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>
18 #include "free-space-cache.h"
19 #include "transaction.h"
21 #include "extent_io.h"
23 #include "space-info.h"
24 #include "delalloc-space.h"
25 #include "block-group.h"
28 #include "inode-item.h"
29 #include "accessors.h"
30 #include "file-item.h"
34 #define BITS_PER_BITMAP (PAGE_SIZE * 8UL)
35 #define MAX_CACHE_BYTES_PER_GIG SZ_64K
36 #define FORCE_EXTENT_THRESHOLD SZ_1M
38 static struct kmem_cache *btrfs_free_space_cachep;
39 static struct kmem_cache *btrfs_free_space_bitmap_cachep;
41 struct btrfs_trim_range {
44 struct list_head list;
47 static int link_free_space(struct btrfs_free_space_ctl *ctl,
48 struct btrfs_free_space *info);
49 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
50 struct btrfs_free_space *info, bool update_stat);
51 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
52 struct btrfs_free_space *bitmap_info, u64 *offset,
53 u64 *bytes, bool for_alloc);
54 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
55 struct btrfs_free_space *bitmap_info);
56 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
57 struct btrfs_free_space *info, u64 offset,
58 u64 bytes, bool update_stats);
60 static void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
62 struct btrfs_free_space *info;
65 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
66 info = rb_entry(node, struct btrfs_free_space, offset_index);
68 unlink_free_space(ctl, info, true);
69 kmem_cache_free(btrfs_free_space_cachep, info);
71 free_bitmap(ctl, info);
74 cond_resched_lock(&ctl->tree_lock);
78 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
79 struct btrfs_path *path,
82 struct btrfs_fs_info *fs_info = root->fs_info;
84 struct btrfs_key location;
85 struct btrfs_disk_key disk_key;
86 struct btrfs_free_space_header *header;
87 struct extent_buffer *leaf;
88 struct inode *inode = NULL;
92 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
96 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
100 btrfs_release_path(path);
101 return ERR_PTR(-ENOENT);
104 leaf = path->nodes[0];
105 header = btrfs_item_ptr(leaf, path->slots[0],
106 struct btrfs_free_space_header);
107 btrfs_free_space_key(leaf, header, &disk_key);
108 btrfs_disk_key_to_cpu(&location, &disk_key);
109 btrfs_release_path(path);
112 * We are often under a trans handle at this point, so we need to make
113 * sure NOFS is set to keep us from deadlocking.
115 nofs_flag = memalloc_nofs_save();
116 inode = btrfs_iget_path(fs_info->sb, location.objectid, root, path);
117 btrfs_release_path(path);
118 memalloc_nofs_restore(nofs_flag);
122 mapping_set_gfp_mask(inode->i_mapping,
123 mapping_gfp_constraint(inode->i_mapping,
124 ~(__GFP_FS | __GFP_HIGHMEM)));
129 struct inode *lookup_free_space_inode(struct btrfs_block_group *block_group,
130 struct btrfs_path *path)
132 struct btrfs_fs_info *fs_info = block_group->fs_info;
133 struct inode *inode = NULL;
134 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
136 spin_lock(&block_group->lock);
137 if (block_group->inode)
138 inode = igrab(block_group->inode);
139 spin_unlock(&block_group->lock);
143 inode = __lookup_free_space_inode(fs_info->tree_root, path,
148 spin_lock(&block_group->lock);
149 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
150 btrfs_info(fs_info, "Old style space inode found, converting.");
151 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
152 BTRFS_INODE_NODATACOW;
153 block_group->disk_cache_state = BTRFS_DC_CLEAR;
156 if (!test_and_set_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags))
157 block_group->inode = igrab(inode);
158 spin_unlock(&block_group->lock);
163 static int __create_free_space_inode(struct btrfs_root *root,
164 struct btrfs_trans_handle *trans,
165 struct btrfs_path *path,
168 struct btrfs_key key;
169 struct btrfs_disk_key disk_key;
170 struct btrfs_free_space_header *header;
171 struct btrfs_inode_item *inode_item;
172 struct extent_buffer *leaf;
173 /* We inline CRCs for the free disk space cache */
174 const u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC |
175 BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
178 ret = btrfs_insert_empty_inode(trans, root, path, ino);
182 leaf = path->nodes[0];
183 inode_item = btrfs_item_ptr(leaf, path->slots[0],
184 struct btrfs_inode_item);
185 btrfs_item_key(leaf, &disk_key, path->slots[0]);
186 memzero_extent_buffer(leaf, (unsigned long)inode_item,
187 sizeof(*inode_item));
188 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
189 btrfs_set_inode_size(leaf, inode_item, 0);
190 btrfs_set_inode_nbytes(leaf, inode_item, 0);
191 btrfs_set_inode_uid(leaf, inode_item, 0);
192 btrfs_set_inode_gid(leaf, inode_item, 0);
193 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
194 btrfs_set_inode_flags(leaf, inode_item, flags);
195 btrfs_set_inode_nlink(leaf, inode_item, 1);
196 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
197 btrfs_set_inode_block_group(leaf, inode_item, offset);
198 btrfs_mark_buffer_dirty(leaf);
199 btrfs_release_path(path);
201 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
204 ret = btrfs_insert_empty_item(trans, root, path, &key,
205 sizeof(struct btrfs_free_space_header));
207 btrfs_release_path(path);
211 leaf = path->nodes[0];
212 header = btrfs_item_ptr(leaf, path->slots[0],
213 struct btrfs_free_space_header);
214 memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
215 btrfs_set_free_space_key(leaf, header, &disk_key);
216 btrfs_mark_buffer_dirty(leaf);
217 btrfs_release_path(path);
222 int create_free_space_inode(struct btrfs_trans_handle *trans,
223 struct btrfs_block_group *block_group,
224 struct btrfs_path *path)
229 ret = btrfs_get_free_objectid(trans->fs_info->tree_root, &ino);
233 return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
234 ino, block_group->start);
238 * inode is an optional sink: if it is NULL, btrfs_remove_free_space_inode
239 * handles lookup, otherwise it takes ownership and iputs the inode.
240 * Don't reuse an inode pointer after passing it into this function.
242 int btrfs_remove_free_space_inode(struct btrfs_trans_handle *trans,
244 struct btrfs_block_group *block_group)
246 struct btrfs_path *path;
247 struct btrfs_key key;
250 path = btrfs_alloc_path();
255 inode = lookup_free_space_inode(block_group, path);
257 if (PTR_ERR(inode) != -ENOENT)
258 ret = PTR_ERR(inode);
261 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
263 btrfs_add_delayed_iput(BTRFS_I(inode));
267 /* One for the block groups ref */
268 spin_lock(&block_group->lock);
269 if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, &block_group->runtime_flags)) {
270 block_group->inode = NULL;
271 spin_unlock(&block_group->lock);
274 spin_unlock(&block_group->lock);
276 /* One for the lookup ref */
277 btrfs_add_delayed_iput(BTRFS_I(inode));
279 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
281 key.offset = block_group->start;
282 ret = btrfs_search_slot(trans, trans->fs_info->tree_root, &key, path,
289 ret = btrfs_del_item(trans, trans->fs_info->tree_root, path);
291 btrfs_free_path(path);
295 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
296 struct btrfs_block_rsv *rsv)
301 /* 1 for slack space, 1 for updating the inode */
302 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
303 btrfs_calc_metadata_size(fs_info, 1);
305 spin_lock(&rsv->lock);
306 if (rsv->reserved < needed_bytes)
310 spin_unlock(&rsv->lock);
314 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
315 struct btrfs_block_group *block_group,
316 struct inode *vfs_inode)
318 struct btrfs_truncate_control control = {
319 .inode = BTRFS_I(vfs_inode),
321 .ino = btrfs_ino(BTRFS_I(vfs_inode)),
322 .min_type = BTRFS_EXTENT_DATA_KEY,
323 .clear_extent_range = true,
325 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
326 struct btrfs_root *root = inode->root;
327 struct extent_state *cached_state = NULL;
332 struct btrfs_path *path = btrfs_alloc_path();
339 mutex_lock(&trans->transaction->cache_write_mutex);
340 if (!list_empty(&block_group->io_list)) {
341 list_del_init(&block_group->io_list);
343 btrfs_wait_cache_io(trans, block_group, path);
344 btrfs_put_block_group(block_group);
348 * now that we've truncated the cache away, its no longer
351 spin_lock(&block_group->lock);
352 block_group->disk_cache_state = BTRFS_DC_CLEAR;
353 spin_unlock(&block_group->lock);
354 btrfs_free_path(path);
357 btrfs_i_size_write(inode, 0);
358 truncate_pagecache(vfs_inode, 0);
360 lock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
361 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
364 * We skip the throttling logic for free space cache inodes, so we don't
365 * need to check for -EAGAIN.
367 ret = btrfs_truncate_inode_items(trans, root, &control);
369 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
370 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
372 unlock_extent(&inode->io_tree, 0, (u64)-1, &cached_state);
376 ret = btrfs_update_inode(trans, root, inode);
380 mutex_unlock(&trans->transaction->cache_write_mutex);
382 btrfs_abort_transaction(trans, ret);
387 static void readahead_cache(struct inode *inode)
389 struct file_ra_state ra;
390 unsigned long last_index;
392 file_ra_state_init(&ra, inode->i_mapping);
393 last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
395 page_cache_sync_readahead(inode->i_mapping, &ra, NULL, 0, last_index);
398 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
403 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
405 /* Make sure we can fit our crcs and generation into the first page */
406 if (write && (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
409 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
411 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
415 io_ctl->num_pages = num_pages;
416 io_ctl->fs_info = btrfs_sb(inode->i_sb);
417 io_ctl->inode = inode;
421 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
423 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
425 kfree(io_ctl->pages);
426 io_ctl->pages = NULL;
429 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
437 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
439 ASSERT(io_ctl->index < io_ctl->num_pages);
440 io_ctl->page = io_ctl->pages[io_ctl->index++];
441 io_ctl->cur = page_address(io_ctl->page);
442 io_ctl->orig = io_ctl->cur;
443 io_ctl->size = PAGE_SIZE;
445 clear_page(io_ctl->cur);
448 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
452 io_ctl_unmap_page(io_ctl);
454 for (i = 0; i < io_ctl->num_pages; i++) {
455 if (io_ctl->pages[i]) {
456 btrfs_page_clear_checked(io_ctl->fs_info,
458 page_offset(io_ctl->pages[i]),
460 unlock_page(io_ctl->pages[i]);
461 put_page(io_ctl->pages[i]);
466 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, bool uptodate)
469 struct inode *inode = io_ctl->inode;
470 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
473 for (i = 0; i < io_ctl->num_pages; i++) {
476 page = find_or_create_page(inode->i_mapping, i, mask);
478 io_ctl_drop_pages(io_ctl);
482 ret = set_page_extent_mapped(page);
486 io_ctl_drop_pages(io_ctl);
490 io_ctl->pages[i] = page;
491 if (uptodate && !PageUptodate(page)) {
492 btrfs_read_folio(NULL, page_folio(page));
494 if (page->mapping != inode->i_mapping) {
495 btrfs_err(BTRFS_I(inode)->root->fs_info,
496 "free space cache page truncated");
497 io_ctl_drop_pages(io_ctl);
500 if (!PageUptodate(page)) {
501 btrfs_err(BTRFS_I(inode)->root->fs_info,
502 "error reading free space cache");
503 io_ctl_drop_pages(io_ctl);
509 for (i = 0; i < io_ctl->num_pages; i++)
510 clear_page_dirty_for_io(io_ctl->pages[i]);
515 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
517 io_ctl_map_page(io_ctl, 1);
520 * Skip the csum areas. If we don't check crcs then we just have a
521 * 64bit chunk at the front of the first page.
523 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
524 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
526 put_unaligned_le64(generation, io_ctl->cur);
527 io_ctl->cur += sizeof(u64);
530 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
535 * Skip the crc area. If we don't check crcs then we just have a 64bit
536 * chunk at the front of the first page.
538 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
539 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
541 cache_gen = get_unaligned_le64(io_ctl->cur);
542 if (cache_gen != generation) {
543 btrfs_err_rl(io_ctl->fs_info,
544 "space cache generation (%llu) does not match inode (%llu)",
545 cache_gen, generation);
546 io_ctl_unmap_page(io_ctl);
549 io_ctl->cur += sizeof(u64);
553 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
560 offset = sizeof(u32) * io_ctl->num_pages;
562 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
563 btrfs_crc32c_final(crc, (u8 *)&crc);
564 io_ctl_unmap_page(io_ctl);
565 tmp = page_address(io_ctl->pages[0]);
570 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
577 offset = sizeof(u32) * io_ctl->num_pages;
579 tmp = page_address(io_ctl->pages[0]);
583 io_ctl_map_page(io_ctl, 0);
584 crc = btrfs_crc32c(crc, io_ctl->orig + offset, PAGE_SIZE - offset);
585 btrfs_crc32c_final(crc, (u8 *)&crc);
587 btrfs_err_rl(io_ctl->fs_info,
588 "csum mismatch on free space cache");
589 io_ctl_unmap_page(io_ctl);
596 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
599 struct btrfs_free_space_entry *entry;
605 put_unaligned_le64(offset, &entry->offset);
606 put_unaligned_le64(bytes, &entry->bytes);
607 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
608 BTRFS_FREE_SPACE_EXTENT;
609 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
610 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
612 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
615 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
617 /* No more pages to map */
618 if (io_ctl->index >= io_ctl->num_pages)
621 /* map the next page */
622 io_ctl_map_page(io_ctl, 1);
626 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
632 * If we aren't at the start of the current page, unmap this one and
633 * map the next one if there is any left.
635 if (io_ctl->cur != io_ctl->orig) {
636 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
637 if (io_ctl->index >= io_ctl->num_pages)
639 io_ctl_map_page(io_ctl, 0);
642 copy_page(io_ctl->cur, bitmap);
643 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
644 if (io_ctl->index < io_ctl->num_pages)
645 io_ctl_map_page(io_ctl, 0);
649 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
652 * If we're not on the boundary we know we've modified the page and we
653 * need to crc the page.
655 if (io_ctl->cur != io_ctl->orig)
656 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
658 io_ctl_unmap_page(io_ctl);
660 while (io_ctl->index < io_ctl->num_pages) {
661 io_ctl_map_page(io_ctl, 1);
662 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
666 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
667 struct btrfs_free_space *entry, u8 *type)
669 struct btrfs_free_space_entry *e;
673 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
679 entry->offset = get_unaligned_le64(&e->offset);
680 entry->bytes = get_unaligned_le64(&e->bytes);
682 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
683 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
685 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
688 io_ctl_unmap_page(io_ctl);
693 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
694 struct btrfs_free_space *entry)
698 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
702 copy_page(entry->bitmap, io_ctl->cur);
703 io_ctl_unmap_page(io_ctl);
708 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
710 struct btrfs_block_group *block_group = ctl->block_group;
714 u64 size = block_group->length;
715 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
716 u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
718 max_bitmaps = max_t(u64, max_bitmaps, 1);
720 if (ctl->total_bitmaps > max_bitmaps)
721 btrfs_err(block_group->fs_info,
722 "invalid free space control: bg start=%llu len=%llu total_bitmaps=%u unit=%u max_bitmaps=%llu bytes_per_bg=%llu",
723 block_group->start, block_group->length,
724 ctl->total_bitmaps, ctl->unit, max_bitmaps,
726 ASSERT(ctl->total_bitmaps <= max_bitmaps);
729 * We are trying to keep the total amount of memory used per 1GiB of
730 * space to be MAX_CACHE_BYTES_PER_GIG. However, with a reclamation
731 * mechanism of pulling extents >= FORCE_EXTENT_THRESHOLD out of
732 * bitmaps, we may end up using more memory than this.
735 max_bytes = MAX_CACHE_BYTES_PER_GIG;
737 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
739 bitmap_bytes = ctl->total_bitmaps * ctl->unit;
742 * we want the extent entry threshold to always be at most 1/2 the max
743 * bytes we can have, or whatever is less than that.
745 extent_bytes = max_bytes - bitmap_bytes;
746 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
748 ctl->extents_thresh =
749 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
752 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
753 struct btrfs_free_space_ctl *ctl,
754 struct btrfs_path *path, u64 offset)
756 struct btrfs_fs_info *fs_info = root->fs_info;
757 struct btrfs_free_space_header *header;
758 struct extent_buffer *leaf;
759 struct btrfs_io_ctl io_ctl;
760 struct btrfs_key key;
761 struct btrfs_free_space *e, *n;
769 /* Nothing in the space cache, goodbye */
770 if (!i_size_read(inode))
773 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
777 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
781 btrfs_release_path(path);
787 leaf = path->nodes[0];
788 header = btrfs_item_ptr(leaf, path->slots[0],
789 struct btrfs_free_space_header);
790 num_entries = btrfs_free_space_entries(leaf, header);
791 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
792 generation = btrfs_free_space_generation(leaf, header);
793 btrfs_release_path(path);
795 if (!BTRFS_I(inode)->generation) {
797 "the free space cache file (%llu) is invalid, skip it",
802 if (BTRFS_I(inode)->generation != generation) {
804 "free space inode generation (%llu) did not match free space cache generation (%llu)",
805 BTRFS_I(inode)->generation, generation);
812 ret = io_ctl_init(&io_ctl, inode, 0);
816 readahead_cache(inode);
818 ret = io_ctl_prepare_pages(&io_ctl, true);
822 ret = io_ctl_check_crc(&io_ctl, 0);
826 ret = io_ctl_check_generation(&io_ctl, generation);
830 while (num_entries) {
831 e = kmem_cache_zalloc(btrfs_free_space_cachep,
838 ret = io_ctl_read_entry(&io_ctl, e, &type);
840 kmem_cache_free(btrfs_free_space_cachep, e);
846 kmem_cache_free(btrfs_free_space_cachep, e);
850 if (type == BTRFS_FREE_SPACE_EXTENT) {
851 spin_lock(&ctl->tree_lock);
852 ret = link_free_space(ctl, e);
853 spin_unlock(&ctl->tree_lock);
856 "Duplicate entries in free space cache, dumping");
857 kmem_cache_free(btrfs_free_space_cachep, e);
863 e->bitmap = kmem_cache_zalloc(
864 btrfs_free_space_bitmap_cachep, GFP_NOFS);
868 btrfs_free_space_cachep, e);
871 spin_lock(&ctl->tree_lock);
872 ret = link_free_space(ctl, e);
874 spin_unlock(&ctl->tree_lock);
876 "Duplicate entries in free space cache, dumping");
877 kmem_cache_free(btrfs_free_space_cachep, e);
880 ctl->total_bitmaps++;
881 recalculate_thresholds(ctl);
882 spin_unlock(&ctl->tree_lock);
883 list_add_tail(&e->list, &bitmaps);
889 io_ctl_unmap_page(&io_ctl);
892 * We add the bitmaps at the end of the entries in order that
893 * the bitmap entries are added to the cache.
895 list_for_each_entry_safe(e, n, &bitmaps, list) {
896 list_del_init(&e->list);
897 ret = io_ctl_read_bitmap(&io_ctl, e);
902 io_ctl_drop_pages(&io_ctl);
905 io_ctl_free(&io_ctl);
908 io_ctl_drop_pages(&io_ctl);
910 spin_lock(&ctl->tree_lock);
911 __btrfs_remove_free_space_cache(ctl);
912 spin_unlock(&ctl->tree_lock);
916 static int copy_free_space_cache(struct btrfs_block_group *block_group,
917 struct btrfs_free_space_ctl *ctl)
919 struct btrfs_free_space *info;
923 while (!ret && (n = rb_first(&ctl->free_space_offset)) != NULL) {
924 info = rb_entry(n, struct btrfs_free_space, offset_index);
926 unlink_free_space(ctl, info, true);
927 ret = btrfs_add_free_space(block_group, info->offset,
929 kmem_cache_free(btrfs_free_space_cachep, info);
931 u64 offset = info->offset;
932 u64 bytes = ctl->unit;
934 while (search_bitmap(ctl, info, &offset, &bytes,
936 ret = btrfs_add_free_space(block_group, offset,
940 bitmap_clear_bits(ctl, info, offset, bytes, true);
941 offset = info->offset;
944 free_bitmap(ctl, info);
951 static struct lock_class_key btrfs_free_space_inode_key;
953 int load_free_space_cache(struct btrfs_block_group *block_group)
955 struct btrfs_fs_info *fs_info = block_group->fs_info;
956 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
957 struct btrfs_free_space_ctl tmp_ctl = {};
959 struct btrfs_path *path;
962 u64 used = block_group->used;
965 * Because we could potentially discard our loaded free space, we want
966 * to load everything into a temporary structure first, and then if it's
967 * valid copy it all into the actual free space ctl.
969 btrfs_init_free_space_ctl(block_group, &tmp_ctl);
972 * If this block group has been marked to be cleared for one reason or
973 * another then we can't trust the on disk cache, so just return.
975 spin_lock(&block_group->lock);
976 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
977 spin_unlock(&block_group->lock);
980 spin_unlock(&block_group->lock);
982 path = btrfs_alloc_path();
985 path->search_commit_root = 1;
986 path->skip_locking = 1;
989 * We must pass a path with search_commit_root set to btrfs_iget in
990 * order to avoid a deadlock when allocating extents for the tree root.
992 * When we are COWing an extent buffer from the tree root, when looking
993 * for a free extent, at extent-tree.c:find_free_extent(), we can find
994 * block group without its free space cache loaded. When we find one
995 * we must load its space cache which requires reading its free space
996 * cache's inode item from the root tree. If this inode item is located
997 * in the same leaf that we started COWing before, then we end up in
998 * deadlock on the extent buffer (trying to read lock it when we
999 * previously write locked it).
1001 * It's safe to read the inode item using the commit root because
1002 * block groups, once loaded, stay in memory forever (until they are
1003 * removed) as well as their space caches once loaded. New block groups
1004 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
1005 * we will never try to read their inode item while the fs is mounted.
1007 inode = lookup_free_space_inode(block_group, path);
1008 if (IS_ERR(inode)) {
1009 btrfs_free_path(path);
1013 /* We may have converted the inode and made the cache invalid. */
1014 spin_lock(&block_group->lock);
1015 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
1016 spin_unlock(&block_group->lock);
1017 btrfs_free_path(path);
1020 spin_unlock(&block_group->lock);
1023 * Reinitialize the class of struct inode's mapping->invalidate_lock for
1024 * free space inodes to prevent false positives related to locks for normal
1027 lockdep_set_class(&(&inode->i_data)->invalidate_lock,
1028 &btrfs_free_space_inode_key);
1030 ret = __load_free_space_cache(fs_info->tree_root, inode, &tmp_ctl,
1031 path, block_group->start);
1032 btrfs_free_path(path);
1036 matched = (tmp_ctl.free_space == (block_group->length - used -
1037 block_group->bytes_super));
1040 ret = copy_free_space_cache(block_group, &tmp_ctl);
1042 * ret == 1 means we successfully loaded the free space cache,
1043 * so we need to re-set it here.
1049 * We need to call the _locked variant so we don't try to update
1050 * the discard counters.
1052 spin_lock(&tmp_ctl.tree_lock);
1053 __btrfs_remove_free_space_cache(&tmp_ctl);
1054 spin_unlock(&tmp_ctl.tree_lock);
1056 "block group %llu has wrong amount of free space",
1057 block_group->start);
1062 /* This cache is bogus, make sure it gets cleared */
1063 spin_lock(&block_group->lock);
1064 block_group->disk_cache_state = BTRFS_DC_CLEAR;
1065 spin_unlock(&block_group->lock);
1069 "failed to load free space cache for block group %llu, rebuilding it now",
1070 block_group->start);
1073 spin_lock(&ctl->tree_lock);
1074 btrfs_discard_update_discardable(block_group);
1075 spin_unlock(&ctl->tree_lock);
1080 static noinline_for_stack
1081 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
1082 struct btrfs_free_space_ctl *ctl,
1083 struct btrfs_block_group *block_group,
1084 int *entries, int *bitmaps,
1085 struct list_head *bitmap_list)
1088 struct btrfs_free_cluster *cluster = NULL;
1089 struct btrfs_free_cluster *cluster_locked = NULL;
1090 struct rb_node *node = rb_first(&ctl->free_space_offset);
1091 struct btrfs_trim_range *trim_entry;
1093 /* Get the cluster for this block_group if it exists */
1094 if (block_group && !list_empty(&block_group->cluster_list)) {
1095 cluster = list_entry(block_group->cluster_list.next,
1096 struct btrfs_free_cluster,
1100 if (!node && cluster) {
1101 cluster_locked = cluster;
1102 spin_lock(&cluster_locked->lock);
1103 node = rb_first(&cluster->root);
1107 /* Write out the extent entries */
1109 struct btrfs_free_space *e;
1111 e = rb_entry(node, struct btrfs_free_space, offset_index);
1114 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
1120 list_add_tail(&e->list, bitmap_list);
1123 node = rb_next(node);
1124 if (!node && cluster) {
1125 node = rb_first(&cluster->root);
1126 cluster_locked = cluster;
1127 spin_lock(&cluster_locked->lock);
1131 if (cluster_locked) {
1132 spin_unlock(&cluster_locked->lock);
1133 cluster_locked = NULL;
1137 * Make sure we don't miss any range that was removed from our rbtree
1138 * because trimming is running. Otherwise after a umount+mount (or crash
1139 * after committing the transaction) we would leak free space and get
1140 * an inconsistent free space cache report from fsck.
1142 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
1143 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
1144 trim_entry->bytes, NULL);
1153 spin_unlock(&cluster_locked->lock);
1157 static noinline_for_stack int
1158 update_cache_item(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root,
1160 struct inode *inode,
1161 struct btrfs_path *path, u64 offset,
1162 int entries, int bitmaps)
1164 struct btrfs_key key;
1165 struct btrfs_free_space_header *header;
1166 struct extent_buffer *leaf;
1169 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1170 key.offset = offset;
1173 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1175 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1176 EXTENT_DELALLOC, NULL);
1179 leaf = path->nodes[0];
1181 struct btrfs_key found_key;
1182 ASSERT(path->slots[0]);
1184 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1185 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1186 found_key.offset != offset) {
1187 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1188 inode->i_size - 1, EXTENT_DELALLOC,
1190 btrfs_release_path(path);
1195 BTRFS_I(inode)->generation = trans->transid;
1196 header = btrfs_item_ptr(leaf, path->slots[0],
1197 struct btrfs_free_space_header);
1198 btrfs_set_free_space_entries(leaf, header, entries);
1199 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1200 btrfs_set_free_space_generation(leaf, header, trans->transid);
1201 btrfs_mark_buffer_dirty(leaf);
1202 btrfs_release_path(path);
1210 static noinline_for_stack int write_pinned_extent_entries(
1211 struct btrfs_trans_handle *trans,
1212 struct btrfs_block_group *block_group,
1213 struct btrfs_io_ctl *io_ctl,
1216 u64 start, extent_start, extent_end, len;
1217 struct extent_io_tree *unpin = NULL;
1224 * We want to add any pinned extents to our free space cache
1225 * so we don't leak the space
1227 * We shouldn't have switched the pinned extents yet so this is the
1230 unpin = &trans->transaction->pinned_extents;
1232 start = block_group->start;
1234 while (start < block_group->start + block_group->length) {
1235 ret = find_first_extent_bit(unpin, start,
1236 &extent_start, &extent_end,
1237 EXTENT_DIRTY, NULL);
1241 /* This pinned extent is out of our range */
1242 if (extent_start >= block_group->start + block_group->length)
1245 extent_start = max(extent_start, start);
1246 extent_end = min(block_group->start + block_group->length,
1248 len = extent_end - extent_start;
1251 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1261 static noinline_for_stack int
1262 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1264 struct btrfs_free_space *entry, *next;
1267 /* Write out the bitmaps */
1268 list_for_each_entry_safe(entry, next, bitmap_list, list) {
1269 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1272 list_del_init(&entry->list);
1278 static int flush_dirty_cache(struct inode *inode)
1282 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1284 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1285 EXTENT_DELALLOC, NULL);
1290 static void noinline_for_stack
1291 cleanup_bitmap_list(struct list_head *bitmap_list)
1293 struct btrfs_free_space *entry, *next;
1295 list_for_each_entry_safe(entry, next, bitmap_list, list)
1296 list_del_init(&entry->list);
1299 static void noinline_for_stack
1300 cleanup_write_cache_enospc(struct inode *inode,
1301 struct btrfs_io_ctl *io_ctl,
1302 struct extent_state **cached_state)
1304 io_ctl_drop_pages(io_ctl);
1305 unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1309 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1310 struct btrfs_trans_handle *trans,
1311 struct btrfs_block_group *block_group,
1312 struct btrfs_io_ctl *io_ctl,
1313 struct btrfs_path *path, u64 offset)
1316 struct inode *inode = io_ctl->inode;
1321 /* Flush the dirty pages in the cache file. */
1322 ret = flush_dirty_cache(inode);
1326 /* Update the cache item to tell everyone this cache file is valid. */
1327 ret = update_cache_item(trans, root, inode, path, offset,
1328 io_ctl->entries, io_ctl->bitmaps);
1331 invalidate_inode_pages2(inode->i_mapping);
1332 BTRFS_I(inode)->generation = 0;
1334 btrfs_debug(root->fs_info,
1335 "failed to write free space cache for block group %llu error %d",
1336 block_group->start, ret);
1338 btrfs_update_inode(trans, root, BTRFS_I(inode));
1341 /* the dirty list is protected by the dirty_bgs_lock */
1342 spin_lock(&trans->transaction->dirty_bgs_lock);
1344 /* the disk_cache_state is protected by the block group lock */
1345 spin_lock(&block_group->lock);
1348 * only mark this as written if we didn't get put back on
1349 * the dirty list while waiting for IO. Otherwise our
1350 * cache state won't be right, and we won't get written again
1352 if (!ret && list_empty(&block_group->dirty_list))
1353 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1355 block_group->disk_cache_state = BTRFS_DC_ERROR;
1357 spin_unlock(&block_group->lock);
1358 spin_unlock(&trans->transaction->dirty_bgs_lock);
1359 io_ctl->inode = NULL;
1367 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1368 struct btrfs_block_group *block_group,
1369 struct btrfs_path *path)
1371 return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1372 block_group, &block_group->io_ctl,
1373 path, block_group->start);
1377 * Write out cached info to an inode.
1379 * @root: root the inode belongs to
1380 * @inode: freespace inode we are writing out
1381 * @ctl: free space cache we are going to write out
1382 * @block_group: block_group for this cache if it belongs to a block_group
1383 * @io_ctl: holds context for the io
1384 * @trans: the trans handle
1386 * This function writes out a free space cache struct to disk for quick recovery
1387 * on mount. This will return 0 if it was successful in writing the cache out,
1388 * or an errno if it was not.
1390 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1391 struct btrfs_free_space_ctl *ctl,
1392 struct btrfs_block_group *block_group,
1393 struct btrfs_io_ctl *io_ctl,
1394 struct btrfs_trans_handle *trans)
1396 struct extent_state *cached_state = NULL;
1397 LIST_HEAD(bitmap_list);
1403 if (!i_size_read(inode))
1406 WARN_ON(io_ctl->pages);
1407 ret = io_ctl_init(io_ctl, inode, 1);
1411 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1412 down_write(&block_group->data_rwsem);
1413 spin_lock(&block_group->lock);
1414 if (block_group->delalloc_bytes) {
1415 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1416 spin_unlock(&block_group->lock);
1417 up_write(&block_group->data_rwsem);
1418 BTRFS_I(inode)->generation = 0;
1423 spin_unlock(&block_group->lock);
1426 /* Lock all pages first so we can lock the extent safely. */
1427 ret = io_ctl_prepare_pages(io_ctl, false);
1431 lock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1434 io_ctl_set_generation(io_ctl, trans->transid);
1436 mutex_lock(&ctl->cache_writeout_mutex);
1437 /* Write out the extent entries in the free space cache */
1438 spin_lock(&ctl->tree_lock);
1439 ret = write_cache_extent_entries(io_ctl, ctl,
1440 block_group, &entries, &bitmaps,
1443 goto out_nospc_locked;
1446 * Some spaces that are freed in the current transaction are pinned,
1447 * they will be added into free space cache after the transaction is
1448 * committed, we shouldn't lose them.
1450 * If this changes while we are working we'll get added back to
1451 * the dirty list and redo it. No locking needed
1453 ret = write_pinned_extent_entries(trans, block_group, io_ctl, &entries);
1455 goto out_nospc_locked;
1458 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1459 * locked while doing it because a concurrent trim can be manipulating
1460 * or freeing the bitmap.
1462 ret = write_bitmap_entries(io_ctl, &bitmap_list);
1463 spin_unlock(&ctl->tree_lock);
1464 mutex_unlock(&ctl->cache_writeout_mutex);
1468 /* Zero out the rest of the pages just to make sure */
1469 io_ctl_zero_remaining_pages(io_ctl);
1471 /* Everything is written out, now we dirty the pages in the file. */
1472 ret = btrfs_dirty_pages(BTRFS_I(inode), io_ctl->pages,
1473 io_ctl->num_pages, 0, i_size_read(inode),
1474 &cached_state, false);
1478 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1479 up_write(&block_group->data_rwsem);
1481 * Release the pages and unlock the extent, we will flush
1484 io_ctl_drop_pages(io_ctl);
1485 io_ctl_free(io_ctl);
1487 unlock_extent(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1491 * at this point the pages are under IO and we're happy,
1492 * The caller is responsible for waiting on them and updating
1493 * the cache and the inode
1495 io_ctl->entries = entries;
1496 io_ctl->bitmaps = bitmaps;
1498 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1505 cleanup_bitmap_list(&bitmap_list);
1506 spin_unlock(&ctl->tree_lock);
1507 mutex_unlock(&ctl->cache_writeout_mutex);
1510 cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1513 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1514 up_write(&block_group->data_rwsem);
1517 io_ctl->inode = NULL;
1518 io_ctl_free(io_ctl);
1520 invalidate_inode_pages2(inode->i_mapping);
1521 BTRFS_I(inode)->generation = 0;
1523 btrfs_update_inode(trans, root, BTRFS_I(inode));
1529 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1530 struct btrfs_block_group *block_group,
1531 struct btrfs_path *path)
1533 struct btrfs_fs_info *fs_info = trans->fs_info;
1534 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1535 struct inode *inode;
1538 spin_lock(&block_group->lock);
1539 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1540 spin_unlock(&block_group->lock);
1543 spin_unlock(&block_group->lock);
1545 inode = lookup_free_space_inode(block_group, path);
1549 ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1550 block_group, &block_group->io_ctl, trans);
1552 btrfs_debug(fs_info,
1553 "failed to write free space cache for block group %llu error %d",
1554 block_group->start, ret);
1555 spin_lock(&block_group->lock);
1556 block_group->disk_cache_state = BTRFS_DC_ERROR;
1557 spin_unlock(&block_group->lock);
1559 block_group->io_ctl.inode = NULL;
1564 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1565 * to wait for IO and put the inode
1571 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1574 ASSERT(offset >= bitmap_start);
1575 offset -= bitmap_start;
1576 return (unsigned long)(div_u64(offset, unit));
1579 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1581 return (unsigned long)(div_u64(bytes, unit));
1584 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1588 u64 bytes_per_bitmap;
1590 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1591 bitmap_start = offset - ctl->start;
1592 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1593 bitmap_start *= bytes_per_bitmap;
1594 bitmap_start += ctl->start;
1596 return bitmap_start;
1599 static int tree_insert_offset(struct rb_root *root, u64 offset,
1600 struct rb_node *node, int bitmap)
1602 struct rb_node **p = &root->rb_node;
1603 struct rb_node *parent = NULL;
1604 struct btrfs_free_space *info;
1608 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1610 if (offset < info->offset) {
1612 } else if (offset > info->offset) {
1613 p = &(*p)->rb_right;
1616 * we could have a bitmap entry and an extent entry
1617 * share the same offset. If this is the case, we want
1618 * the extent entry to always be found first if we do a
1619 * linear search through the tree, since we want to have
1620 * the quickest allocation time, and allocating from an
1621 * extent is faster than allocating from a bitmap. So
1622 * if we're inserting a bitmap and we find an entry at
1623 * this offset, we want to go right, or after this entry
1624 * logically. If we are inserting an extent and we've
1625 * found a bitmap, we want to go left, or before
1633 p = &(*p)->rb_right;
1635 if (!info->bitmap) {
1644 rb_link_node(node, parent, p);
1645 rb_insert_color(node, root);
1651 * This is a little subtle. We *only* have ->max_extent_size set if we actually
1652 * searched through the bitmap and figured out the largest ->max_extent_size,
1653 * otherwise it's 0. In the case that it's 0 we don't want to tell the
1654 * allocator the wrong thing, we want to use the actual real max_extent_size
1655 * we've found already if it's larger, or we want to use ->bytes.
1657 * This matters because find_free_space() will skip entries who's ->bytes is
1658 * less than the required bytes. So if we didn't search down this bitmap, we
1659 * may pick some previous entry that has a smaller ->max_extent_size than we
1660 * have. For example, assume we have two entries, one that has
1661 * ->max_extent_size set to 4K and ->bytes set to 1M. A second entry hasn't set
1662 * ->max_extent_size yet, has ->bytes set to 8K and it's contiguous. We will
1663 * call into find_free_space(), and return with max_extent_size == 4K, because
1664 * that first bitmap entry had ->max_extent_size set, but the second one did
1665 * not. If instead we returned 8K we'd come in searching for 8K, and find the
1666 * 8K contiguous range.
1668 * Consider the other case, we have 2 8K chunks in that second entry and still
1669 * don't have ->max_extent_size set. We'll return 16K, and the next time the
1670 * allocator comes in it'll fully search our second bitmap, and this time it'll
1671 * get an uptodate value of 8K as the maximum chunk size. Then we'll get the
1672 * right allocation the next loop through.
1674 static inline u64 get_max_extent_size(const struct btrfs_free_space *entry)
1676 if (entry->bitmap && entry->max_extent_size)
1677 return entry->max_extent_size;
1678 return entry->bytes;
1682 * We want the largest entry to be leftmost, so this is inverted from what you'd
1685 static bool entry_less(struct rb_node *node, const struct rb_node *parent)
1687 const struct btrfs_free_space *entry, *exist;
1689 entry = rb_entry(node, struct btrfs_free_space, bytes_index);
1690 exist = rb_entry(parent, struct btrfs_free_space, bytes_index);
1691 return get_max_extent_size(exist) < get_max_extent_size(entry);
1695 * searches the tree for the given offset.
1697 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1698 * want a section that has at least bytes size and comes at or after the given
1701 static struct btrfs_free_space *
1702 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1703 u64 offset, int bitmap_only, int fuzzy)
1705 struct rb_node *n = ctl->free_space_offset.rb_node;
1706 struct btrfs_free_space *entry = NULL, *prev = NULL;
1708 /* find entry that is closest to the 'offset' */
1710 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1713 if (offset < entry->offset)
1715 else if (offset > entry->offset)
1730 * bitmap entry and extent entry may share same offset,
1731 * in that case, bitmap entry comes after extent entry.
1736 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1737 if (entry->offset != offset)
1740 WARN_ON(!entry->bitmap);
1743 if (entry->bitmap) {
1745 * if previous extent entry covers the offset,
1746 * we should return it instead of the bitmap entry
1748 n = rb_prev(&entry->offset_index);
1750 prev = rb_entry(n, struct btrfs_free_space,
1752 if (!prev->bitmap &&
1753 prev->offset + prev->bytes > offset)
1763 /* find last entry before the 'offset' */
1765 if (entry->offset > offset) {
1766 n = rb_prev(&entry->offset_index);
1768 entry = rb_entry(n, struct btrfs_free_space,
1770 ASSERT(entry->offset <= offset);
1779 if (entry->bitmap) {
1780 n = rb_prev(&entry->offset_index);
1782 prev = rb_entry(n, struct btrfs_free_space,
1784 if (!prev->bitmap &&
1785 prev->offset + prev->bytes > offset)
1788 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1790 } else if (entry->offset + entry->bytes > offset)
1797 n = rb_next(&entry->offset_index);
1800 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1801 if (entry->bitmap) {
1802 if (entry->offset + BITS_PER_BITMAP *
1806 if (entry->offset + entry->bytes > offset)
1813 static inline void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1814 struct btrfs_free_space *info,
1817 rb_erase(&info->offset_index, &ctl->free_space_offset);
1818 rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1819 ctl->free_extents--;
1821 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1822 ctl->discardable_extents[BTRFS_STAT_CURR]--;
1823 ctl->discardable_bytes[BTRFS_STAT_CURR] -= info->bytes;
1827 ctl->free_space -= info->bytes;
1830 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1831 struct btrfs_free_space *info)
1835 ASSERT(info->bytes || info->bitmap);
1836 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1837 &info->offset_index, (info->bitmap != NULL));
1841 rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1843 if (!info->bitmap && !btrfs_free_space_trimmed(info)) {
1844 ctl->discardable_extents[BTRFS_STAT_CURR]++;
1845 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
1848 ctl->free_space += info->bytes;
1849 ctl->free_extents++;
1853 static void relink_bitmap_entry(struct btrfs_free_space_ctl *ctl,
1854 struct btrfs_free_space *info)
1856 ASSERT(info->bitmap);
1859 * If our entry is empty it's because we're on a cluster and we don't
1860 * want to re-link it into our ctl bytes index.
1862 if (RB_EMPTY_NODE(&info->bytes_index))
1865 rb_erase_cached(&info->bytes_index, &ctl->free_space_bytes);
1866 rb_add_cached(&info->bytes_index, &ctl->free_space_bytes, entry_less);
1869 static inline void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1870 struct btrfs_free_space *info,
1871 u64 offset, u64 bytes, bool update_stat)
1873 unsigned long start, count, end;
1874 int extent_delta = -1;
1876 start = offset_to_bit(info->offset, ctl->unit, offset);
1877 count = bytes_to_bits(bytes, ctl->unit);
1878 end = start + count;
1879 ASSERT(end <= BITS_PER_BITMAP);
1881 bitmap_clear(info->bitmap, start, count);
1883 info->bytes -= bytes;
1884 if (info->max_extent_size > ctl->unit)
1885 info->max_extent_size = 0;
1887 relink_bitmap_entry(ctl, info);
1889 if (start && test_bit(start - 1, info->bitmap))
1892 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1895 info->bitmap_extents += extent_delta;
1896 if (!btrfs_free_space_trimmed(info)) {
1897 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1898 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
1902 ctl->free_space -= bytes;
1905 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1906 struct btrfs_free_space *info, u64 offset,
1909 unsigned long start, count, end;
1910 int extent_delta = 1;
1912 start = offset_to_bit(info->offset, ctl->unit, offset);
1913 count = bytes_to_bits(bytes, ctl->unit);
1914 end = start + count;
1915 ASSERT(end <= BITS_PER_BITMAP);
1917 bitmap_set(info->bitmap, start, count);
1920 * We set some bytes, we have no idea what the max extent size is
1923 info->max_extent_size = 0;
1924 info->bytes += bytes;
1925 ctl->free_space += bytes;
1927 relink_bitmap_entry(ctl, info);
1929 if (start && test_bit(start - 1, info->bitmap))
1932 if (end < BITS_PER_BITMAP && test_bit(end, info->bitmap))
1935 info->bitmap_extents += extent_delta;
1936 if (!btrfs_free_space_trimmed(info)) {
1937 ctl->discardable_extents[BTRFS_STAT_CURR] += extent_delta;
1938 ctl->discardable_bytes[BTRFS_STAT_CURR] += bytes;
1943 * If we can not find suitable extent, we will use bytes to record
1944 * the size of the max extent.
1946 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1947 struct btrfs_free_space *bitmap_info, u64 *offset,
1948 u64 *bytes, bool for_alloc)
1950 unsigned long found_bits = 0;
1951 unsigned long max_bits = 0;
1952 unsigned long bits, i;
1953 unsigned long next_zero;
1954 unsigned long extent_bits;
1957 * Skip searching the bitmap if we don't have a contiguous section that
1958 * is large enough for this allocation.
1961 bitmap_info->max_extent_size &&
1962 bitmap_info->max_extent_size < *bytes) {
1963 *bytes = bitmap_info->max_extent_size;
1967 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1968 max_t(u64, *offset, bitmap_info->offset));
1969 bits = bytes_to_bits(*bytes, ctl->unit);
1971 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1972 if (for_alloc && bits == 1) {
1976 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1977 BITS_PER_BITMAP, i);
1978 extent_bits = next_zero - i;
1979 if (extent_bits >= bits) {
1980 found_bits = extent_bits;
1982 } else if (extent_bits > max_bits) {
1983 max_bits = extent_bits;
1989 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1990 *bytes = (u64)(found_bits) * ctl->unit;
1994 *bytes = (u64)(max_bits) * ctl->unit;
1995 bitmap_info->max_extent_size = *bytes;
1996 relink_bitmap_entry(ctl, bitmap_info);
2000 /* Cache the size of the max extent in bytes */
2001 static struct btrfs_free_space *
2002 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
2003 unsigned long align, u64 *max_extent_size, bool use_bytes_index)
2005 struct btrfs_free_space *entry;
2006 struct rb_node *node;
2011 if (!ctl->free_space_offset.rb_node)
2014 if (use_bytes_index) {
2015 node = rb_first_cached(&ctl->free_space_bytes);
2017 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset),
2021 node = &entry->offset_index;
2024 for (; node; node = rb_next(node)) {
2025 if (use_bytes_index)
2026 entry = rb_entry(node, struct btrfs_free_space,
2029 entry = rb_entry(node, struct btrfs_free_space,
2033 * If we are using the bytes index then all subsequent entries
2034 * in this tree are going to be < bytes, so simply set the max
2035 * extent size and exit the loop.
2037 * If we're using the offset index then we need to keep going
2038 * through the rest of the tree.
2040 if (entry->bytes < *bytes) {
2041 *max_extent_size = max(get_max_extent_size(entry),
2043 if (use_bytes_index)
2048 /* make sure the space returned is big enough
2049 * to match our requested alignment
2051 if (*bytes >= align) {
2052 tmp = entry->offset - ctl->start + align - 1;
2053 tmp = div64_u64(tmp, align);
2054 tmp = tmp * align + ctl->start;
2055 align_off = tmp - entry->offset;
2058 tmp = entry->offset;
2062 * We don't break here if we're using the bytes index because we
2063 * may have another entry that has the correct alignment that is
2064 * the right size, so we don't want to miss that possibility.
2065 * At worst this adds another loop through the logic, but if we
2066 * broke here we could prematurely ENOSPC.
2068 if (entry->bytes < *bytes + align_off) {
2069 *max_extent_size = max(get_max_extent_size(entry),
2074 if (entry->bitmap) {
2075 struct rb_node *old_next = rb_next(node);
2078 ret = search_bitmap(ctl, entry, &tmp, &size, true);
2085 max(get_max_extent_size(entry),
2090 * The bitmap may have gotten re-arranged in the space
2091 * index here because the max_extent_size may have been
2092 * updated. Start from the beginning again if this
2095 if (use_bytes_index && old_next != rb_next(node))
2101 *bytes = entry->bytes - align_off;
2108 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
2109 struct btrfs_free_space *info, u64 offset)
2111 info->offset = offset_to_bitmap(ctl, offset);
2113 info->bitmap_extents = 0;
2114 INIT_LIST_HEAD(&info->list);
2115 link_free_space(ctl, info);
2116 ctl->total_bitmaps++;
2117 recalculate_thresholds(ctl);
2120 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
2121 struct btrfs_free_space *bitmap_info)
2124 * Normally when this is called, the bitmap is completely empty. However,
2125 * if we are blowing up the free space cache for one reason or another
2126 * via __btrfs_remove_free_space_cache(), then it may not be freed and
2127 * we may leave stats on the table.
2129 if (bitmap_info->bytes && !btrfs_free_space_trimmed(bitmap_info)) {
2130 ctl->discardable_extents[BTRFS_STAT_CURR] -=
2131 bitmap_info->bitmap_extents;
2132 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bitmap_info->bytes;
2135 unlink_free_space(ctl, bitmap_info, true);
2136 kmem_cache_free(btrfs_free_space_bitmap_cachep, bitmap_info->bitmap);
2137 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
2138 ctl->total_bitmaps--;
2139 recalculate_thresholds(ctl);
2142 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
2143 struct btrfs_free_space *bitmap_info,
2144 u64 *offset, u64 *bytes)
2147 u64 search_start, search_bytes;
2151 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
2154 * We need to search for bits in this bitmap. We could only cover some
2155 * of the extent in this bitmap thanks to how we add space, so we need
2156 * to search for as much as it as we can and clear that amount, and then
2157 * go searching for the next bit.
2159 search_start = *offset;
2160 search_bytes = ctl->unit;
2161 search_bytes = min(search_bytes, end - search_start + 1);
2162 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
2164 if (ret < 0 || search_start != *offset)
2167 /* We may have found more bits than what we need */
2168 search_bytes = min(search_bytes, *bytes);
2170 /* Cannot clear past the end of the bitmap */
2171 search_bytes = min(search_bytes, end - search_start + 1);
2173 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes, true);
2174 *offset += search_bytes;
2175 *bytes -= search_bytes;
2178 struct rb_node *next = rb_next(&bitmap_info->offset_index);
2179 if (!bitmap_info->bytes)
2180 free_bitmap(ctl, bitmap_info);
2183 * no entry after this bitmap, but we still have bytes to
2184 * remove, so something has gone wrong.
2189 bitmap_info = rb_entry(next, struct btrfs_free_space,
2193 * if the next entry isn't a bitmap we need to return to let the
2194 * extent stuff do its work.
2196 if (!bitmap_info->bitmap)
2200 * Ok the next item is a bitmap, but it may not actually hold
2201 * the information for the rest of this free space stuff, so
2202 * look for it, and if we don't find it return so we can try
2203 * everything over again.
2205 search_start = *offset;
2206 search_bytes = ctl->unit;
2207 ret = search_bitmap(ctl, bitmap_info, &search_start,
2208 &search_bytes, false);
2209 if (ret < 0 || search_start != *offset)
2213 } else if (!bitmap_info->bytes)
2214 free_bitmap(ctl, bitmap_info);
2219 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
2220 struct btrfs_free_space *info, u64 offset,
2221 u64 bytes, enum btrfs_trim_state trim_state)
2223 u64 bytes_to_set = 0;
2227 * This is a tradeoff to make bitmap trim state minimal. We mark the
2228 * whole bitmap untrimmed if at any point we add untrimmed regions.
2230 if (trim_state == BTRFS_TRIM_STATE_UNTRIMMED) {
2231 if (btrfs_free_space_trimmed(info)) {
2232 ctl->discardable_extents[BTRFS_STAT_CURR] +=
2233 info->bitmap_extents;
2234 ctl->discardable_bytes[BTRFS_STAT_CURR] += info->bytes;
2236 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2239 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
2241 bytes_to_set = min(end - offset, bytes);
2243 bitmap_set_bits(ctl, info, offset, bytes_to_set);
2245 return bytes_to_set;
2249 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
2250 struct btrfs_free_space *info)
2252 struct btrfs_block_group *block_group = ctl->block_group;
2253 struct btrfs_fs_info *fs_info = block_group->fs_info;
2254 bool forced = false;
2256 #ifdef CONFIG_BTRFS_DEBUG
2257 if (btrfs_should_fragment_free_space(block_group))
2261 /* This is a way to reclaim large regions from the bitmaps. */
2262 if (!forced && info->bytes >= FORCE_EXTENT_THRESHOLD)
2266 * If we are below the extents threshold then we can add this as an
2267 * extent, and don't have to deal with the bitmap
2269 if (!forced && ctl->free_extents < ctl->extents_thresh) {
2271 * If this block group has some small extents we don't want to
2272 * use up all of our free slots in the cache with them, we want
2273 * to reserve them to larger extents, however if we have plenty
2274 * of cache left then go ahead an dadd them, no sense in adding
2275 * the overhead of a bitmap if we don't have to.
2277 if (info->bytes <= fs_info->sectorsize * 8) {
2278 if (ctl->free_extents * 3 <= ctl->extents_thresh)
2286 * The original block groups from mkfs can be really small, like 8
2287 * megabytes, so don't bother with a bitmap for those entries. However
2288 * some block groups can be smaller than what a bitmap would cover but
2289 * are still large enough that they could overflow the 32k memory limit,
2290 * so allow those block groups to still be allowed to have a bitmap
2293 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->length)
2299 static const struct btrfs_free_space_op free_space_op = {
2300 .use_bitmap = use_bitmap,
2303 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2304 struct btrfs_free_space *info)
2306 struct btrfs_free_space *bitmap_info;
2307 struct btrfs_block_group *block_group = NULL;
2309 u64 bytes, offset, bytes_added;
2310 enum btrfs_trim_state trim_state;
2313 bytes = info->bytes;
2314 offset = info->offset;
2315 trim_state = info->trim_state;
2317 if (!ctl->op->use_bitmap(ctl, info))
2320 if (ctl->op == &free_space_op)
2321 block_group = ctl->block_group;
2324 * Since we link bitmaps right into the cluster we need to see if we
2325 * have a cluster here, and if so and it has our bitmap we need to add
2326 * the free space to that bitmap.
2328 if (block_group && !list_empty(&block_group->cluster_list)) {
2329 struct btrfs_free_cluster *cluster;
2330 struct rb_node *node;
2331 struct btrfs_free_space *entry;
2333 cluster = list_entry(block_group->cluster_list.next,
2334 struct btrfs_free_cluster,
2336 spin_lock(&cluster->lock);
2337 node = rb_first(&cluster->root);
2339 spin_unlock(&cluster->lock);
2340 goto no_cluster_bitmap;
2343 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2344 if (!entry->bitmap) {
2345 spin_unlock(&cluster->lock);
2346 goto no_cluster_bitmap;
2349 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2350 bytes_added = add_bytes_to_bitmap(ctl, entry, offset,
2352 bytes -= bytes_added;
2353 offset += bytes_added;
2355 spin_unlock(&cluster->lock);
2363 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2370 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
2372 bytes -= bytes_added;
2373 offset += bytes_added;
2383 if (info && info->bitmap) {
2384 add_new_bitmap(ctl, info, offset);
2389 spin_unlock(&ctl->tree_lock);
2391 /* no pre-allocated info, allocate a new one */
2393 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2396 spin_lock(&ctl->tree_lock);
2402 /* allocate the bitmap */
2403 info->bitmap = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep,
2405 info->trim_state = BTRFS_TRIM_STATE_TRIMMED;
2406 spin_lock(&ctl->tree_lock);
2407 if (!info->bitmap) {
2417 kmem_cache_free(btrfs_free_space_bitmap_cachep,
2419 kmem_cache_free(btrfs_free_space_cachep, info);
2426 * Free space merging rules:
2427 * 1) Merge trimmed areas together
2428 * 2) Let untrimmed areas coalesce with trimmed areas
2429 * 3) Always pull neighboring regions from bitmaps
2431 * The above rules are for when we merge free space based on btrfs_trim_state.
2432 * Rules 2 and 3 are subtle because they are suboptimal, but are done for the
2433 * same reason: to promote larger extent regions which makes life easier for
2434 * find_free_extent(). Rule 2 enables coalescing based on the common path
2435 * being returning free space from btrfs_finish_extent_commit(). So when free
2436 * space is trimmed, it will prevent aggregating trimmed new region and
2437 * untrimmed regions in the rb_tree. Rule 3 is purely to obtain larger extents
2438 * and provide find_free_extent() with the largest extents possible hoping for
2441 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2442 struct btrfs_free_space *info, bool update_stat)
2444 struct btrfs_free_space *left_info = NULL;
2445 struct btrfs_free_space *right_info;
2446 bool merged = false;
2447 u64 offset = info->offset;
2448 u64 bytes = info->bytes;
2449 const bool is_trimmed = btrfs_free_space_trimmed(info);
2452 * first we want to see if there is free space adjacent to the range we
2453 * are adding, if there is remove that struct and add a new one to
2454 * cover the entire range
2456 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2457 if (right_info && rb_prev(&right_info->offset_index))
2458 left_info = rb_entry(rb_prev(&right_info->offset_index),
2459 struct btrfs_free_space, offset_index);
2460 else if (!right_info)
2461 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2463 /* See try_merge_free_space() comment. */
2464 if (right_info && !right_info->bitmap &&
2465 (!is_trimmed || btrfs_free_space_trimmed(right_info))) {
2466 unlink_free_space(ctl, right_info, update_stat);
2467 info->bytes += right_info->bytes;
2468 kmem_cache_free(btrfs_free_space_cachep, right_info);
2472 /* See try_merge_free_space() comment. */
2473 if (left_info && !left_info->bitmap &&
2474 left_info->offset + left_info->bytes == offset &&
2475 (!is_trimmed || btrfs_free_space_trimmed(left_info))) {
2476 unlink_free_space(ctl, left_info, update_stat);
2477 info->offset = left_info->offset;
2478 info->bytes += left_info->bytes;
2479 kmem_cache_free(btrfs_free_space_cachep, left_info);
2486 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2487 struct btrfs_free_space *info,
2490 struct btrfs_free_space *bitmap;
2493 const u64 end = info->offset + info->bytes;
2494 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2497 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2501 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2502 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2505 bytes = (j - i) * ctl->unit;
2506 info->bytes += bytes;
2508 /* See try_merge_free_space() comment. */
2509 if (!btrfs_free_space_trimmed(bitmap))
2510 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2512 bitmap_clear_bits(ctl, bitmap, end, bytes, update_stat);
2515 free_bitmap(ctl, bitmap);
2520 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2521 struct btrfs_free_space *info,
2524 struct btrfs_free_space *bitmap;
2528 unsigned long prev_j;
2531 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2532 /* If we're on a boundary, try the previous logical bitmap. */
2533 if (bitmap_offset == info->offset) {
2534 if (info->offset == 0)
2536 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2539 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2543 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2545 prev_j = (unsigned long)-1;
2546 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2554 if (prev_j == (unsigned long)-1)
2555 bytes = (i + 1) * ctl->unit;
2557 bytes = (i - prev_j) * ctl->unit;
2559 info->offset -= bytes;
2560 info->bytes += bytes;
2562 /* See try_merge_free_space() comment. */
2563 if (!btrfs_free_space_trimmed(bitmap))
2564 info->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2566 bitmap_clear_bits(ctl, bitmap, info->offset, bytes, update_stat);
2569 free_bitmap(ctl, bitmap);
2575 * We prefer always to allocate from extent entries, both for clustered and
2576 * non-clustered allocation requests. So when attempting to add a new extent
2577 * entry, try to see if there's adjacent free space in bitmap entries, and if
2578 * there is, migrate that space from the bitmaps to the extent.
2579 * Like this we get better chances of satisfying space allocation requests
2580 * because we attempt to satisfy them based on a single cache entry, and never
2581 * on 2 or more entries - even if the entries represent a contiguous free space
2582 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2585 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2586 struct btrfs_free_space *info,
2590 * Only work with disconnected entries, as we can change their offset,
2591 * and must be extent entries.
2593 ASSERT(!info->bitmap);
2594 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2596 if (ctl->total_bitmaps > 0) {
2598 bool stole_front = false;
2600 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2601 if (ctl->total_bitmaps > 0)
2602 stole_front = steal_from_bitmap_to_front(ctl, info,
2605 if (stole_end || stole_front)
2606 try_merge_free_space(ctl, info, update_stat);
2610 int __btrfs_add_free_space(struct btrfs_block_group *block_group,
2611 u64 offset, u64 bytes,
2612 enum btrfs_trim_state trim_state)
2614 struct btrfs_fs_info *fs_info = block_group->fs_info;
2615 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2616 struct btrfs_free_space *info;
2618 u64 filter_bytes = bytes;
2620 ASSERT(!btrfs_is_zoned(fs_info));
2622 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2626 info->offset = offset;
2627 info->bytes = bytes;
2628 info->trim_state = trim_state;
2629 RB_CLEAR_NODE(&info->offset_index);
2630 RB_CLEAR_NODE(&info->bytes_index);
2632 spin_lock(&ctl->tree_lock);
2634 if (try_merge_free_space(ctl, info, true))
2638 * There was no extent directly to the left or right of this new
2639 * extent then we know we're going to have to allocate a new extent, so
2640 * before we do that see if we need to drop this into a bitmap
2642 ret = insert_into_bitmap(ctl, info);
2651 * Only steal free space from adjacent bitmaps if we're sure we're not
2652 * going to add the new free space to existing bitmap entries - because
2653 * that would mean unnecessary work that would be reverted. Therefore
2654 * attempt to steal space from bitmaps if we're adding an extent entry.
2656 steal_from_bitmap(ctl, info, true);
2658 filter_bytes = max(filter_bytes, info->bytes);
2660 ret = link_free_space(ctl, info);
2662 kmem_cache_free(btrfs_free_space_cachep, info);
2664 btrfs_discard_update_discardable(block_group);
2665 spin_unlock(&ctl->tree_lock);
2668 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2669 ASSERT(ret != -EEXIST);
2672 if (trim_state != BTRFS_TRIM_STATE_TRIMMED) {
2673 btrfs_discard_check_filter(block_group, filter_bytes);
2674 btrfs_discard_queue_work(&fs_info->discard_ctl, block_group);
2680 static int __btrfs_add_free_space_zoned(struct btrfs_block_group *block_group,
2681 u64 bytenr, u64 size, bool used)
2683 struct btrfs_space_info *sinfo = block_group->space_info;
2684 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2685 u64 offset = bytenr - block_group->start;
2686 u64 to_free, to_unusable;
2687 int bg_reclaim_threshold = 0;
2688 bool initial = (size == block_group->length);
2689 u64 reclaimable_unusable;
2691 WARN_ON(!initial && offset + size > block_group->zone_capacity);
2694 bg_reclaim_threshold = READ_ONCE(sinfo->bg_reclaim_threshold);
2696 spin_lock(&ctl->tree_lock);
2697 /* Count initial region as zone_unusable until it gets activated. */
2701 test_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &block_group->fs_info->flags) &&
2702 (block_group->flags & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)))
2705 to_free = block_group->zone_capacity;
2706 else if (offset >= block_group->alloc_offset)
2708 else if (offset + size <= block_group->alloc_offset)
2711 to_free = offset + size - block_group->alloc_offset;
2712 to_unusable = size - to_free;
2714 ctl->free_space += to_free;
2716 * If the block group is read-only, we should account freed space into
2719 if (!block_group->ro)
2720 block_group->zone_unusable += to_unusable;
2721 spin_unlock(&ctl->tree_lock);
2723 spin_lock(&block_group->lock);
2724 block_group->alloc_offset -= size;
2725 spin_unlock(&block_group->lock);
2728 reclaimable_unusable = block_group->zone_unusable -
2729 (block_group->length - block_group->zone_capacity);
2730 /* All the region is now unusable. Mark it as unused and reclaim */
2731 if (block_group->zone_unusable == block_group->length &&
2732 block_group->alloc_offset) {
2733 btrfs_mark_bg_unused(block_group);
2734 } else if (bg_reclaim_threshold &&
2735 reclaimable_unusable >=
2736 mult_perc(block_group->zone_capacity, bg_reclaim_threshold)) {
2737 btrfs_mark_bg_to_reclaim(block_group);
2743 int btrfs_add_free_space(struct btrfs_block_group *block_group,
2744 u64 bytenr, u64 size)
2746 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2748 if (btrfs_is_zoned(block_group->fs_info))
2749 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2752 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC))
2753 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2755 return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2758 int btrfs_add_free_space_unused(struct btrfs_block_group *block_group,
2759 u64 bytenr, u64 size)
2761 if (btrfs_is_zoned(block_group->fs_info))
2762 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2765 return btrfs_add_free_space(block_group, bytenr, size);
2769 * This is a subtle distinction because when adding free space back in general,
2770 * we want it to be added as untrimmed for async. But in the case where we add
2771 * it on loading of a block group, we want to consider it trimmed.
2773 int btrfs_add_free_space_async_trimmed(struct btrfs_block_group *block_group,
2774 u64 bytenr, u64 size)
2776 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
2778 if (btrfs_is_zoned(block_group->fs_info))
2779 return __btrfs_add_free_space_zoned(block_group, bytenr, size,
2782 if (btrfs_test_opt(block_group->fs_info, DISCARD_SYNC) ||
2783 btrfs_test_opt(block_group->fs_info, DISCARD_ASYNC))
2784 trim_state = BTRFS_TRIM_STATE_TRIMMED;
2786 return __btrfs_add_free_space(block_group, bytenr, size, trim_state);
2789 int btrfs_remove_free_space(struct btrfs_block_group *block_group,
2790 u64 offset, u64 bytes)
2792 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2793 struct btrfs_free_space *info;
2795 bool re_search = false;
2797 if (btrfs_is_zoned(block_group->fs_info)) {
2799 * This can happen with conventional zones when replaying log.
2800 * Since the allocation info of tree-log nodes are not recorded
2801 * to the extent-tree, calculate_alloc_pointer() failed to
2802 * advance the allocation pointer after last allocated tree log
2805 * This function is called from
2806 * btrfs_pin_extent_for_log_replay() when replaying the log.
2807 * Advance the pointer not to overwrite the tree-log nodes.
2809 if (block_group->start + block_group->alloc_offset <
2811 block_group->alloc_offset =
2812 offset + bytes - block_group->start;
2817 spin_lock(&ctl->tree_lock);
2824 info = tree_search_offset(ctl, offset, 0, 0);
2827 * oops didn't find an extent that matched the space we wanted
2828 * to remove, look for a bitmap instead
2830 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2834 * If we found a partial bit of our free space in a
2835 * bitmap but then couldn't find the other part this may
2836 * be a problem, so WARN about it.
2844 if (!info->bitmap) {
2845 unlink_free_space(ctl, info, true);
2846 if (offset == info->offset) {
2847 u64 to_free = min(bytes, info->bytes);
2849 info->bytes -= to_free;
2850 info->offset += to_free;
2852 ret = link_free_space(ctl, info);
2855 kmem_cache_free(btrfs_free_space_cachep, info);
2862 u64 old_end = info->bytes + info->offset;
2864 info->bytes = offset - info->offset;
2865 ret = link_free_space(ctl, info);
2870 /* Not enough bytes in this entry to satisfy us */
2871 if (old_end < offset + bytes) {
2872 bytes -= old_end - offset;
2875 } else if (old_end == offset + bytes) {
2879 spin_unlock(&ctl->tree_lock);
2881 ret = __btrfs_add_free_space(block_group,
2883 old_end - (offset + bytes),
2890 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2891 if (ret == -EAGAIN) {
2896 btrfs_discard_update_discardable(block_group);
2897 spin_unlock(&ctl->tree_lock);
2902 void btrfs_dump_free_space(struct btrfs_block_group *block_group,
2905 struct btrfs_fs_info *fs_info = block_group->fs_info;
2906 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2907 struct btrfs_free_space *info;
2912 * Zoned btrfs does not use free space tree and cluster. Just print
2913 * out the free space after the allocation offset.
2915 if (btrfs_is_zoned(fs_info)) {
2916 btrfs_info(fs_info, "free space %llu active %d",
2917 block_group->zone_capacity - block_group->alloc_offset,
2918 test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2919 &block_group->runtime_flags));
2923 spin_lock(&ctl->tree_lock);
2924 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2925 info = rb_entry(n, struct btrfs_free_space, offset_index);
2926 if (info->bytes >= bytes && !block_group->ro)
2928 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2929 info->offset, info->bytes,
2930 (info->bitmap) ? "yes" : "no");
2932 spin_unlock(&ctl->tree_lock);
2933 btrfs_info(fs_info, "block group has cluster?: %s",
2934 list_empty(&block_group->cluster_list) ? "no" : "yes");
2936 "%d blocks of free space at or bigger than bytes is", count);
2939 void btrfs_init_free_space_ctl(struct btrfs_block_group *block_group,
2940 struct btrfs_free_space_ctl *ctl)
2942 struct btrfs_fs_info *fs_info = block_group->fs_info;
2944 spin_lock_init(&ctl->tree_lock);
2945 ctl->unit = fs_info->sectorsize;
2946 ctl->start = block_group->start;
2947 ctl->block_group = block_group;
2948 ctl->op = &free_space_op;
2949 ctl->free_space_bytes = RB_ROOT_CACHED;
2950 INIT_LIST_HEAD(&ctl->trimming_ranges);
2951 mutex_init(&ctl->cache_writeout_mutex);
2954 * we only want to have 32k of ram per block group for keeping
2955 * track of free space, and if we pass 1/2 of that we want to
2956 * start converting things over to using bitmaps
2958 ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2962 * for a given cluster, put all of its extents back into the free
2963 * space cache. If the block group passed doesn't match the block group
2964 * pointed to by the cluster, someone else raced in and freed the
2965 * cluster already. In that case, we just return without changing anything
2967 static void __btrfs_return_cluster_to_free_space(
2968 struct btrfs_block_group *block_group,
2969 struct btrfs_free_cluster *cluster)
2971 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2972 struct btrfs_free_space *entry;
2973 struct rb_node *node;
2975 spin_lock(&cluster->lock);
2976 if (cluster->block_group != block_group) {
2977 spin_unlock(&cluster->lock);
2981 cluster->block_group = NULL;
2982 cluster->window_start = 0;
2983 list_del_init(&cluster->block_group_list);
2985 node = rb_first(&cluster->root);
2989 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2990 node = rb_next(&entry->offset_index);
2991 rb_erase(&entry->offset_index, &cluster->root);
2992 RB_CLEAR_NODE(&entry->offset_index);
2994 bitmap = (entry->bitmap != NULL);
2996 /* Merging treats extents as if they were new */
2997 if (!btrfs_free_space_trimmed(entry)) {
2998 ctl->discardable_extents[BTRFS_STAT_CURR]--;
2999 ctl->discardable_bytes[BTRFS_STAT_CURR] -=
3003 try_merge_free_space(ctl, entry, false);
3004 steal_from_bitmap(ctl, entry, false);
3006 /* As we insert directly, update these statistics */
3007 if (!btrfs_free_space_trimmed(entry)) {
3008 ctl->discardable_extents[BTRFS_STAT_CURR]++;
3009 ctl->discardable_bytes[BTRFS_STAT_CURR] +=
3013 tree_insert_offset(&ctl->free_space_offset,
3014 entry->offset, &entry->offset_index, bitmap);
3015 rb_add_cached(&entry->bytes_index, &ctl->free_space_bytes,
3018 cluster->root = RB_ROOT;
3019 spin_unlock(&cluster->lock);
3020 btrfs_put_block_group(block_group);
3023 void btrfs_remove_free_space_cache(struct btrfs_block_group *block_group)
3025 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3026 struct btrfs_free_cluster *cluster;
3027 struct list_head *head;
3029 spin_lock(&ctl->tree_lock);
3030 while ((head = block_group->cluster_list.next) !=
3031 &block_group->cluster_list) {
3032 cluster = list_entry(head, struct btrfs_free_cluster,
3035 WARN_ON(cluster->block_group != block_group);
3036 __btrfs_return_cluster_to_free_space(block_group, cluster);
3038 cond_resched_lock(&ctl->tree_lock);
3040 __btrfs_remove_free_space_cache(ctl);
3041 btrfs_discard_update_discardable(block_group);
3042 spin_unlock(&ctl->tree_lock);
3047 * Walk @block_group's free space rb_tree to determine if everything is trimmed.
3049 bool btrfs_is_free_space_trimmed(struct btrfs_block_group *block_group)
3051 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3052 struct btrfs_free_space *info;
3053 struct rb_node *node;
3056 spin_lock(&ctl->tree_lock);
3057 node = rb_first(&ctl->free_space_offset);
3060 info = rb_entry(node, struct btrfs_free_space, offset_index);
3062 if (!btrfs_free_space_trimmed(info)) {
3067 node = rb_next(node);
3070 spin_unlock(&ctl->tree_lock);
3074 u64 btrfs_find_space_for_alloc(struct btrfs_block_group *block_group,
3075 u64 offset, u64 bytes, u64 empty_size,
3076 u64 *max_extent_size)
3078 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3079 struct btrfs_discard_ctl *discard_ctl =
3080 &block_group->fs_info->discard_ctl;
3081 struct btrfs_free_space *entry = NULL;
3082 u64 bytes_search = bytes + empty_size;
3085 u64 align_gap_len = 0;
3086 enum btrfs_trim_state align_gap_trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3087 bool use_bytes_index = (offset == block_group->start);
3089 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3091 spin_lock(&ctl->tree_lock);
3092 entry = find_free_space(ctl, &offset, &bytes_search,
3093 block_group->full_stripe_len, max_extent_size,
3099 if (entry->bitmap) {
3100 bitmap_clear_bits(ctl, entry, offset, bytes, true);
3102 if (!btrfs_free_space_trimmed(entry))
3103 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3106 free_bitmap(ctl, entry);
3108 unlink_free_space(ctl, entry, true);
3109 align_gap_len = offset - entry->offset;
3110 align_gap = entry->offset;
3111 align_gap_trim_state = entry->trim_state;
3113 if (!btrfs_free_space_trimmed(entry))
3114 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3116 entry->offset = offset + bytes;
3117 WARN_ON(entry->bytes < bytes + align_gap_len);
3119 entry->bytes -= bytes + align_gap_len;
3121 kmem_cache_free(btrfs_free_space_cachep, entry);
3123 link_free_space(ctl, entry);
3126 btrfs_discard_update_discardable(block_group);
3127 spin_unlock(&ctl->tree_lock);
3130 __btrfs_add_free_space(block_group, align_gap, align_gap_len,
3131 align_gap_trim_state);
3136 * given a cluster, put all of its extents back into the free space
3137 * cache. If a block group is passed, this function will only free
3138 * a cluster that belongs to the passed block group.
3140 * Otherwise, it'll get a reference on the block group pointed to by the
3141 * cluster and remove the cluster from it.
3143 void btrfs_return_cluster_to_free_space(
3144 struct btrfs_block_group *block_group,
3145 struct btrfs_free_cluster *cluster)
3147 struct btrfs_free_space_ctl *ctl;
3149 /* first, get a safe pointer to the block group */
3150 spin_lock(&cluster->lock);
3152 block_group = cluster->block_group;
3154 spin_unlock(&cluster->lock);
3157 } else if (cluster->block_group != block_group) {
3158 /* someone else has already freed it don't redo their work */
3159 spin_unlock(&cluster->lock);
3162 btrfs_get_block_group(block_group);
3163 spin_unlock(&cluster->lock);
3165 ctl = block_group->free_space_ctl;
3167 /* now return any extents the cluster had on it */
3168 spin_lock(&ctl->tree_lock);
3169 __btrfs_return_cluster_to_free_space(block_group, cluster);
3170 spin_unlock(&ctl->tree_lock);
3172 btrfs_discard_queue_work(&block_group->fs_info->discard_ctl, block_group);
3174 /* finally drop our ref */
3175 btrfs_put_block_group(block_group);
3178 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group *block_group,
3179 struct btrfs_free_cluster *cluster,
3180 struct btrfs_free_space *entry,
3181 u64 bytes, u64 min_start,
3182 u64 *max_extent_size)
3184 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3186 u64 search_start = cluster->window_start;
3187 u64 search_bytes = bytes;
3190 search_start = min_start;
3191 search_bytes = bytes;
3193 err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
3195 *max_extent_size = max(get_max_extent_size(entry),
3201 bitmap_clear_bits(ctl, entry, ret, bytes, false);
3207 * given a cluster, try to allocate 'bytes' from it, returns 0
3208 * if it couldn't find anything suitably large, or a logical disk offset
3209 * if things worked out
3211 u64 btrfs_alloc_from_cluster(struct btrfs_block_group *block_group,
3212 struct btrfs_free_cluster *cluster, u64 bytes,
3213 u64 min_start, u64 *max_extent_size)
3215 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3216 struct btrfs_discard_ctl *discard_ctl =
3217 &block_group->fs_info->discard_ctl;
3218 struct btrfs_free_space *entry = NULL;
3219 struct rb_node *node;
3222 ASSERT(!btrfs_is_zoned(block_group->fs_info));
3224 spin_lock(&cluster->lock);
3225 if (bytes > cluster->max_size)
3228 if (cluster->block_group != block_group)
3231 node = rb_first(&cluster->root);
3235 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3237 if (entry->bytes < bytes)
3238 *max_extent_size = max(get_max_extent_size(entry),
3241 if (entry->bytes < bytes ||
3242 (!entry->bitmap && entry->offset < min_start)) {
3243 node = rb_next(&entry->offset_index);
3246 entry = rb_entry(node, struct btrfs_free_space,
3251 if (entry->bitmap) {
3252 ret = btrfs_alloc_from_bitmap(block_group,
3253 cluster, entry, bytes,
3254 cluster->window_start,
3257 node = rb_next(&entry->offset_index);
3260 entry = rb_entry(node, struct btrfs_free_space,
3264 cluster->window_start += bytes;
3266 ret = entry->offset;
3268 entry->offset += bytes;
3269 entry->bytes -= bytes;
3275 spin_unlock(&cluster->lock);
3280 spin_lock(&ctl->tree_lock);
3282 if (!btrfs_free_space_trimmed(entry))
3283 atomic64_add(bytes, &discard_ctl->discard_bytes_saved);
3285 ctl->free_space -= bytes;
3286 if (!entry->bitmap && !btrfs_free_space_trimmed(entry))
3287 ctl->discardable_bytes[BTRFS_STAT_CURR] -= bytes;
3289 spin_lock(&cluster->lock);
3290 if (entry->bytes == 0) {
3291 rb_erase(&entry->offset_index, &cluster->root);
3292 ctl->free_extents--;
3293 if (entry->bitmap) {
3294 kmem_cache_free(btrfs_free_space_bitmap_cachep,
3296 ctl->total_bitmaps--;
3297 recalculate_thresholds(ctl);
3298 } else if (!btrfs_free_space_trimmed(entry)) {
3299 ctl->discardable_extents[BTRFS_STAT_CURR]--;
3301 kmem_cache_free(btrfs_free_space_cachep, entry);
3304 spin_unlock(&cluster->lock);
3305 spin_unlock(&ctl->tree_lock);
3310 static int btrfs_bitmap_cluster(struct btrfs_block_group *block_group,
3311 struct btrfs_free_space *entry,
3312 struct btrfs_free_cluster *cluster,
3313 u64 offset, u64 bytes,
3314 u64 cont1_bytes, u64 min_bytes)
3316 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3317 unsigned long next_zero;
3319 unsigned long want_bits;
3320 unsigned long min_bits;
3321 unsigned long found_bits;
3322 unsigned long max_bits = 0;
3323 unsigned long start = 0;
3324 unsigned long total_found = 0;
3327 i = offset_to_bit(entry->offset, ctl->unit,
3328 max_t(u64, offset, entry->offset));
3329 want_bits = bytes_to_bits(bytes, ctl->unit);
3330 min_bits = bytes_to_bits(min_bytes, ctl->unit);
3333 * Don't bother looking for a cluster in this bitmap if it's heavily
3336 if (entry->max_extent_size &&
3337 entry->max_extent_size < cont1_bytes)
3341 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
3342 next_zero = find_next_zero_bit(entry->bitmap,
3343 BITS_PER_BITMAP, i);
3344 if (next_zero - i >= min_bits) {
3345 found_bits = next_zero - i;
3346 if (found_bits > max_bits)
3347 max_bits = found_bits;
3350 if (next_zero - i > max_bits)
3351 max_bits = next_zero - i;
3356 entry->max_extent_size = (u64)max_bits * ctl->unit;
3362 cluster->max_size = 0;
3365 total_found += found_bits;
3367 if (cluster->max_size < found_bits * ctl->unit)
3368 cluster->max_size = found_bits * ctl->unit;
3370 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
3375 cluster->window_start = start * ctl->unit + entry->offset;
3376 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3377 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3380 * We need to know if we're currently on the normal space index when we
3381 * manipulate the bitmap so that we know we need to remove and re-insert
3382 * it into the space_index tree. Clear the bytes_index node here so the
3383 * bitmap manipulation helpers know not to mess with the space_index
3384 * until this bitmap entry is added back into the normal cache.
3386 RB_CLEAR_NODE(&entry->bytes_index);
3388 ret = tree_insert_offset(&cluster->root, entry->offset,
3389 &entry->offset_index, 1);
3390 ASSERT(!ret); /* -EEXIST; Logic error */
3392 trace_btrfs_setup_cluster(block_group, cluster,
3393 total_found * ctl->unit, 1);
3398 * This searches the block group for just extents to fill the cluster with.
3399 * Try to find a cluster with at least bytes total bytes, at least one
3400 * extent of cont1_bytes, and other clusters of at least min_bytes.
3403 setup_cluster_no_bitmap(struct btrfs_block_group *block_group,
3404 struct btrfs_free_cluster *cluster,
3405 struct list_head *bitmaps, u64 offset, u64 bytes,
3406 u64 cont1_bytes, u64 min_bytes)
3408 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3409 struct btrfs_free_space *first = NULL;
3410 struct btrfs_free_space *entry = NULL;
3411 struct btrfs_free_space *last;
3412 struct rb_node *node;
3417 entry = tree_search_offset(ctl, offset, 0, 1);
3422 * We don't want bitmaps, so just move along until we find a normal
3425 while (entry->bitmap || entry->bytes < min_bytes) {
3426 if (entry->bitmap && list_empty(&entry->list))
3427 list_add_tail(&entry->list, bitmaps);
3428 node = rb_next(&entry->offset_index);
3431 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3434 window_free = entry->bytes;
3435 max_extent = entry->bytes;
3439 for (node = rb_next(&entry->offset_index); node;
3440 node = rb_next(&entry->offset_index)) {
3441 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3443 if (entry->bitmap) {
3444 if (list_empty(&entry->list))
3445 list_add_tail(&entry->list, bitmaps);
3449 if (entry->bytes < min_bytes)
3453 window_free += entry->bytes;
3454 if (entry->bytes > max_extent)
3455 max_extent = entry->bytes;
3458 if (window_free < bytes || max_extent < cont1_bytes)
3461 cluster->window_start = first->offset;
3463 node = &first->offset_index;
3466 * now we've found our entries, pull them out of the free space
3467 * cache and put them into the cluster rbtree
3472 entry = rb_entry(node, struct btrfs_free_space, offset_index);
3473 node = rb_next(&entry->offset_index);
3474 if (entry->bitmap || entry->bytes < min_bytes)
3477 rb_erase(&entry->offset_index, &ctl->free_space_offset);
3478 rb_erase_cached(&entry->bytes_index, &ctl->free_space_bytes);
3479 ret = tree_insert_offset(&cluster->root, entry->offset,
3480 &entry->offset_index, 0);
3481 total_size += entry->bytes;
3482 ASSERT(!ret); /* -EEXIST; Logic error */
3483 } while (node && entry != last);
3485 cluster->max_size = max_extent;
3486 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
3491 * This specifically looks for bitmaps that may work in the cluster, we assume
3492 * that we have already failed to find extents that will work.
3495 setup_cluster_bitmap(struct btrfs_block_group *block_group,
3496 struct btrfs_free_cluster *cluster,
3497 struct list_head *bitmaps, u64 offset, u64 bytes,
3498 u64 cont1_bytes, u64 min_bytes)
3500 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3501 struct btrfs_free_space *entry = NULL;
3503 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
3505 if (ctl->total_bitmaps == 0)
3509 * The bitmap that covers offset won't be in the list unless offset
3510 * is just its start offset.
3512 if (!list_empty(bitmaps))
3513 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3515 if (!entry || entry->offset != bitmap_offset) {
3516 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3517 if (entry && list_empty(&entry->list))
3518 list_add(&entry->list, bitmaps);
3521 list_for_each_entry(entry, bitmaps, list) {
3522 if (entry->bytes < bytes)
3524 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3525 bytes, cont1_bytes, min_bytes);
3531 * The bitmaps list has all the bitmaps that record free space
3532 * starting after offset, so no more search is required.
3538 * here we try to find a cluster of blocks in a block group. The goal
3539 * is to find at least bytes+empty_size.
3540 * We might not find them all in one contiguous area.
3542 * returns zero and sets up cluster if things worked out, otherwise
3543 * it returns -enospc
3545 int btrfs_find_space_cluster(struct btrfs_block_group *block_group,
3546 struct btrfs_free_cluster *cluster,
3547 u64 offset, u64 bytes, u64 empty_size)
3549 struct btrfs_fs_info *fs_info = block_group->fs_info;
3550 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3551 struct btrfs_free_space *entry, *tmp;
3558 * Choose the minimum extent size we'll require for this
3559 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3560 * For metadata, allow allocates with smaller extents. For
3561 * data, keep it dense.
3563 if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3564 cont1_bytes = bytes + empty_size;
3565 min_bytes = cont1_bytes;
3566 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3567 cont1_bytes = bytes;
3568 min_bytes = fs_info->sectorsize;
3570 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3571 min_bytes = fs_info->sectorsize;
3574 spin_lock(&ctl->tree_lock);
3577 * If we know we don't have enough space to make a cluster don't even
3578 * bother doing all the work to try and find one.
3580 if (ctl->free_space < bytes) {
3581 spin_unlock(&ctl->tree_lock);
3585 spin_lock(&cluster->lock);
3587 /* someone already found a cluster, hooray */
3588 if (cluster->block_group) {
3593 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3596 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3598 cont1_bytes, min_bytes);
3600 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3601 offset, bytes + empty_size,
3602 cont1_bytes, min_bytes);
3604 /* Clear our temporary list */
3605 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3606 list_del_init(&entry->list);
3609 btrfs_get_block_group(block_group);
3610 list_add_tail(&cluster->block_group_list,
3611 &block_group->cluster_list);
3612 cluster->block_group = block_group;
3614 trace_btrfs_failed_cluster_setup(block_group);
3617 spin_unlock(&cluster->lock);
3618 spin_unlock(&ctl->tree_lock);
3624 * simple code to zero out a cluster
3626 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3628 spin_lock_init(&cluster->lock);
3629 spin_lock_init(&cluster->refill_lock);
3630 cluster->root = RB_ROOT;
3631 cluster->max_size = 0;
3632 cluster->fragmented = false;
3633 INIT_LIST_HEAD(&cluster->block_group_list);
3634 cluster->block_group = NULL;
3637 static int do_trimming(struct btrfs_block_group *block_group,
3638 u64 *total_trimmed, u64 start, u64 bytes,
3639 u64 reserved_start, u64 reserved_bytes,
3640 enum btrfs_trim_state reserved_trim_state,
3641 struct btrfs_trim_range *trim_entry)
3643 struct btrfs_space_info *space_info = block_group->space_info;
3644 struct btrfs_fs_info *fs_info = block_group->fs_info;
3645 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3648 const u64 end = start + bytes;
3649 const u64 reserved_end = reserved_start + reserved_bytes;
3650 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3653 spin_lock(&space_info->lock);
3654 spin_lock(&block_group->lock);
3655 if (!block_group->ro) {
3656 block_group->reserved += reserved_bytes;
3657 space_info->bytes_reserved += reserved_bytes;
3660 spin_unlock(&block_group->lock);
3661 spin_unlock(&space_info->lock);
3663 ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed);
3665 *total_trimmed += trimmed;
3666 trim_state = BTRFS_TRIM_STATE_TRIMMED;
3669 mutex_lock(&ctl->cache_writeout_mutex);
3670 if (reserved_start < start)
3671 __btrfs_add_free_space(block_group, reserved_start,
3672 start - reserved_start,
3673 reserved_trim_state);
3674 if (start + bytes < reserved_start + reserved_bytes)
3675 __btrfs_add_free_space(block_group, end, reserved_end - end,
3676 reserved_trim_state);
3677 __btrfs_add_free_space(block_group, start, bytes, trim_state);
3678 list_del(&trim_entry->list);
3679 mutex_unlock(&ctl->cache_writeout_mutex);
3682 spin_lock(&space_info->lock);
3683 spin_lock(&block_group->lock);
3684 if (block_group->ro)
3685 space_info->bytes_readonly += reserved_bytes;
3686 block_group->reserved -= reserved_bytes;
3687 space_info->bytes_reserved -= reserved_bytes;
3688 spin_unlock(&block_group->lock);
3689 spin_unlock(&space_info->lock);
3696 * If @async is set, then we will trim 1 region and return.
3698 static int trim_no_bitmap(struct btrfs_block_group *block_group,
3699 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3702 struct btrfs_discard_ctl *discard_ctl =
3703 &block_group->fs_info->discard_ctl;
3704 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3705 struct btrfs_free_space *entry;
3706 struct rb_node *node;
3710 enum btrfs_trim_state extent_trim_state;
3712 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3714 while (start < end) {
3715 struct btrfs_trim_range trim_entry;
3717 mutex_lock(&ctl->cache_writeout_mutex);
3718 spin_lock(&ctl->tree_lock);
3720 if (ctl->free_space < minlen)
3723 entry = tree_search_offset(ctl, start, 0, 1);
3727 /* Skip bitmaps and if async, already trimmed entries */
3728 while (entry->bitmap ||
3729 (async && btrfs_free_space_trimmed(entry))) {
3730 node = rb_next(&entry->offset_index);
3733 entry = rb_entry(node, struct btrfs_free_space,
3737 if (entry->offset >= end)
3740 extent_start = entry->offset;
3741 extent_bytes = entry->bytes;
3742 extent_trim_state = entry->trim_state;
3744 start = entry->offset;
3745 bytes = entry->bytes;
3746 if (bytes < minlen) {
3747 spin_unlock(&ctl->tree_lock);
3748 mutex_unlock(&ctl->cache_writeout_mutex);
3751 unlink_free_space(ctl, entry, true);
3753 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3754 * If X < BTRFS_ASYNC_DISCARD_MIN_FILTER, we won't trim
3755 * X when we come back around. So trim it now.
3757 if (max_discard_size &&
3758 bytes >= (max_discard_size +
3759 BTRFS_ASYNC_DISCARD_MIN_FILTER)) {
3760 bytes = max_discard_size;
3761 extent_bytes = max_discard_size;
3762 entry->offset += max_discard_size;
3763 entry->bytes -= max_discard_size;
3764 link_free_space(ctl, entry);
3766 kmem_cache_free(btrfs_free_space_cachep, entry);
3769 start = max(start, extent_start);
3770 bytes = min(extent_start + extent_bytes, end) - start;
3771 if (bytes < minlen) {
3772 spin_unlock(&ctl->tree_lock);
3773 mutex_unlock(&ctl->cache_writeout_mutex);
3777 unlink_free_space(ctl, entry, true);
3778 kmem_cache_free(btrfs_free_space_cachep, entry);
3781 spin_unlock(&ctl->tree_lock);
3782 trim_entry.start = extent_start;
3783 trim_entry.bytes = extent_bytes;
3784 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3785 mutex_unlock(&ctl->cache_writeout_mutex);
3787 ret = do_trimming(block_group, total_trimmed, start, bytes,
3788 extent_start, extent_bytes, extent_trim_state,
3791 block_group->discard_cursor = start + bytes;
3796 block_group->discard_cursor = start;
3797 if (async && *total_trimmed)
3800 if (fatal_signal_pending(current)) {
3811 block_group->discard_cursor = btrfs_block_group_end(block_group);
3812 spin_unlock(&ctl->tree_lock);
3813 mutex_unlock(&ctl->cache_writeout_mutex);
3819 * If we break out of trimming a bitmap prematurely, we should reset the
3820 * trimming bit. In a rather contrieved case, it's possible to race here so
3821 * reset the state to BTRFS_TRIM_STATE_UNTRIMMED.
3823 * start = start of bitmap
3824 * end = near end of bitmap
3826 * Thread 1: Thread 2:
3827 * trim_bitmaps(start)
3829 * end_trimming_bitmap()
3830 * reset_trimming_bitmap()
3832 static void reset_trimming_bitmap(struct btrfs_free_space_ctl *ctl, u64 offset)
3834 struct btrfs_free_space *entry;
3836 spin_lock(&ctl->tree_lock);
3837 entry = tree_search_offset(ctl, offset, 1, 0);
3839 if (btrfs_free_space_trimmed(entry)) {
3840 ctl->discardable_extents[BTRFS_STAT_CURR] +=
3841 entry->bitmap_extents;
3842 ctl->discardable_bytes[BTRFS_STAT_CURR] += entry->bytes;
3844 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3847 spin_unlock(&ctl->tree_lock);
3850 static void end_trimming_bitmap(struct btrfs_free_space_ctl *ctl,
3851 struct btrfs_free_space *entry)
3853 if (btrfs_free_space_trimming_bitmap(entry)) {
3854 entry->trim_state = BTRFS_TRIM_STATE_TRIMMED;
3855 ctl->discardable_extents[BTRFS_STAT_CURR] -=
3856 entry->bitmap_extents;
3857 ctl->discardable_bytes[BTRFS_STAT_CURR] -= entry->bytes;
3862 * If @async is set, then we will trim 1 region and return.
3864 static int trim_bitmaps(struct btrfs_block_group *block_group,
3865 u64 *total_trimmed, u64 start, u64 end, u64 minlen,
3866 u64 maxlen, bool async)
3868 struct btrfs_discard_ctl *discard_ctl =
3869 &block_group->fs_info->discard_ctl;
3870 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3871 struct btrfs_free_space *entry;
3875 u64 offset = offset_to_bitmap(ctl, start);
3876 const u64 max_discard_size = READ_ONCE(discard_ctl->max_discard_size);
3878 while (offset < end) {
3879 bool next_bitmap = false;
3880 struct btrfs_trim_range trim_entry;
3882 mutex_lock(&ctl->cache_writeout_mutex);
3883 spin_lock(&ctl->tree_lock);
3885 if (ctl->free_space < minlen) {
3886 block_group->discard_cursor =
3887 btrfs_block_group_end(block_group);
3888 spin_unlock(&ctl->tree_lock);
3889 mutex_unlock(&ctl->cache_writeout_mutex);
3893 entry = tree_search_offset(ctl, offset, 1, 0);
3895 * Bitmaps are marked trimmed lossily now to prevent constant
3896 * discarding of the same bitmap (the reason why we are bound
3897 * by the filters). So, retrim the block group bitmaps when we
3898 * are preparing to punt to the unused_bgs list. This uses
3899 * @minlen to determine if we are in BTRFS_DISCARD_INDEX_UNUSED
3900 * which is the only discard index which sets minlen to 0.
3902 if (!entry || (async && minlen && start == offset &&
3903 btrfs_free_space_trimmed(entry))) {
3904 spin_unlock(&ctl->tree_lock);
3905 mutex_unlock(&ctl->cache_writeout_mutex);
3911 * Async discard bitmap trimming begins at by setting the start
3912 * to be key.objectid and the offset_to_bitmap() aligns to the
3913 * start of the bitmap. This lets us know we are fully
3914 * scanning the bitmap rather than only some portion of it.
3916 if (start == offset)
3917 entry->trim_state = BTRFS_TRIM_STATE_TRIMMING;
3920 ret2 = search_bitmap(ctl, entry, &start, &bytes, false);
3921 if (ret2 || start >= end) {
3923 * We lossily consider a bitmap trimmed if we only skip
3924 * over regions <= BTRFS_ASYNC_DISCARD_MIN_FILTER.
3926 if (ret2 && minlen <= BTRFS_ASYNC_DISCARD_MIN_FILTER)
3927 end_trimming_bitmap(ctl, entry);
3929 entry->trim_state = BTRFS_TRIM_STATE_UNTRIMMED;
3930 spin_unlock(&ctl->tree_lock);
3931 mutex_unlock(&ctl->cache_writeout_mutex);
3937 * We already trimmed a region, but are using the locking above
3938 * to reset the trim_state.
3940 if (async && *total_trimmed) {
3941 spin_unlock(&ctl->tree_lock);
3942 mutex_unlock(&ctl->cache_writeout_mutex);
3946 bytes = min(bytes, end - start);
3947 if (bytes < minlen || (async && maxlen && bytes > maxlen)) {
3948 spin_unlock(&ctl->tree_lock);
3949 mutex_unlock(&ctl->cache_writeout_mutex);
3954 * Let bytes = BTRFS_MAX_DISCARD_SIZE + X.
3955 * If X < @minlen, we won't trim X when we come back around.
3956 * So trim it now. We differ here from trimming extents as we
3957 * don't keep individual state per bit.
3961 bytes > (max_discard_size + minlen))
3962 bytes = max_discard_size;
3964 bitmap_clear_bits(ctl, entry, start, bytes, true);
3965 if (entry->bytes == 0)
3966 free_bitmap(ctl, entry);
3968 spin_unlock(&ctl->tree_lock);
3969 trim_entry.start = start;
3970 trim_entry.bytes = bytes;
3971 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3972 mutex_unlock(&ctl->cache_writeout_mutex);
3974 ret = do_trimming(block_group, total_trimmed, start, bytes,
3975 start, bytes, 0, &trim_entry);
3977 reset_trimming_bitmap(ctl, offset);
3978 block_group->discard_cursor =
3979 btrfs_block_group_end(block_group);
3984 offset += BITS_PER_BITMAP * ctl->unit;
3989 block_group->discard_cursor = start;
3991 if (fatal_signal_pending(current)) {
3992 if (start != offset)
3993 reset_trimming_bitmap(ctl, offset);
4002 block_group->discard_cursor = end;
4008 int btrfs_trim_block_group(struct btrfs_block_group *block_group,
4009 u64 *trimmed, u64 start, u64 end, u64 minlen)
4011 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
4015 ASSERT(!btrfs_is_zoned(block_group->fs_info));
4019 spin_lock(&block_group->lock);
4020 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4021 spin_unlock(&block_group->lock);
4024 btrfs_freeze_block_group(block_group);
4025 spin_unlock(&block_group->lock);
4027 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, false);
4031 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, 0, false);
4032 div64_u64_rem(end, BITS_PER_BITMAP * ctl->unit, &rem);
4033 /* If we ended in the middle of a bitmap, reset the trimming flag */
4035 reset_trimming_bitmap(ctl, offset_to_bitmap(ctl, end));
4037 btrfs_unfreeze_block_group(block_group);
4041 int btrfs_trim_block_group_extents(struct btrfs_block_group *block_group,
4042 u64 *trimmed, u64 start, u64 end, u64 minlen,
4049 spin_lock(&block_group->lock);
4050 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4051 spin_unlock(&block_group->lock);
4054 btrfs_freeze_block_group(block_group);
4055 spin_unlock(&block_group->lock);
4057 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen, async);
4058 btrfs_unfreeze_block_group(block_group);
4063 int btrfs_trim_block_group_bitmaps(struct btrfs_block_group *block_group,
4064 u64 *trimmed, u64 start, u64 end, u64 minlen,
4065 u64 maxlen, bool async)
4071 spin_lock(&block_group->lock);
4072 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)) {
4073 spin_unlock(&block_group->lock);
4076 btrfs_freeze_block_group(block_group);
4077 spin_unlock(&block_group->lock);
4079 ret = trim_bitmaps(block_group, trimmed, start, end, minlen, maxlen,
4082 btrfs_unfreeze_block_group(block_group);
4087 bool btrfs_free_space_cache_v1_active(struct btrfs_fs_info *fs_info)
4089 return btrfs_super_cache_generation(fs_info->super_copy);
4092 static int cleanup_free_space_cache_v1(struct btrfs_fs_info *fs_info,
4093 struct btrfs_trans_handle *trans)
4095 struct btrfs_block_group *block_group;
4096 struct rb_node *node;
4099 btrfs_info(fs_info, "cleaning free space cache v1");
4101 node = rb_first_cached(&fs_info->block_group_cache_tree);
4103 block_group = rb_entry(node, struct btrfs_block_group, cache_node);
4104 ret = btrfs_remove_free_space_inode(trans, NULL, block_group);
4107 node = rb_next(node);
4113 int btrfs_set_free_space_cache_v1_active(struct btrfs_fs_info *fs_info, bool active)
4115 struct btrfs_trans_handle *trans;
4119 * update_super_roots will appropriately set or unset
4120 * super_copy->cache_generation based on SPACE_CACHE and
4121 * BTRFS_FS_CLEANUP_SPACE_CACHE_V1. For this reason, we need a
4122 * transaction commit whether we are enabling space cache v1 and don't
4123 * have any other work to do, or are disabling it and removing free
4126 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4128 return PTR_ERR(trans);
4131 set_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4132 ret = cleanup_free_space_cache_v1(fs_info, trans);
4134 btrfs_abort_transaction(trans, ret);
4135 btrfs_end_transaction(trans);
4140 ret = btrfs_commit_transaction(trans);
4142 clear_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags);
4147 int __init btrfs_free_space_init(void)
4149 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
4150 sizeof(struct btrfs_free_space), 0,
4151 SLAB_MEM_SPREAD, NULL);
4152 if (!btrfs_free_space_cachep)
4155 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
4156 PAGE_SIZE, PAGE_SIZE,
4157 SLAB_MEM_SPREAD, NULL);
4158 if (!btrfs_free_space_bitmap_cachep) {
4159 kmem_cache_destroy(btrfs_free_space_cachep);
4166 void __cold btrfs_free_space_exit(void)
4168 kmem_cache_destroy(btrfs_free_space_cachep);
4169 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
4172 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4174 * Use this if you need to make a bitmap or extent entry specifically, it
4175 * doesn't do any of the merging that add_free_space does, this acts a lot like
4176 * how the free space cache loading stuff works, so you can get really weird
4179 int test_add_free_space_entry(struct btrfs_block_group *cache,
4180 u64 offset, u64 bytes, bool bitmap)
4182 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4183 struct btrfs_free_space *info = NULL, *bitmap_info;
4185 enum btrfs_trim_state trim_state = BTRFS_TRIM_STATE_TRIMMED;
4191 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
4197 spin_lock(&ctl->tree_lock);
4198 info->offset = offset;
4199 info->bytes = bytes;
4200 info->max_extent_size = 0;
4201 ret = link_free_space(ctl, info);
4202 spin_unlock(&ctl->tree_lock);
4204 kmem_cache_free(btrfs_free_space_cachep, info);
4209 map = kmem_cache_zalloc(btrfs_free_space_bitmap_cachep, GFP_NOFS);
4211 kmem_cache_free(btrfs_free_space_cachep, info);
4216 spin_lock(&ctl->tree_lock);
4217 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4222 add_new_bitmap(ctl, info, offset);
4227 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes,
4230 bytes -= bytes_added;
4231 offset += bytes_added;
4232 spin_unlock(&ctl->tree_lock);
4238 kmem_cache_free(btrfs_free_space_cachep, info);
4240 kmem_cache_free(btrfs_free_space_bitmap_cachep, map);
4245 * Checks to see if the given range is in the free space cache. This is really
4246 * just used to check the absence of space, so if there is free space in the
4247 * range at all we will return 1.
4249 int test_check_exists(struct btrfs_block_group *cache,
4250 u64 offset, u64 bytes)
4252 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
4253 struct btrfs_free_space *info;
4256 spin_lock(&ctl->tree_lock);
4257 info = tree_search_offset(ctl, offset, 0, 0);
4259 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
4267 u64 bit_off, bit_bytes;
4269 struct btrfs_free_space *tmp;
4272 bit_bytes = ctl->unit;
4273 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false);
4275 if (bit_off == offset) {
4278 } else if (bit_off > offset &&
4279 offset + bytes > bit_off) {
4285 n = rb_prev(&info->offset_index);
4287 tmp = rb_entry(n, struct btrfs_free_space,
4289 if (tmp->offset + tmp->bytes < offset)
4291 if (offset + bytes < tmp->offset) {
4292 n = rb_prev(&tmp->offset_index);
4299 n = rb_next(&info->offset_index);
4301 tmp = rb_entry(n, struct btrfs_free_space,
4303 if (offset + bytes < tmp->offset)
4305 if (tmp->offset + tmp->bytes < offset) {
4306 n = rb_next(&tmp->offset_index);
4317 if (info->offset == offset) {
4322 if (offset > info->offset && offset < info->offset + info->bytes)
4325 spin_unlock(&ctl->tree_lock);
4328 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */