Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable

* git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable:
  Btrfs: try to free metadata pages when we free btree blocks
  Btrfs: add extra flushing for renames and truncates
  Btrfs: make sure btrfs_update_delayed_ref doesn't increase ref_mod
  Btrfs: optimize fsyncs on old files
  Btrfs: tree logging unlink/rename fixes
  Btrfs: Make sure i_nlink doesn't hit zero too soon during log replay
  Btrfs: limit balancing work while flushing delayed refs
  Btrfs: readahead checksums during btrfs_finish_ordered_io
  Btrfs: leave btree locks spinning more often
  Btrfs: Only let very young transactions grow during commit
  Btrfs: Check for a blocking lock before taking the spin
  Btrfs: reduce stack in cow_file_range
  Btrfs: reduce stalls during transaction commit
  Btrfs: process the delayed reference queue in clusters
  Btrfs: try to cleanup delayed refs while freeing extents
  Btrfs: reduce stack usage in some crucial tree balancing functions
  Btrfs: do extent allocation and reference count updates in the background
  Btrfs: don't preallocate metadata blocks during btrfs_search_slot

fs/btrfs/Makefile

@@ -8,7 +8,7 @@ btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
 	   extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \
 	   extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
 	   ref-cache.o export.o tree-log.o acl.o free-space-cache.o zlib.o \
-	   compression.o
+	   compression.o delayed-ref.o
 else
 
 # Normal Makefile

fs/btrfs/btrfs_inode.h

@@ -66,6 +66,12 @@ struct btrfs_inode {
 	 */
 	struct list_head delalloc_inodes;
 
+	/*
+	 * list for tracking inodes that must be sent to disk before a
+	 * rename or truncate commit
+	 */
+	struct list_head ordered_operations;
+
 	/* the space_info for where this inode's data allocations are done */
 	struct btrfs_space_info *space_info;
 
@@ -86,12 +92,6 @@ struct btrfs_inode {
 	 */
 	u64 logged_trans;
 
-	/*
-	 * trans that last made a change that should be fully fsync'd.  This
-	 * gets reset to zero each time the inode is logged
-	 */
-	u64 log_dirty_trans;
-
 	/* total number of bytes pending delalloc, used by stat to calc the
 	 * real block usage of the file
 	 */
@@ -121,6 +121,25 @@ struct btrfs_inode {
 	/* the start of block group preferred for allocations. */
 	u64 block_group;
 
+	/* the fsync log has some corner cases that mean we have to check
+	 * directories to see if any unlinks have been done before
+	 * the directory was logged.  See tree-log.c for all the
+	 * details
+	 */
+	u64 last_unlink_trans;
+
+	/*
+	 * ordered_data_close is set by truncate when a file that used
+	 * to have good data has been truncated to zero.  When it is set
+	 * the btrfs file release call will add this inode to the
+	 * ordered operations list so that we make sure to flush out any
+	 * new data the application may have written before commit.
+	 *
+	 * yes, its silly to have a single bitflag, but we might grow more
+	 * of these.
+	 */
+	unsigned ordered_data_close:1;
+
 	struct inode vfs_inode;
 };
 

fs/btrfs/ctree.h

@@ -45,6 +45,13 @@ struct btrfs_ordered_sum;
 
 #define BTRFS_MAX_LEVEL 8
 
+/*
+ * files bigger than this get some pre-flushing when they are added
+ * to the ordered operations list.  That way we limit the total
+ * work done by the commit
+ */
+#define BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT (8 * 1024 * 1024)
+
 /* holds pointers to all of the tree roots */
 #define BTRFS_ROOT_TREE_OBJECTID 1ULL
 
@@ -401,15 +408,16 @@ struct btrfs_path {
 	int locks[BTRFS_MAX_LEVEL];
 	int reada;
 	/* keep some upper locks as we walk down */
-	int keep_locks;
-	int skip_locking;
 	int lowest_level;
 
 	/*
 	 * set by btrfs_split_item, tells search_slot to keep all locks
 	 * and to force calls to keep space in the nodes
 	 */
-	int search_for_split;
+	unsigned int search_for_split:1;
+	unsigned int keep_locks:1;
+	unsigned int skip_locking:1;
+	unsigned int leave_spinning:1;
 };
 
 /*
@@ -688,15 +696,18 @@ struct btrfs_fs_info {
 	struct rb_root block_group_cache_tree;
 
 	struct extent_io_tree pinned_extents;
-	struct extent_io_tree pending_del;
-	struct extent_io_tree extent_ins;
 
 	/* logical->physical extent mapping */
 	struct btrfs_mapping_tree mapping_tree;
 
 	u64 generation;
 	u64 last_trans_committed;
-	u64 last_trans_new_blockgroup;
+
+	/*
+	 * this is updated to the current trans every time a full commit
+	 * is required instead of the faster short fsync log commits
+	 */
+	u64 last_trans_log_full_commit;
 	u64 open_ioctl_trans;
 	unsigned long mount_opt;
 	u64 max_extent;
@@ -717,12 +728,21 @@ struct btrfs_fs_info {
 	struct mutex tree_log_mutex;
 	struct mutex transaction_kthread_mutex;
 	struct mutex cleaner_mutex;
-	struct mutex extent_ins_mutex;
 	struct mutex pinned_mutex;
 	struct mutex chunk_mutex;
 	struct mutex drop_mutex;
 	struct mutex volume_mutex;
 	struct mutex tree_reloc_mutex;
+
+	/*
+	 * this protects the ordered operations list only while we are
+	 * processing all of the entries on it.  This way we make
+	 * sure the commit code doesn't find the list temporarily empty
+	 * because another function happens to be doing non-waiting preflush
+	 * before jumping into the main commit.
+	 */
+	struct mutex ordered_operations_mutex;
+
 	struct list_head trans_list;
 	struct list_head hashers;
 	struct list_head dead_roots;
@@ -737,9 +757,28 @@ struct btrfs_fs_info {
 	 * ordered extents
 	 */
 	spinlock_t ordered_extent_lock;
+
+	/*
+	 * all of the data=ordered extents pending writeback
+	 * these can span multiple transactions and basically include
+	 * every dirty data page that isn't from nodatacow
+	 */
 	struct list_head ordered_extents;
+
+	/*
+	 * all of the inodes that have delalloc bytes.  It is possible for
+	 * this list to be empty even when there is still dirty data=ordered
+	 * extents waiting to finish IO.
+	 */
 	struct list_head delalloc_inodes;
 
+	/*
+	 * special rename and truncate targets that must be on disk before
+	 * we're allowed to commit.  This is basically the ext3 style
+	 * data=ordered list.
+	 */
+	struct list_head ordered_operations;
+
 	/*
 	 * there is a pool of worker threads for checksumming during writes
 	 * and a pool for checksumming after reads.  This is because readers
@@ -781,6 +820,11 @@ struct btrfs_fs_info {
 	atomic_t throttle_gen;
 
 	u64 total_pinned;
+
+	/* protected by the delalloc lock, used to keep from writing
+	 * metadata until there is a nice batch
+	 */
+	u64 dirty_metadata_bytes;
 	struct list_head dirty_cowonly_roots;
 
 	struct btrfs_fs_devices *fs_devices;
@@ -1704,18 +1748,15 @@ static inline struct dentry *fdentry(struct file *file)
 }
 
 /* extent-tree.c */
+int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, unsigned long count);
 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
-int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, u64 bytenr,
-			    u64 num_bytes, u32 *refs);
 int btrfs_update_pinned_extents(struct btrfs_root *root,
 				u64 bytenr, u64 num, int pin);
 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root, struct extent_buffer *leaf);
 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root, u64 objectid, u64 bytenr);
-int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
-			 struct btrfs_root *root);
 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy);
 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 						 struct btrfs_fs_info *info,
@@ -1777,7 +1818,7 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 			 u64 root_objectid, u64 ref_generation,
 			 u64 owner_objectid);
 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
-			    struct btrfs_root *root, u64 bytenr,
+			    struct btrfs_root *root, u64 bytenr, u64 num_bytes,
 			    u64 orig_parent, u64 parent,
 			    u64 root_objectid, u64 ref_generation,
 			    u64 owner_objectid);
@@ -1838,7 +1879,7 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
 int btrfs_cow_block(struct btrfs_trans_handle *trans,
 		    struct btrfs_root *root, struct extent_buffer *buf,
 		    struct extent_buffer *parent, int parent_slot,
-		    struct extent_buffer **cow_ret, u64 prealloc_dest);
+		    struct extent_buffer **cow_ret);
 int btrfs_copy_root(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root,
 		      struct extent_buffer *buf,

fs/btrfs/delayed-ref.h

@@ -0,0 +1,193 @@
+/*
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+#ifndef __DELAYED_REF__
+#define __DELAYED_REF__
+
+/* these are the possible values of struct btrfs_delayed_ref->action */
+#define BTRFS_ADD_DELAYED_REF    1 /* add one backref to the tree */
+#define BTRFS_DROP_DELAYED_REF   2 /* delete one backref from the tree */
+#define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
+#define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
+
+struct btrfs_delayed_ref_node {
+	struct rb_node rb_node;
+
+	/* the starting bytenr of the extent */
+	u64 bytenr;
+
+	/* the parent our backref will point to */
+	u64 parent;
+
+	/* the size of the extent */
+	u64 num_bytes;
+
+	/* ref count on this data structure */
+	atomic_t refs;
+
+	/*
+	 * how many refs is this entry adding or deleting.  For
+	 * head refs, this may be a negative number because it is keeping
+	 * track of the total mods done to the reference count.
+	 * For individual refs, this will always be a positive number
+	 *
+	 * It may be more than one, since it is possible for a single
+	 * parent to have more than one ref on an extent
+	 */
+	int ref_mod;
+
+	/* is this node still in the rbtree? */
+	unsigned int in_tree:1;
+};
+
+/*
+ * the head refs are used to hold a lock on a given extent, which allows us
+ * to make sure that only one process is running the delayed refs
+ * at a time for a single extent.  They also store the sum of all the
+ * reference count modifications we've queued up.
+ */
+struct btrfs_delayed_ref_head {
+	struct btrfs_delayed_ref_node node;
+
+	/*
+	 * the mutex is held while running the refs, and it is also
+	 * held when checking the sum of reference modifications.
+	 */
+	struct mutex mutex;
+
+	struct list_head cluster;
+
+	/*
+	 * when a new extent is allocated, it is just reserved in memory
+	 * The actual extent isn't inserted into the extent allocation tree
+	 * until the delayed ref is processed.  must_insert_reserved is
+	 * used to flag a delayed ref so the accounting can be updated
+	 * when a full insert is done.
+	 *
+	 * It is possible the extent will be freed before it is ever
+	 * inserted into the extent allocation tree.  In this case
+	 * we need to update the in ram accounting to properly reflect
+	 * the free has happened.
+	 */
+	unsigned int must_insert_reserved:1;
+};
+
+struct btrfs_delayed_ref {
+	struct btrfs_delayed_ref_node node;
+
+	/* the root objectid our ref will point to */
+	u64 root;
+
+	/* the generation for the backref */
+	u64 generation;
+
+	/* owner_objectid of the backref  */
+	u64 owner_objectid;
+
+	/* operation done by this entry in the rbtree */
+	u8 action;
+
+	/* if pin == 1, when the extent is freed it will be pinned until
+	 * transaction commit
+	 */
+	unsigned int pin:1;
+};
+
+struct btrfs_delayed_ref_root {
+	struct rb_root root;
+
+	/* this spin lock protects the rbtree and the entries inside */
+	spinlock_t lock;
+
+	/* how many delayed ref updates we've queued, used by the
+	 * throttling code
+	 */
+	unsigned long num_entries;
+
+	/* total number of head nodes in tree */
+	unsigned long num_heads;
+
+	/* total number of head nodes ready for processing */
+	unsigned long num_heads_ready;
+
+	/*
+	 * set when the tree is flushing before a transaction commit,
+	 * used by the throttling code to decide if new updates need
+	 * to be run right away
+	 */
+	int flushing;
+
+	u64 run_delayed_start;
+};
+
+static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
+{
+	WARN_ON(atomic_read(&ref->refs) == 0);
+	if (atomic_dec_and_test(&ref->refs)) {
+		WARN_ON(ref->in_tree);
+		kfree(ref);
+	}
+}
+
+int btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
+			  u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
+			  u64 ref_generation, u64 owner_objectid, int action,
+			  int pin);
+
+struct btrfs_delayed_ref_head *
+btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
+int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr);
+int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 num_bytes, u32 *refs);
+int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans,
+			  u64 bytenr, u64 num_bytes, u64 orig_parent,
+			  u64 parent, u64 orig_ref_root, u64 ref_root,
+			  u64 orig_ref_generation, u64 ref_generation,
+			  u64 owner_objectid, int pin);
+int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
+			   struct btrfs_delayed_ref_head *head);
+int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
+			   struct list_head *cluster, u64 search_start);
+/*
+ * a node might live in a head or a regular ref, this lets you
+ * test for the proper type to use.
+ */
+static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
+{
+	return node->parent == (u64)-1;
+}
+
+/*
+ * helper functions to cast a node into its container
+ */
+static inline struct btrfs_delayed_ref *
+btrfs_delayed_node_to_ref(struct btrfs_delayed_ref_node *node)
+{
+	WARN_ON(btrfs_delayed_ref_is_head(node));
+	return container_of(node, struct btrfs_delayed_ref, node);
+
+}
+
+static inline struct btrfs_delayed_ref_head *
+btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
+{
+	WARN_ON(!btrfs_delayed_ref_is_head(node));
+	return container_of(node, struct btrfs_delayed_ref_head, node);
+
+}
+#endif

fs/btrfs/dir-item.c

@@ -145,7 +145,10 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
 	key.objectid = dir;
 	btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY);
 	key.offset = btrfs_name_hash(name, name_len);
+
 	path = btrfs_alloc_path();
+	path->leave_spinning = 1;
+
 	data_size = sizeof(*dir_item) + name_len;
 	dir_item = insert_with_overflow(trans, root, path, &key, data_size,
 					name, name_len);

fs/btrfs/disk-io.c

@@ -668,14 +668,31 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 {
 	struct extent_io_tree *tree;
-	tree = &BTRFS_I(page->mapping->host)->io_tree;
+	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
+	struct extent_buffer *eb;
+	int was_dirty;
 
-	if (current->flags & PF_MEMALLOC) {
+	tree = &BTRFS_I(page->mapping->host)->io_tree;
-		redirty_page_for_writepage(wbc, page);
+	if (!(current->flags & PF_MEMALLOC)) {
-		unlock_page(page);
+		return extent_write_full_page(tree, page,
-		return 0;
+					      btree_get_extent, wbc);
 	}
-	return extent_write_full_page(tree, page, btree_get_extent, wbc);
+
+	redirty_page_for_writepage(wbc, page);
+	eb = btrfs_find_tree_block(root, page_offset(page),
+				      PAGE_CACHE_SIZE);
+	WARN_ON(!eb);
+
+	was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
+	if (!was_dirty) {
+		spin_lock(&root->fs_info->delalloc_lock);
+		root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
+		spin_unlock(&root->fs_info->delalloc_lock);
+	}
+	free_extent_buffer(eb);
+
+	unlock_page(page);
+	return 0;
 }
 
 static int btree_writepages(struct address_space *mapping,
@@ -684,15 +701,15 @@ static int btree_writepages(struct address_space *mapping,
 	struct extent_io_tree *tree;
 	tree = &BTRFS_I(mapping->host)->io_tree;
 	if (wbc->sync_mode == WB_SYNC_NONE) {
+		struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 		u64 num_dirty;
-		u64 start = 0;
 		unsigned long thresh = 32 * 1024 * 1024;
 
 		if (wbc->for_kupdate)
 			return 0;
 
-		num_dirty = count_range_bits(tree, &start, (u64)-1,
+		/* this is a bit racy, but that's ok */
-					     thresh, EXTENT_DIRTY);
+		num_dirty = root->fs_info->dirty_metadata_bytes;
 		if (num_dirty < thresh)
 			return 0;
 	}
@@ -859,9 +876,17 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	    root->fs_info->running_transaction->transid) {
 		btrfs_assert_tree_locked(buf);
 
-		/* ugh, clear_extent_buffer_dirty can be expensive */
+		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
-		btrfs_set_lock_blocking(buf);
+			spin_lock(&root->fs_info->delalloc_lock);
+			if (root->fs_info->dirty_metadata_bytes >= buf->len)
+				root->fs_info->dirty_metadata_bytes -= buf->len;
+			else
+				WARN_ON(1);
+			spin_unlock(&root->fs_info->delalloc_lock);
+		}
 
+		/* ugh, clear_extent_buffer_dirty needs to lock the page */
+		btrfs_set_lock_blocking(buf);
 		clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 					  buf);
 	}
@@ -1471,12 +1496,6 @@ static int transaction_kthread(void *arg)
 		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
 
-		if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
-			printk(KERN_INFO "btrfs: total reference cache "
-			       "size %llu\n",
-			       root->fs_info->total_ref_cache_size);
-		}
-
 		mutex_lock(&root->fs_info->trans_mutex);
 		cur = root->fs_info->running_transaction;
 		if (!cur) {
@@ -1493,6 +1512,7 @@ static int transaction_kthread(void *arg)
 		mutex_unlock(&root->fs_info->trans_mutex);
 		trans = btrfs_start_transaction(root, 1);
 		ret = btrfs_commit_transaction(trans, root);
+
 sleep:
 		wake_up_process(root->fs_info->cleaner_kthread);
 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
@@ -1552,6 +1572,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	INIT_LIST_HEAD(&fs_info->dead_roots);
 	INIT_LIST_HEAD(&fs_info->hashers);
 	INIT_LIST_HEAD(&fs_info->delalloc_inodes);
+	INIT_LIST_HEAD(&fs_info->ordered_operations);
 	spin_lock_init(&fs_info->delalloc_lock);
 	spin_lock_init(&fs_info->new_trans_lock);
 	spin_lock_init(&fs_info->ref_cache_lock);
@@ -1611,10 +1632,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 
 	extent_io_tree_init(&fs_info->pinned_extents,
 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
-	extent_io_tree_init(&fs_info->pending_del,
-			     fs_info->btree_inode->i_mapping, GFP_NOFS);
-	extent_io_tree_init(&fs_info->extent_ins,
-			     fs_info->btree_inode->i_mapping, GFP_NOFS);
 	fs_info->do_barriers = 1;
 
 	INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
@@ -1627,9 +1644,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	insert_inode_hash(fs_info->btree_inode);
 
 	mutex_init(&fs_info->trans_mutex);
+	mutex_init(&fs_info->ordered_operations_mutex);
 	mutex_init(&fs_info->tree_log_mutex);
 	mutex_init(&fs_info->drop_mutex);
-	mutex_init(&fs_info->extent_ins_mutex);
 	mutex_init(&fs_info->pinned_mutex);
 	mutex_init(&fs_info->chunk_mutex);
 	mutex_init(&fs_info->transaction_kthread_mutex);
@@ -2358,8 +2375,7 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 	u64 transid = btrfs_header_generation(buf);
 	struct inode *btree_inode = root->fs_info->btree_inode;
-
+	int was_dirty;
-	btrfs_set_lock_blocking(buf);
 
 	btrfs_assert_tree_locked(buf);
 	if (transid != root->fs_info->generation) {
@@ -2370,7 +2386,13 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 			(unsigned long long)root->fs_info->generation);
 		WARN_ON(1);
 	}
-	set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
+	was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
+					    buf);
+	if (!was_dirty) {
+		spin_lock(&root->fs_info->delalloc_lock);
+		root->fs_info->dirty_metadata_bytes += buf->len;
+		spin_unlock(&root->fs_info->delalloc_lock);
+	}
 }
 
 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
@@ -2410,6 +2432,7 @@ int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
 int btree_lock_page_hook(struct page *page)
 {
 	struct inode *inode = page->mapping->host;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 	struct extent_buffer *eb;
 	unsigned long len;
@@ -2425,6 +2448,16 @@ int btree_lock_page_hook(struct page *page)
 
 	btrfs_tree_lock(eb);
 	btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
+
+	if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
+		spin_lock(&root->fs_info->delalloc_lock);
+		if (root->fs_info->dirty_metadata_bytes >= eb->len)
+			root->fs_info->dirty_metadata_bytes -= eb->len;
+		else
+			WARN_ON(1);
+		spin_unlock(&root->fs_info->delalloc_lock);
+	}
+
 	btrfs_tree_unlock(eb);
 	free_extent_buffer(eb);
 out:

fs/btrfs/disk-io.h

@@ -72,6 +72,7 @@ int btrfs_insert_dev_radix(struct btrfs_root *root,
 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr);
 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root);
 void btrfs_mark_buffer_dirty(struct extent_buffer *buf);
+void btrfs_mark_buffer_dirty_nonblocking(struct extent_buffer *buf);
 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid);
 int btrfs_set_buffer_uptodate(struct extent_buffer *buf);
 int wait_on_tree_block_writeback(struct btrfs_root *root,

fs/btrfs/extent_io.c

@@ -3124,20 +3124,15 @@ void free_extent_buffer(struct extent_buffer *eb)
 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
 			      struct extent_buffer *eb)
 {
-	int set;
 	unsigned long i;
 	unsigned long num_pages;
 	struct page *page;
 
-	u64 start = eb->start;
-	u64 end = start + eb->len - 1;
-
-	set = clear_extent_dirty(tree, start, end, GFP_NOFS);
 	num_pages = num_extent_pages(eb->start, eb->len);
 
 	for (i = 0; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
-		if (!set && !PageDirty(page))
+		if (!PageDirty(page))
 			continue;
 
 		lock_page(page);
@@ -3146,22 +3141,6 @@ int clear_extent_buffer_dirty(struct extent_io_tree *tree,
 		else
 			set_page_private(page, EXTENT_PAGE_PRIVATE);
 
-		/*
-		 * if we're on the last page or the first page and the
-		 * block isn't aligned on a page boundary, do extra checks
-		 * to make sure we don't clean page that is partially dirty
-		 */
-		if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
-		    ((i == num_pages - 1) &&
-		     ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
-			start = (u64)page->index << PAGE_CACHE_SHIFT;
-			end  = start + PAGE_CACHE_SIZE - 1;
-			if (test_range_bit(tree, start, end,
-					   EXTENT_DIRTY, 0)) {
-				unlock_page(page);
-				continue;
-			}
-		}
 		clear_page_dirty_for_io(page);
 		spin_lock_irq(&page->mapping->tree_lock);
 		if (!PageDirty(page)) {
@@ -3187,29 +3166,13 @@ int set_extent_buffer_dirty(struct extent_io_tree *tree,
 {
 	unsigned long i;
 	unsigned long num_pages;
+	int was_dirty = 0;
 
+	was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 	num_pages = num_extent_pages(eb->start, eb->len);
-	for (i = 0; i < num_pages; i++) {
+	for (i = 0; i < num_pages; i++)
-		struct page *page = extent_buffer_page(eb, i);
-		/* writepage may need to do something special for the
-		 * first page, we have to make sure page->private is
-		 * properly set.  releasepage may drop page->private
-		 * on us if the page isn't already dirty.
-		 */
-		lock_page(page);
-		if (i == 0) {
-			set_page_extent_head(page, eb->len);
-		} else if (PagePrivate(page) &&
-			   page->private != EXTENT_PAGE_PRIVATE) {
-			set_page_extent_mapped(page);
-		}
 		__set_page_dirty_nobuffers(extent_buffer_page(eb, i));
-		set_extent_dirty(tree, page_offset(page),
+	return was_dirty;
-				 page_offset(page) + PAGE_CACHE_SIZE - 1,
-				 GFP_NOFS);
-		unlock_page(page);
-	}
-	return 0;
 }
 
 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
@@ -3789,6 +3752,10 @@ int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
 		ret = 0;
 		goto out;
 	}
+	if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
+		ret = 0;
+		goto out;
+	}
 	/* at this point we can safely release the extent buffer */
 	num_pages = num_extent_pages(eb->start, eb->len);
 	for (i = 0; i < num_pages; i++)

fs/btrfs/extent_io.h

@@ -25,6 +25,7 @@
 /* these are bit numbers for test/set bit */
 #define EXTENT_BUFFER_UPTODATE 0
 #define EXTENT_BUFFER_BLOCKING 1
+#define EXTENT_BUFFER_DIRTY 2
 
 /*
  * page->private values.  Every page that is controlled by the extent
@@ -254,6 +255,8 @@ int clear_extent_buffer_dirty(struct extent_io_tree *tree,
 			      struct extent_buffer *eb);
 int set_extent_buffer_dirty(struct extent_io_tree *tree,
 			     struct extent_buffer *eb);
+int test_extent_buffer_dirty(struct extent_io_tree *tree,
+			     struct extent_buffer *eb);
 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
 			       struct extent_buffer *eb);
 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,

fs/btrfs/file-item.c

@@ -52,6 +52,7 @@ int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
 	file_key.offset = pos;
 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
 
+	path->leave_spinning = 1;
 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 				      sizeof(*item));
 	if (ret < 0)
@@ -523,6 +524,7 @@ int btrfs_del_csums(struct btrfs_trans_handle *trans,
 		key.offset = end_byte - 1;
 		key.type = BTRFS_EXTENT_CSUM_KEY;
 
+		path->leave_spinning = 1;
 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 		if (ret > 0) {
 			if (path->slots[0] == 0)
@@ -757,8 +759,10 @@ insert:
 	} else {
 		ins_size = csum_size;
 	}
+	path->leave_spinning = 1;
 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 				      ins_size);
+	path->leave_spinning = 0;
 	if (ret < 0)
 		goto fail_unlock;
 	if (ret != 0) {
@@ -776,7 +780,6 @@ found:
 	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
 				      btrfs_item_size_nr(leaf, path->slots[0]));
 	eb_token = NULL;
-	cond_resched();
 next_sector:
 
 	if (!eb_token ||
@@ -817,9 +820,9 @@ next_sector:
 		eb_token = NULL;
 	}
 	btrfs_mark_buffer_dirty(path->nodes[0]);
-	cond_resched();
 	if (total_bytes < sums->len) {
 		btrfs_release_path(root, path);
+		cond_resched();
 		goto again;
 	}
 out:

fs/btrfs/file.c

@@ -606,6 +606,7 @@ next_slot:
 			btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 
 			btrfs_release_path(root, path);
+			path->leave_spinning = 1;
 			ret = btrfs_insert_empty_item(trans, root, path, &ins,
 						      sizeof(*extent));
 			BUG_ON(ret);
@@ -639,17 +640,22 @@ next_slot:
 							ram_bytes);
 			btrfs_set_file_extent_type(leaf, extent, found_type);
 
+			btrfs_unlock_up_safe(path, 1);
 			btrfs_mark_buffer_dirty(path->nodes[0]);
+			btrfs_set_lock_blocking(path->nodes[0]);
 
 			if (disk_bytenr != 0) {
 				ret = btrfs_update_extent_ref(trans, root,
-						disk_bytenr, orig_parent,
+						disk_bytenr,
+						le64_to_cpu(old.disk_num_bytes),
+						orig_parent,
 						leaf->start,
 						root->root_key.objectid,
 						trans->transid, ins.objectid);
 
 				BUG_ON(ret);
 			}
+			path->leave_spinning = 0;
 			btrfs_release_path(root, path);
 			if (disk_bytenr != 0)
 				inode_add_bytes(inode, extent_end - end);
@@ -912,7 +918,7 @@ again:
 	btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 
 	if (orig_parent != leaf->start) {
-		ret = btrfs_update_extent_ref(trans, root, bytenr,
+		ret = btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
 					      orig_parent, leaf->start,
 					      root->root_key.objectid,
 					      trans->transid, inode->i_ino);
@@ -1155,6 +1161,20 @@ out_nolock:
 		page_cache_release(pinned[1]);
 	*ppos = pos;
 
+	/*
+	 * we want to make sure fsync finds this change
+	 * but we haven't joined a transaction running right now.
+	 *
+	 * Later on, someone is sure to update the inode and get the
+	 * real transid recorded.
+	 *
+	 * We set last_trans now to the fs_info generation + 1,
+	 * this will either be one more than the running transaction
+	 * or the generation used for the next transaction if there isn't
+	 * one running right now.
+	 */
+	BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
+
 	if (num_written > 0 && will_write) {
 		struct btrfs_trans_handle *trans;
 
@@ -1167,8 +1187,11 @@ out_nolock:
 			ret = btrfs_log_dentry_safe(trans, root,
 						    file->f_dentry);
 			if (ret == 0) {
-				btrfs_sync_log(trans, root);
+				ret = btrfs_sync_log(trans, root);
-				btrfs_end_transaction(trans, root);
+				if (ret == 0)
+					btrfs_end_transaction(trans, root);
+				else
+					btrfs_commit_transaction(trans, root);
 			} else {
 				btrfs_commit_transaction(trans, root);
 			}
@@ -1185,6 +1208,18 @@ out_nolock:
 
 int btrfs_release_file(struct inode *inode, struct file *filp)
 {
+	/*
+	 * ordered_data_close is set by settattr when we are about to truncate
+	 * a file from a non-zero size to a zero size.  This tries to
+	 * flush down new bytes that may have been written if the
+	 * application were using truncate to replace a file in place.
+	 */
+	if (BTRFS_I(inode)->ordered_data_close) {
+		BTRFS_I(inode)->ordered_data_close = 0;
+		btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
+		if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
+			filemap_flush(inode->i_mapping);
+	}
 	if (filp->private_data)
 		btrfs_ioctl_trans_end(filp);
 	return 0;
@@ -1260,8 +1295,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 	if (ret > 0) {
 		ret = btrfs_commit_transaction(trans, root);
 	} else {
-		btrfs_sync_log(trans, root);
+		ret = btrfs_sync_log(trans, root);
-		ret = btrfs_end_transaction(trans, root);
+		if (ret == 0)
+			ret = btrfs_end_transaction(trans, root);
+		else
+			ret = btrfs_commit_transaction(trans, root);
 	}
 	mutex_lock(&dentry->d_inode->i_mutex);
 out:

fs/btrfs/inode-item.c

@@ -73,6 +73,8 @@ int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
 	if (!path)
 		return -ENOMEM;
 
+	path->leave_spinning = 1;
+
 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 	if (ret > 0) {
 		ret = -ENOENT;
@@ -127,6 +129,7 @@ int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
 	if (!path)
 		return -ENOMEM;
 
+	path->leave_spinning = 1;
 	ret = btrfs_insert_empty_item(trans, root, path, &key,
 				      ins_len);
 	if (ret == -EEXIST) {

fs/btrfs/inode.c

@@ -134,6 +134,7 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 	if (!path)
 		return -ENOMEM;
 
+	path->leave_spinning = 1;
 	btrfs_set_trans_block_group(trans, inode);
 
 	key.objectid = inode->i_ino;
@@ -167,9 +168,9 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 			cur_size = min_t(unsigned long, compressed_size,
 				       PAGE_CACHE_SIZE);
 
-			kaddr = kmap(cpage);
+			kaddr = kmap_atomic(cpage, KM_USER0);
 			write_extent_buffer(leaf, kaddr, ptr, cur_size);
-			kunmap(cpage);
+			kunmap_atomic(kaddr, KM_USER0);
 
 			i++;
 			ptr += cur_size;
@@ -204,7 +205,7 @@ fail:
  * does the checks required to make sure the data is small enough
  * to fit as an inline extent.
  */
-static int cow_file_range_inline(struct btrfs_trans_handle *trans,
+static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 				 struct btrfs_root *root,
 				 struct inode *inode, u64 start, u64 end,
 				 size_t compressed_size,
@@ -854,11 +855,6 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 	u64 cur_end;
 	int limit = 10 * 1024 * 1042;
 
-	if (!btrfs_test_opt(root, COMPRESS)) {
-		return cow_file_range(inode, locked_page, start, end,
-				      page_started, nr_written, 1);
-	}
-
 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 			 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
 	while (start < end) {
@@ -935,7 +931,8 @@ static noinline int csum_exist_in_range(struct btrfs_root *root,
  * If no cow copies or snapshots exist, we write directly to the existing
  * blocks on disk
  */
-static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
+static noinline int run_delalloc_nocow(struct inode *inode,
+				       struct page *locked_page,
 			      u64 start, u64 end, int *page_started, int force,
 			      unsigned long *nr_written)
 {
@@ -1133,6 +1130,7 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
 			      unsigned long *nr_written)
 {
 	int ret;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
 
 	if (btrfs_test_flag(inode, NODATACOW))
 		ret = run_delalloc_nocow(inode, locked_page, start, end,
@@ -1140,10 +1138,12 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
 	else if (btrfs_test_flag(inode, PREALLOC))
 		ret = run_delalloc_nocow(inode, locked_page, start, end,
 					 page_started, 0, nr_written);
+	else if (!btrfs_test_opt(root, COMPRESS))
+		ret = cow_file_range(inode, locked_page, start, end,
+				      page_started, nr_written, 1);
 	else
 		ret = cow_file_range_async(inode, locked_page, start, end,
 					   page_started, nr_written);
-
 	return ret;
 }
 
@@ -1453,6 +1453,7 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 
+	path->leave_spinning = 1;
 	ret = btrfs_drop_extents(trans, root, inode, file_pos,
 				 file_pos + num_bytes, file_pos, &hint);
 	BUG_ON(ret);
@@ -1475,6 +1476,10 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	btrfs_set_file_extent_compression(leaf, fi, compression);
 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
+
+	btrfs_unlock_up_safe(path, 1);
+	btrfs_set_lock_blocking(leaf);
+
 	btrfs_mark_buffer_dirty(leaf);
 
 	inode_add_bytes(inode, num_bytes);
@@ -1487,11 +1492,35 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 					  root->root_key.objectid,
 					  trans->transid, inode->i_ino, &ins);
 	BUG_ON(ret);
-
 	btrfs_free_path(path);
+
 	return 0;
 }
 
+/*
+ * helper function for btrfs_finish_ordered_io, this
+ * just reads in some of the csum leaves to prime them into ram
+ * before we start the transaction.  It limits the amount of btree
+ * reads required while inside the transaction.
+ */
+static noinline void reada_csum(struct btrfs_root *root,
+				struct btrfs_path *path,
+				struct btrfs_ordered_extent *ordered_extent)
+{
+	struct btrfs_ordered_sum *sum;
+	u64 bytenr;
+
+	sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
+			 list);
+	bytenr = sum->sums[0].bytenr;
+
+	/*
+	 * we don't care about the results, the point of this search is
+	 * just to get the btree leaves into ram
+	 */
+	btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
+}
+
 /* as ordered data IO finishes, this gets called so we can finish
  * an ordered extent if the range of bytes in the file it covers are
  * fully written.
@@ -1500,8 +1529,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_trans_handle *trans;
-	struct btrfs_ordered_extent *ordered_extent;
+	struct btrfs_ordered_extent *ordered_extent = NULL;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_path *path;
 	int compressed = 0;
 	int ret;
 
@@ -1509,9 +1539,33 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 	if (!ret)
 		return 0;
 
+	/*
+	 * before we join the transaction, try to do some of our IO.
+	 * This will limit the amount of IO that we have to do with
+	 * the transaction running.  We're unlikely to need to do any
+	 * IO if the file extents are new, the disk_i_size checks
+	 * covers the most common case.
+	 */
+	if (start < BTRFS_I(inode)->disk_i_size) {
+		path = btrfs_alloc_path();
+		if (path) {
+			ret = btrfs_lookup_file_extent(NULL, root, path,
+						       inode->i_ino,
+						       start, 0);
+			ordered_extent = btrfs_lookup_ordered_extent(inode,
+								     start);
+			if (!list_empty(&ordered_extent->list)) {
+				btrfs_release_path(root, path);
+				reada_csum(root, path, ordered_extent);
+			}
+			btrfs_free_path(path);
+		}
+	}
+
 	trans = btrfs_join_transaction(root, 1);
 
-	ordered_extent = btrfs_lookup_ordered_extent(inode, start);
+	if (!ordered_extent)
+		ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 	BUG_ON(!ordered_extent);
 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
 		goto nocow;
@@ -2101,6 +2155,7 @@ noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
 
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
+	path->leave_spinning = 1;
 	ret = btrfs_lookup_inode(trans, root, path,
 				 &BTRFS_I(inode)->location, 1);
 	if (ret) {
@@ -2147,6 +2202,7 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
 		goto err;
 	}
 
+	path->leave_spinning = 1;
 	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 				    name, name_len, -1);
 	if (IS_ERR(di)) {
@@ -2190,8 +2246,6 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
 					 inode, dir->i_ino);
 	BUG_ON(ret != 0 && ret != -ENOENT);
-	if (ret != -ENOENT)
-		BTRFS_I(dir)->log_dirty_trans = trans->transid;
 
 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
 					   dir, index);
@@ -2224,6 +2278,9 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 	trans = btrfs_start_transaction(root, 1);
 
 	btrfs_set_trans_block_group(trans, dir);
+
+	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
+
 	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
 				 dentry->d_name.name, dentry->d_name.len);
 
@@ -2498,6 +2555,7 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 	key.type = (u8)-1;
 
 search_again:
+	path->leave_spinning = 1;
 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 	if (ret < 0)
 		goto error;
@@ -2644,6 +2702,7 @@ delete:
 			break;
 		}
 		if (found_extent) {
+			btrfs_set_path_blocking(path);
 			ret = btrfs_free_extent(trans, root, extent_start,
 						extent_num_bytes,
 						leaf->start, root_owner,
@@ -2848,11 +2907,21 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 	if (err)
 		return err;
 
-	if (S_ISREG(inode->i_mode) &&
+	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
-	    attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
+		if (attr->ia_size > inode->i_size) {
-		err = btrfs_cont_expand(inode, attr->ia_size);
+			err = btrfs_cont_expand(inode, attr->ia_size);
-		if (err)
+			if (err)
-			return err;
+				return err;
+		} else if (inode->i_size > 0 &&
+			   attr->ia_size == 0) {
+
+			/* we're truncating a file that used to have good
+			 * data down to zero.  Make sure it gets into
+			 * the ordered flush list so that any new writes
+			 * get down to disk quickly.
+			 */
+			BTRFS_I(inode)->ordered_data_close = 1;
+		}
 	}
 
 	err = inode_setattr(inode, attr);
@@ -2984,13 +3053,14 @@ static noinline void init_btrfs_i(struct inode *inode)
 	bi->disk_i_size = 0;
 	bi->flags = 0;
 	bi->index_cnt = (u64)-1;
-	bi->log_dirty_trans = 0;
+	bi->last_unlink_trans = 0;
 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
 	extent_io_tree_init(&BTRFS_I(inode)->io_tree,
 			     inode->i_mapping, GFP_NOFS);
 	extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 			     inode->i_mapping, GFP_NOFS);
 	INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
+	INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
 	btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
 	mutex_init(&BTRFS_I(inode)->extent_mutex);
 	mutex_init(&BTRFS_I(inode)->log_mutex);
@@ -3449,6 +3519,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 	sizes[0] = sizeof(struct btrfs_inode_item);
 	sizes[1] = name_len + sizeof(*ref);
 
+	path->leave_spinning = 1;
 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
 	if (ret != 0)
 		goto fail;
@@ -3727,6 +3798,8 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
 		drop_inode = 1;
 
 	nr = trans->blocks_used;
+
+	btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
 	btrfs_end_transaction_throttle(trans, root);
 fail:
 	if (drop_inode) {
@@ -4363,6 +4436,8 @@ again:
 	}
 	ClearPageChecked(page);
 	set_page_dirty(page);
+
+	BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
 	unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 
 out_unlock:
@@ -4388,6 +4463,27 @@ static void btrfs_truncate(struct inode *inode)
 	btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
 
 	trans = btrfs_start_transaction(root, 1);
+
+	/*
+	 * setattr is responsible for setting the ordered_data_close flag,
+	 * but that is only tested during the last file release.  That
+	 * could happen well after the next commit, leaving a great big
+	 * window where new writes may get lost if someone chooses to write
+	 * to this file after truncating to zero
+	 *
+	 * The inode doesn't have any dirty data here, and so if we commit
+	 * this is a noop.  If someone immediately starts writing to the inode
+	 * it is very likely we'll catch some of their writes in this
+	 * transaction, and the commit will find this file on the ordered
+	 * data list with good things to send down.
+	 *
+	 * This is a best effort solution, there is still a window where
+	 * using truncate to replace the contents of the file will
+	 * end up with a zero length file after a crash.
+	 */
+	if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
+		btrfs_add_ordered_operation(trans, root, inode);
+
 	btrfs_set_trans_block_group(trans, inode);
 	btrfs_i_size_write(inode, inode->i_size);
 
@@ -4464,12 +4560,15 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
 	ei->i_acl = BTRFS_ACL_NOT_CACHED;
 	ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
 	INIT_LIST_HEAD(&ei->i_orphan);
+	INIT_LIST_HEAD(&ei->ordered_operations);
 	return &ei->vfs_inode;
 }
 
 void btrfs_destroy_inode(struct inode *inode)
 {
 	struct btrfs_ordered_extent *ordered;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+
 	WARN_ON(!list_empty(&inode->i_dentry));
 	WARN_ON(inode->i_data.nrpages);
 
@@ -4480,13 +4579,24 @@ void btrfs_destroy_inode(struct inode *inode)
 	    BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
 		posix_acl_release(BTRFS_I(inode)->i_default_acl);
 
-	spin_lock(&BTRFS_I(inode)->root->list_lock);
+	/*
+	 * Make sure we're properly removed from the ordered operation
+	 * lists.
+	 */
+	smp_mb();
+	if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
+		spin_lock(&root->fs_info->ordered_extent_lock);
+		list_del_init(&BTRFS_I(inode)->ordered_operations);
+		spin_unlock(&root->fs_info->ordered_extent_lock);
+	}
+
+	spin_lock(&root->list_lock);
 	if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 		printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
 		       " list\n", inode->i_ino);
 		dump_stack();
 	}
-	spin_unlock(&BTRFS_I(inode)->root->list_lock);
+	spin_unlock(&root->list_lock);
 
 	while (1) {
 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
@@ -4611,8 +4721,36 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	if (ret)
 		goto out_unlock;
 
+	/*
+	 * we're using rename to replace one file with another.
+	 * and the replacement file is large.  Start IO on it now so
+	 * we don't add too much work to the end of the transaction
+	 */
+	if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
+	    new_inode->i_size &&
+	    old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
+		filemap_flush(old_inode->i_mapping);
+
 	trans = btrfs_start_transaction(root, 1);
 
+	/*
+	 * make sure the inode gets flushed if it is replacing
+	 * something.
+	 */
+	if (new_inode && new_inode->i_size &&
+	    old_inode && S_ISREG(old_inode->i_mode)) {
+		btrfs_add_ordered_operation(trans, root, old_inode);
+	}
+
+	/*
+	 * this is an ugly little race, but the rename is required to make
+	 * sure that if we crash, the inode is either at the old name
+	 * or the new one.  pinning the log transaction lets us make sure
+	 * we don't allow a log commit to come in after we unlink the
+	 * name but before we add the new name back in.
+	 */
+	btrfs_pin_log_trans(root);
+
 	btrfs_set_trans_block_group(trans, new_dir);
 
 	btrfs_inc_nlink(old_dentry->d_inode);
@@ -4620,6 +4758,9 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	new_dir->i_ctime = new_dir->i_mtime = ctime;
 	old_inode->i_ctime = ctime;
 
+	if (old_dentry->d_parent != new_dentry->d_parent)
+		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
+
 	ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
 				 old_dentry->d_name.name,
 				 old_dentry->d_name.len);
@@ -4651,7 +4792,14 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	if (ret)
 		goto out_fail;
 
+	btrfs_log_new_name(trans, old_inode, old_dir,
+				       new_dentry->d_parent);
 out_fail:
+
+	/* this btrfs_end_log_trans just allows the current
+	 * log-sub transaction to complete
+	 */
+	btrfs_end_log_trans(root);
 	btrfs_end_transaction_throttle(trans, root);
 out_unlock:
 	return ret;

fs/btrfs/locking.c

@@ -71,12 +71,13 @@ void btrfs_clear_lock_blocking(struct extent_buffer *eb)
 static int btrfs_spin_on_block(struct extent_buffer *eb)
 {
 	int i;
+
 	for (i = 0; i < 512; i++) {
-		cpu_relax();
 		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
 			return 1;
 		if (need_resched())
 			break;
+		cpu_relax();
 	}
 	return 0;
 }
@@ -95,13 +96,15 @@ int btrfs_try_spin_lock(struct extent_buffer *eb)
 {
 	int i;
 
-	spin_nested(eb);
+	if (btrfs_spin_on_block(eb)) {
-	if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
+		spin_nested(eb);
-		return 1;
+		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
-	spin_unlock(&eb->lock);
+			return 1;
-
+		spin_unlock(&eb->lock);
+	}
 	/* spin for a bit on the BLOCKING flag */
 	for (i = 0; i < 2; i++) {
+		cpu_relax();
 		if (!btrfs_spin_on_block(eb))
 			break;
 
@@ -148,6 +151,9 @@ int btrfs_tree_lock(struct extent_buffer *eb)
 	DEFINE_WAIT(wait);
 	wait.func = btrfs_wake_function;
 
+	if (!btrfs_spin_on_block(eb))
+		goto sleep;
+
 	while(1) {
 		spin_nested(eb);
 
@@ -165,9 +171,10 @@ int btrfs_tree_lock(struct extent_buffer *eb)
 		 * spin for a bit, and if the blocking flag goes away,
 		 * loop around
 		 */
+		cpu_relax();
 		if (btrfs_spin_on_block(eb))
 			continue;
-
+sleep:
 		prepare_to_wait_exclusive(&eb->lock_wq, &wait,
 					  TASK_UNINTERRUPTIBLE);
 

fs/btrfs/ordered-data.c

@@ -310,6 +310,16 @@ int btrfs_remove_ordered_extent(struct inode *inode,
 
 	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 	list_del_init(&entry->root_extent_list);
+
+	/*
+	 * we have no more ordered extents for this inode and
+	 * no dirty pages.  We can safely remove it from the
+	 * list of ordered extents
+	 */
+	if (RB_EMPTY_ROOT(&tree->tree) &&
+	    !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
+		list_del_init(&BTRFS_I(inode)->ordered_operations);
+	}
 	spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 
 	mutex_unlock(&tree->mutex);
@@ -369,6 +379,68 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
 	return 0;
 }
 
+/*
+ * this is used during transaction commit to write all the inodes
+ * added to the ordered operation list.  These files must be fully on
+ * disk before the transaction commits.
+ *
+ * we have two modes here, one is to just start the IO via filemap_flush
+ * and the other is to wait for all the io.  When we wait, we have an
+ * extra check to make sure the ordered operation list really is empty
+ * before we return
+ */
+int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
+{
+	struct btrfs_inode *btrfs_inode;
+	struct inode *inode;
+	struct list_head splice;
+
+	INIT_LIST_HEAD(&splice);
+
+	mutex_lock(&root->fs_info->ordered_operations_mutex);
+	spin_lock(&root->fs_info->ordered_extent_lock);
+again:
+	list_splice_init(&root->fs_info->ordered_operations, &splice);
+
+	while (!list_empty(&splice)) {
+		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
+				   ordered_operations);
+
+		inode = &btrfs_inode->vfs_inode;
+
+		list_del_init(&btrfs_inode->ordered_operations);
+
+		/*
+		 * the inode may be getting freed (in sys_unlink path).
+		 */
+		inode = igrab(inode);
+
+		if (!wait && inode) {
+			list_add_tail(&BTRFS_I(inode)->ordered_operations,
+			      &root->fs_info->ordered_operations);
+		}
+		spin_unlock(&root->fs_info->ordered_extent_lock);
+
+		if (inode) {
+			if (wait)
+				btrfs_wait_ordered_range(inode, 0, (u64)-1);
+			else
+				filemap_flush(inode->i_mapping);
+			iput(inode);
+		}
+
+		cond_resched();
+		spin_lock(&root->fs_info->ordered_extent_lock);
+	}
+	if (wait && !list_empty(&root->fs_info->ordered_operations))
+		goto again;
+
+	spin_unlock(&root->fs_info->ordered_extent_lock);
+	mutex_unlock(&root->fs_info->ordered_operations_mutex);
+
+	return 0;
+}
+
 /*
  * Used to start IO or wait for a given ordered extent to finish.
  *
@@ -726,3 +798,49 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
 
 	return ret;
 }
+
+/*
+ * add a given inode to the list of inodes that must be fully on
+ * disk before a transaction commit finishes.
+ *
+ * This basically gives us the ext3 style data=ordered mode, and it is mostly
+ * used to make sure renamed files are fully on disk.
+ *
+ * It is a noop if the inode is already fully on disk.
+ *
+ * If trans is not null, we'll do a friendly check for a transaction that
+ * is already flushing things and force the IO down ourselves.
+ */
+int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct inode *inode)
+{
+	u64 last_mod;
+
+	last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
+
+	/*
+	 * if this file hasn't been changed since the last transaction
+	 * commit, we can safely return without doing anything
+	 */
+	if (last_mod < root->fs_info->last_trans_committed)
+		return 0;
+
+	/*
+	 * the transaction is already committing.  Just start the IO and
+	 * don't bother with all of this list nonsense
+	 */
+	if (trans && root->fs_info->running_transaction->blocked) {
+		btrfs_wait_ordered_range(inode, 0, (u64)-1);
+		return 0;
+	}
+
+	spin_lock(&root->fs_info->ordered_extent_lock);
+	if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
+		list_add_tail(&BTRFS_I(inode)->ordered_operations,
+			      &root->fs_info->ordered_operations);
+	}
+	spin_unlock(&root->fs_info->ordered_extent_lock);
+
+	return 0;
+}

fs/btrfs/ordered-data.h

@@ -155,4 +155,8 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
 			   loff_t end, int sync_mode);
 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only);
+int btrfs_run_ordered_operations(struct btrfs_root *root, int wait);
+int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				struct inode *inode);
 #endif

fs/btrfs/transaction.c

@@ -65,6 +65,15 @@ static noinline int join_transaction(struct btrfs_root *root)
 		cur_trans->use_count = 1;
 		cur_trans->commit_done = 0;
 		cur_trans->start_time = get_seconds();
+
+		cur_trans->delayed_refs.root.rb_node = NULL;
+		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);
 		extent_io_tree_init(&cur_trans->dirty_pages,
@@ -182,6 +191,8 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 	h->block_group = 0;
 	h->alloc_exclude_nr = 0;
 	h->alloc_exclude_start = 0;
+	h->delayed_ref_updates = 0;
+
 	root->fs_info->running_transaction->use_count++;
 	mutex_unlock(&root->fs_info->trans_mutex);
 	return h;
@@ -271,7 +282,6 @@ void btrfs_throttle(struct btrfs_root *root)
 	if (!root->fs_info->open_ioctl_trans)
 		wait_current_trans(root);
 	mutex_unlock(&root->fs_info->trans_mutex);
-
 	throttle_on_drops(root);
 }
 
@@ -280,6 +290,27 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 {
 	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;
@@ -424,9 +455,10 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
 	u64 old_root_bytenr;
 	struct btrfs_root *tree_root = root->fs_info->tree_root;
 
-	btrfs_extent_post_op(trans, root);
 	btrfs_write_dirty_block_groups(trans, root);
-	btrfs_extent_post_op(trans, root);
+
+	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+	BUG_ON(ret);
 
 	while (1) {
 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
@@ -438,14 +470,14 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
 				     btrfs_header_level(root->node));
 		btrfs_set_root_generation(&root->root_item, trans->transid);
 
-		btrfs_extent_post_op(trans, root);
-
 		ret = btrfs_update_root(trans, tree_root,
 					&root->root_key,
 					&root->root_item);
 		BUG_ON(ret);
 		btrfs_write_dirty_block_groups(trans, root);
-		btrfs_extent_post_op(trans, root);
+
+		ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+		BUG_ON(ret);
 	}
 	return 0;
 }
@@ -459,15 +491,18 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct list_head *next;
 	struct extent_buffer *eb;
+	int ret;
 
-	btrfs_extent_post_op(trans, fs_info->tree_root);
+	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, 0);
+	btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
 	btrfs_tree_unlock(eb);
 	free_extent_buffer(eb);
 
-	btrfs_extent_post_op(trans, fs_info->tree_root);
+	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+	BUG_ON(ret);
 
 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
 		next = fs_info->dirty_cowonly_roots.next;
@@ -475,6 +510,9 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 		root = list_entry(next, struct btrfs_root, dirty_list);
 
 		update_cowonly_root(trans, root);
+
+		ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+		BUG_ON(ret);
 	}
 	return 0;
 }
@@ -634,6 +672,31 @@ int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 	return 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)
+{
+	DEFINE_WAIT(wait);
+
+	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);
+	}
+	mutex_unlock(&info->trans_mutex);
+	return 0;
+}
+
 /*
  * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
  * all of them
@@ -661,7 +724,22 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 		atomic_inc(&root->fs_info->throttles);
 
 		while (1) {
+			/*
+			 * we don't want to jump in and create a bunch of
+			 * delayed refs if the transaction is starting to close
+			 */
+			wait_transaction_pre_flush(tree_root->fs_info);
 			trans = btrfs_start_transaction(tree_root, 1);
+
+			/*
+			 * we've joined a transaction, make sure it isn't
+			 * closing right now
+			 */
+			if (trans->transaction->delayed_refs.flushing) {
+				btrfs_end_transaction(trans, tree_root);
+				continue;
+			}
+
 			mutex_lock(&root->fs_info->drop_mutex);
 			ret = btrfs_drop_snapshot(trans, dirty->root);
 			if (ret != -EAGAIN)
@@ -766,7 +844,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 
 	old = btrfs_lock_root_node(root);
-	btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
+	btrfs_cow_block(trans, root, old, NULL, 0, &old);
 
 	btrfs_copy_root(trans, root, old, &tmp, objectid);
 	btrfs_tree_unlock(old);
@@ -894,12 +972,31 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	struct extent_io_tree *pinned_copy;
 	DEFINE_WAIT(wait);
 	int ret;
+	int should_grow = 0;
+	unsigned long now = get_seconds();
+
+	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;
+
+	ret = btrfs_run_delayed_refs(trans, root, 0);
+	BUG_ON(ret);
 
-	INIT_LIST_HEAD(&dirty_fs_roots);
 	mutex_lock(&root->fs_info->trans_mutex);
-	if (trans->transaction->in_commit) {
+	INIT_LIST_HEAD(&dirty_fs_roots);
-		cur_trans = trans->transaction;
+	if (cur_trans->in_commit) {
-		trans->transaction->use_count++;
+		cur_trans->use_count++;
 		mutex_unlock(&root->fs_info->trans_mutex);
 		btrfs_end_transaction(trans, root);
 
@@ -922,7 +1019,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
 	trans->transaction->in_commit = 1;
 	trans->transaction->blocked = 1;
-	cur_trans = trans->transaction;
 	if (cur_trans->list.prev != &root->fs_info->trans_list) {
 		prev_trans = list_entry(cur_trans->list.prev,
 					struct btrfs_transaction, list);
@@ -937,6 +1033,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *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;
@@ -949,7 +1048,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
 		if (cur_trans->num_writers > 1)
 			timeout = MAX_SCHEDULE_TIMEOUT;
-		else
+		else if (should_grow)
 			timeout = 1;
 
 		mutex_unlock(&root->fs_info->trans_mutex);
@@ -959,16 +1058,30 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 			BUG_ON(ret);
 		}
 
-		schedule_timeout(timeout);
+		/*
+		 * 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->trans_mutex);
 		finish_wait(&cur_trans->writer_wait, &wait);
 	} 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);
 
 	/* btrfs_commit_tree_roots is responsible for getting the
@@ -1032,6 +1145,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	btrfs_copy_pinned(root, pinned_copy);
 
 	trans->transaction->blocked = 0;
+
 	wake_up(&root->fs_info->transaction_throttle);
 	wake_up(&root->fs_info->transaction_wait);
 
@@ -1058,6 +1172,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	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);