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_CACHE_SIZE * 8)
33 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
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);
79 return ERR_PTR(-ENOENT);
82 if (is_bad_inode(inode)) {
84 return ERR_PTR(-ENOENT);
87 mapping_set_gfp_mask(inode->i_mapping,
88 mapping_gfp_constraint(inode->i_mapping,
89 ~(__GFP_FS | __GFP_HIGHMEM)));
94 struct inode *lookup_free_space_inode(struct btrfs_root *root,
95 struct btrfs_block_group_cache
96 *block_group, struct btrfs_path *path)
98 struct inode *inode = NULL;
99 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
101 spin_lock(&block_group->lock);
102 if (block_group->inode)
103 inode = igrab(block_group->inode);
104 spin_unlock(&block_group->lock);
108 inode = __lookup_free_space_inode(root, path,
109 block_group->key.objectid);
113 spin_lock(&block_group->lock);
114 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
115 btrfs_info(root->fs_info,
116 "Old style space inode found, converting.");
117 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
118 BTRFS_INODE_NODATACOW;
119 block_group->disk_cache_state = BTRFS_DC_CLEAR;
122 if (!block_group->iref) {
123 block_group->inode = igrab(inode);
124 block_group->iref = 1;
126 spin_unlock(&block_group->lock);
131 static int __create_free_space_inode(struct btrfs_root *root,
132 struct btrfs_trans_handle *trans,
133 struct btrfs_path *path,
136 struct btrfs_key key;
137 struct btrfs_disk_key disk_key;
138 struct btrfs_free_space_header *header;
139 struct btrfs_inode_item *inode_item;
140 struct extent_buffer *leaf;
141 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
144 ret = btrfs_insert_empty_inode(trans, root, path, ino);
148 /* We inline crc's for the free disk space cache */
149 if (ino != BTRFS_FREE_INO_OBJECTID)
150 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
152 leaf = path->nodes[0];
153 inode_item = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_inode_item);
155 btrfs_item_key(leaf, &disk_key, path->slots[0]);
156 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
157 sizeof(*inode_item));
158 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
159 btrfs_set_inode_size(leaf, inode_item, 0);
160 btrfs_set_inode_nbytes(leaf, inode_item, 0);
161 btrfs_set_inode_uid(leaf, inode_item, 0);
162 btrfs_set_inode_gid(leaf, inode_item, 0);
163 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
164 btrfs_set_inode_flags(leaf, inode_item, flags);
165 btrfs_set_inode_nlink(leaf, inode_item, 1);
166 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
167 btrfs_set_inode_block_group(leaf, inode_item, offset);
168 btrfs_mark_buffer_dirty(leaf);
169 btrfs_release_path(path);
171 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 sizeof(struct btrfs_free_space_header));
177 btrfs_release_path(path);
181 leaf = path->nodes[0];
182 header = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_free_space_header);
184 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
185 btrfs_set_free_space_key(leaf, header, &disk_key);
186 btrfs_mark_buffer_dirty(leaf);
187 btrfs_release_path(path);
192 int create_free_space_inode(struct btrfs_root *root,
193 struct btrfs_trans_handle *trans,
194 struct btrfs_block_group_cache *block_group,
195 struct btrfs_path *path)
200 ret = btrfs_find_free_objectid(root, &ino);
204 return __create_free_space_inode(root, trans, path, ino,
205 block_group->key.objectid);
208 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
209 struct btrfs_block_rsv *rsv)
214 /* 1 for slack space, 1 for updating the inode */
215 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
216 btrfs_calc_trans_metadata_size(root, 1);
218 spin_lock(&rsv->lock);
219 if (rsv->reserved < needed_bytes)
223 spin_unlock(&rsv->lock);
227 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
228 struct btrfs_trans_handle *trans,
229 struct btrfs_block_group_cache *block_group,
233 struct btrfs_path *path = btrfs_alloc_path();
243 mutex_lock(&trans->transaction->cache_write_mutex);
244 if (!list_empty(&block_group->io_list)) {
245 list_del_init(&block_group->io_list);
247 btrfs_wait_cache_io(root, trans, block_group,
248 &block_group->io_ctl, path,
249 block_group->key.objectid);
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);
261 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, root, ret);
288 static int 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_CACHE_SHIFT;
300 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
307 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
308 struct btrfs_root *root, int write)
313 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
315 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
318 /* Make sure we can fit our crcs into the first page */
319 if (write && check_crcs &&
320 (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
323 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
325 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
329 io_ctl->num_pages = num_pages;
331 io_ctl->check_crcs = check_crcs;
332 io_ctl->inode = inode;
337 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
339 kfree(io_ctl->pages);
340 io_ctl->pages = NULL;
343 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
351 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
353 ASSERT(io_ctl->index < io_ctl->num_pages);
354 io_ctl->page = io_ctl->pages[io_ctl->index++];
355 io_ctl->cur = page_address(io_ctl->page);
356 io_ctl->orig = io_ctl->cur;
357 io_ctl->size = PAGE_CACHE_SIZE;
359 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
362 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
366 io_ctl_unmap_page(io_ctl);
368 for (i = 0; i < io_ctl->num_pages; i++) {
369 if (io_ctl->pages[i]) {
370 ClearPageChecked(io_ctl->pages[i]);
371 unlock_page(io_ctl->pages[i]);
372 page_cache_release(io_ctl->pages[i]);
377 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
381 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
384 for (i = 0; i < io_ctl->num_pages; i++) {
385 page = find_or_create_page(inode->i_mapping, i, mask);
387 io_ctl_drop_pages(io_ctl);
390 io_ctl->pages[i] = page;
391 if (uptodate && !PageUptodate(page)) {
392 btrfs_readpage(NULL, page);
394 if (page->mapping != inode->i_mapping) {
395 btrfs_err(BTRFS_I(inode)->root->fs_info,
396 "free space cache page truncated");
397 io_ctl_drop_pages(io_ctl);
400 if (!PageUptodate(page)) {
401 btrfs_err(BTRFS_I(inode)->root->fs_info,
402 "error reading free space cache");
403 io_ctl_drop_pages(io_ctl);
409 for (i = 0; i < io_ctl->num_pages; i++) {
410 clear_page_dirty_for_io(io_ctl->pages[i]);
411 set_page_extent_mapped(io_ctl->pages[i]);
417 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
421 io_ctl_map_page(io_ctl, 1);
424 * Skip the csum areas. If we don't check crcs then we just have a
425 * 64bit chunk at the front of the first page.
427 if (io_ctl->check_crcs) {
428 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
429 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
431 io_ctl->cur += sizeof(u64);
432 io_ctl->size -= sizeof(u64) * 2;
436 *val = cpu_to_le64(generation);
437 io_ctl->cur += sizeof(u64);
440 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
445 * Skip the crc area. If we don't check crcs then we just have a 64bit
446 * chunk at the front of the first page.
448 if (io_ctl->check_crcs) {
449 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
450 io_ctl->size -= sizeof(u64) +
451 (sizeof(u32) * io_ctl->num_pages);
453 io_ctl->cur += sizeof(u64);
454 io_ctl->size -= sizeof(u64) * 2;
458 if (le64_to_cpu(*gen) != generation) {
459 btrfs_err_rl(io_ctl->root->fs_info,
460 "space cache generation (%llu) does not match inode (%llu)",
462 io_ctl_unmap_page(io_ctl);
465 io_ctl->cur += sizeof(u64);
469 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
475 if (!io_ctl->check_crcs) {
476 io_ctl_unmap_page(io_ctl);
481 offset = sizeof(u32) * io_ctl->num_pages;
483 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
484 PAGE_CACHE_SIZE - offset);
485 btrfs_csum_final(crc, (char *)&crc);
486 io_ctl_unmap_page(io_ctl);
487 tmp = page_address(io_ctl->pages[0]);
492 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
498 if (!io_ctl->check_crcs) {
499 io_ctl_map_page(io_ctl, 0);
504 offset = sizeof(u32) * io_ctl->num_pages;
506 tmp = page_address(io_ctl->pages[0]);
510 io_ctl_map_page(io_ctl, 0);
511 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
512 PAGE_CACHE_SIZE - offset);
513 btrfs_csum_final(crc, (char *)&crc);
515 btrfs_err_rl(io_ctl->root->fs_info,
516 "csum mismatch on free space cache");
517 io_ctl_unmap_page(io_ctl);
524 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
527 struct btrfs_free_space_entry *entry;
533 entry->offset = cpu_to_le64(offset);
534 entry->bytes = cpu_to_le64(bytes);
535 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
536 BTRFS_FREE_SPACE_EXTENT;
537 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
538 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
540 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
543 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
545 /* No more pages to map */
546 if (io_ctl->index >= io_ctl->num_pages)
549 /* map the next page */
550 io_ctl_map_page(io_ctl, 1);
554 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
560 * If we aren't at the start of the current page, unmap this one and
561 * map the next one if there is any left.
563 if (io_ctl->cur != io_ctl->orig) {
564 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
565 if (io_ctl->index >= io_ctl->num_pages)
567 io_ctl_map_page(io_ctl, 0);
570 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
571 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
572 if (io_ctl->index < io_ctl->num_pages)
573 io_ctl_map_page(io_ctl, 0);
577 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
580 * If we're not on the boundary we know we've modified the page and we
581 * need to crc the page.
583 if (io_ctl->cur != io_ctl->orig)
584 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
586 io_ctl_unmap_page(io_ctl);
588 while (io_ctl->index < io_ctl->num_pages) {
589 io_ctl_map_page(io_ctl, 1);
590 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
594 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
595 struct btrfs_free_space *entry, u8 *type)
597 struct btrfs_free_space_entry *e;
601 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
607 entry->offset = le64_to_cpu(e->offset);
608 entry->bytes = le64_to_cpu(e->bytes);
610 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
611 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
613 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
616 io_ctl_unmap_page(io_ctl);
621 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
622 struct btrfs_free_space *entry)
626 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
630 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
631 io_ctl_unmap_page(io_ctl);
637 * Since we attach pinned extents after the fact we can have contiguous sections
638 * of free space that are split up in entries. This poses a problem with the
639 * tree logging stuff since it could have allocated across what appears to be 2
640 * entries since we would have merged the entries when adding the pinned extents
641 * back to the free space cache. So run through the space cache that we just
642 * loaded and merge contiguous entries. This will make the log replay stuff not
643 * blow up and it will make for nicer allocator behavior.
645 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
647 struct btrfs_free_space *e, *prev = NULL;
651 spin_lock(&ctl->tree_lock);
652 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
653 e = rb_entry(n, struct btrfs_free_space, offset_index);
656 if (e->bitmap || prev->bitmap)
658 if (prev->offset + prev->bytes == e->offset) {
659 unlink_free_space(ctl, prev);
660 unlink_free_space(ctl, e);
661 prev->bytes += e->bytes;
662 kmem_cache_free(btrfs_free_space_cachep, e);
663 link_free_space(ctl, prev);
665 spin_unlock(&ctl->tree_lock);
671 spin_unlock(&ctl->tree_lock);
674 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
675 struct btrfs_free_space_ctl *ctl,
676 struct btrfs_path *path, u64 offset)
678 struct btrfs_free_space_header *header;
679 struct extent_buffer *leaf;
680 struct btrfs_io_ctl io_ctl;
681 struct btrfs_key key;
682 struct btrfs_free_space *e, *n;
690 /* Nothing in the space cache, goodbye */
691 if (!i_size_read(inode))
694 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
698 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
702 btrfs_release_path(path);
708 leaf = path->nodes[0];
709 header = btrfs_item_ptr(leaf, path->slots[0],
710 struct btrfs_free_space_header);
711 num_entries = btrfs_free_space_entries(leaf, header);
712 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
713 generation = btrfs_free_space_generation(leaf, header);
714 btrfs_release_path(path);
716 if (!BTRFS_I(inode)->generation) {
717 btrfs_info(root->fs_info,
718 "The free space cache file (%llu) is invalid. skip it\n",
723 if (BTRFS_I(inode)->generation != generation) {
724 btrfs_err(root->fs_info,
725 "free space inode generation (%llu) "
726 "did not match free space cache generation (%llu)",
727 BTRFS_I(inode)->generation, generation);
734 ret = io_ctl_init(&io_ctl, inode, root, 0);
738 ret = readahead_cache(inode);
742 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
746 ret = io_ctl_check_crc(&io_ctl, 0);
750 ret = io_ctl_check_generation(&io_ctl, generation);
754 while (num_entries) {
755 e = kmem_cache_zalloc(btrfs_free_space_cachep,
762 ret = io_ctl_read_entry(&io_ctl, e, &type);
764 kmem_cache_free(btrfs_free_space_cachep, e);
770 kmem_cache_free(btrfs_free_space_cachep, e);
774 if (type == BTRFS_FREE_SPACE_EXTENT) {
775 spin_lock(&ctl->tree_lock);
776 ret = link_free_space(ctl, e);
777 spin_unlock(&ctl->tree_lock);
779 btrfs_err(root->fs_info,
780 "Duplicate entries in free space cache, dumping");
781 kmem_cache_free(btrfs_free_space_cachep, e);
787 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
791 btrfs_free_space_cachep, e);
794 spin_lock(&ctl->tree_lock);
795 ret = link_free_space(ctl, e);
796 ctl->total_bitmaps++;
797 ctl->op->recalc_thresholds(ctl);
798 spin_unlock(&ctl->tree_lock);
800 btrfs_err(root->fs_info,
801 "Duplicate entries in free space cache, dumping");
802 kmem_cache_free(btrfs_free_space_cachep, e);
805 list_add_tail(&e->list, &bitmaps);
811 io_ctl_unmap_page(&io_ctl);
814 * We add the bitmaps at the end of the entries in order that
815 * the bitmap entries are added to the cache.
817 list_for_each_entry_safe(e, n, &bitmaps, list) {
818 list_del_init(&e->list);
819 ret = io_ctl_read_bitmap(&io_ctl, e);
824 io_ctl_drop_pages(&io_ctl);
825 merge_space_tree(ctl);
828 io_ctl_free(&io_ctl);
831 io_ctl_drop_pages(&io_ctl);
832 __btrfs_remove_free_space_cache(ctl);
836 int load_free_space_cache(struct btrfs_fs_info *fs_info,
837 struct btrfs_block_group_cache *block_group)
839 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
840 struct btrfs_root *root = fs_info->tree_root;
842 struct btrfs_path *path;
845 u64 used = btrfs_block_group_used(&block_group->item);
848 * If this block group has been marked to be cleared for one reason or
849 * another then we can't trust the on disk cache, so just return.
851 spin_lock(&block_group->lock);
852 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
853 spin_unlock(&block_group->lock);
856 spin_unlock(&block_group->lock);
858 path = btrfs_alloc_path();
861 path->search_commit_root = 1;
862 path->skip_locking = 1;
864 inode = lookup_free_space_inode(root, block_group, path);
866 btrfs_free_path(path);
870 /* We may have converted the inode and made the cache invalid. */
871 spin_lock(&block_group->lock);
872 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
873 spin_unlock(&block_group->lock);
874 btrfs_free_path(path);
877 spin_unlock(&block_group->lock);
879 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
880 path, block_group->key.objectid);
881 btrfs_free_path(path);
885 spin_lock(&ctl->tree_lock);
886 matched = (ctl->free_space == (block_group->key.offset - used -
887 block_group->bytes_super));
888 spin_unlock(&ctl->tree_lock);
891 __btrfs_remove_free_space_cache(ctl);
892 btrfs_warn(fs_info, "block group %llu has wrong amount of free space",
893 block_group->key.objectid);
898 /* This cache is bogus, make sure it gets cleared */
899 spin_lock(&block_group->lock);
900 block_group->disk_cache_state = BTRFS_DC_CLEAR;
901 spin_unlock(&block_group->lock);
904 btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuilding it now",
905 block_group->key.objectid);
912 static noinline_for_stack
913 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
914 struct btrfs_free_space_ctl *ctl,
915 struct btrfs_block_group_cache *block_group,
916 int *entries, int *bitmaps,
917 struct list_head *bitmap_list)
920 struct btrfs_free_cluster *cluster = NULL;
921 struct btrfs_free_cluster *cluster_locked = NULL;
922 struct rb_node *node = rb_first(&ctl->free_space_offset);
923 struct btrfs_trim_range *trim_entry;
925 /* Get the cluster for this block_group if it exists */
926 if (block_group && !list_empty(&block_group->cluster_list)) {
927 cluster = list_entry(block_group->cluster_list.next,
928 struct btrfs_free_cluster,
932 if (!node && cluster) {
933 cluster_locked = cluster;
934 spin_lock(&cluster_locked->lock);
935 node = rb_first(&cluster->root);
939 /* Write out the extent entries */
941 struct btrfs_free_space *e;
943 e = rb_entry(node, struct btrfs_free_space, offset_index);
946 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
952 list_add_tail(&e->list, bitmap_list);
955 node = rb_next(node);
956 if (!node && cluster) {
957 node = rb_first(&cluster->root);
958 cluster_locked = cluster;
959 spin_lock(&cluster_locked->lock);
963 if (cluster_locked) {
964 spin_unlock(&cluster_locked->lock);
965 cluster_locked = NULL;
969 * Make sure we don't miss any range that was removed from our rbtree
970 * because trimming is running. Otherwise after a umount+mount (or crash
971 * after committing the transaction) we would leak free space and get
972 * an inconsistent free space cache report from fsck.
974 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
975 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
976 trim_entry->bytes, NULL);
985 spin_unlock(&cluster_locked->lock);
989 static noinline_for_stack int
990 update_cache_item(struct btrfs_trans_handle *trans,
991 struct btrfs_root *root,
993 struct btrfs_path *path, u64 offset,
994 int entries, int bitmaps)
996 struct btrfs_key key;
997 struct btrfs_free_space_header *header;
998 struct extent_buffer *leaf;
1001 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1002 key.offset = offset;
1005 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1007 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1008 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1012 leaf = path->nodes[0];
1014 struct btrfs_key found_key;
1015 ASSERT(path->slots[0]);
1017 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1018 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1019 found_key.offset != offset) {
1020 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1022 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1024 btrfs_release_path(path);
1029 BTRFS_I(inode)->generation = trans->transid;
1030 header = btrfs_item_ptr(leaf, path->slots[0],
1031 struct btrfs_free_space_header);
1032 btrfs_set_free_space_entries(leaf, header, entries);
1033 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1034 btrfs_set_free_space_generation(leaf, header, trans->transid);
1035 btrfs_mark_buffer_dirty(leaf);
1036 btrfs_release_path(path);
1044 static noinline_for_stack int
1045 write_pinned_extent_entries(struct btrfs_root *root,
1046 struct btrfs_block_group_cache *block_group,
1047 struct btrfs_io_ctl *io_ctl,
1050 u64 start, extent_start, extent_end, len;
1051 struct extent_io_tree *unpin = NULL;
1058 * We want to add any pinned extents to our free space cache
1059 * so we don't leak the space
1061 * We shouldn't have switched the pinned extents yet so this is the
1064 unpin = root->fs_info->pinned_extents;
1066 start = block_group->key.objectid;
1068 while (start < block_group->key.objectid + block_group->key.offset) {
1069 ret = find_first_extent_bit(unpin, start,
1070 &extent_start, &extent_end,
1071 EXTENT_DIRTY, NULL);
1075 /* This pinned extent is out of our range */
1076 if (extent_start >= block_group->key.objectid +
1077 block_group->key.offset)
1080 extent_start = max(extent_start, start);
1081 extent_end = min(block_group->key.objectid +
1082 block_group->key.offset, extent_end + 1);
1083 len = extent_end - extent_start;
1086 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1096 static noinline_for_stack int
1097 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1099 struct list_head *pos, *n;
1102 /* Write out the bitmaps */
1103 list_for_each_safe(pos, n, bitmap_list) {
1104 struct btrfs_free_space *entry =
1105 list_entry(pos, struct btrfs_free_space, list);
1107 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1110 list_del_init(&entry->list);
1116 static int flush_dirty_cache(struct inode *inode)
1120 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1122 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1123 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1129 static void noinline_for_stack
1130 cleanup_bitmap_list(struct list_head *bitmap_list)
1132 struct list_head *pos, *n;
1134 list_for_each_safe(pos, n, bitmap_list) {
1135 struct btrfs_free_space *entry =
1136 list_entry(pos, struct btrfs_free_space, list);
1137 list_del_init(&entry->list);
1141 static void noinline_for_stack
1142 cleanup_write_cache_enospc(struct inode *inode,
1143 struct btrfs_io_ctl *io_ctl,
1144 struct extent_state **cached_state,
1145 struct list_head *bitmap_list)
1147 io_ctl_drop_pages(io_ctl);
1148 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1149 i_size_read(inode) - 1, cached_state,
1153 int btrfs_wait_cache_io(struct btrfs_root *root,
1154 struct btrfs_trans_handle *trans,
1155 struct btrfs_block_group_cache *block_group,
1156 struct btrfs_io_ctl *io_ctl,
1157 struct btrfs_path *path, u64 offset)
1160 struct inode *inode = io_ctl->inode;
1166 root = root->fs_info->tree_root;
1168 /* Flush the dirty pages in the cache file. */
1169 ret = flush_dirty_cache(inode);
1173 /* Update the cache item to tell everyone this cache file is valid. */
1174 ret = update_cache_item(trans, root, inode, path, offset,
1175 io_ctl->entries, io_ctl->bitmaps);
1177 io_ctl_free(io_ctl);
1179 invalidate_inode_pages2(inode->i_mapping);
1180 BTRFS_I(inode)->generation = 0;
1183 btrfs_err(root->fs_info,
1184 "failed to write free space cache for block group %llu",
1185 block_group->key.objectid);
1189 btrfs_update_inode(trans, root, inode);
1192 /* the dirty list is protected by the dirty_bgs_lock */
1193 spin_lock(&trans->transaction->dirty_bgs_lock);
1195 /* the disk_cache_state is protected by the block group lock */
1196 spin_lock(&block_group->lock);
1199 * only mark this as written if we didn't get put back on
1200 * the dirty list while waiting for IO. Otherwise our
1201 * cache state won't be right, and we won't get written again
1203 if (!ret && list_empty(&block_group->dirty_list))
1204 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1206 block_group->disk_cache_state = BTRFS_DC_ERROR;
1208 spin_unlock(&block_group->lock);
1209 spin_unlock(&trans->transaction->dirty_bgs_lock);
1210 io_ctl->inode = NULL;
1219 * __btrfs_write_out_cache - write out cached info to an inode
1220 * @root - the root the inode belongs to
1221 * @ctl - the free space cache we are going to write out
1222 * @block_group - the block_group for this cache if it belongs to a block_group
1223 * @trans - the trans handle
1224 * @path - the path to use
1225 * @offset - the offset for the key we'll insert
1227 * This function writes out a free space cache struct to disk for quick recovery
1228 * on mount. This will return 0 if it was successful in writing the cache out,
1229 * or an errno if it was not.
1231 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1232 struct btrfs_free_space_ctl *ctl,
1233 struct btrfs_block_group_cache *block_group,
1234 struct btrfs_io_ctl *io_ctl,
1235 struct btrfs_trans_handle *trans,
1236 struct btrfs_path *path, u64 offset)
1238 struct extent_state *cached_state = NULL;
1239 LIST_HEAD(bitmap_list);
1245 if (!i_size_read(inode))
1248 WARN_ON(io_ctl->pages);
1249 ret = io_ctl_init(io_ctl, inode, root, 1);
1253 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1254 down_write(&block_group->data_rwsem);
1255 spin_lock(&block_group->lock);
1256 if (block_group->delalloc_bytes) {
1257 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1258 spin_unlock(&block_group->lock);
1259 up_write(&block_group->data_rwsem);
1260 BTRFS_I(inode)->generation = 0;
1265 spin_unlock(&block_group->lock);
1268 /* Lock all pages first so we can lock the extent safely. */
1269 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1273 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1276 io_ctl_set_generation(io_ctl, trans->transid);
1278 mutex_lock(&ctl->cache_writeout_mutex);
1279 /* Write out the extent entries in the free space cache */
1280 spin_lock(&ctl->tree_lock);
1281 ret = write_cache_extent_entries(io_ctl, ctl,
1282 block_group, &entries, &bitmaps,
1285 goto out_nospc_locked;
1288 * Some spaces that are freed in the current transaction are pinned,
1289 * they will be added into free space cache after the transaction is
1290 * committed, we shouldn't lose them.
1292 * If this changes while we are working we'll get added back to
1293 * the dirty list and redo it. No locking needed
1295 ret = write_pinned_extent_entries(root, block_group, io_ctl, &entries);
1297 goto out_nospc_locked;
1300 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1301 * locked while doing it because a concurrent trim can be manipulating
1302 * or freeing the bitmap.
1304 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1305 spin_unlock(&ctl->tree_lock);
1306 mutex_unlock(&ctl->cache_writeout_mutex);
1310 /* Zero out the rest of the pages just to make sure */
1311 io_ctl_zero_remaining_pages(io_ctl);
1313 /* Everything is written out, now we dirty the pages in the file. */
1314 ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages,
1315 0, i_size_read(inode), &cached_state);
1319 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1320 up_write(&block_group->data_rwsem);
1322 * Release the pages and unlock the extent, we will flush
1325 io_ctl_drop_pages(io_ctl);
1327 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1328 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1331 * at this point the pages are under IO and we're happy,
1332 * The caller is responsible for waiting on them and updating the
1333 * the cache and the inode
1335 io_ctl->entries = entries;
1336 io_ctl->bitmaps = bitmaps;
1338 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1345 io_ctl->inode = NULL;
1346 io_ctl_free(io_ctl);
1348 invalidate_inode_pages2(inode->i_mapping);
1349 BTRFS_I(inode)->generation = 0;
1351 btrfs_update_inode(trans, root, inode);
1357 cleanup_bitmap_list(&bitmap_list);
1358 spin_unlock(&ctl->tree_lock);
1359 mutex_unlock(&ctl->cache_writeout_mutex);
1362 cleanup_write_cache_enospc(inode, io_ctl, &cached_state, &bitmap_list);
1365 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1366 up_write(&block_group->data_rwsem);
1371 int btrfs_write_out_cache(struct btrfs_root *root,
1372 struct btrfs_trans_handle *trans,
1373 struct btrfs_block_group_cache *block_group,
1374 struct btrfs_path *path)
1376 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1377 struct inode *inode;
1380 root = root->fs_info->tree_root;
1382 spin_lock(&block_group->lock);
1383 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1384 spin_unlock(&block_group->lock);
1387 spin_unlock(&block_group->lock);
1389 inode = lookup_free_space_inode(root, block_group, path);
1393 ret = __btrfs_write_out_cache(root, inode, ctl, block_group,
1394 &block_group->io_ctl, trans,
1395 path, block_group->key.objectid);
1398 btrfs_err(root->fs_info,
1399 "failed to write free space cache for block group %llu",
1400 block_group->key.objectid);
1402 spin_lock(&block_group->lock);
1403 block_group->disk_cache_state = BTRFS_DC_ERROR;
1404 spin_unlock(&block_group->lock);
1406 block_group->io_ctl.inode = NULL;
1411 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1412 * to wait for IO and put the inode
1418 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1421 ASSERT(offset >= bitmap_start);
1422 offset -= bitmap_start;
1423 return (unsigned long)(div_u64(offset, unit));
1426 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1428 return (unsigned long)(div_u64(bytes, unit));
1431 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1435 u32 bytes_per_bitmap;
1437 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1438 bitmap_start = offset - ctl->start;
1439 bitmap_start = div_u64(bitmap_start, bytes_per_bitmap);
1440 bitmap_start *= bytes_per_bitmap;
1441 bitmap_start += ctl->start;
1443 return bitmap_start;
1446 static int tree_insert_offset(struct rb_root *root, u64 offset,
1447 struct rb_node *node, int bitmap)
1449 struct rb_node **p = &root->rb_node;
1450 struct rb_node *parent = NULL;
1451 struct btrfs_free_space *info;
1455 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1457 if (offset < info->offset) {
1459 } else if (offset > info->offset) {
1460 p = &(*p)->rb_right;
1463 * we could have a bitmap entry and an extent entry
1464 * share the same offset. If this is the case, we want
1465 * the extent entry to always be found first if we do a
1466 * linear search through the tree, since we want to have
1467 * the quickest allocation time, and allocating from an
1468 * extent is faster than allocating from a bitmap. So
1469 * if we're inserting a bitmap and we find an entry at
1470 * this offset, we want to go right, or after this entry
1471 * logically. If we are inserting an extent and we've
1472 * found a bitmap, we want to go left, or before
1480 p = &(*p)->rb_right;
1482 if (!info->bitmap) {
1491 rb_link_node(node, parent, p);
1492 rb_insert_color(node, root);
1498 * searches the tree for the given offset.
1500 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1501 * want a section that has at least bytes size and comes at or after the given
1504 static struct btrfs_free_space *
1505 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1506 u64 offset, int bitmap_only, int fuzzy)
1508 struct rb_node *n = ctl->free_space_offset.rb_node;
1509 struct btrfs_free_space *entry, *prev = NULL;
1511 /* find entry that is closest to the 'offset' */
1518 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1521 if (offset < entry->offset)
1523 else if (offset > entry->offset)
1536 * bitmap entry and extent entry may share same offset,
1537 * in that case, bitmap entry comes after extent entry.
1542 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1543 if (entry->offset != offset)
1546 WARN_ON(!entry->bitmap);
1549 if (entry->bitmap) {
1551 * if previous extent entry covers the offset,
1552 * we should return it instead of the bitmap entry
1554 n = rb_prev(&entry->offset_index);
1556 prev = rb_entry(n, struct btrfs_free_space,
1558 if (!prev->bitmap &&
1559 prev->offset + prev->bytes > offset)
1569 /* find last entry before the 'offset' */
1571 if (entry->offset > offset) {
1572 n = rb_prev(&entry->offset_index);
1574 entry = rb_entry(n, struct btrfs_free_space,
1576 ASSERT(entry->offset <= offset);
1585 if (entry->bitmap) {
1586 n = rb_prev(&entry->offset_index);
1588 prev = rb_entry(n, struct btrfs_free_space,
1590 if (!prev->bitmap &&
1591 prev->offset + prev->bytes > offset)
1594 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1596 } else if (entry->offset + entry->bytes > offset)
1603 if (entry->bitmap) {
1604 if (entry->offset + BITS_PER_BITMAP *
1608 if (entry->offset + entry->bytes > offset)
1612 n = rb_next(&entry->offset_index);
1615 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1621 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1622 struct btrfs_free_space *info)
1624 rb_erase(&info->offset_index, &ctl->free_space_offset);
1625 ctl->free_extents--;
1628 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1629 struct btrfs_free_space *info)
1631 __unlink_free_space(ctl, info);
1632 ctl->free_space -= info->bytes;
1635 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1636 struct btrfs_free_space *info)
1640 ASSERT(info->bytes || info->bitmap);
1641 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1642 &info->offset_index, (info->bitmap != NULL));
1646 ctl->free_space += info->bytes;
1647 ctl->free_extents++;
1651 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1653 struct btrfs_block_group_cache *block_group = ctl->private;
1657 u64 size = block_group->key.offset;
1658 u32 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1659 u32 max_bitmaps = div_u64(size + bytes_per_bg - 1, bytes_per_bg);
1661 max_bitmaps = max_t(u32, max_bitmaps, 1);
1663 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1666 * The goal is to keep the total amount of memory used per 1gb of space
1667 * at or below 32k, so we need to adjust how much memory we allow to be
1668 * used by extent based free space tracking
1670 if (size < 1024 * 1024 * 1024)
1671 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1673 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1674 div_u64(size, 1024 * 1024 * 1024);
1677 * we want to account for 1 more bitmap than what we have so we can make
1678 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1679 * we add more bitmaps.
1681 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1683 if (bitmap_bytes >= max_bytes) {
1684 ctl->extents_thresh = 0;
1689 * we want the extent entry threshold to always be at most 1/2 the max
1690 * bytes we can have, or whatever is less than that.
1692 extent_bytes = max_bytes - bitmap_bytes;
1693 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1695 ctl->extents_thresh =
1696 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1699 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1700 struct btrfs_free_space *info,
1701 u64 offset, u64 bytes)
1703 unsigned long start, count;
1705 start = offset_to_bit(info->offset, ctl->unit, offset);
1706 count = bytes_to_bits(bytes, ctl->unit);
1707 ASSERT(start + count <= BITS_PER_BITMAP);
1709 bitmap_clear(info->bitmap, start, count);
1711 info->bytes -= bytes;
1712 if (info->max_extent_size > ctl->unit)
1713 info->max_extent_size = 0;
1716 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1717 struct btrfs_free_space *info, u64 offset,
1720 __bitmap_clear_bits(ctl, info, offset, bytes);
1721 ctl->free_space -= bytes;
1724 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1725 struct btrfs_free_space *info, u64 offset,
1728 unsigned long start, count;
1730 start = offset_to_bit(info->offset, ctl->unit, offset);
1731 count = bytes_to_bits(bytes, ctl->unit);
1732 ASSERT(start + count <= BITS_PER_BITMAP);
1734 bitmap_set(info->bitmap, start, count);
1736 info->bytes += bytes;
1737 ctl->free_space += bytes;
1741 * If we can not find suitable extent, we will use bytes to record
1742 * the size of the max extent.
1744 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1745 struct btrfs_free_space *bitmap_info, u64 *offset,
1746 u64 *bytes, bool for_alloc)
1748 unsigned long found_bits = 0;
1749 unsigned long max_bits = 0;
1750 unsigned long bits, i;
1751 unsigned long next_zero;
1752 unsigned long extent_bits;
1755 * Skip searching the bitmap if we don't have a contiguous section that
1756 * is large enough for this allocation.
1759 bitmap_info->max_extent_size &&
1760 bitmap_info->max_extent_size < *bytes) {
1761 *bytes = bitmap_info->max_extent_size;
1765 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1766 max_t(u64, *offset, bitmap_info->offset));
1767 bits = bytes_to_bits(*bytes, ctl->unit);
1769 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1770 if (for_alloc && bits == 1) {
1774 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1775 BITS_PER_BITMAP, i);
1776 extent_bits = next_zero - i;
1777 if (extent_bits >= bits) {
1778 found_bits = extent_bits;
1780 } else if (extent_bits > max_bits) {
1781 max_bits = extent_bits;
1787 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1788 *bytes = (u64)(found_bits) * ctl->unit;
1792 *bytes = (u64)(max_bits) * ctl->unit;
1793 bitmap_info->max_extent_size = *bytes;
1797 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1800 return entry->max_extent_size;
1801 return entry->bytes;
1804 /* Cache the size of the max extent in bytes */
1805 static struct btrfs_free_space *
1806 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1807 unsigned long align, u64 *max_extent_size)
1809 struct btrfs_free_space *entry;
1810 struct rb_node *node;
1815 if (!ctl->free_space_offset.rb_node)
1818 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1822 for (node = &entry->offset_index; node; node = rb_next(node)) {
1823 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1824 if (entry->bytes < *bytes) {
1825 *max_extent_size = max(get_max_extent_size(entry),
1830 /* make sure the space returned is big enough
1831 * to match our requested alignment
1833 if (*bytes >= align) {
1834 tmp = entry->offset - ctl->start + align - 1;
1835 tmp = div64_u64(tmp, align);
1836 tmp = tmp * align + ctl->start;
1837 align_off = tmp - entry->offset;
1840 tmp = entry->offset;
1843 if (entry->bytes < *bytes + align_off) {
1844 *max_extent_size = max(get_max_extent_size(entry),
1849 if (entry->bitmap) {
1852 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1859 max(get_max_extent_size(entry),
1866 *bytes = entry->bytes - align_off;
1873 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1874 struct btrfs_free_space *info, u64 offset)
1876 info->offset = offset_to_bitmap(ctl, offset);
1878 INIT_LIST_HEAD(&info->list);
1879 link_free_space(ctl, info);
1880 ctl->total_bitmaps++;
1882 ctl->op->recalc_thresholds(ctl);
1885 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1886 struct btrfs_free_space *bitmap_info)
1888 unlink_free_space(ctl, bitmap_info);
1889 kfree(bitmap_info->bitmap);
1890 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1891 ctl->total_bitmaps--;
1892 ctl->op->recalc_thresholds(ctl);
1895 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1896 struct btrfs_free_space *bitmap_info,
1897 u64 *offset, u64 *bytes)
1900 u64 search_start, search_bytes;
1904 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1907 * We need to search for bits in this bitmap. We could only cover some
1908 * of the extent in this bitmap thanks to how we add space, so we need
1909 * to search for as much as it as we can and clear that amount, and then
1910 * go searching for the next bit.
1912 search_start = *offset;
1913 search_bytes = ctl->unit;
1914 search_bytes = min(search_bytes, end - search_start + 1);
1915 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1917 if (ret < 0 || search_start != *offset)
1920 /* We may have found more bits than what we need */
1921 search_bytes = min(search_bytes, *bytes);
1923 /* Cannot clear past the end of the bitmap */
1924 search_bytes = min(search_bytes, end - search_start + 1);
1926 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1927 *offset += search_bytes;
1928 *bytes -= search_bytes;
1931 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1932 if (!bitmap_info->bytes)
1933 free_bitmap(ctl, bitmap_info);
1936 * no entry after this bitmap, but we still have bytes to
1937 * remove, so something has gone wrong.
1942 bitmap_info = rb_entry(next, struct btrfs_free_space,
1946 * if the next entry isn't a bitmap we need to return to let the
1947 * extent stuff do its work.
1949 if (!bitmap_info->bitmap)
1953 * Ok the next item is a bitmap, but it may not actually hold
1954 * the information for the rest of this free space stuff, so
1955 * look for it, and if we don't find it return so we can try
1956 * everything over again.
1958 search_start = *offset;
1959 search_bytes = ctl->unit;
1960 ret = search_bitmap(ctl, bitmap_info, &search_start,
1961 &search_bytes, false);
1962 if (ret < 0 || search_start != *offset)
1966 } else if (!bitmap_info->bytes)
1967 free_bitmap(ctl, bitmap_info);
1972 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1973 struct btrfs_free_space *info, u64 offset,
1976 u64 bytes_to_set = 0;
1979 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1981 bytes_to_set = min(end - offset, bytes);
1983 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1986 * We set some bytes, we have no idea what the max extent size is
1989 info->max_extent_size = 0;
1991 return bytes_to_set;
1995 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1996 struct btrfs_free_space *info)
1998 struct btrfs_block_group_cache *block_group = ctl->private;
1999 bool forced = false;
2001 #ifdef CONFIG_BTRFS_DEBUG
2002 if (btrfs_should_fragment_free_space(block_group->fs_info->extent_root,
2008 * If we are below the extents threshold then we can add this as an
2009 * extent, and don't have to deal with the bitmap
2011 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2013 * If this block group has some small extents we don't want to
2014 * use up all of our free slots in the cache with them, we want
2015 * to reserve them to larger extents, however if we have plent
2016 * of cache left then go ahead an dadd them, no sense in adding
2017 * the overhead of a bitmap if we don't have to.
2019 if (info->bytes <= block_group->sectorsize * 4) {
2020 if (ctl->free_extents * 2 <= ctl->extents_thresh)
2028 * The original block groups from mkfs can be really small, like 8
2029 * megabytes, so don't bother with a bitmap for those entries. However
2030 * some block groups can be smaller than what a bitmap would cover but
2031 * are still large enough that they could overflow the 32k memory limit,
2032 * so allow those block groups to still be allowed to have a bitmap
2035 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2041 static struct btrfs_free_space_op free_space_op = {
2042 .recalc_thresholds = recalculate_thresholds,
2043 .use_bitmap = use_bitmap,
2046 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2047 struct btrfs_free_space *info)
2049 struct btrfs_free_space *bitmap_info;
2050 struct btrfs_block_group_cache *block_group = NULL;
2052 u64 bytes, offset, bytes_added;
2055 bytes = info->bytes;
2056 offset = info->offset;
2058 if (!ctl->op->use_bitmap(ctl, info))
2061 if (ctl->op == &free_space_op)
2062 block_group = ctl->private;
2065 * Since we link bitmaps right into the cluster we need to see if we
2066 * have a cluster here, and if so and it has our bitmap we need to add
2067 * the free space to that bitmap.
2069 if (block_group && !list_empty(&block_group->cluster_list)) {
2070 struct btrfs_free_cluster *cluster;
2071 struct rb_node *node;
2072 struct btrfs_free_space *entry;
2074 cluster = list_entry(block_group->cluster_list.next,
2075 struct btrfs_free_cluster,
2077 spin_lock(&cluster->lock);
2078 node = rb_first(&cluster->root);
2080 spin_unlock(&cluster->lock);
2081 goto no_cluster_bitmap;
2084 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2085 if (!entry->bitmap) {
2086 spin_unlock(&cluster->lock);
2087 goto no_cluster_bitmap;
2090 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2091 bytes_added = add_bytes_to_bitmap(ctl, entry,
2093 bytes -= bytes_added;
2094 offset += bytes_added;
2096 spin_unlock(&cluster->lock);
2104 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2111 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2112 bytes -= bytes_added;
2113 offset += bytes_added;
2123 if (info && info->bitmap) {
2124 add_new_bitmap(ctl, info, offset);
2129 spin_unlock(&ctl->tree_lock);
2131 /* no pre-allocated info, allocate a new one */
2133 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2136 spin_lock(&ctl->tree_lock);
2142 /* allocate the bitmap */
2143 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
2144 spin_lock(&ctl->tree_lock);
2145 if (!info->bitmap) {
2155 kfree(info->bitmap);
2156 kmem_cache_free(btrfs_free_space_cachep, info);
2162 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2163 struct btrfs_free_space *info, bool update_stat)
2165 struct btrfs_free_space *left_info = NULL;
2166 struct btrfs_free_space *right_info;
2167 bool merged = false;
2168 u64 offset = info->offset;
2169 u64 bytes = info->bytes;
2172 * first we want to see if there is free space adjacent to the range we
2173 * are adding, if there is remove that struct and add a new one to
2174 * cover the entire range
2176 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2177 if (right_info && rb_prev(&right_info->offset_index))
2178 left_info = rb_entry(rb_prev(&right_info->offset_index),
2179 struct btrfs_free_space, offset_index);
2180 else if (!right_info)
2181 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2183 if (right_info && !right_info->bitmap) {
2185 unlink_free_space(ctl, right_info);
2187 __unlink_free_space(ctl, right_info);
2188 info->bytes += right_info->bytes;
2189 kmem_cache_free(btrfs_free_space_cachep, right_info);
2193 if (left_info && !left_info->bitmap &&
2194 left_info->offset + left_info->bytes == offset) {
2196 unlink_free_space(ctl, left_info);
2198 __unlink_free_space(ctl, left_info);
2199 info->offset = left_info->offset;
2200 info->bytes += left_info->bytes;
2201 kmem_cache_free(btrfs_free_space_cachep, left_info);
2208 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2209 struct btrfs_free_space *info,
2212 struct btrfs_free_space *bitmap;
2215 const u64 end = info->offset + info->bytes;
2216 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2219 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2223 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2224 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2227 bytes = (j - i) * ctl->unit;
2228 info->bytes += bytes;
2231 bitmap_clear_bits(ctl, bitmap, end, bytes);
2233 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2236 free_bitmap(ctl, bitmap);
2241 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2242 struct btrfs_free_space *info,
2245 struct btrfs_free_space *bitmap;
2249 unsigned long prev_j;
2252 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2253 /* If we're on a boundary, try the previous logical bitmap. */
2254 if (bitmap_offset == info->offset) {
2255 if (info->offset == 0)
2257 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2260 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2264 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2266 prev_j = (unsigned long)-1;
2267 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2275 if (prev_j == (unsigned long)-1)
2276 bytes = (i + 1) * ctl->unit;
2278 bytes = (i - prev_j) * ctl->unit;
2280 info->offset -= bytes;
2281 info->bytes += bytes;
2284 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2286 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2289 free_bitmap(ctl, bitmap);
2295 * We prefer always to allocate from extent entries, both for clustered and
2296 * non-clustered allocation requests. So when attempting to add a new extent
2297 * entry, try to see if there's adjacent free space in bitmap entries, and if
2298 * there is, migrate that space from the bitmaps to the extent.
2299 * Like this we get better chances of satisfying space allocation requests
2300 * because we attempt to satisfy them based on a single cache entry, and never
2301 * on 2 or more entries - even if the entries represent a contiguous free space
2302 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2305 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2306 struct btrfs_free_space *info,
2310 * Only work with disconnected entries, as we can change their offset,
2311 * and must be extent entries.
2313 ASSERT(!info->bitmap);
2314 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2316 if (ctl->total_bitmaps > 0) {
2318 bool stole_front = false;
2320 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2321 if (ctl->total_bitmaps > 0)
2322 stole_front = steal_from_bitmap_to_front(ctl, info,
2325 if (stole_end || stole_front)
2326 try_merge_free_space(ctl, info, update_stat);
2330 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
2331 u64 offset, u64 bytes)
2333 struct btrfs_free_space *info;
2336 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2340 info->offset = offset;
2341 info->bytes = bytes;
2342 RB_CLEAR_NODE(&info->offset_index);
2344 spin_lock(&ctl->tree_lock);
2346 if (try_merge_free_space(ctl, info, true))
2350 * There was no extent directly to the left or right of this new
2351 * extent then we know we're going to have to allocate a new extent, so
2352 * before we do that see if we need to drop this into a bitmap
2354 ret = insert_into_bitmap(ctl, info);
2363 * Only steal free space from adjacent bitmaps if we're sure we're not
2364 * going to add the new free space to existing bitmap entries - because
2365 * that would mean unnecessary work that would be reverted. Therefore
2366 * attempt to steal space from bitmaps if we're adding an extent entry.
2368 steal_from_bitmap(ctl, info, true);
2370 ret = link_free_space(ctl, info);
2372 kmem_cache_free(btrfs_free_space_cachep, info);
2374 spin_unlock(&ctl->tree_lock);
2377 printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
2378 ASSERT(ret != -EEXIST);
2384 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2385 u64 offset, u64 bytes)
2387 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2388 struct btrfs_free_space *info;
2390 bool re_search = false;
2392 spin_lock(&ctl->tree_lock);
2399 info = tree_search_offset(ctl, offset, 0, 0);
2402 * oops didn't find an extent that matched the space we wanted
2403 * to remove, look for a bitmap instead
2405 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2409 * If we found a partial bit of our free space in a
2410 * bitmap but then couldn't find the other part this may
2411 * be a problem, so WARN about it.
2419 if (!info->bitmap) {
2420 unlink_free_space(ctl, info);
2421 if (offset == info->offset) {
2422 u64 to_free = min(bytes, info->bytes);
2424 info->bytes -= to_free;
2425 info->offset += to_free;
2427 ret = link_free_space(ctl, info);
2430 kmem_cache_free(btrfs_free_space_cachep, info);
2437 u64 old_end = info->bytes + info->offset;
2439 info->bytes = offset - info->offset;
2440 ret = link_free_space(ctl, info);
2445 /* Not enough bytes in this entry to satisfy us */
2446 if (old_end < offset + bytes) {
2447 bytes -= old_end - offset;
2450 } else if (old_end == offset + bytes) {
2454 spin_unlock(&ctl->tree_lock);
2456 ret = btrfs_add_free_space(block_group, offset + bytes,
2457 old_end - (offset + bytes));
2463 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2464 if (ret == -EAGAIN) {
2469 spin_unlock(&ctl->tree_lock);
2474 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2477 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2478 struct btrfs_free_space *info;
2482 spin_lock(&ctl->tree_lock);
2483 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2484 info = rb_entry(n, struct btrfs_free_space, offset_index);
2485 if (info->bytes >= bytes && !block_group->ro)
2487 btrfs_crit(block_group->fs_info,
2488 "entry offset %llu, bytes %llu, bitmap %s",
2489 info->offset, info->bytes,
2490 (info->bitmap) ? "yes" : "no");
2492 spin_unlock(&ctl->tree_lock);
2493 btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2494 list_empty(&block_group->cluster_list) ? "no" : "yes");
2495 btrfs_info(block_group->fs_info,
2496 "%d blocks of free space at or bigger than bytes is", count);
2499 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2501 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2503 spin_lock_init(&ctl->tree_lock);
2504 ctl->unit = block_group->sectorsize;
2505 ctl->start = block_group->key.objectid;
2506 ctl->private = block_group;
2507 ctl->op = &free_space_op;
2508 INIT_LIST_HEAD(&ctl->trimming_ranges);
2509 mutex_init(&ctl->cache_writeout_mutex);
2512 * we only want to have 32k of ram per block group for keeping
2513 * track of free space, and if we pass 1/2 of that we want to
2514 * start converting things over to using bitmaps
2516 ctl->extents_thresh = ((1024 * 32) / 2) /
2517 sizeof(struct btrfs_free_space);
2521 * for a given cluster, put all of its extents back into the free
2522 * space cache. If the block group passed doesn't match the block group
2523 * pointed to by the cluster, someone else raced in and freed the
2524 * cluster already. In that case, we just return without changing anything
2527 __btrfs_return_cluster_to_free_space(
2528 struct btrfs_block_group_cache *block_group,
2529 struct btrfs_free_cluster *cluster)
2531 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2532 struct btrfs_free_space *entry;
2533 struct rb_node *node;
2535 spin_lock(&cluster->lock);
2536 if (cluster->block_group != block_group)
2539 cluster->block_group = NULL;
2540 cluster->window_start = 0;
2541 list_del_init(&cluster->block_group_list);
2543 node = rb_first(&cluster->root);
2547 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2548 node = rb_next(&entry->offset_index);
2549 rb_erase(&entry->offset_index, &cluster->root);
2550 RB_CLEAR_NODE(&entry->offset_index);
2552 bitmap = (entry->bitmap != NULL);
2554 try_merge_free_space(ctl, entry, false);
2555 steal_from_bitmap(ctl, entry, false);
2557 tree_insert_offset(&ctl->free_space_offset,
2558 entry->offset, &entry->offset_index, bitmap);
2560 cluster->root = RB_ROOT;
2563 spin_unlock(&cluster->lock);
2564 btrfs_put_block_group(block_group);
2568 static void __btrfs_remove_free_space_cache_locked(
2569 struct btrfs_free_space_ctl *ctl)
2571 struct btrfs_free_space *info;
2572 struct rb_node *node;
2574 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2575 info = rb_entry(node, struct btrfs_free_space, offset_index);
2576 if (!info->bitmap) {
2577 unlink_free_space(ctl, info);
2578 kmem_cache_free(btrfs_free_space_cachep, info);
2580 free_bitmap(ctl, info);
2583 cond_resched_lock(&ctl->tree_lock);
2587 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2589 spin_lock(&ctl->tree_lock);
2590 __btrfs_remove_free_space_cache_locked(ctl);
2591 spin_unlock(&ctl->tree_lock);
2594 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2596 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2597 struct btrfs_free_cluster *cluster;
2598 struct list_head *head;
2600 spin_lock(&ctl->tree_lock);
2601 while ((head = block_group->cluster_list.next) !=
2602 &block_group->cluster_list) {
2603 cluster = list_entry(head, struct btrfs_free_cluster,
2606 WARN_ON(cluster->block_group != block_group);
2607 __btrfs_return_cluster_to_free_space(block_group, cluster);
2609 cond_resched_lock(&ctl->tree_lock);
2611 __btrfs_remove_free_space_cache_locked(ctl);
2612 spin_unlock(&ctl->tree_lock);
2616 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2617 u64 offset, u64 bytes, u64 empty_size,
2618 u64 *max_extent_size)
2620 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2621 struct btrfs_free_space *entry = NULL;
2622 u64 bytes_search = bytes + empty_size;
2625 u64 align_gap_len = 0;
2627 spin_lock(&ctl->tree_lock);
2628 entry = find_free_space(ctl, &offset, &bytes_search,
2629 block_group->full_stripe_len, max_extent_size);
2634 if (entry->bitmap) {
2635 bitmap_clear_bits(ctl, entry, offset, bytes);
2637 free_bitmap(ctl, entry);
2639 unlink_free_space(ctl, entry);
2640 align_gap_len = offset - entry->offset;
2641 align_gap = entry->offset;
2643 entry->offset = offset + bytes;
2644 WARN_ON(entry->bytes < bytes + align_gap_len);
2646 entry->bytes -= bytes + align_gap_len;
2648 kmem_cache_free(btrfs_free_space_cachep, entry);
2650 link_free_space(ctl, entry);
2653 spin_unlock(&ctl->tree_lock);
2656 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2661 * given a cluster, put all of its extents back into the free space
2662 * cache. If a block group is passed, this function will only free
2663 * a cluster that belongs to the passed block group.
2665 * Otherwise, it'll get a reference on the block group pointed to by the
2666 * cluster and remove the cluster from it.
2668 int btrfs_return_cluster_to_free_space(
2669 struct btrfs_block_group_cache *block_group,
2670 struct btrfs_free_cluster *cluster)
2672 struct btrfs_free_space_ctl *ctl;
2675 /* first, get a safe pointer to the block group */
2676 spin_lock(&cluster->lock);
2678 block_group = cluster->block_group;
2680 spin_unlock(&cluster->lock);
2683 } else if (cluster->block_group != block_group) {
2684 /* someone else has already freed it don't redo their work */
2685 spin_unlock(&cluster->lock);
2688 atomic_inc(&block_group->count);
2689 spin_unlock(&cluster->lock);
2691 ctl = block_group->free_space_ctl;
2693 /* now return any extents the cluster had on it */
2694 spin_lock(&ctl->tree_lock);
2695 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2696 spin_unlock(&ctl->tree_lock);
2698 /* finally drop our ref */
2699 btrfs_put_block_group(block_group);
2703 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2704 struct btrfs_free_cluster *cluster,
2705 struct btrfs_free_space *entry,
2706 u64 bytes, u64 min_start,
2707 u64 *max_extent_size)
2709 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2711 u64 search_start = cluster->window_start;
2712 u64 search_bytes = bytes;
2715 search_start = min_start;
2716 search_bytes = bytes;
2718 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2720 *max_extent_size = max(get_max_extent_size(entry),
2726 __bitmap_clear_bits(ctl, entry, ret, bytes);
2732 * given a cluster, try to allocate 'bytes' from it, returns 0
2733 * if it couldn't find anything suitably large, or a logical disk offset
2734 * if things worked out
2736 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2737 struct btrfs_free_cluster *cluster, u64 bytes,
2738 u64 min_start, u64 *max_extent_size)
2740 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2741 struct btrfs_free_space *entry = NULL;
2742 struct rb_node *node;
2745 spin_lock(&cluster->lock);
2746 if (bytes > cluster->max_size)
2749 if (cluster->block_group != block_group)
2752 node = rb_first(&cluster->root);
2756 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2758 if (entry->bytes < bytes)
2759 *max_extent_size = max(get_max_extent_size(entry),
2762 if (entry->bytes < bytes ||
2763 (!entry->bitmap && entry->offset < min_start)) {
2764 node = rb_next(&entry->offset_index);
2767 entry = rb_entry(node, struct btrfs_free_space,
2772 if (entry->bitmap) {
2773 ret = btrfs_alloc_from_bitmap(block_group,
2774 cluster, entry, bytes,
2775 cluster->window_start,
2778 node = rb_next(&entry->offset_index);
2781 entry = rb_entry(node, struct btrfs_free_space,
2785 cluster->window_start += bytes;
2787 ret = entry->offset;
2789 entry->offset += bytes;
2790 entry->bytes -= bytes;
2793 if (entry->bytes == 0)
2794 rb_erase(&entry->offset_index, &cluster->root);
2798 spin_unlock(&cluster->lock);
2803 spin_lock(&ctl->tree_lock);
2805 ctl->free_space -= bytes;
2806 if (entry->bytes == 0) {
2807 ctl->free_extents--;
2808 if (entry->bitmap) {
2809 kfree(entry->bitmap);
2810 ctl->total_bitmaps--;
2811 ctl->op->recalc_thresholds(ctl);
2813 kmem_cache_free(btrfs_free_space_cachep, entry);
2816 spin_unlock(&ctl->tree_lock);
2821 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2822 struct btrfs_free_space *entry,
2823 struct btrfs_free_cluster *cluster,
2824 u64 offset, u64 bytes,
2825 u64 cont1_bytes, u64 min_bytes)
2827 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2828 unsigned long next_zero;
2830 unsigned long want_bits;
2831 unsigned long min_bits;
2832 unsigned long found_bits;
2833 unsigned long max_bits = 0;
2834 unsigned long start = 0;
2835 unsigned long total_found = 0;
2838 i = offset_to_bit(entry->offset, ctl->unit,
2839 max_t(u64, offset, entry->offset));
2840 want_bits = bytes_to_bits(bytes, ctl->unit);
2841 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2844 * Don't bother looking for a cluster in this bitmap if it's heavily
2847 if (entry->max_extent_size &&
2848 entry->max_extent_size < cont1_bytes)
2852 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2853 next_zero = find_next_zero_bit(entry->bitmap,
2854 BITS_PER_BITMAP, i);
2855 if (next_zero - i >= min_bits) {
2856 found_bits = next_zero - i;
2857 if (found_bits > max_bits)
2858 max_bits = found_bits;
2861 if (next_zero - i > max_bits)
2862 max_bits = next_zero - i;
2867 entry->max_extent_size = (u64)max_bits * ctl->unit;
2873 cluster->max_size = 0;
2876 total_found += found_bits;
2878 if (cluster->max_size < found_bits * ctl->unit)
2879 cluster->max_size = found_bits * ctl->unit;
2881 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2886 cluster->window_start = start * ctl->unit + entry->offset;
2887 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2888 ret = tree_insert_offset(&cluster->root, entry->offset,
2889 &entry->offset_index, 1);
2890 ASSERT(!ret); /* -EEXIST; Logic error */
2892 trace_btrfs_setup_cluster(block_group, cluster,
2893 total_found * ctl->unit, 1);
2898 * This searches the block group for just extents to fill the cluster with.
2899 * Try to find a cluster with at least bytes total bytes, at least one
2900 * extent of cont1_bytes, and other clusters of at least min_bytes.
2903 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2904 struct btrfs_free_cluster *cluster,
2905 struct list_head *bitmaps, u64 offset, u64 bytes,
2906 u64 cont1_bytes, u64 min_bytes)
2908 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2909 struct btrfs_free_space *first = NULL;
2910 struct btrfs_free_space *entry = NULL;
2911 struct btrfs_free_space *last;
2912 struct rb_node *node;
2917 entry = tree_search_offset(ctl, offset, 0, 1);
2922 * We don't want bitmaps, so just move along until we find a normal
2925 while (entry->bitmap || entry->bytes < min_bytes) {
2926 if (entry->bitmap && list_empty(&entry->list))
2927 list_add_tail(&entry->list, bitmaps);
2928 node = rb_next(&entry->offset_index);
2931 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2934 window_free = entry->bytes;
2935 max_extent = entry->bytes;
2939 for (node = rb_next(&entry->offset_index); node;
2940 node = rb_next(&entry->offset_index)) {
2941 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2943 if (entry->bitmap) {
2944 if (list_empty(&entry->list))
2945 list_add_tail(&entry->list, bitmaps);
2949 if (entry->bytes < min_bytes)
2953 window_free += entry->bytes;
2954 if (entry->bytes > max_extent)
2955 max_extent = entry->bytes;
2958 if (window_free < bytes || max_extent < cont1_bytes)
2961 cluster->window_start = first->offset;
2963 node = &first->offset_index;
2966 * now we've found our entries, pull them out of the free space
2967 * cache and put them into the cluster rbtree
2972 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2973 node = rb_next(&entry->offset_index);
2974 if (entry->bitmap || entry->bytes < min_bytes)
2977 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2978 ret = tree_insert_offset(&cluster->root, entry->offset,
2979 &entry->offset_index, 0);
2980 total_size += entry->bytes;
2981 ASSERT(!ret); /* -EEXIST; Logic error */
2982 } while (node && entry != last);
2984 cluster->max_size = max_extent;
2985 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2990 * This specifically looks for bitmaps that may work in the cluster, we assume
2991 * that we have already failed to find extents that will work.
2994 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2995 struct btrfs_free_cluster *cluster,
2996 struct list_head *bitmaps, u64 offset, u64 bytes,
2997 u64 cont1_bytes, u64 min_bytes)
2999 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3000 struct btrfs_free_space *entry = NULL;
3002 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3004 if (ctl->total_bitmaps == 0)
3008 * The bitmap that covers offset won't be in the list unless offset
3009 * is just its start offset.
3011 if (!list_empty(bitmaps))
3012 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3014 if (!entry || entry->offset != bitmap_offset) {
3015 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3016 if (entry && list_empty(&entry->list))
3017 list_add(&entry->list, bitmaps);
3020 list_for_each_entry(entry, bitmaps, list) {
3021 if (entry->bytes < bytes)
3023 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3024 bytes, cont1_bytes, min_bytes);
3030 * The bitmaps list has all the bitmaps that record free space
3031 * starting after offset, so no more search is required.
3037 * here we try to find a cluster of blocks in a block group. The goal
3038 * is to find at least bytes+empty_size.
3039 * We might not find them all in one contiguous area.
3041 * returns zero and sets up cluster if things worked out, otherwise
3042 * it returns -enospc
3044 int btrfs_find_space_cluster(struct btrfs_root *root,
3045 struct btrfs_block_group_cache *block_group,
3046 struct btrfs_free_cluster *cluster,
3047 u64 offset, u64 bytes, u64 empty_size)
3049 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3050 struct btrfs_free_space *entry, *tmp;
3057 * Choose the minimum extent size we'll require for this
3058 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3059 * For metadata, allow allocates with smaller extents. For
3060 * data, keep it dense.
3062 if (btrfs_test_opt(root, SSD_SPREAD)) {
3063 cont1_bytes = min_bytes = bytes + empty_size;
3064 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3065 cont1_bytes = bytes;
3066 min_bytes = block_group->sectorsize;
3068 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3069 min_bytes = block_group->sectorsize;
3072 spin_lock(&ctl->tree_lock);
3075 * If we know we don't have enough space to make a cluster don't even
3076 * bother doing all the work to try and find one.
3078 if (ctl->free_space < bytes) {
3079 spin_unlock(&ctl->tree_lock);
3083 spin_lock(&cluster->lock);
3085 /* someone already found a cluster, hooray */
3086 if (cluster->block_group) {
3091 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3094 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3096 cont1_bytes, min_bytes);
3098 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3099 offset, bytes + empty_size,
3100 cont1_bytes, min_bytes);
3102 /* Clear our temporary list */
3103 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3104 list_del_init(&entry->list);
3107 atomic_inc(&block_group->count);
3108 list_add_tail(&cluster->block_group_list,
3109 &block_group->cluster_list);
3110 cluster->block_group = block_group;
3112 trace_btrfs_failed_cluster_setup(block_group);
3115 spin_unlock(&cluster->lock);
3116 spin_unlock(&ctl->tree_lock);
3122 * simple code to zero out a cluster
3124 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3126 spin_lock_init(&cluster->lock);
3127 spin_lock_init(&cluster->refill_lock);
3128 cluster->root = RB_ROOT;
3129 cluster->max_size = 0;
3130 cluster->fragmented = false;
3131 INIT_LIST_HEAD(&cluster->block_group_list);
3132 cluster->block_group = NULL;
3135 static int do_trimming(struct btrfs_block_group_cache *block_group,
3136 u64 *total_trimmed, u64 start, u64 bytes,
3137 u64 reserved_start, u64 reserved_bytes,
3138 struct btrfs_trim_range *trim_entry)
3140 struct btrfs_space_info *space_info = block_group->space_info;
3141 struct btrfs_fs_info *fs_info = block_group->fs_info;
3142 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3147 spin_lock(&space_info->lock);
3148 spin_lock(&block_group->lock);
3149 if (!block_group->ro) {
3150 block_group->reserved += reserved_bytes;
3151 space_info->bytes_reserved += reserved_bytes;
3154 spin_unlock(&block_group->lock);
3155 spin_unlock(&space_info->lock);
3157 ret = btrfs_discard_extent(fs_info->extent_root,
3158 start, bytes, &trimmed);
3160 *total_trimmed += trimmed;
3162 mutex_lock(&ctl->cache_writeout_mutex);
3163 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3164 list_del(&trim_entry->list);
3165 mutex_unlock(&ctl->cache_writeout_mutex);
3168 spin_lock(&space_info->lock);
3169 spin_lock(&block_group->lock);
3170 if (block_group->ro)
3171 space_info->bytes_readonly += reserved_bytes;
3172 block_group->reserved -= reserved_bytes;
3173 space_info->bytes_reserved -= reserved_bytes;
3174 spin_unlock(&space_info->lock);
3175 spin_unlock(&block_group->lock);
3181 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3182 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3184 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3185 struct btrfs_free_space *entry;
3186 struct rb_node *node;
3192 while (start < end) {
3193 struct btrfs_trim_range trim_entry;
3195 mutex_lock(&ctl->cache_writeout_mutex);
3196 spin_lock(&ctl->tree_lock);
3198 if (ctl->free_space < minlen) {
3199 spin_unlock(&ctl->tree_lock);
3200 mutex_unlock(&ctl->cache_writeout_mutex);
3204 entry = tree_search_offset(ctl, start, 0, 1);
3206 spin_unlock(&ctl->tree_lock);
3207 mutex_unlock(&ctl->cache_writeout_mutex);
3212 while (entry->bitmap) {
3213 node = rb_next(&entry->offset_index);
3215 spin_unlock(&ctl->tree_lock);
3216 mutex_unlock(&ctl->cache_writeout_mutex);
3219 entry = rb_entry(node, struct btrfs_free_space,
3223 if (entry->offset >= end) {
3224 spin_unlock(&ctl->tree_lock);
3225 mutex_unlock(&ctl->cache_writeout_mutex);
3229 extent_start = entry->offset;
3230 extent_bytes = entry->bytes;
3231 start = max(start, extent_start);
3232 bytes = min(extent_start + extent_bytes, end) - start;
3233 if (bytes < minlen) {
3234 spin_unlock(&ctl->tree_lock);
3235 mutex_unlock(&ctl->cache_writeout_mutex);
3239 unlink_free_space(ctl, entry);
3240 kmem_cache_free(btrfs_free_space_cachep, entry);
3242 spin_unlock(&ctl->tree_lock);
3243 trim_entry.start = extent_start;
3244 trim_entry.bytes = extent_bytes;
3245 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3246 mutex_unlock(&ctl->cache_writeout_mutex);
3248 ret = do_trimming(block_group, total_trimmed, start, bytes,
3249 extent_start, extent_bytes, &trim_entry);
3255 if (fatal_signal_pending(current)) {
3266 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3267 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3269 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3270 struct btrfs_free_space *entry;
3274 u64 offset = offset_to_bitmap(ctl, start);
3276 while (offset < end) {
3277 bool next_bitmap = false;
3278 struct btrfs_trim_range trim_entry;
3280 mutex_lock(&ctl->cache_writeout_mutex);
3281 spin_lock(&ctl->tree_lock);
3283 if (ctl->free_space < minlen) {
3284 spin_unlock(&ctl->tree_lock);
3285 mutex_unlock(&ctl->cache_writeout_mutex);
3289 entry = tree_search_offset(ctl, offset, 1, 0);
3291 spin_unlock(&ctl->tree_lock);
3292 mutex_unlock(&ctl->cache_writeout_mutex);
3298 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3299 if (ret2 || start >= end) {
3300 spin_unlock(&ctl->tree_lock);
3301 mutex_unlock(&ctl->cache_writeout_mutex);
3306 bytes = min(bytes, end - start);
3307 if (bytes < minlen) {
3308 spin_unlock(&ctl->tree_lock);
3309 mutex_unlock(&ctl->cache_writeout_mutex);
3313 bitmap_clear_bits(ctl, entry, start, bytes);
3314 if (entry->bytes == 0)
3315 free_bitmap(ctl, entry);
3317 spin_unlock(&ctl->tree_lock);
3318 trim_entry.start = start;
3319 trim_entry.bytes = bytes;
3320 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3321 mutex_unlock(&ctl->cache_writeout_mutex);
3323 ret = do_trimming(block_group, total_trimmed, start, bytes,
3324 start, bytes, &trim_entry);
3329 offset += BITS_PER_BITMAP * ctl->unit;
3332 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3333 offset += BITS_PER_BITMAP * ctl->unit;
3336 if (fatal_signal_pending(current)) {
3347 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3349 atomic_inc(&cache->trimming);
3352 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3354 struct extent_map_tree *em_tree;
3355 struct extent_map *em;
3358 spin_lock(&block_group->lock);
3359 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3360 block_group->removed);
3361 spin_unlock(&block_group->lock);
3364 lock_chunks(block_group->fs_info->chunk_root);
3365 em_tree = &block_group->fs_info->mapping_tree.map_tree;
3366 write_lock(&em_tree->lock);
3367 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3369 BUG_ON(!em); /* logic error, can't happen */
3371 * remove_extent_mapping() will delete us from the pinned_chunks
3372 * list, which is protected by the chunk mutex.
3374 remove_extent_mapping(em_tree, em);
3375 write_unlock(&em_tree->lock);
3376 unlock_chunks(block_group->fs_info->chunk_root);
3378 /* once for us and once for the tree */
3379 free_extent_map(em);
3380 free_extent_map(em);
3383 * We've left one free space entry and other tasks trimming
3384 * this block group have left 1 entry each one. Free them.
3386 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3390 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3391 u64 *trimmed, u64 start, u64 end, u64 minlen)
3397 spin_lock(&block_group->lock);
3398 if (block_group->removed) {
3399 spin_unlock(&block_group->lock);
3402 btrfs_get_block_group_trimming(block_group);
3403 spin_unlock(&block_group->lock);
3405 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3409 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3411 btrfs_put_block_group_trimming(block_group);
3416 * Find the left-most item in the cache tree, and then return the
3417 * smallest inode number in the item.
3419 * Note: the returned inode number may not be the smallest one in
3420 * the tree, if the left-most item is a bitmap.
3422 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3424 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3425 struct btrfs_free_space *entry = NULL;
3428 spin_lock(&ctl->tree_lock);
3430 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3433 entry = rb_entry(rb_first(&ctl->free_space_offset),
3434 struct btrfs_free_space, offset_index);
3436 if (!entry->bitmap) {
3437 ino = entry->offset;
3439 unlink_free_space(ctl, entry);
3443 kmem_cache_free(btrfs_free_space_cachep, entry);
3445 link_free_space(ctl, entry);
3451 ret = search_bitmap(ctl, entry, &offset, &count, true);
3452 /* Logic error; Should be empty if it can't find anything */
3456 bitmap_clear_bits(ctl, entry, offset, 1);
3457 if (entry->bytes == 0)
3458 free_bitmap(ctl, entry);
3461 spin_unlock(&ctl->tree_lock);
3466 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3467 struct btrfs_path *path)
3469 struct inode *inode = NULL;
3471 spin_lock(&root->ino_cache_lock);
3472 if (root->ino_cache_inode)
3473 inode = igrab(root->ino_cache_inode);
3474 spin_unlock(&root->ino_cache_lock);
3478 inode = __lookup_free_space_inode(root, path, 0);
3482 spin_lock(&root->ino_cache_lock);
3483 if (!btrfs_fs_closing(root->fs_info))
3484 root->ino_cache_inode = igrab(inode);
3485 spin_unlock(&root->ino_cache_lock);
3490 int create_free_ino_inode(struct btrfs_root *root,
3491 struct btrfs_trans_handle *trans,
3492 struct btrfs_path *path)
3494 return __create_free_space_inode(root, trans, path,
3495 BTRFS_FREE_INO_OBJECTID, 0);
3498 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3500 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3501 struct btrfs_path *path;
3502 struct inode *inode;
3504 u64 root_gen = btrfs_root_generation(&root->root_item);
3506 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3510 * If we're unmounting then just return, since this does a search on the
3511 * normal root and not the commit root and we could deadlock.
3513 if (btrfs_fs_closing(fs_info))
3516 path = btrfs_alloc_path();
3520 inode = lookup_free_ino_inode(root, path);
3524 if (root_gen != BTRFS_I(inode)->generation)
3527 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3531 "failed to load free ino cache for root %llu",
3532 root->root_key.objectid);
3536 btrfs_free_path(path);
3540 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3541 struct btrfs_trans_handle *trans,
3542 struct btrfs_path *path,
3543 struct inode *inode)
3545 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3547 struct btrfs_io_ctl io_ctl;
3548 bool release_metadata = true;
3550 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3553 memset(&io_ctl, 0, sizeof(io_ctl));
3554 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl,
3558 * At this point writepages() didn't error out, so our metadata
3559 * reservation is released when the writeback finishes, at
3560 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3561 * with or without an error.
3563 release_metadata = false;
3564 ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0);
3568 if (release_metadata)
3569 btrfs_delalloc_release_metadata(inode, inode->i_size);
3571 btrfs_err(root->fs_info,
3572 "failed to write free ino cache for root %llu",
3573 root->root_key.objectid);
3580 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3582 * Use this if you need to make a bitmap or extent entry specifically, it
3583 * doesn't do any of the merging that add_free_space does, this acts a lot like
3584 * how the free space cache loading stuff works, so you can get really weird
3587 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3588 u64 offset, u64 bytes, bool bitmap)
3590 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3591 struct btrfs_free_space *info = NULL, *bitmap_info;
3598 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3604 spin_lock(&ctl->tree_lock);
3605 info->offset = offset;
3606 info->bytes = bytes;
3607 info->max_extent_size = 0;
3608 ret = link_free_space(ctl, info);
3609 spin_unlock(&ctl->tree_lock);
3611 kmem_cache_free(btrfs_free_space_cachep, info);
3616 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3618 kmem_cache_free(btrfs_free_space_cachep, info);
3623 spin_lock(&ctl->tree_lock);
3624 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3629 add_new_bitmap(ctl, info, offset);
3634 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3636 bytes -= bytes_added;
3637 offset += bytes_added;
3638 spin_unlock(&ctl->tree_lock);
3644 kmem_cache_free(btrfs_free_space_cachep, info);
3651 * Checks to see if the given range is in the free space cache. This is really
3652 * just used to check the absence of space, so if there is free space in the
3653 * range at all we will return 1.
3655 int test_check_exists(struct btrfs_block_group_cache *cache,
3656 u64 offset, u64 bytes)
3658 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3659 struct btrfs_free_space *info;
3662 spin_lock(&ctl->tree_lock);
3663 info = tree_search_offset(ctl, offset, 0, 0);
3665 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3673 u64 bit_off, bit_bytes;
3675 struct btrfs_free_space *tmp;
3678 bit_bytes = ctl->unit;
3679 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3681 if (bit_off == offset) {
3684 } else if (bit_off > offset &&
3685 offset + bytes > bit_off) {
3691 n = rb_prev(&info->offset_index);
3693 tmp = rb_entry(n, struct btrfs_free_space,
3695 if (tmp->offset + tmp->bytes < offset)
3697 if (offset + bytes < tmp->offset) {
3698 n = rb_prev(&info->offset_index);
3705 n = rb_next(&info->offset_index);
3707 tmp = rb_entry(n, struct btrfs_free_space,
3709 if (offset + bytes < tmp->offset)
3711 if (tmp->offset + tmp->bytes < offset) {
3712 n = rb_next(&info->offset_index);
3723 if (info->offset == offset) {
3728 if (offset > info->offset && offset < info->offset + info->bytes)
3731 spin_unlock(&ctl->tree_lock);
3734 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */