extent-tree.c

#include <linux/module.h>
#include "ctree.h"
#include "disk-io.h"
#include "print-tree.h"
#include "transaction.h"

static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
			    *orig_root, u64 num_blocks, u64 search_start, u64
			    search_end, struct btrfs_key *ins, int data);
static int finish_current_insert(struct btrfs_trans_handle *trans, struct
				 btrfs_root *extent_root);
static int del_pending_extents(struct btrfs_trans_handle *trans, struct
			       btrfs_root *extent_root);

static int cache_block_group(struct btrfs_root *root,
			     struct btrfs_block_group_cache *block_group)
{
	struct btrfs_path *path;
	int ret;
	struct btrfs_key key;
	struct btrfs_leaf *leaf;
	struct radix_tree_root *extent_radix;
	int slot;
	u64 i;
	u64 last = 0;
	u64 hole_size;
	int found = 0;

	root = root->fs_info->extent_root;
	extent_radix = &root->fs_info->extent_map_radix;

	if (block_group->cached)
		return 0;
	if (block_group->data)
		return 0;
	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;
printk("cache block group %Lu\n", block_group->key.objectid);
	key.objectid = block_group->key.objectid;
	key.flags = 0;
	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)
		return ret;
	if (ret && path->slots[0] > 0)
		path->slots[0]--;
	while(1) {
		leaf = btrfs_buffer_leaf(path->nodes[0]);
		slot = path->slots[0];
		if (slot >= btrfs_header_nritems(&leaf->header)) {
			ret = btrfs_next_leaf(root, path);
			if (ret == 0)
				continue;
			else {
				if (found) {
					hole_size = block_group->key.objectid +
						block_group->key.offset - last;
				} else {
					last = block_group->key.objectid;
					hole_size = block_group->key.offset;
				}
				for (i = 0; i < hole_size; i++) {
					set_radix_bit(extent_radix,
						      last + i);
				}
				break;
			}
		}
		btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
		if (key.objectid >= block_group->key.objectid +
		    block_group->key.offset) {
			if (found) {
				hole_size = block_group->key.objectid +
					block_group->key.offset - last;
			} else {
				last = block_group->key.objectid;
				hole_size = block_group->key.offset;
			}
			for (i = 0; i < hole_size; i++) {
				set_radix_bit(extent_radix, last + i);
			}
			break;
		}
		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
			if (!found) {
				last = key.objectid + key.offset;
				found = 1;
			} else {
				hole_size = key.objectid - last;
				for (i = 0; i < hole_size; i++) {
					set_radix_bit(extent_radix, last + i);
				}
				last = key.objectid + key.offset;
			}
		}
		path->slots[0]++;
	}

	block_group->cached = 1;
	btrfs_free_path(path);
	return 0;
}

static struct btrfs_block_group_cache *lookup_block_group(struct
							  btrfs_fs_info *info,
							  u64 blocknr)
{
	struct btrfs_block_group_cache *block_group;
	int ret;

	ret = radix_tree_gang_lookup(&info->block_group_radix,
				     (void **)&block_group,
				     blocknr, 1);
	if (ret) {
		if (block_group->key.objectid <= blocknr && blocknr <=
		    block_group->key.objectid + block_group->key.offset)
			return block_group;
	}
	ret = radix_tree_gang_lookup(&info->block_group_data_radix,
				     (void **)&block_group,
				     blocknr, 1);
	if (ret) {
		if (block_group->key.objectid <= blocknr && blocknr <=
		    block_group->key.objectid + block_group->key.offset)
			return block_group;
	}
	WARN_ON(1);
	printk("lookup_block_group fails for blocknr %Lu\n", blocknr);
	printk("last ret was %d\n", ret);
	if (ret) {
		printk("last block group was %Lu %Lu\n", block_group->key.objectid, block_group->key.offset);
	}
	return NULL;
}

static u64 leaf_range(struct btrfs_root *root)
{
	u64 size = BTRFS_LEAF_DATA_SIZE(root);
	size = size / (sizeof(struct btrfs_extent_item) +
		       sizeof(struct btrfs_item));
	return size;
}

static u64 find_search_start(struct btrfs_root *root,
			     struct btrfs_block_group_cache **cache_ret,
			     u64 search_start, int num)
{
	unsigned long gang[8];
	int ret;
	struct btrfs_block_group_cache *cache = *cache_ret;
	u64 last = max(search_start, cache->key.objectid);

	if (cache->data)
		goto out;
	if (num > 1) {
		last = max(last, cache->last_prealloc);
	}
again:
	cache_block_group(root, cache);
	while(1) {
		ret = find_first_radix_bit(&root->fs_info->extent_map_radix,
					   gang, last, ARRAY_SIZE(gang));
		if (!ret)
			goto out;
		last = gang[ret-1] + 1;
		if (num > 1) {
			if (ret != ARRAY_SIZE(gang)) {
				goto new_group;
			}
			if (gang[ret-1] - gang[0] > leaf_range(root)) {
				continue;
			}
		}
		if (gang[0] >= cache->key.objectid + cache->key.offset) {
			goto new_group;
		}
		return gang[0];
	}
out:
	return max(cache->last_alloc, search_start);

new_group:
	cache = lookup_block_group(root->fs_info, last + cache->key.offset - 1);
	if (!cache) {
		return max((*cache_ret)->last_alloc, search_start);
	}
	cache = btrfs_find_block_group(root, cache,
				       last + cache->key.offset - 1, 0);
	*cache_ret = cache;
	goto again;
}

struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
						 struct btrfs_block_group_cache
						 *hint, u64 search_start,
						 int data)
{
	struct btrfs_block_group_cache *cache[8];
	struct btrfs_block_group_cache *found_group = NULL;
	struct btrfs_fs_info *info = root->fs_info;
	struct radix_tree_root *radix;
	u64 used;
	u64 last = 0;
	u64 hint_last;
	int i;
	int ret;
	int full_search = 0;

	if (data)
		radix = &info->block_group_data_radix;
	else
		radix = &info->block_group_radix;

	if (search_start) {
		struct btrfs_block_group_cache *shint;
		shint = lookup_block_group(info, search_start);
		if (shint->data == data) {
			used = btrfs_block_group_used(&shint->item);
			if (used + shint->pinned <
			    (shint->key.offset * 8) / 10) {
				return shint;
			}
		}
	}
	if (hint && hint->data == data) {
		used = btrfs_block_group_used(&hint->item);
		if (used + hint->pinned < (hint->key.offset * 8) / 10) {
			return hint;
		}
		if (used >= (hint->key.offset * 8) / 10) {
			radix_tree_tag_clear(radix,
					     hint->key.objectid +
					     hint->key.offset - 1,
					     BTRFS_BLOCK_GROUP_AVAIL);
		}
		last = hint->key.offset * 3;
		if (hint->key.objectid >= last)
			last = max(search_start + hint->key.offset - 1,
				   hint->key.objectid - last);
		else
			last = hint->key.objectid + hint->key.offset;
		hint_last = last;
	} else {
		if (hint)
			hint_last = max(hint->key.objectid, search_start);
		else
			hint_last = search_start;

		last = hint_last;
	}
	while(1) {
		ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
						 last, ARRAY_SIZE(cache),
						 BTRFS_BLOCK_GROUP_AVAIL);
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			last = cache[i]->key.objectid +
				cache[i]->key.offset;
			used = btrfs_block_group_used(&cache[i]->item);
			if (used + cache[i]->pinned <
			    (cache[i]->key.offset * 8) / 10) {
				found_group = cache[i];
				goto found;
			}
			if (used >= (cache[i]->key.offset * 8) / 10) {
				radix_tree_tag_clear(radix,
						     cache[i]->key.objectid +
						     cache[i]->key.offset - 1,
						     BTRFS_BLOCK_GROUP_AVAIL);
			}
		}
	}
	last = hint_last;
again:
	while(1) {
		ret = radix_tree_gang_lookup(radix, (void **)cache,
					     last, ARRAY_SIZE(cache));
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			last = cache[i]->key.objectid +
				cache[i]->key.offset;
			used = btrfs_block_group_used(&cache[i]->item);
			if (used + cache[i]->pinned < cache[i]->key.offset) {
				found_group = cache[i];
				goto found;
			}
			if (used >= cache[i]->key.offset) {
				radix_tree_tag_clear(radix,
						     cache[i]->key.objectid +
						     cache[i]->key.offset - 1,
						     BTRFS_BLOCK_GROUP_AVAIL);
			}
		}
	}
	if (!full_search) {
		last = search_start;
		full_search = 1;
		goto again;
	}
	if (!found_group) {
		ret = radix_tree_gang_lookup(radix,
					     (void **)&found_group, 0, 1);
		BUG_ON(ret != 1);
	}
found:
	return found_group;
}

int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
				struct btrfs_root *root,
				u64 blocknr, u64 num_blocks)
{
	struct btrfs_path *path;
	int ret;
	struct btrfs_key key;
	struct btrfs_leaf *l;
	struct btrfs_extent_item *item;
	struct btrfs_key ins;
	u32 refs;

	find_free_extent(trans, root->fs_info->extent_root, 0, 0, (u64)-1,
			 &ins, 0);
	path = btrfs_alloc_path();
	BUG_ON(!path);
	btrfs_init_path(path);
	key.objectid = blocknr;
	key.flags = 0;
	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
	key.offset = num_blocks;
	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
				0, 1);
	if (ret != 0) {
printk("can't find block %Lu %Lu\n", blocknr, num_blocks);
		BUG();
	}
	BUG_ON(ret != 0);
	l = btrfs_buffer_leaf(path->nodes[0]);
	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
	refs = btrfs_extent_refs(item);
	btrfs_set_extent_refs(item, refs + 1);
	btrfs_mark_buffer_dirty(path->nodes[0]);

	btrfs_release_path(root->fs_info->extent_root, path);
	btrfs_free_path(path);
	finish_current_insert(trans, root->fs_info->extent_root);
	del_pending_extents(trans, root->fs_info->extent_root);
	return 0;
}

static int lookup_extent_ref(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root, u64 blocknr,
			     u64 num_blocks, u32 *refs)
{
	struct btrfs_path *path;
	int ret;
	struct btrfs_key key;
	struct btrfs_leaf *l;
	struct btrfs_extent_item *item;

	path = btrfs_alloc_path();
	btrfs_init_path(path);
	key.objectid = blocknr;
	key.offset = num_blocks;
	key.flags = 0;
	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)
		BUG();
	l = btrfs_buffer_leaf(path->nodes[0]);
	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
	*refs = btrfs_extent_refs(item);
	btrfs_release_path(root->fs_info->extent_root, path);
	btrfs_free_path(path);
	return 0;
}

int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
		       struct btrfs_root *root)
{
	return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1);
}

int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
		  struct buffer_head *buf)
{
	u64 blocknr;
	struct btrfs_node *buf_node;
	struct btrfs_leaf *buf_leaf;
	struct btrfs_disk_key *key;
	struct btrfs_file_extent_item *fi;
	int i;
	int leaf;
	int ret;

	if (!root->ref_cows)
		return 0;
	buf_node = btrfs_buffer_node(buf);
	leaf = btrfs_is_leaf(buf_node);
	buf_leaf = btrfs_buffer_leaf(buf);
	for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) {
		if (leaf) {
			key = &buf_leaf->items[i].key;
			if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
				continue;
			fi = btrfs_item_ptr(buf_leaf, i,
					    struct btrfs_file_extent_item);
			if (btrfs_file_extent_type(fi) ==
			    BTRFS_FILE_EXTENT_INLINE)
				continue;
			ret = btrfs_inc_extent_ref(trans, root,
				    btrfs_file_extent_disk_blocknr(fi),
				    btrfs_file_extent_disk_num_blocks(fi));
			BUG_ON(ret);
		} else {
			blocknr = btrfs_node_blockptr(buf_node, i);
			ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
			BUG_ON(ret);
		}
	}
	return 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;
	int pending_ret;
	struct btrfs_root *extent_root = root->fs_info->extent_root;
	struct btrfs_block_group_item *bi;
	struct btrfs_key ins;

	find_free_extent(trans, extent_root, 0, 0, (u64)-1, &ins, 0);
	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
	BUG_ON(ret);
	bi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
			    struct btrfs_block_group_item);
	memcpy(bi, &cache->item, sizeof(*bi));
	mark_buffer_dirty(path->nodes[0]);
	btrfs_release_path(extent_root, path);

	finish_current_insert(trans, extent_root);
	pending_ret = del_pending_extents(trans, extent_root);
	if (ret)
		return ret;
	if (pending_ret)
		return pending_ret;
	if (cache->data)
		cache->last_alloc = cache->first_free;
	return 0;

}

static int write_dirty_block_radix(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root,
				   struct radix_tree_root *radix)
{
	struct btrfs_block_group_cache *cache[8];
	int ret;
	int err = 0;
	int werr = 0;
	int i;
	struct btrfs_path *path;

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

	while(1) {
		ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
						 0, ARRAY_SIZE(cache),
						 BTRFS_BLOCK_GROUP_DIRTY);
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			radix_tree_tag_clear(radix, cache[i]->key.objectid +
					     cache[i]->key.offset - 1,
					     BTRFS_BLOCK_GROUP_DIRTY);
			err = write_one_cache_group(trans, root,
						    path, cache[i]);
			if (err)
				werr = err;
		}
	}
	btrfs_free_path(path);
	return werr;
}

int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
	int ret;
	int ret2;
	ret = write_dirty_block_radix(trans, root,
				      &root->fs_info->block_group_radix);
	ret2 = write_dirty_block_radix(trans, root,
				      &root->fs_info->block_group_data_radix);
	if (ret)
		return ret;
	if (ret2)
		return ret2;
	return 0;
}

static int update_block_group(struct btrfs_trans_handle *trans,
			      struct btrfs_root *root,
			      u64 blocknr, u64 num, int alloc, int mark_free)
{
	struct btrfs_block_group_cache *cache;
	struct btrfs_fs_info *info = root->fs_info;
	u64 total = num;
	u64 old_val;
	u64 block_in_group;
	u64 i;

	while(total) {
		cache = lookup_block_group(info, blocknr);
		if (!cache) {
			printk(KERN_CRIT "blocknr %Lu lookup failed\n",
			       blocknr);
			return -1;
		}
		block_in_group = blocknr - cache->key.objectid;
		WARN_ON(block_in_group > cache->key.offset);
		radix_tree_tag_set(cache->radix, cache->key.objectid +
				   cache->key.offset - 1,
				   BTRFS_BLOCK_GROUP_DIRTY);

		old_val = btrfs_block_group_used(&cache->item);
		num = min(total, cache->key.offset - block_in_group);
		if (alloc) {
			old_val += num;
			if (blocknr > cache->last_alloc)
				cache->last_alloc = blocknr;
			if (!cache->data) {
				for (i = 0; i < num; i++) {
					clear_radix_bit(&info->extent_map_radix,
						        blocknr + i);
				}
			}
		} else {
			old_val -= num;
			if (blocknr < cache->first_free)
				cache->first_free = blocknr;
			if (!cache->data && mark_free) {
				for (i = 0; i < num; i++) {
					set_radix_bit(&info->extent_map_radix,
						      blocknr + i);
				}
			}
			if (old_val < (cache->key.offset * 6) / 10 &&
			    old_val + num >= (cache->key.offset * 6) / 10) {
printk("group %Lu now available\n", cache->key.objectid);
				radix_tree_tag_set(cache->radix,
						   cache->key.objectid +
						   cache->key.offset - 1,
						   BTRFS_BLOCK_GROUP_AVAIL);
			}
		}
		btrfs_set_block_group_used(&cache->item, old_val);
		total -= num;
		blocknr += num;
	}
	return 0;
}

static int try_remove_page(struct address_space *mapping, unsigned long index)
{
	int ret;
	ret = invalidate_mapping_pages(mapping, index, index);
	return ret;
}

int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
			       btrfs_root *root)
{
	unsigned long gang[8];
	struct inode *btree_inode = root->fs_info->btree_inode;
	struct btrfs_block_group_cache *block_group;
	u64 first = 0;
	int ret;
	int i;
	struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
	struct radix_tree_root *extent_radix = &root->fs_info->extent_map_radix;

	while(1) {
		ret = find_first_radix_bit(pinned_radix, gang, 0,
					   ARRAY_SIZE(gang));
		if (!ret)
			break;
		if (!first)
			first = gang[0];
		for (i = 0; i < ret; i++) {
			clear_radix_bit(pinned_radix, gang[i]);
			block_group = lookup_block_group(root->fs_info,
							 gang[i]);
			if (block_group) {
				WARN_ON(block_group->pinned == 0);
				block_group->pinned--;
				if (gang[i] < block_group->last_alloc)
					block_group->last_alloc = gang[i];
				if (gang[i] < block_group->last_prealloc)
					block_group->last_prealloc = gang[i];
				if (!block_group->data)
					set_radix_bit(extent_radix, gang[i]);
			}
			try_remove_page(btree_inode->i_mapping,
					gang[i] << (PAGE_CACHE_SHIFT -
						    btree_inode->i_blkbits));
		}
	}
	return 0;
}

static int finish_current_insert(struct btrfs_trans_handle *trans, struct
				 btrfs_root *extent_root)
{
	struct btrfs_key ins;
	struct btrfs_extent_item extent_item;
	int i;
	int ret;
	u64 super_blocks_used;
	struct btrfs_fs_info *info = extent_root->fs_info;

	btrfs_set_extent_refs(&extent_item, 1);
	ins.offset = 1;
	ins.flags = 0;
	btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
	btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid);

	for (i = 0; i < extent_root->fs_info->extent_tree_insert_nr; i++) {
		ins.objectid = extent_root->fs_info->extent_tree_insert[i];
		super_blocks_used = btrfs_super_blocks_used(info->disk_super);
		btrfs_set_super_blocks_used(info->disk_super,
					    super_blocks_used + 1);
		ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
					sizeof(extent_item));
		BUG_ON(ret);
	}
	extent_root->fs_info->extent_tree_insert_nr = 0;
	extent_root->fs_info->extent_tree_prealloc_nr = 0;
	return 0;
}

static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
{
	int err;
	struct btrfs_header *header;
	struct buffer_head *bh;

	if (!pending) {
		bh = btrfs_find_tree_block(root, blocknr);
		if (bh) {
			if (buffer_uptodate(bh)) {
				u64 transid =
				    root->fs_info->running_transaction->transid;
				header = btrfs_buffer_header(bh);
				if (btrfs_header_generation(header) ==
				    transid) {
					btrfs_block_release(root, bh);
					return 0;
				}
			}
			btrfs_block_release(root, bh);
		}
		err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
		if (!err) {
			struct btrfs_block_group_cache *cache;
			cache = lookup_block_group(root->fs_info, blocknr);
			if (cache)
				cache->pinned++;
		}
	} else {
		err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
	}
	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 blocknr, u64 num_blocks, 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;
	int ret;
	struct btrfs_extent_item *ei;
	struct btrfs_key ins;
	u32 refs;

	key.objectid = blocknr;
	key.flags = 0;
	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
	key.offset = num_blocks;

	find_free_extent(trans, root, 0, 0, (u64)-1, &ins, 0);
	path = btrfs_alloc_path();
	BUG_ON(!path);
	btrfs_init_path(path);

	ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
	if (ret) {
		printk("failed to find %Lu\n", key.objectid);
		btrfs_print_tree(extent_root, extent_root->node);
		printk("failed to find %Lu\n", key.objectid);
		BUG();
	}
	ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
			    struct btrfs_extent_item);
	BUG_ON(ei->refs == 0);
	refs = btrfs_extent_refs(ei) - 1;
	btrfs_set_extent_refs(ei, refs);
	btrfs_mark_buffer_dirty(path->nodes[0]);
	if (refs == 0) {
		u64 super_blocks_used;

		if (pin) {
			ret = pin_down_block(root, blocknr, 0);
			BUG_ON(ret);
		}

		super_blocks_used = btrfs_super_blocks_used(info->disk_super);
		btrfs_set_super_blocks_used(info->disk_super,
					    super_blocks_used - num_blocks);
		ret = btrfs_del_item(trans, extent_root, path);
		if (ret)
			BUG();
		ret = update_block_group(trans, root, blocknr, num_blocks, 0,
					 mark_free);
		BUG_ON(ret);
	}
	btrfs_free_path(path);
	finish_current_insert(trans, extent_root);
	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 ret;
	int wret;
	int err = 0;
	unsigned long gang[4];
	int i;
	struct radix_tree_root *pending_radix;
	struct radix_tree_root *pinned_radix;
	struct btrfs_block_group_cache *cache;

	pending_radix = &extent_root->fs_info->pending_del_radix;
	pinned_radix = &extent_root->fs_info->pinned_radix;

	while(1) {
		ret = find_first_radix_bit(pending_radix, gang, 0,
					   ARRAY_SIZE(gang));
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			wret = set_radix_bit(pinned_radix, gang[i]);
			if (wret == 0) {
				cache = lookup_block_group(extent_root->fs_info,
							   gang[i]);
				if (cache)
					cache->pinned++;
			}
			if (wret < 0) {
				printk(KERN_CRIT "set_radix_bit, err %d\n",
				       wret);
				BUG_ON(wret < 0);
			}
			wret = clear_radix_bit(pending_radix, gang[i]);
			BUG_ON(wret);
			wret = __free_extent(trans, extent_root,
					     gang[i], 1, 0, 0);
			if (wret)
				err = wret;
		}
	}
	return err;
}

/*
 * remove an extent from the root, returns 0 on success
 */
int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, u64 blocknr, u64 num_blocks, int pin)
{
	struct btrfs_root *extent_root = root->fs_info->extent_root;
	int pending_ret;
	int ret;

	if (root == extent_root) {
		pin_down_block(root, blocknr, 1);
		return 0;
	}
	ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
	pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
	return ret ? ret : pending_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 int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
			    *orig_root, u64 num_blocks, u64 search_start, u64
			    search_end, struct btrfs_key *ins, int data)
{
	struct btrfs_path *path;
	struct btrfs_key key;
	int ret;
	u64 hole_size = 0;
	int slot = 0;
	u64 last_block = 0;
	u64 test_block;
	u64 orig_search_start = search_start;
	int start_found;
	struct btrfs_leaf *l;
	struct btrfs_root * root = orig_root->fs_info->extent_root;
	struct btrfs_fs_info *info = root->fs_info;
	int total_needed = num_blocks;
	int total_found = 0;
	int fill_prealloc = 0;
	int level;
	struct btrfs_block_group_cache *block_group;
	int full_scan = 0;

	path = btrfs_alloc_path();
	ins->flags = 0;
	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);

	level = btrfs_header_level(btrfs_buffer_header(root->node));
	if (num_blocks == 0) {
		fill_prealloc = 1;
		num_blocks = 1;
		total_needed = (min(level + 1, BTRFS_MAX_LEVEL) + 2) * 3;
	}
	if (search_end == (u64)-1)
		search_end = btrfs_super_total_blocks(info->disk_super);
	if (search_start) {
		block_group = lookup_block_group(info, search_start);
		block_group = btrfs_find_block_group(root, block_group,
						     search_start, data);
	} else {
		block_group = btrfs_find_block_group(root,
						     trans->block_group, 0,
						     data);
	}

check_failed:
	if (!full_scan && block_group->data != data)
		WARN_ON(1);

	if (!data)
		search_start = find_search_start(root, &block_group,
						 search_start, total_needed);
	else
		search_start = max(block_group->last_alloc, search_start);

	btrfs_init_path(path);
	ins->objectid = search_start;
	ins->offset = 0;
	start_found = 0;

	ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
	if (ret < 0)
		goto error;

	if (path->slots[0] > 0) {
		path->slots[0]--;
	}

	l = btrfs_buffer_leaf(path->nodes[0]);
	btrfs_disk_key_to_cpu(&key, &l->items[path->slots[0]].key);
	/*
	 * a rare case, go back one key if we hit a block group item
	 * instead of an extent item
	 */
	if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
	    key.objectid + key.offset >= search_start) {
		ins->objectid = key.objectid;
		ins->offset = key.offset - 1;
		btrfs_release_path(root, path);
		ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
		if (ret < 0)
			goto error;

		if (path->slots[0] > 0) {
			path->slots[0]--;
		}
	}

	while (1) {
		l = btrfs_buffer_leaf(path->nodes[0]);
		slot = path->slots[0];
		if (slot >= btrfs_header_nritems(&l->header)) {
			if (fill_prealloc) {
				info->extent_tree_prealloc_nr = 0;
				total_found = 0;
			}
			ret = btrfs_next_leaf(root, path);
			if (ret == 0)
				continue;
			if (ret < 0)
				goto error;
			if (!start_found) {
				ins->objectid = search_start;
				ins->offset = search_end - search_start;
				start_found = 1;
				goto check_pending;
			}
			ins->objectid = last_block > search_start ?
					last_block : search_start;
			ins->offset = search_end - ins->objectid;
			goto check_pending;
		}

		btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
		if (key.objectid >= search_start && key.objectid > last_block &&
		    start_found) {
			if (last_block < search_start)
				last_block = search_start;
			hole_size = key.objectid - last_block;
			if (hole_size >= num_blocks) {
				ins->objectid = last_block;
				ins->offset = hole_size;
				goto check_pending;
			}
		}

		if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
			goto next;

		start_found = 1;
		last_block = key.objectid + key.offset;
		if (last_block >= block_group->key.objectid +
		    block_group->key.offset) {
			btrfs_release_path(root, path);
			search_start = block_group->key.objectid +
				block_group->key.offset * 2;
			goto new_group;
		}
next:
		path->slots[0]++;
	}
	// FIXME -ENOSPC
check_pending:
	/* we have to make sure we didn't find an extent that has already
	 * been allocated by the map tree or the original allocation
	 */
	btrfs_release_path(root, path);
	BUG_ON(ins->objectid < search_start);

	if (ins->objectid + num_blocks >= search_end) {
		if (full_scan)
			return -ENOSPC;
		search_start = orig_search_start;
		full_scan = 1;
		goto new_group;
	}
	for (test_block = ins->objectid;
	     test_block < ins->objectid + num_blocks; test_block++) {
		if (test_radix_bit(&info->pinned_radix, test_block)) {
			search_start = test_block + 1;
			goto new_group;
		}
	}
	if (!fill_prealloc && info->extent_tree_insert_nr) {
		u64 last =
		  info->extent_tree_insert[info->extent_tree_insert_nr - 1];
		if (ins->objectid + num_blocks >
		    info->extent_tree_insert[0] &&
		    ins->objectid <= last) {
			search_start = last + 1;
			WARN_ON(!full_scan);
			goto new_group;
		}
	}
	if (!fill_prealloc && info->extent_tree_prealloc_nr) {
		u64 first =
		  info->extent_tree_prealloc[info->extent_tree_prealloc_nr - 1];
		if (ins->objectid + num_blocks > first &&