fs/btrfs/extent-tree.c

/*
 * 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/sort.h>
#include <linux/rcupdate.h>
#include <linux/kthread.h>
#include "compat.h"
#include "hash.h"
#include "ctree.h"
#include "disk-io.h"
#include "print-tree.h"
#include "transaction.h"
#include "volumes.h"
#include "locking.h"
#include "free-space-cache.h"

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 update_reserved_extents(struct btrfs_block_group_cache *cache,
				   u64 num_bytes, int reserve);
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 owner_objectid,
				u64 owner_offset, int refs_to_drop,
				struct btrfs_delayed_extent_op *extra_op);
static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
				    struct extent_buffer *leaf,
				    struct btrfs_extent_item *ei);
static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
				      struct btrfs_root *root,
				      u64 parent, u64 root_objectid,
				      u64 flags, u64 owner, u64 offset,
				      struct btrfs_key *ins, int ref_mod);
static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
				     struct btrfs_root *root,
				     u64 parent, u64 root_objectid,
				     u64 flags, struct btrfs_disk_key *key,
				     int level, struct btrfs_key *ins);
static int do_chunk_alloc(struct btrfs_trans_handle *trans,
			  struct btrfs_root *extent_root, u64 alloc_bytes,
			  u64 flags, int force);
static int pin_down_bytes(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root,
			  struct btrfs_path *path,
			  u64 bytenr, u64 num_bytes,
			  int is_data, int reserved,
			  struct extent_buffer **must_clean);
static int find_next_key(struct btrfs_path *path, int level,
			 struct btrfs_key *key);
static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
			    int dump_block_groups);

static noinline int
block_group_cache_done(struct btrfs_block_group_cache *cache)
{
	smp_mb();
	return cache->cached == BTRFS_CACHE_FINISHED;
}

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;
}

static int add_excluded_extent(struct btrfs_root *root,
			       u64 start, u64 num_bytes)
{
	u64 end = start + num_bytes - 1;
	set_extent_bits(&root->fs_info->freed_extents[0],
			start, end, EXTENT_UPTODATE, GFP_NOFS);
	set_extent_bits(&root->fs_info->freed_extents[1],
			start, end, EXTENT_UPTODATE, GFP_NOFS);
	return 0;
}

static void free_excluded_extents(struct btrfs_root *root,
				  struct btrfs_block_group_cache *cache)
{
	u64 start, end;

	start = cache->key.objectid;
	end = start + cache->key.offset - 1;

	clear_extent_bits(&root->fs_info->freed_extents[0],
			  start, end, EXTENT_UPTODATE, GFP_NOFS);
	clear_extent_bits(&root->fs_info->freed_extents[1],
			  start, end, EXTENT_UPTODATE, GFP_NOFS);
}

static int exclude_super_stripes(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--) {
			cache->bytes_super += stripe_len;
			ret = add_excluded_extent(root, logical[nr],
						  stripe_len);
			BUG_ON(ret);
		}

		kfree(logical);
	}
	return 0;
}

static struct btrfs_caching_control *
get_caching_control(struct btrfs_block_group_cache *cache)
{
	struct btrfs_caching_control *ctl;

	spin_lock(&cache->lock);
	if (cache->cached != BTRFS_CACHE_STARTED) {
		spin_unlock(&cache->lock);
		return NULL;
	}

	ctl = cache->caching_ctl;
	atomic_inc(&ctl->count);
	spin_unlock(&cache->lock);
	return ctl;
}

static void put_caching_control(struct btrfs_caching_control *ctl)
{
	if (atomic_dec_and_test(&ctl->count))
		kfree(ctl);
}

/*
 * 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 u64 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, total_added = 0;
	int ret;

	while (start < end) {
		ret = find_first_extent_bit(info->pinned_extents, start,
					    &extent_start, &extent_end,
					    EXTENT_DIRTY | EXTENT_UPTODATE);
		if (ret)
			break;

		if (extent_start == start) {
			start = extent_end + 1;
		} else if (extent_start > start && extent_start < end) {
			size = extent_start - start;
			total_added += size;
			ret = btrfs_add_free_space(block_group, start,
						   size);
			BUG_ON(ret);
			start = extent_end + 1;
		} else {
			break;
		}
	}

	if (start < end) {
		size = end - start;
		total_added += size;
		ret = btrfs_add_free_space(block_group, start, size);
		BUG_ON(ret);
	}

	return total_added;
}

static int caching_kthread(void *data)
{
	struct btrfs_block_group_cache *block_group = data;
	struct btrfs_fs_info *fs_info = block_group->fs_info;
	struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
	struct btrfs_root *extent_root = fs_info->extent_root;
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct btrfs_key key;
	u64 total_found = 0;
	u64 last = 0;
	u32 nritems;
	int ret = 0;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	exclude_super_stripes(extent_root, block_group);
	spin_lock(&block_group->space_info->lock);
	block_group->space_info->bytes_super += block_group->bytes_super;
	spin_unlock(&block_group->space_info->lock);

	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);

	/*
	 * We don't want to deadlock with somebody trying to allocate a new
	 * extent for the extent root while also trying to search the extent
	 * root to add free space.  So we skip locking and search the commit
	 * root, since its read-only
	 */
	path->skip_locking = 1;
	path->search_commit_root = 1;
	path->reada = 2;

	key.objectid = last;
	key.offset = 0;
	key.type = BTRFS_EXTENT_ITEM_KEY;
again:
	mutex_lock(&caching_ctl->mutex);
	/* need to make sure the commit_root doesn't disappear */
	down_read(&fs_info->extent_commit_sem);

	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
	if (ret < 0)
		goto err;

	leaf = path->nodes[0];
	nritems = btrfs_header_nritems(leaf);

	while (1) {
		smp_mb();
		if (fs_info->closing > 1) {
			last = (u64)-1;
			break;
		}

		if (path->slots[0] < nritems) {
			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
		} else {
			ret = find_next_key(path, 0, &key);
			if (ret)
				break;

			caching_ctl->progress = last;
			btrfs_release_path(extent_root, path);
			up_read(&fs_info->extent_commit_sem);
			mutex_unlock(&caching_ctl->mutex);
			if (btrfs_transaction_in_commit(fs_info))
				schedule_timeout(1);
			else
				cond_resched();
			goto again;
		}

		if (key.objectid < block_group->key.objectid) {
			path->slots[0]++;
			continue;
		}

		if (key.objectid >= block_group->key.objectid +
		    block_group->key.offset)
			break;

		if (key.type == BTRFS_EXTENT_ITEM_KEY) {
			total_found += add_new_free_space(block_group,
							  fs_info, last,
							  key.objectid);
			last = key.objectid + key.offset;

			if (total_found > (1024 * 1024 * 2)) {
				total_found = 0;
				wake_up(&caching_ctl->wait);
			}
		}
		path->slots[0]++;
	}
	ret = 0;

	total_found += add_new_free_space(block_group, fs_info, last,
					  block_group->key.objectid +
					  block_group->key.offset);
	caching_ctl->progress = (u64)-1;

	spin_lock(&block_group->lock);
	block_group->caching_ctl = NULL;
	block_group->cached = BTRFS_CACHE_FINISHED;
	spin_unlock(&block_group->lock);

err:
	btrfs_free_path(path);
	up_read(&fs_info->extent_commit_sem);

	free_excluded_extents(extent_root, block_group);

	mutex_unlock(&caching_ctl->mutex);
	wake_up(&caching_ctl->wait);

	put_caching_control(caching_ctl);
	atomic_dec(&block_group->space_info->caching_threads);
	return 0;
}

static int cache_block_group(struct btrfs_block_group_cache *cache)
{
	struct btrfs_fs_info *fs_info = cache->fs_info;
	struct btrfs_caching_control *caching_ctl;
	struct task_struct *tsk;
	int ret = 0;

	smp_mb();
	if (cache->cached != BTRFS_CACHE_NO)
		return 0;

	caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
	BUG_ON(!caching_ctl);

	INIT_LIST_HEAD(&caching_ctl->list);
	mutex_init(&caching_ctl->mutex);
	init_waitqueue_head(&caching_ctl->wait);
	caching_ctl->block_group = cache;
	caching_ctl->progress = cache->key.objectid;
	/* one for caching kthread, one for caching block group list */
	atomic_set(&caching_ctl->count, 2);

	spin_lock(&cache->lock);
	if (cache->cached != BTRFS_CACHE_NO) {
		spin_unlock(&cache->lock);
		kfree(caching_ctl);
		return 0;
	}
	cache->caching_ctl = caching_ctl;
	cache->cached = BTRFS_CACHE_STARTED;
	spin_unlock(&cache->lock);

	down_write(&fs_info->extent_commit_sem);
	list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
	up_write(&fs_info->extent_commit_sem);

	atomic_inc(&cache->space_info->caching_threads);

	tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
			  cache->key.objectid);
	if (IS_ERR(tsk)) {
		ret = PTR_ERR(tsk);
		printk(KERN_ERR "error running thread %d\n", ret);
		BUG();
	}

	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 the 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;
}

void btrfs_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 btrfs_space_info *found;

	rcu_read_lock();
	list_for_each_entry_rcu(found, head, list) {
		if (found->flags == flags) {
			rcu_read_unlock();
			return found;
		}
	}
	rcu_read_unlock();
	return NULL;
}

/*
 * after adding space to the filesystem, we need to clear the full flags
 * on all the space infos.
 */
void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
{
	struct list_head *head = &info->space_info;
	struct btrfs_space_info *found;

	rcu_read_lock();
	list_for_each_entry_rcu(found, head, list)
		found->full = 0;
	rcu_read_unlock();
}

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);
				btrfs_put_block_group(cache);
				goto found;
			}
		}
		spin_unlock(&cache->lock);
		btrfs_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.
 *
 * There are two kinds of back refs. The implicit back refs is optimized
 * for pointers in non-shared tree blocks. For a given pointer in a block,
 * back refs of this kind provide information about the block's owner tree
 * and the pointer's key. These information allow us to find the block by
 * b-tree searching. The full back refs is for pointers in tree blocks not
 * referenced by their owner trees. The location of tree block is recorded
 * in the back refs. Actually the full back refs is generic, and can be
 * used in all cases the implicit back refs is used. The major shortcoming
 * of the full back refs is its overhead. Every time a tree block gets
 * COWed, we have to update back refs entry for all pointers in it.
 *
 * For a newly allocated tree block, we use implicit back refs for
 * pointers in it. This means most tree related operations only involve
 * implicit back refs. For a tree block created in old transaction, the
 * only way to drop a reference to it is COW it. So we can detect the
 * event that tree block loses its owner tree's reference and do the
 * back refs conversion.
 *
 * When a tree block is COW'd through a tree, there are four cases:
 *
 * The reference count of the block is one and the tree is the block's
 * owner tree. Nothing to do in this case.
 *
 * The reference count of the block is one and the tree is not the
 * block's owner tree. In this case, full back refs is used for pointers
 * in the block. Remove these full back refs, add implicit back refs for
 * every pointers in the new block.
 *
 * The reference count of the block is greater than one and the tree is
 * the block's owner tree. In this case, implicit back refs is used for
 * pointers in the block. Add full back refs for every pointers in the
 * block, increase lower level extents' reference counts. The original
 * implicit back refs are entailed to the new block.
 *
 * The reference count of the block is greater than one and the tree is
 * not the block's owner tree. Add implicit back refs for every pointer in
 * the new block, increase lower level extents' reference count.
 *
 * Back Reference Key composing:
 *
 * The key objectid corresponds to the first byte in the extent,
 * The key type is used to differentiate between types of back refs.
 * There are different meanings of the key offset for different types
 * of back refs.
 *
 * 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 for the implicit back refs has fields for:
 *
 * - Objectid of the subvolume root
 * - objectid of the file holding the reference
 * - original offset in the file
 * - how many bookend extents
 *
 * The key offset for the implicit back refs is hash of the first
 * three fields.
 *
 * The extent ref structure for the full back refs has field for:
 *
 * - number of pointers in the tree leaf
 *
 * The key offset for the implicit back refs is the first byte of
 * the tree leaf
 *
 * When a file extent is allocated, The implicit back refs is used.
 * the fields are filled in:
 *
 *     (root_key.objectid, inode objectid, offset in file, 1)
 *
 * When a file extent is removed file truncation, we find the
 * corresponding implicit back refs and check the following fields:
 *
 *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 *
 * Btree extents can be referenced by:
 *
 * - Different subvolumes
 *
 * Both the implicit back refs and the full back refs for tree blocks
 * only consist of key. The key offset for the implicit back refs is
 * objectid of block's owner tree. The key offset for the full back refs
 * is the first byte of parent block.
 *
 * When implicit back refs is used, information about the lowest key and
 * level of the tree block are required. These information are stored in
 * tree block info structure.
 */

#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
				  struct btrfs_root *root,
				  struct btrfs_path *path,
				  u64 owner, u32 extra_size)
{
	struct btrfs_extent_item *item;
	struct btrfs_extent_item_v0 *ei0;
	struct btrfs_extent_ref_v0 *ref0;
	struct btrfs_tree_block_info *bi;
	struct extent_buffer *leaf;
	struct btrfs_key key;
	struct btrfs_key found_key;
	u32 new_size = sizeof(*item);
	u64 refs;
	int ret;

	leaf = path->nodes[0];
	BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));

	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
	ei0 = btrfs_item_ptr(leaf, path->slots[0],
			     struct btrfs_extent_item_v0);
	refs = btrfs_extent_refs_v0(leaf, ei0);

	if (owner == (u64)-1) {
		while (1) {
			if (path->slots[0] >= btrfs_header_nritems(leaf)) {
				ret = btrfs_next_leaf(root, path);
				if (ret < 0)
					return ret;
				BUG_ON(ret > 0);
				leaf = path->nodes[0];
			}
			btrfs_item_key_to_cpu(leaf, &found_key,
					      path->slots[0]);
			BUG_ON(key.objectid != found_key.objectid);
			if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
				path->slots[0]++;
				continue;
			}
			ref0 = btrfs_item_ptr(leaf, path->slots[0],
					      struct btrfs_extent_ref_v0);
			owner = btrfs_ref_objectid_v0(leaf, ref0);
			break;
		}
	}
	btrfs_release_path(root, path);

	if (owner < BTRFS_FIRST_FREE_OBJECTID)
		new_size += sizeof(*bi);

	new_size -= sizeof(*ei0);
	ret = btrfs_search_slot(trans, root, &key, path,
				new_size + extra_size, 1);
	if (ret < 0)
		return ret;
	BUG_ON(ret);

	ret = btrfs_extend_item(trans, root, path, new_size);
	BUG_ON(ret);

	leaf = path->nodes[0];
	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	btrfs_set_extent_refs(leaf, item, refs);
	/* FIXME: get real generation */
	btrfs_set_extent_generation(leaf, item, 0);
	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
		btrfs_set_extent_flags(leaf, item,
				       BTRFS_EXTENT_FLAG_TREE_BLOCK |
				       BTRFS_BLOCK_FLAG_FULL_BACKREF);
		bi = (struct btrfs_tree_block_info *)(item + 1);
		/* FIXME: get first key of the block */
		memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
		btrfs_set_tree_block_level(leaf, bi, (int)owner);
	} else {
		btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
	}
	btrfs_mark_buffer_dirty(leaf);
	return 0;
}
#endif

static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
{
	u32 high_crc = ~(u32)0;
	u32 low_crc = ~(u32)0;
	__le64 lenum;

	lenum = cpu_to_le64(root_objectid);
	high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
	lenum = cpu_to_le64(owner);
	low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
	lenum = cpu_to_le64(offset);
	low_crc = crc32c(low_crc, &lenum, sizeof(lenum));

	return ((u64)high_crc << 31) ^ (u64)low_crc;
}

static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
				     struct btrfs_extent_data_ref *ref)
{
	return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
				    btrfs_extent_data_ref_objectid(leaf, ref),
				    btrfs_extent_data_ref_offset(leaf, ref));
}

static int match_extent_data_ref(struct extent_buffer *leaf,
				 struct btrfs_extent_data_ref *ref,
				 u64 root_objectid, u64 owner, u64 offset)
{
	if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
	    btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
		return 0;
	return 1;
}

static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
					   struct btrfs_root *root,
					   struct btrfs_path *path,
					   u64 bytenr, u64 parent,
					   u64 root_objectid,
					   u64 owner, u64 offset)
{
	struct btrfs_key key;
	struct btrfs_extent_data_ref *ref;
	struct extent_buffer *leaf;
	u32 nritems;
	int ret;
	int recow;
	int err = -ENOENT;

	key.objectid = bytenr;
	if (parent) {
		key.type = BTRFS_SHARED_DATA_REF_KEY;
		key.offset = parent;
	} else {
		key.type = BTRFS_EXTENT_DATA_REF_KEY;
		key.offset = hash_extent_data_ref(root_objectid,
						  owner, offset);
	}
again:
	recow = 0;
	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
	if (ret < 0) {
		err = ret;
		goto fail;
	}

	if (parent) {
		if (!ret)
			return 0;
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
		key.type = BTRFS_EXTENT_REF_V0_KEY;
		btrfs_release_path(root, path);
		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
		if (ret < 0) {
			err = ret;
			goto fail;
		}
		if (!ret)
			return 0;
#endif
		goto fail;
	}

	leaf = path->nodes[0];
	nritems = btrfs_header_nritems(leaf);
	while (1) {
		if (path->slots[0] >= nritems) {
			ret = btrfs_next_leaf(root, path);
			if (ret < 0)
				err = ret;
			if (ret)
				goto fail;

			leaf = path->nodes[0];
			nritems = btrfs_header_nritems(leaf);
			recow = 1;
		}

		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
		if (key.objectid != bytenr ||
		    key.type != BTRFS_EXTENT_DATA_REF_KEY)
			goto fail;

		ref = btrfs_item_ptr(leaf, path->slots[0],
				     struct btrfs_extent_data_ref);

		if (match_extent_data_ref(leaf, ref, root_objectid,
					  owner, offset)) {
			if (recow) {
				btrfs_release_path(root, path);
				goto again;
			}
			err = 0;
			break;
		}
		path->slots[0]++;
	}
fail:
	return err;
}

static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
					   struct btrfs_root *root,
					   struct btrfs_path *path,
					   u64 bytenr, u64 parent,
					   u64 root_objectid, u64 owner,
					   u64 offset, int refs_to_add)
{
	struct btrfs_key key;
	struct extent_buffer *leaf;
	u32 size;
	u32 num_refs;
	int ret;

	key.objectid = bytenr;
	if (parent) {
		key.type = BTRFS_SHARED_DATA_REF_KEY;
		key.offset = parent;
		size = sizeof(struct btrfs_shared_data_ref);
	} else {
		key.type = BTRFS_EXTENT_DATA_REF_KEY;
		key.offset = hash_extent_data_ref(root_objectid,
						  owner, offset);
		size = sizeof(struct btrfs_extent_data_ref);
	}

	ret = btrfs_insert_empty_item(trans, root, path, &key, size);
	if (ret && ret != -EEXIST)
		goto fail;

	leaf = path->nodes[0];
	if (parent) {
		struct btrfs_shared_data_ref *ref;
		ref = btrfs_item_ptr(leaf, path->slots[0],
				     struct btrfs_shared_data_ref);
		if (ret == 0) {
			btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
		} else {
			num_refs = btrfs_shared_data_ref_count(leaf, ref);
			num_refs += refs_to_add;
			btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
		}
	} else {
		struct btrfs_extent_data_ref *ref;
		while (ret == -EEXIST) {
			ref = btrfs_item_ptr(leaf, path->slots[0],
					     struct btrfs_extent_data_ref);
			if (match_extent_data_ref(leaf, ref, root_objectid,
						  owner, offset))
				break;
			btrfs_release_path(root, path);
			key.offset++;
			ret = btrfs_insert_empty_item(trans, root, path, &key,
						      size);
			if (ret && ret != -EEXIST)
				goto fail;

			leaf = path->nodes[0];
		}
		ref = btrfs_item_ptr(leaf, path->slots[0],
				     struct btrfs_extent_data_ref);
		if (ret == 0) {
			btrfs_set_extent_data_ref_root(leaf, ref,
						       root_objectid);
			btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
			btrfs_set_extent_data_ref_offset(leaf, ref, offset);
			btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
		} else {
			num_refs = btrfs_extent_data_ref_count(leaf, ref);
			num_refs += refs_to_add;
			btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
		}
	}
	btrfs_mark_buffer_dirty(leaf);
	ret = 0;
fail:
	btrfs_release_path(root, path);
	return ret;
}

static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
					   struct btrfs_root *root,
					   struct btrfs_path *path,
					   int refs_to_drop)
{
	struct btrfs_key key;
	struct btrfs_extent_data_ref *ref1 = NULL;
	struct btrfs_shared_data_ref *ref2 = NULL;
	struct extent_buffer *leaf;
	u32 num_refs = 0;
	int ret = 0;

	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);

	if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
		ref1 = btrfs_item_ptr(leaf, path->slots[0],
				      struct btrfs_extent_data_ref);
		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
		ref2 = btrfs_item_ptr(leaf, path->slots[0],
				      struct btrfs_shared_data_ref);
		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
	} else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
		struct btrfs_extent_ref_v0 *ref0;
		ref0 = btrfs_item_ptr(leaf, path->slots[0],
				      struct btrfs_extent_ref_v0);
		num_refs = btrfs_ref_count_v0(leaf, ref0);
#endif
	} else {
		BUG();
	}

	BUG_ON(num_refs < refs_to_drop);
	num_refs -= refs_to_drop;

	if (num_refs == 0) {
		ret = btrfs_del_item(trans, root, path);
	} else {
		if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
			btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
		else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
			btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
		else {
			struct btrfs_extent_ref_v0 *ref0;
			ref0 = btrfs_item_ptr(leaf, path->slots[0],
					struct btrfs_extent_ref_v0);
			btrfs_set_ref_count_v0(leaf, ref0, num_refs);
		}
#endif
		btrfs_mark_buffer_dirty(leaf);
	}
	return ret;
}

static noinline u32 extent_data_ref_count(struct btrfs_root *root,
					  struct btrfs_path *path,
					  struct btrfs_extent_inline_ref *iref)
{
	struct btrfs_key key;
	struct extent_buffer *leaf;
	struct btrfs_extent_data_ref *ref1;
	struct btrfs_shared_data_ref *ref2;
	u32 num_refs = 0;

	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
	if (iref) {
		if (btrfs_extent_inline_ref_type(leaf, iref) ==
		    BTRFS_EXTENT_DATA_REF_KEY) {
			ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
			num_refs = btrfs_extent_data_ref_count(leaf, ref1);
		} else {
			ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
			num_refs = btrfs_shared_data_ref_count(leaf, ref2);
		}
	} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
		ref1 = btrfs_item_ptr(leaf, path->slots[0],
				      struct btrfs_extent_data_ref);
		num_refs = btrfs_extent_data_ref_count(leaf, ref1);
	} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
		ref2 = btrfs_item_ptr(leaf, path->slots[0],
				      struct btrfs_shared_data_ref);
		num_refs = btrfs_shared_data_ref_count(leaf, ref2);
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
	} else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
		struct btrfs_extent_ref_v0 *ref0;
		ref0 = btrfs_item_ptr(leaf, path->slots[0],
				      struct btrfs_extent_ref_v0);
		num_refs = btrfs_ref_count_v0(leaf, ref0);
#endif
	} else {
		WARN_ON(1);
	}
	return num_refs;
}

static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
					  struct btrfs_root *root,
					  struct btrfs_path *path,
					  u64 bytenr, u64 parent,
					  u64 root_objectid)
{
	struct btrfs_key key;
	int ret;

	key.objectid = bytenr;
	if (parent) {
		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
		key.offset = parent;
	} else {
		key.type = BTRFS_TREE_BLOCK_REF_KEY;
		key.offset = root_objectid;
	}

	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
	if (ret > 0)
		ret = -ENOENT;
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
	if (ret == -ENOENT && parent) {
		btrfs_release_path(root, path);
		key.type = BTRFS_EXTENT_REF_V0_KEY;
		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
		if (ret > 0)
			ret = -ENOENT;
	}
#endif
	return ret;
}

static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
					  struct btrfs_root *root,
					  struct btrfs_path *path,
					  u64 bytenr, u64 parent,
					  u64 root_objectid)
{
	struct btrfs_key key;
	int ret;

	key.objectid = bytenr;
	if (parent) {
		key.type = BTRFS_SHARED_BLOCK_REF_KEY;
		key.offset = parent;
	} else {
		key.type = BTRFS_TREE_BLOCK_REF_KEY;
		key.offset = root_objectid;
	}

	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
	btrfs_release_path(root, path);
	return ret;
}

static inline int extent_ref_type(u64 parent, u64 owner)
{
	int type;
	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
		if (parent > 0)
			type = BTRFS_SHARED_BLOCK_REF_KEY;
		else
			type = BTRFS_TREE_BLOCK_REF_KEY;
	} else {
		if (parent > 0)
			type = BTRFS_SHARED_DATA_REF_KEY;
		else
			type = BTRFS_EXTENT_DATA_REF_KEY;
	}
	return type;
}

static int find_next_key(struct btrfs_path *path, int level,
			 struct btrfs_key *key)

{
	for (; level < BTRFS_MAX_LEVEL; level++) {
		if (!path->nodes[level])
			break;
		if (path->slots[level] + 1 >=
		    btrfs_header_nritems(path->nodes[level]))
			continue;
		if (level == 0)
			btrfs_item_key_to_cpu(path->nodes[level], key,
					      path->slots[level] + 1);
		else
			btrfs_node_key_to_cpu(path->nodes[level], key,
					      path->slots[level] + 1);
		return 0;
	}
	return 1;
}

/*
 * look for inline back ref. if back ref is found, *ref_ret is set
 * to the address of inline back ref, and 0 is returned.
 *
 * if back ref isn't found, *ref_ret is set to the address where it
 * should be inserted, and -ENOENT is returned.
 *
 * if insert is true and there are too many inline back refs, the path
 * points to the extent item, and -EAGAIN is returned.
 *
 * NOTE: inline back refs are ordered in the same way that back ref
 *	 items in the tree are ordered.
 */
static noinline_for_stack
int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_path *path,
				 struct btrfs_extent_inline_ref **ref_ret,
				 u64 bytenr, u64 num_bytes,
				 u64 parent, u64 root_objectid,
				 u64 owner, u64 offset, int insert)
{
	struct btrfs_key key;
	struct extent_buffer *leaf;
	struct btrfs_extent_item *ei;
	struct btrfs_extent_inline_ref *iref;
	u64 flags;
	u64 item_size;
	unsigned long ptr;
	unsigned long end;
	int extra_size;
	int type;
	int want;
	int ret;
	int err = 0;

	key.objectid = bytenr;
	key.type = BTRFS_EXTENT_ITEM_KEY;
	key.offset = num_bytes;

	want = extent_ref_type(parent, owner);
	if (insert) {
		extra_size = btrfs_extent_inline_ref_size(want);
		path->keep_locks = 1;
	} else
		extra_size = -1;
	ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
	if (ret < 0) {
		err = ret;
		goto out;
	}
	BUG_ON(ret);

	leaf = path->nodes[0];
	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
	if (item_size < sizeof(*ei)) {
		if (!insert) {
			err = -ENOENT;
			goto out;
		}
		ret = convert_extent_item_v0(trans, root, path, owner,
					     extra_size);
		if (ret < 0) {
			err = ret;
			goto out;
		}
		leaf = path->nodes[0];
		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
	}
#endif
	BUG_ON(item_size < sizeof(*ei));

	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	flags = btrfs_extent_flags(leaf, ei);

	ptr = (unsigned long)(ei + 1);
	end = (unsigned long)ei + item_size;

	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
		ptr += sizeof(struct btrfs_tree_block_info);
		BUG_ON(ptr > end);
	} else {
		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
	}

	err = -ENOENT;
	while (1) {
		if (ptr >= end) {
			WARN_ON(ptr > end);
			break;
		}
		iref = (struct btrfs_extent_inline_ref *)ptr;
		type = btrfs_extent_inline_ref_type(leaf, iref);
		if (want < type)
			break;
		if (want > type) {
			ptr += btrfs_extent_inline_ref_size(type);
			continue;
		}

		if (type == BTRFS_EXTENT_DATA_REF_KEY) {
			struct btrfs_extent_data_ref *dref;
			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
			if (match_extent_data_ref(leaf, dref, root_objectid,
						  owner, offset)) {
				err = 0;
				break;
			}
			if (hash_extent_data_ref_item(leaf, dref) <
			    hash_extent_data_ref(root_objectid, owner, offset))
				break;
		} else {
			u64 ref_offset;
			ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
			if (parent > 0) {
				if (parent == ref_offset) {
					err = 0;
					break;
				}
				if (ref_offset < parent)
					break;
			} else {
				if (root_objectid == ref_offset) {
					err = 0;
					break;
				}
				if (ref_offset < root_objectid)
					break;
			}
		}
		ptr += btrfs_extent_inline_ref_size(type);
	}
	if (err == -ENOENT && insert) {
		if (item_size + extra_size >=
		    BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
			err = -EAGAIN;
			goto out;
		}
		/*
		 * To add new inline back ref, we have to make sure
		 * there is no corresponding back ref item.
		 * For simplicity, we just do not add new inline back
		 * ref if there is any kind of item for this block
		 */
		if (find_next_key(path, 0, &key) == 0 &&
		    key.objectid == bytenr &&
		    key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
			err = -EAGAIN;
			goto out;
		}
	}
	*ref_ret = (struct btrfs_extent_inline_ref *)ptr;
out:
	if (insert) {
		path->keep_locks = 0;
		btrfs_unlock_up_safe(path, 1);
	}
	return err;
}

/*
 * helper to add new inline back ref
 */
static noinline_for_stack
int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
				struct btrfs_root *root,
				struct btrfs_path *path,
				struct btrfs_extent_inline_ref *iref,
				u64 parent, u64 root_objectid,
				u64 owner, u64 offset, int refs_to_add,
				struct btrfs_delayed_extent_op *extent_op)
{
	struct extent_buffer *leaf;
	struct btrfs_extent_item *ei;
	unsigned long ptr;
	unsigned long end;
	unsigned long item_offset;
	u64 refs;
	int size;
	int type;
	int ret;

	leaf = path->nodes[0];
	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	item_offset = (unsigned long)iref - (unsigned long)ei;

	type = extent_ref_type(parent, owner);
	size = btrfs_extent_inline_ref_size(type);

	ret = btrfs_extend_item(trans, root, path, size);
	BUG_ON(ret);

	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	refs = btrfs_extent_refs(leaf, ei);
	refs += refs_to_add;
	btrfs_set_extent_refs(leaf, ei, refs);
	if (extent_op)
		__run_delayed_extent_op(extent_op, leaf, ei);

	ptr = (unsigned long)ei + item_offset;
	end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
	if (ptr < end - size)
		memmove_extent_buffer(leaf, ptr + size, ptr,
				      end - size - ptr);

	iref = (struct btrfs_extent_inline_ref *)ptr;
	btrfs_set_extent_inline_ref_type(leaf, iref, type);
	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
		struct btrfs_extent_data_ref *dref;
		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
		btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
		btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
		btrfs_set_extent_data_ref_offset(leaf, dref, offset);
		btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
		struct btrfs_shared_data_ref *sref;
		sref = (struct btrfs_shared_data_ref *)(iref + 1);
		btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
	} else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
		btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
	} else {
		btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
	}
	btrfs_mark_buffer_dirty(leaf);
	return 0;
}

static int lookup_extent_backref(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_path *path,
				 struct btrfs_extent_inline_ref **ref_ret,
				 u64 bytenr, u64 num_bytes, u64 parent,
				 u64 root_objectid, u64 owner, u64 offset)
{
	int ret;

	ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
					   bytenr, num_bytes, parent,
					   root_objectid, owner, offset, 0);
	if (ret != -ENOENT)
		return ret;

	btrfs_release_path(root, path);
	*ref_ret = NULL;

	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
		ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
					    root_objectid);
	} else {
		ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
					     root_objectid, owner, offset);
	}
	return ret;
}

/*
 * helper to update/remove inline back ref
 */
static noinline_for_stack
int update_inline_extent_backref(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_path *path,
				 struct btrfs_extent_inline_ref *iref,
				 int refs_to_mod,
				 struct btrfs_delayed_extent_op *extent_op)
{
	struct extent_buffer *leaf;
	struct btrfs_extent_item *ei;
	struct btrfs_extent_data_ref *dref = NULL;
	struct btrfs_shared_data_ref *sref = NULL;
	unsigned long ptr;
	unsigned long end;
	u32 item_size;
	int size;
	int type;
	int ret;
	u64 refs;

	leaf = path->nodes[0];
	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	refs = btrfs_extent_refs(leaf, ei);
	WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
	refs += refs_to_mod;
	btrfs_set_extent_refs(leaf, ei, refs);
	if (extent_op)
		__run_delayed_extent_op(extent_op, leaf, ei);

	type = btrfs_extent_inline_ref_type(leaf, iref);

	if (type == BTRFS_EXTENT_DATA_REF_KEY) {
		dref = (struct btrfs_extent_data_ref *)(&iref->offset);
		refs = btrfs_extent_data_ref_count(leaf, dref);
	} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
		sref = (struct btrfs_shared_data_ref *)(iref + 1);
		refs = btrfs_shared_data_ref_count(leaf, sref);
	} else {
		refs = 1;
		BUG_ON(refs_to_mod != -1);
	}

	BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
	refs += refs_to_mod;

	if (refs > 0) {
		if (type == BTRFS_EXTENT_DATA_REF_KEY)
			btrfs_set_extent_data_ref_count(leaf, dref, refs);
		else
			btrfs_set_shared_data_ref_count(leaf, sref, refs);
	} else {
		size =  btrfs_extent_inline_ref_size(type);
		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
		ptr = (unsigned long)iref;
		end = (unsigned long)ei + item_size;
		if (ptr + size < end)
			memmove_extent_buffer(leaf, ptr, ptr + size,
					      end - ptr - size);
		item_size -= size;
		ret = btrfs_truncate_item(trans, root, path, item_size, 1);
		BUG_ON(ret);
	}
	btrfs_mark_buffer_dirty(leaf);
	return 0;
}

static noinline_for_stack
int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_path *path,
				 u64 bytenr, u64 num_bytes, u64 parent,
				 u64 root_objectid, u64 owner,
				 u64 offset, int refs_to_add,
				 struct btrfs_delayed_extent_op *extent_op)
{
	struct btrfs_extent_inline_ref *iref;
	int ret;

	ret = lookup_inline_extent_backref(trans, root, path, &iref,
					   bytenr, num_bytes, parent,
					   root_objectid, owner, offset, 1);
	if (ret == 0) {
		BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
		ret = update_inline_extent_backref(trans, root, path, iref,
						   refs_to_add, extent_op);
	} else if (ret == -ENOENT) {
		ret = setup_inline_extent_backref(trans, root, path, iref,
						  parent, root_objectid,
						  owner, offset, refs_to_add,
						  extent_op);
	}
	return ret;
}

static int insert_extent_backref(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_path *path,
				 u64 bytenr, u64 parent, u64 root_objectid,
				 u64 owner, u64 offset, int refs_to_add)
{
	int ret;
	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
		BUG_ON(refs_to_add != 1);
		ret = insert_tree_block_ref(trans, root, path, bytenr,
					    parent, root_objectid);
	} else {
		ret = insert_extent_data_ref(trans, root, path, bytenr,
					     parent, root_objectid,
					     owner, offset, refs_to_add);
	}
	return ret;
}

static int remove_extent_backref(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_path *path,
				 struct btrfs_extent_inline_ref *iref,
				 int refs_to_drop, int is_data)
{
	int ret;

	BUG_ON(!is_data && refs_to_drop != 1);
	if (iref) {
		ret = update_inline_extent_backref(trans, root, path, iref,
						   -refs_to_drop, NULL);
	} else if (is_data) {
		ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
	} else {
		ret = btrfs_del_item(trans, root, path);
	}
	return ret;
}

static void btrfs_issue_discard(struct block_device *bdev,
				u64 start, u64 len)
{
	blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
			     DISCARD_FL_BARRIER);
}

static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
				u64 num_bytes)
{
	int ret;
	u64 map_length = num_bytes;
	struct btrfs_multi_bio *multi = NULL;

	if (!btrfs_test_opt(root, DISCARD))
		return 0;

	/* 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;
}

int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
			 struct btrfs_root *root,
			 u64 bytenr, u64 num_bytes, u64 parent,
			 u64 root_objectid, u64 owner, u64 offset)
{
	int ret;
	BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
	       root_objectid == BTRFS_TREE_LOG_OBJECTID);

	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
		ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
					parent, root_objectid, (int)owner,
					BTRFS_ADD_DELAYED_REF, NULL);
	} else {
		ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
					parent, root_objectid, owner, offset,
					BTRFS_ADD_DELAYED_REF, NULL);
	}
	return ret;
}

static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
				  struct btrfs_root *root,
				  u64 bytenr, u64 num_bytes,
				  u64 parent, u64 root_objectid,
				  u64 owner, u64 offset, int refs_to_add,
				  struct btrfs_delayed_extent_op *extent_op)
{
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct btrfs_extent_item *item;
	u64 refs;
	int ret;
	int err = 0;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	path->reada = 1;
	path->leave_spinning = 1;
	/* this will setup the path even if it fails to insert the back ref */
	ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
					   path, bytenr, num_bytes, parent,
					   root_objectid, owner, offset,
					   refs_to_add, extent_op);
	if (ret == 0)
		goto out;

	if (ret != -EAGAIN) {
		err = ret;
		goto out;
	}

	leaf = path->nodes[0];
	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	refs = btrfs_extent_refs(leaf, item);
	btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
	if (extent_op)
		__run_delayed_extent_op(extent_op, leaf, item);

	btrfs_mark_buffer_dirty(leaf);
	btrfs_release_path(root->fs_info->extent_root, path);

	path->reada = 1;
	path->leave_spinning = 1;

	/* now insert the actual backref */
	ret = insert_extent_backref(trans, root->fs_info->extent_root,
				    path, bytenr, parent, root_objectid,
				    owner, offset, refs_to_add);
	BUG_ON(ret);
out:
	btrfs_free_path(path);
	return err;
}

static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
				struct btrfs_root *root,
				struct btrfs_delayed_ref_node *node,
				struct btrfs_delayed_extent_op *extent_op,
				int insert_reserved)
{
	int ret = 0;
	struct btrfs_delayed_data_ref *ref;
	struct btrfs_key ins;
	u64 parent = 0;
	u64 ref_root = 0;
	u64 flags = 0;

	ins.objectid = node->bytenr;
	ins.offset = node->num_bytes;
	ins.type = BTRFS_EXTENT_ITEM_KEY;

	ref = btrfs_delayed_node_to_data_ref(node);
	if (node->type == BTRFS_SHARED_DATA_REF_KEY)
		parent = ref->parent;
	else
		ref_root = ref->root;

	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
		if (extent_op) {
			BUG_ON(extent_op->update_key);
			flags |= extent_op->flags_to_set;
		}
		ret = alloc_reserved_file_extent(trans, root,
						 parent, ref_root, flags,
						 ref->objectid, ref->offset,
						 &ins, node->ref_mod);
	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
					     node->num_bytes, parent,
					     ref_root, ref->objectid,
					     ref->offset, node->ref_mod,
					     extent_op);
	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
		ret = __btrfs_free_extent(trans, root, node->bytenr,
					  node->num_bytes, parent,
					  ref_root, ref->objectid,
					  ref->offset, node->ref_mod,
					  extent_op);
	} else {
		BUG();
	}
	return ret;
}

static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
				    struct extent_buffer *leaf,
				    struct btrfs_extent_item *ei)
{
	u64 flags = btrfs_extent_flags(leaf, ei);
	if (extent_op->update_flags) {
		flags |= extent_op->flags_to_set;
		btrfs_set_extent_flags(leaf, ei, flags);
	}

	if (extent_op->update_key) {
		struct btrfs_tree_block_info *bi;
		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
		bi = (struct btrfs_tree_block_info *)(ei + 1);
		btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
	}
}

static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root,
				 struct btrfs_delayed_ref_node *node,
				 struct btrfs_delayed_extent_op *extent_op)
{
	struct btrfs_key key;
	struct btrfs_path *path;
	struct btrfs_extent_item *ei;
	struct extent_buffer *leaf;
	u32 item_size;
	int ret;
	int err = 0;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	key.objectid = node->bytenr;
	key.type = BTRFS_EXTENT_ITEM_KEY;
	key.offset = node->num_bytes;

	path->reada = 1;
	path->leave_spinning = 1;
	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
				path, 0, 1);
	if (ret < 0) {
		err = ret;
		goto out;
	}
	if (ret > 0) {
		err = -EIO;
		goto out;
	}

	leaf = path->nodes[0];
	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
	if (item_size < sizeof(*ei)) {
		ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
					     path, (u64)-1, 0);
		if (ret < 0) {
			err = ret;
			goto out;
		}
		leaf = path->nodes[0];
		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
	}
#endif
	BUG_ON(item_size < sizeof(*ei));
	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
	__run_delayed_extent_op(extent_op, leaf, ei);

	btrfs_mark_buffer_dirty(leaf);
out:
	btrfs_free_path(path);
	return err;
}

static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
				struct btrfs_root *root,
				struct btrfs_delayed_ref_node *node,
				struct btrfs_delayed_extent_op *extent_op,
				int insert_reserved)
{
	int ret = 0;
	struct btrfs_delayed_tree_ref *ref;
	struct btrfs_key ins;
	u64 parent = 0;
	u64 ref_root = 0;

	ins.objectid = node->bytenr;
	ins.offset = node->num_bytes;
	ins.type = BTRFS_EXTENT_ITEM_KEY;

	ref = btrfs_delayed_node_to_tree_ref(node);
	if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
		parent = ref->parent;
	else
		ref_root = ref->root;

	BUG_ON(node->ref_mod != 1);
	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
		BUG_ON(!extent_op || !extent_op->update_flags ||
		       !extent_op->update_key);
		ret = alloc_reserved_tree_block(trans, root,
						parent, ref_root,
						extent_op->flags_to_set,
						&extent_op->key,
						ref->level, &ins);
	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
					     node->num_bytes, parent, ref_root,
					     ref->level, 0, 1, extent_op);
	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
		ret = __btrfs_free_extent(trans, root, node->bytenr,
					  node->num_bytes, parent, ref_root,
					  ref->level, 0, 1, extent_op);
	} else {
		BUG();
	}
	return ret;
}


/* helper function to actually process a single delayed ref entry */
static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root,
			       struct btrfs_delayed_ref_node *node,
			       struct btrfs_delayed_extent_op *extent_op,
			       int insert_reserved)
{
	int ret;
	if (btrfs_delayed_ref_is_head(node)) {
		struct btrfs_delayed_ref_head *head;
		/*
		 * we've hit the end of the chain and we were supposed
		 * to insert this extent into the tree.  But, it got
		 * deleted before we ever needed to insert it, so all
		 * we have to do is clean up the accounting
		 */
		BUG_ON(extent_op);
		head = btrfs_delayed_node_to_head(node);
		if (insert_reserved) {
			int mark_free = 0;
			struct extent_buffer *must_clean = NULL;

			ret = pin_down_bytes(trans, root, NULL,
					     node->bytenr, node->num_bytes,
					     head->is_data, 1, &must_clean);
			if (ret > 0)
				mark_free = 1;

			if (must_clean) {
				clean_tree_block(NULL, root, must_clean);
				btrfs_tree_unlock(must_clean);
				free_extent_buffer(must_clean);
			}
			if (head->is_data) {
				ret = btrfs_del_csums(trans, root,
						      node->bytenr,
						      node->num_bytes);
				BUG_ON(ret);
			}
			if (mark_free) {
				ret = btrfs_free_reserved_extent(root,
							node->bytenr,
							node->num_bytes);
				BUG_ON(ret);
			}
		}
		mutex_unlock(&head->mutex);
		return 0;
	}

	if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
	    node->type == BTRFS_SHARED_BLOCK_REF_KEY)
		ret = run_delayed_tree_ref(trans, root, node, extent_op,
					   insert_reserved);
	else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
		 node->type == BTRFS_SHARED_DATA_REF_KEY)
		ret = run_delayed_data_ref(trans, root, node, extent_op,
					   insert_reserved);
	else
		BUG();
	return ret;
}

static noinline struct btrfs_delayed_ref_node *
select_delayed_ref(struct btrfs_delayed_ref_head *head)
{
	struct rb_node *node;
	struct btrfs_delayed_ref_node *ref;
	int action = BTRFS_ADD_DELAYED_REF;
again:
	/*
	 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
	 * this prevents ref count from going down to zero when
	 * there still are pending delayed ref.
	 */
	node = rb_prev(&head->node.rb_node);
	while (1) {
		if (!node)
			break;
		ref = rb_entry(node, struct btrfs_delayed_ref_node,
				rb_node);
		if (ref->bytenr != head->node.bytenr)
			break;
		if (ref->action == action)
			return ref;
		node = rb_prev(node);
	}
	if (action == BTRFS_ADD_DELAYED_REF) {
		action = BTRFS_DROP_DELAYED_REF;
		goto again;
	}
	return NULL;
}

static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
				       struct btrfs_root *root,
				       struct list_head *cluster)
{
	struct btrfs_delayed_ref_root *delayed_refs;
	struct btrfs_delayed_ref_node *ref;
	struct btrfs_delayed_ref_head *locked_ref = NULL;
	struct btrfs_delayed_extent_op *extent_op;
	int ret;
	int count = 0;
	int must_insert_reserved = 0;

	delayed_refs = &trans->transaction->delayed_refs;
	while (1) {
		if (!locked_ref) {
			/* pick a new head ref from the cluster list */
			if (list_empty(cluster))
				break;

			locked_ref = list_entry(cluster->next,
				     struct btrfs_delayed_ref_head, cluster);

			/* grab the lock that says we are going to process
			 * all the refs for this head */
			ret = btrfs_delayed_ref_lock(trans, locked_ref);

			/*
			 * we may have dropped the spin lock to get the head
			 * mutex lock, and that might have given someone else
			 * time to free the head.  If that's true, it has been
			 * removed from our list and we can move on.
			 */
			if (ret == -EAGAIN) {
				locked_ref = NULL;
				count++;
				continue;
			}
		}

		/*
		 * record the must insert reserved flag before we
		 * drop the spin lock.
		 */
		must_insert_reserved = locked_ref->must_insert_reserved;
		locked_ref->must_insert_reserved = 0;

		extent_op = locked_ref->extent_op;
		locked_ref->extent_op = NULL;

		/*
		 * locked_ref is the head node, so we have to go one
		 * node back for any delayed ref updates
		 */
		ref = select_delayed_ref(locked_ref);
		if (!ref) {
			/* All delayed refs have been processed, Go ahead
			 * and send the head node to run_one_delayed_ref,
			 * so that any accounting fixes can happen
			 */
			ref = &locked_ref->node;

			if (extent_op && must_insert_reserved) {
				kfree(extent_op);
				extent_op = NULL;
			}

			if (extent_op) {
				spin_unlock(&delayed_refs->lock);

				ret = run_delayed_extent_op(trans, root,
							    ref, extent_op);
				BUG_ON(ret);
				kfree(extent_op);

				cond_resched();
				spin_lock(&delayed_refs->lock);
				continue;
			}

			list_del_init(&locked_ref->cluster);
			locked_ref = NULL;
		}

		ref->in_tree = 0;
		rb_erase(&ref->rb_node, &delayed_refs->root);
		delayed_refs->num_entries--;

		spin_unlock(&delayed_refs->lock);

		ret = run_one_delayed_ref(trans, root, ref, extent_op,
					  must_insert_reserved);
		BUG_ON(ret);

		btrfs_put_delayed_ref(ref);
		kfree(extent_op);
		count++;

		cond_resched();
		spin_lock(&delayed_refs->lock);
	}
	return count;
}

/*
 * this starts processing the delayed reference count updates and
 * extent insertions we have queued up so far.  count can be
 * 0, which means to process everything in the tree at the start
 * of the run (but not newly added entries), or it can be some target
 * number you'd like to process.
 */
int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
			   struct btrfs_root *root, unsigned long count)
{
	struct rb_node *node;
	struct btrfs_delayed_ref_root *delayed_refs;
	struct btrfs_delayed_ref_node *ref;
	struct list_head cluster;
	int ret;
	int run_all = count == (unsigned long)-1;
	int run_most = 0;

	if (root == root->fs_info->extent_root)
		root = root->fs_info->tree_root;

	delayed_refs = &trans->transaction->delayed_refs;
	INIT_LIST_HEAD(&cluster);
again:
	spin_lock(&delayed_refs->lock);
	if (count == 0) {
		count = delayed_refs->num_entries * 2;
		run_most = 1;
	}
	while (1) {
		if (!(run_all || run_most) &&
		    delayed_refs->num_heads_ready < 64)
			break;

		/*
		 * go find something we can process in the rbtree.  We start at
		 * the beginning of the tree, and then build a cluster
		 * of refs to process starting at the first one we are able to
		 * lock
		 */
		ret = btrfs_find_ref_cluster(trans, &cluster,
					     delayed_refs->run_delayed_start);
		if (ret)
			break;

		ret = run_clustered_refs(trans, root, &cluster);
		BUG_ON(ret < 0);

		count -= min_t(unsigned long, ret, count);

		if (count == 0)
			break;
	}

	if (run_all) {
		node = rb_first(&delayed_refs->root);
		if (!node)
			goto out;
		count = (unsigned long)-1;

		while (node) {
			ref = rb_entry(node, struct btrfs_delayed_ref_node,
				       rb_node);
			if (btrfs_delayed_ref_is_head(ref)) {
				struct btrfs_delayed_ref_head *head;

				head = btrfs_delayed_node_to_head(ref);
				atomic_inc(&ref->refs);

				spin_unlock(&delayed_refs->lock);
				mutex_lock(&head->mutex);
				mutex_unlock(&head->mutex);

				btrfs_put_delayed_ref(ref);
				cond_resched();
				goto again;
			}
			node = rb_next(node);
		}
		spin_unlock(&delayed_refs->lock);
		schedule_timeout(1);
		goto again;
	}
out:
	spin_unlock(&delayed_refs->lock);
	return 0;
}

int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
				struct btrfs_root *root,
				u64 bytenr, u64 num_bytes, u64 flags,
				int is_data)
{
	struct btrfs_delayed_extent_op *extent_op;
	int ret;

	extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
	if (!extent_op)
		return -ENOMEM;

	extent_op->flags_to_set = flags;
	extent_op->update_flags = 1;
	extent_op->update_key = 0;
	extent_op->is_data = is_data ? 1 : 0;

	ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
	if (ret)
		kfree(extent_op);
	return ret;
}

static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
				      struct btrfs_root *root,
				      struct btrfs_path *path,
				      u64 objectid, u64 offset, u64 bytenr)
{
	struct btrfs_delayed_ref_head *head;
	struct btrfs_delayed_ref_node *ref;
	struct btrfs_delayed_data_ref *data_ref;
	struct btrfs_delayed_ref_root *delayed_refs;
	struct rb_node *node;
	int ret = 0;

	ret = -ENOENT;
	delayed_refs = &trans->transaction->delayed_refs;
	spin_lock(&delayed_refs->lock);
	head = btrfs_find_delayed_ref_head(trans, bytenr);
	if (!head)
		goto out;

	if (!mutex_trylock(&head->mutex)) {
		atomic_inc(&head->node.refs);
		spin_unlock(&delayed_refs->lock);

		btrfs_release_path(root->fs_info->extent_root, path);

		mutex_lock(&head->mutex);
		mutex_unlock(&head->mutex);
		btrfs_put_delayed_ref(&head->node);
		return -EAGAIN;
	}

	node = rb_prev(&head->node.rb_node);
	if (!node)
		goto out_unlock;

	ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);

	if (ref->bytenr != bytenr)
		goto out_unlock;

	ret = 1;
	if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
		goto out_unlock;

	data_ref = btrfs_delayed_node_to_data_ref(ref);

	node = rb_prev(node);
	if (node) {
		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
		if (ref->bytenr == bytenr)
			goto out_unlock;
	}

	if (data_ref->root != root->root_key.objectid ||
	    data_ref->objectid != objectid || data_ref->offset != offset)
		goto out_unlock;

	ret = 0;
out_unlock:
	mutex_unlock(&head->mutex);
out:
	spin_unlock(&delayed_refs->lock);
	return ret;
}

static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
					struct btrfs_root *root,
					struct btrfs_path *path,
					u64 objectid, u64 offset, u64 bytenr)
{
	struct btrfs_root *extent_root = root->fs_info->extent_root;
	struct extent_buffer *leaf;
	struct btrfs_extent_data_ref *ref;
	struct btrfs_extent_inline_ref *iref;
	struct btrfs_extent_item *ei;
	struct btrfs_key key;
	u32 item_size;
	int ret;

	key.objectid = bytenr;
	key.offset = (u64)-1;
	key.type = BTRFS_EXTENT_ITEM_KEY;

	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, &key, path->slots[0]);

	if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
		goto out;

	ret = 1;
	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
	if (item_size < sizeof(*ei)) {
		WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
		goto out;
	}
#endif
	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);

	if (item_size != sizeof(*ei) +
	    btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
		goto out;

	if (btrfs_extent_generation(leaf, ei) <=
	    btrfs_root_last_snapshot(&root->root_item))
		goto out;

	iref = (struct btrfs_extent_inline_ref *)(ei + 1);
	if (btrfs_extent_inline_ref_type(leaf, iref) !=
	    BTRFS_EXTENT_DATA_REF_KEY)
		goto out;

	ref = (struct btrfs_extent_data_ref *)(&iref->offset);
	if (btrfs_extent_refs(leaf, ei) !=
	    btrfs_extent_data_ref_count(leaf, ref) ||
	    btrfs_extent_data_ref_root(leaf, ref) !=
	    root->root_key.objectid ||
	    btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
	    btrfs_extent_data_ref_offset(leaf, ref) != offset)
		goto out;

	ret = 0;
out:
	return ret;
}

int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root,
			  u64 objectid, u64 offset, u64 bytenr)
{
	struct btrfs_path *path;
	int ret;
	int ret2;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOENT;

	do {
		ret = check_committed_ref(trans, root, path, objectid,
					  offset, bytenr);
		if (ret && ret != -ENOENT)
			goto out;

		ret2 = check_delayed_ref(trans, root, path, objectid,
					 offset, bytenr);
	} while (ret2 == -EAGAIN);

	if (ret2 && ret2 != -ENOENT) {
		ret = ret2;
		goto out;
	}

	if (ret != -ENOENT || ret2 != -ENOENT)
		ret = 0;
out:
	btrfs_free_path(path);
	return ret;
}

#if 0
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;
}

/* when a block goes through cow, we update the reference counts of
 * everything that block points to.  The internal pointers of the block
 * can be in just about any order, and it is likely to have clusters of
 * things that are close together and clusters of things that are not.
 *
 * To help reduce the seeks that come with updating all of these reference
 * counts, sort them by byte number before actual updates are done.
 *
 * struct refsort is used to match byte number to slot in the btree block.
 * we sort based on the byte number and then use the slot to actually
 * find the item.
 *
 * struct refsort is smaller than strcut btrfs_item and smaller than
 * struct btrfs_key_ptr.  Since we're currently limited to the page size
 * for a btree block, there's no way for a kmalloc of refsorts for a
 * single node to be bigger than a page.
 */
struct refsort {
	u64 bytenr;
	u32 slot;
};

/*
 * for passing into sort()
 */
static int refsort_cmp(const void *a_void, const void *b_void)
{
	const struct refsort *a = a_void;
	const struct refsort *b = b_void;

	if (a->bytenr < b->bytenr)
		return -1;
	if (a->bytenr > b->bytenr)
		return 1;
	return 0;
}
#endif

static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
			   struct btrfs_root *root,
			   struct extent_buffer *buf,
			   int full_backref, int inc)
{
	u64 bytenr;
	u64 num_bytes;
	u64 parent;
	u64 ref_root;
	u32 nritems;
	struct btrfs_key key;
	struct btrfs_file_extent_item *fi;
	int i;
	int level;
	int ret = 0;
	int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
			    u64, u64, u64, u64, u64, u64);

	ref_root = btrfs_header_owner(buf);
	nritems = btrfs_header_nritems(buf);
	level = btrfs_header_level(buf);

	if (!root->ref_cows && level == 0)
		return 0;

	if (inc)
		process_func = btrfs_inc_extent_ref;
	else
		process_func = btrfs_free_extent;

	if (full_backref)
		parent = buf->start;
	else
		parent = 0;

	for (i = 0; i < nritems; i++) {
		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;

			num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
			key.offset -= btrfs_file_extent_offset(buf, fi);
			ret = process_func(trans, root, bytenr, num_bytes,
					   parent, ref_root, key.objectid,
					   key.offset);
			if (ret)
				goto fail;
		} else {
			bytenr = btrfs_node_blockptr(buf, i);
			num_bytes = btrfs_level_size(root, level - 1);
			ret = process_func(trans, root, bytenr, num_bytes,
					   parent, ref_root, level - 1, 0);
			if (ret)
				goto fail;
		}
	}
	return 0;
fail:
	BUG();
	return ret;
}

int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
		  struct extent_buffer *buf, int full_backref)
{
	return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
}

int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
		  struct extent_buffer *buf, int full_backref)
{
	return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
}

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;
	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:
	if (ret)
		return ret;
	return 0;

}

static struct btrfs_block_group_cache *
next_block_group(struct btrfs_root *root,
		 struct btrfs_block_group_cache *cache)
{
	struct rb_node *node;
	spin_lock(&root->fs_info->block_group_cache_lock);
	node = rb_next(&cache->cache_node);
	btrfs_put_block_group(cache);
	if (node) {
		cache = rb_entry(node, struct btrfs_block_group_cache,
				 cache_node);
		atomic_inc(&cache->count);
	} else
		cache = NULL;
	spin_unlock(&root->fs_info->block_group_cache_lock);
	return cache;
}

int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
	struct btrfs_block_group_cache *cache;
	int err = 0;
	struct btrfs_path *path;
	u64 last = 0;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	while (1) {
		if (last == 0) {
			err = btrfs_run_delayed_refs(trans, root,
						     (unsigned long)-1);
			BUG_ON(err);
		}

		cache = btrfs_lookup_first_block_group(root->fs_info, last);
		while (cache) {
			if (cache->dirty)
				break;
			cache = next_block_group(root, cache);
		}
		if (!cache) {
			if (last == 0)
				break;
			last = 0;
			continue;
		}

		cache->dirty = 0;
		last = cache->key.objectid + cache->key.offset;

		err = write_one_cache_group(trans, root, path, cache);
		BUG_ON(err);
		btrfs_put_block_group(cache);
	}

	btrfs_free_path(path);
	return 0;
}

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)
		btrfs_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;

	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->bytes_delalloc = 0;
	found->full = 0;
	found->force_alloc = 0;
	*space_info = found;
	list_add_rcu(&found->list, &info->space_info);
	atomic_set(&found->caching_threads, 0);
	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 u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
{
	struct btrfs_fs_info *info = root->fs_info;
	u64 alloc_profile;

	if (data) {
		alloc_profile = info->avail_data_alloc_bits &
			info->data_alloc_profile;
		data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
	} else if (root == root->fs_info->chunk_root) {
		alloc_profile = info->avail_system_alloc_bits &
			info->system_alloc_profile;
		data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
	} else {
		alloc_profile = info->avail_metadata_alloc_bits &
			info->metadata_alloc_profile;
		data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
	}

	return btrfs_reduce_alloc_profile(root, data);
}

void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
{
	u64 alloc_target;

	alloc_target = btrfs_get_alloc_profile(root, 1);
	BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
						       alloc_target);
}

static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
{
	u64 num_bytes;
	int level;

	level = BTRFS_MAX_LEVEL - 2;
	/*
	 * NOTE: these calculations are absolutely the worst possible case.
	 * This assumes that _every_ item we insert will require a new leaf, and
	 * that the tree has grown to its maximum level size.
	 */

	/*
	 * for every item we insert we could insert both an extent item and a
	 * extent ref item.  Then for ever item we insert, we will need to cow
	 * both the original leaf, plus the leaf to the left and right of it.
	 *
	 * Unless we are talking about the extent root, then we just want the
	 * number of items * 2, since we just need the extent item plus its ref.
	 */
	if (root == root->fs_info->extent_root)
		num_bytes = num_items * 2;
	else
		num_bytes = (num_items + (2 * num_items)) * 3;

	/*
	 * num_bytes is total number of leaves we could need times the leaf
	 * size, and then for every leaf we could end up cow'ing 2 nodes per
	 * level, down to the leaf level.
	 */
	num_bytes = (num_bytes * root->leafsize) +
		(num_bytes * (level * 2)) * root->nodesize;

	return num_bytes;
}

/*
 * Unreserve metadata space for delalloc.  If we have less reserved credits than
 * we have extents, this function does nothing.
 */
int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
					  struct inode *inode, int num_items)
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_space_info *meta_sinfo;
	u64 num_bytes;
	u64 alloc_target;
	bool bug = false;

	/* get the space info for where the metadata will live */
	alloc_target = btrfs_get_alloc_profile(root, 0);
	meta_sinfo = __find_space_info(info, alloc_target);

	num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
					   num_items);

	spin_lock(&meta_sinfo->lock);
	spin_lock(&BTRFS_I(inode)->accounting_lock);
	if (BTRFS_I(inode)->reserved_extents <=
	    BTRFS_I(inode)->outstanding_extents) {
		spin_unlock(&BTRFS_I(inode)->accounting_lock);
		spin_unlock(&meta_sinfo->lock);
		return 0;
	}
	spin_unlock(&BTRFS_I(inode)->accounting_lock);

	BTRFS_I(inode)->reserved_extents--;
	BUG_ON(BTRFS_I(inode)->reserved_extents < 0);

	if (meta_sinfo->bytes_delalloc < num_bytes) {
		bug = true;
		meta_sinfo->bytes_delalloc = 0;
	} else {
		meta_sinfo->bytes_delalloc -= num_bytes;
	}
	spin_unlock(&meta_sinfo->lock);

	BUG_ON(bug);

	return 0;
}

static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
{
	u64 thresh;

	thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
		meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
		meta_sinfo->bytes_super + meta_sinfo->bytes_root +
		meta_sinfo->bytes_may_use;

	thresh = meta_sinfo->total_bytes - thresh;
	thresh *= 80;
	do_div(thresh, 100);
	if (thresh <= meta_sinfo->bytes_delalloc)
		meta_sinfo->force_delalloc = 1;
	else
		meta_sinfo->force_delalloc = 0;
}

struct async_flush {
	struct btrfs_root *root;
	struct btrfs_space_info *info;
	struct btrfs_work work;
};

static noinline void flush_delalloc_async(struct btrfs_work *work)
{
	struct async_flush *async;
	struct btrfs_root *root;
	struct btrfs_space_info *info;

	async = container_of(work, struct async_flush, work);
	root = async->root;
	info = async->info;

	btrfs_start_delalloc_inodes(root);
	wake_up(&info->flush_wait);
	btrfs_wait_ordered_extents(root, 0);

	spin_lock(&info->lock);
	info->flushing = 0;
	spin_unlock(&info->lock);
	wake_up(&info->flush_wait);

	kfree(async);
}

static void wait_on_flush(struct btrfs_space_info *info)
{
	DEFINE_WAIT(wait);
	u64 used;

	while (1) {
		prepare_to_wait(&info->flush_wait, &wait,
				TASK_UNINTERRUPTIBLE);
		spin_lock(&info->lock);
		if (!info->flushing) {
			spin_unlock(&info->lock);
			break;
		}

		used = info->bytes_used + info->bytes_reserved +
			info->bytes_pinned + info->bytes_readonly +
			info->bytes_super + info->bytes_root +
			info->bytes_may_use + info->bytes_delalloc;
		if (used < info->total_bytes) {
			spin_unlock(&info->lock);
			break;
		}
		spin_unlock(&info->lock);
		schedule();
	}
	finish_wait(&info->flush_wait, &wait);
}

static void flush_delalloc(struct btrfs_root *root,
				 struct btrfs_space_info *info)
{
	struct async_flush *async;
	bool wait = false;

	spin_lock(&info->lock);

	if (!info->flushing) {
		info->flushing = 1;
		init_waitqueue_head(&info->flush_wait);
	} else {
		wait = true;
	}

	spin_unlock(&info->lock);

	if (wait) {
		wait_on_flush(info);
		return;
	}

	async = kzalloc(sizeof(*async), GFP_NOFS);
	if (!async)
		goto flush;

	async->root = root;
	async->info = info;
	async->work.func = flush_delalloc_async;

	btrfs_queue_worker(&root->fs_info->enospc_workers,
			   &async->work);
	wait_on_flush(info);
	return;

flush:
	btrfs_start_delalloc_inodes(root);
	btrfs_wait_ordered_extents(root, 0);

	spin_lock(&info->lock);
	info->flushing = 0;
	spin_unlock(&info->lock);
	wake_up(&info->flush_wait);
}

static int maybe_allocate_chunk(struct btrfs_root *root,
				 struct btrfs_space_info *info)
{
	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
	struct btrfs_trans_handle *trans;
	bool wait = false;
	int ret = 0;
	u64 min_metadata;
	u64 free_space;

	free_space = btrfs_super_total_bytes(disk_super);
	/*
	 * we allow the metadata to grow to a max of either 10gb or 5% of the
	 * space in the volume.
	 */
	min_metadata = min((u64)10 * 1024 * 1024 * 1024,
			     div64_u64(free_space * 5, 100));
	if (info->total_bytes >= min_metadata) {
		spin_unlock(&info->lock);
		return 0;
	}

	if (info->full) {
		spin_unlock(&info->lock);
		return 0;
	}

	if (!info->allocating_chunk) {
		info->force_alloc = 1;
		info->allocating_chunk = 1;
		init_waitqueue_head(&info->allocate_wait);
	} else {
		wait = true;
	}

	spin_unlock(&info->lock);

	if (wait) {
		wait_event(info->allocate_wait,
			   !info->allocating_chunk);
		return 1;
	}

	trans = btrfs_start_transaction(root, 1);
	if (!trans) {
		ret = -ENOMEM;
		goto out;
	}

	ret = do_chunk_alloc(trans, root->fs_info->extent_root,
			     4096 + 2 * 1024 * 1024,
			     info->flags, 0);
	btrfs_end_transaction(trans, root);
	if (ret)
		goto out;
out:
	spin_lock(&info->lock);
	info->allocating_chunk = 0;
	spin_unlock(&info->lock);
	wake_up(&info->allocate_wait);

	if (ret)
		return 0;
	return 1;
}

/*
 * Reserve metadata space for delalloc.
 */
int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
					struct inode *inode, int num_items)
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_space_info *meta_sinfo;
	u64 num_bytes;
	u64 used;
	u64 alloc_target;
	int flushed = 0;
	int force_delalloc;

	/* get the space info for where the metadata will live */
	alloc_target = btrfs_get_alloc_profile(root, 0);
	meta_sinfo = __find_space_info(info, alloc_target);

	num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
					   num_items);
again:
	spin_lock(&meta_sinfo->lock);

	force_delalloc = meta_sinfo->force_delalloc;

	if (unlikely(!meta_sinfo->bytes_root))