserved + sinfo->bytes_readonly;
+
+	if (sinfo->total_bytes > num_bytes) {
+		num_bytes = sinfo->total_bytes - num_bytes;
+		block_rsv->reserved += num_bytes;
+		sinfo->bytes_reserved += num_bytes;
 	}
-	data_sinfo->bytes_may_use += bytes;
-	BTRFS_I(inode)->reserved_bytes += bytes;
-	spin_unlock(&data_sinfo->lock);
 
-	return 0;
+	if (block_rsv->reserved >= block_rsv->size) {
+		num_bytes = block_rsv->reserved - block_rsv->size;
+		sinfo->bytes_reserved -= num_bytes;
+		block_rsv->reserved = block_rsv->size;
+		block_rsv->full = 1;
+	}
+#if 0
+	printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
+		block_rsv->size, block_rsv->reserved);
+#endif
+	spin_unlock(&sinfo->lock);
+	spin_unlock(&block_rsv->lock);
 }
 
-/*
- * if there was an error for whatever reason after calling
- * btrfs_check_data_free_space, call this so we can cleanup the counters.
- */
-void btrfs_free_reserved_data_space(struct btrfs_root *root,
-				    struct inode *inode, u64 bytes)
+static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
 {
-	struct btrfs_space_info *data_sinfo;
+	struct btrfs_space_info *space_info;
 
-	/* make sure bytes are sectorsize aligned */
-	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
+	space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
+	fs_info->chunk_block_rsv.space_info = space_info;
+	fs_info->chunk_block_rsv.priority = 10;
 
-	data_sinfo = BTRFS_I(inode)->space_info;
-	spin_lock(&data_sinfo->lock);
-	data_sinfo->bytes_may_use -= bytes;
-	BTRFS_I(inode)->reserved_bytes -= bytes;
-	spin_unlock(&data_sinfo->lock);
+	space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
+	fs_info->global_block_rsv.space_info = space_info;
+	fs_info->global_block_rsv.priority = 10;
+	fs_info->global_block_rsv.refill_used = 1;
+	fs_info->delalloc_block_rsv.space_info = space_info;
+	fs_info->trans_block_rsv.space_info = space_info;
+	fs_info->empty_block_rsv.space_info = space_info;
+	fs_info->empty_block_rsv.priority = 10;
+
+	fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
+	fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
+	fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
+	fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
+	fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
+
+	btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
+
+	btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
+
+	update_global_block_rsv(fs_info);
 }
 
-/* called when we are adding a delalloc extent to the inode's io_tree */
-void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
-				  u64 bytes)
+static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
 {
-	struct btrfs_space_info *data_sinfo;
+	block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
+	WARN_ON(fs_info->delalloc_block_rsv.size > 0);
+	WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
+	WARN_ON(fs_info->trans_block_rsv.size > 0);
+	WARN_ON(fs_info->trans_block_rsv.reserved > 0);
+	WARN_ON(fs_info->chunk_block_rsv.size > 0);
+	WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
+}
 
-	/* get the space info for where this inode will be storing its data */
-	data_sinfo = BTRFS_I(inode)->space_info;
+static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
+{
+	return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
+		3 * num_items;
+}
 
-	/* make sure we have enough space to handle the data first */
-	spin_lock(&data_sinfo->lock);
-	data_sinfo->bytes_delalloc += bytes;
+int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 int num_items, int *retries)
+{
+	u64 num_bytes;
+	int ret;
 
-	/*
-	 * we are adding a delalloc extent without calling
-	 * btrfs_check_data_free_space first.  This happens on a weird
-	 * writepage condition, but shouldn't hurt our accounting
-	 */
-	if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
-		data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
-		BTRFS_I(inode)->reserved_bytes = 0;
-	} else {
-		data_sinfo->bytes_may_use -= bytes;
-		BTRFS_I(inode)->reserved_bytes -= bytes;
-	}
+	if (num_items == 0 || root->fs_info->chunk_root == root)
+		return 0;
 
-	spin_unlock(&data_sinfo->lock);
+	num_bytes = calc_trans_metadata_size(root, num_items);
+	ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
+				  num_bytes, retries);
+	if (!ret) {
+		trans->bytes_reserved += num_bytes;
+		trans->block_rsv = &root->fs_info->trans_block_rsv;
+	}
+	return ret;
 }
 
-/* called when we are clearing an delalloc extent from the inode's io_tree */
-void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
-			      u64 bytes)
+void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root)
 {
-	struct btrfs_space_info *info;
+	if (!trans->bytes_reserved)
+		return;
 
-	info = BTRFS_I(inode)->space_info;
+	BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
+	btrfs_block_rsv_release(root, trans->block_rsv,
+				trans->bytes_reserved);
+	trans->bytes_reserved = 0;
+}
 
-	spin_lock(&info->lock);
-	info->bytes_delalloc -= bytes;
-	spin_unlock(&info->lock);
+int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
+				  struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
+	struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
+
+	/*
+	 * one for deleting orphan item, one for updating inode and
+	 * two for calling btrfs_truncate_inode_items.
+	 *
+	 * btrfs_truncate_inode_items is a delete operation, it frees
+	 * more space than it uses in most cases. So two units of
+	 * metadata space should be enough for calling it many times.
+	 * If all of the metadata space is used, we can commit
+	 * transaction and use space it freed.
+	 */
+	u64 num_bytes = calc_trans_metadata_size(root, 4);
+	return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
 }
 
-static void force_metadata_allocation(struct btrfs_fs_info *info)
+void btrfs_orphan_release_metadata(struct inode *inode)
 {
-	struct list_head *head = &info->space_info;
-	struct btrfs_space_info *found;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	u64 num_bytes = calc_trans_metadata_size(root, 4);
+	btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
+}
 
-	rcu_read_lock();
-	list_for_each_entry_rcu(found, head, list) {
-		if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
-			found->force_alloc = 1;
-	}
-	rcu_read_unlock();
+int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
+				struct btrfs_pending_snapshot *pending)
+{
+	struct btrfs_root *root = pending->root;
+	struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
+	struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
+	/*
+	 * two for root back/forward refs, two for directory entries
+	 * and one for root of the snapshot.
+	 */
+	u64 num_bytes = calc_trans_metadata_size(root, 5);
+	dst_rsv->space_info = src_rsv->space_info;
+	return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
 }
 
-static int do_chunk_alloc(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *extent_root, u64 alloc_bytes,
-			  u64 flags, int force)
+static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
 {
-	struct btrfs_space_info *space_info;
-	struct btrfs_fs_info *fs_info = extent_root->fs_info;
-	u64 thresh;
-	int ret = 0;
+	return num_bytes >>= 3;
+}
 
-	mutex_lock(&fs_info->chunk_mutex);
+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 retries = 0;
+	int ret;
 
-	flags = btrfs_reduce_alloc_profile(extent_root, flags);
+	if (btrfs_transaction_in_commit(root->fs_info))
+		schedule_timeout(1);
 
-	space_info = __find_space_info(extent_root->fs_info, flags);
-	if (!space_info) {
-		ret = update_space_info(extent_root->fs_info, flags,
-					0, 0, &space_info);
-		BUG_ON(ret);
+	num_bytes = ALIGN(num_bytes, root->sectorsize);
+again:
+	spin_lock(&BTRFS_I(inode)->accounting_lock);
+	nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
+	if (nr_extents > BTRFS_I(inode)->reserved_extents) {
+		nr_extents -= BTRFS_I(inode)->reserved_extents;
+		to_reserve = calc_trans_metadata_size(root, nr_extents);
+	} else {
+		nr_extents = 0;
+		to_reserve = 0;
 	}
-	BUG_ON(!space_info);
 
-	spin_lock(&space_info->lock);
-	if (space_info->force_alloc)
-		force = 1;
-	if (space_info->full) {
-		spin_unlock(&space_info->lock);
-		goto out;
+	to_reserve += calc_csum_metadata_size(inode, num_bytes);
+	ret = reserve_metadata_bytes(block_rsv, to_reserve);
+	if (ret) {
+		spin_unlock(&BTRFS_I(inode)->accounting_lock);
+		ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
+					   &retries);
+		if (ret > 0)
+			goto again;
+		return ret;
 	}
 
-	thresh = space_info->total_bytes - space_info->bytes_readonly;
-	thresh = div_factor(thresh, 8);
-	if (!force &&
-	   (space_info->bytes_used + space_info->bytes_pinned +
-	    space_info->bytes_reserved + alloc_bytes) < thresh) {
-		spin_unlock(&space_info->lock);
-		goto out;
-	}
-	spin_unlock(&space_info->lock);
+	BTRFS_I(inode)->reserved_extents += nr_extents;
+	atomic_inc(&BTRFS_I(inode)->outstanding_extents);
+	spin_unlock(&BTRFS_I(inode)->accounting_lock);
 
-	/*
-	 * if we're doing a data chunk, go ahead and make sure that
-	 * we keep a reasonable number of metadata chunks allocated in the
-	 * FS as well.
-	 */
-	if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
-		fs_info->data_chunk_allocations++;
-		if (!(fs_info->data_chunk_allocations %
-		      fs_info->metadata_ratio))
-			force_metadata_allocation(fs_info);
+	block_rsv_add_bytes(block_rsv, to_reserve, 1);
+
+	if (block_rsv->size > 512 * 1024 * 1024)
+		shrink_delalloc(NULL, root, to_reserve);
+
+	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;
+
+	num_bytes = ALIGN(num_bytes, root->sectorsize);
+	atomic_dec(&BTRFS_I(inode)->outstanding_extents);
+
+	spin_lock(&BTRFS_I(inode)->accounting_lock);
+	nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
+	if (nr_extents < BTRFS_I(inode)->reserved_extents) {
+		nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
+		BTRFS_I(inode)->reserved_extents -= nr_extents;
+	} else {
+		nr_extents = 0;
 	}
+	spin_unlock(&BTRFS_I(inode)->accounting_lock);
 
-	ret = btrfs_alloc_chunk(trans, extent_root, flags);
-	spin_lock(&space_info->lock);
+	to_free = calc_csum_metadata_size(inode, num_bytes);
+	if (nr_extents > 0)
+		to_free += calc_trans_metadata_size(root, nr_extents);
+
+	btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
+				to_free);
+}
+
+int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
+{
+	int ret;
+
+	ret = btrfs_check_data_free_space(inode, num_bytes);
 	if (ret)
-		space_info->full = 1;
-	space_info->force_alloc = 0;
-	spin_unlock(&space_info->lock);
-out:
-	mutex_unlock(&extent_root->fs_info->chunk_mutex);
-	return ret;
+		return ret;
+
+	ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
+	if (ret) {
+		btrfs_free_reserved_data_space(inode, num_bytes);
+		return ret;
+	}
+
+	return 0;
+}
+
+void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
+{
+	btrfs_delalloc_release_metadata(inode, num_bytes);
+	btrfs_free_reserved_data_space(inode, num_bytes);
 }
 
 static int update_block_group(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
-			      u64 bytenr, u64 num_bytes, int alloc,
-			      int mark_free)
+			      u64 bytenr, u64 num_bytes, int alloc)
 {
 	struct btrfs_block_group_cache *cache;
 	struct btrfs_fs_info *info = root->fs_info;
+	int factor;
 	u64 total = num_bytes;
 	u64 old_val;
 	u64 byte_in_group;
@@ -3486,6 +3798,12 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 		cache = btrfs_lookup_block_group(info, bytenr);
 		if (!cache)
 			return -1;
+		if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
+				    BTRFS_BLOCK_GROUP_RAID1 |
+				    BTRFS_BLOCK_GROUP_RAID10))
+			factor = 2;
+		else
+			factor = 1;
 		byte_in_group = bytenr - cache->key.objectid;
 		WARN_ON(byte_in_group > cache->key.offset);
 
@@ -3498,31 +3816,24 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 			old_val += num_bytes;
 			btrfs_set_block_group_used(&cache->item, old_val);
 			cache->reserved -= num_bytes;
-			cache->space_info->bytes_used += num_bytes;
 			cache->space_info->bytes_reserved -= num_bytes;
-			if (cache->ro)
-				cache->space_info->bytes_readonly -= num_bytes;
+			cache->space_info->bytes_used += num_bytes;
+			cache->space_info->disk_used += num_bytes * factor;
 			spin_unlock(&cache->lock);
 			spin_unlock(&cache->space_info->lock);
 		} else {
 			old_val -= num_bytes;
-			cache->space_info->bytes_used -= num_bytes;
-			if (cache->ro)
-				cache->space_info->bytes_readonly += num_bytes;
 			btrfs_set_block_group_used(&cache->item, old_val);
+			cache->pinned += num_bytes;
+			cache->space_info->bytes_pinned += num_bytes;
+			cache->space_info->bytes_used -= num_bytes;
+			cache->space_info->disk_used -= num_bytes * factor;
 			spin_unlock(&cache->lock);
 			spin_unlock(&cache->space_info->lock);
-			if (mark_free) {
-				int ret;
 
-				ret = btrfs_discard_extent(root, bytenr,
-							   num_bytes);
-				WARN_ON(ret);
-
-				ret = btrfs_add_free_space(cache, bytenr,
-							   num_bytes);
-				WARN_ON(ret);
-			}
+			set_extent_dirty(info->pinned_extents,
+					 bytenr, bytenr + num_bytes - 1,
+					 GFP_NOFS | __GFP_NOFAIL);
 		}
 		btrfs_put_block_group(cache);
 		total -= num_bytes;
@@ -3546,18 +3857,10 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
 	return bytenr;
 }
 
-/*
- * this function must be called within transaction
- */
-int btrfs_pin_extent(struct btrfs_root *root,
-		     u64 bytenr, u64 num_bytes, int reserved)
+static int pin_down_extent(struct btrfs_root *root,
+			   struct btrfs_block_group_cache *cache,
+			   u64 bytenr, u64 num_bytes, int reserved)
 {
-	struct btrfs_fs_info *fs_info = root->fs_info;
-	struct btrfs_block_group_cache *cache;
-
-	cache = btrfs_lookup_block_group(fs_info, bytenr);
-	BUG_ON(!cache);
-
 	spin_lock(&cache->space_info->lock);
 	spin_lock(&cache->lock);
 	cache->pinned += num_bytes;
@@ -3569,28 +3872,68 @@ int btrfs_pin_extent(struct btrfs_root *root,
 	spin_unlock(&cache->lock);
 	spin_unlock(&cache->space_info->lock);
 
-	btrfs_put_block_group(cache);
+	set_extent_dirty(root->fs_info->pinned_extents, bytenr,
+			 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
+	return 0;
+}
+
+/*
+ * this function must be called within transaction
+ */
+int btrfs_pin_extent(struct btrfs_root *root,
+		     u64 bytenr, u64 num_bytes, int reserved)
+{
+	struct btrfs_block_group_cache *cache;
 
-	set_extent_dirty(fs_info->pinned_extents,
-			 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
+	cache = btrfs_lookup_block_group(root->fs_info, bytenr);
+	BUG_ON(!cache);
+
+	pin_down_extent(root, cache, bytenr, num_bytes, reserved);
+
+	btrfs_put_block_group(cache);
 	return 0;
 }
 
-static int update_reserved_extents(struct btrfs_block_group_cache *cache,
-				   u64 num_bytes, int reserve)
+/*
+ * update size of reserved extents. this function may return -EAGAIN
+ * if 'reserve' is true or 'sinfo' is false.
+ */
+static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
+				 u64 num_bytes, int reserve, int sinfo)
 {
-	spin_lock(&cache->space_info->lock);
-	spin_lock(&cache->lock);
-	if (reserve) {
-		cache->reserved += num_bytes;
-		cache->space_info->bytes_reserved += num_bytes;
+	int ret = 0;
+	if (sinfo) {
+		struct btrfs_space_info *space_info = cache->space_info;
+		spin_lock(&space_info->lock);
+		spin_lock(&cache->lock);
+		if (reserve) {
+			if (cache->ro) {
+				ret = -EAGAIN;
+			} else {
+				cache->reserved += num_bytes;
+				space_info->bytes_reserved += num_bytes;
+			}
+		} else {
+			if (cache->ro)
+				space_info->bytes_readonly += num_bytes;
+			cache->reserved -= num_bytes;
+			space_info->bytes_reserved -= num_bytes;
+		}
+		spin_unlock(&cache->lock);
+		spin_unlock(&space_info->lock);
 	} else {
-		cache->reserved -= num_bytes;
-		cache->space_info->bytes_reserved -= num_bytes;
+		spin_lock(&cache->lock);
+		if (cache->ro) {
+			ret = -EAGAIN;
+		} else {
+			if (reserve)
+				cache->reserved += num_bytes;
+			else
+				cache->reserved -= num_bytes;
+		}
+		spin_unlock(&cache->lock);
 	}
-	spin_unlock(&cache->lock);
-	spin_unlock(&cache->space_info->lock);
-	return 0;
+	return ret;
 }
 
 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
@@ -3621,6 +3964,8 @@ int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
 		fs_info->pinned_extents = &fs_info->freed_extents[0];
 
 	up_write(&fs_info->extent_commit_sem);
+
+	update_global_block_rsv(fs_info);
 	return 0;
 }
 
@@ -3647,14 +3992,21 @@ static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
 			btrfs_add_free_space(cache, start, len);
 		}
 
+		start += len;
+
 		spin_lock(&cache->space_info->lock);
 		spin_lock(&cache->lock);
 		cache->pinned -= len;
 		cache->space_info->bytes_pinned -= len;
+		if (cache->ro) {
+			cache->space_info->bytes_readonly += len;
+		} else if (cache->reserved_pinned > 0) {
+			len = min(len, cache->reserved_pinned);
+			cache->reserved_pinned -= len;
+			cache->space_info->bytes_reserved += len;
+		}
 		spin_unlock(&cache->lock);
 		spin_unlock(&cache->space_info->lock);
-
-		start += len;
 	}
 
 	if (cache)
@@ -3667,8 +4019,11 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct extent_io_tree *unpin;
+	struct btrfs_block_rsv *block_rsv;
+	struct btrfs_block_rsv *next_rsv;
 	u64 start;
 	u64 end;
+	int idx;
 	int ret;
 
 	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
@@ -3689,59 +4044,30 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 		cond_resched();
 	}
 
-	return ret;
-}
-
-static int pin_down_bytes(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root,
-			  struct btrfs_path *path,
-			  u64 bytenr, u64 num_bytes,
-			  int is_data, int reserved,
-			  struct extent_buffer **must_clean)
-{
-	int err = 0;
-	struct extent_buffer *buf;
-
-	if (is_data)
-		goto pinit;
-
-	/*
-	 * discard is sloooow, and so triggering discards on
-	 * individual btree blocks isn't a good plan.  Just
-	 * pin everything in discard mode.
-	 */
-	if (btrfs_test_opt(root, DISCARD))
-		goto pinit;
+	mutex_lock(&fs_info->durable_block_rsv_mutex);
+	list_for_each_entry_safe(block_rsv, next_rsv,
+				 &fs_info->durable_block_rsv_list, list) {
 
-	buf = btrfs_find_tree_block(root, bytenr, num_bytes);
-	if (!buf)
-		goto pinit;
+		idx = trans->transid & 0x1;
+		if (block_rsv->freed[idx] > 0) {
+			block_rsv_add_bytes(block_rsv,
+					    block_rsv->freed[idx], 0);
+			block_rsv->freed[idx] = 0;
+		}
+		if (atomic_read(&block_rsv->usage) == 0) {
+			btrfs_block_rsv_release(root, block_rsv, (u64)-1);
 
-	/* we can reuse a block if it hasn't been written
-	 * and it is from this transaction.  We can't
-	 * reuse anything from the tree log root because
-	 * it has tiny sub-transactions.
-	 */
-	if (btrfs_buffer_uptodate(buf, 0) &&
-	    btrfs_try_tree_lock(buf)) {
-		u64 header_owner = btrfs_header_owner(buf);
-		u64 header_transid = btrfs_header_generation(buf);
-		if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
-		    header_transid == trans->transid &&
-		    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
-			*must_clean = buf;
-			return 1;
+			if (block_rsv->freed[0] == 0 &&
+			    block_rsv->freed[1] == 0) {
+				list_del_init(&block_rsv->list);
+				kfree(block_rsv);
+			}
+		} else {
+			btrfs_block_rsv_release(root, block_rsv, 0);
 		}
-		btrfs_tree_unlock(buf);
 	}
-	free_extent_buffer(buf);
-pinit:
-	if (path)
-		btrfs_set_path_blocking(path);
-	/* unlocks the pinned mutex */
-	btrfs_pin_extent(root, bytenr, num_bytes, reserved);
+	mutex_unlock(&fs_info->durable_block_rsv_mutex);
 
-	BUG_ON(err < 0);
 	return 0;
 }
 
@@ -3902,9 +4228,6 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			BUG_ON(ret);
 		}
 	} else {
-		int mark_free = 0;
-		struct extent_buffer *must_clean = NULL;
-
 		if (found_extent) {
 			BUG_ON(is_data && refs_to_drop !=
 			       extent_data_ref_count(root, path, iref));
@@ -3917,31 +4240,11 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			}
 		}
 
-		ret = pin_down_bytes(trans, root, path, bytenr,
-				     num_bytes, is_data, 0, &must_clean);
-		if (ret > 0)
-			mark_free = 1;
-		BUG_ON(ret < 0);
-		/*
-		 * it is going to be very rare for someone to be waiting
-		 * on the block we're freeing.  del_items might need to
-		 * schedule, so rather than get fancy, just force it
-		 * to blocking here
-		 */
-		if (must_clean)
-			btrfs_set_lock_blocking(must_clean);
-
 		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
 				      num_to_del);
 		BUG_ON(ret);
 		btrfs_release_path(extent_root, path);
 
-		if (must_clean) {
-			clean_tree_block(NULL, root, must_clean);
-			btrfs_tree_unlock(must_clean);
-			free_extent_buffer(must_clean);
-		}
-
 		if (is_data) {
 			ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
 			BUG_ON(ret);
@@ -3951,8 +4254,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			     (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
 		}
 
-		ret = update_block_group(trans, root, bytenr, num_bytes, 0,
-					 mark_free);
+		ret = update_block_group(trans, root, bytenr, num_bytes, 0);
 		BUG_ON(ret);
 	}
 	btrfs_free_path(path);
@@ -3960,7 +4262,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 }
 
 /*
- * when we free an extent, it is possible (and likely) that we free the last
+ * when we free an block, it is possible (and likely) that we free the last
  * delayed ref for that extent as well.  This searches the delayed ref tree for
  * a given extent, and if there are no other delayed refs to be processed, it
  * removes it from the tree.
@@ -3972,7 +4274,7 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	struct btrfs_delayed_ref_root *delayed_refs;
 	struct btrfs_delayed_ref_node *ref;
 	struct rb_node *node;
-	int ret;
+	int ret = 0;
 
 	delayed_refs = &trans->transaction->delayed_refs;
 	spin_lock(&delayed_refs->lock);
@@ -4024,17 +4326,99 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	list_del_init(&head->cluster);
 	spin_unlock(&delayed_refs->lock);
 
-	ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
-				  &head->node, head->extent_op,
-				  head->must_insert_reserved);
-	BUG_ON(ret);
+	BUG_ON(head->extent_op);
+	if (head->must_insert_reserved)
+		ret = 1;
+
+	mutex_unlock(&head->mutex);
 	btrfs_put_delayed_ref(&head->node);
-	return 0;
+	return ret;
 out:
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
 
+void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct extent_buffer *buf,
+			   u64 parent, int last_ref)
+{
+	struct btrfs_block_rsv *block_rsv;
+	struct btrfs_block_group_cache *cache = NULL;
+	int ret;
+
+	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
+		ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
+						parent, root->root_key.objectid,
+						btrfs_header_level(buf),
+						BTRFS_DROP_DELAYED_REF, NULL);
+		BUG_ON(ret);
+	}
+
+	if (!last_ref)
+		return;
+
+	block_rsv = get_block_rsv(trans, root);
+	cache = btrfs_lookup_block_group(root->fs_info, buf->start);
+	BUG_ON(block_rsv->space_info != cache->space_info);
+
+	if (btrfs_header_generation(buf) == trans->transid) {
+		if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
+			ret = check_ref_cleanup(trans, root, buf->start);
+			if (!ret)
+				goto pin;
+		}
+
+		if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
+			pin_down_extent(root, cache, buf->start, buf->len, 1);
+			goto pin;
+		}
+
+		WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
+
+		btrfs_add_free_space(cache, buf->start, buf->len);
+		ret = update_reserved_bytes(cache, buf->len, 0, 0);
+		if (ret == -EAGAIN) {
+			/* block group became read-only */
+			update_reserved_bytes(cache, buf->len, 0, 1);
+			goto out;
+		}
+
+		ret = 1;
+		spin_lock(&block_rsv->lock);
+		if (block_rsv->reserved < block_rsv->size) {
+			block_rsv->reserved += buf->len;
+			ret = 0;
+		}
+		spin_unlock(&block_rsv->lock);
+
+		if (ret) {
+			spin_lock(&cache->space_info->lock);
+			cache->space_info->bytes_reserved -= buf->len;
+			spin_unlock(&cache->space_info->lock);
+		}
+		goto out;
+	}
+pin:
+	if (block_rsv->durable && !cache->ro) {
+		ret = 0;
+		spin_lock(&cache->lock);
+		if (!cache->ro) {
+			cache->reserved_pinned += buf->len;
+			ret = 1;
+		}
+		spin_unlock(&cache->lock);
+
+		if (ret) {
+			spin_lock(&block_rsv->lock);
+			block_rsv->freed[trans->transid & 0x1] += buf->len;
+			spin_unlock(&block_rsv->lock);
+		}
+	}
+out:
+	btrfs_put_block_group(cache);
+}
+
 int btrfs_free_extent(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root,
 		      u64 bytenr, u64 num_bytes, u64 parent,
@@ -4056,8 +4440,6 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 					parent, root_objectid, (int)owner,
 					BTRFS_DROP_DELAYED_REF, NULL);
 		BUG_ON(ret);
-		ret = check_ref_cleanup(trans, root, bytenr);
-		BUG_ON(ret);
 	} else {
 		ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
 					parent, root_objectid, owner,
@@ -4067,21 +4449,6 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root,
-			  u64 bytenr, u32 blocksize,
-			  u64 parent, u64 root_objectid, int level)
-{
-	u64 used;
-	spin_lock(&root->node_lock);
-	used = btrfs_root_used(&root->root_item) - blocksize;
-	btrfs_set_root_used(&root->root_item, used);
-	spin_unlock(&root->node_lock);
-
-	return btrfs_free_extent(trans, root, bytenr, blocksize,
-				 parent, root_objectid, level, 0);
-}
-
 static u64 stripe_align(struct btrfs_root *root, u64 val)
 {
 	u64 mask = ((u64)root->stripesize - 1);
@@ -4134,6 +4501,22 @@ wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
 	return 0;
 }
 
+static int get_block_group_index(struct btrfs_block_group_cache *cache)
+{
+	int index;
+	if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
+		index = 0;
+	else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
+		index = 1;
+	else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
+		index = 2;
+	else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
+		index = 3;
+	else
+		index = 4;
+	return index;
+}
+
 enum btrfs_loop_type {
 	LOOP_FIND_IDEAL = 0,
 	LOOP_CACHING_NOWAIT = 1,
@@ -4155,7 +4538,6 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
 				     u64 num_bytes, u64 empty_size,
 				     u64 search_start, u64 search_end,
 				     u64 hint_byte, struct btrfs_key *ins,
-				     u64 exclude_start, u64 exclude_nr,
 				     int data)
 {
 	int ret = 0;
@@ -4168,6 +4550,7 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_space_info *space_info;
 	int last_ptr_loop = 0;
 	int loop = 0;
+	int index = 0;
 	bool found_uncached_bg = false;
 	bool failed_cluster_refill = false;
 	bool failed_alloc = false;
@@ -4237,6 +4620,7 @@ ideal_cache:
 				btrfs_put_block_group(block_group);
 				up_read(&space_info->groups_sem);
 			} else {
+				index = get_block_group_index(block_group);
 				goto have_block_group;
 			}
 		} else if (block_group) {
@@ -4245,7 +4629,8 @@ ideal_cache:
 	}
 search:
 	down_read(&space_info->groups_sem);
-	list_for_each_entry(block_group, &space_info->block_groups, list) {
+	list_for_each_entry(block_group, &space_info->block_groups[index],
+			    list) {
 		u64 offset;
 		int cached;
 
@@ -4436,23 +4821,22 @@ checks:
 			goto loop;
 		}
 
-		if (exclude_nr > 0 &&
-		    (search_start + num_bytes > exclude_start &&
-		     search_start < exclude_start + exclude_nr)) {
-			search_start = exclude_start + exclude_nr;
+		ins->objectid = search_start;
+		ins->offset = num_bytes;
+
+		if (offset < search_start)
+			btrfs_add_free_space(block_group, offset,
+					     search_start - offset);
+		BUG_ON(offset > search_start);
 
+		ret = update_reserved_bytes(block_group, num_bytes, 1,
+					    (data & BTRFS_BLOCK_GROUP_DATA));
+		if (ret == -EAGAIN) {
 			btrfs_add_free_space(block_group, offset, num_bytes);
-			/*
-			 * if search_start is still in this block group
-			 * then we just re-search this block group
-			 */
-			if (search_start >= block_group->key.objectid &&
-			    search_start < (block_group->key.objectid +
-					    block_group->key.offset))
-				goto have_block_group;
 			goto loop;
 		}
 
+		/* we are all good, lets return */
 		ins->objectid = search_start;
 		ins->offset = num_bytes;
 
@@ -4460,18 +4844,18 @@ checks:
 			btrfs_add_free_space(block_group, offset,
 					     search_start - offset);
 		BUG_ON(offset > search_start);
-
-		update_reserved_extents(block_group, num_bytes, 1);
-
-		/* we are all good, lets return */
 		break;
 loop:
 		failed_cluster_refill = false;
 		failed_alloc = false;
+		BUG_ON(index != get_block_group_index(block_group));
 		btrfs_put_block_group(block_group);
 	}
 	up_read(&space_info->groups_sem);
 
+	if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
+		goto search;
+
 	/* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
 	 *			for them to make caching progress.  Also
 	 *			determine the best possible bg to cache
@@ -4485,6 +4869,7 @@ loop:
 	if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
 	    (found_uncached_bg || empty_size || empty_cluster ||
 	     allowed_chunk_alloc)) {
+		index = 0;
 		if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
 			found_uncached_bg = false;
 			loop++;
@@ -4567,31 +4952,30 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 			    int dump_block_groups)
 {
 	struct btrfs_block_group_cache *cache;
+	int index = 0;
 
 	spin_lock(&info->lock);
 	printk(KERN_INFO "space_info has %llu free, is %sfull\n",
 	       (unsigned long long)(info->total_bytes - info->bytes_used -
 				    info->bytes_pinned - info->bytes_reserved -
-				    info->bytes_super),
+				    info->bytes_readonly),
 	       (info->full) ? "" : "not ");
-	printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
-	       " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
-	       "\n",
+	printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
+	       "reserved=%llu, may_use=%llu, readonly=%llu\n",
 	       (unsigned long long)info->total_bytes,
+	       (unsigned long long)info->bytes_used,
 	       (unsigned long long)info->bytes_pinned,
-	       (unsigned long long)info->bytes_delalloc,
+	       (unsigned long long)info->bytes_reserved,
 	       (unsigned long long)info->bytes_may_use,
-	       (unsigned long long)info->bytes_used,
-	       (unsigned long long)info->bytes_root,
-	       (unsigned long long)info->bytes_super,
-	       (unsigned long long)info->bytes_reserved);
+	       (unsigned long long)info->bytes_readonly);
 	spin_unlock(&info->lock);
 
 	if (!dump_block_groups)
 		return;
 
 	down_read(&info->groups_sem);
-	list_for_each_entry(cache, &info->block_groups, list) {
+again:
+	list_for_each_entry(cache, &info->block_groups[index], list) {
 		spin_lock(&cache->lock);
 		printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
 		       "%llu pinned %llu reserved\n",
@@ -4603,6 +4987,8 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 		btrfs_dump_free_space(cache, bytes);
 		spin_unlock(&cache->lock);
 	}
+	if (++index < BTRFS_NR_RAID_TYPES)
+		goto again;
 	up_read(&info->groups_sem);
 }
 
@@ -4628,9 +5014,8 @@ again:
 
 	WARN_ON(num_bytes < root->sectorsize);
 	ret = find_free_extent(trans, root, num_bytes, empty_size,
-			       search_start, search_end, hint_byte, ins,
-			       trans->alloc_exclude_start,
-			       trans->alloc_exclude_nr, data);
+			       search_start, search_end, hint_byte,
+			       ins, data);
 
 	if (ret == -ENOSPC && num_bytes > min_alloc_size) {
 		num_bytes = num_bytes >> 1;
@@ -4668,7 +5053,7 @@ int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
 	ret = btrfs_discard_extent(root, start, len);
 
 	btrfs_add_free_space(cache, start, len);
-	update_reserved_extents(cache, len, 0);
+	update_reserved_bytes(cache, len, 0, 1);
 	btrfs_put_block_group(cache);
 
 	return ret;
@@ -4731,8 +5116,7 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(path->nodes[0]);
 	btrfs_free_path(path);
 
-	ret = update_block_group(trans, root, ins->objectid, ins->offset,
-				 1, 0);
+	ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
 	if (ret) {
 		printk(KERN_ERR "btrfs update block group failed for %llu "
 		       "%llu\n", (unsigned long long)ins->objectid,
@@ -4792,8 +5176,7 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(leaf);
 	btrfs_free_path(path);
 
-	ret = update_block_group(trans, root, ins->objectid, ins->offset,
-				 1, 0);
+	ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
 	if (ret) {
 		printk(KERN_ERR "btrfs update block group failed for %llu "
 		       "%llu\n", (unsigned long long)ins->objectid,
@@ -4869,73 +5252,14 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 		put_caching_control(caching_ctl);
 	}
 
-	update_reserved_extents(block_group, ins->offset, 1);
+	ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
+	BUG_ON(ret);
 	btrfs_put_block_group(block_group);
 	ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
 					 0, owner, offset, ins, 1);
 	return ret;
 }
 
-/*
- * finds a free extent and does all the dirty work required for allocation
- * returns the key for the extent through ins, and a tree buffer for
- * the first block of the extent through buf.
- *
- * returns 0 if everything worked, non-zero otherwise.
- */
-static int alloc_tree_block(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root,
-			    u64 num_bytes, u64 parent, u64 root_objectid,
-			    struct btrfs_disk_key *key, int level,
-			    u64 empty_size, u64 hint_byte, u64 search_end,
-			    struct btrfs_key *ins)
-{
-	int ret;
-	u64 flags = 0;
-
-	ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
-				   empty_size, hint_byte, search_end,
-				   ins, 0);
-	if (ret)
-		return ret;
-
-	if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
-		if (parent == 0)
-			parent = ins->objectid;
-		flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
-	} else
-		BUG_ON(parent > 0);
-
-	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
-		struct btrfs_delayed_extent_op *extent_op;
-		extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
-		BUG_ON(!extent_op);
-		if (key)
-			memcpy(&extent_op->key, key, sizeof(extent_op->key));
-		else
-			memset(&extent_op->key, 0, sizeof(extent_op->key));
-		extent_op->flags_to_set = flags;
-		extent_op->update_key = 1;
-		extent_op->update_flags = 1;
-		extent_op->is_data = 0;
-
-		ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
-					ins->offset, parent, root_objectid,
-					level, BTRFS_ADD_DELAYED_EXTENT,
-					extent_op);
-		BUG_ON(ret);
-	}
-
-	if (root_objectid == root->root_key.objectid) {
-		u64 used;
-		spin_lock(&root->node_lock);
-		used = btrfs_root_used(&root->root_item) + num_bytes;
-		btrfs_set_root_used(&root->root_item, used);
-		spin_unlock(&root->node_lock);
-	}
-	return ret;
-}
-
 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
 					    struct btrfs_root *root,
 					    u64 bytenr, u32 blocksize,
@@ -4974,8 +5298,45 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
 	return buf;
 }
 
+static struct btrfs_block_rsv *
+use_block_rsv(struct btrfs_trans_handle *trans,
+	      struct btrfs_root *root, u32 blocksize)
+{
+	struct btrfs_block_rsv *block_rsv;
+	int ret;
+
+	block_rsv = get_block_rsv(trans, root);
+
+	if (block_rsv->size == 0) {
+		ret = reserve_metadata_bytes(block_rsv, blocksize);
+		if (ret)
+			return ERR_PTR(ret);
+		return block_rsv;
+	}
+
+	ret = block_rsv_use_bytes(block_rsv, blocksize);
+	if (!ret)
+		return block_rsv;
+
+	WARN_ON(1);
+	printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
+		block_rsv->size, block_rsv->reserved,
+		block_rsv->freed[0], block_rsv->freed[1]);
+
+	return ERR_PTR(-ENOSPC);
+}
+
+static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
+{
+	block_rsv_add_bytes(block_rsv, blocksize, 0);
+	block_rsv_release_bytes(block_rsv, NULL, 0);
+}
+
 /*
- * helper function to allocate a block for a given tree
+ * finds a free extent and does all the dirty work required for allocation
+ * returns the key for the extent through ins, and a tree buffer for
+ * the first block of the extent through buf.
+ *
  * returns the tree buffer or NULL.
  */
 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
@@ -4985,18 +5346,53 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 					u64 hint, u64 empty_size)
 {
 	struct btrfs_key ins;
-	int ret;
+	struct btrfs_block_rsv *block_rsv;
 	struct extent_buffer *buf;
+	u64 flags = 0;
+	int ret;
+
 
-	ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
-			       key, level, empty_size, hint, (u64)-1, &ins);
+	block_rsv = use_block_rsv(trans, root, blocksize);
+	if (IS_ERR(block_rsv))
+		return ERR_CAST(block_rsv);
+
+	ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
+				   empty_size, hint, (u64)-1, &ins, 0);
 	if (ret) {
-		BUG_ON(ret > 0);
+		unuse_block_rsv(block_rsv, blocksize);
 		return ERR_PTR(ret);
 	}
 
 	buf = btrfs_init_new_buffer(trans, root, ins.objectid,
 				    blocksize, level);
+	BUG_ON(IS_ERR(buf));
+
+	if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
+		if (parent == 0)
+			parent = ins.objectid;
+		flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
+	} else
+		BUG_ON(parent > 0);
+
+	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
+		struct btrfs_delayed_extent_op *extent_op;
+		extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+		BUG_ON(!extent_op);
+		if (key)
+			memcpy(&extent_op->key, key, sizeof(extent_op->key));
+		else
+			memset(&extent_op->key, 0, sizeof(extent_op->key));
+		extent_op->flags_to_set = flags;
+		extent_op->update_key = 1;
+		extent_op->update_flags = 1;
+		extent_op->is_data = 0;
+
+		ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
+					ins.offset, parent, root_objectid,
+					level, BTRFS_ADD_DELAYED_EXTENT,
+					extent_op);
+		BUG_ON(ret);
+	}
 	return buf;
 }
 
@@ -5321,7 +5717,7 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 				 struct btrfs_path *path,
 				 struct walk_control *wc)
 {
-	int ret = 0;
+	int ret;
 	int level = wc->level;
 	struct extent_buffer *eb = path->nodes[level];
 	u64 parent = 0;
@@ -5399,13 +5795,11 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 			       btrfs_header_owner(path->nodes[level + 1]));
 	}
 
-	ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
-				root->root_key.objectid, level, 0);
-	BUG_ON(ret);
+	btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
 out:
 	wc->refs[level] = 0;
 	wc->flags[level] = 0;
-	return ret;
+	return 0;
 }
 
 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
@@ -5483,7 +5877,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
  * also make sure backrefs for the shared block and all lower level
  * blocks are properly updated.
  */
-int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
+int btrfs_drop_snapshot(struct btrfs_root *root,
+			struct btrfs_block_rsv *block_rsv, int update_ref)
 {
 	struct btrfs_path *path;
 	struct btrfs_trans_handle *trans;
@@ -5501,7 +5896,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
 	wc = kzalloc(sizeof(*wc), GFP_NOFS);
 	BUG_ON(!wc);
 
-	trans = btrfs_start_transaction(tree_root, 1);
+	trans = btrfs_start_transaction(tree_root, 0);
+	if (block_rsv)
+		trans->block_rsv = block_rsv;
 
 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
 		level = btrfs_header_level(root->node);
@@ -5589,22 +5986,16 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
 		}
 
 		BUG_ON(wc->level == 0);
-		if (trans->transaction->in_commit ||
-		    trans->transaction->delayed_refs.flushing) {
+		if (btrfs_should_end_transaction(trans, tree_root)) {
 			ret = btrfs_update_root(trans, tree_root,
 						&root->root_key,
 						root_item);
 			BUG_ON(ret);
 
-			btrfs_end_transaction(trans, tree_root);
-			trans = btrfs_start_transaction(tree_root, 1);
-		} else {
-			unsigned long update;
-			update = trans->delayed_ref_updates;
-			trans->delayed_ref_updates = 0;
-			if (update)
-				btrfs_run_delayed_refs(trans, tree_root,
-						       update);
+			btrfs_end_transaction_throttle(trans, tree_root);
+			trans = btrfs_start_transaction(tree_root, 0);
+			if (block_rsv)
+				trans->block_rsv = block_rsv;
 		}
 	}
 	btrfs_release_path(root, path);
@@ -5632,7 +6023,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
 		kfree(root);
 	}
 out:
-	btrfs_end_transaction(trans, tree_root);
+	btrfs_end_transaction_throttle(trans, tree_root);
 	kfree(wc);
 	btrfs_free_path(path);
 	return err;
@@ -7228,48 +7619,80 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
 	return flags;
 }
 
-static int __alloc_chunk_for_shrink(struct btrfs_root *root,
-		     struct btrfs_block_group_cache *shrink_block_group,
-		     int force)
+static int set_block_group_ro(struct btrfs_block_group_cache *cache)
 {
-	struct btrfs_trans_handle *trans;
-	u64 new_alloc_flags;
-	u64 calc;
+	struct btrfs_space_info *sinfo = cache->space_info;
+	u64 num_bytes;
+	int ret = -ENOSPC;
 
-	spin_lock(&shrink_block_group->lock);
-	if (btrfs_block_group_used(&shrink_block_group->item) +
-	    shrink_block_group->reserved > 0) {
-		spin_unlock(&shrink_block_group->lock);
+	if (cache->ro)
+		return 0;
 
-		trans = btrfs_start_transaction(root, 1);
-		spin_lock(&shrink_block_group->lock);
+	spin_lock(&sinfo->lock);
+	spin_lock(&cache->lock);
+	num_bytes = cache->key.offset - cache->reserved - cache->pinned -
+		    cache->bytes_super - btrfs_block_group_used(&cache->item);
+
+	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) {
+		sinfo->bytes_readonly += num_bytes;
+		sinfo->bytes_reserved += cache->reserved_pinned;
+		cache->reserved_pinned = 0;
+		cache->ro = 1;
+		ret = 0;
+	}
+	spin_unlock(&cache->lock);
+	spin_unlock(&sinfo->lock);
+	return ret;
+}
 
-		new_alloc_flags = update_block_group_flags(root,
-						   shrink_block_group->flags);
-		if (new_alloc_flags != shrink_block_group->flags) {
-			calc =
-			     btrfs_block_group_used(&shrink_block_group->item);
-		} else {
-			calc = shrink_block_group->key.offset;
-		}
-		spin_unlock(&shrink_block_group->lock);
+int btrfs_set_block_group_ro(struct btrfs_root *root,
+			     struct btrfs_block_group_cache *cache)
 
-		do_chunk_alloc(trans, root->fs_info->extent_root,
-			       calc + 2 * 1024 * 1024, new_alloc_flags, force);
+{
+	struct btrfs_trans_handle *trans;
+	u64 alloc_flags;
+	int ret;
 
-		btrfs_end_transaction(trans, root);
-	} else
-		spin_unlock(&shrink_block_group->lock);
-	return 0;
-}
+	BUG_ON(cache->ro);
 
+	trans = btrfs_join_transaction(root, 1);
+	BUG_ON(IS_ERR(trans));
 
-int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
-					 struct btrfs_block_group_cache *group)
+	alloc_flags = update_block_group_flags(root, cache->flags);
+	if (alloc_flags != cache->flags)
+		do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
+
+	ret = set_block_group_ro(cache);
+	if (!ret)
+		goto out;
+	alloc_flags = get_alloc_profile(root, cache->space_info->flags);
+	ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
+	if (ret < 0)
+		goto out;
+	ret = set_block_group_ro(cache);
+out:
+	btrfs_end_transaction(trans, root);
+	return ret;
+}
 
+int btrfs_set_block_group_rw(struct btrfs_root *root,
+			      struct btrfs_block_group_cache *cache)
 {
-	__alloc_chunk_for_shrink(root, group, 1);
-	set_block_group_readonly(group);
+	struct btrfs_space_info *sinfo = cache->space_info;
+	u64 num_bytes;
+
+	BUG_ON(!cache->ro);
+
+	spin_lock(&sinfo->lock);
+	spin_lock(&cache->lock);
+	num_bytes = cache->key.offset - cache->reserved - cache->pinned -
+		    cache->bytes_super - btrfs_block_group_used(&cache->item);
+	sinfo->bytes_readonly -= num_bytes;
+	cache->ro = 0;
+	spin_unlock(&cache->lock);
+	spin_unlock(&sinfo->lock);
 	return 0;
 }
 
@@ -7436,17 +7859,33 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
 	 */
 	synchronize_rcu();
 
+	release_global_block_rsv(info);
+
 	while(!list_empty(&info->space_info)) {
 		space_info = list_entry(info->space_info.next,
 					struct btrfs_space_info,
 					list);
-
+		if (space_info->bytes_pinned > 0 ||
+		    space_info->bytes_reserved > 0) {
+			WARN_ON(1);
+			dump_space_info(space_info, 0, 0);
+		}
 		list_del(&space_info->list);
 		kfree(space_info);
 	}
 	return 0;
 }
 
+static void __link_block_group(struct btrfs_space_info *space_info,
+			       struct btrfs_block_group_cache *cache)
+{
+	int index = get_block_group_index(cache);
+
+	down_write(&space_info->groups_sem);
+	list_add_tail(&cache->list, &space_info->block_groups[index]);
+	up_write(&space_info->groups_sem);
+}
+
 int btrfs_read_block_groups(struct btrfs_root *root)
 {
 	struct btrfs_path *path;
@@ -7468,10 +7907,8 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 
 	while (1) {
 		ret = find_first_block_group(root, path, &key);
-		if (ret > 0) {
-			ret = 0;
-			goto error;
-		}
+		if (ret > 0)
+			break;
 		if (ret != 0)
 			goto error;
 
@@ -7480,7 +7917,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		cache = kzalloc(sizeof(*cache), GFP_NOFS);
 		if (!cache) {
 			ret = -ENOMEM;
-			break;
+			goto error;
 		}
 
 		atomic_set(&cache->count, 1);
@@ -7537,20 +7974,36 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		BUG_ON(ret);
 		cache->space_info = space_info;
 		spin_lock(&cache->space_info->lock);
-		cache->space_info->bytes_super += cache->bytes_super;
+		cache->space_info->bytes_readonly += cache->bytes_super;
 		spin_unlock(&cache->space_info->lock);
 
-		down_write(&space_info->groups_sem);
-		list_add_tail(&cache->list, &space_info->block_groups);
-		up_write(&space_info->groups_sem);
+		__link_block_group(space_info, cache);
 
 		ret = btrfs_add_block_group_cache(root->fs_info, cache);
 		BUG_ON(ret);
 
 		set_avail_alloc_bits(root->fs_info, cache->flags);
 		if (btrfs_chunk_readonly(root, cache->key.objectid))
-			set_block_group_readonly(cache);
+			set_block_group_ro(cache);
+	}
+
+	list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
+		if (!(get_alloc_profile(root, space_info->flags) &
+		      (BTRFS_BLOCK_GROUP_RAID10 |
+		       BTRFS_BLOCK_GROUP_RAID1 |
+		       BTRFS_BLOCK_GROUP_DUP)))
+			continue;
+		/*
+		 * avoid allocating from un-mirrored block group if there are
+		 * mirrored block groups.
+		 */
+		list_for_each_entry(cache, &space_info->block_groups[3], list)
+			set_block_group_ro(cache);
+		list_for_each_entry(cache, &space_info->block_groups[4], list)
+			set_block_group_ro(cache);
 	}
+
+	init_global_block_rsv(info);
 	ret = 0;
 error:
 	btrfs_free_path(path);
@@ -7611,12 +8064,10 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 	BUG_ON(ret);
 
 	spin_lock(&cache->space_info->lock);
-	cache->space_info->bytes_super += cache->bytes_super;
+	cache->space_info->bytes_readonly += cache->bytes_super;
 	spin_unlock(&cache->space_info->lock);
 
-	down_write(&cache->space_info->groups_sem);
-	list_add_tail(&cache->list, &cache->space_info->block_groups);
-	up_write(&cache->space_info->groups_sem);
+	__link_block_group(cache->space_info, cache);
 
 	ret = btrfs_add_block_group_cache(root->fs_info, cache);
 	BUG_ON(ret);

fs/btrfs/extent_io.c

@@ -135,7 +135,7 @@ static struct extent_state *alloc_extent_state(gfp_t mask)
 	return state;
 }
 
-static void free_extent_state(struct extent_state *state)
+void free_extent_state(struct extent_state *state)
 {
 	if (!state)
 		return;
@@ -335,21 +335,18 @@ static int merge_state(struct extent_io_tree *tree,
 }
 
 static int set_state_cb(struct extent_io_tree *tree,
-			 struct extent_state *state,
-			 unsigned long bits)
+			 struct extent_state *state, int *bits)
 {
 	if (tree->ops && tree->ops->set_bit_hook) {
 		return tree->ops->set_bit_hook(tree->mapping->host,
-					       state->start, state->end,
-					       state->state, bits);
+					       state, bits);
 	}
 
 	return 0;
 }
 
 static void clear_state_cb(struct extent_io_tree *tree,
-			   struct extent_state *state,
-			   unsigned long bits)
+			   struct extent_state *state, int *bits)
 {
 	if (tree->ops && tree->ops->clear_bit_hook)
 		tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
@@ -367,9 +364,10 @@ static void clear_state_cb(struct extent_io_tree *tree,
  */
 static int insert_state(struct extent_io_tree *tree,
 			struct extent_state *state, u64 start, u64 end,
-			int bits)
+			int *bits)
 {
 	struct rb_node *node;
+	int bits_to_set = *bits & ~EXTENT_CTLBITS;
 	int ret;
 
 	if (end < start) {
@@ -384,9 +382,9 @@ static int insert_state(struct extent_io_tree *tree,
 	if (ret)
 		return ret;
 
-	if (bits & EXTENT_DIRTY)
+	if (bits_to_set & EXTENT_DIRTY)
 		tree->dirty_bytes += end - start + 1;
-	state->state |= bits;
+	state->state |= bits_to_set;
 	node = tree_insert(&tree->state, end, &state->rb_node);
 	if (node) {
 		struct extent_state *found;
@@ -456,13 +454,13 @@ static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
  * struct is freed and removed from the tree
  */
 static int clear_state_bit(struct extent_io_tree *tree,
-			    struct extent_state *state, int bits, int wake,
-			    int delete)
+			    struct extent_state *state,
+			    int *bits, int wake)
 {
-	int bits_to_clear = bits & ~EXTENT_DO_ACCOUNTING;
+	int bits_to_clear = *bits & ~EXTENT_CTLBITS;
 	int ret = state->state & bits_to_clear;
 
-	if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
+	if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
 		u64 range = state->end - state->start + 1;
 		WARN_ON(range > tree->dirty_bytes);
 		tree->dirty_bytes -= range;
@@ -471,9 +469,8 @@ static int clear_state_bit(struct extent_io_tree *tree,
 	state->state &= ~bits_to_clear;
 	if (wake)
 		wake_up(&state->wq);
-	if (delete || state->state == 0) {
+	if (state->state == 0) {
 		if (state->tree) {
-			clear_state_cb(tree, state, state->state);
 			rb_erase(&state->rb_node, &tree->state);
 			state->tree = NULL;
 			free_extent_state(state);
@@ -514,6 +511,10 @@ int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 	int set = 0;
 	int clear = 0;
 
+	if (delete)
+		bits |= ~EXTENT_CTLBITS;
+	bits |= EXTENT_FIRST_DELALLOC;
+
 	if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
 		clear = 1;
 again:
@@ -580,8 +581,7 @@ hit_next:
 		if (err)
 			goto out;
 		if (state->end <= end) {
-			set |= clear_state_bit(tree, state, bits, wake,
-					       delete);
+			set |= clear_state_bit(tree, state, &bits, wake);
 			if (last_end == (u64)-1)
 				goto out;
 			start = last_end + 1;
@@ -602,7 +602,7 @@ hit_next:
 		if (wake)
 			wake_up(&state->wq);
 
-		set |= clear_state_bit(tree, prealloc, bits, wake, delete);
+		set |= clear_state_bit(tree, prealloc, &bits, wake);
 
 		prealloc = NULL;
 		goto out;
@@ -613,7 +613,7 @@ hit_next:
 	else
 		next_node = NULL;
 
-	set |= clear_state_bit(tree, state, bits, wake, delete);
+	set |= clear_state_bit(tree, state, &bits, wake);
 	if (last_end == (u64)-1)
 		goto out;
 	start = last_end + 1;
@@ -706,19 +706,19 @@ out:
 
 static int set_state_bits(struct extent_io_tree *tree,
 			   struct extent_state *state,
-			   int bits)
+			   int *bits)
 {
 	int ret;
+	int bits_to_set = *bits & ~EXTENT_CTLBITS;
 
 	ret = set_state_cb(tree, state, bits);
 	if (ret)
 		return ret;
-
-	if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
+	if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
 		u64 range = state->end - state->start + 1;
 		tree->dirty_bytes += range;
 	}
-	state->state |= bits;
+	state->state |= bits_to_set;
 
 	return 0;
 }
@@ -745,10 +745,9 @@ static void cache_state(struct extent_state *state,
  * [start, end] is inclusive This takes the tree lock.
  */
 
-static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
-			  int bits, int exclusive_bits, u64 *failed_start,
-			  struct extent_state **cached_state,
-			  gfp_t mask)
+int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
+		   int bits, int exclusive_bits, u64 *failed_start,
+		   struct extent_state **cached_state, gfp_t mask)
 {
 	struct extent_state *state;
 	struct extent_state *prealloc = NULL;
@@ -757,6 +756,7 @@ static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 	u64 last_start;
 	u64 last_end;
 
+	bits |= EXTENT_FIRST_DELALLOC;
 again:
 	if (!prealloc && (mask & __GFP_WAIT)) {
 		prealloc = alloc_extent_state(mask);
@@ -778,7 +778,7 @@ again:
 	 */
 	node = tree_search(tree, start);
 	if (!node) {
-		err = insert_state(tree, prealloc, start, end, bits);
+		err = insert_state(tree, prealloc, start, end, &bits);
 		prealloc = NULL;
 		BUG_ON(err == -EEXIST);
 		goto out;
@@ -802,7 +802,7 @@ hit_next:
 			goto out;
 		}
 
-		err = set_state_bits(tree, state, bits);
+		err = set_state_bits(tree, state, &bits);
 		if (err)
 			goto out;
 
@@ -852,7 +852,7 @@ hit_next:
 		if (err)
 			goto out;
 		if (state->end <= end) {
-			err = set_state_bits(tree, state, bits);
+			err = set_state_bits(tree, state, &bits);
 			if (err)
 				goto out;
 			cache_state(state, cached_state);
@@ -877,7 +877,7 @@ hit_next:
 		else
 			this_end = last_start - 1;
 		err = insert_state(tree, prealloc, start, this_end,
-				   bits);
+				   &bits);
 		BUG_ON(err == -EEXIST);
 		if (err) {
 			prealloc = NULL;
@@ -903,7 +903,7 @@ hit_next:
 		err = split_state(tree, state, prealloc, end + 1);
 		BUG_ON(err == -EEXIST);
 
-		err = set_state_bits(tree, prealloc, bits);
+		err = set_state_bits(tree, prealloc, &bits);
 		if (err) {
 			prealloc = NULL;
 			goto out;
@@ -966,8 +966,7 @@ int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 {
 	return clear_extent_bit(tree, start, end,
 				EXTENT_DIRTY | EXTENT_DELALLOC |
-				EXTENT_DO_ACCOUNTING, 0, 0,
-				NULL, mask);
+				EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
 }
 
 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
@@ -1435,9 +1434,6 @@ int extent_clear_unlock_delalloc(struct inode *inode,
 	if (op & EXTENT_CLEAR_DELALLOC)
 		clear_bits |= EXTENT_DELALLOC;
 
-	if (op & EXTENT_CLEAR_ACCOUNTING)
-		clear_bits |= EXTENT_DO_ACCOUNTING;
-
 	clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
 	if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 		    EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
@@ -1916,7 +1912,7 @@ static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
 
 	if (tree->ops && tree->ops->submit_bio_hook)
 		tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
-					   mirror_num, bio_flags);
+					   mirror_num, bio_flags, start);
 	else
 		submit_bio(rw, bio);
 	if (bio_flagged(bio, BIO_EOPNOTSUPP))
@@ -2020,6 +2016,7 @@ static int __extent_read_full_page(struct extent_io_tree *tree,
 	sector_t sector;
 	struct extent_map *em;
 	struct block_device *bdev;
+	struct btrfs_ordered_extent *ordered;
 	int ret;
 	int nr = 0;
 	size_t page_offset = 0;
@@ -2031,7 +2028,15 @@ static int __extent_read_full_page(struct extent_io_tree *tree,
 	set_page_extent_mapped(page);
 
 	end = page_end;
-	lock_extent(tree, start, end, GFP_NOFS);
+	while (1) {
+		lock_extent(tree, start, end, GFP_NOFS);
+		ordered = btrfs_lookup_ordered_extent(inode, start);
+		if (!ordered)
+			break;
+		unlock_extent(tree, start, end, GFP_NOFS);
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		btrfs_put_ordered_extent(ordered);
+	}
 
 	if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
 		char *userpage;

fs/btrfs/extent_io.h

@@ -16,7 +16,9 @@
 #define EXTENT_BOUNDARY (1 << 9)
 #define EXTENT_NODATASUM (1 << 10)
 #define EXTENT_DO_ACCOUNTING (1 << 11)
+#define EXTENT_FIRST_DELALLOC (1 << 12)
 #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
+#define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
 
 /* flags for bio submission */
 #define EXTENT_BIO_COMPRESSED 1
@@ -47,7 +49,7 @@ struct extent_state;
 
 typedef	int (extent_submit_bio_hook_t)(struct inode *inode, int rw,
 				       struct bio *bio, int mirror_num,
-				       unsigned long bio_flags);
+				       unsigned long bio_flags, u64 bio_offset);
 struct extent_io_ops {
 	int (*fill_delalloc)(struct inode *inode, struct page *locked_page,
 			     u64 start, u64 end, int *page_started,
@@ -69,10 +71,10 @@ struct extent_io_ops {
 				    struct extent_state *state);
 	int (*writepage_end_io_hook)(struct page *page, u64 start, u64 end,
 				      struct extent_state *state, int uptodate);
-	int (*set_bit_hook)(struct inode *inode, u64 start, u64 end,
-			    unsigned long old, unsigned long bits);
+	int (*set_bit_hook)(struct inode *inode, struct extent_state *state,
+			    int *bits);
 	int (*clear_bit_hook)(struct inode *inode, struct extent_state *state,
-			      unsigned long bits);
+			      int *bits);
 	int (*merge_extent_hook)(struct inode *inode,
 				 struct extent_state *new,
 				 struct extent_state *other);
@@ -176,6 +178,7 @@ u64 count_range_bits(struct extent_io_tree *tree,
 		     u64 *start, u64 search_end,
 		     u64 max_bytes, unsigned long bits);
 
+void free_extent_state(struct extent_state *state);
 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
 		   int bits, int filled, struct extent_state *cached_state);
 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
@@ -185,6 +188,9 @@ int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 		     gfp_t mask);
 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 		    int bits, gfp_t mask);
+int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
+		   int bits, int exclusive_bits, u64 *failed_start,
+		   struct extent_state **cached_state, gfp_t mask);
 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 			gfp_t mask);
 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,

fs/btrfs/file-item.c

@@ -149,13 +149,14 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 }
 
 
-int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
-			  struct bio *bio, u32 *dst)
+static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
+				   struct inode *inode, struct bio *bio,
+				   u64 logical_offset, u32 *dst, int dio)
 {
 	u32 sum;
 	struct bio_vec *bvec = bio->bi_io_vec;
 	int bio_index = 0;
-	u64 offset;
+	u64 offset = 0;
 	u64 item_start_offset = 0;
 	u64 item_last_offset = 0;
 	u64 disk_bytenr;
@@ -174,8 +175,11 @@ int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
 	WARN_ON(bio->bi_vcnt <= 0);
 
 	disk_bytenr = (u64)bio->bi_sector << 9;
+	if (dio)
+		offset = logical_offset;
 	while (bio_index < bio->bi_vcnt) {
-		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
+		if (!dio)
+			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
 		ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
 		if (ret == 0)
 			goto found;
@@ -238,6 +242,7 @@ found:
 		else
 			set_state_private(io_tree, offset, sum);
 		disk_bytenr += bvec->bv_len;
+		offset += bvec->bv_len;
 		bio_index++;
 		bvec++;
 	}
@@ -245,6 +250,18 @@ found:
 	return 0;
 }
 
+int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
+			  struct bio *bio, u32 *dst)
+{
+	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
+}
+
+int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
+			      struct bio *bio, u64 offset, u32 *dst)
+{
+	return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
+}
+
 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 			     struct list_head *list)
 {
@@ -657,6 +674,9 @@ again:
 		goto found;
 	}
 	ret = PTR_ERR(item);
+	if (ret != -EFBIG && ret != -ENOENT)
+		goto fail_unlock;
+
 	if (ret == -EFBIG) {
 		u32 item_size;
 		/* we found one, but it isn't big enough yet */

fs/btrfs/file.c

@@ -46,32 +46,42 @@
 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
 					 int write_bytes,
 					 struct page **prepared_pages,
-					 const char __user *buf)
+					 struct iov_iter *i)
 {
-	long page_fault = 0;
-	int i;
+	size_t copied;
+	int pg = 0;
 	int offset = pos & (PAGE_CACHE_SIZE - 1);
 
-	for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
+	while (write_bytes > 0) {
 		size_t count = min_t(size_t,
 				     PAGE_CACHE_SIZE - offset, write_bytes);
-		struct page *page = prepared_pages[i];
-		fault_in_pages_readable(buf, count);
+		struct page *page = prepared_pages[pg];
+again:
+		if (unlikely(iov_iter_fault_in_readable(i, count)))
+			return -EFAULT;
 
 		/* Copy data from userspace to the current page */
-		kmap(page);
-		page_fault = __copy_from_user(page_address(page) + offset,
-					      buf, count);
+		copied = iov_iter_copy_from_user(page, i, offset, count);
+
 		/* Flush processor's dcache for this page */
 		flush_dcache_page(page);
-		kunmap(page);
-		buf += count;
-		write_bytes -= count;
+		iov_iter_advance(i, copied);
+		write_bytes -= copied;
 
-		if (page_fault)
-			break;
+		if (unlikely(copied == 0)) {
+			count = min_t(size_t, PAGE_CACHE_SIZE - offset,
+				      iov_iter_single_seg_count(i));
+			goto again;
+		}
+
+		if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
+			offset += copied;
+		} else {
+			pg++;
+			offset = 0;
+		}
 	}
-	return page_fault ? -EFAULT : 0;
+	return 0;
 }
 
 /*
@@ -126,8 +136,7 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	end_of_last_block = start_pos + num_bytes - 1;
 	err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 					NULL);
-	if (err)
-		return err;
+	BUG_ON(err);
 
 	for (i = 0; i < num_pages; i++) {
 		struct page *p = pages[i];
@@ -142,7 +151,7 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 		 * at this time.
 		 */
 	}
-	return err;
+	return 0;
 }
 
 /*
@@ -823,45 +832,46 @@ again:
 	return 0;
 }
 
-static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
-				size_t count, loff_t *ppos)
+static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
+				    const struct iovec *iov,
+				    unsigned long nr_segs, loff_t pos)
 {
-	loff_t pos;
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = fdentry(file)->d_inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct page *pinned[2];
+	struct page **pages = NULL;
+	struct iov_iter i;
+	loff_t *ppos = &iocb->ki_pos;
 	loff_t start_pos;
 	ssize_t num_written = 0;
 	ssize_t err = 0;
+	size_t count;
+	size_t ocount;
 	int ret = 0;
-	struct inode *inode = fdentry(file)->d_inode;
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct page **pages = NULL;
 	int nrptrs;
-	struct page *pinned[2];
 	unsigned long first_index;
 	unsigned long last_index;
 	int will_write;
+	int buffered = 0;
 
 	will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
 		      (file->f_flags & O_DIRECT));
 
-	nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
-		     PAGE_CACHE_SIZE / (sizeof(struct page *)));
 	pinned[0] = NULL;
 	pinned[1] = NULL;
 
-	pos = *ppos;
 	start_pos = pos;
 
 	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 
-	/* do the reserve before the mutex lock in case we have to do some
-	 * flushing.  We wouldn't deadlock, but this is more polite.
-	 */
-	err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
-	if (err)
-		goto out_nolock;
-
 	mutex_lock(&inode->i_mutex);
 
+	err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
+	if (err)
+		goto out;
+	count = ocount;
+
 	current->backing_dev_info = inode->i_mapping->backing_dev_info;
 	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 	if (err)
@@ -875,15 +885,53 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		goto out;
 
 	file_update_time(file);
+	BTRFS_I(inode)->sequence++;
+
+	if (unlikely(file->f_flags & O_DIRECT)) {
+		num_written = generic_file_direct_write(iocb, iov, &nr_segs,
+							pos, ppos, count,
+							ocount);
+		/*
+		 * the generic O_DIRECT will update in-memory i_size after the
+		 * DIOs are done.  But our endio handlers that update the on
+		 * disk i_size never update past the in memory i_size.  So we
+		 * need one more update here to catch any additions to the
+		 * file
+		 */
+		if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
+			btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
+			mark_inode_dirty(inode);
+		}
 
+		if (num_written < 0) {
+			ret = num_written;
+			num_written = 0;
+			goto out;
+		} else if (num_written == count) {
+			/* pick up pos changes done by the generic code */
+			pos = *ppos;
+			goto out;
+		}
+		/*
+		 * We are going to do buffered for the rest of the range, so we
+		 * need to make sure to invalidate the buffered pages when we're
+		 * done.
+		 */
+		buffered = 1;
+		pos += num_written;
+	}
+
+	iov_iter_init(&i, iov, nr_segs, count, num_written);
+	nrptrs = min((iov_iter_count(&i) + PAGE_CACHE_SIZE - 1) /
+		     PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
+		     (sizeof(struct page *)));
 	pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 
 	/* generic_write_checks can change our pos */
 	start_pos = pos;
 
-	BTRFS_I(inode)->sequence++;
 	first_index = pos >> PAGE_CACHE_SHIFT;
-	last_index = (pos + count) >> PAGE_CACHE_SHIFT;
+	last_index = (pos + iov_iter_count(&i)) >> PAGE_CACHE_SHIFT;
 
 	/*
 	 * there are lots of better ways to do this, but this code
@@ -900,7 +948,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 			unlock_page(pinned[0]);
 		}
 	}
-	if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
+	if ((pos + iov_iter_count(&i)) & (PAGE_CACHE_SIZE - 1)) {
 		pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 		if (!PageUptodate(pinned[1])) {
 			ret = btrfs_readpage(NULL, pinned[1]);
@@ -911,10 +959,10 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		}
 	}
 
-	while (count > 0) {
+	while (iov_iter_count(&i) > 0) {
 		size_t offset = pos & (PAGE_CACHE_SIZE - 1);
-		size_t write_bytes = min(count, nrptrs *
-					(size_t)PAGE_CACHE_SIZE -
+		size_t write_bytes = min(iov_iter_count(&i),
+					 nrptrs * (size_t)PAGE_CACHE_SIZE -
 					 offset);
 		size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 					PAGE_CACHE_SHIFT;
@@ -922,7 +970,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		WARN_ON(num_pages > nrptrs);
 		memset(pages, 0, sizeof(struct page *) * nrptrs);
 
-		ret = btrfs_check_data_free_space(root, inode, write_bytes);
+		ret = btrfs_delalloc_reserve_space(inode, write_bytes);
 		if (ret)
 			goto out;
 
@@ -930,26 +978,20 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 				    pos, first_index, last_index,
 				    write_bytes);
 		if (ret) {
-			btrfs_free_reserved_data_space(root, inode,
-						       write_bytes);
+			btrfs_delalloc_release_space(inode, write_bytes);
 			goto out;
 		}
 
 		ret = btrfs_copy_from_user(pos, num_pages,
-					   write_bytes, pages, buf);
-		if (ret) {
-			btrfs_free_reserved_data_space(root, inode,
-						       write_bytes);
-			btrfs_drop_pages(pages, num_pages);
-			goto out;
+					   write_bytes, pages, &i);
+		if (ret == 0) {
+			dirty_and_release_pages(NULL, root, file, pages,
+						num_pages, pos, write_bytes);
 		}
 
-		ret = dirty_and_release_pages(NULL, root, file, pages,
-					      num_pages, pos, write_bytes);
 		btrfs_drop_pages(pages, num_pages);
 		if (ret) {
-			btrfs_free_reserved_data_space(root, inode,
-						       write_bytes);
+			btrfs_delalloc_release_space(inode, write_bytes);
 			goto out;
 		}
 
@@ -965,8 +1007,6 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 			btrfs_throttle(root);
 		}
 
-		buf += write_bytes;
-		count -= write_bytes;
 		pos += write_bytes;
 		num_written += write_bytes;
 
@@ -976,9 +1016,7 @@ out:
 	mutex_unlock(&inode->i_mutex);
 	if (ret)
 		err = ret;
-	btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
 
-out_nolock:
 	kfree(pages);
 	if (pinned[0])
 		page_cache_release(pinned[0]);
@@ -1008,7 +1046,7 @@ out_nolock:
 			num_written = err;
 
 		if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
-			trans = btrfs_start_transaction(root, 1);
+			trans = btrfs_start_transaction(root, 0);
 			ret = btrfs_log_dentry_safe(trans, root,
 						    file->f_dentry);
 			if (ret == 0) {
@@ -1023,7 +1061,7 @@ out_nolock:
 				btrfs_end_transaction(trans, root);
 			}
 		}
-		if (file->f_flags & O_DIRECT) {
+		if (file->f_flags & O_DIRECT && buffered) {
 			invalidate_mapping_pages(inode->i_mapping,
 			      start_pos >> PAGE_CACHE_SHIFT,
 			     (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
@@ -1104,9 +1142,9 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 	if (file && file->private_data)
 		btrfs_ioctl_trans_end(file);
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans) {
-		ret = -ENOMEM;
+	trans = btrfs_start_transaction(root, 0);
+	if (IS_ERR(trans)) {
+		ret = PTR_ERR(trans);
 		goto out;
 	}
 
@@ -1161,7 +1199,7 @@ const struct file_operations btrfs_file_operations = {
 	.read		= do_sync_read,
 	.aio_read       = generic_file_aio_read,
 	.splice_read	= generic_file_splice_read,
-	.write		= btrfs_file_write,
+	.aio_write	= btrfs_file_aio_write,
 	.mmap		= btrfs_file_mmap,
 	.open		= generic_file_open,
 	.release	= btrfs_release_file,

fs/btrfs/inode-item.c

@@ -49,6 +49,33 @@ static int find_name_in_backref(struct btrfs_path *path, const char *name,
 	return 0;
 }
 
+struct btrfs_inode_ref *
+btrfs_lookup_inode_ref(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			const char *name, int name_len,
+			u64 inode_objectid, u64 ref_objectid, int mod)
+{
+	struct btrfs_key key;
+	struct btrfs_inode_ref *ref;
+	int ins_len = mod < 0 ? -1 : 0;
+	int cow = mod != 0;
+	int ret;
+
+	key.objectid = inode_objectid;
+	key.type = BTRFS_INODE_REF_KEY;
+	key.offset = ref_objectid;
+
+	ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret > 0)
+		return NULL;
+	if (!find_name_in_backref(path, name, name_len, &ref))
+		return NULL;
+	return ref;
+}
+
 int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   const char *name, int name_len,

fs/btrfs/inode.c

@@ -252,6 +252,7 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 				   inline_len, compressed_size,
 				   compressed_pages);
 	BUG_ON(ret);
+	btrfs_delalloc_release_metadata(inode, end + 1 - start);
 	btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 	return 0;
 }
@@ -414,6 +415,7 @@ again:
 		trans = btrfs_join_transaction(root, 1);
 		BUG_ON(!trans);
 		btrfs_set_trans_block_group(trans, inode);
+		trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 
 		/* lets try to make an inline extent */
 		if (ret || total_in < (actual_end - start)) {
@@ -439,7 +441,6 @@ again:
 			     start, end, NULL,
 			     EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 			     EXTENT_CLEAR_DELALLOC |
-			     EXTENT_CLEAR_ACCOUNTING |
 			     EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 
 			btrfs_end_transaction(trans, root);
@@ -697,6 +698,38 @@ retry:
 	return 0;
 }
 
+static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
+				      u64 num_bytes)
+{
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct extent_map *em;
+	u64 alloc_hint = 0;
+
+	read_lock(&em_tree->lock);
+	em = search_extent_mapping(em_tree, start, num_bytes);
+	if (em) {
+		/*
+		 * if block start isn't an actual block number then find the
+		 * first block in this inode and use that as a hint.  If that
+		 * block is also bogus then just don't worry about it.
+		 */
+		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
+			free_extent_map(em);
+			em = search_extent_mapping(em_tree, 0, 0);
+			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
+				alloc_hint = em->block_start;
+			if (em)
+				free_extent_map(em);
+		} else {
+			alloc_hint = em->block_start;
+			free_extent_map(em);
+		}
+	}
+	read_unlock(&em_tree->lock);
+
+	return alloc_hint;
+}
+
 /*
  * 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
@@ -734,6 +767,7 @@ static noinline int cow_file_range(struct inode *inode,
 	trans = btrfs_join_transaction(root, 1);
 	BUG_ON(!trans);
 	btrfs_set_trans_block_group(trans, inode);
+	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 
 	actual_end = min_t(u64, isize, end + 1);
 
@@ -753,7 +787,6 @@ static noinline int cow_file_range(struct inode *inode,
 				     EXTENT_CLEAR_UNLOCK_PAGE |
 				     EXTENT_CLEAR_UNLOCK |
 				     EXTENT_CLEAR_DELALLOC |
-				     EXTENT_CLEAR_ACCOUNTING |
 				     EXTENT_CLEAR_DIRTY |
 				     EXTENT_SET_WRITEBACK |
 				     EXTENT_END_WRITEBACK);
@@ -769,29 +802,7 @@ static noinline int cow_file_range(struct inode *inode,
 	BUG_ON(disk_num_bytes >
 	       btrfs_super_total_bytes(&root->fs_info->super_copy));
 
-
-	read_lock(&BTRFS_I(inode)->extent_tree.lock);
-	em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
-				   start, num_bytes);
-	if (em) {
-		/*
-		 * if block start isn't an actual block number then find the
-		 * first block in this inode and use that as a hint.  If that
-		 * block is also bogus then just don't worry about it.
-		 */
-		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
-			free_extent_map(em);
-			em = search_extent_mapping(em_tree, 0, 0);
-			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
-				alloc_hint = em->block_start;
-			if (em)
-				free_extent_map(em);
-		} else {
-			alloc_hint = em->block_start;
-			free_extent_map(em);
-		}
-	}
-	read_unlock(&BTRFS_I(inode)->extent_tree.lock);
+	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 
 	while (disk_num_bytes > 0) {
@@ -1174,6 +1185,13 @@ out_check:
 					       num_bytes, num_bytes, type);
 		BUG_ON(ret);
 
+		if (root->root_key.objectid ==
+		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
+			ret = btrfs_reloc_clone_csums(inode, cur_offset,
+						      num_bytes);
+			BUG_ON(ret);
+		}
+
 		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 				cur_offset, cur_offset + num_bytes - 1,
 				locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
@@ -1226,15 +1244,13 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
 }
 
 static int btrfs_split_extent_hook(struct inode *inode,
-				    struct extent_state *orig, u64 split)
+				   struct extent_state *orig, u64 split)
 {
+	/* not delalloc, ignore it */
 	if (!(orig->state & EXTENT_DELALLOC))
 		return 0;
 
-	spin_lock(&BTRFS_I(inode)->accounting_lock);
-	BTRFS_I(inode)->outstanding_extents++;
-	spin_unlock(&BTRFS_I(inode)->accounting_lock);
-
+	atomic_inc(&BTRFS_I(inode)->outstanding_extents);
 	return 0;
 }
 
@@ -1252,10 +1268,7 @@ static int btrfs_merge_extent_hook(struct inode *inode,
 	if (!(other->state & EXTENT_DELALLOC))
 		return 0;
 
-	spin_lock(&BTRFS_I(inode)->accounting_lock);
-	BTRFS_I(inode)->outstanding_extents--;
-	spin_unlock(&BTRFS_I(inode)->accounting_lock);
-
+	atomic_dec(&BTRFS_I(inode)->outstanding_extents);
 	return 0;
 }
 
@@ -1264,8 +1277,8 @@ static int btrfs_merge_extent_hook(struct inode *inode,
  * bytes in this file, and to maintain the list of inodes that
  * have pending delalloc work to be done.
  */
-static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
-		       unsigned long old, unsigned long bits)
+static int btrfs_set_bit_hook(struct inode *inode,
+			      struct extent_state *state, int *bits)
 {
 
 	/*
@@ -1273,17 +1286,18 @@ static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 	 * but in this case, we are only testeing for the DELALLOC
 	 * bit, which is only set or cleared with irqs on
 	 */
-	if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
+	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
 		struct btrfs_root *root = BTRFS_I(inode)->root;
+		u64 len = state->end + 1 - state->start;
 
-		spin_lock(&BTRFS_I(inode)->accounting_lock);
-		BTRFS_I(inode)->outstanding_extents++;
-		spin_unlock(&BTRFS_I(inode)->accounting_lock);
-		btrfs_delalloc_reserve_space(root, inode, end - start + 1);
+		if (*bits & EXTENT_FIRST_DELALLOC)
+			*bits &= ~EXTENT_FIRST_DELALLOC;
+		else
+			atomic_inc(&BTRFS_I(inode)->outstanding_extents);
 
 		spin_lock(&root->fs_info->delalloc_lock);
-		BTRFS_I(inode)->delalloc_bytes += end - start + 1;
-		root->fs_info->delalloc_bytes += end - start + 1;
+		BTRFS_I(inode)->delalloc_bytes += len;
+		root->fs_info->delalloc_bytes += len;
 		if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 			list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
 				      &root->fs_info->delalloc_inodes);
@@ -1297,45 +1311,32 @@ static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
  * extent_io.c clear_bit_hook, see set_bit_hook for why
  */
 static int btrfs_clear_bit_hook(struct inode *inode,
-				struct extent_state *state, unsigned long bits)
+				struct extent_state *state, int *bits)
 {
 	/*
 	 * set_bit and clear bit hooks normally require _irqsave/restore
 	 * but in this case, we are only testeing for the DELALLOC
 	 * bit, which is only set or cleared with irqs on
 	 */
-	if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
+	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
 		struct btrfs_root *root = BTRFS_I(inode)->root;
+		u64 len = state->end + 1 - state->start;
 
-		if (bits & EXTENT_DO_ACCOUNTING) {
-			spin_lock(&BTRFS_I(inode)->accounting_lock);
-			WARN_ON(!BTRFS_I(inode)->outstanding_extents);
-			BTRFS_I(inode)->outstanding_extents--;
-			spin_unlock(&BTRFS_I(inode)->accounting_lock);
-			btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
-		}
+		if (*bits & EXTENT_FIRST_DELALLOC)
+			*bits &= ~EXTENT_FIRST_DELALLOC;
+		else if (!(*bits & EXTENT_DO_ACCOUNTING))
+			atomic_dec(&BTRFS_I(inode)->outstanding_extents);
+
+		if (*bits & EXTENT_DO_ACCOUNTING)
+			btrfs_delalloc_release_metadata(inode, len);
+
+		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
+			btrfs_free_reserved_data_space(inode, len);
 
 		spin_lock(&root->fs_info->delalloc_lock);
-		if (state->end - state->start + 1 >
-		    root->fs_info->delalloc_bytes) {
-			printk(KERN_INFO "btrfs warning: delalloc account "
-			       "%llu %llu\n",
-			       (unsigned long long)
-			       state->end - state->start + 1,
-			       (unsigned long long)
-			       root->fs_info->delalloc_bytes);
-			btrfs_delalloc_free_space(root, inode, (u64)-1);
-			root->fs_info->delalloc_bytes = 0;
-			BTRFS_I(inode)->delalloc_bytes = 0;
-		} else {
-			btrfs_delalloc_free_space(root, inode,
-						  state->end -
-						  state->start + 1);
-			root->fs_info->delalloc_bytes -= state->end -
-				state->start + 1;
-			BTRFS_I(inode)->delalloc_bytes -= state->end -
-				state->start + 1;
-		}
+		root->fs_info->delalloc_bytes -= len;
+		BTRFS_I(inode)->delalloc_bytes -= len;
+
 		if (BTRFS_I(inode)->delalloc_bytes == 0 &&
 		    !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 			list_del_init(&BTRFS_I(inode)->delalloc_inodes);
@@ -1384,7 +1385,8 @@ int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
  */
 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
 				    struct bio *bio, int mirror_num,
-				    unsigned long bio_flags)
+				    unsigned long bio_flags,
+				    u64 bio_offset)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	int ret = 0;
@@ -1403,7 +1405,8 @@ static int __btrfs_submit_bio_start(struct inode *inode, int rw,
  * are inserted into the btree
  */
 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
-			  int mirror_num, unsigned long bio_flags)
+			  int mirror_num, unsigned long bio_flags,
+			  u64 bio_offset)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	return btrfs_map_bio(root, rw, bio, mirror_num, 1);
@@ -1414,7 +1417,8 @@ static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
  * on write, or reading the csums from the tree before a read
  */
 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
-			  int mirror_num, unsigned long bio_flags)
+			  int mirror_num, unsigned long bio_flags,
+			  u64 bio_offset)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	int ret = 0;
@@ -1439,7 +1443,8 @@ static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 		/* we're doing a write, do the async checksumming */
 		return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 				   inode, rw, bio, mirror_num,
-				   bio_flags, __btrfs_submit_bio_start,
+				   bio_flags, bio_offset,
+				   __btrfs_submit_bio_start,
 				   __btrfs_submit_bio_done);
 	}
 
@@ -1520,6 +1525,7 @@ again:
 		goto again;
 	}
 
+	BUG();
 	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
 	ClearPageChecked(page);
 out:
@@ -1650,7 +1656,7 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct btrfs_trans_handle *trans;
+	struct btrfs_trans_handle *trans = NULL;
 	struct btrfs_ordered_extent *ordered_extent = NULL;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 	struct extent_state *cached_state = NULL;
@@ -1668,9 +1674,10 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 		ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
 		if (!ret) {
 			trans = btrfs_join_transaction(root, 1);
+			btrfs_set_trans_block_group(trans, inode);
+			trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 			ret = btrfs_update_inode(trans, root, inode);
 			BUG_ON(ret);
-			btrfs_end_transaction(trans, root);
 		}
 		goto out;
 	}
@@ -1680,6 +1687,8 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 			 0, &cached_state, GFP_NOFS);
 
 	trans = btrfs_join_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
+	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 
 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
 		compressed = 1;
@@ -1711,12 +1720,13 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 	add_pending_csums(trans, inode, ordered_extent->file_offset,
 			  &ordered_extent->list);
 
-	/* this also removes the ordered extent from the tree */
 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
 	ret = btrfs_update_inode(trans, root, inode);
 	BUG_ON(ret);
-	btrfs_end_transaction(trans, root);
 out:
+	btrfs_delalloc_release_metadata(inode, ordered_extent->len);
+	if (trans)
+		btrfs_end_transaction(trans, root);
 	/* once for us */
 	btrfs_put_ordered_extent(ordered_extent);
 	/* once for the tree */
@@ -1838,7 +1848,7 @@ static int btrfs_io_failed_hook(struct bio *failed_bio,
 
 	BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 						      failrec->last_mirror,
-						      failrec->bio_flags);
+						      failrec->bio_flags, 0);
 	return 0;
 }
 
@@ -1992,33 +2002,197 @@ void btrfs_run_delayed_iputs(struct btrfs_root *root)
 	up_read(&root->fs_info->cleanup_work_sem);
 }
 
+/*
+ * calculate extra metadata reservation when snapshotting a subvolume
+ * contains orphan files.
+ */
+void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
+				struct btrfs_pending_snapshot *pending,
+				u64 *bytes_to_reserve)
+{
+	struct btrfs_root *root;
+	struct btrfs_block_rsv *block_rsv;
+	u64 num_bytes;
+	int index;
+
+	root = pending->root;
+	if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
+		return;
+
+	block_rsv = root->orphan_block_rsv;
+
+	/* orphan block reservation for the snapshot */
+	num_bytes = block_rsv->size;
+
+	/*
+	 * after the snapshot is created, COWing tree blocks may use more
+	 * space than it frees. So we should make sure there is enough
+	 * reserved space.
+	 */
+	index = trans->transid & 0x1;
+	if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
+		num_bytes += block_rsv->size -
+			     (block_rsv->reserved + block_rsv->freed[index]);
+	}
+
+	*bytes_to_reserve += num_bytes;
+}
+
+void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
+				struct btrfs_pending_snapshot *pending)
+{
+	struct btrfs_root *root = pending->root;
+	struct btrfs_root *snap = pending->snap;
+	struct btrfs_block_rsv *block_rsv;
+	u64 num_bytes;
+	int index;
+	int ret;
+
+	if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
+		return;
+
+	/* refill source subvolume's orphan block reservation */
+	block_rsv = root->orphan_block_rsv;
+	index = trans->transid & 0x1;
+	if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
+		num_bytes = block_rsv->size -
+			    (block_rsv->reserved + block_rsv->freed[index]);
+		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
+					      root->orphan_block_rsv,
+					      num_bytes);
+		BUG_ON(ret);
+	}
+
+	/* setup orphan block reservation for the snapshot */
+	block_rsv = btrfs_alloc_block_rsv(snap);
+	BUG_ON(!block_rsv);
+
+	btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
+	snap->orphan_block_rsv = block_rsv;
+
+	num_bytes = root->orphan_block_rsv->size;
+	ret = btrfs_block_rsv_migrate(&pending->block_rsv,
+				      block_rsv, num_bytes);
+	BUG_ON(ret);
+
+#if 0
+	/* insert orphan item for the snapshot */
+	WARN_ON(!root->orphan_item_inserted);
+	ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
+				       snap->root_key.objectid);
+	BUG_ON(ret);
+	snap->orphan_item_inserted = 1;
+#endif
+}
+
+enum btrfs_orphan_cleanup_state {
+	ORPHAN_CLEANUP_STARTED	= 1,
+	ORPHAN_CLEANUP_DONE	= 2,
+};
+
+/*
+ * This is called in transaction commmit time. If there are no orphan
+ * files in the subvolume, it removes orphan item and frees block_rsv
+ * structure.
+ */
+void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root)
+{
+	int ret;
+
+	if (!list_empty(&root->orphan_list) ||
+	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
+		return;
+
+	if (root->orphan_item_inserted &&
+	    btrfs_root_refs(&root->root_item) > 0) {
+		ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
+					    root->root_key.objectid);
+		BUG_ON(ret);
+		root->orphan_item_inserted = 0;
+	}
+
+	if (root->orphan_block_rsv) {
+		WARN_ON(root->orphan_block_rsv->size > 0);
+		btrfs_free_block_rsv(root, root->orphan_block_rsv);
+		root->orphan_block_rsv = NULL;
+	}
+}
+
 /*
  * This creates an orphan entry for the given inode in case something goes
  * wrong in the middle of an unlink/truncate.
+ *
+ * NOTE: caller of this function should reserve 5 units of metadata for
+ *	 this function.
  */
 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
-	int ret = 0;
+	struct btrfs_block_rsv *block_rsv = NULL;
+	int reserve = 0;
+	int insert = 0;
+	int ret;
+
+	if (!root->orphan_block_rsv) {
+		block_rsv = btrfs_alloc_block_rsv(root);
+		BUG_ON(!block_rsv);
+	}
 
-	spin_lock(&root->list_lock);
+	spin_lock(&root->orphan_lock);
+	if (!root->orphan_block_rsv) {
+		root->orphan_block_rsv = block_rsv;
+	} else if (block_rsv) {
+		btrfs_free_block_rsv(root, block_rsv);
+		block_rsv = NULL;
+	}
 
-	/* already on the orphan list, we're good */
-	if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
-		spin_unlock(&root->list_lock);
-		return 0;
+	if (list_empty(&BTRFS_I(inode)->i_orphan)) {
+		list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
+#if 0
+		/*
+		 * For proper ENOSPC handling, we should do orphan
+		 * cleanup when mounting. But this introduces backward
+		 * compatibility issue.
+		 */
+		if (!xchg(&root->orphan_item_inserted, 1))
+			insert = 2;
+		else
+			insert = 1;
+#endif
+		insert = 1;
+	} else {
+		WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
 	}
 
-	list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
+	if (!BTRFS_I(inode)->orphan_meta_reserved) {
+		BTRFS_I(inode)->orphan_meta_reserved = 1;
+		reserve = 1;
+	}
+	spin_unlock(&root->orphan_lock);
 
-	spin_unlock(&root->list_lock);
+	if (block_rsv)
+		btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
 
-	/*
-	 * insert an orphan item to track this unlinked/truncated file
-	 */
-	ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
+	/* grab metadata reservation from transaction handle */
+	if (reserve) {
+		ret = btrfs_orphan_reserve_metadata(trans, inode);
+		BUG_ON(ret);
+	}
 
-	return ret;
+	/* insert an orphan item to track this unlinked/truncated file */
+	if (insert >= 1) {
+		ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
+		BUG_ON(ret);
+	}
+
+	/* insert an orphan item to track subvolume contains orphan files */
+	if (insert >= 2) {
+		ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
+					       root->root_key.objectid);
+		BUG_ON(ret);
+	}
+	return 0;
 }
 
 /*
@@ -2028,26 +2202,31 @@ int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int delete_item = 0;
+	int release_rsv = 0;
 	int ret = 0;
 
-	spin_lock(&root->list_lock);
-
-	if (list_empty(&BTRFS_I(inode)->i_orphan)) {
-		spin_unlock(&root->list_lock);
-		return 0;
+	spin_lock(&root->orphan_lock);
+	if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
+		list_del_init(&BTRFS_I(inode)->i_orphan);
+		delete_item = 1;
 	}
 
-	list_del_init(&BTRFS_I(inode)->i_orphan);
-	if (!trans) {
-		spin_unlock(&root->list_lock);
-		return 0;
+	if (BTRFS_I(inode)->orphan_meta_reserved) {
+		BTRFS_I(inode)->orphan_meta_reserved = 0;
+		release_rsv = 1;
 	}
+	spin_unlock(&root->orphan_lock);
 
-	spin_unlock(&root->list_lock);
+	if (trans && delete_item) {
+		ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
+		BUG_ON(ret);
+	}
 
-	ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
+	if (release_rsv)
+		btrfs_orphan_release_metadata(inode);
 
-	return ret;
+	return 0;
 }
 
 /*
@@ -2064,7 +2243,7 @@ void btrfs_orphan_cleanup(struct btrfs_root *root)
 	struct inode *inode;
 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
 
-	if (!xchg(&root->clean_orphans, 0))
+	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
 		return;
 
 	path = btrfs_alloc_path();
@@ -2117,16 +2296,15 @@ void btrfs_orphan_cleanup(struct btrfs_root *root)
 		found_key.type = BTRFS_INODE_ITEM_KEY;
 		found_key.offset = 0;
 		inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
-		if (IS_ERR(inode))
-			break;
+		BUG_ON(IS_ERR(inode));
 
 		/*
 		 * add this inode to the orphan list so btrfs_orphan_del does
 		 * the proper thing when we hit it
 		 */
-		spin_lock(&root->list_lock);
+		spin_lock(&root->orphan_lock);
 		list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
-		spin_unlock(&root->list_lock);
+		spin_unlock(&root->orphan_lock);
 
 		/*
 		 * if this is a bad inode, means we actually succeeded in
@@ -2135,7 +2313,7 @@ void btrfs_orphan_cleanup(struct btrfs_root *root)
 		 * do a destroy_inode
 		 */
 		if (is_bad_inode(inode)) {
-			trans = btrfs_start_transaction(root, 1);
+			trans = btrfs_start_transaction(root, 0);
 			btrfs_orphan_del(trans, inode);
 			btrfs_end_transaction(trans, root);
 			iput(inode);
@@ -2153,13 +2331,23 @@ void btrfs_orphan_cleanup(struct btrfs_root *root)
 		/* this will do delete_inode and everything for us */
 		iput(inode);
 	}
+	btrfs_free_path(path);
+
+	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
+
+	if (root->orphan_block_rsv)
+		btrfs_block_rsv_release(root, root->orphan_block_rsv,
+					(u64)-1);
+
+	if (root->orphan_block_rsv || root->orphan_item_inserted) {
+		trans = btrfs_join_transaction(root, 1);
+		btrfs_end_transaction(trans, root);
+	}
 
 	if (nr_unlink)
 		printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 	if (nr_truncate)
 		printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
-
-	btrfs_free_path(path);
 }
 
 /*
@@ -2478,103 +2666,276 @@ out:
 	return ret;
 }
 
-static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
+/* helper to check if there is any shared block in the path */
+static int check_path_shared(struct btrfs_root *root,
+			     struct btrfs_path *path)
 {
-	struct btrfs_root *root;
-	struct btrfs_trans_handle *trans;
-	struct inode *inode = dentry->d_inode;
+	struct extent_buffer *eb;
+	int level;
 	int ret;
-	unsigned long nr = 0;
-
-	root = BTRFS_I(dir)->root;
-
-	/*
-	 * 5 items for unlink inode
-	 * 1 for orphan
-	 */
-	ret = btrfs_reserve_metadata_space(root, 6);
-	if (ret)
-		return ret;
+	u64 refs;
 
-	trans = btrfs_start_transaction(root, 1);
-	if (IS_ERR(trans)) {
-		btrfs_unreserve_metadata_space(root, 6);
-		return PTR_ERR(trans);
+	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
+		if (!path->nodes[level])
+			break;
+		eb = path->nodes[level];
+		if (!btrfs_block_can_be_shared(root, eb))
+			continue;
+		ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
+					       &refs, NULL);
+		if (refs > 1)
+			return 1;
 	}
-
-	btrfs_set_trans_block_group(trans, dir);
-
-	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
-
-	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
-				 dentry->d_name.name, dentry->d_name.len);
-
-	if (inode->i_nlink == 0)
-		ret = btrfs_orphan_add(trans, inode);
-
-	nr = trans->blocks_used;
-
-	btrfs_end_transaction_throttle(trans, root);
-	btrfs_unreserve_metadata_space(root, 6);
-	btrfs_btree_balance_dirty(root, nr);
-	return ret;
+	return 0;
 }
 
-int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
-			struct btrfs_root *root,
-			struct inode *dir, u64 objectid,
-			const char *name, int name_len)
+/*
+ * helper to start transaction for unlink and rmdir.
+ *
+ * unlink and rmdir are special in btrfs, they do not always free space.
+ * so in enospc case, we should make sure they will free space before
+ * allowing them to use the global metadata reservation.
+ */
+static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
+						       struct dentry *dentry)
 {
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
 	struct btrfs_path *path;
-	struct extent_buffer *leaf;
+	struct btrfs_inode_ref *ref;
 	struct btrfs_dir_item *di;
-	struct btrfs_key key;
+	struct inode *inode = dentry->d_inode;
 	u64 index;
+	int check_link = 1;
+	int err = -ENOSPC;
 	int ret;
 
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
+	trans = btrfs_start_transaction(root, 10);
+	if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
+		return trans;
 
-	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
-				   name, name_len, -1);
-	BUG_ON(!di || IS_ERR(di));
+	if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
+		return ERR_PTR(-ENOSPC);
 
-	leaf = path->nodes[0];
-	btrfs_dir_item_key_to_cpu(leaf, di, &key);
-	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
-	ret = btrfs_delete_one_dir_name(trans, root, path, di);
-	BUG_ON(ret);
-	btrfs_release_path(root, path);
+	/* check if there is someone else holds reference */
+	if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
+		return ERR_PTR(-ENOSPC);
 
-	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
-				 objectid, root->root_key.objectid,
-				 dir->i_ino, &index, name, name_len);
-	if (ret < 0) {
-		BUG_ON(ret != -ENOENT);
-		di = btrfs_search_dir_index_item(root, path, dir->i_ino,
-						 name, name_len);
-		BUG_ON(!di || IS_ERR(di));
+	if (atomic_read(&inode->i_count) > 2)
+		return ERR_PTR(-ENOSPC);
 
-		leaf = path->nodes[0];
-		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
-		btrfs_release_path(root, path);
-		index = key.offset;
+	if (xchg(&root->fs_info->enospc_unlink, 1))
+		return ERR_PTR(-ENOSPC);
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		root->fs_info->enospc_unlink = 0;
+		return ERR_PTR(-ENOMEM);
 	}
 
-	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
-					 index, name, name_len, -1);
-	BUG_ON(!di || IS_ERR(di));
+	trans = btrfs_start_transaction(root, 0);
+	if (IS_ERR(trans)) {
+		btrfs_free_path(path);
+		root->fs_info->enospc_unlink = 0;
+		return trans;
+	}
 
-	leaf = path->nodes[0];
-	btrfs_dir_item_key_to_cpu(leaf, di, &key);
-	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
-	ret = btrfs_delete_one_dir_name(trans, root, path, di);
-	BUG_ON(ret);
-	btrfs_release_path(root, path);
+	path->skip_locking = 1;
+	path->search_commit_root = 1;
 
-	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
-	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+	ret = btrfs_lookup_inode(trans, root, path,
+				&BTRFS_I(dir)->location, 0);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
+	if (ret == 0) {
+		if (check_path_shared(root, path))
+			goto out;
+	} else {
+		check_link = 0;
+	}
+	btrfs_release_path(root, path);
+
+	ret = btrfs_lookup_inode(trans, root, path,
+				&BTRFS_I(inode)->location, 0);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
+	if (ret == 0) {
+		if (check_path_shared(root, path))
+			goto out;
+	} else {
+		check_link = 0;
+	}
+	btrfs_release_path(root, path);
+
+	if (ret == 0 && S_ISREG(inode->i_mode)) {
+		ret = btrfs_lookup_file_extent(trans, root, path,
+					       inode->i_ino, (u64)-1, 0);
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+		BUG_ON(ret == 0);
+		if (check_path_shared(root, path))
+			goto out;
+		btrfs_release_path(root, path);
+	}
+
+	if (!check_link) {
+		err = 0;
+		goto out;
+	}
+
+	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+				dentry->d_name.name, dentry->d_name.len, 0);
+	if (IS_ERR(di)) {
+		err = PTR_ERR(di);
+		goto out;
+	}
+	if (di) {
+		if (check_path_shared(root, path))
+			goto out;
+	} else {
+		err = 0;
+		goto out;
+	}
+	btrfs_release_path(root, path);
+
+	ref = btrfs_lookup_inode_ref(trans, root, path,
+				dentry->d_name.name, dentry->d_name.len,
+				inode->i_ino, dir->i_ino, 0);
+	if (IS_ERR(ref)) {
+		err = PTR_ERR(ref);
+		goto out;
+	}
+	BUG_ON(!ref);
+	if (check_path_shared(root, path))
+		goto out;
+	index = btrfs_inode_ref_index(path->nodes[0], ref);
+	btrfs_release_path(root, path);
+
+	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
+				dentry->d_name.name, dentry->d_name.len, 0);
+	if (IS_ERR(di)) {
+		err = PTR_ERR(di);
+		goto out;
+	}
+	BUG_ON(ret == -ENOENT);
+	if (check_path_shared(root, path))
+		goto out;
+
+	err = 0;
+out:
+	btrfs_free_path(path);
+	if (err) {
+		btrfs_end_transaction(trans, root);
+		root->fs_info->enospc_unlink = 0;
+		return ERR_PTR(err);
+	}
+
+	trans->block_rsv = &root->fs_info->global_block_rsv;
+	return trans;
+}
+
+static void __unlink_end_trans(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	if (trans->block_rsv == &root->fs_info->global_block_rsv) {
+		BUG_ON(!root->fs_info->enospc_unlink);
+		root->fs_info->enospc_unlink = 0;
+	}
+	btrfs_end_transaction_throttle(trans, root);
+}
+
+static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
+{
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_trans_handle *trans;
+	struct inode *inode = dentry->d_inode;
+	int ret;
+	unsigned long nr = 0;
+
+	trans = __unlink_start_trans(dir, dentry);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	btrfs_set_trans_block_group(trans, dir);
+
+	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
+
+	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
+				 dentry->d_name.name, dentry->d_name.len);
+	BUG_ON(ret);
+
+	if (inode->i_nlink == 0) {
+		ret = btrfs_orphan_add(trans, inode);
+		BUG_ON(ret);
+	}
+
+	nr = trans->blocks_used;
+	__unlink_end_trans(trans, root);
+	btrfs_btree_balance_dirty(root, nr);
+	return ret;
+}
+
+int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct inode *dir, u64 objectid,
+			const char *name, int name_len)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_dir_item *di;
+	struct btrfs_key key;
+	u64 index;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+				   name, name_len, -1);
+	BUG_ON(!di || IS_ERR(di));
+
+	leaf = path->nodes[0];
+	btrfs_dir_item_key_to_cpu(leaf, di, &key);
+	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
+	ret = btrfs_delete_one_dir_name(trans, root, path, di);
+	BUG_ON(ret);
+	btrfs_release_path(root, path);
+
+	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
+				 objectid, root->root_key.objectid,
+				 dir->i_ino, &index, name, name_len);
+	if (ret < 0) {
+		BUG_ON(ret != -ENOENT);
+		di = btrfs_search_dir_index_item(root, path, dir->i_ino,
+						 name, name_len);
+		BUG_ON(!di || IS_ERR(di));
+
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		btrfs_release_path(root, path);
+		index = key.offset;
+	}
+
+	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
+					 index, name, name_len, -1);
+	BUG_ON(!di || IS_ERR(di));
+
+	leaf = path->nodes[0];
+	btrfs_dir_item_key_to_cpu(leaf, di, &key);
+	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
+	ret = btrfs_delete_one_dir_name(trans, root, path, di);
+	BUG_ON(ret);
+	btrfs_release_path(root, path);
+
+	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
+	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 	ret = btrfs_update_inode(trans, root, dir);
 	BUG_ON(ret);
 	dir->i_sb->s_dirt = 1;
@@ -2587,7 +2948,6 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 {
 	struct inode *inode = dentry->d_inode;
 	int err = 0;
-	int ret;
 	struct btrfs_root *root = BTRFS_I(dir)->root;
 	struct btrfs_trans_handle *trans;
 	unsigned long nr = 0;
@@ -2596,15 +2956,9 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 	    inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 		return -ENOTEMPTY;
 
-	ret = btrfs_reserve_metadata_space(root, 5);
-	if (ret)
-		return ret;
-
-	trans = btrfs_start_transaction(root, 1);
-	if (IS_ERR(trans)) {
-		btrfs_unreserve_metadata_space(root, 5);
+	trans = __unlink_start_trans(dir, dentry);
+	if (IS_ERR(trans))
 		return PTR_ERR(trans);
-	}
 
 	btrfs_set_trans_block_group(trans, dir);
 
@@ -2627,12 +2981,9 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 		btrfs_i_size_write(inode, 0);
 out:
 	nr = trans->blocks_used;
-	ret = btrfs_end_transaction_throttle(trans, root);
-	btrfs_unreserve_metadata_space(root, 5);
+	__unlink_end_trans(trans, root);
 	btrfs_btree_balance_dirty(root, nr);
 
-	if (ret && !err)
-		err = ret;
 	return err;
 }
 
@@ -3029,6 +3380,7 @@ out:
 	if (pending_del_nr) {
 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
 				      pending_del_nr);
+		BUG_ON(ret);
 	}
 	btrfs_free_path(path);
 	return err;
@@ -3056,11 +3408,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 
 	if ((offset & (blocksize - 1)) == 0)
 		goto out;
-	ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
-	if (ret)
-		goto out;
-
-	ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
+	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
 	if (ret)
 		goto out;
 
@@ -3068,8 +3416,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 again:
 	page = grab_cache_page(mapping, index);
 	if (!page) {
-		btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
-		btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
+		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 		goto out;
 	}
 
@@ -3132,8 +3479,7 @@ again:
 
 out_unlock:
 	if (ret)
-		btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
-	btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
+		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 	unlock_page(page);
 	page_cache_release(page);
 out:
@@ -3145,7 +3491,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
 	struct btrfs_trans_handle *trans;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
-	struct extent_map *em;
+	struct extent_map *em = NULL;
 	struct extent_state *cached_state = NULL;
 	u64 mask = root->sectorsize - 1;
 	u64 hole_start = (inode->i_size + mask) & ~mask;
@@ -3183,11 +3529,11 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
 			u64 hint_byte = 0;
 			hole_size = last_byte - cur_offset;
 
-			err = btrfs_reserve_metadata_space(root, 2);
-			if (err)
+			trans = btrfs_start_transaction(root, 2);
+			if (IS_ERR(trans)) {
+				err = PTR_ERR(trans);
 				break;
-
-			trans = btrfs_start_transaction(root, 1);
+			}
 			btrfs_set_trans_block_group(trans, inode);
 
 			err = btrfs_drop_extents(trans, inode, cur_offset,
@@ -3205,14 +3551,15 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
 					last_byte - 1, 0);
 
 			btrfs_end_transaction(trans, root);
-			btrfs_unreserve_metadata_space(root, 2);
 		}
 		free_extent_map(em);
+		em = NULL;
 		cur_offset = last_byte;
 		if (cur_offset >= block_end)
 			break;
 	}
 
+	free_extent_map(em);
 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
 			     GFP_NOFS);
 	return err;
@@ -3239,11 +3586,10 @@ static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
 		}
 	}
 
-	ret = btrfs_reserve_metadata_space(root, 1);
-	if (ret)
-		return ret;
+	trans = btrfs_start_transaction(root, 5);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-	trans = btrfs_start_transaction(root, 1);
 	btrfs_set_trans_block_group(trans, inode);
 
 	ret = btrfs_orphan_add(trans, inode);
@@ -3251,7 +3597,6 @@ static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
 
 	nr = trans->blocks_used;
 	btrfs_end_transaction(trans, root);
-	btrfs_unreserve_metadata_space(root, 1);
 	btrfs_btree_balance_dirty(root, nr);
 
 	if (attr->ia_size > inode->i_size) {
@@ -3264,8 +3609,11 @@ static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
 		i_size_write(inode, attr->ia_size);
 		btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
 
-		trans = btrfs_start_transaction(root, 1);
+		trans = btrfs_start_transaction(root, 0);
+		BUG_ON(IS_ERR(trans));
 		btrfs_set_trans_block_group(trans, inode);
+		trans->block_rsv = root->orphan_block_rsv;
+		BUG_ON(!trans->block_rsv);
 
 		ret = btrfs_update_inode(trans, root, inode);
 		BUG_ON(ret);
@@ -3345,10 +3693,21 @@ void btrfs_delete_inode(struct inode *inode)
 	btrfs_i_size_write(inode, 0);
 
 	while (1) {
-		trans = btrfs_start_transaction(root, 1);
+		trans = btrfs_start_transaction(root, 0);
+		BUG_ON(IS_ERR(trans));
 		btrfs_set_trans_block_group(trans, inode);
-		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
+		trans->block_rsv = root->orphan_block_rsv;
+
+		ret = btrfs_block_rsv_check(trans, root,
+					    root->orphan_block_rsv, 0, 5);
+		if (ret) {
+			BUG_ON(ret != -EAGAIN);
+			ret = btrfs_commit_transaction(trans, root);
+			BUG_ON(ret);
+			continue;
+		}
 
+		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
 		if (ret != -EAGAIN)
 			break;
 
@@ -3356,6 +3715,7 @@ void btrfs_delete_inode(struct inode *inode)
 		btrfs_end_transaction(trans, root);
 		trans = NULL;
 		btrfs_btree_balance_dirty(root, nr);
+
 	}
 
 	if (ret == 0) {
@@ -3596,40 +3956,10 @@ again:
 	return 0;
 }
 
-static noinline void init_btrfs_i(struct inode *inode)
-{
-	struct btrfs_inode *bi = BTRFS_I(inode);
-
-	bi->generation = 0;
-	bi->sequence = 0;
-	bi->last_trans = 0;
-	bi->last_sub_trans = 0;
-	bi->logged_trans = 0;
-	bi->delalloc_bytes = 0;
-	bi->reserved_bytes = 0;
-	bi->disk_i_size = 0;
-	bi->flags = 0;
-	bi->index_cnt = (u64)-1;
-	bi->last_unlink_trans = 0;
-	bi->ordered_data_close = 0;
-	bi->force_compress = 0;
-	extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
-	extent_io_tree_init(&BTRFS_I(inode)->io_tree,
-			     inode->i_mapping, GFP_NOFS);
-	extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
-			     inode->i_mapping, GFP_NOFS);
-	INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
-	INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
-	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
-	btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
-	mutex_init(&BTRFS_I(inode)->log_mutex);
-}
-
 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 {
 	struct btrfs_iget_args *args = p;
 	inode->i_ino = args->ino;
-	init_btrfs_i(inode);
 	BTRFS_I(inode)->root = args->root;
 	btrfs_set_inode_space_info(args->root, inode);
 	return 0;
@@ -3692,8 +4022,6 @@ static struct inode *new_simple_dir(struct super_block *s,
 	if (!inode)
 		return ERR_PTR(-ENOMEM);
 
-	init_btrfs_i(inode);
-
 	BTRFS_I(inode)->root = root;
 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
 	BTRFS_I(inode)->dummy_inode = 1;
@@ -3950,7 +4278,7 @@ int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 	struct btrfs_trans_handle *trans;
 	int ret = 0;
 
-	if (root->fs_info->btree_inode == inode)
+	if (BTRFS_I(inode)->dummy_inode)
 		return 0;
 
 	if (wbc->sync_mode == WB_SYNC_ALL) {
@@ -3971,10 +4299,38 @@ void btrfs_dirty_inode(struct inode *inode)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_trans_handle *trans;
+	int ret;
+
+	if (BTRFS_I(inode)->dummy_inode)
+		return;
 
 	trans = btrfs_join_transaction(root, 1);
 	btrfs_set_trans_block_group(trans, inode);
-	btrfs_update_inode(trans, root, inode);
+
+	ret = btrfs_update_inode(trans, root, inode);
+	if (ret && ret == -ENOSPC) {
+		/* whoops, lets try again with the full transaction */
+		btrfs_end_transaction(trans, root);
+		trans = btrfs_start_transaction(root, 1);
+		if (IS_ERR(trans)) {
+			if (printk_ratelimit()) {
+				printk(KERN_ERR "btrfs: fail to "
+				       "dirty  inode %lu error %ld\n",
+				       inode->i_ino, PTR_ERR(trans));
+			}
+			return;
+		}
+		btrfs_set_trans_block_group(trans, inode);
+
+		ret = btrfs_update_inode(trans, root, inode);
+		if (ret) {
+			if (printk_ratelimit()) {
+				printk(KERN_ERR "btrfs: fail to "
+				       "dirty  inode %lu error %d\n",
+				       inode->i_ino, ret);
+			}
+		}
+	}
 	btrfs_end_transaction(trans, root);
 }
 
@@ -4092,7 +4448,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 	 * btrfs_get_inode_index_count has an explanation for the magic
 	 * number
 	 */
-	init_btrfs_i(inode);
 	BTRFS_I(inode)->index_cnt = 2;
 	BTRFS_I(inode)->root = root;
 	BTRFS_I(inode)->generation = trans->transid;
@@ -4247,26 +4602,21 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
 	if (!new_valid_dev(rdev))
 		return -EINVAL;
 
+	err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
+	if (err)
+		return err;
+
 	/*
 	 * 2 for inode item and ref
 	 * 2 for dir items
 	 * 1 for xattr if selinux is on
 	 */
-	err = btrfs_reserve_metadata_space(root, 5);
-	if (err)
-		return err;
+	trans = btrfs_start_transaction(root, 5);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans)
-		goto fail;
 	btrfs_set_trans_block_group(trans, dir);
 
-	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
-	if (err) {
-		err = -ENOSPC;
-		goto out_unlock;
-	}
-
 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
 				dentry->d_name.len,
 				dentry->d_parent->d_inode->i_ino, objectid,
@@ -4295,13 +4645,11 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
 out_unlock:
 	nr = trans->blocks_used;
 	btrfs_end_transaction_throttle(trans, root);
-fail:
-	btrfs_unreserve_metadata_space(root, 5);
+	btrfs_btree_balance_dirty(root, nr);
 	if (drop_inode) {
 		inode_dec_link_count(inode);
 		iput(inode);
 	}
-	btrfs_btree_balance_dirty(root, nr);
 	return err;
 }
 
@@ -4311,32 +4659,26 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 	struct btrfs_trans_handle *trans;
 	struct btrfs_root *root = BTRFS_I(dir)->root;
 	struct inode *inode = NULL;
-	int err;
 	int drop_inode = 0;
+	int err;
 	unsigned long nr = 0;
 	u64 objectid;
 	u64 index = 0;
 
+	err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
+	if (err)
+		return err;
 	/*
 	 * 2 for inode item and ref
 	 * 2 for dir items
 	 * 1 for xattr if selinux is on
 	 */
-	err = btrfs_reserve_metadata_space(root, 5);
-	if (err)
-		return err;
+	trans = btrfs_start_transaction(root, 5);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans)
-		goto fail;
 	btrfs_set_trans_block_group(trans, dir);
 
-	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
-	if (err) {
-		err = -ENOSPC;
-		goto out_unlock;
-	}
-
 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
 				dentry->d_name.len,
 				dentry->d_parent->d_inode->i_ino,
@@ -4368,8 +4710,6 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 out_unlock:
 	nr = trans->blocks_used;
 	btrfs_end_transaction_throttle(trans, root);
-fail:
-	btrfs_unreserve_metadata_space(root, 5);
 	if (drop_inode) {
 		inode_dec_link_count(inode);
 		iput(inode);
@@ -4396,21 +4736,21 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
 	if (root->objectid != BTRFS_I(inode)->root->objectid)
 		return -EPERM;
 
-	/*
-	 * 1 item for inode ref
-	 * 2 items for dir items
-	 */
-	err = btrfs_reserve_metadata_space(root, 3);
-	if (err)
-		return err;
-
 	btrfs_inc_nlink(inode);
 
 	err = btrfs_set_inode_index(dir, &index);
 	if (err)
 		goto fail;
 
-	trans = btrfs_start_transaction(root, 1);
+	/*
+	 * 1 item for inode ref
+	 * 2 items for dir items
+	 */
+	trans = btrfs_start_transaction(root, 3);
+	if (IS_ERR(trans)) {
+		err = PTR_ERR(trans);
+		goto fail;
+	}
 
 	btrfs_set_trans_block_group(trans, dir);
 	atomic_inc(&inode->i_count);
@@ -4429,7 +4769,6 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
 	nr = trans->blocks_used;
 	btrfs_end_transaction_throttle(trans, root);
 fail:
-	btrfs_unreserve_metadata_space(root, 3);
 	if (drop_inode) {
 		inode_dec_link_count(inode);
 		iput(inode);
@@ -4449,28 +4788,20 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 	u64 index = 0;
 	unsigned long nr = 1;
 
+	err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
+	if (err)
+		return err;
+
 	/*
 	 * 2 items for inode and ref
 	 * 2 items for dir items
 	 * 1 for xattr if selinux is on
 	 */
-	err = btrfs_reserve_metadata_space(root, 5);
-	if (err)
-		return err;
-
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans) {
-		err = -ENOMEM;
-		goto out_unlock;
-	}
+	trans = btrfs_start_transaction(root, 5);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 	btrfs_set_trans_block_group(trans, dir);
 
-	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
-	if (err) {
-		err = -ENOSPC;
-		goto out_fail;
-	}
-
 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
 				dentry->d_name.len,
 				dentry->d_parent->d_inode->i_ino, objectid,
@@ -4510,9 +4841,6 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 out_fail:
 	nr = trans->blocks_used;
 	btrfs_end_transaction_throttle(trans, root);
-
-out_unlock:
-	btrfs_unreserve_metadata_space(root, 5);
 	if (drop_on_err)
 		iput(inode);
 	btrfs_btree_balance_dirty(root, nr);
@@ -4770,6 +5098,7 @@ again:
 			}
 			flush_dcache_page(page);
 		} else if (create && PageUptodate(page)) {
+			WARN_ON(1);
 			if (!trans) {
 				kunmap(page);
 				free_extent_map(em);
@@ -4866,11 +5195,651 @@ out:
 	return em;
 }
 
+static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
+						  u64 start, u64 len)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	struct extent_map *em;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct btrfs_key ins;
+	u64 alloc_hint;
+	int ret;
+
+	btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
+
+	trans = btrfs_join_transaction(root, 0);
+	if (!trans)
+		return ERR_PTR(-ENOMEM);
+
+	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
+
+	alloc_hint = get_extent_allocation_hint(inode, start, len);
+	ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
+				   alloc_hint, (u64)-1, &ins, 1);
+	if (ret) {
+		em = ERR_PTR(ret);
+		goto out;
+	}
+
+	em = alloc_extent_map(GFP_NOFS);
+	if (!em) {
+		em = ERR_PTR(-ENOMEM);
+		goto out;
+	}
+
+	em->start = start;
+	em->orig_start = em->start;
+	em->len = ins.offset;
+
+	em->block_start = ins.objectid;
+	em->block_len = ins.offset;
+	em->bdev = root->fs_info->fs_devices->latest_bdev;
+	set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	while (1) {
+		write_lock(&em_tree->lock);
+		ret = add_extent_mapping(em_tree, em);
+		write_unlock(&em_tree->lock);
+		if (ret != -EEXIST)
+			break;
+		btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
+	}
+
+	ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
+					   ins.offset, ins.offset, 0);
+	if (ret) {
+		btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
+		em = ERR_PTR(ret);
+	}
+out:
+	btrfs_end_transaction(trans, root);
+	return em;
+}
+
+/*
+ * returns 1 when the nocow is safe, < 1 on error, 0 if the
+ * block must be cow'd
+ */
+static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
+				      struct inode *inode, u64 offset, u64 len)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct extent_buffer *leaf;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_key key;
+	u64 disk_bytenr;
+	u64 backref_offset;
+	u64 extent_end;
+	u64 num_bytes;
+	int slot;
+	int found_type;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+				       offset, 0);
+	if (ret < 0)
+		goto out;
+
+	slot = path->slots[0];
+	if (ret == 1) {
+		if (slot == 0) {
+			/* can't find the item, must cow */
+			ret = 0;
+			goto out;
+		}
+		slot--;
+	}
+	ret = 0;
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, slot);
+	if (key.objectid != inode->i_ino ||
+	    key.type != BTRFS_EXTENT_DATA_KEY) {
+		/* not our file or wrong item type, must cow */
+		goto out;
+	}
+
+	if (key.offset > offset) {
+		/* Wrong offset, must cow */
+		goto out;
+	}
+
+	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
+	found_type = btrfs_file_extent_type(leaf, fi);
+	if (found_type != BTRFS_FILE_EXTENT_REG &&
+	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
+		/* not a regular extent, must cow */
+		goto out;
+	}
+	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+	backref_offset = btrfs_file_extent_offset(leaf, fi);
+
+	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
+	if (extent_end < offset + len) {
+		/* extent doesn't include our full range, must cow */
+		goto out;
+	}
+
+	if (btrfs_extent_readonly(root, disk_bytenr))
+		goto out;
+
+	/*
+	 * look for other files referencing this extent, if we
+	 * find any we must cow
+	 */
+	if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
+				  key.offset - backref_offset, disk_bytenr))
+		goto out;
+
+	/*
+	 * adjust disk_bytenr and num_bytes to cover just the bytes
+	 * in this extent we are about to write.  If there
+	 * are any csums in that range we have to cow in order
+	 * to keep the csums correct
+	 */
+	disk_bytenr += backref_offset;
+	disk_bytenr += offset - key.offset;
+	num_bytes = min(offset + len, extent_end) - offset;
+	if (csum_exist_in_range(root, disk_bytenr, num_bytes))
+				goto out;
+	/*
+	 * all of the above have passed, it is safe to overwrite this extent
+	 * without cow
+	 */
+	ret = 1;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
+				   struct buffer_head *bh_result, int create)
+{
+	struct extent_map *em;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	u64 start = iblock << inode->i_blkbits;
+	u64 len = bh_result->b_size;
+	struct btrfs_trans_handle *trans;
+
+	em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
+	if (IS_ERR(em))
+		return PTR_ERR(em);
+
+	/*
+	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
+	 * io.  INLINE is special, and we could probably kludge it in here, but
+	 * it's still buffered so for safety lets just fall back to the generic
+	 * buffered path.
+	 *
+	 * For COMPRESSED we _have_ to read the entire extent in so we can
+	 * decompress it, so there will be buffering required no matter what we
+	 * do, so go ahead and fallback to buffered.
+	 *
+	 * We return -ENOTBLK because thats what makes DIO go ahead and go back
+	 * to buffered IO.  Don't blame me, this is the price we pay for using
+	 * the generic code.
+	 */
+	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
+	    em->block_start == EXTENT_MAP_INLINE) {
+		free_extent_map(em);
+		return -ENOTBLK;
+	}
+
+	/* Just a good old fashioned hole, return */
+	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
+			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
+		free_extent_map(em);
+		/* DIO will do one hole at a time, so just unlock a sector */
+		unlock_extent(&BTRFS_I(inode)->io_tree, start,
+			      start + root->sectorsize - 1, GFP_NOFS);
+		return 0;
+	}
+
+	/*
+	 * We don't allocate a new extent in the following cases
+	 *
+	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
+	 * existing extent.
+	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
+	 * just use the extent.
+	 *
+	 */
+	if (!create) {
+		len = em->len - (start - em->start);
+		goto map;
+	}
+
+	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
+	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
+	     em->block_start != EXTENT_MAP_HOLE)) {
+		int type;
+		int ret;
+		u64 block_start;
+
+		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
+			type = BTRFS_ORDERED_PREALLOC;
+		else
+			type = BTRFS_ORDERED_NOCOW;
+		len = min(len, em->len - (start - em->start));
+		block_start = em->block_start + (start - em->start);
+
+		/*
+		 * we're not going to log anything, but we do need
+		 * to make sure the current transaction stays open
+		 * while we look for nocow cross refs
+		 */
+		trans = btrfs_join_transaction(root, 0);
+		if (!trans)
+			goto must_cow;
+
+		if (can_nocow_odirect(trans, inode, start, len) == 1) {
+			ret = btrfs_add_ordered_extent_dio(inode, start,
+					   block_start, len, len, type);
+			btrfs_end_transaction(trans, root);
+			if (ret) {
+				free_extent_map(em);
+				return ret;
+			}
+			goto unlock;
+		}
+		btrfs_end_transaction(trans, root);
+	}
+must_cow:
+	/*
+	 * this will cow the extent, reset the len in case we changed
+	 * it above
+	 */
+	len = bh_result->b_size;
+	free_extent_map(em);
+	em = btrfs_new_extent_direct(inode, start, len);
+	if (IS_ERR(em))
+		return PTR_ERR(em);
+	len = min(len, em->len - (start - em->start));
+unlock:
+	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
+			  EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
+			  0, NULL, GFP_NOFS);
+map:
+	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
+		inode->i_blkbits;
+	bh_result->b_size = len;
+	bh_result->b_bdev = em->bdev;
+	set_buffer_mapped(bh_result);
+	if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
+		set_buffer_new(bh_result);
+
+	free_extent_map(em);
+
+	return 0;
+}
+
+struct btrfs_dio_private {
+	struct inode *inode;
+	u64 logical_offset;
+	u64 disk_bytenr;
+	u64 bytes;
+	u32 *csums;
+	void *private;
+};
+
+static void btrfs_endio_direct_read(struct bio *bio, int err)
+{
+	struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct bio_vec *bvec = bio->bi_io_vec;
+	struct btrfs_dio_private *dip = bio->bi_private;
+	struct inode *inode = dip->inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	u64 start;
+	u32 *private = dip->csums;
+
+	start = dip->logical_offset;
+	do {
+		if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
+			struct page *page = bvec->bv_page;
+			char *kaddr;
+			u32 csum = ~(u32)0;
+			unsigned long flags;
+
+			local_irq_save(flags);
+			kaddr = kmap_atomic(page, KM_IRQ0);
+			csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
+					       csum, bvec->bv_len);
+			btrfs_csum_final(csum, (char *)&csum);
+			kunmap_atomic(kaddr, KM_IRQ0);
+			local_irq_restore(flags);
+
+			flush_dcache_page(bvec->bv_page);
+			if (csum != *private) {
+				printk(KERN_ERR "btrfs csum failed ino %lu off"
+				      " %llu csum %u private %u\n",
+				      inode->i_ino, (unsigned long long)start,
+				      csum, *private);
+				err = -EIO;
+			}
+		}
+
+		start += bvec->bv_len;
+		private++;
+		bvec++;
+	} while (bvec <= bvec_end);
+
+	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
+		      dip->logical_offset + dip->bytes - 1, GFP_NOFS);
+	bio->bi_private = dip->private;
+
+	kfree(dip->csums);
+	kfree(dip);
+	dio_end_io(bio, err);
+}
+
+static void btrfs_endio_direct_write(struct bio *bio, int err)
+{
+	struct btrfs_dio_private *dip = bio->bi_private;
+	struct inode *inode = dip->inode;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_ordered_extent *ordered = NULL;
+	struct extent_state *cached_state = NULL;
+	int ret;
+
+	if (err)
+		goto out_done;
+
+	ret = btrfs_dec_test_ordered_pending(inode, &ordered,
+					     dip->logical_offset, dip->bytes);
+	if (!ret)
+		goto out_done;
+
+	BUG_ON(!ordered);
+
+	trans = btrfs_join_transaction(root, 1);
+	if (!trans) {
+		err = -ENOMEM;
+		goto out;
+	}
+	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
+
+	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
+		ret = btrfs_ordered_update_i_size(inode, 0, ordered);
+		if (!ret)
+			ret = btrfs_update_inode(trans, root, inode);
+		err = ret;
+		goto out;
+	}
+
+	lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
+			 ordered->file_offset + ordered->len - 1, 0,
+			 &cached_state, GFP_NOFS);
+
+	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
+		ret = btrfs_mark_extent_written(trans, inode,
+						ordered->file_offset,
+						ordered->file_offset +
+						ordered->len);
+		if (ret) {
+			err = ret;
+			goto out_unlock;
+		}
+	} else {
+		ret = insert_reserved_file_extent(trans, inode,
+						  ordered->file_offset,
+						  ordered->start,
+						  ordered->disk_len,
+						  ordered->len,
+						  ordered->len,
+						  0, 0, 0,
+						  BTRFS_FILE_EXTENT_REG);
+		unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
+				   ordered->file_offset, ordered->len);
+		if (ret) {
+			err = ret;
+			WARN_ON(1);
+			goto out_unlock;
+		}
+	}
+
+	add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
+	btrfs_ordered_update_i_size(inode, 0, ordered);
+	btrfs_update_inode(trans, root, inode);
+out_unlock:
+	unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
+			     ordered->file_offset + ordered->len - 1,
+			     &cached_state, GFP_NOFS);
+out:
+	btrfs_delalloc_release_metadata(inode, ordered->len);
+	btrfs_end_transaction(trans, root);
+	btrfs_put_ordered_extent(ordered);
+	btrfs_put_ordered_extent(ordered);
+out_done:
+	bio->bi_private = dip->private;
+
+	kfree(dip->csums);
+	kfree(dip);
+	dio_end_io(bio, err);
+}
+
+static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
+				    struct bio *bio, int mirror_num,
+				    unsigned long bio_flags, u64 offset)
+{
+	int ret;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
+	BUG_ON(ret);
+	return 0;
+}
+
+static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
+				loff_t file_offset)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_dio_private *dip;
+	struct bio_vec *bvec = bio->bi_io_vec;
+	u64 start;
+	int skip_sum;
+	int write = rw & (1 << BIO_RW);
+	int ret = 0;
+
+	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
+
+	dip = kmalloc(sizeof(*dip), GFP_NOFS);
+	if (!dip) {
+		ret = -ENOMEM;
+		goto free_ordered;
+	}
+	dip->csums = NULL;
+
+	if (!skip_sum) {
+		dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
+		if (!dip->csums) {
+			ret = -ENOMEM;
+			goto free_ordered;
+		}
+	}
+
+	dip->private = bio->bi_private;
+	dip->inode = inode;
+	dip->logical_offset = file_offset;
+
+	start = dip->logical_offset;
+	dip->bytes = 0;
+	do {
+		dip->bytes += bvec->bv_len;
+		bvec++;
+	} while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
+
+	dip->disk_bytenr = (u64)bio->bi_sector << 9;
+	bio->bi_private = dip;
+
+	if (write)
+		bio->bi_end_io = btrfs_endio_direct_write;
+	else
+		bio->bi_end_io = btrfs_endio_direct_read;
+
+	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
+	if (ret)
+		goto out_err;
+
+	if (write && !skip_sum) {
+		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
+				   inode, rw, bio, 0, 0,
+				   dip->logical_offset,
+				   __btrfs_submit_bio_start_direct_io,
+				   __btrfs_submit_bio_done);
+		if (ret)
+			goto out_err;
+		return;
+	} else if (!skip_sum)
+		btrfs_lookup_bio_sums_dio(root, inode, bio,
+					  dip->logical_offset, dip->csums);
+
+	ret = btrfs_map_bio(root, rw, bio, 0, 1);
+	if (ret)
+		goto out_err;
+	return;
+out_err:
+	kfree(dip->csums);
+	kfree(dip);
+free_ordered:
+	/*
+	 * If this is a write, we need to clean up the reserved space and kill
+	 * the ordered extent.
+	 */
+	if (write) {
+		struct btrfs_ordered_extent *ordered;
+		ordered = btrfs_lookup_ordered_extent(inode,
+						      dip->logical_offset);
+		if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
+		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
+			btrfs_free_reserved_extent(root, ordered->start,
+						   ordered->disk_len);
+		btrfs_put_ordered_extent(ordered);
+		btrfs_put_ordered_extent(ordered);
+	}
+	bio_endio(bio, ret);
+}
+
+static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
+			const struct iovec *iov, loff_t offset,
+			unsigned long nr_segs)
+{
+	int seg;
+	size_t size;
+	unsigned long addr;
+	unsigned blocksize_mask = root->sectorsize - 1;
+	ssize_t retval = -EINVAL;
+	loff_t end = offset;
+
+	if (offset & blocksize_mask)
+		goto out;
+
+	/* Check the memory alignment.  Blocks cannot straddle pages */
+	for (seg = 0; seg < nr_segs; seg++) {
+		addr = (unsigned long)iov[seg].iov_base;
+		size = iov[seg].iov_len;
+		end += size;
+		if ((addr & blocksize_mask) || (size & blocksize_mask)) 
+			goto out;
+	}
+	retval = 0;
+out:
+	return retval;
+}
 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
 			const struct iovec *iov, loff_t offset,
 			unsigned long nr_segs)
 {
-	return -EINVAL;
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	struct btrfs_ordered_extent *ordered;
+	struct extent_state *cached_state = NULL;
+	u64 lockstart, lockend;
+	ssize_t ret;
+	int writing = rw & WRITE;
+	int write_bits = 0;
+	size_t count = iov_length(iov, nr_segs);
+
+	if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
+			    offset, nr_segs)) {
+		return 0;
+	}
+
+	lockstart = offset;
+	lockend = offset + count - 1;
+
+	if (writing) {
+		ret = btrfs_delalloc_reserve_space(inode, count);
+		if (ret)
+			goto out;
+	}
+
+	while (1) {
+		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
+				 0, &cached_state, GFP_NOFS);
+		/*
+		 * We're concerned with the entire range that we're going to be
+		 * doing DIO to, so we need to make sure theres no ordered
+		 * extents in this range.
+		 */
+		ordered = btrfs_lookup_ordered_range(inode, lockstart,
+						     lockend - lockstart + 1);
+		if (!ordered)
+			break;
+		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
+				     &cached_state, GFP_NOFS);
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		btrfs_put_ordered_extent(ordered);
+		cond_resched();
+	}
+
+	/*
+	 * we don't use btrfs_set_extent_delalloc because we don't want
+	 * the dirty or uptodate bits
+	 */
+	if (writing) {
+		write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
+		ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
+				     EXTENT_DELALLOC, 0, NULL, &cached_state,
+				     GFP_NOFS);
+		if (ret) {
+			clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
+					 lockend, EXTENT_LOCKED | write_bits,
+					 1, 0, &cached_state, GFP_NOFS);
+			goto out;
+		}
+	}
+
+	free_extent_state(cached_state);
+	cached_state = NULL;
+
+	ret = __blockdev_direct_IO(rw, iocb, inode,
+		   BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
+		   iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
+		   btrfs_submit_direct, 0);
+
+	if (ret < 0 && ret != -EIOCBQUEUED) {
+		clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
+			      offset + iov_length(iov, nr_segs) - 1,
+			      EXTENT_LOCKED | write_bits, 1, 0,
+			      &cached_state, GFP_NOFS);
+	} else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
+		/*
+		 * We're falling back to buffered, unlock the section we didn't
+		 * do IO on.
+		 */
+		clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
+			      offset + iov_length(iov, nr_segs) - 1,
+			      EXTENT_LOCKED | write_bits, 1, 0,
+			      &cached_state, GFP_NOFS);
+	}
+out:
+	free_extent_state(cached_state);
+	return ret;
 }
 
 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
@@ -5034,7 +6003,7 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 	u64 page_start;
 	u64 page_end;
 
-	ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
+	ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
 	if (ret) {
 		if (ret == -ENOMEM)
 			ret = VM_FAULT_OOM;
@@ -5043,13 +6012,6 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 		goto out;
 	}
 
-	ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
-	if (ret) {
-		btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
-		ret = VM_FAULT_SIGBUS;
-		goto out;
-	}
-
 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
 again:
 	lock_page(page);
@@ -5059,7 +6021,6 @@ again:
 
 	if ((page->mapping != inode->i_mapping) ||
 	    (page_start >= size)) {
-		btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
 		/* page got truncated out from underneath us */
 		goto out_unlock;
 	}
@@ -5100,7 +6061,6 @@ again:
 		unlock_extent_cached(io_tree, page_start, page_end,
 				     &cached_state, GFP_NOFS);
 		ret = VM_FAULT_SIGBUS;
-		btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
 		goto out_unlock;
 	}
 	ret = 0;
@@ -5127,10 +6087,10 @@ again:
 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
 
 out_unlock:
-	btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
 	if (!ret)
 		return VM_FAULT_LOCKED;
 	unlock_page(page);
+	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 out:
 	return ret;
 }
@@ -5155,8 +6115,10 @@ static void btrfs_truncate(struct inode *inode)
 	btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
 	btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_start_transaction(root, 0);
+	BUG_ON(IS_ERR(trans));
 	btrfs_set_trans_block_group(trans, inode);
+	trans->block_rsv = root->orphan_block_rsv;
 
 	/*
 	 * setattr is responsible for setting the ordered_data_close flag,
@@ -5179,6 +6141,23 @@ static void btrfs_truncate(struct inode *inode)
 		btrfs_add_ordered_operation(trans, root, inode);
 
 	while (1) {
+		if (!trans) {
+			trans = btrfs_start_transaction(root, 0);
+			BUG_ON(IS_ERR(trans));
+			btrfs_set_trans_block_group(trans, inode);
+			trans->block_rsv = root->orphan_block_rsv;
+		}
+
+		ret = btrfs_block_rsv_check(trans, root,
+					    root->orphan_block_rsv, 0, 5);
+		if (ret) {
+			BUG_ON(ret != -EAGAIN);
+			ret = btrfs_commit_transaction(trans, root);
+			BUG_ON(ret);
+			trans = NULL;
+			continue;
+		}
+
 		ret = btrfs_truncate_inode_items(trans, root, inode,
 						 inode->i_size,
 						 BTRFS_EXTENT_DATA_KEY);
@@ -5190,10 +6169,8 @@ static void btrfs_truncate(struct inode *inode)
 
 		nr = trans->blocks_used;
 		btrfs_end_transaction(trans, root);
+		trans = NULL;
 		btrfs_btree_balance_dirty(root, nr);
-
-		trans = btrfs_start_transaction(root, 1);
-		btrfs_set_trans_block_group(trans, inode);
 	}
 
 	if (ret == 0 && inode->i_nlink > 0) {
@@ -5254,21 +6231,47 @@ unsigned long btrfs_force_ra(struct address_space *mapping,
 struct inode *btrfs_alloc_inode(struct super_block *sb)
 {
 	struct btrfs_inode *ei;
+	struct inode *inode;
 
 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
 	if (!ei)
 		return NULL;
+
+	ei->root = NULL;
+	ei->space_info = NULL;
+	ei->generation = 0;
+	ei->sequence = 0;
 	ei->last_trans = 0;
 	ei->last_sub_trans = 0;
 	ei->logged_trans = 0;
-	ei->outstanding_extents = 0;
-	ei->reserved_extents = 0;
-	ei->root = NULL;
+	ei->delalloc_bytes = 0;
+	ei->reserved_bytes = 0;
+	ei->disk_i_size = 0;
+	ei->flags = 0;
+	ei->index_cnt = (u64)-1;
+	ei->last_unlink_trans = 0;
+
 	spin_lock_init(&ei->accounting_lock);
+	atomic_set(&ei->outstanding_extents, 0);
+	ei->reserved_extents = 0;
+
+	ei->ordered_data_close = 0;
+	ei->orphan_meta_reserved = 0;
+	ei->dummy_inode = 0;
+	ei->force_compress = 0;
+
+	inode = &ei->vfs_inode;
+	extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
+	extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
+	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
+	mutex_init(&ei->log_mutex);
 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
 	INIT_LIST_HEAD(&ei->i_orphan);
+	INIT_LIST_HEAD(&ei->delalloc_inodes);
 	INIT_LIST_HEAD(&ei->ordered_operations);
-	return &ei->vfs_inode;
+	RB_CLEAR_NODE(&ei->rb_node);
+
+	return inode;
 }
 
 void btrfs_destroy_inode(struct inode *inode)
@@ -5278,6 +6281,8 @@ 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(BTRFS_I(inode)->reserved_extents);
 
 	/*
 	 * This can happen where we create an inode, but somebody else also
@@ -5298,13 +6303,13 @@ void btrfs_destroy_inode(struct inode *inode)
 		spin_unlock(&root->fs_info->ordered_extent_lock);
 	}
 
-	spin_lock(&root->list_lock);
+	spin_lock(&root->orphan_lock);
 	if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 		printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
 		       inode->i_ino);
 		list_del_init(&BTRFS_I(inode)->i_orphan);
 	}
-	spin_unlock(&root->list_lock);
+	spin_unlock(&root->orphan_lock);
 
 	while (1) {
 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
@@ -5425,19 +6430,6 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 		return -ENOTEMPTY;
-
-	/*
-	 * We want to reserve the absolute worst case amount of items.  So if
-	 * both inodes are subvols and we need to unlink them then that would
-	 * require 4 item modifications, but if they are both normal inodes it
-	 * would require 5 item modifications, so we'll assume their normal
-	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
-	 * should cover the worst case number of items we'll modify.
-	 */
-	ret = btrfs_reserve_metadata_space(root, 11);
-	if (ret)
-		return ret;
-
 	/*
 	 * we're using rename to replace one file with another.
 	 * and the replacement file is large.  Start IO on it now so
@@ -5450,8 +6442,18 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	/* close the racy window with snapshot create/destroy ioctl */
 	if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 		down_read(&root->fs_info->subvol_sem);
+	/*
+	 * We want to reserve the absolute worst case amount of items.  So if
+	 * both inodes are subvols and we need to unlink them then that would
+	 * require 4 item modifications, but if they are both normal inodes it
+	 * would require 5 item modifications, so we'll assume their normal
+	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
+	 * should cover the worst case number of items we'll modify.
+	 */
+	trans = btrfs_start_transaction(root, 20);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-	trans = btrfs_start_transaction(root, 1);
 	btrfs_set_trans_block_group(trans, new_dir);
 
 	if (dest != root)
@@ -5550,7 +6552,6 @@ out_fail:
 	if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 		up_read(&root->fs_info->subvol_sem);
 
-	btrfs_unreserve_metadata_space(root, 11);
 	return ret;
 }
 
@@ -5602,6 +6603,38 @@ int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
 	return 0;
 }
 
+int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput)
+{
+	struct btrfs_inode *binode;
+	struct inode *inode = NULL;
+
+	spin_lock(&root->fs_info->delalloc_lock);
+	while (!list_empty(&root->fs_info->delalloc_inodes)) {
+		binode = list_entry(root->fs_info->delalloc_inodes.next,
+				    struct btrfs_inode, delalloc_inodes);
+		inode = igrab(&binode->vfs_inode);
+		if (inode) {
+			list_move_tail(&binode->delalloc_inodes,
+				       &root->fs_info->delalloc_inodes);
+			break;
+		}
+
+		list_del_init(&binode->delalloc_inodes);
+		cond_resched_lock(&root->fs_info->delalloc_lock);
+	}
+	spin_unlock(&root->fs_info->delalloc_lock);
+
+	if (inode) {
+		write_inode_now(inode, 0);
+		if (delay_iput)
+			btrfs_add_delayed_iput(inode);
+		else
+			iput(inode);
+		return 1;
+	}
+	return 0;
+}
+
 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 			 const char *symname)
 {
@@ -5625,26 +6658,20 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
 		return -ENAMETOOLONG;
 
+	err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
+	if (err)
+		return err;
 	/*
 	 * 2 items for inode item and ref
 	 * 2 items for dir items
 	 * 1 item for xattr if selinux is on
 	 */
-	err = btrfs_reserve_metadata_space(root, 5);
-	if (err)
-		return err;
+	trans = btrfs_start_transaction(root, 5);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans)
-		goto out_fail;
 	btrfs_set_trans_block_group(trans, dir);
 
-	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
-	if (err) {
-		err = -ENOSPC;
-		goto out_unlock;
-	}
-
 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
 				dentry->d_name.len,
 				dentry->d_parent->d_inode->i_ino, objectid,
@@ -5716,8 +6743,6 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 out_unlock:
 	nr = trans->blocks_used;
 	btrfs_end_transaction_throttle(trans, root);
-out_fail:
-	btrfs_unreserve_metadata_space(root, 5);
 	if (drop_inode) {
 		inode_dec_link_count(inode);
 		iput(inode);
@@ -5726,33 +6751,28 @@ out_fail:
 	return err;
 }
 
-static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
-			u64 alloc_hint, int mode, loff_t actual_len)
+int btrfs_prealloc_file_range(struct inode *inode, int mode,
+			      u64 start, u64 num_bytes, u64 min_size,
+			      loff_t actual_len, u64 *alloc_hint)
 {
 	struct btrfs_trans_handle *trans;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_key ins;
 	u64 cur_offset = start;
-	u64 num_bytes = end - start;
 	int ret = 0;
-	u64 i_size;
 
 	while (num_bytes > 0) {
-		trans = btrfs_start_transaction(root, 1);
-
-		ret = btrfs_reserve_extent(trans, root, num_bytes,
-					   root->sectorsize, 0, alloc_hint,
-					   (u64)-1, &ins, 1);
-		if (ret) {
-			WARN_ON(1);
-			goto stop_trans;
+		trans = btrfs_start_transaction(root, 3);
+		if (IS_ERR(trans)) {
+			ret = PTR_ERR(trans);
+			break;
 		}
 
-		ret = btrfs_reserve_metadata_space(root, 3);
+		ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
+					   0, *alloc_hint, (u64)-1, &ins, 1);
 		if (ret) {
-			btrfs_free_reserved_extent(root, ins.objectid,
-						   ins.offset);
-			goto stop_trans;
+			btrfs_end_transaction(trans, root);
+			break;
 		}
 
 		ret = insert_reserved_file_extent(trans, inode,
@@ -5766,34 +6786,27 @@ static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
 
 		num_bytes -= ins.offset;
 		cur_offset += ins.offset;
-		alloc_hint = ins.objectid + ins.offset;
+		*alloc_hint = ins.objectid + ins.offset;
 
 		inode->i_ctime = CURRENT_TIME;
 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
-			(actual_len > inode->i_size) &&
-			(cur_offset > inode->i_size)) {
-
+		    (actual_len > inode->i_size) &&
+		    (cur_offset > inode->i_size)) {
 			if (cur_offset > actual_len)
-				i_size  = actual_len;
+				i_size_write(inode, actual_len);
 			else
-				i_size = cur_offset;
-			i_size_write(inode, i_size);
-			btrfs_ordered_update_i_size(inode, i_size, NULL);
+				i_size_write(inode, cur_offset);
+			i_size_write(inode, cur_offset);
+			btrfs_ordered_update_i_size(inode, cur_offset, NULL);
 		}
 
 		ret = btrfs_update_inode(trans, root, inode);
 		BUG_ON(ret);
 
 		btrfs_end_transaction(trans, root);
-		btrfs_unreserve_metadata_space(root, 3);
 	}
 	return ret;
-
-stop_trans:
-	btrfs_end_transaction(trans, root);
-	return ret;
-
 }
 
 static long btrfs_fallocate(struct inode *inode, int mode,
@@ -5826,8 +6839,7 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 			goto out;
 	}
 
-	ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode,
-					  alloc_end - alloc_start);
+	ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
 	if (ret)
 		goto out;
 
@@ -5872,16 +6884,16 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 		if (em->block_start == EXTENT_MAP_HOLE ||
 		    (cur_offset >= inode->i_size &&
 		     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
-			ret = prealloc_file_range(inode,
-						  cur_offset, last_byte,
-						alloc_hint, mode, offset+len);
+			ret = btrfs_prealloc_file_range(inode, 0, cur_offset,
+							last_byte - cur_offset,
+							1 << inode->i_blkbits,
+							offset + len,
+							&alloc_hint);
 			if (ret < 0) {
 				free_extent_map(em);
 				break;
 			}
 		}
-		if (em->block_start <= EXTENT_MAP_LAST_BYTE)
-			alloc_hint = em->block_start;
 		free_extent_map(em);
 
 		cur_offset = last_byte;
@@ -5893,8 +6905,7 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
 			     &cached_state, GFP_NOFS);
 
-	btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode,
-				       alloc_end - alloc_start);
+	btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
 out:
 	mutex_unlock(&inode->i_mutex);
 	return ret;

fs/btrfs/ioctl.c

@@ -239,23 +239,19 @@ static noinline int create_subvol(struct btrfs_root *root,
 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 	u64 index = 0;
 
+	ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root,
+				       0, &objectid);
+	if (ret)
+		return ret;
 	/*
 	 * 1 - inode item
 	 * 2 - refs
 	 * 1 - root item
 	 * 2 - dir items
 	 */
-	ret = btrfs_reserve_metadata_space(root, 6);
-	if (ret)
-		return ret;
-
-	trans = btrfs_start_transaction(root, 1);
-	BUG_ON(!trans);
-
-	ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
-				       0, &objectid);
-	if (ret)
-		goto fail;
+	trans = btrfs_start_transaction(root, 6);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 				      0, objectid, NULL, 0, 0, 0);
@@ -345,13 +341,10 @@ fail:
 	err = btrfs_commit_transaction(trans, root);
 	if (err && !ret)
 		ret = err;
-
-	btrfs_unreserve_metadata_space(root, 6);
 	return ret;
 }
 
-static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
-			   char *name, int namelen)
+static int create_snapshot(struct btrfs_root *root, struct dentry *dentry)
 {
 	struct inode *inode;
 	struct btrfs_pending_snapshot *pending_snapshot;
@@ -361,40 +354,33 @@ static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 	if (!root->ref_cows)
 		return -EINVAL;
 
-	/*
-	 * 1 - inode item
-	 * 2 - refs
-	 * 1 - root item
-	 * 2 - dir items
-	 */
-	ret = btrfs_reserve_metadata_space(root, 6);
-	if (ret)
-		goto fail;
-
 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
-	if (!pending_snapshot) {
-		ret = -ENOMEM;
-		btrfs_unreserve_metadata_space(root, 6);
-		goto fail;
-	}
-	pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
-	if (!pending_snapshot->name) {
-		ret = -ENOMEM;
-		kfree(pending_snapshot);
-		btrfs_unreserve_metadata_space(root, 6);
-		goto fail;
-	}
-	memcpy(pending_snapshot->name, name, namelen);
-	pending_snapshot->name[namelen] = '\0';
+	if (!pending_snapshot)
+		return -ENOMEM;
+
+	btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 	pending_snapshot->dentry = dentry;
-	trans = btrfs_start_transaction(root, 1);
-	BUG_ON(!trans);
 	pending_snapshot->root = root;
+
+	trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
+	if (IS_ERR(trans)) {
+		ret = PTR_ERR(trans);
+		goto fail;
+	}
+
+	ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
+	BUG_ON(ret);
+
 	list_add(&pending_snapshot->list,
 		 &trans->transaction->pending_snapshots);
-	ret = btrfs_commit_transaction(trans, root);
+	ret = btrfs_commit_transaction(trans, root->fs_info->extent_root);
 	BUG_ON(ret);
-	btrfs_unreserve_metadata_space(root, 6);
+
+	ret = pending_snapshot->error;
+	if (ret)
+		goto fail;
+
+	btrfs_orphan_cleanup(pending_snapshot->snap);
 
 	inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 	if (IS_ERR(inode)) {
@@ -405,6 +391,7 @@ static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 	d_instantiate(dentry, inode);
 	ret = 0;
 fail:
+	kfree(pending_snapshot);
 	return ret;
 }
 
@@ -456,8 +443,7 @@ static noinline int btrfs_mksubvol(struct path *parent,
 		goto out_up_read;
 
 	if (snap_src) {
-		error = create_snapshot(snap_src, dentry,
-					name, namelen);
+		error = create_snapshot(snap_src, dentry);
 	} else {
 		error = create_subvol(BTRFS_I(dir)->root, dentry,
 				      name, namelen);
@@ -601,19 +587,9 @@ static int btrfs_defrag_file(struct file *file,
 		if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
 			BTRFS_I(inode)->force_compress = 1;
 
-		ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
-		if (ret) {
-			ret = -ENOSPC;
-			break;
-		}
-
-		ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
-		if (ret) {
-			btrfs_free_reserved_data_space(root, inode,
-						       PAGE_CACHE_SIZE);
-			ret = -ENOSPC;
-			break;
-		}
+		ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
+		if (ret)
+			goto err_unlock;
 again:
 		if (inode->i_size == 0 ||
 		    i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) {
@@ -622,8 +598,10 @@ again:
 		}
 
 		page = grab_cache_page(inode->i_mapping, i);
-		if (!page)
+		if (!page) {
+			ret = -ENOMEM;
 			goto err_reservations;
+		}
 
 		if (!PageUptodate(page)) {
 			btrfs_readpage(NULL, page);
@@ -631,6 +609,7 @@ again:
 			if (!PageUptodate(page)) {
 				unlock_page(page);
 				page_cache_release(page);
+				ret = -EIO;
 				goto err_reservations;
 			}
 		}
@@ -644,8 +623,7 @@ again:
 		wait_on_page_writeback(page);
 
 		if (PageDirty(page)) {
-			btrfs_free_reserved_data_space(root, inode,
-						       PAGE_CACHE_SIZE);
+			btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 			goto loop_unlock;
 		}
 
@@ -683,7 +661,6 @@ loop_unlock:
 		page_cache_release(page);
 		mutex_unlock(&inode->i_mutex);
 
-		btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
 		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 		i++;
 	}
@@ -713,9 +690,9 @@ loop_unlock:
 	return 0;
 
 err_reservations:
+	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
+err_unlock:
 	mutex_unlock(&inode->i_mutex);
-	btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
-	btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
 	return ret;
 }
 
@@ -811,7 +788,7 @@ static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
 		device->name, (unsigned long long)new_size);
 
 	if (new_size > old_size) {
-		trans = btrfs_start_transaction(root, 1);
+		trans = btrfs_start_transaction(root, 0);
 		ret = btrfs_grow_device(trans, device, new_size);
 		btrfs_commit_transaction(trans, root);
 	} else {
@@ -1300,7 +1277,13 @@ static noinline int btrfs_ioctl_snap_destroy(struct file *file,
 	if (err)
 		goto out_up_write;
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_start_transaction(root, 0);
+	if (IS_ERR(trans)) {
+		err = PTR_ERR(trans);
+		goto out;
+	}
+	trans->block_rsv = &root->fs_info->global_block_rsv;
+
 	ret = btrfs_unlink_subvol(trans, root, dir,
 				dest->root_key.objectid,
 				dentry->d_name.name,
@@ -1314,10 +1297,12 @@ static noinline int btrfs_ioctl_snap_destroy(struct file *file,
 	dest->root_item.drop_level = 0;
 	btrfs_set_root_refs(&dest->root_item, 0);
 
-	ret = btrfs_insert_orphan_item(trans,
-				root->fs_info->tree_root,
-				dest->root_key.objectid);
-	BUG_ON(ret);
+	if (!xchg(&dest->orphan_item_inserted, 1)) {
+		ret = btrfs_insert_orphan_item(trans,
+					root->fs_info->tree_root,
+					dest->root_key.objectid);
+		BUG_ON(ret);
+	}
 
 	ret = btrfs_commit_transaction(trans, root);
 	BUG_ON(ret);
@@ -1358,8 +1343,10 @@ static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
 			ret = -EPERM;
 			goto out;
 		}
-		btrfs_defrag_root(root, 0);
-		btrfs_defrag_root(root->fs_info->extent_root, 0);
+		ret = btrfs_defrag_root(root, 0);
+		if (ret)
+			goto out;
+		ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
 		break;
 	case S_IFREG:
 		if (!(file->f_mode & FMODE_WRITE)) {
@@ -1389,9 +1376,11 @@ static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
 			/* the rest are all set to zero by kzalloc */
 			range->len = (u64)-1;
 		}
-		btrfs_defrag_file(file, range);
+		ret = btrfs_defrag_file(file, range);
 		kfree(range);
 		break;
+	default:
+		ret = -EINVAL;
 	}
 out:
 	mnt_drop_write(file->f_path.mnt);
@@ -1550,12 +1539,6 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 		btrfs_wait_ordered_range(src, off, off+len);
 	}
 
-	trans = btrfs_start_transaction(root, 1);
-	BUG_ON(!trans);
-
-	/* punch hole in destination first */
-	btrfs_drop_extents(trans, inode, off, off + len, &hint_byte, 1);
-
 	/* clone data */
 	key.objectid = src->i_ino;
 	key.type = BTRFS_EXTENT_DATA_KEY;
@@ -1566,7 +1549,7 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 		 * note the key will change type as we walk through the
 		 * tree.
 		 */
-		ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
+		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 		if (ret < 0)
 			goto out;
 
@@ -1629,12 +1612,31 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 			new_key.objectid = inode->i_ino;
 			new_key.offset = key.offset + destoff - off;
 
+			trans = btrfs_start_transaction(root, 1);
+			if (IS_ERR(trans)) {
+				ret = PTR_ERR(trans);
+				goto out;
+			}
+
 			if (type == BTRFS_FILE_EXTENT_REG ||
 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
+				if (off > key.offset) {
+					datao += off - key.offset;
+					datal -= off - key.offset;
+				}
+
+				if (key.offset + datal > off + len)
+					datal = off + len - key.offset;
+
+				ret = btrfs_drop_extents(trans, inode,
+							 new_key.offset,
+							 new_key.offset + datal,
+							 &hint_byte, 1);
+				BUG_ON(ret);
+
 				ret = btrfs_insert_empty_item(trans, root, path,
 							      &new_key, size);
-				if (ret)
-					goto out;
+				BUG_ON(ret);
 
 				leaf = path->nodes[0];
 				slot = path->slots[0];
@@ -1645,14 +1647,6 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 				extent = btrfs_item_ptr(leaf, slot,
 						struct btrfs_file_extent_item);
 
-				if (off > key.offset) {
-					datao += off - key.offset;
-					datal -= off - key.offset;
-				}
-
-				if (key.offset + datal > off + len)
-					datal = off + len - key.offset;
-
 				/* disko == 0 means it's a hole */
 				if (!disko)
 					datao = 0;
@@ -1683,14 +1677,21 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 
 				if (comp && (skip || trim)) {
 					ret = -EINVAL;
+					btrfs_end_transaction(trans, root);
 					goto out;
 				}
 				size -= skip + trim;
 				datal -= skip + trim;
+
+				ret = btrfs_drop_extents(trans, inode,
+							 new_key.offset,
+							 new_key.offset + datal,
+							 &hint_byte, 1);
+				BUG_ON(ret);
+
 				ret = btrfs_insert_empty_item(trans, root, path,
 							      &new_key, size);
-				if (ret)
-					goto out;
+				BUG_ON(ret);
 
 				if (skip) {
 					u32 start =
@@ -1708,8 +1709,17 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 			}
 
 			btrfs_mark_buffer_dirty(leaf);
-		}
+			btrfs_release_path(root, path);
 
+			inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+			if (new_key.offset + datal > inode->i_size)
+				btrfs_i_size_write(inode,
+						   new_key.offset + datal);
+			BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
+			ret = btrfs_update_inode(trans, root, inode);
+			BUG_ON(ret);
+			btrfs_end_transaction(trans, root);
+		}
 next:
 		btrfs_release_path(root, path);
 		key.offset++;
@@ -1717,17 +1727,7 @@ next:
 	ret = 0;
 out:
 	btrfs_release_path(root, path);
-	if (ret == 0) {
-		inode->i_mtime = inode->i_ctime = CURRENT_TIME;
-		if (destoff + olen > inode->i_size)
-			btrfs_i_size_write(inode, destoff + olen);
-		BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
-		ret = btrfs_update_inode(trans, root, inode);
-	}
-	btrfs_end_transaction(trans, root);
 	unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
-	if (ret)
-		vmtruncate(inode, 0);
 out_unlock:
 	mutex_unlock(&src->i_mutex);
 	mutex_unlock(&inode->i_mutex);

fs/btrfs/ordered-data.c

@@ -124,6 +124,15 @@ static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
 	return 1;
 }
 
+static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
+			  u64 len)
+{
+	if (file_offset + len <= entry->file_offset ||
+	    entry->file_offset + entry->len <= file_offset)
+		return 0;
+	return 1;
+}
+
 /*
  * look find the first ordered struct that has this offset, otherwise
  * the first one less than this offset
@@ -161,8 +170,9 @@ static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
  * The tree is given a single reference on the ordered extent that was
  * inserted.
  */
-int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
-			     u64 start, u64 len, u64 disk_len, int type)
+static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
+				      u64 start, u64 len, u64 disk_len,
+				      int type, int dio)
 {
 	struct btrfs_ordered_inode_tree *tree;
 	struct rb_node *node;
@@ -182,6 +192,9 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 		set_bit(type, &entry->flags);
 
+	if (dio)
+		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
+
 	/* one ref for the tree */
 	atomic_set(&entry->refs, 1);
 	init_waitqueue_head(&entry->wait);
@@ -203,6 +216,20 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 	return 0;
 }
 
+int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
+			     u64 start, u64 len, u64 disk_len, int type)
+{
+	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
+					  disk_len, type, 0);
+}
+
+int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
+				 u64 start, u64 len, u64 disk_len, int type)
+{
+	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
+					  disk_len, type, 1);
+}
+
 /*
  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
  * when an ordered extent is finished.  If the list covers more than one
@@ -311,13 +338,6 @@ static int __btrfs_remove_ordered_extent(struct inode *inode,
 	tree->last = NULL;
 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 
-	spin_lock(&BTRFS_I(inode)->accounting_lock);
-	WARN_ON(!BTRFS_I(inode)->outstanding_extents);
-	BTRFS_I(inode)->outstanding_extents--;
-	spin_unlock(&BTRFS_I(inode)->accounting_lock);
-	btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
-					      inode, 1);
-
 	spin_lock(&root->fs_info->ordered_extent_lock);
 	list_del_init(&entry->root_extent_list);
 
@@ -491,7 +511,8 @@ void btrfs_start_ordered_extent(struct inode *inode,
 	 * start IO on any dirty ones so the wait doesn't stall waiting
 	 * for pdflush to find them
 	 */
-	filemap_fdatawrite_range(inode->i_mapping, start, end);
+	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
+		filemap_fdatawrite_range(inode->i_mapping, start, end);
 	if (wait) {
 		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 						 &entry->flags));
@@ -588,6 +609,47 @@ out:
 	return entry;
 }
 
+/* Since the DIO code tries to lock a wide area we need to look for any ordered
+ * extents that exist in the range, rather than just the start of the range.
+ */
+struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
+							u64 file_offset,
+							u64 len)
+{
+	struct btrfs_ordered_inode_tree *tree;
+	struct rb_node *node;
+	struct btrfs_ordered_extent *entry = NULL;
+
+	tree = &BTRFS_I(inode)->ordered_tree;
+	spin_lock(&tree->lock);
+	node = tree_search(tree, file_offset);
+	if (!node) {
+		node = tree_search(tree, file_offset + len);
+		if (!node)
+			goto out;
+	}
+
+	while (1) {
+		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+		if (range_overlaps(entry, file_offset, len))
+			break;
+
+		if (entry->file_offset >= file_offset + len) {
+			entry = NULL;
+			break;
+		}
+		entry = NULL;
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	if (entry)
+		atomic_inc(&entry->refs);
+	spin_unlock(&tree->lock);
+	return entry;
+}
+
 /*
  * lookup and return any extent before 'file_offset'.  NULL is returned
  * if none is found

fs/btrfs/ordered-data.h

@@ -72,6 +72,8 @@ struct btrfs_ordered_sum {
 
 #define BTRFS_ORDERED_PREALLOC 4 /* set when writing to prealloced extent */
 
+#define BTRFS_ORDERED_DIRECT 5 /* set when we're doing DIO with this extent */
+
 struct btrfs_ordered_extent {
 	/* logical offset in the file */
 	u64 file_offset;
@@ -140,7 +142,9 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
 				   struct btrfs_ordered_extent **cached,
 				   u64 file_offset, u64 io_size);
 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
-			     u64 start, u64 len, u64 disk_len, int tyep);
+			     u64 start, u64 len, u64 disk_len, int type);
+int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
+				 u64 start, u64 len, u64 disk_len, int type);
 int btrfs_add_ordered_sum(struct inode *inode,
 			  struct btrfs_ordered_extent *entry,
 			  struct btrfs_ordered_sum *sum);
@@ -151,6 +155,9 @@ void btrfs_start_ordered_extent(struct inode *inode,
 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len);
 struct btrfs_ordered_extent *
 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset);
+struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
+							u64 file_offset,
+							u64 len);
 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 				struct btrfs_ordered_extent *ordered);
 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, u32 *sum);

fs/btrfs/relocation.c

@@ -44,8 +44,12 @@ struct tree_entry {
 struct backref_node {
 	struct rb_node rb_node;
 	u64 bytenr;
-	/* objectid tree block owner */
+
+	u64 new_bytenr;
+	/* objectid of tree block owner, can be not uptodate */
 	u64 owner;
+	/* link to pending, changed or detached list */
+	struct list_head list;
 	/* list of upper level blocks reference this block */
 	struct list_head upper;
 	/* list of child blocks in the cache */
@@ -56,9 +60,9 @@ struct backref_node {
 	struct extent_buffer *eb;
 	/* level of tree block */
 	unsigned int level:8;
-	/* 1 if the block is root of old snapshot */
-	unsigned int old_root:1;
-	/* 1 if no child blocks in the cache */
+	/* is the block in non-reference counted tree */
+	unsigned int cowonly:1;
+	/* 1 if no child node in the cache */
 	unsigned int lowest:1;
 	/* is the extent buffer locked */
 	unsigned int locked:1;
@@ -66,6 +70,16 @@ struct backref_node {
 	unsigned int processed:1;
 	/* have backrefs of this block been checked */
 	unsigned int checked:1;
+	/*
+	 * 1 if corresponding block has been cowed but some upper
+	 * level block pointers may not point to the new location
+	 */
+	unsigned int pending:1;
+	/*
+	 * 1 if the backref node isn't connected to any other
+	 * backref node.
+	 */
+	unsigned int detached:1;
 };
 
 /*
@@ -74,7 +88,6 @@ struct backref_node {
 struct backref_edge {
 	struct list_head list[2];
 	struct backref_node *node[2];
-	u64 blockptr;
 };
 
 #define LOWER	0
@@ -83,9 +96,25 @@ struct backref_edge {
 struct backref_cache {
 	/* red black tree of all backref nodes in the cache */
 	struct rb_root rb_root;
-	/* list of backref nodes with no child block in the cache */
+	/* for passing backref nodes to btrfs_reloc_cow_block */
+	struct backref_node *path[BTRFS_MAX_LEVEL];
+	/*
+	 * list of blocks that have been cowed but some block
+	 * pointers in upper level blocks may not reflect the
+	 * new location
+	 */
 	struct list_head pending[BTRFS_MAX_LEVEL];
-	spinlock_t lock;
+	/* list of backref nodes with no child node */
+	struct list_head leaves;
+	/* list of blocks that have been cowed in current transaction */
+	struct list_head changed;
+	/* list of detached backref node. */
+	struct list_head detached;
+
+	u64 last_trans;
+
+	int nr_nodes;
+	int nr_edges;
 };
 
 /*
@@ -113,15 +142,6 @@ struct tree_block {
 	unsigned int key_ready:1;
 };
 
-/* inode vector */
-#define INODEVEC_SIZE 16
-
-struct inodevec {
-	struct list_head list;
-	struct inode *inode[INODEVEC_SIZE];
-	int nr;
-};
-
 #define MAX_EXTENTS 128
 
 struct file_extent_cluster {
@@ -138,36 +158,43 @@ struct reloc_control {
 	struct btrfs_root *extent_root;
 	/* inode for moving data */
 	struct inode *data_inode;
-	struct btrfs_workers workers;
+
+	struct btrfs_block_rsv *block_rsv;
+
+	struct backref_cache backref_cache;
+
+	struct file_extent_cluster cluster;
 	/* tree blocks have been processed */
 	struct extent_io_tree processed_blocks;
 	/* map start of tree root to corresponding reloc tree */
 	struct mapping_tree reloc_root_tree;
 	/* list of reloc trees */
 	struct list_head reloc_roots;
+	/* size of metadata reservation for merging reloc trees */
+	u64 merging_rsv_size;
+	/* size of relocated tree nodes */
+	u64 nodes_relocated;
+
 	u64 search_start;
 	u64 extents_found;
-	u64 extents_skipped;
-	int stage;
-	int create_reloc_root;
+
+	int block_rsv_retries;
+
+	unsigned int stage:8;
+	unsigned int create_reloc_tree:1;
+	unsigned int merge_reloc_tree:1;
 	unsigned int found_file_extent:1;
-	unsigned int found_old_snapshot:1;
+	unsigned int commit_transaction:1;
 };
 
 /* stages of data relocation */
 #define MOVE_DATA_EXTENTS	0
 #define UPDATE_DATA_PTRS	1
 
-/*
- * merge reloc tree to corresponding fs tree in worker threads
- */
-struct async_merge {
-	struct btrfs_work work;
-	struct reloc_control *rc;
-	struct btrfs_root *root;
-	struct completion *done;
-	atomic_t *num_pending;
-};
+static void remove_backref_node(struct backref_cache *cache,
+				struct backref_node *node);
+static void __mark_block_processed(struct reloc_control *rc,
+				   struct backref_node *node);
 
 static void mapping_tree_init(struct mapping_tree *tree)
 {
@@ -181,15 +208,80 @@ static void backref_cache_init(struct backref_cache *cache)
 	cache->rb_root = RB_ROOT;
 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 		INIT_LIST_HEAD(&cache->pending[i]);
-	spin_lock_init(&cache->lock);
+	INIT_LIST_HEAD(&cache->changed);
+	INIT_LIST_HEAD(&cache->detached);
+	INIT_LIST_HEAD(&cache->leaves);
+}
+
+static void backref_cache_cleanup(struct backref_cache *cache)
+{
+	struct backref_node *node;
+	int i;
+
+	while (!list_empty(&cache->detached)) {
+		node = list_entry(cache->detached.next,
+				  struct backref_node, list);
+		remove_backref_node(cache, node);
+	}
+
+	while (!list_empty(&cache->leaves)) {
+		node = list_entry(cache->leaves.next,
+				  struct backref_node, lower);
+		remove_backref_node(cache, node);
+	}
+
+	cache->last_trans = 0;
+
+	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
+		BUG_ON(!list_empty(&cache->pending[i]));
+	BUG_ON(!list_empty(&cache->changed));
+	BUG_ON(!list_empty(&cache->detached));
+	BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
+	BUG_ON(cache->nr_nodes);
+	BUG_ON(cache->nr_edges);
+}
+
+static struct backref_node *alloc_backref_node(struct backref_cache *cache)
+{
+	struct backref_node *node;
+
+	node = kzalloc(sizeof(*node), GFP_NOFS);
+	if (node) {
+		INIT_LIST_HEAD(&node->list);
+		INIT_LIST_HEAD(&node->upper);
+		INIT_LIST_HEAD(&node->lower);
+		RB_CLEAR_NODE(&node->rb_node);
+		cache->nr_nodes++;
+	}
+	return node;
+}
+
+static void free_backref_node(struct backref_cache *cache,
+			      struct backref_node *node)
+{
+	if (node) {
+		cache->nr_nodes--;
+		kfree(node);
+	}
+}
+
+static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
+{
+	struct backref_edge *edge;
+
+	edge = kzalloc(sizeof(*edge), GFP_NOFS);
+	if (edge)
+		cache->nr_edges++;
+	return edge;
 }
 
-static void backref_node_init(struct backref_node *node)
+static void free_backref_edge(struct backref_cache *cache,
+			      struct backref_edge *edge)
 {
-	memset(node, 0, sizeof(*node));
-	INIT_LIST_HEAD(&node->upper);
-	INIT_LIST_HEAD(&node->lower);
-	RB_CLEAR_NODE(&node->rb_node);
+	if (edge) {
+		cache->nr_edges--;
+		kfree(edge);
+	}
 }
 
 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
@@ -250,6 +342,7 @@ static struct backref_node *walk_up_backref(struct backref_node *node,
 		edges[idx++] = edge;
 		node = edge->node[UPPER];
 	}
+	BUG_ON(node->detached);
 	*index = idx;
 	return node;
 }
@@ -281,13 +374,18 @@ static struct backref_node *walk_down_backref(struct backref_edge *edges[],
 	return NULL;
 }
 
+static void unlock_node_buffer(struct backref_node *node)
+{
+	if (node->locked) {
+		btrfs_tree_unlock(node->eb);
+		node->locked = 0;
+	}
+}
+
 static void drop_node_buffer(struct backref_node *node)
 {
 	if (node->eb) {
-		if (node->locked) {
-			btrfs_tree_unlock(node->eb);
-			node->locked = 0;
-		}
+		unlock_node_buffer(node);
 		free_extent_buffer(node->eb);
 		node->eb = NULL;
 	}
@@ -296,14 +394,14 @@ static void drop_node_buffer(struct backref_node *node)
 static void drop_backref_node(struct backref_cache *tree,
 			      struct backref_node *node)
 {
-	BUG_ON(!node->lowest);
 	BUG_ON(!list_empty(&node->upper));
 
 	drop_node_buffer(node);
+	list_del(&node->list);
 	list_del(&node->lower);
-
-	rb_erase(&node->rb_node, &tree->rb_root);
-	kfree(node);
+	if (!RB_EMPTY_NODE(&node->rb_node))
+		rb_erase(&node->rb_node, &tree->rb_root);
+	free_backref_node(tree, node);
 }
 
 /*
@@ -318,27 +416,121 @@ static void remove_backref_node(struct backref_cache *cache,
 	if (!node)
 		return;
 
-	BUG_ON(!node->lowest);
+	BUG_ON(!node->lowest && !node->detached);
 	while (!list_empty(&node->upper)) {
 		edge = list_entry(node->upper.next, struct backref_edge,
 				  list[LOWER]);
 		upper = edge->node[UPPER];
 		list_del(&edge->list[LOWER]);
 		list_del(&edge->list[UPPER]);
-		kfree(edge);
+		free_backref_edge(cache, edge);
+
+		if (RB_EMPTY_NODE(&upper->rb_node)) {
+			BUG_ON(!list_empty(&node->upper));
+			drop_backref_node(cache, node);
+			node = upper;
+			node->lowest = 1;
+			continue;
+		}
 		/*
-		 * add the node to pending list if no other
+		 * add the node to leaf node list if no other
 		 * child block cached.
 		 */
 		if (list_empty(&upper->lower)) {
-			list_add_tail(&upper->lower,
-				      &cache->pending[upper->level]);
+			list_add_tail(&upper->lower, &cache->leaves);
 			upper->lowest = 1;
 		}
 	}
+
 	drop_backref_node(cache, node);
 }
 
+static void update_backref_node(struct backref_cache *cache,
+				struct backref_node *node, u64 bytenr)
+{
+	struct rb_node *rb_node;
+	rb_erase(&node->rb_node, &cache->rb_root);
+	node->bytenr = bytenr;
+	rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
+	BUG_ON(rb_node);
+}
+
+/*
+ * update backref cache after a transaction commit
+ */
+static int update_backref_cache(struct btrfs_trans_handle *trans,
+				struct backref_cache *cache)
+{
+	struct backref_node *node;
+	int level = 0;
+
+	if (cache->last_trans == 0) {
+		cache->last_trans = trans->transid;
+		return 0;
+	}
+
+	if (cache->last_trans == trans->transid)
+		return 0;
+
+	/*
+	 * detached nodes are used to avoid unnecessary backref
+	 * lookup. transaction commit changes the extent tree.
+	 * so the detached nodes are no longer useful.
+	 */
+	while (!list_empty(&cache->detached)) {
+		node = list_entry(cache->detached.next,
+				  struct backref_node, list);
+		remove_backref_node(cache, node);
+	}
+
+	while (!list_empty(&cache->changed)) {
+		node = list_entry(cache->changed.next,
+				  struct backref_node, list);
+		list_del_init(&node->list);
+		BUG_ON(node->pending);
+		update_backref_node(cache, node, node->new_bytenr);
+	}
+
+	/*
+	 * some nodes can be left in the pending list if there were
+	 * errors during processing the pending nodes.
+	 */
+	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
+		list_for_each_entry(node, &cache->pending[level], list) {
+			BUG_ON(!node->pending);
+			if (node->bytenr == node->new_bytenr)
+				continue;
+			update_backref_node(cache, node, node->new_bytenr);
+		}
+	}
+
+	cache->last_trans = 0;
+	return 1;
+}
+
+static int should_ignore_root(struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+
+	if (!root->ref_cows)
+		return 0;
+
+	reloc_root = root->reloc_root;
+	if (!reloc_root)
+		return 0;
+
+	if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
+	    root->fs_info->running_transaction->transid - 1)
+		return 0;
+	/*
+	 * if there is reloc tree and it was created in previous
+	 * transaction backref lookup can find the reloc tree,
+	 * so backref node for the fs tree root is useless for
+	 * relocation.
+	 */
+	return 1;
+}
+
 /*
  * find reloc tree by address of tree root
  */
@@ -453,11 +645,12 @@ int find_inline_backref(struct extent_buffer *leaf, int slot,
  * for all upper level blocks that directly/indirectly reference the
  * block are also cached.
  */
-static struct backref_node *build_backref_tree(struct reloc_control *rc,
-					       struct backref_cache *cache,
-					       struct btrfs_key *node_key,
-					       int level, u64 bytenr)
+static noinline_for_stack
+struct backref_node *build_backref_tree(struct reloc_control *rc,
+					struct btrfs_key *node_key,
+					int level, u64 bytenr)
 {
+	struct backref_cache *cache = &rc->backref_cache;
 	struct btrfs_path *path1;
 	struct btrfs_path *path2;
 	struct extent_buffer *eb;
@@ -473,6 +666,8 @@ static struct backref_node *build_backref_tree(struct reloc_control *rc,
 	unsigned long end;
 	unsigned long ptr;
 	LIST_HEAD(list);
+	LIST_HEAD(useless);
+	int cowonly;
 	int ret;
 	int err = 0;
 
@@ -483,15 +678,13 @@ static struct backref_node *build_backref_tree(struct reloc_control *rc,
 		goto out;
 	}
 
-	node = kmalloc(sizeof(*node), GFP_NOFS);
+	node = alloc_backref_node(cache);
 	if (!node) {
 		err = -ENOMEM;
 		goto out;
 	}
 
-	backref_node_init(node);
 	node->bytenr = bytenr;
-	node->owner = 0;
 	node->level = level;
 	node->lowest = 1;
 	cur = node;
@@ -587,17 +780,20 @@ again:
 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
 		    key.type == BTRFS_EXTENT_REF_V0_KEY) {
-			if (key.objectid == key.offset &&
-			    key.type == BTRFS_EXTENT_REF_V0_KEY) {
+			if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 				struct btrfs_extent_ref_v0 *ref0;
 				ref0 = btrfs_item_ptr(eb, path1->slots[0],
 						struct btrfs_extent_ref_v0);
 				root = find_tree_root(rc, eb, ref0);
-				if (root)
-					cur->root = root;
-				else
-					cur->old_root = 1;
-				break;
+				if (!root->ref_cows)
+					cur->cowonly = 1;
+				if (key.objectid == key.offset) {
+					if (root && !should_ignore_root(root))
+						cur->root = root;
+					else
+						list_add(&cur->list, &useless);
+					break;
+				}
 			}
 #else
 		BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
@@ -614,22 +810,20 @@ again:
 				break;
 			}
 
-			edge = kzalloc(sizeof(*edge), GFP_NOFS);
+			edge = alloc_backref_edge(cache);
 			if (!edge) {
 				err = -ENOMEM;
 				goto out;
 			}
 			rb_node = tree_search(&cache->rb_root, key.offset);
 			if (!rb_node) {
-				upper = kmalloc(sizeof(*upper), GFP_NOFS);
+				upper = alloc_backref_node(cache);
 				if (!upper) {
-					kfree(edge);
+					free_backref_edge(cache, edge);
 					err = -ENOMEM;
 					goto out;
 				}
-				backref_node_init(upper);
 				upper->bytenr = key.offset;
-				upper->owner = 0;
 				upper->level = cur->level + 1;
 				/*
 				 *  backrefs for the upper level block isn't
@@ -639,11 +833,12 @@ again:
 			} else {
 				upper = rb_entry(rb_node, struct backref_node,
 						 rb_node);
+				BUG_ON(!upper->checked);
 				INIT_LIST_HEAD(&edge->list[UPPER]);
 			}
-			list_add(&edge->list[LOWER], &cur->upper);
-			edge->node[UPPER] = upper;
+			list_add_tail(&edge->list[LOWER], &cur->upper);
 			edge->node[LOWER] = cur;
+			edge->node[UPPER] = upper;
 
 			goto next;
 		} else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
@@ -657,11 +852,17 @@ again:
 			goto out;
 		}
 
+		if (!root->ref_cows)
+			cur->cowonly = 1;
+
 		if (btrfs_root_level(&root->root_item) == cur->level) {
 			/* tree root */
 			BUG_ON(btrfs_root_bytenr(&root->root_item) !=
 			       cur->bytenr);
-			cur->root = root;
+			if (should_ignore_root(root))
+				list_add(&cur->list, &useless);
+			else
+				cur->root = root;
 			break;
 		}
 
@@ -692,11 +893,14 @@ again:
 			if (!path2->nodes[level]) {
 				BUG_ON(btrfs_root_bytenr(&root->root_item) !=
 				       lower->bytenr);
-				lower->root = root;
+				if (should_ignore_root(root))
+					list_add(&lower->list, &useless);
+				else
+					lower->root = root;
 				break;
 			}
 
-			edge = kzalloc(sizeof(*edge), GFP_NOFS);
+			edge = alloc_backref_edge(cache);
 			if (!edge) {
 				err = -ENOMEM;
 				goto out;
@@ -705,16 +909,17 @@ again:
 			eb = path2->nodes[level];
 			rb_node = tree_search(&cache->rb_root, eb->start);
 			if (!rb_node) {
-				upper = kmalloc(sizeof(*upper), GFP_NOFS);
+				upper = alloc_backref_node(cache);
 				if (!upper) {
-					kfree(edge);
+					free_backref_edge(cache, edge);
 					err = -ENOMEM;
 					goto out;
 				}
-				backref_node_init(upper);
 				upper->bytenr = eb->start;
 				upper->owner = btrfs_header_owner(eb);
 				upper->level = lower->level + 1;
+				if (!root->ref_cows)
+					upper->cowonly = 1;
 
 				/*
 				 * if we know the block isn't shared
@@ -744,10 +949,12 @@ again:
 						 rb_node);
 				BUG_ON(!upper->checked);
 				INIT_LIST_HEAD(&edge->list[UPPER]);
+				if (!upper->owner)
+					upper->owner = btrfs_header_owner(eb);
 			}
 			list_add_tail(&edge->list[LOWER], &lower->upper);
-			edge->node[UPPER] = upper;
 			edge->node[LOWER] = lower;
+			edge->node[UPPER] = upper;
 
 			if (rb_node)
 				break;
@@ -785,8 +992,13 @@ next:
 	 * into the cache.
 	 */
 	BUG_ON(!node->checked);
-	rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
-	BUG_ON(rb_node);
+	cowonly = node->cowonly;
+	if (!cowonly) {
+		rb_node = tree_insert(&cache->rb_root, node->bytenr,
+				      &node->rb_node);
+		BUG_ON(rb_node);
+		list_add_tail(&node->lower, &cache->leaves);
+	}
 
 	list_for_each_entry(edge, &node->upper, list[LOWER])
 		list_add_tail(&edge->list[UPPER], &list);
@@ -795,6 +1007,14 @@ next:
 		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
 		list_del_init(&edge->list[UPPER]);
 		upper = edge->node[UPPER];
+		if (upper->detached) {
+			list_del(&edge->list[LOWER]);
+			lower = edge->node[LOWER];
+			free_backref_edge(cache, edge);
+			if (list_empty(&lower->upper))
+				list_add(&lower->list, &useless);
+			continue;
+		}
 
 		if (!RB_EMPTY_NODE(&upper->rb_node)) {
 			if (upper->lowest) {
@@ -807,25 +1027,69 @@ next:
 		}
 
 		BUG_ON(!upper->checked);
-		rb_node = tree_insert(&cache->rb_root, upper->bytenr,
-				      &upper->rb_node);
-		BUG_ON(rb_node);
+		BUG_ON(cowonly != upper->cowonly);
+		if (!cowonly) {
+			rb_node = tree_insert(&cache->rb_root, upper->bytenr,
+					      &upper->rb_node);
+			BUG_ON(rb_node);
+		}
 
 		list_add_tail(&edge->list[UPPER], &upper->lower);
 
 		list_for_each_entry(edge, &upper->upper, list[LOWER])
 			list_add_tail(&edge->list[UPPER], &list);
 	}
+	/*
+	 * process useless backref nodes. backref nodes for tree leaves
+	 * are deleted from the cache. backref nodes for upper level
+	 * tree blocks are left in the cache to avoid unnecessary backref
+	 * lookup.
+	 */
+	while (!list_empty(&useless)) {
+		upper = list_entry(useless.next, struct backref_node, list);
+		list_del_init(&upper->list);
+		BUG_ON(!list_empty(&upper->upper));
+		if (upper == node)
+			node = NULL;
+		if (upper->lowest) {
+			list_del_init(&upper->lower);
+			upper->lowest = 0;
+		}
+		while (!list_empty(&upper->lower)) {
+			edge = list_entry(upper->lower.next,
+					  struct backref_edge, list[UPPER]);
+			list_del(&edge->list[UPPER]);
+			list_del(&edge->list[LOWER]);
+			lower = edge->node[LOWER];
+			free_backref_edge(cache, edge);
+
+			if (list_empty(&lower->upper))
+				list_add(&lower->list, &useless);
+		}
+		__mark_block_processed(rc, upper);
+		if (upper->level > 0) {
+			list_add(&upper->list, &cache->detached);
+			upper->detached = 1;
+		} else {
+			rb_erase(&upper->rb_node, &cache->rb_root);
+			free_backref_node(cache, upper);
+		}
+	}
 out:
 	btrfs_free_path(path1);
 	btrfs_free_path(path2);
 	if (err) {
-		INIT_LIST_HEAD(&list);
+		while (!list_empty(&useless)) {
+			lower = list_entry(useless.next,
+					   struct backref_node, upper);
+			list_del_init(&lower->upper);
+		}
 		upper = node;
+		INIT_LIST_HEAD(&list);
 		while (upper) {
 			if (RB_EMPTY_NODE(&upper->rb_node)) {
 				list_splice_tail(&upper->upper, &list);
-				kfree(upper);
+				free_backref_node(cache, upper);
 			}
 
 			if (list_empty(&list))
@@ -833,14 +1097,103 @@ out:
 
 			edge = list_entry(list.next, struct backref_edge,
 					  list[LOWER]);
+			list_del(&edge->list[LOWER]);
 			upper = edge->node[UPPER];
-			kfree(edge);
+			free_backref_edge(cache, edge);
 		}
 		return ERR_PTR(err);
 	}
+	BUG_ON(node && node->detached);
 	return node;
 }
 
+/*
+ * helper to add backref node for the newly created snapshot.
+ * the backref node is created by cloning backref node that
+ * corresponds to root of source tree
+ */
+static int clone_backref_node(struct btrfs_trans_handle *trans,
+			      struct reloc_control *rc,
+			      struct btrfs_root *src,
+			      struct btrfs_root *dest)
+{
+	struct btrfs_root *reloc_root = src->reloc_root;
+	struct backref_cache *cache = &rc->backref_cache;
+	struct backref_node *node = NULL;
+	struct backref_node *new_node;
+	struct backref_edge *edge;
+	struct backref_edge *new_edge;
+	struct rb_node *rb_node;
+
+	if (cache->last_trans > 0)
+		update_backref_cache(trans, cache);
+
+	rb_node = tree_search(&cache->rb_root, src->commit_root->start);
+	if (rb_node) {
+		node = rb_entry(rb_node, struct backref_node, rb_node);
+		if (node->detached)
+			node = NULL;
+		else
+			BUG_ON(node->new_bytenr != reloc_root->node->start);
+	}
+
+	if (!node) {
+		rb_node = tree_search(&cache->rb_root,
+				      reloc_root->commit_root->start);
+		if (rb_node) {
+			node = rb_entry(rb_node, struct backref_node,
+					rb_node);
+			BUG_ON(node->detached);
+		}
+	}
+
+	if (!node)
+		return 0;
+
+	new_node = alloc_backref_node(cache);
+	if (!new_node)
+		return -ENOMEM;
+
+	new_node->bytenr = dest->node->start;
+	new_node->level = node->level;
+	new_node->lowest = node->lowest;
+	new_node->root = dest;
+
+	if (!node->lowest) {
+		list_for_each_entry(edge, &node->lower, list[UPPER]) {
+			new_edge = alloc_backref_edge(cache);
+			if (!new_edge)
+				goto fail;
+
+			new_edge->node[UPPER] = new_node;
+			new_edge->node[LOWER] = edge->node[LOWER];
+			list_add_tail(&new_edge->list[UPPER],
+				      &new_node->lower);
+		}
+	}
+
+	rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
+			      &new_node->rb_node);
+	BUG_ON(rb_node);
+
+	if (!new_node->lowest) {
+		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
+			list_add_tail(&new_edge->list[LOWER],
+				      &new_edge->node[LOWER]->upper);
+		}
+	}
+	return 0;
+fail:
+	while (!list_empty(&new_node->lower)) {
+		new_edge = list_entry(new_node->lower.next,
+				      struct backref_edge, list[UPPER]);
+		list_del(&new_edge->list[UPPER]);
+		free_backref_edge(cache, new_edge);
+	}
+	free_backref_node(cache, new_node);
+	return -ENOMEM;
+}
+
 /*
  * helper to add 'address of tree root -> reloc tree' mapping
  */
@@ -901,12 +1254,8 @@ static int __update_reloc_root(struct btrfs_root *root, int del)
 	return 0;
 }
 
-/*
- * create reloc tree for a given fs tree. reloc tree is just a
- * snapshot of the fs tree with special root objectid.
- */
-int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root)
+static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root, u64 objectid)
 {
 	struct btrfs_root *reloc_root;
 	struct extent_buffer *eb;
@@ -914,36 +1263,45 @@ int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
 	struct btrfs_key root_key;
 	int ret;
 
-	if (root->reloc_root) {
-		reloc_root = root->reloc_root;
-		reloc_root->last_trans = trans->transid;
-		return 0;
-	}
-
-	if (!root->fs_info->reloc_ctl ||
-	    !root->fs_info->reloc_ctl->create_reloc_root ||
-	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
-		return 0;
-
 	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
 	BUG_ON(!root_item);
 
 	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
 	root_key.type = BTRFS_ROOT_ITEM_KEY;
-	root_key.offset = root->root_key.objectid;
+	root_key.offset = objectid;
 
-	ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
-			      BTRFS_TREE_RELOC_OBJECTID);
-	BUG_ON(ret);
+	if (root->root_key.objectid == objectid) {
+		/* called by btrfs_init_reloc_root */
+		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
+				      BTRFS_TREE_RELOC_OBJECTID);
+		BUG_ON(ret);
+
+		btrfs_set_root_last_snapshot(&root->root_item,
+					     trans->transid - 1);
+	} else {
+		/*
+		 * called by btrfs_reloc_post_snapshot_hook.
+		 * the source tree is a reloc tree, all tree blocks
+		 * modified after it was created have RELOC flag
+		 * set in their headers. so it's OK to not update
+		 * the 'last_snapshot'.
+		 */
+		ret = btrfs_copy_root(trans, root, root->node, &eb,
+				      BTRFS_TREE_RELOC_OBJECTID);
+		BUG_ON(ret);
+	}
 
-	btrfs_set_root_last_snapshot(&root->root_item, trans->transid - 1);
 	memcpy(root_item, &root->root_item, sizeof(*root_item));
-	btrfs_set_root_refs(root_item, 1);
 	btrfs_set_root_bytenr(root_item, eb->start);
 	btrfs_set_root_level(root_item, btrfs_header_level(eb));
 	btrfs_set_root_generation(root_item, trans->transid);
-	memset(&root_item->drop_progress, 0, sizeof(struct btrfs_disk_key));
-	root_item->drop_level = 0;
+
+	if (root->root_key.objectid == objectid) {
+		btrfs_set_root_refs(root_item, 0);
+		memset(&root_item->drop_progress, 0,
+		       sizeof(struct btrfs_disk_key));
+		root_item->drop_level = 0;
+	}
 
 	btrfs_tree_unlock(eb);
 	free_extent_buffer(eb);
@@ -957,6 +1315,37 @@ int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
 						 &root_key);
 	BUG_ON(IS_ERR(reloc_root));
 	reloc_root->last_trans = trans->transid;
+	return reloc_root;
+}
+
+/*
+ * create reloc tree for a given fs tree. reloc tree is just a
+ * snapshot of the fs tree with special root objectid.
+ */
+int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	struct reloc_control *rc = root->fs_info->reloc_ctl;
+	int clear_rsv = 0;
+
+	if (root->reloc_root) {
+		reloc_root = root->reloc_root;
+		reloc_root->last_trans = trans->transid;
+		return 0;
+	}
+
+	if (!rc || !rc->create_reloc_tree ||
+	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
+		return 0;
+
+	if (!trans->block_rsv) {
+		trans->block_rsv = rc->block_rsv;
+		clear_rsv = 1;
+	}
+	reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
+	if (clear_rsv)
+		trans->block_rsv = NULL;
 
 	__add_reloc_root(reloc_root);
 	root->reloc_root = reloc_root;
@@ -980,7 +1369,8 @@ int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 	reloc_root = root->reloc_root;
 	root_item = &reloc_root->root_item;
 
-	if (btrfs_root_refs(root_item) == 0) {
+	if (root->fs_info->reloc_ctl->merge_reloc_tree &&
+	    btrfs_root_refs(root_item) == 0) {
 		root->reloc_root = NULL;
 		del = 1;
 	}
@@ -1102,8 +1492,7 @@ static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
 		goto out;
 	}
 
-	if (new_bytenr)
-		*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 	ret = 0;
 out:
 	btrfs_free_path(path);
@@ -1114,19 +1503,18 @@ out:
  * update file extent items in the tree leaf to point to
  * the new locations.
  */
-static int replace_file_extents(struct btrfs_trans_handle *trans,
-				struct reloc_control *rc,
-				struct btrfs_root *root,
-				struct extent_buffer *leaf,
-				struct list_head *inode_list)
+static noinline_for_stack
+int replace_file_extents(struct btrfs_trans_handle *trans,
+			 struct reloc_control *rc,
+			 struct btrfs_root *root,
+			 struct extent_buffer *leaf)
 {
 	struct btrfs_key key;
 	struct btrfs_file_extent_item *fi;
 	struct inode *inode = NULL;
-	struct inodevec *ivec = NULL;
 	u64 parent;
 	u64 bytenr;
-	u64 new_bytenr;
+	u64 new_bytenr = 0;
 	u64 num_bytes;
 	u64 end;
 	u32 nritems;
@@ -1166,21 +1554,12 @@ static int replace_file_extents(struct btrfs_trans_handle *trans,
 		 * to complete and drop the extent cache
 		 */
 		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
-			if (!ivec || ivec->nr == INODEVEC_SIZE) {
-				ivec = kmalloc(sizeof(*ivec), GFP_NOFS);
-				BUG_ON(!ivec);
-				ivec->nr = 0;
-				list_add_tail(&ivec->list, inode_list);
-			}
 			if (first) {
 				inode = find_next_inode(root, key.objectid);
-				if (inode)
-					ivec->inode[ivec->nr++] = inode;
 				first = 0;
 			} else if (inode && inode->i_ino < key.objectid) {
+				btrfs_add_delayed_iput(inode);
 				inode = find_next_inode(root, key.objectid);
-				if (inode)
-					ivec->inode[ivec->nr++] = inode;
 			}
 			if (inode && inode->i_ino == key.objectid) {
 				end = key.offset +
@@ -1204,8 +1583,10 @@ static int replace_file_extents(struct btrfs_trans_handle *trans,
 
 		ret = get_new_location(rc->data_inode, &new_bytenr,
 				       bytenr, num_bytes);
-		if (ret > 0)
+		if (ret > 0) {
+			WARN_ON(1);
 			continue;
+		}
 		BUG_ON(ret < 0);
 
 		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
@@ -1225,6 +1606,8 @@ static int replace_file_extents(struct btrfs_trans_handle *trans,
 	}
 	if (dirty)
 		btrfs_mark_buffer_dirty(leaf);
+	if (inode)
+		btrfs_add_delayed_iput(inode);
 	return 0;
 }
 
@@ -1248,11 +1631,11 @@ int memcmp_node_keys(struct extent_buffer *eb, int slot,
  * if no block got replaced, 0 is returned. if there are other
  * errors, a negative error number is returned.
  */
-static int replace_path(struct btrfs_trans_handle *trans,
-			struct btrfs_root *dest, struct btrfs_root *src,
-			struct btrfs_path *path, struct btrfs_key *next_key,
-			struct extent_buffer **leaf,
-			int lowest_level, int max_level)
+static noinline_for_stack
+int replace_path(struct btrfs_trans_handle *trans,
+		 struct btrfs_root *dest, struct btrfs_root *src,
+		 struct btrfs_path *path, struct btrfs_key *next_key,
+		 int lowest_level, int max_level)
 {
 	struct extent_buffer *eb;
 	struct extent_buffer *parent;
@@ -1263,16 +1646,16 @@ static int replace_path(struct btrfs_trans_handle *trans,
 	u64 new_ptr_gen;
 	u64 last_snapshot;
 	u32 blocksize;
+	int cow = 0;
 	int level;
 	int ret;
 	int slot;
 
 	BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
 	BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
-	BUG_ON(lowest_level > 1 && leaf);
 
 	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
-
+again:
 	slot = path->slots[lowest_level];
 	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
 
@@ -1286,8 +1669,10 @@ static int replace_path(struct btrfs_trans_handle *trans,
 		return 0;
 	}
 
-	ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
-	BUG_ON(ret);
+	if (cow) {
+		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
+		BUG_ON(ret);
+	}
 	btrfs_set_lock_blocking(eb);
 
 	if (next_key) {
@@ -1331,7 +1716,7 @@ static int replace_path(struct btrfs_trans_handle *trans,
 
 		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
 		    memcmp_node_keys(parent, slot, path, level)) {
-			if (level <= lowest_level && !leaf) {
+			if (level <= lowest_level) {
 				ret = 0;
 				break;
 			}
@@ -1339,16 +1724,12 @@ static int replace_path(struct btrfs_trans_handle *trans,
 			eb = read_tree_block(dest, old_bytenr, blocksize,
 					     old_ptr_gen);
 			btrfs_tree_lock(eb);
-			ret = btrfs_cow_block(trans, dest, eb, parent,
-					      slot, &eb);
-			BUG_ON(ret);
-			btrfs_set_lock_blocking(eb);
-
-			if (level <= lowest_level) {
-				*leaf = eb;
-				ret = 0;
-				break;
+			if (cow) {
+				ret = btrfs_cow_block(trans, dest, eb, parent,
+						      slot, &eb);
+				BUG_ON(ret);
 			}
+			btrfs_set_lock_blocking(eb);
 
 			btrfs_tree_unlock(parent);
 			free_extent_buffer(parent);
@@ -1357,6 +1738,13 @@ static int replace_path(struct btrfs_trans_handle *trans,
 			continue;
 		}
 
+		if (!cow) {
+			btrfs_tree_unlock(parent);
+			free_extent_buffer(parent);
+			cow = 1;
+			goto again;
+		}
+
 		btrfs_node_key_to_cpu(path->nodes[level], &key,
 				      path->slots[level]);
 		btrfs_release_path(src, path);
@@ -1562,20 +1950,6 @@ static int invalidate_extent_cache(struct btrfs_root *root,
 	return 0;
 }
 
-static void put_inodes(struct list_head *list)
-{
-	struct inodevec *ivec;
-	while (!list_empty(list)) {
-		ivec = list_entry(list->next, struct inodevec, list);
-		list_del(&ivec->list);
-		while (ivec->nr > 0) {
-			ivec->nr--;
-			iput(ivec->inode[ivec->nr]);
-		}
-		kfree(ivec);
-	}
-}
-
 static int find_next_key(struct btrfs_path *path, int level,
 			 struct btrfs_key *key)
 
@@ -1608,13 +1982,14 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
 	struct btrfs_root *reloc_root;
 	struct btrfs_root_item *root_item;
 	struct btrfs_path *path;
-	struct extent_buffer *leaf = NULL;
+	struct extent_buffer *leaf;
 	unsigned long nr;
 	int level;
 	int max_level;
 	int replaced = 0;
 	int ret;
 	int err = 0;
+	u32 min_reserved;
 
 	path = btrfs_alloc_path();
 	if (!path)
@@ -1648,34 +2023,23 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
 		btrfs_unlock_up_safe(path, 0);
 	}
 
-	if (level == 0 && rc->stage == UPDATE_DATA_PTRS) {
-		trans = btrfs_start_transaction(root, 1);
+	min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
+	memset(&next_key, 0, sizeof(next_key));
 
-		leaf = path->nodes[0];
-		btrfs_item_key_to_cpu(leaf, &key, 0);
-		btrfs_release_path(reloc_root, path);
+	while (1) {
+		trans = btrfs_start_transaction(root, 0);
+		trans->block_rsv = rc->block_rsv;
 
-		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
-		if (ret < 0) {
-			err = ret;
-			goto out;
+		ret = btrfs_block_rsv_check(trans, root, rc->block_rsv,
+					    min_reserved, 0);
+		if (ret) {
+			BUG_ON(ret != -EAGAIN);
+			ret = btrfs_commit_transaction(trans, root);
+			BUG_ON(ret);
+			continue;
 		}
 
-		leaf = path->nodes[0];
-		btrfs_unlock_up_safe(path, 1);
-		ret = replace_file_extents(trans, rc, root, leaf,
-					   &inode_list);
-		if (ret < 0)
-			err = ret;
-		goto out;
-	}
-
-	memset(&next_key, 0, sizeof(next_key));
-
-	while (1) {
-		leaf = NULL;
 		replaced = 0;
-		trans = btrfs_start_transaction(root, 1);
 		max_level = level;
 
 		ret = walk_down_reloc_tree(reloc_root, path, &level);
@@ -1689,14 +2053,9 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
 		if (!find_next_key(path, level, &key) &&
 		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
 			ret = 0;
-		} else if (level == 1 && rc->stage == UPDATE_DATA_PTRS) {
-			ret = replace_path(trans, root, reloc_root,
-					   path, &next_key, &leaf,
-					   level, max_level);
 		} else {
-			ret = replace_path(trans, root, reloc_root,
-					   path, &next_key, NULL,
-					   level, max_level);
+			ret = replace_path(trans, root, reloc_root, path,
+					   &next_key, level, max_level);
 		}
 		if (ret < 0) {
 			err = ret;
@@ -1708,16 +2067,6 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
 			btrfs_node_key_to_cpu(path->nodes[level], &key,
 					      path->slots[level]);
 			replaced = 1;
-		} else if (leaf) {
-			/*
-			 * no block got replaced, try replacing file extents
-			 */
-			btrfs_item_key_to_cpu(leaf, &key, 0);
-			ret = replace_file_extents(trans, rc, root, leaf,
-						   &inode_list);
-			btrfs_tree_unlock(leaf);
-			free_extent_buffer(leaf);
-			BUG_ON(ret < 0);
 		}
 
 		ret = walk_up_reloc_tree(reloc_root, path, &level);
@@ -1734,15 +2083,10 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
 		root_item->drop_level = level;
 
 		nr = trans->blocks_used;
-		btrfs_end_transaction(trans, root);
+		btrfs_end_transaction_throttle(trans, root);
 
 		btrfs_btree_balance_dirty(root, nr);
 
-		/*
-		 * put inodes outside transaction, otherwise we may deadlock.
-		 */
-		put_inodes(&inode_list);
-
 		if (replaced && rc->stage == UPDATE_DATA_PTRS)
 			invalidate_extent_cache(root, &key, &next_key);
 	}
@@ -1765,87 +2109,125 @@ out:
 		       sizeof(root_item->drop_progress));
 		root_item->drop_level = 0;
 		btrfs_set_root_refs(root_item, 0);
+		btrfs_update_reloc_root(trans, root);
 	}
 
 	nr = trans->blocks_used;
-	btrfs_end_transaction(trans, root);
+	btrfs_end_transaction_throttle(trans, root);
 
 	btrfs_btree_balance_dirty(root, nr);
 
-	put_inodes(&inode_list);
-
 	if (replaced && rc->stage == UPDATE_DATA_PTRS)
 		invalidate_extent_cache(root, &key, &next_key);
 
 	return err;
 }
 
-/*
- * callback for the work threads.
- * this function merges reloc tree with corresponding fs tree,
- * and then drops the reloc tree.
- */
-static void merge_func(struct btrfs_work *work)
+static noinline_for_stack
+int prepare_to_merge(struct reloc_control *rc, int err)
 {
-	struct btrfs_trans_handle *trans;
-	struct btrfs_root *root;
+	struct btrfs_root *root = rc->extent_root;
 	struct btrfs_root *reloc_root;
-	struct async_merge *async;
+	struct btrfs_trans_handle *trans;
+	LIST_HEAD(reloc_roots);
+	u64 num_bytes = 0;
+	int ret;
+	int retries = 0;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
+	rc->merging_rsv_size += rc->nodes_relocated * 2;
+	mutex_unlock(&root->fs_info->trans_mutex);
+again:
+	if (!err) {
+		num_bytes = rc->merging_rsv_size;
+		ret = btrfs_block_rsv_add(NULL, root, rc->block_rsv,
+					  num_bytes, &retries);
+		if (ret)
+			err = ret;
+	}
 
-	async = container_of(work, struct async_merge, work);
-	reloc_root = async->root;
+	trans = btrfs_join_transaction(rc->extent_root, 1);
+
+	if (!err) {
+		if (num_bytes != rc->merging_rsv_size) {
+			btrfs_end_transaction(trans, rc->extent_root);
+			btrfs_block_rsv_release(rc->extent_root,
+						rc->block_rsv, num_bytes);
+			retries = 0;
+			goto again;
+		}
+	}
+
+	rc->merge_reloc_tree = 1;
+
+	while (!list_empty(&rc->reloc_roots)) {
+		reloc_root = list_entry(rc->reloc_roots.next,
+					struct btrfs_root, root_list);
+		list_del_init(&reloc_root->root_list);
 
-	if (btrfs_root_refs(&reloc_root->root_item) > 0) {
 		root = read_fs_root(reloc_root->fs_info,
 				    reloc_root->root_key.offset);
 		BUG_ON(IS_ERR(root));
 		BUG_ON(root->reloc_root != reloc_root);
 
-		merge_reloc_root(async->rc, root);
-
-		trans = btrfs_start_transaction(root, 1);
+		/*
+		 * set reference count to 1, so btrfs_recover_relocation
+		 * knows it should resumes merging
+		 */
+		if (!err)
+			btrfs_set_root_refs(&reloc_root->root_item, 1);
 		btrfs_update_reloc_root(trans, root);
-		btrfs_end_transaction(trans, root);
-	}
 
-	btrfs_drop_snapshot(reloc_root, 0);
+		list_add(&reloc_root->root_list, &reloc_roots);
+	}
 
-	if (atomic_dec_and_test(async->num_pending))
-		complete(async->done);
+	list_splice(&reloc_roots, &rc->reloc_roots);
 
-	kfree(async);
+	if (!err)
+		btrfs_commit_transaction(trans, rc->extent_root);
+	else
+		btrfs_end_transaction(trans, rc->extent_root);
+	return err;
 }
 
-static int merge_reloc_roots(struct reloc_control *rc)
+static noinline_for_stack
+int merge_reloc_roots(struct reloc_control *rc)
 {
-	struct async_merge *async;
 	struct btrfs_root *root;
-	struct completion done;
-	atomic_t num_pending;
+	struct btrfs_root *reloc_root;
+	LIST_HEAD(reloc_roots);
+	int found = 0;
+	int ret;
+again:
+	root = rc->extent_root;
+	mutex_lock(&root->fs_info->trans_mutex);
+	list_splice_init(&rc->reloc_roots, &reloc_roots);
+	mutex_unlock(&root->fs_info->trans_mutex);
 
-	init_completion(&done);
-	atomic_set(&num_pending, 1);
+	while (!list_empty(&reloc_roots)) {
+		found = 1;
+		reloc_root = list_entry(reloc_roots.next,
+					struct btrfs_root, root_list);
 
-	while (!list_empty(&rc->reloc_roots)) {
-		root = list_entry(rc->reloc_roots.next,
-				  struct btrfs_root, root_list);
-		list_del_init(&root->root_list);
+		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
+			root = read_fs_root(reloc_root->fs_info,
+					    reloc_root->root_key.offset);
+			BUG_ON(IS_ERR(root));
+			BUG_ON(root->reloc_root != reloc_root);
 
-		async = kmalloc(sizeof(*async), GFP_NOFS);
-		BUG_ON(!async);
-		async->work.func = merge_func;
-		async->work.flags = 0;
-		async->rc = rc;
-		async->root = root;
-		async->done = &done;
-		async->num_pending = &num_pending;
-		atomic_inc(&num_pending);
-		btrfs_queue_worker(&rc->workers, &async->work);
+			ret = merge_reloc_root(rc, root);
+			BUG_ON(ret);
+		} else {
+			list_del_init(&reloc_root->root_list);
+		}
+		btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0);
 	}
 
-	if (!atomic_dec_and_test(&num_pending))
-		wait_for_completion(&done);
-
+	if (found) {
+		found = 0;
+		goto again;
+	}
 	BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
 	return 0;
 }
@@ -1876,119 +2258,169 @@ static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
 	return btrfs_record_root_in_trans(trans, root);
 }
 
-/*
- * select one tree from trees that references the block.
- * for blocks in refernce counted trees, we preper reloc tree.
- * if no reloc tree found and reloc_only is true, NULL is returned.
- */
-static struct btrfs_root *__select_one_root(struct btrfs_trans_handle *trans,
-					    struct backref_node *node,
-					    struct backref_edge *edges[],
-					    int *nr, int reloc_only)
+static noinline_for_stack
+struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
+				     struct reloc_control *rc,
+				     struct backref_node *node,
+				     struct backref_edge *edges[], int *nr)
 {
 	struct backref_node *next;
 	struct btrfs_root *root;
-	int index;
-	int loop = 0;
-again:
-	index = 0;
+	int index = 0;
+
 	next = node;
 	while (1) {
 		cond_resched();
 		next = walk_up_backref(next, edges, &index);
 		root = next->root;
-		if (!root) {
-			BUG_ON(!node->old_root);
-			goto skip;
-		}
-
-		/* no other choice for non-refernce counted tree */
-		if (!root->ref_cows) {
-			BUG_ON(reloc_only);
-			break;
-		}
+		BUG_ON(!root);
+		BUG_ON(!root->ref_cows);
 
 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
 			record_reloc_root_in_trans(trans, root);
 			break;
 		}
 
-		if (loop) {
-			btrfs_record_root_in_trans(trans, root);
+		btrfs_record_root_in_trans(trans, root);
+		root = root->reloc_root;
+
+		if (next->new_bytenr != root->node->start) {
+			BUG_ON(next->new_bytenr);
+			BUG_ON(!list_empty(&next->list));
+			next->new_bytenr = root->node->start;
+			next->root = root;
+			list_add_tail(&next->list,
+				      &rc->backref_cache.changed);
+			__mark_block_processed(rc, next);
 			break;
 		}
 
-		if (reloc_only || next != node) {
-			if (!root->reloc_root)
-				btrfs_record_root_in_trans(trans, root);
-			root = root->reloc_root;
-			/*
-			 * if the reloc tree was created in current
-			 * transation, there is no node in backref tree
-			 * corresponds to the root of the reloc tree.
-			 */
-			if (btrfs_root_last_snapshot(&root->root_item) ==
-			    trans->transid - 1)
-				break;
-		}
-skip:
+		WARN_ON(1);
 		root = NULL;
 		next = walk_down_backref(edges, &index);
 		if (!next || next->level <= node->level)
 			break;
 	}
+	if (!root)
+		return NULL;
 
-	if (!root && !loop && !reloc_only) {
-		loop = 1;
-		goto again;
+	*nr = index;
+	next = node;
+	/* setup backref node path for btrfs_reloc_cow_block */
+	while (1) {
+		rc->backref_cache.path[next->level] = next;
+		if (--index < 0)
+			break;
+		next = edges[index]->node[UPPER];
 	}
-
-	if (root)
-		*nr = index;
-	else
-		*nr = 0;
-
 	return root;
 }
 
+/*
+ * select a tree root for relocation. return NULL if the block
+ * is reference counted. we should use do_relocation() in this
+ * case. return a tree root pointer if the block isn't reference
+ * counted. return -ENOENT if the block is root of reloc tree.
+ */
 static noinline_for_stack
 struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
 				   struct backref_node *node)
 {
+	struct backref_node *next;
+	struct btrfs_root *root;
+	struct btrfs_root *fs_root = NULL;
 	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
-	int nr;
-	return __select_one_root(trans, node, edges, &nr, 0);
+	int index = 0;
+
+	next = node;
+	while (1) {
+		cond_resched();
+		next = walk_up_backref(next, edges, &index);
+		root = next->root;
+		BUG_ON(!root);
+
+		/* no other choice for non-refernce counted tree */
+		if (!root->ref_cows)
+			return root;
+
+		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
+			fs_root = root;
+
+		if (next != node)
+			return NULL;
+
+		next = walk_down_backref(edges, &index);
+		if (!next || next->level <= node->level)
+			break;
+	}
+
+	if (!fs_root)
+		return ERR_PTR(-ENOENT);
+	return fs_root;
 }
 
 static noinline_for_stack
-struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
-				     struct backref_node *node,
-				     struct backref_edge *edges[], int *nr)
+u64 calcu_metadata_size(struct reloc_control *rc,
+			struct backref_node *node, int reserve)
 {
-	return __select_one_root(trans, node, edges, nr, 1);
+	struct backref_node *next = node;
+	struct backref_edge *edge;
+	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
+	u64 num_bytes = 0;
+	int index = 0;
+
+	BUG_ON(reserve && node->processed);
+
+	while (next) {
+		cond_resched();
+		while (1) {
+			if (next->processed && (reserve || next != node))
+				break;
+
+			num_bytes += btrfs_level_size(rc->extent_root,
+						      next->level);
+
+			if (list_empty(&next->upper))
+				break;
+
+			edge = list_entry(next->upper.next,
+					  struct backref_edge, list[LOWER]);
+			edges[index++] = edge;
+			next = edge->node[UPPER];
+		}
+		next = walk_down_backref(edges, &index);
+	}
+	return num_bytes;
 }
 
-static void grab_path_buffers(struct btrfs_path *path,
-			      struct backref_node *node,
-			      struct backref_edge *edges[], int nr)
+static int reserve_metadata_space(struct btrfs_trans_handle *trans,
+				  struct reloc_control *rc,
+				  struct backref_node *node)
 {
-	int i = 0;
-	while (1) {
-		drop_node_buffer(node);
-		node->eb = path->nodes[node->level];
-		BUG_ON(!node->eb);
-		if (path->locks[node->level])
-			node->locked = 1;
-		path->nodes[node->level] = NULL;
-		path->locks[node->level] = 0;
-
-		if (i >= nr)
-			break;
+	struct btrfs_root *root = rc->extent_root;
+	u64 num_bytes;
+	int ret;
 
-		edges[i]->blockptr = node->eb->start;
-		node = edges[i]->node[UPPER];
-		i++;
+	num_bytes = calcu_metadata_size(rc, node, 1) * 2;
+
+	trans->block_rsv = rc->block_rsv;
+	ret = btrfs_block_rsv_add(trans, root, rc->block_rsv, num_bytes,
+				  &rc->block_rsv_retries);
+	if (ret) {
+		if (ret == -EAGAIN)
+			rc->commit_transaction = 1;
+		return ret;
 	}
+
+	rc->block_rsv_retries = 0;
+	return 0;
+}
+
+static void release_metadata_space(struct reloc_control *rc,
+				   struct backref_node *node)
+{
+	u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2;
+	btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes);
 }
 
 /*
@@ -1999,6 +2431,7 @@ static void grab_path_buffers(struct btrfs_path *path,
  * in that case this function just updates pointers.
  */
 static int do_relocation(struct btrfs_trans_handle *trans,
+			 struct reloc_control *rc,
 			 struct backref_node *node,
 			 struct btrfs_key *key,
 			 struct btrfs_path *path, int lowest)
@@ -2019,18 +2452,25 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 	BUG_ON(lowest && node->eb);
 
 	path->lowest_level = node->level + 1;
+	rc->backref_cache.path[node->level] = node;
 	list_for_each_entry(edge, &node->upper, list[LOWER]) {
 		cond_resched();
-		if (node->eb && node->eb->start == edge->blockptr)
-			continue;
 
 		upper = edge->node[UPPER];
-		root = select_reloc_root(trans, upper, edges, &nr);
-		if (!root)
-			continue;
-
-		if (upper->eb && !upper->locked)
+		root = select_reloc_root(trans, rc, upper, edges, &nr);
+		BUG_ON(!root);
+
+		if (upper->eb && !upper->locked) {
+			if (!lowest) {
+				ret = btrfs_bin_search(upper->eb, key,
+						       upper->level, &slot);
+				BUG_ON(ret);
+				bytenr = btrfs_node_blockptr(upper->eb, slot);
+				if (node->eb->start == bytenr)
+					goto next;
+			}
 			drop_node_buffer(upper);
+		}
 
 		if (!upper->eb) {
 			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
@@ -2040,11 +2480,17 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 			}
 			BUG_ON(ret > 0);
 
-			slot = path->slots[upper->level];
+			if (!upper->eb) {
+				upper->eb = path->nodes[upper->level];
+				path->nodes[upper->level] = NULL;
+			} else {
+				BUG_ON(upper->eb != path->nodes[upper->level]);
+			}
 
-			btrfs_unlock_up_safe(path, upper->level + 1);
-			grab_path_buffers(path, upper, edges, nr);
+			upper->locked = 1;
+			path->locks[upper->level] = 0;
 
+			slot = path->slots[upper->level];
 			btrfs_release_path(NULL, path);
 		} else {
 			ret = btrfs_bin_search(upper->eb, key, upper->level,
@@ -2053,14 +2499,11 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 		}
 
 		bytenr = btrfs_node_blockptr(upper->eb, slot);
-		if (!lowest) {
-			if (node->eb->start == bytenr) {
-				btrfs_tree_unlock(upper->eb);
-				upper->locked = 0;
-				continue;
-			}
+		if (lowest) {
+			BUG_ON(bytenr != node->bytenr);
 		} else {
-			BUG_ON(node->bytenr != bytenr);
+			if (node->eb->start == bytenr)
+				goto next;
 		}
 
 		blocksize = btrfs_level_size(root, node->level);
@@ -2072,13 +2515,13 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 		if (!node->eb) {
 			ret = btrfs_cow_block(trans, root, eb, upper->eb,
 					      slot, &eb);
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
 			if (ret < 0) {
 				err = ret;
-				break;
+				goto next;
 			}
-			btrfs_set_lock_blocking(eb);
-			node->eb = eb;
-			node->locked = 1;
+			BUG_ON(node->eb != eb);
 		} else {
 			btrfs_set_node_blockptr(upper->eb, slot,
 						node->eb->start);
@@ -2096,67 +2539,80 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 			ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
 			BUG_ON(ret);
 		}
-		if (!lowest) {
-			btrfs_tree_unlock(upper->eb);
-			upper->locked = 0;
-		}
+next:
+		if (!upper->pending)
+			drop_node_buffer(upper);
+		else
+			unlock_node_buffer(upper);
+		if (err)
+			break;
+	}
+
+	if (!err && node->pending) {
+		drop_node_buffer(node);
+		list_move_tail(&node->list, &rc->backref_cache.changed);
+		node->pending = 0;
 	}
+
 	path->lowest_level = 0;
+	BUG_ON(err == -ENOSPC);
 	return err;
 }
 
 static int link_to_upper(struct btrfs_trans_handle *trans,
+			 struct reloc_control *rc,
 			 struct backref_node *node,
 			 struct btrfs_path *path)
 {
 	struct btrfs_key key;
-	if (!node->eb || list_empty(&node->upper))
-		return 0;
 
 	btrfs_node_key_to_cpu(node->eb, &key, 0);
-	return do_relocation(trans, node, &key, path, 0);
+	return do_relocation(trans, rc, node, &key, path, 0);
 }
 
 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
-				struct backref_cache *cache,
-				struct btrfs_path *path)
+				struct reloc_control *rc,
+				struct btrfs_path *path, int err)
 {
+	LIST_HEAD(list);
+	struct backref_cache *cache = &rc->backref_cache;
 	struct backref_node *node;
 	int level;
 	int ret;
-	int err = 0;
 
 	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 		while (!list_empty(&cache->pending[level])) {
 			node = list_entry(cache->pending[level].next,
-					  struct backref_node, lower);
-			BUG_ON(node->level != level);
+					  struct backref_node, list);
+			list_move_tail(&node->list, &list);
+			BUG_ON(!node->pending);
 
-			ret = link_to_upper(trans, node, path);
-			if (ret < 0)
-				err = ret;
-			/*
-			 * this remove the node from the pending list and
-			 * may add some other nodes to the level + 1
-			 * pending list
-			 */
-			remove_backref_node(cache, node);
+			if (!err) {
+				ret = link_to_upper(trans, rc, node, path);
+				if (ret < 0)
+					err = ret;
+			}
 		}
+		list_splice_init(&list, &cache->pending[level]);
 	}
-	BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
 	return err;
 }
 
 static void mark_block_processed(struct reloc_control *rc,
-				 struct backref_node *node)
+				 u64 bytenr, u32 blocksize)
+{
+	set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
+			EXTENT_DIRTY, GFP_NOFS);
+}
+
+static void __mark_block_processed(struct reloc_control *rc,
+				   struct backref_node *node)
 {
 	u32 blocksize;
 	if (node->level == 0 ||
 	    in_block_group(node->bytenr, rc->block_group)) {
 		blocksize = btrfs_level_size(rc->extent_root, node->level);
-		set_extent_bits(&rc->processed_blocks, node->bytenr,
-				node->bytenr + blocksize - 1, EXTENT_DIRTY,
-				GFP_NOFS);
+		mark_block_processed(rc, node->bytenr, blocksize);
 	}
 	node->processed = 1;
 }
@@ -2179,7 +2635,7 @@ static void update_processed_blocks(struct reloc_control *rc,
 			if (next->processed)
 				break;
 
-			mark_block_processed(rc, next);
+			__mark_block_processed(rc, next);
 
 			if (list_empty(&next->upper))
 				break;
@@ -2202,138 +2658,6 @@ static int tree_block_processed(u64 bytenr, u32 blocksize,
 	return 0;
 }
 
-/*
- * check if there are any file extent pointers in the leaf point to
- * data require processing
- */
-static int check_file_extents(struct reloc_control *rc,
-			      u64 bytenr, u32 blocksize, u64 ptr_gen)
-{
-	struct btrfs_key found_key;
-	struct btrfs_file_extent_item *fi;
-	struct extent_buffer *leaf;
-	u32 nritems;
-	int i;
-	int ret = 0;
-
-	leaf = read_tree_block(rc->extent_root, bytenr, blocksize, ptr_gen);
-
-	nritems = btrfs_header_nritems(leaf);
-	for (i = 0; i < nritems; i++) {
-		cond_resched();
-		btrfs_item_key_to_cpu(leaf, &found_key, i);
-		if (found_key.type != BTRFS_EXTENT_DATA_KEY)
-			continue;
-		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
-		if (btrfs_file_extent_type(leaf, fi) ==
-		    BTRFS_FILE_EXTENT_INLINE)
-			continue;
-		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
-		if (bytenr == 0)
-			continue;
-		if (in_block_group(bytenr, rc->block_group)) {
-			ret = 1;
-			break;
-		}
-	}
-	free_extent_buffer(leaf);
-	return ret;
-}
-
-/*
- * scan child blocks of a given block to find blocks require processing
- */
-static int add_child_blocks(struct btrfs_trans_handle *trans,
-			    struct reloc_control *rc,
-			    struct backref_node *node,
-			    struct rb_root *blocks)
-{
-	struct tree_block *block;
-	struct rb_node *rb_node;
-	u64 bytenr;
-	u64 ptr_gen;
-	u32 blocksize;
-	u32 nritems;
-	int i;
-	int err = 0;
-
-	nritems = btrfs_header_nritems(node->eb);
-	blocksize = btrfs_level_size(rc->extent_root, node->level - 1);
-	for (i = 0; i < nritems; i++) {
-		cond_resched();
-		bytenr = btrfs_node_blockptr(node->eb, i);
-		ptr_gen = btrfs_node_ptr_generation(node->eb, i);
-		if (ptr_gen == trans->transid)
-			continue;
-		if (!in_block_group(bytenr, rc->block_group) &&
-		    (node->level > 1 || rc->stage == MOVE_DATA_EXTENTS))
-			continue;
-		if (tree_block_processed(bytenr, blocksize, rc))
-			continue;
-
-		readahead_tree_block(rc->extent_root,
-				     bytenr, blocksize, ptr_gen);
-	}
-
-	for (i = 0; i < nritems; i++) {
-		cond_resched();
-		bytenr = btrfs_node_blockptr(node->eb, i);
-		ptr_gen = btrfs_node_ptr_generation(node->eb, i);
-		if (ptr_gen == trans->transid)
-			continue;
-		if (!in_block_group(bytenr, rc->block_group) &&
-		    (node->level > 1 || rc->stage == MOVE_DATA_EXTENTS))
-			continue;
-		if (tree_block_processed(bytenr, blocksize, rc))
-			continue;
-		if (!in_block_group(bytenr, rc->block_group) &&
-		    !check_file_extents(rc, bytenr, blocksize, ptr_gen))
-			continue;
-
-		block = kmalloc(sizeof(*block), GFP_NOFS);
-		if (!block) {
-			err = -ENOMEM;
-			break;
-		}
-		block->bytenr = bytenr;
-		btrfs_node_key_to_cpu(node->eb, &block->key, i);
-		block->level = node->level - 1;
-		block->key_ready = 1;
-		rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
-		BUG_ON(rb_node);
-	}
-	if (err)
-		free_block_list(blocks);
-	return err;
-}
-
-/*
- * find adjacent blocks require processing
- */
-static noinline_for_stack
-int add_adjacent_blocks(struct btrfs_trans_handle *trans,
-			struct reloc_control *rc,
-			struct backref_cache *cache,
-			struct rb_root *blocks, int level,
-			struct backref_node **upper)
-{
-	struct backref_node *node;
-	int ret = 0;
-
-	WARN_ON(!list_empty(&cache->pending[level]));
-
-	if (list_empty(&cache->pending[level + 1]))
-		return 1;
-
-	node = list_entry(cache->pending[level + 1].next,
-			  struct backref_node, lower);
-	if (node->eb)
-		ret = add_child_blocks(trans, rc, node, blocks);
-
-	*upper = node;
-	return ret;
-}
-
 static int get_tree_block_key(struct reloc_control *rc,
 			      struct tree_block *block)
 {
@@ -2371,40 +2695,53 @@ static int relocate_tree_block(struct btrfs_trans_handle *trans,
 				struct btrfs_path *path)
 {
 	struct btrfs_root *root;
-	int ret;
+	int release = 0;
+	int ret = 0;
 
+	if (!node)
+		return 0;
+
+	BUG_ON(node->processed);
 	root = select_one_root(trans, node);
-	if (unlikely(!root)) {
-		rc->found_old_snapshot = 1;
+	if (root == ERR_PTR(-ENOENT)) {
 		update_processed_blocks(rc, node);
-		return 0;
+		goto out;
 	}
 
-	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
-		ret = do_relocation(trans, node, key, path, 1);
-		if (ret < 0)
-			goto out;
-		if (node->level == 0 && rc->stage == UPDATE_DATA_PTRS) {
-			ret = replace_file_extents(trans, rc, root,
-						   node->eb, NULL);
-			if (ret < 0)
-				goto out;
-		}
-		drop_node_buffer(node);
-	} else if (!root->ref_cows) {
-		path->lowest_level = node->level;
-		ret = btrfs_search_slot(trans, root, key, path, 0, 1);
-		btrfs_release_path(root, path);
-		if (ret < 0)
+	if (!root || root->ref_cows) {
+		ret = reserve_metadata_space(trans, rc, node);
+		if (ret)
 			goto out;
-	} else if (root != node->root) {
-		WARN_ON(node->level > 0 || rc->stage != UPDATE_DATA_PTRS);
+		release = 1;
 	}
 
-	update_processed_blocks(rc, node);
-	ret = 0;
+	if (root) {
+		if (root->ref_cows) {
+			BUG_ON(node->new_bytenr);
+			BUG_ON(!list_empty(&node->list));
+			btrfs_record_root_in_trans(trans, root);
+			root = root->reloc_root;
+			node->new_bytenr = root->node->start;
+			node->root = root;
+			list_add_tail(&node->list, &rc->backref_cache.changed);
+		} else {
+			path->lowest_level = node->level;
+			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
+			btrfs_release_path(root, path);
+			if (ret > 0)
+				ret = 0;
+		}
+		if (!ret)
+			update_processed_blocks(rc, node);
+	} else {
+		ret = do_relocation(trans, rc, node, key, path, 1);
+	}
 out:
-	drop_node_buffer(node);
+	if (ret || node->level == 0 || node->cowonly) {
+		if (release)
+			release_metadata_space(rc, node);
+		remove_backref_node(&rc->backref_cache, node);
+	}
 	return ret;
 }
 
@@ -2415,12 +2752,10 @@ static noinline_for_stack
 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
 			 struct reloc_control *rc, struct rb_root *blocks)
 {
-	struct backref_cache *cache;
 	struct backref_node *node;
 	struct btrfs_path *path;
 	struct tree_block *block;
 	struct rb_node *rb_node;
-	int level = -1;
 	int ret;
 	int err = 0;
 
@@ -2428,21 +2763,9 @@ int relocate_tree_blocks(struct btrfs_trans_handle *trans,
 	if (!path)
 		return -ENOMEM;
 
-	cache = kmalloc(sizeof(*cache), GFP_NOFS);
-	if (!cache) {
-		btrfs_free_path(path);
-		return -ENOMEM;
-	}
-
-	backref_cache_init(cache);
-
 	rb_node = rb_first(blocks);
 	while (rb_node) {
 		block = rb_entry(rb_node, struct tree_block, rb_node);
-		if (level == -1)
-			level = block->level;
-		else
-			BUG_ON(level != block->level);
 		if (!block->key_ready)
 			reada_tree_block(rc, block);
 		rb_node = rb_next(rb_node);
@@ -2452,99 +2775,82 @@ int relocate_tree_blocks(struct btrfs_trans_handle *trans,
 	while (rb_node) {
 		block = rb_entry(rb_node, struct tree_block, rb_node);
 		if (!block->key_ready)
-			get_tree_block_key(rc, block);
-		rb_node = rb_next(rb_node);
-	}
-
-	rb_node = rb_first(blocks);
-	while (rb_node) {
-		block = rb_entry(rb_node, struct tree_block, rb_node);
-
-		node = build_backref_tree(rc, cache, &block->key,
-					  block->level, block->bytenr);
-		if (IS_ERR(node)) {
-			err = PTR_ERR(node);
-			goto out;
-		}
-
-		ret = relocate_tree_block(trans, rc, node, &block->key,
-					  path);
-		if (ret < 0) {
-			err = ret;
-			goto out;
-		}
-		remove_backref_node(cache, node);
-		rb_node = rb_next(rb_node);
-	}
-
-	if (level > 0)
-		goto out;
-
-	free_block_list(blocks);
-
-	/*
-	 * now backrefs of some upper level tree blocks have been cached,
-	 * try relocating blocks referenced by these upper level blocks.
-	 */
-	while (1) {
-		struct backref_node *upper = NULL;
-		if (trans->transaction->in_commit ||
-		    trans->transaction->delayed_refs.flushing)
-			break;
-
-		ret = add_adjacent_blocks(trans, rc, cache, blocks, level,
-					  &upper);
-		if (ret < 0)
-			err = ret;
-		if (ret != 0)
-			break;
+			get_tree_block_key(rc, block);
+		rb_node = rb_next(rb_node);
+	}
 
-		rb_node = rb_first(blocks);
-		while (rb_node) {
-			block = rb_entry(rb_node, struct tree_block, rb_node);
-			if (trans->transaction->in_commit ||
-			    trans->transaction->delayed_refs.flushing)
-				goto out;
-			BUG_ON(!block->key_ready);
-			node = build_backref_tree(rc, cache, &block->key,
-						  level, block->bytenr);
-			if (IS_ERR(node)) {
-				err = PTR_ERR(node);
-				goto out;
-			}
+	rb_node = rb_first(blocks);
+	while (rb_node) {
+		block = rb_entry(rb_node, struct tree_block, rb_node);
 
-			ret = relocate_tree_block(trans, rc, node,
-						  &block->key, path);
-			if (ret < 0) {
-				err = ret;
-				goto out;
-			}
-			remove_backref_node(cache, node);
-			rb_node = rb_next(rb_node);
+		node = build_backref_tree(rc, &block->key,
+					  block->level, block->bytenr);
+		if (IS_ERR(node)) {
+			err = PTR_ERR(node);
+			goto out;
 		}
-		free_block_list(blocks);
 
-		if (upper) {
-			ret = link_to_upper(trans, upper, path);
-			if (ret < 0) {
+		ret = relocate_tree_block(trans, rc, node, &block->key,
+					  path);
+		if (ret < 0) {
+			if (ret != -EAGAIN || rb_node == rb_first(blocks))
 				err = ret;
-				break;
-			}
-			remove_backref_node(cache, upper);
+			goto out;
 		}
+		rb_node = rb_next(rb_node);
 	}
 out:
 	free_block_list(blocks);
+	err = finish_pending_nodes(trans, rc, path, err);
 
-	ret = finish_pending_nodes(trans, cache, path);
-	if (ret < 0)
-		err = ret;
-
-	kfree(cache);
 	btrfs_free_path(path);
 	return err;
 }
 
+static noinline_for_stack
+int prealloc_file_extent_cluster(struct inode *inode,
+				 struct file_extent_cluster *cluster)
+{
+	u64 alloc_hint = 0;
+	u64 start;
+	u64 end;
+	u64 offset = BTRFS_I(inode)->index_cnt;
+	u64 num_bytes;
+	int nr = 0;
+	int ret = 0;
+
+	BUG_ON(cluster->start != cluster->boundary[0]);
+	mutex_lock(&inode->i_mutex);
+
+	ret = btrfs_check_data_free_space(inode, cluster->end +
+					  1 - cluster->start);
+	if (ret)
+		goto out;
+
+	while (nr < cluster->nr) {
+		start = cluster->boundary[nr] - offset;
+		if (nr + 1 < cluster->nr)
+			end = cluster->boundary[nr + 1] - 1 - offset;
+		else
+			end = cluster->end - offset;
+
+		lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+		num_bytes = end + 1 - start;
+		ret = btrfs_prealloc_file_range(inode, 0, start,
+						num_bytes, num_bytes,
+						end + 1, &alloc_hint);
+		unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+		if (ret)
+			break;
+		nr++;
+	}
+	btrfs_free_reserved_data_space(inode, cluster->end +
+				       1 - cluster->start);
+out:
+	mutex_unlock(&inode->i_mutex);
+	return ret;
+}
+
 static noinline_for_stack
 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
 			 u64 block_start)
@@ -2588,7 +2894,6 @@ static int relocate_file_extent_cluster(struct inode *inode,
 	u64 offset = BTRFS_I(inode)->index_cnt;
 	unsigned long index;
 	unsigned long last_index;
-	unsigned int dirty_page = 0;
 	struct page *page;
 	struct file_ra_state *ra;
 	int nr = 0;
@@ -2601,21 +2906,24 @@ static int relocate_file_extent_cluster(struct inode *inode,
 	if (!ra)
 		return -ENOMEM;
 
-	index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
-	last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
+	ret = prealloc_file_extent_cluster(inode, cluster);
+	if (ret)
+		goto out;
 
-	mutex_lock(&inode->i_mutex);
+	file_ra_state_init(ra, inode->i_mapping);
 
-	i_size_write(inode, cluster->end + 1 - offset);
 	ret = setup_extent_mapping(inode, cluster->start - offset,
 				   cluster->end - offset, cluster->start);
 	if (ret)
-		goto out_unlock;
-
-	file_ra_state_init(ra, inode->i_mapping);
+		goto out;
 
-	WARN_ON(cluster->start != cluster->boundary[0]);
+	index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
+	last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
 	while (index <= last_index) {
+		ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
+		if (ret)
+			goto out;
+
 		page = find_lock_page(inode->i_mapping, index);
 		if (!page) {
 			page_cache_sync_readahead(inode->i_mapping,
@@ -2623,8 +2931,10 @@ static int relocate_file_extent_cluster(struct inode *inode,
 						  last_index + 1 - index);
 			page = grab_cache_page(inode->i_mapping, index);
 			if (!page) {
+				btrfs_delalloc_release_metadata(inode,
+							PAGE_CACHE_SIZE);
 				ret = -ENOMEM;
-				goto out_unlock;
+				goto out;
 			}
 		}
 
@@ -2640,8 +2950,10 @@ static int relocate_file_extent_cluster(struct inode *inode,
 			if (!PageUptodate(page)) {
 				unlock_page(page);
 				page_cache_release(page);
+				btrfs_delalloc_release_metadata(inode,
+							PAGE_CACHE_SIZE);
 				ret = -EIO;
-				goto out_unlock;
+				goto out;
 			}
 		}
 
@@ -2660,10 +2972,9 @@ static int relocate_file_extent_cluster(struct inode *inode,
 					EXTENT_BOUNDARY, GFP_NOFS);
 			nr++;
 		}
-		btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
 
+		btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
 		set_page_dirty(page);
-		dirty_page++;
 
 		unlock_extent(&BTRFS_I(inode)->io_tree,
 			      page_start, page_end, GFP_NOFS);
@@ -2671,20 +2982,11 @@ static int relocate_file_extent_cluster(struct inode *inode,
 		page_cache_release(page);
 
 		index++;
-		if (nr < cluster->nr &&
-		    page_end + 1 + offset == cluster->boundary[nr]) {
-			balance_dirty_pages_ratelimited_nr(inode->i_mapping,
-							   dirty_page);
-			dirty_page = 0;
-		}
-	}
-	if (dirty_page) {
-		balance_dirty_pages_ratelimited_nr(inode->i_mapping,
-						   dirty_page);
+		balance_dirty_pages_ratelimited(inode->i_mapping);
+		btrfs_throttle(BTRFS_I(inode)->root);
 	}
 	WARN_ON(nr != cluster->nr);
-out_unlock:
-	mutex_unlock(&inode->i_mutex);
+out:
 	kfree(ra);
 	return ret;
 }
@@ -2870,9 +3172,6 @@ out:
 static int block_use_full_backref(struct reloc_control *rc,
 				  struct extent_buffer *eb)
 {
-	struct btrfs_path *path;
-	struct btrfs_extent_item *ei;
-	struct btrfs_key key;
 	u64 flags;
 	int ret;
 
@@ -2880,28 +3179,14 @@ static int block_use_full_backref(struct reloc_control *rc,
 	    btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
 		return 1;
 
-	path = btrfs_alloc_path();
-	BUG_ON(!path);
-
-	key.objectid = eb->start;
-	key.type = BTRFS_EXTENT_ITEM_KEY;
-	key.offset = eb->len;
-
-	path->search_commit_root = 1;
-	path->skip_locking = 1;
-	ret = btrfs_search_slot(NULL, rc->extent_root,
-				&key, path, 0, 0);
+	ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
+				       eb->start, eb->len, NULL, &flags);
 	BUG_ON(ret);
 
-	ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
-			    struct btrfs_extent_item);
-	flags = btrfs_extent_flags(path->nodes[0], ei);
-	BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
 	if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
 		ret = 1;
 	else
 		ret = 0;
-	btrfs_free_path(path);
 	return ret;
 }
 
@@ -3074,22 +3359,10 @@ int add_data_references(struct reloc_control *rc,
 	struct btrfs_extent_inline_ref *iref;
 	unsigned long ptr;
 	unsigned long end;
-	u32 blocksize;
+	u32 blocksize = btrfs_level_size(rc->extent_root, 0);
 	int ret;
 	int err = 0;
 
-	ret = get_new_location(rc->data_inode, NULL, extent_key->objectid,
-			       extent_key->offset);
-	BUG_ON(ret < 0);
-	if (ret > 0) {
-		/* the relocated data is fragmented */
-		rc->extents_skipped++;
-		btrfs_release_path(rc->extent_root, path);
-		return 0;
-	}
-
-	blocksize = btrfs_level_size(rc->extent_root, 0);
-
 	eb = path->nodes[0];
 	ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
 	end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
@@ -3170,7 +3443,8 @@ int add_data_references(struct reloc_control *rc,
  */
 static noinline_for_stack
 int find_next_extent(struct btrfs_trans_handle *trans,
-		     struct reloc_control *rc, struct btrfs_path *path)
+		     struct reloc_control *rc, struct btrfs_path *path,
+		     struct btrfs_key *extent_key)
 {
 	struct btrfs_key key;
 	struct extent_buffer *leaf;
@@ -3225,6 +3499,7 @@ next:
 			rc->search_start = end + 1;
 		} else {
 			rc->search_start = key.objectid + key.offset;
+			memcpy(extent_key, &key, sizeof(key));
 			return 0;
 		}
 	}
@@ -3262,12 +3537,49 @@ static int check_extent_flags(u64 flags)
 	return 0;
 }
 
+static noinline_for_stack
+int prepare_to_relocate(struct reloc_control *rc)
+{
+	struct btrfs_trans_handle *trans;
+	int ret;
+
+	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root);
+	if (!rc->block_rsv)
+		return -ENOMEM;
+
+	/*
+	 * reserve some space for creating reloc trees.
+	 * btrfs_init_reloc_root will use them when there
+	 * is no reservation in transaction handle.
+	 */
+	ret = btrfs_block_rsv_add(NULL, rc->extent_root, rc->block_rsv,
+				  rc->extent_root->nodesize * 256,
+				  &rc->block_rsv_retries);
+	if (ret)
+		return ret;
+
+	rc->block_rsv->refill_used = 1;
+	btrfs_add_durable_block_rsv(rc->extent_root->fs_info, rc->block_rsv);
+
+	memset(&rc->cluster, 0, sizeof(rc->cluster));
+	rc->search_start = rc->block_group->key.objectid;
+	rc->extents_found = 0;
+	rc->nodes_relocated = 0;
+	rc->merging_rsv_size = 0;
+	rc->block_rsv_retries = 0;
+
+	rc->create_reloc_tree = 1;
+	set_reloc_control(rc);
+
+	trans = btrfs_join_transaction(rc->extent_root, 1);
+	btrfs_commit_transaction(trans, rc->extent_root);
+	return 0;
+}
 
 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 {
 	struct rb_root blocks = RB_ROOT;
 	struct btrfs_key key;
-	struct file_extent_cluster *cluster;
 	struct btrfs_trans_handle *trans = NULL;
 	struct btrfs_path *path;
 	struct btrfs_extent_item *ei;
@@ -3277,33 +3589,25 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 	int ret;
 	int err = 0;
 
-	cluster = kzalloc(sizeof(*cluster), GFP_NOFS);
-	if (!cluster)
-		return -ENOMEM;
-
 	path = btrfs_alloc_path();
-	if (!path) {
-		kfree(cluster);
+	if (!path)
 		return -ENOMEM;
-	}
-
-	rc->extents_found = 0;
-	rc->extents_skipped = 0;
 
-	rc->search_start = rc->block_group->key.objectid;
-	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
-			  GFP_NOFS);
-
-	rc->create_reloc_root = 1;
-	set_reloc_control(rc);
-
-	trans = btrfs_start_transaction(rc->extent_root, 1);
-	btrfs_commit_transaction(trans, rc->extent_root);
+	ret = prepare_to_relocate(rc);
+	if (ret) {
+		err = ret;
+		goto out_free;
+	}
 
 	while (1) {
-		trans = btrfs_start_transaction(rc->extent_root, 1);
+		trans = btrfs_start_transaction(rc->extent_root, 0);
+
+		if (update_backref_cache(trans, &rc->backref_cache)) {
+			btrfs_end_transaction(trans, rc->extent_root);
+			continue;
+		}
 
-		ret = find_next_extent(trans, rc, path);
+		ret = find_next_extent(trans, rc, path, &key);
 		if (ret < 0)
 			err = ret;
 		if (ret != 0)
@@ -3313,9 +3617,7 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 
 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
 				    struct btrfs_extent_item);
-		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
-		item_size = btrfs_item_size_nr(path->nodes[0],
-					       path->slots[0]);
+		item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
 		if (item_size >= sizeof(*ei)) {
 			flags = btrfs_extent_flags(path->nodes[0], ei);
 			ret = check_extent_flags(flags);
@@ -3356,73 +3658,100 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 			ret = add_tree_block(rc, &key, path, &blocks);
 		} else if (rc->stage == UPDATE_DATA_PTRS &&
-			 (flags & BTRFS_EXTENT_FLAG_DATA)) {
+			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
 			ret = add_data_references(rc, &key, path, &blocks);
 		} else {
 			btrfs_release_path(rc->extent_root, path);
 			ret = 0;
 		}
 		if (ret < 0) {
-			err = 0;
+			err = ret;
 			break;
 		}
 
 		if (!RB_EMPTY_ROOT(&blocks)) {
 			ret = relocate_tree_blocks(trans, rc, &blocks);
 			if (ret < 0) {
+				if (ret != -EAGAIN) {
+					err = ret;
+					break;
+				}
+				rc->extents_found--;
+				rc->search_start = key.objectid;
+			}
+		}
+
+		ret = btrfs_block_rsv_check(trans, rc->extent_root,
+					    rc->block_rsv, 0, 5);
+		if (ret < 0) {
+			if (ret != -EAGAIN) {
 				err = ret;
+				WARN_ON(1);
 				break;
 			}
+			rc->commit_transaction = 1;
 		}
 
-		nr = trans->blocks_used;
-		btrfs_end_transaction(trans, rc->extent_root);
+		if (rc->commit_transaction) {
+			rc->commit_transaction = 0;
+			ret = btrfs_commit_transaction(trans, rc->extent_root);
+			BUG_ON(ret);
+		} else {
+			nr = trans->blocks_used;
+			btrfs_end_transaction_throttle(trans, rc->extent_root);
+			btrfs_btree_balance_dirty(rc->extent_root, nr);
+		}
 		trans = NULL;
-		btrfs_btree_balance_dirty(rc->extent_root, nr);
 
 		if (rc->stage == MOVE_DATA_EXTENTS &&
 		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
 			rc->found_file_extent = 1;
 			ret = relocate_data_extent(rc->data_inode,
-						   &key, cluster);
+						   &key, &rc->cluster);
 			if (ret < 0) {
 				err = ret;
 				break;
 			}
 		}
 	}
-	btrfs_free_path(path);
+
+	btrfs_release_path(rc->extent_root, path);
+	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
+			  GFP_NOFS);
 
 	if (trans) {
 		nr = trans->blocks_used;
-		btrfs_end_transaction(trans, rc->extent_root);
+		btrfs_end_transaction_throttle(trans, rc->extent_root);
 		btrfs_btree_balance_dirty(rc->extent_root, nr);
 	}
 
 	if (!err) {
-		ret = relocate_file_extent_cluster(rc->data_inode, cluster);
+		ret = relocate_file_extent_cluster(rc->data_inode,
+						   &rc->cluster);
 		if (ret < 0)
 			err = ret;
 	}
 
-	kfree(cluster);
+	rc->create_reloc_tree = 0;
+	set_reloc_control(rc);
 
-	rc->create_reloc_root = 0;
-	smp_mb();
+	backref_cache_cleanup(&rc->backref_cache);
+	btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
 
-	if (rc->extents_found > 0) {
-		trans = btrfs_start_transaction(rc->extent_root, 1);
-		btrfs_commit_transaction(trans, rc->extent_root);
-	}
+	err = prepare_to_merge(rc, err);
 
 	merge_reloc_roots(rc);
 
+	rc->merge_reloc_tree = 0;
 	unset_reloc_control(rc);
+	btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
 
 	/* get rid of pinned extents */
-	trans = btrfs_start_transaction(rc->extent_root, 1);
+	trans = btrfs_join_transaction(rc->extent_root, 1);
 	btrfs_commit_transaction(trans, rc->extent_root);
-
+out_free:
+	btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
+	btrfs_free_path(path);
 	return err;
 }
 
@@ -3448,7 +3777,8 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
 	btrfs_set_inode_generation(leaf, item, 1);
 	btrfs_set_inode_size(leaf, item, 0);
 	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
-	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
+	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
+					  BTRFS_INODE_PREALLOC);
 	btrfs_mark_buffer_dirty(leaf);
 	btrfs_release_path(root, path);
 out:
@@ -3460,8 +3790,9 @@ out:
  * helper to create inode for data relocation.
  * the inode is in data relocation tree and its link count is 0
  */
-static struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
-					struct btrfs_block_group_cache *group)
+static noinline_for_stack
+struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
+				 struct btrfs_block_group_cache *group)
 {
 	struct inode *inode = NULL;
 	struct btrfs_trans_handle *trans;
@@ -3475,8 +3806,9 @@ static struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
 	if (IS_ERR(root))
 		return ERR_CAST(root);
 
-	trans = btrfs_start_transaction(root, 1);
-	BUG_ON(!trans);
+	trans = btrfs_start_transaction(root, 6);
+	if (IS_ERR(trans))
+		return ERR_CAST(trans);
 
 	err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
 	if (err)
@@ -3496,7 +3828,6 @@ static struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
 out:
 	nr = trans->blocks_used;
 	btrfs_end_transaction(trans, root);
-
 	btrfs_btree_balance_dirty(root, nr);
 	if (err) {
 		if (inode)
@@ -3506,6 +3837,21 @@ out:
 	return inode;
 }
 
+static struct reloc_control *alloc_reloc_control(void)
+{
+	struct reloc_control *rc;
+
+	rc = kzalloc(sizeof(*rc), GFP_NOFS);
+	if (!rc)
+		return NULL;
+
+	INIT_LIST_HEAD(&rc->reloc_roots);
+	backref_cache_init(&rc->backref_cache);
+	mapping_tree_init(&rc->reloc_root_tree);
+	extent_io_tree_init(&rc->processed_blocks, NULL, GFP_NOFS);
+	return rc;
+}
+
 /*
  * function to relocate all extents in a block group.
  */
@@ -3514,24 +3860,26 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 	struct btrfs_fs_info *fs_info = extent_root->fs_info;
 	struct reloc_control *rc;
 	int ret;
+	int rw = 0;
 	int err = 0;
 
-	rc = kzalloc(sizeof(*rc), GFP_NOFS);
+	rc = alloc_reloc_control();
 	if (!rc)
 		return -ENOMEM;
 
-	mapping_tree_init(&rc->reloc_root_tree);
-	extent_io_tree_init(&rc->processed_blocks, NULL, GFP_NOFS);
-	INIT_LIST_HEAD(&rc->reloc_roots);
+	rc->extent_root = extent_root;
 
 	rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
 	BUG_ON(!rc->block_group);
 
-	btrfs_init_workers(&rc->workers, "relocate",
-			   fs_info->thread_pool_size, NULL);
-
-	rc->extent_root = extent_root;
-	btrfs_prepare_block_group_relocation(extent_root, rc->block_group);
+	if (!rc->block_group->ro) {
+		ret = btrfs_set_block_group_ro(extent_root, rc->block_group);
+		if (ret) {
+			err = ret;
+			goto out;
+		}
+		rw = 1;
+	}
 
 	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
 	if (IS_ERR(rc->data_inode)) {
@@ -3548,9 +3896,6 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 	btrfs_wait_ordered_extents(fs_info->tree_root, 0, 0);
 
 	while (1) {
-		rc->extents_found = 0;
-		rc->extents_skipped = 0;
-
 		mutex_lock(&fs_info->cleaner_mutex);
 
 		btrfs_clean_old_snapshots(fs_info->tree_root);
@@ -3559,7 +3904,7 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 		mutex_unlock(&fs_info->cleaner_mutex);
 		if (ret < 0) {
 			err = ret;
-			break;
+			goto out;
 		}
 
 		if (rc->extents_found == 0)
@@ -3573,18 +3918,6 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 			invalidate_mapping_pages(rc->data_inode->i_mapping,
 						 0, -1);
 			rc->stage = UPDATE_DATA_PTRS;
-		} else if (rc->stage == UPDATE_DATA_PTRS &&
-			   rc->extents_skipped >= rc->extents_found) {
-			iput(rc->data_inode);
-			rc->data_inode = create_reloc_inode(fs_info,
-							    rc->block_group);
-			if (IS_ERR(rc->data_inode)) {
-				err = PTR_ERR(rc->data_inode);
-				rc->data_inode = NULL;
-				break;
-			}
-			rc->stage = MOVE_DATA_EXTENTS;
-			rc->found_file_extent = 0;
 		}
 	}
 
@@ -3597,8 +3930,9 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 	WARN_ON(rc->block_group->reserved > 0);
 	WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
 out:
+	if (err && rw)
+		btrfs_set_block_group_rw(extent_root, rc->block_group);
 	iput(rc->data_inode);
-	btrfs_stop_workers(&rc->workers);
 	btrfs_put_block_group(rc->block_group);
 	kfree(rc);
 	return err;
@@ -3609,7 +3943,7 @@ static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
 	struct btrfs_trans_handle *trans;
 	int ret;
 
-	trans = btrfs_start_transaction(root->fs_info->tree_root, 1);
+	trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
 
 	memset(&root->root_item.drop_progress, 0,
 		sizeof(root->root_item.drop_progress));
@@ -3702,20 +4036,20 @@ int btrfs_recover_relocation(struct btrfs_root *root)
 	if (list_empty(&reloc_roots))
 		goto out;
 
-	rc = kzalloc(sizeof(*rc), GFP_NOFS);
+	rc = alloc_reloc_control();
 	if (!rc) {
 		err = -ENOMEM;
 		goto out;
 	}
 
-	mapping_tree_init(&rc->reloc_root_tree);
-	INIT_LIST_HEAD(&rc->reloc_roots);
-	btrfs_init_workers(&rc->workers, "relocate",
-			   root->fs_info->thread_pool_size, NULL);
 	rc->extent_root = root->fs_info->extent_root;
 
 	set_reloc_control(rc);
 
+	trans = btrfs_join_transaction(rc->extent_root, 1);
+
+	rc->merge_reloc_tree = 1;
+
 	while (!list_empty(&reloc_roots)) {
 		reloc_root = list_entry(reloc_roots.next,
 					struct btrfs_root, root_list);
@@ -3735,20 +4069,16 @@ int btrfs_recover_relocation(struct btrfs_root *root)
 		fs_root->reloc_root = reloc_root;
 	}
 
-	trans = btrfs_start_transaction(rc->extent_root, 1);
 	btrfs_commit_transaction(trans, rc->extent_root);
 
 	merge_reloc_roots(rc);
 
 	unset_reloc_control(rc);
 
-	trans = btrfs_start_transaction(rc->extent_root, 1);
+	trans = btrfs_join_transaction(rc->extent_root, 1);
 	btrfs_commit_transaction(trans, rc->extent_root);
 out:
-	if (rc) {
-		btrfs_stop_workers(&rc->workers);
-		kfree(rc);
-	}
+	kfree(rc);
 	while (!list_empty(&reloc_roots)) {
 		reloc_root = list_entry(reloc_roots.next,
 					struct btrfs_root, root_list);
@@ -3814,3 +4144,130 @@ int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
 	btrfs_put_ordered_extent(ordered);
 	return 0;
 }
+
+void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, struct extent_buffer *buf,
+			   struct extent_buffer *cow)
+{
+	struct reloc_control *rc;
+	struct backref_node *node;
+	int first_cow = 0;
+	int level;
+	int ret;
+
+	rc = root->fs_info->reloc_ctl;
+	if (!rc)
+		return;
+
+	BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
+	       root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
+
+	level = btrfs_header_level(buf);
+	if (btrfs_header_generation(buf) <=
+	    btrfs_root_last_snapshot(&root->root_item))
+		first_cow = 1;
+
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
+	    rc->create_reloc_tree) {
+		WARN_ON(!first_cow && level == 0);
+
+		node = rc->backref_cache.path[level];
+		BUG_ON(node->bytenr != buf->start &&
+		       node->new_bytenr != buf->start);
+
+		drop_node_buffer(node);
+		extent_buffer_get(cow);
+		node->eb = cow;
+		node->new_bytenr = cow->start;
+
+		if (!node->pending) {
+			list_move_tail(&node->list,
+				       &rc->backref_cache.pending[level]);
+			node->pending = 1;
+		}
+
+		if (first_cow)
+			__mark_block_processed(rc, node);
+
+		if (first_cow && level > 0)
+			rc->nodes_relocated += buf->len;
+	}
+
+	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) {
+		ret = replace_file_extents(trans, rc, root, cow);
+		BUG_ON(ret);
+	}
+}
+
+/*
+ * called before creating snapshot. it calculates metadata reservation
+ * requried for relocating tree blocks in the snapshot
+ */
+void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
+			      struct btrfs_pending_snapshot *pending,
+			      u64 *bytes_to_reserve)
+{
+	struct btrfs_root *root;
+	struct reloc_control *rc;
+
+	root = pending->root;
+	if (!root->reloc_root)
+		return;
+
+	rc = root->fs_info->reloc_ctl;
+	if (!rc->merge_reloc_tree)
+		return;
+
+	root = root->reloc_root;
+	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
+	/*
+	 * relocation is in the stage of merging trees. the space
+	 * used by merging a reloc tree is twice the size of
+	 * relocated tree nodes in the worst case. half for cowing
+	 * the reloc tree, half for cowing the fs tree. the space
+	 * used by cowing the reloc tree will be freed after the
+	 * tree is dropped. if we create snapshot, cowing the fs
+	 * tree may use more space than it frees. so we need
+	 * reserve extra space.
+	 */
+	*bytes_to_reserve += rc->nodes_relocated;
+}
+
+/*
+ * called after snapshot is created. migrate block reservation
+ * and create reloc root for the newly created snapshot
+ */
+void btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
+			       struct btrfs_pending_snapshot *pending)
+{
+	struct btrfs_root *root = pending->root;
+	struct btrfs_root *reloc_root;
+	struct btrfs_root *new_root;
+	struct reloc_control *rc;
+	int ret;
+
+	if (!root->reloc_root)
+		return;
+
+	rc = root->fs_info->reloc_ctl;
+	rc->merging_rsv_size += rc->nodes_relocated;
+
+	if (rc->merge_reloc_tree) {
+		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
+					      rc->block_rsv,
+					      rc->nodes_relocated);
+		BUG_ON(ret);
+	}
+
+	new_root = pending->snap;
+	reloc_root = create_reloc_root(trans, root->reloc_root,
+				       new_root->root_key.objectid);
+
+	__add_reloc_root(reloc_root);
+	new_root->reloc_root = reloc_root;
+
+	if (rc->create_reloc_tree) {
+		ret = clone_backref_node(trans, rc, root, reloc_root);
+		BUG_ON(ret);
+	}
+}

fs/btrfs/root-tree.c

@@ -259,6 +259,8 @@ int btrfs_find_orphan_roots(struct btrfs_root *tree_root)
 	struct extent_buffer *leaf;
 	struct btrfs_path *path;
 	struct btrfs_key key;
+	struct btrfs_key root_key;
+	struct btrfs_root *root;
 	int err = 0;
 	int ret;
 
@@ -270,6 +272,9 @@ int btrfs_find_orphan_roots(struct btrfs_root *tree_root)
 	key.type = BTRFS_ORPHAN_ITEM_KEY;
 	key.offset = 0;
 
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+	root_key.offset = (u64)-1;
+
 	while (1) {
 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
 		if (ret < 0) {
@@ -294,13 +299,25 @@ int btrfs_find_orphan_roots(struct btrfs_root *tree_root)
 		    key.type != BTRFS_ORPHAN_ITEM_KEY)
 			break;
 
-		ret = btrfs_find_dead_roots(tree_root, key.offset);
-		if (ret) {
+		root_key.objectid = key.offset;
+		key.offset++;
+
+		root = btrfs_read_fs_root_no_name(tree_root->fs_info,
+						  &root_key);
+		if (!IS_ERR(root))
+			continue;
+
+		ret = PTR_ERR(root);
+		if (ret != -ENOENT) {
 			err = ret;
 			break;
 		}
 
-		key.offset++;
+		ret = btrfs_find_dead_roots(tree_root, root_key.objectid);
+		if (ret) {
+			err = ret;
+			break;
+		}
 	}
 
 	btrfs_free_path(path);

fs/btrfs/super.c

@@ -498,7 +498,7 @@ int btrfs_sync_fs(struct super_block *sb, int wait)
 	btrfs_start_delalloc_inodes(root, 0);
 	btrfs_wait_ordered_extents(root, 0, 0);
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_start_transaction(root, 0);
 	ret = btrfs_commit_transaction(trans, root);
 	return ret;
 }
@@ -694,11 +694,11 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 		if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
 			return -EINVAL;
 
-		/* recover relocation */
-		ret = btrfs_recover_relocation(root);
+		ret = btrfs_cleanup_fs_roots(root->fs_info);
 		WARN_ON(ret);
 
-		ret = btrfs_cleanup_fs_roots(root->fs_info);
+		/* recover relocation */
+		ret = btrfs_recover_relocation(root);
 		WARN_ON(ret);
 
 		sb->s_flags &= ~MS_RDONLY;
@@ -714,34 +714,18 @@ static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 	struct list_head *head = &root->fs_info->space_info;
 	struct btrfs_space_info *found;
 	u64 total_used = 0;
-	u64 data_used = 0;
 	int bits = dentry->d_sb->s_blocksize_bits;
 	__be32 *fsid = (__be32 *)root->fs_info->fsid;
 
 	rcu_read_lock();
-	list_for_each_entry_rcu(found, head, list) {
-		if (found->flags & (BTRFS_BLOCK_GROUP_DUP|
-				    BTRFS_BLOCK_GROUP_RAID10|
-				    BTRFS_BLOCK_GROUP_RAID1)) {
-			total_used += found->bytes_used;
-			if (found->flags & BTRFS_BLOCK_GROUP_DATA)
-				data_used += found->bytes_used;
-			else
-				data_used += found->total_bytes;
-		}
-
-		total_used += found->bytes_used;
-		if (found->flags & BTRFS_BLOCK_GROUP_DATA)
-			data_used += found->bytes_used;
-		else
-			data_used += found->total_bytes;
-	}
+	list_for_each_entry_rcu(found, head, list)
+		total_used += found->disk_used;
 	rcu_read_unlock();
 
 	buf->f_namelen = BTRFS_NAME_LEN;
 	buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
 	buf->f_bfree = buf->f_blocks - (total_used >> bits);
-	buf->f_bavail = buf->f_blocks - (data_used >> bits);
+	buf->f_bavail = buf->f_bfree;
 	buf->f_bsize = dentry->d_sb->s_blocksize;
 	buf->f_type = BTRFS_SUPER_MAGIC;
 

fs/btrfs/transaction.c

@@ -165,54 +165,89 @@ enum btrfs_trans_type {
 	TRANS_USERSPACE,
 };
 
+static int may_wait_transaction(struct btrfs_root *root, int type)
+{
+	if (!root->fs_info->log_root_recovering &&
+	    ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
+	     type == TRANS_USERSPACE))
+		return 1;
+	return 0;
+}
+
 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
-					     int num_blocks, int type)
+						    u64 num_items, int type)
 {
-	struct btrfs_trans_handle *h =
-		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
+	struct btrfs_trans_handle *h;
+	struct btrfs_transaction *cur_trans;
+	int retries = 0;
 	int ret;
+again:
+	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
+	if (!h)
+		return ERR_PTR(-ENOMEM);
 
 	mutex_lock(&root->fs_info->trans_mutex);
-	if (!root->fs_info->log_root_recovering &&
-	    ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
-	     type == TRANS_USERSPACE))
+	if (may_wait_transaction(root, type))
 		wait_current_trans(root);
+
 	ret = join_transaction(root);
 	BUG_ON(ret);
 
-	h->transid = root->fs_info->running_transaction->transid;
-	h->transaction = root->fs_info->running_transaction;
-	h->blocks_reserved = num_blocks;
+	cur_trans = root->fs_info->running_transaction;
+	cur_trans->use_count++;
+	mutex_unlock(&root->fs_info->trans_mutex);
+
+	h->transid = cur_trans->transid;
+	h->transaction = cur_trans;
 	h->blocks_used = 0;
 	h->block_group = 0;
-	h->alloc_exclude_nr = 0;
-	h->alloc_exclude_start = 0;
+	h->bytes_reserved = 0;
 	h->delayed_ref_updates = 0;
+	h->block_rsv = NULL;
 
-	if (!current->journal_info && type != TRANS_USERSPACE)
-		current->journal_info = h;
+	smp_mb();
+	if (cur_trans->blocked && may_wait_transaction(root, type)) {
+		btrfs_commit_transaction(h, root);
+		goto again;
+	}
+
+	if (num_items > 0) {
+		ret = btrfs_trans_reserve_metadata(h, root, num_items,
+						   &retries);
+		if (ret == -EAGAIN) {
+			btrfs_commit_transaction(h, root);
+			goto again;
+		}
+		if (ret < 0) {
+			btrfs_end_transaction(h, root);
+			return ERR_PTR(ret);
+		}
+	}
 
-	root->fs_info->running_transaction->use_count++;
+	mutex_lock(&root->fs_info->trans_mutex);
 	record_root_in_trans(h, root);
 	mutex_unlock(&root->fs_info->trans_mutex);
+
+	if (!current->journal_info && type != TRANS_USERSPACE)
+		current->journal_info = h;
 	return h;
 }
 
 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
-						   int num_blocks)
+						   int num_items)
 {
-	return start_transaction(root, num_blocks, TRANS_START);
+	return start_transaction(root, num_items, TRANS_START);
 }
 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 						   int num_blocks)
 {
-	return start_transaction(root, num_blocks, TRANS_JOIN);
+	return start_transaction(root, 0, TRANS_JOIN);
 }
 
 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 							 int num_blocks)
 {
-	return start_transaction(r, num_blocks, TRANS_USERSPACE);
+	return start_transaction(r, 0, TRANS_USERSPACE);
 }
 
 /* wait for a transaction commit to be fully complete */
@@ -286,10 +321,36 @@ void btrfs_throttle(struct btrfs_root *root)
 	mutex_unlock(&root->fs_info->trans_mutex);
 }
 
+static int should_end_transaction(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root)
+{
+	int ret;
+	ret = btrfs_block_rsv_check(trans, root,
+				    &root->fs_info->global_block_rsv, 0, 5);
+	return ret ? 1 : 0;
+}
+
+int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root)
+{
+	struct btrfs_transaction *cur_trans = trans->transaction;
+	int updates;
+
+	if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
+		return 1;
+
+	updates = trans->delayed_ref_updates;
+	trans->delayed_ref_updates = 0;
+	if (updates)
+		btrfs_run_delayed_refs(trans, root, updates);
+
+	return should_end_transaction(trans, root);
+}
+
 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root, int throttle)
 {
-	struct btrfs_transaction *cur_trans;
+	struct btrfs_transaction *cur_trans = trans->transaction;
 	struct btrfs_fs_info *info = root->fs_info;
 	int count = 0;
 
@@ -313,9 +374,21 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 		count++;
 	}
 
+	btrfs_trans_release_metadata(trans, root);
+
+	if (!root->fs_info->open_ioctl_trans &&
+	    should_end_transaction(trans, root))
+		trans->transaction->blocked = 1;
+
+	if (cur_trans->blocked && !cur_trans->in_commit) {
+		if (throttle)
+			return btrfs_commit_transaction(trans, root);
+		else
+			wake_up_process(info->transaction_kthread);
+	}
+
 	mutex_lock(&info->trans_mutex);
-	cur_trans = info->running_transaction;
-	WARN_ON(cur_trans != trans->transaction);
+	WARN_ON(cur_trans != info->running_transaction);
 	WARN_ON(cur_trans->num_writers < 1);
 	cur_trans->num_writers--;
 
@@ -603,6 +676,7 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
 
 			btrfs_free_log(trans, root);
 			btrfs_update_reloc_root(trans, root);
+			btrfs_orphan_commit_root(trans, root);
 
 			if (root->commit_root != root->node) {
 				switch_commit_root(root);
@@ -627,30 +701,30 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 {
 	struct btrfs_fs_info *info = root->fs_info;
-	int ret;
 	struct btrfs_trans_handle *trans;
+	int ret;
 	unsigned long nr;
 
-	smp_mb();
-	if (root->defrag_running)
+	if (xchg(&root->defrag_running, 1))
 		return 0;
-	trans = btrfs_start_transaction(root, 1);
+
 	while (1) {
-		root->defrag_running = 1;
+		trans = btrfs_start_transaction(root, 0);
+		if (IS_ERR(trans))
+			return PTR_ERR(trans);
+
 		ret = btrfs_defrag_leaves(trans, root, cacheonly);
+
 		nr = trans->blocks_used;
 		btrfs_end_transaction(trans, root);
 		btrfs_btree_balance_dirty(info->tree_root, nr);
 		cond_resched();
 
-		trans = btrfs_start_transaction(root, 1);
 		if (root->fs_info->closing || ret != -EAGAIN)
 			break;
 	}
 	root->defrag_running = 0;
-	smp_mb();
-	btrfs_end_transaction(trans, root);
-	return 0;
+	return ret;
 }
 
 #if 0
@@ -758,47 +832,63 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	struct btrfs_root *root = pending->root;
 	struct btrfs_root *parent_root;
 	struct inode *parent_inode;
+	struct dentry *dentry;
 	struct extent_buffer *tmp;
 	struct extent_buffer *old;
 	int ret;
-	u64 objectid;
-	int namelen;
+	int retries = 0;
+	u64 to_reserve = 0;
 	u64 index = 0;
-
-	parent_inode = pending->dentry->d_parent->d_inode;
-	parent_root = BTRFS_I(parent_inode)->root;
+	u64 objectid;
 
 	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 	if (!new_root_item) {
-		ret = -ENOMEM;
+		pending->error = -ENOMEM;
 		goto fail;
 	}
+
 	ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
-	if (ret)
+	if (ret) {
+		pending->error = ret;
 		goto fail;
+	}
+
+	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
+	btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
+
+	if (to_reserve > 0) {
+		ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
+					  to_reserve, &retries);
+		if (ret) {
+			pending->error = ret;
+			goto fail;
+		}
+	}
 
 	key.objectid = objectid;
-	/* record when the snapshot was created in key.offset */
-	key.offset = trans->transid;
-	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+	key.offset = (u64)-1;
+	key.type = BTRFS_ROOT_ITEM_KEY;
 
-	memcpy(&pending->root_key, &key, sizeof(key));
-	pending->root_key.offset = (u64)-1;
+	trans->block_rsv = &pending->block_rsv;
 
+	dentry = pending->dentry;
+	parent_inode = dentry->d_parent->d_inode;
+	parent_root = BTRFS_I(parent_inode)->root;
 	record_root_in_trans(trans, parent_root);
+
 	/*
 	 * insert the directory item
 	 */
-	namelen = strlen(pending->name);
 	ret = btrfs_set_inode_index(parent_inode, &index);
 	BUG_ON(ret);
 	ret = btrfs_insert_dir_item(trans, parent_root,
-			    pending->name, namelen,
-			    parent_inode->i_ino,
-			    &pending->root_key, BTRFS_FT_DIR, index);
+				dentry->d_name.name, dentry->d_name.len,
+				parent_inode->i_ino, &key,
+				BTRFS_FT_DIR, index);
 	BUG_ON(ret);
 
-	btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
+	btrfs_i_size_write(parent_inode, parent_inode->i_size +
+					 dentry->d_name.len * 2);
 	ret = btrfs_update_inode(trans, parent_root, parent_inode);
 	BUG_ON(ret);
 
@@ -815,22 +905,32 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	free_extent_buffer(old);
 
 	btrfs_set_root_node(new_root_item, tmp);
-	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
-				new_root_item);
-	BUG_ON(ret);
+	/* record when the snapshot was created in key.offset */
+	key.offset = trans->transid;
+	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
 	btrfs_tree_unlock(tmp);
 	free_extent_buffer(tmp);
+	BUG_ON(ret);
 
-	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
-				 pending->root_key.objectid,
+	/*
+	 * insert root back/forward references
+	 */
+	ret = btrfs_add_root_ref(trans, tree_root, objectid,
 				 parent_root->root_key.objectid,
-				 parent_inode->i_ino, index, pending->name,
-				 namelen);
+				 parent_inode->i_ino, index,
+				 dentry->d_name.name, dentry->d_name.len);
 	BUG_ON(ret);
 
+	key.offset = (u64)-1;
+	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
+	BUG_ON(IS_ERR(pending->snap));
+
+	btrfs_reloc_post_snapshot(trans, pending);
+	btrfs_orphan_post_snapshot(trans, pending);
 fail:
 	kfree(new_root_item);
-	return ret;
+	btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
+	return 0;
 }
 
 /*
@@ -878,6 +978,16 @@ int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
 	return ret;
 }
 
+int btrfs_transaction_blocked(struct btrfs_fs_info *info)
+{
+	int ret = 0;
+	spin_lock(&info->new_trans_lock);
+	if (info->running_transaction)
+		ret = info->running_transaction->blocked;
+	spin_unlock(&info->new_trans_lock);
+	return ret;
+}
+
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root)
 {
@@ -899,6 +1009,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	ret = btrfs_run_delayed_refs(trans, root, 0);
 	BUG_ON(ret);
 
+	btrfs_trans_release_metadata(trans, root);
+
 	cur_trans = trans->transaction;
 	/*
 	 * set the flushing flag so procs in this transaction have to
@@ -951,9 +1063,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			snap_pending = 1;
 
 		WARN_ON(cur_trans != trans->transaction);
-		prepare_to_wait(&cur_trans->writer_wait, &wait,
-				TASK_UNINTERRUPTIBLE);
-
 		if (cur_trans->num_writers > 1)
 			timeout = MAX_SCHEDULE_TIMEOUT;
 		else if (should_grow)
@@ -976,6 +1085,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 		 */
 		btrfs_run_ordered_operations(root, 1);
 
+		prepare_to_wait(&cur_trans->writer_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+
 		smp_mb();
 		if (cur_trans->num_writers > 1 || should_grow)
 			schedule_timeout(timeout);
@@ -1103,9 +1215,9 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
 
 		if (btrfs_header_backref_rev(root->node) <
 		    BTRFS_MIXED_BACKREF_REV)
-			btrfs_drop_snapshot(root, 0);
+			btrfs_drop_snapshot(root, NULL, 0);
 		else
-			btrfs_drop_snapshot(root, 1);
+			btrfs_drop_snapshot(root, NULL, 1);
 	}
 	return 0;
 }

fs/btrfs/transaction.h

@@ -45,20 +45,23 @@ struct btrfs_transaction {
 
 struct btrfs_trans_handle {
 	u64 transid;
+	u64 block_group;
+	u64 bytes_reserved;
 	unsigned long blocks_reserved;
 	unsigned long blocks_used;
-	struct btrfs_transaction *transaction;
-	u64 block_group;
-	u64 alloc_exclude_start;
-	u64 alloc_exclude_nr;
 	unsigned long delayed_ref_updates;
+	struct btrfs_transaction *transaction;
+	struct btrfs_block_rsv *block_rsv;
 };
 
 struct btrfs_pending_snapshot {
 	struct dentry *dentry;
 	struct btrfs_root *root;
-	char *name;
-	struct btrfs_key root_key;
+	struct btrfs_root *snap;
+	/* block reservation for the operation */
+	struct btrfs_block_rsv block_rsv;
+	/* extra metadata reseration for relocation */
+	int error;
 	struct list_head list;
 };
 
@@ -85,11 +88,11 @@ static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root);
 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
-						   int num_blocks);
+						   int num_items);
 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
-						   int num_blocks);
+						  int num_blocks);
 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
-						   int num_blocks);
+							 int num_blocks);
 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 				     struct btrfs_root *root);
 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
@@ -103,6 +106,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root);
 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 				   struct btrfs_root *root);
+int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root);
 void btrfs_throttle(struct btrfs_root *root);
 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root);
@@ -112,5 +117,6 @@ int btrfs_write_marked_extents(struct btrfs_root *root,
 				struct extent_io_tree *dirty_pages, int mark);
 int btrfs_wait_marked_extents(struct btrfs_root *root,
 				struct extent_io_tree *dirty_pages, int mark);
+int btrfs_transaction_blocked(struct btrfs_fs_info *info);
 int btrfs_transaction_in_commit(struct btrfs_fs_info *info);
 #endif

fs/btrfs/tree-defrag.c

@@ -117,13 +117,14 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 				 path->nodes[1], 0,
 				 cache_only, &last_ret,
 				 &root->defrag_progress);
-	WARN_ON(ret && ret != -EAGAIN);
+	if (ret) {
+		WARN_ON(ret == -EAGAIN);
+		goto out;
+	}
 	if (next_key_ret == 0) {
 		memcpy(&root->defrag_progress, &key, sizeof(key));
 		ret = -EAGAIN;
 	}
-
-	btrfs_release_path(root, path);
 out:
 	if (path)
 		btrfs_free_path(path);

fs/btrfs/tree-log.c

@@ -135,6 +135,7 @@ static int start_log_trans(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root)
 {
 	int ret;
+	int err = 0;
 
 	mutex_lock(&root->log_mutex);
 	if (root->log_root) {
@@ -155,17 +156,19 @@ static int start_log_trans(struct btrfs_trans_handle *trans,
 	mutex_lock(&root->fs_info->tree_log_mutex);
 	if (!root->fs_info->log_root_tree) {
 		ret = btrfs_init_log_root_tree(trans, root->fs_info);
-		BUG_ON(ret);
+		if (ret)
+			err = ret;
 	}
-	if (!root->log_root) {
+	if (err == 0 && !root->log_root) {
 		ret = btrfs_add_log_tree(trans, root);
-		BUG_ON(ret);
+		if (ret)
+			err = ret;
 	}
 	mutex_unlock(&root->fs_info->tree_log_mutex);
 	root->log_batch++;
 	atomic_inc(&root->log_writers);
 	mutex_unlock(&root->log_mutex);
-	return 0;
+	return err;
 }
 
 /*
@@ -376,7 +379,7 @@ insert:
 			BUG_ON(ret);
 		}
 	} else if (ret) {
-		BUG();
+		return ret;
 	}
 	dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
 					path->slots[0]);
@@ -1699,9 +1702,9 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
 
 		next = btrfs_find_create_tree_block(root, bytenr, blocksize);
 
-		wc->process_func(root, next, wc, ptr_gen);
-
 		if (*level == 1) {
+			wc->process_func(root, next, wc, ptr_gen);
+
 			path->slots[*level]++;
 			if (wc->free) {
 				btrfs_read_buffer(next, ptr_gen);
@@ -1734,35 +1737,7 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
 	WARN_ON(*level < 0);
 	WARN_ON(*level >= BTRFS_MAX_LEVEL);
 
-	if (path->nodes[*level] == root->node)
-		parent = path->nodes[*level];
-	else
-		parent = path->nodes[*level + 1];
-
-	bytenr = path->nodes[*level]->start;
-
-	blocksize = btrfs_level_size(root, *level);
-	root_owner = btrfs_header_owner(parent);
-	root_gen = btrfs_header_generation(parent);
-
-	wc->process_func(root, path->nodes[*level], wc,
-			 btrfs_header_generation(path->nodes[*level]));
-
-	if (wc->free) {
-		next = path->nodes[*level];
-		btrfs_tree_lock(next);
-		clean_tree_block(trans, root, next);
-		btrfs_set_lock_blocking(next);
-		btrfs_wait_tree_block_writeback(next);
-		btrfs_tree_unlock(next);
-
-		WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
-		ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
-		BUG_ON(ret);
-	}
-	free_extent_buffer(path->nodes[*level]);
-	path->nodes[*level] = NULL;
-	*level += 1;
+	path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
 
 	cond_resched();
 	return 0;
@@ -1781,7 +1756,7 @@ static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
 
 	for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
 		slot = path->slots[i];
-		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
+		if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
 			struct extent_buffer *node;
 			node = path->nodes[i];
 			path->slots[i]++;
@@ -2047,7 +2022,6 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	mutex_unlock(&log_root_tree->log_mutex);
 
 	ret = update_log_root(trans, log);
-	BUG_ON(ret);
 
 	mutex_lock(&log_root_tree->log_mutex);
 	if (atomic_dec_and_test(&log_root_tree->log_writers)) {
@@ -2056,6 +2030,15 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 			wake_up(&log_root_tree->log_writer_wait);
 	}
 
+	if (ret) {
+		BUG_ON(ret != -ENOSPC);
+		root->fs_info->last_trans_log_full_commit = trans->transid;
+		btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
+		mutex_unlock(&log_root_tree->log_mutex);
+		ret = -EAGAIN;
+		goto out;
+	}
+
 	index2 = log_root_tree->log_transid % 2;
 	if (atomic_read(&log_root_tree->log_commit[index2])) {
 		btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
@@ -2129,15 +2112,10 @@ out:
 	return 0;
 }
 
-/*
- * free all the extents used by the tree log.  This should be called
- * at commit time of the full transaction
- */
-int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
+static void free_log_tree(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *log)
 {
 	int ret;
-	struct btrfs_root *log;
-	struct key;
 	u64 start;
 	u64 end;
 	struct walk_control wc = {
@@ -2145,10 +2123,6 @@ int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
 		.process_func = process_one_buffer
 	};
 
-	if (!root->log_root || root->fs_info->log_root_recovering)
-		return 0;
-
-	log = root->log_root;
 	ret = walk_log_tree(trans, log, &wc);
 	BUG_ON(ret);
 
@@ -2162,14 +2136,30 @@ int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
 				  EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
 	}
 
-	if (log->log_transid > 0) {
-		ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
-				     &log->root_key);
-		BUG_ON(ret);
-	}
-	root->log_root = NULL;
 	free_extent_buffer(log->node);
 	kfree(log);
+}
+
+/*
+ * free all the extents used by the tree log.  This should be called
+ * at commit time of the full transaction
+ */
+int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
+{
+	if (root->log_root) {
+		free_log_tree(trans, root->log_root);
+		root->log_root = NULL;
+	}
+	return 0;
+}
+
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info)
+{
+	if (fs_info->log_root_tree) {
+		free_log_tree(trans, fs_info->log_root_tree);
+		fs_info->log_root_tree = NULL;
+	}
 	return 0;
 }
 
@@ -2203,6 +2193,7 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 	struct btrfs_dir_item *di;
 	struct btrfs_path *path;
 	int ret;
+	int err = 0;
 	int bytes_del = 0;
 
 	if (BTRFS_I(dir)->logged_trans < trans->transid)
@@ -2218,7 +2209,11 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 	path = btrfs_alloc_path();
 	di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
 				   name, name_len, -1);
-	if (di && !IS_ERR(di)) {
+	if (IS_ERR(di)) {
+		err = PTR_ERR(di);
+		goto fail;
+	}
+	if (di) {
 		ret = btrfs_delete_one_dir_name(trans, log, path, di);
 		bytes_del += name_len;
 		BUG_ON(ret);
@@ -2226,7 +2221,11 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 	btrfs_release_path(log, path);
 	di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
 					 index, name, name_len, -1);
-	if (di && !IS_ERR(di)) {
+	if (IS_ERR(di)) {
+		err = PTR_ERR(di);
+		goto fail;
+	}
+	if (di) {
 		ret = btrfs_delete_one_dir_name(trans, log, path, di);
 		bytes_del += name_len;
 		BUG_ON(ret);
@@ -2244,6 +2243,10 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 		btrfs_release_path(log, path);
 
 		ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
+		if (ret < 0) {
+			err = ret;
+			goto fail;
+		}
 		if (ret == 0) {
 			struct btrfs_inode_item *item;
 			u64 i_size;
@@ -2261,9 +2264,13 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 			ret = 0;
 		btrfs_release_path(log, path);
 	}
-
+fail:
 	btrfs_free_path(path);
 	mutex_unlock(&BTRFS_I(dir)->log_mutex);
+	if (ret == -ENOSPC) {
+		root->fs_info->last_trans_log_full_commit = trans->transid;
+		ret = 0;
+	}
 	btrfs_end_log_trans(root);
 
 	return 0;
@@ -2291,6 +2298,10 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
 	ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
 				  dirid, &index);
 	mutex_unlock(&BTRFS_I(inode)->log_mutex);
+	if (ret == -ENOSPC) {
+		root->fs_info->last_trans_log_full_commit = trans->transid;
+		ret = 0;
+	}
 	btrfs_end_log_trans(root);
 
 	return ret;
@@ -2318,7 +2329,8 @@ static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
 	else
 		key.type = BTRFS_DIR_LOG_INDEX_KEY;
 	ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
-	BUG_ON(ret);
+	if (ret)
+		return ret;
 
 	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 			      struct btrfs_dir_log_item);
@@ -2343,6 +2355,7 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
 	struct btrfs_key max_key;
 	struct btrfs_root *log = root->log_root;
 	struct extent_buffer *src;
+	int err = 0;
 	int ret;
 	int i;
 	int nritems;
@@ -2405,6 +2418,10 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
 			ret = overwrite_item(trans, log, dst_path,
 					     path->nodes[0], path->slots[0],
 					     &tmp);
+			if (ret) {
+				err = ret;
+				goto done;
+			}
 		}
 	}
 	btrfs_release_path(root, path);
@@ -2432,7 +2449,10 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
 				goto done;
 			ret = overwrite_item(trans, log, dst_path, src, i,
 					     &min_key);
-			BUG_ON(ret);
+			if (ret) {
+				err = ret;
+				goto done;
+			}
 		}
 		path->slots[0] = nritems;
 
@@ -2454,22 +2474,30 @@ static noinline int log_dir_items(struct btrfs_trans_handle *trans,
 			ret = overwrite_item(trans, log, dst_path,
 					     path->nodes[0], path->slots[0],
 					     &tmp);
-
-			BUG_ON(ret);
-			last_offset = tmp.offset;
+			if (ret)
+				err = ret;
+			else
+				last_offset = tmp.offset;
 			goto done;
 		}
 	}
 done:
-	*last_offset_ret = last_offset;
 	btrfs_release_path(root, path);
 	btrfs_release_path(log, dst_path);
 
-	/* insert the log range keys to indicate where the log is valid */
-	ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
-				 first_offset, last_offset);
-	BUG_ON(ret);
-	return 0;
+	if (err == 0) {
+		*last_offset_ret = last_offset;
+		/*
+		 * insert the log range keys to indicate where the log
+		 * is valid
+		 */
+		ret = insert_dir_log_key(trans, log, path, key_type,
+					 inode->i_ino, first_offset,
+					 last_offset);
+		if (ret)
+			err = ret;
+	}
+	return err;
 }
 
 /*
@@ -2501,7 +2529,8 @@ again:
 		ret = log_dir_items(trans, root, inode, path,
 				    dst_path, key_type, min_key,
 				    &max_key);
-		BUG_ON(ret);
+		if (ret)
+			return ret;
 		if (max_key == (u64)-1)
 			break;
 		min_key = max_key + 1;
@@ -2535,8 +2564,8 @@ static int drop_objectid_items(struct btrfs_trans_handle *trans,
 
 	while (1) {
 		ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
-
-		if (ret != 1)
+		BUG_ON(ret == 0);
+		if (ret < 0)
 			break;
 
 		if (path->slots[0] == 0)
@@ -2554,7 +2583,7 @@ static int drop_objectid_items(struct btrfs_trans_handle *trans,
 		btrfs_release_path(log, path);
 	}
 	btrfs_release_path(log, path);
-	return 0;
+	return ret;
 }
 
 static noinline int copy_items(struct btrfs_trans_handle *trans,
@@ -2587,7 +2616,10 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	}
 	ret = btrfs_insert_empty_items(trans, log, dst_path,
 				       ins_keys, ins_sizes, nr);
-	BUG_ON(ret);
+	if (ret) {
+		kfree(ins_data);
+		return ret;
+	}
 
 	for (i = 0; i < nr; i++, dst_path->slots[0]++) {
 		dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
@@ -2660,16 +2692,17 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	 * we have to do this after the loop above to avoid changing the
 	 * log tree while trying to change the log tree.
 	 */
+	ret = 0;
 	while (!list_empty(&ordered_sums)) {
 		struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
 						   struct btrfs_ordered_sum,
 						   list);
-		ret = btrfs_csum_file_blocks(trans, log, sums);
-		BUG_ON(ret);
+		if (!ret)
+			ret = btrfs_csum_file_blocks(trans, log, sums);
 		list_del(&sums->list);
 		kfree(sums);
 	}
-	return 0;
+	return ret;
 }
 
 /* log a single inode in the tree log.
@@ -2697,6 +2730,7 @@ static int btrfs_log_inode(struct btrfs_trans_handle *trans,
 	struct btrfs_root *log = root->log_root;
 	struct extent_buffer *src = NULL;
 	u32 size;
+	int err = 0;
 	int ret;
 	int nritems;
 	int ins_start_slot = 0;
@@ -2739,7 +2773,10 @@ static int btrfs_log_inode(struct btrfs_trans_handle *trans,
 	} else {
 		ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
 	}
-	BUG_ON(ret);
+	if (ret) {
+		err = ret;
+		goto out_unlock;
+	}
 	path->keep_locks = 1;
 
 	while (1) {
@@ -2768,7 +2805,10 @@ again:
 
 		ret = copy_items(trans, log, dst_path, src, ins_start_slot,
 				 ins_nr, inode_only);
-		BUG_ON(ret);
+		if (ret) {
+			err = ret;
+			goto out_unlock;
+		}
 		ins_nr = 1;
 		ins_start_slot = path->slots[0];
 next_slot:
@@ -2784,7 +2824,10 @@ next_slot:
 			ret = copy_items(trans, log, dst_path, src,
 					 ins_start_slot,
 					 ins_nr, inode_only);
-			BUG_ON(ret);
+			if (ret) {
+				err = ret;
+				goto out_unlock;
+			}
 			ins_nr = 0;
 		}
 		btrfs_release_path(root, path);
@@ -2802,7 +2845,10 @@ next_slot:
 		ret = copy_items(trans, log, dst_path, src,
 				 ins_start_slot,
 				 ins_nr, inode_only);
-		BUG_ON(ret);
+		if (ret) {
+			err = ret;
+			goto out_unlock;
+		}
 		ins_nr = 0;
 	}
 	WARN_ON(ins_nr);
@@ -2810,14 +2856,18 @@ next_slot:
 		btrfs_release_path(root, path);
 		btrfs_release_path(log, dst_path);
 		ret = log_directory_changes(trans, root, inode, path, dst_path);
-		BUG_ON(ret);
+		if (ret) {
+			err = ret;
+			goto out_unlock;
+		}
 	}
 	BTRFS_I(inode)->logged_trans = trans->transid;
+out_unlock:
 	mutex_unlock(&BTRFS_I(inode)->log_mutex);
 
 	btrfs_free_path(path);
 	btrfs_free_path(dst_path);
-	return 0;
+	return err;
 }
 
 /*
@@ -2942,10 +2992,13 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 		goto end_no_trans;
 	}
 
-	start_log_trans(trans, root);
+	ret = start_log_trans(trans, root);
+	if (ret)
+		goto end_trans;
 
 	ret = btrfs_log_inode(trans, root, inode, inode_only);
-	BUG_ON(ret);
+	if (ret)
+		goto end_trans;
 
 	/*
 	 * for regular files, if its inode is already on disk, we don't
@@ -2955,8 +3008,10 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 	 */
 	if (S_ISREG(inode->i_mode) &&
 	    BTRFS_I(inode)->generation <= last_committed &&
-	    BTRFS_I(inode)->last_unlink_trans <= last_committed)
-			goto no_parent;
+	    BTRFS_I(inode)->last_unlink_trans <= last_committed) {
+		ret = 0;
+		goto end_trans;
+	}
 
 	inode_only = LOG_INODE_EXISTS;
 	while (1) {
@@ -2970,15 +3025,21 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 		if (BTRFS_I(inode)->generation >
 		    root->fs_info->last_trans_committed) {
 			ret = btrfs_log_inode(trans, root, inode, inode_only);
-			BUG_ON(ret);
+			if (ret)
+				goto end_trans;
 		}
 		if (IS_ROOT(parent))
 			break;
 
 		parent = parent->d_parent;
 	}
-no_parent:
 	ret = 0;
+end_trans:
+	if (ret < 0) {
+		BUG_ON(ret != -ENOSPC);
+		root->fs_info->last_trans_log_full_commit = trans->transid;
+		ret = 1;
+	}
 	btrfs_end_log_trans(root);
 end_no_trans:
 	return ret;
@@ -3020,7 +3081,7 @@ int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 
-	trans = btrfs_start_transaction(fs_info->tree_root, 1);
+	trans = btrfs_start_transaction(fs_info->tree_root, 0);
 
 	wc.trans = trans;
 	wc.pin = 1;

fs/btrfs/tree-log.h

@@ -25,6 +25,8 @@
 int btrfs_sync_log(struct btrfs_trans_handle *trans,
 		   struct btrfs_root *root);
 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root);
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info);
 int btrfs_recover_log_trees(struct btrfs_root *tree_root);
 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root, struct dentry *dentry);

fs/btrfs/volumes.c

@@ -1097,7 +1097,7 @@ static int btrfs_rm_dev_item(struct btrfs_root *root,
 	if (!path)
 		return -ENOMEM;
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_start_transaction(root, 0);
 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 	key.type = BTRFS_DEV_ITEM_KEY;
 	key.offset = device->devid;
@@ -1486,7 +1486,7 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 		goto error;
 	}
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_start_transaction(root, 0);
 	lock_chunks(root);
 
 	device->barriers = 1;
@@ -1751,9 +1751,10 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
 
 	/* step one, relocate all the extents inside this chunk */
 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
-	BUG_ON(ret);
+	if (ret)
+		return ret;
 
-	trans = btrfs_start_transaction(root, 1);
+	trans = btrfs_start_transaction(root, 0);
 	BUG_ON(!trans);
 
 	lock_chunks(root);
@@ -1925,7 +1926,7 @@ int btrfs_balance(struct btrfs_root *dev_root)
 			break;
 		BUG_ON(ret);
 
-		trans = btrfs_start_transaction(dev_root, 1);
+		trans = btrfs_start_transaction(dev_root, 0);
 		BUG_ON(!trans);
 
 		ret = btrfs_grow_device(trans, device, old_size);
@@ -2094,11 +2095,7 @@ again:
 	}
 
 	/* Shrinking succeeded, else we would be at "done". */
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans) {
-		ret = -ENOMEM;
-		goto done;
-	}
+	trans = btrfs_start_transaction(root, 0);
 	lock_chunks(root);
 
 	device->disk_total_bytes = new_size;

fs/btrfs/xattr.c

@@ -154,15 +154,10 @@ int __btrfs_setxattr(struct btrfs_trans_handle *trans,
 	if (trans)
 		return do_setxattr(trans, inode, name, value, size, flags);
 
-	ret = btrfs_reserve_metadata_space(root, 2);
-	if (ret)
-		return ret;
+	trans = btrfs_start_transaction(root, 2);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans) {
-		ret = -ENOMEM;
-		goto out;
-	}
 	btrfs_set_trans_block_group(trans, inode);
 
 	ret = do_setxattr(trans, inode, name, value, size, flags);
@@ -174,7 +169,6 @@ int __btrfs_setxattr(struct btrfs_trans_handle *trans,
 	BUG_ON(ret);
 out:
 	btrfs_end_transaction_throttle(trans, root);
-	btrfs_unreserve_metadata_space(root, 2);
 	return ret;
 }
 

fs/direct-io.c

@@ -82,6 +82,8 @@ struct dio {
 	int reap_counter;		/* rate limit reaping */
 	get_block_t *get_block;		/* block mapping function */
 	dio_iodone_t *end_io;		/* IO completion function */
+	dio_submit_t *submit_io;	/* IO submition function */
+	loff_t logical_offset_in_bio;	/* current first logical block in bio */
 	sector_t final_block_in_bio;	/* current final block in bio + 1 */
 	sector_t next_block_for_io;	/* next block to be put under IO,
 					   in dio_blocks units */
@@ -96,6 +98,7 @@ struct dio {
 	unsigned cur_page_offset;	/* Offset into it, in bytes */
 	unsigned cur_page_len;		/* Nr of bytes at cur_page_offset */
 	sector_t cur_page_block;	/* Where it starts */
+	loff_t cur_page_fs_offset;	/* Offset in file */
 
 	/* BIO completion state */
 	spinlock_t bio_lock;		/* protects BIO fields below */
@@ -300,6 +303,26 @@ static void dio_bio_end_io(struct bio *bio, int error)
 	spin_unlock_irqrestore(&dio->bio_lock, flags);
 }
 
+/**
+ * dio_end_io - handle the end io action for the given bio
+ * @bio: The direct io bio thats being completed
+ * @error: Error if there was one
+ *
+ * This is meant to be called by any filesystem that uses their own dio_submit_t
+ * so that the DIO specific endio actions are dealt with after the filesystem
+ * has done it's completion work.
+ */
+void dio_end_io(struct bio *bio, int error)
+{
+	struct dio *dio = bio->bi_private;
+
+	if (dio->is_async)
+		dio_bio_end_aio(bio, error);
+	else
+		dio_bio_end_io(bio, error);
+}
+EXPORT_SYMBOL_GPL(dio_end_io);
+
 static int
 dio_bio_alloc(struct dio *dio, struct block_device *bdev,
 		sector_t first_sector, int nr_vecs)
@@ -316,6 +339,7 @@ dio_bio_alloc(struct dio *dio, struct block_device *bdev,
 		bio->bi_end_io = dio_bio_end_io;
 
 	dio->bio = bio;
+	dio->logical_offset_in_bio = dio->cur_page_fs_offset;
 	return 0;
 }
 
@@ -340,10 +364,15 @@ static void dio_bio_submit(struct dio *dio)
 	if (dio->is_async && dio->rw == READ)
 		bio_set_pages_dirty(bio);
 
-	submit_bio(dio->rw, bio);
+	if (dio->submit_io)
+		dio->submit_io(dio->rw, bio, dio->inode,
+			       dio->logical_offset_in_bio);
+	else
+		submit_bio(dio->rw, bio);
 
 	dio->bio = NULL;
 	dio->boundary = 0;
+	dio->logical_offset_in_bio = 0;
 }
 
 /*
@@ -603,10 +632,26 @@ static int dio_send_cur_page(struct dio *dio)
 	int ret = 0;
 
 	if (dio->bio) {
+		loff_t cur_offset = dio->block_in_file << dio->blkbits;
+		loff_t bio_next_offset = dio->logical_offset_in_bio +
+			dio->bio->bi_size;
+
 		/*
-		 * See whether this new request is contiguous with the old
+		 * See whether this new request is contiguous with the old.
+		 *
+		 * Btrfs cannot handl having logically non-contiguous requests
+		 * submitted.  For exmple if you have
+		 *
+		 * Logical:  [0-4095][HOLE][8192-12287]
+		 * Phyiscal: [0-4095]      [4096-8181]
+		 *
+		 * We cannot submit those pages together as one BIO.  So if our
+		 * current logical offset in the file does not equal what would
+		 * be the next logical offset in the bio, submit the bio we
+		 * have.
 		 */
-		if (dio->final_block_in_bio != dio->cur_page_block)
+		if (dio->final_block_in_bio != dio->cur_page_block ||
+		    cur_offset != bio_next_offset)
 			dio_bio_submit(dio);
 		/*
 		 * Submit now if the underlying fs is about to perform a
@@ -701,6 +746,7 @@ submit_page_section(struct dio *dio, struct page *page,
 	dio->cur_page_offset = offset;
 	dio->cur_page_len = len;
 	dio->cur_page_block = blocknr;
+	dio->cur_page_fs_offset = dio->block_in_file << dio->blkbits;
 out:
 	return ret;
 }
@@ -935,7 +981,7 @@ static ssize_t
 direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode, 
 	const struct iovec *iov, loff_t offset, unsigned long nr_segs, 
 	unsigned blkbits, get_block_t get_block, dio_iodone_t end_io,
-	struct dio *dio)
+	dio_submit_t submit_io, struct dio *dio)
 {
 	unsigned long user_addr; 
 	unsigned long flags;
@@ -952,6 +998,7 @@ direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode,
 
 	dio->get_block = get_block;
 	dio->end_io = end_io;
+	dio->submit_io = submit_io;
 	dio->final_block_in_bio = -1;
 	dio->next_block_for_io = -1;
 
@@ -1008,7 +1055,7 @@ direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode,
 		}
 	} /* end iovec loop */
 
-	if (ret == -ENOTBLK && (rw & WRITE)) {
+	if (ret == -ENOTBLK) {
 		/*
 		 * The remaining part of the request will be
 		 * be handled by buffered I/O when we return
@@ -1110,7 +1157,7 @@ ssize_t
 __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	struct block_device *bdev, const struct iovec *iov, loff_t offset, 
 	unsigned long nr_segs, get_block_t get_block, dio_iodone_t end_io,
-	int flags)
+	dio_submit_t submit_io,	int flags)
 {
 	int seg;
 	size_t size;
@@ -1197,7 +1244,8 @@ __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 		(end > i_size_read(inode)));
 
 	retval = direct_io_worker(rw, iocb, inode, iov, offset,
-				nr_segs, blkbits, get_block, end_io, dio);
+				nr_segs, blkbits, get_block, end_io,
+				submit_io, dio);
 
 	/*
 	 * In case of error extending write may have instantiated a few

include/linux/fs.h

@@ -2251,10 +2251,15 @@ static inline int xip_truncate_page(struct address_space *mapping, loff_t from)
 #endif
 
 #ifdef CONFIG_BLOCK
+struct bio;
+typedef void (dio_submit_t)(int rw, struct bio *bio, struct inode *inode,
+			    loff_t file_offset);
+void dio_end_io(struct bio *bio, int error);
+
 ssize_t __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	struct block_device *bdev, const struct iovec *iov, loff_t offset,
 	unsigned long nr_segs, get_block_t get_block, dio_iodone_t end_io,
-	int lock_type);
+	dio_submit_t submit_io,	int lock_type);
 
 enum {
 	/* need locking between buffered and direct access */
@@ -2270,7 +2275,7 @@ static inline ssize_t blockdev_direct_IO(int rw, struct kiocb *iocb,
 	dio_iodone_t end_io)
 {
 	return __blockdev_direct_IO(rw, iocb, inode, bdev, iov, offset,
-				    nr_segs, get_block, end_io,
+				    nr_segs, get_block, end_io, NULL,
 				    DIO_LOCKING | DIO_SKIP_HOLES);
 }
 
@@ -2280,7 +2285,7 @@ static inline ssize_t blockdev_direct_IO_no_locking(int rw, struct kiocb *iocb,
 	dio_iodone_t end_io)
 {
 	return __blockdev_direct_IO(rw, iocb, inode, bdev, iov, offset,
-				nr_segs, get_block, end_io, 0);
+				    nr_segs, get_block, end_io, NULL, 0);
 }
 #endif
 

mm/filemap.c

@@ -1275,7 +1275,7 @@ generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
 {
 	struct file *filp = iocb->ki_filp;
 	ssize_t retval;
-	unsigned long seg;
+	unsigned long seg = 0;
 	size_t count;
 	loff_t *ppos = &iocb->ki_pos;
 
@@ -1302,21 +1302,47 @@ generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
 				retval = mapping->a_ops->direct_IO(READ, iocb,
 							iov, pos, nr_segs);
 			}
-			if (retval > 0)
+			if (retval > 0) {
 				*ppos = pos + retval;
-			if (retval) {
+				count -= retval;
+			}
+
+			/*
+			 * Btrfs can have a short DIO read if we encounter
+			 * compressed extents, so if there was an error, or if
+			 * we've already read everything we wanted to, or if
+			 * there was a short read because we hit EOF, go ahead
+			 * and return.  Otherwise fallthrough to buffered io for
+			 * the rest of the read.
+			 */
+			if (retval < 0 || !count || *ppos >= size) {
 				file_accessed(filp);
 				goto out;
 			}
 		}
 	}
 
+	count = retval;
 	for (seg = 0; seg < nr_segs; seg++) {
 		read_descriptor_t desc;
+		loff_t offset = 0;
+
+		/*
+		 * If we did a short DIO read we need to skip the section of the
+		 * iov that we've already read data into.
+		 */
+		if (count) {
+			if (count > iov[seg].iov_len) {
+				count -= iov[seg].iov_len;
+				continue;
+			}
+			offset = count;
+			count = 0;
+		}
 
 		desc.written = 0;
-		desc.arg.buf = iov[seg].iov_base;
-		desc.count = iov[seg].iov_len;
+		desc.arg.buf = iov[seg].iov_base + offset;
+		desc.count = iov[seg].iov_len - offset;
 		if (desc.count == 0)
 			continue;
 		desc.error = 0;