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: (23 commits)
  Btrfs: fix extent_buffer leak during tree log replay
  Btrfs: fix oops when btrfs_inherit_iflags called with a NULL dir
  Btrfs: fix -o nodatasum printk spelling
  Btrfs: check duplicate backrefs for both data and metadata
  Btrfs: init worker struct fields before kthread-run
  Btrfs: pin buffers during write_dev_supers
  Btrfs: avoid races between super writeout and device list updates
  Fix btrfs when ACLs are configured out
  Btrfs: fdatasync should skip metadata writeout
  Btrfs: remove crc32c.h and use libcrc32c directly.
  Btrfs: implement FS_IOC_GETFLAGS/SETFLAGS/GETVERSION
  Btrfs: autodetect SSD devices
  Btrfs: add mount -o ssd_spread to spread allocations out
  Btrfs: avoid allocation clusters that are too spread out
  Btrfs: Add mount -o nossd
  Btrfs: avoid IO stalls behind congested devices in a multi-device FS
  Btrfs: don't allow WRITE_SYNC bios to starve out regular writes
  Btrfs: fix metadata dirty throttling limits
  Btrfs: reduce mount -o ssd CPU usage
  Btrfs: balance btree more often
  ...

fs/btrfs/Makefile

@@ -6,5 +6,5 @@ btrfs-y += super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
 	   transaction.o inode.o file.o tree-defrag.o \
 	   extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \
 	   extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
-	   ref-cache.o export.o tree-log.o acl.o free-space-cache.o zlib.o \
-	   compression.o delayed-ref.o
+	   export.o tree-log.o acl.o free-space-cache.o zlib.o \
+	   compression.o delayed-ref.o relocation.o

fs/btrfs/acl.c

@@ -351,9 +351,4 @@ int btrfs_init_acl(struct inode *inode, struct inode *dir)
 	return 0;
 }
 
-int btrfs_check_acl(struct inode *inode, int mask)
-{
-	return 0;
-}
-
 #endif /* CONFIG_FS_POSIX_ACL */

fs/btrfs/async-thread.c

@@ -294,10 +294,10 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
 		INIT_LIST_HEAD(&worker->worker_list);
 		spin_lock_init(&worker->lock);
 		atomic_set(&worker->num_pending, 0);
+		worker->workers = workers;
 		worker->task = kthread_run(worker_loop, worker,
 					   "btrfs-%s-%d", workers->name,
 					   workers->num_workers + i);
-		worker->workers = workers;
 		if (IS_ERR(worker->task)) {
 			kfree(worker);
 			ret = PTR_ERR(worker->task);

fs/btrfs/btrfs_inode.h

@@ -72,6 +72,9 @@ struct btrfs_inode {
 	 */
 	struct list_head ordered_operations;
 
+	/* node for the red-black tree that links inodes in subvolume root */
+	struct rb_node rb_node;
+
 	/* the space_info for where this inode's data allocations are done */
 	struct btrfs_space_info *space_info;
 
@@ -154,5 +157,4 @@ static inline void btrfs_i_size_write(struct inode *inode, u64 size)
 	BTRFS_I(inode)->disk_i_size = size;
 }
 
-
 #endif

fs/btrfs/compression.c

@@ -123,7 +123,7 @@ static int check_compressed_csum(struct inode *inode,
 	u32 csum;
 	u32 *cb_sum = &cb->sums;
 
-	if (btrfs_test_flag(inode, NODATASUM))
+	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
 		return 0;
 
 	for (i = 0; i < cb->nr_pages; i++) {
@@ -670,7 +670,7 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
 			 */
 			atomic_inc(&cb->pending_bios);
 
-			if (!btrfs_test_flag(inode, NODATASUM)) {
+			if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
 				btrfs_lookup_bio_sums(root, inode, comp_bio,
 						      sums);
 			}
@@ -697,7 +697,7 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
 	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
 	BUG_ON(ret);
 
-	if (!btrfs_test_flag(inode, NODATASUM))
+	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
 		btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
 
 	ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);

fs/btrfs/crc32c.h

@@ -1,29 +0,0 @@
-/*
- * Copyright (C) 2008 Oracle.  All rights reserved.
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public
- * License v2 as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License for more details.
- *
- * You should have received a copy of the GNU General Public
- * License along with this program; if not, write to the
- * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
- * Boston, MA 021110-1307, USA.
- */
-
-#ifndef __BTRFS_CRC32C__
-#define __BTRFS_CRC32C__
-#include <linux/crc32c.h>
-
-/*
- * this file used to do more for selecting the HW version of crc32c,
- * perhaps it will one day again soon.
- */
-#define btrfs_crc32c(seed, data, length) crc32c(seed, data, length)
-#endif
-

fs/btrfs/ctree.c

@@ -197,14 +197,7 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 	u32 nritems;
 	int ret = 0;
 	int level;
-	struct btrfs_root *new_root;
-
-	new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
-	if (!new_root)
-		return -ENOMEM;
-
-	memcpy(new_root, root, sizeof(*new_root));
-	new_root->root_key.objectid = new_root_objectid;
+	struct btrfs_disk_key disk_key;
 
 	WARN_ON(root->ref_cows && trans->transid !=
 		root->fs_info->running_transaction->transid);
@@ -212,28 +205,37 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 
 	level = btrfs_header_level(buf);
 	nritems = btrfs_header_nritems(buf);
+	if (level == 0)
+		btrfs_item_key(buf, &disk_key, 0);
+	else
+		btrfs_node_key(buf, &disk_key, 0);
 
-	cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
-				     new_root_objectid, trans->transid,
-				     level, buf->start, 0);
-	if (IS_ERR(cow)) {
-		kfree(new_root);
+	cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
+				     new_root_objectid, &disk_key, level,
+				     buf->start, 0);
+	if (IS_ERR(cow))
 		return PTR_ERR(cow);
-	}
 
 	copy_extent_buffer(cow, buf, 0, 0, cow->len);
 	btrfs_set_header_bytenr(cow, cow->start);
 	btrfs_set_header_generation(cow, trans->transid);
-	btrfs_set_header_owner(cow, new_root_objectid);
-	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
+	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
+	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
+				     BTRFS_HEADER_FLAG_RELOC);
+	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
+		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
+	else
+		btrfs_set_header_owner(cow, new_root_objectid);
 
 	write_extent_buffer(cow, root->fs_info->fsid,
 			    (unsigned long)btrfs_header_fsid(cow),
 			    BTRFS_FSID_SIZE);
 
 	WARN_ON(btrfs_header_generation(buf) > trans->transid);
-	ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
-	kfree(new_root);
+	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
+		ret = btrfs_inc_ref(trans, root, cow, 1);
+	else
+		ret = btrfs_inc_ref(trans, root, cow, 0);
 
 	if (ret)
 		return ret;
@@ -243,6 +245,125 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+/*
+ * check if the tree block can be shared by multiple trees
+ */
+int btrfs_block_can_be_shared(struct btrfs_root *root,
+			      struct extent_buffer *buf)
+{
+	/*
+	 * Tree blocks not in refernece counted trees and tree roots
+	 * are never shared. If a block was allocated after the last
+	 * snapshot and the block was not allocated by tree relocation,
+	 * we know the block is not shared.
+	 */
+	if (root->ref_cows &&
+	    buf != root->node && buf != root->commit_root &&
+	    (btrfs_header_generation(buf) <=
+	     btrfs_root_last_snapshot(&root->root_item) ||
+	     btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
+		return 1;
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (root->ref_cows &&
+	    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
+		return 1;
+#endif
+	return 0;
+}
+
+static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
+				       struct btrfs_root *root,
+				       struct extent_buffer *buf,
+				       struct extent_buffer *cow)
+{
+	u64 refs;
+	u64 owner;
+	u64 flags;
+	u64 new_flags = 0;
+	int ret;
+
+	/*
+	 * Backrefs update rules:
+	 *
+	 * Always use full backrefs for extent pointers in tree block
+	 * allocated by tree relocation.
+	 *
+	 * If a shared tree block is no longer referenced by its owner
+	 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
+	 * use full backrefs for extent pointers in tree block.
+	 *
+	 * If a tree block is been relocating
+	 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
+	 * use full backrefs for extent pointers in tree block.
+	 * The reason for this is some operations (such as drop tree)
+	 * are only allowed for blocks use full backrefs.
+	 */
+
+	if (btrfs_block_can_be_shared(root, buf)) {
+		ret = btrfs_lookup_extent_info(trans, root, buf->start,
+					       buf->len, &refs, &flags);
+		BUG_ON(ret);
+		BUG_ON(refs == 0);
+	} else {
+		refs = 1;
+		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
+		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
+			flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
+		else
+			flags = 0;
+	}
+
+	owner = btrfs_header_owner(buf);
+	BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
+	       !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
+
+	if (refs > 1) {
+		if ((owner == root->root_key.objectid ||
+		     root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
+		    !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
+			ret = btrfs_inc_ref(trans, root, buf, 1);
+			BUG_ON(ret);
+
+			if (root->root_key.objectid ==
+			    BTRFS_TREE_RELOC_OBJECTID) {
+				ret = btrfs_dec_ref(trans, root, buf, 0);
+				BUG_ON(ret);
+				ret = btrfs_inc_ref(trans, root, cow, 1);
+				BUG_ON(ret);
+			}
+			new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
+		} else {
+
+			if (root->root_key.objectid ==
+			    BTRFS_TREE_RELOC_OBJECTID)
+				ret = btrfs_inc_ref(trans, root, cow, 1);
+			else
+				ret = btrfs_inc_ref(trans, root, cow, 0);
+			BUG_ON(ret);
+		}
+		if (new_flags != 0) {
+			ret = btrfs_set_disk_extent_flags(trans, root,
+							  buf->start,
+							  buf->len,
+							  new_flags, 0);
+			BUG_ON(ret);
+		}
+	} else {
+		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
+			if (root->root_key.objectid ==
+			    BTRFS_TREE_RELOC_OBJECTID)
+				ret = btrfs_inc_ref(trans, root, cow, 1);
+			else
+				ret = btrfs_inc_ref(trans, root, cow, 0);
+			BUG_ON(ret);
+			ret = btrfs_dec_ref(trans, root, buf, 1);
+			BUG_ON(ret);
+		}
+		clean_tree_block(trans, root, buf);
+	}
+	return 0;
+}
+
 /*
  * does the dirty work in cow of a single block.  The parent block (if
  * supplied) is updated to point to the new cow copy.  The new buffer is marked
@@ -262,34 +383,39 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 			     struct extent_buffer **cow_ret,
 			     u64 search_start, u64 empty_size)
 {
-	u64 parent_start;
+	struct btrfs_disk_key disk_key;
 	struct extent_buffer *cow;
-	u32 nritems;
-	int ret = 0;
 	int level;
 	int unlock_orig = 0;
+	u64 parent_start;
 
 	if (*cow_ret == buf)
 		unlock_orig = 1;
 
 	btrfs_assert_tree_locked(buf);
 
-	if (parent)
-		parent_start = parent->start;
-	else
-		parent_start = 0;
-
 	WARN_ON(root->ref_cows && trans->transid !=
 		root->fs_info->running_transaction->transid);
 	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
 
 	level = btrfs_header_level(buf);
-	nritems = btrfs_header_nritems(buf);
 
-	cow = btrfs_alloc_free_block(trans, root, buf->len,
-				     parent_start, root->root_key.objectid,
-				     trans->transid, level,
-				     search_start, empty_size);
+	if (level == 0)
+		btrfs_item_key(buf, &disk_key, 0);
+	else
+		btrfs_node_key(buf, &disk_key, 0);
+
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
+		if (parent)
+			parent_start = parent->start;
+		else
+			parent_start = 0;
+	} else
+		parent_start = 0;
+
+	cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
+				     root->root_key.objectid, &disk_key,
+				     level, search_start, empty_size);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -298,83 +424,53 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	copy_extent_buffer(cow, buf, 0, 0, cow->len);
 	btrfs_set_header_bytenr(cow, cow->start);
 	btrfs_set_header_generation(cow, trans->transid);
-	btrfs_set_header_owner(cow, root->root_key.objectid);
-	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
+	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
+	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
+				     BTRFS_HEADER_FLAG_RELOC);
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
+		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
+	else
+		btrfs_set_header_owner(cow, root->root_key.objectid);
 
 	write_extent_buffer(cow, root->fs_info->fsid,
 			    (unsigned long)btrfs_header_fsid(cow),
 			    BTRFS_FSID_SIZE);
 
-	WARN_ON(btrfs_header_generation(buf) > trans->transid);
-	if (btrfs_header_generation(buf) != trans->transid) {
-		u32 nr_extents;
-		ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
-		if (ret)
-			return ret;
-
-		ret = btrfs_cache_ref(trans, root, buf, nr_extents);
-		WARN_ON(ret);
-	} else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
-		/*
-		 * There are only two places that can drop reference to
-		 * tree blocks owned by living reloc trees, one is here,
-		 * the other place is btrfs_drop_subtree. In both places,
-		 * we check reference count while tree block is locked.
-		 * Furthermore, if reference count is one, it won't get
-		 * increased by someone else.
-		 */
-		u32 refs;
-		ret = btrfs_lookup_extent_ref(trans, root, buf->start,
-					      buf->len, &refs);
-		BUG_ON(ret);
-		if (refs == 1) {
-			ret = btrfs_update_ref(trans, root, buf, cow,
-					       0, nritems);
-			clean_tree_block(trans, root, buf);
-		} else {
-			ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
-		}
-		BUG_ON(ret);
-	} else {
-		ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
-		if (ret)
-			return ret;
-		clean_tree_block(trans, root, buf);
-	}
-
-	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
-		ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
-		WARN_ON(ret);
-	}
+	update_ref_for_cow(trans, root, buf, cow);
 
 	if (buf == root->node) {
 		WARN_ON(parent && parent != buf);
+		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
+		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
+			parent_start = buf->start;
+		else
+			parent_start = 0;
 
 		spin_lock(&root->node_lock);
 		root->node = cow;
 		extent_buffer_get(cow);
 		spin_unlock(&root->node_lock);
 
-		if (buf != root->commit_root) {
-			btrfs_free_extent(trans, root, buf->start,
-					  buf->len, buf->start,
-					  root->root_key.objectid,
-					  btrfs_header_generation(buf),
-					  level, 1);
-		}
+		btrfs_free_extent(trans, root, buf->start, buf->len,
+				  parent_start, root->root_key.objectid,
+				  level, 0);
 		free_extent_buffer(buf);
 		add_root_to_dirty_list(root);
 	} else {
+		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
+			parent_start = parent->start;
+		else
+			parent_start = 0;
+
+		WARN_ON(trans->transid != btrfs_header_generation(parent));
 		btrfs_set_node_blockptr(parent, parent_slot,
 					cow->start);
-		WARN_ON(trans->transid == 0);
 		btrfs_set_node_ptr_generation(parent, parent_slot,
 					      trans->transid);
 		btrfs_mark_buffer_dirty(parent);
-		WARN_ON(btrfs_header_generation(parent) != trans->transid);
 		btrfs_free_extent(trans, root, buf->start, buf->len,
-				  parent_start, btrfs_header_owner(parent),
-				  btrfs_header_generation(parent), level, 1);
+				  parent_start, root->root_key.objectid,
+				  level, 0);
 	}
 	if (unlock_orig)
 		btrfs_tree_unlock(buf);
@@ -384,6 +480,18 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+static inline int should_cow_block(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct extent_buffer *buf)
+{
+	if (btrfs_header_generation(buf) == trans->transid &&
+	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
+	    !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
+	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
+		return 0;
+	return 1;
+}
+
 /*
  * cows a single block, see __btrfs_cow_block for the real work.
  * This version of it has extra checks so that a block isn't cow'd more than
@@ -411,9 +519,7 @@ noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
 		WARN_ON(1);
 	}
 
-	if (btrfs_header_generation(buf) == trans->transid &&
-	    btrfs_header_owner(buf) == root->root_key.objectid &&
-	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
+	if (!should_cow_block(trans, root, buf)) {
 		*cow_ret = buf;
 		return 0;
 	}
@@ -469,7 +575,7 @@ static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
 /*
  * same as comp_keys only with two btrfs_key's
  */
-static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
+int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
 {
 	if (k1->objectid > k2->objectid)
 		return 1;
@@ -845,6 +951,12 @@ static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 	return -1;
 }
 
+int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
+		     int level, int *slot)
+{
+	return bin_search(eb, key, level, slot);
+}
+
 /* 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.
@@ -921,13 +1033,6 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		root->node = child;
 		spin_unlock(&root->node_lock);
 
-		ret = btrfs_update_extent_ref(trans, root, child->start,
-					      child->len,
-					      mid->start, child->start,
-					      root->root_key.objectid,
-					      trans->transid, level - 1);
-		BUG_ON(ret);
-
 		add_root_to_dirty_list(root);
 		btrfs_tree_unlock(child);
 
@@ -938,9 +1043,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		/* once for the path */
 		free_extent_buffer(mid);
 		ret = btrfs_free_extent(trans, root, mid->start, mid->len,
-					mid->start, root->root_key.objectid,
-					btrfs_header_generation(mid),
-					level, 1);
+					0, root->root_key.objectid, level, 1);
 		/* once for the root ptr */
 		free_extent_buffer(mid);
 		return ret;
@@ -949,8 +1052,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 	    BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
 		return 0;
 
-	if (trans->transaction->delayed_refs.flushing &&
-	    btrfs_header_nritems(mid) > 2)
+	if (btrfs_header_nritems(mid) > 2)
 		return 0;
 
 	if (btrfs_header_nritems(mid) < 2)
@@ -998,7 +1100,6 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 			ret = wret;
 		if (btrfs_header_nritems(right) == 0) {
 			u64 bytenr = right->start;
-			u64 generation = btrfs_header_generation(parent);
 			u32 blocksize = right->len;
 
 			clean_tree_block(trans, root, right);
@@ -1010,9 +1111,9 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 			if (wret)
 				ret = wret;
 			wret = btrfs_free_extent(trans, root, bytenr,
-						 blocksize, parent->start,
-						 btrfs_header_owner(parent),
-						 generation, level, 1);
+						 blocksize, 0,
+						 root->root_key.objectid,
+						 level, 0);
 			if (wret)
 				ret = wret;
 		} else {
@@ -1047,7 +1148,6 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 	}
 	if (btrfs_header_nritems(mid) == 0) {
 		/* we've managed to empty the middle node, drop it */
-		u64 root_gen = btrfs_header_generation(parent);
 		u64 bytenr = mid->start;
 		u32 blocksize = mid->len;
 
@@ -1059,9 +1159,8 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		if (wret)
 			ret = wret;
 		wret = btrfs_free_extent(trans, root, bytenr, blocksize,
-					 parent->start,
-					 btrfs_header_owner(parent),
-					 root_gen, level, 1);
+					 0, root->root_key.objectid,
+					 level, 0);
 		if (wret)
 			ret = wret;
 	} else {
@@ -1437,7 +1536,7 @@ noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
 {
 	int i;
 
-	if (path->keep_locks || path->lowest_level)
+	if (path->keep_locks)
 		return;
 
 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
@@ -1552,7 +1651,7 @@ setup_nodes_for_search(struct btrfs_trans_handle *trans,
 		}
 		b = p->nodes[level];
 	} else if (ins_len < 0 && btrfs_header_nritems(b) <
-		   BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
+		   BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
 		int sret;
 
 		sret = reada_for_balance(root, p, level);
@@ -1614,10 +1713,17 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 		lowest_unlock = 2;
 
 again:
-	if (p->skip_locking)
-		b = btrfs_root_node(root);
-	else
-		b = btrfs_lock_root_node(root);
+	if (p->search_commit_root) {
+		b = root->commit_root;
+		extent_buffer_get(b);
+		if (!p->skip_locking)
+			btrfs_tree_lock(b);
+	} else {
+		if (p->skip_locking)
+			b = btrfs_root_node(root);
+		else
+			b = btrfs_lock_root_node(root);
+	}
 
 	while (b) {
 		level = btrfs_header_level(b);
@@ -1638,11 +1744,9 @@ again:
 			 * then we don't want to set the path blocking,
 			 * so we test it here
 			 */
-			if (btrfs_header_generation(b) == trans->transid &&
-			    btrfs_header_owner(b) == root->root_key.objectid &&
-			    !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
+			if (!should_cow_block(trans, root, b))
 				goto cow_done;
-			}
+
 			btrfs_set_path_blocking(p);
 
 			wret = btrfs_cow_block(trans, root, b,
@@ -1764,138 +1868,6 @@ done:
 	return ret;
 }
 
-int btrfs_merge_path(struct btrfs_trans_handle *trans,
-		     struct btrfs_root *root,
-		     struct btrfs_key *node_keys,
-		     u64 *nodes, int lowest_level)
-{
-	struct extent_buffer *eb;
-	struct extent_buffer *parent;
-	struct btrfs_key key;
-	u64 bytenr;
-	u64 generation;
-	u32 blocksize;
-	int level;
-	int slot;
-	int key_match;
-	int ret;
-
-	eb = btrfs_lock_root_node(root);
-	ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
-	BUG_ON(ret);
-
-	btrfs_set_lock_blocking(eb);
-
-	parent = eb;
-	while (1) {
-		level = btrfs_header_level(parent);
-		if (level == 0 || level <= lowest_level)
-			break;
-
-		ret = bin_search(parent, &node_keys[lowest_level], level,
-				 &slot);
-		if (ret && slot > 0)
-			slot--;
-
-		bytenr = btrfs_node_blockptr(parent, slot);
-		if (nodes[level - 1] == bytenr)
-			break;
-
-		blocksize = btrfs_level_size(root, level - 1);
-		generation = btrfs_node_ptr_generation(parent, slot);
-		btrfs_node_key_to_cpu(eb, &key, slot);
-		key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
-
-		if (generation == trans->transid) {
-			eb = read_tree_block(root, bytenr, blocksize,
-					     generation);
-			btrfs_tree_lock(eb);
-			btrfs_set_lock_blocking(eb);
-		}
-
-		/*
-		 * if node keys match and node pointer hasn't been modified
-		 * in the running transaction, we can merge the path. for
-		 * blocks owened by reloc trees, the node pointer check is
-		 * skipped, this is because these blocks are fully controlled
-		 * by the space balance code, no one else can modify them.
-		 */
-		if (!nodes[level - 1] || !key_match ||
-		    (generation == trans->transid &&
-		     btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
-			if (level == 1 || level == lowest_level + 1) {
-				if (generation == trans->transid) {
-					btrfs_tree_unlock(eb);
-					free_extent_buffer(eb);
-				}
-				break;
-			}
-
-			if (generation != trans->transid) {
-				eb = read_tree_block(root, bytenr, blocksize,
-						generation);
-				btrfs_tree_lock(eb);
-				btrfs_set_lock_blocking(eb);
-			}
-
-			ret = btrfs_cow_block(trans, root, eb, parent, slot,
-					      &eb);
-			BUG_ON(ret);
-
-			if (root->root_key.objectid ==
-			    BTRFS_TREE_RELOC_OBJECTID) {
-				if (!nodes[level - 1]) {
-					nodes[level - 1] = eb->start;
-					memcpy(&node_keys[level - 1], &key,
-					       sizeof(node_keys[0]));
-				} else {
-					WARN_ON(1);
-				}
-			}
-
-			btrfs_tree_unlock(parent);
-			free_extent_buffer(parent);
-			parent = eb;
-			continue;
-		}
-
-		btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
-		btrfs_set_node_ptr_generation(parent, slot, trans->transid);
-		btrfs_mark_buffer_dirty(parent);
-
-		ret = btrfs_inc_extent_ref(trans, root,
-					nodes[level - 1],
-					blocksize, parent->start,
-					btrfs_header_owner(parent),
-					btrfs_header_generation(parent),
-					level - 1);
-		BUG_ON(ret);
-
-		/*
-		 * If the block was created in the running transaction,
-		 * it's possible this is the last reference to it, so we
-		 * should drop the subtree.
-		 */
-		if (generation == trans->transid) {
-			ret = btrfs_drop_subtree(trans, root, eb, parent);
-			BUG_ON(ret);
-			btrfs_tree_unlock(eb);
-			free_extent_buffer(eb);
-		} else {
-			ret = btrfs_free_extent(trans, root, bytenr,
-					blocksize, parent->start,
-					btrfs_header_owner(parent),
-					btrfs_header_generation(parent),
-					level - 1, 1);
-			BUG_ON(ret);
-		}
-		break;
-	}
-	btrfs_tree_unlock(parent);
-	free_extent_buffer(parent);
-	return 0;
-}
-
 /*
  * adjust the pointers going up the tree, starting at level
  * making sure the right key of each node is points to 'key'.
@@ -2021,9 +1993,6 @@ static int push_node_left(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(src);
 	btrfs_mark_buffer_dirty(dst);
 
-	ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
-	BUG_ON(ret);
-
 	return ret;
 }
 
@@ -2083,9 +2052,6 @@ static int balance_node_right(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(src);
 	btrfs_mark_buffer_dirty(dst);
 
-	ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
-	BUG_ON(ret);
-
 	return ret;
 }
 
@@ -2105,7 +2071,6 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 	struct extent_buffer *c;
 	struct extent_buffer *old;
 	struct btrfs_disk_key lower_key;
-	int ret;
 
 	BUG_ON(path->nodes[level]);
 	BUG_ON(path->nodes[level-1] != root->node);
@@ -2117,16 +2082,17 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 		btrfs_node_key(lower, &lower_key, 0);
 
 	c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
-				   root->root_key.objectid, trans->transid,
+				   root->root_key.objectid, &lower_key,
 				   level, root->node->start, 0);
 	if (IS_ERR(c))
 		return PTR_ERR(c);
 
-	memset_extent_buffer(c, 0, 0, root->nodesize);
+	memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
 	btrfs_set_header_nritems(c, 1);
 	btrfs_set_header_level(c, level);
 	btrfs_set_header_bytenr(c, c->start);
 	btrfs_set_header_generation(c, trans->transid);
+	btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
 	btrfs_set_header_owner(c, root->root_key.objectid);
 
 	write_extent_buffer(c, root->fs_info->fsid,
@@ -2151,12 +2117,6 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 	root->node = c;
 	spin_unlock(&root->node_lock);
 
-	ret = btrfs_update_extent_ref(trans, root, lower->start,
-				      lower->len, lower->start, c->start,
-				      root->root_key.objectid,
-				      trans->transid, level - 1);
-	BUG_ON(ret);
-
 	/* the super has an extra ref to root->node */
 	free_extent_buffer(old);
 
@@ -2233,7 +2193,7 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 		ret = insert_new_root(trans, root, path, level + 1);
 		if (ret)
 			return ret;
-	} else if (!trans->transaction->delayed_refs.flushing) {
+	} else {
 		ret = push_nodes_for_insert(trans, root, path, level);
 		c = path->nodes[level];
 		if (!ret && btrfs_header_nritems(c) <
@@ -2244,20 +2204,21 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 	}
 
 	c_nritems = btrfs_header_nritems(c);
+	mid = (c_nritems + 1) / 2;
+	btrfs_node_key(c, &disk_key, mid);
 
-	split = btrfs_alloc_free_block(trans, root, root->nodesize,
-					path->nodes[level + 1]->start,
+	split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
 					root->root_key.objectid,
-					trans->transid, level, c->start, 0);
+					&disk_key, level, c->start, 0);
 	if (IS_ERR(split))
 		return PTR_ERR(split);
 
-	btrfs_set_header_flags(split, btrfs_header_flags(c));
+	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);
 	btrfs_set_header_generation(split, trans->transid);
+	btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
 	btrfs_set_header_owner(split, root->root_key.objectid);
-	btrfs_set_header_flags(split, 0);
 	write_extent_buffer(split, root->fs_info->fsid,
 			    (unsigned long)btrfs_header_fsid(split),
 			    BTRFS_FSID_SIZE);
@@ -2265,7 +2226,6 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 			    (unsigned long)btrfs_header_chunk_tree_uuid(split),
 			    BTRFS_UUID_SIZE);
 
-	mid = (c_nritems + 1) / 2;
 
 	copy_extent_buffer(split, c,
 			   btrfs_node_key_ptr_offset(0),
@@ -2278,16 +2238,12 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(c);
 	btrfs_mark_buffer_dirty(split);
 
-	btrfs_node_key(split, &disk_key, 0);
 	wret = insert_ptr(trans, root, path, &disk_key, split->start,
 			  path->slots[level + 1] + 1,
 			  level + 1);
 	if (wret)
 		ret = wret;
 
-	ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
-	BUG_ON(ret);
-
 	if (path->slots[level] >= mid) {
 		path->slots[level] -= mid;
 		btrfs_tree_unlock(c);
@@ -2360,7 +2316,6 @@ static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
 	u32 right_nritems;
 	u32 data_end;
 	u32 this_item_size;
-	int ret;
 
 	if (empty)
 		nr = 0;
@@ -2473,9 +2428,6 @@ static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
 		btrfs_mark_buffer_dirty(left);
 	btrfs_mark_buffer_dirty(right);
 
-	ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
-	BUG_ON(ret);
-
 	btrfs_item_key(right, &disk_key, 0);
 	btrfs_set_node_key(upper, &disk_key, slot + 1);
 	btrfs_mark_buffer_dirty(upper);
@@ -2720,10 +2672,6 @@ static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
 	if (right_nritems)
 		btrfs_mark_buffer_dirty(right);
 
-	ret = btrfs_update_ref(trans, root, right, left,
-			       old_left_nritems, push_items);
-	BUG_ON(ret);
-
 	btrfs_item_key(right, &disk_key, 0);
 	wret = fixup_low_keys(trans, root, path, &disk_key, 1);
 	if (wret)
@@ -2880,9 +2828,6 @@ static noinline int copy_for_split(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(l);
 	BUG_ON(path->slots[0] != slot);
 
-	ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
-	BUG_ON(ret);
-
 	if (mid <= slot) {
 		btrfs_tree_unlock(path->nodes[0]);
 		free_extent_buffer(path->nodes[0]);
@@ -2911,6 +2856,7 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
 			       struct btrfs_path *path, int data_size,
 			       int extend)
 {
+	struct btrfs_disk_key disk_key;
 	struct extent_buffer *l;
 	u32 nritems;
 	int mid;
@@ -2918,12 +2864,11 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
 	struct extent_buffer *right;
 	int ret = 0;
 	int wret;
-	int double_split;
+	int split;
 	int num_doubles = 0;
 
 	/* first try to make some room by pushing left and right */
-	if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY &&
-	    !trans->transaction->delayed_refs.flushing) {
+	if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
 		wret = push_leaf_right(trans, root, path, data_size, 0);
 		if (wret < 0)
 			return wret;
@@ -2945,16 +2890,53 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
 			return ret;
 	}
 again:
-	double_split = 0;
+	split = 1;
 	l = path->nodes[0];
 	slot = path->slots[0];
 	nritems = btrfs_header_nritems(l);
 	mid = (nritems + 1) / 2;
 
-	right = btrfs_alloc_free_block(trans, root, root->leafsize,
-					path->nodes[1]->start,
+	if (mid <= slot) {
+		if (nritems == 1 ||
+		    leaf_space_used(l, mid, nritems - mid) + data_size >
+			BTRFS_LEAF_DATA_SIZE(root)) {
+			if (slot >= nritems) {
+				split = 0;
+			} else {
+				mid = slot;
+				if (mid != nritems &&
+				    leaf_space_used(l, mid, nritems - mid) +
+				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
+					split = 2;
+				}
+			}
+		}
+	} else {
+		if (leaf_space_used(l, 0, mid) + data_size >
+			BTRFS_LEAF_DATA_SIZE(root)) {
+			if (!extend && data_size && slot == 0) {
+				split = 0;
+			} else if ((extend || !data_size) && slot == 0) {
+				mid = 1;
+			} else {
+				mid = slot;
+				if (mid != nritems &&
+				    leaf_space_used(l, mid, nritems - mid) +
+				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
+					split = 2 ;
+				}
+			}
+		}
+	}
+
+	if (split == 0)
+		btrfs_cpu_key_to_disk(&disk_key, ins_key);
+	else
+		btrfs_item_key(l, &disk_key, mid);
+
+	right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
 					root->root_key.objectid,
-					trans->transid, 0, l->start, 0);
+					&disk_key, 0, l->start, 0);
 	if (IS_ERR(right)) {
 		BUG_ON(1);
 		return PTR_ERR(right);
@@ -2963,6 +2945,7 @@ again:
 	memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
 	btrfs_set_header_bytenr(right, right->start);
 	btrfs_set_header_generation(right, trans->transid);
+	btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
 	btrfs_set_header_owner(right, root->root_key.objectid);
 	btrfs_set_header_level(right, 0);
 	write_extent_buffer(right, root->fs_info->fsid,
@@ -2973,79 +2956,47 @@ again:
 			    (unsigned long)btrfs_header_chunk_tree_uuid(right),
 			    BTRFS_UUID_SIZE);
 
-	if (mid <= slot) {
-		if (nritems == 1 ||
-		    leaf_space_used(l, mid, nritems - mid) + data_size >
-			BTRFS_LEAF_DATA_SIZE(root)) {
-			if (slot >= nritems) {
-				struct btrfs_disk_key disk_key;
-
-				btrfs_cpu_key_to_disk(&disk_key, ins_key);
-				btrfs_set_header_nritems(right, 0);
-				wret = insert_ptr(trans, root, path,
-						  &disk_key, right->start,
-						  path->slots[1] + 1, 1);
-				if (wret)
-					ret = wret;
+	if (split == 0) {
+		if (mid <= slot) {
+			btrfs_set_header_nritems(right, 0);
+			wret = insert_ptr(trans, root, path,
+					  &disk_key, right->start,
+					  path->slots[1] + 1, 1);
+			if (wret)
+				ret = wret;
 
-				btrfs_tree_unlock(path->nodes[0]);
-				free_extent_buffer(path->nodes[0]);
-				path->nodes[0] = right;
-				path->slots[0] = 0;
-				path->slots[1] += 1;
-				btrfs_mark_buffer_dirty(right);
-				return ret;
-			}
-			mid = slot;
-			if (mid != nritems &&
-			    leaf_space_used(l, mid, nritems - mid) +
-			    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
-				double_split = 1;
-			}
-		}
-	} else {
-		if (leaf_space_used(l, 0, mid) + data_size >
-			BTRFS_LEAF_DATA_SIZE(root)) {
-			if (!extend && data_size && slot == 0) {
-				struct btrfs_disk_key disk_key;
-
-				btrfs_cpu_key_to_disk(&disk_key, ins_key);
-				btrfs_set_header_nritems(right, 0);
-				wret = insert_ptr(trans, root, path,
-						  &disk_key,
-						  right->start,
-						  path->slots[1], 1);
+			btrfs_tree_unlock(path->nodes[0]);
+			free_extent_buffer(path->nodes[0]);
+			path->nodes[0] = right;
+			path->slots[0] = 0;
+			path->slots[1] += 1;
+		} else {
+			btrfs_set_header_nritems(right, 0);
+			wret = insert_ptr(trans, root, path,
+					  &disk_key,
+					  right->start,
+					  path->slots[1], 1);
+			if (wret)
+				ret = wret;
+			btrfs_tree_unlock(path->nodes[0]);
+			free_extent_buffer(path->nodes[0]);
+			path->nodes[0] = right;
+			path->slots[0] = 0;
+			if (path->slots[1] == 0) {
+				wret = fixup_low_keys(trans, root,
+						path, &disk_key, 1);
 				if (wret)
 					ret = wret;
-				btrfs_tree_unlock(path->nodes[0]);
-				free_extent_buffer(path->nodes[0]);
-				path->nodes[0] = right;
-				path->slots[0] = 0;
-				if (path->slots[1] == 0) {
-					wret = fixup_low_keys(trans, root,
-						      path, &disk_key, 1);
-					if (wret)
-						ret = wret;
-				}
-				btrfs_mark_buffer_dirty(right);
-				return ret;
-			} else if ((extend || !data_size) && slot == 0) {
-				mid = 1;
-			} else {
-				mid = slot;
-				if (mid != nritems &&
-				    leaf_space_used(l, mid, nritems - mid) +
-				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
-					double_split = 1;
-				}
 			}
 		}
+		btrfs_mark_buffer_dirty(right);
+		return ret;
 	}
 
 	ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
 	BUG_ON(ret);
 
-	if (double_split) {
+	if (split == 2) {
 		BUG_ON(num_doubles != 0);
 		num_doubles++;
 		goto again;
@@ -3447,7 +3398,7 @@ int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
 		/* figure out how many keys we can insert in here */
 		total_data = data_size[0];
 		for (i = 1; i < nr; i++) {
-			if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
+			if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
 				break;
 			total_data += data_size[i];
 		}
@@ -3745,9 +3696,7 @@ static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 
 /*
  * a helper function to delete the leaf pointed to by path->slots[1] and
- * path->nodes[1].  bytenr is the node block pointer, but since the callers
- * already know it, it is faster to have them pass it down than to
- * read it out of the node again.
+ * path->nodes[1].
  *
  * This deletes the pointer in path->nodes[1] and frees the leaf
  * block extent.  zero is returned if it all worked out, < 0 otherwise.
@@ -3755,15 +3704,14 @@ static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  * The path must have already been setup for deleting the leaf, including
  * all the proper balancing.  path->nodes[1] must be locked.
  */
-noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root,
-			    struct btrfs_path *path, u64 bytenr)
+static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct btrfs_path *path,
+				   struct extent_buffer *leaf)
 {
 	int ret;
-	u64 root_gen = btrfs_header_generation(path->nodes[1]);
-	u64 parent_start = path->nodes[1]->start;
-	u64 parent_owner = btrfs_header_owner(path->nodes[1]);
 
+	WARN_ON(btrfs_header_generation(leaf) != trans->transid);
 	ret = del_ptr(trans, root, path, 1, path->slots[1]);
 	if (ret)
 		return ret;
@@ -3774,10 +3722,8 @@ noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
 	 */
 	btrfs_unlock_up_safe(path, 0);
 
-	ret = btrfs_free_extent(trans, root, bytenr,
-				btrfs_level_size(root, 0),
-				parent_start, parent_owner,
-				root_gen, 0, 1);
+	ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
+				0, root->root_key.objectid, 0, 0);
 	return ret;
 }
 /*
@@ -3845,7 +3791,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		if (leaf == root->node) {
 			btrfs_set_header_level(leaf, 0);
 		} else {
-			ret = btrfs_del_leaf(trans, root, path, leaf->start);
+			ret = btrfs_del_leaf(trans, root, path, leaf);
 			BUG_ON(ret);
 		}
 	} else {
@@ -3861,8 +3807,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		}
 
 		/* delete the leaf if it is mostly empty */
-		if (used < BTRFS_LEAF_DATA_SIZE(root) / 4 &&
-		    !trans->transaction->delayed_refs.flushing) {
+		if (used < BTRFS_LEAF_DATA_SIZE(root) / 2) {
 			/* push_leaf_left fixes the path.
 			 * make sure the path still points to our leaf
 			 * for possible call to del_ptr below
@@ -3884,8 +3829,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 
 			if (btrfs_header_nritems(leaf) == 0) {
 				path->slots[1] = slot;
-				ret = btrfs_del_leaf(trans, root, path,
-						     leaf->start);
+				ret = btrfs_del_leaf(trans, root, path, leaf);
 				BUG_ON(ret);
 				free_extent_buffer(leaf);
 			} else {

fs/btrfs/ctree.h

@@ -45,6 +45,8 @@ struct btrfs_ordered_sum;
 
 #define BTRFS_MAX_LEVEL 8
 
+#define BTRFS_COMPAT_EXTENT_TREE_V0
+
 /*
  * files bigger than this get some pre-flushing when they are added
  * to the ordered operations list.  That way we limit the total
@@ -267,7 +269,18 @@ static inline unsigned long btrfs_chunk_item_size(int num_stripes)
 }
 
 #define BTRFS_FSID_SIZE 16
-#define BTRFS_HEADER_FLAG_WRITTEN (1 << 0)
+#define BTRFS_HEADER_FLAG_WRITTEN	(1ULL << 0)
+#define BTRFS_HEADER_FLAG_RELOC		(1ULL << 1)
+#define BTRFS_SUPER_FLAG_SEEDING	(1ULL << 32)
+#define BTRFS_SUPER_FLAG_METADUMP	(1ULL << 33)
+
+#define BTRFS_BACKREF_REV_MAX		256
+#define BTRFS_BACKREF_REV_SHIFT		56
+#define BTRFS_BACKREF_REV_MASK		(((u64)BTRFS_BACKREF_REV_MAX - 1) << \
+					 BTRFS_BACKREF_REV_SHIFT)
+
+#define BTRFS_OLD_BACKREF_REV		0
+#define BTRFS_MIXED_BACKREF_REV		1
 
 /*
  * every tree block (leaf or node) starts with this header.
@@ -296,7 +309,6 @@ struct btrfs_header {
 					sizeof(struct btrfs_item) - \
 					sizeof(struct btrfs_file_extent_item))
 
-#define BTRFS_SUPER_FLAG_SEEDING (1ULL << 32)
 
 /*
  * this is a very generous portion of the super block, giving us
@@ -355,9 +367,12 @@ struct btrfs_super_block {
  * Compat flags that we support.  If any incompat flags are set other than the
  * ones specified below then we will fail to mount
  */
-#define BTRFS_FEATURE_COMPAT_SUPP	0x0
-#define BTRFS_FEATURE_COMPAT_RO_SUPP	0x0
-#define BTRFS_FEATURE_INCOMPAT_SUPP	0x0
+#define BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF	(1ULL << 0)
+
+#define BTRFS_FEATURE_COMPAT_SUPP		0ULL
+#define BTRFS_FEATURE_COMPAT_RO_SUPP		0ULL
+#define BTRFS_FEATURE_INCOMPAT_SUPP		\
+	BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
 
 /*
  * A leaf is full of items. offset and size tell us where to find
@@ -421,23 +436,65 @@ struct btrfs_path {
 	unsigned int keep_locks:1;
 	unsigned int skip_locking:1;
 	unsigned int leave_spinning:1;
+	unsigned int search_commit_root:1;
 };
 
 /*
  * items in the extent btree are used to record the objectid of the
  * owner of the block and the number of references
  */
+
 struct btrfs_extent_item {
+	__le64 refs;
+	__le64 generation;
+	__le64 flags;
+} __attribute__ ((__packed__));
+
+struct btrfs_extent_item_v0 {
 	__le32 refs;
 } __attribute__ ((__packed__));
 
-struct btrfs_extent_ref {
+#define BTRFS_MAX_EXTENT_ITEM_SIZE(r) ((BTRFS_LEAF_DATA_SIZE(r) >> 4) - \
+					sizeof(struct btrfs_item))
+
+#define BTRFS_EXTENT_FLAG_DATA		(1ULL << 0)
+#define BTRFS_EXTENT_FLAG_TREE_BLOCK	(1ULL << 1)
+
+/* following flags only apply to tree blocks */
+
+/* use full backrefs for extent pointers in the block */
+#define BTRFS_BLOCK_FLAG_FULL_BACKREF	(1ULL << 8)
+
+struct btrfs_tree_block_info {
+	struct btrfs_disk_key key;
+	u8 level;
+} __attribute__ ((__packed__));
+
+struct btrfs_extent_data_ref {
+	__le64 root;
+	__le64 objectid;
+	__le64 offset;
+	__le32 count;
+} __attribute__ ((__packed__));
+
+struct btrfs_shared_data_ref {
+	__le32 count;
+} __attribute__ ((__packed__));
+
+struct btrfs_extent_inline_ref {
+	u8 type;
+	u64 offset;
+} __attribute__ ((__packed__));
+
+/* old style backrefs item */
+struct btrfs_extent_ref_v0 {
 	__le64 root;
 	__le64 generation;
 	__le64 objectid;
-	__le32 num_refs;
+	__le32 count;
 } __attribute__ ((__packed__));
 
+
 /* dev extents record free space on individual devices.  The owner
  * field points back to the chunk allocation mapping tree that allocated
  * the extent.  The chunk tree uuid field is a way to double check the owner
@@ -695,12 +752,7 @@ struct btrfs_block_group_cache {
 	struct list_head cluster_list;
 };
 
-struct btrfs_leaf_ref_tree {
-	struct rb_root root;
-	struct list_head list;
-	spinlock_t lock;
-};
-
+struct reloc_control;
 struct btrfs_device;
 struct btrfs_fs_devices;
 struct btrfs_fs_info {
@@ -831,18 +883,11 @@ struct btrfs_fs_info {
 	struct task_struct *cleaner_kthread;
 	int thread_pool_size;
 
-	/* tree relocation relocated fields */
-	struct list_head dead_reloc_roots;
-	struct btrfs_leaf_ref_tree reloc_ref_tree;
-	struct btrfs_leaf_ref_tree shared_ref_tree;
-
 	struct kobject super_kobj;
 	struct completion kobj_unregister;
 	int do_barriers;
 	int closing;
 	int log_root_recovering;
-	atomic_t throttles;
-	atomic_t throttle_gen;
 
 	u64 total_pinned;
 
@@ -861,6 +906,8 @@ struct btrfs_fs_info {
 	 */
 	struct list_head space_info;
 
+	struct reloc_control *reloc_ctl;
+
 	spinlock_t delalloc_lock;
 	spinlock_t new_trans_lock;
 	u64 delalloc_bytes;
@@ -891,7 +938,6 @@ struct btrfs_fs_info {
  * in ram representation of the tree.  extent_root is used for all allocations
  * and for the extent tree extent_root root.
  */
-struct btrfs_dirty_root;
 struct btrfs_root {
 	struct extent_buffer *node;
 
@@ -899,9 +945,6 @@ struct btrfs_root {
 	spinlock_t node_lock;
 
 	struct extent_buffer *commit_root;
-	struct btrfs_leaf_ref_tree *ref_tree;
-	struct btrfs_leaf_ref_tree ref_tree_struct;
-	struct btrfs_dirty_root *dirty_root;
 	struct btrfs_root *log_root;
 	struct btrfs_root *reloc_root;
 
@@ -952,10 +995,15 @@ struct btrfs_root {
 	/* the dirty list is only used by non-reference counted roots */
 	struct list_head dirty_list;
 
+	struct list_head root_list;
+
 	spinlock_t list_lock;
-	struct list_head dead_list;
 	struct list_head orphan_list;
 
+	spinlock_t inode_lock;
+	/* red-black tree that keeps track of in-memory inodes */
+	struct rb_root inode_tree;
+
 	/*
 	 * right now this just gets used so that a root has its own devid
 	 * for stat.  It may be used for more later
@@ -1017,7 +1065,16 @@ struct btrfs_root {
  * are used, and how many references there are to each block
  */
 #define BTRFS_EXTENT_ITEM_KEY	168
-#define BTRFS_EXTENT_REF_KEY	180
+
+#define BTRFS_TREE_BLOCK_REF_KEY	176
+
+#define BTRFS_EXTENT_DATA_REF_KEY	178
+
+#define BTRFS_EXTENT_REF_V0_KEY		180
+
+#define BTRFS_SHARED_BLOCK_REF_KEY	182
+
+#define BTRFS_SHARED_DATA_REF_KEY	184
 
 /*
  * block groups give us hints into the extent allocation trees.  Which
@@ -1043,6 +1100,8 @@ struct btrfs_root {
 #define BTRFS_MOUNT_COMPRESS		(1 << 5)
 #define BTRFS_MOUNT_NOTREELOG           (1 << 6)
 #define BTRFS_MOUNT_FLUSHONCOMMIT       (1 << 7)
+#define BTRFS_MOUNT_SSD_SPREAD		(1 << 8)
+#define BTRFS_MOUNT_NOSSD		(1 << 9)
 
 #define btrfs_clear_opt(o, opt)		((o) &= ~BTRFS_MOUNT_##opt)
 #define btrfs_set_opt(o, opt)		((o) |= BTRFS_MOUNT_##opt)
@@ -1056,12 +1115,14 @@ struct btrfs_root {
 #define BTRFS_INODE_READONLY		(1 << 2)
 #define BTRFS_INODE_NOCOMPRESS		(1 << 3)
 #define BTRFS_INODE_PREALLOC		(1 << 4)
-#define btrfs_clear_flag(inode, flag)	(BTRFS_I(inode)->flags &= \
-					 ~BTRFS_INODE_##flag)
-#define btrfs_set_flag(inode, flag)	(BTRFS_I(inode)->flags |= \
-					 BTRFS_INODE_##flag)
-#define btrfs_test_flag(inode, flag)	(BTRFS_I(inode)->flags & \
-					 BTRFS_INODE_##flag)
+#define BTRFS_INODE_SYNC		(1 << 5)
+#define BTRFS_INODE_IMMUTABLE		(1 << 6)
+#define BTRFS_INODE_APPEND		(1 << 7)
+#define BTRFS_INODE_NODUMP		(1 << 8)
+#define BTRFS_INODE_NOATIME		(1 << 9)
+#define BTRFS_INODE_DIRSYNC		(1 << 10)
+
+
 /* some macros to generate set/get funcs for the struct fields.  This
  * assumes there is a lefoo_to_cpu for every type, so lets make a simple
  * one for u8:
@@ -1317,24 +1378,67 @@ static inline u8 *btrfs_dev_extent_chunk_tree_uuid(struct btrfs_dev_extent *dev)
 	return (u8 *)((unsigned long)dev + ptr);
 }
 
-/* struct btrfs_extent_ref */
-BTRFS_SETGET_FUNCS(ref_root, struct btrfs_extent_ref, root, 64);
-BTRFS_SETGET_FUNCS(ref_generation, struct btrfs_extent_ref, generation, 64);
-BTRFS_SETGET_FUNCS(ref_objectid, struct btrfs_extent_ref, objectid, 64);
-BTRFS_SETGET_FUNCS(ref_num_refs, struct btrfs_extent_ref, num_refs, 32);
+BTRFS_SETGET_FUNCS(extent_refs, struct btrfs_extent_item, refs, 64);
+BTRFS_SETGET_FUNCS(extent_generation, struct btrfs_extent_item,
+		   generation, 64);
+BTRFS_SETGET_FUNCS(extent_flags, struct btrfs_extent_item, flags, 64);
 
-BTRFS_SETGET_STACK_FUNCS(stack_ref_root, struct btrfs_extent_ref, root, 64);
-BTRFS_SETGET_STACK_FUNCS(stack_ref_generation, struct btrfs_extent_ref,
-			 generation, 64);
-BTRFS_SETGET_STACK_FUNCS(stack_ref_objectid, struct btrfs_extent_ref,
-			 objectid, 64);
-BTRFS_SETGET_STACK_FUNCS(stack_ref_num_refs, struct btrfs_extent_ref,
-			 num_refs, 32);
+BTRFS_SETGET_FUNCS(extent_refs_v0, struct btrfs_extent_item_v0, refs, 32);
+
+
+BTRFS_SETGET_FUNCS(tree_block_level, struct btrfs_tree_block_info, level, 8);
+
+static inline void btrfs_tree_block_key(struct extent_buffer *eb,
+					struct btrfs_tree_block_info *item,
+					struct btrfs_disk_key *key)
+{
+	read_eb_member(eb, item, struct btrfs_tree_block_info, key, key);
+}
+
+static inline void btrfs_set_tree_block_key(struct extent_buffer *eb,
+					    struct btrfs_tree_block_info *item,
+					    struct btrfs_disk_key *key)
+{
+	write_eb_member(eb, item, struct btrfs_tree_block_info, key, key);
+}
+
+BTRFS_SETGET_FUNCS(extent_data_ref_root, struct btrfs_extent_data_ref,
+		   root, 64);
+BTRFS_SETGET_FUNCS(extent_data_ref_objectid, struct btrfs_extent_data_ref,
+		   objectid, 64);
+BTRFS_SETGET_FUNCS(extent_data_ref_offset, struct btrfs_extent_data_ref,
+		   offset, 64);
+BTRFS_SETGET_FUNCS(extent_data_ref_count, struct btrfs_extent_data_ref,
+		   count, 32);
+
+BTRFS_SETGET_FUNCS(shared_data_ref_count, struct btrfs_shared_data_ref,
+		   count, 32);
 
-/* struct btrfs_extent_item */
-BTRFS_SETGET_FUNCS(extent_refs, struct btrfs_extent_item, refs, 32);
-BTRFS_SETGET_STACK_FUNCS(stack_extent_refs, struct btrfs_extent_item,
-			 refs, 32);
+BTRFS_SETGET_FUNCS(extent_inline_ref_type, struct btrfs_extent_inline_ref,
+		   type, 8);
+BTRFS_SETGET_FUNCS(extent_inline_ref_offset, struct btrfs_extent_inline_ref,
+		   offset, 64);
+
+static inline u32 btrfs_extent_inline_ref_size(int type)
+{
+	if (type == BTRFS_TREE_BLOCK_REF_KEY ||
+	    type == BTRFS_SHARED_BLOCK_REF_KEY)
+		return sizeof(struct btrfs_extent_inline_ref);
+	if (type == BTRFS_SHARED_DATA_REF_KEY)
+		return sizeof(struct btrfs_shared_data_ref) +
+		       sizeof(struct btrfs_extent_inline_ref);
+	if (type == BTRFS_EXTENT_DATA_REF_KEY)
+		return sizeof(struct btrfs_extent_data_ref) +
+		       offsetof(struct btrfs_extent_inline_ref, offset);
+	BUG();
+	return 0;
+}
+
+BTRFS_SETGET_FUNCS(ref_root_v0, struct btrfs_extent_ref_v0, root, 64);
+BTRFS_SETGET_FUNCS(ref_generation_v0, struct btrfs_extent_ref_v0,
+		   generation, 64);
+BTRFS_SETGET_FUNCS(ref_objectid_v0, struct btrfs_extent_ref_v0, objectid, 64);
+BTRFS_SETGET_FUNCS(ref_count_v0, struct btrfs_extent_ref_v0, count, 32);
 
 /* struct btrfs_node */
 BTRFS_SETGET_FUNCS(key_blockptr, struct btrfs_key_ptr, blockptr, 64);
@@ -1558,6 +1662,21 @@ static inline int btrfs_clear_header_flag(struct extent_buffer *eb, u64 flag)
 	return (flags & flag) == flag;
 }
 
+static inline int btrfs_header_backref_rev(struct extent_buffer *eb)
+{
+	u64 flags = btrfs_header_flags(eb);
+	return flags >> BTRFS_BACKREF_REV_SHIFT;
+}
+
+static inline void btrfs_set_header_backref_rev(struct extent_buffer *eb,
+						int rev)
+{
+	u64 flags = btrfs_header_flags(eb);
+	flags &= ~BTRFS_BACKREF_REV_MASK;
+	flags |= (u64)rev << BTRFS_BACKREF_REV_SHIFT;
+	btrfs_set_header_flags(eb, flags);
+}
+
 static inline u8 *btrfs_header_fsid(struct extent_buffer *eb)
 {
 	unsigned long ptr = offsetof(struct btrfs_header, fsid);
@@ -1790,39 +1909,32 @@ int btrfs_update_pinned_extents(struct btrfs_root *root,
 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root, struct extent_buffer *leaf);
 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root, u64 objectid, u64 bytenr);
+			  struct btrfs_root *root,
+			  u64 objectid, u64 offset, u64 bytenr);
 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy);
 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 						 struct btrfs_fs_info *info,
 						 u64 bytenr);
+void btrfs_put_block_group(struct btrfs_block_group_cache *cache);
 u64 btrfs_find_block_group(struct btrfs_root *root,
 			   u64 search_start, u64 search_hint, int owner);
 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
-					     struct btrfs_root *root,
-					     u32 blocksize, u64 parent,
-					     u64 root_objectid,
-					     u64 ref_generation,
-					     int level,
-					     u64 hint,
-					     u64 empty_size);
+					struct btrfs_root *root, u32 blocksize,
+					u64 parent, u64 root_objectid,
+					struct btrfs_disk_key *key, int level,
+					u64 hint, u64 empty_size);
 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
 					    struct btrfs_root *root,
 					    u64 bytenr, u32 blocksize,
 					    int level);
-int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
-		       struct btrfs_root *root,
-		       u64 num_bytes, u64 parent, u64 min_bytes,
-		       u64 root_objectid, u64 ref_generation,
-		       u64 owner, u64 empty_size, u64 hint_byte,
-		       u64 search_end, struct btrfs_key *ins, u64 data);
-int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root, u64 parent,
-				u64 root_objectid, u64 ref_generation,
-				u64 owner, struct btrfs_key *ins);
-int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root, u64 parent,
-				u64 root_objectid, u64 ref_generation,
-				u64 owner, struct btrfs_key *ins);
+int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root,
+				     u64 root_objectid, u64 owner,
+				     u64 offset, struct btrfs_key *ins);
+int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   u64 root_objectid, u64 owner, u64 offset,
+				   struct btrfs_key *ins);
 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
 				  struct btrfs_root *root,
 				  u64 num_bytes, u64 min_alloc_size,
@@ -1830,18 +1942,18 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
 				  u64 search_end, struct btrfs_key *ins,
 				  u64 data);
 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		  struct extent_buffer *orig_buf, struct extent_buffer *buf,
-		  u32 *nr_extents);
-int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		    struct extent_buffer *buf, u32 nr_extents);
-int btrfs_update_ref(struct btrfs_trans_handle *trans,
-		     struct btrfs_root *root, struct extent_buffer *orig_buf,
-		     struct extent_buffer *buf, int start_slot, int nr);
+		  struct extent_buffer *buf, int full_backref);
+int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		  struct extent_buffer *buf, int full_backref);
+int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				u64 bytenr, u64 num_bytes, u64 flags,
+				int is_data);
 int btrfs_free_extent(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root,
 		      u64 bytenr, u64 num_bytes, u64 parent,
-		      u64 root_objectid, u64 ref_generation,
-		      u64 owner_objectid, int pin);
+		      u64 root_objectid, u64 owner, u64 offset);
+
 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len);
 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root,
@@ -1849,13 +1961,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root,
 			 u64 bytenr, u64 num_bytes, u64 parent,
-			 u64 root_objectid, u64 ref_generation,
-			 u64 owner_objectid);
-int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, u64 bytenr, u64 num_bytes,
-			    u64 orig_parent, u64 parent,
-			    u64 root_objectid, u64 ref_generation,
-			    u64 owner_objectid);
+			 u64 root_objectid, u64 owner, u64 offset);
+
 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 				    struct btrfs_root *root);
 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr);
@@ -1867,16 +1974,9 @@ 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_relocate_block_group(struct btrfs_root *root, u64 group_start);
-int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root);
-int btrfs_drop_dead_reloc_roots(struct btrfs_root *root);
-int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
-			       struct btrfs_root *root,
-			       struct extent_buffer *buf, u64 orig_start);
-int btrfs_add_dead_reloc_root(struct btrfs_root *root);
-int btrfs_cleanup_reloc_trees(struct btrfs_root *root);
-int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len);
+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);
@@ -1891,13 +1991,12 @@ void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
 			      u64 bytes);
 /* ctree.c */
+int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
+		     int level, int *slot);
+int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2);
 int btrfs_previous_item(struct btrfs_root *root,
 			struct btrfs_path *path, u64 min_objectid,
 			int type);
-int btrfs_merge_path(struct btrfs_trans_handle *trans,
-		     struct btrfs_root *root,
-		     struct btrfs_key *node_keys,
-		     u64 *nodes, int lowest_level);
 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
 			    struct btrfs_root *root, struct btrfs_path *path,
 			    struct btrfs_key *new_key);
@@ -1918,6 +2017,8 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root,
 		      struct extent_buffer *buf,
 		      struct extent_buffer **cow_ret, u64 new_root_objectid);
+int btrfs_block_can_be_shared(struct btrfs_root *root,
+			      struct extent_buffer *buf);
 int btrfs_extend_item(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, struct btrfs_path *path, u32 data_size);
 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
@@ -1944,9 +2045,6 @@ void btrfs_unlock_up_safe(struct btrfs_path *p, int level);
 
 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		   struct btrfs_path *path, int slot, int nr);
-int btrfs_del_leaf(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root,
-			    struct btrfs_path *path, u64 bytenr);
 static inline int btrfs_del_item(struct btrfs_trans_handle *trans,
 				 struct btrfs_root *root,
 				 struct btrfs_path *path)
@@ -2005,8 +2103,9 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
 			 btrfs_root_item *item, struct btrfs_key *key);
 int btrfs_search_root(struct btrfs_root *root, u64 search_start,
 		      u64 *found_objectid);
-int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid,
-			  struct btrfs_root *latest_root);
+int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid);
+int btrfs_set_root_node(struct btrfs_root_item *item,
+			struct extent_buffer *node);
 /* dir-item.c */
 int btrfs_insert_dir_item(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root, const char *name,
@@ -2139,7 +2238,6 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
 int btrfs_readpage(struct file *file, struct page *page);
 void btrfs_delete_inode(struct inode *inode);
 void btrfs_put_inode(struct inode *inode);
-void btrfs_read_locked_inode(struct inode *inode);
 int btrfs_write_inode(struct inode *inode, int wait);
 void btrfs_dirty_inode(struct inode *inode);
 struct inode *btrfs_alloc_inode(struct super_block *sb);
@@ -2147,12 +2245,8 @@ void btrfs_destroy_inode(struct inode *inode);
 int btrfs_init_cachep(void);
 void btrfs_destroy_cachep(void);
 long btrfs_ioctl_trans_end(struct file *file);
-struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
-			    struct btrfs_root *root, int wait);
-struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
-				struct btrfs_root *root);
 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
-			 struct btrfs_root *root, int *is_new);
+			 struct btrfs_root *root);
 int btrfs_commit_write(struct file *file, struct page *page,
 		       unsigned from, unsigned to);
 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
@@ -2168,6 +2262,8 @@ int btrfs_cont_expand(struct inode *inode, loff_t size);
 
 /* ioctl.c */
 long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
+void btrfs_update_iflags(struct inode *inode);
+void btrfs_inherit_iflags(struct inode *inode, struct inode *dir);
 
 /* file.c */
 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync);
@@ -2205,8 +2301,20 @@ int btrfs_parse_options(struct btrfs_root *root, char *options);
 int btrfs_sync_fs(struct super_block *sb, int wait);
 
 /* acl.c */
+#ifdef CONFIG_FS_POSIX_ACL
 int btrfs_check_acl(struct inode *inode, int mask);
+#else
+#define btrfs_check_acl NULL
+#endif
 int btrfs_init_acl(struct inode *inode, struct inode *dir);
 int btrfs_acl_chmod(struct inode *inode);
 
+/* relocation.c */
+int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start);
+int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root);
+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);
 #endif

fs/btrfs/delayed-ref.c

@@ -29,27 +29,87 @@
  * add extents in the middle of btrfs_search_slot, and it allows
  * us to buffer up frequently modified backrefs in an rb tree instead
  * of hammering updates on the extent allocation tree.
- *
- * Right now this code is only used for reference counted trees, but
- * the long term goal is to get rid of the similar code for delayed
- * extent tree modifications.
  */
 
 /*
- * entries in the rb tree are ordered by the byte number of the extent
- * and by the byte number of the parent block.
+ * compare two delayed tree backrefs with same bytenr and type
+ */
+static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2,
+			  struct btrfs_delayed_tree_ref *ref1)
+{
+	if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) {
+		if (ref1->root < ref2->root)
+			return -1;
+		if (ref1->root > ref2->root)
+			return 1;
+	} else {
+		if (ref1->parent < ref2->parent)
+			return -1;
+		if (ref1->parent > ref2->parent)
+			return 1;
+	}
+	return 0;
+}
+
+/*
+ * compare two delayed data backrefs with same bytenr and type
  */
-static int comp_entry(struct btrfs_delayed_ref_node *ref,
-		      u64 bytenr, u64 parent)
+static int comp_data_refs(struct btrfs_delayed_data_ref *ref2,
+			  struct btrfs_delayed_data_ref *ref1)
 {
-	if (bytenr < ref->bytenr)
+	if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
+		if (ref1->root < ref2->root)
+			return -1;
+		if (ref1->root > ref2->root)
+			return 1;
+		if (ref1->objectid < ref2->objectid)
+			return -1;
+		if (ref1->objectid > ref2->objectid)
+			return 1;
+		if (ref1->offset < ref2->offset)
+			return -1;
+		if (ref1->offset > ref2->offset)
+			return 1;
+	} else {
+		if (ref1->parent < ref2->parent)
+			return -1;
+		if (ref1->parent > ref2->parent)
+			return 1;
+	}
+	return 0;
+}
+
+/*
+ * entries in the rb tree are ordered by the byte number of the extent,
+ * type of the delayed backrefs and content of delayed backrefs.
+ */
+static int comp_entry(struct btrfs_delayed_ref_node *ref2,
+		      struct btrfs_delayed_ref_node *ref1)
+{
+	if (ref1->bytenr < ref2->bytenr)
 		return -1;
-	if (bytenr > ref->bytenr)
+	if (ref1->bytenr > ref2->bytenr)
 		return 1;
-	if (parent < ref->parent)
+	if (ref1->is_head && ref2->is_head)
+		return 0;
+	if (ref2->is_head)
 		return -1;
-	if (parent > ref->parent)
+	if (ref1->is_head)
 		return 1;
+	if (ref1->type < ref2->type)
+		return -1;
+	if (ref1->type > ref2->type)
+		return 1;
+	if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY ||
+	    ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) {
+		return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2),
+				      btrfs_delayed_node_to_tree_ref(ref1));
+	} else if (ref1->type == BTRFS_EXTENT_DATA_REF_KEY ||
+		   ref1->type == BTRFS_SHARED_DATA_REF_KEY) {
+		return comp_data_refs(btrfs_delayed_node_to_data_ref(ref2),
+				      btrfs_delayed_node_to_data_ref(ref1));
+	}
+	BUG();
 	return 0;
 }
 
@@ -59,20 +119,21 @@ static int comp_entry(struct btrfs_delayed_ref_node *ref,
  * inserted.
  */
 static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
-						  u64 bytenr, u64 parent,
 						  struct rb_node *node)
 {
 	struct rb_node **p = &root->rb_node;
 	struct rb_node *parent_node = NULL;
 	struct btrfs_delayed_ref_node *entry;
+	struct btrfs_delayed_ref_node *ins;
 	int cmp;
 
+	ins = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
 	while (*p) {
 		parent_node = *p;
 		entry = rb_entry(parent_node, struct btrfs_delayed_ref_node,
 				 rb_node);
 
-		cmp = comp_entry(entry, bytenr, parent);
+		cmp = comp_entry(entry, ins);
 		if (cmp < 0)
 			p = &(*p)->rb_left;
 		else if (cmp > 0)
@@ -81,18 +142,17 @@ static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
 			return entry;
 	}
 
-	entry = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
 	rb_link_node(node, parent_node, p);
 	rb_insert_color(node, root);
 	return NULL;
 }
 
 /*
- * find an entry based on (bytenr,parent).  This returns the delayed
- * ref if it was able to find one, or NULL if nothing was in that spot
+ * find an head entry based on bytenr. This returns the delayed ref
+ * head if it was able to find one, or NULL if nothing was in that spot
  */
-static struct btrfs_delayed_ref_node *tree_search(struct rb_root *root,
-				  u64 bytenr, u64 parent,
+static struct btrfs_delayed_ref_node *find_ref_head(struct rb_root *root,
+				  u64 bytenr,
 				  struct btrfs_delayed_ref_node **last)
 {
 	struct rb_node *n = root->rb_node;
@@ -105,7 +165,15 @@ static struct btrfs_delayed_ref_node *tree_search(struct rb_root *root,
 		if (last)
 			*last = entry;
 
-		cmp = comp_entry(entry, bytenr, parent);
+		if (bytenr < entry->bytenr)
+			cmp = -1;
+		else if (bytenr > entry->bytenr)
+			cmp = 1;
+		else if (!btrfs_delayed_ref_is_head(entry))
+			cmp = 1;
+		else
+			cmp = 0;
+
 		if (cmp < 0)
 			n = n->rb_left;
 		else if (cmp > 0)
@@ -154,7 +222,7 @@ int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
 		node = rb_first(&delayed_refs->root);
 	} else {
 		ref = NULL;
-		tree_search(&delayed_refs->root, start, (u64)-1, &ref);
+		find_ref_head(&delayed_refs->root, start, &ref);
 		if (ref) {
 			struct btrfs_delayed_ref_node *tmp;
 
@@ -234,7 +302,7 @@ int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr)
 	delayed_refs = &trans->transaction->delayed_refs;
 	spin_lock(&delayed_refs->lock);
 
-	ref = tree_search(&delayed_refs->root, bytenr, (u64)-1, NULL);
+	ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
 	if (ref) {
 		prev_node = rb_prev(&ref->rb_node);
 		if (!prev_node)
@@ -250,25 +318,28 @@ out:
 }
 
 /*
- * helper function to lookup reference count
+ * 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.  This way you
- * can check to see what the reference count would be if all of the
- * delayed refs are processed.
+ * 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_ref(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, u64 bytenr,
-			    u64 num_bytes, u32 *refs)
+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 extent_buffer *leaf;
 	struct btrfs_extent_item *ei;
+	struct extent_buffer *leaf;
 	struct btrfs_key key;
-	u32 num_refs;
+	u32 item_size;
+	u64 num_refs;
+	u64 extent_flags;
 	int ret;
 
 	path = btrfs_alloc_path();
@@ -287,37 +358,60 @@ again:
 
 	if (ret == 0) {
 		leaf = path->nodes[0];
-		ei = btrfs_item_ptr(leaf, path->slots[0],
-				    struct btrfs_extent_item);
-		num_refs = btrfs_extent_refs(leaf, ei);
+		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 = tree_search(&delayed_refs->root, bytenr, (u64)-1, NULL);
+	ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
 	if (ref) {
 		head = btrfs_delayed_node_to_head(ref);
-		if (mutex_trylock(&head->mutex)) {
-			num_refs += ref->ref_mod;
-			mutex_unlock(&head->mutex);
-			*refs = num_refs;
-			goto out;
-		}
+		if (!mutex_trylock(&head->mutex)) {
+			atomic_inc(&ref->refs);
+			spin_unlock(&delayed_refs->lock);
 
-		atomic_inc(&ref->refs);
-		spin_unlock(&delayed_refs->lock);
+			btrfs_release_path(root->fs_info->extent_root, path);
 
-		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);
 
-		mutex_lock(&head->mutex);
+		num_refs += ref->ref_mod;
 		mutex_unlock(&head->mutex);
-		btrfs_put_delayed_ref(ref);
-		goto again;
-	} else {
-		*refs = num_refs;
 	}
+	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);
@@ -338,16 +432,7 @@ update_existing_ref(struct btrfs_trans_handle *trans,
 		    struct btrfs_delayed_ref_node *existing,
 		    struct btrfs_delayed_ref_node *update)
 {
-	struct btrfs_delayed_ref *existing_ref;
-	struct btrfs_delayed_ref *ref;
-
-	existing_ref = btrfs_delayed_node_to_ref(existing);
-	ref = btrfs_delayed_node_to_ref(update);
-
-	if (ref->pin)
-		existing_ref->pin = 1;
-
-	if (ref->action != existing_ref->action) {
+	if (update->action != existing->action) {
 		/*
 		 * this is effectively undoing either an add or a
 		 * drop.  We decrement the ref_mod, and if it goes
@@ -363,20 +448,13 @@ update_existing_ref(struct btrfs_trans_handle *trans,
 			delayed_refs->num_entries--;
 			if (trans->delayed_ref_updates)
 				trans->delayed_ref_updates--;
+		} else {
+			WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
+				existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
 		}
 	} else {
-		if (existing_ref->action == BTRFS_ADD_DELAYED_REF) {
-			/* if we're adding refs, make sure all the
-			 * details match up.  The extent could
-			 * have been totally freed and reallocated
-			 * by a different owner before the delayed
-			 * ref entries were removed.
-			 */
-			existing_ref->owner_objectid = ref->owner_objectid;
-			existing_ref->generation = ref->generation;
-			existing_ref->root = ref->root;
-			existing->num_bytes = update->num_bytes;
-		}
+		WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
+			existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
 		/*
 		 * the action on the existing ref matches
 		 * the action on the ref we're trying to add.
@@ -401,6 +479,7 @@ update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
 
 	existing_ref = btrfs_delayed_node_to_head(existing);
 	ref = btrfs_delayed_node_to_head(update);
+	BUG_ON(existing_ref->is_data != ref->is_data);
 
 	if (ref->must_insert_reserved) {
 		/* if the extent was freed and then
@@ -420,6 +499,24 @@ update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
 
 	}
 
+	if (ref->extent_op) {
+		if (!existing_ref->extent_op) {
+			existing_ref->extent_op = ref->extent_op;
+		} else {
+			if (ref->extent_op->update_key) {
+				memcpy(&existing_ref->extent_op->key,
+				       &ref->extent_op->key,
+				       sizeof(ref->extent_op->key));
+				existing_ref->extent_op->update_key = 1;
+			}
+			if (ref->extent_op->update_flags) {
+				existing_ref->extent_op->flags_to_set |=
+					ref->extent_op->flags_to_set;
+				existing_ref->extent_op->update_flags = 1;
+			}
+			kfree(ref->extent_op);
+		}
+	}
 	/*
 	 * update the reference mod on the head to reflect this new operation
 	 */
@@ -427,19 +524,16 @@ update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
 }
 
 /*
- * helper function to actually insert a delayed ref into the rbtree.
+ * helper function to actually insert a head node into the rbtree.
  * this does all the dirty work in terms of maintaining the correct
- * overall modification count in the head node and properly dealing
- * with updating existing nodes as new modifications are queued.
+ * overall modification count.
  */
-static noinline int __btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
-			  struct btrfs_delayed_ref_node *ref,
-			  u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
-			  u64 ref_generation, u64 owner_objectid, int action,
-			  int pin)
+static noinline int add_delayed_ref_head(struct btrfs_trans_handle *trans,
+					struct btrfs_delayed_ref_node *ref,
+					u64 bytenr, u64 num_bytes,
+					int action, int is_data)
 {
 	struct btrfs_delayed_ref_node *existing;
-	struct btrfs_delayed_ref *full_ref;
 	struct btrfs_delayed_ref_head *head_ref = NULL;
 	struct btrfs_delayed_ref_root *delayed_refs;
 	int count_mod = 1;
@@ -449,12 +543,10 @@ static noinline int __btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
 	 * the head node stores the sum of all the mods, so dropping a ref
 	 * should drop the sum in the head node by one.
 	 */
-	if (parent == (u64)-1) {
-		if (action == BTRFS_DROP_DELAYED_REF)
-			count_mod = -1;
-		else if (action == BTRFS_UPDATE_DELAYED_HEAD)
-			count_mod = 0;
-	}
+	if (action == BTRFS_UPDATE_DELAYED_HEAD)
+		count_mod = 0;
+	else if (action == BTRFS_DROP_DELAYED_REF)
+		count_mod = -1;
 
 	/*
 	 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
@@ -467,57 +559,148 @@ static noinline int __btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
 	 * Once we record must_insert_reserved, switch the action to
 	 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
 	 */
-	if (action == BTRFS_ADD_DELAYED_EXTENT) {
+	if (action == BTRFS_ADD_DELAYED_EXTENT)
 		must_insert_reserved = 1;
-		action = BTRFS_ADD_DELAYED_REF;
-	} else {
+	else
 		must_insert_reserved = 0;
-	}
-
 
 	delayed_refs = &trans->transaction->delayed_refs;
 
 	/* first set the basic ref node struct up */
 	atomic_set(&ref->refs, 1);
 	ref->bytenr = bytenr;
-	ref->parent = parent;
+	ref->num_bytes = num_bytes;
 	ref->ref_mod = count_mod;
+	ref->type  = 0;
+	ref->action  = 0;
+	ref->is_head = 1;
 	ref->in_tree = 1;
+
+	head_ref = btrfs_delayed_node_to_head(ref);
+	head_ref->must_insert_reserved = must_insert_reserved;
+	head_ref->is_data = is_data;
+
+	INIT_LIST_HEAD(&head_ref->cluster);
+	mutex_init(&head_ref->mutex);
+
+	existing = tree_insert(&delayed_refs->root, &ref->rb_node);
+
+	if (existing) {
+		update_existing_head_ref(existing, ref);
+		/*
+		 * we've updated the existing ref, free the newly
+		 * allocated ref
+		 */
+		kfree(ref);
+	} else {
+		delayed_refs->num_heads++;
+		delayed_refs->num_heads_ready++;
+		delayed_refs->num_entries++;
+		trans->delayed_ref_updates++;
+	}
+	return 0;
+}
+
+/*
+ * helper to insert a delayed tree ref into the rbtree.
+ */
+static noinline int add_delayed_tree_ref(struct btrfs_trans_handle *trans,
+					 struct btrfs_delayed_ref_node *ref,
+					 u64 bytenr, u64 num_bytes, u64 parent,
+					 u64 ref_root, int level, int action)
+{
+	struct btrfs_delayed_ref_node *existing;
+	struct btrfs_delayed_tree_ref *full_ref;
+	struct btrfs_delayed_ref_root *delayed_refs;
+
+	if (action == BTRFS_ADD_DELAYED_EXTENT)
+		action = BTRFS_ADD_DELAYED_REF;
+
+	delayed_refs = &trans->transaction->delayed_refs;
+
+	/* first set the basic ref node struct up */
+	atomic_set(&ref->refs, 1);
+	ref->bytenr = bytenr;
 	ref->num_bytes = num_bytes;
+	ref->ref_mod = 1;
+	ref->action = action;
+	ref->is_head = 0;
+	ref->in_tree = 1;
 
-	if (btrfs_delayed_ref_is_head(ref)) {
-		head_ref = btrfs_delayed_node_to_head(ref);
-		head_ref->must_insert_reserved = must_insert_reserved;
-		INIT_LIST_HEAD(&head_ref->cluster);
-		mutex_init(&head_ref->mutex);
+	full_ref = btrfs_delayed_node_to_tree_ref(ref);
+	if (parent) {
+		full_ref->parent = parent;
+		ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
 	} else {
-		full_ref = btrfs_delayed_node_to_ref(ref);
 		full_ref->root = ref_root;
-		full_ref->generation = ref_generation;
-		full_ref->owner_objectid = owner_objectid;
-		full_ref->pin = pin;
-		full_ref->action = action;
+		ref->type = BTRFS_TREE_BLOCK_REF_KEY;
 	}
+	full_ref->level = level;
 
-	existing = tree_insert(&delayed_refs->root, bytenr,
-			       parent, &ref->rb_node);
+	existing = tree_insert(&delayed_refs->root, &ref->rb_node);
 
 	if (existing) {
-		if (btrfs_delayed_ref_is_head(ref))
-			update_existing_head_ref(existing, ref);
-		else
-			update_existing_ref(trans, delayed_refs, existing, ref);
+		update_existing_ref(trans, delayed_refs, existing, ref);
+		/*
+		 * we've updated the existing ref, free the newly
+		 * allocated ref
+		 */
+		kfree(ref);
+	} else {
+		delayed_refs->num_entries++;
+		trans->delayed_ref_updates++;
+	}
+	return 0;
+}
+
+/*
+ * helper to insert a delayed data ref into the rbtree.
+ */
+static noinline int add_delayed_data_ref(struct btrfs_trans_handle *trans,
+					 struct btrfs_delayed_ref_node *ref,
+					 u64 bytenr, u64 num_bytes, u64 parent,
+					 u64 ref_root, u64 owner, u64 offset,
+					 int action)
+{
+	struct btrfs_delayed_ref_node *existing;
+	struct btrfs_delayed_data_ref *full_ref;
+	struct btrfs_delayed_ref_root *delayed_refs;
+
+	if (action == BTRFS_ADD_DELAYED_EXTENT)
+		action = BTRFS_ADD_DELAYED_REF;
+
+	delayed_refs = &trans->transaction->delayed_refs;
+
+	/* first set the basic ref node struct up */
+	atomic_set(&ref->refs, 1);
+	ref->bytenr = bytenr;
+	ref->num_bytes = num_bytes;
+	ref->ref_mod = 1;
+	ref->action = action;
+	ref->is_head = 0;
+	ref->in_tree = 1;
+
+	full_ref = btrfs_delayed_node_to_data_ref(ref);
+	if (parent) {
+		full_ref->parent = parent;
+		ref->type = BTRFS_SHARED_DATA_REF_KEY;
+	} else {
+		full_ref->root = ref_root;
+		ref->type = BTRFS_EXTENT_DATA_REF_KEY;
+	}
+	full_ref->objectid = owner;
+	full_ref->offset = offset;
 
+	existing = tree_insert(&delayed_refs->root, &ref->rb_node);
+
+	if (existing) {
+		update_existing_ref(trans, delayed_refs, existing, ref);
 		/*
 		 * we've updated the existing ref, free the newly
 		 * allocated ref
 		 */
 		kfree(ref);
 	} else {
-		if (btrfs_delayed_ref_is_head(ref)) {
-			delayed_refs->num_heads++;
-			delayed_refs->num_heads_ready++;
-		}
 		delayed_refs->num_entries++;
 		trans->delayed_ref_updates++;
 	}
@@ -525,37 +708,78 @@ static noinline int __btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
 }
 
 /*
- * add a delayed ref to the tree.  This does all of the accounting required
+ * add a delayed tree ref.  This does all of the accounting required
  * to make sure the delayed ref is eventually processed before this
  * transaction commits.
  */
-int btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
-			  u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
-			  u64 ref_generation, u64 owner_objectid, int action,
-			  int pin)
+int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
+			       u64 bytenr, u64 num_bytes, u64 parent,
+			       u64 ref_root,  int level, int action,
+			       struct btrfs_delayed_extent_op *extent_op)
 {
-	struct btrfs_delayed_ref *ref;
+	struct btrfs_delayed_tree_ref *ref;
 	struct btrfs_delayed_ref_head *head_ref;
 	struct btrfs_delayed_ref_root *delayed_refs;
 	int ret;
 
+	BUG_ON(extent_op && extent_op->is_data);
 	ref = kmalloc(sizeof(*ref), GFP_NOFS);
 	if (!ref)
 		return -ENOMEM;
 
+	head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
+	if (!head_ref) {
+		kfree(ref);
+		return -ENOMEM;
+	}
+
+	head_ref->extent_op = extent_op;
+
+	delayed_refs = &trans->transaction->delayed_refs;
+	spin_lock(&delayed_refs->lock);
+
 	/*
-	 * the parent = 0 case comes from cases where we don't actually
-	 * know the parent yet.  It will get updated later via a add/drop
-	 * pair.
+	 * insert both the head node and the new ref without dropping
+	 * the spin lock
 	 */
-	if (parent == 0)
-		parent = bytenr;
+	ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes,
+				   action, 0);
+	BUG_ON(ret);
+
+	ret = add_delayed_tree_ref(trans, &ref->node, bytenr, num_bytes,
+				   parent, ref_root, level, action);
+	BUG_ON(ret);
+	spin_unlock(&delayed_refs->lock);
+	return 0;
+}
+
+/*
+ * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
+ */
+int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
+			       u64 bytenr, u64 num_bytes,
+			       u64 parent, u64 ref_root,
+			       u64 owner, u64 offset, int action,
+			       struct btrfs_delayed_extent_op *extent_op)
+{
+	struct btrfs_delayed_data_ref *ref;
+	struct btrfs_delayed_ref_head *head_ref;
+	struct btrfs_delayed_ref_root *delayed_refs;
+	int ret;
+
+	BUG_ON(extent_op && !extent_op->is_data);
+	ref = kmalloc(sizeof(*ref), GFP_NOFS);
+	if (!ref)
+		return -ENOMEM;
 
 	head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
 	if (!head_ref) {
 		kfree(ref);
 		return -ENOMEM;
 	}
+
+	head_ref->extent_op = extent_op;
+
 	delayed_refs = &trans->transaction->delayed_refs;
 	spin_lock(&delayed_refs->lock);
 
@@ -563,14 +787,39 @@ int btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
 	 * insert both the head node and the new ref without dropping
 	 * the spin lock
 	 */
-	ret = __btrfs_add_delayed_ref(trans, &head_ref->node, bytenr, num_bytes,
-				      (u64)-1, 0, 0, 0, action, pin);
+	ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes,
+				   action, 1);
 	BUG_ON(ret);
 
-	ret = __btrfs_add_delayed_ref(trans, &ref->node, bytenr, num_bytes,
-				      parent, ref_root, ref_generation,
-				      owner_objectid, action, pin);
+	ret = add_delayed_data_ref(trans, &ref->node, bytenr, num_bytes,
+				   parent, ref_root, owner, offset, action);
+	BUG_ON(ret);
+	spin_unlock(&delayed_refs->lock);
+	return 0;
+}
+
+int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
+				u64 bytenr, u64 num_bytes,
+				struct btrfs_delayed_extent_op *extent_op)
+{
+	struct btrfs_delayed_ref_head *head_ref;
+	struct btrfs_delayed_ref_root *delayed_refs;
+	int ret;
+
+	head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
+	if (!head_ref)
+		return -ENOMEM;
+
+	head_ref->extent_op = extent_op;
+
+	delayed_refs = &trans->transaction->delayed_refs;
+	spin_lock(&delayed_refs->lock);
+
+	ret = add_delayed_ref_head(trans, &head_ref->node, bytenr,
+				   num_bytes, BTRFS_UPDATE_DELAYED_HEAD,
+				   extent_op->is_data);
 	BUG_ON(ret);
+
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
@@ -587,7 +836,7 @@ btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
 	struct btrfs_delayed_ref_root *delayed_refs;
 
 	delayed_refs = &trans->transaction->delayed_refs;
-	ref = tree_search(&delayed_refs->root, bytenr, (u64)-1, NULL);
+	ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
 	if (ref)
 		return btrfs_delayed_node_to_head(ref);
 	return NULL;
@@ -603,6 +852,7 @@ btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
  *
  * It is the same as doing a ref add and delete in two separate calls.
  */
+#if 0
 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,
@@ -666,3 +916,4 @@ int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans,
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
+#endif

fs/btrfs/delayed-ref.h

@@ -30,9 +30,6 @@ struct btrfs_delayed_ref_node {
 	/* the starting bytenr of the extent */
 	u64 bytenr;
 
-	/* the parent our backref will point to */
-	u64 parent;
-
 	/* the size of the extent */
 	u64 num_bytes;
 
@@ -50,10 +47,21 @@ struct btrfs_delayed_ref_node {
 	 */
 	int ref_mod;
 
+	unsigned int action:8;
+	unsigned int type:8;
 	/* is this node still in the rbtree? */
+	unsigned int is_head:1;
 	unsigned int in_tree:1;
 };
 
+struct btrfs_delayed_extent_op {
+	struct btrfs_disk_key key;
+	u64 flags_to_set;
+	unsigned int update_key:1;
+	unsigned int update_flags:1;
+	unsigned int is_data:1;
+};
+
 /*
  * the head refs are used to hold a lock on a given extent, which allows us
  * to make sure that only one process is running the delayed refs
@@ -71,6 +79,7 @@ struct btrfs_delayed_ref_head {
 
 	struct list_head cluster;
 
+	struct btrfs_delayed_extent_op *extent_op;
 	/*
 	 * when a new extent is allocated, it is just reserved in memory
 	 * The actual extent isn't inserted into the extent allocation tree
@@ -84,27 +93,26 @@ struct btrfs_delayed_ref_head {
 	 * the free has happened.
 	 */
 	unsigned int must_insert_reserved:1;
+	unsigned int is_data:1;
 };
 
-struct btrfs_delayed_ref {
+struct btrfs_delayed_tree_ref {
 	struct btrfs_delayed_ref_node node;
+	union {
+		u64 root;
+		u64 parent;
+	};
+	int level;
+};
 
-	/* the root objectid our ref will point to */
-	u64 root;
-
-	/* the generation for the backref */
-	u64 generation;
-
-	/* owner_objectid of the backref  */
-	u64 owner_objectid;
-
-	/* operation done by this entry in the rbtree */
-	u8 action;
-
-	/* if pin == 1, when the extent is freed it will be pinned until
-	 * transaction commit
-	 */
-	unsigned int pin:1;
+struct btrfs_delayed_data_ref {
+	struct btrfs_delayed_ref_node node;
+	union {
+		u64 root;
+		u64 parent;
+	};
+	u64 objectid;
+	u64 offset;
 };
 
 struct btrfs_delayed_ref_root {
@@ -143,17 +151,25 @@ static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
 	}
 }
 
-int btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
-			  u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
-			  u64 ref_generation, u64 owner_objectid, int action,
-			  int pin);
+int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
+			       u64 bytenr, u64 num_bytes, u64 parent,
+			       u64 ref_root, int level, int action,
+			       struct btrfs_delayed_extent_op *extent_op);
+int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
+			       u64 bytenr, u64 num_bytes,
+			       u64 parent, u64 ref_root,
+			       u64 owner, u64 offset, int action,
+			       struct btrfs_delayed_extent_op *extent_op);
+int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
+				u64 bytenr, u64 num_bytes,
+				struct btrfs_delayed_extent_op *extent_op);
 
 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_ref(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, u64 bytenr,
-			    u64 num_bytes, u32 *refs);
+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,
@@ -169,18 +185,24 @@ int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
  */
 static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
 {
-	return node->parent == (u64)-1;
+	return node->is_head;
 }
 
 /*
  * helper functions to cast a node into its container
  */
-static inline struct btrfs_delayed_ref *
-btrfs_delayed_node_to_ref(struct btrfs_delayed_ref_node *node)
+static inline struct btrfs_delayed_tree_ref *
+btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
 {
 	WARN_ON(btrfs_delayed_ref_is_head(node));
-	return container_of(node, struct btrfs_delayed_ref, node);
+	return container_of(node, struct btrfs_delayed_tree_ref, node);
+}
 
+static inline struct btrfs_delayed_data_ref *
+btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
+{
+	WARN_ON(btrfs_delayed_ref_is_head(node));
+	return container_of(node, struct btrfs_delayed_data_ref, node);
 }
 
 static inline struct btrfs_delayed_ref_head *
@@ -188,6 +210,5 @@ btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
 {
 	WARN_ON(!btrfs_delayed_ref_is_head(node));
 	return container_of(node, struct btrfs_delayed_ref_head, node);
-
 }
 #endif

fs/btrfs/disk-io.c

@@ -26,8 +26,8 @@
 #include <linux/workqueue.h>
 #include <linux/kthread.h>
 #include <linux/freezer.h>
+#include <linux/crc32c.h>
 #include "compat.h"
-#include "crc32c.h"
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
@@ -36,7 +36,6 @@
 #include "print-tree.h"
 #include "async-thread.h"
 #include "locking.h"
-#include "ref-cache.h"
 #include "tree-log.h"
 #include "free-space-cache.h"
 
@@ -172,7 +171,7 @@ out:
 
 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 {
-	return btrfs_crc32c(seed, data, len);
+	return crc32c(seed, data, len);
 }
 
 void btrfs_csum_final(u32 crc, char *result)
@@ -884,7 +883,6 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 {
 	root->node = NULL;
 	root->commit_root = NULL;
-	root->ref_tree = NULL;
 	root->sectorsize = sectorsize;
 	root->nodesize = nodesize;
 	root->leafsize = leafsize;
@@ -899,12 +897,14 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 	root->last_inode_alloc = 0;
 	root->name = NULL;
 	root->in_sysfs = 0;
+	root->inode_tree.rb_node = NULL;
 
 	INIT_LIST_HEAD(&root->dirty_list);
 	INIT_LIST_HEAD(&root->orphan_list);
-	INIT_LIST_HEAD(&root->dead_list);
+	INIT_LIST_HEAD(&root->root_list);
 	spin_lock_init(&root->node_lock);
 	spin_lock_init(&root->list_lock);
+	spin_lock_init(&root->inode_lock);
 	mutex_init(&root->objectid_mutex);
 	mutex_init(&root->log_mutex);
 	init_waitqueue_head(&root->log_writer_wait);
@@ -918,9 +918,6 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 	extent_io_tree_init(&root->dirty_log_pages,
 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
 
-	btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
-	root->ref_tree = &root->ref_tree_struct;
-
 	memset(&root->root_key, 0, sizeof(root->root_key));
 	memset(&root->root_item, 0, sizeof(root->root_item));
 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
@@ -959,6 +956,7 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 				     blocksize, generation);
+	root->commit_root = btrfs_root_node(root);
 	BUG_ON(!root->node);
 	return 0;
 }
@@ -1025,20 +1023,19 @@ static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 	 */
 	root->ref_cows = 0;
 
-	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
-				      0, BTRFS_TREE_LOG_OBJECTID,
-				      trans->transid, 0, 0, 0);
+	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
+				      BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
 	if (IS_ERR(leaf)) {
 		kfree(root);
 		return ERR_CAST(leaf);
 	}
 
+	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
+	btrfs_set_header_bytenr(leaf, leaf->start);
+	btrfs_set_header_generation(leaf, trans->transid);
+	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
+	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
 	root->node = leaf;
-	btrfs_set_header_nritems(root->node, 0);
-	btrfs_set_header_level(root->node, 0);
-	btrfs_set_header_bytenr(root->node, root->node->start);
-	btrfs_set_header_generation(root->node, trans->transid);
-	btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 
 	write_extent_buffer(root->node, root->fs_info->fsid,
 			    (unsigned long)btrfs_header_fsid(root->node),
@@ -1081,8 +1078,7 @@ int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
 	inode_item->nbytes = cpu_to_le64(root->leafsize);
 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 
-	btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
-	btrfs_set_root_generation(&log_root->root_item, trans->transid);
+	btrfs_set_root_node(&log_root->root_item, log_root->node);
 
 	WARN_ON(root->log_root);
 	root->log_root = log_root;
@@ -1144,6 +1140,7 @@ out:
 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 				     blocksize, generation);
+	root->commit_root = btrfs_root_node(root);
 	BUG_ON(!root->node);
 insert:
 	if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
@@ -1210,7 +1207,7 @@ struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 	}
 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
 		ret = btrfs_find_dead_roots(fs_info->tree_root,
-					    root->root_key.objectid, root);
+					    root->root_key.objectid);
 		BUG_ON(ret);
 		btrfs_orphan_cleanup(root);
 	}
@@ -1569,8 +1566,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	atomic_set(&fs_info->async_delalloc_pages, 0);
 	atomic_set(&fs_info->async_submit_draining, 0);
 	atomic_set(&fs_info->nr_async_bios, 0);
-	atomic_set(&fs_info->throttles, 0);
-	atomic_set(&fs_info->throttle_gen, 0);
 	fs_info->sb = sb;
 	fs_info->max_extent = (u64)-1;
 	fs_info->max_inline = 8192 * 1024;
@@ -1598,6 +1593,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
 
+	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
 			     fs_info->btree_inode->i_mapping,
 			     GFP_NOFS);
@@ -1613,10 +1609,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
 	fs_info->do_barriers = 1;
 
-	INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
-	btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
-	btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
-
 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 	       sizeof(struct btrfs_key));
@@ -1674,6 +1666,12 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 		goto fail_iput;
 	}
 
+	features = btrfs_super_incompat_flags(disk_super);
+	if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
+		features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
+		btrfs_set_super_incompat_flags(disk_super, features);
+	}
+
 	features = btrfs_super_compat_ro_flags(disk_super) &
 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
 	if (!(sb->s_flags & MS_RDONLY) && features) {
@@ -1771,7 +1769,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	if (ret) {
 		printk(KERN_WARNING "btrfs: failed to read the system "
 		       "array on %s\n", sb->s_id);
-		goto fail_sys_array;
+		goto fail_sb_buffer;
 	}
 
 	blocksize = btrfs_level_size(tree_root,
@@ -1785,6 +1783,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 					   btrfs_super_chunk_root(disk_super),
 					   blocksize, generation);
 	BUG_ON(!chunk_root->node);
+	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
+	chunk_root->commit_root = btrfs_root_node(chunk_root);
 
 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
 	   (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
@@ -1810,7 +1810,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 					  blocksize, generation);
 	if (!tree_root->node)
 		goto fail_chunk_root;
-
+	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
+	tree_root->commit_root = btrfs_root_node(tree_root);
 
 	ret = find_and_setup_root(tree_root, fs_info,
 				  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
@@ -1820,14 +1821,14 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 
 	ret = find_and_setup_root(tree_root, fs_info,
 				  BTRFS_DEV_TREE_OBJECTID, dev_root);
-	dev_root->track_dirty = 1;
 	if (ret)
 		goto fail_extent_root;
+	dev_root->track_dirty = 1;
 
 	ret = find_and_setup_root(tree_root, fs_info,
 				  BTRFS_CSUM_TREE_OBJECTID, csum_root);
 	if (ret)
-		goto fail_extent_root;
+		goto fail_dev_root;
 
 	csum_root->track_dirty = 1;
 
@@ -1849,6 +1850,14 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	if (IS_ERR(fs_info->transaction_kthread))
 		goto fail_cleaner;
 
+	if (!btrfs_test_opt(tree_root, SSD) &&
+	    !btrfs_test_opt(tree_root, NOSSD) &&
+	    !fs_info->fs_devices->rotating) {
+		printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
+		       "mode\n");
+		btrfs_set_opt(fs_info->mount_opt, SSD);
+	}
+
 	if (btrfs_super_log_root(disk_super) != 0) {
 		u64 bytenr = btrfs_super_log_root(disk_super);
 
@@ -1881,7 +1890,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	}
 
 	if (!(sb->s_flags & MS_RDONLY)) {
-		ret = btrfs_cleanup_reloc_trees(tree_root);
+		ret = btrfs_recover_relocation(tree_root);
 		BUG_ON(ret);
 	}
 
@@ -1892,6 +1901,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
 	if (!fs_info->fs_root)
 		goto fail_trans_kthread;
+
 	return tree_root;
 
 fail_trans_kthread:
@@ -1908,14 +1918,19 @@ fail_cleaner:
 
 fail_csum_root:
 	free_extent_buffer(csum_root->node);
+	free_extent_buffer(csum_root->commit_root);
+fail_dev_root:
+	free_extent_buffer(dev_root->node);
+	free_extent_buffer(dev_root->commit_root);
 fail_extent_root:
 	free_extent_buffer(extent_root->node);
+	free_extent_buffer(extent_root->commit_root);
 fail_tree_root:
 	free_extent_buffer(tree_root->node);
+	free_extent_buffer(tree_root->commit_root);
 fail_chunk_root:
 	free_extent_buffer(chunk_root->node);
-fail_sys_array:
-	free_extent_buffer(dev_root->node);
+	free_extent_buffer(chunk_root->commit_root);
 fail_sb_buffer:
 	btrfs_stop_workers(&fs_info->fixup_workers);
 	btrfs_stop_workers(&fs_info->delalloc_workers);
@@ -2005,6 +2020,17 @@ struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
 	return latest;
 }
 
+/*
+ * this should be called twice, once with wait == 0 and
+ * once with wait == 1.  When wait == 0 is done, all the buffer heads
+ * we write are pinned.
+ *
+ * They are released when wait == 1 is done.
+ * max_mirrors must be the same for both runs, and it indicates how
+ * many supers on this one device should be written.
+ *
+ * max_mirrors == 0 means to write them all.
+ */
 static int write_dev_supers(struct btrfs_device *device,
 			    struct btrfs_super_block *sb,
 			    int do_barriers, int wait, int max_mirrors)
@@ -2040,12 +2066,16 @@ static int write_dev_supers(struct btrfs_device *device,
 			bh = __find_get_block(device->bdev, bytenr / 4096,
 					      BTRFS_SUPER_INFO_SIZE);
 			BUG_ON(!bh);
-			brelse(bh);
 			wait_on_buffer(bh);
-			if (buffer_uptodate(bh)) {
-				brelse(bh);
-				continue;
-			}
+			if (!buffer_uptodate(bh))
+				errors++;
+
+			/* drop our reference */
+			brelse(bh);
+
+			/* drop the reference from the wait == 0 run */
+			brelse(bh);
+			continue;
 		} else {
 			btrfs_set_super_bytenr(sb, bytenr);
 
@@ -2056,12 +2086,18 @@ static int write_dev_supers(struct btrfs_device *device,
 					      BTRFS_CSUM_SIZE);
 			btrfs_csum_final(crc, sb->csum);
 
+			/*
+			 * one reference for us, and we leave it for the
+			 * caller
+			 */
 			bh = __getblk(device->bdev, bytenr / 4096,
 				      BTRFS_SUPER_INFO_SIZE);
 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
 
-			set_buffer_uptodate(bh);
+			/* one reference for submit_bh */
 			get_bh(bh);
+
+			set_buffer_uptodate(bh);
 			lock_buffer(bh);
 			bh->b_end_io = btrfs_end_buffer_write_sync;
 		}
@@ -2073,6 +2109,7 @@ static int write_dev_supers(struct btrfs_device *device,
 				       device->name);
 				set_buffer_uptodate(bh);
 				device->barriers = 0;
+				/* one reference for submit_bh */
 				get_bh(bh);
 				lock_buffer(bh);
 				ret = submit_bh(WRITE_SYNC, bh);
@@ -2081,22 +2118,15 @@ static int write_dev_supers(struct btrfs_device *device,
 			ret = submit_bh(WRITE_SYNC, bh);
 		}
 
-		if (!ret && wait) {
-			wait_on_buffer(bh);
-			if (!buffer_uptodate(bh))
-				errors++;
-		} else if (ret) {
+		if (ret)
 			errors++;
-		}
-		if (wait)
-			brelse(bh);
 	}
 	return errors < i ? 0 : -1;
 }
 
 int write_all_supers(struct btrfs_root *root, int max_mirrors)
 {
-	struct list_head *head = &root->fs_info->fs_devices->devices;
+	struct list_head *head;
 	struct btrfs_device *dev;
 	struct btrfs_super_block *sb;
 	struct btrfs_dev_item *dev_item;
@@ -2111,6 +2141,9 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors)
 
 	sb = &root->fs_info->super_for_commit;
 	dev_item = &sb->dev_item;
+
+	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
+	head = &root->fs_info->fs_devices->devices;
 	list_for_each_entry(dev, head, dev_list) {
 		if (!dev->bdev) {
 			total_errors++;
@@ -2154,6 +2187,7 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors)
 		if (ret)
 			total_errors++;
 	}
+	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 	if (total_errors > max_errors) {
 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
 		       total_errors);
@@ -2173,6 +2207,7 @@ int write_ctree_super(struct btrfs_trans_handle *trans,
 
 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 {
+	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
 	radix_tree_delete(&fs_info->fs_roots_radix,
 			  (unsigned long)root->root_key.objectid);
 	if (root->anon_super.s_dev) {
@@ -2219,10 +2254,12 @@ int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
 					     ARRAY_SIZE(gang));
 		if (!ret)
 			break;
+
+		root_objectid = gang[ret - 1]->root_key.objectid + 1;
 		for (i = 0; i < ret; i++) {
 			root_objectid = gang[i]->root_key.objectid;
 			ret = btrfs_find_dead_roots(fs_info->tree_root,
-						    root_objectid, gang[i]);
+						    root_objectid);
 			BUG_ON(ret);
 			btrfs_orphan_cleanup(gang[i]);
 		}
@@ -2278,20 +2315,16 @@ int close_ctree(struct btrfs_root *root)
 		       (unsigned long long)fs_info->total_ref_cache_size);
 	}
 
-	if (fs_info->extent_root->node)
-		free_extent_buffer(fs_info->extent_root->node);
-
-	if (fs_info->tree_root->node)
-		free_extent_buffer(fs_info->tree_root->node);
-
-	if (root->fs_info->chunk_root->node)
-		free_extent_buffer(root->fs_info->chunk_root->node);
-
-	if (root->fs_info->dev_root->node)
-		free_extent_buffer(root->fs_info->dev_root->node);
-
-	if (root->fs_info->csum_root->node)
-		free_extent_buffer(root->fs_info->csum_root->node);
+	free_extent_buffer(fs_info->extent_root->node);
+	free_extent_buffer(fs_info->extent_root->commit_root);
+	free_extent_buffer(fs_info->tree_root->node);
+	free_extent_buffer(fs_info->tree_root->commit_root);
+	free_extent_buffer(root->fs_info->chunk_root->node);
+	free_extent_buffer(root->fs_info->chunk_root->commit_root);
+	free_extent_buffer(root->fs_info->dev_root->node);
+	free_extent_buffer(root->fs_info->dev_root->commit_root);
+	free_extent_buffer(root->fs_info->csum_root->node);
+	free_extent_buffer(root->fs_info->csum_root->commit_root);
 
 	btrfs_free_block_groups(root->fs_info);
 
@@ -2373,17 +2406,14 @@ void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
 	 * looks as though older kernels can get into trouble with
 	 * this code, they end up stuck in balance_dirty_pages forever
 	 */
-	struct extent_io_tree *tree;
 	u64 num_dirty;
-	u64 start = 0;
 	unsigned long thresh = 32 * 1024 * 1024;
-	tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 
 	if (current->flags & PF_MEMALLOC)
 		return;
 
-	num_dirty = count_range_bits(tree, &start, (u64)-1,
-				     thresh, EXTENT_DIRTY);
+	num_dirty = root->fs_info->dirty_metadata_bytes;
+
 	if (num_dirty > thresh) {
 		balance_dirty_pages_ratelimited_nr(
 				   root->fs_info->btree_inode->i_mapping, 1);

fs/btrfs/export.c

@@ -78,7 +78,7 @@ static struct dentry *btrfs_get_dentry(struct super_block *sb, u64 objectid,
 	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 	key.offset = 0;
 
-	inode = btrfs_iget(sb, &key, root, NULL);
+	inode = btrfs_iget(sb, &key, root);
 	if (IS_ERR(inode))
 		return (void *)inode;
 
@@ -192,7 +192,7 @@ static struct dentry *btrfs_get_parent(struct dentry *child)
 	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 	key.offset = 0;
 
-	return d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root, NULL));
+	return d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root));
 }
 
 const struct export_operations btrfs_export_ops = {

fs/btrfs/extent-tree.c

@@ -23,50 +23,39 @@
 #include <linux/rcupdate.h>
 #include "compat.h"
 #include "hash.h"
-#include "crc32c.h"
 #include "ctree.h"
 #include "disk-io.h"
 #include "print-tree.h"
 #include "transaction.h"
 #include "volumes.h"
 #include "locking.h"
-#include "ref-cache.h"
 #include "free-space-cache.h"
 
-#define PENDING_EXTENT_INSERT 0
-#define PENDING_EXTENT_DELETE 1
-#define PENDING_BACKREF_UPDATE 2
-
-struct pending_extent_op {
-	int type;
-	u64 bytenr;
-	u64 num_bytes;
-	u64 parent;
-	u64 orig_parent;
-	u64 generation;
-	u64 orig_generation;
-	int level;
-	struct list_head list;
-	int del;
-};
-
-static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
-					 struct btrfs_root *root, u64 parent,
-					 u64 root_objectid, u64 ref_generation,
-					 u64 owner, struct btrfs_key *ins,
-					 int ref_mod);
 static int update_reserved_extents(struct btrfs_root *root,
 				   u64 bytenr, u64 num, int reserve);
 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 noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
-					struct btrfs_root *root,
-					u64 bytenr, u64 num_bytes, u64 parent,
-					u64 root_objectid, u64 ref_generation,
-					u64 owner_objectid, int pin,
-					int ref_to_drop);
+static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				u64 bytenr, u64 num_bytes, u64 parent,
+				u64 root_objectid, u64 owner_objectid,
+				u64 owner_offset, int refs_to_drop,
+				struct btrfs_delayed_extent_op *extra_op);
+static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
+				    struct extent_buffer *leaf,
+				    struct btrfs_extent_item *ei);
+static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      u64 parent, u64 root_objectid,
+				      u64 flags, u64 owner, u64 offset,
+				      struct btrfs_key *ins, int ref_mod);
+static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root,
+				     u64 parent, u64 root_objectid,
+				     u64 flags, struct btrfs_disk_key *key,
+				     int level, struct btrfs_key *ins);
 
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *extent_root, u64 alloc_bytes,
@@ -453,348 +442,1071 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
  *    maintenance.  This is actually the same as #2, but with a slightly
  *    different use case.
  *
+ * There are two kinds of back refs. The implicit back refs is optimized
+ * for pointers in non-shared tree blocks. For a given pointer in a block,
+ * back refs of this kind provide information about the block's owner tree
+ * and the pointer's key. These information allow us to find the block by
+ * b-tree searching. The full back refs is for pointers in tree blocks not
+ * referenced by their owner trees. The location of tree block is recorded
+ * in the back refs. Actually the full back refs is generic, and can be
+ * used in all cases the implicit back refs is used. The major shortcoming
+ * of the full back refs is its overhead. Every time a tree block gets
+ * COWed, we have to update back refs entry for all pointers in it.
+ *
+ * For a newly allocated tree block, we use implicit back refs for
+ * pointers in it. This means most tree related operations only involve
+ * implicit back refs. For a tree block created in old transaction, the
+ * only way to drop a reference to it is COW it. So we can detect the
+ * event that tree block loses its owner tree's reference and do the
+ * back refs conversion.
+ *
+ * When a tree block is COW'd through a tree, there are four cases:
+ *
+ * The reference count of the block is one and the tree is the block's
+ * owner tree. Nothing to do in this case.
+ *
+ * The reference count of the block is one and the tree is not the
+ * block's owner tree. In this case, full back refs is used for pointers
+ * in the block. Remove these full back refs, add implicit back refs for
+ * every pointers in the new block.
+ *
+ * The reference count of the block is greater than one and the tree is
+ * the block's owner tree. In this case, implicit back refs is used for
+ * pointers in the block. Add full back refs for every pointers in the
+ * block, increase lower level extents' reference counts. The original
+ * implicit back refs are entailed to the new block.
+ *
+ * The reference count of the block is greater than one and the tree is
+ * not the block's owner tree. Add implicit back refs for every pointer in
+ * the new block, increase lower level extents' reference count.
+ *
+ * Back Reference Key composing:
+ *
+ * The key objectid corresponds to the first byte in the extent,
+ * The key type is used to differentiate between types of back refs.
+ * There are different meanings of the key offset for different types
+ * of back refs.
+ *
  * File extents can be referenced by:
  *
  * - multiple snapshots, subvolumes, or different generations in one subvol
  * - different files inside a single subvolume
  * - different offsets inside a file (bookend extents in file.c)
  *
- * The extent ref structure has fields for:
+ * The extent ref structure for the implicit back refs has fields for:
  *
  * - Objectid of the subvolume root
- * - Generation number of the tree holding the reference
  * - objectid of the file holding the reference
- * - number of references holding by parent node (alway 1 for tree blocks)
- *
- * Btree leaf may hold multiple references to a file extent. In most cases,
- * these references are from same file and the corresponding offsets inside
- * the file are close together.
- *
- * When a file extent is allocated the fields are filled in:
- *     (root_key.objectid, trans->transid, inode objectid, 1)
+ * - original offset in the file
+ * - how many bookend extents
  *
- * When a leaf is cow'd new references are added for every file extent found
- * in the leaf.  It looks similar to the create case, but trans->transid will
- * be different when the block is cow'd.
+ * The key offset for the implicit back refs is hash of the first
+ * three fields.
  *
- *     (root_key.objectid, trans->transid, inode objectid,
- *      number of references in the leaf)
+ * The extent ref structure for the full back refs has field for:
  *
- * When a file extent is removed either during snapshot deletion or
- * file truncation, we find the corresponding back reference and check
- * the following fields:
+ * - number of pointers in the tree leaf
  *
- *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
- *      inode objectid)
+ * The key offset for the implicit back refs is the first byte of
+ * the tree leaf
  *
- * Btree extents can be referenced by:
- *
- * - Different subvolumes
- * - Different generations of the same subvolume
- *
- * When a tree block is created, back references are inserted:
+ * When a file extent is allocated, The implicit back refs is used.
+ * the fields are filled in:
  *
- * (root->root_key.objectid, trans->transid, level, 1)
+ *     (root_key.objectid, inode objectid, offset in file, 1)
  *
- * When a tree block is cow'd, new back references are added for all the
- * blocks it points to. If the tree block isn't in reference counted root,
- * the old back references are removed. These new back references are of
- * the form (trans->transid will have increased since creation):
+ * When a file extent is removed file truncation, we find the
+ * corresponding implicit back refs and check the following fields:
  *
- * (root->root_key.objectid, trans->transid, level, 1)
+ *     (btrfs_header_owner(leaf), inode objectid, offset in file)
  *
- * When a backref is in deleting, the following fields are checked:
+ * Btree extents can be referenced by:
  *
- * if backref was for a tree root:
- *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
- * else
- *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
+ * - Different subvolumes
  *
- * Back Reference Key composing:
+ * Both the implicit back refs and the full back refs for tree blocks
+ * only consist of key. The key offset for the implicit back refs is
+ * objectid of block's owner tree. The key offset for the full back refs
+ * is the first byte of parent block.
  *
- * The key objectid corresponds to the first byte in the extent, the key
- * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
- * byte of parent extent. If a extent is tree root, the key offset is set
- * to the key objectid.
+ * When implicit back refs is used, information about the lowest key and
+ * level of the tree block are required. These information are stored in
+ * tree block info structure.
  */
 
-static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
-					  struct btrfs_root *root,
-					  struct btrfs_path *path,
-					  u64 bytenr, u64 parent,
-					  u64 ref_root, u64 ref_generation,
-					  u64 owner_objectid, int del)
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  struct btrfs_path *path,
+				  u64 owner, u32 extra_size)
 {
+	struct btrfs_extent_item *item;
+	struct btrfs_extent_item_v0 *ei0;
+	struct btrfs_extent_ref_v0 *ref0;
+	struct btrfs_tree_block_info *bi;
+	struct extent_buffer *leaf;
 	struct btrfs_key key;
-	struct btrfs_extent_ref *ref;
+	struct btrfs_key found_key;
+	u32 new_size = sizeof(*item);
+	u64 refs;
+	int ret;
+
+	leaf = path->nodes[0];
+	BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
+
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	ei0 = btrfs_item_ptr(leaf, path->slots[0],
+			     struct btrfs_extent_item_v0);
+	refs = btrfs_extent_refs_v0(leaf, ei0);
+
+	if (owner == (u64)-1) {
+		while (1) {
+			if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+				ret = btrfs_next_leaf(root, path);
+				if (ret < 0)
+					return ret;
+				BUG_ON(ret > 0);
+				leaf = path->nodes[0];
+			}
+			btrfs_item_key_to_cpu(leaf, &found_key,
+					      path->slots[0]);
+			BUG_ON(key.objectid != found_key.objectid);
+			if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
+				path->slots[0]++;
+				continue;
+			}
+			ref0 = btrfs_item_ptr(leaf, path->slots[0],
+					      struct btrfs_extent_ref_v0);
+			owner = btrfs_ref_objectid_v0(leaf, ref0);
+			break;
+		}
+	}
+	btrfs_release_path(root, path);
+
+	if (owner < BTRFS_FIRST_FREE_OBJECTID)
+		new_size += sizeof(*bi);
+
+	new_size -= sizeof(*ei0);
+	ret = btrfs_search_slot(trans, root, &key, path,
+				new_size + extra_size, 1);
+	if (ret < 0)
+		return ret;
+	BUG_ON(ret);
+
+	ret = btrfs_extend_item(trans, root, path, new_size);
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	btrfs_set_extent_refs(leaf, item, refs);
+	/* FIXME: get real generation */
+	btrfs_set_extent_generation(leaf, item, 0);
+	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
+		btrfs_set_extent_flags(leaf, item,
+				       BTRFS_EXTENT_FLAG_TREE_BLOCK |
+				       BTRFS_BLOCK_FLAG_FULL_BACKREF);
+		bi = (struct btrfs_tree_block_info *)(item + 1);
+		/* FIXME: get first key of the block */
+		memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
+		btrfs_set_tree_block_level(leaf, bi, (int)owner);
+	} else {
+		btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
+	}
+	btrfs_mark_buffer_dirty(leaf);
+	return 0;
+}
+#endif
+
+static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
+{
+	u32 high_crc = ~(u32)0;
+	u32 low_crc = ~(u32)0;
+	__le64 lenum;
+
+	lenum = cpu_to_le64(root_objectid);
+	high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
+	lenum = cpu_to_le64(owner);
+	low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
+	lenum = cpu_to_le64(offset);
+	low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
+
+	return ((u64)high_crc << 31) ^ (u64)low_crc;
+}
+
+static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
+				     struct btrfs_extent_data_ref *ref)
+{
+	return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
+				    btrfs_extent_data_ref_objectid(leaf, ref),
+				    btrfs_extent_data_ref_offset(leaf, ref));
+}
+
+static int match_extent_data_ref(struct extent_buffer *leaf,
+				 struct btrfs_extent_data_ref *ref,
+				 u64 root_objectid, u64 owner, u64 offset)
+{
+	if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
+	    btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
+	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
+		return 0;
+	return 1;
+}
+
+static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
+					   struct btrfs_root *root,
+					   struct btrfs_path *path,
+					   u64 bytenr, u64 parent,
+					   u64 root_objectid,
+					   u64 owner, u64 offset)
+{
+	struct btrfs_key key;
+	struct btrfs_extent_data_ref *ref;
 	struct extent_buffer *leaf;
-	u64 ref_objectid;
+	u32 nritems;
 	int ret;
+	int recow;
+	int err = -ENOENT;
 
 	key.objectid = bytenr;
-	key.type = BTRFS_EXTENT_REF_KEY;
-	key.offset = parent;
+	if (parent) {
+		key.type = BTRFS_SHARED_DATA_REF_KEY;
+		key.offset = parent;
+	} else {
+		key.type = BTRFS_EXTENT_DATA_REF_KEY;
+		key.offset = hash_extent_data_ref(root_objectid,
+						  owner, offset);
+	}
+again:
+	recow = 0;
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret < 0) {
+		err = ret;
+		goto fail;
+	}
 
-	ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
-	if (ret < 0)
-		goto out;
-	if (ret > 0) {
-		ret = -ENOENT;
-		goto out;
+	if (parent) {
+		if (!ret)
+			return 0;
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+		key.type = BTRFS_EXTENT_REF_V0_KEY;
+		btrfs_release_path(root, path);
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (ret < 0) {
+			err = ret;
+			goto fail;
+		}
+		if (!ret)
+			return 0;
+#endif
+		goto fail;
 	}
 
 	leaf = path->nodes[0];
-	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
-	ref_objectid = btrfs_ref_objectid(leaf, ref);
-	if (btrfs_ref_root(leaf, ref) != ref_root ||
-	    btrfs_ref_generation(leaf, ref) != ref_generation ||
-	    (ref_objectid != owner_objectid &&
-	     ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
-		ret = -EIO;
-		WARN_ON(1);
-		goto out;
+	nritems = btrfs_header_nritems(leaf);
+	while (1) {
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				err = ret;
+			if (ret)
+				goto fail;
+
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+			recow = 1;
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid != bytenr ||
+		    key.type != BTRFS_EXTENT_DATA_REF_KEY)
+			goto fail;
+
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_data_ref);
+
+		if (match_extent_data_ref(leaf, ref, root_objectid,
+					  owner, offset)) {
+			if (recow) {
+				btrfs_release_path(root, path);
+				goto again;
+			}
+			err = 0;
+			break;
+		}
+		path->slots[0]++;
 	}
-	ret = 0;
-out:
-	return ret;
+fail:
+	return err;
 }
 
-static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
-					  struct btrfs_root *root,
-					  struct btrfs_path *path,
-					  u64 bytenr, u64 parent,
-					  u64 ref_root, u64 ref_generation,
-					  u64 owner_objectid,
-					  int refs_to_add)
+static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
+					   struct btrfs_root *root,
+					   struct btrfs_path *path,
+					   u64 bytenr, u64 parent,
+					   u64 root_objectid, u64 owner,
+					   u64 offset, int refs_to_add)
 {
 	struct btrfs_key key;
 	struct extent_buffer *leaf;
-	struct btrfs_extent_ref *ref;
+	u32 size;
 	u32 num_refs;
 	int ret;
 
 	key.objectid = bytenr;
-	key.type = BTRFS_EXTENT_REF_KEY;
-	key.offset = parent;
+	if (parent) {
+		key.type = BTRFS_SHARED_DATA_REF_KEY;
+		key.offset = parent;
+		size = sizeof(struct btrfs_shared_data_ref);
+	} else {
+		key.type = BTRFS_EXTENT_DATA_REF_KEY;
+		key.offset = hash_extent_data_ref(root_objectid,
+						  owner, offset);
+		size = sizeof(struct btrfs_extent_data_ref);
+	}
 
-	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
-	if (ret == 0) {
-		leaf = path->nodes[0];
-		ref = btrfs_item_ptr(leaf, path->slots[0],
-				     struct btrfs_extent_ref);
-		btrfs_set_ref_root(leaf, ref, ref_root);
-		btrfs_set_ref_generation(leaf, ref, ref_generation);
-		btrfs_set_ref_objectid(leaf, ref, owner_objectid);
-		btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
-	} else if (ret == -EEXIST) {
-		u64 existing_owner;
-
-		BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
-		leaf = path->nodes[0];
+	ret = btrfs_insert_empty_item(trans, root, path, &key, size);
+	if (ret && ret != -EEXIST)
+		goto fail;
+
+	leaf = path->nodes[0];
+	if (parent) {
+		struct btrfs_shared_data_ref *ref;
 		ref = btrfs_item_ptr(leaf, path->slots[0],
-				     struct btrfs_extent_ref);
-		if (btrfs_ref_root(leaf, ref) != ref_root ||
-		    btrfs_ref_generation(leaf, ref) != ref_generation) {
-			ret = -EIO;
-			WARN_ON(1);
-			goto out;
+				     struct btrfs_shared_data_ref);
+		if (ret == 0) {
+			btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
+		} else {
+			num_refs = btrfs_shared_data_ref_count(leaf, ref);
+			num_refs += refs_to_add;
+			btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 		}
+	} else {
+		struct btrfs_extent_data_ref *ref;
+		while (ret == -EEXIST) {
+			ref = btrfs_item_ptr(leaf, path->slots[0],
+					     struct btrfs_extent_data_ref);
+			if (match_extent_data_ref(leaf, ref, root_objectid,
+						  owner, offset))
+				break;
+			btrfs_release_path(root, path);
+			key.offset++;
+			ret = btrfs_insert_empty_item(trans, root, path, &key,
+						      size);
+			if (ret && ret != -EEXIST)
+				goto fail;
 
-		num_refs = btrfs_ref_num_refs(leaf, ref);
-		BUG_ON(num_refs == 0);
-		btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
-
-		existing_owner = btrfs_ref_objectid(leaf, ref);
-		if (existing_owner != owner_objectid &&
-		    existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
-			btrfs_set_ref_objectid(leaf, ref,
-					BTRFS_MULTIPLE_OBJECTIDS);
+			leaf = path->nodes[0];
+		}
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_data_ref);
+		if (ret == 0) {
+			btrfs_set_extent_data_ref_root(leaf, ref,
+						       root_objectid);
+			btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
+			btrfs_set_extent_data_ref_offset(leaf, ref, offset);
+			btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
+		} else {
+			num_refs = btrfs_extent_data_ref_count(leaf, ref);
+			num_refs += refs_to_add;
+			btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 		}
-		ret = 0;
-	} else {
-		goto out;
 	}
-	btrfs_unlock_up_safe(path, 1);
-	btrfs_mark_buffer_dirty(path->nodes[0]);
-out:
+	btrfs_mark_buffer_dirty(leaf);
+	ret = 0;
+fail:
 	btrfs_release_path(root, path);
 	return ret;
 }
 
-static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
-					  struct btrfs_root *root,
-					  struct btrfs_path *path,
-					  int refs_to_drop)
+static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
+					   struct btrfs_root *root,
+					   struct btrfs_path *path,
+					   int refs_to_drop)
 {
+	struct btrfs_key key;
+	struct btrfs_extent_data_ref *ref1 = NULL;
+	struct btrfs_shared_data_ref *ref2 = NULL;
 	struct extent_buffer *leaf;
-	struct btrfs_extent_ref *ref;
-	u32 num_refs;
+	u32 num_refs = 0;
 	int ret = 0;
 
 	leaf = path->nodes[0];
-	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
-	num_refs = btrfs_ref_num_refs(leaf, ref);
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+	if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
+		ref1 = btrfs_item_ptr(leaf, path->slots[0],
+				      struct btrfs_extent_data_ref);
+		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
+	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
+		ref2 = btrfs_item_ptr(leaf, path->slots[0],
+				      struct btrfs_shared_data_ref);
+		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	} else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
+		struct btrfs_extent_ref_v0 *ref0;
+		ref0 = btrfs_item_ptr(leaf, path->slots[0],
+				      struct btrfs_extent_ref_v0);
+		num_refs = btrfs_ref_count_v0(leaf, ref0);
+#endif
+	} else {
+		BUG();
+	}
+
 	BUG_ON(num_refs < refs_to_drop);
 	num_refs -= refs_to_drop;
+
 	if (num_refs == 0) {
 		ret = btrfs_del_item(trans, root, path);
 	} else {
-		btrfs_set_ref_num_refs(leaf, ref, num_refs);
+		if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
+			btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
+		else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
+			btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+		else {
+			struct btrfs_extent_ref_v0 *ref0;
+			ref0 = btrfs_item_ptr(leaf, path->slots[0],
+					struct btrfs_extent_ref_v0);
+			btrfs_set_ref_count_v0(leaf, ref0, num_refs);
+		}
+#endif
 		btrfs_mark_buffer_dirty(leaf);
 	}
-	btrfs_release_path(root, path);
 	return ret;
 }
 
-#ifdef BIO_RW_DISCARD
-static void btrfs_issue_discard(struct block_device *bdev,
-				u64 start, u64 len)
+static noinline u32 extent_data_ref_count(struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  struct btrfs_extent_inline_ref *iref)
 {
-	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
-}
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_data_ref *ref1;
+	struct btrfs_shared_data_ref *ref2;
+	u32 num_refs = 0;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	if (iref) {
+		if (btrfs_extent_inline_ref_type(leaf, iref) ==
+		    BTRFS_EXTENT_DATA_REF_KEY) {
+			ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
+			num_refs = btrfs_extent_data_ref_count(leaf, ref1);
+		} else {
+			ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
+			num_refs = btrfs_shared_data_ref_count(leaf, ref2);
+		}
+	} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
+		ref1 = btrfs_item_ptr(leaf, path->slots[0],
+				      struct btrfs_extent_data_ref);
+		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
+	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
+		ref2 = btrfs_item_ptr(leaf, path->slots[0],
+				      struct btrfs_shared_data_ref);
+		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	} else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
+		struct btrfs_extent_ref_v0 *ref0;
+		ref0 = btrfs_item_ptr(leaf, path->slots[0],
+				      struct btrfs_extent_ref_v0);
+		num_refs = btrfs_ref_count_v0(leaf, ref0);
 #endif
+	} else {
+		WARN_ON(1);
+	}
+	return num_refs;
+}
 
-static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
-				u64 num_bytes)
+static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, u64 parent,
+					  u64 root_objectid)
 {
-#ifdef BIO_RW_DISCARD
+	struct btrfs_key key;
 	int ret;
-	u64 map_length = num_bytes;
-	struct btrfs_multi_bio *multi = NULL;
-
-	/* Tell the block device(s) that the sectors can be discarded */
-	ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
-			      bytenr, &map_length, &multi, 0);
-	if (!ret) {
-		struct btrfs_bio_stripe *stripe = multi->stripes;
-		int i;
-
-		if (map_length > num_bytes)
-			map_length = num_bytes;
 
-		for (i = 0; i < multi->num_stripes; i++, stripe++) {
-			btrfs_issue_discard(stripe->dev->bdev,
-					    stripe->physical,
-					    map_length);
-		}
-		kfree(multi);
+	key.objectid = bytenr;
+	if (parent) {
+		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
+		key.offset = parent;
+	} else {
+		key.type = BTRFS_TREE_BLOCK_REF_KEY;
+		key.offset = root_objectid;
 	}
 
-	return ret;
-#else
-	return 0;
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret > 0)
+		ret = -ENOENT;
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (ret == -ENOENT && parent) {
+		btrfs_release_path(root, path);
+		key.type = BTRFS_EXTENT_REF_V0_KEY;
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (ret > 0)
+			ret = -ENOENT;
+	}
 #endif
+	return ret;
 }
 
-static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
-				     struct btrfs_root *root, u64 bytenr,
-				     u64 num_bytes,
-				     u64 orig_parent, u64 parent,
-				     u64 orig_root, u64 ref_root,
-				     u64 orig_generation, u64 ref_generation,
-				     u64 owner_objectid)
+static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, u64 parent,
+					  u64 root_objectid)
 {
+	struct btrfs_key key;
 	int ret;
-	int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
 
-	ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
-				       orig_parent, parent, orig_root,
-				       ref_root, orig_generation,
-				       ref_generation, owner_objectid, pin);
-	BUG_ON(ret);
+	key.objectid = bytenr;
+	if (parent) {
+		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
+		key.offset = parent;
+	} else {
+		key.type = BTRFS_TREE_BLOCK_REF_KEY;
+		key.offset = root_objectid;
+	}
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+	btrfs_release_path(root, path);
 	return ret;
 }
 
-int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, u64 bytenr,
-			    u64 num_bytes, u64 orig_parent, u64 parent,
-			    u64 ref_root, u64 ref_generation,
-			    u64 owner_objectid)
+static inline int extent_ref_type(u64 parent, u64 owner)
 {
-	int ret;
-	if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
-	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
-		return 0;
-
-	ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
-					orig_parent, parent, ref_root,
-					ref_root, ref_generation,
-					ref_generation, owner_objectid);
-	return ret;
+	int type;
+	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
+		if (parent > 0)
+			type = BTRFS_SHARED_BLOCK_REF_KEY;
+		else
+			type = BTRFS_TREE_BLOCK_REF_KEY;
+	} else {
+		if (parent > 0)
+			type = BTRFS_SHARED_DATA_REF_KEY;
+		else
+			type = BTRFS_EXTENT_DATA_REF_KEY;
+	}
+	return type;
 }
-static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
-				  struct btrfs_root *root, u64 bytenr,
-				  u64 num_bytes,
-				  u64 orig_parent, u64 parent,
-				  u64 orig_root, u64 ref_root,
-				  u64 orig_generation, u64 ref_generation,
-				  u64 owner_objectid)
-{
-	int ret;
 
-	ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
-				    ref_generation, owner_objectid,
-				    BTRFS_ADD_DELAYED_REF, 0);
-	BUG_ON(ret);
-	return ret;
-}
+static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
 
-static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root, u64 bytenr,
-			  u64 num_bytes, u64 parent, u64 ref_root,
-			  u64 ref_generation, u64 owner_objectid,
-			  int refs_to_add)
 {
-	struct btrfs_path *path;
-	int ret;
-	struct btrfs_key key;
-	struct extent_buffer *l;
-	struct btrfs_extent_item *item;
-	u32 refs;
+	int level;
+	BUG_ON(!path->keep_locks);
+	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
+		if (!path->nodes[level])
+			break;
+		btrfs_assert_tree_locked(path->nodes[level]);
+		if (path->slots[level] + 1 >=
+		    btrfs_header_nritems(path->nodes[level]))
+			continue;
+		if (level == 0)
+			btrfs_item_key_to_cpu(path->nodes[level], key,
+					      path->slots[level] + 1);
+		else
+			btrfs_node_key_to_cpu(path->nodes[level], key,
+					      path->slots[level] + 1);
+		return 0;
+	}
+	return 1;
+}
 
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
+/*
+ * look for inline back ref. if back ref is found, *ref_ret is set
+ * to the address of inline back ref, and 0 is returned.
+ *
+ * if back ref isn't found, *ref_ret is set to the address where it
+ * should be inserted, and -ENOENT is returned.
+ *
+ * if insert is true and there are too many inline back refs, the path
+ * points to the extent item, and -EAGAIN is returned.
+ *
+ * NOTE: inline back refs are ordered in the same way that back ref
+ *	 items in the tree are ordered.
+ */
+static noinline_for_stack
+int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct btrfs_extent_inline_ref **ref_ret,
+				 u64 bytenr, u64 num_bytes,
+				 u64 parent, u64 root_objectid,
+				 u64 owner, u64 offset, int insert)
+{
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_inline_ref *iref;
+	u64 flags;
+	u64 item_size;
+	unsigned long ptr;
+	unsigned long end;
+	int extra_size;
+	int type;
+	int want;
+	int ret;
+	int err = 0;
 
-	path->reada = 1;
-	path->leave_spinning = 1;
 	key.objectid = bytenr;
 	key.type = BTRFS_EXTENT_ITEM_KEY;
 	key.offset = num_bytes;
 
-	/* first find the extent item and update its reference count */
-	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
-				path, 0, 1);
+	want = extent_ref_type(parent, owner);
+	if (insert) {
+		extra_size = btrfs_extent_inline_ref_size(want);
+		path->keep_locks = 1;
+	} else
+		extra_size = -1;
+	ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
 	if (ret < 0) {
-		btrfs_set_path_blocking(path);
+		err = ret;
+		goto out;
+	}
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (item_size < sizeof(*ei)) {
+		if (!insert) {
+			err = -ENOENT;
+			goto out;
+		}
+		ret = convert_extent_item_v0(trans, root, path, owner,
+					     extra_size);
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+		leaf = path->nodes[0];
+		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+	}
+#endif
+	BUG_ON(item_size < sizeof(*ei));
+
+	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	flags = btrfs_extent_flags(leaf, ei);
+
+	ptr = (unsigned long)(ei + 1);
+	end = (unsigned long)ei + item_size;
+
+	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
+		ptr += sizeof(struct btrfs_tree_block_info);
+		BUG_ON(ptr > end);
+	} else {
+		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
+	}
+
+	err = -ENOENT;
+	while (1) {
+		if (ptr >= end) {
+			WARN_ON(ptr > end);
+			break;
+		}
+		iref = (struct btrfs_extent_inline_ref *)ptr;
+		type = btrfs_extent_inline_ref_type(leaf, iref);
+		if (want < type)
+			break;
+		if (want > type) {
+			ptr += btrfs_extent_inline_ref_size(type);
+			continue;
+		}
+
+		if (type == BTRFS_EXTENT_DATA_REF_KEY) {
+			struct btrfs_extent_data_ref *dref;
+			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
+			if (match_extent_data_ref(leaf, dref, root_objectid,
+						  owner, offset)) {
+				err = 0;
+				break;
+			}
+			if (hash_extent_data_ref_item(leaf, dref) <
+			    hash_extent_data_ref(root_objectid, owner, offset))
+				break;
+		} else {
+			u64 ref_offset;
+			ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
+			if (parent > 0) {
+				if (parent == ref_offset) {
+					err = 0;
+					break;
+				}
+				if (ref_offset < parent)
+					break;
+			} else {
+				if (root_objectid == ref_offset) {
+					err = 0;
+					break;
+				}
+				if (ref_offset < root_objectid)
+					break;
+			}
+		}
+		ptr += btrfs_extent_inline_ref_size(type);
+	}
+	if (err == -ENOENT && insert) {
+		if (item_size + extra_size >=
+		    BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
+			err = -EAGAIN;
+			goto out;
+		}
+		/*
+		 * To add new inline back ref, we have to make sure
+		 * there is no corresponding back ref item.
+		 * For simplicity, we just do not add new inline back
+		 * ref if there is any kind of item for this block
+		 */
+		if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
+		    key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
+			err = -EAGAIN;
+			goto out;
+		}
+	}
+	*ref_ret = (struct btrfs_extent_inline_ref *)ptr;
+out:
+	if (insert) {
+		path->keep_locks = 0;
+		btrfs_unlock_up_safe(path, 1);
+	}
+	return err;
+}
+
+/*
+ * helper to add new inline back ref
+ */
+static noinline_for_stack
+int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct btrfs_path *path,
+				struct btrfs_extent_inline_ref *iref,
+				u64 parent, u64 root_objectid,
+				u64 owner, u64 offset, int refs_to_add,
+				struct btrfs_delayed_extent_op *extent_op)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_extent_item *ei;
+	unsigned long ptr;
+	unsigned long end;
+	unsigned long item_offset;
+	u64 refs;
+	int size;
+	int type;
+	int ret;
+
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	item_offset = (unsigned long)iref - (unsigned long)ei;
+
+	type = extent_ref_type(parent, owner);
+	size = btrfs_extent_inline_ref_size(type);
+
+	ret = btrfs_extend_item(trans, root, path, size);
+	BUG_ON(ret);
+
+	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, ei);
+	refs += refs_to_add;
+	btrfs_set_extent_refs(leaf, ei, refs);
+	if (extent_op)
+		__run_delayed_extent_op(extent_op, leaf, ei);
+
+	ptr = (unsigned long)ei + item_offset;
+	end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
+	if (ptr < end - size)
+		memmove_extent_buffer(leaf, ptr + size, ptr,
+				      end - size - ptr);
+
+	iref = (struct btrfs_extent_inline_ref *)ptr;
+	btrfs_set_extent_inline_ref_type(leaf, iref, type);
+	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
+		struct btrfs_extent_data_ref *dref;
+		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
+		btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
+		btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
+		btrfs_set_extent_data_ref_offset(leaf, dref, offset);
+		btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
+	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
+		struct btrfs_shared_data_ref *sref;
+		sref = (struct btrfs_shared_data_ref *)(iref + 1);
+		btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
+		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
+	} else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
+		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
+	} else {
+		btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
+	}
+	btrfs_mark_buffer_dirty(leaf);
+	return 0;
+}
+
+static int lookup_extent_backref(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct btrfs_extent_inline_ref **ref_ret,
+				 u64 bytenr, u64 num_bytes, u64 parent,
+				 u64 root_objectid, u64 owner, u64 offset)
+{
+	int ret;
+
+	ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
+					   bytenr, num_bytes, parent,
+					   root_objectid, owner, offset, 0);
+	if (ret != -ENOENT)
 		return ret;
+
+	btrfs_release_path(root, path);
+	*ref_ret = NULL;
+
+	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
+		ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
+					    root_objectid);
+	} else {
+		ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
+					     root_objectid, owner, offset);
 	}
+	return ret;
+}
 
-	if (ret > 0) {
-		WARN_ON(1);
-		btrfs_free_path(path);
-		return -EIO;
+/*
+ * helper to update/remove inline back ref
+ */
+static noinline_for_stack
+int update_inline_extent_backref(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct btrfs_extent_inline_ref *iref,
+				 int refs_to_mod,
+				 struct btrfs_delayed_extent_op *extent_op)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_data_ref *dref = NULL;
+	struct btrfs_shared_data_ref *sref = NULL;
+	unsigned long ptr;
+	unsigned long end;
+	u32 item_size;
+	int size;
+	int type;
+	int ret;
+	u64 refs;
+
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, ei);
+	WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
+	refs += refs_to_mod;
+	btrfs_set_extent_refs(leaf, ei, refs);
+	if (extent_op)
+		__run_delayed_extent_op(extent_op, leaf, ei);
+
+	type = btrfs_extent_inline_ref_type(leaf, iref);
+
+	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
+		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
+		refs = btrfs_extent_data_ref_count(leaf, dref);
+	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
+		sref = (struct btrfs_shared_data_ref *)(iref + 1);
+		refs = btrfs_shared_data_ref_count(leaf, sref);
+	} else {
+		refs = 1;
+		BUG_ON(refs_to_mod != -1);
 	}
-	l = path->nodes[0];
 
-	btrfs_item_key_to_cpu(l, &key, path->slots[0]);
-	if (key.objectid != bytenr) {
-		btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
-		printk(KERN_ERR "btrfs wanted %llu found %llu\n",
-		       (unsigned long long)bytenr,
-		       (unsigned long long)key.objectid);
-		BUG();
+	BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
+	refs += refs_to_mod;
+
+	if (refs > 0) {
+		if (type == BTRFS_EXTENT_DATA_REF_KEY)
+			btrfs_set_extent_data_ref_count(leaf, dref, refs);
+		else
+			btrfs_set_shared_data_ref_count(leaf, sref, refs);
+	} else {
+		size =  btrfs_extent_inline_ref_size(type);
+		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+		ptr = (unsigned long)iref;
+		end = (unsigned long)ei + item_size;
+		if (ptr + size < end)
+			memmove_extent_buffer(leaf, ptr, ptr + size,
+					      end - ptr - size);
+		item_size -= size;
+		ret = btrfs_truncate_item(trans, root, path, item_size, 1);
+		BUG_ON(ret);
 	}
-	BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
+	btrfs_mark_buffer_dirty(leaf);
+	return 0;
+}
 
-	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
+static noinline_for_stack
+int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 u64 bytenr, u64 num_bytes, u64 parent,
+				 u64 root_objectid, u64 owner,
+				 u64 offset, int refs_to_add,
+				 struct btrfs_delayed_extent_op *extent_op)
+{
+	struct btrfs_extent_inline_ref *iref;
+	int ret;
 
-	refs = btrfs_extent_refs(l, item);
-	btrfs_set_extent_refs(l, item, refs + refs_to_add);
-	btrfs_unlock_up_safe(path, 1);
+	ret = lookup_inline_extent_backref(trans, root, path, &iref,
+					   bytenr, num_bytes, parent,
+					   root_objectid, owner, offset, 1);
+	if (ret == 0) {
+		BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
+		ret = update_inline_extent_backref(trans, root, path, iref,
+						   refs_to_add, extent_op);
+	} else if (ret == -ENOENT) {
+		ret = setup_inline_extent_backref(trans, root, path, iref,
+						  parent, root_objectid,
+						  owner, offset, refs_to_add,
+						  extent_op);
+	}
+	return ret;
+}
 
-	btrfs_mark_buffer_dirty(path->nodes[0]);
+static int insert_extent_backref(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 u64 bytenr, u64 parent, u64 root_objectid,
+				 u64 owner, u64 offset, int refs_to_add)
+{
+	int ret;
+	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
+		BUG_ON(refs_to_add != 1);
+		ret = insert_tree_block_ref(trans, root, path, bytenr,
+					    parent, root_objectid);
+	} else {
+		ret = insert_extent_data_ref(trans, root, path, bytenr,
+					     parent, root_objectid,
+					     owner, offset, refs_to_add);
+	}
+	return ret;
+}
+
+static int remove_extent_backref(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct btrfs_extent_inline_ref *iref,
+				 int refs_to_drop, int is_data)
+{
+	int ret;
+
+	BUG_ON(!is_data && refs_to_drop != 1);
+	if (iref) {
+		ret = update_inline_extent_backref(trans, root, path, iref,
+						   -refs_to_drop, NULL);
+	} else if (is_data) {
+		ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
+	} else {
+		ret = btrfs_del_item(trans, root, path);
+	}
+	return ret;
+}
+
+#ifdef BIO_RW_DISCARD
+static void btrfs_issue_discard(struct block_device *bdev,
+				u64 start, u64 len)
+{
+	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
+}
+#endif
+
+static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
+				u64 num_bytes)
+{
+#ifdef BIO_RW_DISCARD
+	int ret;
+	u64 map_length = num_bytes;
+	struct btrfs_multi_bio *multi = NULL;
+
+	/* Tell the block device(s) that the sectors can be discarded */
+	ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
+			      bytenr, &map_length, &multi, 0);
+	if (!ret) {
+		struct btrfs_bio_stripe *stripe = multi->stripes;
+		int i;
+
+		if (map_length > num_bytes)
+			map_length = num_bytes;
+
+		for (i = 0; i < multi->num_stripes; i++, stripe++) {
+			btrfs_issue_discard(stripe->dev->bdev,
+					    stripe->physical,
+					    map_length);
+		}
+		kfree(multi);
+	}
+
+	return ret;
+#else
+	return 0;
+#endif
+}
+
+int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 bytenr, u64 num_bytes, u64 parent,
+			 u64 root_objectid, u64 owner, u64 offset)
+{
+	int ret;
+	BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
+	       root_objectid == BTRFS_TREE_LOG_OBJECTID);
+
+	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
+		ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
+					parent, root_objectid, (int)owner,
+					BTRFS_ADD_DELAYED_REF, NULL);
+	} else {
+		ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
+					parent, root_objectid, owner, offset,
+					BTRFS_ADD_DELAYED_REF, NULL);
+	}
+	return ret;
+}
+
+static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  u64 bytenr, u64 num_bytes,
+				  u64 parent, u64 root_objectid,
+				  u64 owner, u64 offset, int refs_to_add,
+				  struct btrfs_delayed_extent_op *extent_op)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_item *item;
+	u64 refs;
+	int ret;
+	int err = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 1;
+	path->leave_spinning = 1;
+	/* this will setup the path even if it fails to insert the back ref */
+	ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
+					   path, bytenr, num_bytes, parent,
+					   root_objectid, owner, offset,
+					   refs_to_add, extent_op);
+	if (ret == 0)
+		goto out;
+
+	if (ret != -EAGAIN) {
+		err = ret;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, item);
+	btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
+	if (extent_op)
+		__run_delayed_extent_op(extent_op, leaf, item);
 
+	btrfs_mark_buffer_dirty(leaf);
 	btrfs_release_path(root->fs_info->extent_root, path);
 
 	path->reada = 1;
@@ -802,56 +1514,197 @@ static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
 
 	/* now insert the actual backref */
 	ret = insert_extent_backref(trans, root->fs_info->extent_root,
-				    path, bytenr, parent,
-				    ref_root, ref_generation,
-				    owner_objectid, refs_to_add);
+				    path, bytenr, parent, root_objectid,
+				    owner, offset, refs_to_add);
 	BUG_ON(ret);
+out:
 	btrfs_free_path(path);
-	return 0;
+	return err;
 }
 
-int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
-			 struct btrfs_root *root,
-			 u64 bytenr, u64 num_bytes, u64 parent,
-			 u64 ref_root, u64 ref_generation,
-			 u64 owner_objectid)
+static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct btrfs_delayed_ref_node *node,
+				struct btrfs_delayed_extent_op *extent_op,
+				int insert_reserved)
 {
+	int ret = 0;
+	struct btrfs_delayed_data_ref *ref;
+	struct btrfs_key ins;
+	u64 parent = 0;
+	u64 ref_root = 0;
+	u64 flags = 0;
+
+	ins.objectid = node->bytenr;
+	ins.offset = node->num_bytes;
+	ins.type = BTRFS_EXTENT_ITEM_KEY;
+
+	ref = btrfs_delayed_node_to_data_ref(node);
+	if (node->type == BTRFS_SHARED_DATA_REF_KEY)
+		parent = ref->parent;
+	else
+		ref_root = ref->root;
+
+	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
+		if (extent_op) {
+			BUG_ON(extent_op->update_key);
+			flags |= extent_op->flags_to_set;
+		}
+		ret = alloc_reserved_file_extent(trans, root,
+						 parent, ref_root, flags,
+						 ref->objectid, ref->offset,
+						 &ins, node->ref_mod);
+		update_reserved_extents(root, ins.objectid, ins.offset, 0);
+	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
+		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
+					     node->num_bytes, parent,
+					     ref_root, ref->objectid,
+					     ref->offset, node->ref_mod,
+					     extent_op);
+	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
+		ret = __btrfs_free_extent(trans, root, node->bytenr,
+					  node->num_bytes, parent,
+					  ref_root, ref->objectid,
+					  ref->offset, node->ref_mod,
+					  extent_op);
+	} else {
+		BUG();
+	}
+	return ret;
+}
+
+static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
+				    struct extent_buffer *leaf,
+				    struct btrfs_extent_item *ei)
+{
+	u64 flags = btrfs_extent_flags(leaf, ei);
+	if (extent_op->update_flags) {
+		flags |= extent_op->flags_to_set;
+		btrfs_set_extent_flags(leaf, ei, flags);
+	}
+
+	if (extent_op->update_key) {
+		struct btrfs_tree_block_info *bi;
+		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
+		bi = (struct btrfs_tree_block_info *)(ei + 1);
+		btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
+	}
+}
+
+static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_delayed_ref_node *node,
+				 struct btrfs_delayed_extent_op *extent_op)
+{
+	struct btrfs_key key;
+	struct btrfs_path *path;
+	struct btrfs_extent_item *ei;
+	struct extent_buffer *leaf;
+	u32 item_size;
 	int ret;
-	if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
-	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
-		return 0;
+	int err = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = node->bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = node->num_bytes;
+
+	path->reada = 1;
+	path->leave_spinning = 1;
+	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
+				path, 0, 1);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
+	if (ret > 0) {
+		err = -EIO;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (item_size < sizeof(*ei)) {
+		ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
+					     path, (u64)-1, 0);
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+		leaf = path->nodes[0];
+		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+	}
+#endif
+	BUG_ON(item_size < sizeof(*ei));
+	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	__run_delayed_extent_op(extent_op, leaf, ei);
 
-	ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
-				     0, ref_root, 0, ref_generation,
-				     owner_objectid);
-	return ret;
+	btrfs_mark_buffer_dirty(leaf);
+out:
+	btrfs_free_path(path);
+	return err;
 }
 
-static int drop_delayed_ref(struct btrfs_trans_handle *trans,
-					struct btrfs_root *root,
-					struct btrfs_delayed_ref_node *node)
+static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct btrfs_delayed_ref_node *node,
+				struct btrfs_delayed_extent_op *extent_op,
+				int insert_reserved)
 {
 	int ret = 0;
-	struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
+	struct btrfs_delayed_tree_ref *ref;
+	struct btrfs_key ins;
+	u64 parent = 0;
+	u64 ref_root = 0;
 
-	BUG_ON(node->ref_mod == 0);
-	ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
-				  node->parent, ref->root, ref->generation,
-				  ref->owner_objectid, ref->pin, node->ref_mod);
+	ins.objectid = node->bytenr;
+	ins.offset = node->num_bytes;
+	ins.type = BTRFS_EXTENT_ITEM_KEY;
 
+	ref = btrfs_delayed_node_to_tree_ref(node);
+	if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
+		parent = ref->parent;
+	else
+		ref_root = ref->root;
+
+	BUG_ON(node->ref_mod != 1);
+	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
+		BUG_ON(!extent_op || !extent_op->update_flags ||
+		       !extent_op->update_key);
+		ret = alloc_reserved_tree_block(trans, root,
+						parent, ref_root,
+						extent_op->flags_to_set,
+						&extent_op->key,
+						ref->level, &ins);
+		update_reserved_extents(root, ins.objectid, ins.offset, 0);
+	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
+		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
+					     node->num_bytes, parent, ref_root,
+					     ref->level, 0, 1, extent_op);
+	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
+		ret = __btrfs_free_extent(trans, root, node->bytenr,
+					  node->num_bytes, parent, ref_root,
+					  ref->level, 0, 1, extent_op);
+	} else {
+		BUG();
+	}
 	return ret;
 }
 
+
 /* helper function to actually process a single delayed ref entry */
-static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
-					struct btrfs_root *root,
-					struct btrfs_delayed_ref_node *node,
-					int insert_reserved)
+static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct btrfs_delayed_ref_node *node,
+			       struct btrfs_delayed_extent_op *extent_op,
+			       int insert_reserved)
 {
 	int ret;
-	struct btrfs_delayed_ref *ref;
-
-	if (node->parent == (u64)-1) {
+	if (btrfs_delayed_ref_is_head(node)) {
 		struct btrfs_delayed_ref_head *head;
 		/*
 		 * we've hit the end of the chain and we were supposed
@@ -859,44 +1712,35 @@ static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 		 * deleted before we ever needed to insert it, so all
 		 * we have to do is clean up the accounting
 		 */
+		BUG_ON(extent_op);
+		head = btrfs_delayed_node_to_head(node);
 		if (insert_reserved) {
+			if (head->is_data) {
+				ret = btrfs_del_csums(trans, root,
+						      node->bytenr,
+						      node->num_bytes);
+				BUG_ON(ret);
+			}
+			btrfs_update_pinned_extents(root, node->bytenr,
+						    node->num_bytes, 1);
 			update_reserved_extents(root, node->bytenr,
 						node->num_bytes, 0);
 		}
-		head = btrfs_delayed_node_to_head(node);
 		mutex_unlock(&head->mutex);
 		return 0;
 	}
 
-	ref = btrfs_delayed_node_to_ref(node);
-	if (ref->action == BTRFS_ADD_DELAYED_REF) {
-		if (insert_reserved) {
-			struct btrfs_key ins;
-
-			ins.objectid = node->bytenr;
-			ins.offset = node->num_bytes;
-			ins.type = BTRFS_EXTENT_ITEM_KEY;
-
-			/* record the full extent allocation */
-			ret = __btrfs_alloc_reserved_extent(trans, root,
-					node->parent, ref->root,
-					ref->generation, ref->owner_objectid,
-					&ins, node->ref_mod);
-			update_reserved_extents(root, node->bytenr,
-						node->num_bytes, 0);
-		} else {
-			/* just add one backref */
-			ret = add_extent_ref(trans, root, node->bytenr,
-				     node->num_bytes,
-				     node->parent, ref->root, ref->generation,
-				     ref->owner_objectid, node->ref_mod);
-		}
-		BUG_ON(ret);
-	} else if (ref->action == BTRFS_DROP_DELAYED_REF) {
-		WARN_ON(insert_reserved);
-		ret = drop_delayed_ref(trans, root, node);
-	}
-	return 0;
+	if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
+	    node->type == BTRFS_SHARED_BLOCK_REF_KEY)
+		ret = run_delayed_tree_ref(trans, root, node, extent_op,
+					   insert_reserved);
+	else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
+		 node->type == BTRFS_SHARED_DATA_REF_KEY)
+		ret = run_delayed_data_ref(trans, root, node, extent_op,
+					   insert_reserved);
+	else
+		BUG();
+	return ret;
 }
 
 static noinline struct btrfs_delayed_ref_node *
@@ -919,7 +1763,7 @@ again:
 				rb_node);
 		if (ref->bytenr != head->node.bytenr)
 			break;
-		if (btrfs_delayed_node_to_ref(ref)->action == action)
+		if (ref->action == action)
 			return ref;
 		node = rb_prev(node);
 	}
@@ -937,6 +1781,7 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 	struct btrfs_delayed_ref_root *delayed_refs;
 	struct btrfs_delayed_ref_node *ref;
 	struct btrfs_delayed_ref_head *locked_ref = NULL;
+	struct btrfs_delayed_extent_op *extent_op;
 	int ret;
 	int count = 0;
 	int must_insert_reserved = 0;
@@ -975,6 +1820,9 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 		must_insert_reserved = locked_ref->must_insert_reserved;
 		locked_ref->must_insert_reserved = 0;
 
+		extent_op = locked_ref->extent_op;
+		locked_ref->extent_op = NULL;
+
 		/*
 		 * locked_ref is the head node, so we have to go one
 		 * node back for any delayed ref updates
@@ -986,6 +1834,25 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 			 * so that any accounting fixes can happen
 			 */
 			ref = &locked_ref->node;
+
+			if (extent_op && must_insert_reserved) {
+				kfree(extent_op);
+				extent_op = NULL;
+			}
+
+			if (extent_op) {
+				spin_unlock(&delayed_refs->lock);
+
+				ret = run_delayed_extent_op(trans, root,
+							    ref, extent_op);
+				BUG_ON(ret);
+				kfree(extent_op);
+
+				cond_resched();
+				spin_lock(&delayed_refs->lock);
+				continue;
+			}
+
 			list_del_init(&locked_ref->cluster);
 			locked_ref = NULL;
 		}
@@ -993,14 +1860,17 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 		ref->in_tree = 0;
 		rb_erase(&ref->rb_node, &delayed_refs->root);
 		delayed_refs->num_entries--;
+
 		spin_unlock(&delayed_refs->lock);
 
-		ret = run_one_delayed_ref(trans, root, ref,
+		ret = run_one_delayed_ref(trans, root, ref, extent_op,
 					  must_insert_reserved);
 		BUG_ON(ret);
-		btrfs_put_delayed_ref(ref);
 
+		btrfs_put_delayed_ref(ref);
+		kfree(extent_op);
 		count++;
+
 		cond_resched();
 		spin_lock(&delayed_refs->lock);
 	}
@@ -1095,25 +1965,112 @@ out:
 	return 0;
 }
 
-int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root, u64 objectid, u64 bytenr)
+int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				u64 bytenr, u64 num_bytes, u64 flags,
+				int is_data)
+{
+	struct btrfs_delayed_extent_op *extent_op;
+	int ret;
+
+	extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+	if (!extent_op)
+		return -ENOMEM;
+
+	extent_op->flags_to_set = flags;
+	extent_op->update_flags = 1;
+	extent_op->update_key = 0;
+	extent_op->is_data = is_data ? 1 : 0;
+
+	ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
+	if (ret)
+		kfree(extent_op);
+	return ret;
+}
+
+static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      u64 objectid, u64 offset, u64 bytenr)
+{
+	struct btrfs_delayed_ref_head *head;
+	struct btrfs_delayed_ref_node *ref;
+	struct btrfs_delayed_data_ref *data_ref;
+	struct btrfs_delayed_ref_root *delayed_refs;
+	struct rb_node *node;
+	int ret = 0;
+
+	ret = -ENOENT;
+	delayed_refs = &trans->transaction->delayed_refs;
+	spin_lock(&delayed_refs->lock);
+	head = btrfs_find_delayed_ref_head(trans, bytenr);
+	if (!head)
+		goto out;
+
+	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);
+		return -EAGAIN;
+	}
+
+	node = rb_prev(&head->node.rb_node);
+	if (!node)
+		goto out_unlock;
+
+	ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
+
+	if (ref->bytenr != bytenr)
+		goto out_unlock;
+
+	ret = 1;
+	if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
+		goto out_unlock;
+
+	data_ref = btrfs_delayed_node_to_data_ref(ref);
+
+	node = rb_prev(node);
+	if (node) {
+		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
+		if (ref->bytenr == bytenr)
+			goto out_unlock;
+	}
+
+	if (data_ref->root != root->root_key.objectid ||
+	    data_ref->objectid != objectid || data_ref->offset != offset)
+		goto out_unlock;
+
+	ret = 0;
+out_unlock:
+	mutex_unlock(&head->mutex);
+out:
+	spin_unlock(&delayed_refs->lock);
+	return ret;
+}
+
+static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_path *path,
+					u64 objectid, u64 offset, u64 bytenr)
 {
 	struct btrfs_root *extent_root = root->fs_info->extent_root;
-	struct btrfs_path *path;
 	struct extent_buffer *leaf;
-	struct btrfs_extent_ref *ref_item;
+	struct btrfs_extent_data_ref *ref;
+	struct btrfs_extent_inline_ref *iref;
+	struct btrfs_extent_item *ei;
 	struct btrfs_key key;
-	struct btrfs_key found_key;
-	u64 ref_root;
-	u64 last_snapshot;
-	u32 nritems;
+	u32 item_size;
 	int ret;
 
 	key.objectid = bytenr;
 	key.offset = (u64)-1;
 	key.type = BTRFS_EXTENT_ITEM_KEY;
 
-	path = btrfs_alloc_path();
 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 	if (ret < 0)
 		goto out;
@@ -1125,55 +2082,83 @@ int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
 
 	path->slots[0]--;
 	leaf = path->nodes[0];
-	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 
-	if (found_key.objectid != bytenr ||
-	    found_key.type != BTRFS_EXTENT_ITEM_KEY)
+	if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
 		goto out;
 
-	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
-	while (1) {
-		leaf = path->nodes[0];
-		nritems = btrfs_header_nritems(leaf);
-		if (path->slots[0] >= nritems) {
-			ret = btrfs_next_leaf(extent_root, path);
-			if (ret < 0)
-				goto out;
-			if (ret == 0)
-				continue;
-			break;
-		}
-		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
-		if (found_key.objectid != bytenr)
-			break;
+	ret = 1;
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (item_size < sizeof(*ei)) {
+		WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
+		goto out;
+	}
+#endif
+	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 
-		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
-			path->slots[0]++;
-			continue;
-		}
+	if (item_size != sizeof(*ei) +
+	    btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
+		goto out;
 
-		ref_item = btrfs_item_ptr(leaf, path->slots[0],
-					  struct btrfs_extent_ref);
-		ref_root = btrfs_ref_root(leaf, ref_item);
-		if ((ref_root != root->root_key.objectid &&
-		     ref_root != BTRFS_TREE_LOG_OBJECTID) ||
-		     objectid != btrfs_ref_objectid(leaf, ref_item)) {
-			ret = 1;
-			goto out;
-		}
-		if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
-			ret = 1;
+	if (btrfs_extent_generation(leaf, ei) <=
+	    btrfs_root_last_snapshot(&root->root_item))
+		goto out;
+
+	iref = (struct btrfs_extent_inline_ref *)(ei + 1);
+	if (btrfs_extent_inline_ref_type(leaf, iref) !=
+	    BTRFS_EXTENT_DATA_REF_KEY)
+		goto out;
+
+	ref = (struct btrfs_extent_data_ref *)(&iref->offset);
+	if (btrfs_extent_refs(leaf, ei) !=
+	    btrfs_extent_data_ref_count(leaf, ref) ||
+	    btrfs_extent_data_ref_root(leaf, ref) !=
+	    root->root_key.objectid ||
+	    btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
+	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
+		goto out;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 objectid, u64 offset, u64 bytenr)
+{
+	struct btrfs_path *path;
+	int ret;
+	int ret2;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOENT;
+
+	do {
+		ret = check_committed_ref(trans, root, path, objectid,
+					  offset, bytenr);
+		if (ret && ret != -ENOENT)
 			goto out;
-		}
 
-		path->slots[0]++;
+		ret2 = check_delayed_ref(trans, root, path, objectid,
+					 offset, bytenr);
+	} while (ret2 == -EAGAIN);
+
+	if (ret2 && ret2 != -ENOENT) {
+		ret = ret2;
+		goto out;
 	}
-	ret = 0;
+
+	if (ret != -ENOENT || ret2 != -ENOENT)
+		ret = 0;
 out:
 	btrfs_free_path(path);
 	return ret;
 }
 
+#if 0
 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		    struct extent_buffer *buf, u32 nr_extents)
 {
@@ -1291,191 +2276,49 @@ static int refsort_cmp(const void *a_void, const void *b_void)
 		return 1;
 	return 0;
 }
+#endif
 
-
-noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
+static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
-			   struct extent_buffer *orig_buf,
-			   struct extent_buffer *buf, u32 *nr_extents)
+			   struct extent_buffer *buf,
+			   int full_backref, int inc)
 {
 	u64 bytenr;
+	u64 num_bytes;
+	u64 parent;
 	u64 ref_root;
-	u64 orig_root;
-	u64 ref_generation;
-	u64 orig_generation;
-	struct refsort *sorted;
 	u32 nritems;
-	u32 nr_file_extents = 0;
 	struct btrfs_key key;
 	struct btrfs_file_extent_item *fi;
 	int i;
 	int level;
 	int ret = 0;
-	int faili = 0;
-	int refi = 0;
-	int slot;
 	int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
-			    u64, u64, u64, u64, u64, u64, u64, u64, u64);
+			    u64, u64, u64, u64, u64, u64);
 
 	ref_root = btrfs_header_owner(buf);
-	ref_generation = btrfs_header_generation(buf);
-	orig_root = btrfs_header_owner(orig_buf);
-	orig_generation = btrfs_header_generation(orig_buf);
-
 	nritems = btrfs_header_nritems(buf);
 	level = btrfs_header_level(buf);
 
-	sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
-	BUG_ON(!sorted);
-
-	if (root->ref_cows) {
-		process_func = __btrfs_inc_extent_ref;
-	} else {
-		if (level == 0 &&
-		    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
-			goto out;
-		if (level != 0 &&
-		    root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
-			goto out;
-		process_func = __btrfs_update_extent_ref;
-	}
-
-	/*
-	 * we make two passes through the items.  In the first pass we
-	 * only record the byte number and slot.  Then we sort based on
-	 * byte number and do the actual work based on the sorted results
-	 */
-	for (i = 0; i < nritems; i++) {
-		cond_resched();
-		if (level == 0) {
-			btrfs_item_key_to_cpu(buf, &key, i);
-			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
-				continue;
-			fi = btrfs_item_ptr(buf, i,
-					    struct btrfs_file_extent_item);
-			if (btrfs_file_extent_type(buf, fi) ==
-			    BTRFS_FILE_EXTENT_INLINE)
-				continue;
-			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
-			if (bytenr == 0)
-				continue;
-
-			nr_file_extents++;
-			sorted[refi].bytenr = bytenr;
-			sorted[refi].slot = i;
-			refi++;
-		} else {
-			bytenr = btrfs_node_blockptr(buf, i);
-			sorted[refi].bytenr = bytenr;
-			sorted[refi].slot = i;
-			refi++;
-		}
-	}
-	/*
-	 * if refi == 0, we didn't actually put anything into the sorted
-	 * array and we're done
-	 */
-	if (refi == 0)
-		goto out;
-
-	sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
-
-	for (i = 0; i < refi; i++) {
-		cond_resched();
-		slot = sorted[i].slot;
-		bytenr = sorted[i].bytenr;
-
-		if (level == 0) {
-			btrfs_item_key_to_cpu(buf, &key, slot);
-			fi = btrfs_item_ptr(buf, slot,
-					    struct btrfs_file_extent_item);
-
-			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
-			if (bytenr == 0)
-				continue;
-
-			ret = process_func(trans, root, bytenr,
-				   btrfs_file_extent_disk_num_bytes(buf, fi),
-				   orig_buf->start, buf->start,
-				   orig_root, ref_root,
-				   orig_generation, ref_generation,
-				   key.objectid);
-
-			if (ret) {
-				faili = slot;
-				WARN_ON(1);
-				goto fail;
-			}
-		} else {
-			ret = process_func(trans, root, bytenr, buf->len,
-					   orig_buf->start, buf->start,
-					   orig_root, ref_root,
-					   orig_generation, ref_generation,
-					   level - 1);
-			if (ret) {
-				faili = slot;
-				WARN_ON(1);
-				goto fail;
-			}
-		}
-	}
-out:
-	kfree(sorted);
-	if (nr_extents) {
-		if (level == 0)
-			*nr_extents = nr_file_extents;
-		else
-			*nr_extents = nritems;
-	}
-	return 0;
-fail:
-	kfree(sorted);
-	WARN_ON(1);
-	return ret;
-}
-
-int btrfs_update_ref(struct btrfs_trans_handle *trans,
-		     struct btrfs_root *root, struct extent_buffer *orig_buf,
-		     struct extent_buffer *buf, int start_slot, int nr)
-
-{
-	u64 bytenr;
-	u64 ref_root;
-	u64 orig_root;
-	u64 ref_generation;
-	u64 orig_generation;
-	struct btrfs_key key;
-	struct btrfs_file_extent_item *fi;
-	int i;
-	int ret;
-	int slot;
-	int level;
-
-	BUG_ON(start_slot < 0);
-	BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
-
-	ref_root = btrfs_header_owner(buf);
-	ref_generation = btrfs_header_generation(buf);
-	orig_root = btrfs_header_owner(orig_buf);
-	orig_generation = btrfs_header_generation(orig_buf);
-	level = btrfs_header_level(buf);
+	if (!root->ref_cows && level == 0)
+		return 0;
 
-	if (!root->ref_cows) {
-		if (level == 0 &&
-		    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
-			return 0;
-		if (level != 0 &&
-		    root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
-			return 0;
-	}
+	if (inc)
+		process_func = btrfs_inc_extent_ref;
+	else
+		process_func = btrfs_free_extent;
 
-	for (i = 0, slot = start_slot; i < nr; i++, slot++) {
-		cond_resched();
+	if (full_backref)
+		parent = buf->start;
+	else
+		parent = 0;
+
+	for (i = 0; i < nritems; i++) {
 		if (level == 0) {
-			btrfs_item_key_to_cpu(buf, &key, slot);
+			btrfs_item_key_to_cpu(buf, &key, i);
 			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 				continue;
-			fi = btrfs_item_ptr(buf, slot,
+			fi = btrfs_item_ptr(buf, i,
 					    struct btrfs_file_extent_item);
 			if (btrfs_file_extent_type(buf, fi) ==
 			    BTRFS_FILE_EXTENT_INLINE)
@@ -1483,28 +2326,39 @@ int btrfs_update_ref(struct btrfs_trans_handle *trans,
 			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 			if (bytenr == 0)
 				continue;
-			ret = __btrfs_update_extent_ref(trans, root, bytenr,
-				    btrfs_file_extent_disk_num_bytes(buf, fi),
-				    orig_buf->start, buf->start,
-				    orig_root, ref_root, orig_generation,
-				    ref_generation, key.objectid);
+
+			num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
+			key.offset -= btrfs_file_extent_offset(buf, fi);
+			ret = process_func(trans, root, bytenr, num_bytes,
+					   parent, ref_root, key.objectid,
+					   key.offset);
 			if (ret)
 				goto fail;
 		} else {
-			bytenr = btrfs_node_blockptr(buf, slot);
-			ret = __btrfs_update_extent_ref(trans, root, bytenr,
-					    buf->len, orig_buf->start,
-					    buf->start, orig_root, ref_root,
-					    orig_generation, ref_generation,
-					    level - 1);
+			bytenr = btrfs_node_blockptr(buf, i);
+			num_bytes = btrfs_level_size(root, level - 1);
+			ret = process_func(trans, root, bytenr, num_bytes,
+					   parent, ref_root, level - 1, 0);
 			if (ret)
 				goto fail;
 		}
 	}
 	return 0;
 fail:
-	WARN_ON(1);
-	return -1;
+	BUG();
+	return ret;
+}
+
+int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		  struct extent_buffer *buf, int full_backref)
+{
+	return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
+}
+
+int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		  struct extent_buffer *buf, int full_backref)
+{
+	return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
 }
 
 static int write_one_cache_group(struct btrfs_trans_handle *trans,
@@ -2007,6 +2861,24 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 	u64 old_val;
 	u64 byte_in_group;
 
+	/* block accounting for super block */
+	spin_lock(&info->delalloc_lock);
+	old_val = btrfs_super_bytes_used(&info->super_copy);
+	if (alloc)
+		old_val += num_bytes;
+	else
+		old_val -= num_bytes;
+	btrfs_set_super_bytes_used(&info->super_copy, old_val);
+
+	/* block accounting for root item */
+	old_val = btrfs_root_used(&root->root_item);
+	if (alloc)
+		old_val += num_bytes;
+	else
+		old_val -= num_bytes;
+	btrfs_set_root_used(&root->root_item, old_val);
+	spin_unlock(&info->delalloc_lock);
+
 	while (total) {
 		cache = btrfs_lookup_block_group(info, bytenr);
 		if (!cache)
@@ -2216,8 +3088,6 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
 		u64 header_owner = btrfs_header_owner(buf);
 		u64 header_transid = btrfs_header_generation(buf);
 		if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
-		    header_owner != BTRFS_TREE_RELOC_OBJECTID &&
-		    header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
 		    header_transid == trans->transid &&
 		    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
 			*must_clean = buf;
@@ -2235,63 +3105,77 @@ pinit:
 	return 0;
 }
 
-/*
- * remove an extent from the root, returns 0 on success
- */
-static int __free_extent(struct btrfs_trans_handle *trans,
-			 struct btrfs_root *root,
-			 u64 bytenr, u64 num_bytes, u64 parent,
-			 u64 root_objectid, u64 ref_generation,
-			 u64 owner_objectid, int pin, int mark_free,
-			 int refs_to_drop)
+
+static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				u64 bytenr, u64 num_bytes, u64 parent,
+				u64 root_objectid, u64 owner_objectid,
+				u64 owner_offset, int refs_to_drop,
+				struct btrfs_delayed_extent_op *extent_op)
 {
-	struct btrfs_path *path;
 	struct btrfs_key key;
+	struct btrfs_path *path;
 	struct btrfs_fs_info *info = root->fs_info;
 	struct btrfs_root *extent_root = info->extent_root;
 	struct extent_buffer *leaf;
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_inline_ref *iref;
 	int ret;
+	int is_data;
 	int extent_slot = 0;
 	int found_extent = 0;
 	int num_to_del = 1;
-	struct btrfs_extent_item *ei;
-	u32 refs;
+	u32 item_size;
+	u64 refs;
 
-	key.objectid = bytenr;
-	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
-	key.offset = num_bytes;
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
 
 	path->reada = 1;
 	path->leave_spinning = 1;
-	ret = lookup_extent_backref(trans, extent_root, path,
-				    bytenr, parent, root_objectid,
-				    ref_generation, owner_objectid, 1);
+
+	is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
+	BUG_ON(!is_data && refs_to_drop != 1);
+
+	ret = lookup_extent_backref(trans, extent_root, path, &iref,
+				    bytenr, num_bytes, parent,
+				    root_objectid, owner_objectid,
+				    owner_offset);
 	if (ret == 0) {
-		struct btrfs_key found_key;
 		extent_slot = path->slots[0];
-		while (extent_slot > 0) {
-			extent_slot--;
-			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+		while (extent_slot >= 0) {
+			btrfs_item_key_to_cpu(path->nodes[0], &key,
 					      extent_slot);
-			if (found_key.objectid != bytenr)
+			if (key.objectid != bytenr)
 				break;
-			if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
-			    found_key.offset == num_bytes) {
+			if (key.type == BTRFS_EXTENT_ITEM_KEY &&
+			    key.offset == num_bytes) {
 				found_extent = 1;
 				break;
 			}
 			if (path->slots[0] - extent_slot > 5)
 				break;
+			extent_slot--;
 		}
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+		item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
+		if (found_extent && item_size < sizeof(*ei))
+			found_extent = 0;
+#endif
 		if (!found_extent) {
+			BUG_ON(iref);
 			ret = remove_extent_backref(trans, extent_root, path,
-						    refs_to_drop);
+						    NULL, refs_to_drop,
+						    is_data);
 			BUG_ON(ret);
 			btrfs_release_path(extent_root, path);
 			path->leave_spinning = 1;
+
+			key.objectid = bytenr;
+			key.type = BTRFS_EXTENT_ITEM_KEY;
+			key.offset = num_bytes;
+
 			ret = btrfs_search_slot(trans, extent_root,
 						&key, path, -1, 1);
 			if (ret) {
@@ -2307,82 +3191,98 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 		btrfs_print_leaf(extent_root, path->nodes[0]);
 		WARN_ON(1);
 		printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
-		       "parent %llu root %llu gen %llu owner %llu\n",
+		       "parent %llu root %llu  owner %llu offset %llu\n",
 		       (unsigned long long)bytenr,
 		       (unsigned long long)parent,
 		       (unsigned long long)root_objectid,
-		       (unsigned long long)ref_generation,
-		       (unsigned long long)owner_objectid);
+		       (unsigned long long)owner_objectid,
+		       (unsigned long long)owner_offset);
 	}
 
 	leaf = path->nodes[0];
+	item_size = btrfs_item_size_nr(leaf, extent_slot);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (item_size < sizeof(*ei)) {
+		BUG_ON(found_extent || extent_slot != path->slots[0]);
+		ret = convert_extent_item_v0(trans, extent_root, path,
+					     owner_objectid, 0);
+		BUG_ON(ret < 0);
+
+		btrfs_release_path(extent_root, path);
+		path->leave_spinning = 1;
+
+		key.objectid = bytenr;
+		key.type = BTRFS_EXTENT_ITEM_KEY;
+		key.offset = num_bytes;
+
+		ret = btrfs_search_slot(trans, extent_root, &key, path,
+					-1, 1);
+		if (ret) {
+			printk(KERN_ERR "umm, got %d back from search"
+			       ", was looking for %llu\n", ret,
+			       (unsigned long long)bytenr);
+			btrfs_print_leaf(extent_root, path->nodes[0]);
+		}
+		BUG_ON(ret);
+		extent_slot = path->slots[0];
+		leaf = path->nodes[0];
+		item_size = btrfs_item_size_nr(leaf, extent_slot);
+	}
+#endif
+	BUG_ON(item_size < sizeof(*ei));
 	ei = btrfs_item_ptr(leaf, extent_slot,
 			    struct btrfs_extent_item);
-	refs = btrfs_extent_refs(leaf, ei);
-
-	/*
-	 * we're not allowed to delete the extent item if there
-	 * are other delayed ref updates pending
-	 */
+	if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+		struct btrfs_tree_block_info *bi;
+		BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
+		bi = (struct btrfs_tree_block_info *)(ei + 1);
+		WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
+	}
 
+	refs = btrfs_extent_refs(leaf, ei);
 	BUG_ON(refs < refs_to_drop);
 	refs -= refs_to_drop;
-	btrfs_set_extent_refs(leaf, ei, refs);
-	btrfs_mark_buffer_dirty(leaf);
 
-	if (refs == 0 && found_extent &&
-	    path->slots[0] == extent_slot + 1) {
-		struct btrfs_extent_ref *ref;
-		ref = btrfs_item_ptr(leaf, path->slots[0],
-				     struct btrfs_extent_ref);
-		BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
-		/* if the back ref and the extent are next to each other
-		 * they get deleted below in one shot
+	if (refs > 0) {
+		if (extent_op)
+			__run_delayed_extent_op(extent_op, leaf, ei);
+		/*
+		 * In the case of inline back ref, reference count will
+		 * be updated by remove_extent_backref
 		 */
-		path->slots[0] = extent_slot;
-		num_to_del = 2;
-	} else if (found_extent) {
-		/* otherwise delete the extent back ref */
-		ret = remove_extent_backref(trans, extent_root, path,
-					    refs_to_drop);
-		BUG_ON(ret);
-		/* if refs are 0, we need to setup the path for deletion */
-		if (refs == 0) {
-			btrfs_release_path(extent_root, path);
-			path->leave_spinning = 1;
-			ret = btrfs_search_slot(trans, extent_root, &key, path,
-						-1, 1);
+		if (iref) {
+			BUG_ON(!found_extent);
+		} else {
+			btrfs_set_extent_refs(leaf, ei, refs);
+			btrfs_mark_buffer_dirty(leaf);
+		}
+		if (found_extent) {
+			ret = remove_extent_backref(trans, extent_root, path,
+						    iref, refs_to_drop,
+						    is_data);
 			BUG_ON(ret);
 		}
-	}
-
-	if (refs == 0) {
-		u64 super_used;
-		u64 root_used;
+	} else {
+		int mark_free = 0;
 		struct extent_buffer *must_clean = NULL;
 
-		if (pin) {
-			ret = pin_down_bytes(trans, root, path,
-				bytenr, num_bytes,
-				owner_objectid >= BTRFS_FIRST_FREE_OBJECTID,
-				&must_clean);
-			if (ret > 0)
-				mark_free = 1;
-			BUG_ON(ret < 0);
+		if (found_extent) {
+			BUG_ON(is_data && refs_to_drop !=
+			       extent_data_ref_count(root, path, iref));
+			if (iref) {
+				BUG_ON(path->slots[0] != extent_slot);
+			} else {
+				BUG_ON(path->slots[0] != extent_slot + 1);
+				path->slots[0] = extent_slot;
+				num_to_del = 2;
+			}
 		}
 
-		/* block accounting for super block */
-		spin_lock(&info->delalloc_lock);
-		super_used = btrfs_super_bytes_used(&info->super_copy);
-		btrfs_set_super_bytes_used(&info->super_copy,
-					   super_used - num_bytes);
-
-		/* block accounting for root item */
-		root_used = btrfs_root_used(&root->root_item);
-		btrfs_set_root_used(&root->root_item,
-					   root_used - num_bytes);
-		spin_unlock(&info->delalloc_lock);
-
+		ret = pin_down_bytes(trans, root, path, bytenr,
+				     num_bytes, is_data, &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
@@ -2403,7 +3303,7 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 			free_extent_buffer(must_clean);
 		}
 
-		if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+		if (is_data) {
 			ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
 			BUG_ON(ret);
 		} else {
@@ -2420,34 +3320,6 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-/*
- * remove an extent from the root, returns 0 on success
- */
-static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
-					struct btrfs_root *root,
-					u64 bytenr, u64 num_bytes, u64 parent,
-					u64 root_objectid, u64 ref_generation,
-					u64 owner_objectid, int pin,
-					int refs_to_drop)
-{
-	WARN_ON(num_bytes < root->sectorsize);
-
-	/*
-	 * if metadata always pin
-	 * if data pin when any transaction has committed this
-	 */
-	if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
-	    ref_generation != trans->transid)
-		pin = 1;
-
-	if (ref_generation != trans->transid)
-		pin = 1;
-
-	return __free_extent(trans, root, bytenr, num_bytes, parent,
-			    root_objectid, ref_generation,
-			    owner_objectid, pin, pin == 0, refs_to_drop);
-}
-
 /*
  * when we free an extent, it is possible (and likely) that we free the last
  * delayed ref for that extent as well.  This searches the delayed ref tree for
@@ -2479,6 +3351,13 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (ref->bytenr == bytenr)
 		goto out;
 
+	if (head->extent_op) {
+		if (!head->must_insert_reserved)
+			goto out;
+		kfree(head->extent_op);
+		head->extent_op = NULL;
+	}
+
 	/*
 	 * waiting for the lock here would deadlock.  If someone else has it
 	 * locked they are already in the process of dropping it anyway
@@ -2507,7 +3386,8 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	spin_unlock(&delayed_refs->lock);
 
 	ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
-				  &head->node, head->must_insert_reserved);
+				  &head->node, head->extent_op,
+				  head->must_insert_reserved);
 	BUG_ON(ret);
 	btrfs_put_delayed_ref(&head->node);
 	return 0;
@@ -2519,32 +3399,32 @@ out:
 int btrfs_free_extent(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root,
 		      u64 bytenr, u64 num_bytes, u64 parent,
-		      u64 root_objectid, u64 ref_generation,
-		      u64 owner_objectid, int pin)
+		      u64 root_objectid, u64 owner, u64 offset)
 {
 	int ret;
 
 	/*
 	 * tree log blocks never actually go into the extent allocation
 	 * tree, just update pinning info and exit early.
-	 *
-	 * data extents referenced by the tree log do need to have
-	 * their reference counts bumped.
 	 */
-	if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
-	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+	if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
+		WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
 		/* unlocks the pinned mutex */
 		btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
 		update_reserved_extents(root, bytenr, num_bytes, 0);
 		ret = 0;
-	} else {
-		ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
-				       root_objectid, ref_generation,
-				       owner_objectid,
-				       BTRFS_DROP_DELAYED_REF, 1);
+	} else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
+		ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
+					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,
+					offset, BTRFS_DROP_DELAYED_REF, NULL);
+		BUG_ON(ret);
 	}
 	return ret;
 }
@@ -2719,7 +3599,7 @@ refill_cluster:
 			last_ptr_loop = 0;
 
 			/* allocate a cluster in this block group */
-			ret = btrfs_find_space_cluster(trans,
+			ret = btrfs_find_space_cluster(trans, root,
 					       block_group, last_ptr,
 					       offset, num_bytes,
 					       empty_cluster + empty_size);
@@ -2969,99 +3849,147 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
-					 struct btrfs_root *root, u64 parent,
-					 u64 root_objectid, u64 ref_generation,
-					 u64 owner, struct btrfs_key *ins,
-					 int ref_mod)
+static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      u64 parent, u64 root_objectid,
+				      u64 flags, u64 owner, u64 offset,
+				      struct btrfs_key *ins, int ref_mod)
 {
 	int ret;
-	u64 super_used;
-	u64 root_used;
-	u64 num_bytes = ins->offset;
-	u32 sizes[2];
-	struct btrfs_fs_info *info = root->fs_info;
-	struct btrfs_root *extent_root = info->extent_root;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_extent_item *extent_item;
-	struct btrfs_extent_ref *ref;
+	struct btrfs_extent_inline_ref *iref;
 	struct btrfs_path *path;
-	struct btrfs_key keys[2];
-
-	if (parent == 0)
-		parent = ins->objectid;
-
-	/* block accounting for super block */
-	spin_lock(&info->delalloc_lock);
-	super_used = btrfs_super_bytes_used(&info->super_copy);
-	btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
+	struct extent_buffer *leaf;
+	int type;
+	u32 size;
 
-	/* block accounting for root item */
-	root_used = btrfs_root_used(&root->root_item);
-	btrfs_set_root_used(&root->root_item, root_used + num_bytes);
-	spin_unlock(&info->delalloc_lock);
+	if (parent > 0)
+		type = BTRFS_SHARED_DATA_REF_KEY;
+	else
+		type = BTRFS_EXTENT_DATA_REF_KEY;
 
-	memcpy(&keys[0], ins, sizeof(*ins));
-	keys[1].objectid = ins->objectid;
-	keys[1].type = BTRFS_EXTENT_REF_KEY;
-	keys[1].offset = parent;
-	sizes[0] = sizeof(*extent_item);
-	sizes[1] = sizeof(*ref);
+	size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
 
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 
 	path->leave_spinning = 1;
-	ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
-				       sizes, 2);
+	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
+				      ins, size);
 	BUG_ON(ret);
 
-	extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
+	leaf = path->nodes[0];
+	extent_item = btrfs_item_ptr(leaf, path->slots[0],
 				     struct btrfs_extent_item);
-	btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
-	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
-			     struct btrfs_extent_ref);
-
-	btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
-	btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
-	btrfs_set_ref_objectid(path->nodes[0], ref, owner);
-	btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
+	btrfs_set_extent_refs(leaf, extent_item, ref_mod);
+	btrfs_set_extent_generation(leaf, extent_item, trans->transid);
+	btrfs_set_extent_flags(leaf, extent_item,
+			       flags | BTRFS_EXTENT_FLAG_DATA);
+
+	iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
+	btrfs_set_extent_inline_ref_type(leaf, iref, type);
+	if (parent > 0) {
+		struct btrfs_shared_data_ref *ref;
+		ref = (struct btrfs_shared_data_ref *)(iref + 1);
+		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
+		btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
+	} else {
+		struct btrfs_extent_data_ref *ref;
+		ref = (struct btrfs_extent_data_ref *)(&iref->offset);
+		btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
+		btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
+		btrfs_set_extent_data_ref_offset(leaf, ref, offset);
+		btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
+	}
 
 	btrfs_mark_buffer_dirty(path->nodes[0]);
-
-	trans->alloc_exclude_start = 0;
-	trans->alloc_exclude_nr = 0;
 	btrfs_free_path(path);
 
-	if (ret)
-		goto out;
-
-	ret = update_block_group(trans, root, ins->objectid,
-				 ins->offset, 1, 0);
+	ret = update_block_group(trans, root, ins->objectid, ins->offset,
+				 1, 0);
 	if (ret) {
 		printk(KERN_ERR "btrfs update block group failed for %llu "
 		       "%llu\n", (unsigned long long)ins->objectid,
 		       (unsigned long long)ins->offset);
 		BUG();
 	}
-out:
 	return ret;
 }
 
-int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root, u64 parent,
-				u64 root_objectid, u64 ref_generation,
-				u64 owner, struct btrfs_key *ins)
+static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root,
+				     u64 parent, u64 root_objectid,
+				     u64 flags, struct btrfs_disk_key *key,
+				     int level, struct btrfs_key *ins)
 {
 	int ret;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_extent_item *extent_item;
+	struct btrfs_tree_block_info *block_info;
+	struct btrfs_extent_inline_ref *iref;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
 
-	if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
-		return 0;
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
 
-	ret = btrfs_add_delayed_ref(trans, ins->objectid,
-				    ins->offset, parent, root_objectid,
-				    ref_generation, owner,
-				    BTRFS_ADD_DELAYED_EXTENT, 0);
+	path->leave_spinning = 1;
+	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
+				      ins, size);
 	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	extent_item = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_item);
+	btrfs_set_extent_refs(leaf, extent_item, 1);
+	btrfs_set_extent_generation(leaf, extent_item, trans->transid);
+	btrfs_set_extent_flags(leaf, extent_item,
+			       flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
+	block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
+
+	btrfs_set_tree_block_key(leaf, block_info, key);
+	btrfs_set_tree_block_level(leaf, block_info, level);
+
+	iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
+	if (parent > 0) {
+		BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
+		btrfs_set_extent_inline_ref_type(leaf, iref,
+						 BTRFS_SHARED_BLOCK_REF_KEY);
+		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
+	} else {
+		btrfs_set_extent_inline_ref_type(leaf, iref,
+						 BTRFS_TREE_BLOCK_REF_KEY);
+		btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
+	}
+
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_free_path(path);
+
+	ret = update_block_group(trans, root, ins->objectid, ins->offset,
+				 1, 0);
+	if (ret) {
+		printk(KERN_ERR "btrfs update block group failed for %llu "
+		       "%llu\n", (unsigned long long)ins->objectid,
+		       (unsigned long long)ins->offset);
+		BUG();
+	}
+	return ret;
+}
+
+int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root,
+				     u64 root_objectid, u64 owner,
+				     u64 offset, struct btrfs_key *ins)
+{
+	int ret;
+
+	BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
+
+	ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
+					 0, root_objectid, owner, offset,
+					 BTRFS_ADD_DELAYED_EXTENT, NULL);
 	return ret;
 }
 
@@ -3070,10 +3998,10 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
  * an extent has been allocated and makes sure to clear the free
  * space cache bits as well
  */
-int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root, u64 parent,
-				u64 root_objectid, u64 ref_generation,
-				u64 owner, struct btrfs_key *ins)
+int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   u64 root_objectid, u64 owner, u64 offset,
+				   struct btrfs_key *ins)
 {
 	int ret;
 	struct btrfs_block_group_cache *block_group;
@@ -3087,8 +4015,8 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
 				      ins->offset);
 	BUG_ON(ret);
 	btrfs_put_block_group(block_group);
-	ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
-					    ref_generation, owner, ins, 1);
+	ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
+					 0, owner, offset, ins, 1);
 	return ret;
 }
 
@@ -3099,26 +4027,48 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
  *
  * returns 0 if everything worked, non-zero otherwise.
  */
-int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
-		       struct btrfs_root *root,
-		       u64 num_bytes, u64 parent, u64 min_alloc_size,
-		       u64 root_objectid, u64 ref_generation,
-		       u64 owner_objectid, u64 empty_size, u64 hint_byte,
-		       u64 search_end, struct btrfs_key *ins, u64 data)
+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;
-	ret = __btrfs_reserve_extent(trans, root, num_bytes,
-				     min_alloc_size, empty_size, hint_byte,
-				     search_end, ins, data);
+	u64 flags = 0;
+
+	ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
+				     empty_size, hint_byte, search_end,
+				     ins, 0);
 	BUG_ON(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);
+
+	update_reserved_extents(root, ins->objectid, ins->offset, 1);
 	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
-		ret = btrfs_add_delayed_ref(trans, ins->objectid,
-					    ins->offset, parent, root_objectid,
-					    ref_generation, owner_objectid,
-					    BTRFS_ADD_DELAYED_EXTENT, 0);
+		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);
 	}
-	update_reserved_extents(root, ins->objectid, ins->offset, 1);
 	return ret;
 }
 
@@ -3157,21 +4107,17 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
  * returns the tree buffer or NULL.
  */
 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
-					     struct btrfs_root *root,
-					     u32 blocksize, u64 parent,
-					     u64 root_objectid,
-					     u64 ref_generation,
-					     int level,
-					     u64 hint,
-					     u64 empty_size)
+					struct btrfs_root *root, u32 blocksize,
+					u64 parent, u64 root_objectid,
+					struct btrfs_disk_key *key, int level,
+					u64 hint, u64 empty_size)
 {
 	struct btrfs_key ins;
 	int ret;
 	struct extent_buffer *buf;
 
-	ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
-				 root_objectid, ref_generation, level,
-				 empty_size, hint, (u64)-1, &ins, 0);
+	ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
+			       key, level, empty_size, hint, (u64)-1, &ins);
 	if (ret) {
 		BUG_ON(ret > 0);
 		return ERR_PTR(ret);
@@ -3185,32 +4131,19 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root, struct extent_buffer *leaf)
 {
-	u64 leaf_owner;
-	u64 leaf_generation;
-	struct refsort *sorted;
+	u64 disk_bytenr;
+	u64 num_bytes;
 	struct btrfs_key key;
 	struct btrfs_file_extent_item *fi;
+	u32 nritems;
 	int i;
-	int nritems;
 	int ret;
-	int refi = 0;
-	int slot;
 
 	BUG_ON(!btrfs_is_leaf(leaf));
 	nritems = btrfs_header_nritems(leaf);
-	leaf_owner = btrfs_header_owner(leaf);
-	leaf_generation = btrfs_header_generation(leaf);
 
-	sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
-	/* we do this loop twice.  The first time we build a list
-	 * of the extents we have a reference on, then we sort the list
-	 * by bytenr.  The second time around we actually do the
-	 * extent freeing.
-	 */
 	for (i = 0; i < nritems; i++) {
-		u64 disk_bytenr;
 		cond_resched();
-
 		btrfs_item_key_to_cpu(leaf, &key, i);
 
 		/* only extents have references, skip everything else */
@@ -3230,45 +4163,16 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
 		if (disk_bytenr == 0)
 			continue;
 
-		sorted[refi].bytenr = disk_bytenr;
-		sorted[refi].slot = i;
-		refi++;
-	}
-
-	if (refi == 0)
-		goto out;
-
-	sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
-
-	for (i = 0; i < refi; i++) {
-		u64 disk_bytenr;
-
-		disk_bytenr = sorted[i].bytenr;
-		slot = sorted[i].slot;
-
-		cond_resched();
-
-		btrfs_item_key_to_cpu(leaf, &key, slot);
-		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
-			continue;
-
-		fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
-
-		ret = btrfs_free_extent(trans, root, disk_bytenr,
-				btrfs_file_extent_disk_num_bytes(leaf, fi),
-				leaf->start, leaf_owner, leaf_generation,
-				key.objectid, 0);
+		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
+		ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
+					leaf->start, 0, key.objectid, 0);
 		BUG_ON(ret);
-
-		atomic_inc(&root->fs_info->throttle_gen);
-		wake_up(&root->fs_info->transaction_throttle);
-		cond_resched();
 	}
-out:
-	kfree(sorted);
 	return 0;
 }
 
+#if 0
+
 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
 					struct btrfs_root *root,
 					struct btrfs_leaf_ref *ref)
@@ -3311,13 +4215,14 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+
 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
 				     struct btrfs_root *root, u64 start,
 				     u64 len, u32 *refs)
 {
 	int ret;
 
-	ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
+	ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
 	BUG_ON(ret);
 
 #if 0 /* some debugging code in case we see problems here */
@@ -3352,6 +4257,7 @@ static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
+
 /*
  * this is used while deleting old snapshots, and it drops the refs
  * on a whole subtree starting from a level 1 node.
@@ -3645,32 +4551,36 @@ out:
 	cond_resched();
 	return 0;
 }
+#endif
 
 /*
  * helper function for drop_subtree, this function is similar to
  * walk_down_tree. The main difference is that it checks reference
  * counts while tree blocks are locked.
  */
-static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
-				      struct btrfs_root *root,
-				      struct btrfs_path *path, int *level)
+static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct btrfs_path *path, int *level)
 {
 	struct extent_buffer *next;
 	struct extent_buffer *cur;
 	struct extent_buffer *parent;
 	u64 bytenr;
 	u64 ptr_gen;
+	u64 refs;
+	u64 flags;
 	u32 blocksize;
-	u32 refs;
 	int ret;
 
 	cur = path->nodes[*level];
-	ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
-				      &refs);
+	ret = btrfs_lookup_extent_info(trans, root, cur->start, cur->len,
+				       &refs, &flags);
 	BUG_ON(ret);
 	if (refs > 1)
 		goto out;
 
+	BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
+
 	while (*level >= 0) {
 		cur = path->nodes[*level];
 		if (*level == 0) {
@@ -3692,16 +4602,15 @@ static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
 		btrfs_tree_lock(next);
 		btrfs_set_lock_blocking(next);
 
-		ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
-					      &refs);
+		ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
+					       &refs, &flags);
 		BUG_ON(ret);
 		if (refs > 1) {
 			parent = path->nodes[*level];
 			ret = btrfs_free_extent(trans, root, bytenr,
-					blocksize, parent->start,
-					btrfs_header_owner(parent),
-					btrfs_header_generation(parent),
-					*level - 1, 1);
+						blocksize, parent->start,
+						btrfs_header_owner(parent),
+						*level - 1, 0);
 			BUG_ON(ret);
 			path->slots[*level]++;
 			btrfs_tree_unlock(next);
@@ -3709,6 +4618,8 @@ static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
 			continue;
 		}
 
+		BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
+
 		*level = btrfs_header_level(next);
 		path->nodes[*level] = next;
 		path->slots[*level] = 0;
@@ -3716,13 +4627,15 @@ static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
 		cond_resched();
 	}
 out:
-	parent = path->nodes[*level + 1];
+	if (path->nodes[*level] == root->node)
+		parent = path->nodes[*level];
+	else
+		parent = path->nodes[*level + 1];
 	bytenr = path->nodes[*level]->start;
 	blocksize = path->nodes[*level]->len;
 
-	ret = btrfs_free_extent(trans, root, bytenr, blocksize,
-			parent->start, btrfs_header_owner(parent),
-			btrfs_header_generation(parent), *level, 1);
+	ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent->start,
+				btrfs_header_owner(parent), *level, 0);
 	BUG_ON(ret);
 
 	if (path->locks[*level]) {
@@ -3746,8 +4659,6 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
 				 struct btrfs_path *path,
 				 int *level, int max_level)
 {
-	u64 root_owner;
-	u64 root_gen;
 	struct btrfs_root_item *root_item = &root->root_item;
 	int i;
 	int slot;
@@ -3755,24 +4666,22 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
 
 	for (i = *level; i < max_level && path->nodes[i]; i++) {
 		slot = path->slots[i];
-		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
-			struct extent_buffer *node;
-			struct btrfs_disk_key disk_key;
-
+		if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
 			/*
 			 * there is more work to do in this level.
 			 * Update the drop_progress marker to reflect
 			 * the work we've done so far, and then bump
 			 * the slot number
 			 */
-			node = path->nodes[i];
 			path->slots[i]++;
-			*level = i;
 			WARN_ON(*level == 0);
-			btrfs_node_key(node, &disk_key, path->slots[i]);
-			memcpy(&root_item->drop_progress,
-			       &disk_key, sizeof(disk_key));
-			root_item->drop_level = i;
+			if (max_level == BTRFS_MAX_LEVEL) {
+				btrfs_node_key(path->nodes[i],
+					       &root_item->drop_progress,
+					       path->slots[i]);
+				root_item->drop_level = i;
+			}
+			*level = i;
 			return 0;
 		} else {
 			struct extent_buffer *parent;
@@ -3786,22 +4695,20 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
 			else
 				parent = path->nodes[*level + 1];
 
-			root_owner = btrfs_header_owner(parent);
-			root_gen = btrfs_header_generation(parent);
-
-			clean_tree_block(trans, root, path->nodes[*level]);
+			clean_tree_block(trans, root, path->nodes[i]);
 			ret = btrfs_free_extent(trans, root,
-						path->nodes[*level]->start,
-						path->nodes[*level]->len,
-						parent->start, root_owner,
-						root_gen, *level, 1);
+						path->nodes[i]->start,
+						path->nodes[i]->len,
+						parent->start,
+						btrfs_header_owner(parent),
+						*level, 0);
 			BUG_ON(ret);
 			if (path->locks[*level]) {
-				btrfs_tree_unlock(path->nodes[*level]);
-				path->locks[*level] = 0;
+				btrfs_tree_unlock(path->nodes[i]);
+				path->locks[i] = 0;
 			}
-			free_extent_buffer(path->nodes[*level]);
-			path->nodes[*level] = NULL;
+			free_extent_buffer(path->nodes[i]);
+			path->nodes[i] = NULL;
 			*level = i + 1;
 		}
 	}
@@ -3820,21 +4727,18 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 	int wret;
 	int level;
 	struct btrfs_path *path;
-	int i;
-	int orig_level;
 	int update_count;
 	struct btrfs_root_item *root_item = &root->root_item;
 
-	WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 
 	level = btrfs_header_level(root->node);
-	orig_level = level;
 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
-		path->nodes[level] = root->node;
-		extent_buffer_get(root->node);
+		path->nodes[level] = btrfs_lock_root_node(root);
+		btrfs_set_lock_blocking(path->nodes[level]);
 		path->slots[level] = 0;
+		path->locks[level] = 1;
 	} else {
 		struct btrfs_key key;
 		struct btrfs_disk_key found_key;
@@ -3856,12 +4760,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 		 * unlock our path, this is safe because only this
 		 * function is allowed to delete this snapshot
 		 */
-		for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
-			if (path->nodes[i] && path->locks[i]) {
-				path->locks[i] = 0;
-				btrfs_tree_unlock(path->nodes[i]);
-			}
-		}
+		btrfs_unlock_up_safe(path, 0);
 	}
 	while (1) {
 		unsigned long update;
@@ -3882,8 +4781,6 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 			ret = -EAGAIN;
 			break;
 		}
-		atomic_inc(&root->fs_info->throttle_gen);
-		wake_up(&root->fs_info->transaction_throttle);
 		for (update_count = 0; update_count < 16; update_count++) {
 			update = trans->delayed_ref_updates;
 			trans->delayed_ref_updates = 0;
@@ -3893,12 +4790,6 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 				break;
 		}
 	}
-	for (i = 0; i <= orig_level; i++) {
-		if (path->nodes[i]) {
-			free_extent_buffer(path->nodes[i]);
-			path->nodes[i] = NULL;
-		}
-	}
 out:
 	btrfs_free_path(path);
 	return ret;
@@ -3931,7 +4822,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 	path->slots[level] = 0;
 
 	while (1) {
-		wret = walk_down_subtree(trans, root, path, &level);
+		wret = walk_down_tree(trans, root, path, &level);
 		if (wret < 0)
 			ret = wret;
 		if (wret != 0)
@@ -3948,6 +4839,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
+#if 0
 static unsigned long calc_ra(unsigned long start, unsigned long last,
 			     unsigned long nr)
 {
@@ -5429,6 +6321,7 @@ out:
 	kfree(ref_path);
 	return ret;
 }
+#endif
 
 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
 {
@@ -5477,7 +6370,8 @@ static int __alloc_chunk_for_shrink(struct btrfs_root *root,
 	u64 calc;
 
 	spin_lock(&shrink_block_group->lock);
-	if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
+	if (btrfs_block_group_used(&shrink_block_group->item) +
+	    shrink_block_group->reserved > 0) {
 		spin_unlock(&shrink_block_group->lock);
 
 		trans = btrfs_start_transaction(root, 1);
@@ -5502,6 +6396,17 @@ static int __alloc_chunk_for_shrink(struct btrfs_root *root,
 	return 0;
 }
 
+
+int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
+					 struct btrfs_block_group_cache *group)
+
+{
+	__alloc_chunk_for_shrink(root, group, 1);
+	set_block_group_readonly(group);
+	return 0;
+}
+
+#if 0
 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
 				 struct btrfs_root *root,
 				 u64 objectid, u64 size)
@@ -5781,6 +6686,7 @@ out:
 	btrfs_free_path(path);
 	return ret;
 }
+#endif
 
 static int find_first_block_group(struct btrfs_root *root,
 		struct btrfs_path *path, struct btrfs_key *key)

fs/btrfs/extent_io.c

@@ -476,6 +476,7 @@ int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 	struct extent_state *state;
 	struct extent_state *prealloc = NULL;
 	struct rb_node *node;
+	u64 last_end;
 	int err;
 	int set = 0;
 
@@ -498,6 +499,7 @@ again:
 	if (state->start > end)
 		goto out;
 	WARN_ON(state->end < start);
+	last_end = state->end;
 
 	/*
 	 *     | ---- desired range ---- |
@@ -524,9 +526,11 @@ again:
 		if (err)
 			goto out;
 		if (state->end <= end) {
-			start = state->end + 1;
 			set |= clear_state_bit(tree, state, bits,
 					wake, delete);
+			if (last_end == (u64)-1)
+				goto out;
+			start = last_end + 1;
 		} else {
 			start = state->start;
 		}
@@ -552,8 +556,10 @@ again:
 		goto out;
 	}
 
-	start = state->end + 1;
 	set |= clear_state_bit(tree, state, bits, wake, delete);
+	if (last_end == (u64)-1)
+		goto out;
+	start = last_end + 1;
 	goto search_again;
 
 out:
@@ -707,8 +713,10 @@ again:
 			goto out;
 		}
 		set_state_bits(tree, state, bits);
-		start = state->end + 1;
 		merge_state(tree, state);
+		if (last_end == (u64)-1)
+			goto out;
+		start = last_end + 1;
 		goto search_again;
 	}
 
@@ -742,8 +750,10 @@ again:
 			goto out;
 		if (state->end <= end) {
 			set_state_bits(tree, state, bits);
-			start = state->end + 1;
 			merge_state(tree, state);
+			if (last_end == (u64)-1)
+				goto out;
+			start = last_end + 1;
 		} else {
 			start = state->start;
 		}

fs/btrfs/file.c

@@ -291,16 +291,12 @@ noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 {
 	u64 extent_end = 0;
 	u64 search_start = start;
-	u64 leaf_start;
 	u64 ram_bytes = 0;
-	u64 orig_parent = 0;
 	u64 disk_bytenr = 0;
 	u64 orig_locked_end = locked_end;
 	u8 compression;
 	u8 encryption;
 	u16 other_encoding = 0;
-	u64 root_gen;
-	u64 root_owner;
 	struct extent_buffer *leaf;
 	struct btrfs_file_extent_item *extent;
 	struct btrfs_path *path;
@@ -340,9 +336,6 @@ next_slot:
 		bookend = 0;
 		found_extent = 0;
 		found_inline = 0;
-		leaf_start = 0;
-		root_gen = 0;
-		root_owner = 0;
 		compression = 0;
 		encryption = 0;
 		extent = NULL;
@@ -417,9 +410,6 @@ next_slot:
 		if (found_extent) {
 			read_extent_buffer(leaf, &old, (unsigned long)extent,
 					   sizeof(old));
-			root_gen = btrfs_header_generation(leaf);
-			root_owner = btrfs_header_owner(leaf);
-			leaf_start = leaf->start;
 		}
 
 		if (end < extent_end && end >= key.offset) {
@@ -443,14 +433,14 @@ next_slot:
 				}
 				locked_end = extent_end;
 			}
-			orig_parent = path->nodes[0]->start;
 			disk_bytenr = le64_to_cpu(old.disk_bytenr);
 			if (disk_bytenr != 0) {
 				ret = btrfs_inc_extent_ref(trans, root,
 					   disk_bytenr,
-					   le64_to_cpu(old.disk_num_bytes),
-					   orig_parent, root->root_key.objectid,
-					   trans->transid, inode->i_ino);
+					   le64_to_cpu(old.disk_num_bytes), 0,
+					   root->root_key.objectid,
+					   key.objectid, key.offset -
+					   le64_to_cpu(old.offset));
 				BUG_ON(ret);
 			}
 		}
@@ -568,17 +558,6 @@ next_slot:
 			btrfs_mark_buffer_dirty(path->nodes[0]);
 			btrfs_set_lock_blocking(path->nodes[0]);
 
-			if (disk_bytenr != 0) {
-				ret = btrfs_update_extent_ref(trans, root,
-						disk_bytenr,
-						le64_to_cpu(old.disk_num_bytes),
-						orig_parent,
-						leaf->start,
-						root->root_key.objectid,
-						trans->transid, ins.objectid);
-
-				BUG_ON(ret);
-			}
 			path->leave_spinning = 0;
 			btrfs_release_path(root, path);
 			if (disk_bytenr != 0)
@@ -594,8 +573,9 @@ next_slot:
 				ret = btrfs_free_extent(trans, root,
 						old_disk_bytenr,
 						le64_to_cpu(old.disk_num_bytes),
-						leaf_start, root_owner,
-						root_gen, key.objectid, 0);
+						0, root->root_key.objectid,
+						key.objectid, key.offset -
+						le64_to_cpu(old.offset));
 				BUG_ON(ret);
 				*hint_byte = old_disk_bytenr;
 			}
@@ -664,12 +644,11 @@ int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 	u64 bytenr;
 	u64 num_bytes;
 	u64 extent_end;
-	u64 extent_offset;
+	u64 orig_offset;
 	u64 other_start;
 	u64 other_end;
 	u64 split = start;
 	u64 locked_end = end;
-	u64 orig_parent;
 	int extent_type;
 	int split_end = 1;
 	int ret;
@@ -703,7 +682,7 @@ again:
 
 	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
-	extent_offset = btrfs_file_extent_offset(leaf, fi);
+	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 
 	if (key.offset == start)
 		split = end;
@@ -711,8 +690,6 @@ again:
 	if (key.offset == start && extent_end == end) {
 		int del_nr = 0;
 		int del_slot = 0;
-		u64 leaf_owner = btrfs_header_owner(leaf);
-		u64 leaf_gen = btrfs_header_generation(leaf);
 		other_start = end;
 		other_end = 0;
 		if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
@@ -721,8 +698,8 @@ again:
 			del_slot = path->slots[0] + 1;
 			del_nr++;
 			ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
-						leaf->start, leaf_owner,
-						leaf_gen, inode->i_ino, 0);
+						0, root->root_key.objectid,
+						inode->i_ino, orig_offset);
 			BUG_ON(ret);
 		}
 		other_start = 0;
@@ -733,8 +710,8 @@ again:
 			del_slot = path->slots[0];
 			del_nr++;
 			ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
-						leaf->start, leaf_owner,
-						leaf_gen, inode->i_ino, 0);
+						0, root->root_key.objectid,
+						inode->i_ino, orig_offset);
 			BUG_ON(ret);
 		}
 		split_end = 0;
@@ -768,13 +745,12 @@ again:
 			locked_end = extent_end;
 		}
 		btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
-		extent_offset += split - key.offset;
 	} else  {
 		BUG_ON(key.offset != start);
-		btrfs_set_file_extent_offset(leaf, fi, extent_offset +
-					     split - key.offset);
-		btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
 		key.offset = split;
+		btrfs_set_file_extent_offset(leaf, fi, key.offset -
+					     orig_offset);
+		btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
 		btrfs_set_item_key_safe(trans, root, path, &key);
 		extent_end = split;
 	}
@@ -793,7 +769,8 @@ again:
 					    struct btrfs_file_extent_item);
 			key.offset = split;
 			btrfs_set_item_key_safe(trans, root, path, &key);
-			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
+			btrfs_set_file_extent_offset(leaf, fi, key.offset -
+						     orig_offset);
 			btrfs_set_file_extent_num_bytes(leaf, fi,
 							other_end - split);
 			goto done;
@@ -815,10 +792,9 @@ again:
 
 	btrfs_mark_buffer_dirty(leaf);
 
-	orig_parent = leaf->start;
-	ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
-				   orig_parent, root->root_key.objectid,
-				   trans->transid, inode->i_ino);
+	ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
+				   root->root_key.objectid,
+				   inode->i_ino, orig_offset);
 	BUG_ON(ret);
 	btrfs_release_path(root, path);
 
@@ -833,20 +809,12 @@ again:
 	btrfs_set_file_extent_type(leaf, fi, extent_type);
 	btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
-	btrfs_set_file_extent_offset(leaf, fi, extent_offset);
+	btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
 	btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
 	btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 	btrfs_set_file_extent_compression(leaf, fi, 0);
 	btrfs_set_file_extent_encryption(leaf, fi, 0);
 	btrfs_set_file_extent_other_encoding(leaf, fi, 0);
-
-	if (orig_parent != leaf->start) {
-		ret = btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
-					      orig_parent, leaf->start,
-					      root->root_key.objectid,
-					      trans->transid, inode->i_ino);
-		BUG_ON(ret);
-	}
 done:
 	btrfs_mark_buffer_dirty(leaf);
 
@@ -1189,6 +1157,8 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 	btrfs_wait_ordered_range(inode, 0, (u64)-1);
 	root->log_batch++;
 
+	if (datasync && !(inode->i_state & I_DIRTY_PAGES))
+		goto out;
 	/*
 	 * ok we haven't committed the transaction yet, lets do a commit
 	 */

fs/btrfs/free-space-cache.c

@@ -579,6 +579,7 @@ out:
  * it returns -enospc
  */
 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
 			     struct btrfs_block_group_cache *block_group,
 			     struct btrfs_free_cluster *cluster,
 			     u64 offset, u64 bytes, u64 empty_size)
@@ -595,7 +596,9 @@ int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
 	int ret;
 
 	/* for metadata, allow allocates with more holes */
-	if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
+	if (btrfs_test_opt(root, SSD_SPREAD)) {
+		min_bytes = bytes + empty_size;
+	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
 		/*
 		 * we want to do larger allocations when we are
 		 * flushing out the delayed refs, it helps prevent
@@ -645,14 +648,15 @@ again:
 		 * we haven't filled the empty size and the window is
 		 * very large.  reset and try again
 		 */
-		if (next->offset - window_start > (bytes + empty_size) * 2) {
+		if (next->offset - (last->offset + last->bytes) > 128 * 1024 ||
+		    next->offset - window_start > (bytes + empty_size) * 2) {
 			entry = next;
 			window_start = entry->offset;
 			window_free = entry->bytes;
 			last = entry;
 			max_extent = 0;
 			total_retries++;
-			if (total_retries % 256 == 0) {
+			if (total_retries % 64 == 0) {
 				if (min_bytes >= (bytes + empty_size)) {
 					ret = -ENOSPC;
 					goto out;

fs/btrfs/free-space-cache.h

@@ -31,6 +31,7 @@ void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
 			   u64 bytes);
 u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group);
 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root,
 			     struct btrfs_block_group_cache *block_group,
 			     struct btrfs_free_cluster *cluster,
 			     u64 offset, u64 bytes, u64 empty_size);

fs/btrfs/hash.h

@@ -19,9 +19,9 @@
 #ifndef __HASH__
 #define __HASH__
 
-#include "crc32c.h"
+#include <linux/crc32c.h>
 static inline u64 btrfs_name_hash(const char *name, int len)
 {
-	return btrfs_crc32c((u32)~1, name, len);
+	return crc32c((u32)~1, name, len);
 }
 #endif

fs/btrfs/inode.c

@@ -48,7 +48,6 @@
 #include "ordered-data.h"
 #include "xattr.h"
 #include "tree-log.h"
-#include "ref-cache.h"
 #include "compression.h"
 #include "locking.h"
 
@@ -369,7 +368,7 @@ again:
 	 * inode has not been flagged as nocompress.  This flag can
 	 * change at any time if we discover bad compression ratios.
 	 */
-	if (!btrfs_test_flag(inode, NOCOMPRESS) &&
+	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 	    btrfs_test_opt(root, COMPRESS)) {
 		WARN_ON(pages);
 		pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
@@ -470,7 +469,7 @@ again:
 		nr_pages_ret = 0;
 
 		/* flag the file so we don't compress in the future */
-		btrfs_set_flag(inode, NOCOMPRESS);
+		BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 	}
 	if (will_compress) {
 		*num_added += 1;
@@ -863,7 +862,7 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 		async_cow->locked_page = locked_page;
 		async_cow->start = start;
 
-		if (btrfs_test_flag(inode, NOCOMPRESS))
+		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 			cur_end = end;
 		else
 			cur_end = min(end, start + 512 * 1024 - 1);
@@ -944,6 +943,7 @@ static noinline int run_delalloc_nocow(struct inode *inode,
 	u64 cow_start;
 	u64 cur_offset;
 	u64 extent_end;
+	u64 extent_offset;
 	u64 disk_bytenr;
 	u64 num_bytes;
 	int extent_type;
@@ -1005,6 +1005,7 @@ next_slot:
 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+			extent_offset = btrfs_file_extent_offset(leaf, fi);
 			extent_end = found_key.offset +
 				btrfs_file_extent_num_bytes(leaf, fi);
 			if (extent_end <= start) {
@@ -1022,9 +1023,10 @@ next_slot:
 			if (btrfs_extent_readonly(root, disk_bytenr))
 				goto out_check;
 			if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
-						  disk_bytenr))
+						  found_key.offset -
+						  extent_offset, disk_bytenr))
 				goto out_check;
-			disk_bytenr += btrfs_file_extent_offset(leaf, fi);
+			disk_bytenr += extent_offset;
 			disk_bytenr += cur_offset - found_key.offset;
 			num_bytes = min(end + 1, extent_end) - cur_offset;
 			/*
@@ -1131,10 +1133,10 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
 	int ret;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 
-	if (btrfs_test_flag(inode, NODATACOW))
+	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
 		ret = run_delalloc_nocow(inode, locked_page, start, end,
 					 page_started, 1, nr_written);
-	else if (btrfs_test_flag(inode, PREALLOC))
+	else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
 		ret = run_delalloc_nocow(inode, locked_page, start, end,
 					 page_started, 0, nr_written);
 	else if (!btrfs_test_opt(root, COMPRESS))
@@ -1288,7 +1290,7 @@ static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 	int ret = 0;
 	int skip_sum;
 
-	skip_sum = btrfs_test_flag(inode, NODATASUM);
+	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
 
 	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 	BUG_ON(ret);
@@ -1489,9 +1491,9 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	ins.objectid = disk_bytenr;
 	ins.offset = disk_num_bytes;
 	ins.type = BTRFS_EXTENT_ITEM_KEY;
-	ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
-					  root->root_key.objectid,
-					  trans->transid, inode->i_ino, &ins);
+	ret = btrfs_alloc_reserved_file_extent(trans, root,
+					root->root_key.objectid,
+					inode->i_ino, file_pos, &ins);
 	BUG_ON(ret);
 	btrfs_free_path(path);
 
@@ -1788,7 +1790,8 @@ static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 		ClearPageChecked(page);
 		goto good;
 	}
-	if (btrfs_test_flag(inode, NODATASUM))
+
+	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
 		return 0;
 
 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
@@ -1956,23 +1959,13 @@ void btrfs_orphan_cleanup(struct btrfs_root *root)
 		 * crossing root thing.  we store the inode number in the
 		 * offset of the orphan item.
 		 */
-		inode = btrfs_iget_locked(root->fs_info->sb,
-					  found_key.offset, root);
-		if (!inode)
+		found_key.objectid = found_key.offset;
+		found_key.type = BTRFS_INODE_ITEM_KEY;
+		found_key.offset = 0;
+		inode = btrfs_iget(root->fs_info->sb, &found_key, root);
+		if (IS_ERR(inode))
 			break;
 
-		if (inode->i_state & I_NEW) {
-			BTRFS_I(inode)->root = root;
-
-			/* have to set the location manually */
-			BTRFS_I(inode)->location.objectid = inode->i_ino;
-			BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
-			BTRFS_I(inode)->location.offset = 0;
-
-			btrfs_read_locked_inode(inode);
-			unlock_new_inode(inode);
-		}
-
 		/*
 		 * add this inode to the orphan list so btrfs_orphan_del does
 		 * the proper thing when we hit it
@@ -2069,7 +2062,7 @@ static noinline int acls_after_inode_item(struct extent_buffer *leaf,
 /*
  * read an inode from the btree into the in-memory inode
  */
-void btrfs_read_locked_inode(struct inode *inode)
+static void btrfs_read_locked_inode(struct inode *inode)
 {
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
@@ -2164,6 +2157,8 @@ void btrfs_read_locked_inode(struct inode *inode)
 		init_special_inode(inode, inode->i_mode, rdev);
 		break;
 	}
+
+	btrfs_update_iflags(inode);
 	return;
 
 make_bad:
@@ -2599,9 +2594,8 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 	struct btrfs_file_extent_item *fi;
 	u64 extent_start = 0;
 	u64 extent_num_bytes = 0;
+	u64 extent_offset = 0;
 	u64 item_end = 0;
-	u64 root_gen = 0;
-	u64 root_owner = 0;
 	int found_extent;
 	int del_item;
 	int pending_del_nr = 0;
@@ -2716,6 +2710,9 @@ search_again:
 				extent_num_bytes =
 					btrfs_file_extent_disk_num_bytes(leaf,
 									 fi);
+				extent_offset = found_key.offset -
+					btrfs_file_extent_offset(leaf, fi);
+
 				/* FIXME blocksize != 4096 */
 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
 				if (extent_start != 0) {
@@ -2723,8 +2720,6 @@ search_again:
 					if (root->ref_cows)
 						inode_sub_bytes(inode, num_dec);
 				}
-				root_gen = btrfs_header_generation(leaf);
-				root_owner = btrfs_header_owner(leaf);
 			}
 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 			/*
@@ -2768,12 +2763,12 @@ delete:
 		} else {
 			break;
 		}
-		if (found_extent) {
+		if (found_extent && root->ref_cows) {
 			btrfs_set_path_blocking(path);
 			ret = btrfs_free_extent(trans, root, extent_start,
-						extent_num_bytes,
-						leaf->start, root_owner,
-						root_gen, inode->i_ino, 0);
+						extent_num_bytes, 0,
+						btrfs_header_owner(leaf),
+						inode->i_ino, extent_offset);
 			BUG_ON(ret);
 		}
 next:
@@ -3105,6 +3100,45 @@ static int fixup_tree_root_location(struct btrfs_root *root,
 	return 0;
 }
 
+static void inode_tree_add(struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_inode *entry;
+	struct rb_node **p = &root->inode_tree.rb_node;
+	struct rb_node *parent = NULL;
+
+	spin_lock(&root->inode_lock);
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct btrfs_inode, rb_node);
+
+		if (inode->i_ino < entry->vfs_inode.i_ino)
+			p = &(*p)->rb_left;
+		else if (inode->i_ino > entry->vfs_inode.i_ino)
+			p = &(*p)->rb_right;
+		else {
+			WARN_ON(!(entry->vfs_inode.i_state &
+				  (I_WILL_FREE | I_FREEING | I_CLEAR)));
+			break;
+		}
+	}
+	rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
+	rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
+	spin_unlock(&root->inode_lock);
+}
+
+static void inode_tree_del(struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+
+	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
+		spin_lock(&root->inode_lock);
+		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
+		spin_unlock(&root->inode_lock);
+		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
+	}
+}
+
 static noinline void init_btrfs_i(struct inode *inode)
 {
 	struct btrfs_inode *bi = BTRFS_I(inode);
@@ -3130,6 +3164,7 @@ static noinline void init_btrfs_i(struct inode *inode)
 			     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)->extent_mutex);
 	mutex_init(&BTRFS_I(inode)->log_mutex);
@@ -3152,26 +3187,9 @@ static int btrfs_find_actor(struct inode *inode, void *opaque)
 		args->root == BTRFS_I(inode)->root;
 }
 
-struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
-			    struct btrfs_root *root, int wait)
-{
-	struct inode *inode;
-	struct btrfs_iget_args args;
-	args.ino = objectid;
-	args.root = root;
-
-	if (wait) {
-		inode = ilookup5(s, objectid, btrfs_find_actor,
-				 (void *)&args);
-	} else {
-		inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
-					(void *)&args);
-	}
-	return inode;
-}
-
-struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
-				struct btrfs_root *root)
+static struct inode *btrfs_iget_locked(struct super_block *s,
+				       u64 objectid,
+				       struct btrfs_root *root)
 {
 	struct inode *inode;
 	struct btrfs_iget_args args;
@@ -3188,24 +3206,21 @@ struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
  * Returns in *is_new if the inode was read from disk
  */
 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
-			 struct btrfs_root *root, int *is_new)
+			 struct btrfs_root *root)
 {
 	struct inode *inode;
 
 	inode = btrfs_iget_locked(s, location->objectid, root);
 	if (!inode)
-		return ERR_PTR(-EACCES);
+		return ERR_PTR(-ENOMEM);
 
 	if (inode->i_state & I_NEW) {
 		BTRFS_I(inode)->root = root;
 		memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
 		btrfs_read_locked_inode(inode);
+
+		inode_tree_add(inode);
 		unlock_new_inode(inode);
-		if (is_new)
-			*is_new = 1;
-	} else {
-		if (is_new)
-			*is_new = 0;
 	}
 
 	return inode;
@@ -3218,7 +3233,7 @@ struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
 	struct btrfs_root *root = bi->root;
 	struct btrfs_root *sub_root = root;
 	struct btrfs_key location;
-	int ret, new;
+	int ret;
 
 	if (dentry->d_name.len > BTRFS_NAME_LEN)
 		return ERR_PTR(-ENAMETOOLONG);
@@ -3236,7 +3251,7 @@ struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
 			return ERR_PTR(ret);
 		if (ret > 0)
 			return ERR_PTR(-ENOENT);
-		inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
+		inode = btrfs_iget(dir->i_sb, &location, sub_root);
 		if (IS_ERR(inode))
 			return ERR_CAST(inode);
 	}
@@ -3574,9 +3589,9 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 			btrfs_find_block_group(root, 0, alloc_hint, owner);
 	if ((mode & S_IFREG)) {
 		if (btrfs_test_opt(root, NODATASUM))
-			btrfs_set_flag(inode, NODATASUM);
+			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 		if (btrfs_test_opt(root, NODATACOW))
-			btrfs_set_flag(inode, NODATACOW);
+			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 	}
 
 	key[0].objectid = objectid;
@@ -3630,7 +3645,10 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 	location->offset = 0;
 	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 
+	btrfs_inherit_iflags(inode, dir);
+
 	insert_inode_hash(inode);
+	inode_tree_add(inode);
 	return inode;
 fail:
 	if (dir)
@@ -4683,6 +4701,7 @@ void btrfs_destroy_inode(struct inode *inode)
 			btrfs_put_ordered_extent(ordered);
 		}
 	}
+	inode_tree_del(inode);
 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
 }
@@ -5061,7 +5080,7 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans,
 out:
 	if (cur_offset > start) {
 		inode->i_ctime = CURRENT_TIME;
-		btrfs_set_flag(inode, PREALLOC);
+		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
 		    cur_offset > i_size_read(inode))
 			btrfs_i_size_write(inode, cur_offset);
@@ -5182,7 +5201,7 @@ static int btrfs_set_page_dirty(struct page *page)
 
 static int btrfs_permission(struct inode *inode, int mask)
 {
-	if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
+	if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
 		return -EACCES;
 	return generic_permission(inode, mask, btrfs_check_acl);
 }

fs/btrfs/ioctl.c

@@ -50,7 +50,177 @@
 #include "volumes.h"
 #include "locking.h"
 
+/* Mask out flags that are inappropriate for the given type of inode. */
+static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
+{
+	if (S_ISDIR(mode))
+		return flags;
+	else if (S_ISREG(mode))
+		return flags & ~FS_DIRSYNC_FL;
+	else
+		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
+}
+
+/*
+ * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
+ */
+static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
+{
+	unsigned int iflags = 0;
+
+	if (flags & BTRFS_INODE_SYNC)
+		iflags |= FS_SYNC_FL;
+	if (flags & BTRFS_INODE_IMMUTABLE)
+		iflags |= FS_IMMUTABLE_FL;
+	if (flags & BTRFS_INODE_APPEND)
+		iflags |= FS_APPEND_FL;
+	if (flags & BTRFS_INODE_NODUMP)
+		iflags |= FS_NODUMP_FL;
+	if (flags & BTRFS_INODE_NOATIME)
+		iflags |= FS_NOATIME_FL;
+	if (flags & BTRFS_INODE_DIRSYNC)
+		iflags |= FS_DIRSYNC_FL;
+
+	return iflags;
+}
+
+/*
+ * Update inode->i_flags based on the btrfs internal flags.
+ */
+void btrfs_update_iflags(struct inode *inode)
+{
+	struct btrfs_inode *ip = BTRFS_I(inode);
+
+	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
+
+	if (ip->flags & BTRFS_INODE_SYNC)
+		inode->i_flags |= S_SYNC;
+	if (ip->flags & BTRFS_INODE_IMMUTABLE)
+		inode->i_flags |= S_IMMUTABLE;
+	if (ip->flags & BTRFS_INODE_APPEND)
+		inode->i_flags |= S_APPEND;
+	if (ip->flags & BTRFS_INODE_NOATIME)
+		inode->i_flags |= S_NOATIME;
+	if (ip->flags & BTRFS_INODE_DIRSYNC)
+		inode->i_flags |= S_DIRSYNC;
+}
+
+/*
+ * Inherit flags from the parent inode.
+ *
+ * Unlike extN we don't have any flags we don't want to inherit currently.
+ */
+void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
+{
+	unsigned int flags;
+
+	if (!dir)
+		return;
+
+	flags = BTRFS_I(dir)->flags;
+
+	if (S_ISREG(inode->i_mode))
+		flags &= ~BTRFS_INODE_DIRSYNC;
+	else if (!S_ISDIR(inode->i_mode))
+		flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
+
+	BTRFS_I(inode)->flags = flags;
+	btrfs_update_iflags(inode);
+}
+
+static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
+{
+	struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
+	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
+
+	if (copy_to_user(arg, &flags, sizeof(flags)))
+		return -EFAULT;
+	return 0;
+}
+
+static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
+{
+	struct inode *inode = file->f_path.dentry->d_inode;
+	struct btrfs_inode *ip = BTRFS_I(inode);
+	struct btrfs_root *root = ip->root;
+	struct btrfs_trans_handle *trans;
+	unsigned int flags, oldflags;
+	int ret;
+
+	if (copy_from_user(&flags, arg, sizeof(flags)))
+		return -EFAULT;
+
+	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
+		      FS_NOATIME_FL | FS_NODUMP_FL | \
+		      FS_SYNC_FL | FS_DIRSYNC_FL))
+		return -EOPNOTSUPP;
 
+	if (!is_owner_or_cap(inode))
+		return -EACCES;
+
+	mutex_lock(&inode->i_mutex);
+
+	flags = btrfs_mask_flags(inode->i_mode, flags);
+	oldflags = btrfs_flags_to_ioctl(ip->flags);
+	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
+		if (!capable(CAP_LINUX_IMMUTABLE)) {
+			ret = -EPERM;
+			goto out_unlock;
+		}
+	}
+
+	ret = mnt_want_write(file->f_path.mnt);
+	if (ret)
+		goto out_unlock;
+
+	if (flags & FS_SYNC_FL)
+		ip->flags |= BTRFS_INODE_SYNC;
+	else
+		ip->flags &= ~BTRFS_INODE_SYNC;
+	if (flags & FS_IMMUTABLE_FL)
+		ip->flags |= BTRFS_INODE_IMMUTABLE;
+	else
+		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
+	if (flags & FS_APPEND_FL)
+		ip->flags |= BTRFS_INODE_APPEND;
+	else
+		ip->flags &= ~BTRFS_INODE_APPEND;
+	if (flags & FS_NODUMP_FL)
+		ip->flags |= BTRFS_INODE_NODUMP;
+	else
+		ip->flags &= ~BTRFS_INODE_NODUMP;
+	if (flags & FS_NOATIME_FL)
+		ip->flags |= BTRFS_INODE_NOATIME;
+	else
+		ip->flags &= ~BTRFS_INODE_NOATIME;
+	if (flags & FS_DIRSYNC_FL)
+		ip->flags |= BTRFS_INODE_DIRSYNC;
+	else
+		ip->flags &= ~BTRFS_INODE_DIRSYNC;
+
+
+	trans = btrfs_join_transaction(root, 1);
+	BUG_ON(!trans);
+
+	ret = btrfs_update_inode(trans, root, inode);
+	BUG_ON(ret);
+
+	btrfs_update_iflags(inode);
+	inode->i_ctime = CURRENT_TIME;
+	btrfs_end_transaction(trans, root);
+
+	mnt_drop_write(file->f_path.mnt);
+ out_unlock:
+	mutex_unlock(&inode->i_mutex);
+	return 0;
+}
+
+static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
+{
+	struct inode *inode = file->f_path.dentry->d_inode;
+
+	return put_user(inode->i_generation, arg);
+}
 
 static noinline int create_subvol(struct btrfs_root *root,
 				  struct dentry *dentry,
@@ -82,22 +252,25 @@ static noinline int create_subvol(struct btrfs_root *root,
 	if (ret)
 		goto fail;
 
-	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
-				      objectid, trans->transid, 0, 0, 0);
+	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
+				      0, objectid, NULL, 0, 0, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		goto fail;
 	}
 
-	btrfs_set_header_nritems(leaf, 0);
-	btrfs_set_header_level(leaf, 0);
+	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 	btrfs_set_header_bytenr(leaf, leaf->start);
 	btrfs_set_header_generation(leaf, trans->transid);
+	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 	btrfs_set_header_owner(leaf, objectid);
 
 	write_extent_buffer(leaf, root->fs_info->fsid,
 			    (unsigned long)btrfs_header_fsid(leaf),
 			    BTRFS_FSID_SIZE);
+	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
+			    BTRFS_UUID_SIZE);
 	btrfs_mark_buffer_dirty(leaf);
 
 	inode_item = &root_item.inode;
@@ -125,7 +298,7 @@ static noinline int create_subvol(struct btrfs_root *root,
 	btrfs_set_root_dirid(&root_item, new_dirid);
 
 	key.objectid = objectid;
-	key.offset = 1;
+	key.offset = 0;
 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 				&root_item);
@@ -911,10 +1084,10 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 				if (disko) {
 					inode_add_bytes(inode, datal);
 					ret = btrfs_inc_extent_ref(trans, root,
-						   disko, diskl, leaf->start,
-						   root->root_key.objectid,
-						   trans->transid,
-						   inode->i_ino);
+							disko, diskl, 0,
+							root->root_key.objectid,
+							inode->i_ino,
+							new_key.offset - datao);
 					BUG_ON(ret);
 				}
 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
@@ -1074,6 +1247,12 @@ long btrfs_ioctl(struct file *file, unsigned int
 	void __user *argp = (void __user *)arg;
 
 	switch (cmd) {
+	case FS_IOC_GETFLAGS:
+		return btrfs_ioctl_getflags(file, argp);
+	case FS_IOC_SETFLAGS:
+		return btrfs_ioctl_setflags(file, argp);
+	case FS_IOC_GETVERSION:
+		return btrfs_ioctl_getversion(file, argp);
 	case BTRFS_IOC_SNAP_CREATE:
 		return btrfs_ioctl_snap_create(file, argp, 0);
 	case BTRFS_IOC_SUBVOL_CREATE:

fs/btrfs/print-tree.c

@@ -45,22 +45,132 @@ static void print_dev_item(struct extent_buffer *eb,
 	       (unsigned long long)btrfs_device_total_bytes(eb, dev_item),
 	       (unsigned long long)btrfs_device_bytes_used(eb, dev_item));
 }
+static void print_extent_data_ref(struct extent_buffer *eb,
+				  struct btrfs_extent_data_ref *ref)
+{
+	printk(KERN_INFO "\t\textent data backref root %llu "
+	       "objectid %llu offset %llu count %u\n",
+	       (unsigned long long)btrfs_extent_data_ref_root(eb, ref),
+	       (unsigned long long)btrfs_extent_data_ref_objectid(eb, ref),
+	       (unsigned long long)btrfs_extent_data_ref_offset(eb, ref),
+	       btrfs_extent_data_ref_count(eb, ref));
+}
+
+static void print_extent_item(struct extent_buffer *eb, int slot)
+{
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_inline_ref *iref;
+	struct btrfs_extent_data_ref *dref;
+	struct btrfs_shared_data_ref *sref;
+	struct btrfs_disk_key key;
+	unsigned long end;
+	unsigned long ptr;
+	int type;
+	u32 item_size = btrfs_item_size_nr(eb, slot);
+	u64 flags;
+	u64 offset;
+
+	if (item_size < sizeof(*ei)) {
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+		struct btrfs_extent_item_v0 *ei0;
+		BUG_ON(item_size != sizeof(*ei0));
+		ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
+		printk(KERN_INFO "\t\textent refs %u\n",
+		       btrfs_extent_refs_v0(eb, ei0));
+		return;
+#else
+		BUG();
+#endif
+	}
+
+	ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
+	flags = btrfs_extent_flags(eb, ei);
+
+	printk(KERN_INFO "\t\textent refs %llu gen %llu flags %llu\n",
+	       (unsigned long long)btrfs_extent_refs(eb, ei),
+	       (unsigned long long)btrfs_extent_generation(eb, ei),
+	       (unsigned long long)flags);
+
+	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
+		struct btrfs_tree_block_info *info;
+		info = (struct btrfs_tree_block_info *)(ei + 1);
+		btrfs_tree_block_key(eb, info, &key);
+		printk(KERN_INFO "\t\ttree block key (%llu %x %llu) "
+		       "level %d\n",
+		       (unsigned long long)btrfs_disk_key_objectid(&key),
+		       key.type,
+		       (unsigned long long)btrfs_disk_key_offset(&key),
+		       btrfs_tree_block_level(eb, info));
+		iref = (struct btrfs_extent_inline_ref *)(info + 1);
+	} else {
+		iref = (struct btrfs_extent_inline_ref *)(ei + 1);
+	}
+
+	ptr = (unsigned long)iref;
+	end = (unsigned long)ei + item_size;
+	while (ptr < end) {
+		iref = (struct btrfs_extent_inline_ref *)ptr;
+		type = btrfs_extent_inline_ref_type(eb, iref);
+		offset = btrfs_extent_inline_ref_offset(eb, iref);
+		switch (type) {
+		case BTRFS_TREE_BLOCK_REF_KEY:
+			printk(KERN_INFO "\t\ttree block backref "
+				"root %llu\n", (unsigned long long)offset);
+			break;
+		case BTRFS_SHARED_BLOCK_REF_KEY:
+			printk(KERN_INFO "\t\tshared block backref "
+				"parent %llu\n", (unsigned long long)offset);
+			break;
+		case BTRFS_EXTENT_DATA_REF_KEY:
+			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
+			print_extent_data_ref(eb, dref);
+			break;
+		case BTRFS_SHARED_DATA_REF_KEY:
+			sref = (struct btrfs_shared_data_ref *)(iref + 1);
+			printk(KERN_INFO "\t\tshared data backref "
+			       "parent %llu count %u\n",
+			       (unsigned long long)offset,
+			       btrfs_shared_data_ref_count(eb, sref));
+			break;
+		default:
+			BUG();
+		}
+		ptr += btrfs_extent_inline_ref_size(type);
+	}
+	WARN_ON(ptr > end);
+}
+
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+static void print_extent_ref_v0(struct extent_buffer *eb, int slot)
+{
+	struct btrfs_extent_ref_v0 *ref0;
+
+	ref0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_ref_v0);
+	printk("\t\textent back ref root %llu gen %llu "
+		"owner %llu num_refs %lu\n",
+		(unsigned long long)btrfs_ref_root_v0(eb, ref0),
+		(unsigned long long)btrfs_ref_generation_v0(eb, ref0),
+		(unsigned long long)btrfs_ref_objectid_v0(eb, ref0),
+		(unsigned long)btrfs_ref_count_v0(eb, ref0));
+}
+#endif
+
 void btrfs_print_leaf(struct btrfs_root *root, struct extent_buffer *l)
 {
 	int i;
+	u32 type;
 	u32 nr = btrfs_header_nritems(l);
 	struct btrfs_item *item;
-	struct btrfs_extent_item *ei;
 	struct btrfs_root_item *ri;
 	struct btrfs_dir_item *di;
 	struct btrfs_inode_item *ii;
 	struct btrfs_block_group_item *bi;
 	struct btrfs_file_extent_item *fi;
+	struct btrfs_extent_data_ref *dref;
+	struct btrfs_shared_data_ref *sref;
+	struct btrfs_dev_extent *dev_extent;
 	struct btrfs_key key;
 	struct btrfs_key found_key;
-	struct btrfs_extent_ref *ref;
-	struct btrfs_dev_extent *dev_extent;
-	u32 type;
 
 	printk(KERN_INFO "leaf %llu total ptrs %d free space %d\n",
 		(unsigned long long)btrfs_header_bytenr(l), nr,
@@ -100,20 +210,25 @@ void btrfs_print_leaf(struct btrfs_root *root, struct extent_buffer *l)
 				btrfs_disk_root_refs(l, ri));
 			break;
 		case BTRFS_EXTENT_ITEM_KEY:
-			ei = btrfs_item_ptr(l, i, struct btrfs_extent_item);
-			printk(KERN_INFO "\t\textent data refs %u\n",
-				btrfs_extent_refs(l, ei));
-			break;
-		case BTRFS_EXTENT_REF_KEY:
-			ref = btrfs_item_ptr(l, i, struct btrfs_extent_ref);
-			printk(KERN_INFO "\t\textent back ref root %llu "
-			       "gen %llu owner %llu num_refs %lu\n",
-			       (unsigned long long)btrfs_ref_root(l, ref),
-			       (unsigned long long)btrfs_ref_generation(l, ref),
-			       (unsigned long long)btrfs_ref_objectid(l, ref),
-			       (unsigned long)btrfs_ref_num_refs(l, ref));
+			print_extent_item(l, i);
+			break;
+		case BTRFS_TREE_BLOCK_REF_KEY:
+			printk(KERN_INFO "\t\ttree block backref\n");
+			break;
+		case BTRFS_SHARED_BLOCK_REF_KEY:
+			printk(KERN_INFO "\t\tshared block backref\n");
+			break;
+		case BTRFS_EXTENT_DATA_REF_KEY:
+			dref = btrfs_item_ptr(l, i,
+					      struct btrfs_extent_data_ref);
+			print_extent_data_ref(l, dref);
+			break;
+		case BTRFS_SHARED_DATA_REF_KEY:
+			sref = btrfs_item_ptr(l, i,
+					      struct btrfs_shared_data_ref);
+			printk(KERN_INFO "\t\tshared data backref count %u\n",
+			       btrfs_shared_data_ref_count(l, sref));
 			break;
-
 		case BTRFS_EXTENT_DATA_KEY:
 			fi = btrfs_item_ptr(l, i,
 					    struct btrfs_file_extent_item);
@@ -139,6 +254,12 @@ void btrfs_print_leaf(struct btrfs_root *root, struct extent_buffer *l)
 			       (unsigned long long)
 			       btrfs_file_extent_ram_bytes(l, fi));
 			break;
+		case BTRFS_EXTENT_REF_V0_KEY:
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+			print_extent_ref_v0(l, i);
+#else
+			BUG();
+#endif
 		case BTRFS_BLOCK_GROUP_ITEM_KEY:
 			bi = btrfs_item_ptr(l, i,
 					    struct btrfs_block_group_item);

fs/btrfs/relocation.c

@@ -0,0 +1,3711 @@
+/*
+ * Copyright (C) 2009 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/writeback.h>
+#include <linux/blkdev.h>
+#include <linux/rbtree.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "volumes.h"
+#include "locking.h"
+#include "btrfs_inode.h"
+#include "async-thread.h"
+
+/*
+ * backref_node, mapping_node and tree_block start with this
+ */
+struct tree_entry {
+	struct rb_node rb_node;
+	u64 bytenr;
+};
+
+/*
+ * present a tree block in the backref cache
+ */
+struct backref_node {
+	struct rb_node rb_node;
+	u64 bytenr;
+	/* objectid tree block owner */
+	u64 owner;
+	/* list of upper level blocks reference this block */
+	struct list_head upper;
+	/* list of child blocks in the cache */
+	struct list_head lower;
+	/* NULL if this node is not tree root */
+	struct btrfs_root *root;
+	/* extent buffer got by COW the block */
+	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 */
+	unsigned int lowest:1;
+	/* is the extent buffer locked */
+	unsigned int locked:1;
+	/* has the block been processed */
+	unsigned int processed:1;
+	/* have backrefs of this block been checked */
+	unsigned int checked:1;
+};
+
+/*
+ * present a block pointer in the backref cache
+ */
+struct backref_edge {
+	struct list_head list[2];
+	struct backref_node *node[2];
+	u64 blockptr;
+};
+
+#define LOWER	0
+#define UPPER	1
+
+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 */
+	struct list_head pending[BTRFS_MAX_LEVEL];
+	spinlock_t lock;
+};
+
+/*
+ * map address of tree root to tree
+ */
+struct mapping_node {
+	struct rb_node rb_node;
+	u64 bytenr;
+	void *data;
+};
+
+struct mapping_tree {
+	struct rb_root rb_root;
+	spinlock_t lock;
+};
+
+/*
+ * present a tree block to process
+ */
+struct tree_block {
+	struct rb_node rb_node;
+	u64 bytenr;
+	struct btrfs_key key;
+	unsigned int level:8;
+	unsigned int key_ready:1;
+};
+
+/* inode vector */
+#define INODEVEC_SIZE 16
+
+struct inodevec {
+	struct list_head list;
+	struct inode *inode[INODEVEC_SIZE];
+	int nr;
+};
+
+struct reloc_control {
+	/* block group to relocate */
+	struct btrfs_block_group_cache *block_group;
+	/* extent tree */
+	struct btrfs_root *extent_root;
+	/* inode for moving data */
+	struct inode *data_inode;
+	struct btrfs_workers workers;
+	/* 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;
+	u64 search_start;
+	u64 extents_found;
+	u64 extents_skipped;
+	int stage;
+	int create_reloc_root;
+	unsigned int found_file_extent:1;
+	unsigned int found_old_snapshot: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 mapping_tree_init(struct mapping_tree *tree)
+{
+	tree->rb_root.rb_node = NULL;
+	spin_lock_init(&tree->lock);
+}
+
+static void backref_cache_init(struct backref_cache *cache)
+{
+	int i;
+	cache->rb_root.rb_node = NULL;
+	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
+		INIT_LIST_HEAD(&cache->pending[i]);
+	spin_lock_init(&cache->lock);
+}
+
+static void backref_node_init(struct backref_node *node)
+{
+	memset(node, 0, sizeof(*node));
+	INIT_LIST_HEAD(&node->upper);
+	INIT_LIST_HEAD(&node->lower);
+	RB_CLEAR_NODE(&node->rb_node);
+}
+
+static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
+				   struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct tree_entry *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct tree_entry, rb_node);
+
+		if (bytenr < entry->bytenr)
+			p = &(*p)->rb_left;
+		else if (bytenr > entry->bytenr)
+			p = &(*p)->rb_right;
+		else
+			return parent;
+	}
+
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
+{
+	struct rb_node *n = root->rb_node;
+	struct tree_entry *entry;
+
+	while (n) {
+		entry = rb_entry(n, struct tree_entry, rb_node);
+
+		if (bytenr < entry->bytenr)
+			n = n->rb_left;
+		else if (bytenr > entry->bytenr)
+			n = n->rb_right;
+		else
+			return n;
+	}
+	return NULL;
+}
+
+/*
+ * walk up backref nodes until reach node presents tree root
+ */
+static struct backref_node *walk_up_backref(struct backref_node *node,
+					    struct backref_edge *edges[],
+					    int *index)
+{
+	struct backref_edge *edge;
+	int idx = *index;
+
+	while (!list_empty(&node->upper)) {
+		edge = list_entry(node->upper.next,
+				  struct backref_edge, list[LOWER]);
+		edges[idx++] = edge;
+		node = edge->node[UPPER];
+	}
+	*index = idx;
+	return node;
+}
+
+/*
+ * walk down backref nodes to find start of next reference path
+ */
+static struct backref_node *walk_down_backref(struct backref_edge *edges[],
+					      int *index)
+{
+	struct backref_edge *edge;
+	struct backref_node *lower;
+	int idx = *index;
+
+	while (idx > 0) {
+		edge = edges[idx - 1];
+		lower = edge->node[LOWER];
+		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
+			idx--;
+			continue;
+		}
+		edge = list_entry(edge->list[LOWER].next,
+				  struct backref_edge, list[LOWER]);
+		edges[idx - 1] = edge;
+		*index = idx;
+		return edge->node[UPPER];
+	}
+	*index = 0;
+	return NULL;
+}
+
+static void drop_node_buffer(struct backref_node *node)
+{
+	if (node->eb) {
+		if (node->locked) {
+			btrfs_tree_unlock(node->eb);
+			node->locked = 0;
+		}
+		free_extent_buffer(node->eb);
+		node->eb = NULL;
+	}
+}
+
+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->lower);
+
+	rb_erase(&node->rb_node, &tree->rb_root);
+	kfree(node);
+}
+
+/*
+ * remove a backref node from the backref cache
+ */
+static void remove_backref_node(struct backref_cache *cache,
+				struct backref_node *node)
+{
+	struct backref_node *upper;
+	struct backref_edge *edge;
+
+	if (!node)
+		return;
+
+	BUG_ON(!node->lowest);
+	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);
+		/*
+		 * add the node to pending list if no other
+		 * child block cached.
+		 */
+		if (list_empty(&upper->lower)) {
+			list_add_tail(&upper->lower,
+				      &cache->pending[upper->level]);
+			upper->lowest = 1;
+		}
+	}
+	drop_backref_node(cache, node);
+}
+
+/*
+ * find reloc tree by address of tree root
+ */
+static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
+					  u64 bytenr)
+{
+	struct rb_node *rb_node;
+	struct mapping_node *node;
+	struct btrfs_root *root = NULL;
+
+	spin_lock(&rc->reloc_root_tree.lock);
+	rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
+	if (rb_node) {
+		node = rb_entry(rb_node, struct mapping_node, rb_node);
+		root = (struct btrfs_root *)node->data;
+	}
+	spin_unlock(&rc->reloc_root_tree.lock);
+	return root;
+}
+
+static int is_cowonly_root(u64 root_objectid)
+{
+	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
+	    root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
+	    root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
+	    root_objectid == BTRFS_DEV_TREE_OBJECTID ||
+	    root_objectid == BTRFS_TREE_LOG_OBJECTID ||
+	    root_objectid == BTRFS_CSUM_TREE_OBJECTID)
+		return 1;
+	return 0;
+}
+
+static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
+					u64 root_objectid)
+{
+	struct btrfs_key key;
+
+	key.objectid = root_objectid;
+	key.type = BTRFS_ROOT_ITEM_KEY;
+	if (is_cowonly_root(root_objectid))
+		key.offset = 0;
+	else
+		key.offset = (u64)-1;
+
+	return btrfs_read_fs_root_no_name(fs_info, &key);
+}
+
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+static noinline_for_stack
+struct btrfs_root *find_tree_root(struct reloc_control *rc,
+				  struct extent_buffer *leaf,
+				  struct btrfs_extent_ref_v0 *ref0)
+{
+	struct btrfs_root *root;
+	u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
+	u64 generation = btrfs_ref_generation_v0(leaf, ref0);
+
+	BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
+
+	root = read_fs_root(rc->extent_root->fs_info, root_objectid);
+	BUG_ON(IS_ERR(root));
+
+	if (root->ref_cows &&
+	    generation != btrfs_root_generation(&root->root_item))
+		return NULL;
+
+	return root;
+}
+#endif
+
+static noinline_for_stack
+int find_inline_backref(struct extent_buffer *leaf, int slot,
+			unsigned long *ptr, unsigned long *end)
+{
+	struct btrfs_extent_item *ei;
+	struct btrfs_tree_block_info *bi;
+	u32 item_size;
+
+	item_size = btrfs_item_size_nr(leaf, slot);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (item_size < sizeof(*ei)) {
+		WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
+		return 1;
+	}
+#endif
+	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
+	WARN_ON(!(btrfs_extent_flags(leaf, ei) &
+		  BTRFS_EXTENT_FLAG_TREE_BLOCK));
+
+	if (item_size <= sizeof(*ei) + sizeof(*bi)) {
+		WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
+		return 1;
+	}
+
+	bi = (struct btrfs_tree_block_info *)(ei + 1);
+	*ptr = (unsigned long)(bi + 1);
+	*end = (unsigned long)ei + item_size;
+	return 0;
+}
+
+/*
+ * build backref tree for a given tree block. root of the backref tree
+ * corresponds the tree block, leaves of the backref tree correspond
+ * roots of b-trees that reference the tree block.
+ *
+ * the basic idea of this function is check backrefs of a given block
+ * to find upper level blocks that refernece the block, and then check
+ * bakcrefs of these upper level blocks recursively. the recursion stop
+ * when tree root is reached or backrefs for the block is cached.
+ *
+ * NOTE: if we find backrefs for a block are cached, we know backrefs
+ * 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)
+{
+	struct btrfs_path *path1;
+	struct btrfs_path *path2;
+	struct extent_buffer *eb;
+	struct btrfs_root *root;
+	struct backref_node *cur;
+	struct backref_node *upper;
+	struct backref_node *lower;
+	struct backref_node *node = NULL;
+	struct backref_node *exist = NULL;
+	struct backref_edge *edge;
+	struct rb_node *rb_node;
+	struct btrfs_key key;
+	unsigned long end;
+	unsigned long ptr;
+	LIST_HEAD(list);
+	int ret;
+	int err = 0;
+
+	path1 = btrfs_alloc_path();
+	path2 = btrfs_alloc_path();
+	if (!path1 || !path2) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	node = kmalloc(sizeof(*node), GFP_NOFS);
+	if (!node) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	backref_node_init(node);
+	node->bytenr = bytenr;
+	node->owner = 0;
+	node->level = level;
+	node->lowest = 1;
+	cur = node;
+again:
+	end = 0;
+	ptr = 0;
+	key.objectid = cur->bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	path1->search_commit_root = 1;
+	path1->skip_locking = 1;
+	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
+				0, 0);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
+	BUG_ON(!ret || !path1->slots[0]);
+
+	path1->slots[0]--;
+
+	WARN_ON(cur->checked);
+	if (!list_empty(&cur->upper)) {
+		/*
+		 * the backref was added previously when processsing
+		 * backref of type BTRFS_TREE_BLOCK_REF_KEY
+		 */
+		BUG_ON(!list_is_singular(&cur->upper));
+		edge = list_entry(cur->upper.next, struct backref_edge,
+				  list[LOWER]);
+		BUG_ON(!list_empty(&edge->list[UPPER]));
+		exist = edge->node[UPPER];
+		/*
+		 * add the upper level block to pending list if we need
+		 * check its backrefs
+		 */
+		if (!exist->checked)
+			list_add_tail(&edge->list[UPPER], &list);
+	} else {
+		exist = NULL;
+	}
+
+	while (1) {
+		cond_resched();
+		eb = path1->nodes[0];
+
+		if (ptr >= end) {
+			if (path1->slots[0] >= btrfs_header_nritems(eb)) {
+				ret = btrfs_next_leaf(rc->extent_root, path1);
+				if (ret < 0) {
+					err = ret;
+					goto out;
+				}
+				if (ret > 0)
+					break;
+				eb = path1->nodes[0];
+			}
+
+			btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
+			if (key.objectid != cur->bytenr) {
+				WARN_ON(exist);
+				break;
+			}
+
+			if (key.type == BTRFS_EXTENT_ITEM_KEY) {
+				ret = find_inline_backref(eb, path1->slots[0],
+							  &ptr, &end);
+				if (ret)
+					goto next;
+			}
+		}
+
+		if (ptr < end) {
+			/* update key for inline back ref */
+			struct btrfs_extent_inline_ref *iref;
+			iref = (struct btrfs_extent_inline_ref *)ptr;
+			key.type = btrfs_extent_inline_ref_type(eb, iref);
+			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
+			WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
+				key.type != BTRFS_SHARED_BLOCK_REF_KEY);
+		}
+
+		if (exist &&
+		    ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
+		      exist->owner == key.offset) ||
+		     (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
+		      exist->bytenr == key.offset))) {
+			exist = NULL;
+			goto next;
+		}
+
+#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) {
+				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;
+			}
+#else
+		BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
+		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
+#endif
+			if (key.objectid == key.offset) {
+				/*
+				 * only root blocks of reloc trees use
+				 * backref of this type.
+				 */
+				root = find_reloc_root(rc, cur->bytenr);
+				BUG_ON(!root);
+				cur->root = root;
+				break;
+			}
+
+			edge = kzalloc(sizeof(*edge), GFP_NOFS);
+			if (!edge) {
+				err = -ENOMEM;
+				goto out;
+			}
+			rb_node = tree_search(&cache->rb_root, key.offset);
+			if (!rb_node) {
+				upper = kmalloc(sizeof(*upper), GFP_NOFS);
+				if (!upper) {
+					kfree(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
+				 *  cached, add the block to pending list
+				 */
+				list_add_tail(&edge->list[UPPER], &list);
+			} else {
+				upper = rb_entry(rb_node, struct backref_node,
+						 rb_node);
+				INIT_LIST_HEAD(&edge->list[UPPER]);
+			}
+			list_add(&edge->list[LOWER], &cur->upper);
+			edge->node[UPPER] = upper;
+			edge->node[LOWER] = cur;
+
+			goto next;
+		} else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
+			goto next;
+		}
+
+		/* key.type == BTRFS_TREE_BLOCK_REF_KEY */
+		root = read_fs_root(rc->extent_root->fs_info, key.offset);
+		if (IS_ERR(root)) {
+			err = PTR_ERR(root);
+			goto out;
+		}
+
+		if (btrfs_root_level(&root->root_item) == cur->level) {
+			/* tree root */
+			BUG_ON(btrfs_root_bytenr(&root->root_item) !=
+			       cur->bytenr);
+			cur->root = root;
+			break;
+		}
+
+		level = cur->level + 1;
+
+		/*
+		 * searching the tree to find upper level blocks
+		 * reference the block.
+		 */
+		path2->search_commit_root = 1;
+		path2->skip_locking = 1;
+		path2->lowest_level = level;
+		ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
+		path2->lowest_level = 0;
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+
+		eb = path2->nodes[level];
+		WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) !=
+			cur->bytenr);
+
+		lower = cur;
+		for (; level < BTRFS_MAX_LEVEL; level++) {
+			if (!path2->nodes[level]) {
+				BUG_ON(btrfs_root_bytenr(&root->root_item) !=
+				       lower->bytenr);
+				lower->root = root;
+				break;
+			}
+
+			edge = kzalloc(sizeof(*edge), GFP_NOFS);
+			if (!edge) {
+				err = -ENOMEM;
+				goto out;
+			}
+
+			eb = path2->nodes[level];
+			rb_node = tree_search(&cache->rb_root, eb->start);
+			if (!rb_node) {
+				upper = kmalloc(sizeof(*upper), GFP_NOFS);
+				if (!upper) {
+					kfree(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 we know the block isn't shared
+				 * we can void checking its backrefs.
+				 */
+				if (btrfs_block_can_be_shared(root, eb))
+					upper->checked = 0;
+				else
+					upper->checked = 1;
+
+				/*
+				 * add the block to pending list if we
+				 * need check its backrefs. only block
+				 * at 'cur->level + 1' is added to the
+				 * tail of pending list. this guarantees
+				 * we check backrefs from lower level
+				 * blocks to upper level blocks.
+				 */
+				if (!upper->checked &&
+				    level == cur->level + 1) {
+					list_add_tail(&edge->list[UPPER],
+						      &list);
+				} else
+					INIT_LIST_HEAD(&edge->list[UPPER]);
+			} else {
+				upper = rb_entry(rb_node, struct backref_node,
+						 rb_node);
+				BUG_ON(!upper->checked);
+				INIT_LIST_HEAD(&edge->list[UPPER]);
+			}
+			list_add_tail(&edge->list[LOWER], &lower->upper);
+			edge->node[UPPER] = upper;
+			edge->node[LOWER] = lower;
+
+			if (rb_node)
+				break;
+			lower = upper;
+			upper = NULL;
+		}
+		btrfs_release_path(root, path2);
+next:
+		if (ptr < end) {
+			ptr += btrfs_extent_inline_ref_size(key.type);
+			if (ptr >= end) {
+				WARN_ON(ptr > end);
+				ptr = 0;
+				end = 0;
+			}
+		}
+		if (ptr >= end)
+			path1->slots[0]++;
+	}
+	btrfs_release_path(rc->extent_root, path1);
+
+	cur->checked = 1;
+	WARN_ON(exist);
+
+	/* the pending list isn't empty, take the first block to process */
+	if (!list_empty(&list)) {
+		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
+		list_del_init(&edge->list[UPPER]);
+		cur = edge->node[UPPER];
+		goto again;
+	}
+
+	/*
+	 * everything goes well, connect backref nodes and insert backref nodes
+	 * into the cache.
+	 */
+	BUG_ON(!node->checked);
+	rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
+	BUG_ON(rb_node);
+
+	list_for_each_entry(edge, &node->upper, list[LOWER])
+		list_add_tail(&edge->list[UPPER], &list);
+
+	while (!list_empty(&list)) {
+		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
+		list_del_init(&edge->list[UPPER]);
+		upper = edge->node[UPPER];
+
+		if (!RB_EMPTY_NODE(&upper->rb_node)) {
+			if (upper->lowest) {
+				list_del_init(&upper->lower);
+				upper->lowest = 0;
+			}
+
+			list_add_tail(&edge->list[UPPER], &upper->lower);
+			continue;
+		}
+
+		BUG_ON(!upper->checked);
+		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);
+	}
+out:
+	btrfs_free_path(path1);
+	btrfs_free_path(path2);
+	if (err) {
+		INIT_LIST_HEAD(&list);
+		upper = node;
+		while (upper) {
+			if (RB_EMPTY_NODE(&upper->rb_node)) {
+				list_splice_tail(&upper->upper, &list);
+				kfree(upper);
+			}
+
+			if (list_empty(&list))
+				break;
+
+			edge = list_entry(list.next, struct backref_edge,
+					  list[LOWER]);
+			upper = edge->node[UPPER];
+			kfree(edge);
+		}
+		return ERR_PTR(err);
+	}
+	return node;
+}
+
+/*
+ * helper to add 'address of tree root -> reloc tree' mapping
+ */
+static int __add_reloc_root(struct btrfs_root *root)
+{
+	struct rb_node *rb_node;
+	struct mapping_node *node;
+	struct reloc_control *rc = root->fs_info->reloc_ctl;
+
+	node = kmalloc(sizeof(*node), GFP_NOFS);
+	BUG_ON(!node);
+
+	node->bytenr = root->node->start;
+	node->data = root;
+
+	spin_lock(&rc->reloc_root_tree.lock);
+	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
+			      node->bytenr, &node->rb_node);
+	spin_unlock(&rc->reloc_root_tree.lock);
+	BUG_ON(rb_node);
+
+	list_add_tail(&root->root_list, &rc->reloc_roots);
+	return 0;
+}
+
+/*
+ * helper to update/delete the 'address of tree root -> reloc tree'
+ * mapping
+ */
+static int __update_reloc_root(struct btrfs_root *root, int del)
+{
+	struct rb_node *rb_node;
+	struct mapping_node *node = NULL;
+	struct reloc_control *rc = root->fs_info->reloc_ctl;
+
+	spin_lock(&rc->reloc_root_tree.lock);
+	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
+			      root->commit_root->start);
+	if (rb_node) {
+		node = rb_entry(rb_node, struct mapping_node, rb_node);
+		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
+	}
+	spin_unlock(&rc->reloc_root_tree.lock);
+
+	BUG_ON((struct btrfs_root *)node->data != root);
+
+	if (!del) {
+		spin_lock(&rc->reloc_root_tree.lock);
+		node->bytenr = root->node->start;
+		rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
+				      node->bytenr, &node->rb_node);
+		spin_unlock(&rc->reloc_root_tree.lock);
+		BUG_ON(rb_node);
+	} else {
+		list_del_init(&root->root_list);
+		kfree(node);
+	}
+	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)
+{
+	struct btrfs_root *reloc_root;
+	struct extent_buffer *eb;
+	struct btrfs_root_item *root_item;
+	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;
+
+	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);
+	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;
+
+	btrfs_tree_unlock(eb);
+	free_extent_buffer(eb);
+
+	ret = btrfs_insert_root(trans, root->fs_info->tree_root,
+				&root_key, root_item);
+	BUG_ON(ret);
+	kfree(root_item);
+
+	reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
+						 &root_key);
+	BUG_ON(IS_ERR(reloc_root));
+	reloc_root->last_trans = trans->transid;
+
+	__add_reloc_root(reloc_root);
+	root->reloc_root = reloc_root;
+	return 0;
+}
+
+/*
+ * update root item of reloc tree
+ */
+int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	struct btrfs_root_item *root_item;
+	int del = 0;
+	int ret;
+
+	if (!root->reloc_root)
+		return 0;
+
+	reloc_root = root->reloc_root;
+	root_item = &reloc_root->root_item;
+
+	if (btrfs_root_refs(root_item) == 0) {
+		root->reloc_root = NULL;
+		del = 1;
+	}
+
+	__update_reloc_root(reloc_root, del);
+
+	if (reloc_root->commit_root != reloc_root->node) {
+		btrfs_set_root_node(root_item, reloc_root->node);
+		free_extent_buffer(reloc_root->commit_root);
+		reloc_root->commit_root = btrfs_root_node(reloc_root);
+	}
+
+	ret = btrfs_update_root(trans, root->fs_info->tree_root,
+				&reloc_root->root_key, root_item);
+	BUG_ON(ret);
+	return 0;
+}
+
+/*
+ * helper to find first cached inode with inode number >= objectid
+ * in a subvolume
+ */
+static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
+{
+	struct rb_node *node;
+	struct rb_node *prev;
+	struct btrfs_inode *entry;
+	struct inode *inode;
+
+	spin_lock(&root->inode_lock);
+again:
+	node = root->inode_tree.rb_node;
+	prev = NULL;
+	while (node) {
+		prev = node;
+		entry = rb_entry(node, struct btrfs_inode, rb_node);
+
+		if (objectid < entry->vfs_inode.i_ino)
+			node = node->rb_left;
+		else if (objectid > entry->vfs_inode.i_ino)
+			node = node->rb_right;
+		else
+			break;
+	}
+	if (!node) {
+		while (prev) {
+			entry = rb_entry(prev, struct btrfs_inode, rb_node);
+			if (objectid <= entry->vfs_inode.i_ino) {
+				node = prev;
+				break;
+			}
+			prev = rb_next(prev);
+		}
+	}
+	while (node) {
+		entry = rb_entry(node, struct btrfs_inode, rb_node);
+		inode = igrab(&entry->vfs_inode);
+		if (inode) {
+			spin_unlock(&root->inode_lock);
+			return inode;
+		}
+
+		objectid = entry->vfs_inode.i_ino + 1;
+		if (cond_resched_lock(&root->inode_lock))
+			goto again;
+
+		node = rb_next(node);
+	}
+	spin_unlock(&root->inode_lock);
+	return NULL;
+}
+
+static int in_block_group(u64 bytenr,
+			  struct btrfs_block_group_cache *block_group)
+{
+	if (bytenr >= block_group->key.objectid &&
+	    bytenr < block_group->key.objectid + block_group->key.offset)
+		return 1;
+	return 0;
+}
+
+/*
+ * get new location of data
+ */
+static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
+			    u64 bytenr, u64 num_bytes)
+{
+	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
+	struct btrfs_path *path;
+	struct btrfs_file_extent_item *fi;
+	struct extent_buffer *leaf;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	bytenr -= BTRFS_I(reloc_inode)->index_cnt;
+	ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
+				       bytenr, 0);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	fi = btrfs_item_ptr(leaf, path->slots[0],
+			    struct btrfs_file_extent_item);
+
+	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
+	       btrfs_file_extent_compression(leaf, fi) ||
+	       btrfs_file_extent_encryption(leaf, fi) ||
+	       btrfs_file_extent_other_encoding(leaf, fi));
+
+	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
+		ret = 1;
+		goto out;
+	}
+
+	if (new_bytenr)
+		*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * 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)
+{
+	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 num_bytes;
+	u64 end;
+	u32 nritems;
+	u32 i;
+	int ret;
+	int first = 1;
+	int dirty = 0;
+
+	if (rc->stage != UPDATE_DATA_PTRS)
+		return 0;
+
+	/* reloc trees always use full backref */
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
+		parent = leaf->start;
+	else
+		parent = 0;
+
+	nritems = btrfs_header_nritems(leaf);
+	for (i = 0; i < nritems; i++) {
+		cond_resched();
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (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);
+		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
+		if (bytenr == 0)
+			continue;
+		if (!in_block_group(bytenr, rc->block_group))
+			continue;
+
+		/*
+		 * if we are modifying block in fs tree, wait for readpage
+		 * 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) {
+				inode = find_next_inode(root, key.objectid);
+				if (inode)
+					ivec->inode[ivec->nr++] = inode;
+			}
+			if (inode && inode->i_ino == key.objectid) {
+				end = key.offset +
+				      btrfs_file_extent_num_bytes(leaf, fi);
+				WARN_ON(!IS_ALIGNED(key.offset,
+						    root->sectorsize));
+				WARN_ON(!IS_ALIGNED(end, root->sectorsize));
+				end--;
+				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
+						      key.offset, end,
+						      GFP_NOFS);
+				if (!ret)
+					continue;
+
+				btrfs_drop_extent_cache(inode, key.offset, end,
+							1);
+				unlock_extent(&BTRFS_I(inode)->io_tree,
+					      key.offset, end, GFP_NOFS);
+			}
+		}
+
+		ret = get_new_location(rc->data_inode, &new_bytenr,
+				       bytenr, num_bytes);
+		if (ret > 0)
+			continue;
+		BUG_ON(ret < 0);
+
+		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
+		dirty = 1;
+
+		key.offset -= btrfs_file_extent_offset(leaf, fi);
+		ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
+					   num_bytes, parent,
+					   btrfs_header_owner(leaf),
+					   key.objectid, key.offset);
+		BUG_ON(ret);
+
+		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
+					parent, btrfs_header_owner(leaf),
+					key.objectid, key.offset);
+		BUG_ON(ret);
+	}
+	if (dirty)
+		btrfs_mark_buffer_dirty(leaf);
+	return 0;
+}
+
+static noinline_for_stack
+int memcmp_node_keys(struct extent_buffer *eb, int slot,
+		     struct btrfs_path *path, int level)
+{
+	struct btrfs_disk_key key1;
+	struct btrfs_disk_key key2;
+	btrfs_node_key(eb, &key1, slot);
+	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
+	return memcmp(&key1, &key2, sizeof(key1));
+}
+
+/*
+ * try to replace tree blocks in fs tree with the new blocks
+ * in reloc tree. tree blocks haven't been modified since the
+ * reloc tree was create can be replaced.
+ *
+ * if a block was replaced, level of the block + 1 is returned.
+ * 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)
+{
+	struct extent_buffer *eb;
+	struct extent_buffer *parent;
+	struct btrfs_key key;
+	u64 old_bytenr;
+	u64 new_bytenr;
+	u64 old_ptr_gen;
+	u64 new_ptr_gen;
+	u64 last_snapshot;
+	u32 blocksize;
+	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);
+
+	slot = path->slots[lowest_level];
+	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
+
+	eb = btrfs_lock_root_node(dest);
+	btrfs_set_lock_blocking(eb);
+	level = btrfs_header_level(eb);
+
+	if (level < lowest_level) {
+		btrfs_tree_unlock(eb);
+		free_extent_buffer(eb);
+		return 0;
+	}
+
+	ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
+	BUG_ON(ret);
+	btrfs_set_lock_blocking(eb);
+
+	if (next_key) {
+		next_key->objectid = (u64)-1;
+		next_key->type = (u8)-1;
+		next_key->offset = (u64)-1;
+	}
+
+	parent = eb;
+	while (1) {
+		level = btrfs_header_level(parent);
+		BUG_ON(level < lowest_level);
+
+		ret = btrfs_bin_search(parent, &key, level, &slot);
+		if (ret && slot > 0)
+			slot--;
+
+		if (next_key && slot + 1 < btrfs_header_nritems(parent))
+			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
+
+		old_bytenr = btrfs_node_blockptr(parent, slot);
+		blocksize = btrfs_level_size(dest, level - 1);
+		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
+
+		if (level <= max_level) {
+			eb = path->nodes[level];
+			new_bytenr = btrfs_node_blockptr(eb,
+							path->slots[level]);
+			new_ptr_gen = btrfs_node_ptr_generation(eb,
+							path->slots[level]);
+		} else {
+			new_bytenr = 0;
+			new_ptr_gen = 0;
+		}
+
+		if (new_bytenr > 0 && new_bytenr == old_bytenr) {
+			WARN_ON(1);
+			ret = level;
+			break;
+		}
+
+		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
+		    memcmp_node_keys(parent, slot, path, level)) {
+			if (level <= lowest_level && !leaf) {
+				ret = 0;
+				break;
+			}
+
+			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;
+			}
+
+			btrfs_tree_unlock(parent);
+			free_extent_buffer(parent);
+
+			parent = eb;
+			continue;
+		}
+
+		btrfs_node_key_to_cpu(path->nodes[level], &key,
+				      path->slots[level]);
+		btrfs_release_path(src, path);
+
+		path->lowest_level = level;
+		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
+		path->lowest_level = 0;
+		BUG_ON(ret);
+
+		/*
+		 * swap blocks in fs tree and reloc tree.
+		 */
+		btrfs_set_node_blockptr(parent, slot, new_bytenr);
+		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
+		btrfs_mark_buffer_dirty(parent);
+
+		btrfs_set_node_blockptr(path->nodes[level],
+					path->slots[level], old_bytenr);
+		btrfs_set_node_ptr_generation(path->nodes[level],
+					      path->slots[level], old_ptr_gen);
+		btrfs_mark_buffer_dirty(path->nodes[level]);
+
+		ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
+					path->nodes[level]->start,
+					src->root_key.objectid, level - 1, 0);
+		BUG_ON(ret);
+		ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
+					0, dest->root_key.objectid, level - 1,
+					0);
+		BUG_ON(ret);
+
+		ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
+					path->nodes[level]->start,
+					src->root_key.objectid, level - 1, 0);
+		BUG_ON(ret);
+
+		ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
+					0, dest->root_key.objectid, level - 1,
+					0);
+		BUG_ON(ret);
+
+		btrfs_unlock_up_safe(path, 0);
+
+		ret = level;
+		break;
+	}
+	btrfs_tree_unlock(parent);
+	free_extent_buffer(parent);
+	return ret;
+}
+
+/*
+ * helper to find next relocated block in reloc tree
+ */
+static noinline_for_stack
+int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
+		       int *level)
+{
+	struct extent_buffer *eb;
+	int i;
+	u64 last_snapshot;
+	u32 nritems;
+
+	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
+
+	for (i = 0; i < *level; i++) {
+		free_extent_buffer(path->nodes[i]);
+		path->nodes[i] = NULL;
+	}
+
+	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
+		eb = path->nodes[i];
+		nritems = btrfs_header_nritems(eb);
+		while (path->slots[i] + 1 < nritems) {
+			path->slots[i]++;
+			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
+			    last_snapshot)
+				continue;
+
+			*level = i;
+			return 0;
+		}
+		free_extent_buffer(path->nodes[i]);
+		path->nodes[i] = NULL;
+	}
+	return 1;
+}
+
+/*
+ * walk down reloc tree to find relocated block of lowest level
+ */
+static noinline_for_stack
+int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
+			 int *level)
+{
+	struct extent_buffer *eb = NULL;
+	int i;
+	u64 bytenr;
+	u64 ptr_gen = 0;
+	u64 last_snapshot;
+	u32 blocksize;
+	u32 nritems;
+
+	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
+
+	for (i = *level; i > 0; i--) {
+		eb = path->nodes[i];
+		nritems = btrfs_header_nritems(eb);
+		while (path->slots[i] < nritems) {
+			ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
+			if (ptr_gen > last_snapshot)
+				break;
+			path->slots[i]++;
+		}
+		if (path->slots[i] >= nritems) {
+			if (i == *level)
+				break;
+			*level = i + 1;
+			return 0;
+		}
+		if (i == 1) {
+			*level = i;
+			return 0;
+		}
+
+		bytenr = btrfs_node_blockptr(eb, path->slots[i]);
+		blocksize = btrfs_level_size(root, i - 1);
+		eb = read_tree_block(root, bytenr, blocksize, ptr_gen);
+		BUG_ON(btrfs_header_level(eb) != i - 1);
+		path->nodes[i - 1] = eb;
+		path->slots[i - 1] = 0;
+	}
+	return 1;
+}
+
+/*
+ * invalidate extent cache for file extents whose key in range of
+ * [min_key, max_key)
+ */
+static int invalidate_extent_cache(struct btrfs_root *root,
+				   struct btrfs_key *min_key,
+				   struct btrfs_key *max_key)
+{
+	struct inode *inode = NULL;
+	u64 objectid;
+	u64 start, end;
+
+	objectid = min_key->objectid;
+	while (1) {
+		cond_resched();
+		iput(inode);
+
+		if (objectid > max_key->objectid)
+			break;
+
+		inode = find_next_inode(root, objectid);
+		if (!inode)
+			break;
+
+		if (inode->i_ino > max_key->objectid) {
+			iput(inode);
+			break;
+		}
+
+		objectid = inode->i_ino + 1;
+		if (!S_ISREG(inode->i_mode))
+			continue;
+
+		if (unlikely(min_key->objectid == inode->i_ino)) {
+			if (min_key->type > BTRFS_EXTENT_DATA_KEY)
+				continue;
+			if (min_key->type < BTRFS_EXTENT_DATA_KEY)
+				start = 0;
+			else {
+				start = min_key->offset;
+				WARN_ON(!IS_ALIGNED(start, root->sectorsize));
+			}
+		} else {
+			start = 0;
+		}
+
+		if (unlikely(max_key->objectid == inode->i_ino)) {
+			if (max_key->type < BTRFS_EXTENT_DATA_KEY)
+				continue;
+			if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
+				end = (u64)-1;
+			} else {
+				if (max_key->offset == 0)
+					continue;
+				end = max_key->offset;
+				WARN_ON(!IS_ALIGNED(end, root->sectorsize));
+				end--;
+			}
+		} else {
+			end = (u64)-1;
+		}
+
+		/* the lock_extent waits for readpage to complete */
+		lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+		btrfs_drop_extent_cache(inode, start, end, 1);
+		unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+	}
+	return 0;
+}
+
+static int find_next_key(struct btrfs_path *path, int level,
+			 struct btrfs_key *key)
+
+{
+	while (level < BTRFS_MAX_LEVEL) {
+		if (!path->nodes[level])
+			break;
+		if (path->slots[level] + 1 <
+		    btrfs_header_nritems(path->nodes[level])) {
+			btrfs_node_key_to_cpu(path->nodes[level], key,
+					      path->slots[level] + 1);
+			return 0;
+		}
+		level++;
+	}
+	return 1;
+}
+
+/*
+ * merge the relocated tree blocks in reloc tree with corresponding
+ * fs tree.
+ */
+static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
+					       struct btrfs_root *root)
+{
+	LIST_HEAD(inode_list);
+	struct btrfs_key key;
+	struct btrfs_key next_key;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *reloc_root;
+	struct btrfs_root_item *root_item;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf = NULL;
+	unsigned long nr;
+	int level;
+	int max_level;
+	int replaced = 0;
+	int ret;
+	int err = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	reloc_root = root->reloc_root;
+	root_item = &reloc_root->root_item;
+
+	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
+		level = btrfs_root_level(root_item);
+		extent_buffer_get(reloc_root->node);
+		path->nodes[level] = reloc_root->node;
+		path->slots[level] = 0;
+	} else {
+		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
+
+		level = root_item->drop_level;
+		BUG_ON(level == 0);
+		path->lowest_level = level;
+		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
+		if (ret < 0) {
+			btrfs_free_path(path);
+			return ret;
+		}
+
+		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
+				      path->slots[level]);
+		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
+
+		btrfs_unlock_up_safe(path, 0);
+	}
+
+	if (level == 0 && rc->stage == UPDATE_DATA_PTRS) {
+		trans = btrfs_start_transaction(root, 1);
+
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, 0);
+		btrfs_release_path(reloc_root, path);
+
+		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+
+		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);
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+		if (ret > 0)
+			break;
+
+		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);
+		}
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+
+		if (ret > 0) {
+			level = ret;
+			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);
+		if (ret > 0)
+			break;
+
+		BUG_ON(level == 0);
+		/*
+		 * save the merging progress in the drop_progress.
+		 * this is OK since root refs == 1 in this case.
+		 */
+		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
+			       path->slots[level]);
+		root_item->drop_level = level;
+
+		nr = trans->blocks_used;
+		btrfs_end_transaction(trans, root);
+
+		btrfs_btree_balance_dirty(root, nr);
+
+		if (replaced && rc->stage == UPDATE_DATA_PTRS)
+			invalidate_extent_cache(root, &key, &next_key);
+	}
+
+	/*
+	 * handle the case only one block in the fs tree need to be
+	 * relocated and the block is tree root.
+	 */
+	leaf = btrfs_lock_root_node(root);
+	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
+	btrfs_tree_unlock(leaf);
+	free_extent_buffer(leaf);
+	if (ret < 0)
+		err = ret;
+out:
+	btrfs_free_path(path);
+
+	if (err == 0) {
+		memset(&root_item->drop_progress, 0,
+		       sizeof(root_item->drop_progress));
+		root_item->drop_level = 0;
+		btrfs_set_root_refs(root_item, 0);
+	}
+
+	nr = trans->blocks_used;
+	btrfs_end_transaction(trans, root);
+
+	btrfs_btree_balance_dirty(root, nr);
+
+	/*
+	 * put inodes while we aren't holding the tree locks
+	 */
+	while (!list_empty(&inode_list)) {
+		struct inodevec *ivec;
+		ivec = list_entry(inode_list.next, struct inodevec, list);
+		list_del(&ivec->list);
+		while (ivec->nr > 0) {
+			ivec->nr--;
+			iput(ivec->inode[ivec->nr]);
+		}
+		kfree(ivec);
+	}
+
+	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)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root;
+	struct btrfs_root *reloc_root;
+	struct async_merge *async;
+
+	async = container_of(work, struct async_merge, work);
+	reloc_root = async->root;
+
+	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);
+		btrfs_update_reloc_root(trans, root);
+		btrfs_end_transaction(trans, root);
+	}
+
+	btrfs_drop_dead_root(reloc_root);
+
+	if (atomic_dec_and_test(async->num_pending))
+		complete(async->done);
+
+	kfree(async);
+}
+
+static int merge_reloc_roots(struct reloc_control *rc)
+{
+	struct async_merge *async;
+	struct btrfs_root *root;
+	struct completion done;
+	atomic_t num_pending;
+
+	init_completion(&done);
+	atomic_set(&num_pending, 1);
+
+	while (!list_empty(&rc->reloc_roots)) {
+		root = list_entry(rc->reloc_roots.next,
+				  struct btrfs_root, root_list);
+		list_del_init(&root->root_list);
+
+		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);
+	}
+
+	if (!atomic_dec_and_test(&num_pending))
+		wait_for_completion(&done);
+
+	BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
+	return 0;
+}
+
+static void free_block_list(struct rb_root *blocks)
+{
+	struct tree_block *block;
+	struct rb_node *rb_node;
+	while ((rb_node = rb_first(blocks))) {
+		block = rb_entry(rb_node, struct tree_block, rb_node);
+		rb_erase(rb_node, blocks);
+		kfree(block);
+	}
+}
+
+static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *reloc_root)
+{
+	struct btrfs_root *root;
+
+	if (reloc_root->last_trans == trans->transid)
+		return 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);
+
+	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)
+{
+	struct backref_node *next;
+	struct btrfs_root *root;
+	int index;
+	int loop = 0;
+again:
+	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;
+		}
+
+		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);
+			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:
+		root = NULL;
+		next = walk_down_backref(edges, &index);
+		if (!next || next->level <= node->level)
+			break;
+	}
+
+	if (!root && !loop && !reloc_only) {
+		loop = 1;
+		goto again;
+	}
+
+	if (root)
+		*nr = index;
+	else
+		*nr = 0;
+
+	return root;
+}
+
+static noinline_for_stack
+struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
+				   struct backref_node *node)
+{
+	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
+	int nr;
+	return __select_one_root(trans, node, edges, &nr, 0);
+}
+
+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)
+{
+	return __select_one_root(trans, node, edges, nr, 1);
+}
+
+static void grab_path_buffers(struct btrfs_path *path,
+			      struct backref_node *node,
+			      struct backref_edge *edges[], int nr)
+{
+	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;
+
+		edges[i]->blockptr = node->eb->start;
+		node = edges[i]->node[UPPER];
+		i++;
+	}
+}
+
+/*
+ * relocate a block tree, and then update pointers in upper level
+ * blocks that reference the block to point to the new location.
+ *
+ * if called by link_to_upper, the block has already been relocated.
+ * in that case this function just updates pointers.
+ */
+static int do_relocation(struct btrfs_trans_handle *trans,
+			 struct backref_node *node,
+			 struct btrfs_key *key,
+			 struct btrfs_path *path, int lowest)
+{
+	struct backref_node *upper;
+	struct backref_edge *edge;
+	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
+	struct btrfs_root *root;
+	struct extent_buffer *eb;
+	u32 blocksize;
+	u64 bytenr;
+	u64 generation;
+	int nr;
+	int slot;
+	int ret;
+	int err = 0;
+
+	BUG_ON(lowest && node->eb);
+
+	path->lowest_level = node->level + 1;
+	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)
+			drop_node_buffer(upper);
+
+		if (!upper->eb) {
+			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
+			if (ret < 0) {
+				err = ret;
+				break;
+			}
+			BUG_ON(ret > 0);
+
+			slot = path->slots[upper->level];
+
+			btrfs_unlock_up_safe(path, upper->level + 1);
+			grab_path_buffers(path, upper, edges, nr);
+
+			btrfs_release_path(NULL, path);
+		} else {
+			ret = btrfs_bin_search(upper->eb, key, upper->level,
+					       &slot);
+			BUG_ON(ret);
+		}
+
+		bytenr = btrfs_node_blockptr(upper->eb, slot);
+		if (!lowest) {
+			if (node->eb->start == bytenr) {
+				btrfs_tree_unlock(upper->eb);
+				upper->locked = 0;
+				continue;
+			}
+		} else {
+			BUG_ON(node->bytenr != bytenr);
+		}
+
+		blocksize = btrfs_level_size(root, node->level);
+		generation = btrfs_node_ptr_generation(upper->eb, slot);
+		eb = read_tree_block(root, bytenr, blocksize, generation);
+		btrfs_tree_lock(eb);
+		btrfs_set_lock_blocking(eb);
+
+		if (!node->eb) {
+			ret = btrfs_cow_block(trans, root, eb, upper->eb,
+					      slot, &eb);
+			if (ret < 0) {
+				err = ret;
+				break;
+			}
+			btrfs_set_lock_blocking(eb);
+			node->eb = eb;
+			node->locked = 1;
+		} else {
+			btrfs_set_node_blockptr(upper->eb, slot,
+						node->eb->start);
+			btrfs_set_node_ptr_generation(upper->eb, slot,
+						      trans->transid);
+			btrfs_mark_buffer_dirty(upper->eb);
+
+			ret = btrfs_inc_extent_ref(trans, root,
+						node->eb->start, blocksize,
+						upper->eb->start,
+						btrfs_header_owner(upper->eb),
+						node->level, 0);
+			BUG_ON(ret);
+
+			ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
+			BUG_ON(ret);
+
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+		}
+		if (!lowest) {
+			btrfs_tree_unlock(upper->eb);
+			upper->locked = 0;
+		}
+	}
+	path->lowest_level = 0;
+	return err;
+}
+
+static int link_to_upper(struct btrfs_trans_handle *trans,
+			 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);
+}
+
+static int finish_pending_nodes(struct btrfs_trans_handle *trans,
+				struct backref_cache *cache,
+				struct btrfs_path *path)
+{
+	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);
+
+			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);
+		}
+	}
+	BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
+	return err;
+}
+
+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);
+	}
+	node->processed = 1;
+}
+
+/*
+ * mark a block and all blocks directly/indirectly reference the block
+ * as processed.
+ */
+static void update_processed_blocks(struct reloc_control *rc,
+				    struct backref_node *node)
+{
+	struct backref_node *next = node;
+	struct backref_edge *edge;
+	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
+	int index = 0;
+
+	while (next) {
+		cond_resched();
+		while (1) {
+			if (next->processed)
+				break;
+
+			mark_block_processed(rc, next);
+
+			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);
+	}
+}
+
+static int tree_block_processed(u64 bytenr, u32 blocksize,
+				struct reloc_control *rc)
+{
+	if (test_range_bit(&rc->processed_blocks, bytenr,
+			   bytenr + blocksize - 1, EXTENT_DIRTY, 1))
+		return 1;
+	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)
+{
+	struct extent_buffer *eb;
+
+	BUG_ON(block->key_ready);
+	eb = read_tree_block(rc->extent_root, block->bytenr,
+			     block->key.objectid, block->key.offset);
+	WARN_ON(btrfs_header_level(eb) != block->level);
+	if (block->level == 0)
+		btrfs_item_key_to_cpu(eb, &block->key, 0);
+	else
+		btrfs_node_key_to_cpu(eb, &block->key, 0);
+	free_extent_buffer(eb);
+	block->key_ready = 1;
+	return 0;
+}
+
+static int reada_tree_block(struct reloc_control *rc,
+			    struct tree_block *block)
+{
+	BUG_ON(block->key_ready);
+	readahead_tree_block(rc->extent_root, block->bytenr,
+			     block->key.objectid, block->key.offset);
+	return 0;
+}
+
+/*
+ * helper function to relocate a tree block
+ */
+static int relocate_tree_block(struct btrfs_trans_handle *trans,
+				struct reloc_control *rc,
+				struct backref_node *node,
+				struct btrfs_key *key,
+				struct btrfs_path *path)
+{
+	struct btrfs_root *root;
+	int ret;
+
+	root = select_one_root(trans, node);
+	if (unlikely(!root)) {
+		rc->found_old_snapshot = 1;
+		update_processed_blocks(rc, node);
+		return 0;
+	}
+
+	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)
+			goto out;
+	} else if (root != node->root) {
+		WARN_ON(node->level > 0 || rc->stage != UPDATE_DATA_PTRS);
+	}
+
+	update_processed_blocks(rc, node);
+	ret = 0;
+out:
+	drop_node_buffer(node);
+	return ret;
+}
+
+/*
+ * relocate a list of blocks
+ */
+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;
+
+	path = btrfs_alloc_path();
+	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);
+	}
+
+	rb_node = rb_first(blocks);
+	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;
+
+		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;
+			}
+
+			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);
+		}
+		free_block_list(blocks);
+
+		if (upper) {
+			ret = link_to_upper(trans, upper, path);
+			if (ret < 0) {
+				err = ret;
+				break;
+			}
+			remove_backref_node(cache, upper);
+		}
+	}
+out:
+	free_block_list(blocks);
+
+	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 relocate_inode_pages(struct inode *inode, u64 start, u64 len)
+{
+	u64 page_start;
+	u64 page_end;
+	unsigned long i;
+	unsigned long first_index;
+	unsigned long last_index;
+	unsigned int total_read = 0;
+	unsigned int total_dirty = 0;
+	struct page *page;
+	struct file_ra_state *ra;
+	struct btrfs_ordered_extent *ordered;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	int ret = 0;
+
+	ra = kzalloc(sizeof(*ra), GFP_NOFS);
+	if (!ra)
+		return -ENOMEM;
+
+	mutex_lock(&inode->i_mutex);
+	first_index = start >> PAGE_CACHE_SHIFT;
+	last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
+
+	/* make sure the dirty trick played by the caller work */
+	ret = invalidate_inode_pages2_range(inode->i_mapping,
+					    first_index, last_index);
+	if (ret)
+		goto out_unlock;
+
+	file_ra_state_init(ra, inode->i_mapping);
+
+	for (i = first_index ; i <= last_index; i++) {
+		if (total_read % ra->ra_pages == 0) {
+			btrfs_force_ra(inode->i_mapping, ra, NULL, i,
+				min(last_index, ra->ra_pages + i - 1));
+		}
+		total_read++;
+again:
+		if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
+			BUG_ON(1);
+		page = grab_cache_page(inode->i_mapping, i);
+		if (!page) {
+			ret = -ENOMEM;
+			goto out_unlock;
+		}
+		if (!PageUptodate(page)) {
+			btrfs_readpage(NULL, page);
+			lock_page(page);
+			if (!PageUptodate(page)) {
+				unlock_page(page);
+				page_cache_release(page);
+				ret = -EIO;
+				goto out_unlock;
+			}
+		}
+		wait_on_page_writeback(page);
+
+		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
+		page_end = page_start + PAGE_CACHE_SIZE - 1;
+		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+		ordered = btrfs_lookup_ordered_extent(inode, page_start);
+		if (ordered) {
+			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			goto again;
+		}
+		set_page_extent_mapped(page);
+
+		if (i == first_index)
+			set_extent_bits(io_tree, page_start, page_end,
+					EXTENT_BOUNDARY, GFP_NOFS);
+		btrfs_set_extent_delalloc(inode, page_start, page_end);
+
+		set_page_dirty(page);
+		total_dirty++;
+
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		unlock_page(page);
+		page_cache_release(page);
+	}
+out_unlock:
+	mutex_unlock(&inode->i_mutex);
+	kfree(ra);
+	balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
+	return ret;
+}
+
+static noinline_for_stack
+int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct extent_map *em;
+	u64 start = extent_key->objectid - BTRFS_I(inode)->index_cnt;
+	u64 end = start + extent_key->offset - 1;
+
+	em = alloc_extent_map(GFP_NOFS);
+	em->start = start;
+	em->len = extent_key->offset;
+	em->block_len = extent_key->offset;
+	em->block_start = extent_key->objectid;
+	em->bdev = root->fs_info->fs_devices->latest_bdev;
+	set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	/* setup extent map to cheat btrfs_readpage */
+	lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+	while (1) {
+		int ret;
+		spin_lock(&em_tree->lock);
+		ret = add_extent_mapping(em_tree, em);
+		spin_unlock(&em_tree->lock);
+		if (ret != -EEXIST) {
+			free_extent_map(em);
+			break;
+		}
+		btrfs_drop_extent_cache(inode, start, end, 0);
+	}
+	unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+
+	return relocate_inode_pages(inode, start, extent_key->offset);
+}
+
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+static int get_ref_objectid_v0(struct reloc_control *rc,
+			       struct btrfs_path *path,
+			       struct btrfs_key *extent_key,
+			       u64 *ref_objectid, int *path_change)
+{
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref_v0 *ref0;
+	int ret;
+	int slot;
+
+	leaf = path->nodes[0];
+	slot = path->slots[0];
+	while (1) {
+		if (slot >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(rc->extent_root, path);
+			if (ret < 0)
+				return ret;
+			BUG_ON(ret > 0);
+			leaf = path->nodes[0];
+			slot = path->slots[0];
+			if (path_change)
+				*path_change = 1;
+		}
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (key.objectid != extent_key->objectid)
+			return -ENOENT;
+
+		if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
+			slot++;
+			continue;
+		}
+		ref0 = btrfs_item_ptr(leaf, slot,
+				struct btrfs_extent_ref_v0);
+		*ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
+		break;
+	}
+	return 0;
+}
+#endif
+
+/*
+ * helper to add a tree block to the list.
+ * the major work is getting the generation and level of the block
+ */
+static int add_tree_block(struct reloc_control *rc,
+			  struct btrfs_key *extent_key,
+			  struct btrfs_path *path,
+			  struct rb_root *blocks)
+{
+	struct extent_buffer *eb;
+	struct btrfs_extent_item *ei;
+	struct btrfs_tree_block_info *bi;
+	struct tree_block *block;
+	struct rb_node *rb_node;
+	u32 item_size;
+	int level = -1;
+	int generation;
+
+	eb =  path->nodes[0];
+	item_size = btrfs_item_size_nr(eb, path->slots[0]);
+
+	if (item_size >= sizeof(*ei) + sizeof(*bi)) {
+		ei = btrfs_item_ptr(eb, path->slots[0],
+				struct btrfs_extent_item);
+		bi = (struct btrfs_tree_block_info *)(ei + 1);
+		generation = btrfs_extent_generation(eb, ei);
+		level = btrfs_tree_block_level(eb, bi);
+	} else {
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+		u64 ref_owner;
+		int ret;
+
+		BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
+		ret = get_ref_objectid_v0(rc, path, extent_key,
+					  &ref_owner, NULL);
+		BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
+		level = (int)ref_owner;
+		/* FIXME: get real generation */
+		generation = 0;
+#else
+		BUG();
+#endif
+	}
+
+	btrfs_release_path(rc->extent_root, path);
+
+	BUG_ON(level == -1);
+
+	block = kmalloc(sizeof(*block), GFP_NOFS);
+	if (!block)
+		return -ENOMEM;
+
+	block->bytenr = extent_key->objectid;
+	block->key.objectid = extent_key->offset;
+	block->key.offset = generation;
+	block->level = level;
+	block->key_ready = 0;
+
+	rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
+	BUG_ON(rb_node);
+
+	return 0;
+}
+
+/*
+ * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
+ */
+static int __add_tree_block(struct reloc_control *rc,
+			    u64 bytenr, u32 blocksize,
+			    struct rb_root *blocks)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int ret;
+
+	if (tree_block_processed(bytenr, blocksize, rc))
+		return 0;
+
+	if (tree_search(blocks, bytenr))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = blocksize;
+
+	path->search_commit_root = 1;
+	path->skip_locking = 1;
+	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	BUG_ON(ret);
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+	ret = add_tree_block(rc, &key, path, blocks);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * helper to check if the block use full backrefs for pointers in it
+ */
+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;
+
+	if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
+	    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);
+	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;
+}
+
+/*
+ * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
+ * this function scans fs tree to find blocks reference the data extent
+ */
+static int find_data_references(struct reloc_control *rc,
+				struct btrfs_key *extent_key,
+				struct extent_buffer *leaf,
+				struct btrfs_extent_data_ref *ref,
+				struct rb_root *blocks)
+{
+	struct btrfs_path *path;
+	struct tree_block *block;
+	struct btrfs_root *root;
+	struct btrfs_file_extent_item *fi;
+	struct rb_node *rb_node;
+	struct btrfs_key key;
+	u64 ref_root;
+	u64 ref_objectid;
+	u64 ref_offset;
+	u32 ref_count;
+	u32 nritems;
+	int err = 0;
+	int added = 0;
+	int counted;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ref_root = btrfs_extent_data_ref_root(leaf, ref);
+	ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
+	ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
+	ref_count = btrfs_extent_data_ref_count(leaf, ref);
+
+	root = read_fs_root(rc->extent_root->fs_info, ref_root);
+	if (IS_ERR(root)) {
+		err = PTR_ERR(root);
+		goto out;
+	}
+
+	key.objectid = ref_objectid;
+	key.offset = ref_offset;
+	key.type = BTRFS_EXTENT_DATA_KEY;
+
+	path->search_commit_root = 1;
+	path->skip_locking = 1;
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	nritems = btrfs_header_nritems(leaf);
+	/*
+	 * the references in tree blocks that use full backrefs
+	 * are not counted in
+	 */
+	if (block_use_full_backref(rc, leaf))
+		counted = 0;
+	else
+		counted = 1;
+	rb_node = tree_search(blocks, leaf->start);
+	if (rb_node) {
+		if (counted)
+			added = 1;
+		else
+			path->slots[0] = nritems;
+	}
+
+	while (ref_count > 0) {
+		while (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0) {
+				err = ret;
+				goto out;
+			}
+			if (ret > 0) {
+				WARN_ON(1);
+				goto out;
+			}
+
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+			added = 0;
+
+			if (block_use_full_backref(rc, leaf))
+				counted = 0;
+			else
+				counted = 1;
+			rb_node = tree_search(blocks, leaf->start);
+			if (rb_node) {
+				if (counted)
+					added = 1;
+				else
+					path->slots[0] = nritems;
+			}
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid != ref_objectid ||
+		    key.type != BTRFS_EXTENT_DATA_KEY) {
+			WARN_ON(1);
+			break;
+		}
+
+		fi = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_file_extent_item);
+
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			goto next;
+
+		if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
+		    extent_key->objectid)
+			goto next;
+
+		key.offset -= btrfs_file_extent_offset(leaf, fi);
+		if (key.offset != ref_offset)
+			goto next;
+
+		if (counted)
+			ref_count--;
+		if (added)
+			goto next;
+
+		if (!tree_block_processed(leaf->start, leaf->len, rc)) {
+			block = kmalloc(sizeof(*block), GFP_NOFS);
+			if (!block) {
+				err = -ENOMEM;
+				break;
+			}
+			block->bytenr = leaf->start;
+			btrfs_item_key_to_cpu(leaf, &block->key, 0);
+			block->level = 0;
+			block->key_ready = 1;
+			rb_node = tree_insert(blocks, block->bytenr,
+					      &block->rb_node);
+			BUG_ON(rb_node);
+		}
+		if (counted)
+			added = 1;
+		else
+			path->slots[0] = nritems;
+next:
+		path->slots[0]++;
+
+	}
+out:
+	btrfs_free_path(path);
+	return err;
+}
+
+/*
+ * hepler to find all tree blocks that reference a given data extent
+ */
+static noinline_for_stack
+int add_data_references(struct reloc_control *rc,
+			struct btrfs_key *extent_key,
+			struct btrfs_path *path,
+			struct rb_root *blocks)
+{
+	struct btrfs_key key;
+	struct extent_buffer *eb;
+	struct btrfs_extent_data_ref *dref;
+	struct btrfs_extent_inline_ref *iref;
+	unsigned long ptr;
+	unsigned long end;
+	u32 blocksize;
+	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]);
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+	if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
+		ptr = end;
+	else
+#endif
+		ptr += sizeof(struct btrfs_extent_item);
+
+	while (ptr < end) {
+		iref = (struct btrfs_extent_inline_ref *)ptr;
+		key.type = btrfs_extent_inline_ref_type(eb, iref);
+		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
+			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
+			ret = __add_tree_block(rc, key.offset, blocksize,
+					       blocks);
+		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
+			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
+			ret = find_data_references(rc, extent_key,
+						   eb, dref, blocks);
+		} else {
+			BUG();
+		}
+		ptr += btrfs_extent_inline_ref_size(key.type);
+	}
+	WARN_ON(ptr > end);
+
+	while (1) {
+		cond_resched();
+		eb = path->nodes[0];
+		if (path->slots[0] >= btrfs_header_nritems(eb)) {
+			ret = btrfs_next_leaf(rc->extent_root, path);
+			if (ret < 0) {
+				err = ret;
+				break;
+			}
+			if (ret > 0)
+				break;
+			eb = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
+		if (key.objectid != extent_key->objectid)
+			break;
+
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+		if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
+		    key.type == BTRFS_EXTENT_REF_V0_KEY) {
+#else
+		BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
+		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
+#endif
+			ret = __add_tree_block(rc, key.offset, blocksize,
+					       blocks);
+		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
+			dref = btrfs_item_ptr(eb, path->slots[0],
+					      struct btrfs_extent_data_ref);
+			ret = find_data_references(rc, extent_key,
+						   eb, dref, blocks);
+		} else {
+			ret = 0;
+		}
+		if (ret) {
+			err = ret;
+			break;
+		}
+		path->slots[0]++;
+	}
+	btrfs_release_path(rc->extent_root, path);
+	if (err)
+		free_block_list(blocks);
+	return err;
+}
+
+/*
+ * hepler to find next unprocessed extent
+ */
+static noinline_for_stack
+int find_next_extent(struct btrfs_trans_handle *trans,
+		     struct reloc_control *rc, struct btrfs_path *path)
+{
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	u64 start, end, last;
+	int ret;
+
+	last = rc->block_group->key.objectid + rc->block_group->key.offset;
+	while (1) {
+		cond_resched();
+		if (rc->search_start >= last) {
+			ret = 1;
+			break;
+		}
+
+		key.objectid = rc->search_start;
+		key.type = BTRFS_EXTENT_ITEM_KEY;
+		key.offset = 0;
+
+		path->search_commit_root = 1;
+		path->skip_locking = 1;
+		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
+					0, 0);
+		if (ret < 0)
+			break;
+next:
+		leaf = path->nodes[0];
+		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(rc->extent_root, path);
+			if (ret != 0)
+				break;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid >= last) {
+			ret = 1;
+			break;
+		}
+
+		if (key.type != BTRFS_EXTENT_ITEM_KEY ||
+		    key.objectid + key.offset <= rc->search_start) {
+			path->slots[0]++;
+			goto next;
+		}
+
+		ret = find_first_extent_bit(&rc->processed_blocks,
+					    key.objectid, &start, &end,
+					    EXTENT_DIRTY);
+
+		if (ret == 0 && start <= key.objectid) {
+			btrfs_release_path(rc->extent_root, path);
+			rc->search_start = end + 1;
+		} else {
+			rc->search_start = key.objectid + key.offset;
+			return 0;
+		}
+	}
+	btrfs_release_path(rc->extent_root, path);
+	return ret;
+}
+
+static void set_reloc_control(struct reloc_control *rc)
+{
+	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
+	mutex_lock(&fs_info->trans_mutex);
+	fs_info->reloc_ctl = rc;
+	mutex_unlock(&fs_info->trans_mutex);
+}
+
+static void unset_reloc_control(struct reloc_control *rc)
+{
+	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
+	mutex_lock(&fs_info->trans_mutex);
+	fs_info->reloc_ctl = NULL;
+	mutex_unlock(&fs_info->trans_mutex);
+}
+
+static int check_extent_flags(u64 flags)
+{
+	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
+	    (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
+		return 1;
+	if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
+	    !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
+		return 1;
+	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
+	    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+		return 1;
+	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 btrfs_trans_handle *trans = NULL;
+	struct btrfs_path *path;
+	struct btrfs_extent_item *ei;
+	unsigned long nr;
+	u64 flags;
+	u32 item_size;
+	int ret;
+	int err = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	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);
+
+	while (1) {
+		trans = btrfs_start_transaction(rc->extent_root, 1);
+
+		ret = find_next_extent(trans, rc, path);
+		if (ret < 0)
+			err = ret;
+		if (ret != 0)
+			break;
+
+		rc->extents_found++;
+
+		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]);
+		if (item_size >= sizeof(*ei)) {
+			flags = btrfs_extent_flags(path->nodes[0], ei);
+			ret = check_extent_flags(flags);
+			BUG_ON(ret);
+
+		} else {
+#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
+			u64 ref_owner;
+			int path_change = 0;
+
+			BUG_ON(item_size !=
+			       sizeof(struct btrfs_extent_item_v0));
+			ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
+						  &path_change);
+			if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
+				flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
+			else
+				flags = BTRFS_EXTENT_FLAG_DATA;
+
+			if (path_change) {
+				btrfs_release_path(rc->extent_root, path);
+
+				path->search_commit_root = 1;
+				path->skip_locking = 1;
+				ret = btrfs_search_slot(NULL, rc->extent_root,
+							&key, path, 0, 0);
+				if (ret < 0) {
+					err = ret;
+					break;
+				}
+				BUG_ON(ret > 0);
+			}
+#else
+			BUG();
+#endif
+		}
+
+		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)) {
+			ret = add_data_references(rc, &key, path, &blocks);
+		} else {
+			btrfs_release_path(rc->extent_root, path);
+			ret = 0;
+		}
+		if (ret < 0) {
+			err = 0;
+			break;
+		}
+
+		if (!RB_EMPTY_ROOT(&blocks)) {
+			ret = relocate_tree_blocks(trans, rc, &blocks);
+			if (ret < 0) {
+				err = ret;
+				break;
+			}
+		}
+
+		nr = trans->blocks_used;
+		btrfs_end_transaction_throttle(trans, rc->extent_root);
+		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);
+			if (ret < 0) {
+				err = ret;
+				break;
+			}
+		}
+	}
+	btrfs_free_path(path);
+
+	if (trans) {
+		nr = trans->blocks_used;
+		btrfs_end_transaction(trans, rc->extent_root);
+		btrfs_btree_balance_dirty(rc->extent_root, nr);
+	}
+
+	rc->create_reloc_root = 0;
+	smp_mb();
+
+	if (rc->extents_found > 0) {
+		trans = btrfs_start_transaction(rc->extent_root, 1);
+		btrfs_commit_transaction(trans, rc->extent_root);
+	}
+
+	merge_reloc_roots(rc);
+
+	unset_reloc_control(rc);
+
+	/* get rid of pinned extents */
+	trans = btrfs_start_transaction(rc->extent_root, 1);
+	btrfs_commit_transaction(trans, rc->extent_root);
+
+	return err;
+}
+
+static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 u64 objectid, u64 size)
+{
+	struct btrfs_path *path;
+	struct btrfs_inode_item *item;
+	struct extent_buffer *leaf;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
+	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
+	btrfs_set_inode_generation(leaf, item, 1);
+	btrfs_set_inode_size(leaf, item, size);
+	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
+	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(root, path);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * 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)
+{
+	struct inode *inode = NULL;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root;
+	struct btrfs_key key;
+	unsigned long nr;
+	u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
+	int err = 0;
+
+	root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
+	if (IS_ERR(root))
+		return ERR_CAST(root);
+
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+
+	err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
+	if (err)
+		goto out;
+
+	err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
+	BUG_ON(err);
+
+	err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
+				       group->key.offset, 0, group->key.offset,
+				       0, 0, 0);
+	BUG_ON(err);
+
+	key.objectid = objectid;
+	key.type = BTRFS_INODE_ITEM_KEY;
+	key.offset = 0;
+	inode = btrfs_iget(root->fs_info->sb, &key, root);
+	BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
+	BTRFS_I(inode)->index_cnt = group->key.objectid;
+
+	err = btrfs_orphan_add(trans, inode);
+out:
+	nr = trans->blocks_used;
+	btrfs_end_transaction(trans, root);
+
+	btrfs_btree_balance_dirty(root, nr);
+	if (err) {
+		if (inode)
+			iput(inode);
+		inode = ERR_PTR(err);
+	}
+	return inode;
+}
+
+/*
+ * function to relocate all extents in a block group.
+ */
+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 err = 0;
+
+	rc = kzalloc(sizeof(*rc), GFP_NOFS);
+	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->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);
+
+	rc->extent_root = extent_root;
+	btrfs_prepare_block_group_relocation(extent_root, rc->block_group);
+
+	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;
+		goto out;
+	}
+
+	printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n",
+	       (unsigned long long)rc->block_group->key.objectid,
+	       (unsigned long long)rc->block_group->flags);
+
+	btrfs_start_delalloc_inodes(fs_info->tree_root);
+	btrfs_wait_ordered_extents(fs_info->tree_root, 0);
+
+	while (1) {
+		mutex_lock(&fs_info->cleaner_mutex);
+		btrfs_clean_old_snapshots(fs_info->tree_root);
+		mutex_unlock(&fs_info->cleaner_mutex);
+
+		rc->extents_found = 0;
+		rc->extents_skipped = 0;
+
+		ret = relocate_block_group(rc);
+		if (ret < 0) {
+			err = ret;
+			break;
+		}
+
+		if (rc->extents_found == 0)
+			break;
+
+		printk(KERN_INFO "btrfs: found %llu extents\n",
+			(unsigned long long)rc->extents_found);
+
+		if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
+			btrfs_wait_ordered_range(rc->data_inode, 0, (u64)-1);
+			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;
+		}
+	}
+
+	filemap_fdatawrite_range(fs_info->btree_inode->i_mapping,
+				 rc->block_group->key.objectid,
+				 rc->block_group->key.objectid +
+				 rc->block_group->key.offset - 1);
+
+	WARN_ON(rc->block_group->pinned > 0);
+	WARN_ON(rc->block_group->reserved > 0);
+	WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
+out:
+	iput(rc->data_inode);
+	btrfs_stop_workers(&rc->workers);
+	btrfs_put_block_group(rc->block_group);
+	kfree(rc);
+	return err;
+}
+
+/*
+ * recover relocation interrupted by system crash.
+ *
+ * this function resumes merging reloc trees with corresponding fs trees.
+ * this is important for keeping the sharing of tree blocks
+ */
+int btrfs_recover_relocation(struct btrfs_root *root)
+{
+	LIST_HEAD(reloc_roots);
+	struct btrfs_key key;
+	struct btrfs_root *fs_root;
+	struct btrfs_root *reloc_root;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct reloc_control *rc = NULL;
+	struct btrfs_trans_handle *trans;
+	int ret;
+	int err = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = BTRFS_TREE_RELOC_OBJECTID;
+	key.type = BTRFS_ROOT_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
+					path, 0, 0);
+		if (ret < 0) {
+			err = ret;
+			goto out;
+		}
+		if (ret > 0) {
+			if (path->slots[0] == 0)
+				break;
+			path->slots[0]--;
+		}
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		btrfs_release_path(root->fs_info->tree_root, path);
+
+		if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
+		    key.type != BTRFS_ROOT_ITEM_KEY)
+			break;
+
+		reloc_root = btrfs_read_fs_root_no_radix(root, &key);
+		if (IS_ERR(reloc_root)) {
+			err = PTR_ERR(reloc_root);
+			goto out;
+		}
+
+		list_add(&reloc_root->root_list, &reloc_roots);
+
+		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
+			fs_root = read_fs_root(root->fs_info,
+					       reloc_root->root_key.offset);
+			if (IS_ERR(fs_root)) {
+				err = PTR_ERR(fs_root);
+				goto out;
+			}
+		}
+
+		if (key.offset == 0)
+			break;
+
+		key.offset--;
+	}
+	btrfs_release_path(root->fs_info->tree_root, path);
+
+	if (list_empty(&reloc_roots))
+		goto out;
+
+	rc = kzalloc(sizeof(*rc), GFP_NOFS);
+	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);
+	rc->extent_root = root->fs_info->extent_root;
+
+	set_reloc_control(rc);
+
+	while (!list_empty(&reloc_roots)) {
+		reloc_root = list_entry(reloc_roots.next,
+					struct btrfs_root, root_list);
+		list_del(&reloc_root->root_list);
+
+		if (btrfs_root_refs(&reloc_root->root_item) == 0) {
+			list_add_tail(&reloc_root->root_list,
+				      &rc->reloc_roots);
+			continue;
+		}
+
+		fs_root = read_fs_root(root->fs_info,
+				       reloc_root->root_key.offset);
+		BUG_ON(IS_ERR(fs_root));
+
+		__add_reloc_root(reloc_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);
+	btrfs_commit_transaction(trans, rc->extent_root);
+out:
+	if (rc) {
+		btrfs_stop_workers(&rc->workers);
+		kfree(rc);
+	}
+	while (!list_empty(&reloc_roots)) {
+		reloc_root = list_entry(reloc_roots.next,
+					struct btrfs_root, root_list);
+		list_del(&reloc_root->root_list);
+		free_extent_buffer(reloc_root->node);
+		free_extent_buffer(reloc_root->commit_root);
+		kfree(reloc_root);
+	}
+	btrfs_free_path(path);
+
+	if (err == 0) {
+		/* cleanup orphan inode in data relocation tree */
+		fs_root = read_fs_root(root->fs_info,
+				       BTRFS_DATA_RELOC_TREE_OBJECTID);
+		if (IS_ERR(fs_root))
+			err = PTR_ERR(fs_root);
+	}
+	return err;
+}
+
+/*
+ * helper to add ordered checksum for data relocation.
+ *
+ * cloning checksum properly handles the nodatasum extents.
+ * it also saves CPU time to re-calculate the checksum.
+ */
+int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
+{
+	struct btrfs_ordered_sum *sums;
+	struct btrfs_sector_sum *sector_sum;
+	struct btrfs_ordered_extent *ordered;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	size_t offset;
+	int ret;
+	u64 disk_bytenr;
+	LIST_HEAD(list);
+
+	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
+	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
+
+	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
+	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
+				       disk_bytenr + len - 1, &list);
+
+	while (!list_empty(&list)) {
+		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
+		list_del_init(&sums->list);
+
+		sector_sum = sums->sums;
+		sums->bytenr = ordered->start;
+
+		offset = 0;
+		while (offset < sums->len) {
+			sector_sum->bytenr += ordered->start - disk_bytenr;
+			sector_sum++;
+			offset += root->sectorsize;
+		}
+
+		btrfs_add_ordered_sum(inode, ordered, sums);
+	}
+	btrfs_put_ordered_extent(ordered);
+	return 0;
+}

fs/btrfs/root-tree.c

@@ -111,6 +111,15 @@ out:
 	return ret;
 }
 
+int btrfs_set_root_node(struct btrfs_root_item *item,
+			struct extent_buffer *node)
+{
+	btrfs_set_root_bytenr(item, node->start);
+	btrfs_set_root_level(item, btrfs_header_level(node));
+	btrfs_set_root_generation(item, btrfs_header_generation(node));
+	return 0;
+}
+
 /*
  * copy the data in 'item' into the btree
  */
@@ -164,8 +173,7 @@ int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root
  * offset lower than the latest root.  They need to be queued for deletion to
  * finish what was happening when we crashed.
  */
-int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid,
-			  struct btrfs_root *latest)
+int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid)
 {
 	struct btrfs_root *dead_root;
 	struct btrfs_item *item;
@@ -227,10 +235,7 @@ again:
 			goto err;
 		}
 
-		if (objectid == BTRFS_TREE_RELOC_OBJECTID)
-			ret = btrfs_add_dead_reloc_root(dead_root);
-		else
-			ret = btrfs_add_dead_root(dead_root, latest);
+		ret = btrfs_add_dead_root(dead_root);
 		if (ret)
 			goto err;
 		goto again;

fs/btrfs/super.c

@@ -52,7 +52,6 @@
 #include "export.h"
 #include "compression.h"
 
-
 static struct super_operations btrfs_super_ops;
 
 static void btrfs_put_super(struct super_block *sb)
@@ -67,8 +66,8 @@ static void btrfs_put_super(struct super_block *sb)
 enum {
 	Opt_degraded, Opt_subvol, Opt_device, Opt_nodatasum, Opt_nodatacow,
 	Opt_max_extent, Opt_max_inline, Opt_alloc_start, Opt_nobarrier,
-	Opt_ssd, Opt_thread_pool, Opt_noacl,  Opt_compress, Opt_notreelog,
-	Opt_ratio, Opt_flushoncommit, Opt_err,
+	Opt_ssd, Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl,
+	Opt_compress, Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_err,
 };
 
 static match_table_t tokens = {
@@ -84,6 +83,8 @@ static match_table_t tokens = {
 	{Opt_thread_pool, "thread_pool=%d"},
 	{Opt_compress, "compress"},
 	{Opt_ssd, "ssd"},
+	{Opt_ssd_spread, "ssd_spread"},
+	{Opt_nossd, "nossd"},
 	{Opt_noacl, "noacl"},
 	{Opt_notreelog, "notreelog"},
 	{Opt_flushoncommit, "flushoncommit"},
@@ -158,7 +159,7 @@ int btrfs_parse_options(struct btrfs_root *root, char *options)
 			 */
 			break;
 		case Opt_nodatasum:
-			printk(KERN_INFO "btrfs: setting nodatacsum\n");
+			printk(KERN_INFO "btrfs: setting nodatasum\n");
 			btrfs_set_opt(info->mount_opt, NODATASUM);
 			break;
 		case Opt_nodatacow:
@@ -174,6 +175,19 @@ int btrfs_parse_options(struct btrfs_root *root, char *options)
 			printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 			btrfs_set_opt(info->mount_opt, SSD);
 			break;
+		case Opt_ssd_spread:
+			printk(KERN_INFO "btrfs: use spread ssd "
+			       "allocation scheme\n");
+			btrfs_set_opt(info->mount_opt, SSD);
+			btrfs_set_opt(info->mount_opt, SSD_SPREAD);
+			break;
+		case Opt_nossd:
+			printk(KERN_INFO "btrfs: not using ssd allocation "
+			       "scheme\n");
+			btrfs_set_opt(info->mount_opt, NOSSD);
+			btrfs_clear_opt(info->mount_opt, SSD);
+			btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
+			break;
 		case Opt_nobarrier:
 			printk(KERN_INFO "btrfs: turning off barriers\n");
 			btrfs_set_opt(info->mount_opt, NOBARRIER);
@@ -322,7 +336,7 @@ static int btrfs_fill_super(struct super_block *sb,
 	struct dentry *root_dentry;
 	struct btrfs_super_block *disk_super;
 	struct btrfs_root *tree_root;
-	struct btrfs_inode *bi;
+	struct btrfs_key key;
 	int err;
 
 	sb->s_maxbytes = MAX_LFS_FILESIZE;
@@ -341,23 +355,15 @@ static int btrfs_fill_super(struct super_block *sb,
 	}
 	sb->s_fs_info = tree_root;
 	disk_super = &tree_root->fs_info->super_copy;
-	inode = btrfs_iget_locked(sb, BTRFS_FIRST_FREE_OBJECTID,
-				  tree_root->fs_info->fs_root);
-	bi = BTRFS_I(inode);
-	bi->location.objectid = inode->i_ino;
-	bi->location.offset = 0;
-	bi->root = tree_root->fs_info->fs_root;
 
-	btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
-
-	if (!inode) {
-		err = -ENOMEM;
+	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
+	key.type = BTRFS_INODE_ITEM_KEY;
+	key.offset = 0;
+	inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
 		goto fail_close;
 	}
-	if (inode->i_state & I_NEW) {
-		btrfs_read_locked_inode(inode);
-		unlock_new_inode(inode);
-	}
 
 	root_dentry = d_alloc_root(inode);
 	if (!root_dentry) {
@@ -433,7 +439,11 @@ static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 		seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 	if (btrfs_test_opt(root, COMPRESS))
 		seq_puts(seq, ",compress");
-	if (btrfs_test_opt(root, SSD))
+	if (btrfs_test_opt(root, NOSSD))
+		seq_puts(seq, ",nossd");
+	if (btrfs_test_opt(root, SSD_SPREAD))
+		seq_puts(seq, ",ssd_spread");
+	else if (btrfs_test_opt(root, SSD))
 		seq_puts(seq, ",ssd");
 	if (btrfs_test_opt(root, NOTREELOG))
 		seq_puts(seq, ",notreelog");
@@ -584,7 +594,8 @@ 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;
 
-		ret = btrfs_cleanup_reloc_trees(root);
+		/* recover relocation */
+		ret = btrfs_recover_relocation(root);
 		WARN_ON(ret);
 
 		ret = btrfs_cleanup_fs_roots(root->fs_info);

fs/btrfs/transaction.c

@@ -25,7 +25,6 @@
 #include "disk-io.h"
 #include "transaction.h"
 #include "locking.h"
-#include "ref-cache.h"
 #include "tree-log.h"
 
 #define BTRFS_ROOT_TRANS_TAG 0
@@ -94,45 +93,37 @@ static noinline int join_transaction(struct btrfs_root *root)
  * to make sure the old root from before we joined the transaction is deleted
  * when the transaction commits
  */
-noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
+static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
+					 struct btrfs_root *root)
 {
-	struct btrfs_dirty_root *dirty;
-	u64 running_trans_id = root->fs_info->running_transaction->transid;
-	if (root->ref_cows && root->last_trans < running_trans_id) {
+	if (root->ref_cows && root->last_trans < trans->transid) {
 		WARN_ON(root == root->fs_info->extent_root);
-		if (root->root_item.refs != 0) {
-			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
-				   (unsigned long)root->root_key.objectid,
-				   BTRFS_ROOT_TRANS_TAG);
-
-			dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
-			BUG_ON(!dirty);
-			dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
-			BUG_ON(!dirty->root);
-			dirty->latest_root = root;
-			INIT_LIST_HEAD(&dirty->list);
-
-			root->commit_root = btrfs_root_node(root);
-
-			memcpy(dirty->root, root, sizeof(*root));
-			spin_lock_init(&dirty->root->node_lock);
-			spin_lock_init(&dirty->root->list_lock);
-			mutex_init(&dirty->root->objectid_mutex);
-			mutex_init(&dirty->root->log_mutex);
-			INIT_LIST_HEAD(&dirty->root->dead_list);
-			dirty->root->node = root->commit_root;
-			dirty->root->commit_root = NULL;
+		WARN_ON(root->root_item.refs == 0);
+		WARN_ON(root->commit_root != root->node);
+
+		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+			   (unsigned long)root->root_key.objectid,
+			   BTRFS_ROOT_TRANS_TAG);
+		root->last_trans = trans->transid;
+		btrfs_init_reloc_root(trans, root);
+	}
+	return 0;
+}
 
-			spin_lock(&root->list_lock);
-			list_add(&dirty->root->dead_list, &root->dead_list);
-			spin_unlock(&root->list_lock);
+int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	if (!root->ref_cows)
+		return 0;
 
-			root->dirty_root = dirty;
-		} else {
-			WARN_ON(1);
-		}
-		root->last_trans = running_trans_id;
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (root->last_trans == trans->transid) {
+		mutex_unlock(&root->fs_info->trans_mutex);
+		return 0;
 	}
+
+	record_root_in_trans(trans, root);
+	mutex_unlock(&root->fs_info->trans_mutex);
 	return 0;
 }
 
@@ -181,7 +172,6 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 	ret = join_transaction(root);
 	BUG_ON(ret);
 
-	btrfs_record_root_in_trans(root);
 	h->transid = root->fs_info->running_transaction->transid;
 	h->transaction = root->fs_info->running_transaction;
 	h->blocks_reserved = num_blocks;
@@ -192,6 +182,7 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 	h->delayed_ref_updates = 0;
 
 	root->fs_info->running_transaction->use_count++;
+	record_root_in_trans(h, root);
 	mutex_unlock(&root->fs_info->trans_mutex);
 	return h;
 }
@@ -233,6 +224,7 @@ static noinline int wait_for_commit(struct btrfs_root *root,
 	return 0;
 }
 
+#if 0
 /*
  * rate limit against the drop_snapshot code.  This helps to slow down new
  * operations if the drop_snapshot code isn't able to keep up.
@@ -273,6 +265,7 @@ harder:
 			goto harder;
 	}
 }
+#endif
 
 void btrfs_throttle(struct btrfs_root *root)
 {
@@ -280,7 +273,6 @@ void btrfs_throttle(struct btrfs_root *root)
 	if (!root->fs_info->open_ioctl_trans)
 		wait_current_trans(root);
 	mutex_unlock(&root->fs_info->trans_mutex);
-	throttle_on_drops(root);
 }
 
 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
@@ -323,9 +315,6 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 	memset(trans, 0, sizeof(*trans));
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
 
-	if (throttle)
-		throttle_on_drops(root);
-
 	return 0;
 }
 
@@ -462,12 +451,8 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 		if (old_root_bytenr == root->node->start)
 			break;
-		btrfs_set_root_bytenr(&root->root_item,
-				       root->node->start);
-		btrfs_set_root_level(&root->root_item,
-				     btrfs_header_level(root->node));
-		btrfs_set_root_generation(&root->root_item, trans->transid);
 
+		btrfs_set_root_node(&root->root_item, root->node);
 		ret = btrfs_update_root(trans, tree_root,
 					&root->root_key,
 					&root->root_item);
@@ -477,14 +462,16 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
 		ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 		BUG_ON(ret);
 	}
+	free_extent_buffer(root->commit_root);
+	root->commit_root = btrfs_root_node(root);
 	return 0;
 }
 
 /*
  * update all the cowonly tree roots on disk
  */
-int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root)
+static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
+					 struct btrfs_root *root)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct list_head *next;
@@ -520,118 +507,54 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
  * a dirty root struct and adds it into the list of dead roots that need to
  * be deleted
  */
-int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
+int btrfs_add_dead_root(struct btrfs_root *root)
 {
-	struct btrfs_dirty_root *dirty;
-
-	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
-	if (!dirty)
-		return -ENOMEM;
-	dirty->root = root;
-	dirty->latest_root = latest;
-
 	mutex_lock(&root->fs_info->trans_mutex);
-	list_add(&dirty->list, &latest->fs_info->dead_roots);
+	list_add(&root->root_list, &root->fs_info->dead_roots);
 	mutex_unlock(&root->fs_info->trans_mutex);
 	return 0;
 }
 
 /*
- * at transaction commit time we need to schedule the old roots for
- * deletion via btrfs_drop_snapshot.  This runs through all the
- * reference counted roots that were modified in the current
- * transaction and puts them into the drop list
+ * update all the cowonly tree roots on disk
  */
-static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
-				    struct radix_tree_root *radix,
-				    struct list_head *list)
+static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root)
 {
-	struct btrfs_dirty_root *dirty;
 	struct btrfs_root *gang[8];
-	struct btrfs_root *root;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 	int i;
 	int ret;
 	int err = 0;
-	u32 refs;
 
 	while (1) {
-		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
+		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
+						 (void **)gang, 0,
 						 ARRAY_SIZE(gang),
 						 BTRFS_ROOT_TRANS_TAG);
 		if (ret == 0)
 			break;
 		for (i = 0; i < ret; i++) {
 			root = gang[i];
-			radix_tree_tag_clear(radix,
-				     (unsigned long)root->root_key.objectid,
-				     BTRFS_ROOT_TRANS_TAG);
-
-			BUG_ON(!root->ref_tree);
-			dirty = root->dirty_root;
+			radix_tree_tag_clear(&fs_info->fs_roots_radix,
+					(unsigned long)root->root_key.objectid,
+					BTRFS_ROOT_TRANS_TAG);
 
 			btrfs_free_log(trans, root);
-			btrfs_free_reloc_root(trans, root);
-
-			if (root->commit_root == root->node) {
-				WARN_ON(root->node->start !=
-					btrfs_root_bytenr(&root->root_item));
-
-				free_extent_buffer(root->commit_root);
-				root->commit_root = NULL;
-				root->dirty_root = NULL;
-
-				spin_lock(&root->list_lock);
-				list_del_init(&dirty->root->dead_list);
-				spin_unlock(&root->list_lock);
+			btrfs_update_reloc_root(trans, root);
 
-				kfree(dirty->root);
-				kfree(dirty);
-
-				/* make sure to update the root on disk
-				 * so we get any updates to the block used
-				 * counts
-				 */
-				err = btrfs_update_root(trans,
-						root->fs_info->tree_root,
-						&root->root_key,
-						&root->root_item);
+			if (root->commit_root == root->node)
 				continue;
-			}
 
-			memset(&root->root_item.drop_progress, 0,
-			       sizeof(struct btrfs_disk_key));
-			root->root_item.drop_level = 0;
-			root->commit_root = NULL;
-			root->dirty_root = NULL;
-			root->root_key.offset = root->fs_info->generation;
-			btrfs_set_root_bytenr(&root->root_item,
-					      root->node->start);
-			btrfs_set_root_level(&root->root_item,
-					     btrfs_header_level(root->node));
-			btrfs_set_root_generation(&root->root_item,
-						  root->root_key.offset);
-
-			err = btrfs_insert_root(trans, root->fs_info->tree_root,
+			free_extent_buffer(root->commit_root);
+			root->commit_root = btrfs_root_node(root);
+
+			btrfs_set_root_node(&root->root_item, root->node);
+			err = btrfs_update_root(trans, fs_info->tree_root,
 						&root->root_key,
 						&root->root_item);
 			if (err)
 				break;
-
-			refs = btrfs_root_refs(&dirty->root->root_item);
-			btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
-			err = btrfs_update_root(trans, root->fs_info->tree_root,
-						&dirty->root->root_key,
-						&dirty->root->root_item);
-
-			BUG_ON(err);
-			if (refs == 1) {
-				list_add(&dirty->list, list);
-			} else {
-				WARN_ON(1);
-				free_extent_buffer(dirty->root->node);
-				kfree(dirty->root);
-				kfree(dirty);
-			}
 		}
 	}
 	return err;
@@ -688,12 +611,8 @@ static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
 				TASK_UNINTERRUPTIBLE);
 		mutex_unlock(&info->trans_mutex);
 
-		atomic_dec(&info->throttles);
-		wake_up(&info->transaction_throttle);
-
 		schedule();
 
-		atomic_inc(&info->throttles);
 		mutex_lock(&info->trans_mutex);
 		finish_wait(&info->transaction_wait, &wait);
 	}
@@ -705,111 +624,61 @@ static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
  * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
  * all of them
  */
-static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
-				     struct list_head *list)
+int btrfs_drop_dead_root(struct btrfs_root *root)
 {
-	struct btrfs_dirty_root *dirty;
 	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = root->fs_info->tree_root;
 	unsigned long nr;
-	u64 num_bytes;
-	u64 bytes_used;
-	u64 max_useless;
-	int ret = 0;
-	int err;
-
-	while (!list_empty(list)) {
-		struct btrfs_root *root;
-
-		dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
-		list_del_init(&dirty->list);
-
-		num_bytes = btrfs_root_used(&dirty->root->root_item);
-		root = dirty->latest_root;
-		atomic_inc(&root->fs_info->throttles);
-
-		while (1) {
-			/*
-			 * we don't want to jump in and create a bunch of
-			 * delayed refs if the transaction is starting to close
-			 */
-			wait_transaction_pre_flush(tree_root->fs_info);
-			trans = btrfs_start_transaction(tree_root, 1);
-
-			/*
-			 * we've joined a transaction, make sure it isn't
-			 * closing right now
-			 */
-			if (trans->transaction->delayed_refs.flushing) {
-				btrfs_end_transaction(trans, tree_root);
-				continue;
-			}
-
-			mutex_lock(&root->fs_info->drop_mutex);
-			ret = btrfs_drop_snapshot(trans, dirty->root);
-			if (ret != -EAGAIN)
-				break;
-			mutex_unlock(&root->fs_info->drop_mutex);
+	int ret;
 
-			err = btrfs_update_root(trans,
-					tree_root,
-					&dirty->root->root_key,
-					&dirty->root->root_item);
-			if (err)
-				ret = err;
-			nr = trans->blocks_used;
-			ret = btrfs_end_transaction(trans, tree_root);
-			BUG_ON(ret);
+	while (1) {
+		/*
+		 * we don't want to jump in and create a bunch of
+		 * delayed refs if the transaction is starting to close
+		 */
+		wait_transaction_pre_flush(tree_root->fs_info);
+		trans = btrfs_start_transaction(tree_root, 1);
 
-			btrfs_btree_balance_dirty(tree_root, nr);
-			cond_resched();
+		/*
+		 * we've joined a transaction, make sure it isn't
+		 * closing right now
+		 */
+		if (trans->transaction->delayed_refs.flushing) {
+			btrfs_end_transaction(trans, tree_root);
+			continue;
 		}
-		BUG_ON(ret);
-		atomic_dec(&root->fs_info->throttles);
-		wake_up(&root->fs_info->transaction_throttle);
 
-		num_bytes -= btrfs_root_used(&dirty->root->root_item);
-		bytes_used = btrfs_root_used(&root->root_item);
-		if (num_bytes) {
-			mutex_lock(&root->fs_info->trans_mutex);
-			btrfs_record_root_in_trans(root);
-			mutex_unlock(&root->fs_info->trans_mutex);
-			btrfs_set_root_used(&root->root_item,
-					    bytes_used - num_bytes);
-		}
+		ret = btrfs_drop_snapshot(trans, root);
+		if (ret != -EAGAIN)
+			break;
 
-		ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
-		if (ret) {
-			BUG();
+		ret = btrfs_update_root(trans, tree_root,
+					&root->root_key,
+					&root->root_item);
+		if (ret)
 			break;
-		}
-		mutex_unlock(&root->fs_info->drop_mutex);
-
-		spin_lock(&root->list_lock);
-		list_del_init(&dirty->root->dead_list);
-		if (!list_empty(&root->dead_list)) {
-			struct btrfs_root *oldest;
-			oldest = list_entry(root->dead_list.prev,
-					    struct btrfs_root, dead_list);
-			max_useless = oldest->root_key.offset - 1;
-		} else {
-			max_useless = root->root_key.offset - 1;
-		}
-		spin_unlock(&root->list_lock);
 
 		nr = trans->blocks_used;
 		ret = btrfs_end_transaction(trans, tree_root);
 		BUG_ON(ret);
 
-		ret = btrfs_remove_leaf_refs(root, max_useless, 0);
-		BUG_ON(ret);
-
-		free_extent_buffer(dirty->root->node);
-		kfree(dirty->root);
-		kfree(dirty);
-
 		btrfs_btree_balance_dirty(tree_root, nr);
 		cond_resched();
 	}
+	BUG_ON(ret);
+
+	ret = btrfs_del_root(trans, tree_root, &root->root_key);
+	BUG_ON(ret);
+
+	nr = trans->blocks_used;
+	ret = btrfs_end_transaction(trans, tree_root);
+	BUG_ON(ret);
+
+	free_extent_buffer(root->node);
+	free_extent_buffer(root->commit_root);
+	kfree(root);
+
+	btrfs_btree_balance_dirty(tree_root, nr);
 	return ret;
 }
 
@@ -839,24 +708,23 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	if (ret)
 		goto fail;
 
-	btrfs_record_root_in_trans(root);
+	record_root_in_trans(trans, root);
 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 
 	key.objectid = objectid;
-	key.offset = trans->transid;
+	key.offset = 0;
 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 
 	old = btrfs_lock_root_node(root);
 	btrfs_cow_block(trans, root, old, NULL, 0, &old);
+	btrfs_set_lock_blocking(old);
 
 	btrfs_copy_root(trans, root, old, &tmp, objectid);
 	btrfs_tree_unlock(old);
 	free_extent_buffer(old);
 
-	btrfs_set_root_bytenr(new_root_item, tmp->start);
-	btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
-	btrfs_set_root_generation(new_root_item, trans->transid);
+	btrfs_set_root_node(new_root_item, tmp);
 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 				new_root_item);
 	btrfs_tree_unlock(tmp);
@@ -964,6 +832,24 @@ static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+static void update_super_roots(struct btrfs_root *root)
+{
+	struct btrfs_root_item *root_item;
+	struct btrfs_super_block *super;
+
+	super = &root->fs_info->super_copy;
+
+	root_item = &root->fs_info->chunk_root->root_item;
+	super->chunk_root = root_item->bytenr;
+	super->chunk_root_generation = root_item->generation;
+	super->chunk_root_level = root_item->level;
+
+	root_item = &root->fs_info->tree_root->root_item;
+	super->root = root_item->bytenr;
+	super->generation = root_item->generation;
+	super->root_level = root_item->level;
+}
+
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root)
 {
@@ -971,8 +857,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	unsigned long timeout = 1;
 	struct btrfs_transaction *cur_trans;
 	struct btrfs_transaction *prev_trans = NULL;
-	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
-	struct list_head dirty_fs_roots;
 	struct extent_io_tree *pinned_copy;
 	DEFINE_WAIT(wait);
 	int ret;
@@ -999,7 +883,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	BUG_ON(ret);
 
 	mutex_lock(&root->fs_info->trans_mutex);
-	INIT_LIST_HEAD(&dirty_fs_roots);
 	if (cur_trans->in_commit) {
 		cur_trans->use_count++;
 		mutex_unlock(&root->fs_info->trans_mutex);
@@ -1105,41 +988,36 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	 * with the tree-log code.
 	 */
 	mutex_lock(&root->fs_info->tree_log_mutex);
-	/*
-	 * keep tree reloc code from adding new reloc trees
-	 */
-	mutex_lock(&root->fs_info->tree_reloc_mutex);
-
 
-	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
-			      &dirty_fs_roots);
+	ret = commit_fs_roots(trans, root);
 	BUG_ON(ret);
 
-	/* add_dirty_roots gets rid of all the tree log roots, it is now
+	/* commit_fs_roots gets rid of all the tree log roots, it is now
 	 * safe to free the root of tree log roots
 	 */
 	btrfs_free_log_root_tree(trans, root->fs_info);
 
-	ret = btrfs_commit_tree_roots(trans, root);
+	ret = commit_cowonly_roots(trans, root);
 	BUG_ON(ret);
 
 	cur_trans = root->fs_info->running_transaction;
 	spin_lock(&root->fs_info->new_trans_lock);
 	root->fs_info->running_transaction = NULL;
 	spin_unlock(&root->fs_info->new_trans_lock);
-	btrfs_set_super_generation(&root->fs_info->super_copy,
-				   cur_trans->transid);
-	btrfs_set_super_root(&root->fs_info->super_copy,
-			     root->fs_info->tree_root->node->start);
-	btrfs_set_super_root_level(&root->fs_info->super_copy,
-			   btrfs_header_level(root->fs_info->tree_root->node));
-
-	btrfs_set_super_chunk_root(&root->fs_info->super_copy,
-				   chunk_root->node->start);
-	btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
-					 btrfs_header_level(chunk_root->node));
-	btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
-				btrfs_header_generation(chunk_root->node));
+
+	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
+			    root->fs_info->tree_root->node);
+	free_extent_buffer(root->fs_info->tree_root->commit_root);
+	root->fs_info->tree_root->commit_root =
+				btrfs_root_node(root->fs_info->tree_root);
+
+	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
+			    root->fs_info->chunk_root->node);
+	free_extent_buffer(root->fs_info->chunk_root->commit_root);
+	root->fs_info->chunk_root->commit_root =
+				btrfs_root_node(root->fs_info->chunk_root);
+
+	update_super_roots(root);
 
 	if (!root->fs_info->log_root_recovering) {
 		btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
@@ -1153,7 +1031,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
 	trans->transaction->blocked = 0;
 
-	wake_up(&root->fs_info->transaction_throttle);
 	wake_up(&root->fs_info->transaction_wait);
 
 	mutex_unlock(&root->fs_info->trans_mutex);
@@ -1170,9 +1047,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	btrfs_finish_extent_commit(trans, root, pinned_copy);
 	kfree(pinned_copy);
 
-	btrfs_drop_dead_reloc_roots(root);
-	mutex_unlock(&root->fs_info->tree_reloc_mutex);
-
 	/* do the directory inserts of any pending snapshot creations */
 	finish_pending_snapshots(trans, root->fs_info);
 
@@ -1186,16 +1060,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	put_transaction(cur_trans);
 	put_transaction(cur_trans);
 
-	list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
-	if (root->fs_info->closing)
-		list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
-
 	mutex_unlock(&root->fs_info->trans_mutex);
 
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
-
-	if (root->fs_info->closing)
-		drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 	return ret;
 }
 
@@ -1204,16 +1071,17 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
  */
 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 {
-	struct list_head dirty_roots;
-	INIT_LIST_HEAD(&dirty_roots);
-again:
-	mutex_lock(&root->fs_info->trans_mutex);
-	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
-	mutex_unlock(&root->fs_info->trans_mutex);
+	LIST_HEAD(list);
+	struct btrfs_fs_info *fs_info = root->fs_info;
+
+	mutex_lock(&fs_info->trans_mutex);
+	list_splice_init(&fs_info->dead_roots, &list);
+	mutex_unlock(&fs_info->trans_mutex);
 
-	if (!list_empty(&dirty_roots)) {
-		drop_dirty_roots(root, &dirty_roots);
-		goto again;
+	while (!list_empty(&list)) {
+		root = list_entry(list.next, struct btrfs_root, root_list);
+		list_del_init(&root->root_list);
+		btrfs_drop_dead_root(root);
 	}
 	return 0;
 }

fs/btrfs/transaction.h

@@ -62,12 +62,6 @@ struct btrfs_pending_snapshot {
 	struct list_head list;
 };
 
-struct btrfs_dirty_root {
-	struct list_head list;
-	struct btrfs_root *root;
-	struct btrfs_root *latest_root;
-};
-
 static inline void btrfs_set_trans_block_group(struct btrfs_trans_handle *trans,
 					       struct inode *inode)
 {
@@ -100,7 +94,8 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 			    struct btrfs_root *root);
 
-int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest);
+int btrfs_add_dead_root(struct btrfs_root *root);
+int btrfs_drop_dead_root(struct btrfs_root *root);
 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly);
 int btrfs_clean_old_snapshots(struct btrfs_root *root);
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
@@ -108,7 +103,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 				   struct btrfs_root *root);
 void btrfs_throttle(struct btrfs_root *root);
-int btrfs_record_root_in_trans(struct btrfs_root *root);
+int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root);
 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 					struct extent_io_tree *dirty_pages);
 #endif

fs/btrfs/tree-log.c

@@ -430,18 +430,16 @@ no_copy:
 static noinline struct inode *read_one_inode(struct btrfs_root *root,
 					     u64 objectid)
 {
+	struct btrfs_key key;
 	struct inode *inode;
-	inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
-	if (inode->i_state & I_NEW) {
-		BTRFS_I(inode)->root = root;
-		BTRFS_I(inode)->location.objectid = objectid;
-		BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
-		BTRFS_I(inode)->location.offset = 0;
-		btrfs_read_locked_inode(inode);
-		unlock_new_inode(inode);
 
-	}
-	if (is_bad_inode(inode)) {
+	key.objectid = objectid;
+	key.type = BTRFS_INODE_ITEM_KEY;
+	key.offset = 0;
+	inode = btrfs_iget(root->fs_info->sb, &key, root);
+	if (IS_ERR(inode)) {
+		inode = NULL;
+	} else if (is_bad_inode(inode)) {
 		iput(inode);
 		inode = NULL;
 	}
@@ -541,6 +539,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 
 	if (found_type == BTRFS_FILE_EXTENT_REG ||
 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
+		u64 offset;
 		unsigned long dest_offset;
 		struct btrfs_key ins;
 
@@ -555,6 +554,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 		ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
 		ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
 		ins.type = BTRFS_EXTENT_ITEM_KEY;
+		offset = key->offset - btrfs_file_extent_offset(eb, item);
 
 		if (ins.objectid > 0) {
 			u64 csum_start;
@@ -569,19 +569,16 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 			if (ret == 0) {
 				ret = btrfs_inc_extent_ref(trans, root,
 						ins.objectid, ins.offset,
-						path->nodes[0]->start,
-						root->root_key.objectid,
-						trans->transid, key->objectid);
+						0, root->root_key.objectid,
+						key->objectid, offset);
 			} else {
 				/*
 				 * insert the extent pointer in the extent
 				 * allocation tree
 				 */
-				ret = btrfs_alloc_logged_extent(trans, root,
-						path->nodes[0]->start,
-						root->root_key.objectid,
-						trans->transid, key->objectid,
-						&ins);
+				ret = btrfs_alloc_logged_file_extent(trans,
+						root, root->root_key.objectid,
+						key->objectid, offset, &ins);
 				BUG_ON(ret);
 			}
 			btrfs_release_path(root, path);
@@ -1706,9 +1703,6 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
 				btrfs_wait_tree_block_writeback(next);
 				btrfs_tree_unlock(next);
 
-				ret = btrfs_drop_leaf_ref(trans, root, next);
-				BUG_ON(ret);
-
 				WARN_ON(root_owner !=
 					BTRFS_TREE_LOG_OBJECTID);
 				ret = btrfs_free_reserved_extent(root,
@@ -1753,10 +1747,6 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
 		btrfs_wait_tree_block_writeback(next);
 		btrfs_tree_unlock(next);
 
-		if (*level == 0) {
-			ret = btrfs_drop_leaf_ref(trans, root, next);
-			BUG_ON(ret);
-		}
 		WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
 		ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
 		BUG_ON(ret);
@@ -1811,12 +1801,6 @@ static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
 				btrfs_wait_tree_block_writeback(next);
 				btrfs_tree_unlock(next);
 
-				if (*level == 0) {
-					ret = btrfs_drop_leaf_ref(trans, root,
-								  next);
-					BUG_ON(ret);
-				}
-
 				WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
 				ret = btrfs_free_reserved_extent(root,
 						path->nodes[*level]->start,
@@ -1884,11 +1868,6 @@ static int walk_log_tree(struct btrfs_trans_handle *trans,
 			btrfs_wait_tree_block_writeback(next);
 			btrfs_tree_unlock(next);
 
-			if (orig_level == 0) {
-				ret = btrfs_drop_leaf_ref(trans, log,
-							  next);
-				BUG_ON(ret);
-			}
 			WARN_ON(log->root_key.objectid !=
 				BTRFS_TREE_LOG_OBJECTID);
 			ret = btrfs_free_reserved_extent(log, next->start,
@@ -2027,9 +2006,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
 	BUG_ON(ret);
 
-	btrfs_set_root_bytenr(&log->root_item, log->node->start);
-	btrfs_set_root_generation(&log->root_item, trans->transid);
-	btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
+	btrfs_set_root_node(&log->root_item, log->node);
 
 	root->log_batch = 0;
 	root->log_transid++;
@@ -2581,7 +2558,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 				       ins_keys, ins_sizes, nr);
 	BUG_ON(ret);
 
-	for (i = 0; i < nr; i++) {
+	for (i = 0; i < nr; i++, dst_path->slots[0]++) {
 		dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
 						   dst_path->slots[0]);
 
@@ -2617,36 +2594,31 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 			found_type = btrfs_file_extent_type(src, extent);
 			if (found_type == BTRFS_FILE_EXTENT_REG ||
 			    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
-				u64 ds = btrfs_file_extent_disk_bytenr(src,
-								   extent);
-				u64 dl = btrfs_file_extent_disk_num_bytes(src,
-								      extent);
-				u64 cs = btrfs_file_extent_offset(src, extent);
-				u64 cl = btrfs_file_extent_num_bytes(src,
-								     extent);;
+				u64 ds, dl, cs, cl;
+				ds = btrfs_file_extent_disk_bytenr(src,
+								extent);
+				/* ds == 0 is a hole */
+				if (ds == 0)
+					continue;
+
+				dl = btrfs_file_extent_disk_num_bytes(src,
+								extent);
+				cs = btrfs_file_extent_offset(src, extent);
+				cl = btrfs_file_extent_num_bytes(src,
+								extent);;
 				if (btrfs_file_extent_compression(src,
 								  extent)) {
 					cs = 0;
 					cl = dl;
 				}
-				/* ds == 0 is a hole */
-				if (ds != 0) {
-					ret = btrfs_inc_extent_ref(trans, log,
-						   ds, dl,
-						   dst_path->nodes[0]->start,
-						   BTRFS_TREE_LOG_OBJECTID,
-						   trans->transid,
-						   ins_keys[i].objectid);
-					BUG_ON(ret);
-					ret = btrfs_lookup_csums_range(
-						   log->fs_info->csum_root,
-						   ds + cs, ds + cs + cl - 1,
-						   &ordered_sums);
-					BUG_ON(ret);
-				}
+
+				ret = btrfs_lookup_csums_range(
+						log->fs_info->csum_root,
+						ds + cs, ds + cs + cl - 1,
+						&ordered_sums);
+				BUG_ON(ret);
 			}
 		}
-		dst_path->slots[0]++;
 	}
 
 	btrfs_mark_buffer_dirty(dst_path->nodes[0]);
@@ -3029,9 +3001,7 @@ again:
 		BUG_ON(!wc.replay_dest);
 
 		wc.replay_dest->log_root = log;
-		mutex_lock(&fs_info->trans_mutex);
-		btrfs_record_root_in_trans(wc.replay_dest);
-		mutex_unlock(&fs_info->trans_mutex);
+		btrfs_record_root_in_trans(trans, wc.replay_dest);
 		ret = walk_log_tree(trans, log, &wc);
 		BUG_ON(ret);
 
@@ -3049,6 +3019,7 @@ again:
 		key.offset = found_key.offset - 1;
 		wc.replay_dest->log_root = NULL;
 		free_extent_buffer(log->node);
+		free_extent_buffer(log->commit_root);
 		kfree(log);
 
 		if (found_key.offset == 0)

fs/btrfs/volumes.c

@@ -161,8 +161,10 @@ static noinline int run_scheduled_bios(struct btrfs_device *device)
 	int again = 0;
 	unsigned long num_run;
 	unsigned long num_sync_run;
+	unsigned long batch_run = 0;
 	unsigned long limit;
 	unsigned long last_waited = 0;
+	int force_reg = 0;
 
 	bdi = blk_get_backing_dev_info(device->bdev);
 	fs_info = device->dev_root->fs_info;
@@ -176,19 +178,22 @@ static noinline int run_scheduled_bios(struct btrfs_device *device)
 
 loop:
 	spin_lock(&device->io_lock);
-	num_run = 0;
 
 loop_lock:
+	num_run = 0;
 
 	/* take all the bios off the list at once and process them
 	 * later on (without the lock held).  But, remember the
 	 * tail and other pointers so the bios can be properly reinserted
 	 * into the list if we hit congestion
 	 */
-	if (device->pending_sync_bios.head)
+	if (!force_reg && device->pending_sync_bios.head) {
 		pending_bios = &device->pending_sync_bios;
-	else
+		force_reg = 1;
+	} else {
 		pending_bios = &device->pending_bios;
+		force_reg = 0;
+	}
 
 	pending = pending_bios->head;
 	tail = pending_bios->tail;
@@ -228,10 +233,14 @@ loop_lock:
 	while (pending) {
 
 		rmb();
-		if (pending_bios != &device->pending_sync_bios &&
-		    device->pending_sync_bios.head &&
-		    num_run > 16) {
-			cond_resched();
+		/* we want to work on both lists, but do more bios on the
+		 * sync list than the regular list
+		 */
+		if ((num_run > 32 &&
+		    pending_bios != &device->pending_sync_bios &&
+		    device->pending_sync_bios.head) ||
+		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
+		    device->pending_bios.head)) {
 			spin_lock(&device->io_lock);
 			requeue_list(pending_bios, pending, tail);
 			goto loop_lock;
@@ -249,6 +258,8 @@ loop_lock:
 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 		submit_bio(cur->bi_rw, cur);
 		num_run++;
+		batch_run++;
+
 		if (bio_sync(cur))
 			num_sync_run++;
 
@@ -265,7 +276,7 @@ loop_lock:
 		 * is now congested.  Back off and let other work structs
 		 * run instead
 		 */
-		if (pending && bdi_write_congested(bdi) && num_run > 16 &&
+		if (pending && bdi_write_congested(bdi) && batch_run > 32 &&
 		    fs_info->fs_devices->open_devices > 1) {
 			struct io_context *ioc;
 
@@ -366,6 +377,7 @@ static noinline int device_list_add(const char *path,
 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 		fs_devices->latest_devid = devid;
 		fs_devices->latest_trans = found_transid;
+		mutex_init(&fs_devices->device_list_mutex);
 		device = NULL;
 	} else {
 		device = __find_device(&fs_devices->devices, devid,
@@ -392,7 +404,11 @@ static noinline int device_list_add(const char *path,
 			return -ENOMEM;
 		}
 		INIT_LIST_HEAD(&device->dev_alloc_list);
+
+		mutex_lock(&fs_devices->device_list_mutex);
 		list_add(&device->dev_list, &fs_devices->devices);
+		mutex_unlock(&fs_devices->device_list_mutex);
+
 		device->fs_devices = fs_devices;
 		fs_devices->num_devices++;
 	}
@@ -418,10 +434,12 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 	INIT_LIST_HEAD(&fs_devices->devices);
 	INIT_LIST_HEAD(&fs_devices->alloc_list);
 	INIT_LIST_HEAD(&fs_devices->list);
+	mutex_init(&fs_devices->device_list_mutex);
 	fs_devices->latest_devid = orig->latest_devid;
 	fs_devices->latest_trans = orig->latest_trans;
 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 
+	mutex_lock(&orig->device_list_mutex);
 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 		device = kzalloc(sizeof(*device), GFP_NOFS);
 		if (!device)
@@ -443,8 +461,10 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 		device->fs_devices = fs_devices;
 		fs_devices->num_devices++;
 	}
+	mutex_unlock(&orig->device_list_mutex);
 	return fs_devices;
 error:
+	mutex_unlock(&orig->device_list_mutex);
 	free_fs_devices(fs_devices);
 	return ERR_PTR(-ENOMEM);
 }
@@ -455,6 +475,7 @@ int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 
 	mutex_lock(&uuid_mutex);
 again:
+	mutex_lock(&fs_devices->device_list_mutex);
 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 		if (device->in_fs_metadata)
 			continue;
@@ -474,6 +495,7 @@ again:
 		kfree(device->name);
 		kfree(device);
 	}
+	mutex_unlock(&fs_devices->device_list_mutex);
 
 	if (fs_devices->seed) {
 		fs_devices = fs_devices->seed;
@@ -594,6 +616,9 @@ static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 		device->in_fs_metadata = 0;
 		device->mode = flags;
 
+		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
+			fs_devices->rotating = 1;
+
 		fs_devices->open_devices++;
 		if (device->writeable) {
 			fs_devices->rw_devices++;
@@ -1121,12 +1146,14 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 
 		device = NULL;
 		devices = &root->fs_info->fs_devices->devices;
+		mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 		list_for_each_entry(tmp, devices, dev_list) {
 			if (tmp->in_fs_metadata && !tmp->bdev) {
 				device = tmp;
 				break;
 			}
 		}
+		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 		bdev = NULL;
 		bh = NULL;
 		disk_super = NULL;
@@ -1181,7 +1208,16 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 		goto error_brelse;
 
 	device->in_fs_metadata = 0;
+
+	/*
+	 * the device list mutex makes sure that we don't change
+	 * the device list while someone else is writing out all
+	 * the device supers.
+	 */
+	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 	list_del_init(&device->dev_list);
+	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
+
 	device->fs_devices->num_devices--;
 
 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
@@ -1275,6 +1311,7 @@ static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
 	seed_devices->opened = 1;
 	INIT_LIST_HEAD(&seed_devices->devices);
 	INIT_LIST_HEAD(&seed_devices->alloc_list);
+	mutex_init(&seed_devices->device_list_mutex);
 	list_splice_init(&fs_devices->devices, &seed_devices->devices);
 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
@@ -1400,6 +1437,10 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	mutex_lock(&root->fs_info->volume_mutex);
 
 	devices = &root->fs_info->fs_devices->devices;
+	/*
+	 * we have the volume lock, so we don't need the extra
+	 * device list mutex while reading the list here.
+	 */
 	list_for_each_entry(device, devices, dev_list) {
 		if (device->bdev == bdev) {
 			ret = -EEXIST;
@@ -1454,6 +1495,12 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	}
 
 	device->fs_devices = root->fs_info->fs_devices;
+
+	/*
+	 * we don't want write_supers to jump in here with our device
+	 * half setup
+	 */
+	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 	list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
 	list_add(&device->dev_alloc_list,
 		 &root->fs_info->fs_devices->alloc_list);
@@ -1462,6 +1509,9 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	root->fs_info->fs_devices->rw_devices++;
 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
 
+	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
+		root->fs_info->fs_devices->rotating = 1;
+
 	total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 	btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 				    total_bytes + device->total_bytes);
@@ -1469,6 +1519,7 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 	btrfs_set_super_num_devices(&root->fs_info->super_copy,
 				    total_bytes + 1);
+	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 
 	if (seeding_dev) {
 		ret = init_first_rw_device(trans, root, device);
@@ -1671,8 +1722,6 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
 	int ret;
 	int i;
 
-	printk(KERN_INFO "btrfs relocating chunk %llu\n",
-	       (unsigned long long)chunk_offset);
 	root = root->fs_info->chunk_root;
 	extent_root = root->fs_info->extent_root;
 	em_tree = &root->fs_info->mapping_tree.map_tree;

fs/btrfs/volumes.h

@@ -96,7 +96,12 @@ struct btrfs_fs_devices {
 	u64 rw_devices;
 	u64 total_rw_bytes;
 	struct block_device *latest_bdev;
-	/* all of the devices in the FS */
+
+	/* all of the devices in the FS, protected by a mutex
+	 * so we can safely walk it to write out the supers without
+	 * worrying about add/remove by the multi-device code
+	 */
+	struct mutex device_list_mutex;
 	struct list_head devices;
 
 	/* devices not currently being allocated */
@@ -107,6 +112,11 @@ struct btrfs_fs_devices {
 	int seeding;
 
 	int opened;
+
+	/* set when we find or add a device that doesn't have the
+	 * nonrot flag set
+	 */
+	int rotating;
 };
 
 struct btrfs_bio_stripe {