X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=fs%2Fbtrfs%2Ftransaction.c;h=2a36e236a4924c978d7b74c416e062f7d92dc921;hb=08d925badd63a4e335905066152d96b83266987c;hp=1ed433a7149333621042f1af917efa10e09c3dea;hpb=925baeddc5b0764a53f2214a1253251bab0e0324;p=safe%2Fjmp%2Flinux-2.6

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 1ed433a..2a36e23 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -20,33 +20,35 @@
 #include <linux/sched.h>
 #include <linux/writeback.h>
 #include <linux/pagemap.h>
+#include <linux/blkdev.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
 #include "locking.h"
-
-static int total_trans = 0;
-extern struct kmem_cache *btrfs_trans_handle_cachep;
-extern struct kmem_cache *btrfs_transaction_cachep;
-
-static struct workqueue_struct *trans_wq;
+#include "tree-log.h"
 
 #define BTRFS_ROOT_TRANS_TAG 0
-#define BTRFS_ROOT_DEFRAG_TAG 1
 
 static noinline void put_transaction(struct btrfs_transaction *transaction)
 {
 	WARN_ON(transaction->use_count == 0);
 	transaction->use_count--;
 	if (transaction->use_count == 0) {
-		WARN_ON(total_trans == 0);
-		total_trans--;
 		list_del_init(&transaction->list);
 		memset(transaction, 0, sizeof(*transaction));
 		kmem_cache_free(btrfs_transaction_cachep, transaction);
 	}
 }
 
+static noinline void switch_commit_root(struct btrfs_root *root)
+{
+	free_extent_buffer(root->commit_root);
+	root->commit_root = btrfs_root_node(root);
+}
+
+/*
+ * either allocate a new transaction or hop into the existing one
+ */
 static noinline int join_transaction(struct btrfs_root *root)
 {
 	struct btrfs_transaction *cur_trans;
@@ -54,23 +56,29 @@ static noinline int join_transaction(struct btrfs_root *root)
 	if (!cur_trans) {
 		cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
 					     GFP_NOFS);
-		total_trans++;
 		BUG_ON(!cur_trans);
 		root->fs_info->generation++;
-		root->fs_info->last_alloc = 0;
-		root->fs_info->last_data_alloc = 0;
 		cur_trans->num_writers = 1;
 		cur_trans->num_joined = 0;
 		cur_trans->transid = root->fs_info->generation;
 		init_waitqueue_head(&cur_trans->writer_wait);
 		init_waitqueue_head(&cur_trans->commit_wait);
 		cur_trans->in_commit = 0;
+		cur_trans->blocked = 0;
 		cur_trans->use_count = 1;
 		cur_trans->commit_done = 0;
 		cur_trans->start_time = get_seconds();
+
+		cur_trans->delayed_refs.root = RB_ROOT;
+		cur_trans->delayed_refs.num_entries = 0;
+		cur_trans->delayed_refs.num_heads_ready = 0;
+		cur_trans->delayed_refs.num_heads = 0;
+		cur_trans->delayed_refs.flushing = 0;
+		cur_trans->delayed_refs.run_delayed_start = 0;
+		spin_lock_init(&cur_trans->delayed_refs.lock);
+
 		INIT_LIST_HEAD(&cur_trans->pending_snapshots);
 		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
-		btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
 		extent_io_tree_init(&cur_trans->dirty_pages,
 				     root->fs_info->btree_inode->i_mapping,
 				     GFP_NOFS);
@@ -85,100 +93,302 @@ static noinline int join_transaction(struct btrfs_root *root)
 	return 0;
 }
 
-static noinline int record_root_in_trans(struct btrfs_root *root)
+/*
+ * this does all the record keeping required to make sure that a reference
+ * counted root is properly recorded in a given transaction.  This is required
+ * to make sure the old root from before we joined the transaction is deleted
+ * when the transaction commits
+ */
+static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
+					 struct btrfs_root *root)
 {
-	u64 running_trans_id = root->fs_info->running_transaction->transid;
-	if (root->ref_cows && root->last_trans < running_trans_id) {
+	if (root->ref_cows && root->last_trans < trans->transid) {
 		WARN_ON(root == root->fs_info->extent_root);
-		if (root->root_item.refs != 0) {
-			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
-				   (unsigned long)root->root_key.objectid,
-				   BTRFS_ROOT_TRANS_TAG);
-			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
-				   (unsigned long)root->root_key.objectid,
-				   BTRFS_ROOT_DEFRAG_TAG);
-			root->commit_root = btrfs_root_node(root);
-		} else {
-			WARN_ON(1);
-		}
-		root->last_trans = running_trans_id;
+		WARN_ON(root->commit_root != root->node);
+
+		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+			   (unsigned long)root->root_key.objectid,
+			   BTRFS_ROOT_TRANS_TAG);
+		root->last_trans = trans->transid;
+		btrfs_init_reloc_root(trans, root);
 	}
 	return 0;
 }
 
-struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
-						   int num_blocks)
+int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	if (!root->ref_cows)
+		return 0;
+
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (root->last_trans == trans->transid) {
+		mutex_unlock(&root->fs_info->trans_mutex);
+		return 0;
+	}
+
+	record_root_in_trans(trans, root);
+	mutex_unlock(&root->fs_info->trans_mutex);
+	return 0;
+}
+
+/* wait for commit against the current transaction to become unblocked
+ * when this is done, it is safe to start a new transaction, but the current
+ * transaction might not be fully on disk.
+ */
+static void wait_current_trans(struct btrfs_root *root)
+{
+	struct btrfs_transaction *cur_trans;
+
+	cur_trans = root->fs_info->running_transaction;
+	if (cur_trans && cur_trans->blocked) {
+		DEFINE_WAIT(wait);
+		cur_trans->use_count++;
+		while (1) {
+			prepare_to_wait(&root->fs_info->transaction_wait, &wait,
+					TASK_UNINTERRUPTIBLE);
+			if (cur_trans->blocked) {
+				mutex_unlock(&root->fs_info->trans_mutex);
+				schedule();
+				mutex_lock(&root->fs_info->trans_mutex);
+				finish_wait(&root->fs_info->transaction_wait,
+					    &wait);
+			} else {
+				finish_wait(&root->fs_info->transaction_wait,
+					    &wait);
+				break;
+			}
+		}
+		put_transaction(cur_trans);
+	}
+}
+
+enum btrfs_trans_type {
+	TRANS_START,
+	TRANS_JOIN,
+	TRANS_USERSPACE,
+};
+
+static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
+					     int num_blocks, int type)
 {
 	struct btrfs_trans_handle *h =
 		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 	int ret;
 
 	mutex_lock(&root->fs_info->trans_mutex);
+	if (!root->fs_info->log_root_recovering &&
+	    ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
+	     type == TRANS_USERSPACE))
+		wait_current_trans(root);
 	ret = join_transaction(root);
 	BUG_ON(ret);
 
-	record_root_in_trans(root);
 	h->transid = root->fs_info->running_transaction->transid;
 	h->transaction = root->fs_info->running_transaction;
 	h->blocks_reserved = num_blocks;
 	h->blocks_used = 0;
-	h->block_group = NULL;
+	h->block_group = 0;
 	h->alloc_exclude_nr = 0;
 	h->alloc_exclude_start = 0;
+	h->delayed_ref_updates = 0;
+
+	if (!current->journal_info && type != TRANS_USERSPACE)
+		current->journal_info = h;
+
 	root->fs_info->running_transaction->use_count++;
+	record_root_in_trans(h, root);
 	mutex_unlock(&root->fs_info->trans_mutex);
 	return h;
 }
 
-int btrfs_end_transaction(struct btrfs_trans_handle *trans,
-			  struct btrfs_root *root)
+struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
+						   int num_blocks)
 {
-	struct btrfs_transaction *cur_trans;
+	return start_transaction(root, num_blocks, TRANS_START);
+}
+struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
+						   int num_blocks)
+{
+	return start_transaction(root, num_blocks, TRANS_JOIN);
+}
+
+struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
+							 int num_blocks)
+{
+	return start_transaction(r, num_blocks, TRANS_USERSPACE);
+}
 
+/* wait for a transaction commit to be fully complete */
+static noinline int wait_for_commit(struct btrfs_root *root,
+				    struct btrfs_transaction *commit)
+{
+	DEFINE_WAIT(wait);
 	mutex_lock(&root->fs_info->trans_mutex);
-	cur_trans = root->fs_info->running_transaction;
+	while (!commit->commit_done) {
+		prepare_to_wait(&commit->commit_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+		if (commit->commit_done)
+			break;
+		mutex_unlock(&root->fs_info->trans_mutex);
+		schedule();
+		mutex_lock(&root->fs_info->trans_mutex);
+	}
+	mutex_unlock(&root->fs_info->trans_mutex);
+	finish_wait(&commit->commit_wait, &wait);
+	return 0;
+}
+
+#if 0
+/*
+ * rate limit against the drop_snapshot code.  This helps to slow down new
+ * operations if the drop_snapshot code isn't able to keep up.
+ */
+static void throttle_on_drops(struct btrfs_root *root)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	int harder_count = 0;
+
+harder:
+	if (atomic_read(&info->throttles)) {
+		DEFINE_WAIT(wait);
+		int thr;
+		thr = atomic_read(&info->throttle_gen);
+
+		do {
+			prepare_to_wait(&info->transaction_throttle,
+					&wait, TASK_UNINTERRUPTIBLE);
+			if (!atomic_read(&info->throttles)) {
+				finish_wait(&info->transaction_throttle, &wait);
+				break;
+			}
+			schedule();
+			finish_wait(&info->transaction_throttle, &wait);
+		} while (thr == atomic_read(&info->throttle_gen));
+		harder_count++;
+
+		if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
+		    harder_count < 2)
+			goto harder;
+
+		if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
+		    harder_count < 10)
+			goto harder;
+
+		if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
+		    harder_count < 20)
+			goto harder;
+	}
+}
+#endif
+
+void btrfs_throttle(struct btrfs_root *root)
+{
+	mutex_lock(&root->fs_info->trans_mutex);
+	if (!root->fs_info->open_ioctl_trans)
+		wait_current_trans(root);
+	mutex_unlock(&root->fs_info->trans_mutex);
+}
+
+static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, int throttle)
+{
+	struct btrfs_transaction *cur_trans;
+	struct btrfs_fs_info *info = root->fs_info;
+	int count = 0;
+
+	while (count < 4) {
+		unsigned long cur = trans->delayed_ref_updates;
+		trans->delayed_ref_updates = 0;
+		if (cur &&
+		    trans->transaction->delayed_refs.num_heads_ready > 64) {
+			trans->delayed_ref_updates = 0;
+
+			/*
+			 * do a full flush if the transaction is trying
+			 * to close
+			 */
+			if (trans->transaction->delayed_refs.flushing)
+				cur = 0;
+			btrfs_run_delayed_refs(trans, root, cur);
+		} else {
+			break;
+		}
+		count++;
+	}
+
+	mutex_lock(&info->trans_mutex);
+	cur_trans = info->running_transaction;
 	WARN_ON(cur_trans != trans->transaction);
 	WARN_ON(cur_trans->num_writers < 1);
 	cur_trans->num_writers--;
+
 	if (waitqueue_active(&cur_trans->writer_wait))
 		wake_up(&cur_trans->writer_wait);
 	put_transaction(cur_trans);
-	mutex_unlock(&root->fs_info->trans_mutex);
+	mutex_unlock(&info->trans_mutex);
+
+	if (current->journal_info == trans)
+		current->journal_info = NULL;
 	memset(trans, 0, sizeof(*trans));
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
+
+	if (throttle)
+		btrfs_run_delayed_iputs(root);
+
 	return 0;
 }
 
+int btrfs_end_transaction(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root)
+{
+	return __btrfs_end_transaction(trans, root, 0);
+}
 
-int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
-				     struct btrfs_root *root)
+int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root)
+{
+	return __btrfs_end_transaction(trans, root, 1);
+}
+
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees.  This is used to make sure all of
+ * those extents are sent to disk but does not wait on them
+ */
+int btrfs_write_marked_extents(struct btrfs_root *root,
+			       struct extent_io_tree *dirty_pages, int mark)
 {
 	int ret;
-	int err;
+	int err = 0;
 	int werr = 0;
-	struct extent_io_tree *dirty_pages;
 	struct page *page;
 	struct inode *btree_inode = root->fs_info->btree_inode;
-	u64 start;
+	u64 start = 0;
 	u64 end;
 	unsigned long index;
 
-	if (!trans || !trans->transaction) {
-		return filemap_write_and_wait(btree_inode->i_mapping);
-	}
-	dirty_pages = &trans->transaction->dirty_pages;
-	while(1) {
-		ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
-					    EXTENT_DIRTY);
+	while (1) {
+		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
+					    mark);
 		if (ret)
 			break;
-		clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
-		while(start <= end) {
+		while (start <= end) {
+			cond_resched();
+
 			index = start >> PAGE_CACHE_SHIFT;
 			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
-			page = find_lock_page(btree_inode->i_mapping, index);
+			page = find_get_page(btree_inode->i_mapping, index);
 			if (!page)
 				continue;
+
+			btree_lock_page_hook(page);
+			if (!page->mapping) {
+				unlock_page(page);
+				page_cache_release(page);
+				continue;
+			}
+
 			if (PageWriteback(page)) {
 				if (PageDirty(page))
 					wait_on_page_writeback(page);
@@ -194,175 +404,230 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 			page_cache_release(page);
 		}
 	}
-	err = filemap_fdatawait(btree_inode->i_mapping);
 	if (err)
 		werr = err;
 	return werr;
 }
 
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees.  This is used to make sure all of
+ * those extents are on disk for transaction or log commit.  We wait
+ * on all the pages and clear them from the dirty pages state tree
+ */
+int btrfs_wait_marked_extents(struct btrfs_root *root,
+			      struct extent_io_tree *dirty_pages, int mark)
+{
+	int ret;
+	int err = 0;
+	int werr = 0;
+	struct page *page;
+	struct inode *btree_inode = root->fs_info->btree_inode;
+	u64 start = 0;
+	u64 end;
+	unsigned long index;
+
+	while (1) {
+		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
+					    mark);
+		if (ret)
+			break;
+
+		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
+		while (start <= end) {
+			index = start >> PAGE_CACHE_SHIFT;
+			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
+			page = find_get_page(btree_inode->i_mapping, index);
+			if (!page)
+				continue;
+			if (PageDirty(page)) {
+				btree_lock_page_hook(page);
+				wait_on_page_writeback(page);
+				err = write_one_page(page, 0);
+				if (err)
+					werr = err;
+			}
+			wait_on_page_writeback(page);
+			page_cache_release(page);
+			cond_resched();
+		}
+	}
+	if (err)
+		werr = err;
+	return werr;
+}
+
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees.  This is used to make sure all of
+ * those extents are on disk for transaction or log commit
+ */
+int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
+				struct extent_io_tree *dirty_pages, int mark)
+{
+	int ret;
+	int ret2;
+
+	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
+	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
+	return ret || ret2;
+}
+
+int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root)
+{
+	if (!trans || !trans->transaction) {
+		struct inode *btree_inode;
+		btree_inode = root->fs_info->btree_inode;
+		return filemap_write_and_wait(btree_inode->i_mapping);
+	}
+	return btrfs_write_and_wait_marked_extents(root,
+					   &trans->transaction->dirty_pages,
+					   EXTENT_DIRTY);
+}
+
+/*
+ * this is used to update the root pointer in the tree of tree roots.
+ *
+ * But, in the case of the extent allocation tree, updating the root
+ * pointer may allocate blocks which may change the root of the extent
+ * allocation tree.
+ *
+ * So, this loops and repeats and makes sure the cowonly root didn't
+ * change while the root pointer was being updated in the metadata.
+ */
 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root)
 {
 	int ret;
 	u64 old_root_bytenr;
+	u64 old_root_used;
 	struct btrfs_root *tree_root = root->fs_info->tree_root;
 
+	old_root_used = btrfs_root_used(&root->root_item);
 	btrfs_write_dirty_block_groups(trans, root);
-	while(1) {
+
+	while (1) {
 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
-		if (old_root_bytenr == root->node->start)
+		if (old_root_bytenr == root->node->start &&
+		    old_root_used == btrfs_root_used(&root->root_item))
 			break;
-		btrfs_set_root_bytenr(&root->root_item,
-				       root->node->start);
-		btrfs_set_root_level(&root->root_item,
-				     btrfs_header_level(root->node));
+
+		btrfs_set_root_node(&root->root_item, root->node);
 		ret = btrfs_update_root(trans, tree_root,
 					&root->root_key,
 					&root->root_item);
 		BUG_ON(ret);
-		btrfs_write_dirty_block_groups(trans, root);
+
+		old_root_used = btrfs_root_used(&root->root_item);
+		ret = btrfs_write_dirty_block_groups(trans, root);
+		BUG_ON(ret);
 	}
+
+	if (root != root->fs_info->extent_root)
+		switch_commit_root(root);
+
 	return 0;
 }
 
-int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root)
+/*
+ * update all the cowonly tree roots on disk
+ */
+static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
+					 struct btrfs_root *root)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct list_head *next;
+	struct extent_buffer *eb;
+	int ret;
+
+	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+	BUG_ON(ret);
+
+	eb = btrfs_lock_root_node(fs_info->tree_root);
+	btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
+	btrfs_tree_unlock(eb);
+	free_extent_buffer(eb);
+
+	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+	BUG_ON(ret);
 
-	while(!list_empty(&fs_info->dirty_cowonly_roots)) {
+	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
 		next = fs_info->dirty_cowonly_roots.next;
 		list_del_init(next);
 		root = list_entry(next, struct btrfs_root, dirty_list);
+
 		update_cowonly_root(trans, root);
 	}
+
+	down_write(&fs_info->extent_commit_sem);
+	switch_commit_root(fs_info->extent_root);
+	up_write(&fs_info->extent_commit_sem);
+
 	return 0;
 }
 
-static noinline int wait_for_commit(struct btrfs_root *root,
-				    struct btrfs_transaction *commit)
+/*
+ * dead roots are old snapshots that need to be deleted.  This allocates
+ * a dirty root struct and adds it into the list of dead roots that need to
+ * be deleted
+ */
+int btrfs_add_dead_root(struct btrfs_root *root)
 {
-	DEFINE_WAIT(wait);
 	mutex_lock(&root->fs_info->trans_mutex);
-	while(!commit->commit_done) {
-		prepare_to_wait(&commit->commit_wait, &wait,
-				TASK_UNINTERRUPTIBLE);
-		if (commit->commit_done)
-			break;
-		mutex_unlock(&root->fs_info->trans_mutex);
-		schedule();
-		mutex_lock(&root->fs_info->trans_mutex);
-	}
+	list_add(&root->root_list, &root->fs_info->dead_roots);
 	mutex_unlock(&root->fs_info->trans_mutex);
-	finish_wait(&commit->commit_wait, &wait);
 	return 0;
 }
 
-struct dirty_root {
-	struct list_head list;
-	struct btrfs_root *root;
-	struct btrfs_root *latest_root;
-};
-
-int btrfs_add_dead_root(struct btrfs_root *root,
-			struct btrfs_root *latest,
-			struct list_head *dead_list)
-{
-	struct dirty_root *dirty;
-
-	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
-	if (!dirty)
-		return -ENOMEM;
-	dirty->root = root;
-	dirty->latest_root = latest;
-	list_add(&dirty->list, dead_list);
-	return 0;
-}
-
-static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
-				    struct radix_tree_root *radix,
-				    struct list_head *list)
+/*
+ * update all the cowonly tree roots on disk
+ */
+static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root)
 {
-	struct dirty_root *dirty;
 	struct btrfs_root *gang[8];
-	struct btrfs_root *root;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 	int i;
 	int ret;
 	int err = 0;
-	u32 refs;
 
-	while(1) {
-		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
+	while (1) {
+		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
+						 (void **)gang, 0,
 						 ARRAY_SIZE(gang),
 						 BTRFS_ROOT_TRANS_TAG);
 		if (ret == 0)
 			break;
 		for (i = 0; i < ret; i++) {
 			root = gang[i];
-			radix_tree_tag_clear(radix,
-				     (unsigned long)root->root_key.objectid,
-				     BTRFS_ROOT_TRANS_TAG);
-			if (root->commit_root == root->node) {
-				WARN_ON(root->node->start !=
-					btrfs_root_bytenr(&root->root_item));
-				free_extent_buffer(root->commit_root);
-				root->commit_root = NULL;
-
-				/* make sure to update the root on disk
-				 * so we get any updates to the block used
-				 * counts
-				 */
-				err = btrfs_update_root(trans,
-						root->fs_info->tree_root,
-						&root->root_key,
-						&root->root_item);
-				continue;
+			radix_tree_tag_clear(&fs_info->fs_roots_radix,
+					(unsigned long)root->root_key.objectid,
+					BTRFS_ROOT_TRANS_TAG);
+
+			btrfs_free_log(trans, root);
+			btrfs_update_reloc_root(trans, root);
+
+			if (root->commit_root != root->node) {
+				switch_commit_root(root);
+				btrfs_set_root_node(&root->root_item,
+						    root->node);
 			}
-			dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
-			BUG_ON(!dirty);
-			dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
-			BUG_ON(!dirty->root);
-
-			memset(&root->root_item.drop_progress, 0,
-			       sizeof(struct btrfs_disk_key));
-			root->root_item.drop_level = 0;
-
-			memcpy(dirty->root, root, sizeof(*root));
-			dirty->root->node = root->commit_root;
-			dirty->latest_root = root;
-			root->commit_root = NULL;
-
-			root->root_key.offset = root->fs_info->generation;
-			btrfs_set_root_bytenr(&root->root_item,
-					      root->node->start);
-			btrfs_set_root_level(&root->root_item,
-					     btrfs_header_level(root->node));
-			err = btrfs_insert_root(trans, root->fs_info->tree_root,
+
+			err = btrfs_update_root(trans, fs_info->tree_root,
 						&root->root_key,
 						&root->root_item);
 			if (err)
 				break;
-
-			refs = btrfs_root_refs(&dirty->root->root_item);
-			btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
-			err = btrfs_update_root(trans, root->fs_info->tree_root,
-						&dirty->root->root_key,
-						&dirty->root->root_item);
-
-			BUG_ON(err);
-			if (refs == 1) {
-				list_add(&dirty->list, list);
-			} else {
-				WARN_ON(1);
-				kfree(dirty->root);
-				kfree(dirty);
-			}
 		}
 	}
 	return err;
 }
 
+/*
+ * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
+ * otherwise every leaf in the btree is read and defragged.
+ */
 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 {
 	struct btrfs_fs_info *info = root->fs_info;
@@ -370,6 +635,7 @@ int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 	struct btrfs_trans_handle *trans;
 	unsigned long nr;
 
+	smp_mb();
 	if (root->defrag_running)
 		return 0;
 	trans = btrfs_start_transaction(root, 1);
@@ -378,178 +644,114 @@ int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 		ret = btrfs_defrag_leaves(trans, root, cacheonly);
 		nr = trans->blocks_used;
 		btrfs_end_transaction(trans, root);
-		mutex_unlock(&info->fs_mutex);
 		btrfs_btree_balance_dirty(info->tree_root, nr);
 		cond_resched();
 
-		mutex_lock(&info->fs_mutex);
 		trans = btrfs_start_transaction(root, 1);
-		if (ret != -EAGAIN)
+		if (root->fs_info->closing || ret != -EAGAIN)
 			break;
 	}
 	root->defrag_running = 0;
-	radix_tree_tag_clear(&info->fs_roots_radix,
-		     (unsigned long)root->root_key.objectid,
-		     BTRFS_ROOT_DEFRAG_TAG);
+	smp_mb();
 	btrfs_end_transaction(trans, root);
 	return 0;
 }
 
-int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
+#if 0
+/*
+ * when dropping snapshots, we generate a ton of delayed refs, and it makes
+ * sense not to join the transaction while it is trying to flush the current
+ * queue of delayed refs out.
+ *
+ * This is used by the drop snapshot code only
+ */
+static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
 {
-	struct btrfs_root *gang[1];
-	struct btrfs_root *root;
-	int i;
-	int ret;
-	int err = 0;
-	u64 last = 0;
+	DEFINE_WAIT(wait);
 
-	while(1) {
-		ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
-						 (void **)gang, last,
-						 ARRAY_SIZE(gang),
-						 BTRFS_ROOT_DEFRAG_TAG);
-		if (ret == 0)
-			break;
-		for (i = 0; i < ret; i++) {
-			root = gang[i];
-			last = root->root_key.objectid + 1;
-			btrfs_defrag_root(root, 1);
-		}
+	mutex_lock(&info->trans_mutex);
+	while (info->running_transaction &&
+	       info->running_transaction->delayed_refs.flushing) {
+		prepare_to_wait(&info->transaction_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+		mutex_unlock(&info->trans_mutex);
+
+		schedule();
+
+		mutex_lock(&info->trans_mutex);
+		finish_wait(&info->transaction_wait, &wait);
 	}
-	btrfs_defrag_root(info->extent_root, 1);
-	return err;
+	mutex_unlock(&info->trans_mutex);
+	return 0;
 }
 
-static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
-				     struct list_head *list)
+/*
+ * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
+ * all of them
+ */
+int btrfs_drop_dead_root(struct btrfs_root *root)
 {
-	struct dirty_root *dirty;
 	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = root->fs_info->tree_root;
 	unsigned long nr;
-	u64 num_bytes;
-	u64 bytes_used;
-	int ret = 0;
-	int err;
-
-	while(!list_empty(list)) {
-		struct btrfs_root *root;
-
-		mutex_lock(&tree_root->fs_info->fs_mutex);
-		dirty = list_entry(list->next, struct dirty_root, list);
-		list_del_init(&dirty->list);
+	int ret;
 
-		num_bytes = btrfs_root_used(&dirty->root->root_item);
-		root = dirty->latest_root;
-		root->fs_info->throttles++;
+	while (1) {
+		/*
+		 * we don't want to jump in and create a bunch of
+		 * delayed refs if the transaction is starting to close
+		 */
+		wait_transaction_pre_flush(tree_root->fs_info);
+		trans = btrfs_start_transaction(tree_root, 1);
+
+		/*
+		 * we've joined a transaction, make sure it isn't
+		 * closing right now
+		 */
+		if (trans->transaction->delayed_refs.flushing) {
+			btrfs_end_transaction(trans, tree_root);
+			continue;
+		}
 
-		while(1) {
-			trans = btrfs_start_transaction(tree_root, 1);
-			ret = btrfs_drop_snapshot(trans, dirty->root);
-			if (ret != -EAGAIN) {
-				break;
-			}
+		ret = btrfs_drop_snapshot(trans, root);
+		if (ret != -EAGAIN)
+			break;
 
-			err = btrfs_update_root(trans,
-					tree_root,
-					&dirty->root->root_key,
-					&dirty->root->root_item);
-			if (err)
-				ret = err;
-			nr = trans->blocks_used;
-			ret = btrfs_end_transaction(trans, tree_root);
-			BUG_ON(ret);
-			mutex_unlock(&tree_root->fs_info->fs_mutex);
-			btrfs_btree_balance_dirty(tree_root, nr);
-			cond_resched();
-			mutex_lock(&tree_root->fs_info->fs_mutex);
-		}
-		BUG_ON(ret);
-		root->fs_info->throttles--;
-
-		num_bytes -= btrfs_root_used(&dirty->root->root_item);
-		bytes_used = btrfs_root_used(&root->root_item);
-		if (num_bytes) {
-			record_root_in_trans(root);
-			btrfs_set_root_used(&root->root_item,
-					    bytes_used - num_bytes);
-		}
-		ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
-		if (ret) {
-			BUG();
+		ret = btrfs_update_root(trans, tree_root,
+					&root->root_key,
+					&root->root_item);
+		if (ret)
 			break;
-		}
+
 		nr = trans->blocks_used;
 		ret = btrfs_end_transaction(trans, tree_root);
 		BUG_ON(ret);
 
-		free_extent_buffer(dirty->root->node);
-		kfree(dirty->root);
-		kfree(dirty);
-		mutex_unlock(&tree_root->fs_info->fs_mutex);
-
 		btrfs_btree_balance_dirty(tree_root, nr);
 		cond_resched();
 	}
-	return ret;
-}
-
-int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root)
-{
-	struct btrfs_transaction *cur_trans = trans->transaction;
-	struct inode *inode;
-	u64 root_objectid = 0;
-	u64 objectid = 0;
-	int ret;
-
-	root->fs_info->throttles++;
-	while(1) {
-		ret = btrfs_find_first_ordered_inode(
-				&cur_trans->ordered_inode_tree,
-				&root_objectid, &objectid, &inode);
-		if (!ret)
-			break;
-
-		mutex_unlock(&root->fs_info->trans_mutex);
-		mutex_unlock(&root->fs_info->fs_mutex);
+	BUG_ON(ret);
 
-		if (S_ISREG(inode->i_mode)) {
-			atomic_inc(&BTRFS_I(inode)->ordered_writeback);
-			filemap_fdatawrite(inode->i_mapping);
-			atomic_dec(&BTRFS_I(inode)->ordered_writeback);
-		}
-		iput(inode);
+	ret = btrfs_del_root(trans, tree_root, &root->root_key);
+	BUG_ON(ret);
 
-		mutex_lock(&root->fs_info->fs_mutex);
-		mutex_lock(&root->fs_info->trans_mutex);
-	}
-	while(1) {
-		root_objectid = 0;
-		objectid = 0;
-		ret = btrfs_find_del_first_ordered_inode(
-				&cur_trans->ordered_inode_tree,
-				&root_objectid, &objectid, &inode);
-		if (!ret)
-			break;
-		mutex_unlock(&root->fs_info->trans_mutex);
-		mutex_unlock(&root->fs_info->fs_mutex);
+	nr = trans->blocks_used;
+	ret = btrfs_end_transaction(trans, tree_root);
+	BUG_ON(ret);
 
-		if (S_ISREG(inode->i_mode)) {
-			atomic_inc(&BTRFS_I(inode)->ordered_writeback);
-			filemap_write_and_wait(inode->i_mapping);
-			atomic_dec(&BTRFS_I(inode)->ordered_writeback);
-		}
-		atomic_dec(&inode->i_count);
-		iput(inode);
+	free_extent_buffer(root->node);
+	free_extent_buffer(root->commit_root);
+	kfree(root);
 
-		mutex_lock(&root->fs_info->fs_mutex);
-		mutex_lock(&root->fs_info->trans_mutex);
-	}
-	root->fs_info->throttles--;
-	return 0;
+	btrfs_btree_balance_dirty(tree_root, nr);
+	return ret;
 }
+#endif
 
+/*
+ * new snapshots need to be created at a very specific time in the
+ * transaction commit.  This does the actual creation
+ */
 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 				   struct btrfs_fs_info *fs_info,
 				   struct btrfs_pending_snapshot *pending)
@@ -561,7 +763,6 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	struct extent_buffer *tmp;
 	struct extent_buffer *old;
 	int ret;
-	int namelen;
 	u64 objectid;
 
 	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
@@ -573,21 +774,24 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	if (ret)
 		goto fail;
 
+	record_root_in_trans(trans, root);
+	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 
 	key.objectid = objectid;
-	key.offset = 1;
+	/* record when the snapshot was created in key.offset */
+	key.offset = trans->transid;
 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 
 	old = btrfs_lock_root_node(root);
 	btrfs_cow_block(trans, root, old, NULL, 0, &old);
+	btrfs_set_lock_blocking(old);
 
 	btrfs_copy_root(trans, root, old, &tmp, objectid);
 	btrfs_tree_unlock(old);
 	free_extent_buffer(old);
 
-	btrfs_set_root_bytenr(new_root_item, tmp->start);
-	btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
+	btrfs_set_root_node(new_root_item, tmp);
 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 				new_root_item);
 	btrfs_tree_unlock(tmp);
@@ -595,31 +799,60 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	if (ret)
 		goto fail;
 
+	key.offset = (u64)-1;
+	memcpy(&pending->root_key, &key, sizeof(key));
+fail:
+	kfree(new_root_item);
+	return ret;
+}
+
+static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
+				   struct btrfs_pending_snapshot *pending)
+{
+	int ret;
+	int namelen;
+	u64 index = 0;
+	struct btrfs_trans_handle *trans;
+	struct inode *parent_inode;
+	struct btrfs_root *parent_root;
+
+	parent_inode = pending->dentry->d_parent->d_inode;
+	parent_root = BTRFS_I(parent_inode)->root;
+	trans = btrfs_join_transaction(parent_root, 1);
+
 	/*
 	 * insert the directory item
 	 */
-	key.offset = (u64)-1;
 	namelen = strlen(pending->name);
-	ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
-				    pending->name, namelen,
-				    root->fs_info->sb->s_root->d_inode->i_ino,
-				    &key, BTRFS_FT_DIR);
+	ret = btrfs_set_inode_index(parent_inode, &index);
+	ret = btrfs_insert_dir_item(trans, parent_root,
+			    pending->name, namelen,
+			    parent_inode->i_ino,
+			    &pending->root_key, BTRFS_FT_DIR, index);
 
 	if (ret)
 		goto fail;
 
-	ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
-			     pending->name, strlen(pending->name), objectid,
-			     root->fs_info->sb->s_root->d_inode->i_ino);
+	btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
+	ret = btrfs_update_inode(trans, parent_root, parent_inode);
+	BUG_ON(ret);
 
-	/* Invalidate existing dcache entry for new snapshot. */
-	btrfs_invalidate_dcache_root(root, pending->name, namelen);
+	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+				 pending->root_key.objectid,
+				 parent_root->root_key.objectid,
+				 parent_inode->i_ino, index, pending->name,
+				 namelen);
+
+	BUG_ON(ret);
 
 fail:
-	kfree(new_root_item);
+	btrfs_end_transaction(trans, fs_info->fs_root);
 	return ret;
 }
 
+/*
+ * create all the snapshots we've scheduled for creation
+ */
 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 					     struct btrfs_fs_info *fs_info)
 {
@@ -627,10 +860,24 @@ static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 	struct list_head *head = &trans->transaction->pending_snapshots;
 	int ret;
 
-	while(!list_empty(head)) {
+	list_for_each_entry(pending, head, list) {
+		ret = create_pending_snapshot(trans, fs_info, pending);
+		BUG_ON(ret);
+	}
+	return 0;
+}
+
+static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
+					     struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_pending_snapshot *pending;
+	struct list_head *head = &trans->transaction->pending_snapshots;
+	int ret;
+
+	while (!list_empty(head)) {
 		pending = list_entry(head->next,
 				     struct btrfs_pending_snapshot, list);
-		ret = create_pending_snapshot(trans, fs_info, pending);
+		ret = finish_pending_snapshot(fs_info, pending);
 		BUG_ON(ret);
 		list_del(&pending->list);
 		kfree(pending->name);
@@ -639,6 +886,34 @@ static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+static void update_super_roots(struct btrfs_root *root)
+{
+	struct btrfs_root_item *root_item;
+	struct btrfs_super_block *super;
+
+	super = &root->fs_info->super_copy;
+
+	root_item = &root->fs_info->chunk_root->root_item;
+	super->chunk_root = root_item->bytenr;
+	super->chunk_root_generation = root_item->generation;
+	super->chunk_root_level = root_item->level;
+
+	root_item = &root->fs_info->tree_root->root_item;
+	super->root = root_item->bytenr;
+	super->generation = root_item->generation;
+	super->root_level = root_item->level;
+}
+
+int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
+{
+	int ret = 0;
+	spin_lock(&info->new_trans_lock);
+	if (info->running_transaction)
+		ret = info->running_transaction->in_commit;
+	spin_unlock(&info->new_trans_lock);
+	return ret;
+}
+
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root)
 {
@@ -646,22 +921,36 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	unsigned long timeout = 1;
 	struct btrfs_transaction *cur_trans;
 	struct btrfs_transaction *prev_trans = NULL;
-	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
-	struct list_head dirty_fs_roots;
-	struct extent_io_tree *pinned_copy;
 	DEFINE_WAIT(wait);
 	int ret;
+	int should_grow = 0;
+	unsigned long now = get_seconds();
+	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
+
+	btrfs_run_ordered_operations(root, 0);
+
+	/* make a pass through all the delayed refs we have so far
+	 * any runnings procs may add more while we are here
+	 */
+	ret = btrfs_run_delayed_refs(trans, root, 0);
+	BUG_ON(ret);
+
+	cur_trans = trans->transaction;
+	/*
+	 * set the flushing flag so procs in this transaction have to
+	 * start sending their work down.
+	 */
+	cur_trans->delayed_refs.flushing = 1;
 
-	INIT_LIST_HEAD(&dirty_fs_roots);
+	ret = btrfs_run_delayed_refs(trans, root, 0);
+	BUG_ON(ret);
 
 	mutex_lock(&root->fs_info->trans_mutex);
-	if (trans->transaction->in_commit) {
-		cur_trans = trans->transaction;
-		trans->transaction->use_count++;
+	if (cur_trans->in_commit) {
+		cur_trans->use_count++;
 		mutex_unlock(&root->fs_info->trans_mutex);
 		btrfs_end_transaction(trans, root);
 
-		mutex_unlock(&root->fs_info->fs_mutex);
 		ret = wait_for_commit(root, cur_trans);
 		BUG_ON(ret);
 
@@ -669,205 +958,196 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 		put_transaction(cur_trans);
 		mutex_unlock(&root->fs_info->trans_mutex);
 
-		mutex_lock(&root->fs_info->fs_mutex);
 		return 0;
 	}
 
-	pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
-	if (!pinned_copy)
-		return -ENOMEM;
-
-	extent_io_tree_init(pinned_copy,
-			     root->fs_info->btree_inode->i_mapping, GFP_NOFS);
-
 	trans->transaction->in_commit = 1;
-	cur_trans = trans->transaction;
+	trans->transaction->blocked = 1;
 	if (cur_trans->list.prev != &root->fs_info->trans_list) {
 		prev_trans = list_entry(cur_trans->list.prev,
 					struct btrfs_transaction, list);
 		if (!prev_trans->commit_done) {
 			prev_trans->use_count++;
-			mutex_unlock(&root->fs_info->fs_mutex);
 			mutex_unlock(&root->fs_info->trans_mutex);
 
 			wait_for_commit(root, prev_trans);
 
-			mutex_lock(&root->fs_info->fs_mutex);
 			mutex_lock(&root->fs_info->trans_mutex);
 			put_transaction(prev_trans);
 		}
 	}
 
+	if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
+		should_grow = 1;
+
 	do {
+		int snap_pending = 0;
 		joined = cur_trans->num_joined;
+		if (!list_empty(&trans->transaction->pending_snapshots))
+			snap_pending = 1;
+
 		WARN_ON(cur_trans != trans->transaction);
 		prepare_to_wait(&cur_trans->writer_wait, &wait,
 				TASK_UNINTERRUPTIBLE);
 
 		if (cur_trans->num_writers > 1)
 			timeout = MAX_SCHEDULE_TIMEOUT;
-		else
+		else if (should_grow)
 			timeout = 1;
 
-		mutex_unlock(&root->fs_info->fs_mutex);
 		mutex_unlock(&root->fs_info->trans_mutex);
 
-		schedule_timeout(timeout);
+		if (flush_on_commit) {
+			btrfs_start_delalloc_inodes(root, 1);
+			ret = btrfs_wait_ordered_extents(root, 0, 1);
+			BUG_ON(ret);
+		} else if (snap_pending) {
+			ret = btrfs_wait_ordered_extents(root, 0, 1);
+			BUG_ON(ret);
+		}
+
+		/*
+		 * rename don't use btrfs_join_transaction, so, once we
+		 * set the transaction to blocked above, we aren't going
+		 * to get any new ordered operations.  We can safely run
+		 * it here and no for sure that nothing new will be added
+		 * to the list
+		 */
+		btrfs_run_ordered_operations(root, 1);
+
+		smp_mb();
+		if (cur_trans->num_writers > 1 || should_grow)
+			schedule_timeout(timeout);
 
-		mutex_lock(&root->fs_info->fs_mutex);
 		mutex_lock(&root->fs_info->trans_mutex);
 		finish_wait(&cur_trans->writer_wait, &wait);
-		ret = btrfs_write_ordered_inodes(trans, root);
-
 	} while (cur_trans->num_writers > 1 ||
-		 (cur_trans->num_joined != joined));
+		 (should_grow && cur_trans->num_joined != joined));
 
 	ret = create_pending_snapshots(trans, root->fs_info);
 	BUG_ON(ret);
 
+	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+	BUG_ON(ret);
+
 	WARN_ON(cur_trans != trans->transaction);
 
-	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
-			      &dirty_fs_roots);
+	/* btrfs_commit_tree_roots is responsible for getting the
+	 * various roots consistent with each other.  Every pointer
+	 * in the tree of tree roots has to point to the most up to date
+	 * root for every subvolume and other tree.  So, we have to keep
+	 * the tree logging code from jumping in and changing any
+	 * of the trees.
+	 *
+	 * At this point in the commit, there can't be any tree-log
+	 * writers, but a little lower down we drop the trans mutex
+	 * and let new people in.  By holding the tree_log_mutex
+	 * from now until after the super is written, we avoid races
+	 * with the tree-log code.
+	 */
+	mutex_lock(&root->fs_info->tree_log_mutex);
+
+	ret = commit_fs_roots(trans, root);
 	BUG_ON(ret);
 
-	ret = btrfs_commit_tree_roots(trans, root);
+	/* commit_fs_roots gets rid of all the tree log roots, it is now
+	 * safe to free the root of tree log roots
+	 */
+	btrfs_free_log_root_tree(trans, root->fs_info);
+
+	ret = commit_cowonly_roots(trans, root);
 	BUG_ON(ret);
 
+	btrfs_prepare_extent_commit(trans, root);
+
 	cur_trans = root->fs_info->running_transaction;
 	spin_lock(&root->fs_info->new_trans_lock);
 	root->fs_info->running_transaction = NULL;
 	spin_unlock(&root->fs_info->new_trans_lock);
-	btrfs_set_super_generation(&root->fs_info->super_copy,
-				   cur_trans->transid);
-	btrfs_set_super_root(&root->fs_info->super_copy,
-			     root->fs_info->tree_root->node->start);
-	btrfs_set_super_root_level(&root->fs_info->super_copy,
-			   btrfs_header_level(root->fs_info->tree_root->node));
-
-	btrfs_set_super_chunk_root(&root->fs_info->super_copy,
-				   chunk_root->node->start);
-	btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
-					 btrfs_header_level(chunk_root->node));
+
+	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
+			    root->fs_info->tree_root->node);
+	switch_commit_root(root->fs_info->tree_root);
+
+	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
+			    root->fs_info->chunk_root->node);
+	switch_commit_root(root->fs_info->chunk_root);
+
+	update_super_roots(root);
+
+	if (!root->fs_info->log_root_recovering) {
+		btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
+		btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
+	}
+
 	memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 	       sizeof(root->fs_info->super_copy));
 
-	btrfs_copy_pinned(root, pinned_copy);
+	trans->transaction->blocked = 0;
+
+	wake_up(&root->fs_info->transaction_wait);
 
 	mutex_unlock(&root->fs_info->trans_mutex);
-	mutex_unlock(&root->fs_info->fs_mutex);
 	ret = btrfs_write_and_wait_transaction(trans, root);
 	BUG_ON(ret);
-	write_ctree_super(trans, root);
+	write_ctree_super(trans, root, 0);
 
-	mutex_lock(&root->fs_info->fs_mutex);
-	btrfs_finish_extent_commit(trans, root, pinned_copy);
-	mutex_lock(&root->fs_info->trans_mutex);
+	/*
+	 * the super is written, we can safely allow the tree-loggers
+	 * to go about their business
+	 */
+	mutex_unlock(&root->fs_info->tree_log_mutex);
+
+	btrfs_finish_extent_commit(trans, root);
 
-	kfree(pinned_copy);
+	/* do the directory inserts of any pending snapshot creations */
+	finish_pending_snapshots(trans, root->fs_info);
+
+	mutex_lock(&root->fs_info->trans_mutex);
 
 	cur_trans->commit_done = 1;
+
 	root->fs_info->last_trans_committed = cur_trans->transid;
+
 	wake_up(&cur_trans->commit_wait);
+
 	put_transaction(cur_trans);
 	put_transaction(cur_trans);
 
-	if (root->fs_info->closing)
-		list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
-	else
-		list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
-
 	mutex_unlock(&root->fs_info->trans_mutex);
+
+	if (current->journal_info == trans)
+		current->journal_info = NULL;
+
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
 
-	if (root->fs_info->closing) {
-		mutex_unlock(&root->fs_info->fs_mutex);
-		drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
-		mutex_lock(&root->fs_info->fs_mutex);
-	}
+	if (current != root->fs_info->transaction_kthread)
+		btrfs_run_delayed_iputs(root);
+
 	return ret;
 }
 
+/*
+ * interface function to delete all the snapshots we have scheduled for deletion
+ */
 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 {
-	struct list_head dirty_roots;
-	INIT_LIST_HEAD(&dirty_roots);
-
-	mutex_lock(&root->fs_info->trans_mutex);
-	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
-	mutex_unlock(&root->fs_info->trans_mutex);
-
-	if (!list_empty(&dirty_roots)) {
-		drop_dirty_roots(root, &dirty_roots);
-	}
-	return 0;
-}
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-void btrfs_transaction_cleaner(void *p)
-#else
-void btrfs_transaction_cleaner(struct work_struct *work)
-#endif
-{
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-	struct btrfs_fs_info *fs_info = p;
-#else
-	struct btrfs_fs_info *fs_info = container_of(work,
-						     struct btrfs_fs_info,
-						     trans_work.work);
+	LIST_HEAD(list);
+	struct btrfs_fs_info *fs_info = root->fs_info;
 
-#endif
-	struct btrfs_root *root = fs_info->tree_root;
-	struct btrfs_transaction *cur;
-	struct btrfs_trans_handle *trans;
-	unsigned long now;
-	unsigned long delay = HZ * 30;
-	int ret;
+	mutex_lock(&fs_info->trans_mutex);
+	list_splice_init(&fs_info->dead_roots, &list);
+	mutex_unlock(&fs_info->trans_mutex);
 
-	mutex_lock(&root->fs_info->fs_mutex);
-	if (root->fs_info->closing)
-		goto out;
+	while (!list_empty(&list)) {
+		root = list_entry(list.next, struct btrfs_root, root_list);
+		list_del(&root->root_list);
 
-	mutex_lock(&root->fs_info->trans_mutex);
-	cur = root->fs_info->running_transaction;
-	if (!cur) {
-		mutex_unlock(&root->fs_info->trans_mutex);
-		goto out;
-	}
-	now = get_seconds();
-	if (now < cur->start_time || now - cur->start_time < 30) {
-		mutex_unlock(&root->fs_info->trans_mutex);
-		delay = HZ * 5;
-		goto out;
+		if (btrfs_header_backref_rev(root->node) <
+		    BTRFS_MIXED_BACKREF_REV)
+			btrfs_drop_snapshot(root, 0);
+		else
+			btrfs_drop_snapshot(root, 1);
 	}
-	mutex_unlock(&root->fs_info->trans_mutex);
-	btrfs_defrag_dirty_roots(root->fs_info);
-	trans = btrfs_start_transaction(root, 1);
-	ret = btrfs_commit_transaction(trans, root);
-out:
-	mutex_unlock(&root->fs_info->fs_mutex);
-	btrfs_clean_old_snapshots(root);
-	btrfs_transaction_queue_work(root, delay);
-}
-
-void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
-{
-	if (!root->fs_info->closing)
-		queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
-}
-
-void btrfs_transaction_flush_work(struct btrfs_root *root)
-{
-	cancel_delayed_work(&root->fs_info->trans_work);
-	flush_workqueue(trans_wq);
-}
-
-void __init btrfs_init_transaction_sys(void)
-{
-	trans_wq = create_workqueue("btrfs-transaction");
-}
-
-void btrfs_exit_transaction_sys(void)
-{
-	destroy_workqueue(trans_wq);
+	return 0;
 }
-