2 * Copyright (C) 2008 Red Hat. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
23 #include <linux/ratelimit.h>
25 #include "free-space-cache.h"
26 #include "transaction.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
32 #define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
33 #define MAX_CACHE_BYTES_PER_GIG SZ_32K
35 struct btrfs_trim_range {
38 struct list_head list;
41 static int link_free_space(struct btrfs_free_space_ctl *ctl,
42 struct btrfs_free_space *info);
43 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
44 struct btrfs_free_space *info);
45 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
46 struct btrfs_trans_handle *trans,
47 struct btrfs_io_ctl *io_ctl,
48 struct btrfs_path *path);
50 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
51 struct btrfs_path *path,
54 struct btrfs_fs_info *fs_info = root->fs_info;
56 struct btrfs_key location;
57 struct btrfs_disk_key disk_key;
58 struct btrfs_free_space_header *header;
59 struct extent_buffer *leaf;
60 struct inode *inode = NULL;
63 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
67 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
71 btrfs_release_path(path);
72 return ERR_PTR(-ENOENT);
75 leaf = path->nodes[0];
76 header = btrfs_item_ptr(leaf, path->slots[0],
77 struct btrfs_free_space_header);
78 btrfs_free_space_key(leaf, header, &disk_key);
79 btrfs_disk_key_to_cpu(&location, &disk_key);
80 btrfs_release_path(path);
82 inode = btrfs_iget(fs_info->sb, &location, root, NULL);
85 if (is_bad_inode(inode)) {
87 return ERR_PTR(-ENOENT);
90 mapping_set_gfp_mask(inode->i_mapping,
91 mapping_gfp_constraint(inode->i_mapping,
92 ~(__GFP_FS | __GFP_HIGHMEM)));
97 struct inode *lookup_free_space_inode(struct btrfs_fs_info *fs_info,
98 struct btrfs_block_group_cache
99 *block_group, struct btrfs_path *path)
101 struct inode *inode = NULL;
102 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
104 spin_lock(&block_group->lock);
105 if (block_group->inode)
106 inode = igrab(block_group->inode);
107 spin_unlock(&block_group->lock);
111 inode = __lookup_free_space_inode(fs_info->tree_root, path,
112 block_group->key.objectid);
116 spin_lock(&block_group->lock);
117 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
118 btrfs_info(fs_info, "Old style space inode found, converting.");
119 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
120 BTRFS_INODE_NODATACOW;
121 block_group->disk_cache_state = BTRFS_DC_CLEAR;
124 if (!block_group->iref) {
125 block_group->inode = igrab(inode);
126 block_group->iref = 1;
128 spin_unlock(&block_group->lock);
133 static int __create_free_space_inode(struct btrfs_root *root,
134 struct btrfs_trans_handle *trans,
135 struct btrfs_path *path,
138 struct btrfs_key key;
139 struct btrfs_disk_key disk_key;
140 struct btrfs_free_space_header *header;
141 struct btrfs_inode_item *inode_item;
142 struct extent_buffer *leaf;
143 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
146 ret = btrfs_insert_empty_inode(trans, root, path, ino);
150 /* We inline crc's for the free disk space cache */
151 if (ino != BTRFS_FREE_INO_OBJECTID)
152 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
154 leaf = path->nodes[0];
155 inode_item = btrfs_item_ptr(leaf, path->slots[0],
156 struct btrfs_inode_item);
157 btrfs_item_key(leaf, &disk_key, path->slots[0]);
158 memzero_extent_buffer(leaf, (unsigned long)inode_item,
159 sizeof(*inode_item));
160 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
161 btrfs_set_inode_size(leaf, inode_item, 0);
162 btrfs_set_inode_nbytes(leaf, inode_item, 0);
163 btrfs_set_inode_uid(leaf, inode_item, 0);
164 btrfs_set_inode_gid(leaf, inode_item, 0);
165 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
166 btrfs_set_inode_flags(leaf, inode_item, flags);
167 btrfs_set_inode_nlink(leaf, inode_item, 1);
168 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
169 btrfs_set_inode_block_group(leaf, inode_item, offset);
170 btrfs_mark_buffer_dirty(leaf);
171 btrfs_release_path(path);
173 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
176 ret = btrfs_insert_empty_item(trans, root, path, &key,
177 sizeof(struct btrfs_free_space_header));
179 btrfs_release_path(path);
183 leaf = path->nodes[0];
184 header = btrfs_item_ptr(leaf, path->slots[0],
185 struct btrfs_free_space_header);
186 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
187 btrfs_set_free_space_key(leaf, header, &disk_key);
188 btrfs_mark_buffer_dirty(leaf);
189 btrfs_release_path(path);
194 int create_free_space_inode(struct btrfs_fs_info *fs_info,
195 struct btrfs_trans_handle *trans,
196 struct btrfs_block_group_cache *block_group,
197 struct btrfs_path *path)
202 ret = btrfs_find_free_objectid(fs_info->tree_root, &ino);
206 return __create_free_space_inode(fs_info->tree_root, trans, path, ino,
207 block_group->key.objectid);
210 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
211 struct btrfs_block_rsv *rsv)
216 /* 1 for slack space, 1 for updating the inode */
217 needed_bytes = btrfs_calc_trunc_metadata_size(fs_info, 1) +
218 btrfs_calc_trans_metadata_size(fs_info, 1);
220 spin_lock(&rsv->lock);
221 if (rsv->reserved < needed_bytes)
225 spin_unlock(&rsv->lock);
229 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
230 struct btrfs_block_group_cache *block_group,
233 struct btrfs_root *root = BTRFS_I(inode)->root;
238 struct btrfs_path *path = btrfs_alloc_path();
245 mutex_lock(&trans->transaction->cache_write_mutex);
246 if (!list_empty(&block_group->io_list)) {
247 list_del_init(&block_group->io_list);
249 btrfs_wait_cache_io(trans, block_group, path);
250 btrfs_put_block_group(block_group);
254 * now that we've truncated the cache away, its no longer
257 spin_lock(&block_group->lock);
258 block_group->disk_cache_state = BTRFS_DC_CLEAR;
259 spin_unlock(&block_group->lock);
260 btrfs_free_path(path);
263 btrfs_i_size_write(inode, 0);
264 truncate_pagecache(inode, 0);
267 * We don't need an orphan item because truncating the free space cache
268 * will never be split across transactions.
269 * We don't need to check for -EAGAIN because we're a free space
272 ret = btrfs_truncate_inode_items(trans, root, inode,
273 0, BTRFS_EXTENT_DATA_KEY);
277 ret = btrfs_update_inode(trans, root, inode);
281 mutex_unlock(&trans->transaction->cache_write_mutex);
283 btrfs_abort_transaction(trans, ret);
288 static void readahead_cache(struct inode *inode)
290 struct file_ra_state *ra;
291 unsigned long last_index;
293 ra = kzalloc(sizeof(*ra), GFP_NOFS);
297 file_ra_state_init(ra, inode->i_mapping);
298 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
300 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
305 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
311 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
313 if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FREE_INO_OBJECTID)
316 /* Make sure we can fit our crcs into the first page */
317 if (write && check_crcs &&
318 (num_pages * sizeof(u32)) >= PAGE_SIZE)
321 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
323 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
327 io_ctl->num_pages = num_pages;
328 io_ctl->fs_info = btrfs_sb(inode->i_sb);
329 io_ctl->check_crcs = check_crcs;
330 io_ctl->inode = inode;
335 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
337 kfree(io_ctl->pages);
338 io_ctl->pages = NULL;
341 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
349 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
351 ASSERT(io_ctl->index < io_ctl->num_pages);
352 io_ctl->page = io_ctl->pages[io_ctl->index++];
353 io_ctl->cur = page_address(io_ctl->page);
354 io_ctl->orig = io_ctl->cur;
355 io_ctl->size = PAGE_SIZE;
357 memset(io_ctl->cur, 0, PAGE_SIZE);
360 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
364 io_ctl_unmap_page(io_ctl);
366 for (i = 0; i < io_ctl->num_pages; i++) {
367 if (io_ctl->pages[i]) {
368 ClearPageChecked(io_ctl->pages[i]);
369 unlock_page(io_ctl->pages[i]);
370 put_page(io_ctl->pages[i]);
375 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
379 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
382 for (i = 0; i < io_ctl->num_pages; i++) {
383 page = find_or_create_page(inode->i_mapping, i, mask);
385 io_ctl_drop_pages(io_ctl);
388 io_ctl->pages[i] = page;
389 if (uptodate && !PageUptodate(page)) {
390 btrfs_readpage(NULL, page);
392 if (!PageUptodate(page)) {
393 btrfs_err(BTRFS_I(inode)->root->fs_info,
394 "error reading free space cache");
395 io_ctl_drop_pages(io_ctl);
401 for (i = 0; i < io_ctl->num_pages; i++) {
402 clear_page_dirty_for_io(io_ctl->pages[i]);
403 set_page_extent_mapped(io_ctl->pages[i]);
409 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
413 io_ctl_map_page(io_ctl, 1);
416 * Skip the csum areas. If we don't check crcs then we just have a
417 * 64bit chunk at the front of the first page.
419 if (io_ctl->check_crcs) {
420 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
421 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
423 io_ctl->cur += sizeof(u64);
424 io_ctl->size -= sizeof(u64) * 2;
428 *val = cpu_to_le64(generation);
429 io_ctl->cur += sizeof(u64);
432 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
437 * Skip the crc area. If we don't check crcs then we just have a 64bit
438 * chunk at the front of the first page.
440 if (io_ctl->check_crcs) {
441 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
442 io_ctl->size -= sizeof(u64) +
443 (sizeof(u32) * io_ctl->num_pages);
445 io_ctl->cur += sizeof(u64);
446 io_ctl->size -= sizeof(u64) * 2;
450 if (le64_to_cpu(*gen) != generation) {
451 btrfs_err_rl(io_ctl->fs_info,
452 "space cache generation (%llu) does not match inode (%llu)",
454 io_ctl_unmap_page(io_ctl);
457 io_ctl->cur += sizeof(u64);
461 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
467 if (!io_ctl->check_crcs) {
468 io_ctl_unmap_page(io_ctl);
473 offset = sizeof(u32) * io_ctl->num_pages;
475 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
477 btrfs_csum_final(crc, (u8 *)&crc);
478 io_ctl_unmap_page(io_ctl);
479 tmp = page_address(io_ctl->pages[0]);
484 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
490 if (!io_ctl->check_crcs) {
491 io_ctl_map_page(io_ctl, 0);
496 offset = sizeof(u32) * io_ctl->num_pages;
498 tmp = page_address(io_ctl->pages[0]);
502 io_ctl_map_page(io_ctl, 0);
503 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
505 btrfs_csum_final(crc, (u8 *)&crc);
507 btrfs_err_rl(io_ctl->fs_info,
508 "csum mismatch on free space cache");
509 io_ctl_unmap_page(io_ctl);
516 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
519 struct btrfs_free_space_entry *entry;
525 entry->offset = cpu_to_le64(offset);
526 entry->bytes = cpu_to_le64(bytes);
527 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
528 BTRFS_FREE_SPACE_EXTENT;
529 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
530 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
532 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
535 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
537 /* No more pages to map */
538 if (io_ctl->index >= io_ctl->num_pages)
541 /* map the next page */
542 io_ctl_map_page(io_ctl, 1);
546 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
552 * If we aren't at the start of the current page, unmap this one and
553 * map the next one if there is any left.
555 if (io_ctl->cur != io_ctl->orig) {
556 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
557 if (io_ctl->index >= io_ctl->num_pages)
559 io_ctl_map_page(io_ctl, 0);
562 memcpy(io_ctl->cur, bitmap, PAGE_SIZE);
563 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
564 if (io_ctl->index < io_ctl->num_pages)
565 io_ctl_map_page(io_ctl, 0);
569 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
572 * If we're not on the boundary we know we've modified the page and we
573 * need to crc the page.
575 if (io_ctl->cur != io_ctl->orig)
576 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
578 io_ctl_unmap_page(io_ctl);
580 while (io_ctl->index < io_ctl->num_pages) {
581 io_ctl_map_page(io_ctl, 1);
582 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
586 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
587 struct btrfs_free_space *entry, u8 *type)
589 struct btrfs_free_space_entry *e;
593 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
599 entry->offset = le64_to_cpu(e->offset);
600 entry->bytes = le64_to_cpu(e->bytes);
602 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
603 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
605 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
608 io_ctl_unmap_page(io_ctl);
613 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
614 struct btrfs_free_space *entry)
618 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
622 memcpy(entry->bitmap, io_ctl->cur, PAGE_SIZE);
623 io_ctl_unmap_page(io_ctl);
629 * Since we attach pinned extents after the fact we can have contiguous sections
630 * of free space that are split up in entries. This poses a problem with the
631 * tree logging stuff since it could have allocated across what appears to be 2
632 * entries since we would have merged the entries when adding the pinned extents
633 * back to the free space cache. So run through the space cache that we just
634 * loaded and merge contiguous entries. This will make the log replay stuff not
635 * blow up and it will make for nicer allocator behavior.
637 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
639 struct btrfs_free_space *e, *prev = NULL;
643 spin_lock(&ctl->tree_lock);
644 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
645 e = rb_entry(n, struct btrfs_free_space, offset_index);
648 if (e->bitmap || prev->bitmap)
650 if (prev->offset + prev->bytes == e->offset) {
651 unlink_free_space(ctl, prev);
652 unlink_free_space(ctl, e);
653 prev->bytes += e->bytes;
654 kmem_cache_free(btrfs_free_space_cachep, e);
655 link_free_space(ctl, prev);
657 spin_unlock(&ctl->tree_lock);
663 spin_unlock(&ctl->tree_lock);
666 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
667 struct btrfs_free_space_ctl *ctl,
668 struct btrfs_path *path, u64 offset)
670 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
671 struct btrfs_free_space_header *header;
672 struct extent_buffer *leaf;
673 struct btrfs_io_ctl io_ctl;
674 struct btrfs_key key;
675 struct btrfs_free_space *e, *n;
683 /* Nothing in the space cache, goodbye */
684 if (!i_size_read(inode))
687 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
691 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
695 btrfs_release_path(path);
701 leaf = path->nodes[0];
702 header = btrfs_item_ptr(leaf, path->slots[0],
703 struct btrfs_free_space_header);
704 num_entries = btrfs_free_space_entries(leaf, header);
705 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
706 generation = btrfs_free_space_generation(leaf, header);
707 btrfs_release_path(path);
709 if (!BTRFS_I(inode)->generation) {
711 "The free space cache file (%llu) is invalid. skip it\n",
716 if (BTRFS_I(inode)->generation != generation) {
718 "free space inode generation (%llu) did not match free space cache generation (%llu)",
719 BTRFS_I(inode)->generation, generation);
726 ret = io_ctl_init(&io_ctl, inode, 0);
730 readahead_cache(inode);
732 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
736 ret = io_ctl_check_crc(&io_ctl, 0);
740 ret = io_ctl_check_generation(&io_ctl, generation);
744 while (num_entries) {
745 e = kmem_cache_zalloc(btrfs_free_space_cachep,
750 ret = io_ctl_read_entry(&io_ctl, e, &type);
752 kmem_cache_free(btrfs_free_space_cachep, e);
757 kmem_cache_free(btrfs_free_space_cachep, e);
761 if (type == BTRFS_FREE_SPACE_EXTENT) {
762 spin_lock(&ctl->tree_lock);
763 ret = link_free_space(ctl, e);
764 spin_unlock(&ctl->tree_lock);
767 "Duplicate entries in free space cache, dumping");
768 kmem_cache_free(btrfs_free_space_cachep, e);
774 e->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
777 btrfs_free_space_cachep, e);
780 spin_lock(&ctl->tree_lock);
781 ret = link_free_space(ctl, e);
782 ctl->total_bitmaps++;
783 ctl->op->recalc_thresholds(ctl);
784 spin_unlock(&ctl->tree_lock);
787 "Duplicate entries in free space cache, dumping");
788 kmem_cache_free(btrfs_free_space_cachep, e);
791 list_add_tail(&e->list, &bitmaps);
797 io_ctl_unmap_page(&io_ctl);
800 * We add the bitmaps at the end of the entries in order that
801 * the bitmap entries are added to the cache.
803 list_for_each_entry_safe(e, n, &bitmaps, list) {
804 list_del_init(&e->list);
805 ret = io_ctl_read_bitmap(&io_ctl, e);
810 io_ctl_drop_pages(&io_ctl);
811 merge_space_tree(ctl);
814 io_ctl_free(&io_ctl);
817 io_ctl_drop_pages(&io_ctl);
818 __btrfs_remove_free_space_cache(ctl);
822 int load_free_space_cache(struct btrfs_fs_info *fs_info,
823 struct btrfs_block_group_cache *block_group)
825 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
827 struct btrfs_path *path;
830 u64 used = btrfs_block_group_used(&block_group->item);
833 * If this block group has been marked to be cleared for one reason or
834 * another then we can't trust the on disk cache, so just return.
836 spin_lock(&block_group->lock);
837 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
838 spin_unlock(&block_group->lock);
841 spin_unlock(&block_group->lock);
843 path = btrfs_alloc_path();
846 path->search_commit_root = 1;
847 path->skip_locking = 1;
849 inode = lookup_free_space_inode(fs_info, block_group, path);
851 btrfs_free_path(path);
855 /* We may have converted the inode and made the cache invalid. */
856 spin_lock(&block_group->lock);
857 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
858 spin_unlock(&block_group->lock);
859 btrfs_free_path(path);
862 spin_unlock(&block_group->lock);
864 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
865 path, block_group->key.objectid);
866 btrfs_free_path(path);
870 spin_lock(&ctl->tree_lock);
871 matched = (ctl->free_space == (block_group->key.offset - used -
872 block_group->bytes_super));
873 spin_unlock(&ctl->tree_lock);
876 __btrfs_remove_free_space_cache(ctl);
878 "block group %llu has wrong amount of free space",
879 block_group->key.objectid);
884 /* This cache is bogus, make sure it gets cleared */
885 spin_lock(&block_group->lock);
886 block_group->disk_cache_state = BTRFS_DC_CLEAR;
887 spin_unlock(&block_group->lock);
891 "failed to load free space cache for block group %llu, rebuilding it now",
892 block_group->key.objectid);
899 static noinline_for_stack
900 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
901 struct btrfs_free_space_ctl *ctl,
902 struct btrfs_block_group_cache *block_group,
903 int *entries, int *bitmaps,
904 struct list_head *bitmap_list)
907 struct btrfs_free_cluster *cluster = NULL;
908 struct btrfs_free_cluster *cluster_locked = NULL;
909 struct rb_node *node = rb_first(&ctl->free_space_offset);
910 struct btrfs_trim_range *trim_entry;
912 /* Get the cluster for this block_group if it exists */
913 if (block_group && !list_empty(&block_group->cluster_list)) {
914 cluster = list_entry(block_group->cluster_list.next,
915 struct btrfs_free_cluster,
919 if (!node && cluster) {
920 cluster_locked = cluster;
921 spin_lock(&cluster_locked->lock);
922 node = rb_first(&cluster->root);
926 /* Write out the extent entries */
928 struct btrfs_free_space *e;
930 e = rb_entry(node, struct btrfs_free_space, offset_index);
933 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
939 list_add_tail(&e->list, bitmap_list);
942 node = rb_next(node);
943 if (!node && cluster) {
944 node = rb_first(&cluster->root);
945 cluster_locked = cluster;
946 spin_lock(&cluster_locked->lock);
950 if (cluster_locked) {
951 spin_unlock(&cluster_locked->lock);
952 cluster_locked = NULL;
956 * Make sure we don't miss any range that was removed from our rbtree
957 * because trimming is running. Otherwise after a umount+mount (or crash
958 * after committing the transaction) we would leak free space and get
959 * an inconsistent free space cache report from fsck.
961 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
962 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
963 trim_entry->bytes, NULL);
972 spin_unlock(&cluster_locked->lock);
976 static noinline_for_stack int
977 update_cache_item(struct btrfs_trans_handle *trans,
978 struct btrfs_root *root,
980 struct btrfs_path *path, u64 offset,
981 int entries, int bitmaps)
983 struct btrfs_key key;
984 struct btrfs_free_space_header *header;
985 struct extent_buffer *leaf;
988 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
992 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
994 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
995 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
999 leaf = path->nodes[0];
1001 struct btrfs_key found_key;
1002 ASSERT(path->slots[0]);
1004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1005 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1006 found_key.offset != offset) {
1007 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1009 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1011 btrfs_release_path(path);
1016 BTRFS_I(inode)->generation = trans->transid;
1017 header = btrfs_item_ptr(leaf, path->slots[0],
1018 struct btrfs_free_space_header);
1019 btrfs_set_free_space_entries(leaf, header, entries);
1020 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1021 btrfs_set_free_space_generation(leaf, header, trans->transid);
1022 btrfs_mark_buffer_dirty(leaf);
1023 btrfs_release_path(path);
1031 static noinline_for_stack int
1032 write_pinned_extent_entries(struct btrfs_fs_info *fs_info,
1033 struct btrfs_block_group_cache *block_group,
1034 struct btrfs_io_ctl *io_ctl,
1037 u64 start, extent_start, extent_end, len;
1038 struct extent_io_tree *unpin = NULL;
1045 * We want to add any pinned extents to our free space cache
1046 * so we don't leak the space
1048 * We shouldn't have switched the pinned extents yet so this is the
1051 unpin = fs_info->pinned_extents;
1053 start = block_group->key.objectid;
1055 while (start < block_group->key.objectid + block_group->key.offset) {
1056 ret = find_first_extent_bit(unpin, start,
1057 &extent_start, &extent_end,
1058 EXTENT_DIRTY, NULL);
1062 /* This pinned extent is out of our range */
1063 if (extent_start >= block_group->key.objectid +
1064 block_group->key.offset)
1067 extent_start = max(extent_start, start);
1068 extent_end = min(block_group->key.objectid +
1069 block_group->key.offset, extent_end + 1);
1070 len = extent_end - extent_start;
1073 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1083 static noinline_for_stack int
1084 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1086 struct btrfs_free_space *entry, *next;
1089 /* Write out the bitmaps */
1090 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1091 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1094 list_del_init(&entry->list);
1100 static int flush_dirty_cache(struct inode *inode)
1104 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1106 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1107 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1113 static void noinline_for_stack
1114 cleanup_bitmap_list(struct list_head *bitmap_list)
1116 struct btrfs_free_space *entry, *next;
1118 list_for_each_entry_safe(entry, next, bitmap_list, list)
1119 list_del_init(&entry->list);
1122 static void noinline_for_stack
1123 cleanup_write_cache_enospc(struct inode *inode,
1124 struct btrfs_io_ctl *io_ctl,
1125 struct extent_state **cached_state)
1127 io_ctl_drop_pages(io_ctl);
1128 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1129 i_size_read(inode) - 1, cached_state,
1133 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1134 struct btrfs_trans_handle *trans,
1135 struct btrfs_block_group_cache *block_group,
1136 struct btrfs_io_ctl *io_ctl,
1137 struct btrfs_path *path, u64 offset)
1140 struct inode *inode = io_ctl->inode;
1141 struct btrfs_fs_info *fs_info;
1146 fs_info = btrfs_sb(inode->i_sb);
1148 /* Flush the dirty pages in the cache file. */
1149 ret = flush_dirty_cache(inode);
1153 /* Update the cache item to tell everyone this cache file is valid. */
1154 ret = update_cache_item(trans, root, inode, path, offset,
1155 io_ctl->entries, io_ctl->bitmaps);
1157 io_ctl_free(io_ctl);
1159 invalidate_inode_pages2(inode->i_mapping);
1160 BTRFS_I(inode)->generation = 0;
1164 "failed to write free space cache for block group %llu",
1165 block_group->key.objectid);
1169 btrfs_update_inode(trans, root, inode);
1172 /* the dirty list is protected by the dirty_bgs_lock */
1173 spin_lock(&trans->transaction->dirty_bgs_lock);
1175 /* the disk_cache_state is protected by the block group lock */
1176 spin_lock(&block_group->lock);
1179 * only mark this as written if we didn't get put back on
1180 * the dirty list while waiting for IO. Otherwise our
1181 * cache state won't be right, and we won't get written again
1183 if (!ret && list_empty(&block_group->dirty_list))
1184 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1186 block_group->disk_cache_state = BTRFS_DC_ERROR;
1188 spin_unlock(&block_group->lock);
1189 spin_unlock(&trans->transaction->dirty_bgs_lock);
1190 io_ctl->inode = NULL;
1198 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
1199 struct btrfs_trans_handle *trans,
1200 struct btrfs_io_ctl *io_ctl,
1201 struct btrfs_path *path)
1203 return __btrfs_wait_cache_io(root, trans, NULL, io_ctl, path, 0);
1206 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1207 struct btrfs_block_group_cache *block_group,
1208 struct btrfs_path *path)
1210 return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1211 block_group, &block_group->io_ctl,
1212 path, block_group->key.objectid);
1216 * __btrfs_write_out_cache - write out cached info to an inode
1217 * @root - the root the inode belongs to
1218 * @ctl - the free space cache we are going to write out
1219 * @block_group - the block_group for this cache if it belongs to a block_group
1220 * @trans - the trans handle
1222 * This function writes out a free space cache struct to disk for quick recovery
1223 * on mount. This will return 0 if it was successful in writing the cache out,
1224 * or an errno if it was not.
1226 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1227 struct btrfs_free_space_ctl *ctl,
1228 struct btrfs_block_group_cache *block_group,
1229 struct btrfs_io_ctl *io_ctl,
1230 struct btrfs_trans_handle *trans)
1232 struct btrfs_fs_info *fs_info = root->fs_info;
1233 struct extent_state *cached_state = NULL;
1234 LIST_HEAD(bitmap_list);
1240 if (!i_size_read(inode))
1243 WARN_ON(io_ctl->pages);
1244 ret = io_ctl_init(io_ctl, inode, 1);
1248 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1249 down_write(&block_group->data_rwsem);
1250 spin_lock(&block_group->lock);
1251 if (block_group->delalloc_bytes) {
1252 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1253 spin_unlock(&block_group->lock);
1254 up_write(&block_group->data_rwsem);
1255 BTRFS_I(inode)->generation = 0;
1260 spin_unlock(&block_group->lock);
1263 /* Lock all pages first so we can lock the extent safely. */
1264 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1268 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1271 io_ctl_set_generation(io_ctl, trans->transid);
1273 mutex_lock(&ctl->cache_writeout_mutex);
1274 /* Write out the extent entries in the free space cache */
1275 spin_lock(&ctl->tree_lock);
1276 ret = write_cache_extent_entries(io_ctl, ctl,
1277 block_group, &entries, &bitmaps,
1280 goto out_nospc_locked;
1283 * Some spaces that are freed in the current transaction are pinned,
1284 * they will be added into free space cache after the transaction is
1285 * committed, we shouldn't lose them.
1287 * If this changes while we are working we'll get added back to
1288 * the dirty list and redo it. No locking needed
1290 ret = write_pinned_extent_entries(fs_info, block_group,
1293 goto out_nospc_locked;
1296 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1297 * locked while doing it because a concurrent trim can be manipulating
1298 * or freeing the bitmap.
1300 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1301 spin_unlock(&ctl->tree_lock);
1302 mutex_unlock(&ctl->cache_writeout_mutex);
1306 /* Zero out the rest of the pages just to make sure */
1307 io_ctl_zero_remaining_pages(io_ctl);
1309 /* Everything is written out, now we dirty the pages in the file. */
1310 ret = btrfs_dirty_pages(inode, io_ctl->pages, io_ctl->num_pages, 0,
1311 i_size_read(inode), &cached_state);
1315 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1316 up_write(&block_group->data_rwsem);
1318 * Release the pages and unlock the extent, we will flush
1321 io_ctl_drop_pages(io_ctl);
1323 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1324 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1327 * at this point the pages are under IO and we're happy,
1328 * The caller is responsible for waiting on them and updating the
1329 * the cache and the inode
1331 io_ctl->entries = entries;
1332 io_ctl->bitmaps = bitmaps;
1334 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1341 io_ctl->inode = NULL;
1342 io_ctl_free(io_ctl);
1344 invalidate_inode_pages2(inode->i_mapping);
1345 BTRFS_I(inode)->generation = 0;
1347 btrfs_update_inode(trans, root, inode);
1353 cleanup_bitmap_list(&bitmap_list);
1354 spin_unlock(&ctl->tree_lock);
1355 mutex_unlock(&ctl->cache_writeout_mutex);
1358 cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1360 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1361 up_write(&block_group->data_rwsem);
1366 int btrfs_write_out_cache(struct btrfs_fs_info *fs_info,
1367 struct btrfs_trans_handle *trans,
1368 struct btrfs_block_group_cache *block_group,
1369 struct btrfs_path *path)
1371 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1372 struct inode *inode;
1375 spin_lock(&block_group->lock);
1376 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1377 spin_unlock(&block_group->lock);
1380 spin_unlock(&block_group->lock);
1382 inode = lookup_free_space_inode(fs_info, block_group, path);
1386 ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1387 block_group, &block_group->io_ctl, trans);
1391 "failed to write free space cache for block group %llu",
1392 block_group->key.objectid);
1394 spin_lock(&block_group->lock);
1395 block_group->disk_cache_state = BTRFS_DC_ERROR;
1396 spin_unlock(&block_group->lock);
1398 block_group->io_ctl.inode = NULL;
1403 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1404 * to wait for IO and put the inode
1410 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1413 ASSERT(offset >= bitmap_start);
1414 offset -= bitmap_start;
1415 return (unsigned long)(div_u64(offset, unit));
1418 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1420 return (unsigned long)(div_u64(bytes, unit));
1423 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1427 u64 bytes_per_bitmap;
1429 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1430 bitmap_start = offset - ctl->start;
1431 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1432 bitmap_start *= bytes_per_bitmap;
1433 bitmap_start += ctl->start;
1435 return bitmap_start;
1438 static int tree_insert_offset(struct rb_root *root, u64 offset,
1439 struct rb_node *node, int bitmap)
1441 struct rb_node **p = &root->rb_node;
1442 struct rb_node *parent = NULL;
1443 struct btrfs_free_space *info;
1447 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1449 if (offset < info->offset) {
1451 } else if (offset > info->offset) {
1452 p = &(*p)->rb_right;
1455 * we could have a bitmap entry and an extent entry
1456 * share the same offset. If this is the case, we want
1457 * the extent entry to always be found first if we do a
1458 * linear search through the tree, since we want to have
1459 * the quickest allocation time, and allocating from an
1460 * extent is faster than allocating from a bitmap. So
1461 * if we're inserting a bitmap and we find an entry at
1462 * this offset, we want to go right, or after this entry
1463 * logically. If we are inserting an extent and we've
1464 * found a bitmap, we want to go left, or before
1472 p = &(*p)->rb_right;
1474 if (!info->bitmap) {
1483 rb_link_node(node, parent, p);
1484 rb_insert_color(node, root);
1490 * searches the tree for the given offset.
1492 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1493 * want a section that has at least bytes size and comes at or after the given
1496 static struct btrfs_free_space *
1497 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1498 u64 offset, int bitmap_only, int fuzzy)
1500 struct rb_node *n = ctl->free_space_offset.rb_node;
1501 struct btrfs_free_space *entry, *prev = NULL;
1503 /* find entry that is closest to the 'offset' */
1510 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1513 if (offset < entry->offset)
1515 else if (offset > entry->offset)
1528 * bitmap entry and extent entry may share same offset,
1529 * in that case, bitmap entry comes after extent entry.
1534 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1535 if (entry->offset != offset)
1538 WARN_ON(!entry->bitmap);
1541 if (entry->bitmap) {
1543 * if previous extent entry covers the offset,
1544 * we should return it instead of the bitmap entry
1546 n = rb_prev(&entry->offset_index);
1548 prev = rb_entry(n, struct btrfs_free_space,
1550 if (!prev->bitmap &&
1551 prev->offset + prev->bytes > offset)
1561 /* find last entry before the 'offset' */
1563 if (entry->offset > offset) {
1564 n = rb_prev(&entry->offset_index);
1566 entry = rb_entry(n, struct btrfs_free_space,
1568 ASSERT(entry->offset <= offset);
1577 if (entry->bitmap) {
1578 n = rb_prev(&entry->offset_index);
1580 prev = rb_entry(n, struct btrfs_free_space,
1582 if (!prev->bitmap &&
1583 prev->offset + prev->bytes > offset)
1586 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1588 } else if (entry->offset + entry->bytes > offset)
1595 if (entry->bitmap) {
1596 if (entry->offset + BITS_PER_BITMAP *
1600 if (entry->offset + entry->bytes > offset)
1604 n = rb_next(&entry->offset_index);
1607 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1613 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1614 struct btrfs_free_space *info)
1616 rb_erase(&info->offset_index, &ctl->free_space_offset);
1617 ctl->free_extents--;
1620 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1621 struct btrfs_free_space *info)
1623 __unlink_free_space(ctl, info);
1624 ctl->free_space -= info->bytes;
1627 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1628 struct btrfs_free_space *info)
1632 ASSERT(info->bytes || info->bitmap);
1633 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1634 &info->offset_index, (info->bitmap != NULL));
1638 ctl->free_space += info->bytes;
1639 ctl->free_extents++;
1643 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1645 struct btrfs_block_group_cache *block_group = ctl->private;
1649 u64 size = block_group->key.offset;
1650 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1651 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1653 max_bitmaps = max_t(u64, max_bitmaps, 1);
1655 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1658 * The goal is to keep the total amount of memory used per 1gb of space
1659 * at or below 32k, so we need to adjust how much memory we allow to be
1660 * used by extent based free space tracking
1663 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1665 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
1668 * we want to account for 1 more bitmap than what we have so we can make
1669 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1670 * we add more bitmaps.
1672 bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit;
1674 if (bitmap_bytes >= max_bytes) {
1675 ctl->extents_thresh = 0;
1680 * we want the extent entry threshold to always be at most 1/2 the max
1681 * bytes we can have, or whatever is less than that.
1683 extent_bytes = max_bytes - bitmap_bytes;
1684 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1686 ctl->extents_thresh =
1687 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1690 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1691 struct btrfs_free_space *info,
1692 u64 offset, u64 bytes)
1694 unsigned long start, count;
1696 start = offset_to_bit(info->offset, ctl->unit, offset);
1697 count = bytes_to_bits(bytes, ctl->unit);
1698 ASSERT(start + count <= BITS_PER_BITMAP);
1700 bitmap_clear(info->bitmap, start, count);
1702 info->bytes -= bytes;
1705 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1706 struct btrfs_free_space *info, u64 offset,
1709 __bitmap_clear_bits(ctl, info, offset, bytes);
1710 ctl->free_space -= bytes;
1713 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1714 struct btrfs_free_space *info, u64 offset,
1717 unsigned long start, count;
1719 start = offset_to_bit(info->offset, ctl->unit, offset);
1720 count = bytes_to_bits(bytes, ctl->unit);
1721 ASSERT(start + count <= BITS_PER_BITMAP);
1723 bitmap_set(info->bitmap, start, count);
1725 info->bytes += bytes;
1726 ctl->free_space += bytes;
1730 * If we can not find suitable extent, we will use bytes to record
1731 * the size of the max extent.
1733 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1734 struct btrfs_free_space *bitmap_info, u64 *offset,
1735 u64 *bytes, bool for_alloc)
1737 unsigned long found_bits = 0;
1738 unsigned long max_bits = 0;
1739 unsigned long bits, i;
1740 unsigned long next_zero;
1741 unsigned long extent_bits;
1744 * Skip searching the bitmap if we don't have a contiguous section that
1745 * is large enough for this allocation.
1748 bitmap_info->max_extent_size &&
1749 bitmap_info->max_extent_size < *bytes) {
1750 *bytes = bitmap_info->max_extent_size;
1754 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1755 max_t(u64, *offset, bitmap_info->offset));
1756 bits = bytes_to_bits(*bytes, ctl->unit);
1758 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1759 if (for_alloc && bits == 1) {
1763 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1764 BITS_PER_BITMAP, i);
1765 extent_bits = next_zero - i;
1766 if (extent_bits >= bits) {
1767 found_bits = extent_bits;
1769 } else if (extent_bits > max_bits) {
1770 max_bits = extent_bits;
1776 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1777 *bytes = (u64)(found_bits) * ctl->unit;
1781 *bytes = (u64)(max_bits) * ctl->unit;
1782 bitmap_info->max_extent_size = *bytes;
1786 /* Cache the size of the max extent in bytes */
1787 static struct btrfs_free_space *
1788 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1789 unsigned long align, u64 *max_extent_size)
1791 struct btrfs_free_space *entry;
1792 struct rb_node *node;
1797 if (!ctl->free_space_offset.rb_node)
1800 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1804 for (node = &entry->offset_index; node; node = rb_next(node)) {
1805 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1806 if (entry->bytes < *bytes) {
1807 if (entry->bytes > *max_extent_size)
1808 *max_extent_size = entry->bytes;
1812 /* make sure the space returned is big enough
1813 * to match our requested alignment
1815 if (*bytes >= align) {
1816 tmp = entry->offset - ctl->start + align - 1;
1817 tmp = div64_u64(tmp, align);
1818 tmp = tmp * align + ctl->start;
1819 align_off = tmp - entry->offset;
1822 tmp = entry->offset;
1825 if (entry->bytes < *bytes + align_off) {
1826 if (entry->bytes > *max_extent_size)
1827 *max_extent_size = entry->bytes;
1831 if (entry->bitmap) {
1834 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1839 } else if (size > *max_extent_size) {
1840 *max_extent_size = size;
1846 *bytes = entry->bytes - align_off;
1853 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1854 struct btrfs_free_space *info, u64 offset)
1856 info->offset = offset_to_bitmap(ctl, offset);
1858 INIT_LIST_HEAD(&info->list);
1859 link_free_space(ctl, info);
1860 ctl->total_bitmaps++;
1862 ctl->op->recalc_thresholds(ctl);
1865 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1866 struct btrfs_free_space *bitmap_info)
1868 unlink_free_space(ctl, bitmap_info);
1869 kfree(bitmap_info->bitmap);
1870 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1871 ctl->total_bitmaps--;
1872 ctl->op->recalc_thresholds(ctl);
1875 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1876 struct btrfs_free_space *bitmap_info,
1877 u64 *offset, u64 *bytes)
1880 u64 search_start, search_bytes;
1884 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1887 * We need to search for bits in this bitmap. We could only cover some
1888 * of the extent in this bitmap thanks to how we add space, so we need
1889 * to search for as much as it as we can and clear that amount, and then
1890 * go searching for the next bit.
1892 search_start = *offset;
1893 search_bytes = ctl->unit;
1894 search_bytes = min(search_bytes, end - search_start + 1);
1895 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1897 if (ret < 0 || search_start != *offset)
1900 /* We may have found more bits than what we need */
1901 search_bytes = min(search_bytes, *bytes);
1903 /* Cannot clear past the end of the bitmap */
1904 search_bytes = min(search_bytes, end - search_start + 1);
1906 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1907 *offset += search_bytes;
1908 *bytes -= search_bytes;
1911 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1912 if (!bitmap_info->bytes)
1913 free_bitmap(ctl, bitmap_info);
1916 * no entry after this bitmap, but we still have bytes to
1917 * remove, so something has gone wrong.
1922 bitmap_info = rb_entry(next, struct btrfs_free_space,
1926 * if the next entry isn't a bitmap we need to return to let the
1927 * extent stuff do its work.
1929 if (!bitmap_info->bitmap)
1933 * Ok the next item is a bitmap, but it may not actually hold
1934 * the information for the rest of this free space stuff, so
1935 * look for it, and if we don't find it return so we can try
1936 * everything over again.
1938 search_start = *offset;
1939 search_bytes = ctl->unit;
1940 ret = search_bitmap(ctl, bitmap_info, &search_start,
1941 &search_bytes, false);
1942 if (ret < 0 || search_start != *offset)
1946 } else if (!bitmap_info->bytes)
1947 free_bitmap(ctl, bitmap_info);
1952 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1953 struct btrfs_free_space *info, u64 offset,
1956 u64 bytes_to_set = 0;
1959 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1961 bytes_to_set = min(end - offset, bytes);
1963 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1966 * We set some bytes, we have no idea what the max extent size is
1969 info->max_extent_size = 0;
1971 return bytes_to_set;
1975 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1976 struct btrfs_free_space *info)
1978 struct btrfs_block_group_cache *block_group = ctl->private;
1979 struct btrfs_fs_info *fs_info = block_group->fs_info;
1980 bool forced = false;
1982 #ifdef CONFIG_BTRFS_DEBUG
1983 if (btrfs_should_fragment_free_space(block_group))
1988 * If we are below the extents threshold then we can add this as an
1989 * extent, and don't have to deal with the bitmap
1991 if (!forced && ctl->free_extents < ctl->extents_thresh) {
1993 * If this block group has some small extents we don't want to
1994 * use up all of our free slots in the cache with them, we want
1995 * to reserve them to larger extents, however if we have plenty
1996 * of cache left then go ahead an dadd them, no sense in adding
1997 * the overhead of a bitmap if we don't have to.
1999 if (info->bytes <= fs_info->sectorsize * 4) {
2000 if (ctl->free_extents * 2 <= ctl->extents_thresh)
2008 * The original block groups from mkfs can be really small, like 8
2009 * megabytes, so don't bother with a bitmap for those entries. However
2010 * some block groups can be smaller than what a bitmap would cover but
2011 * are still large enough that they could overflow the 32k memory limit,
2012 * so allow those block groups to still be allowed to have a bitmap
2015 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2021 static const struct btrfs_free_space_op free_space_op = {
2022 .recalc_thresholds = recalculate_thresholds,
2023 .use_bitmap = use_bitmap,
2026 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2027 struct btrfs_free_space *info)
2029 struct btrfs_free_space *bitmap_info;
2030 struct btrfs_block_group_cache *block_group = NULL;
2032 u64 bytes, offset, bytes_added;
2035 bytes = info->bytes;
2036 offset = info->offset;
2038 if (!ctl->op->use_bitmap(ctl, info))
2041 if (ctl->op == &free_space_op)
2042 block_group = ctl->private;
2045 * Since we link bitmaps right into the cluster we need to see if we
2046 * have a cluster here, and if so and it has our bitmap we need to add
2047 * the free space to that bitmap.
2049 if (block_group && !list_empty(&block_group->cluster_list)) {
2050 struct btrfs_free_cluster *cluster;
2051 struct rb_node *node;
2052 struct btrfs_free_space *entry;
2054 cluster = list_entry(block_group->cluster_list.next,
2055 struct btrfs_free_cluster,
2057 spin_lock(&cluster->lock);
2058 node = rb_first(&cluster->root);
2060 spin_unlock(&cluster->lock);
2061 goto no_cluster_bitmap;
2064 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2065 if (!entry->bitmap) {
2066 spin_unlock(&cluster->lock);
2067 goto no_cluster_bitmap;
2070 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2071 bytes_added = add_bytes_to_bitmap(ctl, entry,
2073 bytes -= bytes_added;
2074 offset += bytes_added;
2076 spin_unlock(&cluster->lock);
2084 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2091 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2092 bytes -= bytes_added;
2093 offset += bytes_added;
2103 if (info && info->bitmap) {
2104 add_new_bitmap(ctl, info, offset);
2109 spin_unlock(&ctl->tree_lock);
2111 /* no pre-allocated info, allocate a new one */
2113 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2116 spin_lock(&ctl->tree_lock);
2122 /* allocate the bitmap */
2123 info->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
2124 spin_lock(&ctl->tree_lock);
2125 if (!info->bitmap) {
2135 kfree(info->bitmap);
2136 kmem_cache_free(btrfs_free_space_cachep, info);
2142 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2143 struct btrfs_free_space *info, bool update_stat)
2145 struct btrfs_free_space *left_info;
2146 struct btrfs_free_space *right_info;
2147 bool merged = false;
2148 u64 offset = info->offset;
2149 u64 bytes = info->bytes;
2152 * first we want to see if there is free space adjacent to the range we
2153 * are adding, if there is remove that struct and add a new one to
2154 * cover the entire range
2156 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2157 if (right_info && rb_prev(&right_info->offset_index))
2158 left_info = rb_entry(rb_prev(&right_info->offset_index),
2159 struct btrfs_free_space, offset_index);
2161 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2163 if (right_info && !right_info->bitmap) {
2165 unlink_free_space(ctl, right_info);
2167 __unlink_free_space(ctl, right_info);
2168 info->bytes += right_info->bytes;
2169 kmem_cache_free(btrfs_free_space_cachep, right_info);
2173 if (left_info && !left_info->bitmap &&
2174 left_info->offset + left_info->bytes == offset) {
2176 unlink_free_space(ctl, left_info);
2178 __unlink_free_space(ctl, left_info);
2179 info->offset = left_info->offset;
2180 info->bytes += left_info->bytes;
2181 kmem_cache_free(btrfs_free_space_cachep, left_info);
2188 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2189 struct btrfs_free_space *info,
2192 struct btrfs_free_space *bitmap;
2195 const u64 end = info->offset + info->bytes;
2196 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2199 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2203 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2204 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2207 bytes = (j - i) * ctl->unit;
2208 info->bytes += bytes;
2211 bitmap_clear_bits(ctl, bitmap, end, bytes);
2213 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2216 free_bitmap(ctl, bitmap);
2221 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2222 struct btrfs_free_space *info,
2225 struct btrfs_free_space *bitmap;
2229 unsigned long prev_j;
2232 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2233 /* If we're on a boundary, try the previous logical bitmap. */
2234 if (bitmap_offset == info->offset) {
2235 if (info->offset == 0)
2237 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2240 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2244 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2246 prev_j = (unsigned long)-1;
2247 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2255 if (prev_j == (unsigned long)-1)
2256 bytes = (i + 1) * ctl->unit;
2258 bytes = (i - prev_j) * ctl->unit;
2260 info->offset -= bytes;
2261 info->bytes += bytes;
2264 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2266 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2269 free_bitmap(ctl, bitmap);
2275 * We prefer always to allocate from extent entries, both for clustered and
2276 * non-clustered allocation requests. So when attempting to add a new extent
2277 * entry, try to see if there's adjacent free space in bitmap entries, and if
2278 * there is, migrate that space from the bitmaps to the extent.
2279 * Like this we get better chances of satisfying space allocation requests
2280 * because we attempt to satisfy them based on a single cache entry, and never
2281 * on 2 or more entries - even if the entries represent a contiguous free space
2282 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2285 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2286 struct btrfs_free_space *info,
2290 * Only work with disconnected entries, as we can change their offset,
2291 * and must be extent entries.
2293 ASSERT(!info->bitmap);
2294 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2296 if (ctl->total_bitmaps > 0) {
2298 bool stole_front = false;
2300 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2301 if (ctl->total_bitmaps > 0)
2302 stole_front = steal_from_bitmap_to_front(ctl, info,
2305 if (stole_end || stole_front)
2306 try_merge_free_space(ctl, info, update_stat);
2310 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2311 struct btrfs_free_space_ctl *ctl,
2312 u64 offset, u64 bytes)
2314 struct btrfs_free_space *info;
2317 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2321 info->offset = offset;
2322 info->bytes = bytes;
2323 RB_CLEAR_NODE(&info->offset_index);
2325 spin_lock(&ctl->tree_lock);
2327 if (try_merge_free_space(ctl, info, true))
2331 * There was no extent directly to the left or right of this new
2332 * extent then we know we're going to have to allocate a new extent, so
2333 * before we do that see if we need to drop this into a bitmap
2335 ret = insert_into_bitmap(ctl, info);
2344 * Only steal free space from adjacent bitmaps if we're sure we're not
2345 * going to add the new free space to existing bitmap entries - because
2346 * that would mean unnecessary work that would be reverted. Therefore
2347 * attempt to steal space from bitmaps if we're adding an extent entry.
2349 steal_from_bitmap(ctl, info, true);
2351 ret = link_free_space(ctl, info);
2353 kmem_cache_free(btrfs_free_space_cachep, info);
2355 spin_unlock(&ctl->tree_lock);
2358 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2359 ASSERT(ret != -EEXIST);
2365 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2366 u64 offset, u64 bytes)
2368 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2369 struct btrfs_free_space *info;
2371 bool re_search = false;
2373 spin_lock(&ctl->tree_lock);
2380 info = tree_search_offset(ctl, offset, 0, 0);
2383 * oops didn't find an extent that matched the space we wanted
2384 * to remove, look for a bitmap instead
2386 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2390 * If we found a partial bit of our free space in a
2391 * bitmap but then couldn't find the other part this may
2392 * be a problem, so WARN about it.
2400 if (!info->bitmap) {
2401 unlink_free_space(ctl, info);
2402 if (offset == info->offset) {
2403 u64 to_free = min(bytes, info->bytes);
2405 info->bytes -= to_free;
2406 info->offset += to_free;
2408 ret = link_free_space(ctl, info);
2411 kmem_cache_free(btrfs_free_space_cachep, info);
2418 u64 old_end = info->bytes + info->offset;
2420 info->bytes = offset - info->offset;
2421 ret = link_free_space(ctl, info);
2426 /* Not enough bytes in this entry to satisfy us */
2427 if (old_end < offset + bytes) {
2428 bytes -= old_end - offset;
2431 } else if (old_end == offset + bytes) {
2435 spin_unlock(&ctl->tree_lock);
2437 ret = btrfs_add_free_space(block_group, offset + bytes,
2438 old_end - (offset + bytes));
2444 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2445 if (ret == -EAGAIN) {
2450 spin_unlock(&ctl->tree_lock);
2455 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2458 struct btrfs_fs_info *fs_info = block_group->fs_info;
2459 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2460 struct btrfs_free_space *info;
2464 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2465 info = rb_entry(n, struct btrfs_free_space, offset_index);
2466 if (info->bytes >= bytes && !block_group->ro)
2468 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2469 info->offset, info->bytes,
2470 (info->bitmap) ? "yes" : "no");
2472 btrfs_info(fs_info, "block group has cluster?: %s",
2473 list_empty(&block_group->cluster_list) ? "no" : "yes");
2475 "%d blocks of free space at or bigger than bytes is", count);
2478 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2480 struct btrfs_fs_info *fs_info = block_group->fs_info;
2481 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2483 spin_lock_init(&ctl->tree_lock);
2484 ctl->unit = fs_info->sectorsize;
2485 ctl->start = block_group->key.objectid;
2486 ctl->private = block_group;
2487 ctl->op = &free_space_op;
2488 INIT_LIST_HEAD(&ctl->trimming_ranges);
2489 mutex_init(&ctl->cache_writeout_mutex);
2492 * we only want to have 32k of ram per block group for keeping
2493 * track of free space, and if we pass 1/2 of that we want to
2494 * start converting things over to using bitmaps
2496 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2500 * for a given cluster, put all of its extents back into the free
2501 * space cache. If the block group passed doesn't match the block group
2502 * pointed to by the cluster, someone else raced in and freed the
2503 * cluster already. In that case, we just return without changing anything
2506 __btrfs_return_cluster_to_free_space(
2507 struct btrfs_block_group_cache *block_group,
2508 struct btrfs_free_cluster *cluster)
2510 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2511 struct btrfs_free_space *entry;
2512 struct rb_node *node;
2514 spin_lock(&cluster->lock);
2515 if (cluster->block_group != block_group)
2518 cluster->block_group = NULL;
2519 cluster->window_start = 0;
2520 list_del_init(&cluster->block_group_list);
2522 node = rb_first(&cluster->root);
2526 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2527 node = rb_next(&entry->offset_index);
2528 rb_erase(&entry->offset_index, &cluster->root);
2529 RB_CLEAR_NODE(&entry->offset_index);
2531 bitmap = (entry->bitmap != NULL);
2533 try_merge_free_space(ctl, entry, false);
2534 steal_from_bitmap(ctl, entry, false);
2536 tree_insert_offset(&ctl->free_space_offset,
2537 entry->offset, &entry->offset_index, bitmap);
2539 cluster->root = RB_ROOT;
2542 spin_unlock(&cluster->lock);
2543 btrfs_put_block_group(block_group);
2547 static void __btrfs_remove_free_space_cache_locked(
2548 struct btrfs_free_space_ctl *ctl)
2550 struct btrfs_free_space *info;
2551 struct rb_node *node;
2553 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2554 info = rb_entry(node, struct btrfs_free_space, offset_index);
2555 if (!info->bitmap) {
2556 unlink_free_space(ctl, info);
2557 kmem_cache_free(btrfs_free_space_cachep, info);
2559 free_bitmap(ctl, info);
2562 cond_resched_lock(&ctl->tree_lock);
2566 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2568 spin_lock(&ctl->tree_lock);
2569 __btrfs_remove_free_space_cache_locked(ctl);
2570 spin_unlock(&ctl->tree_lock);
2573 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2575 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2576 struct btrfs_free_cluster *cluster;
2577 struct list_head *head;
2579 spin_lock(&ctl->tree_lock);
2580 while ((head = block_group->cluster_list.next) !=
2581 &block_group->cluster_list) {
2582 cluster = list_entry(head, struct btrfs_free_cluster,
2585 WARN_ON(cluster->block_group != block_group);
2586 __btrfs_return_cluster_to_free_space(block_group, cluster);
2588 cond_resched_lock(&ctl->tree_lock);
2590 __btrfs_remove_free_space_cache_locked(ctl);
2591 spin_unlock(&ctl->tree_lock);
2595 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2596 u64 offset, u64 bytes, u64 empty_size,
2597 u64 *max_extent_size)
2599 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2600 struct btrfs_free_space *entry = NULL;
2601 u64 bytes_search = bytes + empty_size;
2604 u64 align_gap_len = 0;
2606 spin_lock(&ctl->tree_lock);
2607 entry = find_free_space(ctl, &offset, &bytes_search,
2608 block_group->full_stripe_len, max_extent_size);
2613 if (entry->bitmap) {
2614 bitmap_clear_bits(ctl, entry, offset, bytes);
2616 free_bitmap(ctl, entry);
2618 unlink_free_space(ctl, entry);
2619 align_gap_len = offset - entry->offset;
2620 align_gap = entry->offset;
2622 entry->offset = offset + bytes;
2623 WARN_ON(entry->bytes < bytes + align_gap_len);
2625 entry->bytes -= bytes + align_gap_len;
2627 kmem_cache_free(btrfs_free_space_cachep, entry);
2629 link_free_space(ctl, entry);
2632 spin_unlock(&ctl->tree_lock);
2635 __btrfs_add_free_space(block_group->fs_info, ctl,
2636 align_gap, align_gap_len);
2641 * given a cluster, put all of its extents back into the free space
2642 * cache. If a block group is passed, this function will only free
2643 * a cluster that belongs to the passed block group.
2645 * Otherwise, it'll get a reference on the block group pointed to by the
2646 * cluster and remove the cluster from it.
2648 int btrfs_return_cluster_to_free_space(
2649 struct btrfs_block_group_cache *block_group,
2650 struct btrfs_free_cluster *cluster)
2652 struct btrfs_free_space_ctl *ctl;
2655 /* first, get a safe pointer to the block group */
2656 spin_lock(&cluster->lock);
2658 block_group = cluster->block_group;
2660 spin_unlock(&cluster->lock);
2663 } else if (cluster->block_group != block_group) {
2664 /* someone else has already freed it don't redo their work */
2665 spin_unlock(&cluster->lock);
2668 atomic_inc(&block_group->count);
2669 spin_unlock(&cluster->lock);
2671 ctl = block_group->free_space_ctl;
2673 /* now return any extents the cluster had on it */
2674 spin_lock(&ctl->tree_lock);
2675 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2676 spin_unlock(&ctl->tree_lock);
2678 /* finally drop our ref */
2679 btrfs_put_block_group(block_group);
2683 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2684 struct btrfs_free_cluster *cluster,
2685 struct btrfs_free_space *entry,
2686 u64 bytes, u64 min_start,
2687 u64 *max_extent_size)
2689 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2691 u64 search_start = cluster->window_start;
2692 u64 search_bytes = bytes;
2695 search_start = min_start;
2696 search_bytes = bytes;
2698 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2700 if (search_bytes > *max_extent_size)
2701 *max_extent_size = search_bytes;
2706 __bitmap_clear_bits(ctl, entry, ret, bytes);
2712 * given a cluster, try to allocate 'bytes' from it, returns 0
2713 * if it couldn't find anything suitably large, or a logical disk offset
2714 * if things worked out
2716 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2717 struct btrfs_free_cluster *cluster, u64 bytes,
2718 u64 min_start, u64 *max_extent_size)
2720 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2721 struct btrfs_free_space *entry = NULL;
2722 struct rb_node *node;
2725 spin_lock(&cluster->lock);
2726 if (bytes > cluster->max_size)
2729 if (cluster->block_group != block_group)
2732 node = rb_first(&cluster->root);
2736 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2738 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2739 *max_extent_size = entry->bytes;
2741 if (entry->bytes < bytes ||
2742 (!entry->bitmap && entry->offset < min_start)) {
2743 node = rb_next(&entry->offset_index);
2746 entry = rb_entry(node, struct btrfs_free_space,
2751 if (entry->bitmap) {
2752 ret = btrfs_alloc_from_bitmap(block_group,
2753 cluster, entry, bytes,
2754 cluster->window_start,
2757 node = rb_next(&entry->offset_index);
2760 entry = rb_entry(node, struct btrfs_free_space,
2764 cluster->window_start += bytes;
2766 ret = entry->offset;
2768 entry->offset += bytes;
2769 entry->bytes -= bytes;
2772 if (entry->bytes == 0)
2773 rb_erase(&entry->offset_index, &cluster->root);
2777 spin_unlock(&cluster->lock);
2782 spin_lock(&ctl->tree_lock);
2784 ctl->free_space -= bytes;
2785 if (entry->bytes == 0) {
2786 ctl->free_extents--;
2787 if (entry->bitmap) {
2788 kfree(entry->bitmap);
2789 ctl->total_bitmaps--;
2790 ctl->op->recalc_thresholds(ctl);
2792 kmem_cache_free(btrfs_free_space_cachep, entry);
2795 spin_unlock(&ctl->tree_lock);
2800 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2801 struct btrfs_free_space *entry,
2802 struct btrfs_free_cluster *cluster,
2803 u64 offset, u64 bytes,
2804 u64 cont1_bytes, u64 min_bytes)
2806 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2807 unsigned long next_zero;
2809 unsigned long want_bits;
2810 unsigned long min_bits;
2811 unsigned long found_bits;
2812 unsigned long max_bits = 0;
2813 unsigned long start = 0;
2814 unsigned long total_found = 0;
2817 i = offset_to_bit(entry->offset, ctl->unit,
2818 max_t(u64, offset, entry->offset));
2819 want_bits = bytes_to_bits(bytes, ctl->unit);
2820 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2823 * Don't bother looking for a cluster in this bitmap if it's heavily
2826 if (entry->max_extent_size &&
2827 entry->max_extent_size < cont1_bytes)
2831 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2832 next_zero = find_next_zero_bit(entry->bitmap,
2833 BITS_PER_BITMAP, i);
2834 if (next_zero - i >= min_bits) {
2835 found_bits = next_zero - i;
2836 if (found_bits > max_bits)
2837 max_bits = found_bits;
2840 if (next_zero - i > max_bits)
2841 max_bits = next_zero - i;
2846 entry->max_extent_size = (u64)max_bits * ctl->unit;
2852 cluster->max_size = 0;
2855 total_found += found_bits;
2857 if (cluster->max_size < found_bits * ctl->unit)
2858 cluster->max_size = found_bits * ctl->unit;
2860 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2865 cluster->window_start = start * ctl->unit + entry->offset;
2866 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2867 ret = tree_insert_offset(&cluster->root, entry->offset,
2868 &entry->offset_index, 1);
2869 ASSERT(!ret); /* -EEXIST; Logic error */
2871 trace_btrfs_setup_cluster(block_group, cluster,
2872 total_found * ctl->unit, 1);
2877 * This searches the block group for just extents to fill the cluster with.
2878 * Try to find a cluster with at least bytes total bytes, at least one
2879 * extent of cont1_bytes, and other clusters of at least min_bytes.
2882 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2883 struct btrfs_free_cluster *cluster,
2884 struct list_head *bitmaps, u64 offset, u64 bytes,
2885 u64 cont1_bytes, u64 min_bytes)
2887 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2888 struct btrfs_free_space *first = NULL;
2889 struct btrfs_free_space *entry = NULL;
2890 struct btrfs_free_space *last;
2891 struct rb_node *node;
2896 entry = tree_search_offset(ctl, offset, 0, 1);
2901 * We don't want bitmaps, so just move along until we find a normal
2904 while (entry->bitmap || entry->bytes < min_bytes) {
2905 if (entry->bitmap && list_empty(&entry->list))
2906 list_add_tail(&entry->list, bitmaps);
2907 node = rb_next(&entry->offset_index);
2910 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2913 window_free = entry->bytes;
2914 max_extent = entry->bytes;
2918 for (node = rb_next(&entry->offset_index); node;
2919 node = rb_next(&entry->offset_index)) {
2920 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2922 if (entry->bitmap) {
2923 if (list_empty(&entry->list))
2924 list_add_tail(&entry->list, bitmaps);
2928 if (entry->bytes < min_bytes)
2932 window_free += entry->bytes;
2933 if (entry->bytes > max_extent)
2934 max_extent = entry->bytes;
2937 if (window_free < bytes || max_extent < cont1_bytes)
2940 cluster->window_start = first->offset;
2942 node = &first->offset_index;
2945 * now we've found our entries, pull them out of the free space
2946 * cache and put them into the cluster rbtree
2951 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2952 node = rb_next(&entry->offset_index);
2953 if (entry->bitmap || entry->bytes < min_bytes)
2956 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2957 ret = tree_insert_offset(&cluster->root, entry->offset,
2958 &entry->offset_index, 0);
2959 total_size += entry->bytes;
2960 ASSERT(!ret); /* -EEXIST; Logic error */
2961 } while (node && entry != last);
2963 cluster->max_size = max_extent;
2964 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2969 * This specifically looks for bitmaps that may work in the cluster, we assume
2970 * that we have already failed to find extents that will work.
2973 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2974 struct btrfs_free_cluster *cluster,
2975 struct list_head *bitmaps, u64 offset, u64 bytes,
2976 u64 cont1_bytes, u64 min_bytes)
2978 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2979 struct btrfs_free_space *entry = NULL;
2981 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2983 if (ctl->total_bitmaps == 0)
2987 * The bitmap that covers offset won't be in the list unless offset
2988 * is just its start offset.
2990 if (!list_empty(bitmaps))
2991 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2993 if (!entry || entry->offset != bitmap_offset) {
2994 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2995 if (entry && list_empty(&entry->list))
2996 list_add(&entry->list, bitmaps);
2999 list_for_each_entry(entry, bitmaps, list) {
3000 if (entry->bytes < bytes)
3002 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3003 bytes, cont1_bytes, min_bytes);
3009 * The bitmaps list has all the bitmaps that record free space
3010 * starting after offset, so no more search is required.
3016 * here we try to find a cluster of blocks in a block group. The goal
3017 * is to find at least bytes+empty_size.
3018 * We might not find them all in one contiguous area.
3020 * returns zero and sets up cluster if things worked out, otherwise
3021 * it returns -enospc
3023 int btrfs_find_space_cluster(struct btrfs_fs_info *fs_info,
3024 struct btrfs_block_group_cache *block_group,
3025 struct btrfs_free_cluster *cluster,
3026 u64 offset, u64 bytes, u64 empty_size)
3028 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3029 struct btrfs_free_space *entry, *tmp;
3036 * Choose the minimum extent size we'll require for this
3037 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3038 * For metadata, allow allocates with smaller extents. For
3039 * data, keep it dense.
3041 if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3042 cont1_bytes = min_bytes = bytes + empty_size;
3043 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3044 cont1_bytes = bytes;
3045 min_bytes = fs_info->sectorsize;
3047 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3048 min_bytes = fs_info->sectorsize;
3051 spin_lock(&ctl->tree_lock);
3054 * If we know we don't have enough space to make a cluster don't even
3055 * bother doing all the work to try and find one.
3057 if (ctl->free_space < bytes) {
3058 spin_unlock(&ctl->tree_lock);
3062 spin_lock(&cluster->lock);
3064 /* someone already found a cluster, hooray */
3065 if (cluster->block_group) {
3070 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3073 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3075 cont1_bytes, min_bytes);
3077 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3078 offset, bytes + empty_size,
3079 cont1_bytes, min_bytes);
3081 /* Clear our temporary list */
3082 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3083 list_del_init(&entry->list);
3086 atomic_inc(&block_group->count);
3087 list_add_tail(&cluster->block_group_list,
3088 &block_group->cluster_list);
3089 cluster->block_group = block_group;
3091 trace_btrfs_failed_cluster_setup(block_group);
3094 spin_unlock(&cluster->lock);
3095 spin_unlock(&ctl->tree_lock);
3101 * simple code to zero out a cluster
3103 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3105 spin_lock_init(&cluster->lock);
3106 spin_lock_init(&cluster->refill_lock);
3107 cluster->root = RB_ROOT;
3108 cluster->max_size = 0;
3109 cluster->fragmented = false;
3110 INIT_LIST_HEAD(&cluster->block_group_list);
3111 cluster->block_group = NULL;
3114 static int do_trimming(struct btrfs_block_group_cache *block_group,
3115 u64 *total_trimmed, u64 start, u64 bytes,
3116 u64 reserved_start, u64 reserved_bytes,
3117 struct btrfs_trim_range *trim_entry)
3119 struct btrfs_space_info *space_info = block_group->space_info;
3120 struct btrfs_fs_info *fs_info = block_group->fs_info;
3121 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3126 spin_lock(&space_info->lock);
3127 spin_lock(&block_group->lock);
3128 if (!block_group->ro) {
3129 block_group->reserved += reserved_bytes;
3130 space_info->bytes_reserved += reserved_bytes;
3133 spin_unlock(&block_group->lock);
3134 spin_unlock(&space_info->lock);
3136 ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3138 *total_trimmed += trimmed;
3140 mutex_lock(&ctl->cache_writeout_mutex);
3141 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3142 list_del(&trim_entry->list);
3143 mutex_unlock(&ctl->cache_writeout_mutex);
3146 spin_lock(&space_info->lock);
3147 spin_lock(&block_group->lock);
3148 if (block_group->ro)
3149 space_info->bytes_readonly += reserved_bytes;
3150 block_group->reserved -= reserved_bytes;
3151 space_info->bytes_reserved -= reserved_bytes;
3152 spin_unlock(&space_info->lock);
3153 spin_unlock(&block_group->lock);
3159 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3160 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3162 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3163 struct btrfs_free_space *entry;
3164 struct rb_node *node;
3170 while (start < end) {
3171 struct btrfs_trim_range trim_entry;
3173 mutex_lock(&ctl->cache_writeout_mutex);
3174 spin_lock(&ctl->tree_lock);
3176 if (ctl->free_space < minlen) {
3177 spin_unlock(&ctl->tree_lock);
3178 mutex_unlock(&ctl->cache_writeout_mutex);
3182 entry = tree_search_offset(ctl, start, 0, 1);
3184 spin_unlock(&ctl->tree_lock);
3185 mutex_unlock(&ctl->cache_writeout_mutex);
3190 while (entry->bitmap) {
3191 node = rb_next(&entry->offset_index);
3193 spin_unlock(&ctl->tree_lock);
3194 mutex_unlock(&ctl->cache_writeout_mutex);
3197 entry = rb_entry(node, struct btrfs_free_space,
3201 if (entry->offset >= end) {
3202 spin_unlock(&ctl->tree_lock);
3203 mutex_unlock(&ctl->cache_writeout_mutex);
3207 extent_start = entry->offset;
3208 extent_bytes = entry->bytes;
3209 start = max(start, extent_start);
3210 bytes = min(extent_start + extent_bytes, end) - start;
3211 if (bytes < minlen) {
3212 spin_unlock(&ctl->tree_lock);
3213 mutex_unlock(&ctl->cache_writeout_mutex);
3217 unlink_free_space(ctl, entry);
3218 kmem_cache_free(btrfs_free_space_cachep, entry);
3220 spin_unlock(&ctl->tree_lock);
3221 trim_entry.start = extent_start;
3222 trim_entry.bytes = extent_bytes;
3223 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3224 mutex_unlock(&ctl->cache_writeout_mutex);
3226 ret = do_trimming(block_group, total_trimmed, start, bytes,
3227 extent_start, extent_bytes, &trim_entry);
3233 if (fatal_signal_pending(current)) {
3244 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3245 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3247 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3248 struct btrfs_free_space *entry;
3252 u64 offset = offset_to_bitmap(ctl, start);
3254 while (offset < end) {
3255 bool next_bitmap = false;
3256 struct btrfs_trim_range trim_entry;
3258 mutex_lock(&ctl->cache_writeout_mutex);
3259 spin_lock(&ctl->tree_lock);
3261 if (ctl->free_space < minlen) {
3262 spin_unlock(&ctl->tree_lock);
3263 mutex_unlock(&ctl->cache_writeout_mutex);
3267 entry = tree_search_offset(ctl, offset, 1, 0);
3269 spin_unlock(&ctl->tree_lock);
3270 mutex_unlock(&ctl->cache_writeout_mutex);
3276 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3277 if (ret2 || start >= end) {
3278 spin_unlock(&ctl->tree_lock);
3279 mutex_unlock(&ctl->cache_writeout_mutex);
3284 bytes = min(bytes, end - start);
3285 if (bytes < minlen) {
3286 spin_unlock(&ctl->tree_lock);
3287 mutex_unlock(&ctl->cache_writeout_mutex);
3291 bitmap_clear_bits(ctl, entry, start, bytes);
3292 if (entry->bytes == 0)
3293 free_bitmap(ctl, entry);
3295 spin_unlock(&ctl->tree_lock);
3296 trim_entry.start = start;
3297 trim_entry.bytes = bytes;
3298 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3299 mutex_unlock(&ctl->cache_writeout_mutex);
3301 ret = do_trimming(block_group, total_trimmed, start, bytes,
3302 start, bytes, &trim_entry);
3307 offset += BITS_PER_BITMAP * ctl->unit;
3310 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3311 offset += BITS_PER_BITMAP * ctl->unit;
3314 if (fatal_signal_pending(current)) {
3325 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3327 atomic_inc(&cache->trimming);
3330 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3332 struct btrfs_fs_info *fs_info = block_group->fs_info;
3333 struct extent_map_tree *em_tree;
3334 struct extent_map *em;
3337 spin_lock(&block_group->lock);
3338 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3339 block_group->removed);
3340 spin_unlock(&block_group->lock);
3343 mutex_lock(&fs_info->chunk_mutex);
3344 em_tree = &fs_info->mapping_tree.map_tree;
3345 write_lock(&em_tree->lock);
3346 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3348 BUG_ON(!em); /* logic error, can't happen */
3350 * remove_extent_mapping() will delete us from the pinned_chunks
3351 * list, which is protected by the chunk mutex.
3353 remove_extent_mapping(em_tree, em);
3354 write_unlock(&em_tree->lock);
3355 mutex_unlock(&fs_info->chunk_mutex);
3357 /* once for us and once for the tree */
3358 free_extent_map(em);
3359 free_extent_map(em);
3362 * We've left one free space entry and other tasks trimming
3363 * this block group have left 1 entry each one. Free them.
3365 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3369 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3370 u64 *trimmed, u64 start, u64 end, u64 minlen)
3376 spin_lock(&block_group->lock);
3377 if (block_group->removed) {
3378 spin_unlock(&block_group->lock);
3381 btrfs_get_block_group_trimming(block_group);
3382 spin_unlock(&block_group->lock);
3384 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3388 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3390 btrfs_put_block_group_trimming(block_group);
3395 * Find the left-most item in the cache tree, and then return the
3396 * smallest inode number in the item.
3398 * Note: the returned inode number may not be the smallest one in
3399 * the tree, if the left-most item is a bitmap.
3401 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3403 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3404 struct btrfs_free_space *entry = NULL;
3407 spin_lock(&ctl->tree_lock);
3409 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3412 entry = rb_entry(rb_first(&ctl->free_space_offset),
3413 struct btrfs_free_space, offset_index);
3415 if (!entry->bitmap) {
3416 ino = entry->offset;
3418 unlink_free_space(ctl, entry);
3422 kmem_cache_free(btrfs_free_space_cachep, entry);
3424 link_free_space(ctl, entry);
3430 ret = search_bitmap(ctl, entry, &offset, &count, true);
3431 /* Logic error; Should be empty if it can't find anything */
3435 bitmap_clear_bits(ctl, entry, offset, 1);
3436 if (entry->bytes == 0)
3437 free_bitmap(ctl, entry);
3440 spin_unlock(&ctl->tree_lock);
3445 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3446 struct btrfs_path *path)
3448 struct inode *inode = NULL;
3450 spin_lock(&root->ino_cache_lock);
3451 if (root->ino_cache_inode)
3452 inode = igrab(root->ino_cache_inode);
3453 spin_unlock(&root->ino_cache_lock);
3457 inode = __lookup_free_space_inode(root, path, 0);
3461 spin_lock(&root->ino_cache_lock);
3462 if (!btrfs_fs_closing(root->fs_info))
3463 root->ino_cache_inode = igrab(inode);
3464 spin_unlock(&root->ino_cache_lock);
3469 int create_free_ino_inode(struct btrfs_root *root,
3470 struct btrfs_trans_handle *trans,
3471 struct btrfs_path *path)
3473 return __create_free_space_inode(root, trans, path,
3474 BTRFS_FREE_INO_OBJECTID, 0);
3477 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3479 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3480 struct btrfs_path *path;
3481 struct inode *inode;
3483 u64 root_gen = btrfs_root_generation(&root->root_item);
3485 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
3489 * If we're unmounting then just return, since this does a search on the
3490 * normal root and not the commit root and we could deadlock.
3492 if (btrfs_fs_closing(fs_info))
3495 path = btrfs_alloc_path();
3499 inode = lookup_free_ino_inode(root, path);
3503 if (root_gen != BTRFS_I(inode)->generation)
3506 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3510 "failed to load free ino cache for root %llu",
3511 root->root_key.objectid);
3515 btrfs_free_path(path);
3519 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3520 struct btrfs_trans_handle *trans,
3521 struct btrfs_path *path,
3522 struct inode *inode)
3524 struct btrfs_fs_info *fs_info = root->fs_info;
3525 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3527 struct btrfs_io_ctl io_ctl;
3528 bool release_metadata = true;
3530 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
3533 memset(&io_ctl, 0, sizeof(io_ctl));
3534 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl, trans);
3537 * At this point writepages() didn't error out, so our metadata
3538 * reservation is released when the writeback finishes, at
3539 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3540 * with or without an error.
3542 release_metadata = false;
3543 ret = btrfs_wait_cache_io_root(root, trans, &io_ctl, path);
3547 if (release_metadata)
3548 btrfs_delalloc_release_metadata(inode, inode->i_size);
3551 "failed to write free ino cache for root %llu",
3552 root->root_key.objectid);
3559 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3561 * Use this if you need to make a bitmap or extent entry specifically, it
3562 * doesn't do any of the merging that add_free_space does, this acts a lot like
3563 * how the free space cache loading stuff works, so you can get really weird
3566 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3567 u64 offset, u64 bytes, bool bitmap)
3569 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3570 struct btrfs_free_space *info = NULL, *bitmap_info;
3577 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3583 spin_lock(&ctl->tree_lock);
3584 info->offset = offset;
3585 info->bytes = bytes;
3586 info->max_extent_size = 0;
3587 ret = link_free_space(ctl, info);
3588 spin_unlock(&ctl->tree_lock);
3590 kmem_cache_free(btrfs_free_space_cachep, info);
3595 map = kzalloc(PAGE_SIZE, GFP_NOFS);
3597 kmem_cache_free(btrfs_free_space_cachep, info);
3602 spin_lock(&ctl->tree_lock);
3603 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3608 add_new_bitmap(ctl, info, offset);
3613 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3615 bytes -= bytes_added;
3616 offset += bytes_added;
3617 spin_unlock(&ctl->tree_lock);
3623 kmem_cache_free(btrfs_free_space_cachep, info);
3630 * Checks to see if the given range is in the free space cache. This is really
3631 * just used to check the absence of space, so if there is free space in the
3632 * range at all we will return 1.
3634 int test_check_exists(struct btrfs_block_group_cache *cache,
3635 u64 offset, u64 bytes)
3637 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3638 struct btrfs_free_space *info;
3641 spin_lock(&ctl->tree_lock);
3642 info = tree_search_offset(ctl, offset, 0, 0);
3644 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3652 u64 bit_off, bit_bytes;
3654 struct btrfs_free_space *tmp;
3657 bit_bytes = ctl->unit;
3658 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3660 if (bit_off == offset) {
3663 } else if (bit_off > offset &&
3664 offset + bytes > bit_off) {
3670 n = rb_prev(&info->offset_index);
3672 tmp = rb_entry(n, struct btrfs_free_space,
3674 if (tmp->offset + tmp->bytes < offset)
3676 if (offset + bytes < tmp->offset) {
3677 n = rb_prev(&tmp->offset_index);
3684 n = rb_next(&info->offset_index);
3686 tmp = rb_entry(n, struct btrfs_free_space,
3688 if (offset + bytes < tmp->offset)
3690 if (tmp->offset + tmp->bytes < offset) {
3691 n = rb_next(&tmp->offset_index);
3702 if (info->offset == offset) {
3707 if (offset > info->offset && offset < info->offset + info->bytes)
3710 spin_unlock(&ctl->tree_lock);
3713 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */