free-space-cache.c 83.6 KB
Newer Older
J
Josef Bacik 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2008 Red Hat.  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.
 */

19
#include <linux/pagemap.h>
J
Josef Bacik 已提交
20
#include <linux/sched.h>
21
#include <linux/slab.h>
22
#include <linux/math64.h>
23
#include <linux/ratelimit.h>
J
Josef Bacik 已提交
24
#include "ctree.h"
25 26
#include "free-space-cache.h"
#include "transaction.h"
27
#include "disk-io.h"
28
#include "extent_io.h"
29
#include "inode-map.h"
30

31 32
#define BITS_PER_BITMAP		(PAGE_CACHE_SIZE * 8)
#define MAX_CACHE_BYTES_PER_GIG	(32 * 1024)
J
Josef Bacik 已提交
33

34
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
35
			   struct btrfs_free_space *info);
36 37
static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info);
J
Josef Bacik 已提交
38

39 40 41
static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
					       struct btrfs_path *path,
					       u64 offset)
42 43 44 45 46 47 48 49 50 51
{
	struct btrfs_key key;
	struct btrfs_key location;
	struct btrfs_disk_key disk_key;
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
	struct inode *inode = NULL;
	int ret;

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
52
	key.offset = offset;
53 54 55 56 57 58
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	if (ret < 0)
		return ERR_PTR(ret);
	if (ret > 0) {
59
		btrfs_release_path(path);
60 61 62 63 64 65 66 67
		return ERR_PTR(-ENOENT);
	}

	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	btrfs_free_space_key(leaf, header, &disk_key);
	btrfs_disk_key_to_cpu(&location, &disk_key);
68
	btrfs_release_path(path);
69 70 71 72 73 74 75 76 77 78 79

	inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
	if (!inode)
		return ERR_PTR(-ENOENT);
	if (IS_ERR(inode))
		return inode;
	if (is_bad_inode(inode)) {
		iput(inode);
		return ERR_PTR(-ENOENT);
	}

A
Al Viro 已提交
80 81
	mapping_set_gfp_mask(inode->i_mapping,
			mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
82

83 84 85 86 87 88 89 90
	return inode;
}

struct inode *lookup_free_space_inode(struct btrfs_root *root,
				      struct btrfs_block_group_cache
				      *block_group, struct btrfs_path *path)
{
	struct inode *inode = NULL;
91
	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
92 93 94 95 96 97 98 99 100 101 102 103 104

	spin_lock(&block_group->lock);
	if (block_group->inode)
		inode = igrab(block_group->inode);
	spin_unlock(&block_group->lock);
	if (inode)
		return inode;

	inode = __lookup_free_space_inode(root, path,
					  block_group->key.objectid);
	if (IS_ERR(inode))
		return inode;

105
	spin_lock(&block_group->lock);
106
	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
107 108
		btrfs_info(root->fs_info,
			"Old style space inode found, converting.");
109 110
		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
			BTRFS_INODE_NODATACOW;
111 112 113
		block_group->disk_cache_state = BTRFS_DC_CLEAR;
	}

114
	if (!block_group->iref) {
115 116 117 118 119 120 121 122
		block_group->inode = igrab(inode);
		block_group->iref = 1;
	}
	spin_unlock(&block_group->lock);

	return inode;
}

123 124 125 126
static int __create_free_space_inode(struct btrfs_root *root,
				     struct btrfs_trans_handle *trans,
				     struct btrfs_path *path,
				     u64 ino, u64 offset)
127 128 129 130 131 132
{
	struct btrfs_key key;
	struct btrfs_disk_key disk_key;
	struct btrfs_free_space_header *header;
	struct btrfs_inode_item *inode_item;
	struct extent_buffer *leaf;
133
	u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
134 135
	int ret;

136
	ret = btrfs_insert_empty_inode(trans, root, path, ino);
137 138 139
	if (ret)
		return ret;

140 141 142 143
	/* We inline crc's for the free disk space cache */
	if (ino != BTRFS_FREE_INO_OBJECTID)
		flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;

144 145 146 147 148 149 150 151 152 153 154 155
	leaf = path->nodes[0];
	inode_item = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_inode_item);
	btrfs_item_key(leaf, &disk_key, path->slots[0]);
	memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
			     sizeof(*inode_item));
	btrfs_set_inode_generation(leaf, inode_item, trans->transid);
	btrfs_set_inode_size(leaf, inode_item, 0);
	btrfs_set_inode_nbytes(leaf, inode_item, 0);
	btrfs_set_inode_uid(leaf, inode_item, 0);
	btrfs_set_inode_gid(leaf, inode_item, 0);
	btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
156
	btrfs_set_inode_flags(leaf, inode_item, flags);
157 158
	btrfs_set_inode_nlink(leaf, inode_item, 1);
	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
159
	btrfs_set_inode_block_group(leaf, inode_item, offset);
160
	btrfs_mark_buffer_dirty(leaf);
161
	btrfs_release_path(path);
162 163

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
164
	key.offset = offset;
165 166 167 168 169
	key.type = 0;

	ret = btrfs_insert_empty_item(trans, root, path, &key,
				      sizeof(struct btrfs_free_space_header));
	if (ret < 0) {
170
		btrfs_release_path(path);
171 172 173 174 175 176 177 178
		return ret;
	}
	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
	btrfs_set_free_space_key(leaf, header, &disk_key);
	btrfs_mark_buffer_dirty(leaf);
179
	btrfs_release_path(path);
180 181 182 183

	return 0;
}

184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
int create_free_space_inode(struct btrfs_root *root,
			    struct btrfs_trans_handle *trans,
			    struct btrfs_block_group_cache *block_group,
			    struct btrfs_path *path)
{
	int ret;
	u64 ino;

	ret = btrfs_find_free_objectid(root, &ino);
	if (ret < 0)
		return ret;

	return __create_free_space_inode(root, trans, path, ino,
					 block_group->key.objectid);
}

200 201
int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
				       struct btrfs_block_rsv *rsv)
202
{
203
	u64 needed_bytes;
204
	int ret;
205 206 207 208 209

	/* 1 for slack space, 1 for updating the inode */
	needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
		btrfs_calc_trans_metadata_size(root, 1);

210 211 212 213 214 215
	spin_lock(&rsv->lock);
	if (rsv->reserved < needed_bytes)
		ret = -ENOSPC;
	else
		ret = 0;
	spin_unlock(&rsv->lock);
216
	return ret;
217 218 219 220 221 222 223
}

int btrfs_truncate_free_space_cache(struct btrfs_root *root,
				    struct btrfs_trans_handle *trans,
				    struct inode *inode)
{
	int ret = 0;
224 225

	btrfs_i_size_write(inode, 0);
226
	truncate_pagecache(inode, 0);
227 228 229 230 231 232 233 234

	/*
	 * We don't need an orphan item because truncating the free space cache
	 * will never be split across transactions.
	 */
	ret = btrfs_truncate_inode_items(trans, root, inode,
					 0, BTRFS_EXTENT_DATA_KEY);
	if (ret) {
235
		btrfs_abort_transaction(trans, root, ret);
236 237 238
		return ret;
	}

239
	ret = btrfs_update_inode(trans, root, inode);
240 241
	if (ret)
		btrfs_abort_transaction(trans, root, ret);
242

243
	return ret;
244 245
}

246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264
static int readahead_cache(struct inode *inode)
{
	struct file_ra_state *ra;
	unsigned long last_index;

	ra = kzalloc(sizeof(*ra), GFP_NOFS);
	if (!ra)
		return -ENOMEM;

	file_ra_state_init(ra, inode->i_mapping);
	last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;

	page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);

	kfree(ra);

	return 0;
}

265 266 267 268 269 270 271 272
struct io_ctl {
	void *cur, *orig;
	struct page *page;
	struct page **pages;
	struct btrfs_root *root;
	unsigned long size;
	int index;
	int num_pages;
273
	unsigned check_crcs:1;
274 275 276
};

static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
277
		       struct btrfs_root *root, int write)
278
{
279 280 281
	int num_pages;
	int check_crcs = 0;

282
	num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
283 284 285 286 287 288 289 290 291

	if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
		check_crcs = 1;

	/* Make sure we can fit our crcs into the first page */
	if (write && check_crcs &&
	    (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
		return -ENOSPC;

292
	memset(io_ctl, 0, sizeof(struct io_ctl));
293 294

	io_ctl->pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS);
295 296
	if (!io_ctl->pages)
		return -ENOMEM;
297 298

	io_ctl->num_pages = num_pages;
299
	io_ctl->root = root;
300 301
	io_ctl->check_crcs = check_crcs;

302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
	return 0;
}

static void io_ctl_free(struct io_ctl *io_ctl)
{
	kfree(io_ctl->pages);
}

static void io_ctl_unmap_page(struct io_ctl *io_ctl)
{
	if (io_ctl->cur) {
		kunmap(io_ctl->page);
		io_ctl->cur = NULL;
		io_ctl->orig = NULL;
	}
}

static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
{
321
	ASSERT(io_ctl->index < io_ctl->num_pages);
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
	io_ctl->page = io_ctl->pages[io_ctl->index++];
	io_ctl->cur = kmap(io_ctl->page);
	io_ctl->orig = io_ctl->cur;
	io_ctl->size = PAGE_CACHE_SIZE;
	if (clear)
		memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
}

static void io_ctl_drop_pages(struct io_ctl *io_ctl)
{
	int i;

	io_ctl_unmap_page(io_ctl);

	for (i = 0; i < io_ctl->num_pages; i++) {
337 338 339 340 341
		if (io_ctl->pages[i]) {
			ClearPageChecked(io_ctl->pages[i]);
			unlock_page(io_ctl->pages[i]);
			page_cache_release(io_ctl->pages[i]);
		}
342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
	}
}

static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
				int uptodate)
{
	struct page *page;
	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
	int i;

	for (i = 0; i < io_ctl->num_pages; i++) {
		page = find_or_create_page(inode->i_mapping, i, mask);
		if (!page) {
			io_ctl_drop_pages(io_ctl);
			return -ENOMEM;
		}
		io_ctl->pages[i] = page;
		if (uptodate && !PageUptodate(page)) {
			btrfs_readpage(NULL, page);
			lock_page(page);
			if (!PageUptodate(page)) {
363 364
				btrfs_err(BTRFS_I(inode)->root->fs_info,
					   "error reading free space cache");
365 366 367 368 369 370
				io_ctl_drop_pages(io_ctl);
				return -EIO;
			}
		}
	}

371 372 373 374 375
	for (i = 0; i < io_ctl->num_pages; i++) {
		clear_page_dirty_for_io(io_ctl->pages[i]);
		set_page_extent_mapped(io_ctl->pages[i]);
	}

376 377 378 379 380
	return 0;
}

static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
{
A
Al Viro 已提交
381
	__le64 *val;
382 383 384 385

	io_ctl_map_page(io_ctl, 1);

	/*
386 387
	 * Skip the csum areas.  If we don't check crcs then we just have a
	 * 64bit chunk at the front of the first page.
388
	 */
389 390 391 392 393 394 395
	if (io_ctl->check_crcs) {
		io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
		io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
	} else {
		io_ctl->cur += sizeof(u64);
		io_ctl->size -= sizeof(u64) * 2;
	}
396 397 398 399 400 401 402 403

	val = io_ctl->cur;
	*val = cpu_to_le64(generation);
	io_ctl->cur += sizeof(u64);
}

static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
{
A
Al Viro 已提交
404
	__le64 *gen;
405

406 407 408 409 410 411 412 413 414 415 416 417
	/*
	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
	 * chunk at the front of the first page.
	 */
	if (io_ctl->check_crcs) {
		io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
		io_ctl->size -= sizeof(u64) +
			(sizeof(u32) * io_ctl->num_pages);
	} else {
		io_ctl->cur += sizeof(u64);
		io_ctl->size -= sizeof(u64) * 2;
	}
418 419 420

	gen = io_ctl->cur;
	if (le64_to_cpu(*gen) != generation) {
421
		printk_ratelimited(KERN_ERR "BTRFS: space cache generation "
422 423 424 425 426 427
				   "(%Lu) does not match inode (%Lu)\n", *gen,
				   generation);
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}
	io_ctl->cur += sizeof(u64);
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
	return 0;
}

static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
{
	u32 *tmp;
	u32 crc = ~(u32)0;
	unsigned offset = 0;

	if (!io_ctl->check_crcs) {
		io_ctl_unmap_page(io_ctl);
		return;
	}

	if (index == 0)
443
		offset = sizeof(u32) * io_ctl->num_pages;
444

445
	crc = btrfs_csum_data(io_ctl->orig + offset, crc,
446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
			      PAGE_CACHE_SIZE - offset);
	btrfs_csum_final(crc, (char *)&crc);
	io_ctl_unmap_page(io_ctl);
	tmp = kmap(io_ctl->pages[0]);
	tmp += index;
	*tmp = crc;
	kunmap(io_ctl->pages[0]);
}

static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
{
	u32 *tmp, val;
	u32 crc = ~(u32)0;
	unsigned offset = 0;

	if (!io_ctl->check_crcs) {
		io_ctl_map_page(io_ctl, 0);
		return 0;
	}

	if (index == 0)
		offset = sizeof(u32) * io_ctl->num_pages;

	tmp = kmap(io_ctl->pages[0]);
	tmp += index;
	val = *tmp;
	kunmap(io_ctl->pages[0]);

	io_ctl_map_page(io_ctl, 0);
475
	crc = btrfs_csum_data(io_ctl->orig + offset, crc,
476 477 478
			      PAGE_CACHE_SIZE - offset);
	btrfs_csum_final(crc, (char *)&crc);
	if (val != crc) {
479
		printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free "
480 481 482 483 484
				   "space cache\n");
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}

485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506
	return 0;
}

static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
			    void *bitmap)
{
	struct btrfs_free_space_entry *entry;

	if (!io_ctl->cur)
		return -ENOSPC;

	entry = io_ctl->cur;
	entry->offset = cpu_to_le64(offset);
	entry->bytes = cpu_to_le64(bytes);
	entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
		BTRFS_FREE_SPACE_EXTENT;
	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
	io_ctl->size -= sizeof(struct btrfs_free_space_entry);

	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
		return 0;

507
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527

	/* No more pages to map */
	if (io_ctl->index >= io_ctl->num_pages)
		return 0;

	/* map the next page */
	io_ctl_map_page(io_ctl, 1);
	return 0;
}

static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
{
	if (!io_ctl->cur)
		return -ENOSPC;

	/*
	 * If we aren't at the start of the current page, unmap this one and
	 * map the next one if there is any left.
	 */
	if (io_ctl->cur != io_ctl->orig) {
528
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
529 530 531 532 533 534
		if (io_ctl->index >= io_ctl->num_pages)
			return -ENOSPC;
		io_ctl_map_page(io_ctl, 0);
	}

	memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
535
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
536 537 538 539 540 541 542
	if (io_ctl->index < io_ctl->num_pages)
		io_ctl_map_page(io_ctl, 0);
	return 0;
}

static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
{
543 544 545 546 547 548 549 550
	/*
	 * If we're not on the boundary we know we've modified the page and we
	 * need to crc the page.
	 */
	if (io_ctl->cur != io_ctl->orig)
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
	else
		io_ctl_unmap_page(io_ctl);
551 552 553

	while (io_ctl->index < io_ctl->num_pages) {
		io_ctl_map_page(io_ctl, 1);
554
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
555 556 557
	}
}

558 559
static int io_ctl_read_entry(struct io_ctl *io_ctl,
			    struct btrfs_free_space *entry, u8 *type)
560 561
{
	struct btrfs_free_space_entry *e;
562 563 564 565 566 567 568
	int ret;

	if (!io_ctl->cur) {
		ret = io_ctl_check_crc(io_ctl, io_ctl->index);
		if (ret)
			return ret;
	}
569 570 571 572

	e = io_ctl->cur;
	entry->offset = le64_to_cpu(e->offset);
	entry->bytes = le64_to_cpu(e->bytes);
573
	*type = e->type;
574 575 576 577
	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
	io_ctl->size -= sizeof(struct btrfs_free_space_entry);

	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
578
		return 0;
579 580 581

	io_ctl_unmap_page(io_ctl);

582
	return 0;
583 584
}

585 586
static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
			      struct btrfs_free_space *entry)
587
{
588 589 590 591 592 593
	int ret;

	ret = io_ctl_check_crc(io_ctl, io_ctl->index);
	if (ret)
		return ret;

594 595
	memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
	io_ctl_unmap_page(io_ctl);
596 597

	return 0;
598 599
}

600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
/*
 * Since we attach pinned extents after the fact we can have contiguous sections
 * of free space that are split up in entries.  This poses a problem with the
 * tree logging stuff since it could have allocated across what appears to be 2
 * entries since we would have merged the entries when adding the pinned extents
 * back to the free space cache.  So run through the space cache that we just
 * loaded and merge contiguous entries.  This will make the log replay stuff not
 * blow up and it will make for nicer allocator behavior.
 */
static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
{
	struct btrfs_free_space *e, *prev = NULL;
	struct rb_node *n;

again:
	spin_lock(&ctl->tree_lock);
	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
		e = rb_entry(n, struct btrfs_free_space, offset_index);
		if (!prev)
			goto next;
		if (e->bitmap || prev->bitmap)
			goto next;
		if (prev->offset + prev->bytes == e->offset) {
			unlink_free_space(ctl, prev);
			unlink_free_space(ctl, e);
			prev->bytes += e->bytes;
			kmem_cache_free(btrfs_free_space_cachep, e);
			link_free_space(ctl, prev);
			prev = NULL;
			spin_unlock(&ctl->tree_lock);
			goto again;
		}
next:
		prev = e;
	}
	spin_unlock(&ctl->tree_lock);
}

638 639 640
static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
				   struct btrfs_free_space_ctl *ctl,
				   struct btrfs_path *path, u64 offset)
641 642 643
{
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
644
	struct io_ctl io_ctl;
645
	struct btrfs_key key;
646
	struct btrfs_free_space *e, *n;
647 648 649 650
	struct list_head bitmaps;
	u64 num_entries;
	u64 num_bitmaps;
	u64 generation;
651
	u8 type;
652
	int ret = 0;
653 654 655 656

	INIT_LIST_HEAD(&bitmaps);

	/* Nothing in the space cache, goodbye */
657
	if (!i_size_read(inode))
658
		return 0;
659 660

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
661
	key.offset = offset;
662 663 664
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
665
	if (ret < 0)
666
		return 0;
667
	else if (ret > 0) {
668
		btrfs_release_path(path);
669
		return 0;
670 671
	}

672 673
	ret = -1;

674 675 676 677 678 679
	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	num_entries = btrfs_free_space_entries(leaf, header);
	num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
	generation = btrfs_free_space_generation(leaf, header);
680
	btrfs_release_path(path);
681

682 683 684 685 686 687 688
	if (!BTRFS_I(inode)->generation) {
		btrfs_info(root->fs_info,
			   "The free space cache file (%llu) is invalid. skip it\n",
			   offset);
		return 0;
	}

689
	if (BTRFS_I(inode)->generation != generation) {
690 691 692
		btrfs_err(root->fs_info,
			"free space inode generation (%llu) "
			"did not match free space cache generation (%llu)",
693
			BTRFS_I(inode)->generation, generation);
694
		return 0;
695 696 697
	}

	if (!num_entries)
698
		return 0;
699

700
	ret = io_ctl_init(&io_ctl, inode, root, 0);
701 702 703
	if (ret)
		return ret;

704
	ret = readahead_cache(inode);
705
	if (ret)
706 707
		goto out;

708 709 710
	ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
	if (ret)
		goto out;
711

712 713 714 715
	ret = io_ctl_check_crc(&io_ctl, 0);
	if (ret)
		goto free_cache;

716 717 718
	ret = io_ctl_check_generation(&io_ctl, generation);
	if (ret)
		goto free_cache;
719

720 721 722 723
	while (num_entries) {
		e = kmem_cache_zalloc(btrfs_free_space_cachep,
				      GFP_NOFS);
		if (!e)
724 725
			goto free_cache;

726 727 728 729 730 731
		ret = io_ctl_read_entry(&io_ctl, e, &type);
		if (ret) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
		}

732 733 734
		if (!e->bytes) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
735
		}
736 737 738 739 740 741

		if (type == BTRFS_FREE_SPACE_EXTENT) {
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
742 743
				btrfs_err(root->fs_info,
					"Duplicate entries in free space cache, dumping");
744
				kmem_cache_free(btrfs_free_space_cachep, e);
745 746
				goto free_cache;
			}
747
		} else {
748
			ASSERT(num_bitmaps);
749 750 751 752 753
			num_bitmaps--;
			e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
			if (!e->bitmap) {
				kmem_cache_free(
					btrfs_free_space_cachep, e);
754 755
				goto free_cache;
			}
756 757 758 759 760 761
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			ctl->total_bitmaps++;
			ctl->op->recalc_thresholds(ctl);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
762 763
				btrfs_err(root->fs_info,
					"Duplicate entries in free space cache, dumping");
764
				kmem_cache_free(btrfs_free_space_cachep, e);
765 766
				goto free_cache;
			}
767
			list_add_tail(&e->list, &bitmaps);
768 769
		}

770 771
		num_entries--;
	}
772

773 774
	io_ctl_unmap_page(&io_ctl);

775 776 777 778 779
	/*
	 * We add the bitmaps at the end of the entries in order that
	 * the bitmap entries are added to the cache.
	 */
	list_for_each_entry_safe(e, n, &bitmaps, list) {
780
		list_del_init(&e->list);
781 782 783
		ret = io_ctl_read_bitmap(&io_ctl, e);
		if (ret)
			goto free_cache;
784 785
	}

786
	io_ctl_drop_pages(&io_ctl);
787
	merge_space_tree(ctl);
788 789
	ret = 1;
out:
790
	io_ctl_free(&io_ctl);
791 792
	return ret;
free_cache:
793
	io_ctl_drop_pages(&io_ctl);
794
	__btrfs_remove_free_space_cache(ctl);
795 796 797
	goto out;
}

798 799
int load_free_space_cache(struct btrfs_fs_info *fs_info,
			  struct btrfs_block_group_cache *block_group)
J
Josef Bacik 已提交
800
{
801
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
802 803 804
	struct btrfs_root *root = fs_info->tree_root;
	struct inode *inode;
	struct btrfs_path *path;
805
	int ret = 0;
806 807 808 809 810 811 812
	bool matched;
	u64 used = btrfs_block_group_used(&block_group->item);

	/*
	 * If this block group has been marked to be cleared for one reason or
	 * another then we can't trust the on disk cache, so just return.
	 */
813
	spin_lock(&block_group->lock);
814 815 816 817
	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
		spin_unlock(&block_group->lock);
		return 0;
	}
818
	spin_unlock(&block_group->lock);
819 820 821 822

	path = btrfs_alloc_path();
	if (!path)
		return 0;
823 824
	path->search_commit_root = 1;
	path->skip_locking = 1;
825 826 827 828 829 830 831

	inode = lookup_free_space_inode(root, block_group, path);
	if (IS_ERR(inode)) {
		btrfs_free_path(path);
		return 0;
	}

832 833 834 835
	/* We may have converted the inode and made the cache invalid. */
	spin_lock(&block_group->lock);
	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
		spin_unlock(&block_group->lock);
836
		btrfs_free_path(path);
837 838 839 840
		goto out;
	}
	spin_unlock(&block_group->lock);

841 842 843 844 845 846 847 848 849 850 851 852 853
	ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
				      path, block_group->key.objectid);
	btrfs_free_path(path);
	if (ret <= 0)
		goto out;

	spin_lock(&ctl->tree_lock);
	matched = (ctl->free_space == (block_group->key.offset - used -
				       block_group->bytes_super));
	spin_unlock(&ctl->tree_lock);

	if (!matched) {
		__btrfs_remove_free_space_cache(ctl);
854
		btrfs_warn(fs_info, "block group %llu has wrong amount of free space",
855
			block_group->key.objectid);
856 857 858 859 860 861 862 863
		ret = -1;
	}
out:
	if (ret < 0) {
		/* This cache is bogus, make sure it gets cleared */
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_CLEAR;
		spin_unlock(&block_group->lock);
864
		ret = 0;
865

866
		btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuild it now",
867
			block_group->key.objectid);
868 869 870 871
	}

	iput(inode);
	return ret;
872 873
}

874 875 876 877 878 879
static noinline_for_stack
int write_cache_extent_entries(struct io_ctl *io_ctl,
			      struct btrfs_free_space_ctl *ctl,
			      struct btrfs_block_group_cache *block_group,
			      int *entries, int *bitmaps,
			      struct list_head *bitmap_list)
J
Josef Bacik 已提交
880
{
881
	int ret;
882 883
	struct btrfs_free_cluster *cluster = NULL;
	struct rb_node *node = rb_first(&ctl->free_space_offset);
884

885
	/* Get the cluster for this block_group if it exists */
886
	if (block_group && !list_empty(&block_group->cluster_list)) {
887 888 889
		cluster = list_entry(block_group->cluster_list.next,
				     struct btrfs_free_cluster,
				     block_group_list);
890
	}
891

892 893 894 895 896
	if (!node && cluster) {
		node = rb_first(&cluster->root);
		cluster = NULL;
	}

897 898 899
	/* Write out the extent entries */
	while (node) {
		struct btrfs_free_space *e;
J
Josef Bacik 已提交
900

901
		e = rb_entry(node, struct btrfs_free_space, offset_index);
902
		*entries += 1;
J
Josef Bacik 已提交
903

904
		ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
905 906
				       e->bitmap);
		if (ret)
907
			goto fail;
908

909
		if (e->bitmap) {
910 911
			list_add_tail(&e->list, bitmap_list);
			*bitmaps += 1;
912
		}
913 914 915 916
		node = rb_next(node);
		if (!node && cluster) {
			node = rb_first(&cluster->root);
			cluster = NULL;
917
		}
918
	}
919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979
	return 0;
fail:
	return -ENOSPC;
}

static noinline_for_stack int
update_cache_item(struct btrfs_trans_handle *trans,
		  struct btrfs_root *root,
		  struct inode *inode,
		  struct btrfs_path *path, u64 offset,
		  int entries, int bitmaps)
{
	struct btrfs_key key;
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
	int ret;

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
	key.offset = offset;
	key.type = 0;

	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
	if (ret < 0) {
		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
				 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
				 GFP_NOFS);
		goto fail;
	}
	leaf = path->nodes[0];
	if (ret > 0) {
		struct btrfs_key found_key;
		ASSERT(path->slots[0]);
		path->slots[0]--;
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
		if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
		    found_key.offset != offset) {
			clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
					 inode->i_size - 1,
					 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
					 NULL, GFP_NOFS);
			btrfs_release_path(path);
			goto fail;
		}
	}

	BTRFS_I(inode)->generation = trans->transid;
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	btrfs_set_free_space_entries(leaf, header, entries);
	btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
	btrfs_set_free_space_generation(leaf, header, trans->transid);
	btrfs_mark_buffer_dirty(leaf);
	btrfs_release_path(path);

	return 0;

fail:
	return -1;
}

static noinline_for_stack int
980 981 982 983
write_pinned_extent_entries(struct btrfs_root *root,
			    struct btrfs_block_group_cache *block_group,
			    struct io_ctl *io_ctl,
			    int *entries)
984 985 986 987
{
	u64 start, extent_start, extent_end, len;
	struct extent_io_tree *unpin = NULL;
	int ret;
988

989 990 991
	if (!block_group)
		return 0;

992 993 994
	/*
	 * We want to add any pinned extents to our free space cache
	 * so we don't leak the space
995
	 *
996 997 998 999 1000
	 * We shouldn't have switched the pinned extents yet so this is the
	 * right one
	 */
	unpin = root->fs_info->pinned_extents;

1001
	start = block_group->key.objectid;
1002

1003
	while (start < block_group->key.objectid + block_group->key.offset) {
1004 1005
		ret = find_first_extent_bit(unpin, start,
					    &extent_start, &extent_end,
1006
					    EXTENT_DIRTY, NULL);
1007 1008
		if (ret)
			return 0;
J
Josef Bacik 已提交
1009

1010
		/* This pinned extent is out of our range */
1011
		if (extent_start >= block_group->key.objectid +
1012
		    block_group->key.offset)
1013
			return 0;
1014

1015 1016 1017 1018
		extent_start = max(extent_start, start);
		extent_end = min(block_group->key.objectid +
				 block_group->key.offset, extent_end + 1);
		len = extent_end - extent_start;
J
Josef Bacik 已提交
1019

1020 1021
		*entries += 1;
		ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1022
		if (ret)
1023
			return -ENOSPC;
J
Josef Bacik 已提交
1024

1025
		start = extent_end;
1026
	}
J
Josef Bacik 已提交
1027

1028 1029 1030 1031 1032 1033 1034 1035 1036
	return 0;
}

static noinline_for_stack int
write_bitmap_entries(struct io_ctl *io_ctl, struct list_head *bitmap_list)
{
	struct list_head *pos, *n;
	int ret;

J
Josef Bacik 已提交
1037
	/* Write out the bitmaps */
1038
	list_for_each_safe(pos, n, bitmap_list) {
J
Josef Bacik 已提交
1039 1040 1041
		struct btrfs_free_space *entry =
			list_entry(pos, struct btrfs_free_space, list);

1042
		ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1043
		if (ret)
1044
			return -ENOSPC;
J
Josef Bacik 已提交
1045
		list_del_init(&entry->list);
1046 1047
	}

1048 1049
	return 0;
}
J
Josef Bacik 已提交
1050

1051 1052 1053
static int flush_dirty_cache(struct inode *inode)
{
	int ret;
1054

1055
	ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1056
	if (ret)
1057 1058 1059
		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
				 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
				 GFP_NOFS);
J
Josef Bacik 已提交
1060

1061
	return ret;
1062 1063 1064 1065 1066 1067 1068 1069 1070
}

static void noinline_for_stack
cleanup_write_cache_enospc(struct inode *inode,
			   struct io_ctl *io_ctl,
			   struct extent_state **cached_state,
			   struct list_head *bitmap_list)
{
	struct list_head *pos, *n;
1071

1072 1073 1074 1075
	list_for_each_safe(pos, n, bitmap_list) {
		struct btrfs_free_space *entry =
			list_entry(pos, struct btrfs_free_space, list);
		list_del_init(&entry->list);
J
Josef Bacik 已提交
1076
	}
1077 1078 1079 1080 1081
	io_ctl_drop_pages(io_ctl);
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
			     i_size_read(inode) - 1, cached_state,
			     GFP_NOFS);
}
1082

1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
/**
 * __btrfs_write_out_cache - write out cached info to an inode
 * @root - the root the inode belongs to
 * @ctl - the free space cache we are going to write out
 * @block_group - the block_group for this cache if it belongs to a block_group
 * @trans - the trans handle
 * @path - the path to use
 * @offset - the offset for the key we'll insert
 *
 * This function writes out a free space cache struct to disk for quick recovery
 * on mount.  This will return 0 if it was successfull in writing the cache out,
 * and -1 if it was not.
 */
static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
				   struct btrfs_free_space_ctl *ctl,
				   struct btrfs_block_group_cache *block_group,
				   struct btrfs_trans_handle *trans,
				   struct btrfs_path *path, u64 offset)
{
	struct extent_state *cached_state = NULL;
	struct io_ctl io_ctl;
1104
	LIST_HEAD(bitmap_list);
1105 1106 1107 1108 1109 1110 1111
	int entries = 0;
	int bitmaps = 0;
	int ret;

	if (!i_size_read(inode))
		return -1;

1112
	ret = io_ctl_init(&io_ctl, inode, root, 1);
1113 1114 1115
	if (ret)
		return -1;

1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129
	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
		down_write(&block_group->data_rwsem);
		spin_lock(&block_group->lock);
		if (block_group->delalloc_bytes) {
			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
			spin_unlock(&block_group->lock);
			up_write(&block_group->data_rwsem);
			BTRFS_I(inode)->generation = 0;
			ret = 0;
			goto out;
		}
		spin_unlock(&block_group->lock);
	}

1130 1131 1132 1133 1134 1135 1136 1137
	/* Lock all pages first so we can lock the extent safely. */
	io_ctl_prepare_pages(&io_ctl, inode, 0);

	lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
			 0, &cached_state);

	io_ctl_set_generation(&io_ctl, trans->transid);

1138
	/* Write out the extent entries in the free space cache */
1139 1140 1141 1142 1143 1144
	ret = write_cache_extent_entries(&io_ctl, ctl,
					 block_group, &entries, &bitmaps,
					 &bitmap_list);
	if (ret)
		goto out_nospc;

1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160
	/*
	 * Some spaces that are freed in the current transaction are pinned,
	 * they will be added into free space cache after the transaction is
	 * committed, we shouldn't lose them.
	 */
	ret = write_pinned_extent_entries(root, block_group, &io_ctl, &entries);
	if (ret)
		goto out_nospc;

	/* At last, we write out all the bitmaps. */
	ret = write_bitmap_entries(&io_ctl, &bitmap_list);
	if (ret)
		goto out_nospc;

	/* Zero out the rest of the pages just to make sure */
	io_ctl_zero_remaining_pages(&io_ctl);
1161

1162 1163 1164 1165
	/* Everything is written out, now we dirty the pages in the file. */
	ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
				0, i_size_read(inode), &cached_state);
	if (ret)
1166
		goto out_nospc;
1167

1168 1169
	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
		up_write(&block_group->data_rwsem);
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181
	/*
	 * Release the pages and unlock the extent, we will flush
	 * them out later
	 */
	io_ctl_drop_pages(&io_ctl);

	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
			     i_size_read(inode) - 1, &cached_state, GFP_NOFS);

	/* Flush the dirty pages in the cache file. */
	ret = flush_dirty_cache(inode);
	if (ret)
1182 1183
		goto out;

1184 1185
	/* Update the cache item to tell everyone this cache file is valid. */
	ret = update_cache_item(trans, root, inode, path, offset,
1186
				entries, bitmaps);
1187
out:
1188
	io_ctl_free(&io_ctl);
1189
	if (ret) {
1190
		invalidate_inode_pages2(inode->i_mapping);
J
Josef Bacik 已提交
1191 1192 1193
		BTRFS_I(inode)->generation = 0;
	}
	btrfs_update_inode(trans, root, inode);
1194
	return ret;
1195 1196

out_nospc:
1197
	cleanup_write_cache_enospc(inode, &io_ctl, &cached_state, &bitmap_list);
1198 1199 1200 1201

	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
		up_write(&block_group->data_rwsem);

1202
	goto out;
1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
}

int btrfs_write_out_cache(struct btrfs_root *root,
			  struct btrfs_trans_handle *trans,
			  struct btrfs_block_group_cache *block_group,
			  struct btrfs_path *path)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct inode *inode;
	int ret = 0;

	root = root->fs_info->tree_root;

	spin_lock(&block_group->lock);
	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
		spin_unlock(&block_group->lock);
		return 0;
	}
1221 1222 1223 1224 1225 1226

	if (block_group->delalloc_bytes) {
		block_group->disk_cache_state = BTRFS_DC_WRITTEN;
		spin_unlock(&block_group->lock);
		return 0;
	}
1227 1228 1229 1230 1231 1232 1233 1234
	spin_unlock(&block_group->lock);

	inode = lookup_free_space_inode(root, block_group, path);
	if (IS_ERR(inode))
		return 0;

	ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
				      path, block_group->key.objectid);
1235
	if (ret) {
1236 1237 1238
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_ERROR;
		spin_unlock(&block_group->lock);
1239
		ret = 0;
1240
#ifdef DEBUG
1241 1242 1243
		btrfs_err(root->fs_info,
			"failed to write free space cache for block group %llu",
			block_group->key.objectid);
1244
#endif
1245 1246
	}

J
Josef Bacik 已提交
1247 1248 1249 1250
	iput(inode);
	return ret;
}

1251
static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1252
					  u64 offset)
J
Josef Bacik 已提交
1253
{
1254
	ASSERT(offset >= bitmap_start);
1255
	offset -= bitmap_start;
1256
	return (unsigned long)(div_u64(offset, unit));
1257
}
J
Josef Bacik 已提交
1258

1259
static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1260
{
1261
	return (unsigned long)(div_u64(bytes, unit));
1262
}
J
Josef Bacik 已提交
1263

1264
static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1265 1266 1267 1268
				   u64 offset)
{
	u64 bitmap_start;
	u64 bytes_per_bitmap;
J
Josef Bacik 已提交
1269

1270 1271
	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
	bitmap_start = offset - ctl->start;
1272 1273
	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
	bitmap_start *= bytes_per_bitmap;
1274
	bitmap_start += ctl->start;
J
Josef Bacik 已提交
1275

1276
	return bitmap_start;
J
Josef Bacik 已提交
1277 1278
}

1279 1280
static int tree_insert_offset(struct rb_root *root, u64 offset,
			      struct rb_node *node, int bitmap)
J
Josef Bacik 已提交
1281 1282 1283 1284 1285 1286 1287
{
	struct rb_node **p = &root->rb_node;
	struct rb_node *parent = NULL;
	struct btrfs_free_space *info;

	while (*p) {
		parent = *p;
1288
		info = rb_entry(parent, struct btrfs_free_space, offset_index);
J
Josef Bacik 已提交
1289

1290
		if (offset < info->offset) {
J
Josef Bacik 已提交
1291
			p = &(*p)->rb_left;
1292
		} else if (offset > info->offset) {
J
Josef Bacik 已提交
1293
			p = &(*p)->rb_right;
1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308
		} else {
			/*
			 * we could have a bitmap entry and an extent entry
			 * share the same offset.  If this is the case, we want
			 * the extent entry to always be found first if we do a
			 * linear search through the tree, since we want to have
			 * the quickest allocation time, and allocating from an
			 * extent is faster than allocating from a bitmap.  So
			 * if we're inserting a bitmap and we find an entry at
			 * this offset, we want to go right, or after this entry
			 * logically.  If we are inserting an extent and we've
			 * found a bitmap, we want to go left, or before
			 * logically.
			 */
			if (bitmap) {
1309 1310 1311 1312
				if (info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1313 1314
				p = &(*p)->rb_right;
			} else {
1315 1316 1317 1318
				if (!info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1319 1320 1321
				p = &(*p)->rb_left;
			}
		}
J
Josef Bacik 已提交
1322 1323 1324 1325 1326 1327 1328 1329 1330
	}

	rb_link_node(node, parent, p);
	rb_insert_color(node, root);

	return 0;
}

/*
J
Josef Bacik 已提交
1331 1332
 * searches the tree for the given offset.
 *
1333 1334 1335
 * fuzzy - If this is set, then we are trying to make an allocation, and we just
 * want a section that has at least bytes size and comes at or after the given
 * offset.
J
Josef Bacik 已提交
1336
 */
1337
static struct btrfs_free_space *
1338
tree_search_offset(struct btrfs_free_space_ctl *ctl,
1339
		   u64 offset, int bitmap_only, int fuzzy)
J
Josef Bacik 已提交
1340
{
1341
	struct rb_node *n = ctl->free_space_offset.rb_node;
1342 1343 1344 1345 1346 1347 1348 1349
	struct btrfs_free_space *entry, *prev = NULL;

	/* find entry that is closest to the 'offset' */
	while (1) {
		if (!n) {
			entry = NULL;
			break;
		}
J
Josef Bacik 已提交
1350 1351

		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1352
		prev = entry;
J
Josef Bacik 已提交
1353

1354
		if (offset < entry->offset)
J
Josef Bacik 已提交
1355
			n = n->rb_left;
1356
		else if (offset > entry->offset)
J
Josef Bacik 已提交
1357
			n = n->rb_right;
1358
		else
J
Josef Bacik 已提交
1359 1360 1361
			break;
	}

1362 1363 1364 1365 1366
	if (bitmap_only) {
		if (!entry)
			return NULL;
		if (entry->bitmap)
			return entry;
J
Josef Bacik 已提交
1367

1368 1369 1370 1371 1372 1373 1374 1375 1376 1377
		/*
		 * bitmap entry and extent entry may share same offset,
		 * in that case, bitmap entry comes after extent entry.
		 */
		n = rb_next(n);
		if (!n)
			return NULL;
		entry = rb_entry(n, struct btrfs_free_space, offset_index);
		if (entry->offset != offset)
			return NULL;
J
Josef Bacik 已提交
1378

1379 1380 1381 1382
		WARN_ON(!entry->bitmap);
		return entry;
	} else if (entry) {
		if (entry->bitmap) {
J
Josef Bacik 已提交
1383
			/*
1384 1385
			 * if previous extent entry covers the offset,
			 * we should return it instead of the bitmap entry
J
Josef Bacik 已提交
1386
			 */
1387 1388
			n = rb_prev(&entry->offset_index);
			if (n) {
1389 1390
				prev = rb_entry(n, struct btrfs_free_space,
						offset_index);
1391 1392 1393
				if (!prev->bitmap &&
				    prev->offset + prev->bytes > offset)
					entry = prev;
J
Josef Bacik 已提交
1394
			}
1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
		}
		return entry;
	}

	if (!prev)
		return NULL;

	/* find last entry before the 'offset' */
	entry = prev;
	if (entry->offset > offset) {
		n = rb_prev(&entry->offset_index);
		if (n) {
			entry = rb_entry(n, struct btrfs_free_space,
					offset_index);
1409
			ASSERT(entry->offset <= offset);
J
Josef Bacik 已提交
1410
		} else {
1411 1412 1413 1414
			if (fuzzy)
				return entry;
			else
				return NULL;
J
Josef Bacik 已提交
1415 1416 1417
		}
	}

1418
	if (entry->bitmap) {
1419 1420
		n = rb_prev(&entry->offset_index);
		if (n) {
1421 1422
			prev = rb_entry(n, struct btrfs_free_space,
					offset_index);
1423 1424 1425
			if (!prev->bitmap &&
			    prev->offset + prev->bytes > offset)
				return prev;
1426
		}
1427
		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1428 1429 1430 1431 1432 1433 1434 1435 1436 1437
			return entry;
	} else if (entry->offset + entry->bytes > offset)
		return entry;

	if (!fuzzy)
		return NULL;

	while (1) {
		if (entry->bitmap) {
			if (entry->offset + BITS_PER_BITMAP *
1438
			    ctl->unit > offset)
1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
				break;
		} else {
			if (entry->offset + entry->bytes > offset)
				break;
		}

		n = rb_next(&entry->offset_index);
		if (!n)
			return NULL;
		entry = rb_entry(n, struct btrfs_free_space, offset_index);
	}
	return entry;
J
Josef Bacik 已提交
1451 1452
}

1453
static inline void
1454
__unlink_free_space(struct btrfs_free_space_ctl *ctl,
1455
		    struct btrfs_free_space *info)
J
Josef Bacik 已提交
1456
{
1457 1458
	rb_erase(&info->offset_index, &ctl->free_space_offset);
	ctl->free_extents--;
1459 1460
}

1461
static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1462 1463
			      struct btrfs_free_space *info)
{
1464 1465
	__unlink_free_space(ctl, info);
	ctl->free_space -= info->bytes;
J
Josef Bacik 已提交
1466 1467
}

1468
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1469 1470 1471 1472
			   struct btrfs_free_space *info)
{
	int ret = 0;

1473
	ASSERT(info->bytes || info->bitmap);
1474
	ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1475
				 &info->offset_index, (info->bitmap != NULL));
J
Josef Bacik 已提交
1476 1477 1478
	if (ret)
		return ret;

1479 1480
	ctl->free_space += info->bytes;
	ctl->free_extents++;
1481 1482 1483
	return ret;
}

1484
static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1485
{
1486
	struct btrfs_block_group_cache *block_group = ctl->private;
1487 1488 1489
	u64 max_bytes;
	u64 bitmap_bytes;
	u64 extent_bytes;
1490
	u64 size = block_group->key.offset;
1491
	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1492 1493
	int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);

1494 1495
	max_bitmaps = max(max_bitmaps, 1);

1496
	ASSERT(ctl->total_bitmaps <= max_bitmaps);
1497 1498 1499 1500 1501 1502

	/*
	 * The goal is to keep the total amount of memory used per 1gb of space
	 * at or below 32k, so we need to adjust how much memory we allow to be
	 * used by extent based free space tracking
	 */
1503 1504 1505 1506 1507
	if (size < 1024 * 1024 * 1024)
		max_bytes = MAX_CACHE_BYTES_PER_GIG;
	else
		max_bytes = MAX_CACHE_BYTES_PER_GIG *
			div64_u64(size, 1024 * 1024 * 1024);
1508

1509 1510 1511 1512 1513
	/*
	 * we want to account for 1 more bitmap than what we have so we can make
	 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
	 * we add more bitmaps.
	 */
1514
	bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1515

1516
	if (bitmap_bytes >= max_bytes) {
1517
		ctl->extents_thresh = 0;
1518 1519
		return;
	}
1520

1521 1522 1523 1524 1525 1526
	/*
	 * we want the extent entry threshold to always be at most 1/2 the maxw
	 * bytes we can have, or whatever is less than that.
	 */
	extent_bytes = max_bytes - bitmap_bytes;
	extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1527

1528
	ctl->extents_thresh =
1529
		div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1530 1531
}

1532 1533 1534
static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
				       struct btrfs_free_space *info,
				       u64 offset, u64 bytes)
1535
{
L
Li Zefan 已提交
1536
	unsigned long start, count;
1537

1538 1539
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
1540
	ASSERT(start + count <= BITS_PER_BITMAP);
1541

L
Li Zefan 已提交
1542
	bitmap_clear(info->bitmap, start, count);
1543 1544

	info->bytes -= bytes;
1545 1546 1547 1548 1549 1550 1551
}

static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info, u64 offset,
			      u64 bytes)
{
	__bitmap_clear_bits(ctl, info, offset, bytes);
1552
	ctl->free_space -= bytes;
1553 1554
}

1555
static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1556 1557
			    struct btrfs_free_space *info, u64 offset,
			    u64 bytes)
1558
{
L
Li Zefan 已提交
1559
	unsigned long start, count;
1560

1561 1562
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
1563
	ASSERT(start + count <= BITS_PER_BITMAP);
1564

L
Li Zefan 已提交
1565
	bitmap_set(info->bitmap, start, count);
1566 1567

	info->bytes += bytes;
1568
	ctl->free_space += bytes;
1569 1570
}

1571 1572 1573 1574
/*
 * If we can not find suitable extent, we will use bytes to record
 * the size of the max extent.
 */
1575
static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1576 1577 1578 1579
			 struct btrfs_free_space *bitmap_info, u64 *offset,
			 u64 *bytes)
{
	unsigned long found_bits = 0;
1580
	unsigned long max_bits = 0;
1581 1582
	unsigned long bits, i;
	unsigned long next_zero;
1583
	unsigned long extent_bits;
1584

1585
	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1586
			  max_t(u64, *offset, bitmap_info->offset));
1587
	bits = bytes_to_bits(*bytes, ctl->unit);
1588

1589
	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1590 1591
		next_zero = find_next_zero_bit(bitmap_info->bitmap,
					       BITS_PER_BITMAP, i);
1592 1593 1594
		extent_bits = next_zero - i;
		if (extent_bits >= bits) {
			found_bits = extent_bits;
1595
			break;
1596 1597
		} else if (extent_bits > max_bits) {
			max_bits = extent_bits;
1598 1599 1600 1601 1602
		}
		i = next_zero;
	}

	if (found_bits) {
1603 1604
		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
		*bytes = (u64)(found_bits) * ctl->unit;
1605 1606 1607
		return 0;
	}

1608
	*bytes = (u64)(max_bits) * ctl->unit;
1609 1610 1611
	return -1;
}

1612
/* Cache the size of the max extent in bytes */
1613
static struct btrfs_free_space *
D
David Woodhouse 已提交
1614
find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1615
		unsigned long align, u64 *max_extent_size)
1616 1617 1618
{
	struct btrfs_free_space *entry;
	struct rb_node *node;
D
David Woodhouse 已提交
1619 1620
	u64 tmp;
	u64 align_off;
1621 1622
	int ret;

1623
	if (!ctl->free_space_offset.rb_node)
1624
		goto out;
1625

1626
	entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1627
	if (!entry)
1628
		goto out;
1629 1630 1631

	for (node = &entry->offset_index; node; node = rb_next(node)) {
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
1632 1633 1634
		if (entry->bytes < *bytes) {
			if (entry->bytes > *max_extent_size)
				*max_extent_size = entry->bytes;
1635
			continue;
1636
		}
1637

D
David Woodhouse 已提交
1638 1639 1640 1641
		/* make sure the space returned is big enough
		 * to match our requested alignment
		 */
		if (*bytes >= align) {
1642
			tmp = entry->offset - ctl->start + align - 1;
D
David Woodhouse 已提交
1643 1644 1645 1646 1647 1648 1649 1650
			do_div(tmp, align);
			tmp = tmp * align + ctl->start;
			align_off = tmp - entry->offset;
		} else {
			align_off = 0;
			tmp = entry->offset;
		}

1651 1652 1653
		if (entry->bytes < *bytes + align_off) {
			if (entry->bytes > *max_extent_size)
				*max_extent_size = entry->bytes;
D
David Woodhouse 已提交
1654
			continue;
1655
		}
D
David Woodhouse 已提交
1656

1657
		if (entry->bitmap) {
1658 1659 1660
			u64 size = *bytes;

			ret = search_bitmap(ctl, entry, &tmp, &size);
D
David Woodhouse 已提交
1661 1662
			if (!ret) {
				*offset = tmp;
1663
				*bytes = size;
1664
				return entry;
1665 1666
			} else if (size > *max_extent_size) {
				*max_extent_size = size;
D
David Woodhouse 已提交
1667
			}
1668 1669 1670
			continue;
		}

D
David Woodhouse 已提交
1671 1672
		*offset = tmp;
		*bytes = entry->bytes - align_off;
1673 1674
		return entry;
	}
1675
out:
1676 1677 1678
	return NULL;
}

1679
static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1680 1681
			   struct btrfs_free_space *info, u64 offset)
{
1682
	info->offset = offset_to_bitmap(ctl, offset);
J
Josef Bacik 已提交
1683
	info->bytes = 0;
1684
	INIT_LIST_HEAD(&info->list);
1685 1686
	link_free_space(ctl, info);
	ctl->total_bitmaps++;
1687

1688
	ctl->op->recalc_thresholds(ctl);
1689 1690
}

1691
static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1692 1693
			struct btrfs_free_space *bitmap_info)
{
1694
	unlink_free_space(ctl, bitmap_info);
1695
	kfree(bitmap_info->bitmap);
1696
	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1697 1698
	ctl->total_bitmaps--;
	ctl->op->recalc_thresholds(ctl);
1699 1700
}

1701
static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1702 1703 1704 1705
			      struct btrfs_free_space *bitmap_info,
			      u64 *offset, u64 *bytes)
{
	u64 end;
1706 1707
	u64 search_start, search_bytes;
	int ret;
1708 1709

again:
1710
	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1711

1712
	/*
1713 1714 1715 1716
	 * We need to search for bits in this bitmap.  We could only cover some
	 * of the extent in this bitmap thanks to how we add space, so we need
	 * to search for as much as it as we can and clear that amount, and then
	 * go searching for the next bit.
1717 1718
	 */
	search_start = *offset;
1719
	search_bytes = ctl->unit;
1720
	search_bytes = min(search_bytes, end - search_start + 1);
1721
	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1722 1723
	if (ret < 0 || search_start != *offset)
		return -EINVAL;
1724

1725 1726 1727 1728 1729 1730 1731 1732 1733
	/* We may have found more bits than what we need */
	search_bytes = min(search_bytes, *bytes);

	/* Cannot clear past the end of the bitmap */
	search_bytes = min(search_bytes, end - search_start + 1);

	bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
	*offset += search_bytes;
	*bytes -= search_bytes;
1734 1735

	if (*bytes) {
1736
		struct rb_node *next = rb_next(&bitmap_info->offset_index);
1737
		if (!bitmap_info->bytes)
1738
			free_bitmap(ctl, bitmap_info);
1739

1740 1741 1742 1743 1744
		/*
		 * no entry after this bitmap, but we still have bytes to
		 * remove, so something has gone wrong.
		 */
		if (!next)
1745 1746
			return -EINVAL;

1747 1748 1749 1750 1751 1752 1753
		bitmap_info = rb_entry(next, struct btrfs_free_space,
				       offset_index);

		/*
		 * if the next entry isn't a bitmap we need to return to let the
		 * extent stuff do its work.
		 */
1754 1755 1756
		if (!bitmap_info->bitmap)
			return -EAGAIN;

1757 1758 1759 1760 1761 1762 1763
		/*
		 * Ok the next item is a bitmap, but it may not actually hold
		 * the information for the rest of this free space stuff, so
		 * look for it, and if we don't find it return so we can try
		 * everything over again.
		 */
		search_start = *offset;
1764
		search_bytes = ctl->unit;
1765
		ret = search_bitmap(ctl, bitmap_info, &search_start,
1766 1767 1768 1769
				    &search_bytes);
		if (ret < 0 || search_start != *offset)
			return -EAGAIN;

1770
		goto again;
1771
	} else if (!bitmap_info->bytes)
1772
		free_bitmap(ctl, bitmap_info);
1773 1774 1775 1776

	return 0;
}

J
Josef Bacik 已提交
1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
			       struct btrfs_free_space *info, u64 offset,
			       u64 bytes)
{
	u64 bytes_to_set = 0;
	u64 end;

	end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);

	bytes_to_set = min(end - offset, bytes);

	bitmap_set_bits(ctl, info, offset, bytes_to_set);

	return bytes_to_set;

}

1794 1795
static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
		      struct btrfs_free_space *info)
1796
{
1797
	struct btrfs_block_group_cache *block_group = ctl->private;
1798 1799 1800 1801 1802

	/*
	 * If we are below the extents threshold then we can add this as an
	 * extent, and don't have to deal with the bitmap
	 */
1803
	if (ctl->free_extents < ctl->extents_thresh) {
1804 1805 1806 1807 1808 1809 1810 1811
		/*
		 * If this block group has some small extents we don't want to
		 * use up all of our free slots in the cache with them, we want
		 * to reserve them to larger extents, however if we have plent
		 * of cache left then go ahead an dadd them, no sense in adding
		 * the overhead of a bitmap if we don't have to.
		 */
		if (info->bytes <= block_group->sectorsize * 4) {
1812 1813
			if (ctl->free_extents * 2 <= ctl->extents_thresh)
				return false;
1814
		} else {
1815
			return false;
1816 1817
		}
	}
1818 1819

	/*
1820 1821 1822 1823 1824 1825
	 * The original block groups from mkfs can be really small, like 8
	 * megabytes, so don't bother with a bitmap for those entries.  However
	 * some block groups can be smaller than what a bitmap would cover but
	 * are still large enough that they could overflow the 32k memory limit,
	 * so allow those block groups to still be allowed to have a bitmap
	 * entry.
1826
	 */
1827
	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1828 1829 1830 1831 1832
		return false;

	return true;
}

J
Josef Bacik 已提交
1833 1834 1835 1836 1837
static struct btrfs_free_space_op free_space_op = {
	.recalc_thresholds	= recalculate_thresholds,
	.use_bitmap		= use_bitmap,
};

1838 1839 1840 1841
static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info)
{
	struct btrfs_free_space *bitmap_info;
J
Josef Bacik 已提交
1842
	struct btrfs_block_group_cache *block_group = NULL;
1843
	int added = 0;
J
Josef Bacik 已提交
1844
	u64 bytes, offset, bytes_added;
1845
	int ret;
1846 1847 1848 1849

	bytes = info->bytes;
	offset = info->offset;

1850 1851 1852
	if (!ctl->op->use_bitmap(ctl, info))
		return 0;

J
Josef Bacik 已提交
1853 1854
	if (ctl->op == &free_space_op)
		block_group = ctl->private;
1855
again:
J
Josef Bacik 已提交
1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872
	/*
	 * Since we link bitmaps right into the cluster we need to see if we
	 * have a cluster here, and if so and it has our bitmap we need to add
	 * the free space to that bitmap.
	 */
	if (block_group && !list_empty(&block_group->cluster_list)) {
		struct btrfs_free_cluster *cluster;
		struct rb_node *node;
		struct btrfs_free_space *entry;

		cluster = list_entry(block_group->cluster_list.next,
				     struct btrfs_free_cluster,
				     block_group_list);
		spin_lock(&cluster->lock);
		node = rb_first(&cluster->root);
		if (!node) {
			spin_unlock(&cluster->lock);
1873
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
1874 1875 1876 1877 1878
		}

		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		if (!entry->bitmap) {
			spin_unlock(&cluster->lock);
1879
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893
		}

		if (entry->offset == offset_to_bitmap(ctl, offset)) {
			bytes_added = add_bytes_to_bitmap(ctl, entry,
							  offset, bytes);
			bytes -= bytes_added;
			offset += bytes_added;
		}
		spin_unlock(&cluster->lock);
		if (!bytes) {
			ret = 1;
			goto out;
		}
	}
1894 1895

no_cluster_bitmap:
1896
	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1897 1898
					 1, 0);
	if (!bitmap_info) {
1899
		ASSERT(added == 0);
1900 1901 1902
		goto new_bitmap;
	}

J
Josef Bacik 已提交
1903 1904 1905 1906
	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
	bytes -= bytes_added;
	offset += bytes_added;
	added = 0;
1907 1908 1909 1910 1911 1912 1913 1914 1915

	if (!bytes) {
		ret = 1;
		goto out;
	} else
		goto again;

new_bitmap:
	if (info && info->bitmap) {
1916
		add_new_bitmap(ctl, info, offset);
1917 1918 1919 1920
		added = 1;
		info = NULL;
		goto again;
	} else {
1921
		spin_unlock(&ctl->tree_lock);
1922 1923 1924

		/* no pre-allocated info, allocate a new one */
		if (!info) {
1925 1926
			info = kmem_cache_zalloc(btrfs_free_space_cachep,
						 GFP_NOFS);
1927
			if (!info) {
1928
				spin_lock(&ctl->tree_lock);
1929 1930 1931 1932 1933 1934 1935
				ret = -ENOMEM;
				goto out;
			}
		}

		/* allocate the bitmap */
		info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1936
		spin_lock(&ctl->tree_lock);
1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947
		if (!info->bitmap) {
			ret = -ENOMEM;
			goto out;
		}
		goto again;
	}

out:
	if (info) {
		if (info->bitmap)
			kfree(info->bitmap);
1948
		kmem_cache_free(btrfs_free_space_cachep, info);
1949
	}
J
Josef Bacik 已提交
1950 1951 1952 1953

	return ret;
}

1954
static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1955
			  struct btrfs_free_space *info, bool update_stat)
J
Josef Bacik 已提交
1956
{
1957 1958 1959 1960 1961
	struct btrfs_free_space *left_info;
	struct btrfs_free_space *right_info;
	bool merged = false;
	u64 offset = info->offset;
	u64 bytes = info->bytes;
1962

J
Josef Bacik 已提交
1963 1964 1965 1966 1967
	/*
	 * first we want to see if there is free space adjacent to the range we
	 * are adding, if there is remove that struct and add a new one to
	 * cover the entire range
	 */
1968
	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1969 1970 1971 1972
	if (right_info && rb_prev(&right_info->offset_index))
		left_info = rb_entry(rb_prev(&right_info->offset_index),
				     struct btrfs_free_space, offset_index);
	else
1973
		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
J
Josef Bacik 已提交
1974

1975
	if (right_info && !right_info->bitmap) {
1976
		if (update_stat)
1977
			unlink_free_space(ctl, right_info);
1978
		else
1979
			__unlink_free_space(ctl, right_info);
1980
		info->bytes += right_info->bytes;
1981
		kmem_cache_free(btrfs_free_space_cachep, right_info);
1982
		merged = true;
J
Josef Bacik 已提交
1983 1984
	}

1985 1986
	if (left_info && !left_info->bitmap &&
	    left_info->offset + left_info->bytes == offset) {
1987
		if (update_stat)
1988
			unlink_free_space(ctl, left_info);
1989
		else
1990
			__unlink_free_space(ctl, left_info);
1991 1992
		info->offset = left_info->offset;
		info->bytes += left_info->bytes;
1993
		kmem_cache_free(btrfs_free_space_cachep, left_info);
1994
		merged = true;
J
Josef Bacik 已提交
1995 1996
	}

1997 1998 1999
	return merged;
}

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121
static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
				     struct btrfs_free_space *info,
				     bool update_stat)
{
	struct btrfs_free_space *bitmap;
	unsigned long i;
	unsigned long j;
	const u64 end = info->offset + info->bytes;
	const u64 bitmap_offset = offset_to_bitmap(ctl, end);
	u64 bytes;

	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
	if (!bitmap)
		return false;

	i = offset_to_bit(bitmap->offset, ctl->unit, end);
	j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
	if (j == i)
		return false;
	bytes = (j - i) * ctl->unit;
	info->bytes += bytes;

	if (update_stat)
		bitmap_clear_bits(ctl, bitmap, end, bytes);
	else
		__bitmap_clear_bits(ctl, bitmap, end, bytes);

	if (!bitmap->bytes)
		free_bitmap(ctl, bitmap);

	return true;
}

static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
				       struct btrfs_free_space *info,
				       bool update_stat)
{
	struct btrfs_free_space *bitmap;
	u64 bitmap_offset;
	unsigned long i;
	unsigned long j;
	unsigned long prev_j;
	u64 bytes;

	bitmap_offset = offset_to_bitmap(ctl, info->offset);
	/* If we're on a boundary, try the previous logical bitmap. */
	if (bitmap_offset == info->offset) {
		if (info->offset == 0)
			return false;
		bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
	}

	bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
	if (!bitmap)
		return false;

	i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
	j = 0;
	prev_j = (unsigned long)-1;
	for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
		if (j > i)
			break;
		prev_j = j;
	}
	if (prev_j == i)
		return false;

	if (prev_j == (unsigned long)-1)
		bytes = (i + 1) * ctl->unit;
	else
		bytes = (i - prev_j) * ctl->unit;

	info->offset -= bytes;
	info->bytes += bytes;

	if (update_stat)
		bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
	else
		__bitmap_clear_bits(ctl, bitmap, info->offset, bytes);

	if (!bitmap->bytes)
		free_bitmap(ctl, bitmap);

	return true;
}

/*
 * We prefer always to allocate from extent entries, both for clustered and
 * non-clustered allocation requests. So when attempting to add a new extent
 * entry, try to see if there's adjacent free space in bitmap entries, and if
 * there is, migrate that space from the bitmaps to the extent.
 * Like this we get better chances of satisfying space allocation requests
 * because we attempt to satisfy them based on a single cache entry, and never
 * on 2 or more entries - even if the entries represent a contiguous free space
 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
 * ends).
 */
static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info,
			      bool update_stat)
{
	/*
	 * Only work with disconnected entries, as we can change their offset,
	 * and must be extent entries.
	 */
	ASSERT(!info->bitmap);
	ASSERT(RB_EMPTY_NODE(&info->offset_index));

	if (ctl->total_bitmaps > 0) {
		bool stole_end;
		bool stole_front = false;

		stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
		if (ctl->total_bitmaps > 0)
			stole_front = steal_from_bitmap_to_front(ctl, info,
								 update_stat);

		if (stole_end || stole_front)
			try_merge_free_space(ctl, info, update_stat);
	}
}

2122 2123
int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
			   u64 offset, u64 bytes)
2124 2125 2126 2127
{
	struct btrfs_free_space *info;
	int ret = 0;

2128
	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2129 2130 2131 2132 2133
	if (!info)
		return -ENOMEM;

	info->offset = offset;
	info->bytes = bytes;
2134
	RB_CLEAR_NODE(&info->offset_index);
2135

2136
	spin_lock(&ctl->tree_lock);
2137

2138
	if (try_merge_free_space(ctl, info, true))
2139 2140 2141 2142 2143 2144 2145
		goto link;

	/*
	 * There was no extent directly to the left or right of this new
	 * extent then we know we're going to have to allocate a new extent, so
	 * before we do that see if we need to drop this into a bitmap
	 */
2146
	ret = insert_into_bitmap(ctl, info);
2147 2148 2149 2150 2151 2152 2153
	if (ret < 0) {
		goto out;
	} else if (ret) {
		ret = 0;
		goto out;
	}
link:
2154 2155 2156 2157 2158 2159 2160 2161
	/*
	 * Only steal free space from adjacent bitmaps if we're sure we're not
	 * going to add the new free space to existing bitmap entries - because
	 * that would mean unnecessary work that would be reverted. Therefore
	 * attempt to steal space from bitmaps if we're adding an extent entry.
	 */
	steal_from_bitmap(ctl, info, true);

2162
	ret = link_free_space(ctl, info);
J
Josef Bacik 已提交
2163
	if (ret)
2164
		kmem_cache_free(btrfs_free_space_cachep, info);
2165
out:
2166
	spin_unlock(&ctl->tree_lock);
2167

J
Josef Bacik 已提交
2168
	if (ret) {
2169
		printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
2170
		ASSERT(ret != -EEXIST);
J
Josef Bacik 已提交
2171 2172 2173 2174 2175
	}

	return ret;
}

2176 2177
int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
			    u64 offset, u64 bytes)
J
Josef Bacik 已提交
2178
{
2179
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
2180
	struct btrfs_free_space *info;
2181 2182
	int ret;
	bool re_search = false;
J
Josef Bacik 已提交
2183

2184
	spin_lock(&ctl->tree_lock);
2185

2186
again:
2187
	ret = 0;
2188 2189 2190
	if (!bytes)
		goto out_lock;

2191
	info = tree_search_offset(ctl, offset, 0, 0);
2192
	if (!info) {
2193 2194 2195 2196
		/*
		 * oops didn't find an extent that matched the space we wanted
		 * to remove, look for a bitmap instead
		 */
2197
		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2198 2199
					  1, 0);
		if (!info) {
2200 2201 2202 2203
			/*
			 * If we found a partial bit of our free space in a
			 * bitmap but then couldn't find the other part this may
			 * be a problem, so WARN about it.
2204
			 */
2205
			WARN_ON(re_search);
2206 2207
			goto out_lock;
		}
2208 2209
	}

2210
	re_search = false;
2211
	if (!info->bitmap) {
2212
		unlink_free_space(ctl, info);
2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
		if (offset == info->offset) {
			u64 to_free = min(bytes, info->bytes);

			info->bytes -= to_free;
			info->offset += to_free;
			if (info->bytes) {
				ret = link_free_space(ctl, info);
				WARN_ON(ret);
			} else {
				kmem_cache_free(btrfs_free_space_cachep, info);
			}
J
Josef Bacik 已提交
2224

2225 2226 2227 2228 2229
			offset += to_free;
			bytes -= to_free;
			goto again;
		} else {
			u64 old_end = info->bytes + info->offset;
2230

2231
			info->bytes = offset - info->offset;
2232
			ret = link_free_space(ctl, info);
2233 2234 2235 2236
			WARN_ON(ret);
			if (ret)
				goto out_lock;

2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252
			/* Not enough bytes in this entry to satisfy us */
			if (old_end < offset + bytes) {
				bytes -= old_end - offset;
				offset = old_end;
				goto again;
			} else if (old_end == offset + bytes) {
				/* all done */
				goto out_lock;
			}
			spin_unlock(&ctl->tree_lock);

			ret = btrfs_add_free_space(block_group, offset + bytes,
						   old_end - (offset + bytes));
			WARN_ON(ret);
			goto out;
		}
J
Josef Bacik 已提交
2253
	}
2254

2255
	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2256 2257
	if (ret == -EAGAIN) {
		re_search = true;
2258
		goto again;
2259
	}
2260
out_lock:
2261
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
2262
out:
2263 2264 2265
	return ret;
}

J
Josef Bacik 已提交
2266 2267 2268
void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
			   u64 bytes)
{
2269
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
2270 2271 2272 2273
	struct btrfs_free_space *info;
	struct rb_node *n;
	int count = 0;

2274
	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
J
Josef Bacik 已提交
2275
		info = rb_entry(n, struct btrfs_free_space, offset_index);
L
Liu Bo 已提交
2276
		if (info->bytes >= bytes && !block_group->ro)
J
Josef Bacik 已提交
2277
			count++;
2278 2279 2280
		btrfs_crit(block_group->fs_info,
			   "entry offset %llu, bytes %llu, bitmap %s",
			   info->offset, info->bytes,
2281
		       (info->bitmap) ? "yes" : "no");
J
Josef Bacik 已提交
2282
	}
2283
	btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2284
	       list_empty(&block_group->cluster_list) ? "no" : "yes");
2285 2286
	btrfs_info(block_group->fs_info,
		   "%d blocks of free space at or bigger than bytes is", count);
J
Josef Bacik 已提交
2287 2288
}

2289
void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
J
Josef Bacik 已提交
2290
{
2291
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
2292

2293 2294 2295 2296 2297
	spin_lock_init(&ctl->tree_lock);
	ctl->unit = block_group->sectorsize;
	ctl->start = block_group->key.objectid;
	ctl->private = block_group;
	ctl->op = &free_space_op;
J
Josef Bacik 已提交
2298

2299 2300 2301 2302 2303 2304 2305
	/*
	 * we only want to have 32k of ram per block group for keeping
	 * track of free space, and if we pass 1/2 of that we want to
	 * start converting things over to using bitmaps
	 */
	ctl->extents_thresh = ((1024 * 32) / 2) /
				sizeof(struct btrfs_free_space);
J
Josef Bacik 已提交
2306 2307
}

2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318
/*
 * for a given cluster, put all of its extents back into the free
 * space cache.  If the block group passed doesn't match the block group
 * pointed to by the cluster, someone else raced in and freed the
 * cluster already.  In that case, we just return without changing anything
 */
static int
__btrfs_return_cluster_to_free_space(
			     struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster)
{
2319
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2320 2321 2322 2323 2324 2325 2326
	struct btrfs_free_space *entry;
	struct rb_node *node;

	spin_lock(&cluster->lock);
	if (cluster->block_group != block_group)
		goto out;

2327
	cluster->block_group = NULL;
2328
	cluster->window_start = 0;
2329 2330
	list_del_init(&cluster->block_group_list);

2331
	node = rb_first(&cluster->root);
2332
	while (node) {
2333 2334
		bool bitmap;

2335 2336 2337
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		node = rb_next(&entry->offset_index);
		rb_erase(&entry->offset_index, &cluster->root);
2338
		RB_CLEAR_NODE(&entry->offset_index);
2339 2340

		bitmap = (entry->bitmap != NULL);
2341
		if (!bitmap) {
2342
			try_merge_free_space(ctl, entry, false);
2343 2344
			steal_from_bitmap(ctl, entry, false);
		}
2345
		tree_insert_offset(&ctl->free_space_offset,
2346
				   entry->offset, &entry->offset_index, bitmap);
2347
	}
2348
	cluster->root = RB_ROOT;
2349

2350 2351
out:
	spin_unlock(&cluster->lock);
2352
	btrfs_put_block_group(block_group);
2353 2354 2355
	return 0;
}

2356 2357
static void __btrfs_remove_free_space_cache_locked(
				struct btrfs_free_space_ctl *ctl)
J
Josef Bacik 已提交
2358 2359 2360
{
	struct btrfs_free_space *info;
	struct rb_node *node;
2361 2362 2363

	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
		info = rb_entry(node, struct btrfs_free_space, offset_index);
2364 2365 2366 2367 2368 2369
		if (!info->bitmap) {
			unlink_free_space(ctl, info);
			kmem_cache_free(btrfs_free_space_cachep, info);
		} else {
			free_bitmap(ctl, info);
		}
2370 2371 2372 2373 2374 2375
		if (need_resched()) {
			spin_unlock(&ctl->tree_lock);
			cond_resched();
			spin_lock(&ctl->tree_lock);
		}
	}
2376 2377 2378 2379 2380 2381
}

void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
{
	spin_lock(&ctl->tree_lock);
	__btrfs_remove_free_space_cache_locked(ctl);
2382 2383 2384 2385 2386 2387
	spin_unlock(&ctl->tree_lock);
}

void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2388
	struct btrfs_free_cluster *cluster;
2389
	struct list_head *head;
J
Josef Bacik 已提交
2390

2391
	spin_lock(&ctl->tree_lock);
2392 2393 2394 2395
	while ((head = block_group->cluster_list.next) !=
	       &block_group->cluster_list) {
		cluster = list_entry(head, struct btrfs_free_cluster,
				     block_group_list);
2396 2397 2398

		WARN_ON(cluster->block_group != block_group);
		__btrfs_return_cluster_to_free_space(block_group, cluster);
2399
		if (need_resched()) {
2400
			spin_unlock(&ctl->tree_lock);
2401
			cond_resched();
2402
			spin_lock(&ctl->tree_lock);
2403
		}
2404
	}
2405
	__btrfs_remove_free_space_cache_locked(ctl);
2406
	spin_unlock(&ctl->tree_lock);
2407

J
Josef Bacik 已提交
2408 2409
}

2410
u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2411 2412
			       u64 offset, u64 bytes, u64 empty_size,
			       u64 *max_extent_size)
J
Josef Bacik 已提交
2413
{
2414
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2415
	struct btrfs_free_space *entry = NULL;
2416
	u64 bytes_search = bytes + empty_size;
2417
	u64 ret = 0;
D
David Woodhouse 已提交
2418 2419
	u64 align_gap = 0;
	u64 align_gap_len = 0;
J
Josef Bacik 已提交
2420

2421
	spin_lock(&ctl->tree_lock);
D
David Woodhouse 已提交
2422
	entry = find_free_space(ctl, &offset, &bytes_search,
2423
				block_group->full_stripe_len, max_extent_size);
2424
	if (!entry)
2425 2426 2427 2428
		goto out;

	ret = offset;
	if (entry->bitmap) {
2429
		bitmap_clear_bits(ctl, entry, offset, bytes);
2430
		if (!entry->bytes)
2431
			free_bitmap(ctl, entry);
2432
	} else {
2433
		unlink_free_space(ctl, entry);
D
David Woodhouse 已提交
2434 2435 2436 2437 2438 2439 2440
		align_gap_len = offset - entry->offset;
		align_gap = entry->offset;

		entry->offset = offset + bytes;
		WARN_ON(entry->bytes < bytes + align_gap_len);

		entry->bytes -= bytes + align_gap_len;
2441
		if (!entry->bytes)
2442
			kmem_cache_free(btrfs_free_space_cachep, entry);
2443
		else
2444
			link_free_space(ctl, entry);
2445
	}
2446
out:
2447
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
2448

D
David Woodhouse 已提交
2449 2450
	if (align_gap_len)
		__btrfs_add_free_space(ctl, align_gap, align_gap_len);
J
Josef Bacik 已提交
2451 2452
	return ret;
}
2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465

/*
 * given a cluster, put all of its extents back into the free space
 * cache.  If a block group is passed, this function will only free
 * a cluster that belongs to the passed block group.
 *
 * Otherwise, it'll get a reference on the block group pointed to by the
 * cluster and remove the cluster from it.
 */
int btrfs_return_cluster_to_free_space(
			       struct btrfs_block_group_cache *block_group,
			       struct btrfs_free_cluster *cluster)
{
2466
	struct btrfs_free_space_ctl *ctl;
2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484
	int ret;

	/* first, get a safe pointer to the block group */
	spin_lock(&cluster->lock);
	if (!block_group) {
		block_group = cluster->block_group;
		if (!block_group) {
			spin_unlock(&cluster->lock);
			return 0;
		}
	} else if (cluster->block_group != block_group) {
		/* someone else has already freed it don't redo their work */
		spin_unlock(&cluster->lock);
		return 0;
	}
	atomic_inc(&block_group->count);
	spin_unlock(&cluster->lock);

2485 2486
	ctl = block_group->free_space_ctl;

2487
	/* now return any extents the cluster had on it */
2488
	spin_lock(&ctl->tree_lock);
2489
	ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2490
	spin_unlock(&ctl->tree_lock);
2491 2492 2493 2494 2495 2496

	/* finally drop our ref */
	btrfs_put_block_group(block_group);
	return ret;
}

2497 2498
static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
				   struct btrfs_free_cluster *cluster,
2499
				   struct btrfs_free_space *entry,
2500 2501
				   u64 bytes, u64 min_start,
				   u64 *max_extent_size)
2502
{
2503
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2504 2505 2506 2507 2508 2509 2510 2511
	int err;
	u64 search_start = cluster->window_start;
	u64 search_bytes = bytes;
	u64 ret = 0;

	search_start = min_start;
	search_bytes = bytes;

2512
	err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2513 2514 2515
	if (err) {
		if (search_bytes > *max_extent_size)
			*max_extent_size = search_bytes;
2516
		return 0;
2517
	}
2518 2519

	ret = search_start;
2520
	__bitmap_clear_bits(ctl, entry, ret, bytes);
2521 2522 2523 2524

	return ret;
}

2525 2526 2527 2528 2529 2530 2531
/*
 * given a cluster, try to allocate 'bytes' from it, returns 0
 * if it couldn't find anything suitably large, or a logical disk offset
 * if things worked out
 */
u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster, u64 bytes,
2532
			     u64 min_start, u64 *max_extent_size)
2533
{
2534
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550
	struct btrfs_free_space *entry = NULL;
	struct rb_node *node;
	u64 ret = 0;

	spin_lock(&cluster->lock);
	if (bytes > cluster->max_size)
		goto out;

	if (cluster->block_group != block_group)
		goto out;

	node = rb_first(&cluster->root);
	if (!node)
		goto out;

	entry = rb_entry(node, struct btrfs_free_space, offset_index);
2551
	while (1) {
2552 2553 2554
		if (entry->bytes < bytes && entry->bytes > *max_extent_size)
			*max_extent_size = entry->bytes;

2555 2556
		if (entry->bytes < bytes ||
		    (!entry->bitmap && entry->offset < min_start)) {
2557 2558 2559 2560 2561 2562 2563 2564
			node = rb_next(&entry->offset_index);
			if (!node)
				break;
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
			continue;
		}

2565 2566 2567
		if (entry->bitmap) {
			ret = btrfs_alloc_from_bitmap(block_group,
						      cluster, entry, bytes,
2568 2569
						      cluster->window_start,
						      max_extent_size);
2570 2571 2572 2573 2574 2575 2576 2577
			if (ret == 0) {
				node = rb_next(&entry->offset_index);
				if (!node)
					break;
				entry = rb_entry(node, struct btrfs_free_space,
						 offset_index);
				continue;
			}
2578
			cluster->window_start += bytes;
2579 2580 2581 2582 2583 2584
		} else {
			ret = entry->offset;

			entry->offset += bytes;
			entry->bytes -= bytes;
		}
2585

2586
		if (entry->bytes == 0)
2587 2588 2589 2590 2591
			rb_erase(&entry->offset_index, &cluster->root);
		break;
	}
out:
	spin_unlock(&cluster->lock);
2592

2593 2594 2595
	if (!ret)
		return 0;

2596
	spin_lock(&ctl->tree_lock);
2597

2598
	ctl->free_space -= bytes;
2599
	if (entry->bytes == 0) {
2600
		ctl->free_extents--;
2601 2602
		if (entry->bitmap) {
			kfree(entry->bitmap);
2603 2604
			ctl->total_bitmaps--;
			ctl->op->recalc_thresholds(ctl);
2605
		}
2606
		kmem_cache_free(btrfs_free_space_cachep, entry);
2607 2608
	}

2609
	spin_unlock(&ctl->tree_lock);
2610

2611 2612 2613
	return ret;
}

2614 2615 2616
static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
				struct btrfs_free_space *entry,
				struct btrfs_free_cluster *cluster,
2617 2618
				u64 offset, u64 bytes,
				u64 cont1_bytes, u64 min_bytes)
2619
{
2620
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2621 2622
	unsigned long next_zero;
	unsigned long i;
2623 2624
	unsigned long want_bits;
	unsigned long min_bits;
2625 2626 2627
	unsigned long found_bits;
	unsigned long start = 0;
	unsigned long total_found = 0;
2628
	int ret;
2629

2630
	i = offset_to_bit(entry->offset, ctl->unit,
2631
			  max_t(u64, offset, entry->offset));
2632 2633
	want_bits = bytes_to_bits(bytes, ctl->unit);
	min_bits = bytes_to_bits(min_bytes, ctl->unit);
2634 2635 2636

again:
	found_bits = 0;
2637
	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2638 2639
		next_zero = find_next_zero_bit(entry->bitmap,
					       BITS_PER_BITMAP, i);
2640
		if (next_zero - i >= min_bits) {
2641 2642 2643 2644 2645 2646 2647
			found_bits = next_zero - i;
			break;
		}
		i = next_zero;
	}

	if (!found_bits)
2648
		return -ENOSPC;
2649

2650
	if (!total_found) {
2651
		start = i;
2652
		cluster->max_size = 0;
2653 2654 2655 2656
	}

	total_found += found_bits;

2657 2658
	if (cluster->max_size < found_bits * ctl->unit)
		cluster->max_size = found_bits * ctl->unit;
2659

2660 2661
	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
		i = next_zero + 1;
2662 2663 2664
		goto again;
	}

2665
	cluster->window_start = start * ctl->unit + entry->offset;
2666
	rb_erase(&entry->offset_index, &ctl->free_space_offset);
2667 2668
	ret = tree_insert_offset(&cluster->root, entry->offset,
				 &entry->offset_index, 1);
2669
	ASSERT(!ret); /* -EEXIST; Logic error */
2670

J
Josef Bacik 已提交
2671
	trace_btrfs_setup_cluster(block_group, cluster,
2672
				  total_found * ctl->unit, 1);
2673 2674 2675
	return 0;
}

2676 2677
/*
 * This searches the block group for just extents to fill the cluster with.
2678 2679
 * Try to find a cluster with at least bytes total bytes, at least one
 * extent of cont1_bytes, and other clusters of at least min_bytes.
2680
 */
2681 2682 2683 2684
static noinline int
setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
			struct btrfs_free_cluster *cluster,
			struct list_head *bitmaps, u64 offset, u64 bytes,
2685
			u64 cont1_bytes, u64 min_bytes)
2686
{
2687
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2688 2689 2690 2691 2692 2693
	struct btrfs_free_space *first = NULL;
	struct btrfs_free_space *entry = NULL;
	struct btrfs_free_space *last;
	struct rb_node *node;
	u64 window_free;
	u64 max_extent;
J
Josef Bacik 已提交
2694
	u64 total_size = 0;
2695

2696
	entry = tree_search_offset(ctl, offset, 0, 1);
2697 2698 2699 2700 2701 2702 2703
	if (!entry)
		return -ENOSPC;

	/*
	 * We don't want bitmaps, so just move along until we find a normal
	 * extent entry.
	 */
2704 2705
	while (entry->bitmap || entry->bytes < min_bytes) {
		if (entry->bitmap && list_empty(&entry->list))
2706
			list_add_tail(&entry->list, bitmaps);
2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717
		node = rb_next(&entry->offset_index);
		if (!node)
			return -ENOSPC;
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
	}

	window_free = entry->bytes;
	max_extent = entry->bytes;
	first = entry;
	last = entry;

2718 2719
	for (node = rb_next(&entry->offset_index); node;
	     node = rb_next(&entry->offset_index)) {
2720 2721
		entry = rb_entry(node, struct btrfs_free_space, offset_index);

2722 2723 2724
		if (entry->bitmap) {
			if (list_empty(&entry->list))
				list_add_tail(&entry->list, bitmaps);
2725
			continue;
2726 2727
		}

2728 2729 2730 2731 2732 2733
		if (entry->bytes < min_bytes)
			continue;

		last = entry;
		window_free += entry->bytes;
		if (entry->bytes > max_extent)
2734 2735 2736
			max_extent = entry->bytes;
	}

2737 2738 2739
	if (window_free < bytes || max_extent < cont1_bytes)
		return -ENOSPC;

2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752
	cluster->window_start = first->offset;

	node = &first->offset_index;

	/*
	 * now we've found our entries, pull them out of the free space
	 * cache and put them into the cluster rbtree
	 */
	do {
		int ret;

		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		node = rb_next(&entry->offset_index);
2753
		if (entry->bitmap || entry->bytes < min_bytes)
2754 2755
			continue;

2756
		rb_erase(&entry->offset_index, &ctl->free_space_offset);
2757 2758
		ret = tree_insert_offset(&cluster->root, entry->offset,
					 &entry->offset_index, 0);
J
Josef Bacik 已提交
2759
		total_size += entry->bytes;
2760
		ASSERT(!ret); /* -EEXIST; Logic error */
2761 2762 2763
	} while (node && entry != last);

	cluster->max_size = max_extent;
J
Josef Bacik 已提交
2764
	trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2765 2766 2767 2768 2769 2770 2771
	return 0;
}

/*
 * This specifically looks for bitmaps that may work in the cluster, we assume
 * that we have already failed to find extents that will work.
 */
2772 2773 2774 2775
static noinline int
setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
		     struct btrfs_free_cluster *cluster,
		     struct list_head *bitmaps, u64 offset, u64 bytes,
2776
		     u64 cont1_bytes, u64 min_bytes)
2777
{
2778
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2779 2780
	struct btrfs_free_space *entry;
	int ret = -ENOSPC;
2781
	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2782

2783
	if (ctl->total_bitmaps == 0)
2784 2785
		return -ENOSPC;

2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796
	/*
	 * The bitmap that covers offset won't be in the list unless offset
	 * is just its start offset.
	 */
	entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
	if (entry->offset != bitmap_offset) {
		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
		if (entry && list_empty(&entry->list))
			list_add(&entry->list, bitmaps);
	}

2797
	list_for_each_entry(entry, bitmaps, list) {
2798
		if (entry->bytes < bytes)
2799 2800
			continue;
		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2801
					   bytes, cont1_bytes, min_bytes);
2802 2803 2804 2805 2806
		if (!ret)
			return 0;
	}

	/*
2807 2808
	 * The bitmaps list has all the bitmaps that record free space
	 * starting after offset, so no more search is required.
2809
	 */
2810
	return -ENOSPC;
2811 2812
}

2813 2814
/*
 * here we try to find a cluster of blocks in a block group.  The goal
2815
 * is to find at least bytes+empty_size.
2816 2817 2818 2819 2820
 * We might not find them all in one contiguous area.
 *
 * returns zero and sets up cluster if things worked out, otherwise
 * it returns -enospc
 */
2821
int btrfs_find_space_cluster(struct btrfs_root *root,
2822 2823 2824 2825
			     struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster,
			     u64 offset, u64 bytes, u64 empty_size)
{
2826
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2827
	struct btrfs_free_space *entry, *tmp;
2828
	LIST_HEAD(bitmaps);
2829
	u64 min_bytes;
2830
	u64 cont1_bytes;
2831 2832
	int ret;

2833 2834 2835 2836 2837 2838
	/*
	 * Choose the minimum extent size we'll require for this
	 * cluster.  For SSD_SPREAD, don't allow any fragmentation.
	 * For metadata, allow allocates with smaller extents.  For
	 * data, keep it dense.
	 */
2839
	if (btrfs_test_opt(root, SSD_SPREAD)) {
2840
		cont1_bytes = min_bytes = bytes + empty_size;
2841
	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2842 2843 2844 2845 2846 2847
		cont1_bytes = bytes;
		min_bytes = block_group->sectorsize;
	} else {
		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
		min_bytes = block_group->sectorsize;
	}
2848

2849
	spin_lock(&ctl->tree_lock);
2850 2851 2852 2853 2854

	/*
	 * If we know we don't have enough space to make a cluster don't even
	 * bother doing all the work to try and find one.
	 */
2855
	if (ctl->free_space < bytes) {
2856
		spin_unlock(&ctl->tree_lock);
2857 2858 2859
		return -ENOSPC;
	}

2860 2861 2862 2863 2864 2865 2866 2867
	spin_lock(&cluster->lock);

	/* someone already found a cluster, hooray */
	if (cluster->block_group) {
		ret = 0;
		goto out;
	}

J
Josef Bacik 已提交
2868 2869 2870 2871
	trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
				 min_bytes);

	INIT_LIST_HEAD(&bitmaps);
2872
	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2873 2874
				      bytes + empty_size,
				      cont1_bytes, min_bytes);
2875
	if (ret)
2876
		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2877 2878
					   offset, bytes + empty_size,
					   cont1_bytes, min_bytes);
2879 2880 2881 2882

	/* Clear our temporary list */
	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
		list_del_init(&entry->list);
2883

2884 2885 2886 2887 2888
	if (!ret) {
		atomic_inc(&block_group->count);
		list_add_tail(&cluster->block_group_list,
			      &block_group->cluster_list);
		cluster->block_group = block_group;
J
Josef Bacik 已提交
2889 2890
	} else {
		trace_btrfs_failed_cluster_setup(block_group);
2891 2892 2893
	}
out:
	spin_unlock(&cluster->lock);
2894
	spin_unlock(&ctl->tree_lock);
2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905

	return ret;
}

/*
 * simple code to zero out a cluster
 */
void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
{
	spin_lock_init(&cluster->lock);
	spin_lock_init(&cluster->refill_lock);
2906
	cluster->root = RB_ROOT;
2907 2908 2909 2910 2911
	cluster->max_size = 0;
	INIT_LIST_HEAD(&cluster->block_group_list);
	cluster->block_group = NULL;
}

2912 2913 2914
static int do_trimming(struct btrfs_block_group_cache *block_group,
		       u64 *total_trimmed, u64 start, u64 bytes,
		       u64 reserved_start, u64 reserved_bytes)
2915
{
2916
	struct btrfs_space_info *space_info = block_group->space_info;
2917
	struct btrfs_fs_info *fs_info = block_group->fs_info;
2918 2919 2920
	int ret;
	int update = 0;
	u64 trimmed = 0;
2921

2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962
	spin_lock(&space_info->lock);
	spin_lock(&block_group->lock);
	if (!block_group->ro) {
		block_group->reserved += reserved_bytes;
		space_info->bytes_reserved += reserved_bytes;
		update = 1;
	}
	spin_unlock(&block_group->lock);
	spin_unlock(&space_info->lock);

	ret = btrfs_error_discard_extent(fs_info->extent_root,
					 start, bytes, &trimmed);
	if (!ret)
		*total_trimmed += trimmed;

	btrfs_add_free_space(block_group, reserved_start, reserved_bytes);

	if (update) {
		spin_lock(&space_info->lock);
		spin_lock(&block_group->lock);
		if (block_group->ro)
			space_info->bytes_readonly += reserved_bytes;
		block_group->reserved -= reserved_bytes;
		space_info->bytes_reserved -= reserved_bytes;
		spin_unlock(&space_info->lock);
		spin_unlock(&block_group->lock);
	}

	return ret;
}

static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
			  u64 *total_trimmed, u64 start, u64 end, u64 minlen)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct btrfs_free_space *entry;
	struct rb_node *node;
	int ret = 0;
	u64 extent_start;
	u64 extent_bytes;
	u64 bytes;
2963 2964

	while (start < end) {
2965
		spin_lock(&ctl->tree_lock);
2966

2967 2968
		if (ctl->free_space < minlen) {
			spin_unlock(&ctl->tree_lock);
2969 2970 2971
			break;
		}

2972
		entry = tree_search_offset(ctl, start, 0, 1);
2973
		if (!entry) {
2974
			spin_unlock(&ctl->tree_lock);
2975 2976 2977
			break;
		}

2978 2979 2980 2981
		/* skip bitmaps */
		while (entry->bitmap) {
			node = rb_next(&entry->offset_index);
			if (!node) {
2982
				spin_unlock(&ctl->tree_lock);
2983
				goto out;
2984
			}
2985 2986
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
2987 2988
		}

2989 2990 2991
		if (entry->offset >= end) {
			spin_unlock(&ctl->tree_lock);
			break;
2992 2993
		}

2994 2995 2996 2997 2998 2999 3000
		extent_start = entry->offset;
		extent_bytes = entry->bytes;
		start = max(start, extent_start);
		bytes = min(extent_start + extent_bytes, end) - start;
		if (bytes < minlen) {
			spin_unlock(&ctl->tree_lock);
			goto next;
3001 3002
		}

3003 3004 3005
		unlink_free_space(ctl, entry);
		kmem_cache_free(btrfs_free_space_cachep, entry);

3006
		spin_unlock(&ctl->tree_lock);
3007

3008 3009 3010 3011 3012 3013
		ret = do_trimming(block_group, total_trimmed, start, bytes,
				  extent_start, extent_bytes);
		if (ret)
			break;
next:
		start += bytes;
3014

3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083
		if (fatal_signal_pending(current)) {
			ret = -ERESTARTSYS;
			break;
		}

		cond_resched();
	}
out:
	return ret;
}

static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
			u64 *total_trimmed, u64 start, u64 end, u64 minlen)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct btrfs_free_space *entry;
	int ret = 0;
	int ret2;
	u64 bytes;
	u64 offset = offset_to_bitmap(ctl, start);

	while (offset < end) {
		bool next_bitmap = false;

		spin_lock(&ctl->tree_lock);

		if (ctl->free_space < minlen) {
			spin_unlock(&ctl->tree_lock);
			break;
		}

		entry = tree_search_offset(ctl, offset, 1, 0);
		if (!entry) {
			spin_unlock(&ctl->tree_lock);
			next_bitmap = true;
			goto next;
		}

		bytes = minlen;
		ret2 = search_bitmap(ctl, entry, &start, &bytes);
		if (ret2 || start >= end) {
			spin_unlock(&ctl->tree_lock);
			next_bitmap = true;
			goto next;
		}

		bytes = min(bytes, end - start);
		if (bytes < minlen) {
			spin_unlock(&ctl->tree_lock);
			goto next;
		}

		bitmap_clear_bits(ctl, entry, start, bytes);
		if (entry->bytes == 0)
			free_bitmap(ctl, entry);

		spin_unlock(&ctl->tree_lock);

		ret = do_trimming(block_group, total_trimmed, start, bytes,
				  start, bytes);
		if (ret)
			break;
next:
		if (next_bitmap) {
			offset += BITS_PER_BITMAP * ctl->unit;
		} else {
			start += bytes;
			if (start >= offset + BITS_PER_BITMAP * ctl->unit)
				offset += BITS_PER_BITMAP * ctl->unit;
3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095
		}

		if (fatal_signal_pending(current)) {
			ret = -ERESTARTSYS;
			break;
		}

		cond_resched();
	}

	return ret;
}
3096

3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112
int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
			   u64 *trimmed, u64 start, u64 end, u64 minlen)
{
	int ret;

	*trimmed = 0;

	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
	if (ret)
		return ret;

	ret = trim_bitmaps(block_group, trimmed, start, end, minlen);

	return ret;
}

3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149
/*
 * Find the left-most item in the cache tree, and then return the
 * smallest inode number in the item.
 *
 * Note: the returned inode number may not be the smallest one in
 * the tree, if the left-most item is a bitmap.
 */
u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
{
	struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
	struct btrfs_free_space *entry = NULL;
	u64 ino = 0;

	spin_lock(&ctl->tree_lock);

	if (RB_EMPTY_ROOT(&ctl->free_space_offset))
		goto out;

	entry = rb_entry(rb_first(&ctl->free_space_offset),
			 struct btrfs_free_space, offset_index);

	if (!entry->bitmap) {
		ino = entry->offset;

		unlink_free_space(ctl, entry);
		entry->offset++;
		entry->bytes--;
		if (!entry->bytes)
			kmem_cache_free(btrfs_free_space_cachep, entry);
		else
			link_free_space(ctl, entry);
	} else {
		u64 offset = 0;
		u64 count = 1;
		int ret;

		ret = search_bitmap(ctl, entry, &offset, &count);
3150
		/* Logic error; Should be empty if it can't find anything */
3151
		ASSERT(!ret);
3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162

		ino = offset;
		bitmap_clear_bits(ctl, entry, offset, 1);
		if (entry->bytes == 0)
			free_bitmap(ctl, entry);
	}
out:
	spin_unlock(&ctl->tree_lock);

	return ino;
}
3163 3164 3165 3166 3167 3168

struct inode *lookup_free_ino_inode(struct btrfs_root *root,
				    struct btrfs_path *path)
{
	struct inode *inode = NULL;

3169 3170 3171 3172
	spin_lock(&root->ino_cache_lock);
	if (root->ino_cache_inode)
		inode = igrab(root->ino_cache_inode);
	spin_unlock(&root->ino_cache_lock);
3173 3174 3175 3176 3177 3178 3179
	if (inode)
		return inode;

	inode = __lookup_free_space_inode(root, path, 0);
	if (IS_ERR(inode))
		return inode;

3180
	spin_lock(&root->ino_cache_lock);
3181
	if (!btrfs_fs_closing(root->fs_info))
3182 3183
		root->ino_cache_inode = igrab(inode);
	spin_unlock(&root->ino_cache_lock);
3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203

	return inode;
}

int create_free_ino_inode(struct btrfs_root *root,
			  struct btrfs_trans_handle *trans,
			  struct btrfs_path *path)
{
	return __create_free_space_inode(root, trans, path,
					 BTRFS_FREE_INO_OBJECTID, 0);
}

int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_path *path;
	struct inode *inode;
	int ret = 0;
	u64 root_gen = btrfs_root_generation(&root->root_item);

C
Chris Mason 已提交
3204 3205 3206
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

3207 3208 3209 3210
	/*
	 * If we're unmounting then just return, since this does a search on the
	 * normal root and not the commit root and we could deadlock.
	 */
3211
	if (btrfs_fs_closing(fs_info))
3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227
		return 0;

	path = btrfs_alloc_path();
	if (!path)
		return 0;

	inode = lookup_free_ino_inode(root, path);
	if (IS_ERR(inode))
		goto out;

	if (root_gen != BTRFS_I(inode)->generation)
		goto out_put;

	ret = __load_free_space_cache(root, inode, ctl, path, 0);

	if (ret < 0)
3228 3229 3230
		btrfs_err(fs_info,
			"failed to load free ino cache for root %llu",
			root->root_key.objectid);
3231 3232 3233 3234 3235 3236 3237 3238 3239
out_put:
	iput(inode);
out:
	btrfs_free_path(path);
	return ret;
}

int btrfs_write_out_ino_cache(struct btrfs_root *root,
			      struct btrfs_trans_handle *trans,
3240 3241
			      struct btrfs_path *path,
			      struct inode *inode)
3242 3243 3244 3245
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	int ret;

C
Chris Mason 已提交
3246 3247 3248
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

3249
	ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
3250 3251 3252
	if (ret) {
		btrfs_delalloc_release_metadata(inode, inode->i_size);
#ifdef DEBUG
3253 3254 3255
		btrfs_err(root->fs_info,
			"failed to write free ino cache for root %llu",
			root->root_key.objectid);
3256 3257
#endif
	}
3258 3259 3260

	return ret;
}
3261 3262

#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3263 3264 3265 3266 3267 3268 3269 3270
/*
 * Use this if you need to make a bitmap or extent entry specifically, it
 * doesn't do any of the merging that add_free_space does, this acts a lot like
 * how the free space cache loading stuff works, so you can get really weird
 * configurations.
 */
int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
			      u64 offset, u64 bytes, bool bitmap)
3271
{
3272 3273 3274 3275 3276
	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
	struct btrfs_free_space *info = NULL, *bitmap_info;
	void *map = NULL;
	u64 bytes_added;
	int ret;
3277

3278 3279 3280 3281 3282
again:
	if (!info) {
		info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
		if (!info)
			return -ENOMEM;
3283 3284
	}

3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311
	if (!bitmap) {
		spin_lock(&ctl->tree_lock);
		info->offset = offset;
		info->bytes = bytes;
		ret = link_free_space(ctl, info);
		spin_unlock(&ctl->tree_lock);
		if (ret)
			kmem_cache_free(btrfs_free_space_cachep, info);
		return ret;
	}

	if (!map) {
		map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
		if (!map) {
			kmem_cache_free(btrfs_free_space_cachep, info);
			return -ENOMEM;
		}
	}

	spin_lock(&ctl->tree_lock);
	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
					 1, 0);
	if (!bitmap_info) {
		info->bitmap = map;
		map = NULL;
		add_new_bitmap(ctl, info, offset);
		bitmap_info = info;
3312
		info = NULL;
3313
	}
3314

3315 3316 3317 3318
	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
	bytes -= bytes_added;
	offset += bytes_added;
	spin_unlock(&ctl->tree_lock);
3319

3320 3321
	if (bytes)
		goto again;
3322

3323 3324
	if (info)
		kmem_cache_free(btrfs_free_space_cachep, info);
3325 3326 3327
	if (map)
		kfree(map);
	return 0;
3328 3329 3330 3331 3332 3333 3334
}

/*
 * Checks to see if the given range is in the free space cache.  This is really
 * just used to check the absence of space, so if there is free space in the
 * range at all we will return 1.
 */
3335 3336
int test_check_exists(struct btrfs_block_group_cache *cache,
		      u64 offset, u64 bytes)
3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398
{
	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
	struct btrfs_free_space *info;
	int ret = 0;

	spin_lock(&ctl->tree_lock);
	info = tree_search_offset(ctl, offset, 0, 0);
	if (!info) {
		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
					  1, 0);
		if (!info)
			goto out;
	}

have_info:
	if (info->bitmap) {
		u64 bit_off, bit_bytes;
		struct rb_node *n;
		struct btrfs_free_space *tmp;

		bit_off = offset;
		bit_bytes = ctl->unit;
		ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
		if (!ret) {
			if (bit_off == offset) {
				ret = 1;
				goto out;
			} else if (bit_off > offset &&
				   offset + bytes > bit_off) {
				ret = 1;
				goto out;
			}
		}

		n = rb_prev(&info->offset_index);
		while (n) {
			tmp = rb_entry(n, struct btrfs_free_space,
				       offset_index);
			if (tmp->offset + tmp->bytes < offset)
				break;
			if (offset + bytes < tmp->offset) {
				n = rb_prev(&info->offset_index);
				continue;
			}
			info = tmp;
			goto have_info;
		}

		n = rb_next(&info->offset_index);
		while (n) {
			tmp = rb_entry(n, struct btrfs_free_space,
				       offset_index);
			if (offset + bytes < tmp->offset)
				break;
			if (tmp->offset + tmp->bytes < offset) {
				n = rb_next(&info->offset_index);
				continue;
			}
			info = tmp;
			goto have_info;
		}

3399
		ret = 0;
3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413
		goto out;
	}

	if (info->offset == offset) {
		ret = 1;
		goto out;
	}

	if (offset > info->offset && offset < info->offset + info->bytes)
		ret = 1;
out:
	spin_unlock(&ctl->tree_lock);
	return ret;
}
3414
#endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */