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);
46 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
47 struct btrfs_path *path,
51 struct btrfs_key location;
52 struct btrfs_disk_key disk_key;
53 struct btrfs_free_space_header *header;
54 struct extent_buffer *leaf;
55 struct inode *inode = NULL;
58 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
62 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66 btrfs_release_path(path);
67 return ERR_PTR(-ENOENT);
70 leaf = path->nodes[0];
71 header = btrfs_item_ptr(leaf, path->slots[0],
72 struct btrfs_free_space_header);
73 btrfs_free_space_key(leaf, header, &disk_key);
74 btrfs_disk_key_to_cpu(&location, &disk_key);
75 btrfs_release_path(path);
77 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
80 if (is_bad_inode(inode)) {
82 return ERR_PTR(-ENOENT);
85 mapping_set_gfp_mask(inode->i_mapping,
86 mapping_gfp_constraint(inode->i_mapping,
87 ~(__GFP_FS | __GFP_HIGHMEM)));
92 struct inode *lookup_free_space_inode(struct btrfs_root *root,
93 struct btrfs_block_group_cache
94 *block_group, struct btrfs_path *path)
96 struct inode *inode = NULL;
97 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
99 spin_lock(&block_group->lock);
100 if (block_group->inode)
101 inode = igrab(block_group->inode);
102 spin_unlock(&block_group->lock);
106 inode = __lookup_free_space_inode(root, path,
107 block_group->key.objectid);
111 spin_lock(&block_group->lock);
112 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
113 btrfs_info(root->fs_info,
114 "Old style space inode found, converting.");
115 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
116 BTRFS_INODE_NODATACOW;
117 block_group->disk_cache_state = BTRFS_DC_CLEAR;
120 if (!block_group->iref) {
121 block_group->inode = igrab(inode);
122 block_group->iref = 1;
124 spin_unlock(&block_group->lock);
129 static int __create_free_space_inode(struct btrfs_root *root,
130 struct btrfs_trans_handle *trans,
131 struct btrfs_path *path,
134 struct btrfs_key key;
135 struct btrfs_disk_key disk_key;
136 struct btrfs_free_space_header *header;
137 struct btrfs_inode_item *inode_item;
138 struct extent_buffer *leaf;
139 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
142 ret = btrfs_insert_empty_inode(trans, root, path, ino);
146 /* We inline crc's for the free disk space cache */
147 if (ino != BTRFS_FREE_INO_OBJECTID)
148 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
150 leaf = path->nodes[0];
151 inode_item = btrfs_item_ptr(leaf, path->slots[0],
152 struct btrfs_inode_item);
153 btrfs_item_key(leaf, &disk_key, path->slots[0]);
154 memzero_extent_buffer(leaf, (unsigned long)inode_item,
155 sizeof(*inode_item));
156 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
157 btrfs_set_inode_size(leaf, inode_item, 0);
158 btrfs_set_inode_nbytes(leaf, inode_item, 0);
159 btrfs_set_inode_uid(leaf, inode_item, 0);
160 btrfs_set_inode_gid(leaf, inode_item, 0);
161 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
162 btrfs_set_inode_flags(leaf, inode_item, flags);
163 btrfs_set_inode_nlink(leaf, inode_item, 1);
164 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
165 btrfs_set_inode_block_group(leaf, inode_item, offset);
166 btrfs_mark_buffer_dirty(leaf);
167 btrfs_release_path(path);
169 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
172 ret = btrfs_insert_empty_item(trans, root, path, &key,
173 sizeof(struct btrfs_free_space_header));
175 btrfs_release_path(path);
179 leaf = path->nodes[0];
180 header = btrfs_item_ptr(leaf, path->slots[0],
181 struct btrfs_free_space_header);
182 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
183 btrfs_set_free_space_key(leaf, header, &disk_key);
184 btrfs_mark_buffer_dirty(leaf);
185 btrfs_release_path(path);
190 int create_free_space_inode(struct btrfs_root *root,
191 struct btrfs_trans_handle *trans,
192 struct btrfs_block_group_cache *block_group,
193 struct btrfs_path *path)
198 ret = btrfs_find_free_objectid(root, &ino);
202 return __create_free_space_inode(root, trans, path, ino,
203 block_group->key.objectid);
206 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
207 struct btrfs_block_rsv *rsv)
212 /* 1 for slack space, 1 for updating the inode */
213 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
214 btrfs_calc_trans_metadata_size(root, 1);
216 spin_lock(&rsv->lock);
217 if (rsv->reserved < needed_bytes)
221 spin_unlock(&rsv->lock);
225 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
226 struct btrfs_trans_handle *trans,
227 struct btrfs_block_group_cache *block_group,
231 struct btrfs_path *path = btrfs_alloc_path();
241 mutex_lock(&trans->transaction->cache_write_mutex);
242 if (!list_empty(&block_group->io_list)) {
243 list_del_init(&block_group->io_list);
245 btrfs_wait_cache_io(root, trans, block_group,
246 &block_group->io_ctl, path,
247 block_group->key.objectid);
248 btrfs_put_block_group(block_group);
252 * now that we've truncated the cache away, its no longer
255 spin_lock(&block_group->lock);
256 block_group->disk_cache_state = BTRFS_DC_CLEAR;
257 spin_unlock(&block_group->lock);
259 btrfs_free_path(path);
261 btrfs_i_size_write(inode, 0);
262 truncate_pagecache(inode, 0);
265 * We don't need an orphan item because truncating the free space cache
266 * will never be split across transactions.
267 * We don't need to check for -EAGAIN because we're a free space
270 ret = btrfs_truncate_inode_items(trans, root, inode,
271 0, BTRFS_EXTENT_DATA_KEY);
275 ret = btrfs_update_inode(trans, root, inode);
279 mutex_unlock(&trans->transaction->cache_write_mutex);
281 btrfs_abort_transaction(trans, ret);
286 static int readahead_cache(struct inode *inode)
288 struct file_ra_state *ra;
289 unsigned long last_index;
291 ra = kzalloc(sizeof(*ra), GFP_NOFS);
295 file_ra_state_init(ra, inode->i_mapping);
296 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
298 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,
306 struct btrfs_root *root, int write)
311 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
313 if (btrfs_ino(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;
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->root->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->root->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_free_space_header *header;
671 struct extent_buffer *leaf;
672 struct btrfs_io_ctl io_ctl;
673 struct btrfs_key key;
674 struct btrfs_free_space *e, *n;
682 /* Nothing in the space cache, goodbye */
683 if (!i_size_read(inode))
686 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
690 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
694 btrfs_release_path(path);
700 leaf = path->nodes[0];
701 header = btrfs_item_ptr(leaf, path->slots[0],
702 struct btrfs_free_space_header);
703 num_entries = btrfs_free_space_entries(leaf, header);
704 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
705 generation = btrfs_free_space_generation(leaf, header);
706 btrfs_release_path(path);
708 if (!BTRFS_I(inode)->generation) {
709 btrfs_info(root->fs_info,
710 "The free space cache file (%llu) is invalid. skip it\n",
715 if (BTRFS_I(inode)->generation != generation) {
716 btrfs_err(root->fs_info,
717 "free space inode generation (%llu) did not match free space cache generation (%llu)",
718 BTRFS_I(inode)->generation, generation);
725 ret = io_ctl_init(&io_ctl, inode, root, 0);
729 ret = readahead_cache(inode);
733 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
737 ret = io_ctl_check_crc(&io_ctl, 0);
741 ret = io_ctl_check_generation(&io_ctl, generation);
745 while (num_entries) {
746 e = kmem_cache_zalloc(btrfs_free_space_cachep,
751 ret = io_ctl_read_entry(&io_ctl, e, &type);
753 kmem_cache_free(btrfs_free_space_cachep, e);
758 kmem_cache_free(btrfs_free_space_cachep, e);
762 if (type == BTRFS_FREE_SPACE_EXTENT) {
763 spin_lock(&ctl->tree_lock);
764 ret = link_free_space(ctl, e);
765 spin_unlock(&ctl->tree_lock);
767 btrfs_err(root->fs_info,
768 "Duplicate entries in free space cache, dumping");
769 kmem_cache_free(btrfs_free_space_cachep, e);
775 e->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
778 btrfs_free_space_cachep, e);
781 spin_lock(&ctl->tree_lock);
782 ret = link_free_space(ctl, e);
783 ctl->total_bitmaps++;
784 ctl->op->recalc_thresholds(ctl);
785 spin_unlock(&ctl->tree_lock);
787 btrfs_err(root->fs_info,
788 "Duplicate entries in free space cache, dumping");
789 kmem_cache_free(btrfs_free_space_cachep, e);
792 list_add_tail(&e->list, &bitmaps);
798 io_ctl_unmap_page(&io_ctl);
801 * We add the bitmaps at the end of the entries in order that
802 * the bitmap entries are added to the cache.
804 list_for_each_entry_safe(e, n, &bitmaps, list) {
805 list_del_init(&e->list);
806 ret = io_ctl_read_bitmap(&io_ctl, e);
811 io_ctl_drop_pages(&io_ctl);
812 merge_space_tree(ctl);
815 io_ctl_free(&io_ctl);
818 io_ctl_drop_pages(&io_ctl);
819 __btrfs_remove_free_space_cache(ctl);
823 int load_free_space_cache(struct btrfs_fs_info *fs_info,
824 struct btrfs_block_group_cache *block_group)
826 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
827 struct btrfs_root *root = fs_info->tree_root;
829 struct btrfs_path *path;
832 u64 used = btrfs_block_group_used(&block_group->item);
835 * If this block group has been marked to be cleared for one reason or
836 * another then we can't trust the on disk cache, so just return.
838 spin_lock(&block_group->lock);
839 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
840 spin_unlock(&block_group->lock);
843 spin_unlock(&block_group->lock);
845 path = btrfs_alloc_path();
848 path->search_commit_root = 1;
849 path->skip_locking = 1;
851 inode = lookup_free_space_inode(root, block_group, path);
853 btrfs_free_path(path);
857 /* We may have converted the inode and made the cache invalid. */
858 spin_lock(&block_group->lock);
859 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
860 spin_unlock(&block_group->lock);
861 btrfs_free_path(path);
864 spin_unlock(&block_group->lock);
866 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
867 path, block_group->key.objectid);
868 btrfs_free_path(path);
872 spin_lock(&ctl->tree_lock);
873 matched = (ctl->free_space == (block_group->key.offset - used -
874 block_group->bytes_super));
875 spin_unlock(&ctl->tree_lock);
878 __btrfs_remove_free_space_cache(ctl);
880 "block group %llu has wrong amount of free space",
881 block_group->key.objectid);
886 /* This cache is bogus, make sure it gets cleared */
887 spin_lock(&block_group->lock);
888 block_group->disk_cache_state = BTRFS_DC_CLEAR;
889 spin_unlock(&block_group->lock);
893 "failed to load free space cache for block group %llu, rebuilding it now",
894 block_group->key.objectid);
901 static noinline_for_stack
902 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
903 struct btrfs_free_space_ctl *ctl,
904 struct btrfs_block_group_cache *block_group,
905 int *entries, int *bitmaps,
906 struct list_head *bitmap_list)
909 struct btrfs_free_cluster *cluster = NULL;
910 struct btrfs_free_cluster *cluster_locked = NULL;
911 struct rb_node *node = rb_first(&ctl->free_space_offset);
912 struct btrfs_trim_range *trim_entry;
914 /* Get the cluster for this block_group if it exists */
915 if (block_group && !list_empty(&block_group->cluster_list)) {
916 cluster = list_entry(block_group->cluster_list.next,
917 struct btrfs_free_cluster,
921 if (!node && cluster) {
922 cluster_locked = cluster;
923 spin_lock(&cluster_locked->lock);
924 node = rb_first(&cluster->root);
928 /* Write out the extent entries */
930 struct btrfs_free_space *e;
932 e = rb_entry(node, struct btrfs_free_space, offset_index);
935 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
941 list_add_tail(&e->list, bitmap_list);
944 node = rb_next(node);
945 if (!node && cluster) {
946 node = rb_first(&cluster->root);
947 cluster_locked = cluster;
948 spin_lock(&cluster_locked->lock);
952 if (cluster_locked) {
953 spin_unlock(&cluster_locked->lock);
954 cluster_locked = NULL;
958 * Make sure we don't miss any range that was removed from our rbtree
959 * because trimming is running. Otherwise after a umount+mount (or crash
960 * after committing the transaction) we would leak free space and get
961 * an inconsistent free space cache report from fsck.
963 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
964 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
965 trim_entry->bytes, NULL);
974 spin_unlock(&cluster_locked->lock);
978 static noinline_for_stack int
979 update_cache_item(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
982 struct btrfs_path *path, u64 offset,
983 int entries, int bitmaps)
985 struct btrfs_key key;
986 struct btrfs_free_space_header *header;
987 struct extent_buffer *leaf;
990 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
994 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
996 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
997 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1001 leaf = path->nodes[0];
1003 struct btrfs_key found_key;
1004 ASSERT(path->slots[0]);
1006 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1007 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1008 found_key.offset != offset) {
1009 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1011 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1013 btrfs_release_path(path);
1018 BTRFS_I(inode)->generation = trans->transid;
1019 header = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_free_space_header);
1021 btrfs_set_free_space_entries(leaf, header, entries);
1022 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1023 btrfs_set_free_space_generation(leaf, header, trans->transid);
1024 btrfs_mark_buffer_dirty(leaf);
1025 btrfs_release_path(path);
1033 static noinline_for_stack int
1034 write_pinned_extent_entries(struct btrfs_root *root,
1035 struct btrfs_block_group_cache *block_group,
1036 struct btrfs_io_ctl *io_ctl,
1039 u64 start, extent_start, extent_end, len;
1040 struct extent_io_tree *unpin = NULL;
1047 * We want to add any pinned extents to our free space cache
1048 * so we don't leak the space
1050 * We shouldn't have switched the pinned extents yet so this is the
1053 unpin = root->fs_info->pinned_extents;
1055 start = block_group->key.objectid;
1057 while (start < block_group->key.objectid + block_group->key.offset) {
1058 ret = find_first_extent_bit(unpin, start,
1059 &extent_start, &extent_end,
1060 EXTENT_DIRTY, NULL);
1064 /* This pinned extent is out of our range */
1065 if (extent_start >= block_group->key.objectid +
1066 block_group->key.offset)
1069 extent_start = max(extent_start, start);
1070 extent_end = min(block_group->key.objectid +
1071 block_group->key.offset, extent_end + 1);
1072 len = extent_end - extent_start;
1075 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1085 static noinline_for_stack int
1086 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1088 struct btrfs_free_space *entry, *next;
1091 /* Write out the bitmaps */
1092 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1093 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1096 list_del_init(&entry->list);
1102 static int flush_dirty_cache(struct inode *inode)
1106 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1108 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1109 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1115 static void noinline_for_stack
1116 cleanup_bitmap_list(struct list_head *bitmap_list)
1118 struct btrfs_free_space *entry, *next;
1120 list_for_each_entry_safe(entry, next, bitmap_list, list)
1121 list_del_init(&entry->list);
1124 static void noinline_for_stack
1125 cleanup_write_cache_enospc(struct inode *inode,
1126 struct btrfs_io_ctl *io_ctl,
1127 struct extent_state **cached_state,
1128 struct list_head *bitmap_list)
1130 io_ctl_drop_pages(io_ctl);
1131 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1132 i_size_read(inode) - 1, cached_state,
1136 int btrfs_wait_cache_io(struct btrfs_root *root,
1137 struct btrfs_trans_handle *trans,
1138 struct btrfs_block_group_cache *block_group,
1139 struct btrfs_io_ctl *io_ctl,
1140 struct btrfs_path *path, u64 offset)
1143 struct inode *inode = io_ctl->inode;
1149 root = root->fs_info->tree_root;
1151 /* Flush the dirty pages in the cache file. */
1152 ret = flush_dirty_cache(inode);
1156 /* Update the cache item to tell everyone this cache file is valid. */
1157 ret = update_cache_item(trans, root, inode, path, offset,
1158 io_ctl->entries, io_ctl->bitmaps);
1160 io_ctl_free(io_ctl);
1162 invalidate_inode_pages2(inode->i_mapping);
1163 BTRFS_I(inode)->generation = 0;
1166 btrfs_err(root->fs_info,
1167 "failed to write free space cache for block group %llu",
1168 block_group->key.objectid);
1172 btrfs_update_inode(trans, root, inode);
1175 /* the dirty list is protected by the dirty_bgs_lock */
1176 spin_lock(&trans->transaction->dirty_bgs_lock);
1178 /* the disk_cache_state is protected by the block group lock */
1179 spin_lock(&block_group->lock);
1182 * only mark this as written if we didn't get put back on
1183 * the dirty list while waiting for IO. Otherwise our
1184 * cache state won't be right, and we won't get written again
1186 if (!ret && list_empty(&block_group->dirty_list))
1187 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1189 block_group->disk_cache_state = BTRFS_DC_ERROR;
1191 spin_unlock(&block_group->lock);
1192 spin_unlock(&trans->transaction->dirty_bgs_lock);
1193 io_ctl->inode = NULL;
1202 * __btrfs_write_out_cache - write out cached info to an inode
1203 * @root - the root the inode belongs to
1204 * @ctl - the free space cache we are going to write out
1205 * @block_group - the block_group for this cache if it belongs to a block_group
1206 * @trans - the trans handle
1207 * @path - the path to use
1208 * @offset - the offset for the key we'll insert
1210 * This function writes out a free space cache struct to disk for quick recovery
1211 * on mount. This will return 0 if it was successful in writing the cache out,
1212 * or an errno if it was not.
1214 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1215 struct btrfs_free_space_ctl *ctl,
1216 struct btrfs_block_group_cache *block_group,
1217 struct btrfs_io_ctl *io_ctl,
1218 struct btrfs_trans_handle *trans,
1219 struct btrfs_path *path, u64 offset)
1221 struct extent_state *cached_state = NULL;
1222 LIST_HEAD(bitmap_list);
1228 if (!i_size_read(inode))
1231 WARN_ON(io_ctl->pages);
1232 ret = io_ctl_init(io_ctl, inode, root, 1);
1236 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1237 down_write(&block_group->data_rwsem);
1238 spin_lock(&block_group->lock);
1239 if (block_group->delalloc_bytes) {
1240 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1241 spin_unlock(&block_group->lock);
1242 up_write(&block_group->data_rwsem);
1243 BTRFS_I(inode)->generation = 0;
1248 spin_unlock(&block_group->lock);
1251 /* Lock all pages first so we can lock the extent safely. */
1252 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1256 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1259 io_ctl_set_generation(io_ctl, trans->transid);
1261 mutex_lock(&ctl->cache_writeout_mutex);
1262 /* Write out the extent entries in the free space cache */
1263 spin_lock(&ctl->tree_lock);
1264 ret = write_cache_extent_entries(io_ctl, ctl,
1265 block_group, &entries, &bitmaps,
1268 goto out_nospc_locked;
1271 * Some spaces that are freed in the current transaction are pinned,
1272 * they will be added into free space cache after the transaction is
1273 * committed, we shouldn't lose them.
1275 * If this changes while we are working we'll get added back to
1276 * the dirty list and redo it. No locking needed
1278 ret = write_pinned_extent_entries(root, block_group, io_ctl, &entries);
1280 goto out_nospc_locked;
1283 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1284 * locked while doing it because a concurrent trim can be manipulating
1285 * or freeing the bitmap.
1287 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1288 spin_unlock(&ctl->tree_lock);
1289 mutex_unlock(&ctl->cache_writeout_mutex);
1293 /* Zero out the rest of the pages just to make sure */
1294 io_ctl_zero_remaining_pages(io_ctl);
1296 /* Everything is written out, now we dirty the pages in the file. */
1297 ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages,
1298 0, i_size_read(inode), &cached_state);
1302 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1303 up_write(&block_group->data_rwsem);
1305 * Release the pages and unlock the extent, we will flush
1308 io_ctl_drop_pages(io_ctl);
1310 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1311 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1314 * at this point the pages are under IO and we're happy,
1315 * The caller is responsible for waiting on them and updating the
1316 * the cache and the inode
1318 io_ctl->entries = entries;
1319 io_ctl->bitmaps = bitmaps;
1321 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1328 io_ctl->inode = NULL;
1329 io_ctl_free(io_ctl);
1331 invalidate_inode_pages2(inode->i_mapping);
1332 BTRFS_I(inode)->generation = 0;
1334 btrfs_update_inode(trans, root, inode);
1340 cleanup_bitmap_list(&bitmap_list);
1341 spin_unlock(&ctl->tree_lock);
1342 mutex_unlock(&ctl->cache_writeout_mutex);
1345 cleanup_write_cache_enospc(inode, io_ctl, &cached_state, &bitmap_list);
1347 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1348 up_write(&block_group->data_rwsem);
1353 int btrfs_write_out_cache(struct btrfs_root *root,
1354 struct btrfs_trans_handle *trans,
1355 struct btrfs_block_group_cache *block_group,
1356 struct btrfs_path *path)
1358 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1359 struct inode *inode;
1362 root = root->fs_info->tree_root;
1364 spin_lock(&block_group->lock);
1365 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1366 spin_unlock(&block_group->lock);
1369 spin_unlock(&block_group->lock);
1371 inode = lookup_free_space_inode(root, block_group, path);
1375 ret = __btrfs_write_out_cache(root, inode, ctl, block_group,
1376 &block_group->io_ctl, trans,
1377 path, block_group->key.objectid);
1380 btrfs_err(root->fs_info,
1381 "failed to write free space cache for block group %llu",
1382 block_group->key.objectid);
1384 spin_lock(&block_group->lock);
1385 block_group->disk_cache_state = BTRFS_DC_ERROR;
1386 spin_unlock(&block_group->lock);
1388 block_group->io_ctl.inode = NULL;
1393 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1394 * to wait for IO and put the inode
1400 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1403 ASSERT(offset >= bitmap_start);
1404 offset -= bitmap_start;
1405 return (unsigned long)(div_u64(offset, unit));
1408 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1410 return (unsigned long)(div_u64(bytes, unit));
1413 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1417 u64 bytes_per_bitmap;
1419 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1420 bitmap_start = offset - ctl->start;
1421 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1422 bitmap_start *= bytes_per_bitmap;
1423 bitmap_start += ctl->start;
1425 return bitmap_start;
1428 static int tree_insert_offset(struct rb_root *root, u64 offset,
1429 struct rb_node *node, int bitmap)
1431 struct rb_node **p = &root->rb_node;
1432 struct rb_node *parent = NULL;
1433 struct btrfs_free_space *info;
1437 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1439 if (offset < info->offset) {
1441 } else if (offset > info->offset) {
1442 p = &(*p)->rb_right;
1445 * we could have a bitmap entry and an extent entry
1446 * share the same offset. If this is the case, we want
1447 * the extent entry to always be found first if we do a
1448 * linear search through the tree, since we want to have
1449 * the quickest allocation time, and allocating from an
1450 * extent is faster than allocating from a bitmap. So
1451 * if we're inserting a bitmap and we find an entry at
1452 * this offset, we want to go right, or after this entry
1453 * logically. If we are inserting an extent and we've
1454 * found a bitmap, we want to go left, or before
1462 p = &(*p)->rb_right;
1464 if (!info->bitmap) {
1473 rb_link_node(node, parent, p);
1474 rb_insert_color(node, root);
1480 * searches the tree for the given offset.
1482 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1483 * want a section that has at least bytes size and comes at or after the given
1486 static struct btrfs_free_space *
1487 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1488 u64 offset, int bitmap_only, int fuzzy)
1490 struct rb_node *n = ctl->free_space_offset.rb_node;
1491 struct btrfs_free_space *entry, *prev = NULL;
1493 /* find entry that is closest to the 'offset' */
1500 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1503 if (offset < entry->offset)
1505 else if (offset > entry->offset)
1518 * bitmap entry and extent entry may share same offset,
1519 * in that case, bitmap entry comes after extent entry.
1524 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1525 if (entry->offset != offset)
1528 WARN_ON(!entry->bitmap);
1531 if (entry->bitmap) {
1533 * if previous extent entry covers the offset,
1534 * we should return it instead of the bitmap entry
1536 n = rb_prev(&entry->offset_index);
1538 prev = rb_entry(n, struct btrfs_free_space,
1540 if (!prev->bitmap &&
1541 prev->offset + prev->bytes > offset)
1551 /* find last entry before the 'offset' */
1553 if (entry->offset > offset) {
1554 n = rb_prev(&entry->offset_index);
1556 entry = rb_entry(n, struct btrfs_free_space,
1558 ASSERT(entry->offset <= offset);
1567 if (entry->bitmap) {
1568 n = rb_prev(&entry->offset_index);
1570 prev = rb_entry(n, struct btrfs_free_space,
1572 if (!prev->bitmap &&
1573 prev->offset + prev->bytes > offset)
1576 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1578 } else if (entry->offset + entry->bytes > offset)
1585 if (entry->bitmap) {
1586 if (entry->offset + BITS_PER_BITMAP *
1590 if (entry->offset + entry->bytes > offset)
1594 n = rb_next(&entry->offset_index);
1597 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1603 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1604 struct btrfs_free_space *info)
1606 rb_erase(&info->offset_index, &ctl->free_space_offset);
1607 ctl->free_extents--;
1610 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1611 struct btrfs_free_space *info)
1613 __unlink_free_space(ctl, info);
1614 ctl->free_space -= info->bytes;
1617 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1618 struct btrfs_free_space *info)
1622 ASSERT(info->bytes || info->bitmap);
1623 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1624 &info->offset_index, (info->bitmap != NULL));
1628 ctl->free_space += info->bytes;
1629 ctl->free_extents++;
1633 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1635 struct btrfs_block_group_cache *block_group = ctl->private;
1639 u64 size = block_group->key.offset;
1640 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1641 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1643 max_bitmaps = max_t(u64, max_bitmaps, 1);
1645 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1648 * The goal is to keep the total amount of memory used per 1gb of space
1649 * at or below 32k, so we need to adjust how much memory we allow to be
1650 * used by extent based free space tracking
1653 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1655 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
1658 * we want to account for 1 more bitmap than what we have so we can make
1659 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1660 * we add more bitmaps.
1662 bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit;
1664 if (bitmap_bytes >= max_bytes) {
1665 ctl->extents_thresh = 0;
1670 * we want the extent entry threshold to always be at most 1/2 the max
1671 * bytes we can have, or whatever is less than that.
1673 extent_bytes = max_bytes - bitmap_bytes;
1674 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1676 ctl->extents_thresh =
1677 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1680 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1681 struct btrfs_free_space *info,
1682 u64 offset, u64 bytes)
1684 unsigned long start, count;
1686 start = offset_to_bit(info->offset, ctl->unit, offset);
1687 count = bytes_to_bits(bytes, ctl->unit);
1688 ASSERT(start + count <= BITS_PER_BITMAP);
1690 bitmap_clear(info->bitmap, start, count);
1692 info->bytes -= bytes;
1695 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1696 struct btrfs_free_space *info, u64 offset,
1699 __bitmap_clear_bits(ctl, info, offset, bytes);
1700 ctl->free_space -= bytes;
1703 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1704 struct btrfs_free_space *info, u64 offset,
1707 unsigned long start, count;
1709 start = offset_to_bit(info->offset, ctl->unit, offset);
1710 count = bytes_to_bits(bytes, ctl->unit);
1711 ASSERT(start + count <= BITS_PER_BITMAP);
1713 bitmap_set(info->bitmap, start, count);
1715 info->bytes += bytes;
1716 ctl->free_space += bytes;
1720 * If we can not find suitable extent, we will use bytes to record
1721 * the size of the max extent.
1723 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1724 struct btrfs_free_space *bitmap_info, u64 *offset,
1725 u64 *bytes, bool for_alloc)
1727 unsigned long found_bits = 0;
1728 unsigned long max_bits = 0;
1729 unsigned long bits, i;
1730 unsigned long next_zero;
1731 unsigned long extent_bits;
1734 * Skip searching the bitmap if we don't have a contiguous section that
1735 * is large enough for this allocation.
1738 bitmap_info->max_extent_size &&
1739 bitmap_info->max_extent_size < *bytes) {
1740 *bytes = bitmap_info->max_extent_size;
1744 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1745 max_t(u64, *offset, bitmap_info->offset));
1746 bits = bytes_to_bits(*bytes, ctl->unit);
1748 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1749 if (for_alloc && bits == 1) {
1753 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1754 BITS_PER_BITMAP, i);
1755 extent_bits = next_zero - i;
1756 if (extent_bits >= bits) {
1757 found_bits = extent_bits;
1759 } else if (extent_bits > max_bits) {
1760 max_bits = extent_bits;
1766 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1767 *bytes = (u64)(found_bits) * ctl->unit;
1771 *bytes = (u64)(max_bits) * ctl->unit;
1772 bitmap_info->max_extent_size = *bytes;
1776 /* Cache the size of the max extent in bytes */
1777 static struct btrfs_free_space *
1778 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1779 unsigned long align, u64 *max_extent_size)
1781 struct btrfs_free_space *entry;
1782 struct rb_node *node;
1787 if (!ctl->free_space_offset.rb_node)
1790 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1794 for (node = &entry->offset_index; node; node = rb_next(node)) {
1795 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1796 if (entry->bytes < *bytes) {
1797 if (entry->bytes > *max_extent_size)
1798 *max_extent_size = entry->bytes;
1802 /* make sure the space returned is big enough
1803 * to match our requested alignment
1805 if (*bytes >= align) {
1806 tmp = entry->offset - ctl->start + align - 1;
1807 tmp = div64_u64(tmp, align);
1808 tmp = tmp * align + ctl->start;
1809 align_off = tmp - entry->offset;
1812 tmp = entry->offset;
1815 if (entry->bytes < *bytes + align_off) {
1816 if (entry->bytes > *max_extent_size)
1817 *max_extent_size = entry->bytes;
1821 if (entry->bitmap) {
1824 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1829 } else if (size > *max_extent_size) {
1830 *max_extent_size = size;
1836 *bytes = entry->bytes - align_off;
1843 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1844 struct btrfs_free_space *info, u64 offset)
1846 info->offset = offset_to_bitmap(ctl, offset);
1848 INIT_LIST_HEAD(&info->list);
1849 link_free_space(ctl, info);
1850 ctl->total_bitmaps++;
1852 ctl->op->recalc_thresholds(ctl);
1855 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1856 struct btrfs_free_space *bitmap_info)
1858 unlink_free_space(ctl, bitmap_info);
1859 kfree(bitmap_info->bitmap);
1860 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1861 ctl->total_bitmaps--;
1862 ctl->op->recalc_thresholds(ctl);
1865 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1866 struct btrfs_free_space *bitmap_info,
1867 u64 *offset, u64 *bytes)
1870 u64 search_start, search_bytes;
1874 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1877 * We need to search for bits in this bitmap. We could only cover some
1878 * of the extent in this bitmap thanks to how we add space, so we need
1879 * to search for as much as it as we can and clear that amount, and then
1880 * go searching for the next bit.
1882 search_start = *offset;
1883 search_bytes = ctl->unit;
1884 search_bytes = min(search_bytes, end - search_start + 1);
1885 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1887 if (ret < 0 || search_start != *offset)
1890 /* We may have found more bits than what we need */
1891 search_bytes = min(search_bytes, *bytes);
1893 /* Cannot clear past the end of the bitmap */
1894 search_bytes = min(search_bytes, end - search_start + 1);
1896 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1897 *offset += search_bytes;
1898 *bytes -= search_bytes;
1901 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1902 if (!bitmap_info->bytes)
1903 free_bitmap(ctl, bitmap_info);
1906 * no entry after this bitmap, but we still have bytes to
1907 * remove, so something has gone wrong.
1912 bitmap_info = rb_entry(next, struct btrfs_free_space,
1916 * if the next entry isn't a bitmap we need to return to let the
1917 * extent stuff do its work.
1919 if (!bitmap_info->bitmap)
1923 * Ok the next item is a bitmap, but it may not actually hold
1924 * the information for the rest of this free space stuff, so
1925 * look for it, and if we don't find it return so we can try
1926 * everything over again.
1928 search_start = *offset;
1929 search_bytes = ctl->unit;
1930 ret = search_bitmap(ctl, bitmap_info, &search_start,
1931 &search_bytes, false);
1932 if (ret < 0 || search_start != *offset)
1936 } else if (!bitmap_info->bytes)
1937 free_bitmap(ctl, bitmap_info);
1942 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1943 struct btrfs_free_space *info, u64 offset,
1946 u64 bytes_to_set = 0;
1949 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1951 bytes_to_set = min(end - offset, bytes);
1953 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1956 * We set some bytes, we have no idea what the max extent size is
1959 info->max_extent_size = 0;
1961 return bytes_to_set;
1965 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1966 struct btrfs_free_space *info)
1968 struct btrfs_block_group_cache *block_group = ctl->private;
1969 bool forced = false;
1971 #ifdef CONFIG_BTRFS_DEBUG
1972 if (btrfs_should_fragment_free_space(block_group->fs_info->extent_root,
1978 * If we are below the extents threshold then we can add this as an
1979 * extent, and don't have to deal with the bitmap
1981 if (!forced && ctl->free_extents < ctl->extents_thresh) {
1983 * If this block group has some small extents we don't want to
1984 * use up all of our free slots in the cache with them, we want
1985 * to reserve them to larger extents, however if we have plenty
1986 * of cache left then go ahead an dadd them, no sense in adding
1987 * the overhead of a bitmap if we don't have to.
1989 if (info->bytes <= block_group->sectorsize * 4) {
1990 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1998 * The original block groups from mkfs can be really small, like 8
1999 * megabytes, so don't bother with a bitmap for those entries. However
2000 * some block groups can be smaller than what a bitmap would cover but
2001 * are still large enough that they could overflow the 32k memory limit,
2002 * so allow those block groups to still be allowed to have a bitmap
2005 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2011 static const struct btrfs_free_space_op free_space_op = {
2012 .recalc_thresholds = recalculate_thresholds,
2013 .use_bitmap = use_bitmap,
2016 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2017 struct btrfs_free_space *info)
2019 struct btrfs_free_space *bitmap_info;
2020 struct btrfs_block_group_cache *block_group = NULL;
2022 u64 bytes, offset, bytes_added;
2025 bytes = info->bytes;
2026 offset = info->offset;
2028 if (!ctl->op->use_bitmap(ctl, info))
2031 if (ctl->op == &free_space_op)
2032 block_group = ctl->private;
2035 * Since we link bitmaps right into the cluster we need to see if we
2036 * have a cluster here, and if so and it has our bitmap we need to add
2037 * the free space to that bitmap.
2039 if (block_group && !list_empty(&block_group->cluster_list)) {
2040 struct btrfs_free_cluster *cluster;
2041 struct rb_node *node;
2042 struct btrfs_free_space *entry;
2044 cluster = list_entry(block_group->cluster_list.next,
2045 struct btrfs_free_cluster,
2047 spin_lock(&cluster->lock);
2048 node = rb_first(&cluster->root);
2050 spin_unlock(&cluster->lock);
2051 goto no_cluster_bitmap;
2054 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2055 if (!entry->bitmap) {
2056 spin_unlock(&cluster->lock);
2057 goto no_cluster_bitmap;
2060 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2061 bytes_added = add_bytes_to_bitmap(ctl, entry,
2063 bytes -= bytes_added;
2064 offset += bytes_added;
2066 spin_unlock(&cluster->lock);
2074 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2081 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2082 bytes -= bytes_added;
2083 offset += bytes_added;
2093 if (info && info->bitmap) {
2094 add_new_bitmap(ctl, info, offset);
2099 spin_unlock(&ctl->tree_lock);
2101 /* no pre-allocated info, allocate a new one */
2103 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2106 spin_lock(&ctl->tree_lock);
2112 /* allocate the bitmap */
2113 info->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
2114 spin_lock(&ctl->tree_lock);
2115 if (!info->bitmap) {
2125 kfree(info->bitmap);
2126 kmem_cache_free(btrfs_free_space_cachep, info);
2132 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2133 struct btrfs_free_space *info, bool update_stat)
2135 struct btrfs_free_space *left_info;
2136 struct btrfs_free_space *right_info;
2137 bool merged = false;
2138 u64 offset = info->offset;
2139 u64 bytes = info->bytes;
2142 * first we want to see if there is free space adjacent to the range we
2143 * are adding, if there is remove that struct and add a new one to
2144 * cover the entire range
2146 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2147 if (right_info && rb_prev(&right_info->offset_index))
2148 left_info = rb_entry(rb_prev(&right_info->offset_index),
2149 struct btrfs_free_space, offset_index);
2151 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2153 if (right_info && !right_info->bitmap) {
2155 unlink_free_space(ctl, right_info);
2157 __unlink_free_space(ctl, right_info);
2158 info->bytes += right_info->bytes;
2159 kmem_cache_free(btrfs_free_space_cachep, right_info);
2163 if (left_info && !left_info->bitmap &&
2164 left_info->offset + left_info->bytes == offset) {
2166 unlink_free_space(ctl, left_info);
2168 __unlink_free_space(ctl, left_info);
2169 info->offset = left_info->offset;
2170 info->bytes += left_info->bytes;
2171 kmem_cache_free(btrfs_free_space_cachep, left_info);
2178 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2179 struct btrfs_free_space *info,
2182 struct btrfs_free_space *bitmap;
2185 const u64 end = info->offset + info->bytes;
2186 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2189 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2193 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2194 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2197 bytes = (j - i) * ctl->unit;
2198 info->bytes += bytes;
2201 bitmap_clear_bits(ctl, bitmap, end, bytes);
2203 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2206 free_bitmap(ctl, bitmap);
2211 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2212 struct btrfs_free_space *info,
2215 struct btrfs_free_space *bitmap;
2219 unsigned long prev_j;
2222 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2223 /* If we're on a boundary, try the previous logical bitmap. */
2224 if (bitmap_offset == info->offset) {
2225 if (info->offset == 0)
2227 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2230 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2234 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2236 prev_j = (unsigned long)-1;
2237 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2245 if (prev_j == (unsigned long)-1)
2246 bytes = (i + 1) * ctl->unit;
2248 bytes = (i - prev_j) * ctl->unit;
2250 info->offset -= bytes;
2251 info->bytes += bytes;
2254 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2256 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2259 free_bitmap(ctl, bitmap);
2265 * We prefer always to allocate from extent entries, both for clustered and
2266 * non-clustered allocation requests. So when attempting to add a new extent
2267 * entry, try to see if there's adjacent free space in bitmap entries, and if
2268 * there is, migrate that space from the bitmaps to the extent.
2269 * Like this we get better chances of satisfying space allocation requests
2270 * because we attempt to satisfy them based on a single cache entry, and never
2271 * on 2 or more entries - even if the entries represent a contiguous free space
2272 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2275 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2276 struct btrfs_free_space *info,
2280 * Only work with disconnected entries, as we can change their offset,
2281 * and must be extent entries.
2283 ASSERT(!info->bitmap);
2284 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2286 if (ctl->total_bitmaps > 0) {
2288 bool stole_front = false;
2290 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2291 if (ctl->total_bitmaps > 0)
2292 stole_front = steal_from_bitmap_to_front(ctl, info,
2295 if (stole_end || stole_front)
2296 try_merge_free_space(ctl, info, update_stat);
2300 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2301 struct btrfs_free_space_ctl *ctl,
2302 u64 offset, u64 bytes)
2304 struct btrfs_free_space *info;
2307 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2311 info->offset = offset;
2312 info->bytes = bytes;
2313 RB_CLEAR_NODE(&info->offset_index);
2315 spin_lock(&ctl->tree_lock);
2317 if (try_merge_free_space(ctl, info, true))
2321 * There was no extent directly to the left or right of this new
2322 * extent then we know we're going to have to allocate a new extent, so
2323 * before we do that see if we need to drop this into a bitmap
2325 ret = insert_into_bitmap(ctl, info);
2334 * Only steal free space from adjacent bitmaps if we're sure we're not
2335 * going to add the new free space to existing bitmap entries - because
2336 * that would mean unnecessary work that would be reverted. Therefore
2337 * attempt to steal space from bitmaps if we're adding an extent entry.
2339 steal_from_bitmap(ctl, info, true);
2341 ret = link_free_space(ctl, info);
2343 kmem_cache_free(btrfs_free_space_cachep, info);
2345 spin_unlock(&ctl->tree_lock);
2348 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2349 ASSERT(ret != -EEXIST);
2355 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2356 u64 offset, u64 bytes)
2358 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2359 struct btrfs_free_space *info;
2361 bool re_search = false;
2363 spin_lock(&ctl->tree_lock);
2370 info = tree_search_offset(ctl, offset, 0, 0);
2373 * oops didn't find an extent that matched the space we wanted
2374 * to remove, look for a bitmap instead
2376 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2380 * If we found a partial bit of our free space in a
2381 * bitmap but then couldn't find the other part this may
2382 * be a problem, so WARN about it.
2390 if (!info->bitmap) {
2391 unlink_free_space(ctl, info);
2392 if (offset == info->offset) {
2393 u64 to_free = min(bytes, info->bytes);
2395 info->bytes -= to_free;
2396 info->offset += to_free;
2398 ret = link_free_space(ctl, info);
2401 kmem_cache_free(btrfs_free_space_cachep, info);
2408 u64 old_end = info->bytes + info->offset;
2410 info->bytes = offset - info->offset;
2411 ret = link_free_space(ctl, info);
2416 /* Not enough bytes in this entry to satisfy us */
2417 if (old_end < offset + bytes) {
2418 bytes -= old_end - offset;
2421 } else if (old_end == offset + bytes) {
2425 spin_unlock(&ctl->tree_lock);
2427 ret = btrfs_add_free_space(block_group, offset + bytes,
2428 old_end - (offset + bytes));
2434 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2435 if (ret == -EAGAIN) {
2440 spin_unlock(&ctl->tree_lock);
2445 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2448 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2449 struct btrfs_free_space *info;
2453 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2454 info = rb_entry(n, struct btrfs_free_space, offset_index);
2455 if (info->bytes >= bytes && !block_group->ro)
2457 btrfs_crit(block_group->fs_info,
2458 "entry offset %llu, bytes %llu, bitmap %s",
2459 info->offset, info->bytes,
2460 (info->bitmap) ? "yes" : "no");
2462 btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2463 list_empty(&block_group->cluster_list) ? "no" : "yes");
2464 btrfs_info(block_group->fs_info,
2465 "%d blocks of free space at or bigger than bytes is", count);
2468 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2470 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2472 spin_lock_init(&ctl->tree_lock);
2473 ctl->unit = block_group->sectorsize;
2474 ctl->start = block_group->key.objectid;
2475 ctl->private = block_group;
2476 ctl->op = &free_space_op;
2477 INIT_LIST_HEAD(&ctl->trimming_ranges);
2478 mutex_init(&ctl->cache_writeout_mutex);
2481 * we only want to have 32k of ram per block group for keeping
2482 * track of free space, and if we pass 1/2 of that we want to
2483 * start converting things over to using bitmaps
2485 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2489 * for a given cluster, put all of its extents back into the free
2490 * space cache. If the block group passed doesn't match the block group
2491 * pointed to by the cluster, someone else raced in and freed the
2492 * cluster already. In that case, we just return without changing anything
2495 __btrfs_return_cluster_to_free_space(
2496 struct btrfs_block_group_cache *block_group,
2497 struct btrfs_free_cluster *cluster)
2499 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2500 struct btrfs_free_space *entry;
2501 struct rb_node *node;
2503 spin_lock(&cluster->lock);
2504 if (cluster->block_group != block_group)
2507 cluster->block_group = NULL;
2508 cluster->window_start = 0;
2509 list_del_init(&cluster->block_group_list);
2511 node = rb_first(&cluster->root);
2515 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2516 node = rb_next(&entry->offset_index);
2517 rb_erase(&entry->offset_index, &cluster->root);
2518 RB_CLEAR_NODE(&entry->offset_index);
2520 bitmap = (entry->bitmap != NULL);
2522 try_merge_free_space(ctl, entry, false);
2523 steal_from_bitmap(ctl, entry, false);
2525 tree_insert_offset(&ctl->free_space_offset,
2526 entry->offset, &entry->offset_index, bitmap);
2528 cluster->root = RB_ROOT;
2531 spin_unlock(&cluster->lock);
2532 btrfs_put_block_group(block_group);
2536 static void __btrfs_remove_free_space_cache_locked(
2537 struct btrfs_free_space_ctl *ctl)
2539 struct btrfs_free_space *info;
2540 struct rb_node *node;
2542 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2543 info = rb_entry(node, struct btrfs_free_space, offset_index);
2544 if (!info->bitmap) {
2545 unlink_free_space(ctl, info);
2546 kmem_cache_free(btrfs_free_space_cachep, info);
2548 free_bitmap(ctl, info);
2551 cond_resched_lock(&ctl->tree_lock);
2555 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2557 spin_lock(&ctl->tree_lock);
2558 __btrfs_remove_free_space_cache_locked(ctl);
2559 spin_unlock(&ctl->tree_lock);
2562 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2564 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2565 struct btrfs_free_cluster *cluster;
2566 struct list_head *head;
2568 spin_lock(&ctl->tree_lock);
2569 while ((head = block_group->cluster_list.next) !=
2570 &block_group->cluster_list) {
2571 cluster = list_entry(head, struct btrfs_free_cluster,
2574 WARN_ON(cluster->block_group != block_group);
2575 __btrfs_return_cluster_to_free_space(block_group, cluster);
2577 cond_resched_lock(&ctl->tree_lock);
2579 __btrfs_remove_free_space_cache_locked(ctl);
2580 spin_unlock(&ctl->tree_lock);
2584 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2585 u64 offset, u64 bytes, u64 empty_size,
2586 u64 *max_extent_size)
2588 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2589 struct btrfs_free_space *entry = NULL;
2590 u64 bytes_search = bytes + empty_size;
2593 u64 align_gap_len = 0;
2595 spin_lock(&ctl->tree_lock);
2596 entry = find_free_space(ctl, &offset, &bytes_search,
2597 block_group->full_stripe_len, max_extent_size);
2602 if (entry->bitmap) {
2603 bitmap_clear_bits(ctl, entry, offset, bytes);
2605 free_bitmap(ctl, entry);
2607 unlink_free_space(ctl, entry);
2608 align_gap_len = offset - entry->offset;
2609 align_gap = entry->offset;
2611 entry->offset = offset + bytes;
2612 WARN_ON(entry->bytes < bytes + align_gap_len);
2614 entry->bytes -= bytes + align_gap_len;
2616 kmem_cache_free(btrfs_free_space_cachep, entry);
2618 link_free_space(ctl, entry);
2621 spin_unlock(&ctl->tree_lock);
2624 __btrfs_add_free_space(block_group->fs_info, ctl,
2625 align_gap, align_gap_len);
2630 * given a cluster, put all of its extents back into the free space
2631 * cache. If a block group is passed, this function will only free
2632 * a cluster that belongs to the passed block group.
2634 * Otherwise, it'll get a reference on the block group pointed to by the
2635 * cluster and remove the cluster from it.
2637 int btrfs_return_cluster_to_free_space(
2638 struct btrfs_block_group_cache *block_group,
2639 struct btrfs_free_cluster *cluster)
2641 struct btrfs_free_space_ctl *ctl;
2644 /* first, get a safe pointer to the block group */
2645 spin_lock(&cluster->lock);
2647 block_group = cluster->block_group;
2649 spin_unlock(&cluster->lock);
2652 } else if (cluster->block_group != block_group) {
2653 /* someone else has already freed it don't redo their work */
2654 spin_unlock(&cluster->lock);
2657 atomic_inc(&block_group->count);
2658 spin_unlock(&cluster->lock);
2660 ctl = block_group->free_space_ctl;
2662 /* now return any extents the cluster had on it */
2663 spin_lock(&ctl->tree_lock);
2664 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2665 spin_unlock(&ctl->tree_lock);
2667 /* finally drop our ref */
2668 btrfs_put_block_group(block_group);
2672 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2673 struct btrfs_free_cluster *cluster,
2674 struct btrfs_free_space *entry,
2675 u64 bytes, u64 min_start,
2676 u64 *max_extent_size)
2678 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2680 u64 search_start = cluster->window_start;
2681 u64 search_bytes = bytes;
2684 search_start = min_start;
2685 search_bytes = bytes;
2687 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2689 if (search_bytes > *max_extent_size)
2690 *max_extent_size = search_bytes;
2695 __bitmap_clear_bits(ctl, entry, ret, bytes);
2701 * given a cluster, try to allocate 'bytes' from it, returns 0
2702 * if it couldn't find anything suitably large, or a logical disk offset
2703 * if things worked out
2705 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2706 struct btrfs_free_cluster *cluster, u64 bytes,
2707 u64 min_start, u64 *max_extent_size)
2709 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2710 struct btrfs_free_space *entry = NULL;
2711 struct rb_node *node;
2714 spin_lock(&cluster->lock);
2715 if (bytes > cluster->max_size)
2718 if (cluster->block_group != block_group)
2721 node = rb_first(&cluster->root);
2725 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2727 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2728 *max_extent_size = entry->bytes;
2730 if (entry->bytes < bytes ||
2731 (!entry->bitmap && entry->offset < min_start)) {
2732 node = rb_next(&entry->offset_index);
2735 entry = rb_entry(node, struct btrfs_free_space,
2740 if (entry->bitmap) {
2741 ret = btrfs_alloc_from_bitmap(block_group,
2742 cluster, entry, bytes,
2743 cluster->window_start,
2746 node = rb_next(&entry->offset_index);
2749 entry = rb_entry(node, struct btrfs_free_space,
2753 cluster->window_start += bytes;
2755 ret = entry->offset;
2757 entry->offset += bytes;
2758 entry->bytes -= bytes;
2761 if (entry->bytes == 0)
2762 rb_erase(&entry->offset_index, &cluster->root);
2766 spin_unlock(&cluster->lock);
2771 spin_lock(&ctl->tree_lock);
2773 ctl->free_space -= bytes;
2774 if (entry->bytes == 0) {
2775 ctl->free_extents--;
2776 if (entry->bitmap) {
2777 kfree(entry->bitmap);
2778 ctl->total_bitmaps--;
2779 ctl->op->recalc_thresholds(ctl);
2781 kmem_cache_free(btrfs_free_space_cachep, entry);
2784 spin_unlock(&ctl->tree_lock);
2789 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2790 struct btrfs_free_space *entry,
2791 struct btrfs_free_cluster *cluster,
2792 u64 offset, u64 bytes,
2793 u64 cont1_bytes, u64 min_bytes)
2795 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2796 unsigned long next_zero;
2798 unsigned long want_bits;
2799 unsigned long min_bits;
2800 unsigned long found_bits;
2801 unsigned long max_bits = 0;
2802 unsigned long start = 0;
2803 unsigned long total_found = 0;
2806 i = offset_to_bit(entry->offset, ctl->unit,
2807 max_t(u64, offset, entry->offset));
2808 want_bits = bytes_to_bits(bytes, ctl->unit);
2809 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2812 * Don't bother looking for a cluster in this bitmap if it's heavily
2815 if (entry->max_extent_size &&
2816 entry->max_extent_size < cont1_bytes)
2820 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2821 next_zero = find_next_zero_bit(entry->bitmap,
2822 BITS_PER_BITMAP, i);
2823 if (next_zero - i >= min_bits) {
2824 found_bits = next_zero - i;
2825 if (found_bits > max_bits)
2826 max_bits = found_bits;
2829 if (next_zero - i > max_bits)
2830 max_bits = next_zero - i;
2835 entry->max_extent_size = (u64)max_bits * ctl->unit;
2841 cluster->max_size = 0;
2844 total_found += found_bits;
2846 if (cluster->max_size < found_bits * ctl->unit)
2847 cluster->max_size = found_bits * ctl->unit;
2849 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2854 cluster->window_start = start * ctl->unit + entry->offset;
2855 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2856 ret = tree_insert_offset(&cluster->root, entry->offset,
2857 &entry->offset_index, 1);
2858 ASSERT(!ret); /* -EEXIST; Logic error */
2860 trace_btrfs_setup_cluster(block_group, cluster,
2861 total_found * ctl->unit, 1);
2866 * This searches the block group for just extents to fill the cluster with.
2867 * Try to find a cluster with at least bytes total bytes, at least one
2868 * extent of cont1_bytes, and other clusters of at least min_bytes.
2871 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2872 struct btrfs_free_cluster *cluster,
2873 struct list_head *bitmaps, u64 offset, u64 bytes,
2874 u64 cont1_bytes, u64 min_bytes)
2876 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2877 struct btrfs_free_space *first = NULL;
2878 struct btrfs_free_space *entry = NULL;
2879 struct btrfs_free_space *last;
2880 struct rb_node *node;
2885 entry = tree_search_offset(ctl, offset, 0, 1);
2890 * We don't want bitmaps, so just move along until we find a normal
2893 while (entry->bitmap || entry->bytes < min_bytes) {
2894 if (entry->bitmap && list_empty(&entry->list))
2895 list_add_tail(&entry->list, bitmaps);
2896 node = rb_next(&entry->offset_index);
2899 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2902 window_free = entry->bytes;
2903 max_extent = entry->bytes;
2907 for (node = rb_next(&entry->offset_index); node;
2908 node = rb_next(&entry->offset_index)) {
2909 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2911 if (entry->bitmap) {
2912 if (list_empty(&entry->list))
2913 list_add_tail(&entry->list, bitmaps);
2917 if (entry->bytes < min_bytes)
2921 window_free += entry->bytes;
2922 if (entry->bytes > max_extent)
2923 max_extent = entry->bytes;
2926 if (window_free < bytes || max_extent < cont1_bytes)
2929 cluster->window_start = first->offset;
2931 node = &first->offset_index;
2934 * now we've found our entries, pull them out of the free space
2935 * cache and put them into the cluster rbtree
2940 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2941 node = rb_next(&entry->offset_index);
2942 if (entry->bitmap || entry->bytes < min_bytes)
2945 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2946 ret = tree_insert_offset(&cluster->root, entry->offset,
2947 &entry->offset_index, 0);
2948 total_size += entry->bytes;
2949 ASSERT(!ret); /* -EEXIST; Logic error */
2950 } while (node && entry != last);
2952 cluster->max_size = max_extent;
2953 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2958 * This specifically looks for bitmaps that may work in the cluster, we assume
2959 * that we have already failed to find extents that will work.
2962 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2963 struct btrfs_free_cluster *cluster,
2964 struct list_head *bitmaps, u64 offset, u64 bytes,
2965 u64 cont1_bytes, u64 min_bytes)
2967 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2968 struct btrfs_free_space *entry = NULL;
2970 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2972 if (ctl->total_bitmaps == 0)
2976 * The bitmap that covers offset won't be in the list unless offset
2977 * is just its start offset.
2979 if (!list_empty(bitmaps))
2980 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2982 if (!entry || entry->offset != bitmap_offset) {
2983 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2984 if (entry && list_empty(&entry->list))
2985 list_add(&entry->list, bitmaps);
2988 list_for_each_entry(entry, bitmaps, list) {
2989 if (entry->bytes < bytes)
2991 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2992 bytes, cont1_bytes, min_bytes);
2998 * The bitmaps list has all the bitmaps that record free space
2999 * starting after offset, so no more search is required.
3005 * here we try to find a cluster of blocks in a block group. The goal
3006 * is to find at least bytes+empty_size.
3007 * We might not find them all in one contiguous area.
3009 * returns zero and sets up cluster if things worked out, otherwise
3010 * it returns -enospc
3012 int btrfs_find_space_cluster(struct btrfs_root *root,
3013 struct btrfs_block_group_cache *block_group,
3014 struct btrfs_free_cluster *cluster,
3015 u64 offset, u64 bytes, u64 empty_size)
3017 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3018 struct btrfs_free_space *entry, *tmp;
3025 * Choose the minimum extent size we'll require for this
3026 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3027 * For metadata, allow allocates with smaller extents. For
3028 * data, keep it dense.
3030 if (btrfs_test_opt(root->fs_info, SSD_SPREAD)) {
3031 cont1_bytes = min_bytes = bytes + empty_size;
3032 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3033 cont1_bytes = bytes;
3034 min_bytes = block_group->sectorsize;
3036 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3037 min_bytes = block_group->sectorsize;
3040 spin_lock(&ctl->tree_lock);
3043 * If we know we don't have enough space to make a cluster don't even
3044 * bother doing all the work to try and find one.
3046 if (ctl->free_space < bytes) {
3047 spin_unlock(&ctl->tree_lock);
3051 spin_lock(&cluster->lock);
3053 /* someone already found a cluster, hooray */
3054 if (cluster->block_group) {
3059 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3062 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3064 cont1_bytes, min_bytes);
3066 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3067 offset, bytes + empty_size,
3068 cont1_bytes, min_bytes);
3070 /* Clear our temporary list */
3071 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3072 list_del_init(&entry->list);
3075 atomic_inc(&block_group->count);
3076 list_add_tail(&cluster->block_group_list,
3077 &block_group->cluster_list);
3078 cluster->block_group = block_group;
3080 trace_btrfs_failed_cluster_setup(block_group);
3083 spin_unlock(&cluster->lock);
3084 spin_unlock(&ctl->tree_lock);
3090 * simple code to zero out a cluster
3092 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3094 spin_lock_init(&cluster->lock);
3095 spin_lock_init(&cluster->refill_lock);
3096 cluster->root = RB_ROOT;
3097 cluster->max_size = 0;
3098 cluster->fragmented = false;
3099 INIT_LIST_HEAD(&cluster->block_group_list);
3100 cluster->block_group = NULL;
3103 static int do_trimming(struct btrfs_block_group_cache *block_group,
3104 u64 *total_trimmed, u64 start, u64 bytes,
3105 u64 reserved_start, u64 reserved_bytes,
3106 struct btrfs_trim_range *trim_entry)
3108 struct btrfs_space_info *space_info = block_group->space_info;
3109 struct btrfs_fs_info *fs_info = block_group->fs_info;
3110 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3115 spin_lock(&space_info->lock);
3116 spin_lock(&block_group->lock);
3117 if (!block_group->ro) {
3118 block_group->reserved += reserved_bytes;
3119 space_info->bytes_reserved += reserved_bytes;
3122 spin_unlock(&block_group->lock);
3123 spin_unlock(&space_info->lock);
3125 ret = btrfs_discard_extent(fs_info->extent_root,
3126 start, bytes, &trimmed);
3128 *total_trimmed += trimmed;
3130 mutex_lock(&ctl->cache_writeout_mutex);
3131 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3132 list_del(&trim_entry->list);
3133 mutex_unlock(&ctl->cache_writeout_mutex);
3136 spin_lock(&space_info->lock);
3137 spin_lock(&block_group->lock);
3138 if (block_group->ro)
3139 space_info->bytes_readonly += reserved_bytes;
3140 block_group->reserved -= reserved_bytes;
3141 space_info->bytes_reserved -= reserved_bytes;
3142 spin_unlock(&space_info->lock);
3143 spin_unlock(&block_group->lock);
3149 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3150 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3152 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3153 struct btrfs_free_space *entry;
3154 struct rb_node *node;
3160 while (start < end) {
3161 struct btrfs_trim_range trim_entry;
3163 mutex_lock(&ctl->cache_writeout_mutex);
3164 spin_lock(&ctl->tree_lock);
3166 if (ctl->free_space < minlen) {
3167 spin_unlock(&ctl->tree_lock);
3168 mutex_unlock(&ctl->cache_writeout_mutex);
3172 entry = tree_search_offset(ctl, start, 0, 1);
3174 spin_unlock(&ctl->tree_lock);
3175 mutex_unlock(&ctl->cache_writeout_mutex);
3180 while (entry->bitmap) {
3181 node = rb_next(&entry->offset_index);
3183 spin_unlock(&ctl->tree_lock);
3184 mutex_unlock(&ctl->cache_writeout_mutex);
3187 entry = rb_entry(node, struct btrfs_free_space,
3191 if (entry->offset >= end) {
3192 spin_unlock(&ctl->tree_lock);
3193 mutex_unlock(&ctl->cache_writeout_mutex);
3197 extent_start = entry->offset;
3198 extent_bytes = entry->bytes;
3199 start = max(start, extent_start);
3200 bytes = min(extent_start + extent_bytes, end) - start;
3201 if (bytes < minlen) {
3202 spin_unlock(&ctl->tree_lock);
3203 mutex_unlock(&ctl->cache_writeout_mutex);
3207 unlink_free_space(ctl, entry);
3208 kmem_cache_free(btrfs_free_space_cachep, entry);
3210 spin_unlock(&ctl->tree_lock);
3211 trim_entry.start = extent_start;
3212 trim_entry.bytes = extent_bytes;
3213 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3214 mutex_unlock(&ctl->cache_writeout_mutex);
3216 ret = do_trimming(block_group, total_trimmed, start, bytes,
3217 extent_start, extent_bytes, &trim_entry);
3223 if (fatal_signal_pending(current)) {
3234 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3235 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3237 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3238 struct btrfs_free_space *entry;
3242 u64 offset = offset_to_bitmap(ctl, start);
3244 while (offset < end) {
3245 bool next_bitmap = false;
3246 struct btrfs_trim_range trim_entry;
3248 mutex_lock(&ctl->cache_writeout_mutex);
3249 spin_lock(&ctl->tree_lock);
3251 if (ctl->free_space < minlen) {
3252 spin_unlock(&ctl->tree_lock);
3253 mutex_unlock(&ctl->cache_writeout_mutex);
3257 entry = tree_search_offset(ctl, offset, 1, 0);
3259 spin_unlock(&ctl->tree_lock);
3260 mutex_unlock(&ctl->cache_writeout_mutex);
3266 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3267 if (ret2 || start >= end) {
3268 spin_unlock(&ctl->tree_lock);
3269 mutex_unlock(&ctl->cache_writeout_mutex);
3274 bytes = min(bytes, end - start);
3275 if (bytes < minlen) {
3276 spin_unlock(&ctl->tree_lock);
3277 mutex_unlock(&ctl->cache_writeout_mutex);
3281 bitmap_clear_bits(ctl, entry, start, bytes);
3282 if (entry->bytes == 0)
3283 free_bitmap(ctl, entry);
3285 spin_unlock(&ctl->tree_lock);
3286 trim_entry.start = start;
3287 trim_entry.bytes = bytes;
3288 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3289 mutex_unlock(&ctl->cache_writeout_mutex);
3291 ret = do_trimming(block_group, total_trimmed, start, bytes,
3292 start, bytes, &trim_entry);
3297 offset += BITS_PER_BITMAP * ctl->unit;
3300 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3301 offset += BITS_PER_BITMAP * ctl->unit;
3304 if (fatal_signal_pending(current)) {
3315 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3317 atomic_inc(&cache->trimming);
3320 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3322 struct extent_map_tree *em_tree;
3323 struct extent_map *em;
3326 spin_lock(&block_group->lock);
3327 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3328 block_group->removed);
3329 spin_unlock(&block_group->lock);
3332 lock_chunks(block_group->fs_info->chunk_root);
3333 em_tree = &block_group->fs_info->mapping_tree.map_tree;
3334 write_lock(&em_tree->lock);
3335 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3337 BUG_ON(!em); /* logic error, can't happen */
3339 * remove_extent_mapping() will delete us from the pinned_chunks
3340 * list, which is protected by the chunk mutex.
3342 remove_extent_mapping(em_tree, em);
3343 write_unlock(&em_tree->lock);
3344 unlock_chunks(block_group->fs_info->chunk_root);
3346 /* once for us and once for the tree */
3347 free_extent_map(em);
3348 free_extent_map(em);
3351 * We've left one free space entry and other tasks trimming
3352 * this block group have left 1 entry each one. Free them.
3354 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3358 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3359 u64 *trimmed, u64 start, u64 end, u64 minlen)
3365 spin_lock(&block_group->lock);
3366 if (block_group->removed) {
3367 spin_unlock(&block_group->lock);
3370 btrfs_get_block_group_trimming(block_group);
3371 spin_unlock(&block_group->lock);
3373 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3377 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3379 btrfs_put_block_group_trimming(block_group);
3384 * Find the left-most item in the cache tree, and then return the
3385 * smallest inode number in the item.
3387 * Note: the returned inode number may not be the smallest one in
3388 * the tree, if the left-most item is a bitmap.
3390 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3392 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3393 struct btrfs_free_space *entry = NULL;
3396 spin_lock(&ctl->tree_lock);
3398 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3401 entry = rb_entry(rb_first(&ctl->free_space_offset),
3402 struct btrfs_free_space, offset_index);
3404 if (!entry->bitmap) {
3405 ino = entry->offset;
3407 unlink_free_space(ctl, entry);
3411 kmem_cache_free(btrfs_free_space_cachep, entry);
3413 link_free_space(ctl, entry);
3419 ret = search_bitmap(ctl, entry, &offset, &count, true);
3420 /* Logic error; Should be empty if it can't find anything */
3424 bitmap_clear_bits(ctl, entry, offset, 1);
3425 if (entry->bytes == 0)
3426 free_bitmap(ctl, entry);
3429 spin_unlock(&ctl->tree_lock);
3434 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3435 struct btrfs_path *path)
3437 struct inode *inode = NULL;
3439 spin_lock(&root->ino_cache_lock);
3440 if (root->ino_cache_inode)
3441 inode = igrab(root->ino_cache_inode);
3442 spin_unlock(&root->ino_cache_lock);
3446 inode = __lookup_free_space_inode(root, path, 0);
3450 spin_lock(&root->ino_cache_lock);
3451 if (!btrfs_fs_closing(root->fs_info))
3452 root->ino_cache_inode = igrab(inode);
3453 spin_unlock(&root->ino_cache_lock);
3458 int create_free_ino_inode(struct btrfs_root *root,
3459 struct btrfs_trans_handle *trans,
3460 struct btrfs_path *path)
3462 return __create_free_space_inode(root, trans, path,
3463 BTRFS_FREE_INO_OBJECTID, 0);
3466 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3468 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3469 struct btrfs_path *path;
3470 struct inode *inode;
3472 u64 root_gen = btrfs_root_generation(&root->root_item);
3474 if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
3478 * If we're unmounting then just return, since this does a search on the
3479 * normal root and not the commit root and we could deadlock.
3481 if (btrfs_fs_closing(fs_info))
3484 path = btrfs_alloc_path();
3488 inode = lookup_free_ino_inode(root, path);
3492 if (root_gen != BTRFS_I(inode)->generation)
3495 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3499 "failed to load free ino cache for root %llu",
3500 root->root_key.objectid);
3504 btrfs_free_path(path);
3508 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3509 struct btrfs_trans_handle *trans,
3510 struct btrfs_path *path,
3511 struct inode *inode)
3513 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3515 struct btrfs_io_ctl io_ctl;
3516 bool release_metadata = true;
3518 if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
3521 memset(&io_ctl, 0, sizeof(io_ctl));
3522 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl,
3526 * At this point writepages() didn't error out, so our metadata
3527 * reservation is released when the writeback finishes, at
3528 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3529 * with or without an error.
3531 release_metadata = false;
3532 ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0);
3536 if (release_metadata)
3537 btrfs_delalloc_release_metadata(inode, inode->i_size);
3539 btrfs_err(root->fs_info,
3540 "failed to write free ino cache for root %llu",
3541 root->root_key.objectid);
3548 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3550 * Use this if you need to make a bitmap or extent entry specifically, it
3551 * doesn't do any of the merging that add_free_space does, this acts a lot like
3552 * how the free space cache loading stuff works, so you can get really weird
3555 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3556 u64 offset, u64 bytes, bool bitmap)
3558 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3559 struct btrfs_free_space *info = NULL, *bitmap_info;
3566 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3572 spin_lock(&ctl->tree_lock);
3573 info->offset = offset;
3574 info->bytes = bytes;
3575 info->max_extent_size = 0;
3576 ret = link_free_space(ctl, info);
3577 spin_unlock(&ctl->tree_lock);
3579 kmem_cache_free(btrfs_free_space_cachep, info);
3584 map = kzalloc(PAGE_SIZE, GFP_NOFS);
3586 kmem_cache_free(btrfs_free_space_cachep, info);
3591 spin_lock(&ctl->tree_lock);
3592 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3597 add_new_bitmap(ctl, info, offset);
3602 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3604 bytes -= bytes_added;
3605 offset += bytes_added;
3606 spin_unlock(&ctl->tree_lock);
3612 kmem_cache_free(btrfs_free_space_cachep, info);
3619 * Checks to see if the given range is in the free space cache. This is really
3620 * just used to check the absence of space, so if there is free space in the
3621 * range at all we will return 1.
3623 int test_check_exists(struct btrfs_block_group_cache *cache,
3624 u64 offset, u64 bytes)
3626 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3627 struct btrfs_free_space *info;
3630 spin_lock(&ctl->tree_lock);
3631 info = tree_search_offset(ctl, offset, 0, 0);
3633 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3641 u64 bit_off, bit_bytes;
3643 struct btrfs_free_space *tmp;
3646 bit_bytes = ctl->unit;
3647 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3649 if (bit_off == offset) {
3652 } else if (bit_off > offset &&
3653 offset + bytes > bit_off) {
3659 n = rb_prev(&info->offset_index);
3661 tmp = rb_entry(n, struct btrfs_free_space,
3663 if (tmp->offset + tmp->bytes < offset)
3665 if (offset + bytes < tmp->offset) {
3666 n = rb_prev(&tmp->offset_index);
3673 n = rb_next(&info->offset_index);
3675 tmp = rb_entry(n, struct btrfs_free_space,
3677 if (offset + bytes < tmp->offset)
3679 if (tmp->offset + tmp->bytes < offset) {
3680 n = rb_next(&tmp->offset_index);
3691 if (info->offset == offset) {
3696 if (offset > info->offset && offset < info->offset + info->bytes)
3699 spin_unlock(&ctl->tree_lock);
3702 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */