28 files changed, 3250 insertions, 362 deletions

diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index 6d1d0b93b1aa..128ce17a80b0 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -9,11 +9,12 @@ btrfs-y += super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
 	   export.o tree-log.o free-space-cache.o zlib.o lzo.o \
 	   compression.o delayed-ref.o relocation.o delayed-inode.o scrub.o \
 	   reada.o backref.o ulist.o qgroup.o send.o dev-replace.o raid56.o \
-	   uuid-tree.o props.o hash.o
+	   uuid-tree.o props.o hash.o free-space-tree.o
 
 btrfs-$(CONFIG_BTRFS_FS_POSIX_ACL) += acl.o
 btrfs-$(CONFIG_BTRFS_FS_CHECK_INTEGRITY) += check-integrity.o
 
 btrfs-$(CONFIG_BTRFS_FS_RUN_SANITY_TESTS) += tests/free-space-tests.o \
 	tests/extent-buffer-tests.o tests/btrfs-tests.o \
-	tests/extent-io-tests.o tests/inode-tests.o tests/qgroup-tests.o
+	tests/extent-io-tests.o tests/inode-tests.o tests/qgroup-tests.o \
+	tests/free-space-tree-tests.o
diff --git a/fs/btrfs/acl.c b/fs/btrfs/acl.c
index 9a0124a95851..dbbb8ed53a51 100644
--- a/fs/btrfs/acl.c
+++ b/fs/btrfs/acl.c
@@ -48,7 +48,7 @@ struct posix_acl *btrfs_get_acl(struct inode *inode, int type)
 
 	size = __btrfs_getxattr(inode, name, "", 0);
 	if (size > 0) {
-		value = kzalloc(size, GFP_NOFS);
+		value = kzalloc(size, GFP_KERNEL);
 		if (!value)
 			return ERR_PTR(-ENOMEM);
 		size = __btrfs_getxattr(inode, name, value, size);
@@ -102,7 +102,7 @@ static int __btrfs_set_acl(struct btrfs_trans_handle *trans,
 
 	if (acl) {
 		size = posix_acl_xattr_size(acl->a_count);
-		value = kmalloc(size, GFP_NOFS);
+		value = kmalloc(size, GFP_KERNEL);
 		if (!value) {
 			ret = -ENOMEM;
 			goto out;
diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 3e36e4adc4a3..88d9af3d4581 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -97,7 +97,7 @@ static struct __btrfs_workqueue *
 __btrfs_alloc_workqueue(const char *name, unsigned int flags, int limit_active,
 			 int thresh)
 {
-	struct __btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_NOFS);
+	struct __btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_KERNEL);
 
 	if (!ret)
 		return NULL;
@@ -148,7 +148,7 @@ struct btrfs_workqueue *btrfs_alloc_workqueue(const char *name,
 					      int limit_active,
 					      int thresh)
 {
-	struct btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_NOFS);
+	struct btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_KERNEL);
 
 	if (!ret)
 		return NULL;
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 5e0fe0914d54..70c940582482 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -96,6 +96,9 @@ struct btrfs_ordered_sum;
 /* for storing items that use the BTRFS_UUID_KEY* types */
 #define BTRFS_UUID_TREE_OBJECTID 9ULL
 
+/* tracks free space in block groups. */
+#define BTRFS_FREE_SPACE_TREE_OBJECTID 10ULL
+
 /* for storing balance parameters in the root tree */
 #define BTRFS_BALANCE_OBJECTID -4ULL
 
@@ -500,6 +503,8 @@ 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_RO_FREE_SPACE_TREE	(1ULL << 0)
+
 #define BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF	(1ULL << 0)
 #define BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL	(1ULL << 1)
 #define BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS	(1ULL << 2)
@@ -526,7 +531,10 @@ struct btrfs_super_block {
 #define BTRFS_FEATURE_COMPAT_SUPP		0ULL
 #define BTRFS_FEATURE_COMPAT_SAFE_SET		0ULL
 #define BTRFS_FEATURE_COMPAT_SAFE_CLEAR		0ULL
-#define BTRFS_FEATURE_COMPAT_RO_SUPP		0ULL
+
+#define BTRFS_FEATURE_COMPAT_RO_SUPP			\
+	(BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)
+
 #define BTRFS_FEATURE_COMPAT_RO_SAFE_SET	0ULL
 #define BTRFS_FEATURE_COMPAT_RO_SAFE_CLEAR	0ULL
 
@@ -1089,6 +1097,13 @@ struct btrfs_block_group_item {
 	__le64 flags;
 } __attribute__ ((__packed__));
 
+struct btrfs_free_space_info {
+	__le32 extent_count;
+	__le32 flags;
+} __attribute__ ((__packed__));
+
+#define BTRFS_FREE_SPACE_USING_BITMAPS (1ULL << 0)
+
 #define BTRFS_QGROUP_LEVEL_SHIFT		48
 static inline u64 btrfs_qgroup_level(u64 qgroupid)
 {
@@ -1297,6 +1312,9 @@ struct btrfs_caching_control {
 	atomic_t count;
 };
 
+/* Once caching_thread() finds this much free space, it will wake up waiters. */
+#define CACHING_CTL_WAKE_UP (1024 * 1024 * 2)
+
 struct btrfs_io_ctl {
 	void *cur, *orig;
 	struct page *page;
@@ -1322,8 +1340,20 @@ struct btrfs_block_group_cache {
 	u64 delalloc_bytes;
 	u64 bytes_super;
 	u64 flags;
-	u64 sectorsize;
 	u64 cache_generation;
+	u32 sectorsize;
+
+	/*
+	 * If the free space extent count exceeds this number, convert the block
+	 * group to bitmaps.
+	 */
+	u32 bitmap_high_thresh;
+
+	/*
+	 * If the free space extent count drops below this number, convert the
+	 * block group back to extents.
+	 */
+	u32 bitmap_low_thresh;
 
 	/*
 	 * It is just used for the delayed data space allocation because
@@ -1379,6 +1409,15 @@ struct btrfs_block_group_cache {
 	struct list_head io_list;
 
 	struct btrfs_io_ctl io_ctl;
+
+	/* Lock for free space tree operations. */
+	struct mutex free_space_lock;
+
+	/*
+	 * Does the block group need to be added to the free space tree?
+	 * Protected by free_space_lock.
+	 */
+	int needs_free_space;
 };
 
 /* delayed seq elem */
@@ -1430,6 +1469,7 @@ struct btrfs_fs_info {
 	struct btrfs_root *csum_root;
 	struct btrfs_root *quota_root;
 	struct btrfs_root *uuid_root;
+	struct btrfs_root *free_space_root;
 
 	/* the log root tree is a directory of all the other log roots */
 	struct btrfs_root *log_root_tree;
@@ -1817,6 +1857,8 @@ struct btrfs_fs_info {
 	 * and will be latter freed. Protected by fs_info->chunk_mutex.
 	 */
 	struct list_head pinned_chunks;
+
+	int creating_free_space_tree;
 };
 
 struct btrfs_subvolume_writers {
@@ -2093,6 +2135,27 @@ struct btrfs_ioctl_defrag_range_args {
  */
 #define BTRFS_BLOCK_GROUP_ITEM_KEY 192
 
+/*
+ * Every block group is represented in the free space tree by a free space info
+ * item, which stores some accounting information. It is keyed on
+ * (block_group_start, FREE_SPACE_INFO, block_group_length).
+ */
+#define BTRFS_FREE_SPACE_INFO_KEY 198
+
+/*
+ * A free space extent tracks an extent of space that is free in a block group.
+ * It is keyed on (start, FREE_SPACE_EXTENT, length).
+ */
+#define BTRFS_FREE_SPACE_EXTENT_KEY 199
+
+/*
+ * When a block group becomes very fragmented, we convert it to use bitmaps
+ * instead of extents. A free space bitmap is keyed on
+ * (start, FREE_SPACE_BITMAP, length); the corresponding item is a bitmap with
+ * (length / sectorsize) bits.
+ */
+#define BTRFS_FREE_SPACE_BITMAP_KEY 200
+
 #define BTRFS_DEV_EXTENT_KEY	204
 #define BTRFS_DEV_ITEM_KEY	216
 #define BTRFS_CHUNK_ITEM_KEY	228
@@ -2185,6 +2248,7 @@ struct btrfs_ioctl_defrag_range_args {
 #define BTRFS_MOUNT_RESCAN_UUID_TREE	(1 << 23)
 #define BTRFS_MOUNT_FRAGMENT_DATA	(1 << 24)
 #define BTRFS_MOUNT_FRAGMENT_METADATA	(1 << 25)
+#define BTRFS_MOUNT_FREE_SPACE_TREE	(1 << 26)
 
 #define BTRFS_DEFAULT_COMMIT_INTERVAL	(30)
 #define BTRFS_DEFAULT_MAX_INLINE	(8192)
@@ -2507,6 +2571,11 @@ BTRFS_SETGET_FUNCS(disk_block_group_flags,
 BTRFS_SETGET_STACK_FUNCS(block_group_flags,
 			struct btrfs_block_group_item, flags, 64);
 
+/* struct btrfs_free_space_info */
+BTRFS_SETGET_FUNCS(free_space_extent_count, struct btrfs_free_space_info,
+		   extent_count, 32);
+BTRFS_SETGET_FUNCS(free_space_flags, struct btrfs_free_space_info, flags, 32);
+
 /* struct btrfs_inode_ref */
 BTRFS_SETGET_FUNCS(inode_ref_name_len, struct btrfs_inode_ref, name_len, 16);
 BTRFS_SETGET_FUNCS(inode_ref_index, struct btrfs_inode_ref, index, 64);
@@ -3574,6 +3643,9 @@ void btrfs_end_write_no_snapshoting(struct btrfs_root *root);
 void check_system_chunk(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
 			const u64 type);
+u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
+		       struct btrfs_fs_info *info, u64 start, u64 end);
+
 /* ctree.c */
 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 		     int level, int *slot);
@@ -3738,6 +3810,7 @@ static inline void free_fs_info(struct btrfs_fs_info *fs_info)
 	kfree(fs_info->csum_root);
 	kfree(fs_info->quota_root);
 	kfree(fs_info->uuid_root);
+	kfree(fs_info->free_space_root);
 	kfree(fs_info->super_copy);
 	kfree(fs_info->super_for_commit);
 	security_free_mnt_opts(&fs_info->security_opts);
@@ -3907,7 +3980,6 @@ void btrfs_extent_item_to_extent_map(struct inode *inode,
 /* inode.c */
 struct btrfs_delalloc_work {
 	struct inode *inode;
-	int wait;
 	int delay_iput;
 	struct completion completion;
 	struct list_head list;
@@ -3915,7 +3987,7 @@ struct btrfs_delalloc_work {
 };
 
 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
-						    int wait, int delay_iput);
+						    int delay_iput);
 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work);
 
 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
@@ -4248,6 +4320,30 @@ static inline void __btrfs_set_fs_incompat(struct btrfs_fs_info *fs_info,
 	}
 }
 
+#define btrfs_clear_fs_incompat(__fs_info, opt) \
+	__btrfs_clear_fs_incompat((__fs_info), BTRFS_FEATURE_INCOMPAT_##opt)
+
+static inline void __btrfs_clear_fs_incompat(struct btrfs_fs_info *fs_info,
+					     u64 flag)
+{
+	struct btrfs_super_block *disk_super;
+	u64 features;
+
+	disk_super = fs_info->super_copy;
+	features = btrfs_super_incompat_flags(disk_super);
+	if (features & flag) {
+		spin_lock(&fs_info->super_lock);
+		features = btrfs_super_incompat_flags(disk_super);
+		if (features & flag) {
+			features &= ~flag;
+			btrfs_set_super_incompat_flags(disk_super, features);
+			btrfs_info(fs_info, "clearing %llu feature flag",
+					 flag);
+		}
+		spin_unlock(&fs_info->super_lock);
+	}
+}
+
 #define btrfs_fs_incompat(fs_info, opt) \
 	__btrfs_fs_incompat((fs_info), BTRFS_FEATURE_INCOMPAT_##opt)
 
@@ -4258,6 +4354,64 @@ static inline int __btrfs_fs_incompat(struct btrfs_fs_info *fs_info, u64 flag)
 	return !!(btrfs_super_incompat_flags(disk_super) & flag);
 }
 
+#define btrfs_set_fs_compat_ro(__fs_info, opt) \
+	__btrfs_set_fs_compat_ro((__fs_info), BTRFS_FEATURE_COMPAT_RO_##opt)
+
+static inline void __btrfs_set_fs_compat_ro(struct btrfs_fs_info *fs_info,
+					    u64 flag)
+{
+	struct btrfs_super_block *disk_super;
+	u64 features;
+
+	disk_super = fs_info->super_copy;
+	features = btrfs_super_compat_ro_flags(disk_super);
+	if (!(features & flag)) {
+		spin_lock(&fs_info->super_lock);
+		features = btrfs_super_compat_ro_flags(disk_super);
+		if (!(features & flag)) {
+			features |= flag;
+			btrfs_set_super_compat_ro_flags(disk_super, features);
+			btrfs_info(fs_info, "setting %llu ro feature flag",
+				   flag);
+		}
+		spin_unlock(&fs_info->super_lock);
+	}
+}
+
+#define btrfs_clear_fs_compat_ro(__fs_info, opt) \
+	__btrfs_clear_fs_compat_ro((__fs_info), BTRFS_FEATURE_COMPAT_RO_##opt)
+
+static inline void __btrfs_clear_fs_compat_ro(struct btrfs_fs_info *fs_info,
+					      u64 flag)
+{
+	struct btrfs_super_block *disk_super;
+	u64 features;
+
+	disk_super = fs_info->super_copy;
+	features = btrfs_super_compat_ro_flags(disk_super);
+	if (features & flag) {
+		spin_lock(&fs_info->super_lock);
+		features = btrfs_super_compat_ro_flags(disk_super);
+		if (features & flag) {
+			features &= ~flag;
+			btrfs_set_super_compat_ro_flags(disk_super, features);
+			btrfs_info(fs_info, "clearing %llu ro feature flag",
+				   flag);
+		}
+		spin_unlock(&fs_info->super_lock);
+	}
+}
+
+#define btrfs_fs_compat_ro(fs_info, opt) \
+	__btrfs_fs_compat_ro((fs_info), BTRFS_FEATURE_COMPAT_RO_##opt)
+
+static inline int __btrfs_fs_compat_ro(struct btrfs_fs_info *fs_info, u64 flag)
+{
+	struct btrfs_super_block *disk_super;
+	disk_super = fs_info->super_copy;
+	return !!(btrfs_super_compat_ro_flags(disk_super) & flag);
+}
+
 /*
  * Call btrfs_abort_transaction as early as possible when an error condition is
  * detected, that way the exact line number is reported.
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 216f97765b56..c94457fde64d 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -42,6 +42,7 @@
 #include "locking.h"
 #include "tree-log.h"
 #include "free-space-cache.h"
+#include "free-space-tree.h"
 #include "inode-map.h"
 #include "check-integrity.h"
 #include "rcu-string.h"
@@ -362,7 +363,7 @@ static int verify_parent_transid(struct extent_io_tree *io_tree,
 	}
 
 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
-			 0, &cached_state);
+			 &cached_state);
 	if (extent_buffer_uptodate(eb) &&
 	    btrfs_header_generation(eb) == parent_transid) {
 		ret = 0;
@@ -1650,6 +1651,9 @@ struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
 		return fs_info->uuid_root ? fs_info->uuid_root :
 					    ERR_PTR(-ENOENT);
+	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
+		return fs_info->free_space_root ? fs_info->free_space_root :
+						  ERR_PTR(-ENOENT);
 again:
 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
 	if (root) {
@@ -2148,6 +2152,7 @@ static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
 	free_root_extent_buffers(info->uuid_root);
 	if (chunk_root)
 		free_root_extent_buffers(info->chunk_root);
+	free_root_extent_buffers(info->free_space_root);
 }
 
 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
@@ -2448,6 +2453,15 @@ static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
 		fs_info->uuid_root = root;
 	}
 
+	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
+		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
+		root = btrfs_read_tree_root(tree_root, &location);
+		if (IS_ERR(root))
+			return PTR_ERR(root);
+		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
+		fs_info->free_space_root = root;
+	}
+
 	return 0;
 }
 
@@ -3052,6 +3066,18 @@ retry_root_backup:
 	if (sb->s_flags & MS_RDONLY)
 		return 0;
 
+	if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
+	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
+		pr_info("BTRFS: creating free space tree\n");
+		ret = btrfs_create_free_space_tree(fs_info);
+		if (ret) {
+			pr_warn("BTRFS: failed to create free space tree %d\n",
+				ret);
+			close_ctree(tree_root);
+			return ret;
+		}
+	}
+
 	down_read(&fs_info->cleanup_work_sem);
 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
@@ -3077,6 +3103,18 @@ retry_root_backup:
 
 	btrfs_qgroup_rescan_resume(fs_info);
 
+	if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
+	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
+		pr_info("BTRFS: clearing free space tree\n");
+		ret = btrfs_clear_free_space_tree(fs_info);
+		if (ret) {
+			pr_warn("BTRFS: failed to clear free space tree %d\n",
+				ret);
+			close_ctree(tree_root);
+			return ret;
+		}
+	}
+
 	if (!fs_info->uuid_root) {
 		pr_info("BTRFS: creating UUID tree\n");
 		ret = btrfs_create_uuid_tree(fs_info);
@@ -3903,11 +3941,6 @@ int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
 	return !ret;
 }
 
-int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
-{
-	return set_extent_buffer_uptodate(buf);
-}
-
 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 {
 	struct btrfs_root *root;
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index adeb31830b9c..7c52e29fdb0d 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -116,7 +116,6 @@ static inline void btrfs_put_fs_root(struct btrfs_root *root)
 void btrfs_mark_buffer_dirty(struct extent_buffer *buf);
 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
 			  int atomic);
-int btrfs_set_buffer_uptodate(struct extent_buffer *buf);
 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid);
 u32 btrfs_csum_data(char *data, u32 seed, size_t len);
 void btrfs_csum_final(u32 crc, char *result);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 9e3639b21567..46a39336e3d6 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -33,6 +33,7 @@
 #include "raid56.h"
 #include "locking.h"
 #include "free-space-cache.h"
+#include "free-space-tree.h"
 #include "math.h"
 #include "sysfs.h"
 #include "qgroup.h"
@@ -357,8 +358,8 @@ static void fragment_free_space(struct btrfs_root *root,
  * we need to check the pinned_extents for any extents that can't be used yet
  * since their free space will be released as soon as the transaction commits.
  */
-static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
-			      struct btrfs_fs_info *info, u64 start, u64 end)
+u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
+		       struct btrfs_fs_info *info, u64 start, u64 end)
 {
 	u64 extent_start, extent_end, size, total_added = 0;
 	int ret;
@@ -395,11 +396,10 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 	return total_added;
 }
 
-static noinline void caching_thread(struct btrfs_work *work)
+static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
 {
 	struct btrfs_block_group_cache *block_group;
 	struct btrfs_fs_info *fs_info;
-	struct btrfs_caching_control *caching_ctl;
 	struct btrfs_root *extent_root;
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
@@ -407,17 +407,16 @@ static noinline void caching_thread(struct btrfs_work *work)
 	u64 total_found = 0;
 	u64 last = 0;
 	u32 nritems;
-	int ret = -ENOMEM;
+	int ret;
 	bool wakeup = true;
 
-	caching_ctl = container_of(work, struct btrfs_caching_control, work);
 	block_group = caching_ctl->block_group;
 	fs_info = block_group->fs_info;
 	extent_root = fs_info->extent_root;
 
 	path = btrfs_alloc_path();
 	if (!path)
-		goto out;
+		return -ENOMEM;
 
 	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 
@@ -443,15 +442,11 @@ static noinline void caching_thread(struct btrfs_work *work)
 	key.objectid = last;
 	key.offset = 0;
 	key.type = BTRFS_EXTENT_ITEM_KEY;
-again:
-	mutex_lock(&caching_ctl->mutex);
-	/* need to make sure the commit_root doesn't disappear */
-	down_read(&fs_info->commit_root_sem);
 
 next:
 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 	if (ret < 0)
-		goto err;
+		goto out;
 
 	leaf = path->nodes[0];
 	nritems = btrfs_header_nritems(leaf);
@@ -477,12 +472,14 @@ next:
 				up_read(&fs_info->commit_root_sem);
 				mutex_unlock(&caching_ctl->mutex);
 				cond_resched();
-				goto again;
+				mutex_lock(&caching_ctl->mutex);
+				down_read(&fs_info->commit_root_sem);
+				goto next;
 			}
 
 			ret = btrfs_next_leaf(extent_root, path);
 			if (ret < 0)
-				goto err;
+				goto out;
 			if (ret)
 				break;
 			leaf = path->nodes[0];
@@ -521,7 +518,7 @@ next:
 			else
 				last = key.objectid + key.offset;
 
-			if (total_found > (1024 * 1024 * 2)) {
+			if (total_found > CACHING_CTL_WAKE_UP) {
 				total_found = 0;
 				if (wakeup)
 					wake_up(&caching_ctl->wait);
@@ -534,9 +531,37 @@ next:
 	total_found += add_new_free_space(block_group, fs_info, last,
 					  block_group->key.objectid +
 					  block_group->key.offset);
+	caching_ctl->progress = (u64)-1;
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static noinline void caching_thread(struct btrfs_work *work)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_caching_control *caching_ctl;
+	struct btrfs_root *extent_root;
+	int ret;
+
+	caching_ctl = container_of(work, struct btrfs_caching_control, work);
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+	extent_root = fs_info->extent_root;
+
+	mutex_lock(&caching_ctl->mutex);
+	down_read(&fs_info->commit_root_sem);
+
+	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		ret = load_free_space_tree(caching_ctl);
+	else
+		ret = load_extent_tree_free(caching_ctl);
+
 	spin_lock(&block_group->lock);
 	block_group->caching_ctl = NULL;
-	block_group->cached = BTRFS_CACHE_FINISHED;
+	block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
 	spin_unlock(&block_group->lock);
 
 #ifdef CONFIG_BTRFS_DEBUG
@@ -555,20 +580,11 @@ next:
 #endif
 
 	caching_ctl->progress = (u64)-1;
-err:
-	btrfs_free_path(path);
-	up_read(&fs_info->commit_root_sem);
-
-	free_excluded_extents(extent_root, block_group);
 
+	up_read(&fs_info->commit_root_sem);
+	free_excluded_extents(fs_info->extent_root, block_group);
 	mutex_unlock(&caching_ctl->mutex);
-out:
-	if (ret) {
-		spin_lock(&block_group->lock);
-		block_group->caching_ctl = NULL;
-		block_group->cached = BTRFS_CACHE_ERROR;
-		spin_unlock(&block_group->lock);
-	}
+
 	wake_up(&caching_ctl->wait);
 
 	put_caching_control(caching_ctl);
@@ -680,8 +696,8 @@ static int cache_block_group(struct btrfs_block_group_cache *cache,
 		}
 	} else {
 		/*
-		 * We are not going to do the fast caching, set cached to the
-		 * appropriate value and wakeup any waiters.
+		 * We're either using the free space tree or no caching at all.
+		 * Set cached to the appropriate value and wakeup any waiters.
 		 */
 		spin_lock(&cache->lock);
 		if (load_cache_only) {
@@ -2910,6 +2926,9 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 	if (trans->aborted)
 		return 0;
 
+	if (root->fs_info->creating_free_space_tree)
+		return 0;
+
 	if (root == root->fs_info->extent_root)
 		root = root->fs_info->tree_root;
 
@@ -3684,11 +3703,21 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 
 	/*
-	 * We don't need the lock here since we are protected by the transaction
-	 * commit.  We want to do the cache_save_setup first and then run the
+	 * Even though we are in the critical section of the transaction commit,
+	 * we can still have concurrent tasks adding elements to this
+	 * transaction's list of dirty block groups. These tasks correspond to
+	 * endio free space workers started when writeback finishes for a
+	 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
+	 * allocate new block groups as a result of COWing nodes of the root
+	 * tree when updating the free space inode. The writeback for the space
+	 * caches is triggered by an earlier call to
+	 * btrfs_start_dirty_block_groups() and iterations of the following
+	 * loop.
+	 * Also we want to do the cache_save_setup first and then run the
 	 * delayed refs to make sure we have the best chance at doing this all
 	 * in one shot.
 	 */
+	spin_lock(&cur_trans->dirty_bgs_lock);
 	while (!list_empty(&cur_trans->dirty_bgs)) {
 		cache = list_first_entry(&cur_trans->dirty_bgs,
 					 struct btrfs_block_group_cache,
@@ -3700,11 +3729,13 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		 * finish and then do it all again
 		 */
 		if (!list_empty(&cache->io_list)) {
+			spin_unlock(&cur_trans->dirty_bgs_lock);
 			list_del_init(&cache->io_list);
 			btrfs_wait_cache_io(root, trans, cache,
 					    &cache->io_ctl, path,
 					    cache->key.objectid);
 			btrfs_put_block_group(cache);
+			spin_lock(&cur_trans->dirty_bgs_lock);
 		}
 
 		/*
@@ -3712,6 +3743,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		 * on any pending IO
 		 */
 		list_del_init(&cache->dirty_list);
+		spin_unlock(&cur_trans->dirty_bgs_lock);
 		should_put = 1;
 
 		cache_save_setup(cache, trans, path);
@@ -3743,7 +3775,9 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		/* if its not on the io list, we need to put the block group */
 		if (should_put)
 			btrfs_put_block_group(cache);
+		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
+	spin_unlock(&cur_trans->dirty_bgs_lock);
 
 	while (!list_empty(io)) {
 		cache = list_first_entry(io, struct btrfs_block_group_cache,
@@ -6661,6 +6695,13 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			}
 		}
 
+		ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
+					     num_bytes);
+		if (ret) {
+			btrfs_abort_transaction(trans, extent_root, ret);
+			goto out;
+		}
+
 		ret = update_block_group(trans, root, bytenr, num_bytes, 0);
 		if (ret) {
 			btrfs_abort_transaction(trans, extent_root, ret);
@@ -7672,6 +7713,11 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(path->nodes[0]);
 	btrfs_free_path(path);
 
+	ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
+					  ins->offset);
+	if (ret)
+		return ret;
+
 	ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
 	if (ret) { /* -ENOENT, logic error */
 		btrfs_err(fs_info, "update block group failed for %llu %llu",
@@ -7752,6 +7798,11 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(leaf);
 	btrfs_free_path(path);
 
+	ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
+					  num_bytes);
+	if (ret)
+		return ret;
+
 	ret = update_block_group(trans, root, ins->objectid, root->nodesize,
 				 1);
 	if (ret) { /* -ENOENT, logic error */
@@ -7834,7 +7885,7 @@ btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
 
 	btrfs_set_lock_blocking(buf);
-	btrfs_set_buffer_uptodate(buf);
+	set_extent_buffer_uptodate(buf);
 
 	if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
 		buf->log_index = root->log_transid % 2;
@@ -9653,6 +9704,8 @@ btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
 	cache->full_stripe_len = btrfs_full_stripe_len(root,
 					       &root->fs_info->mapping_tree,
 					       start);
+	set_free_space_tree_thresholds(cache);
+
 	atomic_set(&cache->count, 1);
 	spin_lock_init(&cache->lock);
 	init_rwsem(&cache->data_rwsem);
@@ -9664,6 +9717,7 @@ btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
 	INIT_LIST_HEAD(&cache->io_list);
 	btrfs_init_free_space_ctl(cache);
 	atomic_set(&cache->trimming, 0);
+	mutex_init(&cache->free_space_lock);
 
 	return cache;
 }
@@ -9874,6 +9928,8 @@ void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
 					       key.objectid, key.offset);
 		if (ret)
 			btrfs_abort_transaction(trans, extent_root, ret);
+		add_block_group_free_space(trans, root->fs_info, block_group);
+		/* already aborted the transaction if it failed. */
 next:
 		list_del_init(&block_group->bg_list);
 	}
@@ -9904,6 +9960,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 	cache->flags = type;
 	cache->last_byte_to_unpin = (u64)-1;
 	cache->cached = BTRFS_CACHE_FINISHED;
+	cache->needs_free_space = 1;
 	ret = exclude_super_stripes(root, cache);
 	if (ret) {
 		/*
@@ -10274,6 +10331,10 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 
 	unlock_chunks(root);
 
+	ret = remove_block_group_free_space(trans, root->fs_info, block_group);
+	if (ret)
+		goto out;
+
 	btrfs_put_block_group(block_group);
 	btrfs_put_block_group(block_group);
 
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 9abe18763a7f..a2356e261531 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -1285,20 +1285,6 @@ search_again:
 }
 
 /* wrappers around set/clear extent bit */
-int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
-		     gfp_t mask)
-{
-	return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
-			      NULL, mask);
-}
-
-int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
-		    unsigned bits, gfp_t mask)
-{
-	return set_extent_bit(tree, start, end, bits, NULL,
-			      NULL, mask);
-}
-
 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 			   unsigned bits, gfp_t mask,
 			   struct extent_changeset *changeset)
@@ -1323,17 +1309,6 @@ int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 				  cached, mask, NULL);
 }
 
-int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
-		      unsigned bits, gfp_t mask)
-{
-	int wake = 0;
-
-	if (bits & EXTENT_LOCKED)
-		wake = 1;
-
-	return clear_extent_bit(tree, start, end, bits, wake, 0, NULL, mask);
-}
-
 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 			     unsigned bits, gfp_t mask,
 			     struct extent_changeset *changeset)
@@ -1348,63 +1323,18 @@ int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 				  changeset);
 }
 
-int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
-			struct extent_state **cached_state, gfp_t mask)
-{
-	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_UPTODATE,
-			      NULL, cached_state, mask);
-}
-
-int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
-		      struct extent_state **cached_state, gfp_t mask)
-{
-	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
-			      NULL, cached_state, mask);
-}
-
-int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
-		       gfp_t mask)
-{
-	return clear_extent_bit(tree, start, end,
-				EXTENT_DIRTY | EXTENT_DELALLOC |
-				EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
-}
-
-int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
-		     gfp_t mask)
-{
-	return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
-			      NULL, mask);
-}
-
-int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
-			struct extent_state **cached_state, gfp_t mask)
-{
-	return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
-			      cached_state, mask);
-}
-
-int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
-			  struct extent_state **cached_state, gfp_t mask)
-{
-	return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
-				cached_state, mask);
-}
-
 /*
  * either insert or lock state struct between start and end use mask to tell
  * us if waiting is desired.
  */
 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
-		     unsigned bits, struct extent_state **cached_state)
+		     struct extent_state **cached_state)
 {
 	int err;
 	u64 failed_start;
 
 	while (1) {
-		err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
+		err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
 				       EXTENT_LOCKED, &failed_start,
 				       cached_state, GFP_NOFS, NULL);
 		if (err == -EEXIST) {
@@ -1417,11 +1347,6 @@ int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 	return err;
 }
 
-int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
-{
-	return lock_extent_bits(tree, start, end, 0, NULL);
-}
-
 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
 {
 	int err;
@@ -1438,20 +1363,7 @@ int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
 	return 1;
 }
 
-int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
-			 struct extent_state **cached, gfp_t mask)
-{
-	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
-				mask);
-}
-
-int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
-{
-	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
-				GFP_NOFS);
-}
-
-int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
+void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
 {
 	unsigned long index = start >> PAGE_CACHE_SHIFT;
 	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
@@ -1464,10 +1376,9 @@ int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
 		page_cache_release(page);
 		index++;
 	}
-	return 0;
 }
 
-int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
+void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
 {
 	unsigned long index = start >> PAGE_CACHE_SHIFT;
 	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
@@ -1481,13 +1392,12 @@ int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
 		page_cache_release(page);
 		index++;
 	}
-	return 0;
 }
 
 /*
  * helper function to set both pages and extents in the tree writeback
  */
-static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
+static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 {
 	unsigned long index = start >> PAGE_CACHE_SHIFT;
 	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
@@ -1500,7 +1410,6 @@ static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 		page_cache_release(page);
 		index++;
 	}
-	return 0;
 }
 
 /* find the first state struct with 'bits' set after 'start', and
@@ -1800,7 +1709,7 @@ again:
 	BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
 
 	/* step three, lock the state bits for the whole range */
-	lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
+	lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
 
 	/* then test to make sure it is all still delalloc */
 	ret = test_range_bit(tree, delalloc_start, delalloc_end,
@@ -1820,7 +1729,7 @@ out_failed:
 	return found;
 }
 
-int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
+void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
 				 struct page *locked_page,
 				 unsigned clear_bits,
 				 unsigned long page_ops)
@@ -1835,7 +1744,7 @@ int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
 
 	clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
 	if (page_ops == 0)
-		return 0;
+		return;
 
 	if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
 		mapping_set_error(inode->i_mapping, -EIO);
@@ -1869,7 +1778,6 @@ int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
 		index += ret;
 		cond_resched();
 	}
-	return 0;
 }
 
 /*
@@ -2516,7 +2424,7 @@ static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
 
 /* lots and lots of room for performance fixes in the end_bio funcs */
 
-int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
+void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
 {
 	int uptodate = (err == 0);
 	struct extent_io_tree *tree;
@@ -2537,7 +2445,6 @@ int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
 		ret = ret < 0 ? ret : -EIO;
 		mapping_set_error(page->mapping, ret);
 	}
-	return 0;
 }
 
 /*
@@ -2579,9 +2486,7 @@ static void end_bio_extent_writepage(struct bio *bio)
 		start = page_offset(page);
 		end = start + bvec->bv_offset + bvec->bv_len - 1;
 
-		if (end_extent_writepage(page, bio->bi_error, start, end))
-			continue;
-
+		end_extent_writepage(page, bio->bi_error, start, end);
 		end_page_writeback(page);
 	}
 
@@ -4326,7 +4231,7 @@ int extent_invalidatepage(struct extent_io_tree *tree,
 	if (start > end)
 		return 0;
 
-	lock_extent_bits(tree, start, end, 0, &cached_state);
+	lock_extent_bits(tree, start, end, &cached_state);
 	wait_on_page_writeback(page);
 	clear_extent_bit(tree, start, end,
 			 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
@@ -4536,7 +4441,7 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 		last_for_get_extent = isize;
 	}
 
-	lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
+	lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
 			 &cached_state);
 
 	em = get_extent_skip_holes(inode, start, last_for_get_extent,
@@ -4797,24 +4702,14 @@ struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
 	return new;
 }
 
-struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
-						u64 start)
+struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
+						  u64 start, unsigned long len)
 {
 	struct extent_buffer *eb;
-	unsigned long len;
 	unsigned long num_pages;
 	unsigned long i;
 
-	if (!fs_info) {
-		/*
-		 * Called only from tests that don't always have a fs_info
-		 * available, but we know that nodesize is 4096
-		 */
-		len = 4096;
-	} else {
-		len = fs_info->tree_root->nodesize;
-	}
-	num_pages = num_extent_pages(0, len);
+	num_pages = num_extent_pages(start, len);
 
 	eb = __alloc_extent_buffer(fs_info, start, len);
 	if (!eb)
@@ -4837,6 +4732,24 @@ err:
 	return NULL;
 }
 
+struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
+						u64 start)
+{
+	unsigned long len;
+
+	if (!fs_info) {
+		/*
+		 * Called only from tests that don't always have a fs_info
+		 * available, but we know that nodesize is 4096
+		 */
+		len = 4096;
+	} else {
+		len = fs_info->tree_root->nodesize;
+	}
+
+	return __alloc_dummy_extent_buffer(fs_info, start, len);
+}
+
 static void check_buffer_tree_ref(struct extent_buffer *eb)
 {
 	int refs;
@@ -5227,7 +5140,7 @@ int set_extent_buffer_dirty(struct extent_buffer *eb)
 	return was_dirty;
 }
 
-int clear_extent_buffer_uptodate(struct extent_buffer *eb)
+void clear_extent_buffer_uptodate(struct extent_buffer *eb)
 {
 	unsigned long i;
 	struct page *page;
@@ -5240,10 +5153,9 @@ int clear_extent_buffer_uptodate(struct extent_buffer *eb)
 		if (page)
 			ClearPageUptodate(page);
 	}
-	return 0;
 }
 
-int set_extent_buffer_uptodate(struct extent_buffer *eb)
+void set_extent_buffer_uptodate(struct extent_buffer *eb)
 {
 	unsigned long i;
 	struct page *page;
@@ -5255,7 +5167,6 @@ int set_extent_buffer_uptodate(struct extent_buffer *eb)
 		page = eb->pages[i];
 		SetPageUptodate(page);
 	}
-	return 0;
 }
 
 int extent_buffer_uptodate(struct extent_buffer *eb)
@@ -5594,6 +5505,155 @@ void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
 	}
 }
 
+/*
+ * The extent buffer bitmap operations are done with byte granularity because
+ * bitmap items are not guaranteed to be aligned to a word and therefore a
+ * single word in a bitmap may straddle two pages in the extent buffer.
+ */
+#define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
+#define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
+#define BITMAP_FIRST_BYTE_MASK(start) \
+	((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
+#define BITMAP_LAST_BYTE_MASK(nbits) \
+	(BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
+
+/*
+ * eb_bitmap_offset() - calculate the page and offset of the byte containing the
+ * given bit number
+ * @eb: the extent buffer
+ * @start: offset of the bitmap item in the extent buffer
+ * @nr: bit number
+ * @page_index: return index of the page in the extent buffer that contains the
+ * given bit number
+ * @page_offset: return offset into the page given by page_index
+ *
+ * This helper hides the ugliness of finding the byte in an extent buffer which
+ * contains a given bit.
+ */
+static inline void eb_bitmap_offset(struct extent_buffer *eb,
+				    unsigned long start, unsigned long nr,
+				    unsigned long *page_index,
+				    size_t *page_offset)
+{
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	size_t byte_offset = BIT_BYTE(nr);
+	size_t offset;
+
+	/*
+	 * The byte we want is the offset of the extent buffer + the offset of
+	 * the bitmap item in the extent buffer + the offset of the byte in the
+	 * bitmap item.
+	 */
+	offset = start_offset + start + byte_offset;
+
+	*page_index = offset >> PAGE_CACHE_SHIFT;
+	*page_offset = offset & (PAGE_CACHE_SIZE - 1);
+}
+
+/**
+ * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
+ * @eb: the extent buffer
+ * @start: offset of the bitmap item in the extent buffer
+ * @nr: bit number to test
+ */
+int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
+			   unsigned long nr)
+{
+	char *kaddr;
+	struct page *page;
+	unsigned long i;
+	size_t offset;
+
+	eb_bitmap_offset(eb, start, nr, &i, &offset);
+	page = eb->pages[i];
+	WARN_ON(!PageUptodate(page));
+	kaddr = page_address(page);
+	return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
+}
+
+/**
+ * extent_buffer_bitmap_set - set an area of a bitmap
+ * @eb: the extent buffer
+ * @start: offset of the bitmap item in the extent buffer
+ * @pos: bit number of the first bit
+ * @len: number of bits to set
+ */
+void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
+			      unsigned long pos, unsigned long len)
+{
+	char *kaddr;
+	struct page *page;
+	unsigned long i;
+	size_t offset;
+	const unsigned int size = pos + len;
+	int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
+	unsigned int mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
+
+	eb_bitmap_offset(eb, start, pos, &i, &offset);
+	page = eb->pages[i];
+	WARN_ON(!PageUptodate(page));
+	kaddr = page_address(page);
+
+	while (len >= bits_to_set) {
+		kaddr[offset] |= mask_to_set;
+		len -= bits_to_set;
+		bits_to_set = BITS_PER_BYTE;
+		mask_to_set = ~0U;
+		if (++offset >= PAGE_CACHE_SIZE && len > 0) {
+			offset = 0;
+			page = eb->pages[++i];
+			WARN_ON(!PageUptodate(page));
+			kaddr = page_address(page);
+		}
+	}
+	if (len) {
+		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
+		kaddr[offset] |= mask_to_set;
+	}
+}
+
+
+/**
+ * extent_buffer_bitmap_clear - clear an area of a bitmap
+ * @eb: the extent buffer
+ * @start: offset of the bitmap item in the extent buffer
+ * @pos: bit number of the first bit
+ * @len: number of bits to clear
+ */
+void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
+				unsigned long pos, unsigned long len)
+{
+	char *kaddr;
+	struct page *page;
+	unsigned long i;
+	size_t offset;
+	const unsigned int size = pos + len;
+	int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
+	unsigned int mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
+
+	eb_bitmap_offset(eb, start, pos, &i, &offset);
+	page = eb->pages[i];
+	WARN_ON(!PageUptodate(page));
+	kaddr = page_address(page);
+
+	while (len >= bits_to_clear) {
+		kaddr[offset] &= ~mask_to_clear;
+		len -= bits_to_clear;
+		bits_to_clear = BITS_PER_BYTE;
+		mask_to_clear = ~0U;
+		if (++offset >= PAGE_CACHE_SIZE && len > 0) {
+			offset = 0;
+			page = eb->pages[++i];
+			WARN_ON(!PageUptodate(page));
+			kaddr = page_address(page);
+		}
+	}
+	if (len) {
+		mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
+		kaddr[offset] &= ~mask_to_clear;
+	}
+}
+
 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
 {
 	unsigned long distance = (src > dst) ? src - dst : dst - src;
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index f4c1ae11855f..0377413bd4b9 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -199,12 +199,14 @@ int try_release_extent_mapping(struct extent_map_tree *map,
 			       struct extent_io_tree *tree, struct page *page,
 			       gfp_t mask);
 int try_release_extent_buffer(struct page *page);
-int lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
-		     unsigned bits, struct extent_state **cached);
-int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end);
-int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
-			 struct extent_state **cached, gfp_t mask);
+		     struct extent_state **cached);
+
+static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	return lock_extent_bits(tree, start, end, NULL);
+}
+
 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
 			  get_extent_t *get_extent, int mirror_num);
@@ -221,39 +223,105 @@ void free_extent_state(struct extent_state *state);
 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
 		   unsigned bits, int filled,
 		   struct extent_state *cached_state);
-int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
-		      unsigned bits, gfp_t mask);
 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 			     unsigned bits, gfp_t mask,
 			     struct extent_changeset *changeset);
 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 		     unsigned bits, int wake, int delete,
 		     struct extent_state **cached, gfp_t mask);
-int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
-		    unsigned bits, gfp_t mask);
+
+static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
+				GFP_NOFS);
+}
+
+static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start,
+		u64 end, struct extent_state **cached, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
+				mask);
+}
+
+static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start,
+		u64 end, unsigned bits, gfp_t mask)
+{
+	int wake = 0;
+
+	if (bits & EXTENT_LOCKED)
+		wake = 1;
+
+	return clear_extent_bit(tree, start, end, bits, wake, 0, NULL, mask);
+}
+
 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 			   unsigned bits, gfp_t mask,
 			   struct extent_changeset *changeset);
 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 		   unsigned bits, u64 *failed_start,
 		   struct extent_state **cached_state, gfp_t mask);
-int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
-			struct extent_state **cached_state, gfp_t mask);
-int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
-			  struct extent_state **cached_state, gfp_t mask);
-int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
-		   gfp_t mask);
-int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
-		     gfp_t mask);
-int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
-		       gfp_t mask);
+
+static inline int set_extent_bits(struct extent_io_tree *tree, u64 start,
+		u64 end, unsigned bits, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, bits, NULL, NULL, mask);
+}
+
+static inline int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
+		u64 end, struct extent_state **cached_state, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
+				cached_state, mask);
+}
+
+static inline int set_extent_dirty(struct extent_io_tree *tree, u64 start,
+		u64 end, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
+			      NULL, mask);
+}
+
+static inline int clear_extent_dirty(struct extent_io_tree *tree, u64 start,
+		u64 end, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end,
+				EXTENT_DIRTY | EXTENT_DELALLOC |
+				EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
+}
+
 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 		       unsigned bits, unsigned clear_bits,
 		       struct extent_state **cached_state, gfp_t mask);
-int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
-			struct extent_state **cached_state, gfp_t mask);
-int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
-		      struct extent_state **cached_state, gfp_t mask);
+
+static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
+		u64 end, struct extent_state **cached_state, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end,
+			      EXTENT_DELALLOC | EXTENT_UPTODATE,
+			      NULL, cached_state, mask);
+}
+
+static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
+		u64 end, struct extent_state **cached_state, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end,
+			      EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
+			      NULL, cached_state, mask);
+}
+
+static inline int set_extent_new(struct extent_io_tree *tree, u64 start,
+		u64 end, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL, mask);
+}
+
+static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start,
+		u64 end, struct extent_state **cached_state, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
+			      cached_state, mask);
+}
+
 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 			  u64 *start_ret, u64 *end_ret, unsigned bits,
 			  struct extent_state **cached_state);
@@ -282,8 +350,10 @@ void set_page_extent_mapped(struct page *page);
 
 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 					  u64 start);
+struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
+						  u64 start, unsigned long len);
 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
-		u64 start);
+						u64 start);
 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src);
 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
 					 u64 start);
@@ -328,19 +398,25 @@ void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 			   unsigned long src_offset, unsigned long len);
 void memset_extent_buffer(struct extent_buffer *eb, char c,
 			  unsigned long start, unsigned long len);
+int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
+			   unsigned long pos);
+void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
+			      unsigned long pos, unsigned long len);
+void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
+				unsigned long pos, unsigned long len);
 void clear_extent_buffer_dirty(struct extent_buffer *eb);
 int set_extent_buffer_dirty(struct extent_buffer *eb);
-int set_extent_buffer_uptodate(struct extent_buffer *eb);
-int clear_extent_buffer_uptodate(struct extent_buffer *eb);
+void set_extent_buffer_uptodate(struct extent_buffer *eb);
+void clear_extent_buffer_uptodate(struct extent_buffer *eb);
 int extent_buffer_uptodate(struct extent_buffer *eb);
 int extent_buffer_under_io(struct extent_buffer *eb);
 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset,
 		      unsigned long min_len, char **map,
 		      unsigned long *map_start,
 		      unsigned long *map_len);
-int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
-int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
-int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
+void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
+void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
+void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
 				 struct page *locked_page,
 				 unsigned bits_to_clear,
 				 unsigned long page_ops);
@@ -357,7 +433,7 @@ int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
 		      int mirror_num);
 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
 		     unsigned int pg_offset);
-int end_extent_writepage(struct page *page, int err, u64 start, u64 end);
+void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
 			 int mirror_num);
 
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 0f09526aa7d9..364e0f1f61f6 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1394,7 +1394,7 @@ lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
 	if (start_pos < inode->i_size) {
 		struct btrfs_ordered_extent *ordered;
 		lock_extent_bits(&BTRFS_I(inode)->io_tree,
-				 start_pos, last_pos, 0, cached_state);
+				 start_pos, last_pos, cached_state);
 		ordered = btrfs_lookup_ordered_range(inode, start_pos,
 						     last_pos - start_pos + 1);
 		if (ordered &&
@@ -2398,7 +2398,7 @@ static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
 		truncate_pagecache_range(inode, lockstart, lockend);
 
 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
-				 0, &cached_state);
+				 &cached_state);
 		ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
 
 		/*
@@ -2705,7 +2705,7 @@ static long btrfs_fallocate(struct file *file, int mode,
 		 * transaction
 		 */
 		lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
-				 locked_end, 0, &cached_state);
+				 locked_end, &cached_state);
 		ordered = btrfs_lookup_first_ordered_extent(inode,
 							    alloc_end - 1);
 		if (ordered &&
@@ -2852,7 +2852,7 @@ static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
 	lockend--;
 	len = lockend - lockstart + 1;
 
-	lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
+	lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
 			 &cached_state);
 
 	while (start < inode->i_size) {
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index ed8a3b7d3565..d468bfd3debc 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -1261,7 +1261,7 @@ static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
 		goto out;
 
 	lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
-			 0, &cached_state);
+			 &cached_state);
 
 	io_ctl_set_generation(io_ctl, trans->transid);
 
diff --git a/fs/btrfs/free-space-tree.c b/fs/btrfs/free-space-tree.c
new file mode 100644
index 000000000000..393e36bd5845
--- /dev/null
+++ b/fs/btrfs/free-space-tree.c
@@ -0,0 +1,1591 @@
+/*
+ * Copyright (C) 2015 Facebook.  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/kernel.h>
+#include <linux/vmalloc.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "locking.h"
+#include "free-space-tree.h"
+#include "transaction.h"
+
+static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
+					struct btrfs_fs_info *fs_info,
+					struct btrfs_block_group_cache *block_group,
+					struct btrfs_path *path);
+
+void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
+{
+	u32 bitmap_range;
+	size_t bitmap_size;
+	u64 num_bitmaps, total_bitmap_size;
+
+	/*
+	 * We convert to bitmaps when the disk space required for using extents
+	 * exceeds that required for using bitmaps.
+	 */
+	bitmap_range = cache->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
+	num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
+			      bitmap_range);
+	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
+	total_bitmap_size = num_bitmaps * bitmap_size;
+	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
+					    sizeof(struct btrfs_item));
+
+	/*
+	 * We allow for a small buffer between the high threshold and low
+	 * threshold to avoid thrashing back and forth between the two formats.
+	 */
+	if (cache->bitmap_high_thresh > 100)
+		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
+	else
+		cache->bitmap_low_thresh = 0;
+}
+
+static int add_new_free_space_info(struct btrfs_trans_handle *trans,
+				   struct btrfs_fs_info *fs_info,
+				   struct btrfs_block_group_cache *block_group,
+				   struct btrfs_path *path)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	int ret;
+
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_FREE_SPACE_INFO_KEY;
+	key.offset = block_group->key.offset;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	info = btrfs_item_ptr(leaf, path->slots[0],
+			      struct btrfs_free_space_info);
+	btrfs_set_free_space_extent_count(leaf, info, 0);
+	btrfs_set_free_space_flags(leaf, info, 0);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+out:
+	btrfs_release_path(path);
+	return ret;
+}
+
+struct btrfs_free_space_info *
+search_free_space_info(struct btrfs_trans_handle *trans,
+		       struct btrfs_fs_info *fs_info,
+		       struct btrfs_block_group_cache *block_group,
+		       struct btrfs_path *path, int cow)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key;
+	int ret;
+
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_FREE_SPACE_INFO_KEY;
+	key.offset = block_group->key.offset;
+
+	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret != 0) {
+		btrfs_warn(fs_info, "missing free space info for %llu\n",
+			   block_group->key.objectid);
+		ASSERT(0);
+		return ERR_PTR(-ENOENT);
+	}
+
+	return btrfs_item_ptr(path->nodes[0], path->slots[0],
+			      struct btrfs_free_space_info);
+}
+
+/*
+ * btrfs_search_slot() but we're looking for the greatest key less than the
+ * passed key.
+ */
+static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  struct btrfs_key *key, struct btrfs_path *p,
+				  int ins_len, int cow)
+{
+	int ret;
+
+	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
+	if (ret < 0)
+		return ret;
+
+	if (ret == 0) {
+		ASSERT(0);
+		return -EIO;
+	}
+
+	if (p->slots[0] == 0) {
+		ASSERT(0);
+		return -EIO;
+	}
+	p->slots[0]--;
+
+	return 0;
+}
+
+static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
+{
+	return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
+}
+
+static unsigned long *alloc_bitmap(u32 bitmap_size)
+{
+	return __vmalloc(bitmap_size, GFP_NOFS | __GFP_HIGHMEM | __GFP_ZERO,
+			 PAGE_KERNEL);
+}
+
+int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	unsigned long *bitmap;
+	char *bitmap_cursor;
+	u64 start, end;
+	u64 bitmap_range, i;
+	u32 bitmap_size, flags, expected_extent_count;
+	u32 extent_count = 0;
+	int done = 0, nr;
+	int ret;
+
+	bitmap_size = free_space_bitmap_size(block_group->key.offset,
+					     block_group->sectorsize);
+	bitmap = alloc_bitmap(bitmap_size);
+	if (!bitmap) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
+				u64 first, last;
+
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+
+				first = div_u64(found_key.objectid - start,
+						block_group->sectorsize);
+				last = div_u64(found_key.objectid + found_key.offset - start,
+					       block_group->sectorsize);
+				bitmap_set(bitmap, first, last - first);
+
+				extent_count++;
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	info = search_free_space_info(trans, fs_info, block_group, path, 1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	leaf = path->nodes[0];
+	flags = btrfs_free_space_flags(leaf, info);
+	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
+	btrfs_set_free_space_flags(leaf, info, flags);
+	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(path);
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	bitmap_cursor = (char *)bitmap;
+	bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
+	i = start;
+	while (i < end) {
+		unsigned long ptr;
+		u64 extent_size;
+		u32 data_size;
+
+		extent_size = min(end - i, bitmap_range);
+		data_size = free_space_bitmap_size(extent_size,
+						   block_group->sectorsize);
+
+		key.objectid = i;
+		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
+		key.offset = extent_size;
+
+		ret = btrfs_insert_empty_item(trans, root, path, &key,
+					      data_size);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+		write_extent_buffer(leaf, bitmap_cursor, ptr,
+				    data_size);
+		btrfs_mark_buffer_dirty(leaf);
+		btrfs_release_path(path);
+
+		i += extent_size;
+		bitmap_cursor += data_size;
+	}
+
+	ret = 0;
+out:
+	vfree(bitmap);
+	if (ret)
+		btrfs_abort_transaction(trans, root, ret);
+	return ret;
+}
+
+int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	unsigned long *bitmap;
+	u64 start, end;
+	/* Initialize to silence GCC. */
+	u64 extent_start = 0;
+	u64 offset;
+	u32 bitmap_size, flags, expected_extent_count;
+	int prev_bit = 0, bit, bitnr;
+	u32 extent_count = 0;
+	int done = 0, nr;
+	int ret;
+
+	bitmap_size = free_space_bitmap_size(block_group->key.offset,
+					     block_group->sectorsize);
+	bitmap = alloc_bitmap(bitmap_size);
+	if (!bitmap) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
+				unsigned long ptr;
+				char *bitmap_cursor;
+				u32 bitmap_pos, data_size;
+
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+
+				bitmap_pos = div_u64(found_key.objectid - start,
+						     block_group->sectorsize *
+						     BITS_PER_BYTE);
+				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
+				data_size = free_space_bitmap_size(found_key.offset,
+								   block_group->sectorsize);
+
+				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
+				read_extent_buffer(leaf, bitmap_cursor, ptr,
+						   data_size);
+
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	info = search_free_space_info(trans, fs_info, block_group, path, 1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	leaf = path->nodes[0];
+	flags = btrfs_free_space_flags(leaf, info);
+	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
+	btrfs_set_free_space_flags(leaf, info, flags);
+	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(path);
+
+	offset = start;
+	bitnr = 0;
+	while (offset < end) {
+		bit = !!test_bit(bitnr, bitmap);
+		if (prev_bit == 0 && bit == 1) {
+			extent_start = offset;
+		} else if (prev_bit == 1 && bit == 0) {
+			key.objectid = extent_start;
+			key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+			key.offset = offset - extent_start;
+
+			ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+			if (ret)
+				goto out;
+			btrfs_release_path(path);
+
+			extent_count++;
+		}
+		prev_bit = bit;
+		offset += block_group->sectorsize;
+		bitnr++;
+	}
+	if (prev_bit == 1) {
+		key.objectid = extent_start;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = end - extent_start;
+
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+
+		extent_count++;
+	}
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	ret = 0;
+out:
+	vfree(bitmap);
+	if (ret)
+		btrfs_abort_transaction(trans, root, ret);
+	return ret;
+}
+
+static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
+					  struct btrfs_fs_info *fs_info,
+					  struct btrfs_block_group_cache *block_group,
+					  struct btrfs_path *path,
+					  int new_extents)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	u32 extent_count;
+	int ret = 0;
+
+	if (new_extents == 0)
+		return 0;
+
+	info = search_free_space_info(trans, fs_info, block_group, path, 1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
+
+	extent_count += new_extents;
+	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(path);
+
+	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
+	    extent_count > block_group->bitmap_high_thresh) {
+		ret = convert_free_space_to_bitmaps(trans, fs_info, block_group,
+						    path);
+	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
+		   extent_count < block_group->bitmap_low_thresh) {
+		ret = convert_free_space_to_extents(trans, fs_info, block_group,
+						    path);
+	}
+
+out:
+	return ret;
+}
+
+int free_space_test_bit(struct btrfs_block_group_cache *block_group,
+			struct btrfs_path *path, u64 offset)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	u64 found_start, found_end;
+	unsigned long ptr, i;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(offset >= found_start && offset < found_end);
+
+	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	i = div_u64(offset - found_start, block_group->sectorsize);
+	return !!extent_buffer_test_bit(leaf, ptr, i);
+}
+
+static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
+				struct btrfs_path *path, u64 *start, u64 *size,
+				int bit)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	u64 end = *start + *size;
+	u64 found_start, found_end;
+	unsigned long ptr, first, last;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(*start >= found_start && *start < found_end);
+	ASSERT(end > found_start);
+
+	if (end > found_end)
+		end = found_end;
+
+	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	first = div_u64(*start - found_start, block_group->sectorsize);
+	last = div_u64(end - found_start, block_group->sectorsize);
+	if (bit)
+		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
+	else
+		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
+	btrfs_mark_buffer_dirty(leaf);
+
+	*size -= end - *start;
+	*start = end;
+}
+
+/*
+ * We can't use btrfs_next_item() in modify_free_space_bitmap() because
+ * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
+ * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
+ * looking for.
+ */
+static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root, struct btrfs_path *p)
+{
+	struct btrfs_key key;
+
+	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
+		p->slots[0]++;
+		return 0;
+	}
+
+	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
+	btrfs_release_path(p);
+
+	key.objectid += key.offset;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
+}
+
+/*
+ * If remove is 1, then we are removing free space, thus clearing bits in the
+ * bitmap. If remove is 0, then we are adding free space, thus setting bits in
+ * the bitmap.
+ */
+static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *block_group,
+				    struct btrfs_path *path,
+				    u64 start, u64 size, int remove)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key;
+	u64 end = start + size;
+	u64 cur_start, cur_size;
+	int prev_bit, next_bit;
+	int new_extents;
+	int ret;
+
+	/*
+	 * Read the bit for the block immediately before the extent of space if
+	 * that block is within the block group.
+	 */
+	if (start > block_group->key.objectid) {
+		u64 prev_block = start - block_group->sectorsize;
+
+		key.objectid = prev_block;
+		key.type = (u8)-1;
+		key.offset = (u64)-1;
+
+		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
+		if (ret)
+			goto out;
+
+		prev_bit = free_space_test_bit(block_group, path, prev_block);
+
+		/* The previous block may have been in the previous bitmap. */
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+		if (start >= key.objectid + key.offset) {
+			ret = free_space_next_bitmap(trans, root, path);
+			if (ret)
+				goto out;
+		}
+	} else {
+		key.objectid = start;
+		key.type = (u8)-1;
+		key.offset = (u64)-1;
+
+		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
+		if (ret)
+			goto out;
+
+		prev_bit = -1;
+	}
+
+	/*
+	 * Iterate over all of the bitmaps overlapped by the extent of space,
+	 * clearing/setting bits as required.
+	 */
+	cur_start = start;
+	cur_size = size;
+	while (1) {
+		free_space_set_bits(block_group, path, &cur_start, &cur_size,
+				    !remove);
+		if (cur_size == 0)
+			break;
+		ret = free_space_next_bitmap(trans, root, path);
+		if (ret)
+			goto out;
+	}
+
+	/*
+	 * Read the bit for the block immediately after the extent of space if
+	 * that block is within the block group.
+	 */
+	if (end < block_group->key.objectid + block_group->key.offset) {
+		/* The next block may be in the next bitmap. */
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+		if (end >= key.objectid + key.offset) {
+			ret = free_space_next_bitmap(trans, root, path);
+			if (ret)
+				goto out;
+		}
+
+		next_bit = free_space_test_bit(block_group, path, end);
+	} else {
+		next_bit = -1;
+	}
+
+	if (remove) {
+		new_extents = -1;
+		if (prev_bit == 1) {
+			/* Leftover on the left. */
+			new_extents++;
+		}
+		if (next_bit == 1) {
+			/* Leftover on the right. */
+			new_extents++;
+		}
+	} else {
+		new_extents = 1;
+		if (prev_bit == 1) {
+			/* Merging with neighbor on the left. */
+			new_extents--;
+		}
+		if (next_bit == 1) {
+			/* Merging with neighbor on the right. */
+			new_extents--;
+		}
+	}
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+static int remove_free_space_extent(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *block_group,
+				    struct btrfs_path *path,
+				    u64 start, u64 size)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key;
+	u64 found_start, found_end;
+	u64 end = start + size;
+	int new_extents = -1;
+	int ret;
+
+	key.objectid = start;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(start >= found_start && end <= found_end);
+
+	/*
+	 * Okay, now that we've found the free space extent which contains the
+	 * free space that we are removing, there are four cases:
+	 *
+	 * 1. We're using the whole extent: delete the key we found and
+	 * decrement the free space extent count.
+	 * 2. We are using part of the extent starting at the beginning: delete
+	 * the key we found and insert a new key representing the leftover at
+	 * the end. There is no net change in the number of extents.
+	 * 3. We are using part of the extent ending at the end: delete the key
+	 * we found and insert a new key representing the leftover at the
+	 * beginning. There is no net change in the number of extents.
+	 * 4. We are using part of the extent in the middle: delete the key we
+	 * found and insert two new keys representing the leftovers on each
+	 * side. Where we used to have one extent, we now have two, so increment
+	 * the extent count. We may need to convert the block group to bitmaps
+	 * as a result.
+	 */
+
+	/* Delete the existing key (cases 1-4). */
+	ret = btrfs_del_item(trans, root, path);
+	if (ret)
+		goto out;
+
+	/* Add a key for leftovers at the beginning (cases 3 and 4). */
+	if (start > found_start) {
+		key.objectid = found_start;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = start - found_start;
+
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		new_extents++;
+	}
+
+	/* Add a key for leftovers at the end (cases 2 and 4). */
+	if (end < found_end) {
+		key.objectid = end;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = found_end - end;
+
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		new_extents++;
+	}
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path, u64 start, u64 size)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	int ret;
+
+	if (block_group->needs_free_space) {
+		ret = __add_block_group_free_space(trans, fs_info, block_group,
+						   path);
+		if (ret)
+			return ret;
+	}
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info))
+		return PTR_ERR(info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		return modify_free_space_bitmap(trans, fs_info, block_group,
+						path, start, size, 1);
+	} else {
+		return remove_free_space_extent(trans, fs_info, block_group,
+						path, start, size);
+	}
+}
+
+int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info,
+				u64 start, u64 size)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_path *path;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	block_group = btrfs_lookup_block_group(fs_info, start);
+	if (!block_group) {
+		ASSERT(0);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	mutex_lock(&block_group->free_space_lock);
+	ret = __remove_from_free_space_tree(trans, fs_info, block_group, path,
+					    start, size);
+	mutex_unlock(&block_group->free_space_lock);
+
+	btrfs_put_block_group(block_group);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, fs_info->free_space_root, ret);
+	return ret;
+}
+
+static int add_free_space_extent(struct btrfs_trans_handle *trans,
+				 struct btrfs_fs_info *fs_info,
+				 struct btrfs_block_group_cache *block_group,
+				 struct btrfs_path *path,
+				 u64 start, u64 size)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key, new_key;
+	u64 found_start, found_end;
+	u64 end = start + size;
+	int new_extents = 1;
+	int ret;
+
+	/*
+	 * We are adding a new extent of free space, but we need to merge
+	 * extents. There are four cases here:
+	 *
+	 * 1. The new extent does not have any immediate neighbors to merge
+	 * with: add the new key and increment the free space extent count. We
+	 * may need to convert the block group to bitmaps as a result.
+	 * 2. The new extent has an immediate neighbor before it: remove the
+	 * previous key and insert a new key combining both of them. There is no
+	 * net change in the number of extents.
+	 * 3. The new extent has an immediate neighbor after it: remove the next
+	 * key and insert a new key combining both of them. There is no net
+	 * change in the number of extents.
+	 * 4. The new extent has immediate neighbors on both sides: remove both
+	 * of the keys and insert a new key combining all of them. Where we used
+	 * to have two extents, we now have one, so decrement the extent count.
+	 */
+
+	new_key.objectid = start;
+	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+	new_key.offset = size;
+
+	/* Search for a neighbor on the left. */
+	if (start == block_group->key.objectid)
+		goto right;
+	key.objectid = start - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
+		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
+		btrfs_release_path(path);
+		goto right;
+	}
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(found_start >= block_group->key.objectid &&
+	       found_end > block_group->key.objectid);
+	ASSERT(found_start < start && found_end <= start);
+
+	/*
+	 * Delete the neighbor on the left and absorb it into the new key (cases
+	 * 2 and 4).
+	 */
+	if (found_end == start) {
+		ret = btrfs_del_item(trans, root, path);
+		if (ret)
+			goto out;
+		new_key.objectid = found_start;
+		new_key.offset += key.offset;
+		new_extents--;
+	}
+	btrfs_release_path(path);
+
+right:
+	/* Search for a neighbor on the right. */
+	if (end == block_group->key.objectid + block_group->key.offset)
+		goto insert;
+	key.objectid = end;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
+		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
+		btrfs_release_path(path);
+		goto insert;
+	}
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(found_start >= block_group->key.objectid &&
+	       found_end > block_group->key.objectid);
+	ASSERT((found_start < start && found_end <= start) ||
+	       (found_start >= end && found_end > end));
+
+	/*
+	 * Delete the neighbor on the right and absorb it into the new key
+	 * (cases 3 and 4).
+	 */
+	if (found_start == end) {
+		ret = btrfs_del_item(trans, root, path);
+		if (ret)
+			goto out;
+		new_key.offset += key.offset;
+		new_extents--;
+	}
+	btrfs_release_path(path);
+
+insert:
+	/* Insert the new key (cases 1-4). */
+	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
+	if (ret)
+		goto out;
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info,
+			     struct btrfs_block_group_cache *block_group,
+			     struct btrfs_path *path, u64 start, u64 size)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	int ret;
+
+	if (block_group->needs_free_space) {
+		ret = __add_block_group_free_space(trans, fs_info, block_group,
+						   path);
+		if (ret)
+			return ret;
+	}
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info))
+		return PTR_ERR(info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		return modify_free_space_bitmap(trans, fs_info, block_group,
+						path, start, size, 0);
+	} else {
+		return add_free_space_extent(trans, fs_info, block_group, path,
+					     start, size);
+	}
+}
+
+int add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   u64 start, u64 size)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_path *path;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	block_group = btrfs_lookup_block_group(fs_info, start);
+	if (!block_group) {
+		ASSERT(0);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	mutex_lock(&block_group->free_space_lock);
+	ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start,
+				       size);
+	mutex_unlock(&block_group->free_space_lock);
+
+	btrfs_put_block_group(block_group);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, fs_info->free_space_root, ret);
+	return ret;
+}
+
+/*
+ * Populate the free space tree by walking the extent tree. Operations on the
+ * extent tree that happen as a result of writes to the free space tree will go
+ * through the normal add/remove hooks.
+ */
+static int populate_free_space_tree(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_root *extent_root = fs_info->extent_root;
+	struct btrfs_path *path, *path2;
+	struct btrfs_key key;
+	u64 start, end;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	path->reada = 1;
+
+	path2 = btrfs_alloc_path();
+	if (!path2) {
+		btrfs_free_path(path);
+		return -ENOMEM;
+	}
+
+	ret = add_new_free_space_info(trans, fs_info, block_group, path2);
+	if (ret)
+		goto out;
+
+	mutex_lock(&block_group->free_space_lock);
+
+	/*
+	 * Iterate through all of the extent and metadata items in this block
+	 * group, adding the free space between them and the free space at the
+	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
+	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
+	 * contained in.
+	 */
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = 0;
+
+	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
+	if (ret < 0)
+		goto out_locked;
+	ASSERT(ret == 0);
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+	while (1) {
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
+		    key.type == BTRFS_METADATA_ITEM_KEY) {
+			if (key.objectid >= end)
+				break;
+
+			if (start < key.objectid) {
+				ret = __add_to_free_space_tree(trans, fs_info,
+							       block_group,
+							       path2, start,
+							       key.objectid -
+							       start);
+				if (ret)
+					goto out_locked;
+			}
+			start = key.objectid;
+			if (key.type == BTRFS_METADATA_ITEM_KEY)
+				start += fs_info->tree_root->nodesize;
+			else
+				start += key.offset;
+		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
+			if (key.objectid != block_group->key.objectid)
+				break;
+		}
+
+		ret = btrfs_next_item(extent_root, path);
+		if (ret < 0)
+			goto out_locked;
+		if (ret)
+			break;
+	}
+	if (start < end) {
+		ret = __add_to_free_space_tree(trans, fs_info, block_group,
+					       path2, start, end - start);
+		if (ret)
+			goto out_locked;
+	}
+
+	ret = 0;
+out_locked:
+	mutex_unlock(&block_group->free_space_lock);
+out:
+	btrfs_free_path(path2);
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *free_space_root;
+	struct btrfs_block_group_cache *block_group;
+	struct rb_node *node;
+	int ret;
+
+	trans = btrfs_start_transaction(tree_root, 0);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	fs_info->creating_free_space_tree = 1;
+	free_space_root = btrfs_create_tree(trans, fs_info,
+					    BTRFS_FREE_SPACE_TREE_OBJECTID);
+	if (IS_ERR(free_space_root)) {
+		ret = PTR_ERR(free_space_root);
+		goto abort;
+	}
+	fs_info->free_space_root = free_space_root;
+
+	node = rb_first(&fs_info->block_group_cache_tree);
+	while (node) {
+		block_group = rb_entry(node, struct btrfs_block_group_cache,
+				       cache_node);
+		ret = populate_free_space_tree(trans, fs_info, block_group);
+		if (ret)
+			goto abort;
+		node = rb_next(node);
+	}
+
+	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
+	fs_info->creating_free_space_tree = 0;
+
+	ret = btrfs_commit_transaction(trans, tree_root);
+	if (ret)
+		return ret;
+
+	return 0;
+
+abort:
+	fs_info->creating_free_space_tree = 0;
+	btrfs_abort_transaction(trans, tree_root, ret);
+	btrfs_end_transaction(trans, tree_root);
+	return ret;
+}
+
+static int clear_free_space_tree(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int nr;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->leave_spinning = 1;
+
+	key.objectid = 0;
+	key.type = 0;
+	key.offset = 0;
+
+	while (1) {
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (ret < 0)
+			goto out;
+
+		nr = btrfs_header_nritems(path->nodes[0]);
+		if (!nr)
+			break;
+
+		path->slots[0] = 0;
+		ret = btrfs_del_items(trans, root, path, 0, nr);
+		if (ret)
+			goto out;
+
+		btrfs_release_path(path);
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *free_space_root = fs_info->free_space_root;
+	int ret;
+
+	trans = btrfs_start_transaction(tree_root, 0);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
+	fs_info->free_space_root = NULL;
+
+	ret = clear_free_space_tree(trans, free_space_root);
+	if (ret)
+		goto abort;
+
+	ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key);
+	if (ret)
+		goto abort;
+
+	list_del(&free_space_root->dirty_list);
+
+	btrfs_tree_lock(free_space_root->node);
+	clean_tree_block(trans, tree_root->fs_info, free_space_root->node);
+	btrfs_tree_unlock(free_space_root->node);
+	btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
+			      0, 1);
+
+	free_extent_buffer(free_space_root->node);
+	free_extent_buffer(free_space_root->commit_root);
+	kfree(free_space_root);
+
+	ret = btrfs_commit_transaction(trans, tree_root);
+	if (ret)
+		return ret;
+
+	return 0;
+
+abort:
+	btrfs_abort_transaction(trans, tree_root, ret);
+	btrfs_end_transaction(trans, tree_root);
+	return ret;
+}
+
+static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
+					struct btrfs_fs_info *fs_info,
+					struct btrfs_block_group_cache *block_group,
+					struct btrfs_path *path)
+{
+	u64 start, end;
+	int ret;
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	block_group->needs_free_space = 0;
+
+	ret = add_new_free_space_info(trans, fs_info, block_group, path);
+	if (ret)
+		return ret;
+
+	return __add_to_free_space_tree(trans, fs_info, block_group, path,
+					block_group->key.objectid,
+					block_group->key.offset);
+}
+
+int add_block_group_free_space(struct btrfs_trans_handle *trans,
+			       struct btrfs_fs_info *fs_info,
+			       struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_path *path = NULL;
+	int ret = 0;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	mutex_lock(&block_group->free_space_lock);
+	if (!block_group->needs_free_space)
+		goto out;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	ret = __add_block_group_free_space(trans, fs_info, block_group, path);
+
+out:
+	btrfs_free_path(path);
+	mutex_unlock(&block_group->free_space_lock);
+	if (ret)
+		btrfs_abort_transaction(trans, fs_info->free_space_root, ret);
+	return ret;
+}
+
+int remove_block_group_free_space(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_path *path;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	u64 start, end;
+	int done = 0, nr;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	if (block_group->needs_free_space) {
+		/* We never added this block group to the free space tree. */
+		return 0;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				nr++;
+				path->slots[0]--;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
+				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, root, ret);
+	return ret;
+}
+
+static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
+				   struct btrfs_path *path,
+				   u32 expected_extent_count)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_root *root;
+	struct btrfs_key key;
+	int prev_bit = 0, bit;
+	/* Initialize to silence GCC. */
+	u64 extent_start = 0;
+	u64 end, offset;
+	u64 total_found = 0;
+	u32 extent_count = 0;
+	int ret;
+
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+	root = fs_info->free_space_root;
+
+	end = block_group->key.objectid + block_group->key.offset;
+
+	while (1) {
+		ret = btrfs_next_item(root, path);
+		if (ret < 0)
+			goto out;
+		if (ret)
+			break;
+
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
+			break;
+
+		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
+
+		caching_ctl->progress = key.objectid;
+
+		offset = key.objectid;
+		while (offset < key.objectid + key.offset) {
+			bit = free_space_test_bit(block_group, path, offset);
+			if (prev_bit == 0 && bit == 1) {
+				extent_start = offset;
+			} else if (prev_bit == 1 && bit == 0) {
+				total_found += add_new_free_space(block_group,
+								  fs_info,
+								  extent_start,
+								  offset);
+				if (total_found > CACHING_CTL_WAKE_UP) {
+					total_found = 0;
+					wake_up(&caching_ctl->wait);
+				}
+				extent_count++;
+			}
+			prev_bit = bit;
+			offset += block_group->sectorsize;
+		}
+	}
+	if (prev_bit == 1) {
+		total_found += add_new_free_space(block_group, fs_info,
+						  extent_start, end);
+		extent_count++;
+	}
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	caching_ctl->progress = (u64)-1;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
+				   struct btrfs_path *path,
+				   u32 expected_extent_count)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_root *root;
+	struct btrfs_key key;
+	u64 end;
+	u64 total_found = 0;
+	u32 extent_count = 0;
+	int ret;
+
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+	root = fs_info->free_space_root;
+
+	end = block_group->key.objectid + block_group->key.offset;
+
+	while (1) {
+		ret = btrfs_next_item(root, path);
+		if (ret < 0)
+			goto out;
+		if (ret)
+			break;
+
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
+			break;
+
+		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
+		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
+
+		caching_ctl->progress = key.objectid;
+
+		total_found += add_new_free_space(block_group, fs_info,
+						  key.objectid,
+						  key.objectid + key.offset);
+		if (total_found > CACHING_CTL_WAKE_UP) {
+			total_found = 0;
+			wake_up(&caching_ctl->wait);
+		}
+		extent_count++;
+	}
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	caching_ctl->progress = (u64)-1;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_free_space_info *info;
+	struct btrfs_path *path;
+	u32 extent_count, flags;
+	int ret;
+
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	/*
+	 * Just like caching_thread() doesn't want to deadlock on the extent
+	 * tree, we don't want to deadlock on the free space tree.
+	 */
+	path->skip_locking = 1;
+	path->search_commit_root = 1;
+	path->reada = 1;
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+
+	/*
+	 * We left path pointing to the free space info item, so now
+	 * load_free_space_foo can just iterate through the free space tree from
+	 * there.
+	 */
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
+		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
+	else
+		ret = load_free_space_extents(caching_ctl, path, extent_count);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
diff --git a/fs/btrfs/free-space-tree.h b/fs/btrfs/free-space-tree.h
new file mode 100644
index 000000000000..54ffced3bce8
--- /dev/null
+++ b/fs/btrfs/free-space-tree.h
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2015 Facebook.  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_FREE_SPACE_TREE
+#define __BTRFS_FREE_SPACE_TREE
+
+/*
+ * The default size for new free space bitmap items. The last bitmap in a block
+ * group may be truncated, and none of the free space tree code assumes that
+ * existing bitmaps are this size.
+ */
+#define BTRFS_FREE_SPACE_BITMAP_SIZE 256
+#define BTRFS_FREE_SPACE_BITMAP_BITS (BTRFS_FREE_SPACE_BITMAP_SIZE * BITS_PER_BYTE)
+
+void set_free_space_tree_thresholds(struct btrfs_block_group_cache *block_group);
+int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info);
+int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info);
+int load_free_space_tree(struct btrfs_caching_control *caching_ctl);
+int add_block_group_free_space(struct btrfs_trans_handle *trans,
+			       struct btrfs_fs_info *fs_info,
+			       struct btrfs_block_group_cache *block_group);
+int remove_block_group_free_space(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group);
+int add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   u64 start, u64 size);
+int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info,
+				u64 start, u64 size);
+
+/* Exposed for testing. */
+struct btrfs_free_space_info *
+search_free_space_info(struct btrfs_trans_handle *trans,
+		       struct btrfs_fs_info *fs_info,
+		       struct btrfs_block_group_cache *block_group,
+		       struct btrfs_path *path, int cow);
+int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info,
+			     struct btrfs_block_group_cache *block_group,
+			     struct btrfs_path *path, u64 start, u64 size);
+int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path, u64 start, u64 size);
+int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path);
+int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path);
+int free_space_test_bit(struct btrfs_block_group_cache *block_group,
+			struct btrfs_path *path, u64 offset);
+
+#endif
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index dc2c1bedbafa..f09cf27245f9 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -66,6 +66,13 @@ struct btrfs_iget_args {
 	struct btrfs_root *root;
 };
 
+struct btrfs_dio_data {
+	u64 outstanding_extents;
+	u64 reserve;
+	u64 unsubmitted_oe_range_start;
+	u64 unsubmitted_oe_range_end;
+};
+
 static const struct inode_operations btrfs_dir_inode_operations;
 static const struct inode_operations btrfs_symlink_inode_operations;
 static const struct inode_operations btrfs_dir_ro_inode_operations;
@@ -1988,7 +1995,7 @@ again:
 	page_start = page_offset(page);
 	page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 
-	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
+	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
 			 &cached_state);
 
 	/* already ordered? We're done */
@@ -2481,7 +2488,7 @@ static noinline int relink_extent_backref(struct btrfs_path *path,
 	lock_start = backref->file_pos;
 	lock_end = backref->file_pos + backref->num_bytes - 1;
 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
-			 0, &cached);
+			 &cached);
 
 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
 	if (ordered) {
@@ -2873,7 +2880,7 @@ static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
 
 	lock_extent_bits(io_tree, ordered_extent->file_offset,
 			 ordered_extent->file_offset + ordered_extent->len - 1,
-			 0, &cached_state);
+			 &cached_state);
 
 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
 			ordered_extent->file_offset + ordered_extent->len - 1,
@@ -4659,7 +4666,7 @@ again:
 	}
 	wait_on_page_writeback(page);
 
-	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
+	lock_extent_bits(io_tree, page_start, page_end, &cached_state);
 	set_page_extent_mapped(page);
 
 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
@@ -4790,7 +4797,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
 	while (1) {
 		struct btrfs_ordered_extent *ordered;
 
-		lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
+		lock_extent_bits(io_tree, hole_start, block_end - 1,
 				 &cached_state);
 		ordered = btrfs_lookup_ordered_range(inode, hole_start,
 						     block_end - hole_start);
@@ -5102,7 +5109,7 @@ static void evict_inode_truncate_pages(struct inode *inode)
 		end = state->end;
 		spin_unlock(&io_tree->lock);
 
-		lock_extent_bits(io_tree, start, end, 0, &cached_state);
+		lock_extent_bits(io_tree, start, end, &cached_state);
 
 		/*
 		 * If still has DELALLOC flag, the extent didn't reach disk,
@@ -7371,7 +7378,7 @@ static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
 
 	while (1) {
 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
-				 0, cached_state);
+				 cached_state);
 		/*
 		 * We're concerned with the entire range that we're going to be
 		 * doing DIO to, so we need to make sure theres no ordered
@@ -7399,25 +7406,21 @@ static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
 			btrfs_start_ordered_extent(inode, ordered, 1);
 			btrfs_put_ordered_extent(ordered);
 		} else {
-			/* Screw you mmap */
-			ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
-			if (ret)
-				break;
-			ret = filemap_fdatawait_range(inode->i_mapping,
-						      lockstart,
-						      lockend);
-			if (ret)
-				break;
-
 			/*
-			 * If we found a page that couldn't be invalidated just
-			 * fall back to buffered.
+			 * We could trigger writeback for this range (and wait
+			 * for it to complete) and then invalidate the pages for
+			 * this range (through invalidate_inode_pages2_range()),
+			 * but that can lead us to a deadlock with a concurrent
+			 * call to readpages() (a buffered read or a defrag call
+			 * triggered a readahead) on a page lock due to an
+			 * ordered dio extent we created before but did not have
+			 * yet a corresponding bio submitted (whence it can not
+			 * complete), which makes readpages() wait for that
+			 * ordered extent to complete while holding a lock on
+			 * that page.
 			 */
-			ret = invalidate_inode_pages2_range(inode->i_mapping,
-					lockstart >> PAGE_CACHE_SHIFT,
-					lockend >> PAGE_CACHE_SHIFT);
-			if (ret)
-				break;
+			ret = -ENOTBLK;
+			break;
 		}
 
 		cond_resched();
@@ -7473,11 +7476,6 @@ static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 	return em;
 }
 
-struct btrfs_dio_data {
-	u64 outstanding_extents;
-	u64 reserve;
-};
-
 static void adjust_dio_outstanding_extents(struct inode *inode,
 					   struct btrfs_dio_data *dio_data,
 					   const u64 len)
@@ -7661,6 +7659,7 @@ unlock:
 		btrfs_free_reserved_data_space(inode, start, len);
 		WARN_ON(dio_data->reserve < len);
 		dio_data->reserve -= len;
+		dio_data->unsubmitted_oe_range_end = start + len;
 		current->journal_info = dio_data;
 	}
 
@@ -7983,22 +7982,22 @@ static void btrfs_endio_direct_read(struct bio *bio)
 	bio_put(bio);
 }
 
-static void btrfs_endio_direct_write(struct bio *bio)
+static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
+						    const u64 offset,
+						    const u64 bytes,
+						    const int uptodate)
 {
-	struct btrfs_dio_private *dip = bio->bi_private;
-	struct inode *inode = dip->inode;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_ordered_extent *ordered = NULL;
-	u64 ordered_offset = dip->logical_offset;
-	u64 ordered_bytes = dip->bytes;
-	struct bio *dio_bio;
+	u64 ordered_offset = offset;
+	u64 ordered_bytes = bytes;
 	int ret;
 
 again:
 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
 						   &ordered_offset,
 						   ordered_bytes,
-						   !bio->bi_error);
+						   uptodate);
 	if (!ret)
 		goto out_test;
 
@@ -8011,13 +8010,22 @@ out_test:
 	 * our bio might span multiple ordered extents.  If we haven't
 	 * completed the accounting for the whole dio, go back and try again
 	 */
-	if (ordered_offset < dip->logical_offset + dip->bytes) {
-		ordered_bytes = dip->logical_offset + dip->bytes -
-			ordered_offset;
+	if (ordered_offset < offset + bytes) {
+		ordered_bytes = offset + bytes - ordered_offset;
 		ordered = NULL;
 		goto again;
 	}
-	dio_bio = dip->dio_bio;
+}
+
+static void btrfs_endio_direct_write(struct bio *bio)
+{
+	struct btrfs_dio_private *dip = bio->bi_private;
+	struct bio *dio_bio = dip->dio_bio;
+
+	btrfs_endio_direct_write_update_ordered(dip->inode,
+						dip->logical_offset,
+						dip->bytes,
+						!bio->bi_error);
 
 	kfree(dip);
 
@@ -8325,6 +8333,21 @@ static void btrfs_submit_direct(int rw, struct bio *dio_bio,
 		dip->subio_endio = btrfs_subio_endio_read;
 	}
 
+	/*
+	 * Reset the range for unsubmitted ordered extents (to a 0 length range)
+	 * even if we fail to submit a bio, because in such case we do the
+	 * corresponding error handling below and it must not be done a second
+	 * time by btrfs_direct_IO().
+	 */
+	if (write) {
+		struct btrfs_dio_data *dio_data = current->journal_info;
+
+		dio_data->unsubmitted_oe_range_end = dip->logical_offset +
+			dip->bytes;
+		dio_data->unsubmitted_oe_range_start =
+			dio_data->unsubmitted_oe_range_end;
+	}
+
 	ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
 	if (!ret)
 		return;
@@ -8353,24 +8376,15 @@ free_ordered:
 		dip = NULL;
 		io_bio = NULL;
 	} else {
-		if (write) {
-			struct btrfs_ordered_extent *ordered;
-
-			ordered = btrfs_lookup_ordered_extent(inode,
-							      file_offset);
-			set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
-			/*
-			 * Decrements our ref on the ordered extent and removes
-			 * the ordered extent from the inode's ordered tree,
-			 * doing all the proper resource cleanup such as for the
-			 * reserved space and waking up any waiters for this
-			 * ordered extent (through btrfs_remove_ordered_extent).
-			 */
-			btrfs_finish_ordered_io(ordered);
-		} else {
+		if (write)
+			btrfs_endio_direct_write_update_ordered(inode,
+						file_offset,
+						dio_bio->bi_iter.bi_size,
+						0);
+		else
 			unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
 			      file_offset + dio_bio->bi_iter.bi_size - 1);
-		}
+
 		dio_bio->bi_error = -EIO;
 		/*
 		 * Releases and cleans up our dio_bio, no need to bio_put()
@@ -8470,6 +8484,8 @@ static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 		 * originally calculated.  Abuse current->journal_info for this.
 		 */
 		dio_data.reserve = round_up(count, root->sectorsize);
+		dio_data.unsubmitted_oe_range_start = (u64)offset;
+		dio_data.unsubmitted_oe_range_end = (u64)offset;
 		current->journal_info = &dio_data;
 	} else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
 				     &BTRFS_I(inode)->runtime_flags)) {
@@ -8488,6 +8504,19 @@ static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 			if (dio_data.reserve)
 				btrfs_delalloc_release_space(inode, offset,
 							     dio_data.reserve);
+			/*
+			 * On error we might have left some ordered extents
+			 * without submitting corresponding bios for them, so
+			 * cleanup them up to avoid other tasks getting them
+			 * and waiting for them to complete forever.
+			 */
+			if (dio_data.unsubmitted_oe_range_start <
+			    dio_data.unsubmitted_oe_range_end)
+				btrfs_endio_direct_write_update_ordered(inode,
+					dio_data.unsubmitted_oe_range_start,
+					dio_data.unsubmitted_oe_range_end -
+					dio_data.unsubmitted_oe_range_start,
+					0);
 		} else if (ret >= 0 && (size_t)ret < count)
 			btrfs_delalloc_release_space(inode, offset,
 						     count - (size_t)ret);
@@ -8618,7 +8647,7 @@ static void btrfs_invalidatepage(struct page *page, unsigned int offset,
 	}
 
 	if (!inode_evicting)
-		lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
+		lock_extent_bits(tree, page_start, page_end, &cached_state);
 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
 	if (ordered) {
 		/*
@@ -8656,7 +8685,7 @@ static void btrfs_invalidatepage(struct page *page, unsigned int offset,
 		btrfs_put_ordered_extent(ordered);
 		if (!inode_evicting) {
 			cached_state = NULL;
-			lock_extent_bits(tree, page_start, page_end, 0,
+			lock_extent_bits(tree, page_start, page_end,
 					 &cached_state);
 		}
 	}
@@ -8754,7 +8783,7 @@ again:
 	}
 	wait_on_page_writeback(page);
 
-	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
+	lock_extent_bits(io_tree, page_start, page_end, &cached_state);
 	set_page_extent_mapped(page);
 
 	/*
@@ -9415,14 +9444,10 @@ static void btrfs_run_delalloc_work(struct btrfs_work *work)
 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
 				     work);
 	inode = delalloc_work->inode;
-	if (delalloc_work->wait) {
-		btrfs_wait_ordered_range(inode, 0, (u64)-1);
-	} else {
+	filemap_flush(inode->i_mapping);
+	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
+				&BTRFS_I(inode)->runtime_flags))
 		filemap_flush(inode->i_mapping);
-		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
-			     &BTRFS_I(inode)->runtime_flags))
-			filemap_flush(inode->i_mapping);
-	}
 
 	if (delalloc_work->delay_iput)
 		btrfs_add_delayed_iput(inode);
@@ -9432,7 +9457,7 @@ static void btrfs_run_delalloc_work(struct btrfs_work *work)
 }
 
 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
-						    int wait, int delay_iput)
+						    int delay_iput)
 {
 	struct btrfs_delalloc_work *work;
 
@@ -9443,7 +9468,6 @@ struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
 	init_completion(&work->completion);
 	INIT_LIST_HEAD(&work->list);
 	work->inode = inode;
-	work->wait = wait;
 	work->delay_iput = delay_iput;
 	WARN_ON_ONCE(!inode);
 	btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
@@ -9491,7 +9515,7 @@ static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
 		}
 		spin_unlock(&root->delalloc_lock);
 
-		work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
+		work = btrfs_alloc_delalloc_work(inode, delay_iput);
 		if (!work) {
 			if (delay_iput)
 				btrfs_add_delayed_iput(inode);
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 3d9b27d2176e..f8519b866c0a 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1001,7 +1001,7 @@ static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 		u64 end = start + len - 1;
 
 		/* get the big lock and read metadata off disk */
-		lock_extent_bits(io_tree, start, end, 0, &cached);
+		lock_extent_bits(io_tree, start, end, &cached);
 		em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 		unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 
@@ -1149,7 +1149,7 @@ again:
 		page_end = page_start + PAGE_CACHE_SIZE - 1;
 		while (1) {
 			lock_extent_bits(tree, page_start, page_end,
-					 0, &cached_state);
+					 &cached_state);
 			ordered = btrfs_lookup_ordered_extent(inode,
 							      page_start);
 			unlock_extent_cached(tree, page_start, page_end,
@@ -1209,7 +1209,7 @@ again:
 	page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
 
 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
-			 page_start, page_end - 1, 0, &cached_state);
+			 page_start, page_end - 1, &cached_state);
 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
@@ -4156,7 +4156,7 @@ static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
 		return -ENOMEM;
 
 	space_args.total_spaces = 0;
-	dest = kmalloc(alloc_size, GFP_NOFS);
+	dest = kmalloc(alloc_size, GFP_KERNEL);
 	if (!dest)
 		return -ENOMEM;
 	dest_orig = dest;
@@ -4682,7 +4682,7 @@ locked:
 		goto out_bargs;
 	}
 
-	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
+	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
 	if (!bctl) {
 		ret = -ENOMEM;
 		goto out_bargs;
@@ -4768,7 +4768,7 @@ static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
 		goto out;
 	}
 
-	bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
+	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
 	if (!bargs) {
 		ret = -ENOMEM;
 		goto out;
@@ -5028,7 +5028,7 @@ static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
 
-	qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
+	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
 	if (!qsa)
 		return -ENOMEM;
 
@@ -5158,7 +5158,7 @@ static long btrfs_ioctl_set_received_subvol_32(struct file *file,
 		goto out;
 	}
 
-	args64 = kmalloc(sizeof(*args64), GFP_NOFS);
+	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
 	if (!args64) {
 		ret = -ENOMEM;
 		goto out;
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index 143f2b7a8edf..dbfbfb3b86b5 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -4211,7 +4211,7 @@ static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
 
 	io_tree = &BTRFS_I(inode)->io_tree;
 
-	lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
+	lock_extent_bits(io_tree, lockstart, lockend, &cached_state);
 	ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
 	if (ordered) {
 		btrfs_put_ordered_extent(ordered);
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 5099e4633e50..ff15087052f5 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -295,10 +295,11 @@ enum {
 	Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 	Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 	Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
-	Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
-	Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
-	Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
-	Opt_check_integrity, Opt_check_integrity_including_extent_data,
+	Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
+	Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
+	Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
+	Opt_skip_balance, Opt_check_integrity,
+	Opt_check_integrity_including_extent_data,
 	Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 	Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 	Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
@@ -340,6 +341,7 @@ static const match_table_t tokens = {
 	{Opt_discard, "discard"},
 	{Opt_nodiscard, "nodiscard"},
 	{Opt_space_cache, "space_cache"},
+	{Opt_space_cache_version, "space_cache=%s"},
 	{Opt_clear_cache, "clear_cache"},
 	{Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 	{Opt_enospc_debug, "enospc_debug"},
@@ -383,7 +385,9 @@ int btrfs_parse_options(struct btrfs_root *root, char *options)
 	bool compress_force = false;
 
 	cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
-	if (cache_gen)
+	if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
+		btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
+	else if (cache_gen)
 		btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 
 	if (!options)
@@ -617,15 +621,35 @@ int btrfs_parse_options(struct btrfs_root *root, char *options)
 					     "turning off discard");
 			break;
 		case Opt_space_cache:
-			btrfs_set_and_info(root, SPACE_CACHE,
-					   "enabling disk space caching");
+		case Opt_space_cache_version:
+			if (token == Opt_space_cache ||
+			    strcmp(args[0].from, "v1") == 0) {
+				btrfs_clear_opt(root->fs_info->mount_opt,
+						FREE_SPACE_TREE);
+				btrfs_set_and_info(root, SPACE_CACHE,
+						   "enabling disk space caching");
+			} else if (strcmp(args[0].from, "v2") == 0) {
+				btrfs_clear_opt(root->fs_info->mount_opt,
+						SPACE_CACHE);
+				btrfs_set_and_info(root, FREE_SPACE_TREE,
+						   "enabling free space tree");
+			} else {
+				ret = -EINVAL;
+				goto out;
+			}
 			break;
 		case Opt_rescan_uuid_tree:
 			btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 			break;
 		case Opt_no_space_cache:
-			btrfs_clear_and_info(root, SPACE_CACHE,
-					     "disabling disk space caching");
+			if (btrfs_test_opt(root, SPACE_CACHE)) {
+				btrfs_clear_and_info(root, SPACE_CACHE,
+						     "disabling disk space caching");
+			}
+			if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
+				btrfs_clear_and_info(root, FREE_SPACE_TREE,
+						     "disabling free space tree");
+			}
 			break;
 		case Opt_inode_cache:
 			btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
@@ -754,8 +778,17 @@ int btrfs_parse_options(struct btrfs_root *root, char *options)
 		}
 	}
 out:
+	if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
+	    !btrfs_test_opt(root, FREE_SPACE_TREE) &&
+	    !btrfs_test_opt(root, CLEAR_CACHE)) {
+		btrfs_err(root->fs_info, "cannot disable free space tree");
+		ret = -EINVAL;
+
+	}
 	if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 		btrfs_info(root->fs_info, "disk space caching is enabled");
+	if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
+		btrfs_info(root->fs_info, "using free space tree");
 	kfree(orig);
 	return ret;
 }
@@ -1162,6 +1195,8 @@ static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 		seq_puts(seq, ",noacl");
 	if (btrfs_test_opt(root, SPACE_CACHE))
 		seq_puts(seq, ",space_cache");
+	else if (btrfs_test_opt(root, FREE_SPACE_TREE))
+		seq_puts(seq, ",space_cache=v2");
 	else
 		seq_puts(seq, ",nospace_cache");
 	if (btrfs_test_opt(root, RESCAN_UUID_TREE))
@@ -2249,6 +2284,9 @@ static int btrfs_run_sanity_tests(void)
 	if (ret)
 		goto out;
 	ret = btrfs_test_qgroups();
+	if (ret)
+		goto out;
+	ret = btrfs_test_free_space_tree();
 out:
 	btrfs_destroy_test_fs();
 	return ret;
diff --git a/fs/btrfs/tests/btrfs-tests.c b/fs/btrfs/tests/btrfs-tests.c
index 9626252ee6b4..b1d920b30070 100644
--- a/fs/btrfs/tests/btrfs-tests.c
+++ b/fs/btrfs/tests/btrfs-tests.c
@@ -21,6 +21,9 @@
 #include <linux/magic.h>
 #include "btrfs-tests.h"
 #include "../ctree.h"
+#include "../free-space-cache.h"
+#include "../free-space-tree.h"
+#include "../transaction.h"
 #include "../volumes.h"
 #include "../disk-io.h"
 #include "../qgroup.h"
@@ -122,6 +125,9 @@ struct btrfs_fs_info *btrfs_alloc_dummy_fs_info(void)
 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
+	extent_io_tree_init(&fs_info->freed_extents[0], NULL);
+	extent_io_tree_init(&fs_info->freed_extents[1], NULL);
+	fs_info->pinned_extents = &fs_info->freed_extents[0];
 	return fs_info;
 }
 
@@ -169,3 +175,55 @@ void btrfs_free_dummy_root(struct btrfs_root *root)
 	kfree(root);
 }
 
+struct btrfs_block_group_cache *
+btrfs_alloc_dummy_block_group(unsigned long length)
+{
+	struct btrfs_block_group_cache *cache;
+
+	cache = kzalloc(sizeof(*cache), GFP_NOFS);
+	if (!cache)
+		return NULL;
+	cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
+					GFP_NOFS);
+	if (!cache->free_space_ctl) {
+		kfree(cache);
+		return NULL;
+	}
+	cache->fs_info = btrfs_alloc_dummy_fs_info();
+	if (!cache->fs_info) {
+		kfree(cache->free_space_ctl);
+		kfree(cache);
+		return NULL;
+	}
+
+	cache->key.objectid = 0;
+	cache->key.offset = length;
+	cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+	cache->sectorsize = 4096;
+	cache->full_stripe_len = 4096;
+
+	INIT_LIST_HEAD(&cache->list);
+	INIT_LIST_HEAD(&cache->cluster_list);
+	INIT_LIST_HEAD(&cache->bg_list);
+	btrfs_init_free_space_ctl(cache);
+	mutex_init(&cache->free_space_lock);
+
+	return cache;
+}
+
+void btrfs_free_dummy_block_group(struct btrfs_block_group_cache *cache)
+{
+	if (!cache)
+		return;
+	__btrfs_remove_free_space_cache(cache->free_space_ctl);
+	kfree(cache->free_space_ctl);
+	kfree(cache);
+}
+
+void btrfs_init_dummy_trans(struct btrfs_trans_handle *trans)
+{
+	memset(trans, 0, sizeof(*trans));
+	trans->transid = 1;
+	INIT_LIST_HEAD(&trans->qgroup_ref_list);
+	trans->type = __TRANS_DUMMY;
+}
diff --git a/fs/btrfs/tests/btrfs-tests.h b/fs/btrfs/tests/btrfs-tests.h
index fd3954224480..054b8c73c951 100644
--- a/fs/btrfs/tests/btrfs-tests.h
+++ b/fs/btrfs/tests/btrfs-tests.h
@@ -24,17 +24,23 @@
 #define test_msg(fmt, ...) pr_info("BTRFS: selftest: " fmt, ##__VA_ARGS__)
 
 struct btrfs_root;
+struct btrfs_trans_handle;
 
 int btrfs_test_free_space_cache(void);
 int btrfs_test_extent_buffer_operations(void);
 int btrfs_test_extent_io(void);
 int btrfs_test_inodes(void);
 int btrfs_test_qgroups(void);
+int btrfs_test_free_space_tree(void);
 int btrfs_init_test_fs(void);
 void btrfs_destroy_test_fs(void);
 struct inode *btrfs_new_test_inode(void);
 struct btrfs_fs_info *btrfs_alloc_dummy_fs_info(void);
 void btrfs_free_dummy_root(struct btrfs_root *root);
+struct btrfs_block_group_cache *
+btrfs_alloc_dummy_block_group(unsigned long length);
+void btrfs_free_dummy_block_group(struct btrfs_block_group_cache *cache);
+void btrfs_init_dummy_trans(struct btrfs_trans_handle *trans);
 #else
 static inline int btrfs_test_free_space_cache(void)
 {
@@ -63,6 +69,10 @@ static inline int btrfs_test_qgroups(void)
 {
 	return 0;
 }
+static inline int btrfs_test_free_space_tree(void)
+{
+	return 0;
+}
 #endif
 
 #endif
diff --git a/fs/btrfs/tests/extent-io-tests.c b/fs/btrfs/tests/extent-io-tests.c
index 9e9f2368177d..71ab575e7633 100644
--- a/fs/btrfs/tests/extent-io-tests.c
+++ b/fs/btrfs/tests/extent-io-tests.c
@@ -18,6 +18,7 @@
 
 #include <linux/pagemap.h>
 #include <linux/sched.h>
+#include <linux/slab.h>
 #include "btrfs-tests.h"
 #include "../extent_io.h"
 
@@ -76,6 +77,8 @@ static int test_find_delalloc(void)
 	u64 found;
 	int ret = -EINVAL;
 
+	test_msg("Running find delalloc tests\n");
+
 	inode = btrfs_new_test_inode();
 	if (!inode) {
 		test_msg("Failed to allocate test inode\n");
@@ -268,8 +271,139 @@ out:
 	return ret;
 }
 
+static int __test_eb_bitmaps(unsigned long *bitmap, struct extent_buffer *eb,
+			     unsigned long len)
+{
+	unsigned long i, x;
+
+	memset(bitmap, 0, len);
+	memset_extent_buffer(eb, 0, 0, len);
+	if (memcmp_extent_buffer(eb, bitmap, 0, len) != 0) {
+		test_msg("Bitmap was not zeroed\n");
+		return -EINVAL;
+	}
+
+	bitmap_set(bitmap, 0, len * BITS_PER_BYTE);
+	extent_buffer_bitmap_set(eb, 0, 0, len * BITS_PER_BYTE);
+	if (memcmp_extent_buffer(eb, bitmap, 0, len) != 0) {
+		test_msg("Setting all bits failed\n");
+		return -EINVAL;
+	}
+
+	bitmap_clear(bitmap, 0, len * BITS_PER_BYTE);
+	extent_buffer_bitmap_clear(eb, 0, 0, len * BITS_PER_BYTE);
+	if (memcmp_extent_buffer(eb, bitmap, 0, len) != 0) {
+		test_msg("Clearing all bits failed\n");
+		return -EINVAL;
+	}
+
+	bitmap_set(bitmap, (PAGE_CACHE_SIZE - sizeof(long) / 2) * BITS_PER_BYTE,
+		   sizeof(long) * BITS_PER_BYTE);
+	extent_buffer_bitmap_set(eb, PAGE_CACHE_SIZE - sizeof(long) / 2, 0,
+				 sizeof(long) * BITS_PER_BYTE);
+	if (memcmp_extent_buffer(eb, bitmap, 0, len) != 0) {
+		test_msg("Setting straddling pages failed\n");
+		return -EINVAL;
+	}
+
+	bitmap_set(bitmap, 0, len * BITS_PER_BYTE);
+	bitmap_clear(bitmap,
+		     (PAGE_CACHE_SIZE - sizeof(long) / 2) * BITS_PER_BYTE,
+		     sizeof(long) * BITS_PER_BYTE);
+	extent_buffer_bitmap_set(eb, 0, 0, len * BITS_PER_BYTE);
+	extent_buffer_bitmap_clear(eb, PAGE_CACHE_SIZE - sizeof(long) / 2, 0,
+				   sizeof(long) * BITS_PER_BYTE);
+	if (memcmp_extent_buffer(eb, bitmap, 0, len) != 0) {
+		test_msg("Clearing straddling pages failed\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * Generate a wonky pseudo-random bit pattern for the sake of not using
+	 * something repetitive that could miss some hypothetical off-by-n bug.
+	 */
+	x = 0;
+	for (i = 0; i < len / sizeof(long); i++) {
+		x = (0x19660dULL * (u64)x + 0x3c6ef35fULL) & 0xffffffffUL;
+		bitmap[i] = x;
+	}
+	write_extent_buffer(eb, bitmap, 0, len);
+
+	for (i = 0; i < len * BITS_PER_BYTE; i++) {
+		int bit, bit1;
+
+		bit = !!test_bit(i, bitmap);
+		bit1 = !!extent_buffer_test_bit(eb, 0, i);
+		if (bit1 != bit) {
+			test_msg("Testing bit pattern failed\n");
+			return -EINVAL;
+		}
+
+		bit1 = !!extent_buffer_test_bit(eb, i / BITS_PER_BYTE,
+						i % BITS_PER_BYTE);
+		if (bit1 != bit) {
+			test_msg("Testing bit pattern with offset failed\n");
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+static int test_eb_bitmaps(void)
+{
+	unsigned long len = PAGE_CACHE_SIZE * 4;
+	unsigned long *bitmap;
+	struct extent_buffer *eb;
+	int ret;
+
+	test_msg("Running extent buffer bitmap tests\n");
+
+	bitmap = kmalloc(len, GFP_NOFS);
+	if (!bitmap) {
+		test_msg("Couldn't allocate test bitmap\n");
+		return -ENOMEM;
+	}
+
+	eb = __alloc_dummy_extent_buffer(NULL, 0, len);
+	if (!eb) {
+		test_msg("Couldn't allocate test extent buffer\n");
+		kfree(bitmap);
+		return -ENOMEM;
+	}
+
+	ret = __test_eb_bitmaps(bitmap, eb, len);
+	if (ret)
+		goto out;
+
+	/* Do it over again with an extent buffer which isn't page-aligned. */
+	free_extent_buffer(eb);
+	eb = __alloc_dummy_extent_buffer(NULL, PAGE_CACHE_SIZE / 2, len);
+	if (!eb) {
+		test_msg("Couldn't allocate test extent buffer\n");
+		kfree(bitmap);
+		return -ENOMEM;
+	}
+
+	ret = __test_eb_bitmaps(bitmap, eb, len);
+out:
+	free_extent_buffer(eb);
+	kfree(bitmap);
+	return ret;
+}
+
 int btrfs_test_extent_io(void)
 {
-	test_msg("Running find delalloc tests\n");
-	return test_find_delalloc();
+	int ret;
+
+	test_msg("Running extent I/O tests\n");
+
+	ret = test_find_delalloc();
+	if (ret)
+		goto out;
+
+	ret = test_eb_bitmaps();
+out:
+	test_msg("Extent I/O tests finished\n");
+	return ret;
 }
diff --git a/fs/btrfs/tests/free-space-tests.c b/fs/btrfs/tests/free-space-tests.c
index 69a11e63d668..e37d55ecdf1f 100644
--- a/fs/btrfs/tests/free-space-tests.c
+++ b/fs/btrfs/tests/free-space-tests.c
@@ -23,41 +23,6 @@
 #include "../free-space-cache.h"
 
 #define BITS_PER_BITMAP		(PAGE_CACHE_SIZE * 8)
-static struct btrfs_block_group_cache *init_test_block_group(void)
-{
-	struct btrfs_block_group_cache *cache;
-
-	cache = kzalloc(sizeof(*cache), GFP_NOFS);
-	if (!cache)
-		return NULL;
-	cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
-					GFP_NOFS);
-	if (!cache->free_space_ctl) {
-		kfree(cache);
-		return NULL;
-	}
-	cache->fs_info = btrfs_alloc_dummy_fs_info();
-	if (!cache->fs_info) {
-		kfree(cache->free_space_ctl);
-		kfree(cache);
-		return NULL;
-	}
-
-	cache->key.objectid = 0;
-	cache->key.offset = 1024 * 1024 * 1024;
-	cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
-	cache->sectorsize = 4096;
-	cache->full_stripe_len = 4096;
-
-	spin_lock_init(&cache->lock);
-	INIT_LIST_HEAD(&cache->list);
-	INIT_LIST_HEAD(&cache->cluster_list);
-	INIT_LIST_HEAD(&cache->bg_list);
-
-	btrfs_init_free_space_ctl(cache);
-
-	return cache;
-}
 
 /*
  * This test just does basic sanity checking, making sure we can add an exten
@@ -893,7 +858,7 @@ int btrfs_test_free_space_cache(void)
 
 	test_msg("Running btrfs free space cache tests\n");
 
-	cache = init_test_block_group();
+	cache = btrfs_alloc_dummy_block_group(1024 * 1024 * 1024);
 	if (!cache) {
 		test_msg("Couldn't run the tests\n");
 		return 0;
@@ -924,9 +889,7 @@ int btrfs_test_free_space_cache(void)
 
 	ret = test_steal_space_from_bitmap_to_extent(cache);
 out:
-	__btrfs_remove_free_space_cache(cache->free_space_ctl);
-	kfree(cache->free_space_ctl);
-	kfree(cache);
+	btrfs_free_dummy_block_group(cache);
 	btrfs_free_dummy_root(root);
 	test_msg("Free space cache tests finished\n");
 	return ret;
diff --git a/fs/btrfs/tests/free-space-tree-tests.c b/fs/btrfs/tests/free-space-tree-tests.c
new file mode 100644
index 000000000000..d05fe1ab4808
--- /dev/null
+++ b/fs/btrfs/tests/free-space-tree-tests.c
@@ -0,0 +1,571 @@
+/*
+ * Copyright (C) 2015 Facebook.  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 "btrfs-tests.h"
+#include "../ctree.h"
+#include "../disk-io.h"
+#include "../free-space-tree.h"
+#include "../transaction.h"
+
+struct free_space_extent {
+	u64 start, length;
+};
+
+/*
+ * The test cases align their operations to this in order to hit some of the
+ * edge cases in the bitmap code.
+ */
+#define BITMAP_RANGE (BTRFS_FREE_SPACE_BITMAP_BITS * 4096)
+
+static int __check_free_space_extents(struct btrfs_trans_handle *trans,
+				      struct btrfs_fs_info *fs_info,
+				      struct btrfs_block_group_cache *cache,
+				      struct btrfs_path *path,
+				      struct free_space_extent *extents,
+				      unsigned int num_extents)
+{
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key;
+	int prev_bit = 0, bit;
+	u64 extent_start = 0, offset, end;
+	u32 flags, extent_count;
+	unsigned int i;
+	int ret;
+
+	info = search_free_space_info(trans, fs_info, cache, path, 0);
+	if (IS_ERR(info)) {
+		test_msg("Could not find free space info\n");
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
+
+	if (extent_count != num_extents) {
+		test_msg("Extent count is wrong\n");
+		ret = -EINVAL;
+		goto out;
+	}
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		if (path->slots[0] != 0)
+			goto invalid;
+		end = cache->key.objectid + cache->key.offset;
+		i = 0;
+		while (++path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
+			btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+			if (key.type != BTRFS_FREE_SPACE_BITMAP_KEY)
+				goto invalid;
+			offset = key.objectid;
+			while (offset < key.objectid + key.offset) {
+				bit = free_space_test_bit(cache, path, offset);
+				if (prev_bit == 0 && bit == 1) {
+					extent_start = offset;
+				} else if (prev_bit == 1 && bit == 0) {
+					if (i >= num_extents)
+						goto invalid;
+					if (i >= num_extents ||
+					    extent_start != extents[i].start ||
+					    offset - extent_start != extents[i].length)
+						goto invalid;
+					i++;
+				}
+				prev_bit = bit;
+				offset += cache->sectorsize;
+			}
+		}
+		if (prev_bit == 1) {
+			if (i >= num_extents ||
+			    extent_start != extents[i].start ||
+			    end - extent_start != extents[i].length)
+				goto invalid;
+			i++;
+		}
+		if (i != num_extents)
+			goto invalid;
+	} else {
+		if (btrfs_header_nritems(path->nodes[0]) != num_extents + 1 ||
+		    path->slots[0] != 0)
+			goto invalid;
+		for (i = 0; i < num_extents; i++) {
+			path->slots[0]++;
+			btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+			if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY ||
+			    key.objectid != extents[i].start ||
+			    key.offset != extents[i].length)
+				goto invalid;
+		}
+	}
+
+	ret = 0;
+out:
+	btrfs_release_path(path);
+	return ret;
+invalid:
+	test_msg("Free space tree is invalid\n");
+	ret = -EINVAL;
+	goto out;
+}
+
+static int check_free_space_extents(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *cache,
+				    struct btrfs_path *path,
+				    struct free_space_extent *extents,
+				    unsigned int num_extents)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	int ret;
+
+	info = search_free_space_info(trans, fs_info, cache, path, 0);
+	if (IS_ERR(info)) {
+		test_msg("Could not find free space info\n");
+		btrfs_release_path(path);
+		return PTR_ERR(info);
+	}
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	ret = __check_free_space_extents(trans, fs_info, cache, path, extents,
+					 num_extents);
+	if (ret)
+		return ret;
+
+	/* Flip it to the other format and check that for good measure. */
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		ret = convert_free_space_to_extents(trans, fs_info, cache, path);
+		if (ret) {
+			test_msg("Could not convert to extents\n");
+			return ret;
+		}
+	} else {
+		ret = convert_free_space_to_bitmaps(trans, fs_info, cache, path);
+		if (ret) {
+			test_msg("Could not convert to bitmaps\n");
+			return ret;
+		}
+	}
+	return __check_free_space_extents(trans, fs_info, cache, path, extents,
+					  num_extents);
+}
+
+static int test_empty_block_group(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *cache,
+				  struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid, cache->key.offset},
+	};
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_remove_all(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   struct btrfs_block_group_cache *cache,
+			   struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid,
+					    cache->key.offset);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_remove_beginning(struct btrfs_trans_handle *trans,
+				 struct btrfs_fs_info *fs_info,
+				 struct btrfs_block_group_cache *cache,
+				 struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid + BITMAP_RANGE,
+			cache->key.offset - BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid, BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+
+}
+
+static int test_remove_end(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   struct btrfs_block_group_cache *cache,
+			   struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid, cache->key.offset - BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid +
+					    cache->key.offset - BITMAP_RANGE,
+					    BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_remove_middle(struct btrfs_trans_handle *trans,
+			      struct btrfs_fs_info *fs_info,
+			      struct btrfs_block_group_cache *cache,
+			      struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid, BITMAP_RANGE},
+		{cache->key.objectid + 2 * BITMAP_RANGE,
+			cache->key.offset - 2 * BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid + BITMAP_RANGE,
+					    BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_merge_left(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   struct btrfs_block_group_cache *cache,
+			   struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid, 2 * BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid,
+					    cache->key.offset);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid, BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_merge_right(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   struct btrfs_block_group_cache *cache,
+			   struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid + BITMAP_RANGE, 2 * BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid,
+					    cache->key.offset);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + 2 * BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_merge_both(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   struct btrfs_block_group_cache *cache,
+			   struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid, 3 * BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid,
+					    cache->key.offset);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid, BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + 2 * BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+static int test_merge_none(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   struct btrfs_block_group_cache *cache,
+			   struct btrfs_path *path)
+{
+	struct free_space_extent extents[] = {
+		{cache->key.objectid, BITMAP_RANGE},
+		{cache->key.objectid + 2 * BITMAP_RANGE, BITMAP_RANGE},
+		{cache->key.objectid + 4 * BITMAP_RANGE, BITMAP_RANGE},
+	};
+	int ret;
+
+	ret = __remove_from_free_space_tree(trans, fs_info, cache, path,
+					    cache->key.objectid,
+					    cache->key.offset);
+	if (ret) {
+		test_msg("Could not remove free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid, BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + 4 * BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	ret = __add_to_free_space_tree(trans, fs_info, cache, path,
+				       cache->key.objectid + 2 * BITMAP_RANGE,
+				       BITMAP_RANGE);
+	if (ret) {
+		test_msg("Could not add free space\n");
+		return ret;
+	}
+
+	return check_free_space_extents(trans, fs_info, cache, path,
+					extents, ARRAY_SIZE(extents));
+}
+
+typedef int (*test_func_t)(struct btrfs_trans_handle *,
+			   struct btrfs_fs_info *,
+			   struct btrfs_block_group_cache *,
+			   struct btrfs_path *);
+
+static int run_test(test_func_t test_func, int bitmaps)
+{
+	struct btrfs_root *root = NULL;
+	struct btrfs_block_group_cache *cache = NULL;
+	struct btrfs_trans_handle trans;
+	struct btrfs_path *path = NULL;
+	int ret;
+
+	root = btrfs_alloc_dummy_root();
+	if (IS_ERR(root)) {
+		test_msg("Couldn't allocate dummy root\n");
+		ret = PTR_ERR(root);
+		goto out;
+	}
+
+	root->fs_info = btrfs_alloc_dummy_fs_info();
+	if (!root->fs_info) {
+		test_msg("Couldn't allocate dummy fs info\n");
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	btrfs_set_super_compat_ro_flags(root->fs_info->super_copy,
+					BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE);
+	root->fs_info->free_space_root = root;
+	root->fs_info->tree_root = root;
+
+	root->node = alloc_test_extent_buffer(root->fs_info, 4096);
+	if (!root->node) {
+		test_msg("Couldn't allocate dummy buffer\n");
+		ret = -ENOMEM;
+		goto out;
+	}
+	btrfs_set_header_level(root->node, 0);
+	btrfs_set_header_nritems(root->node, 0);
+	root->alloc_bytenr += 8192;
+
+	cache = btrfs_alloc_dummy_block_group(8 * BITMAP_RANGE);
+	if (!cache) {
+		test_msg("Couldn't allocate dummy block group cache\n");
+		ret = -ENOMEM;
+		goto out;
+	}
+	cache->bitmap_low_thresh = 0;
+	cache->bitmap_high_thresh = (u32)-1;
+	cache->needs_free_space = 1;
+
+	btrfs_init_dummy_trans(&trans);
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		test_msg("Couldn't allocate path\n");
+		return -ENOMEM;
+	}
+
+	ret = add_block_group_free_space(&trans, root->fs_info, cache);
+	if (ret) {
+		test_msg("Could not add block group free space\n");
+		goto out;
+	}
+
+	if (bitmaps) {
+		ret = convert_free_space_to_bitmaps(&trans, root->fs_info,
+						    cache, path);
+		if (ret) {
+			test_msg("Could not convert block group to bitmaps\n");
+			goto out;
+		}
+	}
+
+	ret = test_func(&trans, root->fs_info, cache, path);
+	if (ret)
+		goto out;
+
+	ret = remove_block_group_free_space(&trans, root->fs_info, cache);
+	if (ret) {
+		test_msg("Could not remove block group free space\n");
+		goto out;
+	}
+
+	if (btrfs_header_nritems(root->node) != 0) {
+		test_msg("Free space tree has leftover items\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	btrfs_free_dummy_block_group(cache);
+	btrfs_free_dummy_root(root);
+	return ret;
+}
+
+static int run_test_both_formats(test_func_t test_func)
+{
+	int ret;
+
+	ret = run_test(test_func, 0);
+	if (ret)
+		return ret;
+	return run_test(test_func, 1);
+}
+
+int btrfs_test_free_space_tree(void)
+{
+	test_func_t tests[] = {
+		test_empty_block_group,
+		test_remove_all,
+		test_remove_beginning,
+		test_remove_end,
+		test_remove_middle,
+		test_merge_left,
+		test_merge_right,
+		test_merge_both,
+		test_merge_none,
+	};
+	int i;
+
+	test_msg("Running free space tree tests\n");
+	for (i = 0; i < ARRAY_SIZE(tests); i++) {
+		int ret = run_test_both_formats(tests[i]);
+		if (ret) {
+			test_msg("%pf failed\n", tests[i]);
+			return ret;
+		}
+	}
+
+	return 0;
+}
diff --git a/fs/btrfs/tests/qgroup-tests.c b/fs/btrfs/tests/qgroup-tests.c
index 846d277b1901..8ea5d34bc5a2 100644
--- a/fs/btrfs/tests/qgroup-tests.c
+++ b/fs/btrfs/tests/qgroup-tests.c
@@ -23,14 +23,6 @@
 #include "../qgroup.h"
 #include "../backref.h"
 
-static void init_dummy_trans(struct btrfs_trans_handle *trans)
-{
-	memset(trans, 0, sizeof(*trans));
-	trans->transid = 1;
-	INIT_LIST_HEAD(&trans->qgroup_ref_list);
-	trans->type = __TRANS_DUMMY;
-}
-
 static int insert_normal_tree_ref(struct btrfs_root *root, u64 bytenr,
 				  u64 num_bytes, u64 parent, u64 root_objectid)
 {
@@ -44,7 +36,7 @@ static int insert_normal_tree_ref(struct btrfs_root *root, u64 bytenr,
 	u32 size = sizeof(*item) + sizeof(*iref) + sizeof(*block_info);
 	int ret;
 
-	init_dummy_trans(&trans);
+	btrfs_init_dummy_trans(&trans);
 
 	ins.objectid = bytenr;
 	ins.type = BTRFS_EXTENT_ITEM_KEY;
@@ -94,7 +86,7 @@ static int add_tree_ref(struct btrfs_root *root, u64 bytenr, u64 num_bytes,
 	u64 refs;
 	int ret;
 
-	init_dummy_trans(&trans);
+	btrfs_init_dummy_trans(&trans);
 
 	key.objectid = bytenr;
 	key.type = BTRFS_EXTENT_ITEM_KEY;
@@ -144,7 +136,7 @@ static int remove_extent_item(struct btrfs_root *root, u64 bytenr,
 	struct btrfs_path *path;
 	int ret;
 
-	init_dummy_trans(&trans);
+	btrfs_init_dummy_trans(&trans);
 
 	key.objectid = bytenr;
 	key.type = BTRFS_EXTENT_ITEM_KEY;
@@ -178,7 +170,7 @@ static int remove_extent_ref(struct btrfs_root *root, u64 bytenr,
 	u64 refs;
 	int ret;
 
-	init_dummy_trans(&trans);
+	btrfs_init_dummy_trans(&trans);
 
 	key.objectid = bytenr;
 	key.type = BTRFS_EXTENT_ITEM_KEY;
@@ -232,7 +224,7 @@ static int test_no_shared_qgroup(struct btrfs_root *root)
 	struct ulist *new_roots = NULL;
 	int ret;
 
-	init_dummy_trans(&trans);
+	btrfs_init_dummy_trans(&trans);
 
 	test_msg("Qgroup basic add\n");
 	ret = btrfs_create_qgroup(NULL, fs_info, 5);
@@ -326,7 +318,7 @@ static int test_multiple_refs(struct btrfs_root *root)
 	struct ulist *new_roots = NULL;
 	int ret;
 
-	init_dummy_trans(&trans);
+	btrfs_init_dummy_trans(&trans);
 
 	test_msg("Qgroup multiple refs test\n");
 
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index be463b7f1f30..34961f02e58c 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -75,6 +75,23 @@ void btrfs_put_transaction(struct btrfs_transaction *transaction)
 			list_del_init(&em->list);
 			free_extent_map(em);
 		}
+		/*
+		 * If any block groups are found in ->deleted_bgs then it's
+		 * because the transaction was aborted and a commit did not
+		 * happen (things failed before writing the new superblock
+		 * and calling btrfs_finish_extent_commit()), so we can not
+		 * discard the physical locations of the block groups.
+		 */
+		while (!list_empty(&transaction->deleted_bgs)) {
+			struct btrfs_block_group_cache *cache;
+
+			cache = list_first_entry(&transaction->deleted_bgs,
+						 struct btrfs_block_group_cache,
+						 bg_list);
+			list_del_init(&cache->bg_list);
+			btrfs_put_block_group_trimming(cache);
+			btrfs_put_block_group(cache);
+		}
 		kmem_cache_free(btrfs_transaction_cachep, transaction);
 	}
 }
diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index f31db4325339..cb65089127cc 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -89,6 +89,12 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		goto out;
 	}
 	btrfs_release_path(path);
+	/*
+	 * We don't need a lock on a leaf. btrfs_realloc_node() will lock all
+	 * leafs from path->nodes[1], so set lowest_level to 1 to avoid later
+	 * a deadlock (attempting to write lock an already write locked leaf).
+	 */
+	path->lowest_level = 1;
 	wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 
 	if (wret < 0) {
@@ -99,9 +105,12 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		ret = 0;
 		goto out;
 	}
-	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
-	next_key_ret = btrfs_find_next_key(root, path, &key, 1,
-					   min_trans);
+	/*
+	 * The node at level 1 must always be locked when our path has
+	 * keep_locks set and lowest_level is 1, regardless of the value of
+	 * path->slots[1].
+	 */
+	BUG_ON(path->locks[1] == 0);
 	ret = btrfs_realloc_node(trans, root,
 				 path->nodes[1], 0,
 				 &last_ret,
@@ -110,6 +119,18 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		WARN_ON(ret == -EAGAIN);
 		goto out;
 	}
+	/*
+	 * Now that we reallocated the node we can find the next key. Note that
+	 * btrfs_find_next_key() can release our path and do another search
+	 * without COWing, this is because even with path->keep_locks = 1,
+	 * btrfs_search_slot() / ctree.c:unlock_up() does not keeps a lock on a
+	 * node when path->slots[node_level - 1] does not point to the last
+	 * item or a slot beyond the last item (ctree.c:unlock_up()). Therefore
+	 * we search for the next key after reallocating our node.
+	 */
+	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
+	next_key_ret = btrfs_find_next_key(root, path, &key, 1,
+					   min_trans);
 	if (next_key_ret == 0) {
 		memcpy(&root->defrag_progress, &key, sizeof(key));
 		ret = -EAGAIN;
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 574a717fcb6e..2fa5ba814120 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -125,6 +125,7 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
+static void btrfs_close_one_device(struct btrfs_device *device);
 
 DEFINE_MUTEX(uuid_mutex);
 static LIST_HEAD(fs_uuids);
@@ -4792,7 +4793,7 @@ int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
 	u64 dev_offset;
 	u64 stripe_size;
 	int i = 0;
-	int ret;
+	int ret = 0;
 
 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
 	read_lock(&em_tree->lock);
@@ -4823,20 +4824,32 @@ int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
 		goto out;
 	}
 
+	/*
+	 * Take the device list mutex to prevent races with the final phase of
+	 * a device replace operation that replaces the device object associated
+	 * with the map's stripes, because the device object's id can change
+	 * at any time during that final phase of the device replace operation
+	 * (dev-replace.c:btrfs_dev_replace_finishing()).
+	 */
+	mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
 	for (i = 0; i < map->num_stripes; i++) {
 		device = map->stripes[i].dev;
 		dev_offset = map->stripes[i].physical;
 
 		ret = btrfs_update_device(trans, device);
 		if (ret)
-			goto out;
+			break;
 		ret = btrfs_alloc_dev_extent(trans, device,
 					     chunk_root->root_key.objectid,
 					     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 					     chunk_offset, dev_offset,
 					     stripe_size);
 		if (ret)
-			goto out;
+			break;
+	}
+	if (ret) {
+		mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
+		goto out;
 	}
 
 	stripe = &chunk->stripe;
@@ -4849,6 +4862,7 @@ int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 		stripe++;
 	}
+	mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
 
 	btrfs_set_stack_chunk_length(chunk, chunk_size);
 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
@@ -6464,11 +6478,11 @@ int btrfs_read_sys_array(struct btrfs_root *root)
 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
 	if (!sb)
 		return -ENOMEM;
-	btrfs_set_buffer_uptodate(sb);
+	set_extent_buffer_uptodate(sb);
 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
 	/*
 	 * The sb extent buffer is artifical and just used to read the system array.
-	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
+	 * set_extent_buffer_uptodate() call does not properly mark all it's
 	 * pages up-to-date when the page is larger: extent does not cover the
 	 * whole page and consequently check_page_uptodate does not find all
 	 * the page's extents up-to-date (the hole beyond sb),
@@ -6959,7 +6973,7 @@ void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
 	}
 }
 
-void btrfs_close_one_device(struct btrfs_device *device)
+static void btrfs_close_one_device(struct btrfs_device *device)
 {
 	struct btrfs_fs_devices *fs_devices = device->fs_devices;
 	struct btrfs_device *new_device;
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index d5c84f6b1353..6a4375a7c588 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -566,6 +566,5 @@ static inline void unlock_chunks(struct btrfs_root *root)
 struct list_head *btrfs_get_fs_uuids(void);
 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info);
 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info);
-void btrfs_close_one_device(struct btrfs_device *device);
 
 #endif
diff --git a/fs/btrfs/xattr.c b/fs/btrfs/xattr.c
index 28b4e7e71e61..608552ed89c0 100644
--- a/fs/btrfs/xattr.c
+++ b/fs/btrfs/xattr.c
@@ -494,7 +494,7 @@ static int btrfs_initxattrs(struct inode *inode,
 
 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
 		name = kmalloc(XATTR_SECURITY_PREFIX_LEN +
-			       strlen(xattr->name) + 1, GFP_NOFS);
+			       strlen(xattr->name) + 1, GFP_KERNEL);
 		if (!name) {
 			err = -ENOMEM;
 			break;