extent-tree.c 12.6 KB
Newer Older
1 2 3 4 5 6 7 8
#include <stdio.h>
#include <stdlib.h>
#include "kerncompat.h"
#include "radix-tree.h"
#include "ctree.h"
#include "disk-io.h"
#include "print-tree.h"

C
Chris Mason 已提交
9
static int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
C
Chris Mason 已提交
10 11
			    u64 search_start, u64 search_end,
			    struct btrfs_key *ins);
C
Chris Mason 已提交
12 13 14
static int finish_current_insert(struct ctree_root *extent_root);
static int run_pending(struct ctree_root *extent_root);

15 16 17 18 19 20 21
/*
 * 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.
 */
C
Chris Mason 已提交
22
#define CTREE_EXTENT_PENDING_DEL 0
23

C
Chris Mason 已提交
24 25 26 27
static int inc_block_ref(struct ctree_root *root, u64 blocknr)
{
	struct ctree_path path;
	int ret;
C
Chris Mason 已提交
28
	struct btrfs_key key;
C
Chris Mason 已提交
29 30
	struct leaf *l;
	struct extent_item *item;
C
Chris Mason 已提交
31
	struct btrfs_key ins;
C
Chris Mason 已提交
32 33

	find_free_extent(root->extent_root, 0, 0, (u64)-1, &ins);
C
Chris Mason 已提交
34 35 36 37 38
	init_path(&path);
	key.objectid = blocknr;
	key.flags = 0;
	key.offset = 1;
	ret = search_slot(root->extent_root, &key, &path, 0, 1);
39 40
	if (ret != 0)
		BUG();
C
Chris Mason 已提交
41 42
	BUG_ON(ret != 0);
	l = &path.nodes[0]->leaf;
C
Chris Mason 已提交
43 44
	item = (struct extent_item *)(l->data + btrfs_item_offset(l->items +
								path.slots[0]));
C
Chris Mason 已提交
45
	item->refs++;
46

C
Chris Mason 已提交
47 48
	BUG_ON(list_empty(&path.nodes[0]->dirty));
	release_path(root->extent_root, &path);
C
Chris Mason 已提交
49 50
	finish_current_insert(root->extent_root);
	run_pending(root->extent_root);
C
Chris Mason 已提交
51 52 53
	return 0;
}

C
Chris Mason 已提交
54
static int lookup_block_ref(struct ctree_root *root, u64 blocknr, u32 *refs)
55 56 57
{
	struct ctree_path path;
	int ret;
C
Chris Mason 已提交
58
	struct btrfs_key key;
59 60 61 62 63 64 65 66 67 68 69
	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 +
C
Chris Mason 已提交
70 71
				      btrfs_item_offset(l->items +
							path.slots[0]));
72 73 74 75 76
	*refs = item->refs;
	release_path(root->extent_root, &path);
	return 0;
}

C
Chris Mason 已提交
77 78 79 80
int btrfs_inc_ref(struct ctree_root *root, struct tree_buffer *buf)
{
	u64 blocknr;
	int i;
81 82 83

	if (root == root->extent_root)
		return 0;
84
	if (btrfs_is_leaf(&buf->node))
85 86
		return 0;

87
	for (i = 0; i < btrfs_header_nritems(&buf->node.header); i++) {
C
Chris Mason 已提交
88 89 90 91 92 93
		blocknr = buf->node.blockptrs[i];
		inc_block_ref(root, blocknr);
	}
	return 0;
}

94 95 96 97 98 99 100 101 102 103 104 105 106
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;
107
		for (i = 0; i < ret; i++) {
108
			radix_tree_delete(&extent_root->pinned_radix, gang[i]);
109
		}
110
	}
111 112
	extent_root->last_insert.objectid = 0;
	extent_root->last_insert.offset = 0;
113 114 115
	return 0;
}

C
Chris Mason 已提交
116 117
static int finish_current_insert(struct ctree_root *extent_root)
{
C
Chris Mason 已提交
118
	struct btrfs_key ins;
C
Chris Mason 已提交
119 120 121 122 123
	struct extent_item extent_item;
	int i;
	int ret;

	extent_item.refs = 1;
124 125
	extent_item.owner =
		btrfs_header_parentid(&extent_root->node->node.header);
C
Chris Mason 已提交
126 127 128 129 130 131 132 133 134 135 136 137 138
	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;
}

139
/*
140
 * remove an extent from the root, returns 0 on success
141
 */
142 143 144
int __free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
{
	struct ctree_path path;
C
Chris Mason 已提交
145
	struct btrfs_key key;
146 147
	struct ctree_root *extent_root = root->extent_root;
	int ret;
C
Chris Mason 已提交
148
	struct btrfs_item *item;
149
	struct extent_item *ei;
C
Chris Mason 已提交
150
	struct btrfs_key ins;
C
Chris Mason 已提交
151

152 153 154 155
	key.objectid = blocknr;
	key.flags = 0;
	key.offset = num_blocks;

C
Chris Mason 已提交
156
	find_free_extent(root, 0, 0, (u64)-1, &ins);
157 158 159 160 161 162 163 164 165
	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];
C
Chris Mason 已提交
166 167
	ei = (struct extent_item *)(path.nodes[0]->leaf.data +
				    btrfs_item_offset(item));
168 169 170 171 172 173 174 175 176 177 178 179
	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);
180 181 182
		if (root != extent_root &&
		    extent_root->last_insert.objectid < blocknr)
			extent_root->last_insert.objectid = blocknr;
183 184 185 186
		if (ret)
			BUG();
	}
	release_path(extent_root, &path);
C
Chris Mason 已提交
187
	finish_current_insert(extent_root);
188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209
	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);
210 211
			radix_tree_tag_clear(&extent_root->cache_radix,
						gang[i]->blocknr,
212
						CTREE_EXTENT_PENDING_DEL);
213 214 215 216 217 218
			tree_block_release(extent_root, gang[i]);
		}
	}
	return 0;
}

219 220 221
static int run_pending(struct ctree_root *extent_root)
{
	while(radix_tree_tagged(&extent_root->cache_radix,
C
Chris Mason 已提交
222
			        CTREE_EXTENT_PENDING_DEL))
223 224 225 226 227
		del_pending_extents(extent_root);
	return 0;
}


228 229 230 231 232
/*
 * remove an extent from the root, returns 0 on success
 */
int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
{
C
Chris Mason 已提交
233
	struct btrfs_key key;
234 235 236 237
	struct ctree_root *extent_root = root->extent_root;
	struct tree_buffer *t;
	int pending_ret;
	int ret;
238

239
	if (root == extent_root) {
240
		t = find_tree_block(root, blocknr);
C
Chris Mason 已提交
241
		radix_tree_tag_set(&root->cache_radix, blocknr,
242
				   CTREE_EXTENT_PENDING_DEL);
243 244
		return 0;
	}
245 246 247 248 249
	key.objectid = blocknr;
	key.flags = 0;
	key.offset = num_blocks;
	ret = __free_extent(root, blocknr, num_blocks);
	pending_ret = run_pending(root->extent_root);
250 251 252 253 254 255 256 257 258 259 260
	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.
 */
C
Chris Mason 已提交
261
static int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
C
Chris Mason 已提交
262 263
			    u64 search_start, u64 search_end,
			    struct btrfs_key *ins)
264 265
{
	struct ctree_path path;
C
Chris Mason 已提交
266
	struct btrfs_key key;
267 268 269 270
	int ret;
	u64 hole_size = 0;
	int slot = 0;
	u64 last_block;
C
Chris Mason 已提交
271
	u64 test_block;
272 273 274
	int start_found;
	struct leaf *l;
	struct ctree_root * root = orig_root->extent_root;
275
	int total_needed = num_blocks;
276

277
	total_needed += (btrfs_header_level(&root->node->node.header) + 1) * 3;
278 279
	if (root->last_insert.objectid > search_start)
		search_start = root->last_insert.objectid;
280 281 282 283 284 285
check_failed:
	init_path(&path);
	ins->objectid = search_start;
	ins->offset = 0;
	ins->flags = 0;
	start_found = 0;
C
Chris Mason 已提交
286
	ret = search_slot(root, ins, &path, 0, 0);
C
Chris Mason 已提交
287 288
	if (ret < 0)
		goto error;
C
Chris Mason 已提交
289

290 291 292
	if (path.slots[0] > 0)
		path.slots[0]--;

293 294 295
	while (1) {
		l = &path.nodes[0]->leaf;
		slot = path.slots[0];
296
		if (slot >= btrfs_header_nritems(&l->header)) {
297 298 299
			ret = next_leaf(root, &path);
			if (ret == 0)
				continue;
C
Chris Mason 已提交
300 301
			if (ret < 0)
				goto error;
302 303
			if (!start_found) {
				ins->objectid = search_start;
C
Chris Mason 已提交
304
				ins->offset = (u64)-1;
305 306 307 308 309
				start_found = 1;
				goto check_pending;
			}
			ins->objectid = last_block > search_start ?
					last_block : search_start;
C
Chris Mason 已提交
310
			ins->offset = (u64)-1;
311 312
			goto check_pending;
		}
C
Chris Mason 已提交
313 314
		btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
		if (key.objectid >= search_start) {
315
			if (start_found) {
316 317
				if (last_block < search_start)
					last_block = search_start;
C
Chris Mason 已提交
318
				hole_size = key.objectid - last_block;
C
Chris Mason 已提交
319
				if (hole_size > total_needed) {
320
					ins->objectid = last_block;
C
Chris Mason 已提交
321
					ins->offset = hole_size;
322 323
					goto check_pending;
				}
324
			}
325
		}
326
		start_found = 1;
C
Chris Mason 已提交
327
		last_block = key.objectid + key.offset;
328 329 330 331 332 333 334 335 336
		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);
C
Chris Mason 已提交
337 338 339 340
	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;
341 342 343
			goto check_failed;
		}
	}
C
Chris Mason 已提交
344
	BUG_ON(root->current_insert.offset);
345
	root->current_insert.offset = total_needed - num_blocks;
C
Chris Mason 已提交
346 347
	root->current_insert.objectid = ins->objectid + num_blocks;
	root->current_insert.flags = 0;
348
	root->last_insert.objectid = ins->objectid;
C
Chris Mason 已提交
349
	ins->offset = num_blocks;
350
	return 0;
C
Chris Mason 已提交
351 352 353
error:
	release_path(root, &path);
	return ret;
354 355 356 357 358 359 360 361 362 363
}

/*
 * 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,
C
Chris Mason 已提交
364
			 u64 search_end, u64 owner, struct btrfs_key *ins)
365 366 367
{
	int ret;
	int pending_ret;
C
Chris Mason 已提交
368
	struct ctree_root *extent_root = root->extent_root;
369
	struct extent_item extent_item;
C
Chris Mason 已提交
370

371 372 373
	extent_item.refs = 1;
	extent_item.owner = owner;

C
Chris Mason 已提交
374 375 376 377 378 379 380 381
	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++;
382 383
		return 0;
	}
C
Chris Mason 已提交
384 385 386 387
	ret = find_free_extent(root, num_blocks, search_start,
			       search_end, ins);
	if (ret)
		return ret;
388

C
Chris Mason 已提交
389 390 391 392 393 394 395 396 397 398
	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;
399 400 401 402 403 404 405 406
}

/*
 * 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)
{
C
Chris Mason 已提交
407
	struct btrfs_key ins;
408
	int ret;
C
Chris Mason 已提交
409
	struct tree_buffer *buf;
410 411

	ret = alloc_extent(root, 1, 0, (unsigned long)-1,
412
			   btrfs_header_parentid(&root->node->node.header),
C
Chris Mason 已提交
413
			   &ins);
414 415 416 417
	if (ret) {
		BUG();
		return NULL;
	}
C
Chris Mason 已提交
418 419
	buf = find_tree_block(root, ins.objectid);
	dirty_tree_block(root, buf);
420 421
	return buf;
}
422

C
Chris Mason 已提交
423 424 425 426 427 428 429 430 431 432 433 434 435 436
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];
437 438
		if (path->slots[*level] >=
		    btrfs_header_nritems(&cur->node.header))
C
Chris Mason 已提交
439 440 441 442 443 444 445 446 447 448 449
			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);
C
Chris Mason 已提交
450
		if (path->nodes[*level-1])
C
Chris Mason 已提交
451 452
			tree_block_release(root, path->nodes[*level-1]);
		path->nodes[*level-1] = next;
453
		*level = btrfs_header_level(&next->node.header);
C
Chris Mason 已提交
454 455 456 457
		path->slots[*level] = 0;
	}
out:
	ret = free_extent(root, path->nodes[*level]->blocknr, 1);
C
Chris Mason 已提交
458
	tree_block_release(root, path->nodes[*level]);
C
Chris Mason 已提交
459 460 461 462 463 464 465 466 467 468 469 470 471
	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];
472 473
		if (slot <
		    btrfs_header_nritems(&path->nodes[i]->node.header)- 1) {
C
Chris Mason 已提交
474 475 476 477 478 479
			path->slots[i]++;
			*level = i;
			return 0;
		} else {
			ret = free_extent(root,
					  path->nodes[*level]->blocknr, 1);
C
Chris Mason 已提交
480 481
			tree_block_release(root, path->nodes[*level]);
			path->nodes[*level] = NULL;
C
Chris Mason 已提交
482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
			*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);

499
	level = btrfs_header_level(&snap->node.header);
C
Chris Mason 已提交
500 501 502 503 504 505 506 507 508 509 510
	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;
	}
C
Chris Mason 已提交
511 512
	for (i = 0; i <= orig_level; i++) {
		if (path.nodes[i]) {
C
Chris Mason 已提交
513
			tree_block_release(root, path.nodes[i]);
C
Chris Mason 已提交
514
		}
C
Chris Mason 已提交
515 516 517 518
	}

	return 0;
}