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 (!PageUptodate(page)) {
395 btrfs_err(BTRFS_I(inode)->root->fs_info,
396 "error reading free space cache");
397 io_ctl_drop_pages(io_ctl);
403 for (i = 0; i < io_ctl->num_pages; i++) {
404 clear_page_dirty_for_io(io_ctl->pages[i]);
405 set_page_extent_mapped(io_ctl->pages[i]);
411 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
415 io_ctl_map_page(io_ctl, 1);
418 * Skip the csum areas. If we don't check crcs then we just have a
419 * 64bit chunk at the front of the first page.
421 if (io_ctl->check_crcs) {
422 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
423 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
425 io_ctl->cur += sizeof(u64);
426 io_ctl->size -= sizeof(u64) * 2;
430 *val = cpu_to_le64(generation);
431 io_ctl->cur += sizeof(u64);
434 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
439 * Skip the crc area. If we don't check crcs then we just have a 64bit
440 * chunk at the front of the first page.
442 if (io_ctl->check_crcs) {
443 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
444 io_ctl->size -= sizeof(u64) +
445 (sizeof(u32) * io_ctl->num_pages);
447 io_ctl->cur += sizeof(u64);
448 io_ctl->size -= sizeof(u64) * 2;
452 if (le64_to_cpu(*gen) != generation) {
453 btrfs_err_rl(io_ctl->root->fs_info,
454 "space cache generation (%llu) does not match inode (%llu)",
456 io_ctl_unmap_page(io_ctl);
459 io_ctl->cur += sizeof(u64);
463 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
469 if (!io_ctl->check_crcs) {
470 io_ctl_unmap_page(io_ctl);
475 offset = sizeof(u32) * io_ctl->num_pages;
477 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
478 PAGE_CACHE_SIZE - offset);
479 btrfs_csum_final(crc, (char *)&crc);
480 io_ctl_unmap_page(io_ctl);
481 tmp = page_address(io_ctl->pages[0]);
486 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
492 if (!io_ctl->check_crcs) {
493 io_ctl_map_page(io_ctl, 0);
498 offset = sizeof(u32) * io_ctl->num_pages;
500 tmp = page_address(io_ctl->pages[0]);
504 io_ctl_map_page(io_ctl, 0);
505 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
506 PAGE_CACHE_SIZE - offset);
507 btrfs_csum_final(crc, (char *)&crc);
509 btrfs_err_rl(io_ctl->root->fs_info,
510 "csum mismatch on free space cache");
511 io_ctl_unmap_page(io_ctl);
518 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
521 struct btrfs_free_space_entry *entry;
527 entry->offset = cpu_to_le64(offset);
528 entry->bytes = cpu_to_le64(bytes);
529 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
530 BTRFS_FREE_SPACE_EXTENT;
531 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
532 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
534 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
537 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
539 /* No more pages to map */
540 if (io_ctl->index >= io_ctl->num_pages)
543 /* map the next page */
544 io_ctl_map_page(io_ctl, 1);
548 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
554 * If we aren't at the start of the current page, unmap this one and
555 * map the next one if there is any left.
557 if (io_ctl->cur != io_ctl->orig) {
558 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
559 if (io_ctl->index >= io_ctl->num_pages)
561 io_ctl_map_page(io_ctl, 0);
564 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
565 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
566 if (io_ctl->index < io_ctl->num_pages)
567 io_ctl_map_page(io_ctl, 0);
571 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
574 * If we're not on the boundary we know we've modified the page and we
575 * need to crc the page.
577 if (io_ctl->cur != io_ctl->orig)
578 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
580 io_ctl_unmap_page(io_ctl);
582 while (io_ctl->index < io_ctl->num_pages) {
583 io_ctl_map_page(io_ctl, 1);
584 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
588 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
589 struct btrfs_free_space *entry, u8 *type)
591 struct btrfs_free_space_entry *e;
595 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
601 entry->offset = le64_to_cpu(e->offset);
602 entry->bytes = le64_to_cpu(e->bytes);
604 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
605 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
607 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
610 io_ctl_unmap_page(io_ctl);
615 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
616 struct btrfs_free_space *entry)
620 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
624 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
625 io_ctl_unmap_page(io_ctl);
631 * Since we attach pinned extents after the fact we can have contiguous sections
632 * of free space that are split up in entries. This poses a problem with the
633 * tree logging stuff since it could have allocated across what appears to be 2
634 * entries since we would have merged the entries when adding the pinned extents
635 * back to the free space cache. So run through the space cache that we just
636 * loaded and merge contiguous entries. This will make the log replay stuff not
637 * blow up and it will make for nicer allocator behavior.
639 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
641 struct btrfs_free_space *e, *prev = NULL;
645 spin_lock(&ctl->tree_lock);
646 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
647 e = rb_entry(n, struct btrfs_free_space, offset_index);
650 if (e->bitmap || prev->bitmap)
652 if (prev->offset + prev->bytes == e->offset) {
653 unlink_free_space(ctl, prev);
654 unlink_free_space(ctl, e);
655 prev->bytes += e->bytes;
656 kmem_cache_free(btrfs_free_space_cachep, e);
657 link_free_space(ctl, prev);
659 spin_unlock(&ctl->tree_lock);
665 spin_unlock(&ctl->tree_lock);
668 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
669 struct btrfs_free_space_ctl *ctl,
670 struct btrfs_path *path, u64 offset)
672 struct btrfs_free_space_header *header;
673 struct extent_buffer *leaf;
674 struct btrfs_io_ctl io_ctl;
675 struct btrfs_key key;
676 struct btrfs_free_space *e, *n;
684 /* Nothing in the space cache, goodbye */
685 if (!i_size_read(inode))
688 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
692 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
696 btrfs_release_path(path);
702 leaf = path->nodes[0];
703 header = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_free_space_header);
705 num_entries = btrfs_free_space_entries(leaf, header);
706 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
707 generation = btrfs_free_space_generation(leaf, header);
708 btrfs_release_path(path);
710 if (!BTRFS_I(inode)->generation) {
711 btrfs_info(root->fs_info,
712 "The free space cache file (%llu) is invalid. skip it\n",
717 if (BTRFS_I(inode)->generation != generation) {
718 btrfs_err(root->fs_info,
719 "free space inode generation (%llu) "
720 "did not match free space cache generation (%llu)",
721 BTRFS_I(inode)->generation, generation);
728 ret = io_ctl_init(&io_ctl, inode, root, 0);
732 ret = readahead_cache(inode);
736 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
740 ret = io_ctl_check_crc(&io_ctl, 0);
744 ret = io_ctl_check_generation(&io_ctl, generation);
748 while (num_entries) {
749 e = kmem_cache_zalloc(btrfs_free_space_cachep,
754 ret = io_ctl_read_entry(&io_ctl, e, &type);
756 kmem_cache_free(btrfs_free_space_cachep, e);
761 kmem_cache_free(btrfs_free_space_cachep, e);
765 if (type == BTRFS_FREE_SPACE_EXTENT) {
766 spin_lock(&ctl->tree_lock);
767 ret = link_free_space(ctl, e);
768 spin_unlock(&ctl->tree_lock);
770 btrfs_err(root->fs_info,
771 "Duplicate entries in free space cache, dumping");
772 kmem_cache_free(btrfs_free_space_cachep, e);
778 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
781 btrfs_free_space_cachep, e);
784 spin_lock(&ctl->tree_lock);
785 ret = link_free_space(ctl, e);
786 ctl->total_bitmaps++;
787 ctl->op->recalc_thresholds(ctl);
788 spin_unlock(&ctl->tree_lock);
790 btrfs_err(root->fs_info,
791 "Duplicate entries in free space cache, dumping");
792 kmem_cache_free(btrfs_free_space_cachep, e);
795 list_add_tail(&e->list, &bitmaps);
801 io_ctl_unmap_page(&io_ctl);
804 * We add the bitmaps at the end of the entries in order that
805 * the bitmap entries are added to the cache.
807 list_for_each_entry_safe(e, n, &bitmaps, list) {
808 list_del_init(&e->list);
809 ret = io_ctl_read_bitmap(&io_ctl, e);
814 io_ctl_drop_pages(&io_ctl);
815 merge_space_tree(ctl);
818 io_ctl_free(&io_ctl);
821 io_ctl_drop_pages(&io_ctl);
822 __btrfs_remove_free_space_cache(ctl);
826 int load_free_space_cache(struct btrfs_fs_info *fs_info,
827 struct btrfs_block_group_cache *block_group)
829 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
830 struct btrfs_root *root = fs_info->tree_root;
832 struct btrfs_path *path;
835 u64 used = btrfs_block_group_used(&block_group->item);
838 * If this block group has been marked to be cleared for one reason or
839 * another then we can't trust the on disk cache, so just return.
841 spin_lock(&block_group->lock);
842 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
843 spin_unlock(&block_group->lock);
846 spin_unlock(&block_group->lock);
848 path = btrfs_alloc_path();
851 path->search_commit_root = 1;
852 path->skip_locking = 1;
854 inode = lookup_free_space_inode(root, block_group, path);
856 btrfs_free_path(path);
860 /* We may have converted the inode and made the cache invalid. */
861 spin_lock(&block_group->lock);
862 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
863 spin_unlock(&block_group->lock);
864 btrfs_free_path(path);
867 spin_unlock(&block_group->lock);
869 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
870 path, block_group->key.objectid);
871 btrfs_free_path(path);
875 spin_lock(&ctl->tree_lock);
876 matched = (ctl->free_space == (block_group->key.offset - used -
877 block_group->bytes_super));
878 spin_unlock(&ctl->tree_lock);
881 __btrfs_remove_free_space_cache(ctl);
882 btrfs_warn(fs_info, "block group %llu has wrong amount of free space",
883 block_group->key.objectid);
888 /* This cache is bogus, make sure it gets cleared */
889 spin_lock(&block_group->lock);
890 block_group->disk_cache_state = BTRFS_DC_CLEAR;
891 spin_unlock(&block_group->lock);
894 btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuilding it now",
895 block_group->key.objectid);
902 static noinline_for_stack
903 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
904 struct btrfs_free_space_ctl *ctl,
905 struct btrfs_block_group_cache *block_group,
906 int *entries, int *bitmaps,
907 struct list_head *bitmap_list)
910 struct btrfs_free_cluster *cluster = NULL;
911 struct btrfs_free_cluster *cluster_locked = NULL;
912 struct rb_node *node = rb_first(&ctl->free_space_offset);
913 struct btrfs_trim_range *trim_entry;
915 /* Get the cluster for this block_group if it exists */
916 if (block_group && !list_empty(&block_group->cluster_list)) {
917 cluster = list_entry(block_group->cluster_list.next,
918 struct btrfs_free_cluster,
922 if (!node && cluster) {
923 cluster_locked = cluster;
924 spin_lock(&cluster_locked->lock);
925 node = rb_first(&cluster->root);
929 /* Write out the extent entries */
931 struct btrfs_free_space *e;
933 e = rb_entry(node, struct btrfs_free_space, offset_index);
936 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
942 list_add_tail(&e->list, bitmap_list);
945 node = rb_next(node);
946 if (!node && cluster) {
947 node = rb_first(&cluster->root);
948 cluster_locked = cluster;
949 spin_lock(&cluster_locked->lock);
953 if (cluster_locked) {
954 spin_unlock(&cluster_locked->lock);
955 cluster_locked = NULL;
959 * Make sure we don't miss any range that was removed from our rbtree
960 * because trimming is running. Otherwise after a umount+mount (or crash
961 * after committing the transaction) we would leak free space and get
962 * an inconsistent free space cache report from fsck.
964 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
965 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
966 trim_entry->bytes, NULL);
975 spin_unlock(&cluster_locked->lock);
979 static noinline_for_stack int
980 update_cache_item(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
983 struct btrfs_path *path, u64 offset,
984 int entries, int bitmaps)
986 struct btrfs_key key;
987 struct btrfs_free_space_header *header;
988 struct extent_buffer *leaf;
991 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
995 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
997 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
998 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1002 leaf = path->nodes[0];
1004 struct btrfs_key found_key;
1005 ASSERT(path->slots[0]);
1007 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1008 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1009 found_key.offset != offset) {
1010 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1012 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1014 btrfs_release_path(path);
1019 BTRFS_I(inode)->generation = trans->transid;
1020 header = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_free_space_header);
1022 btrfs_set_free_space_entries(leaf, header, entries);
1023 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1024 btrfs_set_free_space_generation(leaf, header, trans->transid);
1025 btrfs_mark_buffer_dirty(leaf);
1026 btrfs_release_path(path);
1034 static noinline_for_stack int
1035 write_pinned_extent_entries(struct btrfs_root *root,
1036 struct btrfs_block_group_cache *block_group,
1037 struct btrfs_io_ctl *io_ctl,
1040 u64 start, extent_start, extent_end, len;
1041 struct extent_io_tree *unpin = NULL;
1048 * We want to add any pinned extents to our free space cache
1049 * so we don't leak the space
1051 * We shouldn't have switched the pinned extents yet so this is the
1054 unpin = root->fs_info->pinned_extents;
1056 start = block_group->key.objectid;
1058 while (start < block_group->key.objectid + block_group->key.offset) {
1059 ret = find_first_extent_bit(unpin, start,
1060 &extent_start, &extent_end,
1061 EXTENT_DIRTY, NULL);
1065 /* This pinned extent is out of our range */
1066 if (extent_start >= block_group->key.objectid +
1067 block_group->key.offset)
1070 extent_start = max(extent_start, start);
1071 extent_end = min(block_group->key.objectid +
1072 block_group->key.offset, extent_end + 1);
1073 len = extent_end - extent_start;
1076 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1086 static noinline_for_stack int
1087 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1089 struct list_head *pos, *n;
1092 /* Write out the bitmaps */
1093 list_for_each_safe(pos, n, bitmap_list) {
1094 struct btrfs_free_space *entry =
1095 list_entry(pos, struct btrfs_free_space, list);
1097 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1100 list_del_init(&entry->list);
1106 static int flush_dirty_cache(struct inode *inode)
1110 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1112 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1113 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1119 static void noinline_for_stack
1120 cleanup_bitmap_list(struct list_head *bitmap_list)
1122 struct list_head *pos, *n;
1124 list_for_each_safe(pos, n, bitmap_list) {
1125 struct btrfs_free_space *entry =
1126 list_entry(pos, struct btrfs_free_space, list);
1127 list_del_init(&entry->list);
1131 static void noinline_for_stack
1132 cleanup_write_cache_enospc(struct inode *inode,
1133 struct btrfs_io_ctl *io_ctl,
1134 struct extent_state **cached_state,
1135 struct list_head *bitmap_list)
1137 io_ctl_drop_pages(io_ctl);
1138 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1139 i_size_read(inode) - 1, cached_state,
1143 int btrfs_wait_cache_io(struct btrfs_root *root,
1144 struct btrfs_trans_handle *trans,
1145 struct btrfs_block_group_cache *block_group,
1146 struct btrfs_io_ctl *io_ctl,
1147 struct btrfs_path *path, u64 offset)
1150 struct inode *inode = io_ctl->inode;
1156 root = root->fs_info->tree_root;
1158 /* Flush the dirty pages in the cache file. */
1159 ret = flush_dirty_cache(inode);
1163 /* Update the cache item to tell everyone this cache file is valid. */
1164 ret = update_cache_item(trans, root, inode, path, offset,
1165 io_ctl->entries, io_ctl->bitmaps);
1167 io_ctl_free(io_ctl);
1169 invalidate_inode_pages2(inode->i_mapping);
1170 BTRFS_I(inode)->generation = 0;
1173 btrfs_err(root->fs_info,
1174 "failed to write free space cache for block group %llu",
1175 block_group->key.objectid);
1179 btrfs_update_inode(trans, root, inode);
1182 /* the dirty list is protected by the dirty_bgs_lock */
1183 spin_lock(&trans->transaction->dirty_bgs_lock);
1185 /* the disk_cache_state is protected by the block group lock */
1186 spin_lock(&block_group->lock);
1189 * only mark this as written if we didn't get put back on
1190 * the dirty list while waiting for IO. Otherwise our
1191 * cache state won't be right, and we won't get written again
1193 if (!ret && list_empty(&block_group->dirty_list))
1194 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1196 block_group->disk_cache_state = BTRFS_DC_ERROR;
1198 spin_unlock(&block_group->lock);
1199 spin_unlock(&trans->transaction->dirty_bgs_lock);
1200 io_ctl->inode = NULL;
1209 * __btrfs_write_out_cache - write out cached info to an inode
1210 * @root - the root the inode belongs to
1211 * @ctl - the free space cache we are going to write out
1212 * @block_group - the block_group for this cache if it belongs to a block_group
1213 * @trans - the trans handle
1214 * @path - the path to use
1215 * @offset - the offset for the key we'll insert
1217 * This function writes out a free space cache struct to disk for quick recovery
1218 * on mount. This will return 0 if it was successful in writing the cache out,
1219 * or an errno if it was not.
1221 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1222 struct btrfs_free_space_ctl *ctl,
1223 struct btrfs_block_group_cache *block_group,
1224 struct btrfs_io_ctl *io_ctl,
1225 struct btrfs_trans_handle *trans,
1226 struct btrfs_path *path, u64 offset)
1228 struct extent_state *cached_state = NULL;
1229 LIST_HEAD(bitmap_list);
1235 if (!i_size_read(inode))
1238 WARN_ON(io_ctl->pages);
1239 ret = io_ctl_init(io_ctl, inode, root, 1);
1243 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1244 down_write(&block_group->data_rwsem);
1245 spin_lock(&block_group->lock);
1246 if (block_group->delalloc_bytes) {
1247 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1248 spin_unlock(&block_group->lock);
1249 up_write(&block_group->data_rwsem);
1250 BTRFS_I(inode)->generation = 0;
1255 spin_unlock(&block_group->lock);
1258 /* Lock all pages first so we can lock the extent safely. */
1259 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1263 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1266 io_ctl_set_generation(io_ctl, trans->transid);
1268 mutex_lock(&ctl->cache_writeout_mutex);
1269 /* Write out the extent entries in the free space cache */
1270 spin_lock(&ctl->tree_lock);
1271 ret = write_cache_extent_entries(io_ctl, ctl,
1272 block_group, &entries, &bitmaps,
1275 goto out_nospc_locked;
1278 * Some spaces that are freed in the current transaction are pinned,
1279 * they will be added into free space cache after the transaction is
1280 * committed, we shouldn't lose them.
1282 * If this changes while we are working we'll get added back to
1283 * the dirty list and redo it. No locking needed
1285 ret = write_pinned_extent_entries(root, block_group, io_ctl, &entries);
1287 goto out_nospc_locked;
1290 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1291 * locked while doing it because a concurrent trim can be manipulating
1292 * or freeing the bitmap.
1294 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1295 spin_unlock(&ctl->tree_lock);
1296 mutex_unlock(&ctl->cache_writeout_mutex);
1300 /* Zero out the rest of the pages just to make sure */
1301 io_ctl_zero_remaining_pages(io_ctl);
1303 /* Everything is written out, now we dirty the pages in the file. */
1304 ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages,
1305 0, i_size_read(inode), &cached_state);
1309 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1310 up_write(&block_group->data_rwsem);
1312 * Release the pages and unlock the extent, we will flush
1315 io_ctl_drop_pages(io_ctl);
1317 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1318 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1321 * at this point the pages are under IO and we're happy,
1322 * The caller is responsible for waiting on them and updating the
1323 * the cache and the inode
1325 io_ctl->entries = entries;
1326 io_ctl->bitmaps = bitmaps;
1328 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1335 io_ctl->inode = NULL;
1336 io_ctl_free(io_ctl);
1338 invalidate_inode_pages2(inode->i_mapping);
1339 BTRFS_I(inode)->generation = 0;
1341 btrfs_update_inode(trans, root, inode);
1347 cleanup_bitmap_list(&bitmap_list);
1348 spin_unlock(&ctl->tree_lock);
1349 mutex_unlock(&ctl->cache_writeout_mutex);
1352 cleanup_write_cache_enospc(inode, io_ctl, &cached_state, &bitmap_list);
1355 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1356 up_write(&block_group->data_rwsem);
1361 int btrfs_write_out_cache(struct btrfs_root *root,
1362 struct btrfs_trans_handle *trans,
1363 struct btrfs_block_group_cache *block_group,
1364 struct btrfs_path *path)
1366 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1367 struct inode *inode;
1370 root = root->fs_info->tree_root;
1372 spin_lock(&block_group->lock);
1373 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1374 spin_unlock(&block_group->lock);
1377 spin_unlock(&block_group->lock);
1379 inode = lookup_free_space_inode(root, block_group, path);
1383 ret = __btrfs_write_out_cache(root, inode, ctl, block_group,
1384 &block_group->io_ctl, trans,
1385 path, block_group->key.objectid);
1388 btrfs_err(root->fs_info,
1389 "failed to write free space cache for block group %llu",
1390 block_group->key.objectid);
1392 spin_lock(&block_group->lock);
1393 block_group->disk_cache_state = BTRFS_DC_ERROR;
1394 spin_unlock(&block_group->lock);
1396 block_group->io_ctl.inode = NULL;
1401 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1402 * to wait for IO and put the inode
1408 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1411 ASSERT(offset >= bitmap_start);
1412 offset -= bitmap_start;
1413 return (unsigned long)(div_u64(offset, unit));
1416 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1418 return (unsigned long)(div_u64(bytes, unit));
1421 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1425 u32 bytes_per_bitmap;
1427 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1428 bitmap_start = offset - ctl->start;
1429 bitmap_start = div_u64(bitmap_start, bytes_per_bitmap);
1430 bitmap_start *= bytes_per_bitmap;
1431 bitmap_start += ctl->start;
1433 return bitmap_start;
1436 static int tree_insert_offset(struct rb_root *root, u64 offset,
1437 struct rb_node *node, int bitmap)
1439 struct rb_node **p = &root->rb_node;
1440 struct rb_node *parent = NULL;
1441 struct btrfs_free_space *info;
1445 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1447 if (offset < info->offset) {
1449 } else if (offset > info->offset) {
1450 p = &(*p)->rb_right;
1453 * we could have a bitmap entry and an extent entry
1454 * share the same offset. If this is the case, we want
1455 * the extent entry to always be found first if we do a
1456 * linear search through the tree, since we want to have
1457 * the quickest allocation time, and allocating from an
1458 * extent is faster than allocating from a bitmap. So
1459 * if we're inserting a bitmap and we find an entry at
1460 * this offset, we want to go right, or after this entry
1461 * logically. If we are inserting an extent and we've
1462 * found a bitmap, we want to go left, or before
1470 p = &(*p)->rb_right;
1472 if (!info->bitmap) {
1481 rb_link_node(node, parent, p);
1482 rb_insert_color(node, root);
1488 * searches the tree for the given offset.
1490 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1491 * want a section that has at least bytes size and comes at or after the given
1494 static struct btrfs_free_space *
1495 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1496 u64 offset, int bitmap_only, int fuzzy)
1498 struct rb_node *n = ctl->free_space_offset.rb_node;
1499 struct btrfs_free_space *entry, *prev = NULL;
1501 /* find entry that is closest to the 'offset' */
1508 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1511 if (offset < entry->offset)
1513 else if (offset > entry->offset)
1526 * bitmap entry and extent entry may share same offset,
1527 * in that case, bitmap entry comes after extent entry.
1532 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1533 if (entry->offset != offset)
1536 WARN_ON(!entry->bitmap);
1539 if (entry->bitmap) {
1541 * if previous extent entry covers the offset,
1542 * we should return it instead of the bitmap entry
1544 n = rb_prev(&entry->offset_index);
1546 prev = rb_entry(n, struct btrfs_free_space,
1548 if (!prev->bitmap &&
1549 prev->offset + prev->bytes > offset)
1559 /* find last entry before the 'offset' */
1561 if (entry->offset > offset) {
1562 n = rb_prev(&entry->offset_index);
1564 entry = rb_entry(n, struct btrfs_free_space,
1566 ASSERT(entry->offset <= offset);
1575 if (entry->bitmap) {
1576 n = rb_prev(&entry->offset_index);
1578 prev = rb_entry(n, struct btrfs_free_space,
1580 if (!prev->bitmap &&
1581 prev->offset + prev->bytes > offset)
1584 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1586 } else if (entry->offset + entry->bytes > offset)
1593 if (entry->bitmap) {
1594 if (entry->offset + BITS_PER_BITMAP *
1598 if (entry->offset + entry->bytes > offset)
1602 n = rb_next(&entry->offset_index);
1605 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1611 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1612 struct btrfs_free_space *info)
1614 rb_erase(&info->offset_index, &ctl->free_space_offset);
1615 ctl->free_extents--;
1618 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1619 struct btrfs_free_space *info)
1621 __unlink_free_space(ctl, info);
1622 ctl->free_space -= info->bytes;
1625 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1626 struct btrfs_free_space *info)
1630 ASSERT(info->bytes || info->bitmap);
1631 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1632 &info->offset_index, (info->bitmap != NULL));
1636 ctl->free_space += info->bytes;
1637 ctl->free_extents++;
1641 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1643 struct btrfs_block_group_cache *block_group = ctl->private;
1647 u64 size = block_group->key.offset;
1648 u32 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1649 u32 max_bitmaps = div_u64(size + bytes_per_bg - 1, bytes_per_bg);
1651 max_bitmaps = max_t(u32, max_bitmaps, 1);
1653 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1656 * The goal is to keep the total amount of memory used per 1gb of space
1657 * at or below 32k, so we need to adjust how much memory we allow to be
1658 * used by extent based free space tracking
1660 if (size < 1024 * 1024 * 1024)
1661 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1663 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1664 div_u64(size, 1024 * 1024 * 1024);
1667 * we want to account for 1 more bitmap than what we have so we can make
1668 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1669 * we add more bitmaps.
1671 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1673 if (bitmap_bytes >= max_bytes) {
1674 ctl->extents_thresh = 0;
1679 * we want the extent entry threshold to always be at most 1/2 the max
1680 * bytes we can have, or whatever is less than that.
1682 extent_bytes = max_bytes - bitmap_bytes;
1683 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1685 ctl->extents_thresh =
1686 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1689 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1690 struct btrfs_free_space *info,
1691 u64 offset, u64 bytes)
1693 unsigned long start, count;
1695 start = offset_to_bit(info->offset, ctl->unit, offset);
1696 count = bytes_to_bits(bytes, ctl->unit);
1697 ASSERT(start + count <= BITS_PER_BITMAP);
1699 bitmap_clear(info->bitmap, start, count);
1701 info->bytes -= bytes;
1702 if (info->max_extent_size > ctl->unit)
1703 info->max_extent_size = 0;
1706 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1707 struct btrfs_free_space *info, u64 offset,
1710 __bitmap_clear_bits(ctl, info, offset, bytes);
1711 ctl->free_space -= bytes;
1714 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1715 struct btrfs_free_space *info, u64 offset,
1718 unsigned long start, count;
1720 start = offset_to_bit(info->offset, ctl->unit, offset);
1721 count = bytes_to_bits(bytes, ctl->unit);
1722 ASSERT(start + count <= BITS_PER_BITMAP);
1724 bitmap_set(info->bitmap, start, count);
1726 info->bytes += bytes;
1727 ctl->free_space += bytes;
1731 * If we can not find suitable extent, we will use bytes to record
1732 * the size of the max extent.
1734 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1735 struct btrfs_free_space *bitmap_info, u64 *offset,
1736 u64 *bytes, bool for_alloc)
1738 unsigned long found_bits = 0;
1739 unsigned long max_bits = 0;
1740 unsigned long bits, i;
1741 unsigned long next_zero;
1742 unsigned long extent_bits;
1745 * Skip searching the bitmap if we don't have a contiguous section that
1746 * is large enough for this allocation.
1749 bitmap_info->max_extent_size &&
1750 bitmap_info->max_extent_size < *bytes) {
1751 *bytes = bitmap_info->max_extent_size;
1755 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1756 max_t(u64, *offset, bitmap_info->offset));
1757 bits = bytes_to_bits(*bytes, ctl->unit);
1759 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1760 if (for_alloc && bits == 1) {
1764 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1765 BITS_PER_BITMAP, i);
1766 extent_bits = next_zero - i;
1767 if (extent_bits >= bits) {
1768 found_bits = extent_bits;
1770 } else if (extent_bits > max_bits) {
1771 max_bits = extent_bits;
1777 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1778 *bytes = (u64)(found_bits) * ctl->unit;
1782 *bytes = (u64)(max_bits) * ctl->unit;
1783 bitmap_info->max_extent_size = *bytes;
1787 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1790 return entry->max_extent_size;
1791 return entry->bytes;
1794 /* Cache the size of the max extent in bytes */
1795 static struct btrfs_free_space *
1796 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1797 unsigned long align, u64 *max_extent_size)
1799 struct btrfs_free_space *entry;
1800 struct rb_node *node;
1805 if (!ctl->free_space_offset.rb_node)
1808 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1812 for (node = &entry->offset_index; node; node = rb_next(node)) {
1813 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1814 if (entry->bytes < *bytes) {
1815 *max_extent_size = max(get_max_extent_size(entry),
1820 /* make sure the space returned is big enough
1821 * to match our requested alignment
1823 if (*bytes >= align) {
1824 tmp = entry->offset - ctl->start + align - 1;
1825 tmp = div64_u64(tmp, align);
1826 tmp = tmp * align + ctl->start;
1827 align_off = tmp - entry->offset;
1830 tmp = entry->offset;
1833 if (entry->bytes < *bytes + align_off) {
1834 *max_extent_size = max(get_max_extent_size(entry),
1839 if (entry->bitmap) {
1842 ret = search_bitmap(ctl, entry, &tmp, &size, true);
1849 max(get_max_extent_size(entry),
1856 *bytes = entry->bytes - align_off;
1863 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1864 struct btrfs_free_space *info, u64 offset)
1866 info->offset = offset_to_bitmap(ctl, offset);
1868 INIT_LIST_HEAD(&info->list);
1869 link_free_space(ctl, info);
1870 ctl->total_bitmaps++;
1872 ctl->op->recalc_thresholds(ctl);
1875 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1876 struct btrfs_free_space *bitmap_info)
1878 unlink_free_space(ctl, bitmap_info);
1879 kfree(bitmap_info->bitmap);
1880 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1881 ctl->total_bitmaps--;
1882 ctl->op->recalc_thresholds(ctl);
1885 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1886 struct btrfs_free_space *bitmap_info,
1887 u64 *offset, u64 *bytes)
1890 u64 search_start, search_bytes;
1894 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1897 * We need to search for bits in this bitmap. We could only cover some
1898 * of the extent in this bitmap thanks to how we add space, so we need
1899 * to search for as much as it as we can and clear that amount, and then
1900 * go searching for the next bit.
1902 search_start = *offset;
1903 search_bytes = ctl->unit;
1904 search_bytes = min(search_bytes, end - search_start + 1);
1905 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1907 if (ret < 0 || search_start != *offset)
1910 /* We may have found more bits than what we need */
1911 search_bytes = min(search_bytes, *bytes);
1913 /* Cannot clear past the end of the bitmap */
1914 search_bytes = min(search_bytes, end - search_start + 1);
1916 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1917 *offset += search_bytes;
1918 *bytes -= search_bytes;
1921 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1922 if (!bitmap_info->bytes)
1923 free_bitmap(ctl, bitmap_info);
1926 * no entry after this bitmap, but we still have bytes to
1927 * remove, so something has gone wrong.
1932 bitmap_info = rb_entry(next, struct btrfs_free_space,
1936 * if the next entry isn't a bitmap we need to return to let the
1937 * extent stuff do its work.
1939 if (!bitmap_info->bitmap)
1943 * Ok the next item is a bitmap, but it may not actually hold
1944 * the information for the rest of this free space stuff, so
1945 * look for it, and if we don't find it return so we can try
1946 * everything over again.
1948 search_start = *offset;
1949 search_bytes = ctl->unit;
1950 ret = search_bitmap(ctl, bitmap_info, &search_start,
1951 &search_bytes, false);
1952 if (ret < 0 || search_start != *offset)
1956 } else if (!bitmap_info->bytes)
1957 free_bitmap(ctl, bitmap_info);
1962 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1963 struct btrfs_free_space *info, u64 offset,
1966 u64 bytes_to_set = 0;
1969 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1971 bytes_to_set = min(end - offset, bytes);
1973 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1976 * We set some bytes, we have no idea what the max extent size is
1979 info->max_extent_size = 0;
1981 return bytes_to_set;
1985 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1986 struct btrfs_free_space *info)
1988 struct btrfs_block_group_cache *block_group = ctl->private;
1989 bool forced = false;
1991 #ifdef CONFIG_BTRFS_DEBUG
1992 if (btrfs_should_fragment_free_space(block_group->fs_info->extent_root,
1998 * If we are below the extents threshold then we can add this as an
1999 * extent, and don't have to deal with the bitmap
2001 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2003 * If this block group has some small extents we don't want to
2004 * use up all of our free slots in the cache with them, we want
2005 * to reserve them to larger extents, however if we have plent
2006 * of cache left then go ahead an dadd them, no sense in adding
2007 * the overhead of a bitmap if we don't have to.
2009 if (info->bytes <= block_group->sectorsize * 4) {
2010 if (ctl->free_extents * 2 <= ctl->extents_thresh)
2018 * The original block groups from mkfs can be really small, like 8
2019 * megabytes, so don't bother with a bitmap for those entries. However
2020 * some block groups can be smaller than what a bitmap would cover but
2021 * are still large enough that they could overflow the 32k memory limit,
2022 * so allow those block groups to still be allowed to have a bitmap
2025 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2031 static struct btrfs_free_space_op free_space_op = {
2032 .recalc_thresholds = recalculate_thresholds,
2033 .use_bitmap = use_bitmap,
2036 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2037 struct btrfs_free_space *info)
2039 struct btrfs_free_space *bitmap_info;
2040 struct btrfs_block_group_cache *block_group = NULL;
2042 u64 bytes, offset, bytes_added;
2045 bytes = info->bytes;
2046 offset = info->offset;
2048 if (!ctl->op->use_bitmap(ctl, info))
2051 if (ctl->op == &free_space_op)
2052 block_group = ctl->private;
2055 * Since we link bitmaps right into the cluster we need to see if we
2056 * have a cluster here, and if so and it has our bitmap we need to add
2057 * the free space to that bitmap.
2059 if (block_group && !list_empty(&block_group->cluster_list)) {
2060 struct btrfs_free_cluster *cluster;
2061 struct rb_node *node;
2062 struct btrfs_free_space *entry;
2064 cluster = list_entry(block_group->cluster_list.next,
2065 struct btrfs_free_cluster,
2067 spin_lock(&cluster->lock);
2068 node = rb_first(&cluster->root);
2070 spin_unlock(&cluster->lock);
2071 goto no_cluster_bitmap;
2074 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2075 if (!entry->bitmap) {
2076 spin_unlock(&cluster->lock);
2077 goto no_cluster_bitmap;
2080 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2081 bytes_added = add_bytes_to_bitmap(ctl, entry,
2083 bytes -= bytes_added;
2084 offset += bytes_added;
2086 spin_unlock(&cluster->lock);
2094 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2101 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2102 bytes -= bytes_added;
2103 offset += bytes_added;
2113 if (info && info->bitmap) {
2114 add_new_bitmap(ctl, info, offset);
2119 spin_unlock(&ctl->tree_lock);
2121 /* no pre-allocated info, allocate a new one */
2123 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2126 spin_lock(&ctl->tree_lock);
2132 /* allocate the bitmap */
2133 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
2134 spin_lock(&ctl->tree_lock);
2135 if (!info->bitmap) {
2145 kfree(info->bitmap);
2146 kmem_cache_free(btrfs_free_space_cachep, info);
2152 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2153 struct btrfs_free_space *info, bool update_stat)
2155 struct btrfs_free_space *left_info;
2156 struct btrfs_free_space *right_info;
2157 bool merged = false;
2158 u64 offset = info->offset;
2159 u64 bytes = info->bytes;
2162 * first we want to see if there is free space adjacent to the range we
2163 * are adding, if there is remove that struct and add a new one to
2164 * cover the entire range
2166 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2167 if (right_info && rb_prev(&right_info->offset_index))
2168 left_info = rb_entry(rb_prev(&right_info->offset_index),
2169 struct btrfs_free_space, offset_index);
2171 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2173 if (right_info && !right_info->bitmap) {
2175 unlink_free_space(ctl, right_info);
2177 __unlink_free_space(ctl, right_info);
2178 info->bytes += right_info->bytes;
2179 kmem_cache_free(btrfs_free_space_cachep, right_info);
2183 if (left_info && !left_info->bitmap &&
2184 left_info->offset + left_info->bytes == offset) {
2186 unlink_free_space(ctl, left_info);
2188 __unlink_free_space(ctl, left_info);
2189 info->offset = left_info->offset;
2190 info->bytes += left_info->bytes;
2191 kmem_cache_free(btrfs_free_space_cachep, left_info);
2198 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2199 struct btrfs_free_space *info,
2202 struct btrfs_free_space *bitmap;
2205 const u64 end = info->offset + info->bytes;
2206 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2209 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2213 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2214 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2217 bytes = (j - i) * ctl->unit;
2218 info->bytes += bytes;
2221 bitmap_clear_bits(ctl, bitmap, end, bytes);
2223 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2226 free_bitmap(ctl, bitmap);
2231 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2232 struct btrfs_free_space *info,
2235 struct btrfs_free_space *bitmap;
2239 unsigned long prev_j;
2242 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2243 /* If we're on a boundary, try the previous logical bitmap. */
2244 if (bitmap_offset == info->offset) {
2245 if (info->offset == 0)
2247 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2250 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2254 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2256 prev_j = (unsigned long)-1;
2257 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2265 if (prev_j == (unsigned long)-1)
2266 bytes = (i + 1) * ctl->unit;
2268 bytes = (i - prev_j) * ctl->unit;
2270 info->offset -= bytes;
2271 info->bytes += bytes;
2274 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2276 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2279 free_bitmap(ctl, bitmap);
2285 * We prefer always to allocate from extent entries, both for clustered and
2286 * non-clustered allocation requests. So when attempting to add a new extent
2287 * entry, try to see if there's adjacent free space in bitmap entries, and if
2288 * there is, migrate that space from the bitmaps to the extent.
2289 * Like this we get better chances of satisfying space allocation requests
2290 * because we attempt to satisfy them based on a single cache entry, and never
2291 * on 2 or more entries - even if the entries represent a contiguous free space
2292 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2295 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2296 struct btrfs_free_space *info,
2300 * Only work with disconnected entries, as we can change their offset,
2301 * and must be extent entries.
2303 ASSERT(!info->bitmap);
2304 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2306 if (ctl->total_bitmaps > 0) {
2308 bool stole_front = false;
2310 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2311 if (ctl->total_bitmaps > 0)
2312 stole_front = steal_from_bitmap_to_front(ctl, info,
2315 if (stole_end || stole_front)
2316 try_merge_free_space(ctl, info, update_stat);
2320 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
2321 u64 offset, u64 bytes)
2323 struct btrfs_free_space *info;
2326 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2330 info->offset = offset;
2331 info->bytes = bytes;
2332 RB_CLEAR_NODE(&info->offset_index);
2334 spin_lock(&ctl->tree_lock);
2336 if (try_merge_free_space(ctl, info, true))
2340 * There was no extent directly to the left or right of this new
2341 * extent then we know we're going to have to allocate a new extent, so
2342 * before we do that see if we need to drop this into a bitmap
2344 ret = insert_into_bitmap(ctl, info);
2353 * Only steal free space from adjacent bitmaps if we're sure we're not
2354 * going to add the new free space to existing bitmap entries - because
2355 * that would mean unnecessary work that would be reverted. Therefore
2356 * attempt to steal space from bitmaps if we're adding an extent entry.
2358 steal_from_bitmap(ctl, info, true);
2360 ret = link_free_space(ctl, info);
2362 kmem_cache_free(btrfs_free_space_cachep, info);
2364 spin_unlock(&ctl->tree_lock);
2367 printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
2368 ASSERT(ret != -EEXIST);
2374 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2375 u64 offset, u64 bytes)
2377 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2378 struct btrfs_free_space *info;
2380 bool re_search = false;
2382 spin_lock(&ctl->tree_lock);
2389 info = tree_search_offset(ctl, offset, 0, 0);
2392 * oops didn't find an extent that matched the space we wanted
2393 * to remove, look for a bitmap instead
2395 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2399 * If we found a partial bit of our free space in a
2400 * bitmap but then couldn't find the other part this may
2401 * be a problem, so WARN about it.
2409 if (!info->bitmap) {
2410 unlink_free_space(ctl, info);
2411 if (offset == info->offset) {
2412 u64 to_free = min(bytes, info->bytes);
2414 info->bytes -= to_free;
2415 info->offset += to_free;
2417 ret = link_free_space(ctl, info);
2420 kmem_cache_free(btrfs_free_space_cachep, info);
2427 u64 old_end = info->bytes + info->offset;
2429 info->bytes = offset - info->offset;
2430 ret = link_free_space(ctl, info);
2435 /* Not enough bytes in this entry to satisfy us */
2436 if (old_end < offset + bytes) {
2437 bytes -= old_end - offset;
2440 } else if (old_end == offset + bytes) {
2444 spin_unlock(&ctl->tree_lock);
2446 ret = btrfs_add_free_space(block_group, offset + bytes,
2447 old_end - (offset + bytes));
2453 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2454 if (ret == -EAGAIN) {
2459 spin_unlock(&ctl->tree_lock);
2464 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2467 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2468 struct btrfs_free_space *info;
2472 spin_lock(&ctl->tree_lock);
2473 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2474 info = rb_entry(n, struct btrfs_free_space, offset_index);
2475 if (info->bytes >= bytes && !block_group->ro)
2477 btrfs_crit(block_group->fs_info,
2478 "entry offset %llu, bytes %llu, bitmap %s",
2479 info->offset, info->bytes,
2480 (info->bitmap) ? "yes" : "no");
2482 spin_unlock(&ctl->tree_lock);
2483 btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2484 list_empty(&block_group->cluster_list) ? "no" : "yes");
2485 btrfs_info(block_group->fs_info,
2486 "%d blocks of free space at or bigger than bytes is", count);
2489 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2491 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2493 spin_lock_init(&ctl->tree_lock);
2494 ctl->unit = block_group->sectorsize;
2495 ctl->start = block_group->key.objectid;
2496 ctl->private = block_group;
2497 ctl->op = &free_space_op;
2498 INIT_LIST_HEAD(&ctl->trimming_ranges);
2499 mutex_init(&ctl->cache_writeout_mutex);
2502 * we only want to have 32k of ram per block group for keeping
2503 * track of free space, and if we pass 1/2 of that we want to
2504 * start converting things over to using bitmaps
2506 ctl->extents_thresh = ((1024 * 32) / 2) /
2507 sizeof(struct btrfs_free_space);
2511 * for a given cluster, put all of its extents back into the free
2512 * space cache. If the block group passed doesn't match the block group
2513 * pointed to by the cluster, someone else raced in and freed the
2514 * cluster already. In that case, we just return without changing anything
2517 __btrfs_return_cluster_to_free_space(
2518 struct btrfs_block_group_cache *block_group,
2519 struct btrfs_free_cluster *cluster)
2521 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2522 struct btrfs_free_space *entry;
2523 struct rb_node *node;
2525 spin_lock(&cluster->lock);
2526 if (cluster->block_group != block_group)
2529 cluster->block_group = NULL;
2530 cluster->window_start = 0;
2531 list_del_init(&cluster->block_group_list);
2533 node = rb_first(&cluster->root);
2537 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2538 node = rb_next(&entry->offset_index);
2539 rb_erase(&entry->offset_index, &cluster->root);
2540 RB_CLEAR_NODE(&entry->offset_index);
2542 bitmap = (entry->bitmap != NULL);
2544 try_merge_free_space(ctl, entry, false);
2545 steal_from_bitmap(ctl, entry, false);
2547 tree_insert_offset(&ctl->free_space_offset,
2548 entry->offset, &entry->offset_index, bitmap);
2550 cluster->root = RB_ROOT;
2553 spin_unlock(&cluster->lock);
2554 btrfs_put_block_group(block_group);
2558 static void __btrfs_remove_free_space_cache_locked(
2559 struct btrfs_free_space_ctl *ctl)
2561 struct btrfs_free_space *info;
2562 struct rb_node *node;
2564 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2565 info = rb_entry(node, struct btrfs_free_space, offset_index);
2566 if (!info->bitmap) {
2567 unlink_free_space(ctl, info);
2568 kmem_cache_free(btrfs_free_space_cachep, info);
2570 free_bitmap(ctl, info);
2573 cond_resched_lock(&ctl->tree_lock);
2577 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2579 spin_lock(&ctl->tree_lock);
2580 __btrfs_remove_free_space_cache_locked(ctl);
2581 spin_unlock(&ctl->tree_lock);
2584 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2586 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2587 struct btrfs_free_cluster *cluster;
2588 struct list_head *head;
2590 spin_lock(&ctl->tree_lock);
2591 while ((head = block_group->cluster_list.next) !=
2592 &block_group->cluster_list) {
2593 cluster = list_entry(head, struct btrfs_free_cluster,
2596 WARN_ON(cluster->block_group != block_group);
2597 __btrfs_return_cluster_to_free_space(block_group, cluster);
2599 cond_resched_lock(&ctl->tree_lock);
2601 __btrfs_remove_free_space_cache_locked(ctl);
2602 spin_unlock(&ctl->tree_lock);
2606 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2607 u64 offset, u64 bytes, u64 empty_size,
2608 u64 *max_extent_size)
2610 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2611 struct btrfs_free_space *entry = NULL;
2612 u64 bytes_search = bytes + empty_size;
2615 u64 align_gap_len = 0;
2617 spin_lock(&ctl->tree_lock);
2618 entry = find_free_space(ctl, &offset, &bytes_search,
2619 block_group->full_stripe_len, max_extent_size);
2624 if (entry->bitmap) {
2625 bitmap_clear_bits(ctl, entry, offset, bytes);
2627 free_bitmap(ctl, entry);
2629 unlink_free_space(ctl, entry);
2630 align_gap_len = offset - entry->offset;
2631 align_gap = entry->offset;
2633 entry->offset = offset + bytes;
2634 WARN_ON(entry->bytes < bytes + align_gap_len);
2636 entry->bytes -= bytes + align_gap_len;
2638 kmem_cache_free(btrfs_free_space_cachep, entry);
2640 link_free_space(ctl, entry);
2643 spin_unlock(&ctl->tree_lock);
2646 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2651 * given a cluster, put all of its extents back into the free space
2652 * cache. If a block group is passed, this function will only free
2653 * a cluster that belongs to the passed block group.
2655 * Otherwise, it'll get a reference on the block group pointed to by the
2656 * cluster and remove the cluster from it.
2658 int btrfs_return_cluster_to_free_space(
2659 struct btrfs_block_group_cache *block_group,
2660 struct btrfs_free_cluster *cluster)
2662 struct btrfs_free_space_ctl *ctl;
2665 /* first, get a safe pointer to the block group */
2666 spin_lock(&cluster->lock);
2668 block_group = cluster->block_group;
2670 spin_unlock(&cluster->lock);
2673 } else if (cluster->block_group != block_group) {
2674 /* someone else has already freed it don't redo their work */
2675 spin_unlock(&cluster->lock);
2678 atomic_inc(&block_group->count);
2679 spin_unlock(&cluster->lock);
2681 ctl = block_group->free_space_ctl;
2683 /* now return any extents the cluster had on it */
2684 spin_lock(&ctl->tree_lock);
2685 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2686 spin_unlock(&ctl->tree_lock);
2688 /* finally drop our ref */
2689 btrfs_put_block_group(block_group);
2693 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2694 struct btrfs_free_cluster *cluster,
2695 struct btrfs_free_space *entry,
2696 u64 bytes, u64 min_start,
2697 u64 *max_extent_size)
2699 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2701 u64 search_start = cluster->window_start;
2702 u64 search_bytes = bytes;
2705 search_start = min_start;
2706 search_bytes = bytes;
2708 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2710 *max_extent_size = max(get_max_extent_size(entry),
2716 __bitmap_clear_bits(ctl, entry, ret, bytes);
2722 * given a cluster, try to allocate 'bytes' from it, returns 0
2723 * if it couldn't find anything suitably large, or a logical disk offset
2724 * if things worked out
2726 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2727 struct btrfs_free_cluster *cluster, u64 bytes,
2728 u64 min_start, u64 *max_extent_size)
2730 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2731 struct btrfs_free_space *entry = NULL;
2732 struct rb_node *node;
2735 spin_lock(&cluster->lock);
2736 if (bytes > cluster->max_size)
2739 if (cluster->block_group != block_group)
2742 node = rb_first(&cluster->root);
2746 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2748 if (entry->bytes < bytes)
2749 *max_extent_size = max(get_max_extent_size(entry),
2752 if (entry->bytes < bytes ||
2753 (!entry->bitmap && entry->offset < min_start)) {
2754 node = rb_next(&entry->offset_index);
2757 entry = rb_entry(node, struct btrfs_free_space,
2762 if (entry->bitmap) {
2763 ret = btrfs_alloc_from_bitmap(block_group,
2764 cluster, entry, bytes,
2765 cluster->window_start,
2768 node = rb_next(&entry->offset_index);
2771 entry = rb_entry(node, struct btrfs_free_space,
2775 cluster->window_start += bytes;
2777 ret = entry->offset;
2779 entry->offset += bytes;
2780 entry->bytes -= bytes;
2783 if (entry->bytes == 0)
2784 rb_erase(&entry->offset_index, &cluster->root);
2788 spin_unlock(&cluster->lock);
2793 spin_lock(&ctl->tree_lock);
2795 ctl->free_space -= bytes;
2796 if (entry->bytes == 0) {
2797 ctl->free_extents--;
2798 if (entry->bitmap) {
2799 kfree(entry->bitmap);
2800 ctl->total_bitmaps--;
2801 ctl->op->recalc_thresholds(ctl);
2803 kmem_cache_free(btrfs_free_space_cachep, entry);
2806 spin_unlock(&ctl->tree_lock);
2811 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2812 struct btrfs_free_space *entry,
2813 struct btrfs_free_cluster *cluster,
2814 u64 offset, u64 bytes,
2815 u64 cont1_bytes, u64 min_bytes)
2817 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2818 unsigned long next_zero;
2820 unsigned long want_bits;
2821 unsigned long min_bits;
2822 unsigned long found_bits;
2823 unsigned long max_bits = 0;
2824 unsigned long start = 0;
2825 unsigned long total_found = 0;
2828 i = offset_to_bit(entry->offset, ctl->unit,
2829 max_t(u64, offset, entry->offset));
2830 want_bits = bytes_to_bits(bytes, ctl->unit);
2831 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2834 * Don't bother looking for a cluster in this bitmap if it's heavily
2837 if (entry->max_extent_size &&
2838 entry->max_extent_size < cont1_bytes)
2842 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2843 next_zero = find_next_zero_bit(entry->bitmap,
2844 BITS_PER_BITMAP, i);
2845 if (next_zero - i >= min_bits) {
2846 found_bits = next_zero - i;
2847 if (found_bits > max_bits)
2848 max_bits = found_bits;
2851 if (next_zero - i > max_bits)
2852 max_bits = next_zero - i;
2857 entry->max_extent_size = (u64)max_bits * ctl->unit;
2863 cluster->max_size = 0;
2866 total_found += found_bits;
2868 if (cluster->max_size < found_bits * ctl->unit)
2869 cluster->max_size = found_bits * ctl->unit;
2871 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2876 cluster->window_start = start * ctl->unit + entry->offset;
2877 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2878 ret = tree_insert_offset(&cluster->root, entry->offset,
2879 &entry->offset_index, 1);
2880 ASSERT(!ret); /* -EEXIST; Logic error */
2882 trace_btrfs_setup_cluster(block_group, cluster,
2883 total_found * ctl->unit, 1);
2888 * This searches the block group for just extents to fill the cluster with.
2889 * Try to find a cluster with at least bytes total bytes, at least one
2890 * extent of cont1_bytes, and other clusters of at least min_bytes.
2893 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2894 struct btrfs_free_cluster *cluster,
2895 struct list_head *bitmaps, u64 offset, u64 bytes,
2896 u64 cont1_bytes, u64 min_bytes)
2898 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2899 struct btrfs_free_space *first = NULL;
2900 struct btrfs_free_space *entry = NULL;
2901 struct btrfs_free_space *last;
2902 struct rb_node *node;
2907 entry = tree_search_offset(ctl, offset, 0, 1);
2912 * We don't want bitmaps, so just move along until we find a normal
2915 while (entry->bitmap || entry->bytes < min_bytes) {
2916 if (entry->bitmap && list_empty(&entry->list))
2917 list_add_tail(&entry->list, bitmaps);
2918 node = rb_next(&entry->offset_index);
2921 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2924 window_free = entry->bytes;
2925 max_extent = entry->bytes;
2929 for (node = rb_next(&entry->offset_index); node;
2930 node = rb_next(&entry->offset_index)) {
2931 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2933 if (entry->bitmap) {
2934 if (list_empty(&entry->list))
2935 list_add_tail(&entry->list, bitmaps);
2939 if (entry->bytes < min_bytes)
2943 window_free += entry->bytes;
2944 if (entry->bytes > max_extent)
2945 max_extent = entry->bytes;
2948 if (window_free < bytes || max_extent < cont1_bytes)
2951 cluster->window_start = first->offset;
2953 node = &first->offset_index;
2956 * now we've found our entries, pull them out of the free space
2957 * cache and put them into the cluster rbtree
2962 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2963 node = rb_next(&entry->offset_index);
2964 if (entry->bitmap || entry->bytes < min_bytes)
2967 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2968 ret = tree_insert_offset(&cluster->root, entry->offset,
2969 &entry->offset_index, 0);
2970 total_size += entry->bytes;
2971 ASSERT(!ret); /* -EEXIST; Logic error */
2972 } while (node && entry != last);
2974 cluster->max_size = max_extent;
2975 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2980 * This specifically looks for bitmaps that may work in the cluster, we assume
2981 * that we have already failed to find extents that will work.
2984 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2985 struct btrfs_free_cluster *cluster,
2986 struct list_head *bitmaps, u64 offset, u64 bytes,
2987 u64 cont1_bytes, u64 min_bytes)
2989 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2990 struct btrfs_free_space *entry = NULL;
2992 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2994 if (ctl->total_bitmaps == 0)
2998 * The bitmap that covers offset won't be in the list unless offset
2999 * is just its start offset.
3001 if (!list_empty(bitmaps))
3002 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3004 if (!entry || entry->offset != bitmap_offset) {
3005 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3006 if (entry && list_empty(&entry->list))
3007 list_add(&entry->list, bitmaps);
3010 list_for_each_entry(entry, bitmaps, list) {
3011 if (entry->bytes < bytes)
3013 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3014 bytes, cont1_bytes, min_bytes);
3020 * The bitmaps list has all the bitmaps that record free space
3021 * starting after offset, so no more search is required.
3027 * here we try to find a cluster of blocks in a block group. The goal
3028 * is to find at least bytes+empty_size.
3029 * We might not find them all in one contiguous area.
3031 * returns zero and sets up cluster if things worked out, otherwise
3032 * it returns -enospc
3034 int btrfs_find_space_cluster(struct btrfs_root *root,
3035 struct btrfs_block_group_cache *block_group,
3036 struct btrfs_free_cluster *cluster,
3037 u64 offset, u64 bytes, u64 empty_size)
3039 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3040 struct btrfs_free_space *entry, *tmp;
3047 * Choose the minimum extent size we'll require for this
3048 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3049 * For metadata, allow allocates with smaller extents. For
3050 * data, keep it dense.
3052 if (btrfs_test_opt(root, SSD_SPREAD)) {
3053 cont1_bytes = min_bytes = bytes + empty_size;
3054 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3055 cont1_bytes = bytes;
3056 min_bytes = block_group->sectorsize;
3058 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3059 min_bytes = block_group->sectorsize;
3062 spin_lock(&ctl->tree_lock);
3065 * If we know we don't have enough space to make a cluster don't even
3066 * bother doing all the work to try and find one.
3068 if (ctl->free_space < bytes) {
3069 spin_unlock(&ctl->tree_lock);
3073 spin_lock(&cluster->lock);
3075 /* someone already found a cluster, hooray */
3076 if (cluster->block_group) {
3081 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3084 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3086 cont1_bytes, min_bytes);
3088 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3089 offset, bytes + empty_size,
3090 cont1_bytes, min_bytes);
3092 /* Clear our temporary list */
3093 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3094 list_del_init(&entry->list);
3097 atomic_inc(&block_group->count);
3098 list_add_tail(&cluster->block_group_list,
3099 &block_group->cluster_list);
3100 cluster->block_group = block_group;
3102 trace_btrfs_failed_cluster_setup(block_group);
3105 spin_unlock(&cluster->lock);
3106 spin_unlock(&ctl->tree_lock);
3112 * simple code to zero out a cluster
3114 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3116 spin_lock_init(&cluster->lock);
3117 spin_lock_init(&cluster->refill_lock);
3118 cluster->root = RB_ROOT;
3119 cluster->max_size = 0;
3120 cluster->fragmented = false;
3121 INIT_LIST_HEAD(&cluster->block_group_list);
3122 cluster->block_group = NULL;
3125 static int do_trimming(struct btrfs_block_group_cache *block_group,
3126 u64 *total_trimmed, u64 start, u64 bytes,
3127 u64 reserved_start, u64 reserved_bytes,
3128 struct btrfs_trim_range *trim_entry)
3130 struct btrfs_space_info *space_info = block_group->space_info;
3131 struct btrfs_fs_info *fs_info = block_group->fs_info;
3132 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3137 spin_lock(&space_info->lock);
3138 spin_lock(&block_group->lock);
3139 if (!block_group->ro) {
3140 block_group->reserved += reserved_bytes;
3141 space_info->bytes_reserved += reserved_bytes;
3144 spin_unlock(&block_group->lock);
3145 spin_unlock(&space_info->lock);
3147 ret = btrfs_discard_extent(fs_info->extent_root,
3148 start, bytes, &trimmed);
3150 *total_trimmed += trimmed;
3152 mutex_lock(&ctl->cache_writeout_mutex);
3153 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
3154 list_del(&trim_entry->list);
3155 mutex_unlock(&ctl->cache_writeout_mutex);
3158 spin_lock(&space_info->lock);
3159 spin_lock(&block_group->lock);
3160 if (block_group->ro)
3161 space_info->bytes_readonly += reserved_bytes;
3162 block_group->reserved -= reserved_bytes;
3163 space_info->bytes_reserved -= reserved_bytes;
3164 spin_unlock(&space_info->lock);
3165 spin_unlock(&block_group->lock);
3171 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3172 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3174 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3175 struct btrfs_free_space *entry;
3176 struct rb_node *node;
3182 while (start < end) {
3183 struct btrfs_trim_range trim_entry;
3185 mutex_lock(&ctl->cache_writeout_mutex);
3186 spin_lock(&ctl->tree_lock);
3188 if (ctl->free_space < minlen) {
3189 spin_unlock(&ctl->tree_lock);
3190 mutex_unlock(&ctl->cache_writeout_mutex);
3194 entry = tree_search_offset(ctl, start, 0, 1);
3196 spin_unlock(&ctl->tree_lock);
3197 mutex_unlock(&ctl->cache_writeout_mutex);
3202 while (entry->bitmap) {
3203 node = rb_next(&entry->offset_index);
3205 spin_unlock(&ctl->tree_lock);
3206 mutex_unlock(&ctl->cache_writeout_mutex);
3209 entry = rb_entry(node, struct btrfs_free_space,
3213 if (entry->offset >= end) {
3214 spin_unlock(&ctl->tree_lock);
3215 mutex_unlock(&ctl->cache_writeout_mutex);
3219 extent_start = entry->offset;
3220 extent_bytes = entry->bytes;
3221 start = max(start, extent_start);
3222 bytes = min(extent_start + extent_bytes, end) - start;
3223 if (bytes < minlen) {
3224 spin_unlock(&ctl->tree_lock);
3225 mutex_unlock(&ctl->cache_writeout_mutex);
3229 unlink_free_space(ctl, entry);
3230 kmem_cache_free(btrfs_free_space_cachep, entry);
3232 spin_unlock(&ctl->tree_lock);
3233 trim_entry.start = extent_start;
3234 trim_entry.bytes = extent_bytes;
3235 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3236 mutex_unlock(&ctl->cache_writeout_mutex);
3238 ret = do_trimming(block_group, total_trimmed, start, bytes,
3239 extent_start, extent_bytes, &trim_entry);
3245 if (fatal_signal_pending(current)) {
3256 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3257 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3259 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3260 struct btrfs_free_space *entry;
3264 u64 offset = offset_to_bitmap(ctl, start);
3266 while (offset < end) {
3267 bool next_bitmap = false;
3268 struct btrfs_trim_range trim_entry;
3270 mutex_lock(&ctl->cache_writeout_mutex);
3271 spin_lock(&ctl->tree_lock);
3273 if (ctl->free_space < minlen) {
3274 spin_unlock(&ctl->tree_lock);
3275 mutex_unlock(&ctl->cache_writeout_mutex);
3279 entry = tree_search_offset(ctl, offset, 1, 0);
3281 spin_unlock(&ctl->tree_lock);
3282 mutex_unlock(&ctl->cache_writeout_mutex);
3288 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3289 if (ret2 || start >= end) {
3290 spin_unlock(&ctl->tree_lock);
3291 mutex_unlock(&ctl->cache_writeout_mutex);
3296 bytes = min(bytes, end - start);
3297 if (bytes < minlen) {
3298 spin_unlock(&ctl->tree_lock);
3299 mutex_unlock(&ctl->cache_writeout_mutex);
3303 bitmap_clear_bits(ctl, entry, start, bytes);
3304 if (entry->bytes == 0)
3305 free_bitmap(ctl, entry);
3307 spin_unlock(&ctl->tree_lock);
3308 trim_entry.start = start;
3309 trim_entry.bytes = bytes;
3310 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3311 mutex_unlock(&ctl->cache_writeout_mutex);
3313 ret = do_trimming(block_group, total_trimmed, start, bytes,
3314 start, bytes, &trim_entry);
3319 offset += BITS_PER_BITMAP * ctl->unit;
3322 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3323 offset += BITS_PER_BITMAP * ctl->unit;
3326 if (fatal_signal_pending(current)) {
3337 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
3339 atomic_inc(&cache->trimming);
3342 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3344 struct extent_map_tree *em_tree;
3345 struct extent_map *em;
3348 spin_lock(&block_group->lock);
3349 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3350 block_group->removed);
3351 spin_unlock(&block_group->lock);
3354 lock_chunks(block_group->fs_info->chunk_root);
3355 em_tree = &block_group->fs_info->mapping_tree.map_tree;
3356 write_lock(&em_tree->lock);
3357 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3359 BUG_ON(!em); /* logic error, can't happen */
3361 * remove_extent_mapping() will delete us from the pinned_chunks
3362 * list, which is protected by the chunk mutex.
3364 remove_extent_mapping(em_tree, em);
3365 write_unlock(&em_tree->lock);
3366 unlock_chunks(block_group->fs_info->chunk_root);
3368 /* once for us and once for the tree */
3369 free_extent_map(em);
3370 free_extent_map(em);
3373 * We've left one free space entry and other tasks trimming
3374 * this block group have left 1 entry each one. Free them.
3376 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3380 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3381 u64 *trimmed, u64 start, u64 end, u64 minlen)
3387 spin_lock(&block_group->lock);
3388 if (block_group->removed) {
3389 spin_unlock(&block_group->lock);
3392 btrfs_get_block_group_trimming(block_group);
3393 spin_unlock(&block_group->lock);
3395 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3399 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3401 btrfs_put_block_group_trimming(block_group);
3406 * Find the left-most item in the cache tree, and then return the
3407 * smallest inode number in the item.
3409 * Note: the returned inode number may not be the smallest one in
3410 * the tree, if the left-most item is a bitmap.
3412 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3414 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3415 struct btrfs_free_space *entry = NULL;
3418 spin_lock(&ctl->tree_lock);
3420 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3423 entry = rb_entry(rb_first(&ctl->free_space_offset),
3424 struct btrfs_free_space, offset_index);
3426 if (!entry->bitmap) {
3427 ino = entry->offset;
3429 unlink_free_space(ctl, entry);
3433 kmem_cache_free(btrfs_free_space_cachep, entry);
3435 link_free_space(ctl, entry);
3441 ret = search_bitmap(ctl, entry, &offset, &count, true);
3442 /* Logic error; Should be empty if it can't find anything */
3446 bitmap_clear_bits(ctl, entry, offset, 1);
3447 if (entry->bytes == 0)
3448 free_bitmap(ctl, entry);
3451 spin_unlock(&ctl->tree_lock);
3456 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3457 struct btrfs_path *path)
3459 struct inode *inode = NULL;
3461 spin_lock(&root->ino_cache_lock);
3462 if (root->ino_cache_inode)
3463 inode = igrab(root->ino_cache_inode);
3464 spin_unlock(&root->ino_cache_lock);
3468 inode = __lookup_free_space_inode(root, path, 0);
3472 spin_lock(&root->ino_cache_lock);
3473 if (!btrfs_fs_closing(root->fs_info))
3474 root->ino_cache_inode = igrab(inode);
3475 spin_unlock(&root->ino_cache_lock);
3480 int create_free_ino_inode(struct btrfs_root *root,
3481 struct btrfs_trans_handle *trans,
3482 struct btrfs_path *path)
3484 return __create_free_space_inode(root, trans, path,
3485 BTRFS_FREE_INO_OBJECTID, 0);
3488 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3490 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3491 struct btrfs_path *path;
3492 struct inode *inode;
3494 u64 root_gen = btrfs_root_generation(&root->root_item);
3496 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3500 * If we're unmounting then just return, since this does a search on the
3501 * normal root and not the commit root and we could deadlock.
3503 if (btrfs_fs_closing(fs_info))
3506 path = btrfs_alloc_path();
3510 inode = lookup_free_ino_inode(root, path);
3514 if (root_gen != BTRFS_I(inode)->generation)
3517 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3521 "failed to load free ino cache for root %llu",
3522 root->root_key.objectid);
3526 btrfs_free_path(path);
3530 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3531 struct btrfs_trans_handle *trans,
3532 struct btrfs_path *path,
3533 struct inode *inode)
3535 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3537 struct btrfs_io_ctl io_ctl;
3538 bool release_metadata = true;
3540 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3543 memset(&io_ctl, 0, sizeof(io_ctl));
3544 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl,
3548 * At this point writepages() didn't error out, so our metadata
3549 * reservation is released when the writeback finishes, at
3550 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3551 * with or without an error.
3553 release_metadata = false;
3554 ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0);
3558 if (release_metadata)
3559 btrfs_delalloc_release_metadata(inode, inode->i_size);
3561 btrfs_err(root->fs_info,
3562 "failed to write free ino cache for root %llu",
3563 root->root_key.objectid);
3570 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3572 * Use this if you need to make a bitmap or extent entry specifically, it
3573 * doesn't do any of the merging that add_free_space does, this acts a lot like
3574 * how the free space cache loading stuff works, so you can get really weird
3577 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3578 u64 offset, u64 bytes, bool bitmap)
3580 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3581 struct btrfs_free_space *info = NULL, *bitmap_info;
3588 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3594 spin_lock(&ctl->tree_lock);
3595 info->offset = offset;
3596 info->bytes = bytes;
3597 info->max_extent_size = 0;
3598 ret = link_free_space(ctl, info);
3599 spin_unlock(&ctl->tree_lock);
3601 kmem_cache_free(btrfs_free_space_cachep, info);
3606 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3608 kmem_cache_free(btrfs_free_space_cachep, info);
3613 spin_lock(&ctl->tree_lock);
3614 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3619 add_new_bitmap(ctl, info, offset);
3624 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3626 bytes -= bytes_added;
3627 offset += bytes_added;
3628 spin_unlock(&ctl->tree_lock);
3634 kmem_cache_free(btrfs_free_space_cachep, info);
3641 * Checks to see if the given range is in the free space cache. This is really
3642 * just used to check the absence of space, so if there is free space in the
3643 * range at all we will return 1.
3645 int test_check_exists(struct btrfs_block_group_cache *cache,
3646 u64 offset, u64 bytes)
3648 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3649 struct btrfs_free_space *info;
3652 spin_lock(&ctl->tree_lock);
3653 info = tree_search_offset(ctl, offset, 0, 0);
3655 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3663 u64 bit_off, bit_bytes;
3665 struct btrfs_free_space *tmp;
3668 bit_bytes = ctl->unit;
3669 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
3671 if (bit_off == offset) {
3674 } else if (bit_off > offset &&
3675 offset + bytes > bit_off) {
3681 n = rb_prev(&info->offset_index);
3683 tmp = rb_entry(n, struct btrfs_free_space,
3685 if (tmp->offset + tmp->bytes < offset)
3687 if (offset + bytes < tmp->offset) {
3688 n = rb_prev(&info->offset_index);
3695 n = rb_next(&info->offset_index);
3697 tmp = rb_entry(n, struct btrfs_free_space,
3699 if (offset + bytes < tmp->offset)
3701 if (tmp->offset + tmp->bytes < offset) {
3702 n = rb_next(&info->offset_index);
3713 if (info->offset == offset) {
3718 if (offset > info->offset && offset < info->offset + info->bytes)
3721 spin_unlock(&ctl->tree_lock);
3724 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */