free-space-cache.c 85.2 KB
Newer Older
J
Josef Bacik 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2008 Red Hat.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

	return inode;
}

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

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

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

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

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

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

	return 0;
}

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

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

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

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

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

210 211 212 213 214 215
	spin_lock(&rsv->lock);
	if (rsv->reserved < needed_bytes)
		ret = -ENOSPC;
	else
		ret = 0;
	spin_unlock(&rsv->lock);
216
	return ret;
217 218 219 220 221 222 223 224 225
}

int btrfs_truncate_free_space_cache(struct btrfs_root *root,
				    struct btrfs_trans_handle *trans,
				    struct btrfs_path *path,
				    struct inode *inode)
{
	loff_t oldsize;
	int ret = 0;
226 227 228 229 230 231 232 233 234 235 236 237

	oldsize = i_size_read(inode);
	btrfs_i_size_write(inode, 0);
	truncate_pagecache(inode, oldsize, 0);

	/*
	 * We don't need an orphan item because truncating the free space cache
	 * will never be split across transactions.
	 */
	ret = btrfs_truncate_inode_items(trans, root, inode,
					 0, BTRFS_EXTENT_DATA_KEY);
	if (ret) {
238
		btrfs_abort_transaction(trans, root, ret);
239 240 241
		return ret;
	}

242
	ret = btrfs_update_inode(trans, root, inode);
243 244
	if (ret)
		btrfs_abort_transaction(trans, root, ret);
245

246
	return ret;
247 248
}

249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267
static int readahead_cache(struct inode *inode)
{
	struct file_ra_state *ra;
	unsigned long last_index;

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

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

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

	kfree(ra);

	return 0;
}

268 269 270 271 272 273 274 275
struct io_ctl {
	void *cur, *orig;
	struct page *page;
	struct page **pages;
	struct btrfs_root *root;
	unsigned long size;
	int index;
	int num_pages;
276
	unsigned check_crcs:1;
277 278 279 280 281 282 283 284 285 286 287 288 289
};

static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
		       struct btrfs_root *root)
{
	memset(io_ctl, 0, sizeof(struct io_ctl));
	io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
		PAGE_CACHE_SHIFT;
	io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
				GFP_NOFS);
	if (!io_ctl->pages)
		return -ENOMEM;
	io_ctl->root = root;
290 291
	if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
		io_ctl->check_crcs = 1;
292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326
	return 0;
}

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

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

static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
{
	BUG_ON(io_ctl->index >= io_ctl->num_pages);
	io_ctl->page = io_ctl->pages[io_ctl->index++];
	io_ctl->cur = kmap(io_ctl->page);
	io_ctl->orig = io_ctl->cur;
	io_ctl->size = PAGE_CACHE_SIZE;
	if (clear)
		memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
}

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

	io_ctl_unmap_page(io_ctl);

	for (i = 0; i < io_ctl->num_pages; i++) {
327 328 329 330 331
		if (io_ctl->pages[i]) {
			ClearPageChecked(io_ctl->pages[i]);
			unlock_page(io_ctl->pages[i]);
			page_cache_release(io_ctl->pages[i]);
		}
332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
	}
}

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

	for (i = 0; i < io_ctl->num_pages; i++) {
		page = find_or_create_page(inode->i_mapping, i, mask);
		if (!page) {
			io_ctl_drop_pages(io_ctl);
			return -ENOMEM;
		}
		io_ctl->pages[i] = page;
		if (uptodate && !PageUptodate(page)) {
			btrfs_readpage(NULL, page);
			lock_page(page);
			if (!PageUptodate(page)) {
				printk(KERN_ERR "btrfs: error reading free "
				       "space cache\n");
				io_ctl_drop_pages(io_ctl);
				return -EIO;
			}
		}
	}

361 362 363 364 365
	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]);
	}

366 367 368 369 370
	return 0;
}

static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
{
A
Al Viro 已提交
371
	__le64 *val;
372 373 374 375

	io_ctl_map_page(io_ctl, 1);

	/*
376 377
	 * Skip the csum areas.  If we don't check crcs then we just have a
	 * 64bit chunk at the front of the first page.
378
	 */
379 380 381 382 383 384 385
	if (io_ctl->check_crcs) {
		io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
		io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
	} else {
		io_ctl->cur += sizeof(u64);
		io_ctl->size -= sizeof(u64) * 2;
	}
386 387 388 389 390 391 392 393

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

static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
{
A
Al Viro 已提交
394
	__le64 *gen;
395

396 397 398 399 400 401 402 403 404 405 406 407
	/*
	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
	 * chunk at the front of the first page.
	 */
	if (io_ctl->check_crcs) {
		io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
		io_ctl->size -= sizeof(u64) +
			(sizeof(u32) * io_ctl->num_pages);
	} else {
		io_ctl->cur += sizeof(u64);
		io_ctl->size -= sizeof(u64) * 2;
	}
408 409 410 411 412 413 414 415 416 417

	gen = io_ctl->cur;
	if (le64_to_cpu(*gen) != generation) {
		printk_ratelimited(KERN_ERR "btrfs: space cache generation "
				   "(%Lu) does not match inode (%Lu)\n", *gen,
				   generation);
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}
	io_ctl->cur += sizeof(u64);
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
	return 0;
}

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

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

	if (index == 0)
433
		offset = sizeof(u32) * io_ctl->num_pages;
434

435
	crc = btrfs_csum_data(io_ctl->orig + offset, crc,
436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464
			      PAGE_CACHE_SIZE - offset);
	btrfs_csum_final(crc, (char *)&crc);
	io_ctl_unmap_page(io_ctl);
	tmp = kmap(io_ctl->pages[0]);
	tmp += index;
	*tmp = crc;
	kunmap(io_ctl->pages[0]);
}

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

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

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

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

	io_ctl_map_page(io_ctl, 0);
465
	crc = btrfs_csum_data(io_ctl->orig + offset, crc,
466 467 468 469 470 471 472 473 474
			      PAGE_CACHE_SIZE - offset);
	btrfs_csum_final(crc, (char *)&crc);
	if (val != crc) {
		printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
				   "space cache\n");
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}

475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
	return 0;
}

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

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

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

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

497
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517

	/* 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;
}

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

	/*
	 * If we aren't at the start of the current page, unmap this one and
	 * map the next one if there is any left.
	 */
	if (io_ctl->cur != io_ctl->orig) {
518
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519 520 521 522 523 524
		if (io_ctl->index >= io_ctl->num_pages)
			return -ENOSPC;
		io_ctl_map_page(io_ctl, 0);
	}

	memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
525
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
526 527 528 529 530 531 532
	if (io_ctl->index < io_ctl->num_pages)
		io_ctl_map_page(io_ctl, 0);
	return 0;
}

static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
{
533 534 535 536 537 538 539 540
	/*
	 * 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);
541 542 543

	while (io_ctl->index < io_ctl->num_pages) {
		io_ctl_map_page(io_ctl, 1);
544
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
545 546 547
	}
}

548 549
static int io_ctl_read_entry(struct io_ctl *io_ctl,
			    struct btrfs_free_space *entry, u8 *type)
550 551
{
	struct btrfs_free_space_entry *e;
552 553 554 555 556 557 558
	int ret;

	if (!io_ctl->cur) {
		ret = io_ctl_check_crc(io_ctl, io_ctl->index);
		if (ret)
			return ret;
	}
559 560 561 562

	e = io_ctl->cur;
	entry->offset = le64_to_cpu(e->offset);
	entry->bytes = le64_to_cpu(e->bytes);
563
	*type = e->type;
564 565 566 567
	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))
568
		return 0;
569 570 571

	io_ctl_unmap_page(io_ctl);

572
	return 0;
573 574
}

575 576
static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
			      struct btrfs_free_space *entry)
577
{
578 579 580 581 582 583
	int ret;

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

584 585
	memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
	io_ctl_unmap_page(io_ctl);
586 587

	return 0;
588 589
}

590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627
/*
 * Since we attach pinned extents after the fact we can have contiguous sections
 * of free space that are split up in entries.  This poses a problem with the
 * tree logging stuff since it could have allocated across what appears to be 2
 * entries since we would have merged the entries when adding the pinned extents
 * back to the free space cache.  So run through the space cache that we just
 * loaded and merge contiguous entries.  This will make the log replay stuff not
 * blow up and it will make for nicer allocator behavior.
 */
static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
{
	struct btrfs_free_space *e, *prev = NULL;
	struct rb_node *n;

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

628 629 630
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)
631 632 633
{
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
634
	struct io_ctl io_ctl;
635
	struct btrfs_key key;
636
	struct btrfs_free_space *e, *n;
637 638 639 640
	struct list_head bitmaps;
	u64 num_entries;
	u64 num_bitmaps;
	u64 generation;
641
	u8 type;
642
	int ret = 0;
643 644 645 646

	INIT_LIST_HEAD(&bitmaps);

	/* Nothing in the space cache, goodbye */
647
	if (!i_size_read(inode))
648
		return 0;
649 650

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
651
	key.offset = offset;
652 653 654
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
655
	if (ret < 0)
656
		return 0;
657
	else if (ret > 0) {
658
		btrfs_release_path(path);
659
		return 0;
660 661
	}

662 663
	ret = -1;

664 665 666 667 668 669
	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);
670
	btrfs_release_path(path);
671 672

	if (BTRFS_I(inode)->generation != generation) {
673 674 675 676 677
		btrfs_err(root->fs_info,
			"free space inode generation (%llu) "
			"did not match free space cache generation (%llu)",
			(unsigned long long)BTRFS_I(inode)->generation,
			(unsigned long long)generation);
678
		return 0;
679 680 681
	}

	if (!num_entries)
682
		return 0;
683

684 685 686 687
	ret = io_ctl_init(&io_ctl, inode, root);
	if (ret)
		return ret;

688
	ret = readahead_cache(inode);
689
	if (ret)
690 691
		goto out;

692 693 694
	ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
	if (ret)
		goto out;
695

696 697 698 699
	ret = io_ctl_check_crc(&io_ctl, 0);
	if (ret)
		goto free_cache;

700 701 702
	ret = io_ctl_check_generation(&io_ctl, generation);
	if (ret)
		goto free_cache;
703

704 705 706 707
	while (num_entries) {
		e = kmem_cache_zalloc(btrfs_free_space_cachep,
				      GFP_NOFS);
		if (!e)
708 709
			goto free_cache;

710 711 712 713 714 715
		ret = io_ctl_read_entry(&io_ctl, e, &type);
		if (ret) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
		}

716 717 718
		if (!e->bytes) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
719
		}
720 721 722 723 724 725

		if (type == BTRFS_FREE_SPACE_EXTENT) {
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
726 727
				btrfs_err(root->fs_info,
					"Duplicate entries in free space cache, dumping");
728
				kmem_cache_free(btrfs_free_space_cachep, e);
729 730
				goto free_cache;
			}
731 732 733 734 735 736 737
		} else {
			BUG_ON(!num_bitmaps);
			num_bitmaps--;
			e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
			if (!e->bitmap) {
				kmem_cache_free(
					btrfs_free_space_cachep, e);
738 739
				goto free_cache;
			}
740 741 742 743 744 745
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			ctl->total_bitmaps++;
			ctl->op->recalc_thresholds(ctl);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
746 747
				btrfs_err(root->fs_info,
					"Duplicate entries in free space cache, dumping");
748
				kmem_cache_free(btrfs_free_space_cachep, e);
749 750
				goto free_cache;
			}
751
			list_add_tail(&e->list, &bitmaps);
752 753
		}

754 755
		num_entries--;
	}
756

757 758
	io_ctl_unmap_page(&io_ctl);

759 760 761 762 763
	/*
	 * 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) {
764
		list_del_init(&e->list);
765 766 767
		ret = io_ctl_read_bitmap(&io_ctl, e);
		if (ret)
			goto free_cache;
768 769
	}

770
	io_ctl_drop_pages(&io_ctl);
771
	merge_space_tree(ctl);
772 773
	ret = 1;
out:
774
	io_ctl_free(&io_ctl);
775 776
	return ret;
free_cache:
777
	io_ctl_drop_pages(&io_ctl);
778
	__btrfs_remove_free_space_cache(ctl);
779 780 781
	goto out;
}

782 783
int load_free_space_cache(struct btrfs_fs_info *fs_info,
			  struct btrfs_block_group_cache *block_group)
J
Josef Bacik 已提交
784
{
785
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
786 787 788
	struct btrfs_root *root = fs_info->tree_root;
	struct inode *inode;
	struct btrfs_path *path;
789
	int ret = 0;
790 791 792 793 794 795 796
	bool matched;
	u64 used = btrfs_block_group_used(&block_group->item);

	/*
	 * If this block group has been marked to be cleared for one reason or
	 * another then we can't trust the on disk cache, so just return.
	 */
797
	spin_lock(&block_group->lock);
798 799 800 801
	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
		spin_unlock(&block_group->lock);
		return 0;
	}
802
	spin_unlock(&block_group->lock);
803 804 805 806

	path = btrfs_alloc_path();
	if (!path)
		return 0;
807 808
	path->search_commit_root = 1;
	path->skip_locking = 1;
809 810 811 812 813 814 815

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

816 817 818 819
	/* 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);
820
		btrfs_free_path(path);
821 822 823 824
		goto out;
	}
	spin_unlock(&block_group->lock);

825 826 827 828 829 830 831 832 833 834 835 836 837
	ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
				      path, block_group->key.objectid);
	btrfs_free_path(path);
	if (ret <= 0)
		goto out;

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

	if (!matched) {
		__btrfs_remove_free_space_cache(ctl);
838 839
		btrfs_err(fs_info, "block group %llu has wrong amount of free space",
			block_group->key.objectid);
840 841 842 843 844 845 846 847
		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);
848
		ret = 0;
849

850 851
		btrfs_err(fs_info, "failed to load free space cache for block group %llu",
			block_group->key.objectid);
852 853 854 855
	}

	iput(inode);
	return ret;
856 857
}

858 859 860 861 862 863 864 865 866 867 868 869 870
/**
 * __btrfs_write_out_cache - write out cached info to an inode
 * @root - the root the inode belongs to
 * @ctl - the free space cache we are going to write out
 * @block_group - the block_group for this cache if it belongs to a block_group
 * @trans - the trans handle
 * @path - the path to use
 * @offset - the offset for the key we'll insert
 *
 * This function writes out a free space cache struct to disk for quick recovery
 * on mount.  This will return 0 if it was successfull in writing the cache out,
 * and -1 if it was not.
 */
871 872 873 874 875
static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
				   struct btrfs_free_space_ctl *ctl,
				   struct btrfs_block_group_cache *block_group,
				   struct btrfs_trans_handle *trans,
				   struct btrfs_path *path, u64 offset)
J
Josef Bacik 已提交
876 877 878 879 880 881
{
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
	struct rb_node *node;
	struct list_head *pos, *n;
	struct extent_state *cached_state = NULL;
882 883
	struct btrfs_free_cluster *cluster = NULL;
	struct extent_io_tree *unpin = NULL;
884
	struct io_ctl io_ctl;
J
Josef Bacik 已提交
885 886
	struct list_head bitmap_list;
	struct btrfs_key key;
887
	u64 start, extent_start, extent_end, len;
J
Josef Bacik 已提交
888 889
	int entries = 0;
	int bitmaps = 0;
890 891
	int ret;
	int err = -1;
J
Josef Bacik 已提交
892 893 894

	INIT_LIST_HEAD(&bitmap_list);

895 896
	if (!i_size_read(inode))
		return -1;
897

898 899 900
	ret = io_ctl_init(&io_ctl, inode, root);
	if (ret)
		return -1;
901

902
	/* Get the cluster for this block_group if it exists */
903
	if (block_group && !list_empty(&block_group->cluster_list))
904 905 906 907
		cluster = list_entry(block_group->cluster_list.next,
				     struct btrfs_free_cluster,
				     block_group_list);

908 909
	/* Lock all pages first so we can lock the extent safely. */
	io_ctl_prepare_pages(&io_ctl, inode, 0);
J
Josef Bacik 已提交
910 911

	lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
912
			 0, &cached_state);
J
Josef Bacik 已提交
913

914 915 916 917 918 919
	node = rb_first(&ctl->free_space_offset);
	if (!node && cluster) {
		node = rb_first(&cluster->root);
		cluster = NULL;
	}

920 921
	/* Make sure we can fit our crcs into the first page */
	if (io_ctl.check_crcs &&
922
	    (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
923 924
		goto out_nospc;

925
	io_ctl_set_generation(&io_ctl, trans->transid);
926

927 928 929
	/* Write out the extent entries */
	while (node) {
		struct btrfs_free_space *e;
J
Josef Bacik 已提交
930

931 932
		e = rb_entry(node, struct btrfs_free_space, offset_index);
		entries++;
J
Josef Bacik 已提交
933

934 935 936 937
		ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
				       e->bitmap);
		if (ret)
			goto out_nospc;
938

939 940 941
		if (e->bitmap) {
			list_add_tail(&e->list, &bitmap_list);
			bitmaps++;
942
		}
943 944 945 946
		node = rb_next(node);
		if (!node && cluster) {
			node = rb_first(&cluster->root);
			cluster = NULL;
947
		}
948
	}
949

950 951 952 953
	/*
	 * We want to add any pinned extents to our free space cache
	 * so we don't leak the space
	 */
954 955 956 957 958 959 960 961 962 963

	/*
	 * We shouldn't have switched the pinned extents yet so this is the
	 * right one
	 */
	unpin = root->fs_info->pinned_extents;

	if (block_group)
		start = block_group->key.objectid;

964 965
	while (block_group && (start < block_group->key.objectid +
			       block_group->key.offset)) {
966 967
		ret = find_first_extent_bit(unpin, start,
					    &extent_start, &extent_end,
968
					    EXTENT_DIRTY, NULL);
969 970 971
		if (ret) {
			ret = 0;
			break;
J
Josef Bacik 已提交
972 973
		}

974
		/* This pinned extent is out of our range */
975
		if (extent_start >= block_group->key.objectid +
976 977
		    block_group->key.offset)
			break;
978

979 980 981 982
		extent_start = max(extent_start, start);
		extent_end = min(block_group->key.objectid +
				 block_group->key.offset, extent_end + 1);
		len = extent_end - extent_start;
J
Josef Bacik 已提交
983

984
		entries++;
985
		ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
986 987
		if (ret)
			goto out_nospc;
J
Josef Bacik 已提交
988

989
		start = extent_end;
990
	}
J
Josef Bacik 已提交
991 992 993 994 995 996

	/* Write out the bitmaps */
	list_for_each_safe(pos, n, &bitmap_list) {
		struct btrfs_free_space *entry =
			list_entry(pos, struct btrfs_free_space, list);

997 998 999
		ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
		if (ret)
			goto out_nospc;
J
Josef Bacik 已提交
1000
		list_del_init(&entry->list);
1001 1002
	}

J
Josef Bacik 已提交
1003
	/* Zero out the rest of the pages just to make sure */
1004
	io_ctl_zero_remaining_pages(&io_ctl);
J
Josef Bacik 已提交
1005

1006 1007 1008
	ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
				0, i_size_read(inode), &cached_state);
	io_ctl_drop_pages(&io_ctl);
J
Josef Bacik 已提交
1009 1010 1011
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
			     i_size_read(inode) - 1, &cached_state, GFP_NOFS);

1012
	if (ret)
1013
		goto out;
1014 1015


1016
	btrfs_wait_ordered_range(inode, 0, (u64)-1);
J
Josef Bacik 已提交
1017 1018

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1019
	key.offset = offset;
J
Josef Bacik 已提交
1020 1021
	key.type = 0;

1022
	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
J
Josef Bacik 已提交
1023
	if (ret < 0) {
1024
		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1025 1026
				 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
				 GFP_NOFS);
1027
		goto out;
J
Josef Bacik 已提交
1028 1029 1030 1031 1032 1033 1034 1035
	}
	leaf = path->nodes[0];
	if (ret > 0) {
		struct btrfs_key found_key;
		BUG_ON(!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 ||
1036
		    found_key.offset != offset) {
1037 1038
			clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
					 inode->i_size - 1,
1039 1040
					 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
					 NULL, GFP_NOFS);
1041
			btrfs_release_path(path);
1042
			goto out;
J
Josef Bacik 已提交
1043 1044
		}
	}
1045 1046

	BTRFS_I(inode)->generation = trans->transid;
J
Josef Bacik 已提交
1047 1048 1049 1050 1051 1052
	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);
1053
	btrfs_release_path(path);
J
Josef Bacik 已提交
1054

1055
	err = 0;
1056
out:
1057
	io_ctl_free(&io_ctl);
1058
	if (err) {
1059
		invalidate_inode_pages2(inode->i_mapping);
J
Josef Bacik 已提交
1060 1061 1062
		BTRFS_I(inode)->generation = 0;
	}
	btrfs_update_inode(trans, root, inode);
1063
	return err;
1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074

out_nospc:
	list_for_each_safe(pos, n, &bitmap_list) {
		struct btrfs_free_space *entry =
			list_entry(pos, struct btrfs_free_space, list);
		list_del_init(&entry->list);
	}
	io_ctl_drop_pages(&io_ctl);
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
			     i_size_read(inode) - 1, &cached_state, GFP_NOFS);
	goto out;
1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100
}

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

	root = root->fs_info->tree_root;

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

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

	ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
				      path, block_group->key.objectid);
1101
	if (ret) {
1102 1103 1104
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_ERROR;
		spin_unlock(&block_group->lock);
1105
		ret = 0;
1106
#ifdef DEBUG
1107 1108 1109
		btrfs_err(root->fs_info,
			"failed to write free space cache for block group %llu",
			block_group->key.objectid);
1110
#endif
1111 1112
	}

J
Josef Bacik 已提交
1113 1114 1115 1116
	iput(inode);
	return ret;
}

1117
static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1118
					  u64 offset)
J
Josef Bacik 已提交
1119
{
1120 1121
	BUG_ON(offset < bitmap_start);
	offset -= bitmap_start;
1122
	return (unsigned long)(div_u64(offset, unit));
1123
}
J
Josef Bacik 已提交
1124

1125
static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1126
{
1127
	return (unsigned long)(div_u64(bytes, unit));
1128
}
J
Josef Bacik 已提交
1129

1130
static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1131 1132 1133 1134
				   u64 offset)
{
	u64 bitmap_start;
	u64 bytes_per_bitmap;
J
Josef Bacik 已提交
1135

1136 1137
	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
	bitmap_start = offset - ctl->start;
1138 1139
	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
	bitmap_start *= bytes_per_bitmap;
1140
	bitmap_start += ctl->start;
J
Josef Bacik 已提交
1141

1142
	return bitmap_start;
J
Josef Bacik 已提交
1143 1144
}

1145 1146
static int tree_insert_offset(struct rb_root *root, u64 offset,
			      struct rb_node *node, int bitmap)
J
Josef Bacik 已提交
1147 1148 1149 1150 1151 1152 1153
{
	struct rb_node **p = &root->rb_node;
	struct rb_node *parent = NULL;
	struct btrfs_free_space *info;

	while (*p) {
		parent = *p;
1154
		info = rb_entry(parent, struct btrfs_free_space, offset_index);
J
Josef Bacik 已提交
1155

1156
		if (offset < info->offset) {
J
Josef Bacik 已提交
1157
			p = &(*p)->rb_left;
1158
		} else if (offset > info->offset) {
J
Josef Bacik 已提交
1159
			p = &(*p)->rb_right;
1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174
		} 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) {
1175 1176 1177 1178
				if (info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1179 1180
				p = &(*p)->rb_right;
			} else {
1181 1182 1183 1184
				if (!info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1185 1186 1187
				p = &(*p)->rb_left;
			}
		}
J
Josef Bacik 已提交
1188 1189 1190 1191 1192 1193 1194 1195 1196
	}

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

	return 0;
}

/*
J
Josef Bacik 已提交
1197 1198
 * searches the tree for the given offset.
 *
1199 1200 1201
 * 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
1202
 */
1203
static struct btrfs_free_space *
1204
tree_search_offset(struct btrfs_free_space_ctl *ctl,
1205
		   u64 offset, int bitmap_only, int fuzzy)
J
Josef Bacik 已提交
1206
{
1207
	struct rb_node *n = ctl->free_space_offset.rb_node;
1208 1209 1210 1211 1212 1213 1214 1215
	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
1216 1217

		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1218
		prev = entry;
J
Josef Bacik 已提交
1219

1220
		if (offset < entry->offset)
J
Josef Bacik 已提交
1221
			n = n->rb_left;
1222
		else if (offset > entry->offset)
J
Josef Bacik 已提交
1223
			n = n->rb_right;
1224
		else
J
Josef Bacik 已提交
1225 1226 1227
			break;
	}

1228 1229 1230 1231 1232
	if (bitmap_only) {
		if (!entry)
			return NULL;
		if (entry->bitmap)
			return entry;
J
Josef Bacik 已提交
1233

1234 1235 1236 1237 1238 1239 1240 1241 1242 1243
		/*
		 * 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
1244

1245 1246 1247 1248
		WARN_ON(!entry->bitmap);
		return entry;
	} else if (entry) {
		if (entry->bitmap) {
J
Josef Bacik 已提交
1249
			/*
1250 1251
			 * if previous extent entry covers the offset,
			 * we should return it instead of the bitmap entry
J
Josef Bacik 已提交
1252
			 */
1253 1254
			n = rb_prev(&entry->offset_index);
			if (n) {
1255 1256
				prev = rb_entry(n, struct btrfs_free_space,
						offset_index);
1257 1258 1259
				if (!prev->bitmap &&
				    prev->offset + prev->bytes > offset)
					entry = prev;
J
Josef Bacik 已提交
1260
			}
1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275
		}
		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);
			BUG_ON(entry->offset > offset);
J
Josef Bacik 已提交
1276
		} else {
1277 1278 1279 1280
			if (fuzzy)
				return entry;
			else
				return NULL;
J
Josef Bacik 已提交
1281 1282 1283
		}
	}

1284
	if (entry->bitmap) {
1285 1286
		n = rb_prev(&entry->offset_index);
		if (n) {
1287 1288
			prev = rb_entry(n, struct btrfs_free_space,
					offset_index);
1289 1290 1291
			if (!prev->bitmap &&
			    prev->offset + prev->bytes > offset)
				return prev;
1292
		}
1293
		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1294 1295 1296 1297 1298 1299 1300 1301 1302 1303
			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 *
1304
			    ctl->unit > offset)
1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316
				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
1317 1318
}

1319
static inline void
1320
__unlink_free_space(struct btrfs_free_space_ctl *ctl,
1321
		    struct btrfs_free_space *info)
J
Josef Bacik 已提交
1322
{
1323 1324
	rb_erase(&info->offset_index, &ctl->free_space_offset);
	ctl->free_extents--;
1325 1326
}

1327
static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1328 1329
			      struct btrfs_free_space *info)
{
1330 1331
	__unlink_free_space(ctl, info);
	ctl->free_space -= info->bytes;
J
Josef Bacik 已提交
1332 1333
}

1334
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1335 1336 1337 1338
			   struct btrfs_free_space *info)
{
	int ret = 0;

1339
	BUG_ON(!info->bitmap && !info->bytes);
1340
	ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1341
				 &info->offset_index, (info->bitmap != NULL));
J
Josef Bacik 已提交
1342 1343 1344
	if (ret)
		return ret;

1345 1346
	ctl->free_space += info->bytes;
	ctl->free_extents++;
1347 1348 1349
	return ret;
}

1350
static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1351
{
1352
	struct btrfs_block_group_cache *block_group = ctl->private;
1353 1354 1355
	u64 max_bytes;
	u64 bitmap_bytes;
	u64 extent_bytes;
1356
	u64 size = block_group->key.offset;
1357
	u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1358 1359
	int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);

1360 1361
	max_bitmaps = max(max_bitmaps, 1);

1362
	BUG_ON(ctl->total_bitmaps > max_bitmaps);
1363 1364 1365 1366 1367 1368

	/*
	 * The goal is to keep the total amount of memory used per 1gb of space
	 * at or below 32k, so we need to adjust how much memory we allow to be
	 * used by extent based free space tracking
	 */
1369 1370 1371 1372 1373
	if (size < 1024 * 1024 * 1024)
		max_bytes = MAX_CACHE_BYTES_PER_GIG;
	else
		max_bytes = MAX_CACHE_BYTES_PER_GIG *
			div64_u64(size, 1024 * 1024 * 1024);
1374

1375 1376 1377 1378 1379
	/*
	 * we want to account for 1 more bitmap than what we have so we can make
	 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
	 * we add more bitmaps.
	 */
1380
	bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1381

1382
	if (bitmap_bytes >= max_bytes) {
1383
		ctl->extents_thresh = 0;
1384 1385
		return;
	}
1386

1387 1388 1389 1390 1391 1392
	/*
	 * we want the extent entry threshold to always be at most 1/2 the maxw
	 * bytes we can have, or whatever is less than that.
	 */
	extent_bytes = max_bytes - bitmap_bytes;
	extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1393

1394
	ctl->extents_thresh =
1395
		div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1396 1397
}

1398 1399 1400
static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
				       struct btrfs_free_space *info,
				       u64 offset, u64 bytes)
1401
{
L
Li Zefan 已提交
1402
	unsigned long start, count;
1403

1404 1405
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
L
Li Zefan 已提交
1406
	BUG_ON(start + count > BITS_PER_BITMAP);
1407

L
Li Zefan 已提交
1408
	bitmap_clear(info->bitmap, start, count);
1409 1410

	info->bytes -= bytes;
1411 1412 1413 1414 1415 1416 1417
}

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);
1418
	ctl->free_space -= bytes;
1419 1420
}

1421
static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1422 1423
			    struct btrfs_free_space *info, u64 offset,
			    u64 bytes)
1424
{
L
Li Zefan 已提交
1425
	unsigned long start, count;
1426

1427 1428
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
L
Li Zefan 已提交
1429
	BUG_ON(start + count > BITS_PER_BITMAP);
1430

L
Li Zefan 已提交
1431
	bitmap_set(info->bitmap, start, count);
1432 1433

	info->bytes += bytes;
1434
	ctl->free_space += bytes;
1435 1436
}

1437
static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1438 1439 1440 1441 1442 1443 1444
			 struct btrfs_free_space *bitmap_info, u64 *offset,
			 u64 *bytes)
{
	unsigned long found_bits = 0;
	unsigned long bits, i;
	unsigned long next_zero;

1445
	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1446
			  max_t(u64, *offset, bitmap_info->offset));
1447
	bits = bytes_to_bits(*bytes, ctl->unit);
1448

1449
	for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1450 1451 1452 1453 1454 1455 1456 1457 1458 1459
		next_zero = find_next_zero_bit(bitmap_info->bitmap,
					       BITS_PER_BITMAP, i);
		if ((next_zero - i) >= bits) {
			found_bits = next_zero - i;
			break;
		}
		i = next_zero;
	}

	if (found_bits) {
1460 1461
		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
		*bytes = (u64)(found_bits) * ctl->unit;
1462 1463 1464 1465 1466 1467
		return 0;
	}

	return -1;
}

1468
static struct btrfs_free_space *
D
David Woodhouse 已提交
1469 1470
find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
		unsigned long align)
1471 1472 1473
{
	struct btrfs_free_space *entry;
	struct rb_node *node;
D
David Woodhouse 已提交
1474 1475 1476
	u64 ctl_off;
	u64 tmp;
	u64 align_off;
1477 1478
	int ret;

1479
	if (!ctl->free_space_offset.rb_node)
1480 1481
		return NULL;

1482
	entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1483 1484 1485 1486 1487 1488 1489 1490
	if (!entry)
		return NULL;

	for (node = &entry->offset_index; node; node = rb_next(node)) {
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		if (entry->bytes < *bytes)
			continue;

D
David Woodhouse 已提交
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507
		/* make sure the space returned is big enough
		 * to match our requested alignment
		 */
		if (*bytes >= align) {
			ctl_off = entry->offset - ctl->start;
			tmp = ctl_off + align - 1;;
			do_div(tmp, align);
			tmp = tmp * align + ctl->start;
			align_off = tmp - entry->offset;
		} else {
			align_off = 0;
			tmp = entry->offset;
		}

		if (entry->bytes < *bytes + align_off)
			continue;

1508
		if (entry->bitmap) {
D
David Woodhouse 已提交
1509 1510 1511
			ret = search_bitmap(ctl, entry, &tmp, bytes);
			if (!ret) {
				*offset = tmp;
1512
				return entry;
D
David Woodhouse 已提交
1513
			}
1514 1515 1516
			continue;
		}

D
David Woodhouse 已提交
1517 1518
		*offset = tmp;
		*bytes = entry->bytes - align_off;
1519 1520 1521 1522 1523 1524
		return entry;
	}

	return NULL;
}

1525
static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1526 1527
			   struct btrfs_free_space *info, u64 offset)
{
1528
	info->offset = offset_to_bitmap(ctl, offset);
J
Josef Bacik 已提交
1529
	info->bytes = 0;
1530
	INIT_LIST_HEAD(&info->list);
1531 1532
	link_free_space(ctl, info);
	ctl->total_bitmaps++;
1533

1534
	ctl->op->recalc_thresholds(ctl);
1535 1536
}

1537
static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1538 1539
			struct btrfs_free_space *bitmap_info)
{
1540
	unlink_free_space(ctl, bitmap_info);
1541
	kfree(bitmap_info->bitmap);
1542
	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1543 1544
	ctl->total_bitmaps--;
	ctl->op->recalc_thresholds(ctl);
1545 1546
}

1547
static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1548 1549 1550 1551
			      struct btrfs_free_space *bitmap_info,
			      u64 *offset, u64 *bytes)
{
	u64 end;
1552 1553
	u64 search_start, search_bytes;
	int ret;
1554 1555

again:
1556
	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1557

1558
	/*
1559 1560 1561 1562
	 * 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.
1563 1564
	 */
	search_start = *offset;
1565
	search_bytes = ctl->unit;
1566
	search_bytes = min(search_bytes, end - search_start + 1);
1567
	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1568 1569
	if (ret < 0 || search_start != *offset)
		return -EINVAL;
1570

1571 1572 1573 1574 1575 1576 1577 1578 1579
	/* 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;
1580 1581

	if (*bytes) {
1582
		struct rb_node *next = rb_next(&bitmap_info->offset_index);
1583
		if (!bitmap_info->bytes)
1584
			free_bitmap(ctl, bitmap_info);
1585

1586 1587 1588 1589 1590
		/*
		 * no entry after this bitmap, but we still have bytes to
		 * remove, so something has gone wrong.
		 */
		if (!next)
1591 1592
			return -EINVAL;

1593 1594 1595 1596 1597 1598 1599
		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.
		 */
1600 1601 1602
		if (!bitmap_info->bitmap)
			return -EAGAIN;

1603 1604 1605 1606 1607 1608 1609
		/*
		 * 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;
1610
		search_bytes = ctl->unit;
1611
		ret = search_bitmap(ctl, bitmap_info, &search_start,
1612 1613 1614 1615
				    &search_bytes);
		if (ret < 0 || search_start != *offset)
			return -EAGAIN;

1616
		goto again;
1617
	} else if (!bitmap_info->bytes)
1618
		free_bitmap(ctl, bitmap_info);
1619 1620 1621 1622

	return 0;
}

J
Josef Bacik 已提交
1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639
static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
			       struct btrfs_free_space *info, u64 offset,
			       u64 bytes)
{
	u64 bytes_to_set = 0;
	u64 end;

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

	bytes_to_set = min(end - offset, bytes);

	bitmap_set_bits(ctl, info, offset, bytes_to_set);

	return bytes_to_set;

}

1640 1641
static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
		      struct btrfs_free_space *info)
1642
{
1643
	struct btrfs_block_group_cache *block_group = ctl->private;
1644 1645 1646 1647 1648

	/*
	 * If we are below the extents threshold then we can add this as an
	 * extent, and don't have to deal with the bitmap
	 */
1649
	if (ctl->free_extents < ctl->extents_thresh) {
1650 1651 1652 1653 1654 1655 1656 1657
		/*
		 * If this block group has some small extents we don't want to
		 * use up all of our free slots in the cache with them, we want
		 * to reserve them to larger extents, however if we have plent
		 * of cache left then go ahead an dadd them, no sense in adding
		 * the overhead of a bitmap if we don't have to.
		 */
		if (info->bytes <= block_group->sectorsize * 4) {
1658 1659
			if (ctl->free_extents * 2 <= ctl->extents_thresh)
				return false;
1660
		} else {
1661
			return false;
1662 1663
		}
	}
1664 1665

	/*
1666 1667 1668 1669 1670 1671
	 * 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.
1672
	 */
1673
	if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1674 1675 1676 1677 1678
		return false;

	return true;
}

J
Josef Bacik 已提交
1679 1680 1681 1682 1683
static struct btrfs_free_space_op free_space_op = {
	.recalc_thresholds	= recalculate_thresholds,
	.use_bitmap		= use_bitmap,
};

1684 1685 1686 1687
static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info)
{
	struct btrfs_free_space *bitmap_info;
J
Josef Bacik 已提交
1688
	struct btrfs_block_group_cache *block_group = NULL;
1689
	int added = 0;
J
Josef Bacik 已提交
1690
	u64 bytes, offset, bytes_added;
1691
	int ret;
1692 1693 1694 1695

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

1696 1697 1698
	if (!ctl->op->use_bitmap(ctl, info))
		return 0;

J
Josef Bacik 已提交
1699 1700
	if (ctl->op == &free_space_op)
		block_group = ctl->private;
1701
again:
J
Josef Bacik 已提交
1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718
	/*
	 * 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);
1719
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
1720 1721 1722 1723 1724
		}

		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		if (!entry->bitmap) {
			spin_unlock(&cluster->lock);
1725
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739
		}

		if (entry->offset == offset_to_bitmap(ctl, offset)) {
			bytes_added = add_bytes_to_bitmap(ctl, entry,
							  offset, bytes);
			bytes -= bytes_added;
			offset += bytes_added;
		}
		spin_unlock(&cluster->lock);
		if (!bytes) {
			ret = 1;
			goto out;
		}
	}
1740 1741

no_cluster_bitmap:
1742
	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1743 1744 1745 1746 1747 1748
					 1, 0);
	if (!bitmap_info) {
		BUG_ON(added);
		goto new_bitmap;
	}

J
Josef Bacik 已提交
1749 1750 1751 1752
	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
	bytes -= bytes_added;
	offset += bytes_added;
	added = 0;
1753 1754 1755 1756 1757 1758 1759 1760 1761

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

new_bitmap:
	if (info && info->bitmap) {
1762
		add_new_bitmap(ctl, info, offset);
1763 1764 1765 1766
		added = 1;
		info = NULL;
		goto again;
	} else {
1767
		spin_unlock(&ctl->tree_lock);
1768 1769 1770

		/* no pre-allocated info, allocate a new one */
		if (!info) {
1771 1772
			info = kmem_cache_zalloc(btrfs_free_space_cachep,
						 GFP_NOFS);
1773
			if (!info) {
1774
				spin_lock(&ctl->tree_lock);
1775 1776 1777 1778 1779 1780 1781
				ret = -ENOMEM;
				goto out;
			}
		}

		/* allocate the bitmap */
		info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1782
		spin_lock(&ctl->tree_lock);
1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
		if (!info->bitmap) {
			ret = -ENOMEM;
			goto out;
		}
		goto again;
	}

out:
	if (info) {
		if (info->bitmap)
			kfree(info->bitmap);
1794
		kmem_cache_free(btrfs_free_space_cachep, info);
1795
	}
J
Josef Bacik 已提交
1796 1797 1798 1799

	return ret;
}

1800
static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1801
			  struct btrfs_free_space *info, bool update_stat)
J
Josef Bacik 已提交
1802
{
1803 1804 1805 1806 1807
	struct btrfs_free_space *left_info;
	struct btrfs_free_space *right_info;
	bool merged = false;
	u64 offset = info->offset;
	u64 bytes = info->bytes;
1808

J
Josef Bacik 已提交
1809 1810 1811 1812 1813
	/*
	 * 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
	 */
1814
	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1815 1816 1817 1818
	if (right_info && rb_prev(&right_info->offset_index))
		left_info = rb_entry(rb_prev(&right_info->offset_index),
				     struct btrfs_free_space, offset_index);
	else
1819
		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
J
Josef Bacik 已提交
1820

1821
	if (right_info && !right_info->bitmap) {
1822
		if (update_stat)
1823
			unlink_free_space(ctl, right_info);
1824
		else
1825
			__unlink_free_space(ctl, right_info);
1826
		info->bytes += right_info->bytes;
1827
		kmem_cache_free(btrfs_free_space_cachep, right_info);
1828
		merged = true;
J
Josef Bacik 已提交
1829 1830
	}

1831 1832
	if (left_info && !left_info->bitmap &&
	    left_info->offset + left_info->bytes == offset) {
1833
		if (update_stat)
1834
			unlink_free_space(ctl, left_info);
1835
		else
1836
			__unlink_free_space(ctl, left_info);
1837 1838
		info->offset = left_info->offset;
		info->bytes += left_info->bytes;
1839
		kmem_cache_free(btrfs_free_space_cachep, left_info);
1840
		merged = true;
J
Josef Bacik 已提交
1841 1842
	}

1843 1844 1845
	return merged;
}

1846 1847
int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
			   u64 offset, u64 bytes)
1848 1849 1850 1851
{
	struct btrfs_free_space *info;
	int ret = 0;

1852
	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1853 1854 1855 1856 1857 1858
	if (!info)
		return -ENOMEM;

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

1859
	spin_lock(&ctl->tree_lock);
1860

1861
	if (try_merge_free_space(ctl, info, true))
1862 1863 1864 1865 1866 1867 1868
		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
	 */
1869
	ret = insert_into_bitmap(ctl, info);
1870 1871 1872 1873 1874 1875 1876
	if (ret < 0) {
		goto out;
	} else if (ret) {
		ret = 0;
		goto out;
	}
link:
1877
	ret = link_free_space(ctl, info);
J
Josef Bacik 已提交
1878
	if (ret)
1879
		kmem_cache_free(btrfs_free_space_cachep, info);
1880
out:
1881
	spin_unlock(&ctl->tree_lock);
1882

J
Josef Bacik 已提交
1883
	if (ret) {
1884
		printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
S
Stoyan Gaydarov 已提交
1885
		BUG_ON(ret == -EEXIST);
J
Josef Bacik 已提交
1886 1887 1888 1889 1890
	}

	return ret;
}

1891 1892
int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
			    u64 offset, u64 bytes)
J
Josef Bacik 已提交
1893
{
1894
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
1895
	struct btrfs_free_space *info;
1896 1897
	int ret;
	bool re_search = false;
J
Josef Bacik 已提交
1898

1899
	spin_lock(&ctl->tree_lock);
1900

1901
again:
1902
	ret = 0;
1903 1904 1905
	if (!bytes)
		goto out_lock;

1906
	info = tree_search_offset(ctl, offset, 0, 0);
1907
	if (!info) {
1908 1909 1910 1911
		/*
		 * oops didn't find an extent that matched the space we wanted
		 * to remove, look for a bitmap instead
		 */
1912
		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1913 1914
					  1, 0);
		if (!info) {
1915 1916 1917 1918
			/*
			 * 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.
1919
			 */
1920
			WARN_ON(re_search);
1921 1922
			goto out_lock;
		}
1923 1924
	}

1925
	re_search = false;
1926
	if (!info->bitmap) {
1927
		unlink_free_space(ctl, info);
1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938
		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
1939

1940 1941 1942 1943 1944
			offset += to_free;
			bytes -= to_free;
			goto again;
		} else {
			u64 old_end = info->bytes + info->offset;
1945

1946
			info->bytes = offset - info->offset;
1947
			ret = link_free_space(ctl, info);
1948 1949 1950 1951
			WARN_ON(ret);
			if (ret)
				goto out_lock;

1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967
			/* Not enough bytes in this entry to satisfy us */
			if (old_end < offset + bytes) {
				bytes -= old_end - offset;
				offset = old_end;
				goto again;
			} else if (old_end == offset + bytes) {
				/* all done */
				goto out_lock;
			}
			spin_unlock(&ctl->tree_lock);

			ret = btrfs_add_free_space(block_group, offset + bytes,
						   old_end - (offset + bytes));
			WARN_ON(ret);
			goto out;
		}
J
Josef Bacik 已提交
1968
	}
1969

1970
	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1971 1972
	if (ret == -EAGAIN) {
		re_search = true;
1973
		goto again;
1974
	}
1975
out_lock:
1976
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
1977
out:
1978 1979 1980
	return ret;
}

J
Josef Bacik 已提交
1981 1982 1983
void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
			   u64 bytes)
{
1984
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
1985 1986 1987 1988
	struct btrfs_free_space *info;
	struct rb_node *n;
	int count = 0;

1989
	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
J
Josef Bacik 已提交
1990
		info = rb_entry(n, struct btrfs_free_space, offset_index);
L
Liu Bo 已提交
1991
		if (info->bytes >= bytes && !block_group->ro)
J
Josef Bacik 已提交
1992
			count++;
1993
		printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1994
		       (unsigned long long)info->offset,
1995 1996
		       (unsigned long long)info->bytes,
		       (info->bitmap) ? "yes" : "no");
J
Josef Bacik 已提交
1997
	}
1998 1999
	printk(KERN_INFO "block group has cluster?: %s\n",
	       list_empty(&block_group->cluster_list) ? "no" : "yes");
J
Josef Bacik 已提交
2000 2001 2002 2003
	printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
	       "\n", count);
}

2004
void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
J
Josef Bacik 已提交
2005
{
2006
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
2007

2008 2009 2010 2011 2012
	spin_lock_init(&ctl->tree_lock);
	ctl->unit = block_group->sectorsize;
	ctl->start = block_group->key.objectid;
	ctl->private = block_group;
	ctl->op = &free_space_op;
J
Josef Bacik 已提交
2013

2014 2015 2016 2017 2018 2019 2020
	/*
	 * we only want to have 32k of ram per block group for keeping
	 * track of free space, and if we pass 1/2 of that we want to
	 * start converting things over to using bitmaps
	 */
	ctl->extents_thresh = ((1024 * 32) / 2) /
				sizeof(struct btrfs_free_space);
J
Josef Bacik 已提交
2021 2022
}

2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033
/*
 * for a given cluster, put all of its extents back into the free
 * space cache.  If the block group passed doesn't match the block group
 * pointed to by the cluster, someone else raced in and freed the
 * cluster already.  In that case, we just return without changing anything
 */
static int
__btrfs_return_cluster_to_free_space(
			     struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster)
{
2034
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2035 2036 2037 2038 2039 2040 2041
	struct btrfs_free_space *entry;
	struct rb_node *node;

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

2042
	cluster->block_group = NULL;
2043
	cluster->window_start = 0;
2044 2045
	list_del_init(&cluster->block_group_list);

2046
	node = rb_first(&cluster->root);
2047
	while (node) {
2048 2049
		bool bitmap;

2050 2051 2052
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		node = rb_next(&entry->offset_index);
		rb_erase(&entry->offset_index, &cluster->root);
2053 2054 2055

		bitmap = (entry->bitmap != NULL);
		if (!bitmap)
2056 2057
			try_merge_free_space(ctl, entry, false);
		tree_insert_offset(&ctl->free_space_offset,
2058
				   entry->offset, &entry->offset_index, bitmap);
2059
	}
2060
	cluster->root = RB_ROOT;
2061

2062 2063
out:
	spin_unlock(&cluster->lock);
2064
	btrfs_put_block_group(block_group);
2065 2066 2067
	return 0;
}

2068 2069
static void __btrfs_remove_free_space_cache_locked(
				struct btrfs_free_space_ctl *ctl)
J
Josef Bacik 已提交
2070 2071 2072
{
	struct btrfs_free_space *info;
	struct rb_node *node;
2073 2074 2075

	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
		info = rb_entry(node, struct btrfs_free_space, offset_index);
2076 2077 2078 2079 2080 2081
		if (!info->bitmap) {
			unlink_free_space(ctl, info);
			kmem_cache_free(btrfs_free_space_cachep, info);
		} else {
			free_bitmap(ctl, info);
		}
2082 2083 2084 2085 2086 2087
		if (need_resched()) {
			spin_unlock(&ctl->tree_lock);
			cond_resched();
			spin_lock(&ctl->tree_lock);
		}
	}
2088 2089 2090 2091 2092 2093
}

void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
{
	spin_lock(&ctl->tree_lock);
	__btrfs_remove_free_space_cache_locked(ctl);
2094 2095 2096 2097 2098 2099
	spin_unlock(&ctl->tree_lock);
}

void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2100
	struct btrfs_free_cluster *cluster;
2101
	struct list_head *head;
J
Josef Bacik 已提交
2102

2103
	spin_lock(&ctl->tree_lock);
2104 2105 2106 2107
	while ((head = block_group->cluster_list.next) !=
	       &block_group->cluster_list) {
		cluster = list_entry(head, struct btrfs_free_cluster,
				     block_group_list);
2108 2109 2110

		WARN_ON(cluster->block_group != block_group);
		__btrfs_return_cluster_to_free_space(block_group, cluster);
2111
		if (need_resched()) {
2112
			spin_unlock(&ctl->tree_lock);
2113
			cond_resched();
2114
			spin_lock(&ctl->tree_lock);
2115
		}
2116
	}
2117
	__btrfs_remove_free_space_cache_locked(ctl);
2118
	spin_unlock(&ctl->tree_lock);
2119

J
Josef Bacik 已提交
2120 2121
}

2122 2123
u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
			       u64 offset, u64 bytes, u64 empty_size)
J
Josef Bacik 已提交
2124
{
2125
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2126
	struct btrfs_free_space *entry = NULL;
2127
	u64 bytes_search = bytes + empty_size;
2128
	u64 ret = 0;
D
David Woodhouse 已提交
2129 2130
	u64 align_gap = 0;
	u64 align_gap_len = 0;
J
Josef Bacik 已提交
2131

2132
	spin_lock(&ctl->tree_lock);
D
David Woodhouse 已提交
2133 2134
	entry = find_free_space(ctl, &offset, &bytes_search,
				block_group->full_stripe_len);
2135
	if (!entry)
2136 2137 2138 2139
		goto out;

	ret = offset;
	if (entry->bitmap) {
2140
		bitmap_clear_bits(ctl, entry, offset, bytes);
2141
		if (!entry->bytes)
2142
			free_bitmap(ctl, entry);
2143
	} else {
D
David Woodhouse 已提交
2144

2145
		unlink_free_space(ctl, entry);
D
David Woodhouse 已提交
2146 2147 2148 2149 2150 2151 2152
		align_gap_len = offset - entry->offset;
		align_gap = entry->offset;

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

		entry->bytes -= bytes + align_gap_len;
2153
		if (!entry->bytes)
2154
			kmem_cache_free(btrfs_free_space_cachep, entry);
2155
		else
2156
			link_free_space(ctl, entry);
2157
	}
J
Josef Bacik 已提交
2158

2159
out:
2160
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
2161

D
David Woodhouse 已提交
2162 2163
	if (align_gap_len)
		__btrfs_add_free_space(ctl, align_gap, align_gap_len);
J
Josef Bacik 已提交
2164 2165
	return ret;
}
2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178

/*
 * given a cluster, put all of its extents back into the free space
 * cache.  If a block group is passed, this function will only free
 * a cluster that belongs to the passed block group.
 *
 * Otherwise, it'll get a reference on the block group pointed to by the
 * cluster and remove the cluster from it.
 */
int btrfs_return_cluster_to_free_space(
			       struct btrfs_block_group_cache *block_group,
			       struct btrfs_free_cluster *cluster)
{
2179
	struct btrfs_free_space_ctl *ctl;
2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197
	int ret;

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

2198 2199
	ctl = block_group->free_space_ctl;

2200
	/* now return any extents the cluster had on it */
2201
	spin_lock(&ctl->tree_lock);
2202
	ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2203
	spin_unlock(&ctl->tree_lock);
2204 2205 2206 2207 2208 2209

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

2210 2211
static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
				   struct btrfs_free_cluster *cluster,
2212
				   struct btrfs_free_space *entry,
2213 2214
				   u64 bytes, u64 min_start)
{
2215
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2216 2217 2218 2219 2220 2221 2222 2223
	int err;
	u64 search_start = cluster->window_start;
	u64 search_bytes = bytes;
	u64 ret = 0;

	search_start = min_start;
	search_bytes = bytes;

2224
	err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2225
	if (err)
2226
		return 0;
2227 2228

	ret = search_start;
2229
	__bitmap_clear_bits(ctl, entry, ret, bytes);
2230 2231 2232 2233

	return ret;
}

2234 2235 2236 2237 2238 2239 2240 2241 2242
/*
 * given a cluster, try to allocate 'bytes' from it, returns 0
 * if it couldn't find anything suitably large, or a logical disk offset
 * if things worked out
 */
u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster, u64 bytes,
			     u64 min_start)
{
2243
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260
	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);
	while(1) {
2261 2262
		if (entry->bytes < bytes ||
		    (!entry->bitmap && entry->offset < min_start)) {
2263 2264 2265 2266 2267 2268 2269 2270
			node = rb_next(&entry->offset_index);
			if (!node)
				break;
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
			continue;
		}

2271 2272 2273
		if (entry->bitmap) {
			ret = btrfs_alloc_from_bitmap(block_group,
						      cluster, entry, bytes,
2274
						      cluster->window_start);
2275 2276 2277 2278 2279 2280 2281 2282
			if (ret == 0) {
				node = rb_next(&entry->offset_index);
				if (!node)
					break;
				entry = rb_entry(node, struct btrfs_free_space,
						 offset_index);
				continue;
			}
2283
			cluster->window_start += bytes;
2284 2285 2286 2287 2288 2289
		} else {
			ret = entry->offset;

			entry->offset += bytes;
			entry->bytes -= bytes;
		}
2290

2291
		if (entry->bytes == 0)
2292 2293 2294 2295 2296
			rb_erase(&entry->offset_index, &cluster->root);
		break;
	}
out:
	spin_unlock(&cluster->lock);
2297

2298 2299 2300
	if (!ret)
		return 0;

2301
	spin_lock(&ctl->tree_lock);
2302

2303
	ctl->free_space -= bytes;
2304
	if (entry->bytes == 0) {
2305
		ctl->free_extents--;
2306 2307
		if (entry->bitmap) {
			kfree(entry->bitmap);
2308 2309
			ctl->total_bitmaps--;
			ctl->op->recalc_thresholds(ctl);
2310
		}
2311
		kmem_cache_free(btrfs_free_space_cachep, entry);
2312 2313
	}

2314
	spin_unlock(&ctl->tree_lock);
2315

2316 2317 2318
	return ret;
}

2319 2320 2321
static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
				struct btrfs_free_space *entry,
				struct btrfs_free_cluster *cluster,
2322 2323
				u64 offset, u64 bytes,
				u64 cont1_bytes, u64 min_bytes)
2324
{
2325
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2326 2327
	unsigned long next_zero;
	unsigned long i;
2328 2329
	unsigned long want_bits;
	unsigned long min_bits;
2330 2331 2332
	unsigned long found_bits;
	unsigned long start = 0;
	unsigned long total_found = 0;
2333
	int ret;
2334

2335
	i = offset_to_bit(entry->offset, ctl->unit,
2336
			  max_t(u64, offset, entry->offset));
2337 2338
	want_bits = bytes_to_bits(bytes, ctl->unit);
	min_bits = bytes_to_bits(min_bytes, ctl->unit);
2339 2340 2341

again:
	found_bits = 0;
2342
	for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2343 2344
		next_zero = find_next_zero_bit(entry->bitmap,
					       BITS_PER_BITMAP, i);
2345
		if (next_zero - i >= min_bits) {
2346 2347 2348 2349 2350 2351 2352
			found_bits = next_zero - i;
			break;
		}
		i = next_zero;
	}

	if (!found_bits)
2353
		return -ENOSPC;
2354

2355
	if (!total_found) {
2356
		start = i;
2357
		cluster->max_size = 0;
2358 2359 2360 2361
	}

	total_found += found_bits;

2362 2363
	if (cluster->max_size < found_bits * ctl->unit)
		cluster->max_size = found_bits * ctl->unit;
2364

2365 2366
	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
		i = next_zero + 1;
2367 2368 2369
		goto again;
	}

2370
	cluster->window_start = start * ctl->unit + entry->offset;
2371
	rb_erase(&entry->offset_index, &ctl->free_space_offset);
2372 2373
	ret = tree_insert_offset(&cluster->root, entry->offset,
				 &entry->offset_index, 1);
2374
	BUG_ON(ret); /* -EEXIST; Logic error */
2375

J
Josef Bacik 已提交
2376
	trace_btrfs_setup_cluster(block_group, cluster,
2377
				  total_found * ctl->unit, 1);
2378 2379 2380
	return 0;
}

2381 2382
/*
 * This searches the block group for just extents to fill the cluster with.
2383 2384
 * 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.
2385
 */
2386 2387 2388 2389
static noinline int
setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
			struct btrfs_free_cluster *cluster,
			struct list_head *bitmaps, u64 offset, u64 bytes,
2390
			u64 cont1_bytes, u64 min_bytes)
2391
{
2392
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2393 2394 2395 2396 2397 2398 2399
	struct btrfs_free_space *first = NULL;
	struct btrfs_free_space *entry = NULL;
	struct btrfs_free_space *last;
	struct rb_node *node;
	u64 window_start;
	u64 window_free;
	u64 max_extent;
J
Josef Bacik 已提交
2400
	u64 total_size = 0;
2401

2402
	entry = tree_search_offset(ctl, offset, 0, 1);
2403 2404 2405 2406 2407 2408 2409
	if (!entry)
		return -ENOSPC;

	/*
	 * We don't want bitmaps, so just move along until we find a normal
	 * extent entry.
	 */
2410 2411
	while (entry->bitmap || entry->bytes < min_bytes) {
		if (entry->bitmap && list_empty(&entry->list))
2412
			list_add_tail(&entry->list, bitmaps);
2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424
		node = rb_next(&entry->offset_index);
		if (!node)
			return -ENOSPC;
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
	}

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

2425 2426
	for (node = rb_next(&entry->offset_index); node;
	     node = rb_next(&entry->offset_index)) {
2427 2428
		entry = rb_entry(node, struct btrfs_free_space, offset_index);

2429 2430 2431
		if (entry->bitmap) {
			if (list_empty(&entry->list))
				list_add_tail(&entry->list, bitmaps);
2432
			continue;
2433 2434
		}

2435 2436 2437 2438 2439 2440
		if (entry->bytes < min_bytes)
			continue;

		last = entry;
		window_free += entry->bytes;
		if (entry->bytes > max_extent)
2441 2442 2443
			max_extent = entry->bytes;
	}

2444 2445 2446
	if (window_free < bytes || max_extent < cont1_bytes)
		return -ENOSPC;

2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459
	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);
2460
		if (entry->bitmap || entry->bytes < min_bytes)
2461 2462
			continue;

2463
		rb_erase(&entry->offset_index, &ctl->free_space_offset);
2464 2465
		ret = tree_insert_offset(&cluster->root, entry->offset,
					 &entry->offset_index, 0);
J
Josef Bacik 已提交
2466
		total_size += entry->bytes;
2467
		BUG_ON(ret); /* -EEXIST; Logic error */
2468 2469 2470
	} while (node && entry != last);

	cluster->max_size = max_extent;
J
Josef Bacik 已提交
2471
	trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2472 2473 2474 2475 2476 2477 2478
	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.
 */
2479 2480 2481 2482
static noinline int
setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
		     struct btrfs_free_cluster *cluster,
		     struct list_head *bitmaps, u64 offset, u64 bytes,
2483
		     u64 cont1_bytes, u64 min_bytes)
2484
{
2485
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2486 2487
	struct btrfs_free_space *entry;
	int ret = -ENOSPC;
2488
	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2489

2490
	if (ctl->total_bitmaps == 0)
2491 2492
		return -ENOSPC;

2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503
	/*
	 * The bitmap that covers offset won't be in the list unless offset
	 * is just its start offset.
	 */
	entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
	if (entry->offset != bitmap_offset) {
		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
		if (entry && list_empty(&entry->list))
			list_add(&entry->list, bitmaps);
	}

2504
	list_for_each_entry(entry, bitmaps, list) {
2505
		if (entry->bytes < bytes)
2506 2507
			continue;
		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2508
					   bytes, cont1_bytes, min_bytes);
2509 2510 2511 2512 2513
		if (!ret)
			return 0;
	}

	/*
2514 2515
	 * The bitmaps list has all the bitmaps that record free space
	 * starting after offset, so no more search is required.
2516
	 */
2517
	return -ENOSPC;
2518 2519
}

2520 2521
/*
 * here we try to find a cluster of blocks in a block group.  The goal
2522
 * is to find at least bytes+empty_size.
2523 2524 2525 2526 2527
 * We might not find them all in one contiguous area.
 *
 * returns zero and sets up cluster if things worked out, otherwise
 * it returns -enospc
 */
2528
int btrfs_find_space_cluster(struct btrfs_root *root,
2529 2530 2531 2532
			     struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster,
			     u64 offset, u64 bytes, u64 empty_size)
{
2533
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2534
	struct btrfs_free_space *entry, *tmp;
2535
	LIST_HEAD(bitmaps);
2536
	u64 min_bytes;
2537
	u64 cont1_bytes;
2538 2539
	int ret;

2540 2541 2542 2543 2544 2545
	/*
	 * 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.
	 */
2546
	if (btrfs_test_opt(root, SSD_SPREAD)) {
2547
		cont1_bytes = min_bytes = bytes + empty_size;
2548
	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2549 2550 2551 2552 2553 2554
		cont1_bytes = bytes;
		min_bytes = block_group->sectorsize;
	} else {
		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
		min_bytes = block_group->sectorsize;
	}
2555

2556
	spin_lock(&ctl->tree_lock);
2557 2558 2559 2560 2561

	/*
	 * 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.
	 */
2562
	if (ctl->free_space < bytes) {
2563
		spin_unlock(&ctl->tree_lock);
2564 2565 2566
		return -ENOSPC;
	}

2567 2568 2569 2570 2571 2572 2573 2574
	spin_lock(&cluster->lock);

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

J
Josef Bacik 已提交
2575 2576 2577 2578
	trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
				 min_bytes);

	INIT_LIST_HEAD(&bitmaps);
2579
	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2580 2581
				      bytes + empty_size,
				      cont1_bytes, min_bytes);
2582
	if (ret)
2583
		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2584 2585
					   offset, bytes + empty_size,
					   cont1_bytes, min_bytes);
2586 2587 2588 2589

	/* Clear our temporary list */
	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
		list_del_init(&entry->list);
2590

2591 2592 2593 2594 2595
	if (!ret) {
		atomic_inc(&block_group->count);
		list_add_tail(&cluster->block_group_list,
			      &block_group->cluster_list);
		cluster->block_group = block_group;
J
Josef Bacik 已提交
2596 2597
	} else {
		trace_btrfs_failed_cluster_setup(block_group);
2598 2599 2600
	}
out:
	spin_unlock(&cluster->lock);
2601
	spin_unlock(&ctl->tree_lock);
2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612

	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);
2613
	cluster->root = RB_ROOT;
2614 2615 2616 2617 2618
	cluster->max_size = 0;
	INIT_LIST_HEAD(&cluster->block_group_list);
	cluster->block_group = NULL;
}

2619 2620 2621
static int do_trimming(struct btrfs_block_group_cache *block_group,
		       u64 *total_trimmed, u64 start, u64 bytes,
		       u64 reserved_start, u64 reserved_bytes)
2622
{
2623
	struct btrfs_space_info *space_info = block_group->space_info;
2624
	struct btrfs_fs_info *fs_info = block_group->fs_info;
2625 2626 2627
	int ret;
	int update = 0;
	u64 trimmed = 0;
2628

2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669
	spin_lock(&space_info->lock);
	spin_lock(&block_group->lock);
	if (!block_group->ro) {
		block_group->reserved += reserved_bytes;
		space_info->bytes_reserved += reserved_bytes;
		update = 1;
	}
	spin_unlock(&block_group->lock);
	spin_unlock(&space_info->lock);

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

	btrfs_add_free_space(block_group, reserved_start, reserved_bytes);

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

	return ret;
}

static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
			  u64 *total_trimmed, u64 start, u64 end, u64 minlen)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct btrfs_free_space *entry;
	struct rb_node *node;
	int ret = 0;
	u64 extent_start;
	u64 extent_bytes;
	u64 bytes;
2670 2671

	while (start < end) {
2672
		spin_lock(&ctl->tree_lock);
2673

2674 2675
		if (ctl->free_space < minlen) {
			spin_unlock(&ctl->tree_lock);
2676 2677 2678
			break;
		}

2679
		entry = tree_search_offset(ctl, start, 0, 1);
2680
		if (!entry) {
2681
			spin_unlock(&ctl->tree_lock);
2682 2683 2684
			break;
		}

2685 2686 2687 2688
		/* skip bitmaps */
		while (entry->bitmap) {
			node = rb_next(&entry->offset_index);
			if (!node) {
2689
				spin_unlock(&ctl->tree_lock);
2690
				goto out;
2691
			}
2692 2693
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
2694 2695
		}

2696 2697 2698
		if (entry->offset >= end) {
			spin_unlock(&ctl->tree_lock);
			break;
2699 2700
		}

2701 2702 2703 2704 2705 2706 2707
		extent_start = entry->offset;
		extent_bytes = entry->bytes;
		start = max(start, extent_start);
		bytes = min(extent_start + extent_bytes, end) - start;
		if (bytes < minlen) {
			spin_unlock(&ctl->tree_lock);
			goto next;
2708 2709
		}

2710 2711 2712
		unlink_free_space(ctl, entry);
		kmem_cache_free(btrfs_free_space_cachep, entry);

2713
		spin_unlock(&ctl->tree_lock);
2714

2715 2716 2717 2718 2719 2720
		ret = do_trimming(block_group, total_trimmed, start, bytes,
				  extent_start, extent_bytes);
		if (ret)
			break;
next:
		start += bytes;
2721

2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 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 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790
		if (fatal_signal_pending(current)) {
			ret = -ERESTARTSYS;
			break;
		}

		cond_resched();
	}
out:
	return ret;
}

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

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

		spin_lock(&ctl->tree_lock);

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

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

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

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

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

		spin_unlock(&ctl->tree_lock);

		ret = do_trimming(block_group, total_trimmed, start, bytes,
				  start, bytes);
		if (ret)
			break;
next:
		if (next_bitmap) {
			offset += BITS_PER_BITMAP * ctl->unit;
		} else {
			start += bytes;
			if (start >= offset + BITS_PER_BITMAP * ctl->unit)
				offset += BITS_PER_BITMAP * ctl->unit;
2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802
		}

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

		cond_resched();
	}

	return ret;
}
2803

2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819
int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
			   u64 *trimmed, u64 start, u64 end, u64 minlen)
{
	int ret;

	*trimmed = 0;

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

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

	return ret;
}

2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856
/*
 * Find the left-most item in the cache tree, and then return the
 * smallest inode number in the item.
 *
 * Note: the returned inode number may not be the smallest one in
 * the tree, if the left-most item is a bitmap.
 */
u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
{
	struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
	struct btrfs_free_space *entry = NULL;
	u64 ino = 0;

	spin_lock(&ctl->tree_lock);

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

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

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

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

		ret = search_bitmap(ctl, entry, &offset, &count);
2857
		/* Logic error; Should be empty if it can't find anything */
2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869
		BUG_ON(ret);

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

	return ino;
}
2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887

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

	spin_lock(&root->cache_lock);
	if (root->cache_inode)
		inode = igrab(root->cache_inode);
	spin_unlock(&root->cache_lock);
	if (inode)
		return inode;

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

	spin_lock(&root->cache_lock);
2888
	if (!btrfs_fs_closing(root->fs_info))
2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910
		root->cache_inode = igrab(inode);
	spin_unlock(&root->cache_lock);

	return inode;
}

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

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

C
Chris Mason 已提交
2911 2912 2913
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

2914 2915 2916 2917
	/*
	 * If we're unmounting then just return, since this does a search on the
	 * normal root and not the commit root and we could deadlock.
	 */
2918
	if (btrfs_fs_closing(fs_info))
2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934
		return 0;

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

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

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

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

	if (ret < 0)
2935 2936 2937
		btrfs_err(fs_info,
			"failed to load free ino cache for root %llu",
			root->root_key.objectid);
2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952
out_put:
	iput(inode);
out:
	btrfs_free_path(path);
	return ret;
}

int btrfs_write_out_ino_cache(struct btrfs_root *root,
			      struct btrfs_trans_handle *trans,
			      struct btrfs_path *path)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct inode *inode;
	int ret;

C
Chris Mason 已提交
2953 2954 2955
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

2956 2957 2958 2959 2960
	inode = lookup_free_ino_inode(root, path);
	if (IS_ERR(inode))
		return 0;

	ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2961 2962 2963
	if (ret) {
		btrfs_delalloc_release_metadata(inode, inode->i_size);
#ifdef DEBUG
2964 2965 2966
		btrfs_err(root->fs_info,
			"failed to write free ino cache for root %llu",
			root->root_key.objectid);
2967 2968
#endif
	}
2969 2970 2971 2972

	iput(inode);
	return ret;
}
2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151

#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
static struct btrfs_block_group_cache *init_test_block_group(void)
{
	struct btrfs_block_group_cache *cache;

	cache = kzalloc(sizeof(*cache), GFP_NOFS);
	if (!cache)
		return NULL;
	cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
					GFP_NOFS);
	if (!cache->free_space_ctl) {
		kfree(cache);
		return NULL;
	}

	cache->key.objectid = 0;
	cache->key.offset = 1024 * 1024 * 1024;
	cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
	cache->sectorsize = 4096;

	spin_lock_init(&cache->lock);
	INIT_LIST_HEAD(&cache->list);
	INIT_LIST_HEAD(&cache->cluster_list);
	INIT_LIST_HEAD(&cache->new_bg_list);

	btrfs_init_free_space_ctl(cache);

	return cache;
}

/*
 * 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.
 */
static int check_exists(struct btrfs_block_group_cache *cache, u64 offset,
			u64 bytes)
{
	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
	struct btrfs_free_space *info;
	int ret = 0;

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

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

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

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

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

		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;
}

/*
 * 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.
 */
static int add_free_space_entry(struct btrfs_block_group_cache *cache,
				u64 offset, u64 bytes, bool bitmap)
{
	struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
	struct btrfs_free_space *info = NULL, *bitmap_info;
	void *map = NULL;
	u64 bytes_added;
	int ret;

again:
	if (!info) {
		info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
		if (!info)
			return -ENOMEM;
	}

	if (!bitmap) {
		spin_lock(&ctl->tree_lock);
		info->offset = offset;
		info->bytes = bytes;
		ret = link_free_space(ctl, info);
		spin_unlock(&ctl->tree_lock);
		if (ret)
			kmem_cache_free(btrfs_free_space_cachep, info);
		return ret;
	}

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

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

	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
	bytes -= bytes_added;
	offset += bytes_added;
	spin_unlock(&ctl->tree_lock);

	if (bytes)
		goto again;

	if (map)
		kfree(map);
	return 0;
}

3152 3153
#define test_msg(fmt, ...) printk(KERN_INFO "btrfs: selftest: " fmt, ##__VA_ARGS__)

3154 3155 3156 3157 3158 3159 3160 3161 3162
/*
 * This test just does basic sanity checking, making sure we can add an exten
 * entry and remove space from either end and the middle, and make sure we can
 * remove space that covers adjacent extent entries.
 */
static int test_extents(struct btrfs_block_group_cache *cache)
{
	int ret = 0;

3163
	test_msg("Running extent only tests\n");
3164 3165 3166 3167

	/* First just make sure we can remove an entire entry */
	ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
	if (ret) {
3168
		test_msg("Error adding initial extents %d\n", ret);
3169 3170 3171 3172 3173
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
	if (ret) {
3174
		test_msg("Error removing extent %d\n", ret);
3175 3176 3177 3178
		return ret;
	}

	if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3179
		test_msg("Full remove left some lingering space\n");
3180 3181 3182 3183 3184 3185
		return -1;
	}

	/* Ok edge and middle cases now */
	ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
	if (ret) {
3186
		test_msg("Error adding half extent %d\n", ret);
3187 3188 3189 3190 3191
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 1 * 1024 * 1024);
	if (ret) {
3192
		test_msg("Error removing tail end %d\n", ret);
3193 3194 3195 3196 3197
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
	if (ret) {
3198
		test_msg("Error removing front end %d\n", ret);
3199 3200 3201 3202 3203
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 2 * 1024 * 1024, 4096);
	if (ret) {
3204
		test_msg("Error removing middle piece %d\n", ret);
3205 3206 3207 3208
		return ret;
	}

	if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3209
		test_msg("Still have space at the front\n");
3210 3211 3212 3213
		return -1;
	}

	if (check_exists(cache, 2 * 1024 * 1024, 4096)) {
3214
		test_msg("Still have space in the middle\n");
3215 3216 3217 3218
		return -1;
	}

	if (check_exists(cache, 3 * 1024 * 1024, 1 * 1024 * 1024)) {
3219
		test_msg("Still have space at the end\n");
3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233
		return -1;
	}

	/* Cleanup */
	__btrfs_remove_free_space_cache(cache->free_space_ctl);

	return 0;
}

static int test_bitmaps(struct btrfs_block_group_cache *cache)
{
	u64 next_bitmap_offset;
	int ret;

3234
	test_msg("Running bitmap only tests\n");
3235 3236 3237

	ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
	if (ret) {
3238
		test_msg("Couldn't create a bitmap entry %d\n", ret);
3239 3240 3241 3242 3243
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
	if (ret) {
3244
		test_msg("Error removing bitmap full range %d\n", ret);
3245 3246 3247 3248
		return ret;
	}

	if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3249
		test_msg("Left some space in bitmap\n");
3250 3251 3252 3253 3254
		return -1;
	}

	ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
	if (ret) {
3255
		test_msg("Couldn't add to our bitmap entry %d\n", ret);
3256 3257 3258 3259 3260
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 2 * 1024 * 1024);
	if (ret) {
3261
		test_msg("Couldn't remove middle chunk %d\n", ret);
3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274
		return ret;
	}

	/*
	 * The first bitmap we have starts at offset 0 so the next one is just
	 * at the end of the first bitmap.
	 */
	next_bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);

	/* Test a bit straddling two bitmaps */
	ret = add_free_space_entry(cache, next_bitmap_offset -
				   (2 * 1024 * 1024), 4 * 1024 * 1024, 1);
	if (ret) {
3275 3276
		test_msg("Couldn't add space that straddles two bitmaps %d\n",
				ret);
3277 3278 3279 3280 3281 3282
		return ret;
	}

	ret = btrfs_remove_free_space(cache, next_bitmap_offset -
				      (1 * 1024 * 1024), 2 * 1024 * 1024);
	if (ret) {
3283
		test_msg("Couldn't remove overlapping space %d\n", ret);
3284 3285 3286 3287 3288
		return ret;
	}

	if (check_exists(cache, next_bitmap_offset - (1 * 1024 * 1024),
			 2 * 1024 * 1024)) {
3289
		test_msg("Left some space when removing overlapping\n");
3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303
		return -1;
	}

	__btrfs_remove_free_space_cache(cache->free_space_ctl);

	return 0;
}

/* This is the high grade jackassery */
static int test_bitmaps_and_extents(struct btrfs_block_group_cache *cache)
{
	u64 bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
	int ret;

3304
	test_msg("Running bitmap and extent tests\n");
3305 3306 3307 3308 3309 3310 3311 3312

	/*
	 * First let's do something simple, an extent at the same offset as the
	 * bitmap, but the free space completely in the extent and then
	 * completely in the bitmap.
	 */
	ret = add_free_space_entry(cache, 4 * 1024 * 1024, 1 * 1024 * 1024, 1);
	if (ret) {
3313
		test_msg("Couldn't create bitmap entry %d\n", ret);
3314 3315 3316 3317 3318
		return ret;
	}

	ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
	if (ret) {
3319
		test_msg("Couldn't add extent entry %d\n", ret);
3320 3321 3322 3323 3324
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
	if (ret) {
3325
		test_msg("Couldn't remove extent entry %d\n", ret);
3326 3327 3328 3329
		return ret;
	}

	if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3330
		test_msg("Left remnants after our remove\n");
3331 3332 3333 3334 3335 3336
		return -1;
	}

	/* Now to add back the extent entry and remove from the bitmap */
	ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
	if (ret) {
3337
		test_msg("Couldn't re-add extent entry %d\n", ret);
3338 3339 3340 3341 3342
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 4 * 1024 * 1024, 1 * 1024 * 1024);
	if (ret) {
3343
		test_msg("Couldn't remove from bitmap %d\n", ret);
3344 3345 3346 3347
		return ret;
	}

	if (check_exists(cache, 4 * 1024 * 1024, 1 * 1024 * 1024)) {
3348
		test_msg("Left remnants in the bitmap\n");
3349 3350 3351 3352 3353 3354 3355 3356 3357
		return -1;
	}

	/*
	 * Ok so a little more evil, extent entry and bitmap at the same offset,
	 * removing an overlapping chunk.
	 */
	ret = add_free_space_entry(cache, 1 * 1024 * 1024, 4 * 1024 * 1024, 1);
	if (ret) {
3358
		test_msg("Couldn't add to a bitmap %d\n", ret);
3359 3360 3361 3362 3363
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 512 * 1024, 3 * 1024 * 1024);
	if (ret) {
3364
		test_msg("Couldn't remove overlapping space %d\n", ret);
3365 3366 3367 3368
		return ret;
	}

	if (check_exists(cache, 512 * 1024, 3 * 1024 * 1024)) {
3369
		test_msg("Left over peices after removing overlapping\n");
3370 3371 3372 3373 3374 3375 3376 3377
		return -1;
	}

	__btrfs_remove_free_space_cache(cache->free_space_ctl);

	/* Now with the extent entry offset into the bitmap */
	ret = add_free_space_entry(cache, 4 * 1024 * 1024, 4 * 1024 * 1024, 1);
	if (ret) {
3378
		test_msg("Couldn't add space to the bitmap %d\n", ret);
3379 3380 3381 3382 3383
		return ret;
	}

	ret = add_free_space_entry(cache, 2 * 1024 * 1024, 2 * 1024 * 1024, 0);
	if (ret) {
3384
		test_msg("Couldn't add extent to the cache %d\n", ret);
3385 3386 3387 3388 3389
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 4 * 1024 * 1024);
	if (ret) {
3390
		test_msg("Problem removing overlapping space %d\n", ret);
3391 3392 3393 3394
		return ret;
	}

	if (check_exists(cache, 3 * 1024 * 1024, 4 * 1024 * 1024)) {
3395
		test_msg("Left something behind when removing space");
3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412
		return -1;
	}

	/*
	 * This has blown up in the past, the extent entry starts before the
	 * bitmap entry, but we're trying to remove an offset that falls
	 * completely within the bitmap range and is in both the extent entry
	 * and the bitmap entry, looks like this
	 *
	 *   [ extent ]
	 *      [ bitmap ]
	 *        [ del ]
	 */
	__btrfs_remove_free_space_cache(cache->free_space_ctl);
	ret = add_free_space_entry(cache, bitmap_offset + 4 * 1024 * 1024,
				   4 * 1024 * 1024, 1);
	if (ret) {
3413
		test_msg("Couldn't add bitmap %d\n", ret);
3414 3415 3416 3417 3418 3419
		return ret;
	}

	ret = add_free_space_entry(cache, bitmap_offset - 1 * 1024 * 1024,
				   5 * 1024 * 1024, 0);
	if (ret) {
3420
		test_msg("Couldn't add extent entry %d\n", ret);
3421 3422 3423 3424 3425 3426
		return ret;
	}

	ret = btrfs_remove_free_space(cache, bitmap_offset + 1 * 1024 * 1024,
				      5 * 1024 * 1024);
	if (ret) {
3427
		test_msg("Failed to free our space %d\n", ret);
3428 3429 3430 3431 3432
		return ret;
	}

	if (check_exists(cache, bitmap_offset + 1 * 1024 * 1024,
			 5 * 1024 * 1024)) {
3433
		test_msg("Left stuff over\n");
3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446
		return -1;
	}

	__btrfs_remove_free_space_cache(cache->free_space_ctl);

	/*
	 * This blew up before, we have part of the free space in a bitmap and
	 * then the entirety of the rest of the space in an extent.  This used
	 * to return -EAGAIN back from btrfs_remove_extent, make sure this
	 * doesn't happen.
	 */
	ret = add_free_space_entry(cache, 1 * 1024 * 1024, 2 * 1024 * 1024, 1);
	if (ret) {
3447
		test_msg("Couldn't add bitmap entry %d\n", ret);
3448 3449 3450 3451 3452
		return ret;
	}

	ret = add_free_space_entry(cache, 3 * 1024 * 1024, 1 * 1024 * 1024, 0);
	if (ret) {
3453
		test_msg("Couldn't add extent entry %d\n", ret);
3454 3455 3456 3457 3458
		return ret;
	}

	ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 3 * 1024 * 1024);
	if (ret) {
3459
		test_msg("Error removing bitmap and extent overlapping %d\n", ret);
3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470
		return ret;
	}

	__btrfs_remove_free_space_cache(cache->free_space_ctl);
	return 0;
}

void btrfs_test_free_space_cache(void)
{
	struct btrfs_block_group_cache *cache;

3471
	test_msg("Running btrfs free space cache tests\n");
3472 3473 3474

	cache = init_test_block_group();
	if (!cache) {
3475
		test_msg("Couldn't run the tests\n");
3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488
		return;
	}

	if (test_extents(cache))
		goto out;
	if (test_bitmaps(cache))
		goto out;
	if (test_bitmaps_and_extents(cache))
		goto out;
out:
	__btrfs_remove_free_space_cache(cache->free_space_ctl);
	kfree(cache->free_space_ctl);
	kfree(cache);
3489
	test_msg("Free space cache tests finished\n");
3490
}
3491
#undef test_msg
3492 3493 3494
#else /* !CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
void btrfs_test_free_space_cache(void) {}
#endif /* !CONFIG_BTRFS_FS_RUN_SANITY_TESTS */