Btrfs: groundwork for subvolume and snapshot roots

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

6 files changed, 282 insertions, 99 deletions

diff --git a/fs/btrfs/bit-radix.c b/fs/btrfs/bit-radix.c
index 783f54c..9fc42e9 100644
--- a/fs/btrfs/bit-radix.c
+++ b/fs/btrfs/bit-radix.c
@@ -86,7 +86,7 @@ int find_first_radix_bit(struct radix_tree_root *radix, unsigned long *retbits,
 	int i;
 	int total_found = 0;
 
-	ret = radix_tree_gang_lookup(radix, (void *)&gang, 0, ARRAY_SIZE(gang));
+	ret = radix_tree_gang_lookup(radix, (void **)gang, 0, ARRAY_SIZE(gang));
 	for (i = 0; i < ret && nr > 0; i++) {
 		found = 0;
 		bits = gang[i];
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 1ff5b99..5460030 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -232,12 +232,12 @@ struct btrfs_inode_map_item {
 
 struct crypto_hash;
 struct btrfs_fs_info {
-	struct btrfs_root *fs_root;
 	struct btrfs_root *extent_root;
 	struct btrfs_root *tree_root;
 	struct btrfs_root *inode_root;
 	struct btrfs_key current_insert;
 	struct btrfs_key last_insert;
+	struct radix_tree_root fs_roots_radix;
 	struct radix_tree_root pending_del_radix;
 	struct radix_tree_root pinned_radix;
 	u64 last_inode_alloc;
@@ -266,6 +266,9 @@ struct btrfs_root {
 	struct btrfs_root_item root_item;
 	struct btrfs_key root_key;
 	struct btrfs_fs_info *fs_info;
+	struct inode *inode;
+	u64 objectid;
+	u64 last_trans;
 	u32 blocksize;
 	int ref_cows;
 	u32 type;
@@ -595,7 +598,7 @@ static inline u32 btrfs_key_overflow(struct btrfs_key *key)
 
 static inline void btrfs_set_key_overflow(struct btrfs_key *key, u32 over)
 {
-	BUG_ON(over > BTRFS_KEY_OVERFLOW_MAX);
+	BUG_ON(over >= BTRFS_KEY_OVERFLOW_MAX);
 	over = over << BTRFS_KEY_OVERFLOW_SHIFT;
 	key->flags = (key->flags & ~((u64)BTRFS_KEY_OVERFLOW_MASK)) | over;
 }
@@ -634,7 +637,7 @@ static inline void btrfs_set_disK_key_overflow(struct btrfs_disk_key *key,
 					       u32 over)
 {
 	u32 flags = btrfs_disk_key_flags(key);
-	BUG_ON(over > BTRFS_KEY_OVERFLOW_MAX);
+	BUG_ON(over >= BTRFS_KEY_OVERFLOW_MAX);
 	over = over << BTRFS_KEY_OVERFLOW_SHIFT;
 	flags = (flags & ~((u64)BTRFS_KEY_OVERFLOW_MASK)) | over;
 	btrfs_set_disk_key_flags(key, flags);
@@ -748,6 +751,17 @@ static inline void btrfs_set_super_blocknr(struct btrfs_super_block *s, u64 val)
 	s->blocknr = cpu_to_le64(val);
 }
 
+static inline u64 btrfs_super_generation(struct btrfs_super_block *s)
+{
+	return le64_to_cpu(s->generation);
+}
+
+static inline void btrfs_set_super_generation(struct btrfs_super_block *s,
+					      u64 val)
+{
+	s->generation = cpu_to_le64(val);
+}
+
 static inline u64 btrfs_super_root(struct btrfs_super_block *s)
 {
 	return le64_to_cpu(s->root);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index b9301a5..b557bdd 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4,9 +4,11 @@
 #include <linux/crypto.h>
 #include <linux/scatterlist.h>
 #include <linux/swap.h>
+#include <linux/radix-tree.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
+#include "btrfs_inode.h"
 
 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
 {
@@ -180,7 +182,7 @@ static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 {
 	struct buffer_head *bh;
-	struct btrfs_root *root = btrfs_sb(page->mapping->host->i_sb);
+	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 	struct buffer_head *head;
 	if (!page_has_buffers(page)) {
 		create_empty_buffers(page, root->fs_info->sb->s_blocksize,
@@ -259,10 +261,13 @@ static int __setup_root(int blocksize,
 			u64 objectid)
 {
 	root->node = NULL;
+	root->inode = NULL;
 	root->commit_root = NULL;
 	root->blocksize = blocksize;
 	root->ref_cows = 0;
 	root->fs_info = fs_info;
+	root->objectid = objectid;
+	root->last_trans = 0;
 	memset(&root->root_key, 0, sizeof(root->root_key));
 	memset(&root->root_item, 0, sizeof(root->root_item));
 	return 0;
@@ -287,10 +292,78 @@ static int find_and_setup_root(int blocksize,
 	return 0;
 }
 
+struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
+				      struct btrfs_key *location)
+{
+	struct btrfs_root *root;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_path *path;
+	struct btrfs_leaf *l;
+	int ret = 0;
+
+printk("read_fs_root looking for %Lu %Lu %u\n", location->objectid, location->offset, location->flags);
+	root = kmalloc(sizeof(*root), GFP_NOFS);
+	if (!root) {
+printk("failed1\n");
+		return ERR_PTR(-ENOMEM);
+	}
+	if (location->offset == (u64)-1) {
+		ret = find_and_setup_root(fs_info->sb->s_blocksize,
+					  fs_info->tree_root, fs_info,
+					  location->objectid, root);
+		if (ret) {
+printk("failed2\n");
+			kfree(root);
+			return ERR_PTR(ret);
+		}
+		goto insert;
+	}
+
+	__setup_root(fs_info->sb->s_blocksize, root, fs_info,
+		     location->objectid);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
+	if (ret != 0) {
+printk("internal search_slot gives us %d\n", ret);
+		if (ret > 0)
+			ret = -ENOENT;
+		goto out;
+	}
+	l = btrfs_buffer_leaf(path->nodes[0]);
+	memcpy(&root->root_item,
+	       btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
+	       sizeof(root->root_item));
+	memcpy(&root->root_key, location, sizeof(*location));
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	btrfs_free_path(path);
+	if (ret) {
+		kfree(root);
+		return ERR_PTR(ret);
+	}
+	root->node = read_tree_block(root,
+				     btrfs_root_blocknr(&root->root_item));
+	BUG_ON(!root->node);
+insert:
+printk("inserting %p\n", root);
+	root->ref_cows = 1;
+	ret = radix_tree_insert(&fs_info->fs_roots_radix, (unsigned long)root,
+				root);
+	if (ret) {
+printk("radix_tree_insert gives us %d\n", ret);
+		brelse(root->node);
+		kfree(root);
+		return ERR_PTR(ret);
+	}
+printk("all worked\n");
+	return root;
+}
+
 struct btrfs_root *open_ctree(struct super_block *sb)
 {
-	struct btrfs_root *root = kmalloc(sizeof(struct btrfs_root),
-					  GFP_NOFS);
 	struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 						 GFP_NOFS);
 	struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
@@ -304,9 +377,9 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 
 	init_bit_radix(&fs_info->pinned_radix);
 	init_bit_radix(&fs_info->pending_del_radix);
+	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 	sb_set_blocksize(sb, 4096);
 	fs_info->running_transaction = NULL;
-	fs_info->fs_root = root;
 	fs_info->tree_root = tree_root;
 	fs_info->extent_root = extent_root;
 	fs_info->inode_root = inode_root;
@@ -318,6 +391,9 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 	fs_info->btree_inode->i_nlink = 1;
 	fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
+	BTRFS_I(fs_info->btree_inode)->root = tree_root;
+	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
+	       sizeof(struct btrfs_key));
 	insert_inode_hash(fs_info->btree_inode);
 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 	fs_info->hash_tfm = crypto_alloc_hash("sha256", 0, CRYPTO_ALG_ASYNC);
@@ -337,13 +413,12 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 					     BTRFS_SUPER_INFO_OFFSET /
 					     sb->s_blocksize);
 
-	if (!fs_info->sb_buffer) {
+	if (!fs_info->sb_buffer)
 		return NULL;
-	}
 	disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
-	if (!btrfs_super_root(disk_super)) {
+	if (!btrfs_super_root(disk_super))
 		return NULL;
-	}
+
 	fs_info->disk_super = disk_super;
 	tree_root->node = read_tree_block(tree_root,
 					  btrfs_super_root(disk_super));
@@ -358,14 +433,8 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 				  BTRFS_INODE_MAP_OBJECTID, inode_root);
 	BUG_ON(ret);
 
-	ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
-				  BTRFS_FS_TREE_OBJECTID, root);
-	BUG_ON(ret);
-	root->commit_root = root->node;
-	get_bh(root->node);
-	root->ref_cows = 1;
-	root->fs_info->generation = root->root_key.offset + 1;
-	ret = btrfs_find_highest_inode(root, &root->fs_info->last_inode_alloc);
+	fs_info->generation = btrfs_super_generation(disk_super) + 1;
+	ret = btrfs_find_highest_inode(tree_root, &fs_info->last_inode_alloc);
 	if (ret == 0)
 		fs_info->highest_inode = fs_info->last_inode_alloc;
 	memset(&fs_info->kobj, 0, sizeof(fs_info->kobj));
@@ -373,7 +442,7 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 	kobject_set_name(&fs_info->kobj, "%s", sb->s_id);
 	kobject_register(&fs_info->kobj);
 	mutex_unlock(&fs_info->fs_mutex);
-	return root;
+	return tree_root;
 }
 
 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
@@ -398,12 +467,42 @@ int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 	return 0;
 }
 
+int del_fs_roots(struct btrfs_fs_info *fs_info)
+{
+	int ret;
+	struct btrfs_root *gang[8];
+	int i;
+
+	while(1) {
+		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
+					     (void **)gang, 0,
+					     ARRAY_SIZE(gang));
+		if (!ret)
+			break;
+		for (i = 0; i < ret; i++) {
+			radix_tree_delete(&fs_info->fs_roots_radix,
+					  (unsigned long)gang[i]);
+			if (gang[i]->inode)
+				iput(gang[i]->inode);
+			else
+				printk("no inode for root %p\n", gang[i]);
+			if (gang[i]->node)
+				brelse(gang[i]->node);
+			if (gang[i]->commit_root)
+				brelse(gang[i]->commit_root);
+			kfree(gang[i]);
+		}
+	}
+	return 0;
+}
+
 int close_ctree(struct btrfs_root *root)
 {
 	int ret;
 	struct btrfs_trans_handle *trans;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 
-	mutex_lock(&root->fs_info->fs_mutex);
+	mutex_lock(&fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
 	btrfs_commit_transaction(trans, root);
 	/* run commit again to  drop the original snapshot */
@@ -412,29 +511,26 @@ int close_ctree(struct btrfs_root *root)
 	ret = btrfs_write_and_wait_transaction(NULL, root);
 	BUG_ON(ret);
 	write_ctree_super(NULL, root);
-	mutex_unlock(&root->fs_info->fs_mutex);
-
-	if (root->node)
-		btrfs_block_release(root, root->node);
-	if (root->fs_info->extent_root->node)
-		btrfs_block_release(root->fs_info->extent_root,
-				    root->fs_info->extent_root->node);
-	if (root->fs_info->inode_root->node)
-		btrfs_block_release(root->fs_info->inode_root,
-				    root->fs_info->inode_root->node);
-	if (root->fs_info->tree_root->node)
-		btrfs_block_release(root->fs_info->tree_root,
-				    root->fs_info->tree_root->node);
-	btrfs_block_release(root, root->commit_root);
-	btrfs_block_release(root, root->fs_info->sb_buffer);
-	crypto_free_hash(root->fs_info->hash_tfm);
-	truncate_inode_pages(root->fs_info->btree_inode->i_mapping, 0);
-	iput(root->fs_info->btree_inode);
-	kfree(root->fs_info->extent_root);
-	kfree(root->fs_info->inode_root);
-	kfree(root->fs_info->tree_root);
-	kobject_unregister(&root->fs_info->kobj);
-	kfree(root);
+	mutex_unlock(&fs_info->fs_mutex);
+
+	if (fs_info->extent_root->node)
+		btrfs_block_release(fs_info->extent_root,
+				    fs_info->extent_root->node);
+	if (fs_info->inode_root->node)
+		btrfs_block_release(fs_info->inode_root,
+				    fs_info->inode_root->node);
+	if (fs_info->tree_root->node)
+		btrfs_block_release(fs_info->tree_root,
+				    fs_info->tree_root->node);
+	btrfs_block_release(root, fs_info->sb_buffer);
+	crypto_free_hash(fs_info->hash_tfm);
+	truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
+	iput(fs_info->btree_inode);
+	del_fs_roots(fs_info);
+	kfree(fs_info->extent_root);
+	kfree(fs_info->inode_root);
+	kfree(fs_info->tree_root);
+	kobject_unregister(&fs_info->kobj);
 	return 0;
 }
 
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index ac6764b..0ef6e6f 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -39,4 +39,6 @@ int write_ctree_super(struct btrfs_trans_handle *trans,
 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr);
 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 		    char *result);
+struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
+				      struct btrfs_key *location);
 #endif
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 66d9fb2..3c9236c 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -436,7 +436,6 @@ int fixup_tree_root_location(struct btrfs_root *root,
 {
 	struct btrfs_path *path;
 	struct btrfs_root_item *ri;
-	int ret;
 
 	if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 		return 0;
@@ -447,22 +446,19 @@ int fixup_tree_root_location(struct btrfs_root *root,
 	BUG_ON(!path);
 	mutex_lock(&root->fs_info->fs_mutex);
 
-	ret = btrfs_lookup_inode(NULL, root, path, location, 0);
-	if (ret)
-		goto out;
-	ri = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
-			  path->slots[0],
-			  struct btrfs_root_item);
+	*sub_root = btrfs_read_fs_root(root->fs_info, location);
+	if (IS_ERR(*sub_root))
+		return PTR_ERR(*sub_root);
+
+	ri = &(*sub_root)->root_item;
 	location->objectid = btrfs_root_dirid(ri);
 	location->flags = 0;
 	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 	location->offset = 0;
-	/* FIXME properly select the root */
-	*sub_root = root->fs_info->fs_root;
-out:
+
 	btrfs_free_path(path);
 	mutex_unlock(&root->fs_info->fs_mutex);
-	return ret;
+	return 0;
 }
 
 
@@ -494,6 +490,15 @@ static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 		if (!inode)
 			return ERR_PTR(-EACCES);
 		if (inode->i_state & I_NEW) {
+			if (sub_root != root) {
+				ret = radix_tree_insert(
+						&root->fs_info->fs_roots_radix,
+						(unsigned long)sub_root,
+						sub_root);
+printk("adding new root for inode %lu\n", inode->i_ino);
+				igrab(inode);
+				sub_root->inode = inode;
+			}
 			BTRFS_I(inode)->root = sub_root;
 			memcpy(&BTRFS_I(inode)->location, &location,
 			       sizeof(location));
@@ -605,7 +610,7 @@ static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 	struct inode * inode;
 	struct dentry * root_dentry;
 	struct btrfs_super_block *disk_super;
-	struct btrfs_root *root;
+	struct btrfs_root *tree_root;
 	struct btrfs_inode *bi;
 
 	sb->s_maxbytes = MAX_LFS_FILESIZE;
@@ -613,14 +618,14 @@ static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 	sb->s_op = &btrfs_super_ops;
 	sb->s_time_gran = 1;
 
-	root = open_ctree(sb);
+	tree_root = open_ctree(sb);
 
-	if (!root) {
+	if (!tree_root) {
 		printk("btrfs: open_ctree failed\n");
 		return -EIO;
 	}
-	sb->s_fs_info = root;
-	disk_super = root->fs_info->disk_super;
+	sb->s_fs_info = tree_root;
+	disk_super = tree_root->fs_info->disk_super;
 	printk("read in super total blocks %Lu root %Lu\n",
 	       btrfs_super_total_blocks(disk_super),
 	       btrfs_super_root_dir(disk_super));
@@ -630,7 +635,7 @@ static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 	bi->location.objectid = inode->i_ino;
 	bi->location.offset = 0;
 	bi->location.flags = 0;
-	bi->root = root->fs_info->tree_root;
+	bi->root = tree_root;
 	btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
 
 	if (!inode)
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 20d84bd..83a0194 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -8,6 +8,8 @@ static int total_trans = 0;
 extern struct kmem_cache *btrfs_trans_handle_cachep;
 extern struct kmem_cache *btrfs_transaction_cachep;
 
+#define BTRFS_ROOT_TRANS_TAG 0
+
 #define TRANS_MAGIC 0xE1E10E
 static void put_transaction(struct btrfs_transaction *transaction)
 {
@@ -31,9 +33,10 @@ static int join_transaction(struct btrfs_root *root)
 					     GFP_NOFS);
 		total_trans++;
 		BUG_ON(!cur_trans);
+		root->fs_info->generation++;
 		root->fs_info->running_transaction = cur_trans;
 		cur_trans->num_writers = 0;
-		cur_trans->transid = root->root_key.offset + 1;
+		cur_trans->transid = root->fs_info->generation;
 		init_waitqueue_head(&cur_trans->writer_wait);
 		init_waitqueue_head(&cur_trans->commit_wait);
 		cur_trans->magic = TRANS_MAGIC;
@@ -51,13 +54,22 @@ struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 	struct btrfs_trans_handle *h =
 		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 	int ret;
+	u64 running_trans_id;
 
-	/* FIXME, use the right root */
-	root = root->fs_info->fs_root;
 	mutex_lock(&root->fs_info->trans_mutex);
 	ret = join_transaction(root);
 	BUG_ON(ret);
-	h->transid = root->fs_info->running_transaction->transid;
+	running_trans_id = root->fs_info->running_transaction->transid;
+
+	if (root != root->fs_info->tree_root && root->last_trans <
+	    running_trans_id) {
+		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+				   (unsigned long)root, BTRFS_ROOT_TRANS_TAG);
+		root->commit_root = root->node;
+		get_bh(root->node);
+	}
+	root->last_trans = running_trans_id;
+	h->transid = running_trans_id;
 	h->transaction = root->fs_info->running_transaction;
 	h->blocks_reserved = num_blocks;
 	h->blocks_used = 0;
@@ -72,9 +84,6 @@ int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 {
 	struct btrfs_transaction *cur_trans;
 
-	/* FIXME, use the right root */
-	root = root->fs_info->fs_root;
-
 	WARN_ON(trans->magic != TRANS_MAGIC);
 	WARN_ON(trans->magic2 != TRANS_MAGIC);
 	mutex_lock(&root->fs_info->trans_mutex);
@@ -145,17 +154,96 @@ static int wait_for_commit(struct btrfs_root *root,
 	return 0;
 }
 
+struct dirty_root {
+	struct list_head list;
+	struct btrfs_key snap_key;
+	struct buffer_head *commit_root;
+	struct btrfs_root *root;
+};
+
+int add_dirty_roots(struct btrfs_trans_handle *trans,
+		    struct radix_tree_root *radix, struct list_head *list)
+{
+	struct dirty_root *dirty;
+	struct btrfs_root *gang[8];
+	struct btrfs_root *root;
+	int i;
+	int ret;
+	int err;
+printk("add dirty\n");
+	while(1) {
+		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
+						 ARRAY_SIZE(gang),
+						 BTRFS_ROOT_TRANS_TAG);
+		if (ret == 0)
+			break;
+		for (i = 0; i < ret; i++) {
+			root = gang[i];
+			radix_tree_tag_clear(radix, (unsigned long)root,
+					     BTRFS_ROOT_TRANS_TAG);
+			if (root->commit_root == root->node) {
+				WARN_ON(root->node->b_blocknr !=
+					btrfs_root_blocknr(&root->root_item));
+				brelse(root->commit_root);
+				root->commit_root = NULL;
+				continue;
+			}
+			dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
+			BUG_ON(!dirty);
+			memcpy(&dirty->snap_key, &root->root_key,
+			       sizeof(root->root_key));
+			dirty->commit_root = root->commit_root;
+			root->commit_root = NULL;
+			dirty->root = root;
+printk("adding dirty root %Lu gen %Lu blocknr %Lu\n", root->root_key.objectid, root->root_key.offset, dirty->commit_root->b_blocknr);
+			root->root_key.offset = root->fs_info->generation;
+			btrfs_set_root_blocknr(&root->root_item,
+					       root->node->b_blocknr);
+			err = btrfs_insert_root(trans, root->fs_info->tree_root,
+						&root->root_key,
+						&root->root_item);
+			BUG_ON(err);
+			list_add(&dirty->list, list);
+		}
+	}
+printk("add dirty done\n");
+	return 0;
+}
+
+int drop_dirty_roots(struct btrfs_root *tree_root, struct list_head *list)
+{
+	struct dirty_root *dirty;
+	struct btrfs_trans_handle *trans;
+	int ret;
+
+	while(!list_empty(list)) {
+		dirty = list_entry(list->next, struct dirty_root, list);
+		list_del_init(&dirty->list);
+		trans = btrfs_start_transaction(tree_root, 1);
+printk("drop snapshot root %p, commit_root blocknr %Lu generation %Lu\n", dirty->root, dirty->commit_root->b_blocknr, dirty->snap_key.offset);
+		ret = btrfs_drop_snapshot(trans, dirty->root,
+					  dirty->commit_root);
+		BUG_ON(ret);
+
+printk("del root objectid %Lu, offset %Lu\n", dirty->snap_key.objectid, dirty->snap_key.offset);
+		ret = btrfs_del_root(trans, tree_root, &dirty->snap_key);
+		BUG_ON(ret);
+		ret = btrfs_end_transaction(trans, tree_root);
+		BUG_ON(ret);
+		kfree(dirty);
+	}
+	return 0;
+}
+
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root)
 {
 	int ret = 0;
-	struct buffer_head *snap;
-	struct btrfs_key snap_key;
 	struct btrfs_transaction *cur_trans;
+	struct list_head dirty_fs_roots;
 	DEFINE_WAIT(wait);
 
-	/* FIXME, use the right root */
-	root = root->fs_info->fs_root;
+	INIT_LIST_HEAD(&dirty_fs_roots);
 
 	mutex_lock(&root->fs_info->trans_mutex);
 	if (trans->transaction->in_commit) {
@@ -184,22 +272,13 @@ printk("already in commit!, waiting\n");
 	}
 	finish_wait(&trans->transaction->writer_wait, &wait);
 	WARN_ON(cur_trans != trans->transaction);
-	if (root->node != root->commit_root) {
-		memcpy(&snap_key, &root->root_key, sizeof(snap_key));
-		root->root_key.offset++;
-	}
-
-	if (btrfs_root_blocknr(&root->root_item) != root->node->b_blocknr) {
-		btrfs_set_root_blocknr(&root->root_item, root->node->b_blocknr);
-		ret = btrfs_insert_root(trans, root->fs_info->tree_root,
-					&root->root_key, &root->root_item);
-		BUG_ON(ret);
-	}
-
+	add_dirty_roots(trans, &root->fs_info->fs_roots_radix, &dirty_fs_roots);
 	ret = btrfs_commit_tree_roots(trans, root);
 	BUG_ON(ret);
 	cur_trans = root->fs_info->running_transaction;
 	root->fs_info->running_transaction = NULL;
+	btrfs_set_super_generation(root->fs_info->disk_super,
+				   root->fs_info->generation + 1);
 	mutex_unlock(&root->fs_info->trans_mutex);
 	ret = btrfs_write_and_wait_transaction(trans, root);
 	BUG_ON(ret);
@@ -213,21 +292,8 @@ printk("already in commit!, waiting\n");
 	put_transaction(cur_trans);
 	mutex_unlock(&root->fs_info->trans_mutex);
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
-	if (root->node != root->commit_root) {
-		trans = btrfs_start_transaction(root, 1);
-		snap = root->commit_root;
-		root->commit_root = root->node;
-		get_bh(root->node);
-		ret = btrfs_drop_snapshot(trans, root, snap);
-		BUG_ON(ret);
 
-		ret = btrfs_del_root(trans, root->fs_info->tree_root,
-				     &snap_key);
-		BUG_ON(ret);
-		root->fs_info->generation = root->root_key.offset + 1;
-		ret = btrfs_end_transaction(trans, root);
-		BUG_ON(ret);
-	}
+	drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 	return ret;
 }