Btrfs: allocator improvements, inode block groups

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/btrfs_inode.h

@@ -3,6 +3,7 @@
 
 struct btrfs_inode {
 	struct btrfs_root *root;
+	struct btrfs_block_group_cache *block_group;
 	struct btrfs_key location;
 	struct inode vfs_inode;
 };

fs/btrfs/ctree.c

@@ -61,7 +61,7 @@ static int btrfs_cow_block(struct btrfs_trans_handle *trans, struct btrfs_root
 		*cow_ret = buf;
 		return 0;
 	}
-	cow = btrfs_alloc_free_block(trans, root);
+	cow = btrfs_alloc_free_block(trans, root, buf->b_blocknr);
 	cow_node = btrfs_buffer_node(cow);
 	if (buf->b_size != root->blocksize || cow->b_size != root->blocksize)
 		WARN_ON(1);
@@ -800,7 +800,7 @@ static int insert_new_root(struct btrfs_trans_handle *trans, struct btrfs_root
 	BUG_ON(path->nodes[level]);
 	BUG_ON(path->nodes[level-1] != root->node);
 
-	t = btrfs_alloc_free_block(trans, root);
+	t = btrfs_alloc_free_block(trans, root, root->node->b_blocknr);
 	c = btrfs_buffer_node(t);
 	memset(c, 0, root->blocksize);
 	btrfs_set_header_nritems(&c->header, 1);
@@ -905,7 +905,7 @@ static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
 	}
 
 	c_nritems = btrfs_header_nritems(&c->header);
-	split_buffer = btrfs_alloc_free_block(trans, root);
+	split_buffer = btrfs_alloc_free_block(trans, root, t->b_blocknr);
 	split = btrfs_buffer_node(split_buffer);
 	btrfs_set_header_flags(&split->header, btrfs_header_flags(&c->header));
 	btrfs_set_header_level(&split->header, btrfs_header_level(&c->header));
@@ -1277,7 +1277,7 @@ static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
 	slot = path->slots[0];
 	nritems = btrfs_header_nritems(&l->header);
 	mid = (nritems + 1)/ 2;
-	right_buffer = btrfs_alloc_free_block(trans, root);
+	right_buffer = btrfs_alloc_free_block(trans, root, l_buf->b_blocknr);
 	BUG_ON(!right_buffer);
 	right = btrfs_buffer_leaf(right_buffer);
 	memset(&right->header, 0, sizeof(right->header));
@@ -1374,7 +1374,7 @@ static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
 
 	if (!double_split)
 		return ret;
-	right_buffer = btrfs_alloc_free_block(trans, root);
+	right_buffer = btrfs_alloc_free_block(trans, root, l_buf->b_blocknr);
 	BUG_ON(!right_buffer);
 	right = btrfs_buffer_leaf(right_buffer);
 	memset(&right->header, 0, sizeof(right->header));

fs/btrfs/ctree.h

@@ -174,6 +174,7 @@ struct btrfs_inode_item {
 	__le64 generation;
 	__le64 size;
 	__le64 nblocks;
+	__le64 block_group;
 	__le32 nlink;
 	__le32 uid;
 	__le32 gid;
@@ -241,6 +242,7 @@ struct btrfs_device_item {
 
 /* tag for the radix tree of block groups in ram */
 #define BTRFS_BLOCK_GROUP_DIRTY 0
+#define BTRFS_BLOCK_GROUP_AVAIL 1
 #define BTRFS_BLOCK_GROUP_HINTS 8
 #define BTRFS_BLOCK_GROUP_SIZE (256 * 1024 * 1024)
 struct btrfs_block_group_item {
@@ -410,6 +412,17 @@ static inline void btrfs_set_inode_nblocks(struct btrfs_inode_item *i, u64 val)
 	i->nblocks = cpu_to_le64(val);
 }
 
+static inline u64 btrfs_inode_block_group(struct btrfs_inode_item *i)
+{
+	return le64_to_cpu(i->block_group);
+}
+
+static inline void btrfs_set_inode_block_group(struct btrfs_inode_item *i,
+						u64 val)
+{
+	i->block_group = cpu_to_le64(val);
+}
+
 static inline u32 btrfs_inode_nlink(struct btrfs_inode_item *i)
 {
 	return le32_to_cpu(i->nlink);
@@ -1054,10 +1067,13 @@ static inline void btrfs_mark_buffer_dirty(struct buffer_head *bh)
 	btrfs_item_offset((leaf)->items + (slot))))
 
 /* extent-tree.c */
+struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
+						 struct btrfs_block_group_cache
+						 *hint, int data);
 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root);
 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
-					    struct btrfs_root *root);
+					    struct btrfs_root *root, u64 hint);
 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, u64 owner,
 		       u64 num_blocks, u64 search_start,

fs/btrfs/extent-tree.c

@@ -12,42 +12,57 @@ static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 			       btrfs_root *extent_root);
 
-static int find_search_start(struct btrfs_root *root, int data)
+struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
+						 struct btrfs_block_group_cache
+						 *hint, int data)
 {
 	struct btrfs_block_group_cache *cache[8];
+	struct btrfs_block_group_cache *found_group = NULL;
 	struct btrfs_fs_info *info = root->fs_info;
 	u64 used;
-	u64 last;
+	u64 last = 0;
+	u64 hint_last;
 	int i;
 	int ret;
-
-	cache[0] = info->block_group_cache;
-	if (!cache[0])
-		goto find_new;
-	used = btrfs_block_group_used(&cache[0]->item);
-	if (used < (cache[0]->key.offset * 3 / 2))
-		return 0;
-find_new:
+	int full_search = 0;
+	if (hint) {
+		used = btrfs_block_group_used(&hint->item);
+		if (used < (hint->key.offset * 2) / 3) {
+			return hint;
+		}
+		radix_tree_tag_clear(&info->block_group_radix,
+				     hint->key.objectid + hint->key.offset - 1,
+				     BTRFS_BLOCK_GROUP_AVAIL);
+		last = hint->key.objectid + hint->key.offset;
+		hint_last = last;
+	} else {
+		hint_last = 0;
 		last = 0;
+	}
 	while(1) {
 		ret = radix_tree_gang_lookup_tag(&info->block_group_radix,
 						 (void **)cache,
 						 last, ARRAY_SIZE(cache),
-						 BTRFS_BLOCK_GROUP_DIRTY);
+						 BTRFS_BLOCK_GROUP_AVAIL);
 		if (!ret)
 			break;
 		for (i = 0; i < ret; i++) {
 			used = btrfs_block_group_used(&cache[i]->item);
-			if (used < (cache[i]->key.offset * 3 / 2)) {
+			if (used < (cache[i]->key.offset * 2) / 3) {
 				info->block_group_cache = cache[i];
-				cache[i]->last_alloc = cache[i]->first_free;
-				return 0;
+				found_group = cache[i];
+				goto found;
 			}
+			radix_tree_tag_clear(&info->block_group_radix,
+					   cache[i]->key.objectid +
+					   cache[i]->key.offset - 1,
+					   BTRFS_BLOCK_GROUP_AVAIL);
 			last = cache[i]->key.objectid +
-				cache[i]->key.offset - 1;
+				cache[i]->key.offset;
 		}
 	}
-	last = 0;
+	last = hint_last;
+again:
 	while(1) {
 		ret = radix_tree_gang_lookup(&info->block_group_radix,
 						 (void **)cache,
@@ -56,17 +71,32 @@ static int find_search_start(struct btrfs_root *root, int data)
 			break;
 		for (i = 0; i < ret; i++) {
 			used = btrfs_block_group_used(&cache[i]->item);
-			if (used < (cache[i]->key.offset * 3 / 2)) {
+			if (used < cache[i]->key.offset) {
 				info->block_group_cache = cache[i];
-				cache[i]->last_alloc = cache[i]->first_free;
-				return 0;
+				found_group = cache[i];
+				goto found;
 			}
+			radix_tree_tag_clear(&info->block_group_radix,
+					   cache[i]->key.objectid +
+					   cache[i]->key.offset - 1,
+					   BTRFS_BLOCK_GROUP_AVAIL);
 			last = cache[i]->key.objectid +
-				cache[i]->key.offset - 1;
+				cache[i]->key.offset;
 		}
 	}
 	info->block_group_cache = NULL;
-	return 0;
+	if (!full_search) {
+		last = 0;
+		full_search = 1;
+		goto again;
+	}
+found:
+	if (!found_group) {
+		ret = radix_tree_gang_lookup(&info->block_group_radix,
+					     (void **)&found_group, 0, 1);
+		BUG_ON(ret != 1);
+	}
+	return found_group;
 }
 
 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
@@ -243,6 +273,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 						    path, cache[i]);
 			if (err)
 				werr = err;
+			cache[i]->last_alloc = cache[i]->first_free;
 		}
 	}
 	btrfs_free_path(path);
@@ -322,10 +353,6 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
 						    btree_inode->i_blkbits));
 		}
 	}
-	if (root->fs_info->block_group_cache) {
-		root->fs_info->block_group_cache->last_alloc =
-			root->fs_info->block_group_cache->first_free;
-	}
 	return 0;
 }
 
@@ -532,22 +559,43 @@ static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 	int total_found = 0;
 	int fill_prealloc = 0;
 	int level;
+	int update_block_group = 0;
+	struct btrfs_block_group_cache *hint_block_group;
 
 	path = btrfs_alloc_path();
 	ins->flags = 0;
 	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 
 	level = btrfs_header_level(btrfs_buffer_header(root->node));
+	/* find search start here */
+	if (0 && search_start && num_blocks) {
+		u64 used;
+		ret = radix_tree_gang_lookup(&info->block_group_radix,
+					     (void **)&hint_block_group,
+					     search_start, 1);
+		if (ret) {
+			used = btrfs_block_group_used(&hint_block_group->item);
+			if (used > (hint_block_group->key.offset * 9) / 10)
+				search_start = 0;
+			else if (search_start < hint_block_group->last_alloc)
+				search_start = hint_block_group->last_alloc;
+		} else {
+			search_start = 0;
+		}
+	}
 	if (num_blocks == 0) {
 		fill_prealloc = 1;
 		num_blocks = 1;
 		total_needed = (min(level + 1, BTRFS_MAX_LEVEL) + 2) * 3;
 	}
-	find_search_start(root, 0);
-	if (info->block_group_cache &&
-	    info->block_group_cache->last_alloc > search_start)
-		search_start = info->block_group_cache->last_alloc;
-
+	if (1 || !search_start) {
+		trans->block_group = btrfs_find_block_group(root,
+							    trans->block_group,
+							    0);
+		if (trans->block_group->last_alloc > search_start)
+			search_start = trans->block_group->last_alloc;
+		update_block_group = 1;
+	}
 check_failed:
 	btrfs_init_path(path);
 	ins->objectid = search_start;
@@ -662,11 +710,13 @@ static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 		}
 		info->extent_tree_prealloc_nr = total_found;
 	}
+	if (update_block_group) {
 		ret = radix_tree_gang_lookup(&info->block_group_radix,
-				     (void **)&info->block_group_cache,
+					     (void **)&trans->block_group,
 					     ins->objectid, 1);
 		if (ret) {
-		info->block_group_cache->last_alloc = ins->objectid;
+			trans->block_group->last_alloc = ins->objectid;
+		}
 	}
 	ins->offset = num_blocks;
 	btrfs_free_path(path);
@@ -747,14 +797,14 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
  * returns the tree buffer or NULL.
  */
 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
-					   struct btrfs_root *root)
+					   struct btrfs_root *root, u64 hint)
 {
 	struct btrfs_key ins;
 	int ret;
 	struct buffer_head *buf;
 
 	ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
-				 1, 0, (unsigned long)-1, &ins);
+				 1, hint, (unsigned long)-1, &ins);
 	if (ret) {
 		BUG();
 		return NULL;
@@ -975,6 +1025,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 	struct btrfs_key found_key;
 	struct btrfs_leaf *leaf;
 	u64 group_size_blocks = BTRFS_BLOCK_GROUP_SIZE / root->blocksize;
+	u64 used;
 
 	root = root->fs_info->extent_root;
 	key.objectid = 0;
@@ -1005,8 +1056,8 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 				    struct btrfs_block_group_item);
 		memcpy(&cache->item, bi, sizeof(*bi));
 		memcpy(&cache->key, &found_key, sizeof(found_key));
-		cache->last_alloc = 0;
-		cache->first_free = 0;
+		cache->last_alloc = cache->key.objectid;
+		cache->first_free = cache->key.objectid;
 		key.objectid = found_key.objectid + found_key.offset;
 		btrfs_release_path(root, path);
 		ret = radix_tree_insert(&root->fs_info->block_group_radix,
@@ -1014,6 +1065,13 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 					found_key.offset - 1,
 					(void *)cache);
 		BUG_ON(ret);
+		used = btrfs_block_group_used(bi);
+		if (used < (key.offset * 2) / 3) {
+			radix_tree_tag_set(&root->fs_info->block_group_radix,
+					   found_key.objectid +
+					   found_key.offset - 1,
+					   BTRFS_BLOCK_GROUP_AVAIL);
+		}
 		if (key.objectid >=
 		    btrfs_super_total_blocks(root->fs_info->disk_super))
 			break;

fs/btrfs/super.c

@@ -52,6 +52,8 @@ static void btrfs_read_locked_inode(struct inode *inode)
 	struct btrfs_inode_item *inode_item;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_key location;
+	struct btrfs_block_group_cache *alloc_group;
+	u64 alloc_group_block;
 	int ret;
 
 	path = btrfs_alloc_path();
@@ -82,6 +84,12 @@ static void btrfs_read_locked_inode(struct inode *inode)
 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
 	inode->i_blocks = btrfs_inode_nblocks(inode_item);
 	inode->i_generation = btrfs_inode_generation(inode_item);
+	alloc_group_block = btrfs_inode_block_group(inode_item);
+	ret = radix_tree_gang_lookup(&root->fs_info->block_group_radix,
+				     (void **)&alloc_group,
+				     alloc_group_block, 1);
+	BUG_ON(!ret);
+	BTRFS_I(inode)->block_group = alloc_group;
 
 	btrfs_free_path(path);
 	inode_item = NULL;
@@ -136,6 +144,8 @@ static void fill_inode_item(struct btrfs_inode_item *item,
 	btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 	btrfs_set_inode_nblocks(item, inode->i_blocks);
 	btrfs_set_inode_generation(item, inode->i_generation);
+	btrfs_set_inode_block_group(item,
+				    BTRFS_I(inode)->block_group->key.objectid);
 }
 
 
@@ -237,6 +247,7 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 	root = BTRFS_I(dir)->root;
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
 	ret = btrfs_unlink_trans(trans, root, dir, dentry);
 	btrfs_end_transaction(trans, root);
 	mutex_unlock(&root->fs_info->fs_mutex);
@@ -262,6 +273,7 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 	btrfs_init_path(path);
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
 	key.objectid = inode->i_ino;
 	key.offset = (u64)-1;
 	key.flags = (u32)-1;
@@ -429,6 +441,7 @@ static void btrfs_delete_inode(struct inode *inode)
 	inode->i_size = 0;
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
 	if (S_ISREG(inode->i_mode)) {
 		ret = btrfs_truncate_in_trans(trans, root, inode);
 		BUG_ON(ret);
@@ -731,6 +744,7 @@ static int btrfs_write_inode(struct inode *inode, int wait)
 	if (wait) {
 		mutex_lock(&root->fs_info->fs_mutex);
 		trans = btrfs_start_transaction(root, 1);
+		btrfs_set_trans_block_group(trans, inode);
 		ret = btrfs_commit_transaction(trans, root);
 		mutex_unlock(&root->fs_info->fs_mutex);
 	}
@@ -744,6 +758,7 @@ static void btrfs_dirty_inode(struct inode *inode)
 
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
 	btrfs_update_inode(trans, root, inode);
 	btrfs_end_transaction(trans, root);
 	mutex_unlock(&root->fs_info->fs_mutex);
@@ -751,7 +766,9 @@ static void btrfs_dirty_inode(struct inode *inode)
 
 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 				     struct btrfs_root *root,
-				     u64 objectid, int mode)
+				     u64 objectid,
+				     struct btrfs_block_group_cache *group,
+				     int mode)
 {
 	struct inode *inode;
 	struct btrfs_inode_item inode_item;
@@ -763,6 +780,8 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 		return ERR_PTR(-ENOMEM);
 
 	BTRFS_I(inode)->root = root;
+	group = btrfs_find_block_group(root, group, 0);
+	BTRFS_I(inode)->block_group = group;
 
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
@@ -832,6 +851,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
 
 	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 	if (err) {
@@ -839,11 +859,13 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 		goto out_unlock;
 	}
 
-	inode = btrfs_new_inode(trans, root, objectid, mode);
+	inode = btrfs_new_inode(trans, root, objectid,
+				BTRFS_I(dir)->block_group, mode);
 	err = PTR_ERR(inode);
 	if (IS_ERR(inode))
 		goto out_unlock;
-	// FIXME mark the inode dirty
+
+	btrfs_set_trans_block_group(trans, inode);
 	err = btrfs_add_nondir(trans, dentry, inode);
 	if (err)
 		drop_inode = 1;
@@ -853,6 +875,8 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 		inode->i_op = &btrfs_file_inode_operations;
 	}
 	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, inode);
+	btrfs_update_inode_block_group(trans, dir);
 out_unlock:
 	btrfs_end_transaction(trans, root);
 	mutex_unlock(&root->fs_info->fs_mutex);
@@ -904,6 +928,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, dir);
 	if (IS_ERR(trans)) {
 		err = PTR_ERR(trans);
 		goto out_unlock;
@@ -915,7 +940,8 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 		goto out_unlock;
 	}
 
-	inode = btrfs_new_inode(trans, root, objectid, S_IFDIR | mode);
+	inode = btrfs_new_inode(trans, root, objectid,
+				BTRFS_I(dir)->block_group, S_IFDIR | mode);
 	if (IS_ERR(inode)) {
 		err = PTR_ERR(inode);
 		goto out_fail;
@@ -923,6 +949,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 	drop_on_err = 1;
 	inode->i_op = &btrfs_dir_inode_operations;
 	inode->i_fop = &btrfs_dir_file_operations;
+	btrfs_set_trans_block_group(trans, inode);
 
 	err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
 	if (err)
@@ -938,6 +965,8 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 	d_instantiate(dentry, inode);
 	drop_on_err = 0;
 	dir->i_sb->s_dirt = 1;
+	btrfs_update_inode_block_group(trans, inode);
+	btrfs_update_inode_block_group(trans, dir);
 
 out_fail:
 	btrfs_end_transaction(trans, root);
@@ -1349,6 +1378,7 @@ static void btrfs_truncate(struct inode *inode)
 	/* FIXME, add redo link to tree so we don't leak on crash */
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
 	ret = btrfs_truncate_in_trans(trans, root, inode);
 	BUG_ON(ret);
 	ret = btrfs_end_transaction(trans, root);
@@ -1445,6 +1475,7 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 
 		mutex_lock(&root->fs_info->fs_mutex);
 		trans = btrfs_start_transaction(root, 1);
+		btrfs_set_trans_block_group(trans, inode);
 
 		bh = page_buffers(pages[i]);
 		if (buffer_mapped(bh) && bh->b_blocknr == 0) {
@@ -1481,6 +1512,7 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 			kunmap(pages[i]);
 		}
 		SetPageChecked(pages[i]);
+		btrfs_update_inode_block_group(trans, inode);
 		ret = btrfs_end_transaction(trans, root);
 		BUG_ON(ret);
 		mutex_unlock(&root->fs_info->fs_mutex);
@@ -1821,6 +1853,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		mutex_unlock(&root->fs_info->fs_mutex);
 		goto out_unlock;
 	}
+	btrfs_set_trans_block_group(trans, inode);
 	/* FIXME blocksize != 4096 */
 	inode->i_blocks += num_blocks << 3;
 	if (start_pos < inode->i_size) {
@@ -1845,6 +1878,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	}
 	BUG_ON(ret);
 	alloc_extent_start = ins.objectid;
+	btrfs_update_inode_block_group(trans, inode);
 	ret = btrfs_end_transaction(trans, root);
 	mutex_unlock(&root->fs_info->fs_mutex);
 
@@ -2017,6 +2051,7 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
 	struct btrfs_leaf *leaf;
 	struct btrfs_root *new_root;
 	struct inode *inode;
+	struct inode *dir;
 	int ret;
 	u64 objectid;
 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
@@ -2025,7 +2060,7 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
 	trans = btrfs_start_transaction(root, 1);
 	BUG_ON(!trans);
 
-	subvol = btrfs_alloc_free_block(trans, root);
+	subvol = btrfs_alloc_free_block(trans, root, 0);
 	if (subvol == NULL)
 		return -ENOSPC;
 	leaf = btrfs_buffer_leaf(subvol);
@@ -2069,10 +2104,9 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
 	 * insert the directory item
 	 */
 	key.offset = (u64)-1;
+	dir = root->fs_info->sb->s_root->d_inode;
 	ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
-				    name, namelen,
-				    root->fs_info->sb->s_root->d_inode->i_ino,
-				    &key, 0);
+				    name, namelen, dir->i_ino, &key, 0);
 	BUG_ON(ret);
 
 	ret = btrfs_commit_transaction(trans, root);
@@ -2084,7 +2118,8 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
 	trans = btrfs_start_transaction(new_root, 1);
 	BUG_ON(!trans);
 
-	inode = btrfs_new_inode(trans, new_root, new_dirid, S_IFDIR | 0700);
+	inode = btrfs_new_inode(trans, new_root, new_dirid,
+				BTRFS_I(dir)->block_group, S_IFDIR | 0700);
 	inode->i_op = &btrfs_dir_inode_operations;
 	inode->i_fop = &btrfs_dir_file_operations;
 

fs/btrfs/transaction.c

@@ -77,6 +77,7 @@ struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 	h->transaction = root->fs_info->running_transaction;
 	h->blocks_reserved = num_blocks;
 	h->blocks_used = 0;
+	h->block_group = NULL;
 	root->fs_info->running_transaction->use_count++;
 	mutex_unlock(&root->fs_info->trans_mutex);
 	h->magic = h->magic2 = TRANS_MAGIC;

fs/btrfs/transaction.h

 #ifndef __TRANSACTION__
 #define __TRANSACTION__
+#include "btrfs_inode.h"
 
 struct btrfs_transaction {
 	u64 transid;
@@ -20,10 +21,24 @@ struct btrfs_trans_handle {
 	unsigned long blocks_reserved;
 	unsigned long blocks_used;
 	struct btrfs_transaction *transaction;
+	struct btrfs_block_group_cache *block_group;
 	int magic2;
 };
 
 
+static inline void btrfs_set_trans_block_group(struct btrfs_trans_handle *trans,
+					       struct inode *inode)
+{
+	trans->block_group = BTRFS_I(inode)->block_group;
+}
+
+static inline void btrfs_update_inode_block_group(struct
+						  btrfs_trans_handle *trans,
+						  struct inode *inode)
+{
+	BTRFS_I(inode)->block_group = trans->block_group;
+}
+
 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root);
 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,