free-space-cache.c 102.2 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
J
Josef Bacik 已提交
2 3 4 5
/*
 * Copyright (C) 2008 Red Hat.  All rights reserved.
 */

6
#include <linux/pagemap.h>
J
Josef Bacik 已提交
7
#include <linux/sched.h>
8
#include <linux/sched/signal.h>
9
#include <linux/slab.h>
10
#include <linux/math64.h>
11
#include <linux/ratelimit.h>
12
#include <linux/error-injection.h>
13
#include <linux/sched/mm.h>
J
Josef Bacik 已提交
14
#include "ctree.h"
15 16
#include "free-space-cache.h"
#include "transaction.h"
17
#include "disk-io.h"
18
#include "extent_io.h"
19
#include "volumes.h"
20
#include "space-info.h"
21
#include "delalloc-space.h"
22
#include "block-group.h"
23
#include "discard.h"
24

25
#define BITS_PER_BITMAP		(PAGE_SIZE * 8UL)
26 27
#define MAX_CACHE_BYTES_PER_GIG	SZ_64K
#define FORCE_EXTENT_THRESHOLD	SZ_1M
J
Josef Bacik 已提交
28

29 30 31 32 33 34
struct btrfs_trim_range {
	u64 start;
	u64 bytes;
	struct list_head list;
};

35
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
36
			   struct btrfs_free_space *info);
37 38
static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info);
39 40 41 42 43 44 45 46
static int search_bitmap(struct btrfs_free_space_ctl *ctl,
			 struct btrfs_free_space *bitmap_info, u64 *offset,
			 u64 *bytes, bool for_alloc);
static void free_bitmap(struct btrfs_free_space_ctl *ctl,
			struct btrfs_free_space *bitmap_info);
static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info, u64 offset,
			      u64 bytes);
J
Josef Bacik 已提交
47

48 49 50
static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
					       struct btrfs_path *path,
					       u64 offset)
51
{
52
	struct btrfs_fs_info *fs_info = root->fs_info;
53 54 55 56 57 58
	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;
59
	unsigned nofs_flag;
60 61 62
	int ret;

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
63
	key.offset = offset;
64 65 66 67 68 69
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	if (ret < 0)
		return ERR_PTR(ret);
	if (ret > 0) {
70
		btrfs_release_path(path);
71 72 73 74 75 76 77 78
		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);
79
	btrfs_release_path(path);
80

81 82 83 84 85
	/*
	 * We are often under a trans handle at this point, so we need to make
	 * sure NOFS is set to keep us from deadlocking.
	 */
	nofs_flag = memalloc_nofs_save();
D
David Sterba 已提交
86
	inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
87
	btrfs_release_path(path);
88
	memalloc_nofs_restore(nofs_flag);
89 90 91
	if (IS_ERR(inode))
		return inode;

A
Al Viro 已提交
92
	mapping_set_gfp_mask(inode->i_mapping,
93 94
			mapping_gfp_constraint(inode->i_mapping,
			~(__GFP_FS | __GFP_HIGHMEM)));
95

96 97 98
	return inode;
}

99
struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
100
		struct btrfs_path *path)
101
{
102
	struct btrfs_fs_info *fs_info = block_group->fs_info;
103
	struct inode *inode = NULL;
104
	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
105 106 107 108 109 110 111 112

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

113
	inode = __lookup_free_space_inode(fs_info->tree_root, path,
114
					  block_group->start);
115 116 117
	if (IS_ERR(inode))
		return inode;

118
	spin_lock(&block_group->lock);
119
	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
120
		btrfs_info(fs_info, "Old style space inode found, converting.");
121 122
		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
			BTRFS_INODE_NODATACOW;
123 124 125
		block_group->disk_cache_state = BTRFS_DC_CLEAR;
	}

126
	if (!block_group->iref) {
127 128 129 130 131 132 133 134
		block_group->inode = igrab(inode);
		block_group->iref = 1;
	}
	spin_unlock(&block_group->lock);

	return inode;
}

135 136 137 138
static int __create_free_space_inode(struct btrfs_root *root,
				     struct btrfs_trans_handle *trans,
				     struct btrfs_path *path,
				     u64 ino, u64 offset)
139 140 141 142 143 144
{
	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;
145 146 147
	/* We inline CRCs for the free disk space cache */
	const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
			  BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
148 149
	int ret;

150
	ret = btrfs_insert_empty_inode(trans, root, path, ino);
151 152 153 154 155 156 157
	if (ret)
		return ret;

	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]);
158
	memzero_extent_buffer(leaf, (unsigned long)inode_item,
159 160 161 162 163 164 165
			     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);
166
	btrfs_set_inode_flags(leaf, inode_item, flags);
167 168
	btrfs_set_inode_nlink(leaf, inode_item, 1);
	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
169
	btrfs_set_inode_block_group(leaf, inode_item, offset);
170
	btrfs_mark_buffer_dirty(leaf);
171
	btrfs_release_path(path);
172 173

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
174
	key.offset = offset;
175 176 177 178
	key.type = 0;
	ret = btrfs_insert_empty_item(trans, root, path, &key,
				      sizeof(struct btrfs_free_space_header));
	if (ret < 0) {
179
		btrfs_release_path(path);
180 181
		return ret;
	}
182

183 184 185
	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
186
	memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
187 188
	btrfs_set_free_space_key(leaf, header, &disk_key);
	btrfs_mark_buffer_dirty(leaf);
189
	btrfs_release_path(path);
190 191 192 193

	return 0;
}

194
int create_free_space_inode(struct btrfs_trans_handle *trans,
195
			    struct btrfs_block_group *block_group,
196 197 198 199 200
			    struct btrfs_path *path)
{
	int ret;
	u64 ino;

201
	ret = btrfs_find_free_objectid(trans->fs_info->tree_root, &ino);
202 203 204
	if (ret < 0)
		return ret;

205
	return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
206
					 ino, block_group->start);
207 208
}

209
int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
210
				       struct btrfs_block_rsv *rsv)
211
{
212
	u64 needed_bytes;
213
	int ret;
214 215

	/* 1 for slack space, 1 for updating the inode */
216 217
	needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
		btrfs_calc_metadata_size(fs_info, 1);
218

219 220 221 222 223 224
	spin_lock(&rsv->lock);
	if (rsv->reserved < needed_bytes)
		ret = -ENOSPC;
	else
		ret = 0;
	spin_unlock(&rsv->lock);
225
	return ret;
226 227
}

228
int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
229
				    struct btrfs_block_group *block_group,
230 231
				    struct inode *inode)
{
232
	struct btrfs_root *root = BTRFS_I(inode)->root;
233
	int ret = 0;
234
	bool locked = false;
235 236

	if (block_group) {
237 238 239 240 241 242
		struct btrfs_path *path = btrfs_alloc_path();

		if (!path) {
			ret = -ENOMEM;
			goto fail;
		}
243
		locked = true;
244 245 246 247
		mutex_lock(&trans->transaction->cache_write_mutex);
		if (!list_empty(&block_group->io_list)) {
			list_del_init(&block_group->io_list);

248
			btrfs_wait_cache_io(trans, block_group, path);
249 250 251 252 253 254 255 256 257 258
			btrfs_put_block_group(block_group);
		}

		/*
		 * now that we've truncated the cache away, its no longer
		 * setup or written
		 */
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_CLEAR;
		spin_unlock(&block_group->lock);
259
		btrfs_free_path(path);
260
	}
261

262
	btrfs_i_size_write(BTRFS_I(inode), 0);
263
	truncate_pagecache(inode, 0);
264 265

	/*
266 267
	 * We skip the throttling logic for free space cache inodes, so we don't
	 * need to check for -EAGAIN.
268
	 */
269
	ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
270
					 0, BTRFS_EXTENT_DATA_KEY);
271 272
	if (ret)
		goto fail;
273

274
	ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
275 276

fail:
277 278
	if (locked)
		mutex_unlock(&trans->transaction->cache_write_mutex);
279
	if (ret)
280
		btrfs_abort_transaction(trans, ret);
281

282
	return ret;
283 284
}

285
static void readahead_cache(struct inode *inode)
286 287 288 289 290 291
{
	struct file_ra_state *ra;
	unsigned long last_index;

	ra = kzalloc(sizeof(*ra), GFP_NOFS);
	if (!ra)
292
		return;
293 294

	file_ra_state_init(ra, inode->i_mapping);
295
	last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
296 297 298 299 300 301

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

	kfree(ra);
}

302
static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
303
		       int write)
304
{
305 306
	int num_pages;

307
	num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
308

309
	/* Make sure we can fit our crcs and generation into the first page */
310
	if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
311 312
		return -ENOSPC;

313
	memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
314

315
	io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
316 317
	if (!io_ctl->pages)
		return -ENOMEM;
318 319

	io_ctl->num_pages = num_pages;
320
	io_ctl->fs_info = btrfs_sb(inode->i_sb);
321
	io_ctl->inode = inode;
322

323 324
	return 0;
}
325
ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
326

327
static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
328 329
{
	kfree(io_ctl->pages);
330
	io_ctl->pages = NULL;
331 332
}

333
static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
334 335 336 337 338 339 340
{
	if (io_ctl->cur) {
		io_ctl->cur = NULL;
		io_ctl->orig = NULL;
	}
}

341
static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
342
{
343
	ASSERT(io_ctl->index < io_ctl->num_pages);
344
	io_ctl->page = io_ctl->pages[io_ctl->index++];
345
	io_ctl->cur = page_address(io_ctl->page);
346
	io_ctl->orig = io_ctl->cur;
347
	io_ctl->size = PAGE_SIZE;
348
	if (clear)
349
		clear_page(io_ctl->cur);
350 351
}

352
static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
353 354 355 356 357 358
{
	int i;

	io_ctl_unmap_page(io_ctl);

	for (i = 0; i < io_ctl->num_pages; i++) {
359 360 361
		if (io_ctl->pages[i]) {
			ClearPageChecked(io_ctl->pages[i]);
			unlock_page(io_ctl->pages[i]);
362
			put_page(io_ctl->pages[i]);
363
		}
364 365 366
	}
}

367
static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
368 369
{
	struct page *page;
370
	struct inode *inode = io_ctl->inode;
371 372 373 374 375 376 377 378 379 380 381 382 383
	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);
384 385 386 387 388 389
			if (page->mapping != inode->i_mapping) {
				btrfs_err(BTRFS_I(inode)->root->fs_info,
					  "free space cache page truncated");
				io_ctl_drop_pages(io_ctl);
				return -EIO;
			}
390
			if (!PageUptodate(page)) {
391 392
				btrfs_err(BTRFS_I(inode)->root->fs_info,
					   "error reading free space cache");
393 394 395 396 397 398
				io_ctl_drop_pages(io_ctl);
				return -EIO;
			}
		}
	}

399 400 401 402 403
	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]);
	}

404 405 406
	return 0;
}

407
static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
408 409 410 411
{
	io_ctl_map_page(io_ctl, 1);

	/*
412 413
	 * Skip the csum areas.  If we don't check crcs then we just have a
	 * 64bit chunk at the front of the first page.
414
	 */
415 416
	io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
417

418
	put_unaligned_le64(generation, io_ctl->cur);
419 420 421
	io_ctl->cur += sizeof(u64);
}

422
static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
423
{
424
	u64 cache_gen;
425

426 427 428 429
	/*
	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
	 * chunk at the front of the first page.
	 */
430 431
	io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
	io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
432

433 434
	cache_gen = get_unaligned_le64(io_ctl->cur);
	if (cache_gen != generation) {
435
		btrfs_err_rl(io_ctl->fs_info,
436
			"space cache generation (%llu) does not match inode (%llu)",
437
				cache_gen, generation);
438 439 440 441
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}
	io_ctl->cur += sizeof(u64);
442 443 444
	return 0;
}

445
static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
446 447 448 449 450 451
{
	u32 *tmp;
	u32 crc = ~(u32)0;
	unsigned offset = 0;

	if (index == 0)
452
		offset = sizeof(u32) * io_ctl->num_pages;
453

454 455
	crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
	btrfs_crc32c_final(crc, (u8 *)&crc);
456
	io_ctl_unmap_page(io_ctl);
457
	tmp = page_address(io_ctl->pages[0]);
458 459 460 461
	tmp += index;
	*tmp = crc;
}

462
static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
463 464 465 466 467 468 469 470
{
	u32 *tmp, val;
	u32 crc = ~(u32)0;
	unsigned offset = 0;

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

471
	tmp = page_address(io_ctl->pages[0]);
472 473 474 475
	tmp += index;
	val = *tmp;

	io_ctl_map_page(io_ctl, 0);
476 477
	crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
	btrfs_crc32c_final(crc, (u8 *)&crc);
478
	if (val != crc) {
479
		btrfs_err_rl(io_ctl->fs_info,
480
			"csum mismatch on free space cache");
481 482 483 484
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}

485 486 487
	return 0;
}

488
static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
489 490 491 492 493 494 495 496
			    void *bitmap)
{
	struct btrfs_free_space_entry *entry;

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

	entry = io_ctl->cur;
497 498
	put_unaligned_le64(offset, &entry->offset);
	put_unaligned_le64(bytes, &entry->bytes);
499 500 501 502 503 504 505 506
	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

	/* 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;
}

518
static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
519 520 521 522 523 524 525 526 527
{
	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
		if (io_ctl->index >= io_ctl->num_pages)
			return -ENOSPC;
		io_ctl_map_page(io_ctl, 0);
	}

534
	copy_page(io_ctl->cur, bitmap);
535
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
536 537 538 539 540
	if (io_ctl->index < io_ctl->num_pages)
		io_ctl_map_page(io_ctl, 0);
	return 0;
}

541
static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
542
{
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
static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
559
			    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

	e = io_ctl->cur;
571 572
	entry->offset = get_unaligned_le64(&e->offset);
	entry->bytes = get_unaligned_le64(&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
static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
586
			      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
	copy_page(entry->bitmap, io_ctl->cur);
595
	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
static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
{
	struct btrfs_block_group *block_group = ctl->private;
	u64 max_bytes;
	u64 bitmap_bytes;
	u64 extent_bytes;
	u64 size = block_group->length;
	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
	u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);

	max_bitmaps = max_t(u64, max_bitmaps, 1);

	ASSERT(ctl->total_bitmaps <= max_bitmaps);

	/*
	 * We are trying to keep the total amount of memory used per 1GiB of
	 * space to be MAX_CACHE_BYTES_PER_GIG.  However, with a reclamation
	 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
	 * bitmaps, we may end up using more memory than this.
	 */
	if (size < SZ_1G)
		max_bytes = MAX_CACHE_BYTES_PER_GIG;
	else
		max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);

	bitmap_bytes = ctl->total_bitmaps * ctl->unit;

	/*
	 * we want the extent entry threshold to always be at most 1/2 the max
	 * bytes we can have, or whatever is less than that.
	 */
	extent_bytes = max_bytes - bitmap_bytes;
	extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);

	ctl->extents_thresh =
		div_u64(extent_bytes, sizeof(struct btrfs_free_space));
}

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
	struct btrfs_fs_info *fs_info = root->fs_info;
643 644
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
645
	struct btrfs_io_ctl io_ctl;
646
	struct btrfs_key key;
647
	struct btrfs_free_space *e, *n;
648
	LIST_HEAD(bitmaps);
649 650 651
	u64 num_entries;
	u64 num_bitmaps;
	u64 generation;
652
	u8 type;
653
	int ret = 0;
654 655

	/* Nothing in the space cache, goodbye */
656
	if (!i_size_read(inode))
657
		return 0;
658 659

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
660
	key.offset = offset;
661 662 663
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
664
	if (ret < 0)
665
		return 0;
666
	else if (ret > 0) {
667
		btrfs_release_path(path);
668
		return 0;
669 670
	}

671 672
	ret = -1;

673 674 675 676 677 678
	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);
679
	btrfs_release_path(path);
680

681
	if (!BTRFS_I(inode)->generation) {
682
		btrfs_info(fs_info,
683
			   "the free space cache file (%llu) is invalid, skip it",
684 685 686 687
			   offset);
		return 0;
	}

688
	if (BTRFS_I(inode)->generation != generation) {
689 690 691
		btrfs_err(fs_info,
			  "free space inode generation (%llu) did not match free space cache generation (%llu)",
			  BTRFS_I(inode)->generation, generation);
692
		return 0;
693 694 695
	}

	if (!num_entries)
696
		return 0;
697

698
	ret = io_ctl_init(&io_ctl, inode, 0);
699 700 701
	if (ret)
		return ret;

702
	readahead_cache(inode);
703

704
	ret = io_ctl_prepare_pages(&io_ctl, true);
705 706
	if (ret)
		goto out;
707

708 709 710 711
	ret = io_ctl_check_crc(&io_ctl, 0);
	if (ret)
		goto free_cache;

712 713 714
	ret = io_ctl_check_generation(&io_ctl, generation);
	if (ret)
		goto free_cache;
715

716 717 718 719
	while (num_entries) {
		e = kmem_cache_zalloc(btrfs_free_space_cachep,
				      GFP_NOFS);
		if (!e)
720 721
			goto free_cache;

722 723 724 725 726 727
		ret = io_ctl_read_entry(&io_ctl, e, &type);
		if (ret) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
		}

728 729 730
		if (!e->bytes) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
731
		}
732 733 734 735 736 737

		if (type == BTRFS_FREE_SPACE_EXTENT) {
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
738
				btrfs_err(fs_info,
739
					"Duplicate entries in free space cache, dumping");
740
				kmem_cache_free(btrfs_free_space_cachep, e);
741 742
				goto free_cache;
			}
743
		} else {
744
			ASSERT(num_bitmaps);
745
			num_bitmaps--;
746 747
			e->bitmap = kmem_cache_zalloc(
					btrfs_free_space_bitmap_cachep, GFP_NOFS);
748 749 750
			if (!e->bitmap) {
				kmem_cache_free(
					btrfs_free_space_cachep, e);
751 752
				goto free_cache;
			}
753 754 755
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			ctl->total_bitmaps++;
756
			recalculate_thresholds(ctl);
757 758
			spin_unlock(&ctl->tree_lock);
			if (ret) {
759
				btrfs_err(fs_info,
760
					"Duplicate entries in free space cache, dumping");
761
				kmem_cache_free(btrfs_free_space_cachep, e);
762 763
				goto free_cache;
			}
764
			list_add_tail(&e->list, &bitmaps);
765 766
		}

767 768
		num_entries--;
	}
769

770 771
	io_ctl_unmap_page(&io_ctl);

772 773 774 775 776
	/*
	 * 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) {
777
		list_del_init(&e->list);
778 779 780
		ret = io_ctl_read_bitmap(&io_ctl, e);
		if (ret)
			goto free_cache;
781 782
	}

783
	io_ctl_drop_pages(&io_ctl);
784 785
	ret = 1;
out:
786
	io_ctl_free(&io_ctl);
787 788
	return ret;
free_cache:
789
	io_ctl_drop_pages(&io_ctl);
790
	__btrfs_remove_free_space_cache(ctl);
791 792 793
	goto out;
}

794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828
static int copy_free_space_cache(struct btrfs_block_group *block_group,
				 struct btrfs_free_space_ctl *ctl)
{
	struct btrfs_free_space *info;
	struct rb_node *n;
	int ret = 0;

	while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
		info = rb_entry(n, struct btrfs_free_space, offset_index);
		if (!info->bitmap) {
			unlink_free_space(ctl, info);
			ret = btrfs_add_free_space(block_group, info->offset,
						   info->bytes);
			kmem_cache_free(btrfs_free_space_cachep, info);
		} else {
			u64 offset = info->offset;
			u64 bytes = ctl->unit;

			while (search_bitmap(ctl, info, &offset, &bytes,
					     false) == 0) {
				ret = btrfs_add_free_space(block_group, offset,
							   bytes);
				if (ret)
					break;
				bitmap_clear_bits(ctl, info, offset, bytes);
				offset = info->offset;
				bytes = ctl->unit;
			}
			free_bitmap(ctl, info);
		}
		cond_resched();
	}
	return ret;
}

829
int load_free_space_cache(struct btrfs_block_group *block_group)
J
Josef Bacik 已提交
830
{
831
	struct btrfs_fs_info *fs_info = block_group->fs_info;
832
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
833
	struct btrfs_free_space_ctl tmp_ctl = {};
834 835
	struct inode *inode;
	struct btrfs_path *path;
836
	int ret = 0;
837
	bool matched;
838
	u64 used = block_group->used;
839

840 841 842 843 844 845 846
	/*
	 * Because we could potentially discard our loaded free space, we want
	 * to load everything into a temporary structure first, and then if it's
	 * valid copy it all into the actual free space ctl.
	 */
	btrfs_init_free_space_ctl(block_group, &tmp_ctl);

847 848 849 850
	/*
	 * 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.
	 */
851
	spin_lock(&block_group->lock);
852 853 854 855
	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
		spin_unlock(&block_group->lock);
		return 0;
	}
856
	spin_unlock(&block_group->lock);
857 858 859 860

	path = btrfs_alloc_path();
	if (!path)
		return 0;
861 862
	path->search_commit_root = 1;
	path->skip_locking = 1;
863

864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882
	/*
	 * We must pass a path with search_commit_root set to btrfs_iget in
	 * order to avoid a deadlock when allocating extents for the tree root.
	 *
	 * When we are COWing an extent buffer from the tree root, when looking
	 * for a free extent, at extent-tree.c:find_free_extent(), we can find
	 * block group without its free space cache loaded. When we find one
	 * we must load its space cache which requires reading its free space
	 * cache's inode item from the root tree. If this inode item is located
	 * in the same leaf that we started COWing before, then we end up in
	 * deadlock on the extent buffer (trying to read lock it when we
	 * previously write locked it).
	 *
	 * It's safe to read the inode item using the commit root because
	 * block groups, once loaded, stay in memory forever (until they are
	 * removed) as well as their space caches once loaded. New block groups
	 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
	 * we will never try to read their inode item while the fs is mounted.
	 */
883
	inode = lookup_free_space_inode(block_group, path);
884 885 886 887 888
	if (IS_ERR(inode)) {
		btrfs_free_path(path);
		return 0;
	}

889 890 891 892
	/* 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);
893
		btrfs_free_path(path);
894 895 896 897
		goto out;
	}
	spin_unlock(&block_group->lock);

898
	ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
899
				      path, block_group->start);
900 901 902 903
	btrfs_free_path(path);
	if (ret <= 0)
		goto out;

904 905
	matched = (tmp_ctl.free_space == (block_group->length - used -
					  block_group->bytes_super));
906

907 908 909 910 911 912 913 914 915 916
	if (matched) {
		ret = copy_free_space_cache(block_group, &tmp_ctl);
		/*
		 * ret == 1 means we successfully loaded the free space cache,
		 * so we need to re-set it here.
		 */
		if (ret == 0)
			ret = 1;
	} else {
		__btrfs_remove_free_space_cache(&tmp_ctl);
J
Jeff Mahoney 已提交
917 918
		btrfs_warn(fs_info,
			   "block group %llu has wrong amount of free space",
919
			   block_group->start);
920 921 922 923 924 925 926 927
		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);
928
		ret = 0;
929

J
Jeff Mahoney 已提交
930 931
		btrfs_warn(fs_info,
			   "failed to load free space cache for block group %llu, rebuilding it now",
932
			   block_group->start);
933 934
	}

935 936 937
	spin_lock(&ctl->tree_lock);
	btrfs_discard_update_discardable(block_group);
	spin_unlock(&ctl->tree_lock);
938 939
	iput(inode);
	return ret;
940 941
}

942
static noinline_for_stack
943
int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
944
			      struct btrfs_free_space_ctl *ctl,
945
			      struct btrfs_block_group *block_group,
946 947
			      int *entries, int *bitmaps,
			      struct list_head *bitmap_list)
J
Josef Bacik 已提交
948
{
949
	int ret;
950
	struct btrfs_free_cluster *cluster = NULL;
951
	struct btrfs_free_cluster *cluster_locked = NULL;
952
	struct rb_node *node = rb_first(&ctl->free_space_offset);
953
	struct btrfs_trim_range *trim_entry;
954

955
	/* Get the cluster for this block_group if it exists */
956
	if (block_group && !list_empty(&block_group->cluster_list)) {
957 958 959
		cluster = list_entry(block_group->cluster_list.next,
				     struct btrfs_free_cluster,
				     block_group_list);
960
	}
961

962
	if (!node && cluster) {
963 964
		cluster_locked = cluster;
		spin_lock(&cluster_locked->lock);
965 966 967 968
		node = rb_first(&cluster->root);
		cluster = NULL;
	}

969 970 971
	/* Write out the extent entries */
	while (node) {
		struct btrfs_free_space *e;
J
Josef Bacik 已提交
972

973
		e = rb_entry(node, struct btrfs_free_space, offset_index);
974
		*entries += 1;
J
Josef Bacik 已提交
975

976
		ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
977 978
				       e->bitmap);
		if (ret)
979
			goto fail;
980

981
		if (e->bitmap) {
982 983
			list_add_tail(&e->list, bitmap_list);
			*bitmaps += 1;
984
		}
985 986 987
		node = rb_next(node);
		if (!node && cluster) {
			node = rb_first(&cluster->root);
988 989
			cluster_locked = cluster;
			spin_lock(&cluster_locked->lock);
990
			cluster = NULL;
991
		}
992
	}
993 994 995 996
	if (cluster_locked) {
		spin_unlock(&cluster_locked->lock);
		cluster_locked = NULL;
	}
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011

	/*
	 * Make sure we don't miss any range that was removed from our rbtree
	 * because trimming is running. Otherwise after a umount+mount (or crash
	 * after committing the transaction) we would leak free space and get
	 * an inconsistent free space cache report from fsck.
	 */
	list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
		ret = io_ctl_add_entry(io_ctl, trim_entry->start,
				       trim_entry->bytes, NULL);
		if (ret)
			goto fail;
		*entries += 1;
	}

1012 1013
	return 0;
fail:
1014 1015
	if (cluster_locked)
		spin_unlock(&cluster_locked->lock);
1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
	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,
1038
				 EXTENT_DELALLOC, 0, 0, NULL);
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
		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,
1050 1051
					 inode->i_size - 1, EXTENT_DELALLOC, 0,
					 0, NULL);
1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071
			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;
}

1072
static noinline_for_stack int write_pinned_extent_entries(
1073
			    struct btrfs_trans_handle *trans,
1074
			    struct btrfs_block_group *block_group,
1075
			    struct btrfs_io_ctl *io_ctl,
1076
			    int *entries)
1077 1078 1079 1080
{
	u64 start, extent_start, extent_end, len;
	struct extent_io_tree *unpin = NULL;
	int ret;
1081

1082 1083 1084
	if (!block_group)
		return 0;

1085 1086 1087
	/*
	 * We want to add any pinned extents to our free space cache
	 * so we don't leak the space
1088
	 *
1089 1090 1091
	 * We shouldn't have switched the pinned extents yet so this is the
	 * right one
	 */
1092
	unpin = &trans->transaction->pinned_extents;
1093

1094
	start = block_group->start;
1095

1096
	while (start < block_group->start + block_group->length) {
1097 1098
		ret = find_first_extent_bit(unpin, start,
					    &extent_start, &extent_end,
1099
					    EXTENT_DIRTY, NULL);
1100 1101
		if (ret)
			return 0;
J
Josef Bacik 已提交
1102

1103
		/* This pinned extent is out of our range */
1104
		if (extent_start >= block_group->start + block_group->length)
1105
			return 0;
1106

1107
		extent_start = max(extent_start, start);
1108 1109
		extent_end = min(block_group->start + block_group->length,
				 extent_end + 1);
1110
		len = extent_end - extent_start;
J
Josef Bacik 已提交
1111

1112 1113
		*entries += 1;
		ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1114
		if (ret)
1115
			return -ENOSPC;
J
Josef Bacik 已提交
1116

1117
		start = extent_end;
1118
	}
J
Josef Bacik 已提交
1119

1120 1121 1122 1123
	return 0;
}

static noinline_for_stack int
1124
write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1125
{
1126
	struct btrfs_free_space *entry, *next;
1127 1128
	int ret;

J
Josef Bacik 已提交
1129
	/* Write out the bitmaps */
1130
	list_for_each_entry_safe(entry, next, bitmap_list, list) {
1131
		ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1132
		if (ret)
1133
			return -ENOSPC;
J
Josef Bacik 已提交
1134
		list_del_init(&entry->list);
1135 1136
	}

1137 1138
	return 0;
}
J
Josef Bacik 已提交
1139

1140 1141 1142
static int flush_dirty_cache(struct inode *inode)
{
	int ret;
1143

1144
	ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1145
	if (ret)
1146
		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1147
				 EXTENT_DELALLOC, 0, 0, NULL);
J
Josef Bacik 已提交
1148

1149
	return ret;
1150 1151 1152
}

static void noinline_for_stack
1153
cleanup_bitmap_list(struct list_head *bitmap_list)
1154
{
1155
	struct btrfs_free_space *entry, *next;
1156

1157
	list_for_each_entry_safe(entry, next, bitmap_list, list)
1158
		list_del_init(&entry->list);
1159 1160 1161 1162 1163
}

static void noinline_for_stack
cleanup_write_cache_enospc(struct inode *inode,
			   struct btrfs_io_ctl *io_ctl,
1164
			   struct extent_state **cached_state)
1165
{
1166 1167
	io_ctl_drop_pages(io_ctl);
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1168
			     i_size_read(inode) - 1, cached_state);
1169
}
1170

1171 1172
static int __btrfs_wait_cache_io(struct btrfs_root *root,
				 struct btrfs_trans_handle *trans,
1173
				 struct btrfs_block_group *block_group,
1174 1175
				 struct btrfs_io_ctl *io_ctl,
				 struct btrfs_path *path, u64 offset)
1176 1177 1178 1179
{
	int ret;
	struct inode *inode = io_ctl->inode;

1180 1181 1182
	if (!inode)
		return 0;

1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194
	/* Flush the dirty pages in the cache file. */
	ret = flush_dirty_cache(inode);
	if (ret)
		goto out;

	/* Update the cache item to tell everyone this cache file is valid. */
	ret = update_cache_item(trans, root, inode, path, offset,
				io_ctl->entries, io_ctl->bitmaps);
out:
	if (ret) {
		invalidate_inode_pages2(inode->i_mapping);
		BTRFS_I(inode)->generation = 0;
1195 1196
		if (block_group)
			btrfs_debug(root->fs_info,
1197 1198
	  "failed to write free space cache for block group %llu error %d",
				  block_group->start, ret);
1199
	}
1200
	btrfs_update_inode(trans, root, BTRFS_I(inode));
1201 1202

	if (block_group) {
1203 1204 1205 1206
		/* the dirty list is protected by the dirty_bgs_lock */
		spin_lock(&trans->transaction->dirty_bgs_lock);

		/* the disk_cache_state is protected by the block group lock */
1207 1208 1209 1210
		spin_lock(&block_group->lock);

		/*
		 * only mark this as written if we didn't get put back on
1211 1212
		 * the dirty list while waiting for IO.   Otherwise our
		 * cache state won't be right, and we won't get written again
1213 1214 1215 1216 1217 1218 1219
		 */
		if (!ret && list_empty(&block_group->dirty_list))
			block_group->disk_cache_state = BTRFS_DC_WRITTEN;
		else if (ret)
			block_group->disk_cache_state = BTRFS_DC_ERROR;

		spin_unlock(&block_group->lock);
1220
		spin_unlock(&trans->transaction->dirty_bgs_lock);
1221 1222 1223 1224 1225 1226 1227 1228
		io_ctl->inode = NULL;
		iput(inode);
	}

	return ret;

}

1229
int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1230
			struct btrfs_block_group *block_group,
1231 1232 1233 1234
			struct btrfs_path *path)
{
	return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
				     block_group, &block_group->io_ctl,
1235
				     path, block_group->start);
1236 1237
}

1238 1239 1240 1241 1242 1243 1244 1245
/**
 * __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
 *
 * This function writes out a free space cache struct to disk for quick recovery
G
Geliang Tang 已提交
1246
 * on mount.  This will return 0 if it was successful in writing the cache out,
1247
 * or an errno if it was not.
1248 1249 1250
 */
static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
				   struct btrfs_free_space_ctl *ctl,
1251
				   struct btrfs_block_group *block_group,
1252
				   struct btrfs_io_ctl *io_ctl,
1253
				   struct btrfs_trans_handle *trans)
1254 1255
{
	struct extent_state *cached_state = NULL;
1256
	LIST_HEAD(bitmap_list);
1257 1258 1259
	int entries = 0;
	int bitmaps = 0;
	int ret;
1260
	int must_iput = 0;
1261 1262

	if (!i_size_read(inode))
1263
		return -EIO;
1264

1265
	WARN_ON(io_ctl->pages);
1266
	ret = io_ctl_init(io_ctl, inode, 1);
1267
	if (ret)
1268
		return ret;
1269

1270 1271 1272 1273 1274 1275 1276 1277 1278
	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;
1279
			must_iput = 1;
1280 1281 1282 1283 1284
			goto out;
		}
		spin_unlock(&block_group->lock);
	}

1285
	/* Lock all pages first so we can lock the extent safely. */
1286
	ret = io_ctl_prepare_pages(io_ctl, false);
1287
	if (ret)
1288
		goto out_unlock;
1289 1290

	lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1291
			 &cached_state);
1292

1293
	io_ctl_set_generation(io_ctl, trans->transid);
1294

1295
	mutex_lock(&ctl->cache_writeout_mutex);
1296
	/* Write out the extent entries in the free space cache */
1297
	spin_lock(&ctl->tree_lock);
1298
	ret = write_cache_extent_entries(io_ctl, ctl,
1299 1300
					 block_group, &entries, &bitmaps,
					 &bitmap_list);
1301 1302
	if (ret)
		goto out_nospc_locked;
1303

1304 1305 1306 1307
	/*
	 * 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.
1308 1309 1310
	 *
	 * If this changes while we are working we'll get added back to
	 * the dirty list and redo it.  No locking needed
1311
	 */
1312
	ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1313 1314
	if (ret)
		goto out_nospc_locked;
1315

1316 1317 1318 1319 1320
	/*
	 * At last, we write out all the bitmaps and keep cache_writeout_mutex
	 * locked while doing it because a concurrent trim can be manipulating
	 * or freeing the bitmap.
	 */
1321
	ret = write_bitmap_entries(io_ctl, &bitmap_list);
1322
	spin_unlock(&ctl->tree_lock);
1323
	mutex_unlock(&ctl->cache_writeout_mutex);
1324 1325 1326 1327
	if (ret)
		goto out_nospc;

	/* Zero out the rest of the pages just to make sure */
1328
	io_ctl_zero_remaining_pages(io_ctl);
1329

1330
	/* Everything is written out, now we dirty the pages in the file. */
1331 1332
	ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
				io_ctl->num_pages, 0, i_size_read(inode),
1333
				&cached_state, false);
1334
	if (ret)
1335
		goto out_nospc;
1336

1337 1338
	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
		up_write(&block_group->data_rwsem);
1339 1340 1341 1342
	/*
	 * Release the pages and unlock the extent, we will flush
	 * them out later
	 */
1343
	io_ctl_drop_pages(io_ctl);
1344
	io_ctl_free(io_ctl);
1345 1346

	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1347
			     i_size_read(inode) - 1, &cached_state);
1348

1349 1350
	/*
	 * at this point the pages are under IO and we're happy,
1351
	 * The caller is responsible for waiting on them and updating
1352 1353 1354 1355 1356 1357
	 * the cache and the inode
	 */
	io_ctl->entries = entries;
	io_ctl->bitmaps = bitmaps;

	ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1358
	if (ret)
1359 1360
		goto out;

1361 1362
	return 0;

1363 1364 1365 1366 1367
out_nospc_locked:
	cleanup_bitmap_list(&bitmap_list);
	spin_unlock(&ctl->tree_lock);
	mutex_unlock(&ctl->cache_writeout_mutex);

1368
out_nospc:
1369
	cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1370

1371
out_unlock:
1372 1373 1374
	if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
		up_write(&block_group->data_rwsem);

1375 1376 1377 1378 1379 1380 1381
out:
	io_ctl->inode = NULL;
	io_ctl_free(io_ctl);
	if (ret) {
		invalidate_inode_pages2(inode->i_mapping);
		BTRFS_I(inode)->generation = 0;
	}
1382
	btrfs_update_inode(trans, root, BTRFS_I(inode));
1383 1384 1385
	if (must_iput)
		iput(inode);
	return ret;
1386 1387
}

1388
int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1389
			  struct btrfs_block_group *block_group,
1390 1391
			  struct btrfs_path *path)
{
1392
	struct btrfs_fs_info *fs_info = trans->fs_info;
1393 1394 1395 1396 1397 1398 1399
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct inode *inode;
	int ret = 0;

	spin_lock(&block_group->lock);
	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
		spin_unlock(&block_group->lock);
1400 1401
		return 0;
	}
1402 1403
	spin_unlock(&block_group->lock);

1404
	inode = lookup_free_space_inode(block_group, path);
1405 1406 1407
	if (IS_ERR(inode))
		return 0;

1408 1409
	ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
				block_group, &block_group->io_ctl, trans);
1410
	if (ret) {
1411
		btrfs_debug(fs_info,
1412 1413
	  "failed to write free space cache for block group %llu error %d",
			  block_group->start, ret);
1414 1415 1416 1417 1418 1419
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_ERROR;
		spin_unlock(&block_group->lock);

		block_group->io_ctl.inode = NULL;
		iput(inode);
1420 1421
	}

1422 1423 1424 1425 1426
	/*
	 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
	 * to wait for IO and put the inode
	 */

J
Josef Bacik 已提交
1427 1428 1429
	return ret;
}

1430
static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1431
					  u64 offset)
J
Josef Bacik 已提交
1432
{
1433
	ASSERT(offset >= bitmap_start);
1434
	offset -= bitmap_start;
1435
	return (unsigned long)(div_u64(offset, unit));
1436
}
J
Josef Bacik 已提交
1437

1438
static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1439
{
1440
	return (unsigned long)(div_u64(bytes, unit));
1441
}
J
Josef Bacik 已提交
1442

1443
static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1444 1445 1446
				   u64 offset)
{
	u64 bitmap_start;
1447
	u64 bytes_per_bitmap;
J
Josef Bacik 已提交
1448

1449 1450
	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
	bitmap_start = offset - ctl->start;
1451
	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1452
	bitmap_start *= bytes_per_bitmap;
1453
	bitmap_start += ctl->start;
J
Josef Bacik 已提交
1454

1455
	return bitmap_start;
J
Josef Bacik 已提交
1456 1457
}

1458 1459
static int tree_insert_offset(struct rb_root *root, u64 offset,
			      struct rb_node *node, int bitmap)
J
Josef Bacik 已提交
1460 1461 1462 1463 1464 1465 1466
{
	struct rb_node **p = &root->rb_node;
	struct rb_node *parent = NULL;
	struct btrfs_free_space *info;

	while (*p) {
		parent = *p;
1467
		info = rb_entry(parent, struct btrfs_free_space, offset_index);
J
Josef Bacik 已提交
1468

1469
		if (offset < info->offset) {
J
Josef Bacik 已提交
1470
			p = &(*p)->rb_left;
1471
		} else if (offset > info->offset) {
J
Josef Bacik 已提交
1472
			p = &(*p)->rb_right;
1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487
		} 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) {
1488 1489 1490 1491
				if (info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1492 1493
				p = &(*p)->rb_right;
			} else {
1494 1495 1496 1497
				if (!info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1498 1499 1500
				p = &(*p)->rb_left;
			}
		}
J
Josef Bacik 已提交
1501 1502 1503 1504 1505 1506 1507 1508 1509
	}

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

	return 0;
}

/*
J
Josef Bacik 已提交
1510 1511
 * searches the tree for the given offset.
 *
1512 1513 1514
 * 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
1515
 */
1516
static struct btrfs_free_space *
1517
tree_search_offset(struct btrfs_free_space_ctl *ctl,
1518
		   u64 offset, int bitmap_only, int fuzzy)
J
Josef Bacik 已提交
1519
{
1520
	struct rb_node *n = ctl->free_space_offset.rb_node;
1521 1522 1523 1524 1525 1526 1527 1528
	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
1529 1530

		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1531
		prev = entry;
J
Josef Bacik 已提交
1532

1533
		if (offset < entry->offset)
J
Josef Bacik 已提交
1534
			n = n->rb_left;
1535
		else if (offset > entry->offset)
J
Josef Bacik 已提交
1536
			n = n->rb_right;
1537
		else
J
Josef Bacik 已提交
1538 1539 1540
			break;
	}

1541 1542 1543 1544 1545
	if (bitmap_only) {
		if (!entry)
			return NULL;
		if (entry->bitmap)
			return entry;
J
Josef Bacik 已提交
1546

1547 1548 1549 1550 1551 1552 1553 1554 1555 1556
		/*
		 * 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
1557

1558 1559 1560 1561
		WARN_ON(!entry->bitmap);
		return entry;
	} else if (entry) {
		if (entry->bitmap) {
J
Josef Bacik 已提交
1562
			/*
1563 1564
			 * if previous extent entry covers the offset,
			 * we should return it instead of the bitmap entry
J
Josef Bacik 已提交
1565
			 */
1566 1567
			n = rb_prev(&entry->offset_index);
			if (n) {
1568 1569
				prev = rb_entry(n, struct btrfs_free_space,
						offset_index);
1570 1571 1572
				if (!prev->bitmap &&
				    prev->offset + prev->bytes > offset)
					entry = prev;
J
Josef Bacik 已提交
1573
			}
1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587
		}
		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);
1588
			ASSERT(entry->offset <= offset);
J
Josef Bacik 已提交
1589
		} else {
1590 1591 1592 1593
			if (fuzzy)
				return entry;
			else
				return NULL;
J
Josef Bacik 已提交
1594 1595 1596
		}
	}

1597
	if (entry->bitmap) {
1598 1599
		n = rb_prev(&entry->offset_index);
		if (n) {
1600 1601
			prev = rb_entry(n, struct btrfs_free_space,
					offset_index);
1602 1603 1604
			if (!prev->bitmap &&
			    prev->offset + prev->bytes > offset)
				return prev;
1605
		}
1606
		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1607 1608 1609 1610 1611 1612 1613 1614 1615 1616
			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 *
1617
			    ctl->unit > offset)
1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629
				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
1630 1631
}

1632
static inline void
1633
__unlink_free_space(struct btrfs_free_space_ctl *ctl,
1634
		    struct btrfs_free_space *info)
J
Josef Bacik 已提交
1635
{
1636 1637
	rb_erase(&info->offset_index, &ctl->free_space_offset);
	ctl->free_extents--;
1638

1639
	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1640
		ctl->discardable_extents[BTRFS_STAT_CURR]--;
1641 1642
		ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
	}
1643 1644
}

1645
static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1646 1647
			      struct btrfs_free_space *info)
{
1648 1649
	__unlink_free_space(ctl, info);
	ctl->free_space -= info->bytes;
J
Josef Bacik 已提交
1650 1651
}

1652
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1653 1654 1655 1656
			   struct btrfs_free_space *info)
{
	int ret = 0;

1657
	ASSERT(info->bytes || info->bitmap);
1658
	ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1659
				 &info->offset_index, (info->bitmap != NULL));
J
Josef Bacik 已提交
1660 1661 1662
	if (ret)
		return ret;

1663
	if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1664
		ctl->discardable_extents[BTRFS_STAT_CURR]++;
1665 1666
		ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
	}
1667

1668 1669
	ctl->free_space += info->bytes;
	ctl->free_extents++;
1670 1671 1672
	return ret;
}

1673 1674 1675
static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
				       struct btrfs_free_space *info,
				       u64 offset, u64 bytes)
1676
{
1677 1678
	unsigned long start, count, end;
	int extent_delta = -1;
1679

1680 1681
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
1682 1683
	end = start + count;
	ASSERT(end <= BITS_PER_BITMAP);
1684

L
Li Zefan 已提交
1685
	bitmap_clear(info->bitmap, start, count);
1686 1687

	info->bytes -= bytes;
1688 1689
	if (info->max_extent_size > ctl->unit)
		info->max_extent_size = 0;
1690 1691 1692 1693 1694 1695 1696 1697

	if (start && test_bit(start - 1, info->bitmap))
		extent_delta++;

	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
		extent_delta++;

	info->bitmap_extents += extent_delta;
1698
	if (!btrfs_free_space_trimmed(info)) {
1699
		ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1700 1701
		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
	}
1702 1703 1704 1705 1706 1707 1708
}

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);
1709
	ctl->free_space -= bytes;
1710 1711
}

1712
static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1713 1714
			    struct btrfs_free_space *info, u64 offset,
			    u64 bytes)
1715
{
1716 1717
	unsigned long start, count, end;
	int extent_delta = 1;
1718

1719 1720
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
1721 1722
	end = start + count;
	ASSERT(end <= BITS_PER_BITMAP);
1723

L
Li Zefan 已提交
1724
	bitmap_set(info->bitmap, start, count);
1725 1726

	info->bytes += bytes;
1727
	ctl->free_space += bytes;
1728 1729 1730 1731 1732 1733 1734 1735

	if (start && test_bit(start - 1, info->bitmap))
		extent_delta--;

	if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
		extent_delta--;

	info->bitmap_extents += extent_delta;
1736
	if (!btrfs_free_space_trimmed(info)) {
1737
		ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1738 1739
		ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
	}
1740 1741
}

1742 1743 1744 1745
/*
 * If we can not find suitable extent, we will use bytes to record
 * the size of the max extent.
 */
1746
static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1747
			 struct btrfs_free_space *bitmap_info, u64 *offset,
1748
			 u64 *bytes, bool for_alloc)
1749 1750
{
	unsigned long found_bits = 0;
1751
	unsigned long max_bits = 0;
1752 1753
	unsigned long bits, i;
	unsigned long next_zero;
1754
	unsigned long extent_bits;
1755

1756 1757 1758 1759
	/*
	 * Skip searching the bitmap if we don't have a contiguous section that
	 * is large enough for this allocation.
	 */
1760 1761
	if (for_alloc &&
	    bitmap_info->max_extent_size &&
1762 1763 1764 1765 1766
	    bitmap_info->max_extent_size < *bytes) {
		*bytes = bitmap_info->max_extent_size;
		return -1;
	}

1767
	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1768
			  max_t(u64, *offset, bitmap_info->offset));
1769
	bits = bytes_to_bits(*bytes, ctl->unit);
1770

1771
	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1772 1773 1774 1775
		if (for_alloc && bits == 1) {
			found_bits = 1;
			break;
		}
1776 1777
		next_zero = find_next_zero_bit(bitmap_info->bitmap,
					       BITS_PER_BITMAP, i);
1778 1779 1780
		extent_bits = next_zero - i;
		if (extent_bits >= bits) {
			found_bits = extent_bits;
1781
			break;
1782 1783
		} else if (extent_bits > max_bits) {
			max_bits = extent_bits;
1784 1785 1786 1787 1788
		}
		i = next_zero;
	}

	if (found_bits) {
1789 1790
		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
		*bytes = (u64)(found_bits) * ctl->unit;
1791 1792 1793
		return 0;
	}

1794
	*bytes = (u64)(max_bits) * ctl->unit;
1795
	bitmap_info->max_extent_size = *bytes;
1796 1797 1798
	return -1;
}

J
Josef Bacik 已提交
1799 1800 1801 1802 1803 1804 1805
static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
{
	if (entry->bitmap)
		return entry->max_extent_size;
	return entry->bytes;
}

1806
/* Cache the size of the max extent in bytes */
1807
static struct btrfs_free_space *
D
David Woodhouse 已提交
1808
find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1809
		unsigned long align, u64 *max_extent_size)
1810 1811 1812
{
	struct btrfs_free_space *entry;
	struct rb_node *node;
D
David Woodhouse 已提交
1813 1814
	u64 tmp;
	u64 align_off;
1815 1816
	int ret;

1817
	if (!ctl->free_space_offset.rb_node)
1818
		goto out;
1819

1820
	entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1821
	if (!entry)
1822
		goto out;
1823 1824 1825

	for (node = &entry->offset_index; node; node = rb_next(node)) {
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
1826
		if (entry->bytes < *bytes) {
J
Josef Bacik 已提交
1827 1828
			*max_extent_size = max(get_max_extent_size(entry),
					       *max_extent_size);
1829
			continue;
1830
		}
1831

D
David Woodhouse 已提交
1832 1833 1834 1835
		/* make sure the space returned is big enough
		 * to match our requested alignment
		 */
		if (*bytes >= align) {
1836
			tmp = entry->offset - ctl->start + align - 1;
1837
			tmp = div64_u64(tmp, align);
D
David Woodhouse 已提交
1838 1839 1840 1841 1842 1843 1844
			tmp = tmp * align + ctl->start;
			align_off = tmp - entry->offset;
		} else {
			align_off = 0;
			tmp = entry->offset;
		}

1845
		if (entry->bytes < *bytes + align_off) {
J
Josef Bacik 已提交
1846 1847
			*max_extent_size = max(get_max_extent_size(entry),
					       *max_extent_size);
D
David Woodhouse 已提交
1848
			continue;
1849
		}
D
David Woodhouse 已提交
1850

1851
		if (entry->bitmap) {
1852 1853
			u64 size = *bytes;

1854
			ret = search_bitmap(ctl, entry, &tmp, &size, true);
D
David Woodhouse 已提交
1855 1856
			if (!ret) {
				*offset = tmp;
1857
				*bytes = size;
1858
				return entry;
J
Josef Bacik 已提交
1859 1860 1861 1862
			} else {
				*max_extent_size =
					max(get_max_extent_size(entry),
					    *max_extent_size);
D
David Woodhouse 已提交
1863
			}
1864 1865 1866
			continue;
		}

D
David Woodhouse 已提交
1867 1868
		*offset = tmp;
		*bytes = entry->bytes - align_off;
1869 1870
		return entry;
	}
1871
out:
1872 1873 1874
	return NULL;
}

1875
static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1876 1877
			   struct btrfs_free_space *info, u64 offset)
{
1878
	info->offset = offset_to_bitmap(ctl, offset);
J
Josef Bacik 已提交
1879
	info->bytes = 0;
1880
	info->bitmap_extents = 0;
1881
	INIT_LIST_HEAD(&info->list);
1882 1883
	link_free_space(ctl, info);
	ctl->total_bitmaps++;
1884
	recalculate_thresholds(ctl);
1885 1886
}

1887
static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1888 1889
			struct btrfs_free_space *bitmap_info)
{
1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901
	/*
	 * Normally when this is called, the bitmap is completely empty. However,
	 * if we are blowing up the free space cache for one reason or another
	 * via __btrfs_remove_free_space_cache(), then it may not be freed and
	 * we may leave stats on the table.
	 */
	if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
		ctl->discardable_extents[BTRFS_STAT_CURR] -=
			bitmap_info->bitmap_extents;
		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;

	}
1902
	unlink_free_space(ctl, bitmap_info);
1903
	kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
1904
	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1905
	ctl->total_bitmaps--;
1906
	recalculate_thresholds(ctl);
1907 1908
}

1909
static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1910 1911 1912 1913
			      struct btrfs_free_space *bitmap_info,
			      u64 *offset, u64 *bytes)
{
	u64 end;
1914 1915
	u64 search_start, search_bytes;
	int ret;
1916 1917

again:
1918
	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1919

1920
	/*
1921 1922 1923 1924
	 * 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.
1925 1926
	 */
	search_start = *offset;
1927
	search_bytes = ctl->unit;
1928
	search_bytes = min(search_bytes, end - search_start + 1);
1929 1930
	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
			    false);
1931 1932
	if (ret < 0 || search_start != *offset)
		return -EINVAL;
1933

1934 1935 1936 1937 1938 1939 1940 1941 1942
	/* 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;
1943 1944

	if (*bytes) {
1945
		struct rb_node *next = rb_next(&bitmap_info->offset_index);
1946
		if (!bitmap_info->bytes)
1947
			free_bitmap(ctl, bitmap_info);
1948

1949 1950 1951 1952 1953
		/*
		 * no entry after this bitmap, but we still have bytes to
		 * remove, so something has gone wrong.
		 */
		if (!next)
1954 1955
			return -EINVAL;

1956 1957 1958 1959 1960 1961 1962
		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.
		 */
1963 1964 1965
		if (!bitmap_info->bitmap)
			return -EAGAIN;

1966 1967 1968 1969 1970 1971 1972
		/*
		 * 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;
1973
		search_bytes = ctl->unit;
1974
		ret = search_bitmap(ctl, bitmap_info, &search_start,
1975
				    &search_bytes, false);
1976 1977 1978
		if (ret < 0 || search_start != *offset)
			return -EAGAIN;

1979
		goto again;
1980
	} else if (!bitmap_info->bytes)
1981
		free_bitmap(ctl, bitmap_info);
1982 1983 1984 1985

	return 0;
}

J
Josef Bacik 已提交
1986 1987
static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
			       struct btrfs_free_space *info, u64 offset,
1988
			       u64 bytes, enum btrfs_trim_state trim_state)
J
Josef Bacik 已提交
1989 1990 1991 1992
{
	u64 bytes_to_set = 0;
	u64 end;

1993 1994 1995 1996
	/*
	 * This is a tradeoff to make bitmap trim state minimal.  We mark the
	 * whole bitmap untrimmed if at any point we add untrimmed regions.
	 */
1997
	if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
1998
		if (btrfs_free_space_trimmed(info)) {
1999 2000
			ctl->discardable_extents[BTRFS_STAT_CURR] +=
				info->bitmap_extents;
2001 2002
			ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
		}
2003
		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2004
	}
2005

J
Josef Bacik 已提交
2006 2007 2008 2009 2010 2011
	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);

2012 2013 2014 2015 2016 2017
	/*
	 * We set some bytes, we have no idea what the max extent size is
	 * anymore.
	 */
	info->max_extent_size = 0;

J
Josef Bacik 已提交
2018 2019 2020 2021
	return bytes_to_set;

}

2022 2023
static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
		      struct btrfs_free_space *info)
2024
{
2025
	struct btrfs_block_group *block_group = ctl->private;
2026
	struct btrfs_fs_info *fs_info = block_group->fs_info;
2027 2028 2029
	bool forced = false;

#ifdef CONFIG_BTRFS_DEBUG
2030
	if (btrfs_should_fragment_free_space(block_group))
2031 2032
		forced = true;
#endif
2033

2034 2035 2036 2037
	/* This is a way to reclaim large regions from the bitmaps. */
	if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
		return false;

2038 2039 2040 2041
	/*
	 * If we are below the extents threshold then we can add this as an
	 * extent, and don't have to deal with the bitmap
	 */
2042
	if (!forced && ctl->free_extents < ctl->extents_thresh) {
2043 2044 2045
		/*
		 * 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
2046
		 * to reserve them to larger extents, however if we have plenty
2047 2048 2049
		 * of cache left then go ahead an dadd them, no sense in adding
		 * the overhead of a bitmap if we don't have to.
		 */
2050 2051
		if (info->bytes <= fs_info->sectorsize * 8) {
			if (ctl->free_extents * 3 <= ctl->extents_thresh)
2052
				return false;
2053
		} else {
2054
			return false;
2055 2056
		}
	}
2057 2058

	/*
2059 2060 2061 2062 2063 2064
	 * 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.
2065
	 */
2066
	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2067 2068 2069 2070 2071
		return false;

	return true;
}

2072
static const struct btrfs_free_space_op free_space_op = {
J
Josef Bacik 已提交
2073 2074 2075
	.use_bitmap		= use_bitmap,
};

2076 2077 2078 2079
static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info)
{
	struct btrfs_free_space *bitmap_info;
2080
	struct btrfs_block_group *block_group = NULL;
2081
	int added = 0;
J
Josef Bacik 已提交
2082
	u64 bytes, offset, bytes_added;
2083
	enum btrfs_trim_state trim_state;
2084
	int ret;
2085 2086 2087

	bytes = info->bytes;
	offset = info->offset;
2088
	trim_state = info->trim_state;
2089

2090 2091 2092
	if (!ctl->op->use_bitmap(ctl, info))
		return 0;

J
Josef Bacik 已提交
2093 2094
	if (ctl->op == &free_space_op)
		block_group = ctl->private;
2095
again:
J
Josef Bacik 已提交
2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112
	/*
	 * 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);
2113
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
2114 2115 2116 2117 2118
		}

		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		if (!entry->bitmap) {
			spin_unlock(&cluster->lock);
2119
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
2120 2121 2122
		}

		if (entry->offset == offset_to_bitmap(ctl, offset)) {
2123 2124
			bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
							  bytes, trim_state);
J
Josef Bacik 已提交
2125 2126 2127 2128 2129 2130 2131 2132 2133
			bytes -= bytes_added;
			offset += bytes_added;
		}
		spin_unlock(&cluster->lock);
		if (!bytes) {
			ret = 1;
			goto out;
		}
	}
2134 2135

no_cluster_bitmap:
2136
	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2137 2138
					 1, 0);
	if (!bitmap_info) {
2139
		ASSERT(added == 0);
2140 2141 2142
		goto new_bitmap;
	}

2143 2144
	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
					  trim_state);
J
Josef Bacik 已提交
2145 2146 2147
	bytes -= bytes_added;
	offset += bytes_added;
	added = 0;
2148 2149 2150 2151 2152 2153 2154 2155 2156

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

new_bitmap:
	if (info && info->bitmap) {
2157
		add_new_bitmap(ctl, info, offset);
2158 2159 2160 2161
		added = 1;
		info = NULL;
		goto again;
	} else {
2162
		spin_unlock(&ctl->tree_lock);
2163 2164 2165

		/* no pre-allocated info, allocate a new one */
		if (!info) {
2166 2167
			info = kmem_cache_zalloc(btrfs_free_space_cachep,
						 GFP_NOFS);
2168
			if (!info) {
2169
				spin_lock(&ctl->tree_lock);
2170 2171 2172 2173 2174 2175
				ret = -ENOMEM;
				goto out;
			}
		}

		/* allocate the bitmap */
2176 2177
		info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
						 GFP_NOFS);
2178
		info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2179
		spin_lock(&ctl->tree_lock);
2180 2181 2182 2183 2184 2185 2186 2187 2188
		if (!info->bitmap) {
			ret = -ENOMEM;
			goto out;
		}
		goto again;
	}

out:
	if (info) {
2189 2190 2191
		if (info->bitmap)
			kmem_cache_free(btrfs_free_space_bitmap_cachep,
					info->bitmap);
2192
		kmem_cache_free(btrfs_free_space_cachep, info);
2193
	}
J
Josef Bacik 已提交
2194 2195 2196 2197

	return ret;
}

2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
/*
 * Free space merging rules:
 *  1) Merge trimmed areas together
 *  2) Let untrimmed areas coalesce with trimmed areas
 *  3) Always pull neighboring regions from bitmaps
 *
 * The above rules are for when we merge free space based on btrfs_trim_state.
 * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
 * same reason: to promote larger extent regions which makes life easier for
 * find_free_extent().  Rule 2 enables coalescing based on the common path
 * being returning free space from btrfs_finish_extent_commit().  So when free
 * space is trimmed, it will prevent aggregating trimmed new region and
 * untrimmed regions in the rb_tree.  Rule 3 is purely to obtain larger extents
 * and provide find_free_extent() with the largest extents possible hoping for
 * the reuse path.
 */
2214
static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2215
			  struct btrfs_free_space *info, bool update_stat)
J
Josef Bacik 已提交
2216
{
2217
	struct btrfs_free_space *left_info = NULL;
2218 2219 2220 2221
	struct btrfs_free_space *right_info;
	bool merged = false;
	u64 offset = info->offset;
	u64 bytes = info->bytes;
2222
	const bool is_trimmed = btrfs_free_space_trimmed(info);
2223

J
Josef Bacik 已提交
2224 2225 2226 2227 2228
	/*
	 * 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
	 */
2229
	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2230 2231 2232
	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);
2233
	else if (!right_info)
2234
		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
J
Josef Bacik 已提交
2235

2236 2237 2238
	/* See try_merge_free_space() comment. */
	if (right_info && !right_info->bitmap &&
	    (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2239
		if (update_stat)
2240
			unlink_free_space(ctl, right_info);
2241
		else
2242
			__unlink_free_space(ctl, right_info);
2243
		info->bytes += right_info->bytes;
2244
		kmem_cache_free(btrfs_free_space_cachep, right_info);
2245
		merged = true;
J
Josef Bacik 已提交
2246 2247
	}

2248
	/* See try_merge_free_space() comment. */
2249
	if (left_info && !left_info->bitmap &&
2250 2251
	    left_info->offset + left_info->bytes == offset &&
	    (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2252
		if (update_stat)
2253
			unlink_free_space(ctl, left_info);
2254
		else
2255
			__unlink_free_space(ctl, left_info);
2256 2257
		info->offset = left_info->offset;
		info->bytes += left_info->bytes;
2258
		kmem_cache_free(btrfs_free_space_cachep, left_info);
2259
		merged = true;
J
Josef Bacik 已提交
2260 2261
	}

2262 2263 2264
	return merged;
}

2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286
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;

2287 2288 2289 2290
	/* See try_merge_free_space() comment. */
	if (!btrfs_free_space_trimmed(bitmap))
		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;

2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343
	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;

2344 2345 2346 2347
	/* See try_merge_free_space() comment. */
	if (!btrfs_free_space_trimmed(bitmap))
		info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;

2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394
	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);
	}
}

2395 2396
int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
			   struct btrfs_free_space_ctl *ctl,
2397 2398
			   u64 offset, u64 bytes,
			   enum btrfs_trim_state trim_state)
2399
{
2400
	struct btrfs_block_group *block_group = ctl->private;
2401 2402
	struct btrfs_free_space *info;
	int ret = 0;
D
Dennis Zhou 已提交
2403
	u64 filter_bytes = bytes;
2404

2405
	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2406 2407 2408 2409 2410
	if (!info)
		return -ENOMEM;

	info->offset = offset;
	info->bytes = bytes;
2411
	info->trim_state = trim_state;
2412
	RB_CLEAR_NODE(&info->offset_index);
2413

2414
	spin_lock(&ctl->tree_lock);
2415

2416
	if (try_merge_free_space(ctl, info, true))
2417 2418 2419 2420 2421 2422 2423
		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
	 */
2424
	ret = insert_into_bitmap(ctl, info);
2425 2426 2427 2428 2429 2430 2431
	if (ret < 0) {
		goto out;
	} else if (ret) {
		ret = 0;
		goto out;
	}
link:
2432 2433 2434 2435 2436 2437 2438 2439
	/*
	 * 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);

D
Dennis Zhou 已提交
2440 2441
	filter_bytes = max(filter_bytes, info->bytes);

2442
	ret = link_free_space(ctl, info);
J
Josef Bacik 已提交
2443
	if (ret)
2444
		kmem_cache_free(btrfs_free_space_cachep, info);
2445
out:
2446
	btrfs_discard_update_discardable(block_group);
2447
	spin_unlock(&ctl->tree_lock);
2448

J
Josef Bacik 已提交
2449
	if (ret) {
2450
		btrfs_crit(fs_info, "unable to add free space :%d", ret);
2451
		ASSERT(ret != -EEXIST);
J
Josef Bacik 已提交
2452 2453
	}

D
Dennis Zhou 已提交
2454 2455
	if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
		btrfs_discard_check_filter(block_group, filter_bytes);
2456
		btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
D
Dennis Zhou 已提交
2457
	}
2458

J
Josef Bacik 已提交
2459 2460 2461
	return ret;
}

2462
int btrfs_add_free_space(struct btrfs_block_group *block_group,
2463 2464
			 u64 bytenr, u64 size)
{
2465 2466 2467 2468 2469
	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;

	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
		trim_state = BTRFS_TRIM_STATE_TRIMMED;

2470 2471
	return __btrfs_add_free_space(block_group->fs_info,
				      block_group->free_space_ctl,
2472
				      bytenr, size, trim_state);
2473 2474
}

2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493
/*
 * This is a subtle distinction because when adding free space back in general,
 * we want it to be added as untrimmed for async. But in the case where we add
 * it on loading of a block group, we want to consider it trimmed.
 */
int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
				       u64 bytenr, u64 size)
{
	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;

	if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
	    btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
		trim_state = BTRFS_TRIM_STATE_TRIMMED;

	return __btrfs_add_free_space(block_group->fs_info,
				      block_group->free_space_ctl,
				      bytenr, size, trim_state);
}

2494
int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2495
			    u64 offset, u64 bytes)
J
Josef Bacik 已提交
2496
{
2497
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
2498
	struct btrfs_free_space *info;
2499 2500
	int ret;
	bool re_search = false;
J
Josef Bacik 已提交
2501

2502
	spin_lock(&ctl->tree_lock);
2503

2504
again:
2505
	ret = 0;
2506 2507 2508
	if (!bytes)
		goto out_lock;

2509
	info = tree_search_offset(ctl, offset, 0, 0);
2510
	if (!info) {
2511 2512 2513 2514
		/*
		 * oops didn't find an extent that matched the space we wanted
		 * to remove, look for a bitmap instead
		 */
2515
		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2516 2517
					  1, 0);
		if (!info) {
2518 2519 2520 2521
			/*
			 * 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.
2522
			 */
2523
			WARN_ON(re_search);
2524 2525
			goto out_lock;
		}
2526 2527
	}

2528
	re_search = false;
2529
	if (!info->bitmap) {
2530
		unlink_free_space(ctl, info);
2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541
		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
2542

2543 2544 2545 2546 2547
			offset += to_free;
			bytes -= to_free;
			goto again;
		} else {
			u64 old_end = info->bytes + info->offset;
2548

2549
			info->bytes = offset - info->offset;
2550
			ret = link_free_space(ctl, info);
2551 2552 2553 2554
			WARN_ON(ret);
			if (ret)
				goto out_lock;

2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565
			/* 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);

2566 2567 2568 2569
			ret = __btrfs_add_free_space(block_group->fs_info, ctl,
						     offset + bytes,
						     old_end - (offset + bytes),
						     info->trim_state);
2570 2571 2572
			WARN_ON(ret);
			goto out;
		}
J
Josef Bacik 已提交
2573
	}
2574

2575
	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2576 2577
	if (ret == -EAGAIN) {
		re_search = true;
2578
		goto again;
2579
	}
2580
out_lock:
2581
	btrfs_discard_update_discardable(block_group);
2582
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
2583
out:
2584 2585 2586
	return ret;
}

2587
void btrfs_dump_free_space(struct btrfs_block_group *block_group,
J
Josef Bacik 已提交
2588 2589
			   u64 bytes)
{
2590
	struct btrfs_fs_info *fs_info = block_group->fs_info;
2591
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
2592 2593 2594 2595
	struct btrfs_free_space *info;
	struct rb_node *n;
	int count = 0;

2596
	spin_lock(&ctl->tree_lock);
2597
	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
J
Josef Bacik 已提交
2598
		info = rb_entry(n, struct btrfs_free_space, offset_index);
L
Liu Bo 已提交
2599
		if (info->bytes >= bytes && !block_group->ro)
J
Josef Bacik 已提交
2600
			count++;
2601
		btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2602
			   info->offset, info->bytes,
2603
		       (info->bitmap) ? "yes" : "no");
J
Josef Bacik 已提交
2604
	}
2605
	spin_unlock(&ctl->tree_lock);
2606
	btrfs_info(fs_info, "block group has cluster?: %s",
2607
	       list_empty(&block_group->cluster_list) ? "no" : "yes");
2608
	btrfs_info(fs_info,
2609
		   "%d blocks of free space at or bigger than bytes is", count);
J
Josef Bacik 已提交
2610 2611
}

2612 2613
void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
			       struct btrfs_free_space_ctl *ctl)
J
Josef Bacik 已提交
2614
{
2615
	struct btrfs_fs_info *fs_info = block_group->fs_info;
J
Josef Bacik 已提交
2616

2617
	spin_lock_init(&ctl->tree_lock);
2618
	ctl->unit = fs_info->sectorsize;
2619
	ctl->start = block_group->start;
2620 2621
	ctl->private = block_group;
	ctl->op = &free_space_op;
2622 2623
	INIT_LIST_HEAD(&ctl->trimming_ranges);
	mutex_init(&ctl->cache_writeout_mutex);
J
Josef Bacik 已提交
2624

2625 2626 2627 2628 2629
	/*
	 * 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
	 */
2630
	ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
J
Josef Bacik 已提交
2631 2632
}

2633 2634 2635 2636 2637 2638
/*
 * 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
 */
2639
static void __btrfs_return_cluster_to_free_space(
2640
			     struct btrfs_block_group *block_group,
2641 2642
			     struct btrfs_free_cluster *cluster)
{
2643
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2644 2645 2646 2647 2648 2649 2650
	struct btrfs_free_space *entry;
	struct rb_node *node;

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

2651
	cluster->block_group = NULL;
2652
	cluster->window_start = 0;
2653 2654
	list_del_init(&cluster->block_group_list);

2655
	node = rb_first(&cluster->root);
2656
	while (node) {
2657 2658
		bool bitmap;

2659 2660 2661
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		node = rb_next(&entry->offset_index);
		rb_erase(&entry->offset_index, &cluster->root);
2662
		RB_CLEAR_NODE(&entry->offset_index);
2663 2664

		bitmap = (entry->bitmap != NULL);
2665
		if (!bitmap) {
2666
			/* Merging treats extents as if they were new */
2667
			if (!btrfs_free_space_trimmed(entry)) {
2668
				ctl->discardable_extents[BTRFS_STAT_CURR]--;
2669 2670 2671
				ctl->discardable_bytes[BTRFS_STAT_CURR] -=
					entry->bytes;
			}
2672

2673
			try_merge_free_space(ctl, entry, false);
2674
			steal_from_bitmap(ctl, entry, false);
2675 2676

			/* As we insert directly, update these statistics */
2677
			if (!btrfs_free_space_trimmed(entry)) {
2678
				ctl->discardable_extents[BTRFS_STAT_CURR]++;
2679 2680 2681
				ctl->discardable_bytes[BTRFS_STAT_CURR] +=
					entry->bytes;
			}
2682
		}
2683
		tree_insert_offset(&ctl->free_space_offset,
2684
				   entry->offset, &entry->offset_index, bitmap);
2685
	}
2686
	cluster->root = RB_ROOT;
2687

2688 2689
out:
	spin_unlock(&cluster->lock);
2690
	btrfs_put_block_group(block_group);
2691 2692
}

2693 2694
static void __btrfs_remove_free_space_cache_locked(
				struct btrfs_free_space_ctl *ctl)
J
Josef Bacik 已提交
2695 2696 2697
{
	struct btrfs_free_space *info;
	struct rb_node *node;
2698 2699 2700

	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
		info = rb_entry(node, struct btrfs_free_space, offset_index);
2701 2702 2703 2704 2705 2706
		if (!info->bitmap) {
			unlink_free_space(ctl, info);
			kmem_cache_free(btrfs_free_space_cachep, info);
		} else {
			free_bitmap(ctl, info);
		}
2707 2708

		cond_resched_lock(&ctl->tree_lock);
2709
	}
2710 2711 2712 2713 2714 2715
}

void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
{
	spin_lock(&ctl->tree_lock);
	__btrfs_remove_free_space_cache_locked(ctl);
2716
	if (ctl->private)
2717
		btrfs_discard_update_discardable(ctl->private);
2718 2719 2720
	spin_unlock(&ctl->tree_lock);
}

2721
void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
2722 2723
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2724
	struct btrfs_free_cluster *cluster;
2725
	struct list_head *head;
J
Josef Bacik 已提交
2726

2727
	spin_lock(&ctl->tree_lock);
2728 2729 2730 2731
	while ((head = block_group->cluster_list.next) !=
	       &block_group->cluster_list) {
		cluster = list_entry(head, struct btrfs_free_cluster,
				     block_group_list);
2732 2733 2734

		WARN_ON(cluster->block_group != block_group);
		__btrfs_return_cluster_to_free_space(block_group, cluster);
2735 2736

		cond_resched_lock(&ctl->tree_lock);
2737
	}
2738
	__btrfs_remove_free_space_cache_locked(ctl);
2739
	btrfs_discard_update_discardable(block_group);
2740
	spin_unlock(&ctl->tree_lock);
2741

J
Josef Bacik 已提交
2742 2743
}

2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774
/**
 * btrfs_is_free_space_trimmed - see if everything is trimmed
 * @block_group: block_group of interest
 *
 * Walk @block_group's free space rb_tree to determine if everything is trimmed.
 */
bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct btrfs_free_space *info;
	struct rb_node *node;
	bool ret = true;

	spin_lock(&ctl->tree_lock);
	node = rb_first(&ctl->free_space_offset);

	while (node) {
		info = rb_entry(node, struct btrfs_free_space, offset_index);

		if (!btrfs_free_space_trimmed(info)) {
			ret = false;
			break;
		}

		node = rb_next(node);
	}

	spin_unlock(&ctl->tree_lock);
	return ret;
}

2775
u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
2776 2777
			       u64 offset, u64 bytes, u64 empty_size,
			       u64 *max_extent_size)
J
Josef Bacik 已提交
2778
{
2779
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2780 2781
	struct btrfs_discard_ctl *discard_ctl =
					&block_group->fs_info->discard_ctl;
2782
	struct btrfs_free_space *entry = NULL;
2783
	u64 bytes_search = bytes + empty_size;
2784
	u64 ret = 0;
D
David Woodhouse 已提交
2785 2786
	u64 align_gap = 0;
	u64 align_gap_len = 0;
2787
	enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
J
Josef Bacik 已提交
2788

2789
	spin_lock(&ctl->tree_lock);
D
David Woodhouse 已提交
2790
	entry = find_free_space(ctl, &offset, &bytes_search,
2791
				block_group->full_stripe_len, max_extent_size);
2792
	if (!entry)
2793 2794 2795 2796
		goto out;

	ret = offset;
	if (entry->bitmap) {
2797
		bitmap_clear_bits(ctl, entry, offset, bytes);
2798 2799 2800 2801

		if (!btrfs_free_space_trimmed(entry))
			atomic64_add(bytes, &discard_ctl->discard_bytes_saved);

2802
		if (!entry->bytes)
2803
			free_bitmap(ctl, entry);
2804
	} else {
2805
		unlink_free_space(ctl, entry);
D
David Woodhouse 已提交
2806 2807
		align_gap_len = offset - entry->offset;
		align_gap = entry->offset;
2808
		align_gap_trim_state = entry->trim_state;
D
David Woodhouse 已提交
2809

2810 2811 2812
		if (!btrfs_free_space_trimmed(entry))
			atomic64_add(bytes, &discard_ctl->discard_bytes_saved);

D
David Woodhouse 已提交
2813 2814 2815 2816
		entry->offset = offset + bytes;
		WARN_ON(entry->bytes < bytes + align_gap_len);

		entry->bytes -= bytes + align_gap_len;
2817
		if (!entry->bytes)
2818
			kmem_cache_free(btrfs_free_space_cachep, entry);
2819
		else
2820
			link_free_space(ctl, entry);
2821
	}
2822
out:
2823
	btrfs_discard_update_discardable(block_group);
2824
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
2825

D
David Woodhouse 已提交
2826
	if (align_gap_len)
2827
		__btrfs_add_free_space(block_group->fs_info, ctl,
2828 2829
				       align_gap, align_gap_len,
				       align_gap_trim_state);
J
Josef Bacik 已提交
2830 2831
	return ret;
}
2832 2833 2834 2835 2836 2837 2838 2839 2840

/*
 * 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.
 */
2841
void btrfs_return_cluster_to_free_space(
2842
			       struct btrfs_block_group *block_group,
2843 2844
			       struct btrfs_free_cluster *cluster)
{
2845
	struct btrfs_free_space_ctl *ctl;
2846 2847 2848 2849 2850 2851 2852

	/* 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);
2853
			return;
2854 2855 2856 2857
		}
	} else if (cluster->block_group != block_group) {
		/* someone else has already freed it don't redo their work */
		spin_unlock(&cluster->lock);
2858
		return;
2859
	}
2860
	btrfs_get_block_group(block_group);
2861 2862
	spin_unlock(&cluster->lock);

2863 2864
	ctl = block_group->free_space_ctl;

2865
	/* now return any extents the cluster had on it */
2866
	spin_lock(&ctl->tree_lock);
2867
	__btrfs_return_cluster_to_free_space(block_group, cluster);
2868
	spin_unlock(&ctl->tree_lock);
2869

2870 2871
	btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);

2872 2873 2874 2875
	/* finally drop our ref */
	btrfs_put_block_group(block_group);
}

2876
static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
2877
				   struct btrfs_free_cluster *cluster,
2878
				   struct btrfs_free_space *entry,
2879 2880
				   u64 bytes, u64 min_start,
				   u64 *max_extent_size)
2881
{
2882
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2883 2884 2885 2886 2887 2888 2889 2890
	int err;
	u64 search_start = cluster->window_start;
	u64 search_bytes = bytes;
	u64 ret = 0;

	search_start = min_start;
	search_bytes = bytes;

2891
	err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2892
	if (err) {
J
Josef Bacik 已提交
2893 2894
		*max_extent_size = max(get_max_extent_size(entry),
				       *max_extent_size);
2895
		return 0;
2896
	}
2897 2898

	ret = search_start;
2899
	__bitmap_clear_bits(ctl, entry, ret, bytes);
2900 2901 2902 2903

	return ret;
}

2904 2905 2906 2907 2908
/*
 * 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
 */
2909
u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
2910
			     struct btrfs_free_cluster *cluster, u64 bytes,
2911
			     u64 min_start, u64 *max_extent_size)
2912
{
2913
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2914 2915
	struct btrfs_discard_ctl *discard_ctl =
					&block_group->fs_info->discard_ctl;
2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931
	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);
2932
	while (1) {
J
Josef Bacik 已提交
2933 2934 2935
		if (entry->bytes < bytes)
			*max_extent_size = max(get_max_extent_size(entry),
					       *max_extent_size);
2936

2937 2938
		if (entry->bytes < bytes ||
		    (!entry->bitmap && entry->offset < min_start)) {
2939 2940 2941 2942 2943 2944 2945 2946
			node = rb_next(&entry->offset_index);
			if (!node)
				break;
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
			continue;
		}

2947 2948 2949
		if (entry->bitmap) {
			ret = btrfs_alloc_from_bitmap(block_group,
						      cluster, entry, bytes,
2950 2951
						      cluster->window_start,
						      max_extent_size);
2952 2953 2954 2955 2956 2957 2958 2959
			if (ret == 0) {
				node = rb_next(&entry->offset_index);
				if (!node)
					break;
				entry = rb_entry(node, struct btrfs_free_space,
						 offset_index);
				continue;
			}
2960
			cluster->window_start += bytes;
2961 2962 2963 2964 2965 2966
		} else {
			ret = entry->offset;

			entry->offset += bytes;
			entry->bytes -= bytes;
		}
2967

2968
		if (entry->bytes == 0)
2969 2970 2971 2972 2973
			rb_erase(&entry->offset_index, &cluster->root);
		break;
	}
out:
	spin_unlock(&cluster->lock);
2974

2975 2976 2977
	if (!ret)
		return 0;

2978
	spin_lock(&ctl->tree_lock);
2979

2980 2981 2982
	if (!btrfs_free_space_trimmed(entry))
		atomic64_add(bytes, &discard_ctl->discard_bytes_saved);

2983
	ctl->free_space -= bytes;
2984 2985
	if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
		ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
2986
	if (entry->bytes == 0) {
2987
		ctl->free_extents--;
2988
		if (entry->bitmap) {
2989 2990
			kmem_cache_free(btrfs_free_space_bitmap_cachep,
					entry->bitmap);
2991
			ctl->total_bitmaps--;
2992
			recalculate_thresholds(ctl);
2993 2994
		} else if (!btrfs_free_space_trimmed(entry)) {
			ctl->discardable_extents[BTRFS_STAT_CURR]--;
2995
		}
2996
		kmem_cache_free(btrfs_free_space_cachep, entry);
2997 2998
	}

2999
	spin_unlock(&ctl->tree_lock);
3000

3001 3002 3003
	return ret;
}

3004
static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3005 3006
				struct btrfs_free_space *entry,
				struct btrfs_free_cluster *cluster,
3007 3008
				u64 offset, u64 bytes,
				u64 cont1_bytes, u64 min_bytes)
3009
{
3010
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3011 3012
	unsigned long next_zero;
	unsigned long i;
3013 3014
	unsigned long want_bits;
	unsigned long min_bits;
3015
	unsigned long found_bits;
3016
	unsigned long max_bits = 0;
3017 3018
	unsigned long start = 0;
	unsigned long total_found = 0;
3019
	int ret;
3020

3021
	i = offset_to_bit(entry->offset, ctl->unit,
3022
			  max_t(u64, offset, entry->offset));
3023 3024
	want_bits = bytes_to_bits(bytes, ctl->unit);
	min_bits = bytes_to_bits(min_bytes, ctl->unit);
3025

3026 3027 3028 3029 3030 3031 3032
	/*
	 * Don't bother looking for a cluster in this bitmap if it's heavily
	 * fragmented.
	 */
	if (entry->max_extent_size &&
	    entry->max_extent_size < cont1_bytes)
		return -ENOSPC;
3033 3034
again:
	found_bits = 0;
3035
	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3036 3037
		next_zero = find_next_zero_bit(entry->bitmap,
					       BITS_PER_BITMAP, i);
3038
		if (next_zero - i >= min_bits) {
3039
			found_bits = next_zero - i;
3040 3041
			if (found_bits > max_bits)
				max_bits = found_bits;
3042 3043
			break;
		}
3044 3045
		if (next_zero - i > max_bits)
			max_bits = next_zero - i;
3046 3047 3048
		i = next_zero;
	}

3049 3050
	if (!found_bits) {
		entry->max_extent_size = (u64)max_bits * ctl->unit;
3051
		return -ENOSPC;
3052
	}
3053

3054
	if (!total_found) {
3055
		start = i;
3056
		cluster->max_size = 0;
3057 3058 3059 3060
	}

	total_found += found_bits;

3061 3062
	if (cluster->max_size < found_bits * ctl->unit)
		cluster->max_size = found_bits * ctl->unit;
3063

3064 3065
	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
		i = next_zero + 1;
3066 3067 3068
		goto again;
	}

3069
	cluster->window_start = start * ctl->unit + entry->offset;
3070
	rb_erase(&entry->offset_index, &ctl->free_space_offset);
3071 3072
	ret = tree_insert_offset(&cluster->root, entry->offset,
				 &entry->offset_index, 1);
3073
	ASSERT(!ret); /* -EEXIST; Logic error */
3074

J
Josef Bacik 已提交
3075
	trace_btrfs_setup_cluster(block_group, cluster,
3076
				  total_found * ctl->unit, 1);
3077 3078 3079
	return 0;
}

3080 3081
/*
 * This searches the block group for just extents to fill the cluster with.
3082 3083
 * 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.
3084
 */
3085
static noinline int
3086
setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3087 3088
			struct btrfs_free_cluster *cluster,
			struct list_head *bitmaps, u64 offset, u64 bytes,
3089
			u64 cont1_bytes, u64 min_bytes)
3090
{
3091
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3092 3093 3094 3095 3096 3097
	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
3098
	u64 total_size = 0;
3099

3100
	entry = tree_search_offset(ctl, offset, 0, 1);
3101 3102 3103 3104 3105 3106 3107
	if (!entry)
		return -ENOSPC;

	/*
	 * We don't want bitmaps, so just move along until we find a normal
	 * extent entry.
	 */
3108 3109
	while (entry->bitmap || entry->bytes < min_bytes) {
		if (entry->bitmap && list_empty(&entry->list))
3110
			list_add_tail(&entry->list, bitmaps);
3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121
		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;

3122 3123
	for (node = rb_next(&entry->offset_index); node;
	     node = rb_next(&entry->offset_index)) {
3124 3125
		entry = rb_entry(node, struct btrfs_free_space, offset_index);

3126 3127 3128
		if (entry->bitmap) {
			if (list_empty(&entry->list))
				list_add_tail(&entry->list, bitmaps);
3129
			continue;
3130 3131
		}

3132 3133 3134 3135 3136 3137
		if (entry->bytes < min_bytes)
			continue;

		last = entry;
		window_free += entry->bytes;
		if (entry->bytes > max_extent)
3138 3139 3140
			max_extent = entry->bytes;
	}

3141 3142 3143
	if (window_free < bytes || max_extent < cont1_bytes)
		return -ENOSPC;

3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156
	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);
3157
		if (entry->bitmap || entry->bytes < min_bytes)
3158 3159
			continue;

3160
		rb_erase(&entry->offset_index, &ctl->free_space_offset);
3161 3162
		ret = tree_insert_offset(&cluster->root, entry->offset,
					 &entry->offset_index, 0);
J
Josef Bacik 已提交
3163
		total_size += entry->bytes;
3164
		ASSERT(!ret); /* -EEXIST; Logic error */
3165 3166 3167
	} while (node && entry != last);

	cluster->max_size = max_extent;
J
Josef Bacik 已提交
3168
	trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3169 3170 3171 3172 3173 3174 3175
	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.
 */
3176
static noinline int
3177
setup_cluster_bitmap(struct btrfs_block_group *block_group,
3178 3179
		     struct btrfs_free_cluster *cluster,
		     struct list_head *bitmaps, u64 offset, u64 bytes,
3180
		     u64 cont1_bytes, u64 min_bytes)
3181
{
3182
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3183
	struct btrfs_free_space *entry = NULL;
3184
	int ret = -ENOSPC;
3185
	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3186

3187
	if (ctl->total_bitmaps == 0)
3188 3189
		return -ENOSPC;

3190 3191 3192 3193
	/*
	 * The bitmap that covers offset won't be in the list unless offset
	 * is just its start offset.
	 */
3194 3195 3196 3197
	if (!list_empty(bitmaps))
		entry = list_first_entry(bitmaps, struct btrfs_free_space, list);

	if (!entry || entry->offset != bitmap_offset) {
3198 3199 3200 3201 3202
		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
		if (entry && list_empty(&entry->list))
			list_add(&entry->list, bitmaps);
	}

3203
	list_for_each_entry(entry, bitmaps, list) {
3204
		if (entry->bytes < bytes)
3205 3206
			continue;
		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3207
					   bytes, cont1_bytes, min_bytes);
3208 3209 3210 3211 3212
		if (!ret)
			return 0;
	}

	/*
3213 3214
	 * The bitmaps list has all the bitmaps that record free space
	 * starting after offset, so no more search is required.
3215
	 */
3216
	return -ENOSPC;
3217 3218
}

3219 3220
/*
 * here we try to find a cluster of blocks in a block group.  The goal
3221
 * is to find at least bytes+empty_size.
3222 3223 3224 3225 3226
 * We might not find them all in one contiguous area.
 *
 * returns zero and sets up cluster if things worked out, otherwise
 * it returns -enospc
 */
3227
int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3228 3229 3230
			     struct btrfs_free_cluster *cluster,
			     u64 offset, u64 bytes, u64 empty_size)
{
3231
	struct btrfs_fs_info *fs_info = block_group->fs_info;
3232
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3233
	struct btrfs_free_space *entry, *tmp;
3234
	LIST_HEAD(bitmaps);
3235
	u64 min_bytes;
3236
	u64 cont1_bytes;
3237 3238
	int ret;

3239 3240 3241 3242 3243 3244
	/*
	 * 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.
	 */
3245
	if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3246
		cont1_bytes = min_bytes = bytes + empty_size;
3247
	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3248
		cont1_bytes = bytes;
3249
		min_bytes = fs_info->sectorsize;
3250 3251
	} else {
		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3252
		min_bytes = fs_info->sectorsize;
3253
	}
3254

3255
	spin_lock(&ctl->tree_lock);
3256 3257 3258 3259 3260

	/*
	 * 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.
	 */
3261
	if (ctl->free_space < bytes) {
3262
		spin_unlock(&ctl->tree_lock);
3263 3264 3265
		return -ENOSPC;
	}

3266 3267 3268 3269 3270 3271 3272 3273
	spin_lock(&cluster->lock);

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

J
Josef Bacik 已提交
3274 3275 3276
	trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
				 min_bytes);

3277
	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3278 3279
				      bytes + empty_size,
				      cont1_bytes, min_bytes);
3280
	if (ret)
3281
		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3282 3283
					   offset, bytes + empty_size,
					   cont1_bytes, min_bytes);
3284 3285 3286 3287

	/* Clear our temporary list */
	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
		list_del_init(&entry->list);
3288

3289
	if (!ret) {
3290
		btrfs_get_block_group(block_group);
3291 3292 3293
		list_add_tail(&cluster->block_group_list,
			      &block_group->cluster_list);
		cluster->block_group = block_group;
J
Josef Bacik 已提交
3294 3295
	} else {
		trace_btrfs_failed_cluster_setup(block_group);
3296 3297 3298
	}
out:
	spin_unlock(&cluster->lock);
3299
	spin_unlock(&ctl->tree_lock);
3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310

	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);
3311
	cluster->root = RB_ROOT;
3312
	cluster->max_size = 0;
3313
	cluster->fragmented = false;
3314 3315 3316 3317
	INIT_LIST_HEAD(&cluster->block_group_list);
	cluster->block_group = NULL;
}

3318
static int do_trimming(struct btrfs_block_group *block_group,
3319
		       u64 *total_trimmed, u64 start, u64 bytes,
3320
		       u64 reserved_start, u64 reserved_bytes,
3321
		       enum btrfs_trim_state reserved_trim_state,
3322
		       struct btrfs_trim_range *trim_entry)
3323
{
3324
	struct btrfs_space_info *space_info = block_group->space_info;
3325
	struct btrfs_fs_info *fs_info = block_group->fs_info;
3326
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3327 3328
	int ret;
	int update = 0;
3329 3330 3331
	const u64 end = start + bytes;
	const u64 reserved_end = reserved_start + reserved_bytes;
	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3332
	u64 trimmed = 0;
3333

3334 3335 3336 3337 3338 3339 3340 3341 3342 3343
	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);

3344
	ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3345
	if (!ret) {
3346
		*total_trimmed += trimmed;
3347 3348
		trim_state = BTRFS_TRIM_STATE_TRIMMED;
	}
3349

3350
	mutex_lock(&ctl->cache_writeout_mutex);
3351 3352 3353 3354 3355 3356 3357 3358
	if (reserved_start < start)
		__btrfs_add_free_space(fs_info, ctl, reserved_start,
				       start - reserved_start,
				       reserved_trim_state);
	if (start + bytes < reserved_start + reserved_bytes)
		__btrfs_add_free_space(fs_info, ctl, end, reserved_end - end,
				       reserved_trim_state);
	__btrfs_add_free_space(fs_info, ctl, start, bytes, trim_state);
3359 3360
	list_del(&trim_entry->list);
	mutex_unlock(&ctl->cache_writeout_mutex);
3361 3362 3363 3364 3365 3366 3367 3368 3369

	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(&block_group->lock);
3370
		spin_unlock(&space_info->lock);
3371 3372 3373 3374 3375
	}

	return ret;
}

3376 3377 3378
/*
 * If @async is set, then we will trim 1 region and return.
 */
3379
static int trim_no_bitmap(struct btrfs_block_group *block_group,
3380 3381
			  u64 *total_trimmed, u64 start, u64 end, u64 minlen,
			  bool async)
3382
{
3383 3384
	struct btrfs_discard_ctl *discard_ctl =
					&block_group->fs_info->discard_ctl;
3385 3386 3387 3388 3389 3390
	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;
3391
	enum btrfs_trim_state extent_trim_state;
3392
	u64 bytes;
3393
	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3394 3395

	while (start < end) {
3396 3397 3398
		struct btrfs_trim_range trim_entry;

		mutex_lock(&ctl->cache_writeout_mutex);
3399
		spin_lock(&ctl->tree_lock);
3400

3401 3402
		if (ctl->free_space < minlen)
			goto out_unlock;
3403

3404
		entry = tree_search_offset(ctl, start, 0, 1);
3405 3406
		if (!entry)
			goto out_unlock;
3407

3408 3409 3410
		/* Skip bitmaps and if async, already trimmed entries */
		while (entry->bitmap ||
		       (async && btrfs_free_space_trimmed(entry))) {
3411
			node = rb_next(&entry->offset_index);
3412 3413
			if (!node)
				goto out_unlock;
3414 3415
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
3416 3417
		}

3418 3419
		if (entry->offset >= end)
			goto out_unlock;
3420

3421 3422
		extent_start = entry->offset;
		extent_bytes = entry->bytes;
3423
		extent_trim_state = entry->trim_state;
3424 3425 3426 3427 3428 3429 3430 3431 3432
		if (async) {
			start = entry->offset;
			bytes = entry->bytes;
			if (bytes < minlen) {
				spin_unlock(&ctl->tree_lock);
				mutex_unlock(&ctl->cache_writeout_mutex);
				goto next;
			}
			unlink_free_space(ctl, entry);
D
Dennis Zhou 已提交
3433 3434 3435 3436 3437 3438 3439 3440
			/*
			 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
			 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
			 * X when we come back around.  So trim it now.
			 */
			if (max_discard_size &&
			    bytes >= (max_discard_size +
				      BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3441 3442 3443 3444
				bytes = max_discard_size;
				extent_bytes = max_discard_size;
				entry->offset += max_discard_size;
				entry->bytes -= max_discard_size;
3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456
				link_free_space(ctl, entry);
			} else {
				kmem_cache_free(btrfs_free_space_cachep, entry);
			}
		} else {
			start = max(start, extent_start);
			bytes = min(extent_start + extent_bytes, end) - start;
			if (bytes < minlen) {
				spin_unlock(&ctl->tree_lock);
				mutex_unlock(&ctl->cache_writeout_mutex);
				goto next;
			}
3457

3458 3459 3460
			unlink_free_space(ctl, entry);
			kmem_cache_free(btrfs_free_space_cachep, entry);
		}
3461

3462
		spin_unlock(&ctl->tree_lock);
3463 3464 3465 3466
		trim_entry.start = extent_start;
		trim_entry.bytes = extent_bytes;
		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
		mutex_unlock(&ctl->cache_writeout_mutex);
3467

3468
		ret = do_trimming(block_group, total_trimmed, start, bytes,
3469 3470
				  extent_start, extent_bytes, extent_trim_state,
				  &trim_entry);
3471 3472
		if (ret) {
			block_group->discard_cursor = start + bytes;
3473
			break;
3474
		}
3475 3476
next:
		start += bytes;
3477 3478 3479
		block_group->discard_cursor = start;
		if (async && *total_trimmed)
			break;
3480

3481 3482 3483 3484 3485 3486 3487
		if (fatal_signal_pending(current)) {
			ret = -ERESTARTSYS;
			break;
		}

		cond_resched();
	}
3488 3489 3490 3491 3492 3493 3494 3495

	return ret;

out_unlock:
	block_group->discard_cursor = btrfs_block_group_end(block_group);
	spin_unlock(&ctl->tree_lock);
	mutex_unlock(&ctl->cache_writeout_mutex);

3496 3497 3498
	return ret;
}

3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518
/*
 * If we break out of trimming a bitmap prematurely, we should reset the
 * trimming bit.  In a rather contrieved case, it's possible to race here so
 * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
 *
 * start = start of bitmap
 * end = near end of bitmap
 *
 * Thread 1:			Thread 2:
 * trim_bitmaps(start)
 *				trim_bitmaps(end)
 *				end_trimming_bitmap()
 * reset_trimming_bitmap()
 */
static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
{
	struct btrfs_free_space *entry;

	spin_lock(&ctl->tree_lock);
	entry = tree_search_offset(ctl, offset, 1, 0);
3519
	if (entry) {
3520
		if (btrfs_free_space_trimmed(entry)) {
3521 3522
			ctl->discardable_extents[BTRFS_STAT_CURR] +=
				entry->bitmap_extents;
3523 3524
			ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
		}
3525
		entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3526 3527
	}

3528 3529 3530
	spin_unlock(&ctl->tree_lock);
}

3531 3532
static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
				struct btrfs_free_space *entry)
3533
{
3534
	if (btrfs_free_space_trimming_bitmap(entry)) {
3535
		entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3536 3537
		ctl->discardable_extents[BTRFS_STAT_CURR] -=
			entry->bitmap_extents;
3538
		ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3539
	}
3540 3541
}

3542 3543 3544
/*
 * If @async is set, then we will trim 1 region and return.
 */
3545
static int trim_bitmaps(struct btrfs_block_group *block_group,
3546
			u64 *total_trimmed, u64 start, u64 end, u64 minlen,
D
Dennis Zhou 已提交
3547
			u64 maxlen, bool async)
3548
{
3549 3550
	struct btrfs_discard_ctl *discard_ctl =
					&block_group->fs_info->discard_ctl;
3551 3552 3553 3554 3555 3556
	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);
3557
	const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3558 3559 3560

	while (offset < end) {
		bool next_bitmap = false;
3561
		struct btrfs_trim_range trim_entry;
3562

3563
		mutex_lock(&ctl->cache_writeout_mutex);
3564 3565 3566
		spin_lock(&ctl->tree_lock);

		if (ctl->free_space < minlen) {
3567 3568
			block_group->discard_cursor =
				btrfs_block_group_end(block_group);
3569
			spin_unlock(&ctl->tree_lock);
3570
			mutex_unlock(&ctl->cache_writeout_mutex);
3571 3572 3573 3574
			break;
		}

		entry = tree_search_offset(ctl, offset, 1, 0);
D
Dennis Zhou 已提交
3575 3576 3577 3578 3579 3580 3581 3582 3583
		/*
		 * Bitmaps are marked trimmed lossily now to prevent constant
		 * discarding of the same bitmap (the reason why we are bound
		 * by the filters).  So, retrim the block group bitmaps when we
		 * are preparing to punt to the unused_bgs list.  This uses
		 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
		 * which is the only discard index which sets minlen to 0.
		 */
		if (!entry || (async && minlen && start == offset &&
3584
			       btrfs_free_space_trimmed(entry))) {
3585
			spin_unlock(&ctl->tree_lock);
3586
			mutex_unlock(&ctl->cache_writeout_mutex);
3587 3588 3589 3590
			next_bitmap = true;
			goto next;
		}

3591 3592 3593 3594 3595 3596 3597 3598 3599
		/*
		 * Async discard bitmap trimming begins at by setting the start
		 * to be key.objectid and the offset_to_bitmap() aligns to the
		 * start of the bitmap.  This lets us know we are fully
		 * scanning the bitmap rather than only some portion of it.
		 */
		if (start == offset)
			entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;

3600
		bytes = minlen;
3601
		ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3602
		if (ret2 || start >= end) {
3603
			/*
D
Dennis Zhou 已提交
3604 3605
			 * We lossily consider a bitmap trimmed if we only skip
			 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3606
			 */
D
Dennis Zhou 已提交
3607
			if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3608
				end_trimming_bitmap(ctl, entry);
3609 3610
			else
				entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3611
			spin_unlock(&ctl->tree_lock);
3612
			mutex_unlock(&ctl->cache_writeout_mutex);
3613 3614 3615 3616
			next_bitmap = true;
			goto next;
		}

3617 3618 3619 3620 3621 3622 3623 3624 3625 3626
		/*
		 * We already trimmed a region, but are using the locking above
		 * to reset the trim_state.
		 */
		if (async && *total_trimmed) {
			spin_unlock(&ctl->tree_lock);
			mutex_unlock(&ctl->cache_writeout_mutex);
			goto out;
		}

3627
		bytes = min(bytes, end - start);
D
Dennis Zhou 已提交
3628
		if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3629
			spin_unlock(&ctl->tree_lock);
3630
			mutex_unlock(&ctl->cache_writeout_mutex);
3631 3632 3633
			goto next;
		}

D
Dennis Zhou 已提交
3634 3635 3636 3637 3638 3639 3640 3641 3642
		/*
		 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
		 * If X < @minlen, we won't trim X when we come back around.
		 * So trim it now.  We differ here from trimming extents as we
		 * don't keep individual state per bit.
		 */
		if (async &&
		    max_discard_size &&
		    bytes > (max_discard_size + minlen))
3643
			bytes = max_discard_size;
3644

3645 3646 3647 3648 3649
		bitmap_clear_bits(ctl, entry, start, bytes);
		if (entry->bytes == 0)
			free_bitmap(ctl, entry);

		spin_unlock(&ctl->tree_lock);
3650 3651 3652 3653
		trim_entry.start = start;
		trim_entry.bytes = bytes;
		list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
		mutex_unlock(&ctl->cache_writeout_mutex);
3654 3655

		ret = do_trimming(block_group, total_trimmed, start, bytes,
3656
				  start, bytes, 0, &trim_entry);
3657 3658
		if (ret) {
			reset_trimming_bitmap(ctl, offset);
3659 3660
			block_group->discard_cursor =
				btrfs_block_group_end(block_group);
3661
			break;
3662
		}
3663 3664 3665
next:
		if (next_bitmap) {
			offset += BITS_PER_BITMAP * ctl->unit;
3666
			start = offset;
3667 3668
		} else {
			start += bytes;
3669
		}
3670
		block_group->discard_cursor = start;
3671 3672

		if (fatal_signal_pending(current)) {
3673 3674
			if (start != offset)
				reset_trimming_bitmap(ctl, offset);
3675 3676 3677 3678 3679 3680 3681
			ret = -ERESTARTSYS;
			break;
		}

		cond_resched();
	}

3682 3683 3684 3685
	if (offset >= end)
		block_group->discard_cursor = end;

out:
3686 3687
	return ret;
}
3688

3689
int btrfs_trim_block_group(struct btrfs_block_group *block_group,
3690 3691
			   u64 *trimmed, u64 start, u64 end, u64 minlen)
{
3692
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3693
	int ret;
3694
	u64 rem = 0;
3695 3696 3697 3698 3699

	*trimmed = 0;

	spin_lock(&block_group->lock);
	if (block_group->removed) {
3700
		spin_unlock(&block_group->lock);
3701
		return 0;
3702
	}
3703
	btrfs_freeze_block_group(block_group);
3704 3705
	spin_unlock(&block_group->lock);

3706
	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
3707 3708
	if (ret)
		goto out;
3709

D
Dennis Zhou 已提交
3710
	ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
3711 3712 3713 3714
	div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
	/* If we ended in the middle of a bitmap, reset the trimming flag */
	if (rem)
		reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
3715
out:
3716
	btrfs_unfreeze_block_group(block_group);
3717 3718 3719
	return ret;
}

3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732
int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
				   u64 *trimmed, u64 start, u64 end, u64 minlen,
				   bool async)
{
	int ret;

	*trimmed = 0;

	spin_lock(&block_group->lock);
	if (block_group->removed) {
		spin_unlock(&block_group->lock);
		return 0;
	}
3733
	btrfs_freeze_block_group(block_group);
3734 3735 3736
	spin_unlock(&block_group->lock);

	ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
3737
	btrfs_unfreeze_block_group(block_group);
3738 3739 3740 3741 3742 3743

	return ret;
}

int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
				   u64 *trimmed, u64 start, u64 end, u64 minlen,
D
Dennis Zhou 已提交
3744
				   u64 maxlen, bool async)
3745 3746 3747 3748 3749 3750 3751 3752 3753 3754
{
	int ret;

	*trimmed = 0;

	spin_lock(&block_group->lock);
	if (block_group->removed) {
		spin_unlock(&block_group->lock);
		return 0;
	}
3755
	btrfs_freeze_block_group(block_group);
3756 3757
	spin_unlock(&block_group->lock);

D
Dennis Zhou 已提交
3758 3759 3760
	ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
			   async);

3761
	btrfs_unfreeze_block_group(block_group);
3762 3763 3764 3765

	return ret;
}

3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793
bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
{
	return btrfs_super_cache_generation(fs_info->super_copy);
}

int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
{
	struct btrfs_trans_handle *trans;
	int ret;

	/*
	 * update_super_roots will appropriately set
	 * super_copy->cache_generation based on the SPACE_CACHE option, so all
	 * we have to do is trigger a transaction commit.
	 */
	trans = btrfs_start_transaction(fs_info->tree_root, 0);
	if (IS_ERR(trans))
		return PTR_ERR(trans);

	if (!active)
		set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);

	ret = btrfs_commit_transaction(trans);
	clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);

	return ret;
}

3794
#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3795 3796 3797 3798 3799 3800
/*
 * 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.
 */
3801
int test_add_free_space_entry(struct btrfs_block_group *cache,
3802
			      u64 offset, u64 bytes, bool bitmap)
3803
{
3804 3805 3806
	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
	struct btrfs_free_space *info = NULL, *bitmap_info;
	void *map = NULL;
3807
	enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
3808 3809
	u64 bytes_added;
	int ret;
3810

3811 3812 3813 3814 3815
again:
	if (!info) {
		info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
		if (!info)
			return -ENOMEM;
3816 3817
	}

3818 3819 3820 3821
	if (!bitmap) {
		spin_lock(&ctl->tree_lock);
		info->offset = offset;
		info->bytes = bytes;
3822
		info->max_extent_size = 0;
3823 3824 3825 3826 3827 3828 3829 3830
		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) {
3831
		map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845
		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;
3846
		info = NULL;
3847
	}
3848

3849 3850
	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
					  trim_state);
3851

3852 3853 3854
	bytes -= bytes_added;
	offset += bytes_added;
	spin_unlock(&ctl->tree_lock);
3855

3856 3857
	if (bytes)
		goto again;
3858

3859 3860
	if (info)
		kmem_cache_free(btrfs_free_space_cachep, info);
3861 3862
	if (map)
		kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
3863
	return 0;
3864 3865 3866 3867 3868 3869 3870
}

/*
 * 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.
 */
3871
int test_check_exists(struct btrfs_block_group *cache,
3872
		      u64 offset, u64 bytes)
3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894
{
	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;
3895
		ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913
		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) {
3914
				n = rb_prev(&tmp->offset_index);
3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927
				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) {
3928
				n = rb_next(&tmp->offset_index);
3929 3930 3931 3932 3933 3934
				continue;
			}
			info = tmp;
			goto have_info;
		}

3935
		ret = 0;
3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949
		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;
}
3950
#endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */