1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Red Hat. All rights reserved.
6 #include <linux/pagemap.h>
7 #include <linux/sched.h>
8 #include <linux/sched/signal.h>
9 #include <linux/slab.h>
10 #include <linux/math64.h>
11 #include <linux/ratelimit.h>
12 #include <linux/error-injection.h>
13 #include <linux/sched/mm.h>
16 #include "free-space-cache.h"
17 #include "transaction.h"
19 #include "extent_io.h"
21 #include "space-info.h"
22 #include "delalloc-space.h"
23 #include "block-group.h"
26 #include "inode-item.h"
28 #define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
29 #define MAX_CACHE_BYTES_PER_GIG SZ_64K
30 #define FORCE_EXTENT_THRESHOLD SZ_1M
32 struct btrfs_trim_range {
35 struct list_head list;
38 static int link_free_space(struct btrfs_free_space_ctl *ctl,
39 struct btrfs_free_space *info);
40 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
41 struct btrfs_free_space *info, bool update_stat);
42 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
43 struct btrfs_free_space *bitmap_info, u64 *offset,
44 u64 *bytes, bool for_alloc);
45 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
46 struct btrfs_free_space *bitmap_info);
47 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
48 struct btrfs_free_space *info, u64 offset,
49 u64 bytes, bool update_stats);
51 static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
53 struct btrfs_free_space *info;
56 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
57 info = rb_entry(node, struct btrfs_free_space, offset_index);
59 unlink_free_space(ctl, info, true);
60 kmem_cache_free(btrfs_free_space_cachep, info);
62 free_bitmap(ctl, info);
65 cond_resched_lock(&ctl->tree_lock);
69 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
70 struct btrfs_path *path,
73 struct btrfs_fs_info *fs_info = root->fs_info;
75 struct btrfs_key location;
76 struct btrfs_disk_key disk_key;
77 struct btrfs_free_space_header *header;
78 struct extent_buffer *leaf;
79 struct inode *inode = NULL;
83 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
87 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
91 btrfs_release_path(path);
92 return ERR_PTR(-ENOENT);
95 leaf = path->nodes[0];
96 header = btrfs_item_ptr(leaf, path->slots[0],
97 struct btrfs_free_space_header);
98 btrfs_free_space_key(leaf, header, &disk_key);
99 btrfs_disk_key_to_cpu(&location, &disk_key);
100 btrfs_release_path(path);
103 * We are often under a trans handle at this point, so we need to make
104 * sure NOFS is set to keep us from deadlocking.
106 nofs_flag = memalloc_nofs_save();
107 inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
108 btrfs_release_path(path);
109 memalloc_nofs_restore(nofs_flag);
113 mapping_set_gfp_mask(inode->i_mapping,
114 mapping_gfp_constraint(inode->i_mapping,
115 ~(__GFP_FS | __GFP_HIGHMEM)));
120 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
121 struct btrfs_path *path)
123 struct btrfs_fs_info *fs_info = block_group->fs_info;
124 struct inode *inode = NULL;
125 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
127 spin_lock(&block_group->lock);
128 if (block_group->inode)
129 inode = igrab(block_group->inode);
130 spin_unlock(&block_group->lock);
134 inode = __lookup_free_space_inode(fs_info->tree_root, path,
139 spin_lock(&block_group->lock);
140 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
141 btrfs_info(fs_info, "Old style space inode found, converting.");
142 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
143 BTRFS_INODE_NODATACOW;
144 block_group->disk_cache_state = BTRFS_DC_CLEAR;
147 if (!test_and_set_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags))
148 block_group->inode = igrab(inode);
149 spin_unlock(&block_group->lock);
154 static int __create_free_space_inode(struct btrfs_root *root,
155 struct btrfs_trans_handle *trans,
156 struct btrfs_path *path,
159 struct btrfs_key key;
160 struct btrfs_disk_key disk_key;
161 struct btrfs_free_space_header *header;
162 struct btrfs_inode_item *inode_item;
163 struct extent_buffer *leaf;
164 /* We inline CRCs for the free disk space cache */
165 const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
166 BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
169 ret = btrfs_insert_empty_inode(trans, root, path, ino);
173 leaf = path->nodes[0];
174 inode_item = btrfs_item_ptr(leaf, path->slots[0],
175 struct btrfs_inode_item);
176 btrfs_item_key(leaf, &disk_key, path->slots[0]);
177 memzero_extent_buffer(leaf, (unsigned long)inode_item,
178 sizeof(*inode_item));
179 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
180 btrfs_set_inode_size(leaf, inode_item, 0);
181 btrfs_set_inode_nbytes(leaf, inode_item, 0);
182 btrfs_set_inode_uid(leaf, inode_item, 0);
183 btrfs_set_inode_gid(leaf, inode_item, 0);
184 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
185 btrfs_set_inode_flags(leaf, inode_item, flags);
186 btrfs_set_inode_nlink(leaf, inode_item, 1);
187 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
188 btrfs_set_inode_block_group(leaf, inode_item, offset);
189 btrfs_mark_buffer_dirty(leaf);
190 btrfs_release_path(path);
192 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
195 ret = btrfs_insert_empty_item(trans, root, path, &key,
196 sizeof(struct btrfs_free_space_header));
198 btrfs_release_path(path);
202 leaf = path->nodes[0];
203 header = btrfs_item_ptr(leaf, path->slots[0],
204 struct btrfs_free_space_header);
205 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
206 btrfs_set_free_space_key(leaf, header, &disk_key);
207 btrfs_mark_buffer_dirty(leaf);
208 btrfs_release_path(path);
213 int create_free_space_inode(struct btrfs_trans_handle *trans,
214 struct btrfs_block_group *block_group,
215 struct btrfs_path *path)
220 ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
224 return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
225 ino, block_group->start);
229 * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
230 * handles lookup, otherwise it takes ownership and iputs the inode.
231 * Don't reuse an inode pointer after passing it into this function.
233 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
235 struct btrfs_block_group *block_group)
237 struct btrfs_path *path;
238 struct btrfs_key key;
241 path = btrfs_alloc_path();
246 inode = lookup_free_space_inode(block_group, path);
248 if (PTR_ERR(inode) != -ENOENT)
249 ret = PTR_ERR(inode);
252 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
254 btrfs_add_delayed_iput(inode);
258 /* One for the block groups ref */
259 spin_lock(&block_group->lock);
260 if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags)) {
261 block_group->inode = NULL;
262 spin_unlock(&block_group->lock);
265 spin_unlock(&block_group->lock);
267 /* One for the lookup ref */
268 btrfs_add_delayed_iput(inode);
270 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
272 key.offset = block_group->start;
273 ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
280 ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
282 btrfs_free_path(path);
286 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
287 struct btrfs_block_rsv *rsv)
292 /* 1 for slack space, 1 for updating the inode */
293 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
294 btrfs_calc_metadata_size(fs_info, 1);
296 spin_lock(&rsv->lock);
297 if (rsv->reserved < needed_bytes)
301 spin_unlock(&rsv->lock);
305 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
306 struct btrfs_block_group *block_group,
307 struct inode *vfs_inode)
309 struct btrfs_truncate_control control = {
310 .inode = BTRFS_I(vfs_inode),
312 .ino = btrfs_ino(BTRFS_I(vfs_inode)),
313 .min_type = BTRFS_EXTENT_DATA_KEY,
314 .clear_extent_range = true,
316 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
317 struct btrfs_root *root = inode->root;
318 struct extent_state *cached_state = NULL;
323 struct btrfs_path *path = btrfs_alloc_path();
330 mutex_lock(&trans->transaction->cache_write_mutex);
331 if (!list_empty(&block_group->io_list)) {
332 list_del_init(&block_group->io_list);
334 btrfs_wait_cache_io(trans, block_group, path);
335 btrfs_put_block_group(block_group);
339 * now that we've truncated the cache away, its no longer
342 spin_lock(&block_group->lock);
343 block_group->disk_cache_state = BTRFS_DC_CLEAR;
344 spin_unlock(&block_group->lock);
345 btrfs_free_path(path);
348 btrfs_i_size_write(inode, 0);
349 truncate_pagecache(vfs_inode, 0);
351 lock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
352 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
355 * We skip the throttling logic for free space cache inodes, so we don't
356 * need to check for -EAGAIN.
358 ret = btrfs_truncate_inode_items(trans, root, &control);
360 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
361 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
363 unlock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
367 ret = btrfs_update_inode(trans, root, inode);
371 mutex_unlock(&trans->transaction->cache_write_mutex);
373 btrfs_abort_transaction(trans, ret);
378 static void readahead_cache(struct inode *inode)
380 struct file_ra_state ra;
381 unsigned long last_index;
383 file_ra_state_init(&ra, inode->i_mapping);
384 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
386 page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
389 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
394 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
396 /* Make sure we can fit our crcs and generation into the first page */
397 if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
400 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
402 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
406 io_ctl->num_pages = num_pages;
407 io_ctl->fs_info = btrfs_sb(inode->i_sb);
408 io_ctl->inode = inode;
412 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
414 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
416 kfree(io_ctl->pages);
417 io_ctl->pages = NULL;
420 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
428 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
430 ASSERT(io_ctl->index < io_ctl->num_pages);
431 io_ctl->page = io_ctl->pages[io_ctl->index++];
432 io_ctl->cur = page_address(io_ctl->page);
433 io_ctl->orig = io_ctl->cur;
434 io_ctl->size = PAGE_SIZE;
436 clear_page(io_ctl->cur);
439 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
443 io_ctl_unmap_page(io_ctl);
445 for (i = 0; i < io_ctl->num_pages; i++) {
446 if (io_ctl->pages[i]) {
447 btrfs_page_clear_checked(io_ctl->fs_info,
449 page_offset(io_ctl->pages[i]),
451 unlock_page(io_ctl->pages[i]);
452 put_page(io_ctl->pages[i]);
457 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
460 struct inode *inode = io_ctl->inode;
461 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
464 for (i = 0; i < io_ctl->num_pages; i++) {
467 page = find_or_create_page(inode->i_mapping, i, mask);
469 io_ctl_drop_pages(io_ctl);
473 ret = set_page_extent_mapped(page);
477 io_ctl_drop_pages(io_ctl);
481 io_ctl->pages[i] = page;
482 if (uptodate && !PageUptodate(page)) {
483 btrfs_read_folio(NULL, page_folio(page));
485 if (page->mapping != inode->i_mapping) {
486 btrfs_err(BTRFS_I(inode)->root->fs_info,
487 "free space cache page truncated");
488 io_ctl_drop_pages(io_ctl);
491 if (!PageUptodate(page)) {
492 btrfs_err(BTRFS_I(inode)->root->fs_info,
493 "error reading free space cache");
494 io_ctl_drop_pages(io_ctl);
500 for (i = 0; i < io_ctl->num_pages; i++)
501 clear_page_dirty_for_io(io_ctl->pages[i]);
506 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
508 io_ctl_map_page(io_ctl, 1);
511 * Skip the csum areas. If we don't check crcs then we just have a
512 * 64bit chunk at the front of the first page.
514 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
515 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
517 put_unaligned_le64(generation, io_ctl->cur);
518 io_ctl->cur += sizeof(u64);
521 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
526 * Skip the crc area. If we don't check crcs then we just have a 64bit
527 * chunk at the front of the first page.
529 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
530 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
532 cache_gen = get_unaligned_le64(io_ctl->cur);
533 if (cache_gen != generation) {
534 btrfs_err_rl(io_ctl->fs_info,
535 "space cache generation (%llu) does not match inode (%llu)",
536 cache_gen, generation);
537 io_ctl_unmap_page(io_ctl);
540 io_ctl->cur += sizeof(u64);
544 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
551 offset = sizeof(u32) * io_ctl->num_pages;
553 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
554 btrfs_crc32c_final(crc, (u8 *)&crc);
555 io_ctl_unmap_page(io_ctl);
556 tmp = page_address(io_ctl->pages[0]);
561 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
568 offset = sizeof(u32) * io_ctl->num_pages;
570 tmp = page_address(io_ctl->pages[0]);
574 io_ctl_map_page(io_ctl, 0);
575 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
576 btrfs_crc32c_final(crc, (u8 *)&crc);
578 btrfs_err_rl(io_ctl->fs_info,
579 "csum mismatch on free space cache");
580 io_ctl_unmap_page(io_ctl);
587 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
590 struct btrfs_free_space_entry *entry;
596 put_unaligned_le64(offset, &entry->offset);
597 put_unaligned_le64(bytes, &entry->bytes);
598 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
599 BTRFS_FREE_SPACE_EXTENT;
600 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
601 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
603 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
606 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
608 /* No more pages to map */
609 if (io_ctl->index >= io_ctl->num_pages)
612 /* map the next page */
613 io_ctl_map_page(io_ctl, 1);
617 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
623 * If we aren't at the start of the current page, unmap this one and
624 * map the next one if there is any left.
626 if (io_ctl->cur != io_ctl->orig) {
627 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
628 if (io_ctl->index >= io_ctl->num_pages)
630 io_ctl_map_page(io_ctl, 0);
633 copy_page(io_ctl->cur, bitmap);
634 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
635 if (io_ctl->index < io_ctl->num_pages)
636 io_ctl_map_page(io_ctl, 0);
640 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
643 * If we're not on the boundary we know we've modified the page and we
644 * need to crc the page.
646 if (io_ctl->cur != io_ctl->orig)
647 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
649 io_ctl_unmap_page(io_ctl);
651 while (io_ctl->index < io_ctl->num_pages) {
652 io_ctl_map_page(io_ctl, 1);
653 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
657 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
658 struct btrfs_free_space *entry, u8 *type)
660 struct btrfs_free_space_entry *e;
664 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
670 entry->offset = get_unaligned_le64(&e->offset);
671 entry->bytes = get_unaligned_le64(&e->bytes);
673 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
674 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
676 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
679 io_ctl_unmap_page(io_ctl);
684 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
685 struct btrfs_free_space *entry)
689 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
693 copy_page(entry->bitmap, io_ctl->cur);
694 io_ctl_unmap_page(io_ctl);
699 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
701 struct btrfs_block_group *block_group = ctl->block_group;
705 u64 size = block_group->length;
706 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
707 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
709 max_bitmaps = max_t(u64, max_bitmaps, 1);
711 if (ctl->total_bitmaps > max_bitmaps)
712 btrfs_err(block_group->fs_info,
713 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
714 block_group->start, block_group->length,
715 ctl->total_bitmaps, ctl->unit, max_bitmaps,
717 ASSERT(ctl->total_bitmaps <= max_bitmaps);
720 * We are trying to keep the total amount of memory used per 1GiB of
721 * space to be MAX_CACHE_BYTES_PER_GIG. However, with a reclamation
722 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
723 * bitmaps, we may end up using more memory than this.
726 max_bytes = MAX_CACHE_BYTES_PER_GIG;
728 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
730 bitmap_bytes = ctl->total_bitmaps * ctl->unit;
733 * we want the extent entry threshold to always be at most 1/2 the max
734 * bytes we can have, or whatever is less than that.
736 extent_bytes = max_bytes - bitmap_bytes;
737 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
739 ctl->extents_thresh =
740 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
743 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
744 struct btrfs_free_space_ctl *ctl,
745 struct btrfs_path *path, u64 offset)
747 struct btrfs_fs_info *fs_info = root->fs_info;
748 struct btrfs_free_space_header *header;
749 struct extent_buffer *leaf;
750 struct btrfs_io_ctl io_ctl;
751 struct btrfs_key key;
752 struct btrfs_free_space *e, *n;
760 /* Nothing in the space cache, goodbye */
761 if (!i_size_read(inode))
764 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
768 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
772 btrfs_release_path(path);
778 leaf = path->nodes[0];
779 header = btrfs_item_ptr(leaf, path->slots[0],
780 struct btrfs_free_space_header);
781 num_entries = btrfs_free_space_entries(leaf, header);
782 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
783 generation = btrfs_free_space_generation(leaf, header);
784 btrfs_release_path(path);
786 if (!BTRFS_I(inode)->generation) {
788 "the free space cache file (%llu) is invalid, skip it",
793 if (BTRFS_I(inode)->generation != generation) {
795 "free space inode generation (%llu) did not match free space cache generation (%llu)",
796 BTRFS_I(inode)->generation, generation);
803 ret = io_ctl_init(&io_ctl, inode, 0);
807 readahead_cache(inode);
809 ret = io_ctl_prepare_pages(&io_ctl, true);
813 ret = io_ctl_check_crc(&io_ctl, 0);
817 ret = io_ctl_check_generation(&io_ctl, generation);
821 while (num_entries) {
822 e = kmem_cache_zalloc(btrfs_free_space_cachep,
829 ret = io_ctl_read_entry(&io_ctl, e, &type);
831 kmem_cache_free(btrfs_free_space_cachep, e);
837 kmem_cache_free(btrfs_free_space_cachep, e);
841 if (type == BTRFS_FREE_SPACE_EXTENT) {
842 spin_lock(&ctl->tree_lock);
843 ret = link_free_space(ctl, e);
844 spin_unlock(&ctl->tree_lock);
847 "Duplicate entries in free space cache, dumping");
848 kmem_cache_free(btrfs_free_space_cachep, e);
854 e->bitmap = kmem_cache_zalloc(
855 btrfs_free_space_bitmap_cachep, GFP_NOFS);
859 btrfs_free_space_cachep, e);
862 spin_lock(&ctl->tree_lock);
863 ret = link_free_space(ctl, e);
865 spin_unlock(&ctl->tree_lock);
867 "Duplicate entries in free space cache, dumping");
868 kmem_cache_free(btrfs_free_space_cachep, e);
871 ctl->total_bitmaps++;
872 recalculate_thresholds(ctl);
873 spin_unlock(&ctl->tree_lock);
874 list_add_tail(&e->list, &bitmaps);
880 io_ctl_unmap_page(&io_ctl);
883 * We add the bitmaps at the end of the entries in order that
884 * the bitmap entries are added to the cache.
886 list_for_each_entry_safe(e, n, &bitmaps, list) {
887 list_del_init(&e->list);
888 ret = io_ctl_read_bitmap(&io_ctl, e);
893 io_ctl_drop_pages(&io_ctl);
896 io_ctl_free(&io_ctl);
899 io_ctl_drop_pages(&io_ctl);
901 spin_lock(&ctl->tree_lock);
902 __btrfs_remove_free_space_cache(ctl);
903 spin_unlock(&ctl->tree_lock);
907 static int copy_free_space_cache(struct btrfs_block_group *block_group,
908 struct btrfs_free_space_ctl *ctl)
910 struct btrfs_free_space *info;
914 while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
915 info = rb_entry(n, struct btrfs_free_space, offset_index);
917 unlink_free_space(ctl, info, true);
918 ret = btrfs_add_free_space(block_group, info->offset,
920 kmem_cache_free(btrfs_free_space_cachep, info);
922 u64 offset = info->offset;
923 u64 bytes = ctl->unit;
925 while (search_bitmap(ctl, info, &offset, &bytes,
927 ret = btrfs_add_free_space(block_group, offset,
931 bitmap_clear_bits(ctl, info, offset, bytes, true);
932 offset = info->offset;
935 free_bitmap(ctl, info);
942 static struct lock_class_key btrfs_free_space_inode_key;
944 int load_free_space_cache(struct btrfs_block_group *block_group)
946 struct btrfs_fs_info *fs_info = block_group->fs_info;
947 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
948 struct btrfs_free_space_ctl tmp_ctl = {};
950 struct btrfs_path *path;
953 u64 used = block_group->used;
956 * Because we could potentially discard our loaded free space, we want
957 * to load everything into a temporary structure first, and then if it's
958 * valid copy it all into the actual free space ctl.
960 btrfs_init_free_space_ctl(block_group, &tmp_ctl);
963 * If this block group has been marked to be cleared for one reason or
964 * another then we can't trust the on disk cache, so just return.
966 spin_lock(&block_group->lock);
967 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
968 spin_unlock(&block_group->lock);
971 spin_unlock(&block_group->lock);
973 path = btrfs_alloc_path();
976 path->search_commit_root = 1;
977 path->skip_locking = 1;
980 * We must pass a path with search_commit_root set to btrfs_iget in
981 * order to avoid a deadlock when allocating extents for the tree root.
983 * When we are COWing an extent buffer from the tree root, when looking
984 * for a free extent, at extent-tree.c:find_free_extent(), we can find
985 * block group without its free space cache loaded. When we find one
986 * we must load its space cache which requires reading its free space
987 * cache's inode item from the root tree. If this inode item is located
988 * in the same leaf that we started COWing before, then we end up in
989 * deadlock on the extent buffer (trying to read lock it when we
990 * previously write locked it).
992 * It's safe to read the inode item using the commit root because
993 * block groups, once loaded, stay in memory forever (until they are
994 * removed) as well as their space caches once loaded. New block groups
995 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
996 * we will never try to read their inode item while the fs is mounted.
998 inode = lookup_free_space_inode(block_group, path);
1000 btrfs_free_path(path);
1004 /* We may have converted the inode and made the cache invalid. */
1005 spin_lock(&block_group->lock);
1006 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1007 spin_unlock(&block_group->lock);
1008 btrfs_free_path(path);
1011 spin_unlock(&block_group->lock);
1014 * Reinitialize the class of struct inode's mapping->invalidate_lock for
1015 * free space inodes to prevent false positives related to locks for normal
1018 lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1019 &btrfs_free_space_inode_key);
1021 ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
1022 path, block_group->start);
1023 btrfs_free_path(path);
1027 matched = (tmp_ctl.free_space == (block_group->length - used -
1028 block_group->bytes_super));
1031 ret = copy_free_space_cache(block_group, &tmp_ctl);
1033 * ret == 1 means we successfully loaded the free space cache,
1034 * so we need to re-set it here.
1040 * We need to call the _locked variant so we don't try to update
1041 * the discard counters.
1043 spin_lock(&tmp_ctl.tree_lock);
1044 __btrfs_remove_free_space_cache(&tmp_ctl);
1045 spin_unlock(&tmp_ctl.tree_lock);
1047 "block group %llu has wrong amount of free space",
1048 block_group->start);
1053 /* This cache is bogus, make sure it gets cleared */
1054 spin_lock(&block_group->lock);
1055 block_group->disk_cache_state = BTRFS_DC_CLEAR;
1056 spin_unlock(&block_group->lock);
1060 "failed to load free space cache for block group %llu, rebuilding it now",
1061 block_group->start);
1064 spin_lock(&ctl->tree_lock);
1065 btrfs_discard_update_discardable(block_group);
1066 spin_unlock(&ctl->tree_lock);
1071 static noinline_for_stack
1072 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1073 struct btrfs_free_space_ctl *ctl,
1074 struct btrfs_block_group *block_group,
1075 int *entries, int *bitmaps,
1076 struct list_head *bitmap_list)
1079 struct btrfs_free_cluster *cluster = NULL;
1080 struct btrfs_free_cluster *cluster_locked = NULL;
1081 struct rb_node *node = rb_first(&ctl->free_space_offset);
1082 struct btrfs_trim_range *trim_entry;
1084 /* Get the cluster for this block_group if it exists */
1085 if (block_group && !list_empty(&block_group->cluster_list)) {
1086 cluster = list_entry(block_group->cluster_list.next,
1087 struct btrfs_free_cluster,
1091 if (!node && cluster) {
1092 cluster_locked = cluster;
1093 spin_lock(&cluster_locked->lock);
1094 node = rb_first(&cluster->root);
1098 /* Write out the extent entries */
1100 struct btrfs_free_space *e;
1102 e = rb_entry(node, struct btrfs_free_space, offset_index);
1105 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1111 list_add_tail(&e->list, bitmap_list);
1114 node = rb_next(node);
1115 if (!node && cluster) {
1116 node = rb_first(&cluster->root);
1117 cluster_locked = cluster;
1118 spin_lock(&cluster_locked->lock);
1122 if (cluster_locked) {
1123 spin_unlock(&cluster_locked->lock);
1124 cluster_locked = NULL;
1128 * Make sure we don't miss any range that was removed from our rbtree
1129 * because trimming is running. Otherwise after a umount+mount (or crash
1130 * after committing the transaction) we would leak free space and get
1131 * an inconsistent free space cache report from fsck.
1133 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1134 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1135 trim_entry->bytes, NULL);
1144 spin_unlock(&cluster_locked->lock);
1148 static noinline_for_stack int
1149 update_cache_item(struct btrfs_trans_handle *trans,
1150 struct btrfs_root *root,
1151 struct inode *inode,
1152 struct btrfs_path *path, u64 offset,
1153 int entries, int bitmaps)
1155 struct btrfs_key key;
1156 struct btrfs_free_space_header *header;
1157 struct extent_buffer *leaf;
1160 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1161 key.offset = offset;
1164 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1166 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1167 EXTENT_DELALLOC, NULL);
1170 leaf = path->nodes[0];
1172 struct btrfs_key found_key;
1173 ASSERT(path->slots[0]);
1175 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1176 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1177 found_key.offset != offset) {
1178 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1179 inode->i_size - 1, EXTENT_DELALLOC,
1181 btrfs_release_path(path);
1186 BTRFS_I(inode)->generation = trans->transid;
1187 header = btrfs_item_ptr(leaf, path->slots[0],
1188 struct btrfs_free_space_header);
1189 btrfs_set_free_space_entries(leaf, header, entries);
1190 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1191 btrfs_set_free_space_generation(leaf, header, trans->transid);
1192 btrfs_mark_buffer_dirty(leaf);
1193 btrfs_release_path(path);
1201 static noinline_for_stack int write_pinned_extent_entries(
1202 struct btrfs_trans_handle *trans,
1203 struct btrfs_block_group *block_group,
1204 struct btrfs_io_ctl *io_ctl,
1207 u64 start, extent_start, extent_end, len;
1208 struct extent_io_tree *unpin = NULL;
1215 * We want to add any pinned extents to our free space cache
1216 * so we don't leak the space
1218 * We shouldn't have switched the pinned extents yet so this is the
1221 unpin = &trans->transaction->pinned_extents;
1223 start = block_group->start;
1225 while (start < block_group->start + block_group->length) {
1226 ret = find_first_extent_bit(unpin, start,
1227 &extent_start, &extent_end,
1228 EXTENT_DIRTY, NULL);
1232 /* This pinned extent is out of our range */
1233 if (extent_start >= block_group->start + block_group->length)
1236 extent_start = max(extent_start, start);
1237 extent_end = min(block_group->start + block_group->length,
1239 len = extent_end - extent_start;
1242 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1252 static noinline_for_stack int
1253 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1255 struct btrfs_free_space *entry, *next;
1258 /* Write out the bitmaps */
1259 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1260 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1263 list_del_init(&entry->list);
1269 static int flush_dirty_cache(struct inode *inode)
1273 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1275 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1276 EXTENT_DELALLOC, NULL);
1281 static void noinline_for_stack
1282 cleanup_bitmap_list(struct list_head *bitmap_list)
1284 struct btrfs_free_space *entry, *next;
1286 list_for_each_entry_safe(entry, next, bitmap_list, list)
1287 list_del_init(&entry->list);
1290 static void noinline_for_stack
1291 cleanup_write_cache_enospc(struct inode *inode,
1292 struct btrfs_io_ctl *io_ctl,
1293 struct extent_state **cached_state)
1295 io_ctl_drop_pages(io_ctl);
1296 unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1300 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1301 struct btrfs_trans_handle *trans,
1302 struct btrfs_block_group *block_group,
1303 struct btrfs_io_ctl *io_ctl,
1304 struct btrfs_path *path, u64 offset)
1307 struct inode *inode = io_ctl->inode;
1312 /* Flush the dirty pages in the cache file. */
1313 ret = flush_dirty_cache(inode);
1317 /* Update the cache item to tell everyone this cache file is valid. */
1318 ret = update_cache_item(trans, root, inode, path, offset,
1319 io_ctl->entries, io_ctl->bitmaps);
1322 invalidate_inode_pages2(inode->i_mapping);
1323 BTRFS_I(inode)->generation = 0;
1325 btrfs_debug(root->fs_info,
1326 "failed to write free space cache for block group %llu error %d",
1327 block_group->start, ret);
1329 btrfs_update_inode(trans, root, BTRFS_I(inode));
1332 /* the dirty list is protected by the dirty_bgs_lock */
1333 spin_lock(&trans->transaction->dirty_bgs_lock);
1335 /* the disk_cache_state is protected by the block group lock */
1336 spin_lock(&block_group->lock);
1339 * only mark this as written if we didn't get put back on
1340 * the dirty list while waiting for IO. Otherwise our
1341 * cache state won't be right, and we won't get written again
1343 if (!ret && list_empty(&block_group->dirty_list))
1344 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1346 block_group->disk_cache_state = BTRFS_DC_ERROR;
1348 spin_unlock(&block_group->lock);
1349 spin_unlock(&trans->transaction->dirty_bgs_lock);
1350 io_ctl->inode = NULL;
1358 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1359 struct btrfs_block_group *block_group,
1360 struct btrfs_path *path)
1362 return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1363 block_group, &block_group->io_ctl,
1364 path, block_group->start);
1368 * Write out cached info to an inode
1370 * @root: root the inode belongs to
1371 * @inode: freespace inode we are writing out
1372 * @ctl: free space cache we are going to write out
1373 * @block_group: block_group for this cache if it belongs to a block_group
1374 * @io_ctl: holds context for the io
1375 * @trans: the trans handle
1377 * This function writes out a free space cache struct to disk for quick recovery
1378 * on mount. This will return 0 if it was successful in writing the cache out,
1379 * or an errno if it was not.
1381 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1382 struct btrfs_free_space_ctl *ctl,
1383 struct btrfs_block_group *block_group,
1384 struct btrfs_io_ctl *io_ctl,
1385 struct btrfs_trans_handle *trans)
1387 struct extent_state *cached_state = NULL;
1388 LIST_HEAD(bitmap_list);
1394 if (!i_size_read(inode))
1397 WARN_ON(io_ctl->pages);
1398 ret = io_ctl_init(io_ctl, inode, 1);
1402 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1403 down_write(&block_group->data_rwsem);
1404 spin_lock(&block_group->lock);
1405 if (block_group->delalloc_bytes) {
1406 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1407 spin_unlock(&block_group->lock);
1408 up_write(&block_group->data_rwsem);
1409 BTRFS_I(inode)->generation = 0;
1414 spin_unlock(&block_group->lock);
1417 /* Lock all pages first so we can lock the extent safely. */
1418 ret = io_ctl_prepare_pages(io_ctl, false);
1422 lock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1425 io_ctl_set_generation(io_ctl, trans->transid);
1427 mutex_lock(&ctl->cache_writeout_mutex);
1428 /* Write out the extent entries in the free space cache */
1429 spin_lock(&ctl->tree_lock);
1430 ret = write_cache_extent_entries(io_ctl, ctl,
1431 block_group, &entries, &bitmaps,
1434 goto out_nospc_locked;
1437 * Some spaces that are freed in the current transaction are pinned,
1438 * they will be added into free space cache after the transaction is
1439 * committed, we shouldn't lose them.
1441 * If this changes while we are working we'll get added back to
1442 * the dirty list and redo it. No locking needed
1444 ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1446 goto out_nospc_locked;
1449 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1450 * locked while doing it because a concurrent trim can be manipulating
1451 * or freeing the bitmap.
1453 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1454 spin_unlock(&ctl->tree_lock);
1455 mutex_unlock(&ctl->cache_writeout_mutex);
1459 /* Zero out the rest of the pages just to make sure */
1460 io_ctl_zero_remaining_pages(io_ctl);
1462 /* Everything is written out, now we dirty the pages in the file. */
1463 ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1464 io_ctl->num_pages, 0, i_size_read(inode),
1465 &cached_state, false);
1469 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1470 up_write(&block_group->data_rwsem);
1472 * Release the pages and unlock the extent, we will flush
1475 io_ctl_drop_pages(io_ctl);
1476 io_ctl_free(io_ctl);
1478 unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1482 * at this point the pages are under IO and we're happy,
1483 * The caller is responsible for waiting on them and updating
1484 * the cache and the inode
1486 io_ctl->entries = entries;
1487 io_ctl->bitmaps = bitmaps;
1489 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1496 cleanup_bitmap_list(&bitmap_list);
1497 spin_unlock(&ctl->tree_lock);
1498 mutex_unlock(&ctl->cache_writeout_mutex);
1501 cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1504 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1505 up_write(&block_group->data_rwsem);
1508 io_ctl->inode = NULL;
1509 io_ctl_free(io_ctl);
1511 invalidate_inode_pages2(inode->i_mapping);
1512 BTRFS_I(inode)->generation = 0;
1514 btrfs_update_inode(trans, root, BTRFS_I(inode));
1520 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1521 struct btrfs_block_group *block_group,
1522 struct btrfs_path *path)
1524 struct btrfs_fs_info *fs_info = trans->fs_info;
1525 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1526 struct inode *inode;
1529 spin_lock(&block_group->lock);
1530 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1531 spin_unlock(&block_group->lock);
1534 spin_unlock(&block_group->lock);
1536 inode = lookup_free_space_inode(block_group, path);
1540 ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1541 block_group, &block_group->io_ctl, trans);
1543 btrfs_debug(fs_info,
1544 "failed to write free space cache for block group %llu error %d",
1545 block_group->start, ret);
1546 spin_lock(&block_group->lock);
1547 block_group->disk_cache_state = BTRFS_DC_ERROR;
1548 spin_unlock(&block_group->lock);
1550 block_group->io_ctl.inode = NULL;
1555 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1556 * to wait for IO and put the inode
1562 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1565 ASSERT(offset >= bitmap_start);
1566 offset -= bitmap_start;
1567 return (unsigned long)(div_u64(offset, unit));
1570 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1572 return (unsigned long)(div_u64(bytes, unit));
1575 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1579 u64 bytes_per_bitmap;
1581 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1582 bitmap_start = offset - ctl->start;
1583 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1584 bitmap_start *= bytes_per_bitmap;
1585 bitmap_start += ctl->start;
1587 return bitmap_start;
1590 static int tree_insert_offset(struct rb_root *root, u64 offset,
1591 struct rb_node *node, int bitmap)
1593 struct rb_node **p = &root->rb_node;
1594 struct rb_node *parent = NULL;
1595 struct btrfs_free_space *info;
1599 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1601 if (offset < info->offset) {
1603 } else if (offset > info->offset) {
1604 p = &(*p)->rb_right;
1607 * we could have a bitmap entry and an extent entry
1608 * share the same offset. If this is the case, we want
1609 * the extent entry to always be found first if we do a
1610 * linear search through the tree, since we want to have
1611 * the quickest allocation time, and allocating from an
1612 * extent is faster than allocating from a bitmap. So
1613 * if we're inserting a bitmap and we find an entry at
1614 * this offset, we want to go right, or after this entry
1615 * logically. If we are inserting an extent and we've
1616 * found a bitmap, we want to go left, or before
1624 p = &(*p)->rb_right;
1626 if (!info->bitmap) {
1635 rb_link_node(node, parent, p);
1636 rb_insert_color(node, root);
1642 * This is a little subtle. We *only* have ->max_extent_size set if we actually
1643 * searched through the bitmap and figured out the largest ->max_extent_size,
1644 * otherwise it's 0. In the case that it's 0 we don't want to tell the
1645 * allocator the wrong thing, we want to use the actual real max_extent_size
1646 * we've found already if it's larger, or we want to use ->bytes.
1648 * This matters because find_free_space() will skip entries who's ->bytes is
1649 * less than the required bytes. So if we didn't search down this bitmap, we
1650 * may pick some previous entry that has a smaller ->max_extent_size than we
1651 * have. For example, assume we have two entries, one that has
1652 * ->max_extent_size set to 4K and ->bytes set to 1M. A second entry hasn't set
1653 * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous. We will
1654 * call into find_free_space(), and return with max_extent_size == 4K, because
1655 * that first bitmap entry had ->max_extent_size set, but the second one did
1656 * not. If instead we returned 8K we'd come in searching for 8K, and find the
1657 * 8K contiguous range.
1659 * Consider the other case, we have 2 8K chunks in that second entry and still
1660 * don't have ->max_extent_size set. We'll return 16K, and the next time the
1661 * allocator comes in it'll fully search our second bitmap, and this time it'll
1662 * get an uptodate value of 8K as the maximum chunk size. Then we'll get the
1663 * right allocation the next loop through.
1665 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1667 if (entry->bitmap && entry->max_extent_size)
1668 return entry->max_extent_size;
1669 return entry->bytes;
1673 * We want the largest entry to be leftmost, so this is inverted from what you'd
1676 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1678 const struct btrfs_free_space *entry, *exist;
1680 entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1681 exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1682 return get_max_extent_size(exist) < get_max_extent_size(entry);
1686 * searches the tree for the given offset.
1688 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1689 * want a section that has at least bytes size and comes at or after the given
1692 static struct btrfs_free_space *
1693 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1694 u64 offset, int bitmap_only, int fuzzy)
1696 struct rb_node *n = ctl->free_space_offset.rb_node;
1697 struct btrfs_free_space *entry = NULL, *prev = NULL;
1699 /* find entry that is closest to the 'offset' */
1701 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1704 if (offset < entry->offset)
1706 else if (offset > entry->offset)
1721 * bitmap entry and extent entry may share same offset,
1722 * in that case, bitmap entry comes after extent entry.
1727 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1728 if (entry->offset != offset)
1731 WARN_ON(!entry->bitmap);
1734 if (entry->bitmap) {
1736 * if previous extent entry covers the offset,
1737 * we should return it instead of the bitmap entry
1739 n = rb_prev(&entry->offset_index);
1741 prev = rb_entry(n, struct btrfs_free_space,
1743 if (!prev->bitmap &&
1744 prev->offset + prev->bytes > offset)
1754 /* find last entry before the 'offset' */
1756 if (entry->offset > offset) {
1757 n = rb_prev(&entry->offset_index);
1759 entry = rb_entry(n, struct btrfs_free_space,
1761 ASSERT(entry->offset <= offset);
1770 if (entry->bitmap) {
1771 n = rb_prev(&entry->offset_index);
1773 prev = rb_entry(n, struct btrfs_free_space,
1775 if (!prev->bitmap &&
1776 prev->offset + prev->bytes > offset)
1779 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1781 } else if (entry->offset + entry->bytes > offset)
1788 n = rb_next(&entry->offset_index);
1791 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1792 if (entry->bitmap) {
1793 if (entry->offset + BITS_PER_BITMAP *
1797 if (entry->offset + entry->bytes > offset)
1804 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1805 struct btrfs_free_space *info,
1808 rb_erase(&info->offset_index, &ctl->free_space_offset);
1809 rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1810 ctl->free_extents--;
1812 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1813 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1814 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1818 ctl->free_space -= info->bytes;
1821 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1822 struct btrfs_free_space *info)
1826 ASSERT(info->bytes || info->bitmap);
1827 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1828 &info->offset_index, (info->bitmap != NULL));
1832 rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1834 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1835 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1836 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1839 ctl->free_space += info->bytes;
1840 ctl->free_extents++;
1844 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1845 struct btrfs_free_space *info)
1847 ASSERT(info->bitmap);
1850 * If our entry is empty it's because we're on a cluster and we don't
1851 * want to re-link it into our ctl bytes index.
1853 if (RB_EMPTY_NODE(&info->bytes_index))
1856 rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1857 rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1860 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1861 struct btrfs_free_space *info,
1862 u64 offset, u64 bytes, bool update_stat)
1864 unsigned long start, count, end;
1865 int extent_delta = -1;
1867 start = offset_to_bit(info->offset, ctl->unit, offset);
1868 count = bytes_to_bits(bytes, ctl->unit);
1869 end = start + count;
1870 ASSERT(end <= BITS_PER_BITMAP);
1872 bitmap_clear(info->bitmap, start, count);
1874 info->bytes -= bytes;
1875 if (info->max_extent_size > ctl->unit)
1876 info->max_extent_size = 0;
1878 relink_bitmap_entry(ctl, info);
1880 if (start && test_bit(start - 1, info->bitmap))
1883 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1886 info->bitmap_extents += extent_delta;
1887 if (!btrfs_free_space_trimmed(info)) {
1888 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1889 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1893 ctl->free_space -= bytes;
1896 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1897 struct btrfs_free_space *info, u64 offset,
1900 unsigned long start, count, end;
1901 int extent_delta = 1;
1903 start = offset_to_bit(info->offset, ctl->unit, offset);
1904 count = bytes_to_bits(bytes, ctl->unit);
1905 end = start + count;
1906 ASSERT(end <= BITS_PER_BITMAP);
1908 bitmap_set(info->bitmap, start, count);
1911 * We set some bytes, we have no idea what the max extent size is
1914 info->max_extent_size = 0;
1915 info->bytes += bytes;
1916 ctl->free_space += bytes;
1918 relink_bitmap_entry(ctl, info);
1920 if (start && test_bit(start - 1, info->bitmap))
1923 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1926 info->bitmap_extents += extent_delta;
1927 if (!btrfs_free_space_trimmed(info)) {
1928 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1929 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1934 * If we can not find suitable extent, we will use bytes to record
1935 * the size of the max extent.
1937 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1938 struct btrfs_free_space *bitmap_info, u64 *offset,
1939 u64 *bytes, bool for_alloc)
1941 unsigned long found_bits = 0;
1942 unsigned long max_bits = 0;
1943 unsigned long bits, i;
1944 unsigned long next_zero;
1945 unsigned long extent_bits;
1948 * Skip searching the bitmap if we don't have a contiguous section that
1949 * is large enough for this allocation.
1952 bitmap_info->max_extent_size &&
1953 bitmap_info->max_extent_size < *bytes) {
1954 *bytes = bitmap_info->max_extent_size;
1958 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1959 max_t(u64, *offset, bitmap_info->offset));
1960 bits = bytes_to_bits(*bytes, ctl->unit);
1962 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1963 if (for_alloc && bits == 1) {
1967 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1968 BITS_PER_BITMAP, i);
1969 extent_bits = next_zero - i;
1970 if (extent_bits >= bits) {
1971 found_bits = extent_bits;
1973 } else if (extent_bits > max_bits) {
1974 max_bits = extent_bits;
1980 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1981 *bytes = (u64)(found_bits) * ctl->unit;
1985 *bytes = (u64)(max_bits) * ctl->unit;
1986 bitmap_info->max_extent_size = *bytes;
1987 relink_bitmap_entry(ctl, bitmap_info);
1991 /* Cache the size of the max extent in bytes */
1992 static struct btrfs_free_space *
1993 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1994 unsigned long align, u64 *max_extent_size, bool use_bytes_index)
1996 struct btrfs_free_space *entry;
1997 struct rb_node *node;
2002 if (!ctl->free_space_offset.rb_node)
2005 if (use_bytes_index) {
2006 node = rb_first_cached(&ctl->free_space_bytes);
2008 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
2012 node = &entry->offset_index;
2015 for (; node; node = rb_next(node)) {
2016 if (use_bytes_index)
2017 entry = rb_entry(node, struct btrfs_free_space,
2020 entry = rb_entry(node, struct btrfs_free_space,
2024 * If we are using the bytes index then all subsequent entries
2025 * in this tree are going to be < bytes, so simply set the max
2026 * extent size and exit the loop.
2028 * If we're using the offset index then we need to keep going
2029 * through the rest of the tree.
2031 if (entry->bytes < *bytes) {
2032 *max_extent_size = max(get_max_extent_size(entry),
2034 if (use_bytes_index)
2039 /* make sure the space returned is big enough
2040 * to match our requested alignment
2042 if (*bytes >= align) {
2043 tmp = entry->offset - ctl->start + align - 1;
2044 tmp = div64_u64(tmp, align);
2045 tmp = tmp * align + ctl->start;
2046 align_off = tmp - entry->offset;
2049 tmp = entry->offset;
2053 * We don't break here if we're using the bytes index because we
2054 * may have another entry that has the correct alignment that is
2055 * the right size, so we don't want to miss that possibility.
2056 * At worst this adds another loop through the logic, but if we
2057 * broke here we could prematurely ENOSPC.
2059 if (entry->bytes < *bytes + align_off) {
2060 *max_extent_size = max(get_max_extent_size(entry),
2065 if (entry->bitmap) {
2066 struct rb_node *old_next = rb_next(node);
2069 ret = search_bitmap(ctl, entry, &tmp, &size, true);
2076 max(get_max_extent_size(entry),
2081 * The bitmap may have gotten re-arranged in the space
2082 * index here because the max_extent_size may have been
2083 * updated. Start from the beginning again if this
2086 if (use_bytes_index && old_next != rb_next(node))
2092 *bytes = entry->bytes - align_off;
2099 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2100 struct btrfs_free_space *info, u64 offset)
2102 info->offset = offset_to_bitmap(ctl, offset);
2104 info->bitmap_extents = 0;
2105 INIT_LIST_HEAD(&info->list);
2106 link_free_space(ctl, info);
2107 ctl->total_bitmaps++;
2108 recalculate_thresholds(ctl);
2111 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2112 struct btrfs_free_space *bitmap_info)
2115 * Normally when this is called, the bitmap is completely empty. However,
2116 * if we are blowing up the free space cache for one reason or another
2117 * via __btrfs_remove_free_space_cache(), then it may not be freed and
2118 * we may leave stats on the table.
2120 if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2121 ctl->discardable_extents[BTRFS_STAT_CURR] -=
2122 bitmap_info->bitmap_extents;
2123 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2126 unlink_free_space(ctl, bitmap_info, true);
2127 kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2128 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2129 ctl->total_bitmaps--;
2130 recalculate_thresholds(ctl);
2133 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2134 struct btrfs_free_space *bitmap_info,
2135 u64 *offset, u64 *bytes)
2138 u64 search_start, search_bytes;
2142 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2145 * We need to search for bits in this bitmap. We could only cover some
2146 * of the extent in this bitmap thanks to how we add space, so we need
2147 * to search for as much as it as we can and clear that amount, and then
2148 * go searching for the next bit.
2150 search_start = *offset;
2151 search_bytes = ctl->unit;
2152 search_bytes = min(search_bytes, end - search_start + 1);
2153 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2155 if (ret < 0 || search_start != *offset)
2158 /* We may have found more bits than what we need */
2159 search_bytes = min(search_bytes, *bytes);
2161 /* Cannot clear past the end of the bitmap */
2162 search_bytes = min(search_bytes, end - search_start + 1);
2164 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2165 *offset += search_bytes;
2166 *bytes -= search_bytes;
2169 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2170 if (!bitmap_info->bytes)
2171 free_bitmap(ctl, bitmap_info);
2174 * no entry after this bitmap, but we still have bytes to
2175 * remove, so something has gone wrong.
2180 bitmap_info = rb_entry(next, struct btrfs_free_space,
2184 * if the next entry isn't a bitmap we need to return to let the
2185 * extent stuff do its work.
2187 if (!bitmap_info->bitmap)
2191 * Ok the next item is a bitmap, but it may not actually hold
2192 * the information for the rest of this free space stuff, so
2193 * look for it, and if we don't find it return so we can try
2194 * everything over again.
2196 search_start = *offset;
2197 search_bytes = ctl->unit;
2198 ret = search_bitmap(ctl, bitmap_info, &search_start,
2199 &search_bytes, false);
2200 if (ret < 0 || search_start != *offset)
2204 } else if (!bitmap_info->bytes)
2205 free_bitmap(ctl, bitmap_info);
2210 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2211 struct btrfs_free_space *info, u64 offset,
2212 u64 bytes, enum btrfs_trim_state trim_state)
2214 u64 bytes_to_set = 0;
2218 * This is a tradeoff to make bitmap trim state minimal. We mark the
2219 * whole bitmap untrimmed if at any point we add untrimmed regions.
2221 if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2222 if (btrfs_free_space_trimmed(info)) {
2223 ctl->discardable_extents[BTRFS_STAT_CURR] +=
2224 info->bitmap_extents;
2225 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2227 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2230 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2232 bytes_to_set = min(end - offset, bytes);
2234 bitmap_set_bits(ctl, info, offset, bytes_to_set);
2236 return bytes_to_set;
2240 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2241 struct btrfs_free_space *info)
2243 struct btrfs_block_group *block_group = ctl->block_group;
2244 struct btrfs_fs_info *fs_info = block_group->fs_info;
2245 bool forced = false;
2247 #ifdef CONFIG_BTRFS_DEBUG
2248 if (btrfs_should_fragment_free_space(block_group))
2252 /* This is a way to reclaim large regions from the bitmaps. */
2253 if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2257 * If we are below the extents threshold then we can add this as an
2258 * extent, and don't have to deal with the bitmap
2260 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2262 * If this block group has some small extents we don't want to
2263 * use up all of our free slots in the cache with them, we want
2264 * to reserve them to larger extents, however if we have plenty
2265 * of cache left then go ahead an dadd them, no sense in adding
2266 * the overhead of a bitmap if we don't have to.
2268 if (info->bytes <= fs_info->sectorsize * 8) {
2269 if (ctl->free_extents * 3 <= ctl->extents_thresh)
2277 * The original block groups from mkfs can be really small, like 8
2278 * megabytes, so don't bother with a bitmap for those entries. However
2279 * some block groups can be smaller than what a bitmap would cover but
2280 * are still large enough that they could overflow the 32k memory limit,
2281 * so allow those block groups to still be allowed to have a bitmap
2284 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2290 static const struct btrfs_free_space_op free_space_op = {
2291 .use_bitmap = use_bitmap,
2294 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2295 struct btrfs_free_space *info)
2297 struct btrfs_free_space *bitmap_info;
2298 struct btrfs_block_group *block_group = NULL;
2300 u64 bytes, offset, bytes_added;
2301 enum btrfs_trim_state trim_state;
2304 bytes = info->bytes;
2305 offset = info->offset;
2306 trim_state = info->trim_state;
2308 if (!ctl->op->use_bitmap(ctl, info))
2311 if (ctl->op == &free_space_op)
2312 block_group = ctl->block_group;
2315 * Since we link bitmaps right into the cluster we need to see if we
2316 * have a cluster here, and if so and it has our bitmap we need to add
2317 * the free space to that bitmap.
2319 if (block_group && !list_empty(&block_group->cluster_list)) {
2320 struct btrfs_free_cluster *cluster;
2321 struct rb_node *node;
2322 struct btrfs_free_space *entry;
2324 cluster = list_entry(block_group->cluster_list.next,
2325 struct btrfs_free_cluster,
2327 spin_lock(&cluster->lock);
2328 node = rb_first(&cluster->root);
2330 spin_unlock(&cluster->lock);
2331 goto no_cluster_bitmap;
2334 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2335 if (!entry->bitmap) {
2336 spin_unlock(&cluster->lock);
2337 goto no_cluster_bitmap;
2340 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2341 bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2343 bytes -= bytes_added;
2344 offset += bytes_added;
2346 spin_unlock(&cluster->lock);
2354 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2361 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2363 bytes -= bytes_added;
2364 offset += bytes_added;
2374 if (info && info->bitmap) {
2375 add_new_bitmap(ctl, info, offset);
2380 spin_unlock(&ctl->tree_lock);
2382 /* no pre-allocated info, allocate a new one */
2384 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2387 spin_lock(&ctl->tree_lock);
2393 /* allocate the bitmap */
2394 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2396 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2397 spin_lock(&ctl->tree_lock);
2398 if (!info->bitmap) {
2408 kmem_cache_free(btrfs_free_space_bitmap_cachep,
2410 kmem_cache_free(btrfs_free_space_cachep, info);
2417 * Free space merging rules:
2418 * 1) Merge trimmed areas together
2419 * 2) Let untrimmed areas coalesce with trimmed areas
2420 * 3) Always pull neighboring regions from bitmaps
2422 * The above rules are for when we merge free space based on btrfs_trim_state.
2423 * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2424 * same reason: to promote larger extent regions which makes life easier for
2425 * find_free_extent(). Rule 2 enables coalescing based on the common path
2426 * being returning free space from btrfs_finish_extent_commit(). So when free
2427 * space is trimmed, it will prevent aggregating trimmed new region and
2428 * untrimmed regions in the rb_tree. Rule 3 is purely to obtain larger extents
2429 * and provide find_free_extent() with the largest extents possible hoping for
2432 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2433 struct btrfs_free_space *info, bool update_stat)
2435 struct btrfs_free_space *left_info = NULL;
2436 struct btrfs_free_space *right_info;
2437 bool merged = false;
2438 u64 offset = info->offset;
2439 u64 bytes = info->bytes;
2440 const bool is_trimmed = btrfs_free_space_trimmed(info);
2443 * first we want to see if there is free space adjacent to the range we
2444 * are adding, if there is remove that struct and add a new one to
2445 * cover the entire range
2447 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2448 if (right_info && rb_prev(&right_info->offset_index))
2449 left_info = rb_entry(rb_prev(&right_info->offset_index),
2450 struct btrfs_free_space, offset_index);
2451 else if (!right_info)
2452 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2454 /* See try_merge_free_space() comment. */
2455 if (right_info && !right_info->bitmap &&
2456 (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2457 unlink_free_space(ctl, right_info, update_stat);
2458 info->bytes += right_info->bytes;
2459 kmem_cache_free(btrfs_free_space_cachep, right_info);
2463 /* See try_merge_free_space() comment. */
2464 if (left_info && !left_info->bitmap &&
2465 left_info->offset + left_info->bytes == offset &&
2466 (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2467 unlink_free_space(ctl, left_info, update_stat);
2468 info->offset = left_info->offset;
2469 info->bytes += left_info->bytes;
2470 kmem_cache_free(btrfs_free_space_cachep, left_info);
2477 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2478 struct btrfs_free_space *info,
2481 struct btrfs_free_space *bitmap;
2484 const u64 end = info->offset + info->bytes;
2485 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2488 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2492 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2493 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2496 bytes = (j - i) * ctl->unit;
2497 info->bytes += bytes;
2499 /* See try_merge_free_space() comment. */
2500 if (!btrfs_free_space_trimmed(bitmap))
2501 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2503 bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2506 free_bitmap(ctl, bitmap);
2511 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2512 struct btrfs_free_space *info,
2515 struct btrfs_free_space *bitmap;
2519 unsigned long prev_j;
2522 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2523 /* If we're on a boundary, try the previous logical bitmap. */
2524 if (bitmap_offset == info->offset) {
2525 if (info->offset == 0)
2527 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2530 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2534 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2536 prev_j = (unsigned long)-1;
2537 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2545 if (prev_j == (unsigned long)-1)
2546 bytes = (i + 1) * ctl->unit;
2548 bytes = (i - prev_j) * ctl->unit;
2550 info->offset -= bytes;
2551 info->bytes += bytes;
2553 /* See try_merge_free_space() comment. */
2554 if (!btrfs_free_space_trimmed(bitmap))
2555 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2557 bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2560 free_bitmap(ctl, bitmap);
2566 * We prefer always to allocate from extent entries, both for clustered and
2567 * non-clustered allocation requests. So when attempting to add a new extent
2568 * entry, try to see if there's adjacent free space in bitmap entries, and if
2569 * there is, migrate that space from the bitmaps to the extent.
2570 * Like this we get better chances of satisfying space allocation requests
2571 * because we attempt to satisfy them based on a single cache entry, and never
2572 * on 2 or more entries - even if the entries represent a contiguous free space
2573 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2576 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2577 struct btrfs_free_space *info,
2581 * Only work with disconnected entries, as we can change their offset,
2582 * and must be extent entries.
2584 ASSERT(!info->bitmap);
2585 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2587 if (ctl->total_bitmaps > 0) {
2589 bool stole_front = false;
2591 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2592 if (ctl->total_bitmaps > 0)
2593 stole_front = steal_from_bitmap_to_front(ctl, info,
2596 if (stole_end || stole_front)
2597 try_merge_free_space(ctl, info, update_stat);
2601 int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2602 u64 offset, u64 bytes,
2603 enum btrfs_trim_state trim_state)
2605 struct btrfs_fs_info *fs_info = block_group->fs_info;
2606 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2607 struct btrfs_free_space *info;
2609 u64 filter_bytes = bytes;
2611 ASSERT(!btrfs_is_zoned(fs_info));
2613 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2617 info->offset = offset;
2618 info->bytes = bytes;
2619 info->trim_state = trim_state;
2620 RB_CLEAR_NODE(&info->offset_index);
2621 RB_CLEAR_NODE(&info->bytes_index);
2623 spin_lock(&ctl->tree_lock);
2625 if (try_merge_free_space(ctl, info, true))
2629 * There was no extent directly to the left or right of this new
2630 * extent then we know we're going to have to allocate a new extent, so
2631 * before we do that see if we need to drop this into a bitmap
2633 ret = insert_into_bitmap(ctl, info);
2642 * Only steal free space from adjacent bitmaps if we're sure we're not
2643 * going to add the new free space to existing bitmap entries - because
2644 * that would mean unnecessary work that would be reverted. Therefore
2645 * attempt to steal space from bitmaps if we're adding an extent entry.
2647 steal_from_bitmap(ctl, info, true);
2649 filter_bytes = max(filter_bytes, info->bytes);
2651 ret = link_free_space(ctl, info);
2653 kmem_cache_free(btrfs_free_space_cachep, info);
2655 btrfs_discard_update_discardable(block_group);
2656 spin_unlock(&ctl->tree_lock);
2659 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2660 ASSERT(ret != -EEXIST);
2663 if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2664 btrfs_discard_check_filter(block_group, filter_bytes);
2665 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2671 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2672 u64 bytenr, u64 size, bool used)
2674 struct btrfs_space_info *sinfo = block_group->space_info;
2675 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2676 u64 offset = bytenr - block_group->start;
2677 u64 to_free, to_unusable;
2678 int bg_reclaim_threshold = 0;
2679 bool initial = (size == block_group->length);
2680 u64 reclaimable_unusable;
2682 WARN_ON(!initial && offset + size > block_group->zone_capacity);
2685 bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2687 spin_lock(&ctl->tree_lock);
2688 /* Count initial region as zone_unusable until it gets activated. */
2692 test_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &block_group->fs_info->flags) &&
2693 (block_group->flags & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)))
2696 to_free = block_group->zone_capacity;
2697 else if (offset >= block_group->alloc_offset)
2699 else if (offset + size <= block_group->alloc_offset)
2702 to_free = offset + size - block_group->alloc_offset;
2703 to_unusable = size - to_free;
2705 ctl->free_space += to_free;
2707 * If the block group is read-only, we should account freed space into
2710 if (!block_group->ro)
2711 block_group->zone_unusable += to_unusable;
2712 spin_unlock(&ctl->tree_lock);
2714 spin_lock(&block_group->lock);
2715 block_group->alloc_offset -= size;
2716 spin_unlock(&block_group->lock);
2719 reclaimable_unusable = block_group->zone_unusable -
2720 (block_group->length - block_group->zone_capacity);
2721 /* All the region is now unusable. Mark it as unused and reclaim */
2722 if (block_group->zone_unusable == block_group->length &&
2723 block_group->alloc_offset) {
2724 btrfs_mark_bg_unused(block_group);
2725 } else if (bg_reclaim_threshold &&
2726 reclaimable_unusable >=
2727 div_factor_fine(block_group->zone_capacity,
2728 bg_reclaim_threshold)) {
2729 btrfs_mark_bg_to_reclaim(block_group);
2735 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2736 u64 bytenr, u64 size)
2738 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2740 if (btrfs_is_zoned(block_group->fs_info))
2741 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2744 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2745 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2747 return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2750 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2751 u64 bytenr, u64 size)
2753 if (btrfs_is_zoned(block_group->fs_info))
2754 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2757 return btrfs_add_free_space(block_group, bytenr, size);
2761 * This is a subtle distinction because when adding free space back in general,
2762 * we want it to be added as untrimmed for async. But in the case where we add
2763 * it on loading of a block group, we want to consider it trimmed.
2765 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2766 u64 bytenr, u64 size)
2768 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2770 if (btrfs_is_zoned(block_group->fs_info))
2771 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2774 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2775 btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2776 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2778 return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2781 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2782 u64 offset, u64 bytes)
2784 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2785 struct btrfs_free_space *info;
2787 bool re_search = false;
2789 if (btrfs_is_zoned(block_group->fs_info)) {
2791 * This can happen with conventional zones when replaying log.
2792 * Since the allocation info of tree-log nodes are not recorded
2793 * to the extent-tree, calculate_alloc_pointer() failed to
2794 * advance the allocation pointer after last allocated tree log
2797 * This function is called from
2798 * btrfs_pin_extent_for_log_replay() when replaying the log.
2799 * Advance the pointer not to overwrite the tree-log nodes.
2801 if (block_group->start + block_group->alloc_offset <
2803 block_group->alloc_offset =
2804 offset + bytes - block_group->start;
2809 spin_lock(&ctl->tree_lock);
2816 info = tree_search_offset(ctl, offset, 0, 0);
2819 * oops didn't find an extent that matched the space we wanted
2820 * to remove, look for a bitmap instead
2822 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2826 * If we found a partial bit of our free space in a
2827 * bitmap but then couldn't find the other part this may
2828 * be a problem, so WARN about it.
2836 if (!info->bitmap) {
2837 unlink_free_space(ctl, info, true);
2838 if (offset == info->offset) {
2839 u64 to_free = min(bytes, info->bytes);
2841 info->bytes -= to_free;
2842 info->offset += to_free;
2844 ret = link_free_space(ctl, info);
2847 kmem_cache_free(btrfs_free_space_cachep, info);
2854 u64 old_end = info->bytes + info->offset;
2856 info->bytes = offset - info->offset;
2857 ret = link_free_space(ctl, info);
2862 /* Not enough bytes in this entry to satisfy us */
2863 if (old_end < offset + bytes) {
2864 bytes -= old_end - offset;
2867 } else if (old_end == offset + bytes) {
2871 spin_unlock(&ctl->tree_lock);
2873 ret = __btrfs_add_free_space(block_group,
2875 old_end - (offset + bytes),
2882 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2883 if (ret == -EAGAIN) {
2888 btrfs_discard_update_discardable(block_group);
2889 spin_unlock(&ctl->tree_lock);
2894 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2897 struct btrfs_fs_info *fs_info = block_group->fs_info;
2898 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2899 struct btrfs_free_space *info;
2904 * Zoned btrfs does not use free space tree and cluster. Just print
2905 * out the free space after the allocation offset.
2907 if (btrfs_is_zoned(fs_info)) {
2908 btrfs_info(fs_info, "free space %llu active %d",
2909 block_group->zone_capacity - block_group->alloc_offset,
2910 test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2911 &block_group->runtime_flags));
2915 spin_lock(&ctl->tree_lock);
2916 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2917 info = rb_entry(n, struct btrfs_free_space, offset_index);
2918 if (info->bytes >= bytes && !block_group->ro)
2920 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2921 info->offset, info->bytes,
2922 (info->bitmap) ? "yes" : "no");
2924 spin_unlock(&ctl->tree_lock);
2925 btrfs_info(fs_info, "block group has cluster?: %s",
2926 list_empty(&block_group->cluster_list) ? "no" : "yes");
2928 "%d blocks of free space at or bigger than bytes is", count);
2931 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2932 struct btrfs_free_space_ctl *ctl)
2934 struct btrfs_fs_info *fs_info = block_group->fs_info;
2936 spin_lock_init(&ctl->tree_lock);
2937 ctl->unit = fs_info->sectorsize;
2938 ctl->start = block_group->start;
2939 ctl->block_group = block_group;
2940 ctl->op = &free_space_op;
2941 ctl->free_space_bytes = RB_ROOT_CACHED;
2942 INIT_LIST_HEAD(&ctl->trimming_ranges);
2943 mutex_init(&ctl->cache_writeout_mutex);
2946 * we only want to have 32k of ram per block group for keeping
2947 * track of free space, and if we pass 1/2 of that we want to
2948 * start converting things over to using bitmaps
2950 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2954 * for a given cluster, put all of its extents back into the free
2955 * space cache. If the block group passed doesn't match the block group
2956 * pointed to by the cluster, someone else raced in and freed the
2957 * cluster already. In that case, we just return without changing anything
2959 static void __btrfs_return_cluster_to_free_space(
2960 struct btrfs_block_group *block_group,
2961 struct btrfs_free_cluster *cluster)
2963 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2964 struct btrfs_free_space *entry;
2965 struct rb_node *node;
2967 spin_lock(&cluster->lock);
2968 if (cluster->block_group != block_group) {
2969 spin_unlock(&cluster->lock);
2973 cluster->block_group = NULL;
2974 cluster->window_start = 0;
2975 list_del_init(&cluster->block_group_list);
2977 node = rb_first(&cluster->root);
2981 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2982 node = rb_next(&entry->offset_index);
2983 rb_erase(&entry->offset_index, &cluster->root);
2984 RB_CLEAR_NODE(&entry->offset_index);
2986 bitmap = (entry->bitmap != NULL);
2988 /* Merging treats extents as if they were new */
2989 if (!btrfs_free_space_trimmed(entry)) {
2990 ctl->discardable_extents[BTRFS_STAT_CURR]--;
2991 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
2995 try_merge_free_space(ctl, entry, false);
2996 steal_from_bitmap(ctl, entry, false);
2998 /* As we insert directly, update these statistics */
2999 if (!btrfs_free_space_trimmed(entry)) {
3000 ctl->discardable_extents[BTRFS_STAT_CURR]++;
3001 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3005 tree_insert_offset(&ctl->free_space_offset,
3006 entry->offset, &entry->offset_index, bitmap);
3007 rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
3010 cluster->root = RB_ROOT;
3011 spin_unlock(&cluster->lock);
3012 btrfs_put_block_group(block_group);
3015 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3017 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3018 struct btrfs_free_cluster *cluster;
3019 struct list_head *head;
3021 spin_lock(&ctl->tree_lock);
3022 while ((head = block_group->cluster_list.next) !=
3023 &block_group->cluster_list) {
3024 cluster = list_entry(head, struct btrfs_free_cluster,
3027 WARN_ON(cluster->block_group != block_group);
3028 __btrfs_return_cluster_to_free_space(block_group, cluster);
3030 cond_resched_lock(&ctl->tree_lock);
3032 __btrfs_remove_free_space_cache(ctl);
3033 btrfs_discard_update_discardable(block_group);
3034 spin_unlock(&ctl->tree_lock);
3039 * btrfs_is_free_space_trimmed - see if everything is trimmed
3040 * @block_group: block_group of interest
3042 * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3044 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3046 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3047 struct btrfs_free_space *info;
3048 struct rb_node *node;
3051 spin_lock(&ctl->tree_lock);
3052 node = rb_first(&ctl->free_space_offset);
3055 info = rb_entry(node, struct btrfs_free_space, offset_index);
3057 if (!btrfs_free_space_trimmed(info)) {
3062 node = rb_next(node);
3065 spin_unlock(&ctl->tree_lock);
3069 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3070 u64 offset, u64 bytes, u64 empty_size,
3071 u64 *max_extent_size)
3073 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3074 struct btrfs_discard_ctl *discard_ctl =
3075 &block_group->fs_info->discard_ctl;
3076 struct btrfs_free_space *entry = NULL;
3077 u64 bytes_search = bytes + empty_size;
3080 u64 align_gap_len = 0;
3081 enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3082 bool use_bytes_index = (offset == block_group->start);
3084 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3086 spin_lock(&ctl->tree_lock);
3087 entry = find_free_space(ctl, &offset, &bytes_search,
3088 block_group->full_stripe_len, max_extent_size,
3094 if (entry->bitmap) {
3095 bitmap_clear_bits(ctl, entry, offset, bytes, true);
3097 if (!btrfs_free_space_trimmed(entry))
3098 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3101 free_bitmap(ctl, entry);
3103 unlink_free_space(ctl, entry, true);
3104 align_gap_len = offset - entry->offset;
3105 align_gap = entry->offset;
3106 align_gap_trim_state = entry->trim_state;
3108 if (!btrfs_free_space_trimmed(entry))
3109 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3111 entry->offset = offset + bytes;
3112 WARN_ON(entry->bytes < bytes + align_gap_len);
3114 entry->bytes -= bytes + align_gap_len;
3116 kmem_cache_free(btrfs_free_space_cachep, entry);
3118 link_free_space(ctl, entry);
3121 btrfs_discard_update_discardable(block_group);
3122 spin_unlock(&ctl->tree_lock);
3125 __btrfs_add_free_space(block_group, align_gap, align_gap_len,
3126 align_gap_trim_state);
3131 * given a cluster, put all of its extents back into the free space
3132 * cache. If a block group is passed, this function will only free
3133 * a cluster that belongs to the passed block group.
3135 * Otherwise, it'll get a reference on the block group pointed to by the
3136 * cluster and remove the cluster from it.
3138 void btrfs_return_cluster_to_free_space(
3139 struct btrfs_block_group *block_group,
3140 struct btrfs_free_cluster *cluster)
3142 struct btrfs_free_space_ctl *ctl;
3144 /* first, get a safe pointer to the block group */
3145 spin_lock(&cluster->lock);
3147 block_group = cluster->block_group;
3149 spin_unlock(&cluster->lock);
3152 } else if (cluster->block_group != block_group) {
3153 /* someone else has already freed it don't redo their work */
3154 spin_unlock(&cluster->lock);
3157 btrfs_get_block_group(block_group);
3158 spin_unlock(&cluster->lock);
3160 ctl = block_group->free_space_ctl;
3162 /* now return any extents the cluster had on it */
3163 spin_lock(&ctl->tree_lock);
3164 __btrfs_return_cluster_to_free_space(block_group, cluster);
3165 spin_unlock(&ctl->tree_lock);
3167 btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3169 /* finally drop our ref */
3170 btrfs_put_block_group(block_group);
3173 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3174 struct btrfs_free_cluster *cluster,
3175 struct btrfs_free_space *entry,
3176 u64 bytes, u64 min_start,
3177 u64 *max_extent_size)
3179 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3181 u64 search_start = cluster->window_start;
3182 u64 search_bytes = bytes;
3185 search_start = min_start;
3186 search_bytes = bytes;
3188 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3190 *max_extent_size = max(get_max_extent_size(entry),
3196 bitmap_clear_bits(ctl, entry, ret, bytes, false);
3202 * given a cluster, try to allocate 'bytes' from it, returns 0
3203 * if it couldn't find anything suitably large, or a logical disk offset
3204 * if things worked out
3206 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3207 struct btrfs_free_cluster *cluster, u64 bytes,
3208 u64 min_start, u64 *max_extent_size)
3210 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3211 struct btrfs_discard_ctl *discard_ctl =
3212 &block_group->fs_info->discard_ctl;
3213 struct btrfs_free_space *entry = NULL;
3214 struct rb_node *node;
3217 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3219 spin_lock(&cluster->lock);
3220 if (bytes > cluster->max_size)
3223 if (cluster->block_group != block_group)
3226 node = rb_first(&cluster->root);
3230 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3232 if (entry->bytes < bytes)
3233 *max_extent_size = max(get_max_extent_size(entry),
3236 if (entry->bytes < bytes ||
3237 (!entry->bitmap && entry->offset < min_start)) {
3238 node = rb_next(&entry->offset_index);
3241 entry = rb_entry(node, struct btrfs_free_space,
3246 if (entry->bitmap) {
3247 ret = btrfs_alloc_from_bitmap(block_group,
3248 cluster, entry, bytes,
3249 cluster->window_start,
3252 node = rb_next(&entry->offset_index);
3255 entry = rb_entry(node, struct btrfs_free_space,
3259 cluster->window_start += bytes;
3261 ret = entry->offset;
3263 entry->offset += bytes;
3264 entry->bytes -= bytes;
3270 spin_unlock(&cluster->lock);
3275 spin_lock(&ctl->tree_lock);
3277 if (!btrfs_free_space_trimmed(entry))
3278 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3280 ctl->free_space -= bytes;
3281 if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3282 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3284 spin_lock(&cluster->lock);
3285 if (entry->bytes == 0) {
3286 rb_erase(&entry->offset_index, &cluster->root);
3287 ctl->free_extents--;
3288 if (entry->bitmap) {
3289 kmem_cache_free(btrfs_free_space_bitmap_cachep,
3291 ctl->total_bitmaps--;
3292 recalculate_thresholds(ctl);
3293 } else if (!btrfs_free_space_trimmed(entry)) {
3294 ctl->discardable_extents[BTRFS_STAT_CURR]--;
3296 kmem_cache_free(btrfs_free_space_cachep, entry);
3299 spin_unlock(&cluster->lock);
3300 spin_unlock(&ctl->tree_lock);
3305 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3306 struct btrfs_free_space *entry,
3307 struct btrfs_free_cluster *cluster,
3308 u64 offset, u64 bytes,
3309 u64 cont1_bytes, u64 min_bytes)
3311 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3312 unsigned long next_zero;
3314 unsigned long want_bits;
3315 unsigned long min_bits;
3316 unsigned long found_bits;
3317 unsigned long max_bits = 0;
3318 unsigned long start = 0;
3319 unsigned long total_found = 0;
3322 i = offset_to_bit(entry->offset, ctl->unit,
3323 max_t(u64, offset, entry->offset));
3324 want_bits = bytes_to_bits(bytes, ctl->unit);
3325 min_bits = bytes_to_bits(min_bytes, ctl->unit);
3328 * Don't bother looking for a cluster in this bitmap if it's heavily
3331 if (entry->max_extent_size &&
3332 entry->max_extent_size < cont1_bytes)
3336 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3337 next_zero = find_next_zero_bit(entry->bitmap,
3338 BITS_PER_BITMAP, i);
3339 if (next_zero - i >= min_bits) {
3340 found_bits = next_zero - i;
3341 if (found_bits > max_bits)
3342 max_bits = found_bits;
3345 if (next_zero - i > max_bits)
3346 max_bits = next_zero - i;
3351 entry->max_extent_size = (u64)max_bits * ctl->unit;
3357 cluster->max_size = 0;
3360 total_found += found_bits;
3362 if (cluster->max_size < found_bits * ctl->unit)
3363 cluster->max_size = found_bits * ctl->unit;
3365 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3370 cluster->window_start = start * ctl->unit + entry->offset;
3371 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3372 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3375 * We need to know if we're currently on the normal space index when we
3376 * manipulate the bitmap so that we know we need to remove and re-insert
3377 * it into the space_index tree. Clear the bytes_index node here so the
3378 * bitmap manipulation helpers know not to mess with the space_index
3379 * until this bitmap entry is added back into the normal cache.
3381 RB_CLEAR_NODE(&entry->bytes_index);
3383 ret = tree_insert_offset(&cluster->root, entry->offset,
3384 &entry->offset_index, 1);
3385 ASSERT(!ret); /* -EEXIST; Logic error */
3387 trace_btrfs_setup_cluster(block_group, cluster,
3388 total_found * ctl->unit, 1);
3393 * This searches the block group for just extents to fill the cluster with.
3394 * Try to find a cluster with at least bytes total bytes, at least one
3395 * extent of cont1_bytes, and other clusters of at least min_bytes.
3398 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3399 struct btrfs_free_cluster *cluster,
3400 struct list_head *bitmaps, u64 offset, u64 bytes,
3401 u64 cont1_bytes, u64 min_bytes)
3403 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3404 struct btrfs_free_space *first = NULL;
3405 struct btrfs_free_space *entry = NULL;
3406 struct btrfs_free_space *last;
3407 struct rb_node *node;
3412 entry = tree_search_offset(ctl, offset, 0, 1);
3417 * We don't want bitmaps, so just move along until we find a normal
3420 while (entry->bitmap || entry->bytes < min_bytes) {
3421 if (entry->bitmap && list_empty(&entry->list))
3422 list_add_tail(&entry->list, bitmaps);
3423 node = rb_next(&entry->offset_index);
3426 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3429 window_free = entry->bytes;
3430 max_extent = entry->bytes;
3434 for (node = rb_next(&entry->offset_index); node;
3435 node = rb_next(&entry->offset_index)) {
3436 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3438 if (entry->bitmap) {
3439 if (list_empty(&entry->list))
3440 list_add_tail(&entry->list, bitmaps);
3444 if (entry->bytes < min_bytes)
3448 window_free += entry->bytes;
3449 if (entry->bytes > max_extent)
3450 max_extent = entry->bytes;
3453 if (window_free < bytes || max_extent < cont1_bytes)
3456 cluster->window_start = first->offset;
3458 node = &first->offset_index;
3461 * now we've found our entries, pull them out of the free space
3462 * cache and put them into the cluster rbtree
3467 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3468 node = rb_next(&entry->offset_index);
3469 if (entry->bitmap || entry->bytes < min_bytes)
3472 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3473 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3474 ret = tree_insert_offset(&cluster->root, entry->offset,
3475 &entry->offset_index, 0);
3476 total_size += entry->bytes;
3477 ASSERT(!ret); /* -EEXIST; Logic error */
3478 } while (node && entry != last);
3480 cluster->max_size = max_extent;
3481 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3486 * This specifically looks for bitmaps that may work in the cluster, we assume
3487 * that we have already failed to find extents that will work.
3490 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3491 struct btrfs_free_cluster *cluster,
3492 struct list_head *bitmaps, u64 offset, u64 bytes,
3493 u64 cont1_bytes, u64 min_bytes)
3495 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3496 struct btrfs_free_space *entry = NULL;
3498 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3500 if (ctl->total_bitmaps == 0)
3504 * The bitmap that covers offset won't be in the list unless offset
3505 * is just its start offset.
3507 if (!list_empty(bitmaps))
3508 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3510 if (!entry || entry->offset != bitmap_offset) {
3511 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3512 if (entry && list_empty(&entry->list))
3513 list_add(&entry->list, bitmaps);
3516 list_for_each_entry(entry, bitmaps, list) {
3517 if (entry->bytes < bytes)
3519 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3520 bytes, cont1_bytes, min_bytes);
3526 * The bitmaps list has all the bitmaps that record free space
3527 * starting after offset, so no more search is required.
3533 * here we try to find a cluster of blocks in a block group. The goal
3534 * is to find at least bytes+empty_size.
3535 * We might not find them all in one contiguous area.
3537 * returns zero and sets up cluster if things worked out, otherwise
3538 * it returns -enospc
3540 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3541 struct btrfs_free_cluster *cluster,
3542 u64 offset, u64 bytes, u64 empty_size)
3544 struct btrfs_fs_info *fs_info = block_group->fs_info;
3545 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3546 struct btrfs_free_space *entry, *tmp;
3553 * Choose the minimum extent size we'll require for this
3554 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3555 * For metadata, allow allocates with smaller extents. For
3556 * data, keep it dense.
3558 if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3559 cont1_bytes = bytes + empty_size;
3560 min_bytes = cont1_bytes;
3561 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3562 cont1_bytes = bytes;
3563 min_bytes = fs_info->sectorsize;
3565 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3566 min_bytes = fs_info->sectorsize;
3569 spin_lock(&ctl->tree_lock);
3572 * If we know we don't have enough space to make a cluster don't even
3573 * bother doing all the work to try and find one.
3575 if (ctl->free_space < bytes) {
3576 spin_unlock(&ctl->tree_lock);
3580 spin_lock(&cluster->lock);
3582 /* someone already found a cluster, hooray */
3583 if (cluster->block_group) {
3588 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3591 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3593 cont1_bytes, min_bytes);
3595 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3596 offset, bytes + empty_size,
3597 cont1_bytes, min_bytes);
3599 /* Clear our temporary list */
3600 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3601 list_del_init(&entry->list);
3604 btrfs_get_block_group(block_group);
3605 list_add_tail(&cluster->block_group_list,
3606 &block_group->cluster_list);
3607 cluster->block_group = block_group;
3609 trace_btrfs_failed_cluster_setup(block_group);
3612 spin_unlock(&cluster->lock);
3613 spin_unlock(&ctl->tree_lock);
3619 * simple code to zero out a cluster
3621 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3623 spin_lock_init(&cluster->lock);
3624 spin_lock_init(&cluster->refill_lock);
3625 cluster->root = RB_ROOT;
3626 cluster->max_size = 0;
3627 cluster->fragmented = false;
3628 INIT_LIST_HEAD(&cluster->block_group_list);
3629 cluster->block_group = NULL;
3632 static int do_trimming(struct btrfs_block_group *block_group,
3633 u64 *total_trimmed, u64 start, u64 bytes,
3634 u64 reserved_start, u64 reserved_bytes,
3635 enum btrfs_trim_state reserved_trim_state,
3636 struct btrfs_trim_range *trim_entry)
3638 struct btrfs_space_info *space_info = block_group->space_info;
3639 struct btrfs_fs_info *fs_info = block_group->fs_info;
3640 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3643 const u64 end = start + bytes;
3644 const u64 reserved_end = reserved_start + reserved_bytes;
3645 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3648 spin_lock(&space_info->lock);
3649 spin_lock(&block_group->lock);
3650 if (!block_group->ro) {
3651 block_group->reserved += reserved_bytes;
3652 space_info->bytes_reserved += reserved_bytes;
3655 spin_unlock(&block_group->lock);
3656 spin_unlock(&space_info->lock);
3658 ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3660 *total_trimmed += trimmed;
3661 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3664 mutex_lock(&ctl->cache_writeout_mutex);
3665 if (reserved_start < start)
3666 __btrfs_add_free_space(block_group, reserved_start,
3667 start - reserved_start,
3668 reserved_trim_state);
3669 if (start + bytes < reserved_start + reserved_bytes)
3670 __btrfs_add_free_space(block_group, end, reserved_end - end,
3671 reserved_trim_state);
3672 __btrfs_add_free_space(block_group, start, bytes, trim_state);
3673 list_del(&trim_entry->list);
3674 mutex_unlock(&ctl->cache_writeout_mutex);
3677 spin_lock(&space_info->lock);
3678 spin_lock(&block_group->lock);
3679 if (block_group->ro)
3680 space_info->bytes_readonly += reserved_bytes;
3681 block_group->reserved -= reserved_bytes;
3682 space_info->bytes_reserved -= reserved_bytes;
3683 spin_unlock(&block_group->lock);
3684 spin_unlock(&space_info->lock);
3691 * If @async is set, then we will trim 1 region and return.
3693 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3694 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3697 struct btrfs_discard_ctl *discard_ctl =
3698 &block_group->fs_info->discard_ctl;
3699 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3700 struct btrfs_free_space *entry;
3701 struct rb_node *node;
3705 enum btrfs_trim_state extent_trim_state;
3707 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3709 while (start < end) {
3710 struct btrfs_trim_range trim_entry;
3712 mutex_lock(&ctl->cache_writeout_mutex);
3713 spin_lock(&ctl->tree_lock);
3715 if (ctl->free_space < minlen)
3718 entry = tree_search_offset(ctl, start, 0, 1);
3722 /* Skip bitmaps and if async, already trimmed entries */
3723 while (entry->bitmap ||
3724 (async && btrfs_free_space_trimmed(entry))) {
3725 node = rb_next(&entry->offset_index);
3728 entry = rb_entry(node, struct btrfs_free_space,
3732 if (entry->offset >= end)
3735 extent_start = entry->offset;
3736 extent_bytes = entry->bytes;
3737 extent_trim_state = entry->trim_state;
3739 start = entry->offset;
3740 bytes = entry->bytes;
3741 if (bytes < minlen) {
3742 spin_unlock(&ctl->tree_lock);
3743 mutex_unlock(&ctl->cache_writeout_mutex);
3746 unlink_free_space(ctl, entry, true);
3748 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3749 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3750 * X when we come back around. So trim it now.
3752 if (max_discard_size &&
3753 bytes >= (max_discard_size +
3754 BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3755 bytes = max_discard_size;
3756 extent_bytes = max_discard_size;
3757 entry->offset += max_discard_size;
3758 entry->bytes -= max_discard_size;
3759 link_free_space(ctl, entry);
3761 kmem_cache_free(btrfs_free_space_cachep, entry);
3764 start = max(start, extent_start);
3765 bytes = min(extent_start + extent_bytes, end) - start;
3766 if (bytes < minlen) {
3767 spin_unlock(&ctl->tree_lock);
3768 mutex_unlock(&ctl->cache_writeout_mutex);
3772 unlink_free_space(ctl, entry, true);
3773 kmem_cache_free(btrfs_free_space_cachep, entry);
3776 spin_unlock(&ctl->tree_lock);
3777 trim_entry.start = extent_start;
3778 trim_entry.bytes = extent_bytes;
3779 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3780 mutex_unlock(&ctl->cache_writeout_mutex);
3782 ret = do_trimming(block_group, total_trimmed, start, bytes,
3783 extent_start, extent_bytes, extent_trim_state,
3786 block_group->discard_cursor = start + bytes;
3791 block_group->discard_cursor = start;
3792 if (async && *total_trimmed)
3795 if (fatal_signal_pending(current)) {
3806 block_group->discard_cursor = btrfs_block_group_end(block_group);
3807 spin_unlock(&ctl->tree_lock);
3808 mutex_unlock(&ctl->cache_writeout_mutex);
3814 * If we break out of trimming a bitmap prematurely, we should reset the
3815 * trimming bit. In a rather contrieved case, it's possible to race here so
3816 * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3818 * start = start of bitmap
3819 * end = near end of bitmap
3821 * Thread 1: Thread 2:
3822 * trim_bitmaps(start)
3824 * end_trimming_bitmap()
3825 * reset_trimming_bitmap()
3827 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3829 struct btrfs_free_space *entry;
3831 spin_lock(&ctl->tree_lock);
3832 entry = tree_search_offset(ctl, offset, 1, 0);
3834 if (btrfs_free_space_trimmed(entry)) {
3835 ctl->discardable_extents[BTRFS_STAT_CURR] +=
3836 entry->bitmap_extents;
3837 ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3839 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3842 spin_unlock(&ctl->tree_lock);
3845 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3846 struct btrfs_free_space *entry)
3848 if (btrfs_free_space_trimming_bitmap(entry)) {
3849 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3850 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3851 entry->bitmap_extents;
3852 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3857 * If @async is set, then we will trim 1 region and return.
3859 static int trim_bitmaps(struct btrfs_block_group *block_group,
3860 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3861 u64 maxlen, bool async)
3863 struct btrfs_discard_ctl *discard_ctl =
3864 &block_group->fs_info->discard_ctl;
3865 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3866 struct btrfs_free_space *entry;
3870 u64 offset = offset_to_bitmap(ctl, start);
3871 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3873 while (offset < end) {
3874 bool next_bitmap = false;
3875 struct btrfs_trim_range trim_entry;
3877 mutex_lock(&ctl->cache_writeout_mutex);
3878 spin_lock(&ctl->tree_lock);
3880 if (ctl->free_space < minlen) {
3881 block_group->discard_cursor =
3882 btrfs_block_group_end(block_group);
3883 spin_unlock(&ctl->tree_lock);
3884 mutex_unlock(&ctl->cache_writeout_mutex);
3888 entry = tree_search_offset(ctl, offset, 1, 0);
3890 * Bitmaps are marked trimmed lossily now to prevent constant
3891 * discarding of the same bitmap (the reason why we are bound
3892 * by the filters). So, retrim the block group bitmaps when we
3893 * are preparing to punt to the unused_bgs list. This uses
3894 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3895 * which is the only discard index which sets minlen to 0.
3897 if (!entry || (async && minlen && start == offset &&
3898 btrfs_free_space_trimmed(entry))) {
3899 spin_unlock(&ctl->tree_lock);
3900 mutex_unlock(&ctl->cache_writeout_mutex);
3906 * Async discard bitmap trimming begins at by setting the start
3907 * to be key.objectid and the offset_to_bitmap() aligns to the
3908 * start of the bitmap. This lets us know we are fully
3909 * scanning the bitmap rather than only some portion of it.
3911 if (start == offset)
3912 entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3915 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3916 if (ret2 || start >= end) {
3918 * We lossily consider a bitmap trimmed if we only skip
3919 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3921 if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3922 end_trimming_bitmap(ctl, entry);
3924 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3925 spin_unlock(&ctl->tree_lock);
3926 mutex_unlock(&ctl->cache_writeout_mutex);
3932 * We already trimmed a region, but are using the locking above
3933 * to reset the trim_state.
3935 if (async && *total_trimmed) {
3936 spin_unlock(&ctl->tree_lock);
3937 mutex_unlock(&ctl->cache_writeout_mutex);
3941 bytes = min(bytes, end - start);
3942 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3943 spin_unlock(&ctl->tree_lock);
3944 mutex_unlock(&ctl->cache_writeout_mutex);
3949 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3950 * If X < @minlen, we won't trim X when we come back around.
3951 * So trim it now. We differ here from trimming extents as we
3952 * don't keep individual state per bit.
3956 bytes > (max_discard_size + minlen))
3957 bytes = max_discard_size;
3959 bitmap_clear_bits(ctl, entry, start, bytes, true);
3960 if (entry->bytes == 0)
3961 free_bitmap(ctl, entry);
3963 spin_unlock(&ctl->tree_lock);
3964 trim_entry.start = start;
3965 trim_entry.bytes = bytes;
3966 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3967 mutex_unlock(&ctl->cache_writeout_mutex);
3969 ret = do_trimming(block_group, total_trimmed, start, bytes,
3970 start, bytes, 0, &trim_entry);
3972 reset_trimming_bitmap(ctl, offset);
3973 block_group->discard_cursor =
3974 btrfs_block_group_end(block_group);
3979 offset += BITS_PER_BITMAP * ctl->unit;
3984 block_group->discard_cursor = start;
3986 if (fatal_signal_pending(current)) {
3987 if (start != offset)
3988 reset_trimming_bitmap(ctl, offset);
3997 block_group->discard_cursor = end;
4003 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4004 u64 *trimmed, u64 start, u64 end, u64 minlen)
4006 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4010 ASSERT(!btrfs_is_zoned(block_group->fs_info));
4014 spin_lock(&block_group->lock);
4015 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4016 spin_unlock(&block_group->lock);
4019 btrfs_freeze_block_group(block_group);
4020 spin_unlock(&block_group->lock);
4022 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4026 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4027 div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4028 /* If we ended in the middle of a bitmap, reset the trimming flag */
4030 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4032 btrfs_unfreeze_block_group(block_group);
4036 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4037 u64 *trimmed, u64 start, u64 end, u64 minlen,
4044 spin_lock(&block_group->lock);
4045 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4046 spin_unlock(&block_group->lock);
4049 btrfs_freeze_block_group(block_group);
4050 spin_unlock(&block_group->lock);
4052 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4053 btrfs_unfreeze_block_group(block_group);
4058 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4059 u64 *trimmed, u64 start, u64 end, u64 minlen,
4060 u64 maxlen, bool async)
4066 spin_lock(&block_group->lock);
4067 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4068 spin_unlock(&block_group->lock);
4071 btrfs_freeze_block_group(block_group);
4072 spin_unlock(&block_group->lock);
4074 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4077 btrfs_unfreeze_block_group(block_group);
4082 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4084 return btrfs_super_cache_generation(fs_info->super_copy);
4087 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4088 struct btrfs_trans_handle *trans)
4090 struct btrfs_block_group *block_group;
4091 struct rb_node *node;
4094 btrfs_info(fs_info, "cleaning free space cache v1");
4096 node = rb_first_cached(&fs_info->block_group_cache_tree);
4098 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4099 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4102 node = rb_next(node);
4108 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4110 struct btrfs_trans_handle *trans;
4114 * update_super_roots will appropriately set or unset
4115 * super_copy->cache_generation based on SPACE_CACHE and
4116 * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4117 * transaction commit whether we are enabling space cache v1 and don't
4118 * have any other work to do, or are disabling it and removing free
4121 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4123 return PTR_ERR(trans);
4126 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4127 ret = cleanup_free_space_cache_v1(fs_info, trans);
4129 btrfs_abort_transaction(trans, ret);
4130 btrfs_end_transaction(trans);
4135 ret = btrfs_commit_transaction(trans);
4137 clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4142 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4144 * Use this if you need to make a bitmap or extent entry specifically, it
4145 * doesn't do any of the merging that add_free_space does, this acts a lot like
4146 * how the free space cache loading stuff works, so you can get really weird
4149 int test_add_free_space_entry(struct btrfs_block_group *cache,
4150 u64 offset, u64 bytes, bool bitmap)
4152 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4153 struct btrfs_free_space *info = NULL, *bitmap_info;
4155 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4161 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4167 spin_lock(&ctl->tree_lock);
4168 info->offset = offset;
4169 info->bytes = bytes;
4170 info->max_extent_size = 0;
4171 ret = link_free_space(ctl, info);
4172 spin_unlock(&ctl->tree_lock);
4174 kmem_cache_free(btrfs_free_space_cachep, info);
4179 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4181 kmem_cache_free(btrfs_free_space_cachep, info);
4186 spin_lock(&ctl->tree_lock);
4187 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4192 add_new_bitmap(ctl, info, offset);
4197 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4200 bytes -= bytes_added;
4201 offset += bytes_added;
4202 spin_unlock(&ctl->tree_lock);
4208 kmem_cache_free(btrfs_free_space_cachep, info);
4210 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4215 * Checks to see if the given range is in the free space cache. This is really
4216 * just used to check the absence of space, so if there is free space in the
4217 * range at all we will return 1.
4219 int test_check_exists(struct btrfs_block_group *cache,
4220 u64 offset, u64 bytes)
4222 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4223 struct btrfs_free_space *info;
4226 spin_lock(&ctl->tree_lock);
4227 info = tree_search_offset(ctl, offset, 0, 0);
4229 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4237 u64 bit_off, bit_bytes;
4239 struct btrfs_free_space *tmp;
4242 bit_bytes = ctl->unit;
4243 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4245 if (bit_off == offset) {
4248 } else if (bit_off > offset &&
4249 offset + bytes > bit_off) {
4255 n = rb_prev(&info->offset_index);
4257 tmp = rb_entry(n, struct btrfs_free_space,
4259 if (tmp->offset + tmp->bytes < offset)
4261 if (offset + bytes < tmp->offset) {
4262 n = rb_prev(&tmp->offset_index);
4269 n = rb_next(&info->offset_index);
4271 tmp = rb_entry(n, struct btrfs_free_space,
4273 if (offset + bytes < tmp->offset)
4275 if (tmp->offset + tmp->bytes < offset) {
4276 n = rb_next(&tmp->offset_index);
4287 if (info->offset == offset) {
4292 if (offset > info->offset && offset < info->offset + info->bytes)
4295 spin_unlock(&ctl->tree_lock);
4298 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */