Skip to contentSkip to conversation

Btrfs for review

PATCH In review Chris Mason 7 participants 58 messages Jan 8, 2009 – Jan 30, 2009 Page 3 of 3

Subsystems (2)
lore
Share & tools
lore
#49 lore raw

PATCH 1/1 Btrfs extent allocation code

In reply to Chris Mason

All metadata and data extents are allocated here. A single btree holds records for all of the extents allocated on the FS. It also records backreferences from those extents to the objects that have references on them.

Btree blocks for the extent allocation tree are allocated from the extent allocation tree. This leads to some ugly recursion, which is avoided by building up a list of pending allocations/deletions for the extent allocation tree and processing them at safe times.

fs/btrfs/extent-tree.c new file +3336 -0
@@ -0,0 +1,3336 @@
+/*
+ * Copyright (C) 2007 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.
+ */
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/writeback.h>
+#include <linux/blkdev.h>
+#include <linux/version.h>
+#include "compat.h"
+#include "hash.h"
+#include "crc32c.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+#include "transaction.h"
+#include "volumes.h"
+#include "locking.h"
+#include "ref-cache.h"
+#include "compat.h"
+
+#define PENDING_EXTENT_INSERT 0
+#define PENDING_EXTENT_DELETE 1
+#define PENDING_BACKREF_UPDATE 2
+
+struct pending_extent_op {
+	int type;
+	u64 bytenr;
+	u64 num_bytes;
+	u64 parent;
+	u64 orig_parent;
+	u64 generation;
+	u64 orig_generation;
+	int level;
+	struct list_head list;
+	int del;
+};
+
+static int finish_current_insert(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *extent_root, int all);
+static int del_pending_extents(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root, int all);
+static int pin_down_bytes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 bytenr, u64 num_bytes, int is_data);
+static int update_block_group(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      u64 bytenr, u64 num_bytes, int alloc,
+			      int mark_free);
+
+static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
+{
+	return (cache->flags & bits) == bits;
+}
+
+/*
+ * this adds the block group to the fs_info rb tree for the block group
+ * cache
+ */
+static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
+				struct btrfs_block_group_cache *block_group)
+{
+	struct rb_node **p;
+	struct rb_node *parent = NULL;
+	struct btrfs_block_group_cache *cache;
+
+	spin_lock(&info->block_group_cache_lock);
+	p = &info->block_group_cache_tree.rb_node;
+
+	while (*p) {
+		parent = *p;
+		cache = rb_entry(parent, struct btrfs_block_group_cache,
+				 cache_node);
+		if (block_group->key.objectid < cache->key.objectid) {
+			p = &(*p)->rb_left;
+		} else if (block_group->key.objectid > cache->key.objectid) {
+			p = &(*p)->rb_right;
+		} else {
+			spin_unlock(&info->block_group_cache_lock);
+			return -EEXIST;
+		}
+	}
+
+	rb_link_node(&block_group->cache_node, parent, p);
+	rb_insert_color(&block_group->cache_node,
+			&info->block_group_cache_tree);
+	spin_unlock(&info->block_group_cache_lock);
+
+	return 0;
+}
+
+/*
+ * This will return the block group at or after bytenr if contains is 0, else
+ * it will return the block group that contains the bytenr
+ */
+static struct btrfs_block_group_cache *
+block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
+			      int contains)
+{
+	struct btrfs_block_group_cache *cache, *ret = NULL;
+	struct rb_node *n;
+	u64 end, start;
+
+	spin_lock(&info->block_group_cache_lock);
+	n = info->block_group_cache_tree.rb_node;
+
+	while (n) {
+		cache = rb_entry(n, struct btrfs_block_group_cache,
+				 cache_node);
+		end = cache->key.objectid + cache->key.offset - 1;
+		start = cache->key.objectid;
+
+		if (bytenr < start) {
+			if (!contains && (!ret || start < ret->key.objectid))
+				ret = cache;
+			n = n->rb_left;
+		} else if (bytenr > start) {
+			if (contains && bytenr <= end) {
+				ret = cache;
+				break;
+			}
+			n = n->rb_right;
+		} else {
+			ret = cache;
+			break;
+		}
+	}
+	if (ret)
+		atomic_inc(&ret->count);
+	spin_unlock(&info->block_group_cache_lock);
+
+	return ret;
+}
+
+/*
+ * this is only called by cache_block_group, since we could have freed extents
+ * we need to check the pinned_extents for any extents that can't be used yet
+ * since their free space will be released as soon as the transaction commits.
+ */
+static int add_new_free_space(struct btrfs_block_group_cache *block_group,
+			      struct btrfs_fs_info *info, u64 start, u64 end)
+{
+	u64 extent_start, extent_end, size;
+	int ret;
+
+	mutex_lock(&info->pinned_mutex);
+	while (start < end) {
+		ret = find_first_extent_bit(&info->pinned_extents, start,
+					    &extent_start, &extent_end,
+					    EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		if (extent_start == start) {
+			start = extent_end + 1;
+		} else if (extent_start > start && extent_start < end) {
+			size = extent_start - start;
+			ret = btrfs_add_free_space(block_group, start,
+						   size);
+			BUG_ON(ret);
+			start = extent_end + 1;
+		} else {
+			break;
+		}
+	}
+
+	if (start < end) {
+		size = end - start;
+		ret = btrfs_add_free_space(block_group, start, size);
+		BUG_ON(ret);
+	}
+	mutex_unlock(&info->pinned_mutex);
+
+	return 0;
+}
+
+static int remove_sb_from_cache(struct btrfs_root *root,
+				struct btrfs_block_group_cache *cache)
+{
+	u64 bytenr;
+	u64 *logical;
+	int stripe_len;
+	int i, nr, ret;
+
+	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+		bytenr = btrfs_sb_offset(i);
+		ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
+				       cache->key.objectid, bytenr, 0,
+				       &logical, &nr, &stripe_len);
+		BUG_ON(ret);
+		while (nr--) {
+			btrfs_remove_free_space(cache, logical[nr],
+						stripe_len);
+		}
+		kfree(logical);
+	}
+	return 0;
+}
+
+static int cache_block_group(struct btrfs_root *root,
+			     struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_path *path;
+	int ret = 0;
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	int slot;
+	u64 last;
+
+	if (!block_group)
+		return 0;
+
+	root = root->fs_info->extent_root;
+
+	if (block_group->cached)
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 2;
+	/*
+	 * we get into deadlocks with paths held by callers of this function.
+	 * since the alloc_mutex is protecting things right now, just
+	 * skip the locking here
+	 */
+	path->skip_locking = 1;
+	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
+	key.objectid = last;
+	key.offset = 0;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto err;
+
+	while (1) {
+		leaf = path->nodes[0];
+		slot = path->slots[0];
+		if (slot >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto err;
+			if (ret == 0)
+				continue;
+			else
+				break;
+		}
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (key.objectid < block_group->key.objectid)
+			goto next;
+
+		if (key.objectid >= block_group->key.objectid +
+		    block_group->key.offset)
+			break;
+
+		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
+			add_new_free_space(block_group, root->fs_info, last,
+					   key.objectid);
+
+			last = key.objectid + key.offset;
+		}
+next:
+		path->slots[0]++;
+	}
+
+	add_new_free_space(block_group, root->fs_info, last,
+			   block_group->key.objectid +
+			   block_group->key.offset);
+
+	remove_sb_from_cache(root, block_group);
+	block_group->cached = 1;
+	ret = 0;
+err:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * return the block group that starts at or after bytenr
+ */
+static struct btrfs_block_group_cache *
+btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
+{
+	struct btrfs_block_group_cache *cache;
+
+	cache = block_group_cache_tree_search(info, bytenr, 0);
+
+	return cache;
+}
+
+/*
+ * return the block group that contains teh given bytenr
+ */
+struct btrfs_block_group_cache *btrfs_lookup_block_group(
+						 struct btrfs_fs_info *info,
+						 u64 bytenr)
+{
+	struct btrfs_block_group_cache *cache;
+
+	cache = block_group_cache_tree_search(info, bytenr, 1);
+
+	return cache;
+}
+
+static inline void put_block_group(struct btrfs_block_group_cache *cache)
+{
+	if (atomic_dec_and_test(&cache->count))
+		kfree(cache);
+}
+
+static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
+						  u64 flags)
+{
+	struct list_head *head = &info->space_info;
+	struct list_head *cur;
+	struct btrfs_space_info *found;
+	list_for_each(cur, head) {
+		found = list_entry(cur, struct btrfs_space_info, list);
+		if (found->flags == flags)
+			return found;
+	}
+	return NULL;
+}
+
+static u64 div_factor(u64 num, int factor)
+{
+	if (factor == 10)
+		return num;
+	num *= factor;
+	do_div(num, 10);
+	return num;
+}
+
+u64 btrfs_find_block_group(struct btrfs_root *root,
+			   u64 search_start, u64 search_hint, int owner)
+{
+	struct btrfs_block_group_cache *cache;
+	u64 used;
+	u64 last = max(search_hint, search_start);
+	u64 group_start = 0;
+	int full_search = 0;
+	int factor = 9;
+	int wrapped = 0;
+again:
+	while (1) {
+		cache = btrfs_lookup_first_block_group(root->fs_info, last);
+		if (!cache)
+			break;
+
+		spin_lock(&cache->lock);
+		last = cache->key.objectid + cache->key.offset;
+		used = btrfs_block_group_used(&cache->item);
+
+		if ((full_search || !cache->ro) &&
+		    block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
+			if (used + cache->pinned + cache->reserved <
+			    div_factor(cache->key.offset, factor)) {
+				group_start = cache->key.objectid;
+				spin_unlock(&cache->lock);
+				put_block_group(cache);
+				goto found;
+			}
+		}
+		spin_unlock(&cache->lock);
+		put_block_group(cache);
+		cond_resched();
+	}
+	if (!wrapped) {
+		last = search_start;
+		wrapped = 1;
+		goto again;
+	}
+	if (!full_search && factor < 10) {
+		last = search_start;
+		full_search = 1;
+		factor = 10;
+		goto again;
+	}
+found:
+	return group_start;
+}
+
+/* simple helper to search for an existing extent at a given offset */
+int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_path *path;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	key.objectid = start;
+	key.offset = len;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
+				0, 0);
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * Back reference rules.  Back refs have three main goals:
+ *
+ * 1) differentiate between all holders of references to an extent so that
+ *    when a reference is dropped we can make sure it was a valid reference
+ *    before freeing the extent.
+ *
+ * 2) Provide enough information to quickly find the holders of an extent
+ *    if we notice a given block is corrupted or bad.
+ *
+ * 3) Make it easy to migrate blocks for FS shrinking or storage pool
+ *    maintenance.  This is actually the same as #2, but with a slightly
+ *    different use case.
+ *
+ * File extents can be referenced by:
+ *
+ * - multiple snapshots, subvolumes, or different generations in one subvol
+ * - different files inside a single subvolume
+ * - different offsets inside a file (bookend extents in file.c)
+ *
+ * The extent ref structure has fields for:
+ *
+ * - Objectid of the subvolume root
+ * - Generation number of the tree holding the reference
+ * - objectid of the file holding the reference
+ * - number of references holding by parent node (alway 1 for tree blocks)
+ *
+ * Btree leaf may hold multiple references to a file extent. In most cases,
+ * these references are from same file and the corresponding offsets inside
+ * the file are close together.
+ *
+ * When a file extent is allocated the fields are filled in:
+ *     (root_key.objectid, trans->transid, inode objectid, 1)
+ *
+ * When a leaf is cow'd new references are added for every file extent found
+ * in the leaf.  It looks similar to the create case, but trans->transid will
+ * be different when the block is cow'd.
+ *
+ *     (root_key.objectid, trans->transid, inode objectid,
+ *      number of references in the leaf)
+ *
+ * When a file extent is removed either during snapshot deletion or
+ * file truncation, we find the corresponding back reference and check
+ * the following fields:
+ *
+ *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
+ *      inode objectid)
+ *
+ * Btree extents can be referenced by:
+ *
+ * - Different subvolumes
+ * - Different generations of the same subvolume
+ *
+ * When a tree block is created, back references are inserted:
+ *
+ * (root->root_key.objectid, trans->transid, level, 1)
+ *
+ * When a tree block is cow'd, new back references are added for all the
+ * blocks it points to. If the tree block isn't in reference counted root,
+ * the old back references are removed. These new back references are of
+ * the form (trans->transid will have increased since creation):
+ *
+ * (root->root_key.objectid, trans->transid, level, 1)
+ *
+ * When a backref is in deleting, the following fields are checked:
+ *
+ * if backref was for a tree root:
+ *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
+ * else
+ *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
+ *
+ * Back Reference Key composing:
+ *
+ * The key objectid corresponds to the first byte in the extent, the key
+ * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
+ * byte of parent extent. If a extent is tree root, the key offset is set
+ * to the key objectid.
+ */
+
+static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, u64 parent,
+					  u64 ref_root, u64 ref_generation,
+					  u64 owner_objectid, int del)
+{
+	struct btrfs_key key;
+	struct btrfs_extent_ref *ref;
+	struct extent_buffer *leaf;
+	u64 ref_objectid;
+	int ret;
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	key.offset = parent;
+
+	ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	leaf = path->nodes[0];
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+	ref_objectid = btrfs_ref_objectid(leaf, ref);
+	if (btrfs_ref_root(leaf, ref) != ref_root ||
+	    btrfs_ref_generation(leaf, ref) != ref_generation ||
+	    (ref_objectid != owner_objectid &&
+	     ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
+		ret = -EIO;
+		WARN_ON(1);
+		goto out;
+	}
+	ret = 0;
+out:
+	return ret;
+}
+
+/*
+ * updates all the backrefs that are pending on update_list for the
+ * extent_root
+ */
+static noinline int update_backrefs(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *extent_root,
+				    struct btrfs_path *path,
+				    struct list_head *update_list)
+{
+	struct btrfs_key key;
+	struct btrfs_extent_ref *ref;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct pending_extent_op *op;
+	struct extent_buffer *leaf;
+	int ret = 0;
+	struct list_head *cur = update_list->next;
+	u64 ref_objectid;
+	u64 ref_root = extent_root->root_key.objectid;
+
+	op = list_entry(cur, struct pending_extent_op, list);
+
+search:
+	key.objectid = op->bytenr;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	key.offset = op->orig_parent;
+
+	ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+
+loop:
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+
+	ref_objectid = btrfs_ref_objectid(leaf, ref);
+
+	if (btrfs_ref_root(leaf, ref) != ref_root ||
+	    btrfs_ref_generation(leaf, ref) != op->orig_generation ||
+	    (ref_objectid != op->level &&
+	     ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
+		printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
+		       "root %llu, owner %u\n",
+		       (unsigned long long)op->bytenr,
+		       (unsigned long long)op->orig_parent,
+		       (unsigned long long)ref_root, op->level);
+		btrfs_print_leaf(extent_root, leaf);
+		BUG();
+	}
+
+	key.objectid = op->bytenr;
+	key.offset = op->parent;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
+	BUG_ON(ret);
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+	btrfs_set_ref_generation(leaf, ref, op->generation);
+
+	cur = cur->next;
+
+	list_del_init(&op->list);
+	unlock_extent(&info->extent_ins, op->bytenr,
+		      op->bytenr + op->num_bytes - 1, GFP_NOFS);
+	kfree(op);
+
+	if (cur == update_list) {
+		btrfs_mark_buffer_dirty(path->nodes[0]);
+		btrfs_release_path(extent_root, path);
+		goto out;
+	}
+
+	op = list_entry(cur, struct pending_extent_op, list);
+
+	path->slots[0]++;
+	while (path->slots[0] < btrfs_header_nritems(leaf)) {
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid == op->bytenr &&
+		    key.type == BTRFS_EXTENT_REF_KEY)
+			goto loop;
+		path->slots[0]++;
+	}
+
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(extent_root, path);
+	goto search;
+
+out:
+	return 0;
+}
+
+static noinline int insert_extents(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *extent_root,
+				   struct btrfs_path *path,
+				   struct list_head *insert_list, int nr)
+{
+	struct btrfs_key *keys;
+	u32 *data_size;
+	struct pending_extent_op *op;
+	struct extent_buffer *leaf;
+	struct list_head *cur = insert_list->next;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	u64 ref_root = extent_root->root_key.objectid;
+	int i = 0, last = 0, ret;
+	int total = nr * 2;
+
+	if (!nr)
+		return 0;
+
+	keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
+	if (!keys)
+		return -ENOMEM;
+
+	data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
+	if (!data_size) {
+		kfree(keys);
+		return -ENOMEM;
+	}
+
+	list_for_each_entry(op, insert_list, list) {
+		keys[i].objectid = op->bytenr;
+		keys[i].offset = op->num_bytes;
+		keys[i].type = BTRFS_EXTENT_ITEM_KEY;
+		data_size[i] = sizeof(struct btrfs_extent_item);
+		i++;
+
+		keys[i].objectid = op->bytenr;
+		keys[i].offset = op->parent;
+		keys[i].type = BTRFS_EXTENT_REF_KEY;
+		data_size[i] = sizeof(struct btrfs_extent_ref);
+		i++;
+	}
+
+	op = list_entry(cur, struct pending_extent_op, list);
+	i = 0;
+	while (i < total) {
+		int c;
+		ret = btrfs_insert_some_items(trans, extent_root, path,
+					      keys+i, data_size+i, total-i);
+		BUG_ON(ret < 0);
+
+		if (last && ret > 1)
+			BUG();
+
+		leaf = path->nodes[0];
+		for (c = 0; c < ret; c++) {
+			int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
+
+			/*
+			 * if the first item we inserted was a backref, then
+			 * the EXTENT_ITEM will be the odd c's, else it will
+			 * be the even c's
+			 */
+			if ((ref_first && (c % 2)) ||
+			    (!ref_first && !(c % 2))) {
+				struct btrfs_extent_item *itm;
+
+				itm = btrfs_item_ptr(leaf, path->slots[0] + c,
+						     struct btrfs_extent_item);
+				btrfs_set_extent_refs(path->nodes[0], itm, 1);
+				op->del++;
+			} else {
+				struct btrfs_extent_ref *ref;
+
+				ref = btrfs_item_ptr(leaf, path->slots[0] + c,
+						     struct btrfs_extent_ref);
+				btrfs_set_ref_root(leaf, ref, ref_root);
+				btrfs_set_ref_generation(leaf, ref,
+							 op->generation);
+				btrfs_set_ref_objectid(leaf, ref, op->level);
+				btrfs_set_ref_num_refs(leaf, ref, 1);
+				op->del++;
+			}
+
+			/*
+			 * using del to see when its ok to free up the
+			 * pending_extent_op.  In the case where we insert the
+			 * last item on the list in order to help do batching
+			 * we need to not free the extent op until we actually
+			 * insert the extent_item
+			 */
+			if (op->del == 2) {
+				unlock_extent(&info->extent_ins, op->bytenr,
+					      op->bytenr + op->num_bytes - 1,
+					      GFP_NOFS);
+				cur = cur->next;
+				list_del_init(&op->list);
+				kfree(op);
+				if (cur != insert_list)
+					op = list_entry(cur,
+						struct pending_extent_op,
+						list);
+			}
+		}
+		btrfs_mark_buffer_dirty(leaf);
+		btrfs_release_path(extent_root, path);
+
+		/*
+		 * Ok backref's and items usually go right next to eachother,
+		 * but if we could only insert 1 item that means that we
+		 * inserted on the end of a leaf, and we have no idea what may
+		 * be on the next leaf so we just play it safe.  In order to
+		 * try and help this case we insert the last thing on our
+		 * insert list so hopefully it will end up being the last
+		 * thing on the leaf and everything else will be before it,
+		 * which will let us insert a whole bunch of items at the same
+		 * time.
+		 */
+		if (ret == 1 && !last && (i + ret < total)) {
+			/*
+			 * last: where we will pick up the next time around
+			 * i: our current key to insert, will be total - 1
+			 * cur: the current op we are screwing with
+			 * op: duh
+			 */
+			last = i + ret;
+			i = total - 1;
+			cur = insert_list->prev;
+			op = list_entry(cur, struct pending_extent_op, list);
+		} else if (last) {
+			/*
+			 * ok we successfully inserted the last item on the
+			 * list, lets reset everything
+			 *
+			 * i: our current key to insert, so where we left off
+			 *    last time
+			 * last: done with this
+			 * cur: the op we are messing with
+			 * op: duh
+			 * total: since we inserted the last key, we need to
+			 *        decrement total so we dont overflow
+			 */
+			i = last;
+			last = 0;
+			total--;
+			if (i < total) {
+				cur = insert_list->next;
+				op = list_entry(cur, struct pending_extent_op,
+						list);
+			}
+		} else {
+			i += ret;
+		}
+
+		cond_resched();
+	}
+	ret = 0;
+	kfree(keys);
+	kfree(data_size);
+	return ret;
+}
+
+static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 bytenr, u64 parent,
+					  u64 ref_root, u64 ref_generation,
+					  u64 owner_objectid)
+{
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref *ref;
+	u32 num_refs;
+	int ret;
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_REF_KEY;
+	key.offset = parent;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
+	if (ret == 0) {
+		leaf = path->nodes[0];
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_ref);
+		btrfs_set_ref_root(leaf, ref, ref_root);
+		btrfs_set_ref_generation(leaf, ref, ref_generation);
+		btrfs_set_ref_objectid(leaf, ref, owner_objectid);
+		btrfs_set_ref_num_refs(leaf, ref, 1);
+	} else if (ret == -EEXIST) {
+		u64 existing_owner;
+		BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
+		leaf = path->nodes[0];
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_ref);
+		if (btrfs_ref_root(leaf, ref) != ref_root ||
+		    btrfs_ref_generation(leaf, ref) != ref_generation) {
+			ret = -EIO;
+			WARN_ON(1);
+			goto out;
+		}
+
+		num_refs = btrfs_ref_num_refs(leaf, ref);
+		BUG_ON(num_refs == 0);
+		btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
+
+		existing_owner = btrfs_ref_objectid(leaf, ref);
+		if (existing_owner != owner_objectid &&
+		    existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
+			btrfs_set_ref_objectid(leaf, ref,
+					BTRFS_MULTIPLE_OBJECTIDS);
+		}
+		ret = 0;
+	} else {
+		goto out;
+	}
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+out:
+	btrfs_release_path(root, path);
+	return ret;
+}
+
+static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref *ref;
+	u32 num_refs;
+	int ret = 0;
+
+	leaf = path->nodes[0];
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
+	num_refs = btrfs_ref_num_refs(leaf, ref);
+	BUG_ON(num_refs == 0);
+	num_refs -= 1;
+	if (num_refs == 0) {
+		ret = btrfs_del_item(trans, root, path);
+	} else {
+		btrfs_set_ref_num_refs(leaf, ref, num_refs);
+		btrfs_mark_buffer_dirty(leaf);
+	}
+	btrfs_release_path(root, path);
+	return ret;
+}
+
+#ifdef BIO_RW_DISCARD
+static void btrfs_issue_discard(struct block_device *bdev,
+				u64 start, u64 len)
+{
+	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
+}
+#endif
+
+static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
+				u64 num_bytes)
+{
+#ifdef BIO_RW_DISCARD
+	int ret;
+	u64 map_length = num_bytes;
+	struct btrfs_multi_bio *multi = NULL;
+
+	/* Tell the block device(s) that the sectors can be discarded */
+	ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
+			      bytenr, &map_length, &multi, 0);
+	if (!ret) {
+		struct btrfs_bio_stripe *stripe = multi->stripes;
+		int i;
+
+		if (map_length > num_bytes)
+			map_length = num_bytes;
+
+		for (i = 0; i < multi->num_stripes; i++, stripe++) {
+			btrfs_issue_discard(stripe->dev->bdev,
+					    stripe->physical,
+					    map_length);
+		}
+		kfree(multi);
+	}
+
+	return ret;
+#else
+	return 0;
+#endif
+}
+
+static noinline int free_extents(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *extent_root,
+				 struct list_head *del_list)
+{
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct btrfs_path *path;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	struct list_head *cur;
+	struct pending_extent_op *op;
+	struct btrfs_extent_item *ei;
+	int ret, num_to_del, extent_slot = 0, found_extent = 0;
+	u32 refs;
+	u64 bytes_freed = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	path->reada = 1;
+
+search:
+	/* search for the backref for the current ref we want to delete */
+	cur = del_list->next;
+	op = list_entry(cur, struct pending_extent_op, list);
+	ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
+				    op->orig_parent,
+				    extent_root->root_key.objectid,
+				    op->orig_generation, op->level, 1);
+	if (ret) {
+		printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
+		       "root %llu gen %llu owner %u\n",
+		       (unsigned long long)op->bytenr,
+		       (unsigned long long)extent_root->root_key.objectid,
+		       (unsigned long long)op->orig_generation, op->level);
+		btrfs_print_leaf(extent_root, path->nodes[0]);
+		WARN_ON(1);
+		goto out;
+	}
+
+	extent_slot = path->slots[0];
+	num_to_del = 1;
+	found_extent = 0;
+
+	/*
+	 * if we aren't the first item on the leaf we can move back one and see
+	 * if our ref is right next to our extent item
+	 */
+	if (likely(extent_slot)) {
+		extent_slot--;
+		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+				      extent_slot);
+		if (found_key.objectid == op->bytenr &&
+		    found_key.type == BTRFS_EXTENT_ITEM_KEY &&
+		    found_key.offset == op->num_bytes) {
+			num_to_del++;
+			found_extent = 1;
+		}
+	}
+
+	/*
+	 * if we didn't find the extent we need to delete the backref and then
+	 * search for the extent item key so we can update its ref count
+	 */
+	if (!found_extent) {
+		key.objectid = op->bytenr;
+		key.type = BTRFS_EXTENT_ITEM_KEY;
+		key.offset = op->num_bytes;
+
+		ret = remove_extent_backref(trans, extent_root, path);
+		BUG_ON(ret);
+		btrfs_release_path(extent_root, path);
+		ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
+		BUG_ON(ret);
+		extent_slot = path->slots[0];
+	}
+
+	/* this is where we update the ref count for the extent */
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, ei);
+	BUG_ON(refs == 0);
+	refs--;
+	btrfs_set_extent_refs(leaf, ei, refs);
+
+	btrfs_mark_buffer_dirty(leaf);
+
+	/*
+	 * This extent needs deleting.  The reason cur_slot is extent_slot +
+	 * num_to_del is because extent_slot points to the slot where the extent
+	 * is, and if the backref was not right next to the extent we will be
+	 * deleting at least 1 item, and will want to start searching at the
+	 * slot directly next to extent_slot.  However if we did find the
+	 * backref next to the extent item them we will be deleting at least 2
+	 * items and will want to start searching directly after the ref slot
+	 */
+	if (!refs) {
+		struct list_head *pos, *n, *end;
+		int cur_slot = extent_slot+num_to_del;
+		u64 super_used;
+		u64 root_used;
+
+		path->slots[0] = extent_slot;
+		bytes_freed = op->num_bytes;
+
+		mutex_lock(&info->pinned_mutex);
+		ret = pin_down_bytes(trans, extent_root, op->bytenr,
+				     op->num_bytes, op->level >=
+				     BTRFS_FIRST_FREE_OBJECTID);
+		mutex_unlock(&info->pinned_mutex);
+		BUG_ON(ret < 0);
+		op->del = ret;
+
+		/*
+		 * we need to see if we can delete multiple things at once, so
+		 * start looping through the list of extents we are wanting to
+		 * delete and see if their extent/backref's are right next to
+		 * eachother and the extents only have 1 ref
+		 */
+		for (pos = cur->next; pos != del_list; pos = pos->next) {
+			struct pending_extent_op *tmp;
+
+			tmp = list_entry(pos, struct pending_extent_op, list);
+
+			/* we only want to delete extent+ref at this stage */
+			if (cur_slot >= btrfs_header_nritems(leaf) - 1)
+				break;
+
+			btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
+			if (found_key.objectid != tmp->bytenr ||
+			    found_key.type != BTRFS_EXTENT_ITEM_KEY ||
+			    found_key.offset != tmp->num_bytes)
+				break;
+
+			/* check to make sure this extent only has one ref */
+			ei = btrfs_item_ptr(leaf, cur_slot,
+					    struct btrfs_extent_item);
+			if (btrfs_extent_refs(leaf, ei) != 1)
+				break;
+
+			btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
+			if (found_key.objectid != tmp->bytenr ||
+			    found_key.type != BTRFS_EXTENT_REF_KEY ||
+			    found_key.offset != tmp->orig_parent)
+				break;
+
+			/*
+			 * the ref is right next to the extent, we can set the
+			 * ref count to 0 since we will delete them both now
+			 */
+			btrfs_set_extent_refs(leaf, ei, 0);
+
+			/* pin down the bytes for this extent */
+			mutex_lock(&info->pinned_mutex);
+			ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
+					     tmp->num_bytes, tmp->level >=
+					     BTRFS_FIRST_FREE_OBJECTID);
+			mutex_unlock(&info->pinned_mutex);
+			BUG_ON(ret < 0);
+
+			/*
+			 * use the del field to tell if we need to go ahead and
+			 * free up the extent when we delete the item or not.
+			 */
+			tmp->del = ret;
+			bytes_freed += tmp->num_bytes;
+
+			num_to_del += 2;
+			cur_slot += 2;
+		}
+		end = pos;
+
+		/* update the free space counters */
+		spin_lock(&info->delalloc_lock);
+		super_used = btrfs_super_bytes_used(&info->super_copy);
+		btrfs_set_super_bytes_used(&info->super_copy,
+					   super_used - bytes_freed);
+
+		root_used = btrfs_root_used(&extent_root->root_item);
+		btrfs_set_root_used(&extent_root->root_item,
+				    root_used - bytes_freed);
+		spin_unlock(&info->delalloc_lock);
+
+		/* delete the items */
+		ret = btrfs_del_items(trans, extent_root, path,
+				      path->slots[0], num_to_del);
+		BUG_ON(ret);
+
+		/*
+		 * loop through the extents we deleted and do the cleanup work
+		 * on them
+		 */
+		for (pos = cur, n = pos->next; pos != end;
+		     pos = n, n = pos->next) {
+			struct pending_extent_op *tmp;
+			tmp = list_entry(pos, struct pending_extent_op, list);
+
+			/*
+			 * remember tmp->del tells us wether or not we pinned
+			 * down the extent
+			 */
+			ret = update_block_group(trans, extent_root,
+						 tmp->bytenr, tmp->num_bytes, 0,
+						 tmp->del);
+			BUG_ON(ret);
+
+			list_del_init(&tmp->list);
+			unlock_extent(&info->extent_ins, tmp->bytenr,
+				      tmp->bytenr + tmp->num_bytes - 1,
+				      GFP_NOFS);
+			kfree(tmp);
+		}
+	} else if (refs && found_extent) {
+		/*
+		 * the ref and extent were right next to eachother, but the
+		 * extent still has a ref, so just free the backref and keep
+		 * going
+		 */
+		ret = remove_extent_backref(trans, extent_root, path);
+		BUG_ON(ret);
+
+		list_del_init(&op->list);
+		unlock_extent(&info->extent_ins, op->bytenr,
+			      op->bytenr + op->num_bytes - 1, GFP_NOFS);
+		kfree(op);
+	} else {
+		/*
+		 * the extent has multiple refs and the backref we were looking
+		 * for was not right next to it, so just unlock and go next,
+		 * we're good to go
+		 */
+		list_del_init(&op->list);
+		unlock_extent(&info->extent_ins, op->bytenr,
+			      op->bytenr + op->num_bytes - 1, GFP_NOFS);
+		kfree(op);
+	}
+
+	btrfs_release_path(extent_root, path);
+	if (!list_empty(del_list))
+		goto search;
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root, u64 bytenr,
+				     u64 orig_parent, u64 parent,
+				     u64 orig_root, u64 ref_root,
+				     u64 orig_generation, u64 ref_generation,
+				     u64 owner_objectid)
+{
+	int ret;
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	struct btrfs_path *path;
+
+	if (root == root->fs_info->extent_root) {
+		struct pending_extent_op *extent_op;
+		u64 num_bytes;
+
+		BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
+		num_bytes = btrfs_level_size(root, (int)owner_objectid);
+		mutex_lock(&root->fs_info->extent_ins_mutex);
+		if (test_range_bit(&root->fs_info->extent_ins, bytenr,
+				bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
+			u64 priv;
+			ret = get_state_private(&root->fs_info->extent_ins,
+						bytenr, &priv);
+			BUG_ON(ret);
+			extent_op = (struct pending_extent_op *)
+							(unsigned long)priv;
+			BUG_ON(extent_op->parent != orig_parent);
+			BUG_ON(extent_op->generation != orig_generation);
+
+			extent_op->parent = parent;
+			extent_op->generation = ref_generation;
+		} else {
+			extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+			BUG_ON(!extent_op);
+
+			extent_op->type = PENDING_BACKREF_UPDATE;
+			extent_op->bytenr = bytenr;
+			extent_op->num_bytes = num_bytes;
+			extent_op->parent = parent;
+			extent_op->orig_parent = orig_parent;
+			extent_op->generation = ref_generation;
+			extent_op->orig_generation = orig_generation;
+			extent_op->level = (int)owner_objectid;
+			INIT_LIST_HEAD(&extent_op->list);
+			extent_op->del = 0;
+
+			set_extent_bits(&root->fs_info->extent_ins,
+					bytenr, bytenr + num_bytes - 1,
+					EXTENT_WRITEBACK, GFP_NOFS);
+			set_state_private(&root->fs_info->extent_ins,
+					  bytenr, (unsigned long)extent_op);
+		}
+		mutex_unlock(&root->fs_info->extent_ins_mutex);
+		return 0;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	ret = lookup_extent_backref(trans, extent_root, path,
+				    bytenr, orig_parent, orig_root,
+				    orig_generation, owner_objectid, 1);
+	if (ret)
+		goto out;
+	ret = remove_extent_backref(trans, extent_root, path);
+	if (ret)
+		goto out;
+	ret = insert_extent_backref(trans, extent_root, path, bytenr,
+				    parent, ref_root, ref_generation,
+				    owner_objectid);
+	BUG_ON(ret);
+	finish_current_insert(trans, extent_root, 0);
+	del_pending_extents(trans, extent_root, 0);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 orig_parent, u64 parent,
+			    u64 ref_root, u64 ref_generation,
+			    u64 owner_objectid)
+{
+	int ret;
+	if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
+	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
+		return 0;
+	ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
+					parent, ref_root, ref_root,
+					ref_generation, ref_generation,
+					owner_objectid);
+	return ret;
+}
+
+static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root, u64 bytenr,
+				  u64 orig_parent, u64 parent,
+				  u64 orig_root, u64 ref_root,
+				  u64 orig_generation, u64 ref_generation,
+				  u64 owner_objectid)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	struct extent_buffer *l;
+	struct btrfs_extent_item *item;
+	u32 refs;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 1;
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
+				0, 1);
+	if (ret < 0)
+		return ret;
+	BUG_ON(ret == 0 || path->slots[0] == 0);
+
+	path->slots[0]--;
+	l = path->nodes[0];
+
+	btrfs_item_key_to_cpu(l, &key, path->slots[0]);
+	if (key.objectid != bytenr) {
+		btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
+		printk(KERN_ERR "btrfs wanted %llu found %llu\n",
+		       (unsigned long long)bytenr,
+		       (unsigned long long)key.objectid);
+		BUG();
+	}
+	BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
+
+	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
+	refs = btrfs_extent_refs(l, item);
+	btrfs_set_extent_refs(l, item, refs + 1);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+
+	btrfs_release_path(root->fs_info->extent_root, path);
+
+	path->reada = 1;
+	ret = insert_extent_backref(trans, root->fs_info->extent_root,
+				    path, bytenr, parent,
+				    ref_root, ref_generation,
+				    owner_objectid);
+	BUG_ON(ret);
+	finish_current_insert(trans, root->fs_info->extent_root, 0);
+	del_pending_extents(trans, root->fs_info->extent_root, 0);
+
+	btrfs_free_path(path);
+	return 0;
+}
+
+int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 bytenr, u64 num_bytes, u64 parent,
+			 u64 ref_root, u64 ref_generation,
+			 u64 owner_objectid)
+{
+	int ret;
+	if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
+	    owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
+		return 0;
+	ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
+				     0, ref_root, 0, ref_generation,
+				     owner_objectid);
+	return ret;
+}
+
+int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root)
+{
+	finish_current_insert(trans, root->fs_info->extent_root, 1);
+	del_pending_extents(trans, root->fs_info->extent_root, 1);
+	return 0;
+}
+
+int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, u64 bytenr,
+			    u64 num_bytes, u32 *refs)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	struct extent_buffer *l;
+	struct btrfs_extent_item *item;
+
+	WARN_ON(num_bytes < root->sectorsize);
+	path = btrfs_alloc_path();
+	path->reada = 1;
+	key.objectid = bytenr;
+	key.offset = num_bytes;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
+				0, 0);
+	if (ret < 0)
+		goto out;
+	if (ret != 0) {
+		btrfs_print_leaf(root, path->nodes[0]);
+		printk(KERN_INFO "btrfs failed to find block number %llu\n",
+		       (unsigned long long)bytenr);
+		BUG();
+	}
+	l = path->nodes[0];
+	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
+	*refs = btrfs_extent_refs(l, item);
+out:
+	btrfs_free_path(path);
+	return 0;
+}
+
+int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, u64 objectid, u64 bytenr)
+{
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_extent_ref *ref_item;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	u64 ref_root;
+	u64 last_snapshot;
+	u32 nritems;
+	int ret;
+
+	key.objectid = bytenr;
+	key.offset = (u64)-1;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+
+	path = btrfs_alloc_path();
+	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	BUG_ON(ret == 0);
+
+	ret = -ENOENT;
+	if (path->slots[0] == 0)
+		goto out;
+
+	path->slots[0]--;
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+	if (found_key.objectid != bytenr ||
+	    found_key.type != BTRFS_EXTENT_ITEM_KEY)
+		goto out;
+
+	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(extent_root, path);
+			if (ret < 0)
+				goto out;
+			if (ret == 0)
+				continue;
+			break;
+		}
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid != bytenr)
+			break;
+
+		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
+			path->slots[0]++;
+			continue;
+		}
+
+		ref_item = btrfs_item_ptr(leaf, path->slots[0],
+					  struct btrfs_extent_ref);
+		ref_root = btrfs_ref_root(leaf, ref_item);
+		if ((ref_root != root->root_key.objectid &&
+		     ref_root != BTRFS_TREE_LOG_OBJECTID) ||
+		     objectid != btrfs_ref_objectid(leaf, ref_item)) {
+			ret = 1;
+			goto out;
+		}
+		if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
+			ret = 1;
+			goto out;
+		}
+
+		path->slots[0]++;
+	}
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		    struct extent_buffer *buf, u32 nr_extents)
+{
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	u64 root_gen;
+	u32 nritems;
+	int i;
+	int level;
+	int ret = 0;
+	int shared = 0;
+
+	if (!root->ref_cows)
+		return 0;
+
+	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
+		shared = 0;
+		root_gen = root->root_key.offset;
+	} else {
+		shared = 1;
+		root_gen = trans->transid - 1;
+	}
+
+	level = btrfs_header_level(buf);
+	nritems = btrfs_header_nritems(buf);
+
+	if (level == 0) {
+		struct btrfs_leaf_ref *ref;
+		struct btrfs_extent_info *info;
+
+		ref = btrfs_alloc_leaf_ref(root, nr_extents);
+		if (!ref) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ref->root_gen = root_gen;
+		ref->bytenr = buf->start;
+		ref->owner = btrfs_header_owner(buf);
+		ref->generation = btrfs_header_generation(buf);
+		ref->nritems = nr_extents;
+		info = ref->extents;
+
+		for (i = 0; nr_extents > 0 && i < nritems; i++) {
+			u64 disk_bytenr;
+			btrfs_item_key_to_cpu(buf, &key, i);
+			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+				continue;
+			fi = btrfs_item_ptr(buf, i,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(buf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE)
+				continue;
+			disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
+			if (disk_bytenr == 0)
+				continue;
+
+			info->bytenr = disk_bytenr;
+			info->num_bytes =
+				btrfs_file_extent_disk_num_bytes(buf, fi);
+			info->objectid = key.objectid;
+			info->offset = key.offset;
+			info++;
+		}
+
+		ret = btrfs_add_leaf_ref(root, ref, shared);
+		if (ret == -EEXIST && shared) {
+			struct btrfs_leaf_ref *old;
+			old = btrfs_lookup_leaf_ref(root, ref->bytenr);
+			BUG_ON(!old);
+			btrfs_remove_leaf_ref(root, old);
+			btrfs_free_leaf_ref(root, old);
+			ret = btrfs_add_leaf_ref(root, ref, shared);
+		}
+		WARN_ON(ret);
+		btrfs_free_leaf_ref(root, ref);
+	}
+out:
+	return ret;
+}
+
+int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		  struct extent_buffer *orig_buf, struct extent_buffer *buf,
+		  u32 *nr_extents)
+{
+	u64 bytenr;
+	u64 ref_root;
+	u64 orig_root;
+	u64 ref_generation;
+	u64 orig_generation;
+	u32 nritems;
+	u32 nr_file_extents = 0;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int i;
+	int level;
+	int ret = 0;
+	int faili = 0;
+	int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
+			    u64, u64, u64, u64, u64, u64, u64, u64);
+
+	ref_root = btrfs_header_owner(buf);
+	ref_generation = btrfs_header_generation(buf);
+	orig_root = btrfs_header_owner(orig_buf);
+	orig_generation = btrfs_header_generation(orig_buf);
+
+	nritems = btrfs_header_nritems(buf);
+	level = btrfs_header_level(buf);
+
+	if (root->ref_cows) {
+		process_func = __btrfs_inc_extent_ref;
+	} else {
+		if (level == 0 &&
+		    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+			goto out;
+		if (level != 0 &&
+		    root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
+			goto out;
+		process_func = __btrfs_update_extent_ref;
+	}
+
+	for (i = 0; i < nritems; i++) {
+		cond_resched();
+		if (level == 0) {
+			btrfs_item_key_to_cpu(buf, &key, i);
+			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+				continue;
+			fi = btrfs_item_ptr(buf, i,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(buf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE)
+				continue;
+			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
+			if (bytenr == 0)
+				continue;
+
+			nr_file_extents++;
+
+			ret = process_func(trans, root, bytenr,
+					   orig_buf->start, buf->start,
+					   orig_root, ref_root,
+					   orig_generation, ref_generation,
+					   key.objectid);
+
+			if (ret) {
+				faili = i;
+				WARN_ON(1);
+				goto fail;
+			}
+		} else {
+			bytenr = btrfs_node_blockptr(buf, i);
+			ret = process_func(trans, root, bytenr,
+					   orig_buf->start, buf->start,
+					   orig_root, ref_root,
+					   orig_generation, ref_generation,
+					   level - 1);
+			if (ret) {
+				faili = i;
+				WARN_ON(1);
+				goto fail;
+			}
+		}
+	}
+out:
+	if (nr_extents) {
+		if (level == 0)
+			*nr_extents = nr_file_extents;
+		else
+			*nr_extents = nritems;
+	}
+	return 0;
+fail:
+	WARN_ON(1);
+	return ret;
+}
+
+int btrfs_update_ref(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root, struct extent_buffer *orig_buf,
+		     struct extent_buffer *buf, int start_slot, int nr)
+
+{
+	u64 bytenr;
+	u64 ref_root;
+	u64 orig_root;
+	u64 ref_generation;
+	u64 orig_generation;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int i;
+	int ret;
+	int slot;
+	int level;
+
+	BUG_ON(start_slot < 0);
+	BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
+
+	ref_root = btrfs_header_owner(buf);
+	ref_generation = btrfs_header_generation(buf);
+	orig_root = btrfs_header_owner(orig_buf);
+	orig_generation = btrfs_header_generation(orig_buf);
+	level = btrfs_header_level(buf);
+
+	if (!root->ref_cows) {
+		if (level == 0 &&
+		    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
+			return 0;
+		if (level != 0 &&
+		    root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
+			return 0;
+	}
+
+	for (i = 0, slot = start_slot; i < nr; i++, slot++) {
+		cond_resched();
+		if (level == 0) {
+			btrfs_item_key_to_cpu(buf, &key, slot);
+			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+				continue;
+			fi = btrfs_item_ptr(buf, slot,
+					    struct btrfs_file_extent_item);
+			if (btrfs_file_extent_type(buf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE)
+				continue;
+			bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
+			if (bytenr == 0)
+				continue;
+			ret = __btrfs_update_extent_ref(trans, root, bytenr,
+					    orig_buf->start, buf->start,
+					    orig_root, ref_root,
+					    orig_generation, ref_generation,
+					    key.objectid);
+			if (ret)
+				goto fail;
+		} else {
+			bytenr = btrfs_node_blockptr(buf, slot);
+			ret = __btrfs_update_extent_ref(trans, root, bytenr,
+					    orig_buf->start, buf->start,
+					    orig_root, ref_root,
+					    orig_generation, ref_generation,
+					    level - 1);
+			if (ret)
+				goto fail;
+		}
+	}
+	return 0;
+fail:
+	WARN_ON(1);
+	return -1;
+}
+
+static int write_one_cache_group(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct btrfs_block_group_cache *cache)
+{
+	int ret;
+	int pending_ret;
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	unsigned long bi;
+	struct extent_buffer *leaf;
+
+	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
+	if (ret < 0)
+		goto fail;
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(extent_root, path);
+fail:
+	finish_current_insert(trans, extent_root, 0);
+	pending_ret = del_pending_extents(trans, extent_root, 0);
+	if (ret)
+		return ret;
+	if (pending_ret)
+		return pending_ret;
+	return 0;
+
+}
+
+int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root)
+{
+	struct btrfs_block_group_cache *cache, *entry;
+	struct rb_node *n;
+	int err = 0;
+	int werr = 0;
+	struct btrfs_path *path;
+	u64 last = 0;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	while (1) {
+		cache = NULL;
+		spin_lock(&root->fs_info->block_group_cache_lock);
+		for (n = rb_first(&root->fs_info->block_group_cache_tree);
+		     n; n = rb_next(n)) {
+			entry = rb_entry(n, struct btrfs_block_group_cache,
+					 cache_node);
+			if (entry->dirty) {
+				cache = entry;
+				break;
+			}
+		}
+		spin_unlock(&root->fs_info->block_group_cache_lock);
+
+		if (!cache)
+			break;
+
+		cache->dirty = 0;
+		last += cache->key.offset;
+
+		err = write_one_cache_group(trans, root,
+					    path, cache);
+		/*
+		 * if we fail to write the cache group, we want
+		 * to keep it marked dirty in hopes that a later
+		 * write will work
+		 */
+		if (err) {
+			werr = err;
+			continue;
+		}
+	}
+	btrfs_free_path(path);
+	return werr;
+}
+
+int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
+{
+	struct btrfs_block_group_cache *block_group;
+	int readonly = 0;
+
+	block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
+	if (!block_group || block_group->ro)
+		readonly = 1;
+	if (block_group)
+		put_block_group(block_group);
+	return readonly;
+}
+
+static int update_space_info(struct btrfs_fs_info *info, u64 flags,
+			     u64 total_bytes, u64 bytes_used,
+			     struct btrfs_space_info **space_info)
+{
+	struct btrfs_space_info *found;
+
+	found = __find_space_info(info, flags);
+	if (found) {
+		spin_lock(&found->lock);
+		found->total_bytes += total_bytes;
+		found->bytes_used += bytes_used;
+		found->full = 0;
+		spin_unlock(&found->lock);
+		*space_info = found;
+		return 0;
+	}
+	found = kzalloc(sizeof(*found), GFP_NOFS);
+	if (!found)
+		return -ENOMEM;
+
+	list_add(&found->list, &info->space_info);
+	INIT_LIST_HEAD(&found->block_groups);
+	init_rwsem(&found->groups_sem);
+	spin_lock_init(&found->lock);
+	found->flags = flags;
+	found->total_bytes = total_bytes;
+	found->bytes_used = bytes_used;
+	found->bytes_pinned = 0;
+	found->bytes_reserved = 0;
+	found->bytes_readonly = 0;
+	found->full = 0;
+	found->force_alloc = 0;
+	*space_info = found;
+	return 0;
+}
+
+static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
+{
+	u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
+				   BTRFS_BLOCK_GROUP_RAID1 |
+				   BTRFS_BLOCK_GROUP_RAID10 |
+				   BTRFS_BLOCK_GROUP_DUP);
+	if (extra_flags) {
+		if (flags & BTRFS_BLOCK_GROUP_DATA)
+			fs_info->avail_data_alloc_bits |= extra_flags;
+		if (flags & BTRFS_BLOCK_GROUP_METADATA)
+			fs_info->avail_metadata_alloc_bits |= extra_flags;
+		if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
+			fs_info->avail_system_alloc_bits |= extra_flags;
+	}
+}
+
+static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
+{
+	spin_lock(&cache->space_info->lock);
+	spin_lock(&cache->lock);
+	if (!cache->ro) {
+		cache->space_info->bytes_readonly += cache->key.offset -
+					btrfs_block_group_used(&cache->item);
+		cache->ro = 1;
+	}
+	spin_unlock(&cache->lock);
+	spin_unlock(&cache->space_info->lock);
+}
+
+u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
+{
+	u64 num_devices = root->fs_info->fs_devices->rw_devices;
+
+	if (num_devices == 1)
+		flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
+	if (num_devices < 4)
+		flags &= ~BTRFS_BLOCK_GROUP_RAID10;
+
+	if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
+	    (flags & (BTRFS_BLOCK_GROUP_RAID1 |
+		      BTRFS_BLOCK_GROUP_RAID10))) {
+		flags &= ~BTRFS_BLOCK_GROUP_DUP;
+	}
+
+	if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
+	    (flags & BTRFS_BLOCK_GROUP_RAID10)) {
+		flags &= ~BTRFS_BLOCK_GROUP_RAID1;
+	}
+
+	if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
+	    ((flags & BTRFS_BLOCK_GROUP_RAID1) |
+	     (flags & BTRFS_BLOCK_GROUP_RAID10) |
+	     (flags & BTRFS_BLOCK_GROUP_DUP)))
+		flags &= ~BTRFS_BLOCK_GROUP_RAID0;
+	return flags;
+}
+
+static int update_block_group(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      u64 bytenr, u64 num_bytes, int alloc,
+			      int mark_free)
+{
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *info = root->fs_info;
+	u64 total = num_bytes;
+	u64 old_val;
+	u64 byte_in_group;
+
+	while (total) {
+		cache = btrfs_lookup_block_group(info, bytenr);
+		if (!cache)
+			return -1;
+		byte_in_group = bytenr - cache->key.objectid;
+		WARN_ON(byte_in_group > cache->key.offset);
+
+		spin_lock(&cache->space_info->lock);
+		spin_lock(&cache->lock);
+		cache->dirty = 1;
+		old_val = btrfs_block_group_used(&cache->item);
+		num_bytes = min(total, cache->key.offset - byte_in_group);
+		if (alloc) {
+			old_val += num_bytes;
+			cache->space_info->bytes_used += num_bytes;
+			if (cache->ro)
+				cache->space_info->bytes_readonly -= num_bytes;
+			btrfs_set_block_group_used(&cache->item, old_val);
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+		} else {
+			old_val -= num_bytes;
+			cache->space_info->bytes_used -= num_bytes;
+			if (cache->ro)
+				cache->space_info->bytes_readonly += num_bytes;
+			btrfs_set_block_group_used(&cache->item, old_val);
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+			if (mark_free) {
+				int ret;
+
+				ret = btrfs_discard_extent(root, bytenr,
+							   num_bytes);
+				WARN_ON(ret);
+
+				ret = btrfs_add_free_space(cache, bytenr,
+							   num_bytes);
+				WARN_ON(ret);
+			}
+		}
+		put_block_group(cache);
+		total -= num_bytes;
+		bytenr += num_bytes;
+	}
+	return 0;
+}
+
+static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
+{
+	struct btrfs_block_group_cache *cache;
+	u64 bytenr;
+
+	cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
+	if (!cache)
+		return 0;
+
+	bytenr = cache->key.objectid;
+	put_block_group(cache);
+
+	return bytenr;
+}
+
+int btrfs_update_pinned_extents(struct btrfs_root *root,
+				u64 bytenr, u64 num, int pin)
+{
+	u64 len;
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+
+	WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
+	if (pin) {
+		set_extent_dirty(&fs_info->pinned_extents,
+				bytenr, bytenr + num - 1, GFP_NOFS);
+	} else {
+		clear_extent_dirty(&fs_info->pinned_extents,
+				bytenr, bytenr + num - 1, GFP_NOFS);
+	}
+	while (num > 0) {
+		cache = btrfs_lookup_block_group(fs_info, bytenr);
+		BUG_ON(!cache);
+		len = min(num, cache->key.offset -
+			  (bytenr - cache->key.objectid));
+		if (pin) {
+			spin_lock(&cache->space_info->lock);
+			spin_lock(&cache->lock);
+			cache->pinned += len;
+			cache->space_info->bytes_pinned += len;
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+			fs_info->total_pinned += len;
+		} else {
+			spin_lock(&cache->space_info->lock);
+			spin_lock(&cache->lock);
+			cache->pinned -= len;
+			cache->space_info->bytes_pinned -= len;
+			spin_unlock(&cache->lock);
+			spin_unlock(&cache->space_info->lock);
+			fs_info->total_pinned -= len;
+			if (cache->cached)
+				btrfs_add_free_space(cache, bytenr, len);
+		}
+		put_block_group(cache);
+		bytenr += len;
+		num -= len;
+	}
+	return 0;
+}
+
+static int update_reserved_extents(struct btrfs_root *root,
+				   u64 bytenr, u64 num, int reserve)
+{
+	u64 len;
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+
+	while (num > 0) {
+		cache = btrfs_lookup_block_group(fs_info, bytenr);
+		BUG_ON(!cache);
+		len = min(num, cache->key.offset -
+			  (bytenr - cache->key.objectid));
+
+		spin_lock(&cache->space_info->lock);
+		spin_lock(&cache->lock);
+		if (reserve) {
+			cache->reserved += len;
+			cache->space_info->bytes_reserved += len;
+		} else {
+			cache->reserved -= len;
+			cache->space_info->bytes_reserved -= len;
+		}
+		spin_unlock(&cache->lock);
+		spin_unlock(&cache->space_info->lock);
+		put_block_group(cache);
+		bytenr += len;
+		num -= len;
+	}
+	return 0;
+}
+
+int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
+{
+	u64 last = 0;
+	u64 start;
+	u64 end;
+	struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
+	int ret;
+
+	mutex_lock(&root->fs_info->pinned_mutex);
+	while (1) {
+		ret = find_first_extent_bit(pinned_extents, last,
+					    &start, &end, EXTENT_DIRTY);
+		if (ret)
+			break;
+		set_extent_dirty(copy, start, end, GFP_NOFS);
+		last = end + 1;
+	}
+	mutex_unlock(&root->fs_info->pinned_mutex);
+	return 0;
+}
+
+int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_io_tree *unpin)
+{
+	u64 start;
+	u64 end;
+	int ret;
+
+	mutex_lock(&root->fs_info->pinned_mutex);
+	while (1) {
+		ret = find_first_extent_bit(unpin, 0, &start, &end,
+					    EXTENT_DIRTY);
+		if (ret)
+			break;
+
+		ret = btrfs_discard_extent(root, start, end + 1 - start);
+
+		btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
+		clear_extent_dirty(unpin, start, end, GFP_NOFS);
+
+		if (need_resched()) {
+			mutex_unlock(&root->fs_info->pinned_mutex);
+			cond_resched();
+			mutex_lock(&root->fs_info->pinned_mutex);
+		}
+	}
+	mutex_unlock(&root->fs_info->pinned_mutex);
+	return ret;
+}
+
+static int finish_current_insert(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *extent_root, int all)
+{
+	u64 start;
+	u64 end;
+	u64 priv;
+	u64 search = 0;
+	u64 skipped = 0;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct btrfs_path *path;
+	struct pending_extent_op *extent_op, *tmp;
+	struct list_head insert_list, update_list;
+	int ret;
+	int num_inserts = 0, max_inserts;
+
+	path = btrfs_alloc_path();
+	INIT_LIST_HEAD(&insert_list);
+	INIT_LIST_HEAD(&update_list);
+
+	max_inserts = extent_root->leafsize /
+		(2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
+		 sizeof(struct btrfs_extent_ref) +
+		 sizeof(struct btrfs_extent_item));
+again:
+	mutex_lock(&info->extent_ins_mutex);
+	while (1) {
+		ret = find_first_extent_bit(&info->extent_ins, search, &start,
+					    &end, EXTENT_WRITEBACK);
+		if (ret) {
+			if (skipped && all && !num_inserts) {
+				skipped = 0;
+				search = 0;
+				continue;
+			}
+			mutex_unlock(&info->extent_ins_mutex);
+			break;
+		}
+
+		ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
+		if (!ret) {
+			skipped = 1;
+			search = end + 1;
+			if (need_resched()) {
+				mutex_unlock(&info->extent_ins_mutex);
+				cond_resched();
+				mutex_lock(&info->extent_ins_mutex);
+			}
+			continue;
+		}
+
+		ret = get_state_private(&info->extent_ins, start, &priv);
+		BUG_ON(ret);
+		extent_op = (struct pending_extent_op *)(unsigned long) priv;
+
+		if (extent_op->type == PENDING_EXTENT_INSERT) {
+			num_inserts++;
+			list_add_tail(&extent_op->list, &insert_list);
+			search = end + 1;
+			if (num_inserts == max_inserts) {
+				mutex_unlock(&info->extent_ins_mutex);
+				break;
+			}
+		} else if (extent_op->type == PENDING_BACKREF_UPDATE) {
+			list_add_tail(&extent_op->list, &update_list);
+			search = end + 1;
+		} else {
+			BUG();
+		}
+	}
+
+	/*
+	 * process the update list, clear the writeback bit for it, and if
+	 * somebody marked this thing for deletion then just unlock it and be
+	 * done, the free_extents will handle it
+	 */
+	mutex_lock(&info->extent_ins_mutex);
+	list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
+		clear_extent_bits(&info->extent_ins, extent_op->bytenr,
+				  extent_op->bytenr + extent_op->num_bytes - 1,
+				  EXTENT_WRITEBACK, GFP_NOFS);
+		if (extent_op->del) {
+			list_del_init(&extent_op->list);
+			unlock_extent(&info->extent_ins, extent_op->bytenr,
+				      extent_op->bytenr + extent_op->num_bytes
+				      - 1, GFP_NOFS);
+			kfree(extent_op);
+		}
+	}
+	mutex_unlock(&info->extent_ins_mutex);
+
+	/*
+	 * still have things left on the update list, go ahead an update
+	 * everything
+	 */
+	if (!list_empty(&update_list)) {
+		ret = update_backrefs(trans, extent_root, path, &update_list);
+		BUG_ON(ret);
+	}
+
+	/*
+	 * if no inserts need to be done, but we skipped some extents and we
+	 * need to make sure everything is cleaned then reset everything and
+	 * go back to the beginning
+	 */
+	if (!num_inserts && all && skipped) {
+		search = 0;
+		skipped = 0;
+		INIT_LIST_HEAD(&update_list);
+		INIT_LIST_HEAD(&insert_list);
+		goto again;
+	} else if (!num_inserts) {
+		goto out;
+	}
+
+	/*
+	 * process the insert extents list.  Again if we are deleting this
+	 * extent, then just unlock it, pin down the bytes if need be, and be
+	 * done with it.  Saves us from having to actually insert the extent
+	 * into the tree and then subsequently come along and delete it
+	 */
+	mutex_lock(&info->extent_ins_mutex);
+	list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
+		clear_extent_bits(&info->extent_ins, extent_op->bytenr,
+				  extent_op->bytenr + extent_op->num_bytes - 1,
+				  EXTENT_WRITEBACK, GFP_NOFS);
+		if (extent_op->del) {
+			u64 used;
+			list_del_init(&extent_op->list);
+			unlock_extent(&info->extent_ins, extent_op->bytenr,
+				      extent_op->bytenr + extent_op->num_bytes
+				      - 1, GFP_NOFS);
+
+			mutex_lock(&extent_root->fs_info->pinned_mutex);
+			ret = pin_down_bytes(trans, extent_root,
+					     extent_op->bytenr,
+					     extent_op->num_bytes, 0);
+			mutex_unlock(&extent_root->fs_info->pinned_mutex);
+
+			spin_lock(&info->delalloc_lock);
+			used = btrfs_super_bytes_used(&info->super_copy);
+			btrfs_set_super_bytes_used(&info->super_copy,
+					used - extent_op->num_bytes);
+			used = btrfs_root_used(&extent_root->root_item);
+			btrfs_set_root_used(&extent_root->root_item,
+					used - extent_op->num_bytes);
+			spin_unlock(&info->delalloc_lock);
+
+			ret = update_block_group(trans, extent_root,
+						 extent_op->bytenr,
+						 extent_op->num_bytes,
+						 0, ret > 0);
+			BUG_ON(ret);
+			kfree(extent_op);
+			num_inserts--;
+		}
+	}
+	mutex_unlock(&info->extent_ins_mutex);
+
+	ret = insert_extents(trans, extent_root, path, &insert_list,
+			     num_inserts);
+	BUG_ON(ret);
+
+	/*
+	 * if we broke out of the loop in order to insert stuff because we hit
+	 * the maximum number of inserts at a time we can handle, then loop
+	 * back and pick up where we left off
+	 */
+	if (num_inserts == max_inserts) {
+		INIT_LIST_HEAD(&insert_list);
+		INIT_LIST_HEAD(&update_list);
+		num_inserts = 0;
+		goto again;
+	}
+
+	/*
+	 * again, if we need to make absolutely sure there are no more pending
+	 * extent operations left and we know that we skipped some, go back to
+	 * the beginning and do it all again
+	 */
+	if (all && skipped) {
+		INIT_LIST_HEAD(&insert_list);
+		INIT_LIST_HEAD(&update_list);
+		search = 0;
+		skipped = 0;
+		num_inserts = 0;
+		goto again;
+	}
+out:
+	btrfs_free_path(path);
+	return 0;
+}
+
+static int pin_down_bytes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  u64 bytenr, u64 num_bytes, int is_data)
+{
+	int err = 0;
+	struct extent_buffer *buf;
+
+	if (is_data)
+		goto pinit;
+
+	buf = btrfs_find_tree_block(root, bytenr, num_bytes);
+	if (!buf)
+		goto pinit;
+
+	/* we can reuse a block if it hasn't been written
+	 * and it is from this transaction.  We can't
+	 * reuse anything from the tree log root because
+	 * it has tiny sub-transactions.
+	 */
+	if (btrfs_buffer_uptodate(buf, 0) &&
+	    btrfs_try_tree_lock(buf)) {
+		u64 header_owner = btrfs_header_owner(buf);
+		u64 header_transid = btrfs_header_generation(buf);
+		if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
+		    header_owner != BTRFS_TREE_RELOC_OBJECTID &&
+		    header_transid == trans->transid &&
+		    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
+			clean_tree_block(NULL, root, buf);
+			btrfs_tree_unlock(buf);
+			free_extent_buffer(buf);
+			return 1;
+		}
+		btrfs_tree_unlock(buf);
+	}
+	free_extent_buffer(buf);
+pinit:
+	btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
+
+	BUG_ON(err < 0);
+	return 0;
+}
+
+/*
+ * remove an extent from the root, returns 0 on success
+ */
+static int __free_extent(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 bytenr, u64 num_bytes, u64 parent,
+			 u64 root_objectid, u64 ref_generation,
+			 u64 owner_objectid, int pin, int mark_free)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_root *extent_root = info->extent_root;
+	struct extent_buffer *leaf;
+	int ret;
+	int extent_slot = 0;
+	int found_extent = 0;
+	int num_to_del = 1;
+	struct btrfs_extent_item *ei;
+	u32 refs;
+
+	key.objectid = bytenr;
+	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	key.offset = num_bytes;
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->reada = 1;
+	ret = lookup_extent_backref(trans, extent_root, path,
+				    bytenr, parent, root_objectid,
+				    ref_generation, owner_objectid, 1);
+	if (ret == 0) {
+		struct btrfs_key found_key;
+		extent_slot = path->slots[0];
+		while (extent_slot > 0) {
+			extent_slot--;
+			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
+					      extent_slot);
+			if (found_key.objectid != bytenr)
+				break;
+			if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
+			    found_key.offset == num_bytes) {
+				found_extent = 1;
+				break;
+			}
+			if (path->slots[0] - extent_slot > 5)
+				break;
+		}
+		if (!found_extent) {
+			ret = remove_extent_backref(trans, extent_root, path);
+			BUG_ON(ret);
+			btrfs_release_path(extent_root, path);
+			ret = btrfs_search_slot(trans, extent_root,
+						&key, path, -1, 1);
+			if (ret) {
+				printk(KERN_ERR "umm, got %d back from search"
+				       ", was looking for %llu\n", ret,
+				       (unsigned long long)bytenr);
+				btrfs_print_leaf(extent_root, path->nodes[0]);
+			}
+			BUG_ON(ret);
+			extent_slot = path->slots[0];
+		}
+	} else {
+		btrfs_print_leaf(extent_root, path->nodes[0]);
+		WARN_ON(1);
+		printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
+		       "root %llu gen %llu owner %llu\n",
+		       (unsigned long long)bytenr,
+		       (unsigned long long)root_objectid,
+		       (unsigned long long)ref_generation,
+		       (unsigned long long)owner_objectid);
+	}
+
+	leaf = path->nodes[0];
+	ei = btrfs_item_ptr(leaf, extent_slot,
+			    struct btrfs_extent_item);
+	refs = btrfs_extent_refs(leaf, ei);
+	BUG_ON(refs == 0);
+	refs -= 1;
+	btrfs_set_extent_refs(leaf, ei, refs);
+
+	btrfs_mark_buffer_dirty(leaf);
+
+	if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
+		struct btrfs_extent_ref *ref;
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_extent_ref);
+		BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
+		/* if the back ref and the extent are next to each other
+		 * they get deleted below in one shot
+		 */
+		path->slots[0] = extent_slot;
+		num_to_del = 2;
+	} else if (found_extent) {
+		/* otherwise delete the extent back ref */
+		ret = remove_extent_backref(trans, extent_root, path);
+		BUG_ON(ret);
+		/* if refs are 0, we need to setup the path for deletion */
+		if (refs == 0) {
+			btrfs_release_path(extent_root, path);
+			ret = btrfs_search_slot(trans, extent_root, &key, path,
+						-1, 1);
+			BUG_ON(ret);
+		}
+	}
+
+	if (refs == 0) {
+		u64 super_used;
+		u64 root_used;
+
+		if (pin) {
+			mutex_lock(&root->fs_info->pinned_mutex);
+			ret = pin_down_bytes(trans, root, bytenr, num_bytes,
+				owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
+			mutex_unlock(&root->fs_info->pinned_mutex);
+			if (ret > 0)
+				mark_free = 1;
+			BUG_ON(ret < 0);
+		}
+		/* block accounting for super block */
+		spin_lock(&info->delalloc_lock);
+		super_used = btrfs_super_bytes_used(&info->super_copy);
+		btrfs_set_super_bytes_used(&info->super_copy,
+					   super_used - num_bytes);
+
+		/* block accounting for root item */
+		root_used = btrfs_root_used(&root->root_item);
+		btrfs_set_root_used(&root->root_item,
+					   root_used - num_bytes);
+		spin_unlock(&info->delalloc_lock);
+		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
+				      num_to_del);
+		BUG_ON(ret);
+		btrfs_release_path(extent_root, path);
+
+		if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
+			BUG_ON(ret);
+		}
+
+		ret = update_block_group(trans, root, bytenr, num_bytes, 0,
+					 mark_free);
+		BUG_ON(ret);
+	}
+	btrfs_free_path(path);
+	finish_current_insert(trans, extent_root, 0);
+	return ret;
+}
+
+/*
+ * find all the blocks marked as pending in the radix tree and remove
+ * them from the extent map
+ */
+static int del_pending_extents(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root, int all)
+{
+	int ret;
+	int err = 0;
+	u64 start;
+	u64 end;
+	u64 priv;
+	u64 search = 0;
+	int nr = 0, skipped = 0;
+	struct extent_io_tree *pending_del;
+	struct extent_io_tree *extent_ins;
+	struct pending_extent_op *extent_op;
+	struct btrfs_fs_info *info = extent_root->fs_info;
+	struct list_head delete_list;
+
+	INIT_LIST_HEAD(&delete_list);
+	extent_ins = &extent_root->fs_info->extent_ins;
+	pending_del = &extent_root->fs_info->pending_del;
+
+again:
+	mutex_lock(&info->extent_ins_mutex);
+	while (1) {
+		ret = find_first_extent_bit(pending_del, search, &start, &end,
+					    EXTENT_WRITEBACK);
+		if (ret) {
+			if (all && skipped && !nr) {
+				search = 0;
+				continue;
+			}
+			mutex_unlock(&info->extent_ins_mutex);
+			break;
+		}
+
+		ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
+		if (!ret) {
+			search = end+1;
+			skipped = 1;
+
+			if (need_resched()) {
+				mutex_unlock(&info->extent_ins_mutex);
+				cond_resched();
+				mutex_lock(&info->extent_ins_mutex);
+			}
+
+			continue;
+		}
+		BUG_ON(ret < 0);
+
+		ret = get_state_private(pending_del, start, &priv);
+		BUG_ON(ret);
+		extent_op = (struct pending_extent_op *)(unsigned long)priv;
+
+		clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
+				  GFP_NOFS);
+		if (!test_range_bit(extent_ins, start, end,
+				    EXTENT_WRITEBACK, 0)) {
+			list_add_tail(&extent_op->list, &delete_list);
+			nr++;
+		} else {
+			kfree(extent_op);
+
+			ret = get_state_private(&info->extent_ins, start,
+						&priv);
+			BUG_ON(ret);
+			extent_op = (struct pending_extent_op *)
+						(unsigned long)priv;
+
+			clear_extent_bits(&info->extent_ins, start, end,
+					  EXTENT_WRITEBACK, GFP_NOFS);
+
+			if (extent_op->type == PENDING_BACKREF_UPDATE) {
+				list_add_tail(&extent_op->list, &delete_list);
+				search = end + 1;
+				nr++;
+				continue;
+			}
+
+			mutex_lock(&extent_root->fs_info->pinned_mutex);
+			ret = pin_down_bytes(trans, extent_root, start,
+					     end + 1 - start, 0);
+			mutex_unlock(&extent_root->fs_info->pinned_mutex);
+
+			ret = update_block_group(trans, extent_root, start,
+						end + 1 - start, 0, ret > 0);
+
+			unlock_extent(extent_ins, start, end, GFP_NOFS);
+			BUG_ON(ret);
+			kfree(extent_op);
+		}
+		if (ret)
+			err = ret;
+
+		search = end + 1;
+
+		if (need_resched()) {
+			mutex_unlock(&info->extent_ins_mutex);
+			cond_resched();
+			mutex_lock(&info->extent_ins_mutex);
+		}
+	}
+
+	if (nr) {
+		ret = free_extents(trans, extent_root, &delete_list);
+		BUG_ON(ret);
+	}
+
+	if (all && skipped) {
+		INIT_LIST_HEAD(&delete_list);
+		search = 0;
+		nr = 0;
+		goto again;
+	}
+
+	return err;
+}
+
+/*
+ * remove an extent from the root, returns 0 on success
+ */
+static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       u64 bytenr, u64 num_bytes, u64 parent,
+			       u64 root_objectid, u64 ref_generation,
+			       u64 owner_objectid, int pin)
+{
+	struct btrfs_root *extent_root = root->fs_info->extent_root;
+	int pending_ret;
+	int ret;
+
+	WARN_ON(num_bytes < root->sectorsize);
+	if (root == extent_root) {
+		struct pending_extent_op *extent_op = NULL;
+
+		mutex_lock(&root->fs_info->extent_ins_mutex);
+		if (test_range_bit(&root->fs_info->extent_ins, bytenr,
+				bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
+			u64 priv;
+			ret = get_state_private(&root->fs_info->extent_ins,
+						bytenr, &priv);
+			BUG_ON(ret);
+			extent_op = (struct pending_extent_op *)
+						(unsigned long)priv;
+
+			extent_op->del = 1;
+			if (extent_op->type == PENDING_EXTENT_INSERT) {
+				mutex_unlock(&root->fs_info->extent_ins_mutex);
+				return 0;
+			}
+		}
+
+		if (extent_op) {
+			ref_generation = extent_op->orig_generation;
+			parent = extent_op->orig_parent;
+		}
+
+		extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
+		BUG_ON(!extent_op);
+
+		extent_op->type = PENDING_EXTENT_DELETE;
+		extent_op->bytenr = bytenr;
+		extent_op->num_bytes = num_bytes;
+		extent_op->parent = parent;
+		extent_op->orig_parent = parent;
+		extent_op->generation = ref_generation;
+		extent_op->orig_generation = ref_generation;
+		extent_op->level = (int)owner_objectid;
+		INIT_LIST_HEAD(&extent_op->list);
+		extent_op->del = 0;
+
+		set_extent_bits(&root->fs_info->pending_del,
+				bytenr, bytenr + num_bytes - 1,
+				EXTENT_WRITEBACK, GFP_NOFS);
+		set_state_private(&root->fs_info->pending_del,
+				  bytenr, (unsigned long)extent_op);
+		mutex_unlock(&root->fs_info->extent_ins_mutex);
+		return 0;
+	}
+	/* if metadata always pin */
+	if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+		if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
+			struct btrfs_block_group_cache *cache;
+
+			/* btrfs_free_reserved_extent */
+			cache = btrfs_lookup_block_group(root->fs_info, bytenr);
+			BUG_ON(!cache);
+			btrfs_add_free_space(cache, bytenr, num_bytes);
+			put_block_group(cache);
+			update_reserved_extents(root, bytenr, num_bytes, 0);
+			return 0;
+		}
+		pin = 1;
+	}
+
+	/* if data pin when any transaction has committed this */
+	if (ref_generation != trans->transid)
+		pin = 1;
+
+	ret = __free_extent(trans, root, bytenr, num_bytes, parent,
+			    root_objectid, ref_generation,
+			    owner_objectid, pin, pin == 0);
+
+	finish_current_insert(trans, root->fs_info->extent_root, 0);
+	pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
+	return ret ? ret : pending_ret;
+}
+
+int btrfs_free_extent(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root,
+		      u64 bytenr, u64 num_bytes, u64 parent,
+		      u64 root_objectid, u64 ref_generation,
+		      u64 owner_objectid, int pin)
+{
+	int ret;
+
+	ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
+				  root_objectid, ref_generation,
+				  owner_objectid, pin);
+	return ret;
+}
+
+static u64 stripe_align(struct btrfs_root *root, u64 val)
+{
+	u64 mask = ((u64)root->stripesize - 1);
+	u64 ret = (val + mask) & ~mask;
+	return ret;
+}
+
+/*
+ * walks the btree of allocated extents and find a hole of a given size.
+ * The key ins is changed to record the hole:
+ * ins->objectid == block start
+ * ins->flags = BTRFS_EXTENT_ITEM_KEY
+ * ins->offset == number of blocks
+ * Any available blocks before search_start are skipped.
+ */
+static noinline int find_free_extent(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *orig_root,
+				     u64 num_bytes, u64 empty_size,
+				     u64 search_start, u64 search_end,
+				     u64 hint_byte, struct btrfs_key *ins,
+				     u64 exclude_start, u64 exclude_nr,
+				     int data)
+{
+	int ret = 0;
+	struct btrfs_root *root = orig_root->fs_info->extent_root;
+	u64 total_needed = num_bytes;
+	u64 *last_ptr = NULL;
+	u64 last_wanted = 0;
+	struct btrfs_block_group_cache *block_group = NULL;
+	int chunk_alloc_done = 0;
+	int empty_cluster = 2 * 1024 * 1024;
+	int allowed_chunk_alloc = 0;
+	struct list_head *head = NULL, *cur = NULL;
+	int loop = 0;
+	int extra_loop = 0;
+	struct btrfs_space_info *space_info;
+
+	WARN_ON(num_bytes < root->sectorsize);
+	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
+	ins->objectid = 0;
+	ins->offset = 0;
+
+	if (orig_root->ref_cows || empty_size)
+		allowed_chunk_alloc = 1;
+
+	if (data & BTRFS_BLOCK_GROUP_METADATA) {
+		last_ptr = &root->fs_info->last_alloc;
+		empty_cluster = 64 * 1024;
+	}
+
+	if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
+		last_ptr = &root->fs_info->last_data_alloc;
+
+	if (last_ptr) {
+		if (*last_ptr) {
+			hint_byte = *last_ptr;
+			last_wanted = *last_ptr;
+		} else
+			empty_size += empty_cluster;
+	} else {
+		empty_cluster = 0;
+	}
+	search_start = max(search_start, first_logical_byte(root, 0));
+	search_start = max(search_start, hint_byte);
+
+	if (last_wanted && search_start != last_wanted) {
+		last_wanted = 0;
+		empty_size += empty_cluster;
+	}
+
+	total_needed += empty_size;
+	block_group = btrfs_lookup_block_group(root->fs_info, search_start);
+	if (!block_group)
+		block_group = btrfs_lookup_first_block_group(root->fs_info,
+							     search_start);
+	space_info = __find_space_info(root->fs_info, data);
+
+	down_read(&space_info->groups_sem);
+	while (1) {
+		struct btrfs_free_space *free_space;
+		/*
+		 * the only way this happens if our hint points to a block
+		 * group thats not of the proper type, while looping this
+		 * should never happen
+		 */
+		if (empty_size)
+			extra_loop = 1;
+
+		if (!block_group)
+			goto new_group_no_lock;
+
+		if (unlikely(!block_group->cached)) {
+			mutex_lock(&block_group->cache_mutex);
+			ret = cache_block_group(root, block_group);
+			mutex_unlock(&block_group->cache_mutex);
+			if (ret)
+				break;
+		}
+
+		mutex_lock(&block_group->alloc_mutex);
+		if (unlikely(!block_group_bits(block_group, data)))
+			goto new_group;
+
+		if (unlikely(block_group->ro))
+			goto new_group;
+
+		free_space = btrfs_find_free_space(block_group, search_start,
+						   total_needed);
+		if (free_space) {
+			u64 start = block_group->key.objectid;
+			u64 end = block_group->key.objectid +
+				block_group->key.offset;
+
+			search_start = stripe_align(root, free_space->offset);
+
+			/* move on to the next group */
+			if (search_start + num_bytes >= search_end)
+				goto new_group;
+
+			/* move on to the next group */
+			if (search_start + num_bytes > end)
+				goto new_group;
+
+			if (last_wanted && search_start != last_wanted) {
+				total_needed += empty_cluster;
+				empty_size += empty_cluster;
+				last_wanted = 0;
+				/*
+				 * if search_start is still in this block group
+				 * then we just re-search this block group
+				 */
+				if (search_start >= start &&
+				    search_start < end) {
+					mutex_unlock(&block_group->alloc_mutex);
+					continue;
+				}
+
+				/* else we go to the next block group */
+				goto new_group;
+			}
+
+			if (exclude_nr > 0 &&
+			    (search_start + num_bytes > exclude_start &&
+			     search_start < exclude_start + exclude_nr)) {
+				search_start = exclude_start + exclude_nr;
+				/*
+				 * if search_start is still in this block group
+				 * then we just re-search this block group
+				 */
+				if (search_start >= start &&
+				    search_start < end) {
+					mutex_unlock(&block_group->alloc_mutex);
+					last_wanted = 0;
+					continue;
+				}
+
+				/* else we go to the next block group */
+				goto new_group;
+			}
+
+			ins->objectid = search_start;
+			ins->offset = num_bytes;
+
+			btrfs_remove_free_space_lock(block_group, search_start,
+						     num_bytes);
+			/* we are all good, lets return */
+			mutex_unlock(&block_group->alloc_mutex);
+			break;
+		}
+new_group:
+		mutex_unlock(&block_group->alloc_mutex);
+		put_block_group(block_group);
+		block_group = NULL;
+new_group_no_lock:
+		/* don't try to compare new allocations against the
+		 * last allocation any more
+		 */
+		last_wanted = 0;
+
+		/*
+		 * Here's how this works.
+		 * loop == 0: we were searching a block group via a hint
+		 *		and didn't find anything, so we start at
+		 *		the head of the block groups and keep searching
+		 * loop == 1: we're searching through all of the block groups
+		 *		if we hit the head again we have searched
+		 *		all of the block groups for this space and we
+		 *		need to try and allocate, if we cant error out.
+		 * loop == 2: we allocated more space and are looping through
+		 *		all of the block groups again.
+		 */
+		if (loop == 0) {
+			head = &space_info->block_groups;
+			cur = head->next;
+			loop++;
+		} else if (loop == 1 && cur == head) {
+			int keep_going;
+
+			/* at this point we give up on the empty_size
+			 * allocations and just try to allocate the min
+			 * space.
+			 *
+			 * The extra_loop field was set if an empty_size
+			 * allocation was attempted above, and if this
+			 * is try we need to try the loop again without
+			 * the additional empty_size.
+			 */
+			total_needed -= empty_size;
+			empty_size = 0;
+			keep_going = extra_loop;
+			loop++;
+
+			if (allowed_chunk_alloc && !chunk_alloc_done) {
+				up_read(&space_info->groups_sem);
+				ret = do_chunk_alloc(trans, root, num_bytes +
+						     2 * 1024 * 1024, data, 1);
+				down_read(&space_info->groups_sem);
+				if (ret < 0)
+					goto loop_check;
+				head = &space_info->block_groups;
+				/*
+				 * we've allocated a new chunk, keep
+				 * trying
+				 */
+				keep_going = 1;
+				chunk_alloc_done = 1;
+			} else if (!allowed_chunk_alloc) {
+				space_info->force_alloc = 1;
+			}
+loop_check:
+			if (keep_going) {
+				cur = head->next;
+				extra_loop = 0;
+			} else {
+				break;
+			}
+		} else if (cur == head) {
+			break;
+		}
+
+		block_group = list_entry(cur, struct btrfs_block_group_cache,
+					 list);
+		atomic_inc(&block_group->count);
+
+		search_start = block_group->key.objectid;
+		cur = cur->next;
+	}
+
+	/* we found what we needed */
+	if (ins->objectid) {
+		if (!(data & BTRFS_BLOCK_GROUP_DATA))
+			trans->block_group = block_group->key.objectid;
+
+		if (last_ptr)
+			*last_ptr = ins->objectid + ins->offset;
+		ret = 0;
+	} else if (!ret) {
+		printk(KERN_ERR "btrfs searching for %llu bytes, "
+		       "num_bytes %llu, loop %d, allowed_alloc %d\n",
+		       (unsigned long long)total_needed,
+		       (unsigned long long)num_bytes,
+		       loop, allowed_chunk_alloc);
+		ret = -ENOSPC;
+	}
+	if (block_group)
+		put_block_group(block_group);
+
+	up_read(&space_info->groups_sem);
+	return ret;
+}
+
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
+{
+	struct btrfs_block_group_cache *cache;
+	struct list_head *l;
+
+	printk(KERN_INFO "space_info has %llu free, is %sfull\n",
+	       (unsigned long long)(info->total_bytes - info->bytes_used -
+				    info->bytes_pinned - info->bytes_reserved),
+	       (info->full) ? "" : "not ");
+
+	down_read(&info->groups_sem);
+	list_for_each(l, &info->block_groups) {
+		cache = list_entry(l, struct btrfs_block_group_cache, list);
+		spin_lock(&cache->lock);
+		printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
+		       "%llu pinned %llu reserved\n",
+		       (unsigned long long)cache->key.objectid,
+		       (unsigned long long)cache->key.offset,
+		       (unsigned long long)btrfs_block_group_used(&cache->item),
+		       (unsigned long long)cache->pinned,
+		       (unsigned long long)cache->reserved);
+		btrfs_dump_free_space(cache, bytes);
+		spin_unlock(&cache->lock);
+	}
+	up_read(&info->groups_sem);
+}
+
+static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  u64 num_bytes, u64 min_alloc_size,
+				  u64 empty_size, u64 hint_byte,
+				  u64 search_end, struct btrfs_key *ins,
+				  u64 data)
+{
+	int ret;
[… more lines in the raw message]

Lines of this file are not shown. View the raw message.

The rest of this diff is not shown. View the raw message.

Part of the diff is not shown here. View the raw message.

#50 lore raw

PATCH 1/1 Btrfs snapshot deletion

In reply to Chris Mason

Snapshot deletion is done by walking down the btree and dropping references as you go. For a given node or leaf, if dropping the reference count will free the block, all of the extents that block points to will also have their reference counts dropped.

If dropping the reference count will not free a block, the entire subtree pointed to by that block is skipped.

Since most of the metadata lives in leaves, a reference cache is maintained in ram of the extents a given leaf has references on. This limits the number of btree leaves that need to be read while deleting old transactions at run time.

Snapshot deletion happens incrementally. The metadata for each root indicates how far the last delete run got into the tree. This allows smaller atomic units for snapshot deletion, and it doesn't stall transaction commit with long running operations. It also means that once snapshot deletion starts, that tree must not be accessed for reading or writing outside of the snapshot deletion code.

fs/btrfs/extent-tree.c +495 -0
@@ -3334,3 +3334,498 @@
 	buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
 	return buf;
 }
+
+int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root, struct extent_buffer *leaf)
+{
+	u64 leaf_owner;
+	u64 leaf_generation;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int i;
+	int nritems;
+	int ret;
+
+	BUG_ON(!btrfs_is_leaf(leaf));
+	nritems = btrfs_header_nritems(leaf);
+	leaf_owner = btrfs_header_owner(leaf);
+	leaf_generation = btrfs_header_generation(leaf);
+
+	for (i = 0; i < nritems; i++) {
+		u64 disk_bytenr;
+		cond_resched();
+
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		/*
+		 * FIXME make sure to insert a trans record that
+		 * repeats the snapshot del on crash
+		 */
+		disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+		if (disk_bytenr == 0)
+			continue;
+
+		ret = __btrfs_free_extent(trans, root, disk_bytenr,
+				btrfs_file_extent_disk_num_bytes(leaf, fi),
+				leaf->start, leaf_owner, leaf_generation,
+				key.objectid, 0);
+		BUG_ON(ret);
+
+		atomic_inc(&root->fs_info->throttle_gen);
+		wake_up(&root->fs_info->transaction_throttle);
+		cond_resched();
+	}
+	return 0;
+}
+
+static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_leaf_ref *ref)
+{
+	int i;
+	int ret;
+	struct btrfs_extent_info *info = ref->extents;
+
+	for (i = 0; i < ref->nritems; i++) {
+		ret = __btrfs_free_extent(trans, root, info->bytenr,
+					  info->num_bytes, ref->bytenr,
+					  ref->owner, ref->generation,
+					  info->objectid, 0);
+
+		atomic_inc(&root->fs_info->throttle_gen);
+		wake_up(&root->fs_info->transaction_throttle);
+		cond_resched();
+
+		BUG_ON(ret);
+		info++;
+	}
+
+	return 0;
+}
+
+static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
+				     u64 len, u32 *refs)
+{
+	int ret;
+
+	ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
+	BUG_ON(ret);
+
+#if 0 /* some debugging code in case we see problems here */
+	/* if the refs count is one, it won't get increased again.  But
+	 * if the ref count is > 1, someone may be decreasing it at
+	 * the same time we are.
+	 */
+	if (*refs != 1) {
+		struct extent_buffer *eb = NULL;
+		eb = btrfs_find_create_tree_block(root, start, len);
+		if (eb)
+			btrfs_tree_lock(eb);
+
+		mutex_lock(&root->fs_info->alloc_mutex);
+		ret = lookup_extent_ref(NULL, root, start, len, refs);
+		BUG_ON(ret);
+		mutex_unlock(&root->fs_info->alloc_mutex);
+
+		if (eb) {
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+		}
+		if (*refs == 1) {
+			printk(KERN_ERR "btrfs block %llu went down to one "
+			       "during drop_snap\n", (unsigned long long)start);
+		}
+
+	}
+#endif
+
+	cond_resched();
+	return ret;
+}
+
+/*
+ * helper function for drop_snapshot, this walks down the tree dropping ref
+ * counts as it goes.
+ */
+static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
+				   struct btrfs_root *root,
+				   struct btrfs_path *path, int *level)
+{
+	u64 root_owner;
+	u64 root_gen;
+	u64 bytenr;
+	u64 ptr_gen;
+	struct extent_buffer *next;
+	struct extent_buffer *cur;
+	struct extent_buffer *parent;
+	struct btrfs_leaf_ref *ref;
+	u32 blocksize;
+	int ret;
+	u32 refs;
+
+	WARN_ON(*level < 0);
+	WARN_ON(*level >= BTRFS_MAX_LEVEL);
+	ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
+				path->nodes[*level]->len, &refs);
+	BUG_ON(ret);
+	if (refs > 1)
+		goto out;
+
+	/*
+	 * walk down to the last node level and free all the leaves
+	 */
+	while (*level >= 0) {
+		WARN_ON(*level < 0);
+		WARN_ON(*level >= BTRFS_MAX_LEVEL);
+		cur = path->nodes[*level];
+
+		if (btrfs_header_level(cur) != *level)
+			WARN_ON(1);
+
+		if (path->slots[*level] >=
+		    btrfs_header_nritems(cur))
+			break;
+		if (*level == 0) {
+			ret = btrfs_drop_leaf_ref(trans, root, cur);
+			BUG_ON(ret);
+			break;
+		}
+		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
+		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+		blocksize = btrfs_level_size(root, *level - 1);
+
+		ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
+		BUG_ON(ret);
+		if (refs != 1) {
+			parent = path->nodes[*level];
+			root_owner = btrfs_header_owner(parent);
+			root_gen = btrfs_header_generation(parent);
+			path->slots[*level]++;
+
+			ret = __btrfs_free_extent(trans, root, bytenr,
+						blocksize, parent->start,
+						root_owner, root_gen,
+						*level - 1, 1);
+			BUG_ON(ret);
+
+			atomic_inc(&root->fs_info->throttle_gen);
+			wake_up(&root->fs_info->transaction_throttle);
+			cond_resched();
+
+			continue;
+		}
+		/*
+		 * at this point, we have a single ref, and since the
+		 * only place referencing this extent is a dead root
+		 * the reference count should never go higher.
+		 * So, we don't need to check it again
+		 */
+		if (*level == 1) {
+			ref = btrfs_lookup_leaf_ref(root, bytenr);
+			if (ref && ref->generation != ptr_gen) {
+				btrfs_free_leaf_ref(root, ref);
+				ref = NULL;
+			}
+			if (ref) {
+				ret = cache_drop_leaf_ref(trans, root, ref);
+				BUG_ON(ret);
+				btrfs_remove_leaf_ref(root, ref);
+				btrfs_free_leaf_ref(root, ref);
+				*level = 0;
+				break;
+			}
+		}
+		next = btrfs_find_tree_block(root, bytenr, blocksize);
+		if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
+			free_extent_buffer(next);
+
+			next = read_tree_block(root, bytenr, blocksize,
+					       ptr_gen);
+			cond_resched();
+#if 0
+			/*
+			 * this is a debugging check and can go away
+			 * the ref should never go all the way down to 1
+			 * at this point
+			 */
+			ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
+						&refs);
+			BUG_ON(ret);
+			WARN_ON(refs != 1);
+#endif
+		}
+		WARN_ON(*level <= 0);
+		if (path->nodes[*level-1])
+			free_extent_buffer(path->nodes[*level-1]);
+		path->nodes[*level-1] = next;
+		*level = btrfs_header_level(next);
+		path->slots[*level] = 0;
+		cond_resched();
+	}
+out:
+	WARN_ON(*level < 0);
+	WARN_ON(*level >= BTRFS_MAX_LEVEL);
+
+	if (path->nodes[*level] == root->node) {
+		parent = path->nodes[*level];
+		bytenr = path->nodes[*level]->start;
+	} else {
+		parent = path->nodes[*level + 1];
+		bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
+	}
+
+	blocksize = btrfs_level_size(root, *level);
+	root_owner = btrfs_header_owner(parent);
+	root_gen = btrfs_header_generation(parent);
+
+	ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
+				  parent->start, root_owner, root_gen,
+				  *level, 1);
+	free_extent_buffer(path->nodes[*level]);
+	path->nodes[*level] = NULL;
+	*level += 1;
+	BUG_ON(ret);
+
+	cond_resched();
+	return 0;
+}
+
+/*
+ * helper function for drop_subtree, this function is similar to
+ * walk_down_tree. The main difference is that it checks reference
+ * counts while tree blocks are locked.
+ */
+static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path, int *level)
+{
+	struct extent_buffer *next;
+	struct extent_buffer *cur;
+	struct extent_buffer *parent;
+	u64 bytenr;
+	u64 ptr_gen;
+	u32 blocksize;
+	u32 refs;
+	int ret;
+
+	cur = path->nodes[*level];
+	ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
+				      &refs);
+	BUG_ON(ret);
+	if (refs > 1)
+		goto out;
+
+	while (*level >= 0) {
+		cur = path->nodes[*level];
+		if (*level == 0) {
+			ret = btrfs_drop_leaf_ref(trans, root, cur);
+			BUG_ON(ret);
+			clean_tree_block(trans, root, cur);
+			break;
+		}
+		if (path->slots[*level] >= btrfs_header_nritems(cur)) {
+			clean_tree_block(trans, root, cur);
+			break;
+		}
+
+		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
+		blocksize = btrfs_level_size(root, *level - 1);
+		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+
+		next = read_tree_block(root, bytenr, blocksize, ptr_gen);
+		btrfs_tree_lock(next);
+
+		ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
+					      &refs);
+		BUG_ON(ret);
+		if (refs > 1) {
+			parent = path->nodes[*level];
+			ret = btrfs_free_extent(trans, root, bytenr,
+					blocksize, parent->start,
+					btrfs_header_owner(parent),
+					btrfs_header_generation(parent),
+					*level - 1, 1);
+			BUG_ON(ret);
+			path->slots[*level]++;
+			btrfs_tree_unlock(next);
+			free_extent_buffer(next);
+			continue;
+		}
+
+		*level = btrfs_header_level(next);
+		path->nodes[*level] = next;
+		path->slots[*level] = 0;
+		path->locks[*level] = 1;
+		cond_resched();
+	}
+out:
+	parent = path->nodes[*level + 1];
+	bytenr = path->nodes[*level]->start;
+	blocksize = path->nodes[*level]->len;
+
+	ret = btrfs_free_extent(trans, root, bytenr, blocksize,
+			parent->start, btrfs_header_owner(parent),
+			btrfs_header_generation(parent), *level, 1);
+	BUG_ON(ret);
+
+	if (path->locks[*level]) {
+		btrfs_tree_unlock(path->nodes[*level]);
+		path->locks[*level] = 0;
+	}
+	free_extent_buffer(path->nodes[*level]);
+	path->nodes[*level] = NULL;
+	*level += 1;
+	cond_resched();
+	return 0;
+}
+
+/*
+ * helper for dropping snapshots.  This walks back up the tree in the path
+ * to find the first node higher up where we haven't yet gone through
+ * all the slots
+ */
+static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 int *level, int max_level)
+{
+	u64 root_owner;
+	u64 root_gen;
+	struct btrfs_root_item *root_item = &root->root_item;
+	int i;
+	int slot;
+	int ret;
+
+	for (i = *level; i < max_level && path->nodes[i]; i++) {
+		slot = path->slots[i];
+		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
+			struct extent_buffer *node;
+			struct btrfs_disk_key disk_key;
+			node = path->nodes[i];
+			path->slots[i]++;
+			*level = i;
+			WARN_ON(*level == 0);
+			btrfs_node_key(node, &disk_key, path->slots[i]);
+			memcpy(&root_item->drop_progress,
+			       &disk_key, sizeof(disk_key));
+			root_item->drop_level = i;
+			return 0;
+		} else {
+			struct extent_buffer *parent;
+			if (path->nodes[*level] == root->node)
+				parent = path->nodes[*level];
+			else
+				parent = path->nodes[*level + 1];
+
+			root_owner = btrfs_header_owner(parent);
+			root_gen = btrfs_header_generation(parent);
+
+			clean_tree_block(trans, root, path->nodes[*level]);
+			ret = btrfs_free_extent(trans, root,
+						path->nodes[*level]->start,
+						path->nodes[*level]->len,
+						parent->start, root_owner,
+						root_gen, *level, 1);
+			BUG_ON(ret);
+			if (path->locks[*level]) {
+				btrfs_tree_unlock(path->nodes[*level]);
+				path->locks[*level] = 0;
+			}
+			free_extent_buffer(path->nodes[*level]);
+			path->nodes[*level] = NULL;
+			*level = i + 1;
+		}
+	}
+	return 1;
+}
+
+/*
+ * drop the reference count on the tree rooted at 'snap'.  This traverses
+ * the tree freeing any blocks that have a ref count of zero after being
+ * decremented.
+ */
+int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
+			*root)
+{
+	int ret = 0;
+	int wret;
+	int level;
+	struct btrfs_path *path;
+	int i;
+	int orig_level;
+	struct btrfs_root_item *root_item = &root->root_item;
+
+	WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	level = btrfs_header_level(root->node);
+	orig_level = level;
+	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
+		path->nodes[level] = root->node;
+		extent_buffer_get(root->node);
+		path->slots[level] = 0;
+	} else {
+		struct btrfs_key key;
+		struct btrfs_disk_key found_key;
+		struct extent_buffer *node;
+
+		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
+		level = root_item->drop_level;
+		path->lowest_level = level;
+		wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (wret < 0) {
+			ret = wret;
+			goto out;
+		}
+		node = path->nodes[level];
+		btrfs_node_key(node, &found_key, path->slots[level]);
+		WARN_ON(memcmp(&found_key, &root_item->drop_progress,
+			       sizeof(found_key)));
+		/*
+		 * unlock our path, this is safe because only this
+		 * function is allowed to delete this snapshot
+		 */
+		for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+			if (path->nodes[i] && path->locks[i]) {
+				path->locks[i] = 0;
+				btrfs_tree_unlock(path->nodes[i]);
+			}
+		}
+	}
+	while (1) {
+		wret = walk_down_tree(trans, root, path, &level);
+		if (wret > 0)
+			break;
+		if (wret < 0)
+			ret = wret;
+
+		wret = walk_up_tree(trans, root, path, &level,
+				    BTRFS_MAX_LEVEL);
+		if (wret > 0)
+			break;
+		if (wret < 0)
+			ret = wret;
+		if (trans->transaction->in_commit) {
+			ret = -EAGAIN;
+			break;
+		}
+		atomic_inc(&root->fs_info->throttle_gen);
+		wake_up(&root->fs_info->transaction_throttle);
+	}
+	for (i = 0; i <= orig_level; i++) {
+		if (path->nodes[i]) {
+			free_extent_buffer(path->nodes[i]);
+			path->nodes[i] = NULL;
+		}
+	}
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
#51 lore raw

PATCH 1/1 Btrfs extent relocation

In reply to Chris Mason

Block groups can be balanced and relocated. Today the relocation happens by following back references and forcing COW on all of the extents that live inside a given block group.

Other the long term relocation code will be added that blindly shifts extents around without changing their logical addresses. That way it won't have to follow the backreferences at all.

This code has the advantage of heavily testing the backrefs, and it can restripe into new RAID levels. When two or more snapshots have references on an extent, the extent is COW'd only once and all of the reference holders are updated to point to the new location.

fs/btrfs/extent-tree.c +2155 -0
@@ -1881,6 +1881,54 @@
 	return flags;
 }
 
+static int do_chunk_alloc(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *extent_root, u64 alloc_bytes,
+			  u64 flags, int force)
+{
+	struct btrfs_space_info *space_info;
+	u64 thresh;
+	int ret = 0;
+
+	mutex_lock(&extent_root->fs_info->chunk_mutex);
+
+	flags = btrfs_reduce_alloc_profile(extent_root, flags);
+
+	space_info = __find_space_info(extent_root->fs_info, flags);
+	if (!space_info) {
+		ret = update_space_info(extent_root->fs_info, flags,
+					0, 0, &space_info);
+		BUG_ON(ret);
+	}
+	BUG_ON(!space_info);
+
+	spin_lock(&space_info->lock);
+	if (space_info->force_alloc) {
+		force = 1;
+		space_info->force_alloc = 0;
+	}
+	if (space_info->full) {
+		spin_unlock(&space_info->lock);
+		goto out;
+	}
+
+	thresh = space_info->total_bytes - space_info->bytes_readonly;
+	thresh = div_factor(thresh, 6);
+	if (!force &&
+	   (space_info->bytes_used + space_info->bytes_pinned +
+	    space_info->bytes_reserved + alloc_bytes) < thresh) {
+		spin_unlock(&space_info->lock);
+		goto out;
+	}
+	spin_unlock(&space_info->lock);
+
+	ret = btrfs_alloc_chunk(trans, extent_root, flags);
+	if (ret)
+		space_info->full = 1;
+out:
+	mutex_unlock(&extent_root->fs_info->chunk_mutex);
+	return ret;
+}
+
 static int update_block_group(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
 			      u64 bytenr, u64 num_bytes, int alloc,
@@ -3829,3 +3877,2110 @@
 	return ret;
 }
 
+int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct extent_buffer *node,
+			struct extent_buffer *parent)
+{
+	struct btrfs_path *path;
+	int level;
+	int parent_level;
+	int ret = 0;
+	int wret;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	BUG_ON(!btrfs_tree_locked(parent));
+	parent_level = btrfs_header_level(parent);
+	extent_buffer_get(parent);
+	path->nodes[parent_level] = parent;
+	path->slots[parent_level] = btrfs_header_nritems(parent);
+
+	BUG_ON(!btrfs_tree_locked(node));
+	level = btrfs_header_level(node);
+	extent_buffer_get(node);
+	path->nodes[level] = node;
+	path->slots[level] = 0;
+
+	while (1) {
+		wret = walk_down_subtree(trans, root, path, &level);
+		if (wret < 0)
+			ret = wret;
+		if (wret != 0)
+			break;
+
+		wret = walk_up_tree(trans, root, path, &level, parent_level);
+		if (wret < 0)
+			ret = wret;
+		if (wret != 0)
+			break;
+	}
+
+	btrfs_free_path(path);
+	return ret;
+}
+
+static unsigned long calc_ra(unsigned long start, unsigned long last,
+			     unsigned long nr)
+{
+	return min(last, start + nr - 1);
+}
+
+static noinline int relocate_inode_pages(struct inode *inode, u64 start,
+					 u64 len)
+{
+	u64 page_start;
+	u64 page_end;
+	unsigned long first_index;
+	unsigned long last_index;
+	unsigned long i;
+	struct page *page;
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct file_ra_state *ra;
+	struct btrfs_ordered_extent *ordered;
+	unsigned int total_read = 0;
+	unsigned int total_dirty = 0;
+	int ret = 0;
+
+	ra = kzalloc(sizeof(*ra), GFP_NOFS);
+
+	mutex_lock(&inode->i_mutex);
+	first_index = start >> PAGE_CACHE_SHIFT;
+	last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
+
+	/* make sure the dirty trick played by the caller work */
+	ret = invalidate_inode_pages2_range(inode->i_mapping,
+					    first_index, last_index);
+	if (ret)
+		goto out_unlock;
+
+	file_ra_state_init(ra, inode->i_mapping);
+
+	for (i = first_index ; i <= last_index; i++) {
+		if (total_read % ra->ra_pages == 0) {
+			btrfs_force_ra(inode->i_mapping, ra, NULL, i,
+				       calc_ra(i, last_index, ra->ra_pages));
+		}
+		total_read++;
+again:
+		if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
+			BUG_ON(1);
+		page = grab_cache_page(inode->i_mapping, i);
+		if (!page) {
+			ret = -ENOMEM;
+			goto out_unlock;
+		}
+		if (!PageUptodate(page)) {
+			btrfs_readpage(NULL, page);
+			lock_page(page);
+			if (!PageUptodate(page)) {
+				unlock_page(page);
+				page_cache_release(page);
+				ret = -EIO;
+				goto out_unlock;
+			}
+		}
+		wait_on_page_writeback(page);
+
+		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
+		page_end = page_start + PAGE_CACHE_SIZE - 1;
+		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+		ordered = btrfs_lookup_ordered_extent(inode, page_start);
+		if (ordered) {
+			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			goto again;
+		}
+		set_page_extent_mapped(page);
+
+		if (i == first_index)
+			set_extent_bits(io_tree, page_start, page_end,
+					EXTENT_BOUNDARY, GFP_NOFS);
+		btrfs_set_extent_delalloc(inode, page_start, page_end);
+
+		set_page_dirty(page);
+		total_dirty++;
+
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		unlock_page(page);
+		page_cache_release(page);
+	}
+
+out_unlock:
+	kfree(ra);
+	mutex_unlock(&inode->i_mutex);
+	balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
+	return ret;
+}
+
+static noinline int relocate_data_extent(struct inode *reloc_inode,
+					 struct btrfs_key *extent_key,
+					 u64 offset)
+{
+	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
+	struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
+	struct extent_map *em;
+	u64 start = extent_key->objectid - offset;
+	u64 end = start + extent_key->offset - 1;
+
+	em = alloc_extent_map(GFP_NOFS);
+	BUG_ON(!em || IS_ERR(em));
+
+	em->start = start;
+	em->len = extent_key->offset;
+	em->block_len = extent_key->offset;
+	em->block_start = extent_key->objectid;
+	em->bdev = root->fs_info->fs_devices->latest_bdev;
+	set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	/* setup extent map to cheat btrfs_readpage */
+	lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
+	while (1) {
+		int ret;
+		spin_lock(&em_tree->lock);
+		ret = add_extent_mapping(em_tree, em);
+		spin_unlock(&em_tree->lock);
+		if (ret != -EEXIST) {
+			free_extent_map(em);
+			break;
+		}
+		btrfs_drop_extent_cache(reloc_inode, start, end, 0);
+	}
+	unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
+
+	return relocate_inode_pages(reloc_inode, start, extent_key->offset);
+}
+
+struct btrfs_ref_path {
+	u64 extent_start;
+	u64 nodes[BTRFS_MAX_LEVEL];
+	u64 root_objectid;
+	u64 root_generation;
+	u64 owner_objectid;
+	u32 num_refs;
+	int lowest_level;
+	int current_level;
+	int shared_level;
+
+	struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
+	u64 new_nodes[BTRFS_MAX_LEVEL];
+};
+
+struct disk_extent {
+	u64 ram_bytes;
+	u64 disk_bytenr;
+	u64 disk_num_bytes;
+	u64 offset;
+	u64 num_bytes;
+	u8 compression;
+	u8 encryption;
+	u16 other_encoding;
+};
+
+static int is_cowonly_root(u64 root_objectid)
+{
+	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
+	    root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
+	    root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
+	    root_objectid == BTRFS_DEV_TREE_OBJECTID ||
+	    root_objectid == BTRFS_TREE_LOG_OBJECTID ||
+	    root_objectid == BTRFS_CSUM_TREE_OBJECTID)
+		return 1;
+	return 0;
+}
+
+static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *extent_root,
+				    struct btrfs_ref_path *ref_path,
+				    int first_time)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_path *path;
+	struct btrfs_extent_ref *ref;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	u64 bytenr;
+	u32 nritems;
+	int level;
+	int ret = 1;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	if (first_time) {
+		ref_path->lowest_level = -1;
+		ref_path->current_level = -1;
+		ref_path->shared_level = -1;
+		goto walk_up;
+	}
+walk_down:
+	level = ref_path->current_level - 1;
+	while (level >= -1) {
+		u64 parent;
+		if (level < ref_path->lowest_level)
+			break;
+
+		if (level >= 0)
+			bytenr = ref_path->nodes[level];
+		else
+			bytenr = ref_path->extent_start;
+		BUG_ON(bytenr == 0);
+
+		parent = ref_path->nodes[level + 1];
+		ref_path->nodes[level + 1] = 0;
+		ref_path->current_level = level;
+		BUG_ON(parent == 0);
+
+		key.objectid = bytenr;
+		key.offset = parent + 1;
+		key.type = BTRFS_EXTENT_REF_KEY;
+
+		ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+		BUG_ON(ret == 0);
+
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(extent_root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				goto next;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid == bytenr &&
+		    found_key.type == BTRFS_EXTENT_REF_KEY) {
+			if (level < ref_path->shared_level)
+				ref_path->shared_level = level;
+			goto found;
+		}
+next:
+		level--;
+		btrfs_release_path(extent_root, path);
+		cond_resched();
+	}
+	/* reached lowest level */
+	ret = 1;
+	goto out;
+walk_up:
+	level = ref_path->current_level;
+	while (level < BTRFS_MAX_LEVEL - 1) {
+		u64 ref_objectid;
+
+		if (level >= 0)
+			bytenr = ref_path->nodes[level];
+		else
+			bytenr = ref_path->extent_start;
+
+		BUG_ON(bytenr == 0);
+
+		key.objectid = bytenr;
+		key.offset = 0;
+		key.type = BTRFS_EXTENT_REF_KEY;
+
+		ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(extent_root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0) {
+				/* the extent was freed by someone */
+				if (ref_path->lowest_level == level)
+					goto out;
+				btrfs_release_path(extent_root, path);
+				goto walk_down;
+			}
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.objectid != bytenr ||
+				found_key.type != BTRFS_EXTENT_REF_KEY) {
+			/* the extent was freed by someone */
+			if (ref_path->lowest_level == level) {
+				ret = 1;
+				goto out;
+			}
+			btrfs_release_path(extent_root, path);
+			goto walk_down;
+		}
+found:
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				struct btrfs_extent_ref);
+		ref_objectid = btrfs_ref_objectid(leaf, ref);
+		if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+			if (first_time) {
+				level = (int)ref_objectid;
+				BUG_ON(level >= BTRFS_MAX_LEVEL);
+				ref_path->lowest_level = level;
+				ref_path->current_level = level;
+				ref_path->nodes[level] = bytenr;
+			} else {
+				WARN_ON(ref_objectid != level);
+			}
+		} else {
+			WARN_ON(level != -1);
+		}
+		first_time = 0;
+
+		if (ref_path->lowest_level == level) {
+			ref_path->owner_objectid = ref_objectid;
+			ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
+		}
+
+		/*
+		 * the block is tree root or the block isn't in reference
+		 * counted tree.
+		 */
+		if (found_key.objectid == found_key.offset ||
+		    is_cowonly_root(btrfs_ref_root(leaf, ref))) {
+			ref_path->root_objectid = btrfs_ref_root(leaf, ref);
+			ref_path->root_generation =
+				btrfs_ref_generation(leaf, ref);
+			if (level < 0) {
+				/* special reference from the tree log */
+				ref_path->nodes[0] = found_key.offset;
+				ref_path->current_level = 0;
+			}
+			ret = 0;
+			goto out;
+		}
+
+		level++;
+		BUG_ON(ref_path->nodes[level] != 0);
+		ref_path->nodes[level] = found_key.offset;
+		ref_path->current_level = level;
+
+		/*
+		 * the reference was created in the running transaction,
+		 * no need to continue walking up.
+		 */
+		if (btrfs_ref_generation(leaf, ref) == trans->transid) {
+			ref_path->root_objectid = btrfs_ref_root(leaf, ref);
+			ref_path->root_generation =
+				btrfs_ref_generation(leaf, ref);
+			ret = 0;
+			goto out;
+		}
+
+		btrfs_release_path(extent_root, path);
+		cond_resched();
+	}
+	/* reached max tree level, but no tree root found. */
+	BUG();
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
+				struct btrfs_root *extent_root,
+				struct btrfs_ref_path *ref_path,
+				u64 extent_start)
+{
+	memset(ref_path, 0, sizeof(*ref_path));
+	ref_path->extent_start = extent_start;
+
+	return __next_ref_path(trans, extent_root, ref_path, 1);
+}
+
+static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *extent_root,
+			       struct btrfs_ref_path *ref_path)
+{
+	return __next_ref_path(trans, extent_root, ref_path, 0);
+}
+
+static noinline int get_new_locations(struct inode *reloc_inode,
+				      struct btrfs_key *extent_key,
+				      u64 offset, int no_fragment,
+				      struct disk_extent **extents,
+				      int *nr_extents)
+{
+	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
+	struct btrfs_path *path;
+	struct btrfs_file_extent_item *fi;
+	struct extent_buffer *leaf;
+	struct disk_extent *exts = *extents;
+	struct btrfs_key found_key;
+	u64 cur_pos;
+	u64 last_byte;
+	u32 nritems;
+	int nr = 0;
+	int max = *nr_extents;
+	int ret;
+
+	WARN_ON(!no_fragment && *extents);
+	if (!exts) {
+		max = 1;
+		exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
+		if (!exts)
+			return -ENOMEM;
+	}
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	cur_pos = extent_key->objectid - offset;
+	last_byte = extent_key->objectid + extent_key->offset;
+	ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
+				       cur_pos, 0);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	while (1) {
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.offset != cur_pos ||
+		    found_key.type != BTRFS_EXTENT_DATA_KEY ||
+		    found_key.objectid != reloc_inode->i_ino)
+			break;
+
+		fi = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) !=
+		    BTRFS_FILE_EXTENT_REG ||
+		    btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
+			break;
+
+		if (nr == max) {
+			struct disk_extent *old = exts;
+			max *= 2;
+			exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
+			memcpy(exts, old, sizeof(*exts) * nr);
+			if (old != *extents)
+				kfree(old);
+		}
+
+		exts[nr].disk_bytenr =
+			btrfs_file_extent_disk_bytenr(leaf, fi);
+		exts[nr].disk_num_bytes =
+			btrfs_file_extent_disk_num_bytes(leaf, fi);
+		exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
+		exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+		exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
+		exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
+		exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
+		exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
+									   fi);
+		BUG_ON(exts[nr].offset > 0);
+		BUG_ON(exts[nr].compression || exts[nr].encryption);
+		BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
+
+		cur_pos += exts[nr].num_bytes;
+		nr++;
+
+		if (cur_pos + offset >= last_byte)
+			break;
+
+		if (no_fragment) {
+			ret = 1;
+			goto out;
+		}
+		path->slots[0]++;
+	}
+
+	BUG_ON(cur_pos + offset > last_byte);
+	if (cur_pos + offset < last_byte) {
+		ret = -ENOENT;
+		goto out;
+	}
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	if (ret) {
+		if (exts != *extents)
+			kfree(exts);
+	} else {
+		*extents = exts;
+		*nr_extents = nr;
+	}
+	return ret;
+}
+
+static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_path *path,
+					struct btrfs_key *extent_key,
+					struct btrfs_key *leaf_key,
+					struct btrfs_ref_path *ref_path,
+					struct disk_extent *new_extents,
+					int nr_extents)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *fi;
+	struct inode *inode = NULL;
+	struct btrfs_key key;
+	u64 lock_start = 0;
+	u64 lock_end = 0;
+	u64 num_bytes;
+	u64 ext_offset;
+	u64 first_pos;
+	u32 nritems;
+	int nr_scaned = 0;
+	int extent_locked = 0;
+	int extent_type;
+	int ret;
+
+	memcpy(&key, leaf_key, sizeof(key));
+	first_pos = INT_LIMIT(loff_t) - extent_key->offset;
+	if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
+		if (key.objectid < ref_path->owner_objectid ||
+		    (key.objectid == ref_path->owner_objectid &&
+		     key.type < BTRFS_EXTENT_DATA_KEY)) {
+			key.objectid = ref_path->owner_objectid;
+			key.type = BTRFS_EXTENT_DATA_KEY;
+			key.offset = 0;
+		}
+	}
+
+	while (1) {
+		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+		if (ret < 0)
+			goto out;
+
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+next:
+		if (extent_locked && ret > 0) {
+			/*
+			 * the file extent item was modified by someone
+			 * before the extent got locked.
+			 */
+			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				      lock_end, GFP_NOFS);
+			extent_locked = 0;
+		}
+
+		if (path->slots[0] >= nritems) {
+			if (++nr_scaned > 2)
+				break;
+
+			BUG_ON(extent_locked);
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+		if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
+			if ((key.objectid > ref_path->owner_objectid) ||
+			    (key.objectid == ref_path->owner_objectid &&
+			     key.type > BTRFS_EXTENT_DATA_KEY) ||
+			    (key.offset >= first_pos + extent_key->offset))
+				break;
+		}
+
+		if (inode && key.objectid != inode->i_ino) {
+			BUG_ON(extent_locked);
+			btrfs_release_path(root, path);
+			mutex_unlock(&inode->i_mutex);
+			iput(inode);
+			inode = NULL;
+			continue;
+		}
+
+		if (key.type != BTRFS_EXTENT_DATA_KEY) {
+			path->slots[0]++;
+			ret = 1;
+			goto next;
+		}
+		fi = btrfs_item_ptr(leaf, path->slots[0],
+				    struct btrfs_file_extent_item);
+		extent_type = btrfs_file_extent_type(leaf, fi);
+		if ((extent_type != BTRFS_FILE_EXTENT_REG &&
+		     extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
+		    (btrfs_file_extent_disk_bytenr(leaf, fi) !=
+		     extent_key->objectid)) {
+			path->slots[0]++;
+			ret = 1;
+			goto next;
+		}
+
+		num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+		ext_offset = btrfs_file_extent_offset(leaf, fi);
+
+		if (first_pos > key.offset - ext_offset)
+			first_pos = key.offset - ext_offset;
+
+		if (!extent_locked) {
+			lock_start = key.offset;
+			lock_end = lock_start + num_bytes - 1;
+		} else {
+			if (lock_start > key.offset ||
+			    lock_end + 1 < key.offset + num_bytes) {
+				unlock_extent(&BTRFS_I(inode)->io_tree,
+					      lock_start, lock_end, GFP_NOFS);
+				extent_locked = 0;
+			}
+		}
+
+		if (!inode) {
+			btrfs_release_path(root, path);
+
+			inode = btrfs_iget_locked(root->fs_info->sb,
+						  key.objectid, root);
+			if (inode->i_state & I_NEW) {
+				BTRFS_I(inode)->root = root;
+				BTRFS_I(inode)->location.objectid =
+					key.objectid;
+				BTRFS_I(inode)->location.type =
+					BTRFS_INODE_ITEM_KEY;
+				BTRFS_I(inode)->location.offset = 0;
+				btrfs_read_locked_inode(inode);
+				unlock_new_inode(inode);
+			}
+			/*
+			 * some code call btrfs_commit_transaction while
+			 * holding the i_mutex, so we can't use mutex_lock
+			 * here.
+			 */
+			if (is_bad_inode(inode) ||
+			    !mutex_trylock(&inode->i_mutex)) {
+				iput(inode);
+				inode = NULL;
+				key.offset = (u64)-1;
+				goto skip;
+			}
+		}
+
+		if (!extent_locked) {
+			struct btrfs_ordered_extent *ordered;
+
+			btrfs_release_path(root, path);
+
+			lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				    lock_end, GFP_NOFS);
+			ordered = btrfs_lookup_first_ordered_extent(inode,
+								    lock_end);
+			if (ordered &&
+			    ordered->file_offset <= lock_end &&
+			    ordered->file_offset + ordered->len > lock_start) {
+				unlock_extent(&BTRFS_I(inode)->io_tree,
+					      lock_start, lock_end, GFP_NOFS);
+				btrfs_start_ordered_extent(inode, ordered, 1);
+				btrfs_put_ordered_extent(ordered);
+				key.offset += num_bytes;
+				goto skip;
+			}
+			if (ordered)
+				btrfs_put_ordered_extent(ordered);
+
+			extent_locked = 1;
+			continue;
+		}
+
+		if (nr_extents == 1) {
+			/* update extent pointer in place */
+			btrfs_set_file_extent_disk_bytenr(leaf, fi,
+						new_extents[0].disk_bytenr);
+			btrfs_set_file_extent_disk_num_bytes(leaf, fi,
+						new_extents[0].disk_num_bytes);
+			btrfs_mark_buffer_dirty(leaf);
+
+			btrfs_drop_extent_cache(inode, key.offset,
+						key.offset + num_bytes - 1, 0);
+
+			ret = btrfs_inc_extent_ref(trans, root,
+						new_extents[0].disk_bytenr,
+						new_extents[0].disk_num_bytes,
+						leaf->start,
+						root->root_key.objectid,
+						trans->transid,
+						key.objectid);
+			BUG_ON(ret);
+
+			ret = btrfs_free_extent(trans, root,
+						extent_key->objectid,
+						extent_key->offset,
+						leaf->start,
+						btrfs_header_owner(leaf),
+						btrfs_header_generation(leaf),
+						key.objectid, 0);
+			BUG_ON(ret);
+
+			btrfs_release_path(root, path);
+			key.offset += num_bytes;
+		} else {
+			BUG_ON(1);
+#if 0
+			u64 alloc_hint;
+			u64 extent_len;
+			int i;
+			/*
+			 * drop old extent pointer at first, then insert the
+			 * new pointers one bye one
+			 */
+			btrfs_release_path(root, path);
+			ret = btrfs_drop_extents(trans, root, inode, key.offset,
+						 key.offset + num_bytes,
+						 key.offset, &alloc_hint);
+			BUG_ON(ret);
+
+			for (i = 0; i < nr_extents; i++) {
+				if (ext_offset >= new_extents[i].num_bytes) {
+					ext_offset -= new_extents[i].num_bytes;
+					continue;
+				}
+				extent_len = min(new_extents[i].num_bytes -
+						 ext_offset, num_bytes);
+
+				ret = btrfs_insert_empty_item(trans, root,
+							      path, &key,
+							      sizeof(*fi));
+				BUG_ON(ret);
+
+				leaf = path->nodes[0];
+				fi = btrfs_item_ptr(leaf, path->slots[0],
+						struct btrfs_file_extent_item);
+				btrfs_set_file_extent_generation(leaf, fi,
+							trans->transid);
+				btrfs_set_file_extent_type(leaf, fi,
+							BTRFS_FILE_EXTENT_REG);
+				btrfs_set_file_extent_disk_bytenr(leaf, fi,
+						new_extents[i].disk_bytenr);
+				btrfs_set_file_extent_disk_num_bytes(leaf, fi,
+						new_extents[i].disk_num_bytes);
+				btrfs_set_file_extent_ram_bytes(leaf, fi,
+						new_extents[i].ram_bytes);
+
+				btrfs_set_file_extent_compression(leaf, fi,
+						new_extents[i].compression);
+				btrfs_set_file_extent_encryption(leaf, fi,
+						new_extents[i].encryption);
+				btrfs_set_file_extent_other_encoding(leaf, fi,
+						new_extents[i].other_encoding);
+
+				btrfs_set_file_extent_num_bytes(leaf, fi,
+							extent_len);
+				ext_offset += new_extents[i].offset;
+				btrfs_set_file_extent_offset(leaf, fi,
+							ext_offset);
+				btrfs_mark_buffer_dirty(leaf);
+
+				btrfs_drop_extent_cache(inode, key.offset,
+						key.offset + extent_len - 1, 0);
+
+				ret = btrfs_inc_extent_ref(trans, root,
+						new_extents[i].disk_bytenr,
+						new_extents[i].disk_num_bytes,
+						leaf->start,
+						root->root_key.objectid,
+						trans->transid, key.objectid);
+				BUG_ON(ret);
+				btrfs_release_path(root, path);
+
+				inode_add_bytes(inode, extent_len);
+
+				ext_offset = 0;
+				num_bytes -= extent_len;
+				key.offset += extent_len;
+
+				if (num_bytes == 0)
+					break;
+			}
+			BUG_ON(i >= nr_extents);
+#endif
+		}
+
+		if (extent_locked) {
+			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				      lock_end, GFP_NOFS);
+			extent_locked = 0;
+		}
+skip:
+		if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
+		    key.offset >= first_pos + extent_key->offset)
+			break;
+
+		cond_resched();
+	}
+	ret = 0;
+out:
+	btrfs_release_path(root, path);
+	if (inode) {
+		mutex_unlock(&inode->i_mutex);
+		if (extent_locked) {
+			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
+				      lock_end, GFP_NOFS);
+		}
+		iput(inode);
+	}
+	return ret;
+}
+
+int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_buffer *buf, u64 orig_start)
+{
+	int level;
+	int ret;
+
+	BUG_ON(btrfs_header_generation(buf) != trans->transid);
+	BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
+
+	level = btrfs_header_level(buf);
+	if (level == 0) {
+		struct btrfs_leaf_ref *ref;
+		struct btrfs_leaf_ref *orig_ref;
+
+		orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
+		if (!orig_ref)
+			return -ENOENT;
+
+		ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
+		if (!ref) {
+			btrfs_free_leaf_ref(root, orig_ref);
+			return -ENOMEM;
+		}
+
+		ref->nritems = orig_ref->nritems;
+		memcpy(ref->extents, orig_ref->extents,
+			sizeof(ref->extents[0]) * ref->nritems);
+
+		btrfs_free_leaf_ref(root, orig_ref);
+
+		ref->root_gen = trans->transid;
+		ref->bytenr = buf->start;
+		ref->owner = btrfs_header_owner(buf);
+		ref->generation = btrfs_header_generation(buf);
+		ret = btrfs_add_leaf_ref(root, ref, 0);
+		WARN_ON(ret);
+		btrfs_free_leaf_ref(root, ref);
+	}
+	return 0;
+}
+
+static noinline int invalidate_extent_cache(struct btrfs_root *root,
+					struct extent_buffer *leaf,
+					struct btrfs_block_group_cache *group,
+					struct btrfs_root *target_root)
+{
+	struct btrfs_key key;
+	struct inode *inode = NULL;
+	struct btrfs_file_extent_item *fi;
+	u64 num_bytes;
+	u64 skip_objectid = 0;
+	u32 nritems;
+	u32 i;
+
+	nritems = btrfs_header_nritems(leaf);
+	for (i = 0; i < nritems; i++) {
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (key.objectid == skip_objectid ||
+		    key.type != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
+			continue;
+		if (!inode || inode->i_ino != key.objectid) {
+			iput(inode);
+			inode = btrfs_ilookup(target_root->fs_info->sb,
+					      key.objectid, target_root, 1);
+		}
+		if (!inode) {
+			skip_objectid = key.objectid;
+			continue;
+		}
+		num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+
+		lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
+			    key.offset + num_bytes - 1, GFP_NOFS);
+		btrfs_drop_extent_cache(inode, key.offset,
+					key.offset + num_bytes - 1, 1);
+		unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
+			      key.offset + num_bytes - 1, GFP_NOFS);
+		cond_resched();
+	}
+	iput(inode);
+	return 0;
+}
+
+static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct extent_buffer *leaf,
+					struct btrfs_block_group_cache *group,
+					struct inode *reloc_inode)
+{
+	struct btrfs_key key;
+	struct btrfs_key extent_key;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_leaf_ref *ref;
+	struct disk_extent *new_extent;
+	u64 bytenr;
+	u64 num_bytes;
+	u32 nritems;
+	u32 i;
+	int ext_index;
+	int nr_extent;
+	int ret;
+
+	new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
+	BUG_ON(!new_extent);
+
+	ref = btrfs_lookup_leaf_ref(root, leaf->start);
+	BUG_ON(!ref);
+
+	ext_index = -1;
+	nritems = btrfs_header_nritems(leaf);
+	for (i = 0; i < nritems; i++) {
+		btrfs_item_key_to_cpu(leaf, &key, i);
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(leaf, fi) ==
+		    BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
+		if (bytenr == 0)
+			continue;
+
+		ext_index++;
+		if (bytenr >= group->key.objectid + group->key.offset ||
+		    bytenr + num_bytes <= group->key.objectid)
+			continue;
+
+		extent_key.objectid = bytenr;
+		extent_key.offset = num_bytes;
+		extent_key.type = BTRFS_EXTENT_ITEM_KEY;
+		nr_extent = 1;
+		ret = get_new_locations(reloc_inode, &extent_key,
+					group->key.objectid, 1,
+					&new_extent, &nr_extent);
+		if (ret > 0)
+			continue;
+		BUG_ON(ret < 0);
+
+		BUG_ON(ref->extents[ext_index].bytenr != bytenr);
+		BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
+		ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
+		ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
+
+		btrfs_set_file_extent_disk_bytenr(leaf, fi,
+						new_extent->disk_bytenr);
+		btrfs_set_file_extent_disk_num_bytes(leaf, fi,
+						new_extent->disk_num_bytes);
+		btrfs_mark_buffer_dirty(leaf);
+
+		ret = btrfs_inc_extent_ref(trans, root,
+					new_extent->disk_bytenr,
+					new_extent->disk_num_bytes,
+					leaf->start,
+					root->root_key.objectid,
+					trans->transid, key.objectid);
+		BUG_ON(ret);
+		ret = btrfs_free_extent(trans, root,
+					bytenr, num_bytes, leaf->start,
+					btrfs_header_owner(leaf),
+					btrfs_header_generation(leaf),
+					key.objectid, 0);
+		BUG_ON(ret);
+		cond_resched();
+	}
+	kfree(new_extent);
+	BUG_ON(ext_index + 1 != ref->nritems);
+	btrfs_free_leaf_ref(root, ref);
+	return 0;
+}
+
+int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	int ret;
+
+	if (root->reloc_root) {
+		reloc_root = root->reloc_root;
+		root->reloc_root = NULL;
+		list_add(&reloc_root->dead_list,
+			 &root->fs_info->dead_reloc_roots);
+
+		btrfs_set_root_bytenr(&reloc_root->root_item,
+				      reloc_root->node->start);
+		btrfs_set_root_level(&root->root_item,
+				     btrfs_header_level(reloc_root->node));
+		memset(&reloc_root->root_item.drop_progress, 0,
+			sizeof(struct btrfs_disk_key));
+		reloc_root->root_item.drop_level = 0;
+
+		ret = btrfs_update_root(trans, root->fs_info->tree_root,
+					&reloc_root->root_key,
+					&reloc_root->root_item);
+		BUG_ON(ret);
+	}
+	return 0;
+}
+
+int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *reloc_root;
+	struct btrfs_root *prev_root = NULL;
+	struct list_head dead_roots;
+	int ret;
+	unsigned long nr;
+
+	INIT_LIST_HEAD(&dead_roots);
+	list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
+
+	while (!list_empty(&dead_roots)) {
+		reloc_root = list_entry(dead_roots.prev,
+					struct btrfs_root, dead_list);
+		list_del_init(&reloc_root->dead_list);
+
+		BUG_ON(reloc_root->commit_root != NULL);
+		while (1) {
+			trans = btrfs_join_transaction(root, 1);
+			BUG_ON(!trans);
+
+			mutex_lock(&root->fs_info->drop_mutex);
+			ret = btrfs_drop_snapshot(trans, reloc_root);
+			if (ret != -EAGAIN)
+				break;
+			mutex_unlock(&root->fs_info->drop_mutex);
+
+			nr = trans->blocks_used;
+			ret = btrfs_end_transaction(trans, root);
+			BUG_ON(ret);
+			btrfs_btree_balance_dirty(root, nr);
+		}
+
+		free_extent_buffer(reloc_root->node);
+
+		ret = btrfs_del_root(trans, root->fs_info->tree_root,
+				     &reloc_root->root_key);
+		BUG_ON(ret);
+		mutex_unlock(&root->fs_info->drop_mutex);
+
+		nr = trans->blocks_used;
+		ret = btrfs_end_transaction(trans, root);
+		BUG_ON(ret);
+		btrfs_btree_balance_dirty(root, nr);
+
+		kfree(prev_root);
+		prev_root = reloc_root;
+	}
+	if (prev_root) {
+		btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
+		kfree(prev_root);
+	}
+	return 0;
+}
+
+int btrfs_add_dead_reloc_root(struct btrfs_root *root)
+{
+	list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
+	return 0;
+}
+
+int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_key location;
+	int found;
+	int ret;
+
+	mutex_lock(&root->fs_info->tree_reloc_mutex);
+	ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
+	BUG_ON(ret);
+	found = !list_empty(&root->fs_info->dead_reloc_roots);
+	mutex_unlock(&root->fs_info->tree_reloc_mutex);
+
+	if (found) {
+		trans = btrfs_start_transaction(root, 1);
+		BUG_ON(!trans);
+		ret = btrfs_commit_transaction(trans, root);
+		BUG_ON(ret);
+	}
+
+	location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
+	location.offset = (u64)-1;
+	location.type = BTRFS_ROOT_ITEM_KEY;
+
+	reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
+	BUG_ON(!reloc_root);
+	btrfs_orphan_cleanup(reloc_root);
+	return 0;
+}
+
+static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *root)
+{
+	struct btrfs_root *reloc_root;
+	struct extent_buffer *eb;
+	struct btrfs_root_item *root_item;
+	struct btrfs_key root_key;
+	int ret;
+
+	BUG_ON(!root->ref_cows);
+	if (root->reloc_root)
+		return 0;
+
+	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
+	BUG_ON(!root_item);
+
+	ret = btrfs_copy_root(trans, root, root->commit_root,
+			      &eb, BTRFS_TREE_RELOC_OBJECTID);
+	BUG_ON(ret);
+
+	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
+	root_key.offset = root->root_key.objectid;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+
+	memcpy(root_item, &root->root_item, sizeof(root_item));
+	btrfs_set_root_refs(root_item, 0);
+	btrfs_set_root_bytenr(root_item, eb->start);
+	btrfs_set_root_level(root_item, btrfs_header_level(eb));
+	btrfs_set_root_generation(root_item, trans->transid);
+
+	btrfs_tree_unlock(eb);
+	free_extent_buffer(eb);
+
+	ret = btrfs_insert_root(trans, root->fs_info->tree_root,
+				&root_key, root_item);
+	BUG_ON(ret);
+	kfree(root_item);
+
+	reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
+						 &root_key);
+	BUG_ON(!reloc_root);
+	reloc_root->last_trans = trans->transid;
+	reloc_root->commit_root = NULL;
+	reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
+
+	root->reloc_root = reloc_root;
+	return 0;
+}
+
+/*
+ * Core function of space balance.
+ *
+ * The idea is using reloc trees to relocate tree blocks in reference
+ * counted roots. There is one reloc tree for each subvol, and all
+ * reloc trees share same root key objectid. Reloc trees are snapshots
+ * of the latest committed roots of subvols (root->commit_root).
+ *
+ * To relocate a tree block referenced by a subvol, there are two steps.
+ * COW the block through subvol's reloc tree, then update block pointer
+ * in the subvol to point to the new block. Since all reloc trees share
+ * same root key objectid, doing special handing for tree blocks owned
+ * by them is easy. Once a tree block has been COWed in one reloc tree,
+ * we can use the resulting new block directly when the same block is
+ * required to COW again through other reloc trees. By this way, relocated
+ * tree blocks are shared between reloc trees, so they are also shared
+ * between subvols.
+ */
+static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
+				      struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      struct btrfs_key *first_key,
+				      struct btrfs_ref_path *ref_path,
+				      struct btrfs_block_group_cache *group,
+				      struct inode *reloc_inode)
+{
+	struct btrfs_root *reloc_root;
+	struct extent_buffer *eb = NULL;
+	struct btrfs_key *keys;
+	u64 *nodes;
+	int level;
+	int shared_level;
+	int lowest_level = 0;
+	int ret;
+
+	if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
+		lowest_level = ref_path->owner_objectid;
+
+	if (!root->ref_cows) {
+		path->lowest_level = lowest_level;
+		ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
+		BUG_ON(ret < 0);
+		path->lowest_level = 0;
+		btrfs_release_path(root, path);
+		return 0;
+	}
+
+	mutex_lock(&root->fs_info->tree_reloc_mutex);
+	ret = init_reloc_tree(trans, root);
+	BUG_ON(ret);
+	reloc_root = root->reloc_root;
+
+	shared_level = ref_path->shared_level;
+	ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
+
+	keys = ref_path->node_keys;
+	nodes = ref_path->new_nodes;
+	memset(&keys[shared_level + 1], 0,
+	       sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
+	memset(&nodes[shared_level + 1], 0,
+	       sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
+
+	if (nodes[lowest_level] == 0) {
+		path->lowest_level = lowest_level;
+		ret = btrfs_search_slot(trans, reloc_root, first_key, path,
+					0, 1);
+		BUG_ON(ret);
+		for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
+			eb = path->nodes[level];
+			if (!eb || eb == reloc_root->node)
+				break;
+			nodes[level] = eb->start;
+			if (level == 0)
+				btrfs_item_key_to_cpu(eb, &keys[level], 0);
+			else
+				btrfs_node_key_to_cpu(eb, &keys[level], 0);
+		}
+		if (nodes[0] &&
+		    ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			eb = path->nodes[0];
+			ret = replace_extents_in_leaf(trans, reloc_root, eb,
+						      group, reloc_inode);
+			BUG_ON(ret);
+		}
+		btrfs_release_path(reloc_root, path);
+	} else {
+		ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
+				       lowest_level);
+		BUG_ON(ret);
+	}
+
+	/*
+	 * replace tree blocks in the fs tree with tree blocks in
+	 * the reloc tree.
+	 */
+	ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
+	BUG_ON(ret < 0);
+
+	if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+		ret = btrfs_search_slot(trans, reloc_root, first_key, path,
+					0, 0);
+		BUG_ON(ret);
+		extent_buffer_get(path->nodes[0]);
+		eb = path->nodes[0];
+		btrfs_release_path(reloc_root, path);
+		ret = invalidate_extent_cache(reloc_root, eb, group, root);
+		BUG_ON(ret);
+		free_extent_buffer(eb);
+	}
+
+	mutex_unlock(&root->fs_info->tree_reloc_mutex);
+	path->lowest_level = 0;
+	return 0;
+}
+
+static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
+					struct btrfs_root *root,
+					struct btrfs_path *path,
+					struct btrfs_key *first_key,
+					struct btrfs_ref_path *ref_path)
+{
+	int ret;
+
+	ret = relocate_one_path(trans, root, path, first_key,
+				ref_path, NULL, NULL);
+	BUG_ON(ret);
+
+	if (root == root->fs_info->extent_root)
+		btrfs_extent_post_op(trans, root);
+
+	return 0;
+}
+
+static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
+				    struct btrfs_root *extent_root,
+				    struct btrfs_path *path,
+				    struct btrfs_key *extent_key)
+{
+	int ret;
+
+	ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
+	if (ret)
+		goto out;
+	ret = btrfs_del_item(trans, extent_root, path);
+out:
+	btrfs_release_path(extent_root, path);
+	return ret;
+}
+
+static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
+						struct btrfs_ref_path *ref_path)
+{
+	struct btrfs_key root_key;
+
+	root_key.objectid = ref_path->root_objectid;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+	if (is_cowonly_root(ref_path->root_objectid))
+		root_key.offset = 0;
+	else
+		root_key.offset = (u64)-1;
+
+	return btrfs_read_fs_root_no_name(fs_info, &root_key);
+}
+
+static noinline int relocate_one_extent(struct btrfs_root *extent_root,
+					struct btrfs_path *path,
+					struct btrfs_key *extent_key,
+					struct btrfs_block_group_cache *group,
+					struct inode *reloc_inode, int pass)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *found_root;
+	struct btrfs_ref_path *ref_path = NULL;
+	struct disk_extent *new_extents = NULL;
+	int nr_extents = 0;
+	int loops;
+	int ret;
+	int level;
+	struct btrfs_key first_key;
+	u64 prev_block = 0;
+
+
+	trans = btrfs_start_transaction(extent_root, 1);
+	BUG_ON(!trans);
+
+	if (extent_key->objectid == 0) {
+		ret = del_extent_zero(trans, extent_root, path, extent_key);
+		goto out;
+	}
+
+	ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
+	if (!ref_path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	for (loops = 0; ; loops++) {
+		if (loops == 0) {
+			ret = btrfs_first_ref_path(trans, extent_root, ref_path,
+						   extent_key->objectid);
+		} else {
+			ret = btrfs_next_ref_path(trans, extent_root, ref_path);
+		}
+		if (ret < 0)
+			goto out;
+		if (ret > 0)
+			break;
+
+		if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
+		    ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
+			continue;
+
+		found_root = read_ref_root(extent_root->fs_info, ref_path);
+		BUG_ON(!found_root);
+		/*
+		 * for reference counted tree, only process reference paths
+		 * rooted at the latest committed root.
+		 */
+		if (found_root->ref_cows &&
+		    ref_path->root_generation != found_root->root_key.offset)
+			continue;
+
+		if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			if (pass == 0) {
+				/*
+				 * copy data extents to new locations
+				 */
+				u64 group_start = group->key.objectid;
+				ret = relocate_data_extent(reloc_inode,
+							   extent_key,
+							   group_start);
+				if (ret < 0)
+					goto out;
+				break;
+			}
+			level = 0;
+		} else {
+			level = ref_path->owner_objectid;
+		}
+
+		if (prev_block != ref_path->nodes[level]) {
+			struct extent_buffer *eb;
+			u64 block_start = ref_path->nodes[level];
+			u64 block_size = btrfs_level_size(found_root, level);
+
+			eb = read_tree_block(found_root, block_start,
+					     block_size, 0);
+			btrfs_tree_lock(eb);
+			BUG_ON(level != btrfs_header_level(eb));
+
+			if (level == 0)
+				btrfs_item_key_to_cpu(eb, &first_key, 0);
+			else
+				btrfs_node_key_to_cpu(eb, &first_key, 0);
+
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+			prev_block = block_start;
+		}
+
+		btrfs_record_root_in_trans(found_root);
+		if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
+			/*
+			 * try to update data extent references while
+			 * keeping metadata shared between snapshots.
+			 */
+			if (pass == 1) {
+				ret = relocate_one_path(trans, found_root,
+						path, &first_key, ref_path,
+						group, reloc_inode);
+				if (ret < 0)
+					goto out;
+				continue;
+			}
+			/*
+			 * use fallback method to process the remaining
+			 * references.
+			 */
+			if (!new_extents) {
+				u64 group_start = group->key.objectid;
+				new_extents = kmalloc(sizeof(*new_extents),
+						      GFP_NOFS);
+				nr_extents = 1;
+				ret = get_new_locations(reloc_inode,
+							extent_key,
+							group_start, 1,
+							&new_extents,
+							&nr_extents);
+				if (ret)
+					goto out;
+			}
+			ret = replace_one_extent(trans, found_root,
+						path, extent_key,
+						&first_key, ref_path,
+						new_extents, nr_extents);
+		} else {
+			ret = relocate_tree_block(trans, found_root, path,
+						  &first_key, ref_path);
+		}
+		if (ret < 0)
+			goto out;
+	}
+	ret = 0;
+out:
+	btrfs_end_transaction(trans, extent_root);
+	kfree(new_extents);
+	kfree(ref_path);
+	return ret;
+}
+
+static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
+{
+	u64 num_devices;
+	u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
+		BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
+
+	num_devices = root->fs_info->fs_devices->rw_devices;
+	if (num_devices == 1) {
+		stripped |= BTRFS_BLOCK_GROUP_DUP;
+		stripped = flags & ~stripped;
+
+		/* turn raid0 into single device chunks */
+		if (flags & BTRFS_BLOCK_GROUP_RAID0)
+			return stripped;
+
+		/* turn mirroring into duplication */
+		if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
+			     BTRFS_BLOCK_GROUP_RAID10))
+			return stripped | BTRFS_BLOCK_GROUP_DUP;
+		return flags;
+	} else {
+		/* they already had raid on here, just return */
+		if (flags & stripped)
+			return flags;
+
+		stripped |= BTRFS_BLOCK_GROUP_DUP;
+		stripped = flags & ~stripped;
+
+		/* switch duplicated blocks with raid1 */
+		if (flags & BTRFS_BLOCK_GROUP_DUP)
+			return stripped | BTRFS_BLOCK_GROUP_RAID1;
+
+		/* turn single device chunks into raid0 */
+		return stripped | BTRFS_BLOCK_GROUP_RAID0;
+	}
+	return flags;
+}
+
+static int __alloc_chunk_for_shrink(struct btrfs_root *root,
+		     struct btrfs_block_group_cache *shrink_block_group,
+		     int force)
+{
+	struct btrfs_trans_handle *trans;
+	u64 new_alloc_flags;
+	u64 calc;
+
+	spin_lock(&shrink_block_group->lock);
+	if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
+		spin_unlock(&shrink_block_group->lock);
+
+		trans = btrfs_start_transaction(root, 1);
+		spin_lock(&shrink_block_group->lock);
+
+		new_alloc_flags = update_block_group_flags(root,
+						   shrink_block_group->flags);
+		if (new_alloc_flags != shrink_block_group->flags) {
+			calc =
+			     btrfs_block_group_used(&shrink_block_group->item);
+		} else {
+			calc = shrink_block_group->key.offset;
+		}
+		spin_unlock(&shrink_block_group->lock);
+
+		do_chunk_alloc(trans, root->fs_info->extent_root,
+			       calc + 2 * 1024 * 1024, new_alloc_flags, force);
+
+		btrfs_end_transaction(trans, root);
+	} else
+		spin_unlock(&shrink_block_group->lock);
+	return 0;
+}
+
+static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 u64 objectid, u64 size)
+{
+	struct btrfs_path *path;
+	struct btrfs_inode_item *item;
+	struct extent_buffer *leaf;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
+	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
+	btrfs_set_inode_generation(leaf, item, 1);
+	btrfs_set_inode_size(leaf, item, size);
+	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
+	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(root, path);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
+					struct btrfs_block_group_cache *group)
+{
+	struct inode *inode = NULL;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *root;
+	struct btrfs_key root_key;
+	u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
+	int err = 0;
+
+	root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+	root_key.offset = (u64)-1;
+	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
+	if (IS_ERR(root))
+		return ERR_CAST(root);
+
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(!trans);
+
+	err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
+	if (err)
+		goto out;
+
+	err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
+	BUG_ON(err);
+
+	err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
+				       group->key.offset, 0, group->key.offset,
+				       0, 0, 0);
+	BUG_ON(err);
+
+	inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
+	if (inode->i_state & I_NEW) {
+		BTRFS_I(inode)->root = root;
+		BTRFS_I(inode)->location.objectid = objectid;
+		BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
+		BTRFS_I(inode)->location.offset = 0;
+		btrfs_read_locked_inode(inode);
+		unlock_new_inode(inode);
+		BUG_ON(is_bad_inode(inode));
+	} else {
+		BUG_ON(1);
+	}
+	BTRFS_I(inode)->index_cnt = group->key.objectid;
+
+	err = btrfs_orphan_add(trans, inode);
+out:
+	btrfs_end_transaction(trans, root);
+	if (err) {
+		if (inode)
+			iput(inode);
+		inode = ERR_PTR(err);
+	}
+	return inode;
+}
+
+int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
+{
+
+	struct btrfs_ordered_sum *sums;
+	struct btrfs_sector_sum *sector_sum;
+	struct btrfs_ordered_extent *ordered;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct list_head list;
+	size_t offset;
+	int ret;
+	u64 disk_bytenr;
+
+	INIT_LIST_HEAD(&list);
+
+	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
+	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
+
+	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
+	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
+				       disk_bytenr + len - 1, &list);
+
+	while (!list_empty(&list)) {
+		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
+		list_del_init(&sums->list);
+
+		sector_sum = sums->sums;
+		sums->bytenr = ordered->start;
+
+		offset = 0;
+		while (offset < sums->len) {
+			sector_sum->bytenr += ordered->start - disk_bytenr;
+			sector_sum++;
+			offset += root->sectorsize;
+		}
+
+		btrfs_add_ordered_sum(inode, ordered, sums);
+	}
+	btrfs_put_ordered_extent(ordered);
+	return 0;
+}
+
+int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_path *path;
+	struct btrfs_fs_info *info = root->fs_info;
+	struct extent_buffer *leaf;
+	struct inode *reloc_inode;
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_key key;
+	u64 skipped;
+	u64 cur_byte;
+	u64 total_found;
+	u32 nritems;
+	int ret;
+	int progress;
+	int pass = 0;
+
+	root = root->fs_info->extent_root;
+
+	block_group = btrfs_lookup_block_group(info, group_start);
+	BUG_ON(!block_group);
+
+	printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
+	       (unsigned long long)block_group->key.objectid,
+	       (unsigned long long)block_group->flags);
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	reloc_inode = create_reloc_inode(info, block_group);
+	BUG_ON(IS_ERR(reloc_inode));
+
+	__alloc_chunk_for_shrink(root, block_group, 1);
+	set_block_group_readonly(block_group);
+
+	btrfs_start_delalloc_inodes(info->tree_root);
+	btrfs_wait_ordered_extents(info->tree_root, 0);
+again:
+	skipped = 0;
+	total_found = 0;
+	progress = 0;
+	key.objectid = block_group->key.objectid;
+	key.offset = 0;
+	key.type = 0;
+	cur_byte = key.objectid;
+
+	trans = btrfs_start_transaction(info->tree_root, 1);
+	btrfs_commit_transaction(trans, info->tree_root);
+
+	mutex_lock(&root->fs_info->cleaner_mutex);
+	btrfs_clean_old_snapshots(info->tree_root);
+	btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
+	mutex_unlock(&root->fs_info->cleaner_mutex);
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+		if (ret < 0)
+			goto out;
+next:
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				goto out;
+			if (ret == 1) {
+				ret = 0;
+				break;
+			}
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+		if (key.objectid >= block_group->key.objectid +
+		    block_group->key.offset)
+			break;
+
+		if (progress && need_resched()) {
+			btrfs_release_path(root, path);
+			cond_resched();
+			progress = 0;
+			continue;
+		}
+		progress = 1;
+
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
+		    key.objectid + key.offset <= cur_byte) {
+			path->slots[0]++;
+			goto next;
+		}
+
+		total_found++;
+		cur_byte = key.objectid + key.offset;
+		btrfs_release_path(root, path);
+
+		__alloc_chunk_for_shrink(root, block_group, 0);
+		ret = relocate_one_extent(root, path, &key, block_group,
+					  reloc_inode, pass);
+		BUG_ON(ret < 0);
+		if (ret > 0)
+			skipped++;
+
+		key.objectid = cur_byte;
+		key.type = 0;
+		key.offset = 0;
+	}
+
+	btrfs_release_path(root, path);
+
+	if (pass == 0) {
+		btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
+		invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
+	}
+
+	if (total_found > 0) {
+		printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
+		       (unsigned long long)total_found, pass);
+		pass++;
+		if (total_found == skipped && pass > 2) {
+			iput(reloc_inode);
+			reloc_inode = create_reloc_inode(info, block_group);
+			pass = 0;
+		}
+		goto again;
+	}
+
+	/* delete reloc_inode */
+	iput(reloc_inode);
+
+	/* unpin extents in this range */
+	trans = btrfs_start_transaction(info->tree_root, 1);
+	btrfs_commit_transaction(trans, info->tree_root);
+
+	spin_lock(&block_group->lock);
+	WARN_ON(block_group->pinned > 0);
+	WARN_ON(block_group->reserved > 0);
+	WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
+	spin_unlock(&block_group->lock);
+	put_block_group(block_group);
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int find_first_block_group(struct btrfs_root *root,
+		struct btrfs_path *path, struct btrfs_key *key)
+{
+	int ret = 0;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+	int slot;
+
+	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+
+	while (1) {
+		slot = path->slots[0];
+		leaf = path->nodes[0];
+		if (slot >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret == 0)
+				continue;
+			if (ret < 0)
+				goto out;
+			break;
+		}
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+
+		if (found_key.objectid >= key->objectid &&
+		    found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
+			ret = 0;
+			goto out;
+		}
+		path->slots[0]++;
+	}
+	ret = -ENOENT;
+out:
+	return ret;
+}
+
+int btrfs_free_block_groups(struct btrfs_fs_info *info)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct rb_node *n;
+
+	spin_lock(&info->block_group_cache_lock);
+	while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
+		block_group = rb_entry(n, struct btrfs_block_group_cache,
+				       cache_node);
+		rb_erase(&block_group->cache_node,
+			 &info->block_group_cache_tree);
+		spin_unlock(&info->block_group_cache_lock);
+
+		btrfs_remove_free_space_cache(block_group);
+		down_write(&block_group->space_info->groups_sem);
+		list_del(&block_group->list);
+		up_write(&block_group->space_info->groups_sem);
+
+		WARN_ON(atomic_read(&block_group->count) != 1);
+		kfree(block_group);
+
+		spin_lock(&info->block_group_cache_lock);
+	}
+	spin_unlock(&info->block_group_cache_lock);
+	return 0;
+}
+
+int btrfs_read_block_groups(struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_block_group_cache *cache;
+	struct btrfs_fs_info *info = root->fs_info;
+	struct btrfs_space_info *space_info;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+
+	root = info->extent_root;
+	key.objectid = 0;
+	key.offset = 0;
+	btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	while (1) {
+		ret = find_first_block_group(root, path, &key);
+		if (ret > 0) {
+			ret = 0;
+			goto error;
+		}
+		if (ret != 0)
+			goto error;
+
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		cache = kzalloc(sizeof(*cache), GFP_NOFS);
+		if (!cache) {
+			ret = -ENOMEM;
+			break;
+		}
+
+		atomic_set(&cache->count, 1);
+		spin_lock_init(&cache->lock);
+		mutex_init(&cache->alloc_mutex);
+		mutex_init(&cache->cache_mutex);
+		INIT_LIST_HEAD(&cache->list);
+		read_extent_buffer(leaf, &cache->item,
+				   btrfs_item_ptr_offset(leaf, path->slots[0]),
+				   sizeof(cache->item));
+		memcpy(&cache->key, &found_key, sizeof(found_key));
+
+		key.objectid = found_key.objectid + found_key.offset;
+		btrfs_release_path(root, path);
+		cache->flags = btrfs_block_group_flags(&cache->item);
+
+		ret = update_space_info(info, cache->flags, found_key.offset,
+					btrfs_block_group_used(&cache->item),
+					&space_info);
+		BUG_ON(ret);
+		cache->space_info = space_info;
+		down_write(&space_info->groups_sem);
+		list_add_tail(&cache->list, &space_info->block_groups);
+		up_write(&space_info->groups_sem);
+
+		ret = btrfs_add_block_group_cache(root->fs_info, cache);
+		BUG_ON(ret);
+
+		set_avail_alloc_bits(root->fs_info, cache->flags);
+		if (btrfs_chunk_readonly(root, cache->key.objectid))
+			set_block_group_readonly(cache);
+	}
+	ret = 0;
+error:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_make_block_group(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root, u64 bytes_used,
+			   u64 type, u64 chunk_objectid, u64 chunk_offset,
+			   u64 size)
+{
+	int ret;
+	struct btrfs_root *extent_root;
+	struct btrfs_block_group_cache *cache;
+
+	extent_root = root->fs_info->extent_root;
+
+	root->fs_info->last_trans_new_blockgroup = trans->transid;
+
+	cache = kzalloc(sizeof(*cache), GFP_NOFS);
+	if (!cache)
+		return -ENOMEM;
+
+	cache->key.objectid = chunk_offset;
+	cache->key.offset = size;
+	cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+	atomic_set(&cache->count, 1);
+	spin_lock_init(&cache->lock);
+	mutex_init(&cache->alloc_mutex);
+	mutex_init(&cache->cache_mutex);
+	INIT_LIST_HEAD(&cache->list);
+
+	btrfs_set_block_group_used(&cache->item, bytes_used);
+	btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
+	cache->flags = type;
+	btrfs_set_block_group_flags(&cache->item, type);
+
+	ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
+				&cache->space_info);
+	BUG_ON(ret);
+	down_write(&cache->space_info->groups_sem);
+	list_add_tail(&cache->list, &cache->space_info->block_groups);
+	up_write(&cache->space_info->groups_sem);
+
+	ret = btrfs_add_block_group_cache(root->fs_info, cache);
+	BUG_ON(ret);
+
+	ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
+				sizeof(cache->item));
+	BUG_ON(ret);
+
+	finish_current_insert(trans, extent_root, 0);
+	ret = del_pending_extents(trans, extent_root, 0);
+	BUG_ON(ret);
+	set_avail_alloc_bits(extent_root->fs_info, type);
+
+	return 0;
+}
+
+int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 group_start)
+{
+	struct btrfs_path *path;
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_key key;
+	int ret;
+
+	root = root->fs_info->extent_root;
+
+	block_group = btrfs_lookup_block_group(root->fs_info, group_start);
+	BUG_ON(!block_group);
+	BUG_ON(!block_group->ro);
+
+	memcpy(&key, &block_group->key, sizeof(key));
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+
+	btrfs_remove_free_space_cache(block_group);
+	rb_erase(&block_group->cache_node,
+		 &root->fs_info->block_group_cache_tree);
+	down_write(&block_group->space_info->groups_sem);
+	list_del(&block_group->list);
+	up_write(&block_group->space_info->groups_sem);
+
+	spin_lock(&block_group->space_info->lock);
+	block_group->space_info->total_bytes -= block_group->key.offset;
+	block_group->space_info->bytes_readonly -= block_group->key.offset;
+	spin_unlock(&block_group->space_info->lock);
+	block_group->space_info->full = 0;
+
+	put_block_group(block_group);
+	put_block_group(block_group);
+
+	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+	if (ret > 0)
+		ret = -EIO;
+	if (ret < 0)
+		goto out;
+
+	ret = btrfs_del_item(trans, root, path);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
#52 lore raw

PATCH 1/1 Btrfs btree core

In reply to Chris Mason

This implements ctrees, which are checksummed btrees. The btrfs btree stores all data in the leaves and only has pointers to leaves or nodes in the nodes. Data stored in the leaves is broken up into two areas.

* Fixed sized item headers contain the keys, the offsets, and size of the item data in the leaf

* Variable sized data

These two areas grow toward each other. Searches are done on the fixed sized item headers, but all the useful data lives in the variable sized data portion of the leaf.

Btrfs keys have three parts: [ u64 objectid, u8 type, u64 offset ]

The type field controls what the offset field means, and the objectid field is generally the inode number, but this varies by the metadata being stored.

All metadata in btrfs is indexed in one of these trees. The fields of the key are used in different ways depending on which btree it is.

fs/btrfs/ctree.c new file +2441 -0
@@ -0,0 +1,2441 @@
+/*
+ * Copyright (C) 2007,2008 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.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "transaction.h"
+#include "print-tree.h"
+#include "locking.h"
+
+static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_path *path, int level);
+static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *ins_key,
+		      struct btrfs_path *path, int data_size, int extend);
+static int push_node_left(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct extent_buffer *dst,
+			  struct extent_buffer *src, int empty);
+static int balance_node_right(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct extent_buffer *dst_buf,
+			      struct extent_buffer *src_buf);
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int level, int slot);
+
+inline void btrfs_init_path(struct btrfs_path *p)
+{
+	memset(p, 0, sizeof(*p));
+}
+
+struct btrfs_path *btrfs_alloc_path(void)
+{
+	struct btrfs_path *path;
+	path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
+	if (path) {
+		btrfs_init_path(path);
+		path->reada = 1;
+	}
+	return path;
+}
+
+/* this also releases the path */
+void btrfs_free_path(struct btrfs_path *p)
+{
+	btrfs_release_path(NULL, p);
+	kmem_cache_free(btrfs_path_cachep, p);
+}
+
+/*
+ * path release drops references on the extent buffers in the path
+ * and it drops any locks held by this path
+ *
+ * It is safe to call this on paths that no locks or extent buffers held.
+ */
+noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
+{
+	int i;
+
+	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+		p->slots[i] = 0;
+		if (!p->nodes[i])
+			continue;
+		if (p->locks[i]) {
+			btrfs_tree_unlock(p->nodes[i]);
+			p->locks[i] = 0;
+		}
+		free_extent_buffer(p->nodes[i]);
+		p->nodes[i] = NULL;
+	}
+}
+
+/*
+ * safely gets a reference on the root node of a tree.  A lock
+ * is not taken, so a concurrent writer may put a different node
+ * at the root of the tree.  See btrfs_lock_root_node for the
+ * looping required.
+ *
+ * The extent buffer returned by this has a reference taken, so
+ * it won't disappear.  It may stop being the root of the tree
+ * at any time because there are no locks held.
+ */
+struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+	spin_lock(&root->node_lock);
+	eb = root->node;
+	extent_buffer_get(eb);
+	spin_unlock(&root->node_lock);
+	return eb;
+}
+
+/* loop around taking references on and locking the root node of the
+ * tree until you end up with a lock on the root.  A locked buffer
+ * is returned, with a reference held.
+ */
+struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+
+	while (1) {
+		eb = btrfs_root_node(root);
+		btrfs_tree_lock(eb);
+
+		spin_lock(&root->node_lock);
+		if (eb == root->node) {
+			spin_unlock(&root->node_lock);
+			break;
+		}
+		spin_unlock(&root->node_lock);
+
+		btrfs_tree_unlock(eb);
+		free_extent_buffer(eb);
+	}
+	return eb;
+}
+
+/* cowonly root (everything not a reference counted cow subvolume), just get
+ * put onto a simple dirty list.  transaction.c walks this to make sure they
+ * get properly updated on disk.
+ */
+static void add_root_to_dirty_list(struct btrfs_root *root)
+{
+	if (root->track_dirty && list_empty(&root->dirty_list)) {
+		list_add(&root->dirty_list,
+			 &root->fs_info->dirty_cowonly_roots);
+	}
+}
+
+/*
+ * used by snapshot creation to make a copy of a root for a tree with
+ * a given objectid.  The buffer with the new root node is returned in
+ * cow_ret, and this func returns zero on success or a negative error code.
+ */
+int btrfs_copy_root(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root,
+		      struct extent_buffer *buf,
+		      struct extent_buffer **cow_ret, u64 new_root_objectid)
+{
+	struct extent_buffer *cow;
+	u32 nritems;
+	int ret = 0;
+	int level;
+	struct btrfs_root *new_root;
+
+	new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
+	if (!new_root)
+		return -ENOMEM;
+
+	memcpy(new_root, root, sizeof(*new_root));
+	new_root->root_key.objectid = new_root_objectid;
+
+	WARN_ON(root->ref_cows && trans->transid !=
+		root->fs_info->running_transaction->transid);
+	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
+
+	level = btrfs_header_level(buf);
+	nritems = btrfs_header_nritems(buf);
+
+	cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
+				     new_root_objectid, trans->transid,
+				     level, buf->start, 0);
+	if (IS_ERR(cow)) {
+		kfree(new_root);
+		return PTR_ERR(cow);
+	}
+
+	copy_extent_buffer(cow, buf, 0, 0, cow->len);
+	btrfs_set_header_bytenr(cow, cow->start);
+	btrfs_set_header_generation(cow, trans->transid);
+	btrfs_set_header_owner(cow, new_root_objectid);
+	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
+
+	write_extent_buffer(cow, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(cow),
+			    BTRFS_FSID_SIZE);
+
+	WARN_ON(btrfs_header_generation(buf) > trans->transid);
+	ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
+	kfree(new_root);
+
+	if (ret)
+		return ret;
+
+	btrfs_mark_buffer_dirty(cow);
+	*cow_ret = cow;
+	return 0;
+}
+
+/*
+ * does the dirty work in cow of a single block.  The parent block (if
+ * supplied) is updated to point to the new cow copy.  The new buffer is marked
+ * dirty and returned locked.  If you modify the block it needs to be marked
+ * dirty again.
+ *
+ * search_start -- an allocation hint for the new block
+ *
+ * empty_size -- a hint that you plan on doing more cow.  This is the size in
+ * bytes the allocator should try to find free next to the block it returns.
+ * This is just a hint and may be ignored by the allocator.
+ *
+ * prealloc_dest -- if you have already reserved a destination for the cow,
+ * this uses that block instead of allocating a new one.
+ * btrfs_alloc_reserved_extent is used to finish the allocation.
+ */
+static noinline 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 search_start, u64 empty_size,
+			     u64 prealloc_dest)
+{
+	u64 parent_start;
+	struct extent_buffer *cow;
+	u32 nritems;
+	int ret = 0;
+	int level;
+	int unlock_orig = 0;
+
+	if (*cow_ret == buf)
+		unlock_orig = 1;
+
+	WARN_ON(!btrfs_tree_locked(buf));
+
+	if (parent)
+		parent_start = parent->start;
+	else
+		parent_start = 0;
+
+	WARN_ON(root->ref_cows && trans->transid !=
+		root->fs_info->running_transaction->transid);
+	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
+
+	level = btrfs_header_level(buf);
+	nritems = btrfs_header_nritems(buf);
+
+	if (prealloc_dest) {
+		struct btrfs_key ins;
+
+		ins.objectid = prealloc_dest;
+		ins.offset = buf->len;
+		ins.type = BTRFS_EXTENT_ITEM_KEY;
+
+		ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
+						  root->root_key.objectid,
+						  trans->transid, level, &ins);
+		BUG_ON(ret);
+		cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
+					    buf->len);
+	} else {
+		cow = btrfs_alloc_free_block(trans, root, buf->len,
+					     parent_start,
+					     root->root_key.objectid,
+					     trans->transid, level,
+					     search_start, empty_size);
+	}
+	if (IS_ERR(cow))
+		return PTR_ERR(cow);
+
+	copy_extent_buffer(cow, buf, 0, 0, cow->len);
+	btrfs_set_header_bytenr(cow, cow->start);
+	btrfs_set_header_generation(cow, trans->transid);
+	btrfs_set_header_owner(cow, root->root_key.objectid);
+	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
+
+	write_extent_buffer(cow, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(cow),
+			    BTRFS_FSID_SIZE);
+
+	WARN_ON(btrfs_header_generation(buf) > trans->transid);
+	if (btrfs_header_generation(buf) != trans->transid) {
+		u32 nr_extents;
+		ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
+		if (ret)
+			return ret;
+
+		ret = btrfs_cache_ref(trans, root, buf, nr_extents);
+		WARN_ON(ret);
+	} else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
+		/*
+		 * There are only two places that can drop reference to
+		 * tree blocks owned by living reloc trees, one is here,
+		 * the other place is btrfs_drop_subtree. In both places,
+		 * we check reference count while tree block is locked.
+		 * Furthermore, if reference count is one, it won't get
+		 * increased by someone else.
+		 */
+		u32 refs;
+		ret = btrfs_lookup_extent_ref(trans, root, buf->start,
+					      buf->len, &refs);
+		BUG_ON(ret);
+		if (refs == 1) {
+			ret = btrfs_update_ref(trans, root, buf, cow,
+					       0, nritems);
+			clean_tree_block(trans, root, buf);
+		} else {
+			ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
+		}
+		BUG_ON(ret);
+	} else {
+		ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
+		if (ret)
+			return ret;
+		clean_tree_block(trans, root, buf);
+	}
+
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
+		ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
+		WARN_ON(ret);
+	}
+
+	if (buf == root->node) {
+		WARN_ON(parent && parent != buf);
+
+		spin_lock(&root->node_lock);
+		root->node = cow;
+		extent_buffer_get(cow);
+		spin_unlock(&root->node_lock);
+
+		if (buf != root->commit_root) {
+			btrfs_free_extent(trans, root, buf->start,
+					  buf->len, buf->start,
+					  root->root_key.objectid,
+					  btrfs_header_generation(buf),
+					  level, 1);
+		}
+		free_extent_buffer(buf);
+		add_root_to_dirty_list(root);
+	} else {
+		btrfs_set_node_blockptr(parent, parent_slot,
+					cow->start);
+		WARN_ON(trans->transid == 0);
+		btrfs_set_node_ptr_generation(parent, parent_slot,
+					      trans->transid);
+		btrfs_mark_buffer_dirty(parent);
+		WARN_ON(btrfs_header_generation(parent) != trans->transid);
+		btrfs_free_extent(trans, root, buf->start, buf->len,
+				  parent_start, btrfs_header_owner(parent),
+				  btrfs_header_generation(parent), level, 1);
+	}
+	if (unlock_orig)
+		btrfs_tree_unlock(buf);
+	free_extent_buffer(buf);
+	btrfs_mark_buffer_dirty(cow);
+	*cow_ret = cow;
+	return 0;
+}
+
+/*
+ * cows a single block, see __btrfs_cow_block for the real work.
+ * This version of it has extra checks so that a block isn't cow'd more than
+ * once per transaction, as long as it hasn't been written yet
+ */
+noinline 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)
+{
+	u64 search_start;
+	int ret;
+
+	if (trans->transaction != root->fs_info->running_transaction) {
+		printk(KERN_CRIT "trans %llu running %llu\n",
+		       (unsigned long long)trans->transid,
+		       (unsigned long long)
+		       root->fs_info->running_transaction->transid);
+		WARN_ON(1);
+	}
+	if (trans->transid != root->fs_info->generation) {
+		printk(KERN_CRIT "trans %llu running %llu\n",
+		       (unsigned long long)trans->transid,
+		       (unsigned long long)root->fs_info->generation);
+		WARN_ON(1);
+	}
+
+	spin_lock(&root->fs_info->hash_lock);
+	if (btrfs_header_generation(buf) == trans->transid &&
+	    btrfs_header_owner(buf) == root->root_key.objectid &&
+	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
+		*cow_ret = buf;
+		spin_unlock(&root->fs_info->hash_lock);
+		WARN_ON(prealloc_dest);
+		return 0;
+	}
+	spin_unlock(&root->fs_info->hash_lock);
+	search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
+	ret = __btrfs_cow_block(trans, root, buf, parent,
+				 parent_slot, cow_ret, search_start, 0,
+				 prealloc_dest);
+	return ret;
+}
+
+/*
+ * helper function for defrag to decide if two blocks pointed to by a
+ * node are actually close by
+ */
+static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
+{
+	if (blocknr < other && other - (blocknr + blocksize) < 32768)
+		return 1;
+	if (blocknr > other && blocknr - (other + blocksize) < 32768)
+		return 1;
+	return 0;
+}
+
+/*
+ * compare two keys in a memcmp fashion
+ */
+static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
+{
+	struct btrfs_key k1;
+
+	btrfs_disk_key_to_cpu(&k1, disk);
+
+	if (k1.objectid > k2->objectid)
+		return 1;
+	if (k1.objectid < k2->objectid)
+		return -1;
+	if (k1.type > k2->type)
+		return 1;
+	if (k1.type < k2->type)
+		return -1;
+	if (k1.offset > k2->offset)
+		return 1;
+	if (k1.offset < k2->offset)
+		return -1;
+	return 0;
+}
+
+/*
+ * same as comp_keys only with two btrfs_key's
+ */
+static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
+{
+	if (k1->objectid > k2->objectid)
+		return 1;
+	if (k1->objectid < k2->objectid)
+		return -1;
+	if (k1->type > k2->type)
+		return 1;
+	if (k1->type < k2->type)
+		return -1;
+	if (k1->offset > k2->offset)
+		return 1;
+	if (k1->offset < k2->offset)
+		return -1;
+	return 0;
+}
+
+/*
+ * The leaf data grows from end-to-front in the node.
+ * this returns the address of the start of the last item,
+ * which is the stop of the leaf data stack
+ */
+static inline unsigned int leaf_data_end(struct btrfs_root *root,
+					 struct extent_buffer *leaf)
+{
+	u32 nr = btrfs_header_nritems(leaf);
+	if (nr == 0)
+		return BTRFS_LEAF_DATA_SIZE(root);
+	return btrfs_item_offset_nr(leaf, nr - 1);
+}
+
+/*
+ * extra debugging checks to make sure all the items in a key are
+ * well formed and in the proper order
+ */
+static int check_node(struct btrfs_root *root, struct btrfs_path *path,
+		      int level)
+{
+	struct extent_buffer *parent = NULL;
+	struct extent_buffer *node = path->nodes[level];
+	struct btrfs_disk_key parent_key;
+	struct btrfs_disk_key node_key;
+	int parent_slot;
+	int slot;
+	struct btrfs_key cpukey;
+	u32 nritems = btrfs_header_nritems(node);
+
+	if (path->nodes[level + 1])
+		parent = path->nodes[level + 1];
+
+	slot = path->slots[level];
+	BUG_ON(nritems == 0);
+	if (parent) {
+		parent_slot = path->slots[level + 1];
+		btrfs_node_key(parent, &parent_key, parent_slot);
+		btrfs_node_key(node, &node_key, 0);
+		BUG_ON(memcmp(&parent_key, &node_key,
+			      sizeof(struct btrfs_disk_key)));
+		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
+		       btrfs_header_bytenr(node));
+	}
+	BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
+	if (slot != 0) {
+		btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
+		btrfs_node_key(node, &node_key, slot);
+		BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
+	}
+	if (slot < nritems - 1) {
+		btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
+		btrfs_node_key(node, &node_key, slot);
+		BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
+	}
+	return 0;
+}
+
+/*
+ * extra checking to make sure all the items in a leaf are
+ * well formed and in the proper order
+ */
+static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
+		      int level)
+{
+	struct extent_buffer *leaf = path->nodes[level];
+	struct extent_buffer *parent = NULL;
+	int parent_slot;
+	struct btrfs_key cpukey;
+	struct btrfs_disk_key parent_key;
+	struct btrfs_disk_key leaf_key;
+	int slot = path->slots[0];
+
+	u32 nritems = btrfs_header_nritems(leaf);
+
+	if (path->nodes[level + 1])
+		parent = path->nodes[level + 1];
+
+	if (nritems == 0)
+		return 0;
+
+	if (parent) {
+		parent_slot = path->slots[level + 1];
+		btrfs_node_key(parent, &parent_key, parent_slot);
+		btrfs_item_key(leaf, &leaf_key, 0);
+
+		BUG_ON(memcmp(&parent_key, &leaf_key,
+		       sizeof(struct btrfs_disk_key)));
+		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
+		       btrfs_header_bytenr(leaf));
+	}
+	if (slot != 0 && slot < nritems - 1) {
+		btrfs_item_key(leaf, &leaf_key, slot);
+		btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
+		if (comp_keys(&leaf_key, &cpukey) <= 0) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d offset bad key\n", slot);
+			BUG_ON(1);
+		}
+		if (btrfs_item_offset_nr(leaf, slot - 1) !=
+		       btrfs_item_end_nr(leaf, slot)) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d offset bad\n", slot);
+			BUG_ON(1);
+		}
+	}
+	if (slot < nritems - 1) {
+		btrfs_item_key(leaf, &leaf_key, slot);
+		btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
+		BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
+		if (btrfs_item_offset_nr(leaf, slot) !=
+			btrfs_item_end_nr(leaf, slot + 1)) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d offset bad\n", slot);
+			BUG_ON(1);
+		}
+	}
+	BUG_ON(btrfs_item_offset_nr(leaf, 0) +
+	       btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
+	return 0;
+}
+
+static noinline int check_block(struct btrfs_root *root,
+				struct btrfs_path *path, int level)
+{
+	return 0;
+	if (level == 0)
+		return check_leaf(root, path, level);
+	return check_node(root, path, level);
+}
+
+/*
+ * search for key in the extent_buffer.  The items start at offset p,
+ * and they are item_size apart.  There are 'max' items in p.
+ *
+ * the slot in the array is returned via slot, and it points to
+ * the place where you would insert key if it is not found in
+ * the array.
+ *
+ * slot may point to max if the key is bigger than all of the keys
+ */
+static noinline int generic_bin_search(struct extent_buffer *eb,
+				       unsigned long p,
+				       int item_size, struct btrfs_key *key,
+				       int max, int *slot)
+{
+	int low = 0;
+	int high = max;
+	int mid;
+	int ret;
+	struct btrfs_disk_key *tmp = NULL;
+	struct btrfs_disk_key unaligned;
+	unsigned long offset;
+	char *map_token = NULL;
+	char *kaddr = NULL;
+	unsigned long map_start = 0;
+	unsigned long map_len = 0;
+	int err;
+
+	while (low < high) {
+		mid = (low + high) / 2;
+		offset = p + mid * item_size;
+
+		if (!map_token || offset < map_start ||
+		    (offset + sizeof(struct btrfs_disk_key)) >
+		    map_start + map_len) {
+			if (map_token) {
+				unmap_extent_buffer(eb, map_token, KM_USER0);
+				map_token = NULL;
+			}
+
+			err = map_private_extent_buffer(eb, offset,
+						sizeof(struct btrfs_disk_key),
+						&map_token, &kaddr,
+						&map_start, &map_len, KM_USER0);
+
+			if (!err) {
+				tmp = (struct btrfs_disk_key *)(kaddr + offset -
+							map_start);
+			} else {
+				read_extent_buffer(eb, &unaligned,
+						   offset, sizeof(unaligned));
+				tmp = &unaligned;
+			}
+
+		} else {
+			tmp = (struct btrfs_disk_key *)(kaddr + offset -
+							map_start);
+		}
+		ret = comp_keys(tmp, key);
+
+		if (ret < 0)
+			low = mid + 1;
+		else if (ret > 0)
+			high = mid;
+		else {
+			*slot = mid;
+			if (map_token)
+				unmap_extent_buffer(eb, map_token, KM_USER0);
+			return 0;
+		}
+	}
+	*slot = low;
+	if (map_token)
+		unmap_extent_buffer(eb, map_token, KM_USER0);
+	return 1;
+}
+
+/*
+ * simple bin_search frontend that does the right thing for
+ * leaves vs nodes
+ */
+static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
+		      int level, int *slot)
+{
+	if (level == 0) {
+		return generic_bin_search(eb,
+					  offsetof(struct btrfs_leaf, items),
+					  sizeof(struct btrfs_item),
+					  key, btrfs_header_nritems(eb),
+					  slot);
+	} else {
+		return generic_bin_search(eb,
+					  offsetof(struct btrfs_node, ptrs),
+					  sizeof(struct btrfs_key_ptr),
+					  key, btrfs_header_nritems(eb),
+					  slot);
+	}
+	return -1;
+}
+
+/* given a node and slot number, this reads the blocks it points to.  The
+ * extent buffer is returned with a reference taken (but unlocked).
+ * NULL is returned on error.
+ */
+static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
+				   struct extent_buffer *parent, int slot)
+{
+	int level = btrfs_header_level(parent);
+	if (slot < 0)
+		return NULL;
+	if (slot >= btrfs_header_nritems(parent))
+		return NULL;
+
+	BUG_ON(level == 0);
+
+	return read_tree_block(root, btrfs_node_blockptr(parent, slot),
+		       btrfs_level_size(root, level - 1),
+		       btrfs_node_ptr_generation(parent, slot));
+}
+
+/*
+ * node level balancing, used to make sure nodes are in proper order for
+ * item deletion.  We balance from the top down, so we have to make sure
+ * that a deletion won't leave an node completely empty later on.
+ */
+static noinline int balance_level(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 struct btrfs_path *path, int level)
+{
+	struct extent_buffer *right = NULL;
+	struct extent_buffer *mid;
+	struct extent_buffer *left = NULL;
+	struct extent_buffer *parent = NULL;
+	int ret = 0;
+	int wret;
+	int pslot;
+	int orig_slot = path->slots[level];
+	int err_on_enospc = 0;
+	u64 orig_ptr;
+
+	if (level == 0)
+		return 0;
+
+	mid = path->nodes[level];
+	WARN_ON(!path->locks[level]);
+	WARN_ON(btrfs_header_generation(mid) != trans->transid);
+
+	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
+
+	if (level < BTRFS_MAX_LEVEL - 1)
+		parent = path->nodes[level + 1];
+	pslot = path->slots[level + 1];
+
+	/*
+	 * deal with the case where there is only one pointer in the root
+	 * by promoting the node below to a root
+	 */
+	if (!parent) {
+		struct extent_buffer *child;
+
+		if (btrfs_header_nritems(mid) != 1)
+			return 0;
+
+		/* promote the child to a root */
+		child = read_node_slot(root, mid, 0);
+		btrfs_tree_lock(child);
+		BUG_ON(!child);
+		ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
+		BUG_ON(ret);
+
+		spin_lock(&root->node_lock);
+		root->node = child;
+		spin_unlock(&root->node_lock);
+
+		ret = btrfs_update_extent_ref(trans, root, child->start,
+					      mid->start, child->start,
+					      root->root_key.objectid,
+					      trans->transid, level - 1);
+		BUG_ON(ret);
+
+		add_root_to_dirty_list(root);
+		btrfs_tree_unlock(child);
+		path->locks[level] = 0;
+		path->nodes[level] = NULL;
+		clean_tree_block(trans, root, mid);
+		btrfs_tree_unlock(mid);
+		/* once for the path */
+		free_extent_buffer(mid);
+		ret = btrfs_free_extent(trans, root, mid->start, mid->len,
+					mid->start, root->root_key.objectid,
+					btrfs_header_generation(mid),
+					level, 1);
+		/* once for the root ptr */
+		free_extent_buffer(mid);
+		return ret;
+	}
+	if (btrfs_header_nritems(mid) >
+	    BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
+		return 0;
+
+	if (btrfs_header_nritems(mid) < 2)
+		err_on_enospc = 1;
+
+	left = read_node_slot(root, parent, pslot - 1);
+	if (left) {
+		btrfs_tree_lock(left);
+		wret = btrfs_cow_block(trans, root, left,
+				       parent, pslot - 1, &left, 0);
+		if (wret) {
+			ret = wret;
+			goto enospc;
+		}
+	}
+	right = read_node_slot(root, parent, pslot + 1);
+	if (right) {
+		btrfs_tree_lock(right);
+		wret = btrfs_cow_block(trans, root, right,
+				       parent, pslot + 1, &right, 0);
+		if (wret) {
+			ret = wret;
+			goto enospc;
+		}
+	}
+
+	/* first, try to make some room in the middle buffer */
+	if (left) {
+		orig_slot += btrfs_header_nritems(left);
+		wret = push_node_left(trans, root, left, mid, 1);
+		if (wret < 0)
+			ret = wret;
+		if (btrfs_header_nritems(mid) < 2)
+			err_on_enospc = 1;
+	}
+
+	/*
+	 * then try to empty the right most buffer into the middle
+	 */
+	if (right) {
+		wret = push_node_left(trans, root, mid, right, 1);
+		if (wret < 0 && wret != -ENOSPC)
+			ret = wret;
+		if (btrfs_header_nritems(right) == 0) {
+			u64 bytenr = right->start;
+			u64 generation = btrfs_header_generation(parent);
+			u32 blocksize = right->len;
+
+			clean_tree_block(trans, root, right);
+			btrfs_tree_unlock(right);
+			free_extent_buffer(right);
+			right = NULL;
+			wret = del_ptr(trans, root, path, level + 1, pslot +
+				       1);
+			if (wret)
+				ret = wret;
+			wret = btrfs_free_extent(trans, root, bytenr,
+						 blocksize, parent->start,
+						 btrfs_header_owner(parent),
+						 generation, level, 1);
+			if (wret)
+				ret = wret;
+		} else {
+			struct btrfs_disk_key right_key;
+			btrfs_node_key(right, &right_key, 0);
+			btrfs_set_node_key(parent, &right_key, pslot + 1);
+			btrfs_mark_buffer_dirty(parent);
+		}
+	}
+	if (btrfs_header_nritems(mid) == 1) {
+		/*
+		 * we're not allowed to leave a node with one item in the
+		 * tree during a delete.  A deletion from lower in the tree
+		 * could try to delete the only pointer in this node.
+		 * So, pull some keys from the left.
+		 * There has to be a left pointer at this point because
+		 * otherwise we would have pulled some pointers from the
+		 * right
+		 */
+		BUG_ON(!left);
+		wret = balance_node_right(trans, root, mid, left);
+		if (wret < 0) {
+			ret = wret;
+			goto enospc;
+		}
+		if (wret == 1) {
+			wret = push_node_left(trans, root, left, mid, 1);
+			if (wret < 0)
+				ret = wret;
+		}
+		BUG_ON(wret == 1);
+	}
+	if (btrfs_header_nritems(mid) == 0) {
+		/* we've managed to empty the middle node, drop it */
+		u64 root_gen = btrfs_header_generation(parent);
+		u64 bytenr = mid->start;
+		u32 blocksize = mid->len;
+
+		clean_tree_block(trans, root, mid);
+		btrfs_tree_unlock(mid);
+		free_extent_buffer(mid);
+		mid = NULL;
+		wret = del_ptr(trans, root, path, level + 1, pslot);
+		if (wret)
+			ret = wret;
+		wret = btrfs_free_extent(trans, root, bytenr, blocksize,
+					 parent->start,
+					 btrfs_header_owner(parent),
+					 root_gen, level, 1);
+		if (wret)
+			ret = wret;
+	} else {
+		/* update the parent key to reflect our changes */
+		struct btrfs_disk_key mid_key;
+		btrfs_node_key(mid, &mid_key, 0);
+		btrfs_set_node_key(parent, &mid_key, pslot);
+		btrfs_mark_buffer_dirty(parent);
+	}
+
+	/* update the path */
+	if (left) {
+		if (btrfs_header_nritems(left) > orig_slot) {
+			extent_buffer_get(left);
+			/* left was locked after cow */
+			path->nodes[level] = left;
+			path->slots[level + 1] -= 1;
+			path->slots[level] = orig_slot;
+			if (mid) {
+				btrfs_tree_unlock(mid);
+				free_extent_buffer(mid);
+			}
+		} else {
+			orig_slot -= btrfs_header_nritems(left);
+			path->slots[level] = orig_slot;
+		}
+	}
+	/* double check we haven't messed things up */
+	check_block(root, path, level);
+	if (orig_ptr !=
+	    btrfs_node_blockptr(path->nodes[level], path->slots[level]))
+		BUG();
+enospc:
+	if (right) {
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+	if (left) {
+		if (path->nodes[level] != left)
+			btrfs_tree_unlock(left);
+		free_extent_buffer(left);
+	}
+	return ret;
+}
+
+/* Node balancing for insertion.  Here we only split or push nodes around
+ * when they are completely full.  This is also done top down, so we
+ * have to be pessimistic.
+ */
+static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path, int level)
+{
+	struct extent_buffer *right = NULL;
+	struct extent_buffer *mid;
+	struct extent_buffer *left = NULL;
+	struct extent_buffer *parent = NULL;
+	int ret = 0;
+	int wret;
+	int pslot;
+	int orig_slot = path->slots[level];
+	u64 orig_ptr;
+
+	if (level == 0)
+		return 1;
+
+	mid = path->nodes[level];
+	WARN_ON(btrfs_header_generation(mid) != trans->transid);
+	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
+
+	if (level < BTRFS_MAX_LEVEL - 1)
+		parent = path->nodes[level + 1];
+	pslot = path->slots[level + 1];
+
+	if (!parent)
+		return 1;
+
+	left = read_node_slot(root, parent, pslot - 1);
+
+	/* first, try to make some room in the middle buffer */
+	if (left) {
+		u32 left_nr;
+
+		btrfs_tree_lock(left);
+		left_nr = btrfs_header_nritems(left);
+		if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
+			wret = 1;
+		} else {
+			ret = btrfs_cow_block(trans, root, left, parent,
+					      pslot - 1, &left, 0);
+			if (ret)
+				wret = 1;
+			else {
+				wret = push_node_left(trans, root,
+						      left, mid, 0);
+			}
+		}
+		if (wret < 0)
+			ret = wret;
+		if (wret == 0) {
+			struct btrfs_disk_key disk_key;
+			orig_slot += left_nr;
+			btrfs_node_key(mid, &disk_key, 0);
+			btrfs_set_node_key(parent, &disk_key, pslot);
+			btrfs_mark_buffer_dirty(parent);
+			if (btrfs_header_nritems(left) > orig_slot) {
+				path->nodes[level] = left;
+				path->slots[level + 1] -= 1;
+				path->slots[level] = orig_slot;
+				btrfs_tree_unlock(mid);
+				free_extent_buffer(mid);
+			} else {
+				orig_slot -=
+					btrfs_header_nritems(left);
+				path->slots[level] = orig_slot;
+				btrfs_tree_unlock(left);
+				free_extent_buffer(left);
+			}
+			return 0;
+		}
+		btrfs_tree_unlock(left);
+		free_extent_buffer(left);
+	}
+	right = read_node_slot(root, parent, pslot + 1);
+
+	/*
+	 * then try to empty the right most buffer into the middle
+	 */
+	if (right) {
+		u32 right_nr;
+		btrfs_tree_lock(right);
+		right_nr = btrfs_header_nritems(right);
+		if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
+			wret = 1;
+		} else {
+			ret = btrfs_cow_block(trans, root, right,
+					      parent, pslot + 1,
+					      &right, 0);
+			if (ret)
+				wret = 1;
+			else {
+				wret = balance_node_right(trans, root,
+							  right, mid);
+			}
+		}
+		if (wret < 0)
+			ret = wret;
+		if (wret == 0) {
+			struct btrfs_disk_key disk_key;
+
+			btrfs_node_key(right, &disk_key, 0);
+			btrfs_set_node_key(parent, &disk_key, pslot + 1);
+			btrfs_mark_buffer_dirty(parent);
+
+			if (btrfs_header_nritems(mid) <= orig_slot) {
+				path->nodes[level] = right;
+				path->slots[level + 1] += 1;
+				path->slots[level] = orig_slot -
+					btrfs_header_nritems(mid);
+				btrfs_tree_unlock(mid);
+				free_extent_buffer(mid);
+			} else {
+				btrfs_tree_unlock(right);
+				free_extent_buffer(right);
+			}
+			return 0;
+		}
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+	return 1;
+}
+
+/*
+ * readahead one full node of leaves, finding things that are close
+ * to the block in 'slot', and triggering ra on them.
+ */
+static noinline void reada_for_search(struct btrfs_root *root,
+				      struct btrfs_path *path,
+				      int level, int slot, u64 objectid)
+{
+	struct extent_buffer *node;
+	struct btrfs_disk_key disk_key;
+	u32 nritems;
+	u64 search;
+	u64 lowest_read;
+	u64 highest_read;
+	u64 nread = 0;
+	int direction = path->reada;
+	struct extent_buffer *eb;
+	u32 nr;
+	u32 blocksize;
+	u32 nscan = 0;
+
+	if (level != 1)
+		return;
+
+	if (!path->nodes[level])
+		return;
+
+	node = path->nodes[level];
+
+	search = btrfs_node_blockptr(node, slot);
+	blocksize = btrfs_level_size(root, level - 1);
+	eb = btrfs_find_tree_block(root, search, blocksize);
+	if (eb) {
+		free_extent_buffer(eb);
+		return;
+	}
+
+	highest_read = search;
+	lowest_read = search;
+
+	nritems = btrfs_header_nritems(node);
+	nr = slot;
+	while (1) {
+		if (direction < 0) {
+			if (nr == 0)
+				break;
+			nr--;
+		} else if (direction > 0) {
+			nr++;
+			if (nr >= nritems)
+				break;
+		}
+		if (path->reada < 0 && objectid) {
+			btrfs_node_key(node, &disk_key, nr);
+			if (btrfs_disk_key_objectid(&disk_key) != objectid)
+				break;
+		}
+		search = btrfs_node_blockptr(node, nr);
+		if ((search >= lowest_read && search <= highest_read) ||
+		    (search < lowest_read && lowest_read - search <= 16384) ||
+		    (search > highest_read && search - highest_read <= 16384)) {
+			readahead_tree_block(root, search, blocksize,
+				     btrfs_node_ptr_generation(node, nr));
+			nread += blocksize;
+		}
+		nscan++;
+		if (path->reada < 2 && (nread > (64 * 1024) || nscan > 32))
+			break;
+
+		if (nread > (256 * 1024) || nscan > 128)
+			break;
+
+		if (search < lowest_read)
+			lowest_read = search;
+		if (search > highest_read)
+			highest_read = search;
+	}
+}
+
+/*
+ * when we walk down the tree, it is usually safe to unlock the higher layers
+ * in the tree.  The exceptions are when our path goes through slot 0, because
+ * operations on the tree might require changing key pointers higher up in the
+ * tree.
+ *
+ * callers might also have set path->keep_locks, which tells this code to keep
+ * the lock if the path points to the last slot in the block.  This is part of
+ * walking through the tree, and selecting the next slot in the higher block.
+ *
+ * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
+ * if lowest_unlock is 1, level 0 won't be unlocked
+ */
+static noinline void unlock_up(struct btrfs_path *path, int level,
+			       int lowest_unlock)
+{
+	int i;
+	int skip_level = level;
+	int no_skips = 0;
+	struct extent_buffer *t;
+
+	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+		if (!path->nodes[i])
+			break;
+		if (!path->locks[i])
+			break;
+		if (!no_skips && path->slots[i] == 0) {
+			skip_level = i + 1;
+			continue;
+		}
+		if (!no_skips && path->keep_locks) {
+			u32 nritems;
+			t = path->nodes[i];
+			nritems = btrfs_header_nritems(t);
+			if (nritems < 1 || path->slots[i] >= nritems - 1) {
+				skip_level = i + 1;
+				continue;
+			}
+		}
+		if (skip_level < i && i >= lowest_unlock)
+			no_skips = 1;
+
+		t = path->nodes[i];
+		if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
+			btrfs_tree_unlock(t);
+			path->locks[i] = 0;
+		}
+	}
+}
+
+/*
+ * look for key in the tree.  path is filled in with nodes along the way
+ * if key is found, we return zero and you can find the item in the leaf
+ * level of the path (level 0)
+ *
+ * If the key isn't found, the path points to the slot where it should
+ * be inserted, and 1 is returned.  If there are other errors during the
+ * search a negative error number is returned.
+ *
+ * if ins_len > 0, nodes and leaves will be split as we walk down the
+ * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
+ * possible)
+ */
+int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *key, struct btrfs_path *p, int
+		      ins_len, int cow)
+{
+	struct extent_buffer *b;
+	struct extent_buffer *tmp;
+	int slot;
+	int ret;
+	int level;
+	int should_reada = p->reada;
+	int lowest_unlock = 1;
+	int blocksize;
+	u8 lowest_level = 0;
+	u64 blocknr;
+	u64 gen;
+	struct btrfs_key prealloc_block;
+
+	lowest_level = p->lowest_level;
+	WARN_ON(lowest_level && ins_len > 0);
+	WARN_ON(p->nodes[0] != NULL);
+
+	if (ins_len < 0)
+		lowest_unlock = 2;
+
+	prealloc_block.objectid = 0;
+
+again:
+	if (p->skip_locking)
+		b = btrfs_root_node(root);
+	else
+		b = btrfs_lock_root_node(root);
+
+	while (b) {
+		level = btrfs_header_level(b);
+
+		/*
+		 * setup the path here so we can release it under lock
+		 * contention with the cow code
+		 */
+		p->nodes[level] = b;
+		if (!p->skip_locking)
+			p->locks[level] = 1;
+
+		if (cow) {
+			int wret;
+
+			/* is a cow on this block not required */
+			spin_lock(&root->fs_info->hash_lock);
+			if (btrfs_header_generation(b) == trans->transid &&
+			    btrfs_header_owner(b) == root->root_key.objectid &&
+			    !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
+				spin_unlock(&root->fs_info->hash_lock);
+				goto cow_done;
+			}
+			spin_unlock(&root->fs_info->hash_lock);
+
+			/* ok, we have to cow, is our old prealloc the right
+			 * size?
+			 */
+			if (prealloc_block.objectid &&
+			    prealloc_block.offset != b->len) {
+				btrfs_free_reserved_extent(root,
+					   prealloc_block.objectid,
+					   prealloc_block.offset);
+				prealloc_block.objectid = 0;
+			}
+
+			/*
+			 * for higher level blocks, try not to allocate blocks
+			 * with the block and the parent locks held.
+			 */
+			if (level > 1 && !prealloc_block.objectid &&
+			    btrfs_path_lock_waiting(p, level)) {
+				u32 size = b->len;
+				u64 hint = b->start;
+
+				btrfs_release_path(root, p);
+				ret = btrfs_reserve_extent(trans, root,
+							   size, size, 0,
+							   hint, (u64)-1,
+							   &prealloc_block, 0);
+				BUG_ON(ret);
+				goto again;
+			}
+
+			wret = btrfs_cow_block(trans, root, b,
+					       p->nodes[level + 1],
+					       p->slots[level + 1],
+					       &b, prealloc_block.objectid);
+			prealloc_block.objectid = 0;
+			if (wret) {
+				free_extent_buffer(b);
+				ret = wret;
+				goto done;
+			}
+		}
+cow_done:
+		BUG_ON(!cow && ins_len);
+		if (level != btrfs_header_level(b))
+			WARN_ON(1);
+		level = btrfs_header_level(b);
+
+		p->nodes[level] = b;
+		if (!p->skip_locking)
+			p->locks[level] = 1;
+
+		ret = check_block(root, p, level);
+		if (ret) {
+			ret = -1;
+			goto done;
+		}
+
+		ret = bin_search(b, key, level, &slot);
+		if (level != 0) {
+			if (ret && slot > 0)
+				slot -= 1;
+			p->slots[level] = slot;
+			if ((p->search_for_split || ins_len > 0) &&
+			    btrfs_header_nritems(b) >=
+			    BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
+				int sret = split_node(trans, root, p, level);
+				BUG_ON(sret > 0);
+				if (sret) {
+					ret = sret;
+					goto done;
+				}
+				b = p->nodes[level];
+				slot = p->slots[level];
+			} else if (ins_len < 0) {
+				int sret = balance_level(trans, root, p,
+							 level);
+				if (sret) {
+					ret = sret;
+					goto done;
+				}
+				b = p->nodes[level];
+				if (!b) {
+					btrfs_release_path(NULL, p);
+					goto again;
+				}
+				slot = p->slots[level];
+				BUG_ON(btrfs_header_nritems(b) == 1);
+			}
+			unlock_up(p, level, lowest_unlock);
+
+			/* this is only true while dropping a snapshot */
+			if (level == lowest_level) {
+				ret = 0;
+				goto done;
+			}
+
+			blocknr = btrfs_node_blockptr(b, slot);
+			gen = btrfs_node_ptr_generation(b, slot);
+			blocksize = btrfs_level_size(root, level - 1);
+
+			tmp = btrfs_find_tree_block(root, blocknr, blocksize);
+			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
+				b = tmp;
+			} else {
+				/*
+				 * reduce lock contention at high levels
+				 * of the btree by dropping locks before
+				 * we read.
+				 */
+				if (level > 1) {
+					btrfs_release_path(NULL, p);
+					if (tmp)
+						free_extent_buffer(tmp);
+					if (should_reada)
+						reada_for_search(root, p,
+								 level, slot,
+								 key->objectid);
+
+					tmp = read_tree_block(root, blocknr,
+							 blocksize, gen);
+					if (tmp)
+						free_extent_buffer(tmp);
+					goto again;
+				} else {
+					if (tmp)
+						free_extent_buffer(tmp);
+					if (should_reada)
+						reada_for_search(root, p,
+								 level, slot,
+								 key->objectid);
+					b = read_node_slot(root, b, slot);
+				}
+			}
+			if (!p->skip_locking)
+				btrfs_tree_lock(b);
+		} else {
+			p->slots[level] = slot;
+			if (ins_len > 0 &&
+			    btrfs_leaf_free_space(root, b) < ins_len) {
+				int sret = split_leaf(trans, root, key,
+						      p, ins_len, ret == 0);
+				BUG_ON(sret > 0);
+				if (sret) {
+					ret = sret;
+					goto done;
+				}
+			}
+			if (!p->search_for_split)
+				unlock_up(p, level, lowest_unlock);
+			goto done;
+		}
+	}
+	ret = 1;
+done:
+	if (prealloc_block.objectid) {
+		btrfs_free_reserved_extent(root,
+			   prealloc_block.objectid,
+			   prealloc_block.offset);
+	}
+
+	return ret;
+}
+
+/*
+ * adjust the pointers going up the tree, starting at level
+ * making sure the right key of each node is points to 'key'.
+ * This is used after shifting pointers to the left, so it stops
+ * fixing up pointers when a given leaf/node is not in slot 0 of the
+ * higher levels
+ *
+ * If this fails to write a tree block, it returns -1, but continues
+ * fixing up the blocks in ram so the tree is consistent.
+ */
+static int fixup_low_keys(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct btrfs_path *path,
+			  struct btrfs_disk_key *key, int level)
+{
+	int i;
+	int ret = 0;
+	struct extent_buffer *t;
+
+	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+		int tslot = path->slots[i];
+		if (!path->nodes[i])
+			break;
+		t = path->nodes[i];
+		btrfs_set_node_key(t, key, tslot);
+		btrfs_mark_buffer_dirty(path->nodes[i]);
+		if (tslot != 0)
+			break;
+	}
+	return ret;
+}
+
+
+/*
+ * try to push data from one node into the next node left in the
+ * tree.
+ *
+ * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
+ * error, and > 0 if there was no room in the left hand block.
+ */
+static int push_node_left(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root, struct extent_buffer *dst,
+			  struct extent_buffer *src, int empty)
+{
+	int push_items = 0;
+	int src_nritems;
+	int dst_nritems;
+	int ret = 0;
+
+	src_nritems = btrfs_header_nritems(src);
+	dst_nritems = btrfs_header_nritems(dst);
+	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
+	WARN_ON(btrfs_header_generation(src) != trans->transid);
+	WARN_ON(btrfs_header_generation(dst) != trans->transid);
+
+	if (!empty && src_nritems <= 8)
+		return 1;
+
+	if (push_items <= 0)
+		return 1;
+
+	if (empty) {
+		push_items = min(src_nritems, push_items);
+		if (push_items < src_nritems) {
+			/* leave at least 8 pointers in the node if
+			 * we aren't going to empty it
+			 */
+			if (src_nritems - push_items < 8) {
+				if (push_items <= 8)
+					return 1;
+				push_items -= 8;
+			}
+		}
+	} else
+		push_items = min(src_nritems - 8, push_items);
+
+	copy_extent_buffer(dst, src,
+			   btrfs_node_key_ptr_offset(dst_nritems),
+			   btrfs_node_key_ptr_offset(0),
+			   push_items * sizeof(struct btrfs_key_ptr));
+
+	if (push_items < src_nritems) {
+		memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
+				      btrfs_node_key_ptr_offset(push_items),
+				      (src_nritems - push_items) *
+				      sizeof(struct btrfs_key_ptr));
+	}
+	btrfs_set_header_nritems(src, src_nritems - push_items);
+	btrfs_set_header_nritems(dst, dst_nritems + push_items);
+	btrfs_mark_buffer_dirty(src);
+	btrfs_mark_buffer_dirty(dst);
+
+	ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
+	BUG_ON(ret);
+
+	return ret;
+}
+
+/*
+ * try to push data from one node into the next node right in the
+ * tree.
+ *
+ * returns 0 if some ptrs were pushed, < 0 if there was some horrible
+ * error, and > 0 if there was no room in the right hand block.
+ *
+ * this will  only push up to 1/2 the contents of the left node over
+ */
+static int balance_node_right(struct btrfs_trans_handle *trans,
+			      struct btrfs_root *root,
+			      struct extent_buffer *dst,
+			      struct extent_buffer *src)
+{
+	int push_items = 0;
+	int max_push;
+	int src_nritems;
+	int dst_nritems;
+	int ret = 0;
+
+	WARN_ON(btrfs_header_generation(src) != trans->transid);
+	WARN_ON(btrfs_header_generation(dst) != trans->transid);
+
+	src_nritems = btrfs_header_nritems(src);
+	dst_nritems = btrfs_header_nritems(dst);
+	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
+	if (push_items <= 0)
+		return 1;
+
+	if (src_nritems < 4)
+		return 1;
+
+	max_push = src_nritems / 2 + 1;
+	/* don't try to empty the node */
+	if (max_push >= src_nritems)
+		return 1;
+
+	if (max_push < push_items)
+		push_items = max_push;
+
+	memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
+				      btrfs_node_key_ptr_offset(0),
+				      (dst_nritems) *
+				      sizeof(struct btrfs_key_ptr));
+
+	copy_extent_buffer(dst, src,
+			   btrfs_node_key_ptr_offset(0),
+			   btrfs_node_key_ptr_offset(src_nritems - push_items),
+			   push_items * sizeof(struct btrfs_key_ptr));
+
+	btrfs_set_header_nritems(src, src_nritems - push_items);
+	btrfs_set_header_nritems(dst, dst_nritems + push_items);
+
+	btrfs_mark_buffer_dirty(src);
+	btrfs_mark_buffer_dirty(dst);
+
+	ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
+	BUG_ON(ret);
+
+	return ret;
+}
+
+/*
+ * worker function to insert a single pointer in a node.
+ * the node should have enough room for the pointer already
+ *
+ * slot and level indicate where you want the key to go, and
+ * blocknr is the block the key points to.
+ *
+ * returns zero on success and < 0 on any error
+ */
+static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_path *path, struct btrfs_disk_key
+		      *key, u64 bytenr, int slot, int level)
+{
+	struct extent_buffer *lower;
+	int nritems;
+
+	BUG_ON(!path->nodes[level]);
+	lower = path->nodes[level];
+	nritems = btrfs_header_nritems(lower);
+	if (slot > nritems)
+		BUG();
+	if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
+		BUG();
+	if (slot != nritems) {
+		memmove_extent_buffer(lower,
+			      btrfs_node_key_ptr_offset(slot + 1),
+			      btrfs_node_key_ptr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_key_ptr));
+	}
+	btrfs_set_node_key(lower, key, slot);
+	btrfs_set_node_blockptr(lower, slot, bytenr);
+	WARN_ON(trans->transid == 0);
+	btrfs_set_node_ptr_generation(lower, slot, trans->transid);
+	btrfs_set_header_nritems(lower, nritems + 1);
+	btrfs_mark_buffer_dirty(lower);
+	return 0;
+}
+
+/*
+ * split the node at the specified level in path in two.
+ * The path is corrected to point to the appropriate node after the split
+ *
+ * Before splitting this tries to make some room in the node by pushing
+ * left and right, if either one works, it returns right away.
+ *
+ * returns 0 on success and < 0 on failure
+ */
+static noinline int split_node(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct btrfs_path *path, int level)
+{
+	struct extent_buffer *c;
+	struct extent_buffer *split;
+	struct btrfs_disk_key disk_key;
+	int mid;
+	int ret;
+	int wret;
+	u32 c_nritems;
+
+	c = path->nodes[level];
+	WARN_ON(btrfs_header_generation(c) != trans->transid);
+	if (c == root->node) {
+		/* trying to split the root, lets make a new one */
+		ret = insert_new_root(trans, root, path, level + 1);
+		if (ret)
+			return ret;
+	} else {
+		ret = push_nodes_for_insert(trans, root, path, level);
+		c = path->nodes[level];
+		if (!ret && btrfs_header_nritems(c) <
+		    BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
+			return 0;
+		if (ret < 0)
+			return ret;
+	}
+
+	c_nritems = btrfs_header_nritems(c);
+
+	split = btrfs_alloc_free_block(trans, root, root->nodesize,
+					path->nodes[level + 1]->start,
+					root->root_key.objectid,
+					trans->transid, level, c->start, 0);
+	if (IS_ERR(split))
+		return PTR_ERR(split);
+
+	btrfs_set_header_flags(split, btrfs_header_flags(c));
+	btrfs_set_header_level(split, btrfs_header_level(c));
+	btrfs_set_header_bytenr(split, split->start);
+	btrfs_set_header_generation(split, trans->transid);
+	btrfs_set_header_owner(split, root->root_key.objectid);
+	btrfs_set_header_flags(split, 0);
+	write_extent_buffer(split, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(split),
+			    BTRFS_FSID_SIZE);
+	write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(split),
+			    BTRFS_UUID_SIZE);
+
+	mid = (c_nritems + 1) / 2;
+
+	copy_extent_buffer(split, c,
+			   btrfs_node_key_ptr_offset(0),
+			   btrfs_node_key_ptr_offset(mid),
+			   (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
+	btrfs_set_header_nritems(split, c_nritems - mid);
+	btrfs_set_header_nritems(c, mid);
+	ret = 0;
+
+	btrfs_mark_buffer_dirty(c);
+	btrfs_mark_buffer_dirty(split);
+
+	btrfs_node_key(split, &disk_key, 0);
+	wret = insert_ptr(trans, root, path, &disk_key, split->start,
+			  path->slots[level + 1] + 1,
+			  level + 1);
+	if (wret)
+		ret = wret;
+
+	ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
+	BUG_ON(ret);
+
+	if (path->slots[level] >= mid) {
+		path->slots[level] -= mid;
+		btrfs_tree_unlock(c);
+		free_extent_buffer(c);
+		path->nodes[level] = split;
+		path->slots[level + 1] += 1;
+	} else {
+		btrfs_tree_unlock(split);
+		free_extent_buffer(split);
+	}
+	return ret;
+}
+
+/*
+ * how many bytes are required to store the items in a leaf.  start
+ * and nr indicate which items in the leaf to check.  This totals up the
+ * space used both by the item structs and the item data
+ */
+static int leaf_space_used(struct extent_buffer *l, int start, int nr)
+{
+	int data_len;
+	int nritems = btrfs_header_nritems(l);
+	int end = min(nritems, start + nr) - 1;
+
+	if (!nr)
+		return 0;
+	data_len = btrfs_item_end_nr(l, start);
+	data_len = data_len - btrfs_item_offset_nr(l, end);
+	data_len += sizeof(struct btrfs_item) * nr;
+	WARN_ON(data_len < 0);
+	return data_len;
+}
+
+/*
+ * The space between the end of the leaf items and
+ * the start of the leaf data.  IOW, how much room
+ * the leaf has left for both items and data
+ */
+noinline int btrfs_leaf_free_space(struct btrfs_root *root,
+				   struct extent_buffer *leaf)
+{
+	int nritems = btrfs_header_nritems(leaf);
+	int ret;
+	ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
+	if (ret < 0) {
+		printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
+		       "used %d nritems %d\n",
+		       ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
+		       leaf_space_used(leaf, 0, nritems), nritems);
+	}
+	return ret;
+}
+
+/*
+ * push some data in the path leaf to the right, trying to free up at
+ * least data_size bytes.  returns zero if the push worked, nonzero otherwise
+ *
+ * returns 1 if the push failed because the other node didn't have enough
+ * room, 0 if everything worked out and < 0 if there were major errors.
+ */
+static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
+			   *root, struct btrfs_path *path, int data_size,
+			   int empty)
+{
+	struct extent_buffer *left = path->nodes[0];
+	struct extent_buffer *right;
+	struct extent_buffer *upper;
+	struct btrfs_disk_key disk_key;
+	int slot;
+	u32 i;
+	int free_space;
+	int push_space = 0;
+	int push_items = 0;
+	struct btrfs_item *item;
+	u32 left_nritems;
+	u32 nr;
+	u32 right_nritems;
+	u32 data_end;
+	u32 this_item_size;
+	int ret;
+
+	slot = path->slots[1];
+	if (!path->nodes[1])
+		return 1;
+
+	upper = path->nodes[1];
+	if (slot >= btrfs_header_nritems(upper) - 1)
+		return 1;
+
+	WARN_ON(!btrfs_tree_locked(path->nodes[1]));
+
+	right = read_node_slot(root, upper, slot + 1);
+	btrfs_tree_lock(right);
+	free_space = btrfs_leaf_free_space(root, right);
+	if (free_space < data_size)
+		goto out_unlock;
+
+	/* cow and double check */
+	ret = btrfs_cow_block(trans, root, right, upper,
+			      slot + 1, &right, 0);
+	if (ret)
+		goto out_unlock;
+
+	free_space = btrfs_leaf_free_space(root, right);
+	if (free_space < data_size)
+		goto out_unlock;
+
+	left_nritems = btrfs_header_nritems(left);
+	if (left_nritems == 0)
+		goto out_unlock;
+
+	if (empty)
+		nr = 0;
+	else
+		nr = 1;
+
+	if (path->slots[0] >= left_nritems)
+		push_space += data_size;
+
+	i = left_nritems - 1;
+	while (i >= nr) {
+		item = btrfs_item_nr(left, i);
+
+		if (!empty && push_items > 0) {
+			if (path->slots[0] > i)
+				break;
+			if (path->slots[0] == i) {
+				int space = btrfs_leaf_free_space(root, left);
+				if (space + push_space * 2 > free_space)
+					break;
+			}
+		}
+
+		if (path->slots[0] == i)
+			push_space += data_size;
+
+		if (!left->map_token) {
+			map_extent_buffer(left, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&left->map_token, &left->kaddr,
+					&left->map_start, &left->map_len,
+					KM_USER1);
+		}
+
+		this_item_size = btrfs_item_size(left, item);
+		if (this_item_size + sizeof(*item) + push_space > free_space)
+			break;
+
+		push_items++;
+		push_space += this_item_size + sizeof(*item);
+		if (i == 0)
+			break;
+		i--;
+	}
+	if (left->map_token) {
+		unmap_extent_buffer(left, left->map_token, KM_USER1);
+		left->map_token = NULL;
+	}
+
+	if (push_items == 0)
+		goto out_unlock;
+
+	if (!empty && push_items == left_nritems)
+		WARN_ON(1);
+
+	/* push left to right */
+	right_nritems = btrfs_header_nritems(right);
+
+	push_space = btrfs_item_end_nr(left, left_nritems - push_items);
+	push_space -= leaf_data_end(root, left);
+
+	/* make room in the right data area */
+	data_end = leaf_data_end(root, right);
+	memmove_extent_buffer(right,
+			      btrfs_leaf_data(right) + data_end - push_space,
+			      btrfs_leaf_data(right) + data_end,
+			      BTRFS_LEAF_DATA_SIZE(root) - data_end);
+
+	/* copy from the left data area */
+	copy_extent_buffer(right, left, btrfs_leaf_data(right) +
+		     BTRFS_LEAF_DATA_SIZE(root) - push_space,
+		     btrfs_leaf_data(left) + leaf_data_end(root, left),
+		     push_space);
+
+	memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
+			      btrfs_item_nr_offset(0),
+			      right_nritems * sizeof(struct btrfs_item));
+
+	/* copy the items from left to right */
+	copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
+		   btrfs_item_nr_offset(left_nritems - push_items),
+		   push_items * sizeof(struct btrfs_item));
+
+	/* update the item pointers */
+	right_nritems += push_items;
+	btrfs_set_header_nritems(right, right_nritems);
+	push_space = BTRFS_LEAF_DATA_SIZE(root);
+	for (i = 0; i < right_nritems; i++) {
+		item = btrfs_item_nr(right, i);
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+		push_space -= btrfs_item_size(right, item);
+		btrfs_set_item_offset(right, item, push_space);
+	}
+
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+	left_nritems -= push_items;
+	btrfs_set_header_nritems(left, left_nritems);
+
+	if (left_nritems)
+		btrfs_mark_buffer_dirty(left);
+	btrfs_mark_buffer_dirty(right);
+
+	ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
+	BUG_ON(ret);
+
+	btrfs_item_key(right, &disk_key, 0);
+	btrfs_set_node_key(upper, &disk_key, slot + 1);
+	btrfs_mark_buffer_dirty(upper);
+
+	/* then fixup the leaf pointer in the path */
+	if (path->slots[0] >= left_nritems) {
+		path->slots[0] -= left_nritems;
+		if (btrfs_header_nritems(path->nodes[0]) == 0)
+			clean_tree_block(trans, root, path->nodes[0]);
+		btrfs_tree_unlock(path->nodes[0]);
+		free_extent_buffer(path->nodes[0]);
+		path->nodes[0] = right;
+		path->slots[1] += 1;
+	} else {
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+	return 0;
+
+out_unlock:
+	btrfs_tree_unlock(right);
+	free_extent_buffer(right);
+	return 1;
+}
+
+/*
+ * push some data in the path leaf to the left, trying to free up at
+ * least data_size bytes.  returns zero if the push worked, nonzero otherwise
+ */
+static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
+			  *root, struct btrfs_path *path, int data_size,
+			  int empty)
+{
+	struct btrfs_disk_key disk_key;
+	struct extent_buffer *right = path->nodes[0];
+	struct extent_buffer *left;
+	int slot;
+	int i;
+	int free_space;
+	int push_space = 0;
+	int push_items = 0;
+	struct btrfs_item *item;
+	u32 old_left_nritems;
+	u32 right_nritems;
+	u32 nr;
+	int ret = 0;
+	int wret;
+	u32 this_item_size;
+	u32 old_left_item_size;
+
+	slot = path->slots[1];
+	if (slot == 0)
+		return 1;
+	if (!path->nodes[1])
+		return 1;
+
+	right_nritems = btrfs_header_nritems(right);
+	if (right_nritems == 0)
+		return 1;
+
+	WARN_ON(!btrfs_tree_locked(path->nodes[1]));
+
+	left = read_node_slot(root, path->nodes[1], slot - 1);
+	btrfs_tree_lock(left);
+	free_space = btrfs_leaf_free_space(root, left);
+	if (free_space < data_size) {
+		ret = 1;
+		goto out;
+	}
+
+	/* cow and double check */
+	ret = btrfs_cow_block(trans, root, left,
+			      path->nodes[1], slot - 1, &left, 0);
+	if (ret) {
+		/* we hit -ENOSPC, but it isn't fatal here */
+		ret = 1;
+		goto out;
+	}
+
+	free_space = btrfs_leaf_free_space(root, left);
+	if (free_space < data_size) {
+		ret = 1;
+		goto out;
+	}
+
+	if (empty)
+		nr = right_nritems;
+	else
+		nr = right_nritems - 1;
+
+	for (i = 0; i < nr; i++) {
+		item = btrfs_item_nr(right, i);
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+
+		if (!empty && push_items > 0) {
+			if (path->slots[0] < i)
+				break;
+			if (path->slots[0] == i) {
+				int space = btrfs_leaf_free_space(root, right);
+				if (space + push_space * 2 > free_space)
+					break;
+			}
+		}
+
+		if (path->slots[0] == i)
+			push_space += data_size;
+
+		this_item_size = btrfs_item_size(right, item);
+		if (this_item_size + sizeof(*item) + push_space > free_space)
+			break;
+
+		push_items++;
+		push_space += this_item_size + sizeof(*item);
+	}
+
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+
+	if (push_items == 0) {
+		ret = 1;
+		goto out;
+	}
+	if (!empty && push_items == btrfs_header_nritems(right))
+		WARN_ON(1);
+
+	/* push data from right to left */
+	copy_extent_buffer(left, right,
+			   btrfs_item_nr_offset(btrfs_header_nritems(left)),
+			   btrfs_item_nr_offset(0),
+			   push_items * sizeof(struct btrfs_item));
+
+	push_space = BTRFS_LEAF_DATA_SIZE(root) -
+		     btrfs_item_offset_nr(right, push_items - 1);
+
+	copy_extent_buffer(left, right, btrfs_leaf_data(left) +
+		     leaf_data_end(root, left) - push_space,
+		     btrfs_leaf_data(right) +
+		     btrfs_item_offset_nr(right, push_items - 1),
+		     push_space);
+	old_left_nritems = btrfs_header_nritems(left);
+	BUG_ON(old_left_nritems <= 0);
+
+	old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
+	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
+		u32 ioff;
+
+		item = btrfs_item_nr(left, i);
+		if (!left->map_token) {
+			map_extent_buffer(left, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&left->map_token, &left->kaddr,
+					&left->map_start, &left->map_len,
+					KM_USER1);
+		}
+
+		ioff = btrfs_item_offset(left, item);
+		btrfs_set_item_offset(left, item,
+		      ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
+	}
+	btrfs_set_header_nritems(left, old_left_nritems + push_items);
+	if (left->map_token) {
+		unmap_extent_buffer(left, left->map_token, KM_USER1);
+		left->map_token = NULL;
+	}
+
+	/* fixup right node */
+	if (push_items > right_nritems) {
+		printk(KERN_CRIT "push items %d nr %u\n", push_items,
+		       right_nritems);
+		WARN_ON(1);
+	}
+
+	if (push_items < right_nritems) {
+		push_space = btrfs_item_offset_nr(right, push_items - 1) -
+						  leaf_data_end(root, right);
+		memmove_extent_buffer(right, btrfs_leaf_data(right) +
+				      BTRFS_LEAF_DATA_SIZE(root) - push_space,
+				      btrfs_leaf_data(right) +
+				      leaf_data_end(root, right), push_space);
+
+		memmove_extent_buffer(right, btrfs_item_nr_offset(0),
+			      btrfs_item_nr_offset(push_items),
+			     (btrfs_header_nritems(right) - push_items) *
+			     sizeof(struct btrfs_item));
+	}
+	right_nritems -= push_items;
+	btrfs_set_header_nritems(right, right_nritems);
+	push_space = BTRFS_LEAF_DATA_SIZE(root);
+	for (i = 0; i < right_nritems; i++) {
+		item = btrfs_item_nr(right, i);
+
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+
+		push_space = push_space - btrfs_item_size(right, item);
+		btrfs_set_item_offset(right, item, push_space);
+	}
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+
+	btrfs_mark_buffer_dirty(left);
+	if (right_nritems)
+		btrfs_mark_buffer_dirty(right);
+
+	ret = btrfs_update_ref(trans, root, right, left,
+			       old_left_nritems, push_items);
+	BUG_ON(ret);
+
+	btrfs_item_key(right, &disk_key, 0);
+	wret = fixup_low_keys(trans, root, path, &disk_key, 1);
+	if (wret)
+		ret = wret;
+
+	/* then fixup the leaf pointer in the path */
+	if (path->slots[0] < push_items) {
+		path->slots[0] += old_left_nritems;
+		if (btrfs_header_nritems(path->nodes[0]) == 0)
+			clean_tree_block(trans, root, path->nodes[0]);
+		btrfs_tree_unlock(path->nodes[0]);
+		free_extent_buffer(path->nodes[0]);
+		path->nodes[0] = left;
+		path->slots[1] -= 1;
+	} else {
+		btrfs_tree_unlock(left);
+		free_extent_buffer(left);
+		path->slots[0] -= push_items;
+	}
+	BUG_ON(path->slots[0] < 0);
+	return ret;
+out:
+	btrfs_tree_unlock(left);
+	free_extent_buffer(left);
+	return ret;
+}
+
+/*
+ * split the path's leaf in two, making sure there is at least data_size
+ * available for the resulting leaf level of the path.
+ *
+ * returns 0 if all went well and < 0 on failure.
+ */
+static noinline int split_leaf(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct btrfs_key *ins_key,
+			       struct btrfs_path *path, int data_size,
+			       int extend)
+{
+	struct extent_buffer *l;
+	u32 nritems;
+	int mid;
+	int slot;
+	struct extent_buffer *right;
+	int data_copy_size;
+	int rt_data_off;
+	int i;
+	int ret = 0;
+	int wret;
+	int double_split;
+	int num_doubles = 0;
+	struct btrfs_disk_key disk_key;
+
+	/* first try to make some room by pushing left and right */
+	if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
+		wret = push_leaf_right(trans, root, path, data_size, 0);
+		if (wret < 0)
+			return wret;
+		if (wret) {
+			wret = push_leaf_left(trans, root, path, data_size, 0);
+			if (wret < 0)
+				return wret;
+		}
+		l = path->nodes[0];
+
+		/* did the pushes work? */
+		if (btrfs_leaf_free_space(root, l) >= data_size)
+			return 0;
+	}
+
+	if (!path->nodes[1]) {
+		ret = insert_new_root(trans, root, path, 1);
+		if (ret)
+			return ret;
+	}
+again:
+	double_split = 0;
+	l = path->nodes[0];
+	slot = path->slots[0];
+	nritems = btrfs_header_nritems(l);
+	mid = (nritems + 1) / 2;
+
+	right = btrfs_alloc_free_block(trans, root, root->leafsize,
+					path->nodes[1]->start,
+					root->root_key.objectid,
+					trans->transid, 0, l->start, 0);
+	if (IS_ERR(right)) {
+		BUG_ON(1);
+		return PTR_ERR(right);
+	}
+
+	memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
+	btrfs_set_header_bytenr(right, right->start);
+	btrfs_set_header_generation(right, trans->transid);
+	btrfs_set_header_owner(right, root->root_key.objectid);
+	btrfs_set_header_level(right, 0);
+	write_extent_buffer(right, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(right),
+			    BTRFS_FSID_SIZE);
+
+	write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(right),
+			    BTRFS_UUID_SIZE);
+	if (mid <= slot) {
+		if (nritems == 1 ||
+		    leaf_space_used(l, mid, nritems - mid) + data_size >
+			BTRFS_LEAF_DATA_SIZE(root)) {
+			if (slot >= nritems) {
+				btrfs_cpu_key_to_disk(&disk_key, ins_key);
+				btrfs_set_header_nritems(right, 0);
+				wret = insert_ptr(trans, root, path,
+						  &disk_key, right->start,
+						  path->slots[1] + 1, 1);
+				if (wret)
+					ret = wret;
+
+				btrfs_tree_unlock(path->nodes[0]);
+				free_extent_buffer(path->nodes[0]);
+				path->nodes[0] = right;
+				path->slots[0] = 0;
+				path->slots[1] += 1;
+				btrfs_mark_buffer_dirty(right);
+				return ret;
+			}
+			mid = slot;
+			if (mid != nritems &&
+			    leaf_space_used(l, mid, nritems - mid) +
+			    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
+				double_split = 1;
+			}
+		}
+	} else {
+		if (leaf_space_used(l, 0, mid) + data_size >
+			BTRFS_LEAF_DATA_SIZE(root)) {
+			if (!extend && data_size && slot == 0) {
+				btrfs_cpu_key_to_disk(&disk_key, ins_key);
+				btrfs_set_header_nritems(right, 0);
+				wret = insert_ptr(trans, root, path,
+						  &disk_key,
+						  right->start,
+						  path->slots[1], 1);
+				if (wret)
+					ret = wret;
+				btrfs_tree_unlock(path->nodes[0]);
+				free_extent_buffer(path->nodes[0]);
+				path->nodes[0] = right;
+				path->slots[0] = 0;
+				if (path->slots[1] == 0) {
+					wret = fixup_low_keys(trans, root,
+						      path, &disk_key, 1);
+					if (wret)
+						ret = wret;
+				}
+				btrfs_mark_buffer_dirty(right);
+				return ret;
+			} else if ((extend || !data_size) && slot == 0) {
+				mid = 1;
+			} else {
+				mid = slot;
+				if (mid != nritems &&
+				    leaf_space_used(l, mid, nritems - mid) +
+				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
+					double_split = 1;
+				}
+			}
+		}
+	}
+	nritems = nritems - mid;
+	btrfs_set_header_nritems(right, nritems);
+	data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
+
+	copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
+			   btrfs_item_nr_offset(mid),
+			   nritems * sizeof(struct btrfs_item));
+
+	copy_extent_buffer(right, l,
+		     btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
+		     data_copy_size, btrfs_leaf_data(l) +
+		     leaf_data_end(root, l), data_copy_size);
+
+	rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
+		      btrfs_item_end_nr(l, mid);
+
+	for (i = 0; i < nritems; i++) {
+		struct btrfs_item *item = btrfs_item_nr(right, i);
+		u32 ioff;
+
+		if (!right->map_token) {
+			map_extent_buffer(right, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&right->map_token, &right->kaddr,
+					&right->map_start, &right->map_len,
+					KM_USER1);
+		}
+
+		ioff = btrfs_item_offset(right, item);
+		btrfs_set_item_offset(right, item, ioff + rt_data_off);
+	}
+
+	if (right->map_token) {
+		unmap_extent_buffer(right, right->map_token, KM_USER1);
+		right->map_token = NULL;
+	}
+
+	btrfs_set_header_nritems(l, mid);
+	ret = 0;
+	btrfs_item_key(right, &disk_key, 0);
+	wret = insert_ptr(trans, root, path, &disk_key, right->start,
+			  path->slots[1] + 1, 1);
+	if (wret)
+		ret = wret;
+
+	btrfs_mark_buffer_dirty(right);
+	btrfs_mark_buffer_dirty(l);
+	BUG_ON(path->slots[0] != slot);
+
+	ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
+	BUG_ON(ret);
+
+	if (mid <= slot) {
+		btrfs_tree_unlock(path->nodes[0]);
+		free_extent_buffer(path->nodes[0]);
+		path->nodes[0] = right;
+		path->slots[0] -= mid;
+		path->slots[1] += 1;
+	} else {
+		btrfs_tree_unlock(right);
+		free_extent_buffer(right);
+	}
+
+	BUG_ON(path->slots[0] < 0);
+
+	if (double_split) {
+		BUG_ON(num_doubles != 0);
+		num_doubles++;
+		goto again;
+	}
+	return ret;
+}
+
+/*
+ * delete the pointer from a given node.
+ *
+ * the tree should have been previously balanced so the deletion does not
+ * empty a node.
+ */
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int level, int slot)
+{
+	struct extent_buffer *parent = path->nodes[level];
+	u32 nritems;
+	int ret = 0;
+	int wret;
+
+	nritems = btrfs_header_nritems(parent);
+	if (slot != nritems - 1) {
+		memmove_extent_buffer(parent,
+			      btrfs_node_key_ptr_offset(slot),
+			      btrfs_node_key_ptr_offset(slot + 1),
+			      sizeof(struct btrfs_key_ptr) *
+			      (nritems - slot - 1));
+	}
+	nritems--;
+	btrfs_set_header_nritems(parent, nritems);
+	if (nritems == 0 && parent == root->node) {
+		BUG_ON(btrfs_header_level(root->node) != 1);
+		/* just turn the root into a leaf and break */
+		btrfs_set_header_level(root->node, 0);
+	} else if (slot == 0) {
+		struct btrfs_disk_key disk_key;
+
+		btrfs_node_key(parent, &disk_key, 0);
+		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
+		if (wret)
+			ret = wret;
+	}
+	btrfs_mark_buffer_dirty(parent);
+	return ret;
+}
#53 lore raw

PATCH 1/1 Btrfs btree item routines

In reply to Chris Mason

Btrfs btree item manipulation.

This includes the helpers to actually insert, remove, extent, truncate and do other basic operations on the btree items.

It also has helpers for walking forward and backward in the btree.

All metadata updates eventually hit this code.

fs/btrfs/ctree.c +1539 -27
@@ -463,6 +463,131 @@
 }
 
 /*
+ * this is used by the defrag code to go through all the
+ * leaves pointed to by a node and reallocate them so that
+ * disk order is close to key order
+ */
+int btrfs_realloc_node(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *root, struct extent_buffer *parent,
+		       int start_slot, int cache_only, u64 *last_ret,
+		       struct btrfs_key *progress)
+{
+	struct extent_buffer *cur;
+	u64 blocknr;
+	u64 gen;
+	u64 search_start = *last_ret;
+	u64 last_block = 0;
+	u64 other;
+	u32 parent_nritems;
+	int end_slot;
+	int i;
+	int err = 0;
+	int parent_level;
+	int uptodate;
+	u32 blocksize;
+	int progress_passed = 0;
+	struct btrfs_disk_key disk_key;
+
+	parent_level = btrfs_header_level(parent);
+	if (cache_only && parent_level != 1)
+		return 0;
+
+	if (trans->transaction != root->fs_info->running_transaction)
+		WARN_ON(1);
+	if (trans->transid != root->fs_info->generation)
+		WARN_ON(1);
+
+	parent_nritems = btrfs_header_nritems(parent);
+	blocksize = btrfs_level_size(root, parent_level - 1);
+	end_slot = parent_nritems;
+
+	if (parent_nritems == 1)
+		return 0;
+
+	for (i = start_slot; i < end_slot; i++) {
+		int close = 1;
+
+		if (!parent->map_token) {
+			map_extent_buffer(parent,
+					btrfs_node_key_ptr_offset(i),
+					sizeof(struct btrfs_key_ptr),
+					&parent->map_token, &parent->kaddr,
+					&parent->map_start, &parent->map_len,
+					KM_USER1);
+		}
+		btrfs_node_key(parent, &disk_key, i);
+		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
+			continue;
+
+		progress_passed = 1;
+		blocknr = btrfs_node_blockptr(parent, i);
+		gen = btrfs_node_ptr_generation(parent, i);
+		if (last_block == 0)
+			last_block = blocknr;
+
+		if (i > 0) {
+			other = btrfs_node_blockptr(parent, i - 1);
+			close = close_blocks(blocknr, other, blocksize);
+		}
+		if (!close && i < end_slot - 2) {
+			other = btrfs_node_blockptr(parent, i + 1);
+			close = close_blocks(blocknr, other, blocksize);
+		}
+		if (close) {
+			last_block = blocknr;
+			continue;
+		}
+		if (parent->map_token) {
+			unmap_extent_buffer(parent, parent->map_token,
+					    KM_USER1);
+			parent->map_token = NULL;
+		}
+
+		cur = btrfs_find_tree_block(root, blocknr, blocksize);
+		if (cur)
+			uptodate = btrfs_buffer_uptodate(cur, gen);
+		else
+			uptodate = 0;
+		if (!cur || !uptodate) {
+			if (cache_only) {
+				free_extent_buffer(cur);
+				continue;
+			}
+			if (!cur) {
+				cur = read_tree_block(root, blocknr,
+							 blocksize, gen);
+			} else if (!uptodate) {
+				btrfs_read_buffer(cur, gen);
+			}
+		}
+		if (search_start == 0)
+			search_start = last_block;
+
+		btrfs_tree_lock(cur);
+		err = __btrfs_cow_block(trans, root, cur, parent, i,
+					&cur, search_start,
+					min(16 * blocksize,
+					    (end_slot - i) * blocksize), 0);
+		if (err) {
+			btrfs_tree_unlock(cur);
+			free_extent_buffer(cur);
+			break;
+		}
+		search_start = cur->start;
+		last_block = cur->start;
+		*last_ret = search_start;
+		btrfs_tree_unlock(cur);
+		free_extent_buffer(cur);
+	}
+	if (parent->map_token) {
+		unmap_extent_buffer(parent, parent->map_token,
+				    KM_USER1);
+		parent->map_token = NULL;
+	}
+	return err;
+}
+
+/*
  * The leaf data grows from end-to-front in the node.
  * this returns the address of the start of the last item,
  * which is the stop of the leaf data stack
@@ -1433,6 +1558,134 @@
 	return ret;
 }
 
+int btrfs_merge_path(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_key *node_keys,
+		     u64 *nodes, int lowest_level)
+{
+	struct extent_buffer *eb;
+	struct extent_buffer *parent;
+	struct btrfs_key key;
+	u64 bytenr;
+	u64 generation;
+	u32 blocksize;
+	int level;
+	int slot;
+	int key_match;
+	int ret;
+
+	eb = btrfs_lock_root_node(root);
+	ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0);
+	BUG_ON(ret);
+
+	parent = eb;
+	while (1) {
+		level = btrfs_header_level(parent);
+		if (level == 0 || level <= lowest_level)
+			break;
+
+		ret = bin_search(parent, &node_keys[lowest_level], level,
+				 &slot);
+		if (ret && slot > 0)
+			slot--;
+
+		bytenr = btrfs_node_blockptr(parent, slot);
+		if (nodes[level - 1] == bytenr)
+			break;
+
+		blocksize = btrfs_level_size(root, level - 1);
+		generation = btrfs_node_ptr_generation(parent, slot);
+		btrfs_node_key_to_cpu(eb, &key, slot);
+		key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
+
+		if (generation == trans->transid) {
+			eb = read_tree_block(root, bytenr, blocksize,
+					     generation);
+			btrfs_tree_lock(eb);
+		}
+
+		/*
+		 * if node keys match and node pointer hasn't been modified
+		 * in the running transaction, we can merge the path. for
+		 * blocks owened by reloc trees, the node pointer check is
+		 * skipped, this is because these blocks are fully controlled
+		 * by the space balance code, no one else can modify them.
+		 */
+		if (!nodes[level - 1] || !key_match ||
+		    (generation == trans->transid &&
+		     btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
+			if (level == 1 || level == lowest_level + 1) {
+				if (generation == trans->transid) {
+					btrfs_tree_unlock(eb);
+					free_extent_buffer(eb);
+				}
+				break;
+			}
+
+			if (generation != trans->transid) {
+				eb = read_tree_block(root, bytenr, blocksize,
+						generation);
+				btrfs_tree_lock(eb);
+			}
+
+			ret = btrfs_cow_block(trans, root, eb, parent, slot,
+					      &eb, 0);
+			BUG_ON(ret);
+
+			if (root->root_key.objectid ==
+			    BTRFS_TREE_RELOC_OBJECTID) {
+				if (!nodes[level - 1]) {
+					nodes[level - 1] = eb->start;
+					memcpy(&node_keys[level - 1], &key,
+					       sizeof(node_keys[0]));
+				} else {
+					WARN_ON(1);
+				}
+			}
+
+			btrfs_tree_unlock(parent);
+			free_extent_buffer(parent);
+			parent = eb;
+			continue;
+		}
+
+		btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
+		btrfs_set_node_ptr_generation(parent, slot, trans->transid);
+		btrfs_mark_buffer_dirty(parent);
+
+		ret = btrfs_inc_extent_ref(trans, root,
+					nodes[level - 1],
+					blocksize, parent->start,
+					btrfs_header_owner(parent),
+					btrfs_header_generation(parent),
+					level - 1);
+		BUG_ON(ret);
+
+		/*
+		 * If the block was created in the running transaction,
+		 * it's possible this is the last reference to it, so we
+		 * should drop the subtree.
+		 */
+		if (generation == trans->transid) {
+			ret = btrfs_drop_subtree(trans, root, eb, parent);
+			BUG_ON(ret);
+			btrfs_tree_unlock(eb);
+			free_extent_buffer(eb);
+		} else {
+			ret = btrfs_free_extent(trans, root, bytenr,
+					blocksize, parent->start,
+					btrfs_header_owner(parent),
+					btrfs_header_generation(parent),
+					level - 1, 1);
+			BUG_ON(ret);
+		}
+		break;
+	}
+	btrfs_tree_unlock(parent);
+	free_extent_buffer(parent);
+	return 0;
+}
+
 /*
  * adjust the pointers going up the tree, starting at level
  * making sure the right key of each node is points to 'key'.
@@ -1464,6 +1717,40 @@
 	return ret;
 }
 
+/*
+ * update item key.
+ *
+ * This function isn't completely safe. It's the caller's responsibility
+ * that the new key won't break the order
+ */
+int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root, struct btrfs_path *path,
+			    struct btrfs_key *new_key)
+{
+	struct btrfs_disk_key disk_key;
+	struct extent_buffer *eb;
+	int slot;
+
+	eb = path->nodes[0];
+	slot = path->slots[0];
+	if (slot > 0) {
+		btrfs_item_key(eb, &disk_key, slot - 1);
+		if (comp_keys(&disk_key, new_key) >= 0)
+			return -1;
+	}
+	if (slot < btrfs_header_nritems(eb) - 1) {
+		btrfs_item_key(eb, &disk_key, slot + 1);
+		if (comp_keys(&disk_key, new_key) <= 0)
+			return -1;
+	}
+
+	btrfs_cpu_key_to_disk(&disk_key, new_key);
+	btrfs_set_item_key(eb, &disk_key, slot);
+	btrfs_mark_buffer_dirty(eb);
+	if (slot == 0)
+		fixup_low_keys(trans, root, path, &disk_key, 1);
+	return 0;
+}
 
 /*
  * try to push data from one node into the next node left in the
@@ -1593,6 +1880,85 @@
 }
 
 /*
+ * helper function to insert a new root level in the tree.
+ * A new node is allocated, and a single item is inserted to
+ * point to the existing root
+ *
+ * returns zero on success or < 0 on failure.
+ */
+static noinline int insert_new_root(struct btrfs_trans_handle *trans,
+			   struct btrfs_root *root,
+			   struct btrfs_path *path, int level)
+{
+	u64 lower_gen;
+	struct extent_buffer *lower;
+	struct extent_buffer *c;
+	struct extent_buffer *old;
+	struct btrfs_disk_key lower_key;
+	int ret;
+
+	BUG_ON(path->nodes[level]);
+	BUG_ON(path->nodes[level-1] != root->node);
+
+	lower = path->nodes[level-1];
+	if (level == 1)
+		btrfs_item_key(lower, &lower_key, 0);
+	else
+		btrfs_node_key(lower, &lower_key, 0);
+
+	c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
+				   root->root_key.objectid, trans->transid,
+				   level, root->node->start, 0);
+	if (IS_ERR(c))
+		return PTR_ERR(c);
+
+	memset_extent_buffer(c, 0, 0, root->nodesize);
+	btrfs_set_header_nritems(c, 1);
+	btrfs_set_header_level(c, level);
+	btrfs_set_header_bytenr(c, c->start);
+	btrfs_set_header_generation(c, trans->transid);
+	btrfs_set_header_owner(c, root->root_key.objectid);
+
+	write_extent_buffer(c, root->fs_info->fsid,
+			    (unsigned long)btrfs_header_fsid(c),
+			    BTRFS_FSID_SIZE);
+
+	write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
+			    (unsigned long)btrfs_header_chunk_tree_uuid(c),
+			    BTRFS_UUID_SIZE);
+
+	btrfs_set_node_key(c, &lower_key, 0);
+	btrfs_set_node_blockptr(c, 0, lower->start);
+	lower_gen = btrfs_header_generation(lower);
+	WARN_ON(lower_gen != trans->transid);
+
+	btrfs_set_node_ptr_generation(c, 0, lower_gen);
+
+	btrfs_mark_buffer_dirty(c);
+
+	spin_lock(&root->node_lock);
+	old = root->node;
+	root->node = c;
+	spin_unlock(&root->node_lock);
+
+	ret = btrfs_update_extent_ref(trans, root, lower->start,
+				      lower->start, c->start,
+				      root->root_key.objectid,
+				      trans->transid, level - 1);
+	BUG_ON(ret);
+
+	/* the super has an extra ref to root->node */
+	free_extent_buffer(old);
+
+	add_root_to_dirty_list(root);
+	extent_buffer_get(c);
+	path->nodes[level] = c;
+	path->locks[level] = 1;
+	path->slots[level] = 0;
+	return 0;
+}
+
+/*
  * worker function to insert a single pointer in a node.
  * the node should have enough room for the pointer already
  *
@@ -2401,41 +2767,1187 @@
 }
 
 /*
- * delete the pointer from a given node.
+ * This function splits a single item into two items,
+ * giving 'new_key' to the new item and splitting the
+ * old one at split_offset (from the start of the item).
  *
- * the tree should have been previously balanced so the deletion does not
- * empty a node.
+ * The path may be released by this operation.  After
+ * the split, the path is pointing to the old item.  The
+ * new item is going to be in the same node as the old one.
+ *
+ * Note, the item being split must be smaller enough to live alone on
+ * a tree block with room for one extra struct btrfs_item
+ *
+ * This allows us to split the item in place, keeping a lock on the
+ * leaf the entire time.
  */
-static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		   struct btrfs_path *path, int level, int slot)
-{
-	struct extent_buffer *parent = path->nodes[level];
+int btrfs_split_item(struct btrfs_trans_handle *trans,
+		     struct btrfs_root *root,
+		     struct btrfs_path *path,
+		     struct btrfs_key *new_key,
+		     unsigned long split_offset)
+{
+	u32 item_size;
+	struct extent_buffer *leaf;
+	struct btrfs_key orig_key;
+	struct btrfs_item *item;
+	struct btrfs_item *new_item;
+	int ret = 0;
+	int slot;
 	u32 nritems;
+	u32 orig_offset;
+	struct btrfs_disk_key disk_key;
+	char *buf;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
+	if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
+		goto split;
+
+	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+	btrfs_release_path(root, path);
+
+	path->search_for_split = 1;
+	path->keep_locks = 1;
+
+	ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
+	path->search_for_split = 0;
+
+	/* if our item isn't there or got smaller, return now */
+	if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
+							path->slots[0])) {
+		path->keep_locks = 0;
+		return -EAGAIN;
+	}
+
+	ret = split_leaf(trans, root, &orig_key, path,
+			 sizeof(struct btrfs_item), 1);
+	path->keep_locks = 0;
+	BUG_ON(ret);
+
+	leaf = path->nodes[0];
+	BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
+
+split:
+	item = btrfs_item_nr(leaf, path->slots[0]);
+	orig_offset = btrfs_item_offset(leaf, item);
+	item_size = btrfs_item_size(leaf, item);
+
+
+	buf = kmalloc(item_size, GFP_NOFS);
+	read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
+			    path->slots[0]), item_size);
+	slot = path->slots[0] + 1;
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+
+	if (slot != nritems) {
+		/* shift the items */
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
+			      btrfs_item_nr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_item));
+
+	}
+
+	btrfs_cpu_key_to_disk(&disk_key, new_key);
+	btrfs_set_item_key(leaf, &disk_key, slot);
+
+	new_item = btrfs_item_nr(leaf, slot);
+
+	btrfs_set_item_offset(leaf, new_item, orig_offset);
+	btrfs_set_item_size(leaf, new_item, item_size - split_offset);
+
+	btrfs_set_item_offset(leaf, item,
+			      orig_offset + item_size - split_offset);
+	btrfs_set_item_size(leaf, item, split_offset);
+
+	btrfs_set_header_nritems(leaf, nritems + 1);
+
+	/* write the data for the start of the original item */
+	write_extent_buffer(leaf, buf,
+			    btrfs_item_ptr_offset(leaf, path->slots[0]),
+			    split_offset);
+
+	/* write the data for the new item */
+	write_extent_buffer(leaf, buf + split_offset,
+			    btrfs_item_ptr_offset(leaf, slot),
+			    item_size - split_offset);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+	kfree(buf);
+	return ret;
+}
+
+/*
+ * make the item pointed to by the path smaller.  new_size indicates
+ * how small to make it, and from_end tells us if we just chop bytes
+ * off the end of the item or if we shift the item to chop bytes off
+ * the front.
+ */
+int btrfs_truncate_item(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			u32 new_size, int from_end)
+{
 	int ret = 0;
-	int wret;
+	int slot;
+	int slot_orig;
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	u32 nritems;
+	unsigned int data_end;
+	unsigned int old_data_start;
+	unsigned int old_size;
+	unsigned int size_diff;
+	int i;
 
-	nritems = btrfs_header_nritems(parent);
-	if (slot != nritems - 1) {
-		memmove_extent_buffer(parent,
-			      btrfs_node_key_ptr_offset(slot),
-			      btrfs_node_key_ptr_offset(slot + 1),
-			      sizeof(struct btrfs_key_ptr) *
-			      (nritems - slot - 1));
+	slot_orig = path->slots[0];
+	leaf = path->nodes[0];
+	slot = path->slots[0];
+
+	old_size = btrfs_item_size_nr(leaf, slot);
+	if (old_size == new_size)
+		return 0;
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	old_data_start = btrfs_item_offset_nr(leaf, slot);
+
+	size_diff = old_size - new_size;
+
+	BUG_ON(slot < 0);
+	BUG_ON(slot >= nritems);
+
+	/*
+	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+	 */
+	/* first correct the data pointers */
+	for (i = slot; i < nritems; i++) {
+		u32 ioff;
+		item = btrfs_item_nr(leaf, i);
+
+		if (!leaf->map_token) {
+			map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+		}
+
+		ioff = btrfs_item_offset(leaf, item);
+		btrfs_set_item_offset(leaf, item, ioff + size_diff);
 	}
-	nritems--;
-	btrfs_set_header_nritems(parent, nritems);
-	if (nritems == 0 && parent == root->node) {
-		BUG_ON(btrfs_header_level(root->node) != 1);
-		/* just turn the root into a leaf and break */
-		btrfs_set_header_level(root->node, 0);
-	} else if (slot == 0) {
+
+	if (leaf->map_token) {
+		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+		leaf->map_token = NULL;
+	}
+
+	/* shift the data */
+	if (from_end) {
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + size_diff, btrfs_leaf_data(leaf) +
+			      data_end, old_data_start + new_size - data_end);
+	} else {
 		struct btrfs_disk_key disk_key;
+		u64 offset;
 
-		btrfs_node_key(parent, &disk_key, 0);
-		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
-		if (wret)
-			ret = wret;
+		btrfs_item_key(leaf, &disk_key, slot);
+
+		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
+			unsigned long ptr;
+			struct btrfs_file_extent_item *fi;
+
+			fi = btrfs_item_ptr(leaf, slot,
+					    struct btrfs_file_extent_item);
+			fi = (struct btrfs_file_extent_item *)(
+			     (unsigned long)fi - size_diff);
+
+			if (btrfs_file_extent_type(leaf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE) {
+				ptr = btrfs_item_ptr_offset(leaf, slot);
+				memmove_extent_buffer(leaf, ptr,
+				      (unsigned long)fi,
+				      offsetof(struct btrfs_file_extent_item,
+						 disk_bytenr));
+			}
+		}
+
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + size_diff, btrfs_leaf_data(leaf) +
+			      data_end, old_data_start - data_end);
+
+		offset = btrfs_disk_key_offset(&disk_key);
+		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
+		btrfs_set_item_key(leaf, &disk_key, slot);
+		if (slot == 0)
+			fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
+
+	item = btrfs_item_nr(leaf, slot);
+	btrfs_set_item_size(leaf, item, new_size);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+	return ret;
+}
+
+/*
+ * make the item pointed to by the path bigger, data_size is the new size.
+ */
+int btrfs_extend_item(struct btrfs_trans_handle *trans,
+		      struct btrfs_root *root, struct btrfs_path *path,
+		      u32 data_size)
+{
+	int ret = 0;
+	int slot;
+	int slot_orig;
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	u32 nritems;
+	unsigned int data_end;
+	unsigned int old_data;
+	unsigned int old_size;
+	int i;
+
+	slot_orig = path->slots[0];
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	if (btrfs_leaf_free_space(root, leaf) < data_size) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+	slot = path->slots[0];
+	old_data = btrfs_item_end_nr(leaf, slot);
+
+	BUG_ON(slot < 0);
+	if (slot >= nritems) {
+		btrfs_print_leaf(root, leaf);
+		printk(KERN_CRIT "slot %d too large, nritems %d\n",
+		       slot, nritems);
+		BUG_ON(1);
+	}
+
+	/*
+	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+	 */
+	/* first correct the data pointers */
+	for (i = slot; i < nritems; i++) {
+		u32 ioff;
+		item = btrfs_item_nr(leaf, i);
+
+		if (!leaf->map_token) {
+			map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+		}
+		ioff = btrfs_item_offset(leaf, item);
+		btrfs_set_item_offset(leaf, item, ioff - data_size);
+	}
+
+	if (leaf->map_token) {
+		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+		leaf->map_token = NULL;
+	}
+
+	/* shift the data */
+	memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+		      data_end - data_size, btrfs_leaf_data(leaf) +
+		      data_end, old_data - data_end);
+
+	data_end = old_data;
+	old_size = btrfs_item_size_nr(leaf, slot);
+	item = btrfs_item_nr(leaf, slot);
+	btrfs_set_item_size(leaf, item, old_size + data_size);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
 	}
-	btrfs_mark_buffer_dirty(parent);
 	return ret;
 }
+
+/*
+ * Given a key and some data, insert items into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ * Returns the number of keys that were inserted.
+ */
+int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path,
+			    struct btrfs_key *cpu_key, u32 *data_size,
+			    int nr)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	int ret = 0;
+	int slot;
+	int i;
+	u32 nritems;
+	u32 total_data = 0;
+	u32 total_size = 0;
+	unsigned int data_end;
+	struct btrfs_disk_key disk_key;
+	struct btrfs_key found_key;
+
+	for (i = 0; i < nr; i++) {
+		if (total_size + data_size[i] + sizeof(struct btrfs_item) >
+		    BTRFS_LEAF_DATA_SIZE(root)) {
+			break;
+			nr = i;
+		}
+		total_data += data_size[i];
+		total_size += data_size[i] + sizeof(struct btrfs_item);
+	}
+	BUG_ON(nr == 0);
+
+	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
+	if (ret == 0)
+		return -EEXIST;
+	if (ret < 0)
+		goto out;
+
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	if (btrfs_leaf_free_space(root, leaf) < total_size) {
+		for (i = nr; i >= 0; i--) {
+			total_data -= data_size[i];
+			total_size -= data_size[i] + sizeof(struct btrfs_item);
+			if (total_size < btrfs_leaf_free_space(root, leaf))
+				break;
+		}
+		nr = i;
+	}
+
+	slot = path->slots[0];
+	BUG_ON(slot < 0);
+
+	if (slot != nritems) {
+		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
+
+		item = btrfs_item_nr(leaf, slot);
+		btrfs_item_key_to_cpu(leaf, &found_key, slot);
+
+		/* figure out how many keys we can insert in here */
+		total_data = data_size[0];
+		for (i = 1; i < nr; i++) {
+			if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
+				break;
+			total_data += data_size[i];
+		}
+		nr = i;
+
+		if (old_data < data_end) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
+			       slot, old_data, data_end);
+			BUG_ON(1);
+		}
+		/*
+		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+		 */
+		/* first correct the data pointers */
+		WARN_ON(leaf->map_token);
+		for (i = slot; i < nritems; i++) {
+			u32 ioff;
+
+			item = btrfs_item_nr(leaf, i);
+			if (!leaf->map_token) {
+				map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+			}
+
+			ioff = btrfs_item_offset(leaf, item);
+			btrfs_set_item_offset(leaf, item, ioff - total_data);
+		}
+		if (leaf->map_token) {
+			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+			leaf->map_token = NULL;
+		}
+
+		/* shift the items */
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
+			      btrfs_item_nr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_item));
+
+		/* shift the data */
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end - total_data, btrfs_leaf_data(leaf) +
+			      data_end, old_data - data_end);
+		data_end = old_data;
+	} else {
+		/*
+		 * this sucks but it has to be done, if we are inserting at
+		 * the end of the leaf only insert 1 of the items, since we
+		 * have no way of knowing whats on the next leaf and we'd have
+		 * to drop our current locks to figure it out
+		 */
+		nr = 1;
+	}
+
+	/* setup the item for the new data */
+	for (i = 0; i < nr; i++) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
+		btrfs_set_item_key(leaf, &disk_key, slot + i);
+		item = btrfs_item_nr(leaf, slot + i);
+		btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
+		data_end -= data_size[i];
+		btrfs_set_item_size(leaf, item, data_size[i]);
+	}
+	btrfs_set_header_nritems(leaf, nritems + nr);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (slot == 0) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
+		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
+
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+out:
+	if (!ret)
+		ret = nr;
+	return ret;
+}
+
+/*
+ * Given a key and some data, insert items into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ */
+int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path,
+			    struct btrfs_key *cpu_key, u32 *data_size,
+			    int nr)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	int ret = 0;
+	int slot;
+	int slot_orig;
+	int i;
+	u32 nritems;
+	u32 total_size = 0;
+	u32 total_data = 0;
+	unsigned int data_end;
+	struct btrfs_disk_key disk_key;
+
+	for (i = 0; i < nr; i++)
+		total_data += data_size[i];
+
+	total_size = total_data + (nr * sizeof(struct btrfs_item));
+	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
+	if (ret == 0)
+		return -EEXIST;
+	if (ret < 0)
+		goto out;
+
+	slot_orig = path->slots[0];
+	leaf = path->nodes[0];
+
+	nritems = btrfs_header_nritems(leaf);
+	data_end = leaf_data_end(root, leaf);
+
+	if (btrfs_leaf_free_space(root, leaf) < total_size) {
+		btrfs_print_leaf(root, leaf);
+		printk(KERN_CRIT "not enough freespace need %u have %d\n",
+		       total_size, btrfs_leaf_free_space(root, leaf));
+		BUG();
+	}
+
+	slot = path->slots[0];
+	BUG_ON(slot < 0);
+
+	if (slot != nritems) {
+		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
+
+		if (old_data < data_end) {
+			btrfs_print_leaf(root, leaf);
+			printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
+			       slot, old_data, data_end);
+			BUG_ON(1);
+		}
+		/*
+		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+		 */
+		/* first correct the data pointers */
+		WARN_ON(leaf->map_token);
+		for (i = slot; i < nritems; i++) {
+			u32 ioff;
+
+			item = btrfs_item_nr(leaf, i);
+			if (!leaf->map_token) {
+				map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+			}
+
+			ioff = btrfs_item_offset(leaf, item);
+			btrfs_set_item_offset(leaf, item, ioff - total_data);
+		}
+		if (leaf->map_token) {
+			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+			leaf->map_token = NULL;
+		}
+
+		/* shift the items */
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
+			      btrfs_item_nr_offset(slot),
+			      (nritems - slot) * sizeof(struct btrfs_item));
+
+		/* shift the data */
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end - total_data, btrfs_leaf_data(leaf) +
+			      data_end, old_data - data_end);
+		data_end = old_data;
+	}
+
+	/* setup the item for the new data */
+	for (i = 0; i < nr; i++) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
+		btrfs_set_item_key(leaf, &disk_key, slot + i);
+		item = btrfs_item_nr(leaf, slot + i);
+		btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
+		data_end -= data_size[i];
+		btrfs_set_item_size(leaf, item, data_size[i]);
+	}
+	btrfs_set_header_nritems(leaf, nritems + nr);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+	if (slot == 0) {
+		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
+		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
+
+	if (btrfs_leaf_free_space(root, leaf) < 0) {
+		btrfs_print_leaf(root, leaf);
+		BUG();
+	}
+out:
+	return ret;
+}
+
+/*
+ * Given a key and some data, insert an item into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ */
+int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
+		      *root, struct btrfs_key *cpu_key, void *data, u32
+		      data_size)
+{
+	int ret = 0;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	unsigned long ptr;
+
+	path = btrfs_alloc_path();
+	BUG_ON(!path);
+	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
+	if (!ret) {
+		leaf = path->nodes[0];
+		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+		write_extent_buffer(leaf, data, ptr, data_size);
+		btrfs_mark_buffer_dirty(leaf);
+	}
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * delete the pointer from a given node.
+ *
+ * the tree should have been previously balanced so the deletion does not
+ * empty a node.
+ */
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		   struct btrfs_path *path, int level, int slot)
+{
+	struct extent_buffer *parent = path->nodes[level];
+	u32 nritems;
+	int ret = 0;
+	int wret;
+
+	nritems = btrfs_header_nritems(parent);
+	if (slot != nritems - 1) {
+		memmove_extent_buffer(parent,
+			      btrfs_node_key_ptr_offset(slot),
+			      btrfs_node_key_ptr_offset(slot + 1),
+			      sizeof(struct btrfs_key_ptr) *
+			      (nritems - slot - 1));
+	}
+	nritems--;
+	btrfs_set_header_nritems(parent, nritems);
+	if (nritems == 0 && parent == root->node) {
+		BUG_ON(btrfs_header_level(root->node) != 1);
+		/* just turn the root into a leaf and break */
+		btrfs_set_header_level(root->node, 0);
+	} else if (slot == 0) {
+		struct btrfs_disk_key disk_key;
+
+		btrfs_node_key(parent, &disk_key, 0);
+		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
+		if (wret)
+			ret = wret;
+	}
+	btrfs_mark_buffer_dirty(parent);
+	return ret;
+}
+
+/*
+ * a helper function to delete the leaf pointed to by path->slots[1] and
+ * path->nodes[1].  bytenr is the node block pointer, but since the callers
+ * already know it, it is faster to have them pass it down than to
+ * read it out of the node again.
+ *
+ * This deletes the pointer in path->nodes[1] and frees the leaf
+ * block extent.  zero is returned if it all worked out, < 0 otherwise.
+ *
+ * The path must have already been setup for deleting the leaf, including
+ * all the proper balancing.  path->nodes[1] must be locked.
+ */
+noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path, u64 bytenr)
+{
+	int ret;
+	u64 root_gen = btrfs_header_generation(path->nodes[1]);
+
+	ret = del_ptr(trans, root, path, 1, path->slots[1]);
+	if (ret)
+		return ret;
+
+	ret = btrfs_free_extent(trans, root, bytenr,
+				btrfs_level_size(root, 0),
+				path->nodes[1]->start,
+				btrfs_header_owner(path->nodes[1]),
+				root_gen, 0, 1);
+	return ret;
+}
+/*
+ * delete the item at the leaf level in path.  If that empties
+ * the leaf, remove it from the tree
+ */
+int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		    struct btrfs_path *path, int slot, int nr)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_item *item;
+	int last_off;
+	int dsize = 0;
+	int ret = 0;
+	int wret;
+	int i;
+	u32 nritems;
+
+	leaf = path->nodes[0];
+	last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
+
+	for (i = 0; i < nr; i++)
+		dsize += btrfs_item_size_nr(leaf, slot + i);
+
+	nritems = btrfs_header_nritems(leaf);
+
+	if (slot + nr != nritems) {
+		int data_end = leaf_data_end(root, leaf);
+
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + dsize,
+			      btrfs_leaf_data(leaf) + data_end,
+			      last_off - data_end);
+
+		for (i = slot + nr; i < nritems; i++) {
+			u32 ioff;
+
+			item = btrfs_item_nr(leaf, i);
+			if (!leaf->map_token) {
+				map_extent_buffer(leaf, (unsigned long)item,
+					sizeof(struct btrfs_item),
+					&leaf->map_token, &leaf->kaddr,
+					&leaf->map_start, &leaf->map_len,
+					KM_USER1);
+			}
+			ioff = btrfs_item_offset(leaf, item);
+			btrfs_set_item_offset(leaf, item, ioff + dsize);
+		}
+
+		if (leaf->map_token) {
+			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
+			leaf->map_token = NULL;
+		}
+
+		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
+			      btrfs_item_nr_offset(slot + nr),
+			      sizeof(struct btrfs_item) *
+			      (nritems - slot - nr));
+	}
+	btrfs_set_header_nritems(leaf, nritems - nr);
+	nritems -= nr;
+
+	/* delete the leaf if we've emptied it */
+	if (nritems == 0) {
+		if (leaf == root->node) {
+			btrfs_set_header_level(leaf, 0);
+		} else {
+			ret = btrfs_del_leaf(trans, root, path, leaf->start);
+			BUG_ON(ret);
+		}
+	} else {
+		int used = leaf_space_used(leaf, 0, nritems);
+		if (slot == 0) {
+			struct btrfs_disk_key disk_key;
+
+			btrfs_item_key(leaf, &disk_key, 0);
+			wret = fixup_low_keys(trans, root, path,
+					      &disk_key, 1);
+			if (wret)
+				ret = wret;
+		}
+
+		/* delete the leaf if it is mostly empty */
+		if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
+			/* push_leaf_left fixes the path.
+			 * make sure the path still points to our leaf
+			 * for possible call to del_ptr below
+			 */
+			slot = path->slots[1];
+			extent_buffer_get(leaf);
+
+			wret = push_leaf_left(trans, root, path, 1, 1);
+			if (wret < 0 && wret != -ENOSPC)
+				ret = wret;
+
+			if (path->nodes[0] == leaf &&
+			    btrfs_header_nritems(leaf)) {
+				wret = push_leaf_right(trans, root, path, 1, 1);
+				if (wret < 0 && wret != -ENOSPC)
+					ret = wret;
+			}
+
+			if (btrfs_header_nritems(leaf) == 0) {
+				path->slots[1] = slot;
+				ret = btrfs_del_leaf(trans, root, path,
+						     leaf->start);
+				BUG_ON(ret);
+				free_extent_buffer(leaf);
+			} else {
+				/* if we're still in the path, make sure
+				 * we're dirty.  Otherwise, one of the
+				 * push_leaf functions must have already
+				 * dirtied this buffer
+				 */
+				if (path->nodes[0] == leaf)
+					btrfs_mark_buffer_dirty(leaf);
+				free_extent_buffer(leaf);
+			}
+		} else {
+			btrfs_mark_buffer_dirty(leaf);
+		}
+	}
+	return ret;
+}
+
+/*
+ * search the tree again to find a leaf with lesser keys
+ * returns 0 if it found something or 1 if there are no lesser leaves.
+ * returns < 0 on io errors.
+ *
+ * This may release the path, and so you may lose any locks held at the
+ * time you call it.
+ */
+int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
+{
+	struct btrfs_key key;
+	struct btrfs_disk_key found_key;
+	int ret;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
+
+	if (key.offset > 0)
+		key.offset--;
+	else if (key.type > 0)
+		key.type--;
+	else if (key.objectid > 0)
+		key.objectid--;
+	else
+		return 1;
+
+	btrfs_release_path(root, path);
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		return ret;
+	btrfs_item_key(path->nodes[0], &found_key, 0);
+	ret = comp_keys(&found_key, &key);
+	if (ret < 0)
+		return 0;
+	return 1;
+}
+
+/*
+ * A helper function to walk down the tree starting at min_key, and looking
+ * for nodes or leaves that are either in cache or have a minimum
+ * transaction id.  This is used by the btree defrag code, and tree logging
+ *
+ * This does not cow, but it does stuff the starting key it finds back
+ * into min_key, so you can call btrfs_search_slot with cow=1 on the
+ * key and get a writable path.
+ *
+ * This does lock as it descends, and path->keep_locks should be set
+ * to 1 by the caller.
+ *
+ * This honors path->lowest_level to prevent descent past a given level
+ * of the tree.
+ *
+ * min_trans indicates the oldest transaction that you are interested
+ * in walking through.  Any nodes or leaves older than min_trans are
+ * skipped over (without reading them).
+ *
+ * returns zero if something useful was found, < 0 on error and 1 if there
+ * was nothing in the tree that matched the search criteria.
+ */
+int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+			 struct btrfs_key *max_key,
+			 struct btrfs_path *path, int cache_only,
+			 u64 min_trans)
+{
+	struct extent_buffer *cur;
+	struct btrfs_key found_key;
+	int slot;
+	int sret;
+	u32 nritems;
+	int level;
+	int ret = 1;
+
+	WARN_ON(!path->keep_locks);
+again:
+	cur = btrfs_lock_root_node(root);
+	level = btrfs_header_level(cur);
+	WARN_ON(path->nodes[level]);
+	path->nodes[level] = cur;
+	path->locks[level] = 1;
+
+	if (btrfs_header_generation(cur) < min_trans) {
+		ret = 1;
+		goto out;
+	}
+	while (1) {
+		nritems = btrfs_header_nritems(cur);
+		level = btrfs_header_level(cur);
+		sret = bin_search(cur, min_key, level, &slot);
+
+		/* at the lowest level, we're done, setup the path and exit */
+		if (level == path->lowest_level) {
+			if (slot >= nritems)
+				goto find_next_key;
+			ret = 0;
+			path->slots[level] = slot;
+			btrfs_item_key_to_cpu(cur, &found_key, slot);
+			goto out;
+		}
+		if (sret && slot > 0)
+			slot--;
+		/*
+		 * check this node pointer against the cache_only and
+		 * min_trans parameters.  If it isn't in cache or is too
+		 * old, skip to the next one.
+		 */
+		while (slot < nritems) {
+			u64 blockptr;
+			u64 gen;
+			struct extent_buffer *tmp;
+			struct btrfs_disk_key disk_key;
+
+			blockptr = btrfs_node_blockptr(cur, slot);
+			gen = btrfs_node_ptr_generation(cur, slot);
+			if (gen < min_trans) {
+				slot++;
+				continue;
+			}
+			if (!cache_only)
+				break;
+
+			if (max_key) {
+				btrfs_node_key(cur, &disk_key, slot);
+				if (comp_keys(&disk_key, max_key) >= 0) {
+					ret = 1;
+					goto out;
+				}
+			}
+
+			tmp = btrfs_find_tree_block(root, blockptr,
+					    btrfs_level_size(root, level - 1));
+
+			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
+				free_extent_buffer(tmp);
+				break;
+			}
+			if (tmp)
+				free_extent_buffer(tmp);
+			slot++;
+		}
+find_next_key:
+		/*
+		 * we didn't find a candidate key in this node, walk forward
+		 * and find another one
+		 */
+		if (slot >= nritems) {
+			path->slots[level] = slot;
+			sret = btrfs_find_next_key(root, path, min_key, level,
+						  cache_only, min_trans);
+			if (sret == 0) {
+				btrfs_release_path(root, path);
+				goto again;
+			} else {
+				goto out;
+			}
+		}
+		/* save our key for returning back */
+		btrfs_node_key_to_cpu(cur, &found_key, slot);
+		path->slots[level] = slot;
+		if (level == path->lowest_level) {
+			ret = 0;
+			unlock_up(path, level, 1);
+			goto out;
+		}
+		cur = read_node_slot(root, cur, slot);
+
+		btrfs_tree_lock(cur);
+		path->locks[level - 1] = 1;
+		path->nodes[level - 1] = cur;
+		unlock_up(path, level, 1);
+	}
+out:
+	if (ret == 0)
+		memcpy(min_key, &found_key, sizeof(found_key));
+	return ret;
+}
+
+/*
+ * this is similar to btrfs_next_leaf, but does not try to preserve
+ * and fixup the path.  It looks for and returns the next key in the
+ * tree based on the current path and the cache_only and min_trans
+ * parameters.
+ *
+ * 0 is returned if another key is found, < 0 if there are any errors
+ * and 1 is returned if there are no higher keys in the tree
+ *
+ * path->keep_locks should be set to 1 on the search made before
+ * calling this function.
+ */
+int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
+			struct btrfs_key *key, int lowest_level,
+			int cache_only, u64 min_trans)
+{
+	int level = lowest_level;
+	int slot;
+	struct extent_buffer *c;
+
+	WARN_ON(!path->keep_locks);
+	while (level < BTRFS_MAX_LEVEL) {
+		if (!path->nodes[level])
+			return 1;
+
+		slot = path->slots[level] + 1;
+		c = path->nodes[level];
+next:
+		if (slot >= btrfs_header_nritems(c)) {
+			level++;
+			if (level == BTRFS_MAX_LEVEL)
+				return 1;
+			continue;
+		}
+		if (level == 0)
+			btrfs_item_key_to_cpu(c, key, slot);
+		else {
+			u64 blockptr = btrfs_node_blockptr(c, slot);
+			u64 gen = btrfs_node_ptr_generation(c, slot);
+
+			if (cache_only) {
+				struct extent_buffer *cur;
+				cur = btrfs_find_tree_block(root, blockptr,
+					    btrfs_level_size(root, level - 1));
+				if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
+					slot++;
+					if (cur)
+						free_extent_buffer(cur);
+					goto next;
+				}
+				free_extent_buffer(cur);
+			}
+			if (gen < min_trans) {
+				slot++;
+				goto next;
+			}
+			btrfs_node_key_to_cpu(c, key, slot);
+		}
+		return 0;
+	}
+	return 1;
+}
+
+/*
+ * search the tree again to find a leaf with greater keys
+ * returns 0 if it found something or 1 if there are no greater leaves.
+ * returns < 0 on io errors.
+ */
+int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
+{
+	int slot;
+	int level = 1;
+	struct extent_buffer *c;
+	struct extent_buffer *next = NULL;
+	struct btrfs_key key;
+	u32 nritems;
+	int ret;
+
+	nritems = btrfs_header_nritems(path->nodes[0]);
+	if (nritems == 0)
+		return 1;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
+
+	btrfs_release_path(root, path);
+	path->keep_locks = 1;
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	path->keep_locks = 0;
+
+	if (ret < 0)
+		return ret;
+
+	nritems = btrfs_header_nritems(path->nodes[0]);
+	/*
+	 * by releasing the path above we dropped all our locks.  A balance
+	 * could have added more items next to the key that used to be
+	 * at the very end of the block.  So, check again here and
+	 * advance the path if there are now more items available.
+	 */
+	if (nritems > 0 && path->slots[0] < nritems - 1) {
+		path->slots[0]++;
+		goto done;
+	}
+
+	while (level < BTRFS_MAX_LEVEL) {
+		if (!path->nodes[level])
+			return 1;
+
+		slot = path->slots[level] + 1;
+		c = path->nodes[level];
+		if (slot >= btrfs_header_nritems(c)) {
+			level++;
+			if (level == BTRFS_MAX_LEVEL)
+				return 1;
+			continue;
+		}
+
+		if (next) {
+			btrfs_tree_unlock(next);
+			free_extent_buffer(next);
+		}
+
+		if (level == 1 && (path->locks[1] || path->skip_locking) &&
+		    path->reada)
+			reada_for_search(root, path, level, slot, 0);
+
+		next = read_node_slot(root, c, slot);
+		if (!path->skip_locking) {
+			WARN_ON(!btrfs_tree_locked(c));
+			btrfs_tree_lock(next);
+		}
+		break;
+	}
+	path->slots[level] = slot;
+	while (1) {
+		level--;
+		c = path->nodes[level];
+		if (path->locks[level])
+			btrfs_tree_unlock(c);
+		free_extent_buffer(c);
+		path->nodes[level] = next;
+		path->slots[level] = 0;
+		if (!path->skip_locking)
+			path->locks[level] = 1;
+		if (!level)
+			break;
+		if (level == 1 && path->locks[1] && path->reada)
+			reada_for_search(root, path, level, slot, 0);
+		next = read_node_slot(root, next, 0);
+		if (!path->skip_locking) {
+			WARN_ON(!btrfs_tree_locked(path->nodes[level]));
+			btrfs_tree_lock(next);
+		}
+	}
+done:
+	unlock_up(path, 0, 1);
+	return 0;
+}
+
+/*
+ * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
+ * searching until it gets past min_objectid or finds an item of 'type'
+ *
+ * returns 0 if something is found, 1 if nothing was found and < 0 on error
+ */
+int btrfs_previous_item(struct btrfs_root *root,
+			struct btrfs_path *path, u64 min_objectid,
+			int type)
+{
+	struct btrfs_key found_key;
+	struct extent_buffer *leaf;
+	u32 nritems;
+	int ret;
+
+	while (1) {
+		if (path->slots[0] == 0) {
+			ret = btrfs_prev_leaf(root, path);
+			if (ret != 0)
+				return ret;
+		} else {
+			path->slots[0]--;
+		}
+		leaf = path->nodes[0];
+		nritems = btrfs_header_nritems(leaf);
+		if (nritems == 0)
+			return 1;
+		if (path->slots[0] == nritems)
+			path->slots[0]--;
+
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+		if (found_key.type == type)
+			return 0;
+		if (found_key.objectid < min_objectid)
+			break;
+		if (found_key.objectid == min_objectid &&
+		    found_key.type < type)
+			break;
+	}
+	return 1;
+}
#54 lore raw

PATCH 1/1 Btrfs extent io operations

In reply to Chris Mason

extent_io includes two interfaces. Each one should live in its own file, but for now they share a single one.

extent_state: These track state bits across a range of bytes instead of a single page. So, btrfs can say bytes 0-1MB are marked for delalloc, and do searches based on the ranges. An rbtree is used to track state per inode.

This code also implements readpages and writepages based on the extent_state code. No buffer heads are used, even as temporary containers for IO. This only works with page sized IO today.

fs/btrfs/extent_io.c new file +2795 -0
@@ -0,0 +1,2795 @@
+#include <linux/bitops.h>
+#include <linux/slab.h>
+#include <linux/bio.h>
+#include <linux/mm.h>
+#include <linux/gfp.h>
+#include <linux/pagemap.h>
+#include <linux/page-flags.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/blkdev.h>
+#include <linux/swap.h>
+#include <linux/version.h>
+#include <linux/writeback.h>
+#include <linux/pagevec.h>
+#include "extent_io.h"
+#include "extent_map.h"
+#include "compat.h"
+#include "ctree.h"
+#include "btrfs_inode.h"
+
+/* temporary define until extent_map moves out of btrfs */
+struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
+				       unsigned long extra_flags,
+				       void (*ctor)(void *, struct kmem_cache *,
+						    unsigned long));
+
+static struct kmem_cache *extent_state_cache;
+
+static LIST_HEAD(buffers);
+static LIST_HEAD(states);
+
+#define LEAK_DEBUG 0
+#ifdef LEAK_DEBUG
+static DEFINE_SPINLOCK(leak_lock);
+#endif
+
+#define BUFFER_LRU_MAX 64
+
+struct tree_entry {
+	u64 start;
+	u64 end;
+	struct rb_node rb_node;
+};
+
+struct extent_page_data {
+	struct bio *bio;
+	struct extent_io_tree *tree;
+	get_extent_t *get_extent;
+
+	/* tells writepage not to lock the state bits for this range
+	 * it still does the unlocking
+	 */
+	int extent_locked;
+};
+
+int __init extent_io_init(void)
+{
+	extent_state_cache = btrfs_cache_create("extent_state",
+					    sizeof(struct extent_state), 0,
+					    NULL);
+	if (!extent_state_cache)
+		return -ENOMEM;
+
+	return 0;
+
+free_state_cache:
+	kmem_cache_destroy(extent_state_cache);
+	return -ENOMEM;
+}
+
+void extent_io_exit(void)
+{
+	struct extent_state *state;
+
+	while (!list_empty(&states)) {
+		state = list_entry(states.next, struct extent_state, leak_list);
+		printk(KERN_ERR "btrfs state leak: start %llu end %llu "
+		       "state %lu in tree %p refs %d\n",
+		       (unsigned long long)state->start,
+		       (unsigned long long)state->end,
+		       state->state, state->tree, atomic_read(&state->refs));
+		list_del(&state->leak_list);
+		kmem_cache_free(extent_state_cache, state);
+
+	}
+
+	if (extent_state_cache)
+		kmem_cache_destroy(extent_state_cache);
+}
+
+void extent_io_tree_init(struct extent_io_tree *tree,
+			  struct address_space *mapping, gfp_t mask)
+{
+	tree->state.rb_node = NULL;
+	tree->buffer.rb_node = NULL;
+	tree->ops = NULL;
+	tree->dirty_bytes = 0;
+	spin_lock_init(&tree->lock);
+	spin_lock_init(&tree->buffer_lock);
+	tree->mapping = mapping;
+}
+
+static struct extent_state *alloc_extent_state(gfp_t mask)
+{
+	struct extent_state *state;
+#ifdef LEAK_DEBUG
+	unsigned long flags;
+#endif
+
+	state = kmem_cache_alloc(extent_state_cache, mask);
+	if (!state)
+		return state;
+	state->state = 0;
+	state->private = 0;
+	state->tree = NULL;
+#ifdef LEAK_DEBUG
+	spin_lock_irqsave(&leak_lock, flags);
+	list_add(&state->leak_list, &states);
+	spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+	atomic_set(&state->refs, 1);
+	init_waitqueue_head(&state->wq);
+	return state;
+}
+
+static void free_extent_state(struct extent_state *state)
+{
+	if (!state)
+		return;
+	if (atomic_dec_and_test(&state->refs)) {
+#ifdef LEAK_DEBUG
+		unsigned long flags;
+#endif
+		WARN_ON(state->tree);
+#ifdef LEAK_DEBUG
+		spin_lock_irqsave(&leak_lock, flags);
+		list_del(&state->leak_list);
+		spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+		kmem_cache_free(extent_state_cache, state);
+	}
+}
+
+static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
+				   struct rb_node *node)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct tree_entry *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct tree_entry, rb_node);
+
+		if (offset < entry->start)
+			p = &(*p)->rb_left;
+		else if (offset > entry->end)
+			p = &(*p)->rb_right;
+		else
+			return parent;
+	}
+
+	entry = rb_entry(node, struct tree_entry, rb_node);
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
+				     struct rb_node **prev_ret,
+				     struct rb_node **next_ret)
+{
+	struct rb_root *root = &tree->state;
+	struct rb_node *n = root->rb_node;
+	struct rb_node *prev = NULL;
+	struct rb_node *orig_prev = NULL;
+	struct tree_entry *entry;
+	struct tree_entry *prev_entry = NULL;
+
+	while (n) {
+		entry = rb_entry(n, struct tree_entry, rb_node);
+		prev = n;
+		prev_entry = entry;
+
+		if (offset < entry->start)
+			n = n->rb_left;
+		else if (offset > entry->end)
+			n = n->rb_right;
+		else
+			return n;
+	}
+
+	if (prev_ret) {
+		orig_prev = prev;
+		while (prev && offset > prev_entry->end) {
+			prev = rb_next(prev);
+			prev_entry = rb_entry(prev, struct tree_entry, rb_node);
+		}
+		*prev_ret = prev;
+		prev = orig_prev;
+	}
+
+	if (next_ret) {
+		prev_entry = rb_entry(prev, struct tree_entry, rb_node);
+		while (prev && offset < prev_entry->start) {
+			prev = rb_prev(prev);
+			prev_entry = rb_entry(prev, struct tree_entry, rb_node);
+		}
+		*next_ret = prev;
+	}
+	return NULL;
+}
+
+static inline struct rb_node *tree_search(struct extent_io_tree *tree,
+					  u64 offset)
+{
+	struct rb_node *prev = NULL;
+	struct rb_node *ret;
+
+	ret = __etree_search(tree, offset, &prev, NULL);
+	if (!ret)
+		return prev;
+	return ret;
+}
+
+/*
+ * utility function to look for merge candidates inside a given range.
+ * Any extents with matching state are merged together into a single
+ * extent in the tree.  Extents with EXTENT_IO in their state field
+ * are not merged because the end_io handlers need to be able to do
+ * operations on them without sleeping (or doing allocations/splits).
+ *
+ * This should be called with the tree lock held.
+ */
+static int merge_state(struct extent_io_tree *tree,
+		       struct extent_state *state)
+{
+	struct extent_state *other;
+	struct rb_node *other_node;
+
+	if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
+		return 0;
+
+	other_node = rb_prev(&state->rb_node);
+	if (other_node) {
+		other = rb_entry(other_node, struct extent_state, rb_node);
+		if (other->end == state->start - 1 &&
+		    other->state == state->state) {
+			state->start = other->start;
+			other->tree = NULL;
+			rb_erase(&other->rb_node, &tree->state);
+			free_extent_state(other);
+		}
+	}
+	other_node = rb_next(&state->rb_node);
+	if (other_node) {
+		other = rb_entry(other_node, struct extent_state, rb_node);
+		if (other->start == state->end + 1 &&
+		    other->state == state->state) {
+			other->start = state->start;
+			state->tree = NULL;
+			rb_erase(&state->rb_node, &tree->state);
+			free_extent_state(state);
+		}
+	}
+	return 0;
+}
+
+static void set_state_cb(struct extent_io_tree *tree,
+			 struct extent_state *state,
+			 unsigned long bits)
+{
+	if (tree->ops && tree->ops->set_bit_hook) {
+		tree->ops->set_bit_hook(tree->mapping->host, state->start,
+					state->end, state->state, bits);
+	}
+}
+
+static void clear_state_cb(struct extent_io_tree *tree,
+			   struct extent_state *state,
+			   unsigned long bits)
+{
+	if (tree->ops && tree->ops->clear_bit_hook) {
+		tree->ops->clear_bit_hook(tree->mapping->host, state->start,
+					  state->end, state->state, bits);
+	}
+}
+
+/*
+ * insert an extent_state struct into the tree.  'bits' are set on the
+ * struct before it is inserted.
+ *
+ * This may return -EEXIST if the extent is already there, in which case the
+ * state struct is freed.
+ *
+ * The tree lock is not taken internally.  This is a utility function and
+ * probably isn't what you want to call (see set/clear_extent_bit).
+ */
+static int insert_state(struct extent_io_tree *tree,
+			struct extent_state *state, u64 start, u64 end,
+			int bits)
+{
+	struct rb_node *node;
+
+	if (end < start) {
+		printk(KERN_ERR "btrfs end < start %llu %llu\n",
+		       (unsigned long long)end,
+		       (unsigned long long)start);
+		WARN_ON(1);
+	}
+	if (bits & EXTENT_DIRTY)
+		tree->dirty_bytes += end - start + 1;
+	set_state_cb(tree, state, bits);
+	state->state |= bits;
+	state->start = start;
+	state->end = end;
+	node = tree_insert(&tree->state, end, &state->rb_node);
+	if (node) {
+		struct extent_state *found;
+		found = rb_entry(node, struct extent_state, rb_node);
+		printk(KERN_ERR "btrfs found node %llu %llu on insert of "
+		       "%llu %llu\n", (unsigned long long)found->start,
+		       (unsigned long long)found->end,
+		       (unsigned long long)start, (unsigned long long)end);
+		free_extent_state(state);
+		return -EEXIST;
+	}
+	state->tree = tree;
+	merge_state(tree, state);
+	return 0;
+}
+
+/*
+ * split a given extent state struct in two, inserting the preallocated
+ * struct 'prealloc' as the newly created second half.  'split' indicates an
+ * offset inside 'orig' where it should be split.
+ *
+ * Before calling,
+ * the tree has 'orig' at [orig->start, orig->end].  After calling, there
+ * are two extent state structs in the tree:
+ * prealloc: [orig->start, split - 1]
+ * orig: [ split, orig->end ]
+ *
+ * The tree locks are not taken by this function. They need to be held
+ * by the caller.
+ */
+static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
+		       struct extent_state *prealloc, u64 split)
+{
+	struct rb_node *node;
+	prealloc->start = orig->start;
+	prealloc->end = split - 1;
+	prealloc->state = orig->state;
+	orig->start = split;
+
+	node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
+	if (node) {
+		struct extent_state *found;
+		found = rb_entry(node, struct extent_state, rb_node);
+		free_extent_state(prealloc);
+		return -EEXIST;
+	}
+	prealloc->tree = tree;
+	return 0;
+}
+
+/*
+ * utility function to clear some bits in an extent state struct.
+ * it will optionally wake up any one waiting on this state (wake == 1), or
+ * forcibly remove the state from the tree (delete == 1).
+ *
+ * If no bits are set on the state struct after clearing things, the
+ * struct is freed and removed from the tree
+ */
+static int clear_state_bit(struct extent_io_tree *tree,
+			    struct extent_state *state, int bits, int wake,
+			    int delete)
+{
+	int ret = state->state & bits;
+
+	if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
+		u64 range = state->end - state->start + 1;
+		WARN_ON(range > tree->dirty_bytes);
+		tree->dirty_bytes -= range;
+	}
+	clear_state_cb(tree, state, bits);
+	state->state &= ~bits;
+	if (wake)
+		wake_up(&state->wq);
+	if (delete || state->state == 0) {
+		if (state->tree) {
+			clear_state_cb(tree, state, state->state);
+			rb_erase(&state->rb_node, &tree->state);
+			state->tree = NULL;
+			free_extent_state(state);
+		} else {
+			WARN_ON(1);
+		}
+	} else {
+		merge_state(tree, state);
+	}
+	return ret;
+}
+
+/*
+ * clear some bits on a range in the tree.  This may require splitting
+ * or inserting elements in the tree, so the gfp mask is used to
+ * indicate which allocations or sleeping are allowed.
+ *
+ * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
+ * the given range from the tree regardless of state (ie for truncate).
+ *
+ * the range [start, end] is inclusive.
+ *
+ * This takes the tree lock, and returns < 0 on error, > 0 if any of the
+ * bits were already set, or zero if none of the bits were already set.
+ */
+int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
+		     int bits, int wake, int delete, gfp_t mask)
+{
+	struct extent_state *state;
+	struct extent_state *prealloc = NULL;
+	struct rb_node *node;
+	int err;
+	int set = 0;
+
+again:
+	if (!prealloc && (mask & __GFP_WAIT)) {
+		prealloc = alloc_extent_state(mask);
+		if (!prealloc)
+			return -ENOMEM;
+	}
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find the extents that end after
+	 * our range starts
+	 */
+	node = tree_search(tree, start);
+	if (!node)
+		goto out;
+	state = rb_entry(node, struct extent_state, rb_node);
+	if (state->start > end)
+		goto out;
+	WARN_ON(state->end < start);
+
+	/*
+	 *     | ---- desired range ---- |
+	 *  | state | or
+	 *  | ------------- state -------------- |
+	 *
+	 * We need to split the extent we found, and may flip
+	 * bits on second half.
+	 *
+	 * If the extent we found extends past our range, we
+	 * just split and search again.  It'll get split again
+	 * the next time though.
+	 *
+	 * If the extent we found is inside our range, we clear
+	 * the desired bit on it.
+	 */
+
+	if (state->start < start) {
+		if (!prealloc)
+			prealloc = alloc_extent_state(GFP_ATOMIC);
+		err = split_state(tree, state, prealloc, start);
+		BUG_ON(err == -EEXIST);
+		prealloc = NULL;
+		if (err)
+			goto out;
+		if (state->end <= end) {
+			start = state->end + 1;
+			set |= clear_state_bit(tree, state, bits,
+					wake, delete);
+		} else {
+			start = state->start;
+		}
+		goto search_again;
+	}
+	/*
+	 * | ---- desired range ---- |
+	 *                        | state |
+	 * We need to split the extent, and clear the bit
+	 * on the first half
+	 */
+	if (state->start <= end && state->end > end) {
+		if (!prealloc)
+			prealloc = alloc_extent_state(GFP_ATOMIC);
+		err = split_state(tree, state, prealloc, end + 1);
+		BUG_ON(err == -EEXIST);
+
+		if (wake)
+			wake_up(&state->wq);
+		set |= clear_state_bit(tree, prealloc, bits,
+				       wake, delete);
+		prealloc = NULL;
+		goto out;
+	}
+
+	start = state->end + 1;
+	set |= clear_state_bit(tree, state, bits, wake, delete);
+	goto search_again;
+
+out:
+	spin_unlock(&tree->lock);
+	if (prealloc)
+		free_extent_state(prealloc);
+
+	return set;
+
+search_again:
+	if (start > end)
+		goto out;
+	spin_unlock(&tree->lock);
+	if (mask & __GFP_WAIT)
+		cond_resched();
+	goto again;
+}
+
+static int wait_on_state(struct extent_io_tree *tree,
+			 struct extent_state *state)
+		__releases(tree->lock)
+		__acquires(tree->lock)
+{
+	DEFINE_WAIT(wait);
+	prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
+	spin_unlock(&tree->lock);
+	schedule();
+	spin_lock(&tree->lock);
+	finish_wait(&state->wq, &wait);
+	return 0;
+}
+
+/*
+ * waits for one or more bits to clear on a range in the state tree.
+ * The range [start, end] is inclusive.
+ * The tree lock is taken by this function
+ */
+int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
+{
+	struct extent_state *state;
+	struct rb_node *node;
+
+	spin_lock(&tree->lock);
+again:
+	while (1) {
+		/*
+		 * this search will find all the extents that end after
+		 * our range starts
+		 */
+		node = tree_search(tree, start);
+		if (!node)
+			break;
+
+		state = rb_entry(node, struct extent_state, rb_node);
+
+		if (state->start > end)
+			goto out;
+
+		if (state->state & bits) {
+			start = state->start;
+			atomic_inc(&state->refs);
+			wait_on_state(tree, state);
+			free_extent_state(state);
+			goto again;
+		}
+		start = state->end + 1;
+
+		if (start > end)
+			break;
+
+		if (need_resched()) {
+			spin_unlock(&tree->lock);
+			cond_resched();
+			spin_lock(&tree->lock);
+		}
+	}
+out:
+	spin_unlock(&tree->lock);
+	return 0;
+}
+
+static void set_state_bits(struct extent_io_tree *tree,
+			   struct extent_state *state,
+			   int bits)
+{
+	if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
+		u64 range = state->end - state->start + 1;
+		tree->dirty_bytes += range;
+	}
+	set_state_cb(tree, state, bits);
+	state->state |= bits;
+}
+
+/*
+ * set some bits on a range in the tree.  This may require allocations
+ * or sleeping, so the gfp mask is used to indicate what is allowed.
+ *
+ * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
+ * range already has the desired bits set.  The start of the existing
+ * range is returned in failed_start in this case.
+ *
+ * [start, end] is inclusive
+ * This takes the tree lock.
+ */
+static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
+			  int bits, int exclusive, u64 *failed_start,
+			  gfp_t mask)
+{
+	struct extent_state *state;
+	struct extent_state *prealloc = NULL;
+	struct rb_node *node;
+	int err = 0;
+	int set;
+	u64 last_start;
+	u64 last_end;
+again:
+	if (!prealloc && (mask & __GFP_WAIT)) {
+		prealloc = alloc_extent_state(mask);
+		if (!prealloc)
+			return -ENOMEM;
+	}
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node) {
+		err = insert_state(tree, prealloc, start, end, bits);
+		prealloc = NULL;
+		BUG_ON(err == -EEXIST);
+		goto out;
+	}
+
+	state = rb_entry(node, struct extent_state, rb_node);
+	last_start = state->start;
+	last_end = state->end;
+
+	/*
+	 * | ---- desired range ---- |
+	 * | state |
+	 *
+	 * Just lock what we found and keep going
+	 */
+	if (state->start == start && state->end <= end) {
+		set = state->state & bits;
+		if (set && exclusive) {
+			*failed_start = state->start;
+			err = -EEXIST;
+			goto out;
+		}
+		set_state_bits(tree, state, bits);
+		start = state->end + 1;
+		merge_state(tree, state);
+		goto search_again;
+	}
+
+	/*
+	 *     | ---- desired range ---- |
+	 * | state |
+	 *   or
+	 * | ------------- state -------------- |
+	 *
+	 * We need to split the extent we found, and may flip bits on
+	 * second half.
+	 *
+	 * If the extent we found extends past our
+	 * range, we just split and search again.  It'll get split
+	 * again the next time though.
+	 *
+	 * If the extent we found is inside our range, we set the
+	 * desired bit on it.
+	 */
+	if (state->start < start) {
+		set = state->state & bits;
+		if (exclusive && set) {
+			*failed_start = start;
+			err = -EEXIST;
+			goto out;
+		}
+		err = split_state(tree, state, prealloc, start);
+		BUG_ON(err == -EEXIST);
+		prealloc = NULL;
+		if (err)
+			goto out;
+		if (state->end <= end) {
+			set_state_bits(tree, state, bits);
+			start = state->end + 1;
+			merge_state(tree, state);
+		} else {
+			start = state->start;
+		}
+		goto search_again;
+	}
+	/*
+	 * | ---- desired range ---- |
+	 *     | state | or               | state |
+	 *
+	 * There's a hole, we need to insert something in it and
+	 * ignore the extent we found.
+	 */
+	if (state->start > start) {
+		u64 this_end;
+		if (end < last_start)
+			this_end = end;
+		else
+			this_end = last_start - 1;
+		err = insert_state(tree, prealloc, start, this_end,
+				   bits);
+		prealloc = NULL;
+		BUG_ON(err == -EEXIST);
+		if (err)
+			goto out;
+		start = this_end + 1;
+		goto search_again;
+	}
+	/*
+	 * | ---- desired range ---- |
+	 *                        | state |
+	 * We need to split the extent, and set the bit
+	 * on the first half
+	 */
+	if (state->start <= end && state->end > end) {
+		set = state->state & bits;
+		if (exclusive && set) {
+			*failed_start = start;
+			err = -EEXIST;
+			goto out;
+		}
+		err = split_state(tree, state, prealloc, end + 1);
+		BUG_ON(err == -EEXIST);
+
+		set_state_bits(tree, prealloc, bits);
+		merge_state(tree, prealloc);
+		prealloc = NULL;
+		goto out;
+	}
+
+	goto search_again;
+
+out:
+	spin_unlock(&tree->lock);
+	if (prealloc)
+		free_extent_state(prealloc);
+
+	return err;
+
+search_again:
+	if (start > end)
+		goto out;
+	spin_unlock(&tree->lock);
+	if (mask & __GFP_WAIT)
+		cond_resched();
+	goto again;
+}
+
+/* wrappers around set/clear extent bit */
+int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
+			      mask);
+}
+
+int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
+		       gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
+}
+
+int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
+		    int bits, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, bits, 0, NULL,
+			      mask);
+}
+
+int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
+		      int bits, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
+}
+
+int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end,
+			      EXTENT_DELALLOC | EXTENT_DIRTY,
+			      0, NULL, mask);
+}
+
+int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
+		       gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end,
+				EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
+}
+
+int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
+			 gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
+}
+
+int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
+		     gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
+			      mask);
+}
+
+static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
+		       gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
+}
+
+int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
+			gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
+			      mask);
+}
+
+static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
+				 u64 end, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
+}
+
+static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
+			 gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
+			      0, NULL, mask);
+}
+
+static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
+				  u64 end, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
+}
+
+int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
+}
+
+/*
+ * either insert or lock state struct between start and end use mask to tell
+ * us if waiting is desired.
+ */
+int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
+{
+	int err;
+	u64 failed_start;
+	while (1) {
+		err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
+				     &failed_start, mask);
+		if (err == -EEXIST && (mask & __GFP_WAIT)) {
+			wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
+			start = failed_start;
+		} else {
+			break;
+		}
+		WARN_ON(start > end);
+	}
+	return err;
+}
+
+int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
+		    gfp_t mask)
+{
+	int err;
+	u64 failed_start;
+
+	err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
+			     &failed_start, mask);
+	if (err == -EEXIST) {
+		if (failed_start > start)
+			clear_extent_bit(tree, start, failed_start - 1,
+					 EXTENT_LOCKED, 1, 0, mask);
+		return 0;
+	}
+	return 1;
+}
+
+int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
+		  gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
+}
+
+/*
+ * helper function to set pages and extents in the tree dirty
+ */
+int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+
+	while (index <= end_index) {
+		page = find_get_page(tree->mapping, index);
+		BUG_ON(!page);
+		__set_page_dirty_nobuffers(page);
+		page_cache_release(page);
+		index++;
+	}
+	set_extent_dirty(tree, start, end, GFP_NOFS);
+	return 0;
+}
+
+/*
+ * helper function to set both pages and extents in the tree writeback
+ */
+static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+
+	while (index <= end_index) {
+		page = find_get_page(tree->mapping, index);
+		BUG_ON(!page);
+		set_page_writeback(page);
+		page_cache_release(page);
+		index++;
+	}
+	set_extent_writeback(tree, start, end, GFP_NOFS);
+	return 0;
+}
+
+/*
+ * find the first offset in the io tree with 'bits' set. zero is
+ * returned if we find something, and *start_ret and *end_ret are
+ * set to reflect the state struct that was found.
+ *
+ * If nothing was found, 1 is returned, < 0 on error
+ */
+int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
+			  u64 *start_ret, u64 *end_ret, int bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	int ret = 1;
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node)
+		goto out;
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->end >= start && (state->state & bits)) {
+			*start_ret = state->start;
+			*end_ret = state->end;
+			ret = 0;
+			break;
+		}
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+/* find the first state struct with 'bits' set after 'start', and
+ * return it.  tree->lock must be held.  NULL will returned if
+ * nothing was found after 'start'
+ */
+struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
+						 u64 start, int bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node)
+		goto out;
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->end >= start && (state->state & bits))
+			return state;
+
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	return NULL;
+}
+
+/*
+ * find a contiguous range of bytes in the file marked as delalloc, not
+ * more than 'max_bytes'.  start and end are used to return the range,
+ *
+ * 1 is returned if we find something, 0 if nothing was in the tree
+ */
+static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
+					u64 *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;
+
+	spin_lock(&tree->lock);
+
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, cur_start);
+	if (!node) {
+		if (!found)
+			*end = (u64)-1;
+		goto out;
+	}
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (found && (state->start != cur_start ||
+			      (state->state & EXTENT_BOUNDARY))) {
+			goto out;
+		}
+		if (!(state->state & EXTENT_DELALLOC)) {
+			if (!found)
+				*end = state->end;
+			goto out;
+		}
+		if (!found)
+			*start = state->start;
+		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:
+	spin_unlock(&tree->lock);
+	return found;
+}
+
+static noinline int __unlock_for_delalloc(struct inode *inode,
+					  struct page *locked_page,
+					  u64 start, u64 end)
+{
+	int ret;
+	struct page *pages[16];
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	unsigned long nr_pages = end_index - index + 1;
+	int i;
+
+	if (index == locked_page->index && end_index == index)
+		return 0;
+
+	while (nr_pages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long, nr_pages,
+				     ARRAY_SIZE(pages)), pages);
+		for (i = 0; i < ret; i++) {
+			if (pages[i] != locked_page)
+				unlock_page(pages[i]);
+			page_cache_release(pages[i]);
+		}
+		nr_pages -= ret;
+		index += ret;
+		cond_resched();
+	}
+	return 0;
+}
+
+static noinline int lock_delalloc_pages(struct inode *inode,
+					struct page *locked_page,
+					u64 delalloc_start,
+					u64 delalloc_end)
+{
+	unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
+	unsigned long start_index = index;
+	unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
+	unsigned long pages_locked = 0;
+	struct page *pages[16];
+	unsigned long nrpages;
+	int ret;
+	int i;
+
+	/* the caller is responsible for locking the start index */
+	if (index == locked_page->index && index == end_index)
+		return 0;
+
+	/* skip the page at the start index */
+	nrpages = end_index - index + 1;
+	while (nrpages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long,
+				     nrpages, ARRAY_SIZE(pages)), pages);
+		if (ret == 0) {
+			ret = -EAGAIN;
+			goto done;
+		}
+		/* now we have an array of pages, lock them all */
+		for (i = 0; i < ret; i++) {
+			/*
+			 * the caller is taking responsibility for
+			 * locked_page
+			 */
+			if (pages[i] != locked_page) {
+				lock_page(pages[i]);
+				if (!PageDirty(pages[i]) ||
+				    pages[i]->mapping != inode->i_mapping) {
+					ret = -EAGAIN;
+					unlock_page(pages[i]);
+					page_cache_release(pages[i]);
+					goto done;
+				}
+			}
+			page_cache_release(pages[i]);
+			pages_locked++;
+		}
+		nrpages -= ret;
+		index += ret;
+		cond_resched();
+	}
+	ret = 0;
+done:
+	if (ret && pages_locked) {
+		__unlock_for_delalloc(inode, locked_page,
+			      delalloc_start,
+			      ((u64)(start_index + pages_locked - 1)) <<
+			      PAGE_CACHE_SHIFT);
+	}
+	return ret;
+}
+
+/*
+ * find a contiguous range of bytes in the file marked as delalloc, not
+ * more than 'max_bytes'.  start and end are used to return the range,
+ *
+ * 1 is returned if we find something, 0 if nothing was in the tree
+ */
+static noinline u64 find_lock_delalloc_range(struct inode *inode,
+					     struct extent_io_tree *tree,
+					     struct page *locked_page,
+					     u64 *start, u64 *end,
+					     u64 max_bytes)
+{
+	u64 delalloc_start;
+	u64 delalloc_end;
+	u64 found;
+	int ret;
+	int loops = 0;
+
+again:
+	/* step one, find a bunch of delalloc bytes starting at start */
+	delalloc_start = *start;
+	delalloc_end = 0;
+	found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
+				    max_bytes);
+	if (!found || delalloc_end <= *start) {
+		*start = delalloc_start;
+		*end = delalloc_end;
+		return found;
+	}
+
+	/*
+	 * start comes from the offset of locked_page.  We have to lock
+	 * pages in order, so we can't process delalloc bytes before
+	 * locked_page
+	 */
+	if (delalloc_start < *start)
+		delalloc_start = *start;
+
+	/*
+	 * make sure to limit the number of pages we try to lock down
+	 * if we're looping.
+	 */
+	if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
+		delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
+
+	/* step two, lock all the pages after the page that has start */
+	ret = lock_delalloc_pages(inode, locked_page,
+				  delalloc_start, delalloc_end);
+	if (ret == -EAGAIN) {
+		/* some of the pages are gone, lets avoid looping by
+		 * shortening the size of the delalloc range we're searching
+		 */
+		if (!loops) {
+			unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
+			max_bytes = PAGE_CACHE_SIZE - offset;
+			loops = 1;
+			goto again;
+		} else {
+			found = 0;
+			goto out_failed;
+		}
+	}
+	BUG_ON(ret);
+
+	/* step three, lock the state bits for the whole range */
+	lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
+
+	/* then test to make sure it is all still delalloc */
+	ret = test_range_bit(tree, delalloc_start, delalloc_end,
+			     EXTENT_DELALLOC, 1);
+	if (!ret) {
+		unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
+		__unlock_for_delalloc(inode, locked_page,
+			      delalloc_start, delalloc_end);
+		cond_resched();
+		goto again;
+	}
+	*start = delalloc_start;
+	*end = delalloc_end;
+out_failed:
+	return found;
+}
+
+int extent_clear_unlock_delalloc(struct inode *inode,
+				struct extent_io_tree *tree,
+				u64 start, u64 end, struct page *locked_page,
+				int unlock_pages,
+				int clear_unlock,
+				int clear_delalloc, int clear_dirty,
+				int set_writeback,
+				int end_writeback)
+{
+	int ret;
+	struct page *pages[16];
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	unsigned long nr_pages = end_index - index + 1;
+	int i;
+	int clear_bits = 0;
+
+	if (clear_unlock)
+		clear_bits |= EXTENT_LOCKED;
+	if (clear_dirty)
+		clear_bits |= EXTENT_DIRTY;
+
+	if (clear_delalloc)
+		clear_bits |= EXTENT_DELALLOC;
+
+	clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
+	if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
+		return 0;
+
+	while (nr_pages > 0) {
+		ret = find_get_pages_contig(inode->i_mapping, index,
+				     min_t(unsigned long,
+				     nr_pages, ARRAY_SIZE(pages)), pages);
+		for (i = 0; i < ret; i++) {
+			if (pages[i] == locked_page) {
+				page_cache_release(pages[i]);
+				continue;
+			}
+			if (clear_dirty)
+				clear_page_dirty_for_io(pages[i]);
+			if (set_writeback)
+				set_page_writeback(pages[i]);
+			if (end_writeback)
+				end_page_writeback(pages[i]);
+			if (unlock_pages)
+				unlock_page(pages[i]);
+			page_cache_release(pages[i]);
+		}
+		nr_pages -= ret;
+		index += ret;
+		cond_resched();
+	}
+	return 0;
+}
+
+/*
+ * count the number of bytes in the tree that have a given bit(s)
+ * set.  This can be fairly slow, except for EXTENT_DIRTY which is
+ * cached.  The total number found is returned.
+ */
+u64 count_range_bits(struct extent_io_tree *tree,
+		     u64 *start, u64 search_end, u64 max_bytes,
+		     unsigned long bits)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	u64 cur_start = *start;
+	u64 total_bytes = 0;
+	int found = 0;
+
+	if (search_end <= cur_start) {
+		WARN_ON(1);
+		return 0;
+	}
+
+	spin_lock(&tree->lock);
+	if (cur_start == 0 && bits == EXTENT_DIRTY) {
+		total_bytes = tree->dirty_bytes;
+		goto out;
+	}
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, cur_start);
+	if (!node)
+		goto out;
+
+	while (1) {
+		state = rb_entry(node, struct extent_state, rb_node);
+		if (state->start > search_end)
+			break;
+		if (state->end >= cur_start && (state->state & bits)) {
+			total_bytes += min(search_end, state->end) + 1 -
+				       max(cur_start, state->start);
+			if (total_bytes >= max_bytes)
+				break;
+			if (!found) {
+				*start = state->start;
+				found = 1;
+			}
+		}
+		node = rb_next(node);
+		if (!node)
+			break;
+	}
+out:
+	spin_unlock(&tree->lock);
+	return total_bytes;
+}
+
+#if 0
+/*
+ * helper function to lock both pages and extents in the tree.
+ * pages must be locked first.
+ */
+static int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+	int err;
+
+	while (index <= end_index) {
+		page = grab_cache_page(tree->mapping, index);
+		if (!page) {
+			err = -ENOMEM;
+			goto failed;
+		}
+		if (IS_ERR(page)) {
+			err = PTR_ERR(page);
+			goto failed;
+		}
+		index++;
+	}
+	lock_extent(tree, start, end, GFP_NOFS);
+	return 0;
+
+failed:
+	/*
+	 * we failed above in getting the page at 'index', so we undo here
+	 * up to but not including the page at 'index'
+	 */
+	end_index = index;
+	index = start >> PAGE_CACHE_SHIFT;
+	while (index < end_index) {
+		page = find_get_page(tree->mapping, index);
+		unlock_page(page);
+		page_cache_release(page);
+		index++;
+	}
+	return err;
+}
+
+/*
+ * helper function to unlock both pages and extents in the tree.
+ */
+static int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
+{
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	unsigned long end_index = end >> PAGE_CACHE_SHIFT;
+	struct page *page;
+
+	while (index <= end_index) {
+		page = find_get_page(tree->mapping, index);
+		unlock_page(page);
+		page_cache_release(page);
+		index++;
+	}
+	unlock_extent(tree, start, end, GFP_NOFS);
+	return 0;
+}
+#endif
+
+/*
+ * set the private field for a given byte offset in the tree.  If there isn't
+ * an extent_state there already, this does nothing.
+ */
+int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	int ret = 0;
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node) {
+		ret = -ENOENT;
+		goto out;
+	}
+	state = rb_entry(node, struct extent_state, rb_node);
+	if (state->start != start) {
+		ret = -ENOENT;
+		goto out;
+	}
+	state->private = private;
+out:
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
+{
+	struct rb_node *node;
+	struct extent_state *state;
+	int ret = 0;
+
+	spin_lock(&tree->lock);
+	/*
+	 * this search will find all the extents that end after
+	 * our range starts.
+	 */
+	node = tree_search(tree, start);
+	if (!node) {
+		ret = -ENOENT;
+		goto out;
+	}
+	state = rb_entry(node, struct extent_state, rb_node);
+	if (state->start != start) {
+		ret = -ENOENT;
+		goto out;
+	}
+	*private = state->private;
+out:
+	spin_unlock(&tree->lock);
+	return ret;
+}
+
+/*
+ * searches a range in the state tree for a given mask.
+ * If 'filled' == 1, this returns 1 only if every extent in the tree
+ * has the bits set.  Otherwise, 1 is returned if any bit in the
+ * range is found set.
+ */
+int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
+		   int bits, int filled)
+{
+	struct extent_state *state = NULL;
+	struct rb_node *node;
+	int bitset = 0;
+
+	spin_lock(&tree->lock);
+	node = tree_search(tree, start);
+	while (node && start <= end) {
+		state = rb_entry(node, struct extent_state, rb_node);
+
+		if (filled && state->start > start) {
+			bitset = 0;
+			break;
+		}
+
+		if (state->start > end)
+			break;
+
+		if (state->state & bits) {
+			bitset = 1;
+			if (!filled)
+				break;
+		} else if (filled) {
+			bitset = 0;
+			break;
+		}
+		start = state->end + 1;
+		if (start > end)
+			break;
+		node = rb_next(node);
+		if (!node) {
+			if (filled)
+				bitset = 0;
+			break;
+		}
+	}
+	spin_unlock(&tree->lock);
+	return bitset;
+}
+
+/*
+ * helper function to set a given page up to date if all the
+ * extents in the tree for that page are up to date
+ */
+static int check_page_uptodate(struct extent_io_tree *tree,
+			       struct page *page)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
+		SetPageUptodate(page);
+	return 0;
+}
+
+/*
+ * helper function to unlock a page if all the extents in the tree
+ * for that page are unlocked
+ */
+static int check_page_locked(struct extent_io_tree *tree,
+			     struct page *page)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
+		unlock_page(page);
+	return 0;
+}
+
+/*
+ * helper function to end page writeback if all the extents
+ * in the tree for that page are done with writeback
+ */
+static int check_page_writeback(struct extent_io_tree *tree,
+			     struct page *page)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
+		end_page_writeback(page);
+	return 0;
+}
+
+/* lots and lots of room for performance fixes in the end_bio funcs */
+
+/*
+ * after a writepage IO is done, we need to:
+ * clear the uptodate bits on error
+ * clear the writeback bits in the extent tree for this IO
+ * end_page_writeback if the page has no more pending IO
+ *
+ * Scheduling is not allowed, so the extent state tree is expected
+ * to have one and only one object corresponding to this IO.
+ */
+static void end_bio_extent_writepage(struct bio *bio, int err)
+{
+	int uptodate = err == 0;
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct extent_io_tree *tree;
+	u64 start;
+	u64 end;
+	int whole_page;
+	int ret;
+
+	do {
+		struct page *page = bvec->bv_page;
+		tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+		start = ((u64)page->index << PAGE_CACHE_SHIFT) +
+			 bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
+
+		if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
+			whole_page = 1;
+		else
+			whole_page = 0;
+
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+		if (tree->ops && tree->ops->writepage_end_io_hook) {
+			ret = tree->ops->writepage_end_io_hook(page, start,
+						       end, NULL, uptodate);
+			if (ret)
+				uptodate = 0;
+		}
+
+		if (!uptodate && tree->ops &&
+		    tree->ops->writepage_io_failed_hook) {
+			ret = tree->ops->writepage_io_failed_hook(bio, page,
+							 start, end, NULL);
+			if (ret == 0) {
+				uptodate = (err == 0);
+				continue;
+			}
+		}
+
+		if (!uptodate) {
+			clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
+			ClearPageUptodate(page);
+			SetPageError(page);
+		}
+
+		clear_extent_writeback(tree, start, end, GFP_ATOMIC);
+
+		if (whole_page)
+			end_page_writeback(page);
+		else
+			check_page_writeback(tree, page);
+	} while (bvec >= bio->bi_io_vec);
+
+	bio_put(bio);
+}
+
+/*
+ * after a readpage IO is done, we need to:
+ * clear the uptodate bits on error
+ * set the uptodate bits if things worked
+ * set the page up to date if all extents in the tree are uptodate
+ * clear the lock bit in the extent tree
+ * unlock the page if there are no other extents locked for it
+ *
+ * Scheduling is not allowed, so the extent state tree is expected
+ * to have one and only one object corresponding to this IO.
+ */
+static void end_bio_extent_readpage(struct bio *bio, int err)
+{
+	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct extent_io_tree *tree;
+	u64 start;
+	u64 end;
+	int whole_page;
+	int ret;
+
+	if (err)
+		uptodate = 0;
+
+	do {
+		struct page *page = bvec->bv_page;
+		tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+		start = ((u64)page->index << PAGE_CACHE_SHIFT) +
+			bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
+
+		if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
+			whole_page = 1;
+		else
+			whole_page = 0;
+
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+
+		if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
+			ret = tree->ops->readpage_end_io_hook(page, start, end,
+							      NULL);
+			if (ret)
+				uptodate = 0;
+		}
+		if (!uptodate && tree->ops &&
+		    tree->ops->readpage_io_failed_hook) {
+			ret = tree->ops->readpage_io_failed_hook(bio, page,
+							 start, end, NULL);
+			if (ret == 0) {
+				uptodate =
+					test_bit(BIO_UPTODATE, &bio->bi_flags);
+				if (err)
+					uptodate = 0;
+				continue;
+			}
+		}
+
+		if (uptodate) {
+			set_extent_uptodate(tree, start, end,
+					    GFP_ATOMIC);
+		}
+		unlock_extent(tree, start, end, GFP_ATOMIC);
+
+		if (whole_page) {
+			if (uptodate) {
+				SetPageUptodate(page);
+			} else {
+				ClearPageUptodate(page);
+				SetPageError(page);
+			}
+			unlock_page(page);
+		} else {
+			if (uptodate) {
+				check_page_uptodate(tree, page);
+			} else {
+				ClearPageUptodate(page);
+				SetPageError(page);
+			}
+			check_page_locked(tree, page);
+		}
+	} while (bvec >= bio->bi_io_vec);
+
+	bio_put(bio);
+}
+
+/*
+ * IO done from prepare_write is pretty simple, we just unlock
+ * the structs in the extent tree when done, and set the uptodate bits
+ * as appropriate.
+ */
+static void end_bio_extent_preparewrite(struct bio *bio, int err)
+{
+	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct extent_io_tree *tree;
+	u64 start;
+	u64 end;
+
+	do {
+		struct page *page = bvec->bv_page;
+		tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+		start = ((u64)page->index << PAGE_CACHE_SHIFT) +
+			bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
+
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+
+		if (uptodate) {
+			set_extent_uptodate(tree, start, end, GFP_ATOMIC);
+		} else {
+			ClearPageUptodate(page);
+			SetPageError(page);
+		}
+
+		unlock_extent(tree, start, end, GFP_ATOMIC);
+
+	} while (bvec >= bio->bi_io_vec);
+
+	bio_put(bio);
+}
+
+static struct bio *
+extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
+		 gfp_t gfp_flags)
+{
+	struct bio *bio;
+
+	bio = bio_alloc(gfp_flags, nr_vecs);
+
+	if (bio == NULL && (current->flags & PF_MEMALLOC)) {
+		while (!bio && (nr_vecs /= 2))
+			bio = bio_alloc(gfp_flags, nr_vecs);
+	}
+
+	if (bio) {
+		bio->bi_size = 0;
+		bio->bi_bdev = bdev;
+		bio->bi_sector = first_sector;
+	}
+	return bio;
+}
+
+static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
+			  unsigned long bio_flags)
+{
+	int ret = 0;
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct page *page = bvec->bv_page;
+	struct extent_io_tree *tree = bio->bi_private;
+	u64 start;
+	u64 end;
+
+	start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
+	end = start + bvec->bv_len - 1;
+
+	bio->bi_private = NULL;
+
+	bio_get(bio);
+
+	if (tree->ops && tree->ops->submit_bio_hook)
+		tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
+					   mirror_num, bio_flags);
+	else
+		submit_bio(rw, bio);
+	if (bio_flagged(bio, BIO_EOPNOTSUPP))
+		ret = -EOPNOTSUPP;
+	bio_put(bio);
+	return ret;
+}
+
+static int submit_extent_page(int rw, struct extent_io_tree *tree,
+			      struct page *page, sector_t sector,
+			      size_t size, unsigned long offset,
+			      struct block_device *bdev,
+			      struct bio **bio_ret,
+			      unsigned long max_pages,
+			      bio_end_io_t end_io_func,
+			      int mirror_num,
+			      unsigned long prev_bio_flags,
+			      unsigned long bio_flags)
+{
+	int ret = 0;
+	struct bio *bio;
+	int nr;
+	int contig = 0;
+	int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
+	int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
+	size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
+
+	if (bio_ret && *bio_ret) {
+		bio = *bio_ret;
+		if (old_compressed)
+			contig = bio->bi_sector == sector;
+		else
+			contig = bio->bi_sector + (bio->bi_size >> 9) ==
+				sector;
+
+		if (prev_bio_flags != bio_flags || !contig ||
+		    (tree->ops && tree->ops->merge_bio_hook &&
+		     tree->ops->merge_bio_hook(page, offset, page_size, bio,
+					       bio_flags)) ||
+		    bio_add_page(bio, page, page_size, offset) < page_size) {
+			ret = submit_one_bio(rw, bio, mirror_num,
+					     prev_bio_flags);
+			bio = NULL;
+		} else {
+			return 0;
+		}
+	}
+	if (this_compressed)
+		nr = BIO_MAX_PAGES;
+	else
+		nr = bio_get_nr_vecs(bdev);
+
+	bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
+
+	bio_add_page(bio, page, page_size, offset);
+	bio->bi_end_io = end_io_func;
+	bio->bi_private = tree;
+
+	if (bio_ret)
+		*bio_ret = bio;
+	else
+		ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
+
+	return ret;
+}
+
+void set_page_extent_mapped(struct page *page)
+{
+	if (!PagePrivate(page)) {
+		SetPagePrivate(page);
+		page_cache_get(page);
+		set_page_private(page, EXTENT_PAGE_PRIVATE);
+	}
+}
+
+static void set_page_extent_head(struct page *page, unsigned long len)
+{
+	set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
+}
+
+/*
+ * basic readpage implementation.  Locked extent state structs are inserted
+ * into the tree that are removed when the IO is done (by the end_io
+ * handlers)
+ */
+static int __extent_read_full_page(struct extent_io_tree *tree,
+				   struct page *page,
+				   get_extent_t *get_extent,
+				   struct bio **bio, int mirror_num,
+				   unsigned long *bio_flags)
+{
+	struct inode *inode = page->mapping->host;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 page_end = start + PAGE_CACHE_SIZE - 1;
+	u64 end;
+	u64 cur = start;
+	u64 extent_offset;
+	u64 last_byte = i_size_read(inode);
+	u64 block_start;
+	u64 cur_end;
+	sector_t sector;
+	struct extent_map *em;
+	struct block_device *bdev;
+	int ret;
+	int nr = 0;
+	size_t page_offset = 0;
+	size_t iosize;
+	size_t disk_io_size;
+	size_t blocksize = inode->i_sb->s_blocksize;
+	unsigned long this_bio_flag = 0;
+
+	set_page_extent_mapped(page);
+
+	end = page_end;
+	lock_extent(tree, start, end, GFP_NOFS);
+
+	if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
+		char *userpage;
+		size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
+
+		if (zero_offset) {
+			iosize = PAGE_CACHE_SIZE - zero_offset;
+			userpage = kmap_atomic(page, KM_USER0);
+			memset(userpage + zero_offset, 0, iosize);
+			flush_dcache_page(page);
+			kunmap_atomic(userpage, KM_USER0);
+		}
+	}
+	while (cur <= end) {
+		if (cur >= last_byte) {
+			char *userpage;
+			iosize = PAGE_CACHE_SIZE - page_offset;
+			userpage = kmap_atomic(page, KM_USER0);
+			memset(userpage + page_offset, 0, iosize);
+			flush_dcache_page(page);
+			kunmap_atomic(userpage, KM_USER0);
+			set_extent_uptodate(tree, cur, cur + iosize - 1,
+					    GFP_NOFS);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			break;
+		}
+		em = get_extent(inode, page, page_offset, cur,
+				end - cur + 1, 0);
+		if (IS_ERR(em) || !em) {
+			SetPageError(page);
+			unlock_extent(tree, cur, end, GFP_NOFS);
+			break;
+		}
+		extent_offset = cur - em->start;
+		BUG_ON(extent_map_end(em) <= cur);
+		BUG_ON(end < cur);
+
+		if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
+			this_bio_flag = EXTENT_BIO_COMPRESSED;
+
+		iosize = min(extent_map_end(em) - cur, end - cur + 1);
+		cur_end = min(extent_map_end(em) - 1, end);
+		iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
+		if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
+			disk_io_size = em->block_len;
+			sector = em->block_start >> 9;
+		} else {
+			sector = (em->block_start + extent_offset) >> 9;
+			disk_io_size = iosize;
+		}
+		bdev = em->bdev;
+		block_start = em->block_start;
+		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
+			block_start = EXTENT_MAP_HOLE;
+		free_extent_map(em);
+		em = NULL;
+
+		/* we've found a hole, just zero and go on */
+		if (block_start == EXTENT_MAP_HOLE) {
+			char *userpage;
+			userpage = kmap_atomic(page, KM_USER0);
+			memset(userpage + page_offset, 0, iosize);
+			flush_dcache_page(page);
+			kunmap_atomic(userpage, KM_USER0);
+
+			set_extent_uptodate(tree, cur, cur + iosize - 1,
+					    GFP_NOFS);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			cur = cur + iosize;
+			page_offset += iosize;
+			continue;
+		}
+		/* the get_extent function already copied into the page */
+		if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
+			check_page_uptodate(tree, page);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			cur = cur + iosize;
+			page_offset += iosize;
+			continue;
+		}
+		/* we have an inline extent but it didn't get marked up
+		 * to date.  Error out
+		 */
+		if (block_start == EXTENT_MAP_INLINE) {
+			SetPageError(page);
+			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
+			cur = cur + iosize;
+			page_offset += iosize;
+			continue;
+		}
+
+		ret = 0;
+		if (tree->ops && tree->ops->readpage_io_hook) {
+			ret = tree->ops->readpage_io_hook(page, cur,
+							  cur + iosize - 1);
+		}
+		if (!ret) {
+			unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
+			pnr -= page->index;
+			ret = submit_extent_page(READ, tree, page,
+					 sector, disk_io_size, page_offset,
+					 bdev, bio, pnr,
+					 end_bio_extent_readpage, mirror_num,
+					 *bio_flags,
+					 this_bio_flag);
+			nr++;
+			*bio_flags = this_bio_flag;
+		}
+		if (ret)
+			SetPageError(page);
+		cur = cur + iosize;
+		page_offset += iosize;
+	}
+	if (!nr) {
+		if (!PageError(page))
+			SetPageUptodate(page);
+		unlock_page(page);
+	}
+	return 0;
+}
+
+int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
+			    get_extent_t *get_extent)
+{
+	struct bio *bio = NULL;
+	unsigned long bio_flags = 0;
+	int ret;
+
+	ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
+				      &bio_flags);
+	if (bio)
+		submit_one_bio(READ, bio, 0, bio_flags);
+	return ret;
+}
+
+/*
+ * the writepage semantics are similar to regular writepage.  extent
+ * records are inserted to lock ranges in the tree, and as dirty areas
+ * are found, they are marked writeback.  Then the lock bits are removed
+ * and the end_io handler clears the writeback ranges
+ */
+static int __extent_writepage(struct page *page, struct writeback_control *wbc,
+			      void *data)
+{
+	struct inode *inode = page->mapping->host;
+	struct extent_page_data *epd = data;
+	struct extent_io_tree *tree = epd->tree;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 delalloc_start;
+	u64 page_end = start + PAGE_CACHE_SIZE - 1;
+	u64 end;
+	u64 cur = start;
+	u64 extent_offset;
+	u64 last_byte = i_size_read(inode);
+	u64 block_start;
+	u64 iosize;
+	u64 unlock_start;
+	sector_t sector;
+	struct extent_map *em;
+	struct block_device *bdev;
+	int ret;
+	int nr = 0;
+	size_t pg_offset = 0;
+	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;
+	int page_started;
+	int compressed;
+	unsigned long nr_written = 0;
+
+	WARN_ON(!PageLocked(page));
+	pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
+	if (page->index > end_index ||
+	   (page->index == end_index && !pg_offset)) {
+		page->mapping->a_ops->invalidatepage(page, 0);
+		unlock_page(page);
+		return 0;
+	}
+
+	if (page->index == end_index) {
+		char *userpage;
+
+		userpage = kmap_atomic(page, KM_USER0);
+		memset(userpage + pg_offset, 0,
+		       PAGE_CACHE_SIZE - pg_offset);
+		kunmap_atomic(userpage, KM_USER0);
+		flush_dcache_page(page);
+	}
+	pg_offset = 0;
+
+	set_page_extent_mapped(page);
+
+	delalloc_start = start;
+	delalloc_end = 0;
+	page_started = 0;
+	if (!epd->extent_locked) {
+		while (delalloc_end < page_end) {
+			nr_delalloc = find_lock_delalloc_range(inode, tree,
+						       page,
+						       &delalloc_start,
+						       &delalloc_end,
+						       128 * 1024 * 1024);
+			if (nr_delalloc == 0) {
+				delalloc_start = delalloc_end + 1;
+				continue;
+			}
+			tree->ops->fill_delalloc(inode, page, delalloc_start,
+						 delalloc_end, &page_started,
+						 &nr_written);
+			delalloc_start = delalloc_end + 1;
+		}
+
+		/* did the fill delalloc function already unlock and start
+		 * the IO?
+		 */
+		if (page_started) {
+			ret = 0;
+			goto update_nr_written;
+		}
+	}
+	lock_extent(tree, start, page_end, GFP_NOFS);
+
+	unlock_start = start;
+
+	if (tree->ops && tree->ops->writepage_start_hook) {
+		ret = tree->ops->writepage_start_hook(page, start,
+						      page_end);
+		if (ret == -EAGAIN) {
+			unlock_extent(tree, start, page_end, GFP_NOFS);
+			redirty_page_for_writepage(wbc, page);
+			unlock_page(page);
+			ret = 0;
+			goto update_nr_written;
+		}
+	}
+
+	nr_written++;
+
+	end = page_end;
+	if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
+		printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
+
+	if (last_byte <= start) {
+		clear_extent_dirty(tree, start, page_end, GFP_NOFS);
+		unlock_extent(tree, start, page_end, GFP_NOFS);
+		if (tree->ops && tree->ops->writepage_end_io_hook)
+			tree->ops->writepage_end_io_hook(page, start,
+							 page_end, NULL, 1);
+		unlock_start = page_end + 1;
+		goto done;
+	}
+
+	set_extent_uptodate(tree, start, page_end, GFP_NOFS);
+	blocksize = inode->i_sb->s_blocksize;
+
+	while (cur <= end) {
+		if (cur >= last_byte) {
+			clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
+			unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
+			if (tree->ops && tree->ops->writepage_end_io_hook)
+				tree->ops->writepage_end_io_hook(page, cur,
+							 page_end, NULL, 1);
+			unlock_start = page_end + 1;
+			break;
+		}
+		em = epd->get_extent(inode, page, pg_offset, cur,
+				     end - cur + 1, 1);
+		if (IS_ERR(em) || !em) {
+			SetPageError(page);
+			break;
+		}
+
+		extent_offset = cur - em->start;
+		BUG_ON(extent_map_end(em) <= cur);
+		BUG_ON(end < cur);
+		iosize = min(extent_map_end(em) - cur, end - cur + 1);
+		iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
+		sector = (em->block_start + extent_offset) >> 9;
+		bdev = em->bdev;
+		block_start = em->block_start;
+		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+		free_extent_map(em);
+		em = NULL;
+
+		/*
+		 * compressed and inline extents are written through other
+		 * paths in the FS
+		 */
+		if (compressed || block_start == EXTENT_MAP_HOLE ||
+		    block_start == EXTENT_MAP_INLINE) {
+			clear_extent_dirty(tree, cur,
+					   cur + iosize - 1, GFP_NOFS);
+
+			unlock_extent(tree, unlock_start, cur + iosize - 1,
+				      GFP_NOFS);
+
+			/*
+			 * end_io notification does not happen here for
+			 * compressed extents
+			 */
+			if (!compressed && tree->ops &&
+			    tree->ops->writepage_end_io_hook)
+				tree->ops->writepage_end_io_hook(page, cur,
+							 cur + iosize - 1,
+							 NULL, 1);
+			else if (compressed) {
+				/* we don't want to end_page_writeback on
+				 * a compressed extent.  this happens
+				 * elsewhere
+				 */
+				nr++;
+			}
+
+			cur += iosize;
+			pg_offset += iosize;
+			unlock_start = cur;
+			continue;
+		}
+		/* leave this out until we have a page_mkwrite call */
+		if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
+				   EXTENT_DIRTY, 0)) {
+			cur = cur + iosize;
+			pg_offset += iosize;
+			continue;
+		}
+
+		clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
+		if (tree->ops && tree->ops->writepage_io_hook) {
+			ret = tree->ops->writepage_io_hook(page, cur,
+						cur + iosize - 1);
+		} else {
+			ret = 0;
+		}
+		if (ret) {
+			SetPageError(page);
+		} else {
+			unsigned long max_nr = end_index + 1;
+
+			set_range_writeback(tree, cur, cur + iosize - 1);
+			if (!PageWriteback(page)) {
+				printk(KERN_ERR "btrfs warning page %lu not "
+				       "writeback, cur %llu end %llu\n",
+				       page->index, (unsigned long long)cur,
+				       (unsigned long long)end);
+			}
+
+			ret = submit_extent_page(WRITE, tree, page, sector,
+						 iosize, pg_offset, bdev,
+						 &epd->bio, max_nr,
+						 end_bio_extent_writepage,
+						 0, 0, 0);
+			if (ret)
+				SetPageError(page);
+		}
+		cur = cur + iosize;
+		pg_offset += iosize;
+		nr++;
+	}
+done:
+	if (nr == 0) {
+		/* make sure the mapping tag for page dirty gets cleared */
+		set_page_writeback(page);
+		end_page_writeback(page);
+	}
+	if (unlock_start <= page_end)
+		unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
+	unlock_page(page);
+
+update_nr_written:
+	wbc->nr_to_write -= nr_written;
+	if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
+	    wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
+		page->mapping->writeback_index = page->index + nr_written;
+	return 0;
+}
+
+/**
+ * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
+ * @mapping: address space structure to write
+ * @wbc: subtract the number of written pages from *@wbc->nr_to_write
+ * @writepage: function called for each page
+ * @data: data passed to writepage function
+ *
+ * If a page is already under I/O, write_cache_pages() skips it, even
+ * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
+ * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
+ * and msync() need to guarantee that all the data which was dirty at the time
+ * the call was made get new I/O started against them.  If wbc->sync_mode is
+ * WB_SYNC_ALL then we were called for data integrity and we must wait for
+ * existing IO to complete.
+ */
+static int extent_write_cache_pages(struct extent_io_tree *tree,
+			     struct address_space *mapping,
+			     struct writeback_control *wbc,
+			     writepage_t writepage, void *data,
+			     void (*flush_fn)(void *))
+{
+	struct backing_dev_info *bdi = mapping->backing_dev_info;
+	int ret = 0;
+	int done = 0;
+	struct pagevec pvec;
+	int nr_pages;
+	pgoff_t index;
+	pgoff_t end;		/* Inclusive */
+	int scanned = 0;
+	int range_whole = 0;
+
+	if (wbc->nonblocking && bdi_write_congested(bdi)) {
+		wbc->encountered_congestion = 1;
+		return 0;
+	}
+
+	pagevec_init(&pvec, 0);
+	if (wbc->range_cyclic) {
+		index = mapping->writeback_index; /* Start from prev offset */
+		end = -1;
+	} else {
+		index = wbc->range_start >> PAGE_CACHE_SHIFT;
+		end = wbc->range_end >> PAGE_CACHE_SHIFT;
+		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+			range_whole = 1;
+		scanned = 1;
+	}
+retry:
+	while (!done && (index <= end) &&
+	       (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
+			      PAGECACHE_TAG_DIRTY, min(end - index,
+				  (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
+		unsigned i;
+
+		scanned = 1;
+		for (i = 0; i < nr_pages; i++) {
+			struct page *page = pvec.pages[i];
+
+			/*
+			 * At this point we hold neither mapping->tree_lock nor
+			 * lock on the page itself: the page may be truncated or
+			 * invalidated (changing page->mapping to NULL), or even
+			 * swizzled back from swapper_space to tmpfs file
+			 * mapping
+			 */
+			if (tree->ops && tree->ops->write_cache_pages_lock_hook)
+				tree->ops->write_cache_pages_lock_hook(page);
+			else
+				lock_page(page);
+
+			if (unlikely(page->mapping != mapping)) {
+				unlock_page(page);
+				continue;
+			}
+
+			if (!wbc->range_cyclic && page->index > end) {
+				done = 1;
+				unlock_page(page);
+				continue;
+			}
+
+			if (wbc->sync_mode != WB_SYNC_NONE) {
+				if (PageWriteback(page))
+					flush_fn(data);
+				wait_on_page_writeback(page);
+			}
+
+			if (PageWriteback(page) ||
+			    !clear_page_dirty_for_io(page)) {
+				unlock_page(page);
+				continue;
+			}
+
+			ret = (*writepage)(page, wbc, data);
+
+			if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
+				unlock_page(page);
+				ret = 0;
+			}
+			if (ret || wbc->nr_to_write <= 0)
+				done = 1;
+			if (wbc->nonblocking && bdi_write_congested(bdi)) {
+				wbc->encountered_congestion = 1;
+				done = 1;
+			}
+		}
+		pagevec_release(&pvec);
+		cond_resched();
+	}
+	if (!scanned && !done) {
+		/*
+		 * We hit the last page and there is more work to be done: wrap
+		 * back to the start of the file
+		 */
+		scanned = 1;
+		index = 0;
+		goto retry;
+	}
+	return ret;
+}
+
+static noinline void flush_write_bio(void *data)
+{
+	struct extent_page_data *epd = data;
+	if (epd->bio) {
+		submit_one_bio(WRITE, epd->bio, 0, 0);
+		epd->bio = NULL;
+	}
+}
+
+int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
+			  get_extent_t *get_extent,
+			  struct writeback_control *wbc)
+{
+	int ret;
+	struct address_space *mapping = page->mapping;
+	struct extent_page_data epd = {
+		.bio = NULL,
+		.tree = tree,
+		.get_extent = get_extent,
+		.extent_locked = 0,
+	};
+	struct writeback_control wbc_writepages = {
+		.bdi		= wbc->bdi,
+		.sync_mode	= WB_SYNC_NONE,
+		.older_than_this = NULL,
+		.nr_to_write	= 64,
+		.range_start	= page_offset(page) + PAGE_CACHE_SIZE,
+		.range_end	= (loff_t)-1,
+	};
+
+
+	ret = __extent_writepage(page, wbc, &epd);
+
+	extent_write_cache_pages(tree, mapping, &wbc_writepages,
+				 __extent_writepage, &epd, flush_write_bio);
+	if (epd.bio)
+		submit_one_bio(WRITE, epd.bio, 0, 0);
+	return ret;
+}
+
+int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
+			      u64 start, u64 end, get_extent_t *get_extent,
+			      int mode)
+{
+	int ret = 0;
+	struct address_space *mapping = inode->i_mapping;
+	struct page *page;
+	unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
+		PAGE_CACHE_SHIFT;
+
+	struct extent_page_data epd = {
+		.bio = NULL,
+		.tree = tree,
+		.get_extent = get_extent,
+		.extent_locked = 1,
+	};
+	struct writeback_control wbc_writepages = {
+		.bdi		= inode->i_mapping->backing_dev_info,
+		.sync_mode	= mode,
+		.older_than_this = NULL,
+		.nr_to_write	= nr_pages * 2,
+		.range_start	= start,
+		.range_end	= end + 1,
+	};
+
+	while (start <= end) {
+		page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
+		if (clear_page_dirty_for_io(page))
+			ret = __extent_writepage(page, &wbc_writepages, &epd);
+		else {
+			if (tree->ops && tree->ops->writepage_end_io_hook)
+				tree->ops->writepage_end_io_hook(page, start,
+						 start + PAGE_CACHE_SIZE - 1,
+						 NULL, 1);
+			unlock_page(page);
+		}
+		page_cache_release(page);
+		start += PAGE_CACHE_SIZE;
+	}
+
+	if (epd.bio)
+		submit_one_bio(WRITE, epd.bio, 0, 0);
+	return ret;
+}
+
+int extent_writepages(struct extent_io_tree *tree,
+		      struct address_space *mapping,
+		      get_extent_t *get_extent,
+		      struct writeback_control *wbc)
+{
+	int ret = 0;
+	struct extent_page_data epd = {
+		.bio = NULL,
+		.tree = tree,
+		.get_extent = get_extent,
+		.extent_locked = 0,
+	};
+
+	ret = extent_write_cache_pages(tree, mapping, wbc,
+				       __extent_writepage, &epd,
+				       flush_write_bio);
+	if (epd.bio)
+		submit_one_bio(WRITE, epd.bio, 0, 0);
+	return ret;
+}
+
+int extent_readpages(struct extent_io_tree *tree,
+		     struct address_space *mapping,
+		     struct list_head *pages, unsigned nr_pages,
+		     get_extent_t get_extent)
+{
+	struct bio *bio = NULL;
+	unsigned page_idx;
+	struct pagevec pvec;
+	unsigned long bio_flags = 0;
+
+	pagevec_init(&pvec, 0);
+	for (page_idx = 0; page_idx < nr_pages; page_idx++) {
+		struct page *page = list_entry(pages->prev, struct page, lru);
+
+		prefetchw(&page->flags);
+		list_del(&page->lru);
+		/*
+		 * what we want to do here is call add_to_page_cache_lru,
+		 * but that isn't exported, so we reproduce it here
+		 */
+		if (!add_to_page_cache(page, mapping,
+					page->index, GFP_KERNEL)) {
+
+			/* open coding of lru_cache_add, also not exported */
+			page_cache_get(page);
+			if (!pagevec_add(&pvec, page))
+				__pagevec_lru_add_file(&pvec);
+			__extent_read_full_page(tree, page, get_extent,
+						&bio, 0, &bio_flags);
+		}
+		page_cache_release(page);
+	}
+	if (pagevec_count(&pvec))
+		__pagevec_lru_add_file(&pvec);
+	BUG_ON(!list_empty(pages));
+	if (bio)
+		submit_one_bio(READ, bio, 0, bio_flags);
+	return 0;
+}
+
+/*
+ * basic invalidatepage code, this waits on any locked or writeback
+ * ranges corresponding to the page, and then deletes any extent state
+ * records from the tree
+ */
+int extent_invalidatepage(struct extent_io_tree *tree,
+			  struct page *page, unsigned long offset)
+{
+	u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	size_t blocksize = page->mapping->host->i_sb->s_blocksize;
+
+	start += (offset + blocksize - 1) & ~(blocksize - 1);
+	if (start > end)
+		return 0;
+
+	lock_extent(tree, start, end, GFP_NOFS);
+	wait_on_extent_writeback(tree, start, end);
+	clear_extent_bit(tree, start, end,
+			 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
+			 1, 1, GFP_NOFS);
+	return 0;
+}
+
+/*
+ * simple commit_write call, set_range_dirty is used to mark both
+ * the pages and the extent records as dirty
+ */
+int extent_commit_write(struct extent_io_tree *tree,
+			struct inode *inode, struct page *page,
+			unsigned from, unsigned to)
+{
+	loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
+
+	set_page_extent_mapped(page);
+	set_page_dirty(page);
+
+	if (pos > inode->i_size) {
+		i_size_write(inode, pos);
+		mark_inode_dirty(inode);
+	}
+	return 0;
+}
+
+int extent_prepare_write(struct extent_io_tree *tree,
+			 struct inode *inode, struct page *page,
+			 unsigned from, unsigned to, get_extent_t *get_extent)
+{
+	u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
+	u64 block_start;
+	u64 orig_block_start;
+	u64 block_end;
+	u64 cur_end;
+	struct extent_map *em;
+	unsigned blocksize = 1 << inode->i_blkbits;
+	size_t page_offset = 0;
+	size_t block_off_start;
+	size_t block_off_end;
+	int err = 0;
+	int iocount = 0;
+	int ret = 0;
+	int isnew;
+
+	set_page_extent_mapped(page);
+
+	block_start = (page_start + from) & ~((u64)blocksize - 1);
+	block_end = (page_start + to - 1) | (blocksize - 1);
+	orig_block_start = block_start;
+
+	lock_extent(tree, page_start, page_end, GFP_NOFS);
+	while (block_start <= block_end) {
+		em = get_extent(inode, page, page_offset, block_start,
+				block_end - block_start + 1, 1);
+		if (IS_ERR(em) || !em)
+			goto err;
+
+		cur_end = min(block_end, extent_map_end(em) - 1);
+		block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
+		block_off_end = block_off_start + blocksize;
+		isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
+
+		if (!PageUptodate(page) && isnew &&
+		    (block_off_end > to || block_off_start < from)) {
+			void *kaddr;
+
+			kaddr = kmap_atomic(page, KM_USER0);
+			if (block_off_end > to)
+				memset(kaddr + to, 0, block_off_end - to);
+			if (block_off_start < from)
+				memset(kaddr + block_off_start, 0,
+				       from - block_off_start);
+			flush_dcache_page(page);
+			kunmap_atomic(kaddr, KM_USER0);
+		}
+		if ((em->block_start != EXTENT_MAP_HOLE &&
+		     em->block_start != EXTENT_MAP_INLINE) &&
+		    !isnew && !PageUptodate(page) &&
+		    (block_off_end > to || block_off_start < from) &&
+		    !test_range_bit(tree, block_start, cur_end,
+				    EXTENT_UPTODATE, 1)) {
+			u64 sector;
+			u64 extent_offset = block_start - em->start;
+			size_t iosize;
+			sector = (em->block_start + extent_offset) >> 9;
+			iosize = (cur_end - block_start + blocksize) &
+				~((u64)blocksize - 1);
+			/*
+			 * we've already got the extent locked, but we
+			 * need to split the state such that our end_bio
+			 * handler can clear the lock.
+			 */
+			set_extent_bit(tree, block_start,
+				       block_start + iosize - 1,
+				       EXTENT_LOCKED, 0, NULL, GFP_NOFS);
+			ret = submit_extent_page(READ, tree, page,
+					 sector, iosize, page_offset, em->bdev,
+					 NULL, 1,
+					 end_bio_extent_preparewrite, 0,
+					 0, 0);
+			iocount++;
+			block_start = block_start + iosize;
+		} else {
+			set_extent_uptodate(tree, block_start, cur_end,
+					    GFP_NOFS);
+			unlock_extent(tree, block_start, cur_end, GFP_NOFS);
+			block_start = cur_end + 1;
+		}
+		page_offset = block_start & (PAGE_CACHE_SIZE - 1);
+		free_extent_map(em);
+	}
+	if (iocount) {
+		wait_extent_bit(tree, orig_block_start,
+				block_end, EXTENT_LOCKED);
+	}
+	check_page_uptodate(tree, page);
+err:
+	/* FIXME, zero out newly allocated blocks on error */
+	return err;
+}
+
+/*
+ * a helper for releasepage, this tests for areas of the page that
+ * are locked or under IO and drops the related state bits if it is safe
+ * to drop the page.
+ */
+int try_release_extent_state(struct extent_map_tree *map,
+			     struct extent_io_tree *tree, struct page *page,
+			     gfp_t mask)
+{
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+	int ret = 1;
+
+	if (test_range_bit(tree, start, end,
+			   EXTENT_IOBITS | EXTENT_ORDERED, 0))
+		ret = 0;
+	else {
+		if ((mask & GFP_NOFS) == GFP_NOFS)
+			mask = GFP_NOFS;
+		clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
+				 1, 1, mask);
+	}
+	return ret;
+}
+
+/*
+ * a helper for releasepage.  As long as there are no locked extents
+ * in the range corresponding to the page, both state records and extent
+ * map records are removed
+ */
+int try_release_extent_mapping(struct extent_map_tree *map,
+			       struct extent_io_tree *tree, struct page *page,
+			       gfp_t mask)
+{
+	struct extent_map *em;
+	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+
+	if ((mask & __GFP_WAIT) &&
+	    page->mapping->host->i_size > 16 * 1024 * 1024) {
+		u64 len;
+		while (start <= end) {
+			len = end - start + 1;
+			spin_lock(&map->lock);
+			em = lookup_extent_mapping(map, start, len);
+			if (!em || IS_ERR(em)) {
+				spin_unlock(&map->lock);
+				break;
+			}
+			if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
+			    em->start != start) {
+				spin_unlock(&map->lock);
+				free_extent_map(em);
+				break;
+			}
+			if (!test_range_bit(tree, em->start,
+					    extent_map_end(em) - 1,
+					    EXTENT_LOCKED | EXTENT_WRITEBACK |
+					    EXTENT_ORDERED,
+					    0)) {
+				remove_extent_mapping(map, em);
+				/* once for the rb tree */
+				free_extent_map(em);
+			}
+			start = extent_map_end(em);
+			spin_unlock(&map->lock);
+
+			/* once for us */
+			free_extent_map(em);
+		}
+	}
+	return try_release_extent_state(map, tree, page, mask);
+}
+
+sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
+		get_extent_t *get_extent)
+{
+	struct inode *inode = mapping->host;
+	u64 start = iblock << inode->i_blkbits;
+	sector_t sector = 0;
+	size_t blksize = (1 << inode->i_blkbits);
+	struct extent_map *em;
+
+	lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
+		    GFP_NOFS);
+	em = get_extent(inode, NULL, 0, start, blksize, 0);
+	unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
+		      GFP_NOFS);
+	if (!em || IS_ERR(em))
+		return 0;
+
+	if (em->block_start > EXTENT_MAP_LAST_BYTE)
+		goto out;
+
+	sector = (em->block_start + start - em->start) >> inode->i_blkbits;
+out:
+	free_extent_map(em);
+	return sector;
+}
#55 lore raw

PATCH 1/1 Btrfs extent_buffer code

In reply to Chris Mason

Extent buffers provide an API to access pages or collections of pages. They are meant to store the btrfs metadata and to handle cases where the block size is different from the OS page size.

Right now, it only works for the OS page size, but it does work with highmem pages.

The core usage here is an explicit read and write function call to access the contents of a tree block. The contents may be in a highmem page or may span multiple different pages, this is supposed to do the right thing.

Btrfs defines helper functions to access almost all of the on disk structures which also handle endian translation. Most of the btrfs code only knows about the helpers and does not directly call the extent_buffer read/write functions.

See ctree.h and struct-funcs.c for more details on the helpers themselves.

This also contains the code for doing IO on extent buffers.

fs/btrfs/extent_io.c +922 -0
@@ -25,6 +25,7 @@
 						    unsigned long));
 
 static struct kmem_cache *extent_state_cache;
+static struct kmem_cache *extent_buffer_cache;
 
 static LIST_HEAD(buffers);
 static LIST_HEAD(states);
@@ -61,6 +62,11 @@
 	if (!extent_state_cache)
 		return -ENOMEM;
 
+	extent_buffer_cache = btrfs_cache_create("extent_buffers",
+					    sizeof(struct extent_buffer), 0,
+					    NULL);
+	if (!extent_buffer_cache)
+		goto free_state_cache;
 	return 0;
 
 free_state_cache:
@@ -71,6 +77,7 @@
 void extent_io_exit(void)
 {
 	struct extent_state *state;
+	struct extent_buffer *eb;
 
 	while (!list_empty(&states)) {
 		state = list_entry(states.next, struct extent_state, leak_list);
@@ -84,8 +91,18 @@
 
 	}
 
+	while (!list_empty(&buffers)) {
+		eb = list_entry(buffers.next, struct extent_buffer, leak_list);
+		printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
+		       "refs %d\n", (unsigned long long)eb->start,
+		       eb->len, atomic_read(&eb->refs));
+		list_del(&eb->leak_list);
+		kmem_cache_free(extent_buffer_cache, eb);
+	}
 	if (extent_state_cache)
 		kmem_cache_destroy(extent_state_cache);
+	if (extent_buffer_cache)
+		kmem_cache_destroy(extent_buffer_cache);
 }
 
 void extent_io_tree_init(struct extent_io_tree *tree,
@@ -223,6 +240,50 @@
 	return ret;
 }
 
+static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
+					  u64 offset, struct rb_node *node)
+{
+	struct rb_root *root = &tree->buffer;
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct extent_buffer *eb;
+
+	while (*p) {
+		parent = *p;
+		eb = rb_entry(parent, struct extent_buffer, rb_node);
+
+		if (offset < eb->start)
+			p = &(*p)->rb_left;
+		else if (offset > eb->start)
+			p = &(*p)->rb_right;
+		else
+			return eb;
+	}
+
+	rb_link_node(node, parent, p);
+	rb_insert_color(node, root);
+	return NULL;
+}
+
+static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
+					   u64 offset)
+{
+	struct rb_root *root = &tree->buffer;
+	struct rb_node *n = root->rb_node;
+	struct extent_buffer *eb;
+
+	while (n) {
+		eb = rb_entry(n, struct extent_buffer, rb_node);
+		if (offset < eb->start)
+			n = n->rb_left;
+		else if (offset > eb->start)
+			n = n->rb_right;
+		else
+			return eb;
+	}
+	return NULL;
+}
+
 /*
  * utility function to look for merge candidates inside a given range.
  * Any extents with matching state are merged together into a single
@@ -2793,3 +2854,864 @@
 	free_extent_map(em);
 	return sector;
 }
+
+static inline struct page *extent_buffer_page(struct extent_buffer *eb,
+					      unsigned long i)
+{
+	struct page *p;
+	struct address_space *mapping;
+
+	if (i == 0)
+		return eb->first_page;
+	i += eb->start >> PAGE_CACHE_SHIFT;
+	mapping = eb->first_page->mapping;
+	if (!mapping)
+		return NULL;
+
+	/*
+	 * extent_buffer_page is only called after pinning the page
+	 * by increasing the reference count.  So we know the page must
+	 * be in the radix tree.
+	 */
+	rcu_read_lock();
+	p = radix_tree_lookup(&mapping->page_tree, i);
+	rcu_read_unlock();
+
+	return p;
+}
+
+static inline unsigned long num_extent_pages(u64 start, u64 len)
+{
+	return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
+		(start >> PAGE_CACHE_SHIFT);
+}
+
+static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
+						   u64 start,
+						   unsigned long len,
+						   gfp_t mask)
+{
+	struct extent_buffer *eb = NULL;
+#ifdef LEAK_DEBUG
+	unsigned long flags;
+#endif
+
+	eb = kmem_cache_zalloc(extent_buffer_cache, mask);
+	eb->start = start;
+	eb->len = len;
+	mutex_init(&eb->mutex);
+#ifdef LEAK_DEBUG
+	spin_lock_irqsave(&leak_lock, flags);
+	list_add(&eb->leak_list, &buffers);
+	spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+	atomic_set(&eb->refs, 1);
+
+	return eb;
+}
+
+static void __free_extent_buffer(struct extent_buffer *eb)
+{
+#ifdef LEAK_DEBUG
+	unsigned long flags;
+	spin_lock_irqsave(&leak_lock, flags);
+	list_del(&eb->leak_list);
+	spin_unlock_irqrestore(&leak_lock, flags);
+#endif
+	kmem_cache_free(extent_buffer_cache, eb);
+}
+
+struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
+					  u64 start, unsigned long len,
+					  struct page *page0,
+					  gfp_t mask)
+{
+	unsigned long num_pages = num_extent_pages(start, len);
+	unsigned long i;
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	struct extent_buffer *eb;
+	struct extent_buffer *exists = NULL;
+	struct page *p;
+	struct address_space *mapping = tree->mapping;
+	int uptodate = 1;
+
+	spin_lock(&tree->buffer_lock);
+	eb = buffer_search(tree, start);
+	if (eb) {
+		atomic_inc(&eb->refs);
+		spin_unlock(&tree->buffer_lock);
+		mark_page_accessed(eb->first_page);
+		return eb;
+	}
+	spin_unlock(&tree->buffer_lock);
+
+	eb = __alloc_extent_buffer(tree, start, len, mask);
+	if (!eb)
+		return NULL;
+
+	if (page0) {
+		eb->first_page = page0;
+		i = 1;
+		index++;
+		page_cache_get(page0);
+		mark_page_accessed(page0);
+		set_page_extent_mapped(page0);
+		set_page_extent_head(page0, len);
+		uptodate = PageUptodate(page0);
+	} else {
+		i = 0;
+	}
+	for (; i < num_pages; i++, index++) {
+		p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
+		if (!p) {
+			WARN_ON(1);
+			goto free_eb;
+		}
+		set_page_extent_mapped(p);
+		mark_page_accessed(p);
+		if (i == 0) {
+			eb->first_page = p;
+			set_page_extent_head(p, len);
+		} else {
+			set_page_private(p, EXTENT_PAGE_PRIVATE);
+		}
+		if (!PageUptodate(p))
+			uptodate = 0;
+		unlock_page(p);
+	}
+	if (uptodate)
+		eb->flags |= EXTENT_UPTODATE;
+	eb->flags |= EXTENT_BUFFER_FILLED;
+
+	spin_lock(&tree->buffer_lock);
+	exists = buffer_tree_insert(tree, start, &eb->rb_node);
+	if (exists) {
+		/* add one reference for the caller */
+		atomic_inc(&exists->refs);
+		spin_unlock(&tree->buffer_lock);
+		goto free_eb;
+	}
+	spin_unlock(&tree->buffer_lock);
+
+	/* add one reference for the tree */
+	atomic_inc(&eb->refs);
+	return eb;
+
+free_eb:
+	if (!atomic_dec_and_test(&eb->refs))
+		return exists;
+	for (index = 1; index < i; index++)
+		page_cache_release(extent_buffer_page(eb, index));
+	page_cache_release(extent_buffer_page(eb, 0));
+	__free_extent_buffer(eb);
+	return exists;
+}
+
+struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
+					 u64 start, unsigned long len,
+					  gfp_t mask)
+{
+	struct extent_buffer *eb;
+
+	spin_lock(&tree->buffer_lock);
+	eb = buffer_search(tree, start);
+	if (eb)
+		atomic_inc(&eb->refs);
+	spin_unlock(&tree->buffer_lock);
+
+	if (eb)
+		mark_page_accessed(eb->first_page);
+
+	return eb;
+}
+
+void free_extent_buffer(struct extent_buffer *eb)
+{
+	if (!eb)
+		return;
+
+	if (!atomic_dec_and_test(&eb->refs))
+		return;
+
+	WARN_ON(1);
+}
+
+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))
+			continue;
+
+		lock_page(page);
+		if (i == 0)
+			set_page_extent_head(page, eb->len);
+		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)) {
+			radix_tree_tag_clear(&page->mapping->page_tree,
+						page_index(page),
+						PAGECACHE_TAG_DIRTY);
+		}
+		spin_unlock_irq(&page->mapping->tree_lock);
+		unlock_page(page);
+	}
+	return 0;
+}
+
+int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
+				    struct extent_buffer *eb)
+{
+	return wait_on_extent_writeback(tree, eb->start,
+					eb->start + eb->len - 1);
+}
+
+int set_extent_buffer_dirty(struct extent_io_tree *tree,
+			     struct extent_buffer *eb)
+{
+	unsigned long i;
+	unsigned long num_pages;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	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),
+				 page_offset(page) + PAGE_CACHE_SIZE - 1,
+				 GFP_NOFS);
+		unlock_page(page);
+	}
+	return 0;
+}
+
+int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
+				struct extent_buffer *eb)
+{
+	unsigned long i;
+	struct page *page;
+	unsigned long num_pages;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	eb->flags &= ~EXTENT_UPTODATE;
+
+	clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
+			      GFP_NOFS);
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (page)
+			ClearPageUptodate(page);
+	}
+	return 0;
+}
+
+int set_extent_buffer_uptodate(struct extent_io_tree *tree,
+				struct extent_buffer *eb)
+{
+	unsigned long i;
+	struct page *page;
+	unsigned long num_pages;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+
+	set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
+			    GFP_NOFS);
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
+		    ((i == num_pages - 1) &&
+		     ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
+			check_page_uptodate(tree, page);
+			continue;
+		}
+		SetPageUptodate(page);
+	}
+	return 0;
+}
+
+int extent_range_uptodate(struct extent_io_tree *tree,
+			  u64 start, u64 end)
+{
+	struct page *page;
+	int ret;
+	int pg_uptodate = 1;
+	int uptodate;
+	unsigned long index;
+
+	ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
+	if (ret)
+		return 1;
+	while (start <= end) {
+		index = start >> PAGE_CACHE_SHIFT;
+		page = find_get_page(tree->mapping, index);
+		uptodate = PageUptodate(page);
+		page_cache_release(page);
+		if (!uptodate) {
+			pg_uptodate = 0;
+			break;
+		}
+		start += PAGE_CACHE_SIZE;
+	}
+	return pg_uptodate;
+}
+
+int extent_buffer_uptodate(struct extent_io_tree *tree,
+			   struct extent_buffer *eb)
+{
+	int ret = 0;
+	unsigned long num_pages;
+	unsigned long i;
+	struct page *page;
+	int pg_uptodate = 1;
+
+	if (eb->flags & EXTENT_UPTODATE)
+		return 1;
+
+	ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
+			   EXTENT_UPTODATE, 1);
+	if (ret)
+		return ret;
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	for (i = 0; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (!PageUptodate(page)) {
+			pg_uptodate = 0;
+			break;
+		}
+	}
+	return pg_uptodate;
+}
+
+int read_extent_buffer_pages(struct extent_io_tree *tree,
+			     struct extent_buffer *eb,
+			     u64 start, int wait,
+			     get_extent_t *get_extent, int mirror_num)
+{
+	unsigned long i;
+	unsigned long start_i;
+	struct page *page;
+	int err;
+	int ret = 0;
+	int locked_pages = 0;
+	int all_uptodate = 1;
+	int inc_all_pages = 0;
+	unsigned long num_pages;
+	struct bio *bio = NULL;
+	unsigned long bio_flags = 0;
+
+	if (eb->flags & EXTENT_UPTODATE)
+		return 0;
+
+	if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
+			   EXTENT_UPTODATE, 1)) {
+		return 0;
+	}
+
+	if (start) {
+		WARN_ON(start < eb->start);
+		start_i = (start >> PAGE_CACHE_SHIFT) -
+			(eb->start >> PAGE_CACHE_SHIFT);
+	} else {
+		start_i = 0;
+	}
+
+	num_pages = num_extent_pages(eb->start, eb->len);
+	for (i = start_i; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (!wait) {
+			if (!trylock_page(page))
+				goto unlock_exit;
+		} else {
+			lock_page(page);
+		}
+		locked_pages++;
+		if (!PageUptodate(page))
+			all_uptodate = 0;
+	}
+	if (all_uptodate) {
+		if (start_i == 0)
+			eb->flags |= EXTENT_UPTODATE;
+		goto unlock_exit;
+	}
+
+	for (i = start_i; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		if (inc_all_pages)
+			page_cache_get(page);
+		if (!PageUptodate(page)) {
+			if (start_i == 0)
+				inc_all_pages = 1;
+			ClearPageError(page);
+			err = __extent_read_full_page(tree, page,
+						      get_extent, &bio,
+						      mirror_num, &bio_flags);
+			if (err)
+				ret = err;
+		} else {
+			unlock_page(page);
+		}
+	}
+
+	if (bio)
+		submit_one_bio(READ, bio, mirror_num, bio_flags);
+
+	if (ret || !wait)
+		return ret;
+
+	for (i = start_i; i < num_pages; i++) {
+		page = extent_buffer_page(eb, i);
+		wait_on_page_locked(page);
+		if (!PageUptodate(page))
+			ret = -EIO;
+	}
+
+	if (!ret)
+		eb->flags |= EXTENT_UPTODATE;
+	return ret;
+
+unlock_exit:
+	i = start_i;
+	while (locked_pages > 0) {
+		page = extent_buffer_page(eb, i);
+		i++;
+		unlock_page(page);
+		locked_pages--;
+	}
+	return ret;
+}
+
+void read_extent_buffer(struct extent_buffer *eb, void *dstv,
+			unsigned long start,
+			unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	char *dst = (char *)dstv;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+
+		cur = min(len, (PAGE_CACHE_SIZE - offset));
+		kaddr = kmap_atomic(page, KM_USER1);
+		memcpy(dst, kaddr + offset, cur);
+		kunmap_atomic(kaddr, KM_USER1);
+
+		dst += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
+			       unsigned long min_len, char **token, char **map,
+			       unsigned long *map_start,
+			       unsigned long *map_len, int km)
+{
+	size_t offset = start & (PAGE_CACHE_SIZE - 1);
+	char *kaddr;
+	struct page *p;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+	unsigned long end_i = (start_offset + start + min_len - 1) >>
+		PAGE_CACHE_SHIFT;
+
+	if (i != end_i)
+		return -EINVAL;
+
+	if (i == 0) {
+		offset = start_offset;
+		*map_start = 0;
+	} else {
+		offset = 0;
+		*map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
+	}
+
+	if (start + min_len > eb->len) {
+		printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
+		       "wanted %lu %lu\n", (unsigned long long)eb->start,
+		       eb->len, start, min_len);
+		WARN_ON(1);
+	}
+
+	p = extent_buffer_page(eb, i);
+	kaddr = kmap_atomic(p, km);
+	*token = kaddr;
+	*map = kaddr + offset;
+	*map_len = PAGE_CACHE_SIZE - offset;
+	return 0;
+}
+
+int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
+		      unsigned long min_len,
+		      char **token, char **map,
+		      unsigned long *map_start,
+		      unsigned long *map_len, int km)
+{
+	int err;
+	int save = 0;
+	if (eb->map_token) {
+		unmap_extent_buffer(eb, eb->map_token, km);
+		eb->map_token = NULL;
+		save = 1;
+		WARN_ON(!mutex_is_locked(&eb->mutex));
+	}
+	err = map_private_extent_buffer(eb, start, min_len, token, map,
+				       map_start, map_len, km);
+	if (!err && save) {
+		eb->map_token = *token;
+		eb->kaddr = *map;
+		eb->map_start = *map_start;
+		eb->map_len = *map_len;
+	}
+	return err;
+}
+
+void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
+{
+	kunmap_atomic(token, km);
+}
+
+int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
+			  unsigned long start,
+			  unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	char *ptr = (char *)ptrv;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+	int ret = 0;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+
+		cur = min(len, (PAGE_CACHE_SIZE - offset));
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		ret = memcmp(ptr, kaddr + offset, cur);
+		kunmap_atomic(kaddr, KM_USER0);
+		if (ret)
+			break;
+
+		ptr += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+	return ret;
+}
+
+void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
+			 unsigned long start, unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	char *src = (char *)srcv;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+		WARN_ON(!PageUptodate(page));
+
+		cur = min(len, PAGE_CACHE_SIZE - offset);
+		kaddr = kmap_atomic(page, KM_USER1);
+		memcpy(kaddr + offset, src, cur);
+		kunmap_atomic(kaddr, KM_USER1);
+
+		src += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+void memset_extent_buffer(struct extent_buffer *eb, char c,
+			  unsigned long start, unsigned long len)
+{
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(start > eb->len);
+	WARN_ON(start + len > eb->start + eb->len);
+
+	offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(eb, i);
+		WARN_ON(!PageUptodate(page));
+
+		cur = min(len, PAGE_CACHE_SIZE - offset);
+		kaddr = kmap_atomic(page, KM_USER0);
+		memset(kaddr + offset, c, cur);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
+			unsigned long dst_offset, unsigned long src_offset,
+			unsigned long len)
+{
+	u64 dst_len = dst->len;
+	size_t cur;
+	size_t offset;
+	struct page *page;
+	char *kaddr;
+	size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
+
+	WARN_ON(src->len != dst_len);
+
+	offset = (start_offset + dst_offset) &
+		((unsigned long)PAGE_CACHE_SIZE - 1);
+
+	while (len > 0) {
+		page = extent_buffer_page(dst, i);
+		WARN_ON(!PageUptodate(page));
+
+		cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		read_extent_buffer(src, kaddr + offset, src_offset, cur);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		src_offset += cur;
+		len -= cur;
+		offset = 0;
+		i++;
+	}
+}
+
+static void move_pages(struct page *dst_page, struct page *src_page,
+		       unsigned long dst_off, unsigned long src_off,
+		       unsigned long len)
+{
+	char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
+	if (dst_page == src_page) {
+		memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
+	} else {
+		char *src_kaddr = kmap_atomic(src_page, KM_USER1);
+		char *p = dst_kaddr + dst_off + len;
+		char *s = src_kaddr + src_off + len;
+
+		while (len--)
+			*--p = *--s;
+
+		kunmap_atomic(src_kaddr, KM_USER1);
+	}
+	kunmap_atomic(dst_kaddr, KM_USER0);
+}
+
+static void copy_pages(struct page *dst_page, struct page *src_page,
+		       unsigned long dst_off, unsigned long src_off,
+		       unsigned long len)
+{
+	char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
+	char *src_kaddr;
+
+	if (dst_page != src_page)
+		src_kaddr = kmap_atomic(src_page, KM_USER1);
+	else
+		src_kaddr = dst_kaddr;
+
+	memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
+	kunmap_atomic(dst_kaddr, KM_USER0);
+	if (dst_page != src_page)
+		kunmap_atomic(src_kaddr, KM_USER1);
+}
+
+void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
+			   unsigned long src_offset, unsigned long len)
+{
+	size_t cur;
+	size_t dst_off_in_page;
+	size_t src_off_in_page;
+	size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long dst_i;
+	unsigned long src_i;
+
+	if (src_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
+		       "len %lu dst len %lu\n", src_offset, len, dst->len);
+		BUG_ON(1);
+	}
+	if (dst_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
+		       "len %lu dst len %lu\n", dst_offset, len, dst->len);
+		BUG_ON(1);
+	}
+
+	while (len > 0) {
+		dst_off_in_page = (start_offset + dst_offset) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+		src_off_in_page = (start_offset + src_offset) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+
+		dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
+		src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
+
+		cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
+					       src_off_in_page));
+		cur = min_t(unsigned long, cur,
+			(unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
+
+		copy_pages(extent_buffer_page(dst, dst_i),
+			   extent_buffer_page(dst, src_i),
+			   dst_off_in_page, src_off_in_page, cur);
+
+		src_offset += cur;
+		dst_offset += cur;
+		len -= cur;
+	}
+}
+
+void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
+			   unsigned long src_offset, unsigned long len)
+{
+	size_t cur;
+	size_t dst_off_in_page;
+	size_t src_off_in_page;
+	unsigned long dst_end = dst_offset + len - 1;
+	unsigned long src_end = src_offset + len - 1;
+	size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
+	unsigned long dst_i;
+	unsigned long src_i;
+
+	if (src_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
+		       "len %lu len %lu\n", src_offset, len, dst->len);
+		BUG_ON(1);
+	}
+	if (dst_offset + len > dst->len) {
+		printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
+		       "len %lu len %lu\n", dst_offset, len, dst->len);
+		BUG_ON(1);
+	}
+	if (dst_offset < src_offset) {
+		memcpy_extent_buffer(dst, dst_offset, src_offset, len);
+		return;
+	}
+	while (len > 0) {
+		dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
+		src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
+
+		dst_off_in_page = (start_offset + dst_end) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+		src_off_in_page = (start_offset + src_end) &
+			((unsigned long)PAGE_CACHE_SIZE - 1);
+
+		cur = min_t(unsigned long, len, src_off_in_page + 1);
+		cur = min(cur, dst_off_in_page + 1);
+		move_pages(extent_buffer_page(dst, dst_i),
+			   extent_buffer_page(dst, src_i),
+			   dst_off_in_page - cur + 1,
+			   src_off_in_page - cur + 1, cur);
+
+		dst_end -= cur;
+		src_end -= cur;
+		len -= cur;
+	}
+}
+
+int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
+{
+	u64 start = page_offset(page);
+	struct extent_buffer *eb;
+	int ret = 1;
+	unsigned long i;
+	unsigned long num_pages;
+
+	spin_lock(&tree->buffer_lock);
+	eb = buffer_search(tree, start);
+	if (!eb)
+		goto out;
+
+	if (atomic_read(&eb->refs) > 1) {
+		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++)
+		page_cache_release(extent_buffer_page(eb, i));
+	rb_erase(&eb->rb_node, &tree->buffer);
+	__free_extent_buffer(eb);
+out:
+	spin_unlock(&tree->buffer_lock);
+	return ret;
+}
#56 lore raw
In reply to Chris Mason
Quoted 4 lines · Chris Mason

For actually running the sources, downloading from the git repo will work better, but hopefully this will be easier for people to review and ask questions about the code.

Could you perhaps put the descriptions you wrote for each .c file patch as comments into the respective source files?

That would address one of the items in my earlier list.

-Andi

Signature · 1 line
[email protected]
#57 lore raw
Quoted 10 lines · Andi Kleen
Quoted 4 lines · Chris Mason

For actually running the sources, downloading from the git repo will work better, but hopefully this will be easier for people to review and ask questions about the code.

Could you perhaps put the descriptions you wrote for each .c file patch as comments into the respective source files?

That would address one of the items in my earlier list.

Good point, will do.

-chris