Btrfs: Add delayed allocation to the extent based page tree code

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

6 files changed, 216 insertions, 135 deletions

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 4049aadbeda9..c31e84d42653 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -39,7 +39,6 @@ static int cache_block_group(struct btrfs_root *root,
 	u64 i;
 	u64 last = 0;
 	u64 hole_size;
-	u64 limit;
 	int found = 0;
 
 	root = root->fs_info->extent_root;
@@ -62,7 +61,6 @@ static int cache_block_group(struct btrfs_root *root,
 		return ret;
 	if (ret && path->slots[0] > 0)
 		path->slots[0]--;
-	limit = block_group->key.objectid + block_group->key.offset;
 	while(1) {
 		leaf = btrfs_buffer_leaf(path->nodes[0]);
 		slot = path->slots[0];
@@ -982,7 +980,6 @@ static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 	struct btrfs_block_group_cache *block_group;
 	int full_scan = 0;
 	int wrapped = 0;
-	u64 limit;
 
 	WARN_ON(num_blocks < 1);
 	ins->flags = 0;
@@ -1049,12 +1046,6 @@ check_failed:
 		l = btrfs_buffer_leaf(path->nodes[0]);
 		slot = path->slots[0];
 		if (slot >= btrfs_header_nritems(&l->header)) {
-			if (start_found)
-				limit = last_block +
-					(block_group->key.offset >> 1);
-			else
-				limit = search_start +
-					(block_group->key.offset >> 1);
 			ret = btrfs_next_leaf(root, path);
 			if (ret == 0)
 				continue;
diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index d378edf0964e..a9c7419615b9 100644
--- a/fs/btrfs/extent_map.c
+++ b/fs/btrfs/extent_map.c
@@ -30,9 +30,6 @@ struct tree_entry {
 
 #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
 
-static LIST_HEAD(all_states);
-spinlock_t state_lock = SPIN_LOCK_UNLOCKED;
-
 void __init extent_map_init(void)
 {
 	extent_map_cache = kmem_cache_create("extent_map",
@@ -49,15 +46,6 @@ void __init extent_map_init(void)
 
 void __exit extent_map_exit(void)
 {
-	while(!list_empty(&all_states)) {
-		struct extent_state *state;
-		struct list_head *cur = all_states.next;
-		state = list_entry(cur, struct extent_state, list);
-		printk("found leaked state %Lu %Lu state %d in_tree %d\n",
-		       state->start, state->end, state->state, state->in_tree);
-		list_del(&state->list);
-		kfree(state);
-	}
 	if (extent_map_cache)
 		kmem_cache_destroy(extent_map_cache);
 	if (extent_state_cache)
@@ -69,6 +57,7 @@ void extent_map_tree_init(struct extent_map_tree *tree,
 {
 	tree->map.rb_node = NULL;
 	tree->state.rb_node = NULL;
+	tree->fill_delalloc = NULL;
 	rwlock_init(&tree->lock);
 	tree->mapping = mapping;
 }
@@ -106,9 +95,6 @@ struct extent_state *alloc_extent_state(gfp_t mask)
 	state->in_tree = 0;
 	atomic_set(&state->refs, 1);
 	init_waitqueue_head(&state->wq);
-	spin_lock_irq(&state_lock);
-	list_add(&state->list, &all_states);
-	spin_unlock_irq(&state_lock);
 	return state;
 }
 EXPORT_SYMBOL(alloc_extent_state);
@@ -117,9 +103,6 @@ void free_extent_state(struct extent_state *state)
 {
 	if (atomic_dec_and_test(&state->refs)) {
 		WARN_ON(state->in_tree);
-		spin_lock_irq(&state_lock);
-		list_del_init(&state->list);
-		spin_unlock_irq(&state_lock);
 		kmem_cache_free(extent_state_cache, state);
 	}
 }
@@ -369,7 +352,7 @@ static int insert_state(struct extent_map_tree *tree,
 	if (node) {
 		struct extent_state *found;
 		found = rb_entry(node, struct extent_state, rb_node);
-printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
+		printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
 		free_extent_state(state);
 		return -EEXIST;
 	}
@@ -408,7 +391,7 @@ static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
 	if (node) {
 		struct extent_state *found;
 		found = rb_entry(node, struct extent_state, rb_node);
-printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
+		printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
 		free_extent_state(prealloc);
 		return -EEXIST;
 	}
@@ -792,10 +775,20 @@ int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 }
 EXPORT_SYMBOL(set_extent_dirty);
 
+int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end,
+			      EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
+			      mask);
+}
+EXPORT_SYMBOL(set_extent_delalloc);
+
 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 		       gfp_t mask)
 {
-	return clear_extent_bit(tree, start, end, EXTENT_DIRTY, 0, 0, mask);
+	return clear_extent_bit(tree, start, end,
+				EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
 }
 EXPORT_SYMBOL(clear_extent_dirty);
 
@@ -922,6 +915,62 @@ int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
 }
 EXPORT_SYMBOL(set_range_writeback);
 
+u64 find_lock_delalloc_range(struct extent_map_tree *tree,
+			     u64 start, u64 lock_start, u64 *end, u64 max_bytes)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	u64 cur_start = start;
+	u64 found = 0;
+	u64 total_bytes = 0;
+
+	write_lock_irq(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+search_again:
+	node = tree_search(&tree->state, cur_start);
+	if (!node || IS_ERR(node)) {
+		goto out;
+	}
+
+	while(1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->start != cur_start) {
+			goto out;
+		}
+		if (!(state->state & EXTENT_DELALLOC)) {
+			goto out;
+		}
+		if (state->start >= lock_start) {
+			if (state->state & EXTENT_LOCKED) {
+				DEFINE_WAIT(wait);
+				atomic_inc(&state->refs);
+				write_unlock_irq(&tree->lock);
+				schedule();
+				write_lock_irq(&tree->lock);
+				finish_wait(&state->wq, &wait);
+				free_extent_state(state);
+				goto search_again;
+			}
+			state->state |= EXTENT_LOCKED;
+		}
+		found++;
+		*end = state->end;
+		cur_start = state->end + 1;
+		node = rb_next(node);
+		if (!node)
+			break;
+		total_bytes = state->end - state->start + 1;
+		if (total_bytes >= max_bytes)
+			break;
+	}
+out:
+	write_unlock_irq(&tree->lock);
+	return found;
+}
+
 /*
  * helper function to lock both pages and extents in the tree.
  * pages must be locked first.
@@ -1285,6 +1334,7 @@ int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
 	if (!PagePrivate(page)) {
 		SetPagePrivate(page);
 		set_page_private(page, 1);
+		WARN_ON(!page->mapping->a_ops->invalidatepage);
 		page_cache_get(page);
 	}
 
@@ -1384,7 +1434,10 @@ int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
 	size_t blocksize;
 	loff_t i_size = i_size_read(inode);
 	unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
+	u64 nr_delalloc;
+	u64 delalloc_end;
 
+	WARN_ON(!PageLocked(page));
 	if (page->index > end_index) {
 		clear_extent_dirty(tree, start, page_end, GFP_NOFS);
 		unlock_page(page);
@@ -1400,11 +1453,34 @@ int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
 	if (!PagePrivate(page)) {
 		SetPagePrivate(page);
 		set_page_private(page, 1);
+		WARN_ON(!page->mapping->a_ops->invalidatepage);
 		page_cache_get(page);
 	}
 
-	end = page_end;
 	lock_extent(tree, start, page_end, GFP_NOFS);
+	nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1,
+					       &delalloc_end,
+					       128 * 1024 * 1024);
+	if (nr_delalloc) {
+		tree->fill_delalloc(inode, start, delalloc_end);
+		if (delalloc_end >= page_end + 1) {
+			clear_extent_bit(tree, page_end + 1, delalloc_end,
+					 EXTENT_LOCKED | EXTENT_DELALLOC,
+					 1, 0, GFP_NOFS);
+		}
+		clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC,
+				 0, 0, GFP_NOFS);
+		if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
+			printk("found delalloc bits after clear extent_bit\n");
+		}
+	} else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
+		printk("found delalloc bits after find_delalloc_range returns 0\n");
+	}
+
+	end = page_end;
+	if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
+		printk("found delalloc bits after lock_extent\n");
+	}
 
 	if (last_byte <= start) {
 		clear_extent_dirty(tree, start, page_end, GFP_NOFS);
@@ -1419,7 +1495,7 @@ int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
 			clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
 			break;
 		}
-		em = get_extent(inode, page, page_offset, cur, end, 1);
+		em = get_extent(inode, page, page_offset, cur, end, 0);
 		if (IS_ERR(em) || !em) {
 			SetPageError(page);
 			break;
@@ -1507,6 +1583,7 @@ int extent_commit_write(struct extent_map_tree *tree,
 	if (!PagePrivate(page)) {
 		SetPagePrivate(page);
 		set_page_private(page, 1);
+		WARN_ON(!page->mapping->a_ops->invalidatepage);
 		page_cache_get(page);
 	}
 
@@ -1543,6 +1620,7 @@ int extent_prepare_write(struct extent_map_tree *tree,
 	if (!PagePrivate(page)) {
 		SetPagePrivate(page);
 		set_page_private(page, 1);
+		WARN_ON(!page->mapping->a_ops->invalidatepage);
 		page_cache_get(page);
 	}
 	block_start = (page_start + from) & ~((u64)blocksize - 1);
@@ -1628,29 +1706,28 @@ int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
 	u64 start = page->index << PAGE_CACHE_SHIFT;
 	u64 end = start + PAGE_CACHE_SIZE - 1;
 	u64 orig_start = start;
+	int ret = 1;
 
 	while (start <= end) {
 		em = lookup_extent_mapping(tree, start, end);
 		if (!em || IS_ERR(em))
 			break;
-		if (test_range_bit(tree, em->start, em->end,
-				   EXTENT_LOCKED, 0)) {
+		if (!test_range_bit(tree, em->start, em->end,
+				    EXTENT_LOCKED, 0)) {
+			remove_extent_mapping(tree, em);
+			/* once for the rb tree */
 			free_extent_map(em);
-			start = em->end + 1;
-printk("range still locked %Lu %Lu\n", em->start, em->end);
-			break;
 		}
-		remove_extent_mapping(tree, em);
 		start = em->end + 1;
-		/* once for the rb tree */
-		free_extent_map(em);
 		/* once for us */
 		free_extent_map(em);
 	}
-	WARN_ON(test_range_bit(tree, orig_start, end, EXTENT_WRITEBACK, 0));
-	clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
-			 1, 1, GFP_NOFS);
-	return 1;
+	if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0))
+		ret = 0;
+	else
+		clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
+				 1, 1, GFP_NOFS);
+	return ret;
 }
 EXPORT_SYMBOL(try_release_extent_mapping);
 
diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h
index 108944aab4bd..e91a2e9619ee 100644
--- a/fs/btrfs/extent_map.h
+++ b/fs/btrfs/extent_map.h
@@ -11,6 +11,7 @@ struct extent_map_tree {
 	struct rb_root state;
 	struct address_space *mapping;
 	rwlock_t lock;
+	int (*fill_delalloc)(struct inode *inode, u64 start, u64 end);
 };
 
 /* note, this must start with the same fields as fs/extent_map.c:tree_entry */
@@ -74,6 +75,8 @@ int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
 		   gfp_t mask);
 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 		     gfp_t mask);
+int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end,
+		     gfp_t mask);
 int extent_invalidatepage(struct extent_map_tree *tree,
 			  struct page *page, unsigned long offset);
 int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 71a481894ab6..d3d39e4a2797 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -143,7 +143,6 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct extent_map *em;
 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
-	struct btrfs_key ins;
 	u64 hint_block;
 	u64 num_blocks;
 	u64 start_pos;
@@ -162,6 +161,7 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 			inode->i_blkbits;
 
 	end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
+	lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
 	if (!trans) {
@@ -179,16 +179,6 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 
 	/* FIXME...EIEIO, ENOSPC and more */
 
-	/* step one, delete the existing extents in this range */
-	/* FIXME blocksize != pagesize */
-	if (start_pos < inode->i_size) {
-		err = btrfs_drop_extents(trans, root, inode,
-			 start_pos, (pos + write_bytes + root->blocksize -1) &
-			 ~((u64)root->blocksize - 1), &hint_block);
-		if (err)
-			goto failed;
-	}
-
 	/* insert any holes we need to create */
 	if (inode->i_size < start_pos) {
 		u64 last_pos_in_file;
@@ -213,29 +203,28 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	 */
 	if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
 	    pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
-		err = btrfs_alloc_extent(trans, root, inode->i_ino,
-					 num_blocks, 0, hint_block, (u64)-1,
-					 &ins, 1);
-		BUG_ON(err);
-		err = btrfs_insert_file_extent(trans, root, inode->i_ino,
-				       start_pos, ins.objectid, ins.offset,
-				       ins.offset);
-		BUG_ON(err);
-		em->start = start_pos;
-		em->end = end_of_last_block;
-		em->block_start = ins.objectid << inode->i_blkbits;
-		em->block_end = em->block_start +
-			(ins.offset << inode->i_blkbits) - 1;
-		set_extent_dirty(em_tree, start_pos, end_of_last_block,
-				 GFP_NOFS);
-		err = add_extent_mapping(em_tree, em);
+		u64 last_end;
 		for (i = 0; i < num_pages; i++) {
 			struct page *p = pages[i];
 			SetPageUptodate(p);
-			__set_page_dirty_nobuffers(p);
+			set_page_dirty(p);
 		}
+		last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
+		last_end += PAGE_CACHE_SIZE - 1;
+		set_extent_delalloc(em_tree, start_pos, end_of_last_block,
+				 GFP_NOFS);
 	} else {
 		struct page *p = pages[0];
+		/* step one, delete the existing extents in this range */
+		/* FIXME blocksize != pagesize */
+		if (start_pos < inode->i_size) {
+			err = btrfs_drop_extents(trans, root, inode, start_pos,
+				 (pos + write_bytes + root->blocksize -1) &
+				 ~((u64)root->blocksize - 1), &hint_block);
+			if (err)
+				goto failed;
+		}
+
 		err = insert_inline_extent(trans, root, inode, start_pos,
 					   end_pos - start_pos, p, 0);
 		BUG_ON(err);
@@ -253,6 +242,7 @@ failed:
 	err = btrfs_end_transaction(trans, root);
 out_unlock:
 	mutex_unlock(&root->fs_info->fs_mutex);
+	unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 	free_extent_map(em);
 	return err;
 }
@@ -531,7 +521,7 @@ static int prepare_pages(struct btrfs_root *root,
 	u64 num_blocks;
 	u64 start_pos;
 
-	start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
+	start_pos = pos & ~((u64)root->blocksize - 1);
 	num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
 			inode->i_blkbits;
 
@@ -548,8 +538,10 @@ static int prepare_pages(struct btrfs_root *root,
 		if (!PagePrivate(pages[i])) {
 			SetPagePrivate(pages[i]);
 			set_page_private(pages[i], 1);
+			WARN_ON(!pages[i]->mapping->a_ops->invalidatepage);
 			page_cache_get(pages[i]);
 		}
+		WARN_ON(!PageLocked(pages[i]));
 	}
 	return 0;
 }
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index fd4cb65fcf1f..dc181089aa74 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -70,6 +70,40 @@ static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
 };
 
+static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key ins;
+	u64 alloc_hint = 0;
+	u64 num_blocks;
+	int ret;
+	u64 blocksize = 1 << inode->i_blkbits;
+
+	mutex_lock(&root->fs_info->fs_mutex);
+	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, inode);
+	BUG_ON(!trans);
+	num_blocks = (end - start + blocksize) & ~(blocksize - 1);
+	ret = btrfs_drop_extents(trans, root, inode,
+				 start, start + num_blocks, &alloc_hint);
+	num_blocks = num_blocks >> inode->i_blkbits;
+	ret = btrfs_alloc_extent(trans, root, inode->i_ino, num_blocks, 0,
+				 alloc_hint, (u64)-1, &ins, 1);
+	if (ret) {
+		WARN_ON(1);
+		goto out;
+	}
+	ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
+				       start, ins.objectid, ins.offset,
+				       ins.offset);
+out:
+	btrfs_end_transaction(trans, root);
+	mutex_unlock(&root->fs_info->fs_mutex);
+	return ret;
+}
+
+
 void btrfs_read_locked_inode(struct inode *inode)
 {
 	struct btrfs_path *path;
@@ -121,8 +155,7 @@ void btrfs_read_locked_inode(struct inode *inode)
 	switch (inode->i_mode & S_IFMT) {
 	case S_IFREG:
 		inode->i_mapping->a_ops = &btrfs_aops;
-		extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
-				     inode->i_mapping, GFP_NOFS);
+		BTRFS_I(inode)->extent_tree.fill_delalloc = run_delalloc_range;
 		inode->i_fop = &btrfs_file_operations;
 		inode->i_op = &btrfs_file_inode_operations;
 		break;
@@ -450,11 +483,17 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 			BUG_ON(ret);
 		}
 		if (item_end < inode->i_size) {
-			if (found_type) {
-				btrfs_set_key_type(&key, found_type - 1);
-				continue;
+			if (found_type == BTRFS_DIR_ITEM_KEY) {
+				found_type = BTRFS_INODE_ITEM_KEY;
+			} else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
+				found_type = BTRFS_CSUM_ITEM_KEY;
+			} else if (found_type) {
+				found_type--;
+			} else {
+				break;
 			}
-			break;
+			btrfs_set_key_type(&key, found_type - 1);
+			continue;
 		}
 		if (btrfs_disk_key_offset(found_key) >= inode->i_size)
 			del_item = 1;
@@ -514,47 +553,34 @@ error:
 	return ret;
 }
 
-static int btrfs_cow_one_page(struct btrfs_trans_handle *trans,
-			      struct inode *inode, struct page *page,
+static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 			      size_t zero_start)
 {
 	char *kaddr;
 	int ret = 0;
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	u64 alloc_hint = 0;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 	u64 page_start = page->index << PAGE_CACHE_SHIFT;
-	struct btrfs_key ins;
+	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 
 	if (!PagePrivate(page)) {
 		SetPagePrivate(page);
 		set_page_private(page, 1);
+		WARN_ON(!page->mapping->a_ops->invalidatepage);
 		page_cache_get(page);
 	}
 
-	btrfs_set_trans_block_group(trans, inode);
-
-	ret = btrfs_drop_extents(trans, root, inode,
-				 page_start, page_start + PAGE_CACHE_SIZE,
-				 &alloc_hint);
-	if (ret)
-		goto out;
-	ret = btrfs_alloc_extent(trans, root, inode->i_ino, 1, 0,
-				 alloc_hint, (u64)-1, &ins, 1);
-	if (ret)
-		goto out;
-	ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
-				       page_start, ins.objectid, 1, 1);
-	if (ret)
-		goto out;
-	SetPageChecked(page);
-	kaddr = kmap(page);
+	lock_extent(em_tree, page_start, page_end, GFP_NOFS);
+	set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
+			    page_end, GFP_NOFS);
 	if (zero_start != PAGE_CACHE_SIZE) {
+		kaddr = kmap(page);
 		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 		flush_dcache_page(page);
+		kunmap(page);
 	}
-	kunmap(page);
+	set_page_dirty(page);
+	unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 
-out:
 	return ret;
 }
 
@@ -565,8 +591,6 @@ out:
 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 {
 	struct inode *inode = mapping->host;
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct btrfs_trans_handle *trans;
 	unsigned blocksize = 1 << inode->i_blkbits;
 	pgoff_t index = from >> PAGE_CACHE_SHIFT;
 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
@@ -591,21 +615,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 	}
 	page_start = page->index << PAGE_CACHE_SHIFT;
 
-	mutex_lock(&root->fs_info->fs_mutex);
-	trans = btrfs_start_transaction(root, 1);
-	ret = btrfs_cow_one_page(trans, inode, page, offset);
-	if (!ret) {
-		char *kaddr = kmap(page);
-		ret = btrfs_csum_file_block(trans, root, inode->i_ino,
-					    page_start, kaddr, PAGE_CACHE_SIZE);
-		kunmap(page);
-	}
-	set_extent_dirty(&BTRFS_I(inode)->extent_tree,
-			 page_start, page_start + PAGE_CACHE_SIZE - 1,
-			 GFP_NOFS);
-	set_page_dirty(page);
-	btrfs_end_transaction(trans, root);
-	mutex_unlock(&root->fs_info->fs_mutex);
+	ret = btrfs_cow_one_page(inode, page, offset);
 
 	unlock_page(page);
 	page_cache_release(page);
@@ -757,6 +767,8 @@ static int btrfs_init_locked_inode(struct inode *inode, void *p)
 	struct btrfs_iget_args *args = p;
 	inode->i_ino = args->ino;
 	BTRFS_I(inode)->root = args->root;
+	extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
+			     inode->i_mapping, GFP_NOFS);
 	return 0;
 }
 
@@ -968,7 +980,10 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 	if (!inode)
 		return ERR_PTR(-ENOMEM);
 
+	extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
+			     inode->i_mapping, GFP_NOFS);
 	BTRFS_I(inode)->root = root;
+
 	if (mode & S_IFDIR)
 		owner = 0;
 	else
@@ -1128,6 +1143,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 		inode->i_op = &btrfs_file_inode_operations;
 		extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
 				     inode->i_mapping, GFP_NOFS);
+		BTRFS_I(inode)->extent_tree.fill_delalloc = run_delalloc_range;
 	}
 	dir->i_sb->s_dirt = 1;
 	btrfs_update_inode_block_group(trans, inode);
@@ -1344,9 +1360,11 @@ again:
 		extent_end = extent_start +
 		       (btrfs_file_extent_num_blocks(item) << inode->i_blkbits);
 		err = 0;
-		if (start < extent_start || start > extent_end) {
+		if (start < extent_start || start >= extent_end) {
 			em->start = start;
 			if (start < extent_start) {
+				if (end < extent_start)
+					goto not_found;
 				em->end = extent_end - 1;
 			} else {
 				em->end = end;
@@ -1375,9 +1393,11 @@ again:
 		size = btrfs_file_extent_inline_len(leaf->items +
 						    path->slots[0]);
 		extent_end = extent_start + size;
-		if (start < extent_start || start > extent_end) {
+		if (start < extent_start || start >= extent_end) {
 			em->start = start;
 			if (start < extent_start) {
+				if (end < extent_start)
+					goto not_found;
 				em->end = extent_end - 1;
 			} else {
 				em->end = end;
@@ -1412,8 +1432,7 @@ not_found_em:
 insert:
 	btrfs_release_path(root, path);
 	if (em->start > start || em->end < start) {
-		printk("bad extent! %Lu %Lu start %Lu end %Lu\n", em->start, em->end, start, end);
-		WARN_ON(1);
+		printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
 		err = -EIO;
 		goto out;
 	}
@@ -1648,6 +1667,13 @@ int btrfs_readpage(struct file *file, struct page *page)
 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
 {
 	struct extent_map_tree *tree;
+
+
+	if (current->flags & PF_MEMALLOC) {
+		redirty_page_for_writepage(wbc, page);
+		unlock_page(page);
+		return 0;
+	}
 	tree = &BTRFS_I(page->mapping->host)->extent_tree;
 	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
 }
@@ -1698,8 +1724,6 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 {
 	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct btrfs_trans_handle *trans;
 	unsigned long end;
 	loff_t size;
 	int ret = -EINVAL;
@@ -1722,15 +1746,7 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 	else
 		end = PAGE_CACHE_SIZE;
 
-	mutex_lock(&root->fs_info->fs_mutex);
-	trans = btrfs_start_transaction(root, 1);
-	ret = btrfs_cow_one_page(trans, inode, page, end);
-	btrfs_end_transaction(trans, root);
-	mutex_unlock(&root->fs_info->fs_mutex);
-	set_extent_dirty(&BTRFS_I(inode)->extent_tree,
-			 page_start, page_start + PAGE_CACHE_SIZE - 1,
-			 GFP_NOFS);
-	set_page_dirty(page);
+	ret = btrfs_cow_one_page(inode, page, end);
 
 out_unlock:
 	unlock_page(page);
@@ -2264,6 +2280,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 		inode->i_op = &btrfs_file_inode_operations;
 		extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
 				     inode->i_mapping, GFP_NOFS);
+		BTRFS_I(inode)->extent_tree.fill_delalloc = run_delalloc_range;
 	}
 	dir->i_sb->s_dirt = 1;
 	btrfs_update_inode_block_group(trans, inode);
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 7b7c32c73a2c..7da95538745f 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -83,6 +83,7 @@ static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 	bi->location.offset = 0;
 	bi->location.flags = 0;
 	bi->root = tree_root;
+
 	btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
 
 	if (!inode) {