[linux-2.6-block.git] / fs / btrfs / extent-tree.c

#include <stdio.h>
#include <stdlib.h>
#include "kerncompat.h"
#include "radix-tree.h"
#include "ctree.h"
#include "disk-io.h"
#include "print-tree.h"

static int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
			    u64 search_start, u64 search_end,
			    struct btrfs_key *ins);
static int finish_current_insert(struct ctree_root *extent_root);
static int run_pending(struct ctree_root *extent_root);

/*
 * pending extents are blocks that we're trying to allocate in the extent
 * map while trying to grow the map because of other allocations.  To avoid
 * recursing, they are tagged in the radix tree and cleaned up after
 * other allocations are done.  The pending tag is also used in the same
 * manner for deletes.
 */
#define CTREE_EXTENT_PENDING_DEL 0

static int inc_block_ref(struct ctree_root *root, u64 blocknr)
{
	struct ctree_path path;
	int ret;
	struct btrfs_key key;
	struct leaf *l;
	struct extent_item *item;
	struct btrfs_key ins;

	find_free_extent(root->extent_root, 0, 0, (u64)-1, &ins);
	init_path(&path);
	key.objectid = blocknr;
	key.flags = 0;
	key.offset = 1;
	ret = search_slot(root->extent_root, &key, &path, 0, 1);
	if (ret != 0)
		BUG();
	BUG_ON(ret != 0);
	l = &path.nodes[0]->leaf;
	item = (struct extent_item *)(l->data + btrfs_item_offset(l->items +
								path.slots[0]));
	item->refs++;

	BUG_ON(list_empty(&path.nodes[0]->dirty));
	release_path(root->extent_root, &path);
	finish_current_insert(root->extent_root);
	run_pending(root->extent_root);
	return 0;
}

static int lookup_block_ref(struct ctree_root *root, u64 blocknr, u32 *refs)
{
	struct ctree_path path;
	int ret;
	struct btrfs_key key;
	struct leaf *l;
	struct extent_item *item;
	init_path(&path);
	key.objectid = blocknr;
	key.flags = 0;
	key.offset = 1;
	ret = search_slot(root->extent_root, &key, &path, 0, 0);
	if (ret != 0)
		BUG();
	l = &path.nodes[0]->leaf;
	item = (struct extent_item *)(l->data +
				      btrfs_item_offset(l->items +
							path.slots[0]));
	*refs = item->refs;
	release_path(root->extent_root, &path);
	return 0;
}

int btrfs_inc_ref(struct ctree_root *root, struct tree_buffer *buf)
{
	u64 blocknr;
	int i;

	if (root == root->extent_root)
		return 0;
	if (btrfs_is_leaf(&buf->node))
		return 0;

	for (i = 0; i < btrfs_header_nritems(&buf->node.header); i++) {
		blocknr = buf->node.blockptrs[i];
		inc_block_ref(root, blocknr);
	}
	return 0;
}

int btrfs_finish_extent_commit(struct ctree_root *root)
{
	struct ctree_root *extent_root = root->extent_root;
	unsigned long gang[8];
	int ret;
	int i;

	while(1) {
		ret = radix_tree_gang_lookup(&extent_root->pinned_radix,
						 (void **)gang, 0,
						 ARRAY_SIZE(gang));
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			radix_tree_delete(&extent_root->pinned_radix, gang[i]);
		}
	}
	extent_root->last_insert.objectid = 0;
	extent_root->last_insert.offset = 0;
	return 0;
}

static int finish_current_insert(struct ctree_root *extent_root)
{
	struct btrfs_key ins;
	struct extent_item extent_item;
	int i;
	int ret;

	extent_item.refs = 1;
	extent_item.owner =
		btrfs_header_parentid(&extent_root->node->node.header);
	ins.offset = 1;
	ins.flags = 0;

	for (i = 0; i < extent_root->current_insert.flags; i++) {
		ins.objectid = extent_root->current_insert.objectid + i;
		ret = insert_item(extent_root, &ins, &extent_item,
				  sizeof(extent_item));
		BUG_ON(ret);
	}
	extent_root->current_insert.offset = 0;
	return 0;
}

/*
 * remove an extent from the root, returns 0 on success
 */
int __free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
{
	struct ctree_path path;
	struct btrfs_key key;
	struct ctree_root *extent_root = root->extent_root;
	int ret;
	struct btrfs_item *item;
	struct extent_item *ei;
	struct btrfs_key ins;

	key.objectid = blocknr;
	key.flags = 0;
	key.offset = num_blocks;

	find_free_extent(root, 0, 0, (u64)-1, &ins);
	init_path(&path);
	ret = search_slot(extent_root, &key, &path, -1, 1);
	if (ret) {
		printf("failed to find %Lu\n", key.objectid);
		print_tree(extent_root, extent_root->node);
		printf("failed to find %Lu\n", key.objectid);
		BUG();
	}
	item = path.nodes[0]->leaf.items + path.slots[0];
	ei = (struct extent_item *)(path.nodes[0]->leaf.data +
				    btrfs_item_offset(item));
	BUG_ON(ei->refs == 0);
	ei->refs--;
	if (ei->refs == 0) {
		if (root == extent_root) {
			int err;
			radix_tree_preload(GFP_KERNEL);
			err = radix_tree_insert(&extent_root->pinned_radix,
					  blocknr, (void *)blocknr);
			BUG_ON(err);
			radix_tree_preload_end();
		}
		ret = del_item(extent_root, &path);
		if (root != extent_root &&
		    extent_root->last_insert.objectid < blocknr)
			extent_root->last_insert.objectid = blocknr;
		if (ret)
			BUG();
	}
	release_path(extent_root, &path);
	finish_current_insert(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 ctree_root *extent_root)
{
	int ret;
	struct tree_buffer *gang[4];
	int i;

	while(1) {
		ret = radix_tree_gang_lookup_tag(&extent_root->cache_radix,
						 (void **)gang, 0,
						 ARRAY_SIZE(gang),
						 CTREE_EXTENT_PENDING_DEL);
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			ret = __free_extent(extent_root, gang[i]->blocknr, 1);
			radix_tree_tag_clear(&extent_root->cache_radix,
						gang[i]->blocknr,
						CTREE_EXTENT_PENDING_DEL);
			tree_block_release(extent_root, gang[i]);
		}
	}
	return 0;
}

static int run_pending(struct ctree_root *extent_root)
{
	while(radix_tree_tagged(&extent_root->cache_radix,
			        CTREE_EXTENT_PENDING_DEL))
		del_pending_extents(extent_root);
	return 0;
}


/*
 * remove an extent from the root, returns 0 on success
 */
int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
{
	struct btrfs_key key;
	struct ctree_root *extent_root = root->extent_root;
	struct tree_buffer *t;
	int pending_ret;
	int ret;

	if (root == extent_root) {
		t = find_tree_block(root, blocknr);
		radix_tree_tag_set(&root->cache_radix, blocknr,
				   CTREE_EXTENT_PENDING_DEL);
		return 0;
	}
	key.objectid = blocknr;
	key.flags = 0;
	key.offset = num_blocks;
	ret = __free_extent(root, blocknr, num_blocks);
	pending_ret = run_pending(root->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 = 0
 * ins->offset == number of blocks
 * Any available blocks before search_start are skipped.
 */
static int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
			    u64 search_start, u64 search_end,
			    struct btrfs_key *ins)
{
	struct ctree_path path;
	struct btrfs_key key;
	int ret;
	u64 hole_size = 0;
	int slot = 0;
	u64 last_block;
	u64 test_block;
	int start_found;
	struct leaf *l;
	struct ctree_root * root = orig_root->extent_root;
	int total_needed = num_blocks;

	total_needed += (btrfs_header_level(&root->node->node.header) + 1) * 3;
	if (root->last_insert.objectid > search_start)
		search_start = root->last_insert.objectid;
check_failed:
	init_path(&path);
	ins->objectid = search_start;
	ins->offset = 0;
	ins->flags = 0;
	start_found = 0;
	ret = search_slot(root, ins, &path, 0, 0);
	if (ret < 0)
		goto error;

	if (path.slots[0] > 0)
		path.slots[0]--;

	while (1) {
		l = &path.nodes[0]->leaf;
		slot = path.slots[0];
		if (slot >= btrfs_header_nritems(&l->header)) {
			ret = next_leaf(root, &path);
			if (ret == 0)
				continue;
			if (ret < 0)
				goto error;
			if (!start_found) {
				ins->objectid = search_start;
				ins->offset = (u64)-1;
				start_found = 1;
				goto check_pending;
			}
			ins->objectid = last_block > search_start ?
					last_block : search_start;
			ins->offset = (u64)-1;
			goto check_pending;
		}
		btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
		if (key.objectid >= search_start) {
			if (start_found) {
				if (last_block < search_start)
					last_block = search_start;
				hole_size = key.objectid - last_block;
				if (hole_size > total_needed) {
					ins->objectid = last_block;
					ins->offset = hole_size;
					goto check_pending;
				}
			}
		}
		start_found = 1;
		last_block = key.objectid + key.offset;
		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
	 */
	release_path(root, &path);
	BUG_ON(ins->objectid < search_start);
	for (test_block = ins->objectid;
	     test_block < ins->objectid + total_needed; test_block++) {
		if (radix_tree_lookup(&root->pinned_radix, test_block)) {
			search_start = test_block + 1;
			goto check_failed;
		}
	}
	BUG_ON(root->current_insert.offset);
	root->current_insert.offset = total_needed - num_blocks;
	root->current_insert.objectid = ins->objectid + num_blocks;
	root->current_insert.flags = 0;
	root->last_insert.objectid = ins->objectid;
	ins->offset = num_blocks;
	return 0;
error:
	release_path(root, &path);
	return ret;
}

/*
 * finds a free extent and does all the dirty work required for allocation
 * returns the key for the extent through ins, and a tree buffer for
 * the first block of the extent through buf.
 *
 * returns 0 if everything worked, non-zero otherwise.
 */
int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
			 u64 search_end, u64 owner, struct btrfs_key *ins)
{
	int ret;
	int pending_ret;
	struct ctree_root *extent_root = root->extent_root;
	struct extent_item extent_item;

	extent_item.refs = 1;
	extent_item.owner = owner;

	if (root == extent_root) {
		BUG_ON(extent_root->current_insert.offset == 0);
		BUG_ON(num_blocks != 1);
		BUG_ON(extent_root->current_insert.flags ==
		       extent_root->current_insert.offset);
		ins->offset = 1;
		ins->objectid = extent_root->current_insert.objectid +
				extent_root->current_insert.flags++;
		return 0;
	}
	ret = find_free_extent(root, num_blocks, search_start,
			       search_end, ins);
	if (ret)
		return ret;

	ret = insert_item(extent_root, ins, &extent_item,
			  sizeof(extent_item));

	finish_current_insert(extent_root);
	pending_ret = run_pending(extent_root);
	if (ret)
		return ret;
	if (pending_ret)
		return pending_ret;
	return 0;
}

/*
 * helper function to allocate a block for a given tree
 * returns the tree buffer or NULL.
 */
struct tree_buffer *alloc_free_block(struct ctree_root *root)
{
	struct btrfs_key ins;
	int ret;
	struct tree_buffer *buf;

	ret = alloc_extent(root, 1, 0, (unsigned long)-1,
			   btrfs_header_parentid(&root->node->node.header),
			   &ins);
	if (ret) {
		BUG();
		return NULL;
	}
	buf = find_tree_block(root, ins.objectid);
	dirty_tree_block(root, buf);
	return buf;
}

int walk_down_tree(struct ctree_root *root, struct ctree_path *path, int *level)
{
	struct tree_buffer *next;
	struct tree_buffer *cur;
	u64 blocknr;
	int ret;
	u32 refs;

	ret = lookup_block_ref(root, path->nodes[*level]->blocknr, &refs);
	BUG_ON(ret);
	if (refs > 1)
		goto out;
	while(*level > 0) {
		cur = path->nodes[*level];
		if (path->slots[*level] >=
		    btrfs_header_nritems(&cur->node.header))
			break;
		blocknr = cur->node.blockptrs[path->slots[*level]];
		ret = lookup_block_ref(root, blocknr, &refs);
		if (refs != 1 || *level == 1) {
			path->slots[*level]++;
			ret = free_extent(root, blocknr, 1);
			BUG_ON(ret);
			continue;
		}
		BUG_ON(ret);
		next = read_tree_block(root, blocknr);
		if (path->nodes[*level-1])
			tree_block_release(root, path->nodes[*level-1]);
		path->nodes[*level-1] = next;
		*level = btrfs_header_level(&next->node.header);
		path->slots[*level] = 0;
	}
out:
	ret = free_extent(root, path->nodes[*level]->blocknr, 1);
	tree_block_release(root, path->nodes[*level]);
	path->nodes[*level] = NULL;
	*level += 1;
	BUG_ON(ret);
	return 0;
}

int walk_up_tree(struct ctree_root *root, struct ctree_path *path, int *level)
{
	int i;
	int slot;
	int ret;
	for(i = *level; i < MAX_LEVEL - 1 && path->nodes[i]; i++) {
		slot = path->slots[i];
		if (slot <
		    btrfs_header_nritems(&path->nodes[i]->node.header)- 1) {
			path->slots[i]++;
			*level = i;
			return 0;
		} else {
			ret = free_extent(root,
					  path->nodes[*level]->blocknr, 1);
			tree_block_release(root, path->nodes[*level]);
			path->nodes[*level] = NULL;
			*level = i + 1;
			BUG_ON(ret);
		}
	}
	return 1;
}

int btrfs_drop_snapshot(struct ctree_root *root, struct tree_buffer *snap)
{
	int ret;
	int level;
	struct ctree_path path;
	int i;
	int orig_level;

	init_path(&path);

	level = btrfs_header_level(&snap->node.header);
	orig_level = level;
	path.nodes[level] = snap;
	path.slots[level] = 0;
	while(1) {
		ret = walk_down_tree(root, &path, &level);
		if (ret > 0)
			break;
		ret = walk_up_tree(root, &path, &level);
		if (ret > 0)
			break;
	}
	for (i = 0; i <= orig_level; i++) {
		if (path.nodes[i]) {
			tree_block_release(root, path.nodes[i]);
		}
	}

	return 0;
}
Commit	Line	Data
fec577fb CM	1	#include <stdio.h>
	2	#include <stdlib.h>
	3	#include "kerncompat.h"
	4	#include "radix-tree.h"
	5	#include "ctree.h"
	6	#include "disk-io.h"
	7	#include "print-tree.h"
	8
037e6390	9	static int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
e2fa7227 CM	10	u64 search_start, u64 search_end,
e2fa7227 CM	11	struct btrfs_key *ins);
037e6390 CM	12	static int finish_current_insert(struct ctree_root *extent_root);
	13	static int run_pending(struct ctree_root *extent_root);
	14
fec577fb CM	15	/*
	16	* pending extents are blocks that we're trying to allocate in the extent
	17	* map while trying to grow the map because of other allocations. To avoid
	18	* recursing, they are tagged in the radix tree and cleaned up after
	19	* other allocations are done. The pending tag is also used in the same
	20	* manner for deletes.
	21	*/
037e6390	22	#define CTREE_EXTENT_PENDING_DEL 0
fec577fb	23
02217ed2 CM	24	static int inc_block_ref(struct ctree_root *root, u64 blocknr)
	25	{
	26	struct ctree_path path;
	27	int ret;
e2fa7227	28	struct btrfs_key key;
02217ed2 CM	29	struct leaf *l;
02217ed2 CM	30	struct extent_item *item;
e2fa7227	31	struct btrfs_key ins;
037e6390 CM	32
037e6390 CM	33	find_free_extent(root->extent_root, 0, 0, (u64)-1, &ins);
02217ed2 CM	34	init_path(&path);
	35	key.objectid = blocknr;
	36	key.flags = 0;
	37	key.offset = 1;
	38	ret = search_slot(root->extent_root, &key, &path, 0, 1);
a28ec197 CM	39	if (ret != 0)
a28ec197 CM	40	BUG();
02217ed2 CM	41	BUG_ON(ret != 0);
02217ed2 CM	42	l = &path.nodes[0]->leaf;
0783fcfc CM	43	item = (struct extent_item *)(l->data + btrfs_item_offset(l->items +
0783fcfc CM	44	path.slots[0]));
02217ed2	45	item->refs++;
a28ec197	46
02217ed2 CM	47	BUG_ON(list_empty(&path.nodes[0]->dirty));
02217ed2 CM	48	release_path(root->extent_root, &path);
037e6390 CM	49	finish_current_insert(root->extent_root);
037e6390 CM	50	run_pending(root->extent_root);
02217ed2 CM	51	return 0;
	52	}
	53
20524f02	54	static int lookup_block_ref(struct ctree_root root, u64 blocknr, u32 refs)
a28ec197 CM	55	{
	56	struct ctree_path path;
	57	int ret;
e2fa7227	58	struct btrfs_key key;
a28ec197 CM	59	struct leaf *l;
	60	struct extent_item *item;
	61	init_path(&path);
	62	key.objectid = blocknr;
	63	key.flags = 0;
	64	key.offset = 1;
	65	ret = search_slot(root->extent_root, &key, &path, 0, 0);
	66	if (ret != 0)
	67	BUG();
	68	l = &path.nodes[0]->leaf;
	69	item = (struct extent_item *)(l->data +
0783fcfc CM	70	btrfs_item_offset(l->items +
0783fcfc CM	71	path.slots[0]));
a28ec197 CM	72	*refs = item->refs;
	73	release_path(root->extent_root, &path);
	74	return 0;
	75	}
	76
02217ed2 CM	77	int btrfs_inc_ref(struct ctree_root root, struct tree_buffer buf)
	78	{
	79	u64 blocknr;
	80	int i;
a28ec197 CM	81
	82	if (root == root->extent_root)
	83	return 0;
7518a238	84	if (btrfs_is_leaf(&buf->node))
a28ec197 CM	85	return 0;
a28ec197 CM	86
7518a238	87	for (i = 0; i < btrfs_header_nritems(&buf->node.header); i++) {
02217ed2 CM	88	blocknr = buf->node.blockptrs[i];
	89	inc_block_ref(root, blocknr);
	90	}
	91	return 0;
	92	}
	93
a28ec197 CM	94	int btrfs_finish_extent_commit(struct ctree_root *root)
	95	{
	96	struct ctree_root *extent_root = root->extent_root;
	97	unsigned long gang[8];
	98	int ret;
	99	int i;
	100
	101	while(1) {
	102	ret = radix_tree_gang_lookup(&extent_root->pinned_radix,
	103	(void **)gang, 0,
	104	ARRAY_SIZE(gang));
	105	if (!ret)
	106	break;
0579da42	107	for (i = 0; i < ret; i++) {
a28ec197	108	radix_tree_delete(&extent_root->pinned_radix, gang[i]);
0579da42	109	}
a28ec197	110	}
0579da42 CM	111	extent_root->last_insert.objectid = 0;
0579da42 CM	112	extent_root->last_insert.offset = 0;
a28ec197 CM	113	return 0;
	114	}
	115
037e6390 CM	116	static int finish_current_insert(struct ctree_root *extent_root)
037e6390 CM	117	{
e2fa7227	118	struct btrfs_key ins;
037e6390 CM	119	struct extent_item extent_item;
	120	int i;
	121	int ret;
	122
	123	extent_item.refs = 1;
7518a238 CM	124	extent_item.owner =
7518a238 CM	125	btrfs_header_parentid(&extent_root->node->node.header);
037e6390 CM	126	ins.offset = 1;
	127	ins.flags = 0;
	128
	129	for (i = 0; i < extent_root->current_insert.flags; i++) {
	130	ins.objectid = extent_root->current_insert.objectid + i;
	131	ret = insert_item(extent_root, &ins, &extent_item,
	132	sizeof(extent_item));
	133	BUG_ON(ret);
	134	}
	135	extent_root->current_insert.offset = 0;
	136	return 0;
	137	}
	138
fec577fb	139	/*
a28ec197	140	* remove an extent from the root, returns 0 on success
fec577fb	141	*/
a28ec197 CM	142	int __free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
	143	{
	144	struct ctree_path path;
e2fa7227	145	struct btrfs_key key;
a28ec197 CM	146	struct ctree_root *extent_root = root->extent_root;
a28ec197 CM	147	int ret;
0783fcfc	148	struct btrfs_item *item;
a28ec197	149	struct extent_item *ei;
e2fa7227	150	struct btrfs_key ins;
037e6390	151
a28ec197 CM	152	key.objectid = blocknr;
	153	key.flags = 0;
	154	key.offset = num_blocks;
	155
037e6390	156	find_free_extent(root, 0, 0, (u64)-1, &ins);
a28ec197 CM	157	init_path(&path);
	158	ret = search_slot(extent_root, &key, &path, -1, 1);
	159	if (ret) {
	160	printf("failed to find %Lu\n", key.objectid);
	161	print_tree(extent_root, extent_root->node);
	162	printf("failed to find %Lu\n", key.objectid);
	163	BUG();
	164	}
	165	item = path.nodes[0]->leaf.items + path.slots[0];
0783fcfc CM	166	ei = (struct extent_item *)(path.nodes[0]->leaf.data +
0783fcfc CM	167	btrfs_item_offset(item));
a28ec197 CM	168	BUG_ON(ei->refs == 0);
	169	ei->refs--;
	170	if (ei->refs == 0) {
	171	if (root == extent_root) {
	172	int err;
	173	radix_tree_preload(GFP_KERNEL);
	174	err = radix_tree_insert(&extent_root->pinned_radix,
	175	blocknr, (void *)blocknr);
	176	BUG_ON(err);
	177	radix_tree_preload_end();
	178	}
	179	ret = del_item(extent_root, &path);
0579da42 CM	180	if (root != extent_root &&
	181	extent_root->last_insert.objectid < blocknr)
	182	extent_root->last_insert.objectid = blocknr;
a28ec197 CM	183	if (ret)
	184	BUG();
	185	}
	186	release_path(extent_root, &path);
037e6390	187	finish_current_insert(extent_root);
a28ec197 CM	188	return ret;
	189	}
	190
a28ec197 CM	191	/*
	192	* find all the blocks marked as pending in the radix tree and remove
	193	* them from the extent map
	194	*/
	195	static int del_pending_extents(struct ctree_root *extent_root)
	196	{
	197	int ret;
	198	struct tree_buffer *gang[4];
	199	int i;
	200
	201	while(1) {
	202	ret = radix_tree_gang_lookup_tag(&extent_root->cache_radix,
	203	(void **)gang, 0,
	204	ARRAY_SIZE(gang),
	205	CTREE_EXTENT_PENDING_DEL);
	206	if (!ret)
	207	break;
	208	for (i = 0; i < ret; i++) {
	209	ret = __free_extent(extent_root, gang[i]->blocknr, 1);
fec577fb CM	210	radix_tree_tag_clear(&extent_root->cache_radix,
fec577fb CM	211	gang[i]->blocknr,
a28ec197	212	CTREE_EXTENT_PENDING_DEL);
fec577fb CM	213	tree_block_release(extent_root, gang[i]);
	214	}
	215	}
	216	return 0;
	217	}
	218
a28ec197 CM	219	static int run_pending(struct ctree_root *extent_root)
	220	{
	221	while(radix_tree_tagged(&extent_root->cache_radix,
037e6390	222	CTREE_EXTENT_PENDING_DEL))
a28ec197	223	del_pending_extents(extent_root);
a28ec197 CM	224	return 0;
	225	}
	226
	227
fec577fb CM	228	/*
	229	* remove an extent from the root, returns 0 on success
	230	*/
	231	int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
	232	{
e2fa7227	233	struct btrfs_key key;
fec577fb CM	234	struct ctree_root *extent_root = root->extent_root;
	235	struct tree_buffer *t;
	236	int pending_ret;
	237	int ret;
a28ec197	238
fec577fb	239	if (root == extent_root) {
a28ec197	240	t = find_tree_block(root, blocknr);
037e6390	241	radix_tree_tag_set(&root->cache_radix, blocknr,
a28ec197	242	CTREE_EXTENT_PENDING_DEL);
fec577fb CM	243	return 0;
fec577fb CM	244	}
a28ec197 CM	245	key.objectid = blocknr;
	246	key.flags = 0;
	247	key.offset = num_blocks;
	248	ret = __free_extent(root, blocknr, num_blocks);
	249	pending_ret = run_pending(root->extent_root);
fec577fb CM	250	return ret ? ret : pending_ret;
	251	}
	252
	253	/*
	254	* walks the btree of allocated extents and find a hole of a given size.
	255	* The key ins is changed to record the hole:
	256	* ins->objectid == block start
	257	* ins->flags = 0
	258	* ins->offset == number of blocks
	259	* Any available blocks before search_start are skipped.
	260	*/
0f70abe2	261	static int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
e2fa7227 CM	262	u64 search_start, u64 search_end,
e2fa7227 CM	263	struct btrfs_key *ins)
fec577fb CM	264	{
fec577fb CM	265	struct ctree_path path;
e2fa7227	266	struct btrfs_key key;
fec577fb CM	267	int ret;
	268	u64 hole_size = 0;
	269	int slot = 0;
	270	u64 last_block;
037e6390	271	u64 test_block;
fec577fb CM	272	int start_found;
	273	struct leaf *l;
	274	struct ctree_root * root = orig_root->extent_root;
0579da42	275	int total_needed = num_blocks;
fec577fb	276
7518a238	277	total_needed += (btrfs_header_level(&root->node->node.header) + 1) * 3;
0579da42 CM	278	if (root->last_insert.objectid > search_start)
0579da42 CM	279	search_start = root->last_insert.objectid;
fec577fb CM	280	check_failed:
	281	init_path(&path);
	282	ins->objectid = search_start;
	283	ins->offset = 0;
	284	ins->flags = 0;
	285	start_found = 0;
02217ed2	286	ret = search_slot(root, ins, &path, 0, 0);
0f70abe2 CM	287	if (ret < 0)
0f70abe2 CM	288	goto error;
aa5d6bed	289
0579da42 CM	290	if (path.slots[0] > 0)
	291	path.slots[0]--;
	292
fec577fb CM	293	while (1) {
	294	l = &path.nodes[0]->leaf;
	295	slot = path.slots[0];
7518a238	296	if (slot >= btrfs_header_nritems(&l->header)) {
fec577fb CM	297	ret = next_leaf(root, &path);
	298	if (ret == 0)
	299	continue;
0f70abe2 CM	300	if (ret < 0)
0f70abe2 CM	301	goto error;
fec577fb CM	302	if (!start_found) {
fec577fb CM	303	ins->objectid = search_start;
037e6390	304	ins->offset = (u64)-1;
fec577fb CM	305	start_found = 1;
	306	goto check_pending;
	307	}
	308	ins->objectid = last_block > search_start ?
	309	last_block : search_start;
037e6390	310	ins->offset = (u64)-1;
fec577fb CM	311	goto check_pending;
fec577fb CM	312	}
e2fa7227 CM	313	btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
e2fa7227 CM	314	if (key.objectid >= search_start) {
fec577fb	315	if (start_found) {
0579da42 CM	316	if (last_block < search_start)
0579da42 CM	317	last_block = search_start;
e2fa7227	318	hole_size = key.objectid - last_block;
037e6390	319	if (hole_size > total_needed) {
fec577fb	320	ins->objectid = last_block;
037e6390	321	ins->offset = hole_size;
fec577fb CM	322	goto check_pending;
fec577fb CM	323	}
0579da42	324	}
fec577fb	325	}
0579da42	326	start_found = 1;
e2fa7227	327	last_block = key.objectid + key.offset;
fec577fb CM	328	path.slots[0]++;
	329	}
	330	// FIXME -ENOSPC
	331	check_pending:
	332	/* we have to make sure we didn't find an extent that has already
	333	* been allocated by the map tree or the original allocation
	334	*/
	335	release_path(root, &path);
	336	BUG_ON(ins->objectid < search_start);
037e6390 CM	337	for (test_block = ins->objectid;
	338	test_block < ins->objectid + total_needed; test_block++) {
	339	if (radix_tree_lookup(&root->pinned_radix, test_block)) {
	340	search_start = test_block + 1;
fec577fb CM	341	goto check_failed;
	342	}
	343	}
037e6390	344	BUG_ON(root->current_insert.offset);
0579da42	345	root->current_insert.offset = total_needed - num_blocks;
037e6390 CM	346	root->current_insert.objectid = ins->objectid + num_blocks;
037e6390 CM	347	root->current_insert.flags = 0;
0579da42	348	root->last_insert.objectid = ins->objectid;
037e6390	349	ins->offset = num_blocks;
fec577fb	350	return 0;
0f70abe2 CM	351	error:
	352	release_path(root, &path);
	353	return ret;
fec577fb CM	354	}
fec577fb CM	355
fec577fb CM	356	/*
	357	* finds a free extent and does all the dirty work required for allocation
	358	* returns the key for the extent through ins, and a tree buffer for
	359	* the first block of the extent through buf.
	360	*
	361	* returns 0 if everything worked, non-zero otherwise.
	362	*/
	363	int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
e2fa7227	364	u64 search_end, u64 owner, struct btrfs_key *ins)
fec577fb CM	365	{
	366	int ret;
	367	int pending_ret;
037e6390	368	struct ctree_root *extent_root = root->extent_root;
fec577fb	369	struct extent_item extent_item;
037e6390	370
fec577fb CM	371	extent_item.refs = 1;
	372	extent_item.owner = owner;
	373
037e6390 CM	374	if (root == extent_root) {
	375	BUG_ON(extent_root->current_insert.offset == 0);
	376	BUG_ON(num_blocks != 1);
	377	BUG_ON(extent_root->current_insert.flags ==
	378	extent_root->current_insert.offset);
	379	ins->offset = 1;
	380	ins->objectid = extent_root->current_insert.objectid +
	381	extent_root->current_insert.flags++;
fec577fb CM	382	return 0;
fec577fb CM	383	}
037e6390 CM	384	ret = find_free_extent(root, num_blocks, search_start,
	385	search_end, ins);
	386	if (ret)
	387	return ret;
fec577fb	388
037e6390 CM	389	ret = insert_item(extent_root, ins, &extent_item,
	390	sizeof(extent_item));
	391
	392	finish_current_insert(extent_root);
	393	pending_ret = run_pending(extent_root);
	394	if (ret)
	395	return ret;
	396	if (pending_ret)
	397	return pending_ret;
	398	return 0;
fec577fb CM	399	}
	400
	401	/*
	402	* helper function to allocate a block for a given tree
	403	* returns the tree buffer or NULL.
	404	*/
	405	struct tree_buffer alloc_free_block(struct ctree_root root)
	406	{
e2fa7227	407	struct btrfs_key ins;
fec577fb	408	int ret;
037e6390	409	struct tree_buffer *buf;
fec577fb CM	410
fec577fb CM	411	ret = alloc_extent(root, 1, 0, (unsigned long)-1,
7518a238	412	btrfs_header_parentid(&root->node->node.header),
037e6390	413	&ins);
fec577fb CM	414	if (ret) {
	415	BUG();
	416	return NULL;
	417	}
037e6390 CM	418	buf = find_tree_block(root, ins.objectid);
037e6390 CM	419	dirty_tree_block(root, buf);
fec577fb CM	420	return buf;
fec577fb CM	421	}
a28ec197	422
20524f02 CM	423	int walk_down_tree(struct ctree_root root, struct ctree_path path, int *level)
	424	{
	425	struct tree_buffer *next;
	426	struct tree_buffer *cur;
	427	u64 blocknr;
	428	int ret;
	429	u32 refs;
	430
	431	ret = lookup_block_ref(root, path->nodes[*level]->blocknr, &refs);
	432	BUG_ON(ret);
	433	if (refs > 1)
	434	goto out;
	435	while(*level > 0) {
	436	cur = path->nodes[*level];
7518a238 CM	437	if (path->slots[*level] >=
7518a238 CM	438	btrfs_header_nritems(&cur->node.header))
20524f02 CM	439	break;
	440	blocknr = cur->node.blockptrs[path->slots[*level]];
	441	ret = lookup_block_ref(root, blocknr, &refs);
	442	if (refs != 1 \|\| *level == 1) {
	443	path->slots[*level]++;
	444	ret = free_extent(root, blocknr, 1);
	445	BUG_ON(ret);
	446	continue;
	447	}
	448	BUG_ON(ret);
	449	next = read_tree_block(root, blocknr);
83e15a28	450	if (path->nodes[*level-1])
20524f02	451	tree_block_release(root, path->nodes[*level-1]);
20524f02	452	path->nodes[*level-1] = next;
7518a238	453	*level = btrfs_header_level(&next->node.header);
20524f02 CM	454	path->slots[*level] = 0;
	455	}
	456	out:
	457	ret = free_extent(root, path->nodes[*level]->blocknr, 1);
83e15a28	458	tree_block_release(root, path->nodes[*level]);
20524f02 CM	459	path->nodes[*level] = NULL;
	460	*level += 1;
	461	BUG_ON(ret);
	462	return 0;
	463	}
	464
	465	int walk_up_tree(struct ctree_root root, struct ctree_path path, int *level)
	466	{
	467	int i;
	468	int slot;
	469	int ret;
	470	for(i = *level; i < MAX_LEVEL - 1 && path->nodes[i]; i++) {
	471	slot = path->slots[i];
7518a238 CM	472	if (slot <
7518a238 CM	473	btrfs_header_nritems(&path->nodes[i]->node.header)- 1) {
20524f02 CM	474	path->slots[i]++;
	475	*level = i;
	476	return 0;
	477	} else {
	478	ret = free_extent(root,
	479	path->nodes[*level]->blocknr, 1);
83e15a28 CM	480	tree_block_release(root, path->nodes[*level]);
83e15a28 CM	481	path->nodes[*level] = NULL;
20524f02 CM	482	*level = i + 1;
	483	BUG_ON(ret);
	484	}
	485	}
	486	return 1;
	487	}
	488
	489	int btrfs_drop_snapshot(struct ctree_root root, struct tree_buffer snap)
	490	{
	491	int ret;
	492	int level;
	493	struct ctree_path path;
	494	int i;
	495	int orig_level;
	496
	497	init_path(&path);
	498
7518a238	499	level = btrfs_header_level(&snap->node.header);
20524f02 CM	500	orig_level = level;
	501	path.nodes[level] = snap;
	502	path.slots[level] = 0;
	503	while(1) {
	504	ret = walk_down_tree(root, &path, &level);
	505	if (ret > 0)
	506	break;
	507	ret = walk_up_tree(root, &path, &level);
	508	if (ret > 0)
	509	break;
	510	}
83e15a28 CM	511	for (i = 0; i <= orig_level; i++) {
83e15a28 CM	512	if (path.nodes[i]) {
20524f02	513	tree_block_release(root, path.nodes[i]);
83e15a28	514	}
20524f02 CM	515	}
	516
	517	return 0;
	518	}