2 * Copyright (C) 2008 Red Hat. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
24 #include "free-space-cache.h"
25 #include "transaction.h"
27 #include "extent_io.h"
29 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
30 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
32 static void recalculate_thresholds(struct btrfs_block_group_cache
34 static int link_free_space(struct btrfs_block_group_cache *block_group,
35 struct btrfs_free_space *info);
37 struct inode *lookup_free_space_inode(struct btrfs_root *root,
38 struct btrfs_block_group_cache
39 *block_group, struct btrfs_path *path)
42 struct btrfs_key location;
43 struct btrfs_disk_key disk_key;
44 struct btrfs_free_space_header *header;
45 struct extent_buffer *leaf;
46 struct inode *inode = NULL;
49 spin_lock(&block_group->lock);
50 if (block_group->inode)
51 inode = igrab(block_group->inode);
52 spin_unlock(&block_group->lock);
56 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
57 key.offset = block_group->key.objectid;
60 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
64 btrfs_release_path(root, path);
65 return ERR_PTR(-ENOENT);
68 leaf = path->nodes[0];
69 header = btrfs_item_ptr(leaf, path->slots[0],
70 struct btrfs_free_space_header);
71 btrfs_free_space_key(leaf, header, &disk_key);
72 btrfs_disk_key_to_cpu(&location, &disk_key);
73 btrfs_release_path(root, path);
75 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
77 return ERR_PTR(-ENOENT);
80 if (is_bad_inode(inode)) {
82 return ERR_PTR(-ENOENT);
85 inode->i_mapping->flags &= ~__GFP_FS;
87 spin_lock(&block_group->lock);
88 if (!root->fs_info->closing) {
89 block_group->inode = igrab(inode);
90 block_group->iref = 1;
92 spin_unlock(&block_group->lock);
97 int create_free_space_inode(struct btrfs_root *root,
98 struct btrfs_trans_handle *trans,
99 struct btrfs_block_group_cache *block_group,
100 struct btrfs_path *path)
102 struct btrfs_key key;
103 struct btrfs_disk_key disk_key;
104 struct btrfs_free_space_header *header;
105 struct btrfs_inode_item *inode_item;
106 struct extent_buffer *leaf;
110 ret = btrfs_find_free_objectid(trans, root, 0, &objectid);
114 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
118 leaf = path->nodes[0];
119 inode_item = btrfs_item_ptr(leaf, path->slots[0],
120 struct btrfs_inode_item);
121 btrfs_item_key(leaf, &disk_key, path->slots[0]);
122 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
123 sizeof(*inode_item));
124 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
125 btrfs_set_inode_size(leaf, inode_item, 0);
126 btrfs_set_inode_nbytes(leaf, inode_item, 0);
127 btrfs_set_inode_uid(leaf, inode_item, 0);
128 btrfs_set_inode_gid(leaf, inode_item, 0);
129 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
130 btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS |
131 BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM);
132 btrfs_set_inode_nlink(leaf, inode_item, 1);
133 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
134 btrfs_set_inode_block_group(leaf, inode_item,
135 block_group->key.objectid);
136 btrfs_mark_buffer_dirty(leaf);
137 btrfs_release_path(root, path);
139 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
140 key.offset = block_group->key.objectid;
143 ret = btrfs_insert_empty_item(trans, root, path, &key,
144 sizeof(struct btrfs_free_space_header));
146 btrfs_release_path(root, path);
149 leaf = path->nodes[0];
150 header = btrfs_item_ptr(leaf, path->slots[0],
151 struct btrfs_free_space_header);
152 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
153 btrfs_set_free_space_key(leaf, header, &disk_key);
154 btrfs_mark_buffer_dirty(leaf);
155 btrfs_release_path(root, path);
160 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
161 struct btrfs_trans_handle *trans,
162 struct btrfs_path *path,
168 trans->block_rsv = root->orphan_block_rsv;
169 ret = btrfs_block_rsv_check(trans, root,
170 root->orphan_block_rsv,
175 oldsize = i_size_read(inode);
176 btrfs_i_size_write(inode, 0);
177 truncate_pagecache(inode, oldsize, 0);
180 * We don't need an orphan item because truncating the free space cache
181 * will never be split across transactions.
183 ret = btrfs_truncate_inode_items(trans, root, inode,
184 0, BTRFS_EXTENT_DATA_KEY);
190 return btrfs_update_inode(trans, root, inode);
193 static int readahead_cache(struct inode *inode)
195 struct file_ra_state *ra;
196 unsigned long last_index;
198 ra = kzalloc(sizeof(*ra), GFP_NOFS);
202 file_ra_state_init(ra, inode->i_mapping);
203 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
205 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
212 int load_free_space_cache(struct btrfs_fs_info *fs_info,
213 struct btrfs_block_group_cache *block_group)
215 struct btrfs_root *root = fs_info->tree_root;
217 struct btrfs_free_space_header *header;
218 struct extent_buffer *leaf;
220 struct btrfs_path *path;
221 u32 *checksums = NULL, *crc;
222 char *disk_crcs = NULL;
223 struct btrfs_key key;
224 struct list_head bitmaps;
228 u64 used = btrfs_block_group_used(&block_group->item);
229 u32 cur_crc = ~(u32)0;
231 unsigned long first_page_offset;
236 * If we're unmounting then just return, since this does a search on the
237 * normal root and not the commit root and we could deadlock.
240 if (fs_info->closing)
244 * If this block group has been marked to be cleared for one reason or
245 * another then we can't trust the on disk cache, so just return.
247 spin_lock(&block_group->lock);
248 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
249 spin_unlock(&block_group->lock);
252 spin_unlock(&block_group->lock);
254 INIT_LIST_HEAD(&bitmaps);
256 path = btrfs_alloc_path();
260 inode = lookup_free_space_inode(root, block_group, path);
262 btrfs_free_path(path);
266 /* Nothing in the space cache, goodbye */
267 if (!i_size_read(inode)) {
268 btrfs_free_path(path);
272 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
273 key.offset = block_group->key.objectid;
276 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
278 btrfs_free_path(path);
282 leaf = path->nodes[0];
283 header = btrfs_item_ptr(leaf, path->slots[0],
284 struct btrfs_free_space_header);
285 num_entries = btrfs_free_space_entries(leaf, header);
286 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
287 generation = btrfs_free_space_generation(leaf, header);
288 btrfs_free_path(path);
290 if (BTRFS_I(inode)->generation != generation) {
291 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
292 " not match free space cache generation (%llu) for "
293 "block group %llu\n",
294 (unsigned long long)BTRFS_I(inode)->generation,
295 (unsigned long long)generation,
296 (unsigned long long)block_group->key.objectid);
303 /* Setup everything for doing checksumming */
304 num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
305 checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
308 first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
309 disk_crcs = kzalloc(first_page_offset, GFP_NOFS);
313 ret = readahead_cache(inode);
320 struct btrfs_free_space_entry *entry;
321 struct btrfs_free_space *e;
323 unsigned long offset = 0;
324 unsigned long start_offset = 0;
327 if (!num_entries && !num_bitmaps)
331 start_offset = first_page_offset;
332 offset = start_offset;
335 page = grab_cache_page(inode->i_mapping, index);
341 if (!PageUptodate(page)) {
342 btrfs_readpage(NULL, page);
344 if (!PageUptodate(page)) {
346 page_cache_release(page);
347 printk(KERN_ERR "btrfs: error reading free "
348 "space cache: %llu\n",
350 block_group->key.objectid);
359 memcpy(disk_crcs, addr, first_page_offset);
360 gen = addr + (sizeof(u32) * num_checksums);
361 if (*gen != BTRFS_I(inode)->generation) {
362 printk(KERN_ERR "btrfs: space cache generation"
363 " (%llu) does not match inode (%llu) "
364 "for block group %llu\n",
365 (unsigned long long)*gen,
367 BTRFS_I(inode)->generation,
369 block_group->key.objectid);
372 page_cache_release(page);
375 crc = (u32 *)disk_crcs;
377 entry = addr + start_offset;
379 /* First lets check our crc before we do anything fun */
381 cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc,
382 PAGE_CACHE_SIZE - start_offset);
383 btrfs_csum_final(cur_crc, (char *)&cur_crc);
384 if (cur_crc != *crc) {
385 printk(KERN_ERR "btrfs: crc mismatch for page %lu in "
386 "block group %llu\n", index,
387 (unsigned long long)block_group->key.objectid);
390 page_cache_release(page);
400 e = kmem_cache_zalloc(btrfs_free_space_cachep,
405 page_cache_release(page);
409 e->offset = le64_to_cpu(entry->offset);
410 e->bytes = le64_to_cpu(entry->bytes);
413 kmem_cache_free(btrfs_free_space_cachep, e);
415 page_cache_release(page);
419 if (entry->type == BTRFS_FREE_SPACE_EXTENT) {
420 spin_lock(&block_group->tree_lock);
421 ret = link_free_space(block_group, e);
422 spin_unlock(&block_group->tree_lock);
425 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
429 btrfs_free_space_cachep, e);
431 page_cache_release(page);
434 spin_lock(&block_group->tree_lock);
435 ret = link_free_space(block_group, e);
436 block_group->total_bitmaps++;
437 recalculate_thresholds(block_group);
438 spin_unlock(&block_group->tree_lock);
439 list_add_tail(&e->list, &bitmaps);
443 offset += sizeof(struct btrfs_free_space_entry);
444 if (offset + sizeof(struct btrfs_free_space_entry) >=
451 * We read an entry out of this page, we need to move on to the
460 * We add the bitmaps at the end of the entries in order that
461 * the bitmap entries are added to the cache.
463 e = list_entry(bitmaps.next, struct btrfs_free_space, list);
464 list_del_init(&e->list);
465 memcpy(e->bitmap, addr, PAGE_CACHE_SIZE);
470 page_cache_release(page);
474 spin_lock(&block_group->tree_lock);
475 if (block_group->free_space != (block_group->key.offset - used -
476 block_group->bytes_super)) {
477 spin_unlock(&block_group->tree_lock);
478 printk(KERN_ERR "block group %llu has an wrong amount of free "
479 "space\n", block_group->key.objectid);
483 spin_unlock(&block_group->tree_lock);
493 /* This cache is bogus, make sure it gets cleared */
494 spin_lock(&block_group->lock);
495 block_group->disk_cache_state = BTRFS_DC_CLEAR;
496 spin_unlock(&block_group->lock);
497 btrfs_remove_free_space_cache(block_group);
501 int btrfs_write_out_cache(struct btrfs_root *root,
502 struct btrfs_trans_handle *trans,
503 struct btrfs_block_group_cache *block_group,
504 struct btrfs_path *path)
506 struct btrfs_free_space_header *header;
507 struct extent_buffer *leaf;
509 struct rb_node *node;
510 struct list_head *pos, *n;
513 struct extent_state *cached_state = NULL;
514 struct btrfs_free_cluster *cluster = NULL;
515 struct extent_io_tree *unpin = NULL;
516 struct list_head bitmap_list;
517 struct btrfs_key key;
520 u32 *crc, *checksums;
521 unsigned long first_page_offset;
522 int index = 0, num_pages = 0;
526 bool next_page = false;
527 bool out_of_space = false;
529 root = root->fs_info->tree_root;
531 INIT_LIST_HEAD(&bitmap_list);
533 spin_lock(&block_group->lock);
534 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
535 spin_unlock(&block_group->lock);
538 spin_unlock(&block_group->lock);
540 inode = lookup_free_space_inode(root, block_group, path);
544 if (!i_size_read(inode)) {
549 node = rb_first(&block_group->free_space_offset);
555 num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
557 filemap_write_and_wait(inode->i_mapping);
558 btrfs_wait_ordered_range(inode, inode->i_size &
559 ~(root->sectorsize - 1), (u64)-1);
561 /* We need a checksum per page. */
562 crc = checksums = kzalloc(sizeof(u32) * num_pages, GFP_NOFS);
568 pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS);
575 /* Since the first page has all of our checksums and our generation we
576 * need to calculate the offset into the page that we can start writing
579 first_page_offset = (sizeof(u32) * num_pages) + sizeof(u64);
581 /* Get the cluster for this block_group if it exists */
582 if (!list_empty(&block_group->cluster_list))
583 cluster = list_entry(block_group->cluster_list.next,
584 struct btrfs_free_cluster,
588 * We shouldn't have switched the pinned extents yet so this is the
591 unpin = root->fs_info->pinned_extents;
594 * Lock all pages first so we can lock the extent safely.
596 * NOTE: Because we hold the ref the entire time we're going to write to
597 * the page find_get_page should never fail, so we don't do a check
598 * after find_get_page at this point. Just putting this here so people
599 * know and don't freak out.
601 while (index < num_pages) {
602 page = grab_cache_page(inode->i_mapping, index);
606 for (i = 0; i < num_pages; i++) {
607 unlock_page(pages[i]);
608 page_cache_release(pages[i]);
617 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
618 0, &cached_state, GFP_NOFS);
621 * When searching for pinned extents, we need to start at our start
624 start = block_group->key.objectid;
626 /* Write out the extent entries */
628 struct btrfs_free_space_entry *entry;
630 unsigned long offset = 0;
631 unsigned long start_offset = 0;
636 start_offset = first_page_offset;
637 offset = start_offset;
640 if (index >= num_pages) {
648 entry = addr + start_offset;
650 memset(addr, 0, PAGE_CACHE_SIZE);
651 while (node && !next_page) {
652 struct btrfs_free_space *e;
654 e = rb_entry(node, struct btrfs_free_space, offset_index);
657 entry->offset = cpu_to_le64(e->offset);
658 entry->bytes = cpu_to_le64(e->bytes);
660 entry->type = BTRFS_FREE_SPACE_BITMAP;
661 list_add_tail(&e->list, &bitmap_list);
664 entry->type = BTRFS_FREE_SPACE_EXTENT;
666 node = rb_next(node);
667 if (!node && cluster) {
668 node = rb_first(&cluster->root);
671 offset += sizeof(struct btrfs_free_space_entry);
672 if (offset + sizeof(struct btrfs_free_space_entry) >=
679 * We want to add any pinned extents to our free space cache
680 * so we don't leak the space
682 while (!next_page && (start < block_group->key.objectid +
683 block_group->key.offset)) {
684 ret = find_first_extent_bit(unpin, start, &start, &end,
691 /* This pinned extent is out of our range */
692 if (start >= block_group->key.objectid +
693 block_group->key.offset)
696 len = block_group->key.objectid +
697 block_group->key.offset - start;
698 len = min(len, end + 1 - start);
701 entry->offset = cpu_to_le64(start);
702 entry->bytes = cpu_to_le64(len);
703 entry->type = BTRFS_FREE_SPACE_EXTENT;
706 offset += sizeof(struct btrfs_free_space_entry);
707 if (offset + sizeof(struct btrfs_free_space_entry) >=
713 *crc = btrfs_csum_data(root, addr + start_offset, *crc,
714 PAGE_CACHE_SIZE - start_offset);
717 btrfs_csum_final(*crc, (char *)crc);
720 bytes += PAGE_CACHE_SIZE;
723 } while (node || next_page);
725 /* Write out the bitmaps */
726 list_for_each_safe(pos, n, &bitmap_list) {
728 struct btrfs_free_space *entry =
729 list_entry(pos, struct btrfs_free_space, list);
731 if (index >= num_pages) {
738 memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE);
740 *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE);
742 btrfs_csum_final(*crc, (char *)crc);
744 bytes += PAGE_CACHE_SIZE;
746 list_del_init(&entry->list);
751 btrfs_drop_pages(pages, num_pages);
752 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
753 i_size_read(inode) - 1, &cached_state,
759 /* Zero out the rest of the pages just to make sure */
760 while (index < num_pages) {
765 memset(addr, 0, PAGE_CACHE_SIZE);
767 bytes += PAGE_CACHE_SIZE;
771 /* Write the checksums and trans id to the first page */
779 memcpy(addr, checksums, sizeof(u32) * num_pages);
780 gen = addr + (sizeof(u32) * num_pages);
781 *gen = trans->transid;
785 ret = btrfs_dirty_pages(root, inode, pages, num_pages, 0,
786 bytes, &cached_state);
787 btrfs_drop_pages(pages, num_pages);
788 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
789 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
796 BTRFS_I(inode)->generation = trans->transid;
798 filemap_write_and_wait(inode->i_mapping);
800 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
801 key.offset = block_group->key.objectid;
804 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
807 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
808 EXTENT_DIRTY | EXTENT_DELALLOC |
809 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
812 leaf = path->nodes[0];
814 struct btrfs_key found_key;
815 BUG_ON(!path->slots[0]);
817 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
818 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
819 found_key.offset != block_group->key.objectid) {
821 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
822 EXTENT_DIRTY | EXTENT_DELALLOC |
823 EXTENT_DO_ACCOUNTING, 0, 0, NULL,
825 btrfs_release_path(root, path);
829 header = btrfs_item_ptr(leaf, path->slots[0],
830 struct btrfs_free_space_header);
831 btrfs_set_free_space_entries(leaf, header, entries);
832 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
833 btrfs_set_free_space_generation(leaf, header, trans->transid);
834 btrfs_mark_buffer_dirty(leaf);
835 btrfs_release_path(root, path);
841 invalidate_inode_pages2_range(inode->i_mapping, 0, index);
842 spin_lock(&block_group->lock);
843 block_group->disk_cache_state = BTRFS_DC_ERROR;
844 spin_unlock(&block_group->lock);
845 BTRFS_I(inode)->generation = 0;
849 btrfs_update_inode(trans, root, inode);
854 static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize,
857 BUG_ON(offset < bitmap_start);
858 offset -= bitmap_start;
859 return (unsigned long)(div64_u64(offset, sectorsize));
862 static inline unsigned long bytes_to_bits(u64 bytes, u64 sectorsize)
864 return (unsigned long)(div64_u64(bytes, sectorsize));
867 static inline u64 offset_to_bitmap(struct btrfs_block_group_cache *block_group,
871 u64 bytes_per_bitmap;
873 bytes_per_bitmap = BITS_PER_BITMAP * block_group->sectorsize;
874 bitmap_start = offset - block_group->key.objectid;
875 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
876 bitmap_start *= bytes_per_bitmap;
877 bitmap_start += block_group->key.objectid;
882 static int tree_insert_offset(struct rb_root *root, u64 offset,
883 struct rb_node *node, int bitmap)
885 struct rb_node **p = &root->rb_node;
886 struct rb_node *parent = NULL;
887 struct btrfs_free_space *info;
891 info = rb_entry(parent, struct btrfs_free_space, offset_index);
893 if (offset < info->offset) {
895 } else if (offset > info->offset) {
899 * we could have a bitmap entry and an extent entry
900 * share the same offset. If this is the case, we want
901 * the extent entry to always be found first if we do a
902 * linear search through the tree, since we want to have
903 * the quickest allocation time, and allocating from an
904 * extent is faster than allocating from a bitmap. So
905 * if we're inserting a bitmap and we find an entry at
906 * this offset, we want to go right, or after this entry
907 * logically. If we are inserting an extent and we've
908 * found a bitmap, we want to go left, or before
912 WARN_ON(info->bitmap);
915 WARN_ON(!info->bitmap);
921 rb_link_node(node, parent, p);
922 rb_insert_color(node, root);
928 * searches the tree for the given offset.
930 * fuzzy - If this is set, then we are trying to make an allocation, and we just
931 * want a section that has at least bytes size and comes at or after the given
934 static struct btrfs_free_space *
935 tree_search_offset(struct btrfs_block_group_cache *block_group,
936 u64 offset, int bitmap_only, int fuzzy)
938 struct rb_node *n = block_group->free_space_offset.rb_node;
939 struct btrfs_free_space *entry, *prev = NULL;
941 /* find entry that is closest to the 'offset' */
948 entry = rb_entry(n, struct btrfs_free_space, offset_index);
951 if (offset < entry->offset)
953 else if (offset > entry->offset)
966 * bitmap entry and extent entry may share same offset,
967 * in that case, bitmap entry comes after extent entry.
972 entry = rb_entry(n, struct btrfs_free_space, offset_index);
973 if (entry->offset != offset)
976 WARN_ON(!entry->bitmap);
981 * if previous extent entry covers the offset,
982 * we should return it instead of the bitmap entry
984 n = &entry->offset_index;
989 prev = rb_entry(n, struct btrfs_free_space,
992 if (prev->offset + prev->bytes > offset)
1004 /* find last entry before the 'offset' */
1006 if (entry->offset > offset) {
1007 n = rb_prev(&entry->offset_index);
1009 entry = rb_entry(n, struct btrfs_free_space,
1011 BUG_ON(entry->offset > offset);
1020 if (entry->bitmap) {
1021 n = &entry->offset_index;
1026 prev = rb_entry(n, struct btrfs_free_space,
1028 if (!prev->bitmap) {
1029 if (prev->offset + prev->bytes > offset)
1034 if (entry->offset + BITS_PER_BITMAP *
1035 block_group->sectorsize > offset)
1037 } else if (entry->offset + entry->bytes > offset)
1044 if (entry->bitmap) {
1045 if (entry->offset + BITS_PER_BITMAP *
1046 block_group->sectorsize > offset)
1049 if (entry->offset + entry->bytes > offset)
1053 n = rb_next(&entry->offset_index);
1056 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1062 __unlink_free_space(struct btrfs_block_group_cache *block_group,
1063 struct btrfs_free_space *info)
1065 rb_erase(&info->offset_index, &block_group->free_space_offset);
1066 block_group->free_extents--;
1069 static void unlink_free_space(struct btrfs_block_group_cache *block_group,
1070 struct btrfs_free_space *info)
1072 __unlink_free_space(block_group, info);
1073 block_group->free_space -= info->bytes;
1076 static int link_free_space(struct btrfs_block_group_cache *block_group,
1077 struct btrfs_free_space *info)
1081 BUG_ON(!info->bitmap && !info->bytes);
1082 ret = tree_insert_offset(&block_group->free_space_offset, info->offset,
1083 &info->offset_index, (info->bitmap != NULL));
1087 block_group->free_space += info->bytes;
1088 block_group->free_extents++;
1092 static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
1097 u64 size = block_group->key.offset;
1100 * The goal is to keep the total amount of memory used per 1gb of space
1101 * at or below 32k, so we need to adjust how much memory we allow to be
1102 * used by extent based free space tracking
1104 if (size < 1024 * 1024 * 1024)
1105 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1107 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1108 div64_u64(size, 1024 * 1024 * 1024);
1111 * we want to account for 1 more bitmap than what we have so we can make
1112 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1113 * we add more bitmaps.
1115 bitmap_bytes = (block_group->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1117 if (bitmap_bytes >= max_bytes) {
1118 block_group->extents_thresh = 0;
1123 * we want the extent entry threshold to always be at most 1/2 the maxw
1124 * bytes we can have, or whatever is less than that.
1126 extent_bytes = max_bytes - bitmap_bytes;
1127 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1129 block_group->extents_thresh =
1130 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1133 static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group,
1134 struct btrfs_free_space *info, u64 offset,
1137 unsigned long start, end;
1140 start = offset_to_bit(info->offset, block_group->sectorsize, offset);
1141 end = start + bytes_to_bits(bytes, block_group->sectorsize);
1142 BUG_ON(end > BITS_PER_BITMAP);
1144 for (i = start; i < end; i++)
1145 clear_bit(i, info->bitmap);
1147 info->bytes -= bytes;
1148 block_group->free_space -= bytes;
1151 static void bitmap_set_bits(struct btrfs_block_group_cache *block_group,
1152 struct btrfs_free_space *info, u64 offset,
1155 unsigned long start, end;
1158 start = offset_to_bit(info->offset, block_group->sectorsize, offset);
1159 end = start + bytes_to_bits(bytes, block_group->sectorsize);
1160 BUG_ON(end > BITS_PER_BITMAP);
1162 for (i = start; i < end; i++)
1163 set_bit(i, info->bitmap);
1165 info->bytes += bytes;
1166 block_group->free_space += bytes;
1169 static int search_bitmap(struct btrfs_block_group_cache *block_group,
1170 struct btrfs_free_space *bitmap_info, u64 *offset,
1173 unsigned long found_bits = 0;
1174 unsigned long bits, i;
1175 unsigned long next_zero;
1177 i = offset_to_bit(bitmap_info->offset, block_group->sectorsize,
1178 max_t(u64, *offset, bitmap_info->offset));
1179 bits = bytes_to_bits(*bytes, block_group->sectorsize);
1181 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1182 i < BITS_PER_BITMAP;
1183 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1184 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1185 BITS_PER_BITMAP, i);
1186 if ((next_zero - i) >= bits) {
1187 found_bits = next_zero - i;
1194 *offset = (u64)(i * block_group->sectorsize) +
1195 bitmap_info->offset;
1196 *bytes = (u64)(found_bits) * block_group->sectorsize;
1203 static struct btrfs_free_space *find_free_space(struct btrfs_block_group_cache
1204 *block_group, u64 *offset,
1205 u64 *bytes, int debug)
1207 struct btrfs_free_space *entry;
1208 struct rb_node *node;
1211 if (!block_group->free_space_offset.rb_node)
1214 entry = tree_search_offset(block_group,
1215 offset_to_bitmap(block_group, *offset),
1220 for (node = &entry->offset_index; node; node = rb_next(node)) {
1221 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1222 if (entry->bytes < *bytes)
1225 if (entry->bitmap) {
1226 ret = search_bitmap(block_group, entry, offset, bytes);
1232 *offset = entry->offset;
1233 *bytes = entry->bytes;
1240 static void add_new_bitmap(struct btrfs_block_group_cache *block_group,
1241 struct btrfs_free_space *info, u64 offset)
1243 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1244 int max_bitmaps = (int)div64_u64(block_group->key.offset +
1245 bytes_per_bg - 1, bytes_per_bg);
1246 BUG_ON(block_group->total_bitmaps >= max_bitmaps);
1248 info->offset = offset_to_bitmap(block_group, offset);
1250 link_free_space(block_group, info);
1251 block_group->total_bitmaps++;
1253 recalculate_thresholds(block_group);
1256 static void free_bitmap(struct btrfs_block_group_cache *block_group,
1257 struct btrfs_free_space *bitmap_info)
1259 unlink_free_space(block_group, bitmap_info);
1260 kfree(bitmap_info->bitmap);
1261 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1262 block_group->total_bitmaps--;
1263 recalculate_thresholds(block_group);
1266 static noinline int remove_from_bitmap(struct btrfs_block_group_cache *block_group,
1267 struct btrfs_free_space *bitmap_info,
1268 u64 *offset, u64 *bytes)
1271 u64 search_start, search_bytes;
1275 end = bitmap_info->offset +
1276 (u64)(BITS_PER_BITMAP * block_group->sectorsize) - 1;
1279 * XXX - this can go away after a few releases.
1281 * since the only user of btrfs_remove_free_space is the tree logging
1282 * stuff, and the only way to test that is under crash conditions, we
1283 * want to have this debug stuff here just in case somethings not
1284 * working. Search the bitmap for the space we are trying to use to
1285 * make sure its actually there. If its not there then we need to stop
1286 * because something has gone wrong.
1288 search_start = *offset;
1289 search_bytes = *bytes;
1290 search_bytes = min(search_bytes, end - search_start + 1);
1291 ret = search_bitmap(block_group, bitmap_info, &search_start,
1293 BUG_ON(ret < 0 || search_start != *offset);
1295 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1296 bitmap_clear_bits(block_group, bitmap_info, *offset,
1298 *bytes -= end - *offset + 1;
1300 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1301 bitmap_clear_bits(block_group, bitmap_info, *offset, *bytes);
1306 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1307 if (!bitmap_info->bytes)
1308 free_bitmap(block_group, bitmap_info);
1311 * no entry after this bitmap, but we still have bytes to
1312 * remove, so something has gone wrong.
1317 bitmap_info = rb_entry(next, struct btrfs_free_space,
1321 * if the next entry isn't a bitmap we need to return to let the
1322 * extent stuff do its work.
1324 if (!bitmap_info->bitmap)
1328 * Ok the next item is a bitmap, but it may not actually hold
1329 * the information for the rest of this free space stuff, so
1330 * look for it, and if we don't find it return so we can try
1331 * everything over again.
1333 search_start = *offset;
1334 search_bytes = *bytes;
1335 ret = search_bitmap(block_group, bitmap_info, &search_start,
1337 if (ret < 0 || search_start != *offset)
1341 } else if (!bitmap_info->bytes)
1342 free_bitmap(block_group, bitmap_info);
1347 static int insert_into_bitmap(struct btrfs_block_group_cache *block_group,
1348 struct btrfs_free_space *info)
1350 struct btrfs_free_space *bitmap_info;
1352 u64 bytes, offset, end;
1356 * If we are below the extents threshold then we can add this as an
1357 * extent, and don't have to deal with the bitmap
1359 if (block_group->free_extents < block_group->extents_thresh) {
1361 * If this block group has some small extents we don't want to
1362 * use up all of our free slots in the cache with them, we want
1363 * to reserve them to larger extents, however if we have plent
1364 * of cache left then go ahead an dadd them, no sense in adding
1365 * the overhead of a bitmap if we don't have to.
1367 if (info->bytes <= block_group->sectorsize * 4) {
1368 if (block_group->free_extents * 2 <=
1369 block_group->extents_thresh)
1377 * some block groups are so tiny they can't be enveloped by a bitmap, so
1378 * don't even bother to create a bitmap for this
1380 if (BITS_PER_BITMAP * block_group->sectorsize >
1381 block_group->key.offset)
1384 bytes = info->bytes;
1385 offset = info->offset;
1388 bitmap_info = tree_search_offset(block_group,
1389 offset_to_bitmap(block_group, offset),
1396 end = bitmap_info->offset +
1397 (u64)(BITS_PER_BITMAP * block_group->sectorsize);
1399 if (offset >= bitmap_info->offset && offset + bytes > end) {
1400 bitmap_set_bits(block_group, bitmap_info, offset,
1402 bytes -= end - offset;
1405 } else if (offset >= bitmap_info->offset && offset + bytes <= end) {
1406 bitmap_set_bits(block_group, bitmap_info, offset, bytes);
1419 if (info && info->bitmap) {
1420 add_new_bitmap(block_group, info, offset);
1425 spin_unlock(&block_group->tree_lock);
1427 /* no pre-allocated info, allocate a new one */
1429 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1432 spin_lock(&block_group->tree_lock);
1438 /* allocate the bitmap */
1439 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1440 spin_lock(&block_group->tree_lock);
1441 if (!info->bitmap) {
1451 kfree(info->bitmap);
1452 kmem_cache_free(btrfs_free_space_cachep, info);
1458 bool try_merge_free_space(struct btrfs_block_group_cache *block_group,
1459 struct btrfs_free_space *info, bool update_stat)
1461 struct btrfs_free_space *left_info;
1462 struct btrfs_free_space *right_info;
1463 bool merged = false;
1464 u64 offset = info->offset;
1465 u64 bytes = info->bytes;
1468 * first we want to see if there is free space adjacent to the range we
1469 * are adding, if there is remove that struct and add a new one to
1470 * cover the entire range
1472 right_info = tree_search_offset(block_group, offset + bytes, 0, 0);
1473 if (right_info && rb_prev(&right_info->offset_index))
1474 left_info = rb_entry(rb_prev(&right_info->offset_index),
1475 struct btrfs_free_space, offset_index);
1477 left_info = tree_search_offset(block_group, offset - 1, 0, 0);
1479 if (right_info && !right_info->bitmap) {
1481 unlink_free_space(block_group, right_info);
1483 __unlink_free_space(block_group, right_info);
1484 info->bytes += right_info->bytes;
1485 kmem_cache_free(btrfs_free_space_cachep, right_info);
1489 if (left_info && !left_info->bitmap &&
1490 left_info->offset + left_info->bytes == offset) {
1492 unlink_free_space(block_group, left_info);
1494 __unlink_free_space(block_group, left_info);
1495 info->offset = left_info->offset;
1496 info->bytes += left_info->bytes;
1497 kmem_cache_free(btrfs_free_space_cachep, left_info);
1504 int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
1505 u64 offset, u64 bytes)
1507 struct btrfs_free_space *info;
1510 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1514 info->offset = offset;
1515 info->bytes = bytes;
1517 spin_lock(&block_group->tree_lock);
1519 if (try_merge_free_space(block_group, info, true))
1523 * There was no extent directly to the left or right of this new
1524 * extent then we know we're going to have to allocate a new extent, so
1525 * before we do that see if we need to drop this into a bitmap
1527 ret = insert_into_bitmap(block_group, info);
1535 ret = link_free_space(block_group, info);
1537 kmem_cache_free(btrfs_free_space_cachep, info);
1539 spin_unlock(&block_group->tree_lock);
1542 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1543 BUG_ON(ret == -EEXIST);
1549 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1550 u64 offset, u64 bytes)
1552 struct btrfs_free_space *info;
1553 struct btrfs_free_space *next_info = NULL;
1556 spin_lock(&block_group->tree_lock);
1559 info = tree_search_offset(block_group, offset, 0, 0);
1562 * oops didn't find an extent that matched the space we wanted
1563 * to remove, look for a bitmap instead
1565 info = tree_search_offset(block_group,
1566 offset_to_bitmap(block_group, offset),
1574 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1576 next_info = rb_entry(rb_next(&info->offset_index),
1577 struct btrfs_free_space,
1580 if (next_info->bitmap)
1581 end = next_info->offset + BITS_PER_BITMAP *
1582 block_group->sectorsize - 1;
1584 end = next_info->offset + next_info->bytes;
1586 if (next_info->bytes < bytes ||
1587 next_info->offset > offset || offset > end) {
1588 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1589 " trying to use %llu\n",
1590 (unsigned long long)info->offset,
1591 (unsigned long long)info->bytes,
1592 (unsigned long long)bytes);
1601 if (info->bytes == bytes) {
1602 unlink_free_space(block_group, info);
1604 kfree(info->bitmap);
1605 block_group->total_bitmaps--;
1607 kmem_cache_free(btrfs_free_space_cachep, info);
1611 if (!info->bitmap && info->offset == offset) {
1612 unlink_free_space(block_group, info);
1613 info->offset += bytes;
1614 info->bytes -= bytes;
1615 link_free_space(block_group, info);
1619 if (!info->bitmap && info->offset <= offset &&
1620 info->offset + info->bytes >= offset + bytes) {
1621 u64 old_start = info->offset;
1623 * we're freeing space in the middle of the info,
1624 * this can happen during tree log replay
1626 * first unlink the old info and then
1627 * insert it again after the hole we're creating
1629 unlink_free_space(block_group, info);
1630 if (offset + bytes < info->offset + info->bytes) {
1631 u64 old_end = info->offset + info->bytes;
1633 info->offset = offset + bytes;
1634 info->bytes = old_end - info->offset;
1635 ret = link_free_space(block_group, info);
1640 /* the hole we're creating ends at the end
1641 * of the info struct, just free the info
1643 kmem_cache_free(btrfs_free_space_cachep, info);
1645 spin_unlock(&block_group->tree_lock);
1647 /* step two, insert a new info struct to cover
1648 * anything before the hole
1650 ret = btrfs_add_free_space(block_group, old_start,
1651 offset - old_start);
1656 ret = remove_from_bitmap(block_group, info, &offset, &bytes);
1661 spin_unlock(&block_group->tree_lock);
1666 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1669 struct btrfs_free_space *info;
1673 for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) {
1674 info = rb_entry(n, struct btrfs_free_space, offset_index);
1675 if (info->bytes >= bytes)
1677 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1678 (unsigned long long)info->offset,
1679 (unsigned long long)info->bytes,
1680 (info->bitmap) ? "yes" : "no");
1682 printk(KERN_INFO "block group has cluster?: %s\n",
1683 list_empty(&block_group->cluster_list) ? "no" : "yes");
1684 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1688 u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group)
1690 struct btrfs_free_space *info;
1694 for (n = rb_first(&block_group->free_space_offset); n;
1696 info = rb_entry(n, struct btrfs_free_space, offset_index);
1704 * for a given cluster, put all of its extents back into the free
1705 * space cache. If the block group passed doesn't match the block group
1706 * pointed to by the cluster, someone else raced in and freed the
1707 * cluster already. In that case, we just return without changing anything
1710 __btrfs_return_cluster_to_free_space(
1711 struct btrfs_block_group_cache *block_group,
1712 struct btrfs_free_cluster *cluster)
1714 struct btrfs_free_space *entry;
1715 struct rb_node *node;
1717 spin_lock(&cluster->lock);
1718 if (cluster->block_group != block_group)
1721 cluster->block_group = NULL;
1722 cluster->window_start = 0;
1723 list_del_init(&cluster->block_group_list);
1725 node = rb_first(&cluster->root);
1729 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1730 node = rb_next(&entry->offset_index);
1731 rb_erase(&entry->offset_index, &cluster->root);
1733 bitmap = (entry->bitmap != NULL);
1735 try_merge_free_space(block_group, entry, false);
1736 tree_insert_offset(&block_group->free_space_offset,
1737 entry->offset, &entry->offset_index, bitmap);
1739 cluster->root = RB_ROOT;
1742 spin_unlock(&cluster->lock);
1743 btrfs_put_block_group(block_group);
1747 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
1749 struct btrfs_free_space *info;
1750 struct rb_node *node;
1751 struct btrfs_free_cluster *cluster;
1752 struct list_head *head;
1754 spin_lock(&block_group->tree_lock);
1755 while ((head = block_group->cluster_list.next) !=
1756 &block_group->cluster_list) {
1757 cluster = list_entry(head, struct btrfs_free_cluster,
1760 WARN_ON(cluster->block_group != block_group);
1761 __btrfs_return_cluster_to_free_space(block_group, cluster);
1762 if (need_resched()) {
1763 spin_unlock(&block_group->tree_lock);
1765 spin_lock(&block_group->tree_lock);
1769 while ((node = rb_last(&block_group->free_space_offset)) != NULL) {
1770 info = rb_entry(node, struct btrfs_free_space, offset_index);
1771 if (!info->bitmap) {
1772 unlink_free_space(block_group, info);
1773 kmem_cache_free(btrfs_free_space_cachep, info);
1775 free_bitmap(block_group, info);
1778 if (need_resched()) {
1779 spin_unlock(&block_group->tree_lock);
1781 spin_lock(&block_group->tree_lock);
1785 spin_unlock(&block_group->tree_lock);
1788 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
1789 u64 offset, u64 bytes, u64 empty_size)
1791 struct btrfs_free_space *entry = NULL;
1792 u64 bytes_search = bytes + empty_size;
1795 spin_lock(&block_group->tree_lock);
1796 entry = find_free_space(block_group, &offset, &bytes_search, 0);
1801 if (entry->bitmap) {
1802 bitmap_clear_bits(block_group, entry, offset, bytes);
1804 free_bitmap(block_group, entry);
1806 unlink_free_space(block_group, entry);
1807 entry->offset += bytes;
1808 entry->bytes -= bytes;
1810 kmem_cache_free(btrfs_free_space_cachep, entry);
1812 link_free_space(block_group, entry);
1816 spin_unlock(&block_group->tree_lock);
1822 * given a cluster, put all of its extents back into the free space
1823 * cache. If a block group is passed, this function will only free
1824 * a cluster that belongs to the passed block group.
1826 * Otherwise, it'll get a reference on the block group pointed to by the
1827 * cluster and remove the cluster from it.
1829 int btrfs_return_cluster_to_free_space(
1830 struct btrfs_block_group_cache *block_group,
1831 struct btrfs_free_cluster *cluster)
1835 /* first, get a safe pointer to the block group */
1836 spin_lock(&cluster->lock);
1838 block_group = cluster->block_group;
1840 spin_unlock(&cluster->lock);
1843 } else if (cluster->block_group != block_group) {
1844 /* someone else has already freed it don't redo their work */
1845 spin_unlock(&cluster->lock);
1848 atomic_inc(&block_group->count);
1849 spin_unlock(&cluster->lock);
1851 /* now return any extents the cluster had on it */
1852 spin_lock(&block_group->tree_lock);
1853 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
1854 spin_unlock(&block_group->tree_lock);
1856 /* finally drop our ref */
1857 btrfs_put_block_group(block_group);
1861 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
1862 struct btrfs_free_cluster *cluster,
1863 struct btrfs_free_space *entry,
1864 u64 bytes, u64 min_start)
1867 u64 search_start = cluster->window_start;
1868 u64 search_bytes = bytes;
1871 search_start = min_start;
1872 search_bytes = bytes;
1874 err = search_bitmap(block_group, entry, &search_start,
1880 bitmap_clear_bits(block_group, entry, ret, bytes);
1886 * given a cluster, try to allocate 'bytes' from it, returns 0
1887 * if it couldn't find anything suitably large, or a logical disk offset
1888 * if things worked out
1890 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
1891 struct btrfs_free_cluster *cluster, u64 bytes,
1894 struct btrfs_free_space *entry = NULL;
1895 struct rb_node *node;
1898 spin_lock(&cluster->lock);
1899 if (bytes > cluster->max_size)
1902 if (cluster->block_group != block_group)
1905 node = rb_first(&cluster->root);
1909 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1911 if (entry->bytes < bytes ||
1912 (!entry->bitmap && entry->offset < min_start)) {
1913 struct rb_node *node;
1915 node = rb_next(&entry->offset_index);
1918 entry = rb_entry(node, struct btrfs_free_space,
1923 if (entry->bitmap) {
1924 ret = btrfs_alloc_from_bitmap(block_group,
1925 cluster, entry, bytes,
1928 struct rb_node *node;
1929 node = rb_next(&entry->offset_index);
1932 entry = rb_entry(node, struct btrfs_free_space,
1938 ret = entry->offset;
1940 entry->offset += bytes;
1941 entry->bytes -= bytes;
1944 if (entry->bytes == 0)
1945 rb_erase(&entry->offset_index, &cluster->root);
1949 spin_unlock(&cluster->lock);
1954 spin_lock(&block_group->tree_lock);
1956 block_group->free_space -= bytes;
1957 if (entry->bytes == 0) {
1958 block_group->free_extents--;
1959 if (entry->bitmap) {
1960 kfree(entry->bitmap);
1961 block_group->total_bitmaps--;
1962 recalculate_thresholds(block_group);
1964 kmem_cache_free(btrfs_free_space_cachep, entry);
1967 spin_unlock(&block_group->tree_lock);
1972 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
1973 struct btrfs_free_space *entry,
1974 struct btrfs_free_cluster *cluster,
1975 u64 offset, u64 bytes, u64 min_bytes)
1977 unsigned long next_zero;
1979 unsigned long search_bits;
1980 unsigned long total_bits;
1981 unsigned long found_bits;
1982 unsigned long start = 0;
1983 unsigned long total_found = 0;
1987 i = offset_to_bit(entry->offset, block_group->sectorsize,
1988 max_t(u64, offset, entry->offset));
1989 search_bits = bytes_to_bits(bytes, block_group->sectorsize);
1990 total_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
1994 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
1995 i < BITS_PER_BITMAP;
1996 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
1997 next_zero = find_next_zero_bit(entry->bitmap,
1998 BITS_PER_BITMAP, i);
1999 if (next_zero - i >= search_bits) {
2000 found_bits = next_zero - i;
2014 total_found += found_bits;
2016 if (cluster->max_size < found_bits * block_group->sectorsize)
2017 cluster->max_size = found_bits * block_group->sectorsize;
2019 if (total_found < total_bits) {
2020 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
2021 if (i - start > total_bits * 2) {
2023 cluster->max_size = 0;
2029 cluster->window_start = start * block_group->sectorsize +
2031 rb_erase(&entry->offset_index, &block_group->free_space_offset);
2032 ret = tree_insert_offset(&cluster->root, entry->offset,
2033 &entry->offset_index, 1);
2040 * This searches the block group for just extents to fill the cluster with.
2042 static int setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2043 struct btrfs_free_cluster *cluster,
2044 u64 offset, u64 bytes, u64 min_bytes)
2046 struct btrfs_free_space *first = NULL;
2047 struct btrfs_free_space *entry = NULL;
2048 struct btrfs_free_space *prev = NULL;
2049 struct btrfs_free_space *last;
2050 struct rb_node *node;
2054 u64 max_gap = 128 * 1024;
2056 entry = tree_search_offset(block_group, offset, 0, 1);
2061 * We don't want bitmaps, so just move along until we find a normal
2064 while (entry->bitmap) {
2065 node = rb_next(&entry->offset_index);
2068 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2071 window_start = entry->offset;
2072 window_free = entry->bytes;
2073 max_extent = entry->bytes;
2078 while (window_free <= min_bytes) {
2079 node = rb_next(&entry->offset_index);
2082 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2087 * we haven't filled the empty size and the window is
2088 * very large. reset and try again
2090 if (entry->offset - (prev->offset + prev->bytes) > max_gap ||
2091 entry->offset - window_start > (min_bytes * 2)) {
2093 window_start = entry->offset;
2094 window_free = entry->bytes;
2096 max_extent = entry->bytes;
2099 window_free += entry->bytes;
2100 if (entry->bytes > max_extent)
2101 max_extent = entry->bytes;
2106 cluster->window_start = first->offset;
2108 node = &first->offset_index;
2111 * now we've found our entries, pull them out of the free space
2112 * cache and put them into the cluster rbtree
2117 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2118 node = rb_next(&entry->offset_index);
2122 rb_erase(&entry->offset_index, &block_group->free_space_offset);
2123 ret = tree_insert_offset(&cluster->root, entry->offset,
2124 &entry->offset_index, 0);
2126 } while (node && entry != last);
2128 cluster->max_size = max_extent;
2134 * This specifically looks for bitmaps that may work in the cluster, we assume
2135 * that we have already failed to find extents that will work.
2137 static int setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2138 struct btrfs_free_cluster *cluster,
2139 u64 offset, u64 bytes, u64 min_bytes)
2141 struct btrfs_free_space *entry;
2142 struct rb_node *node;
2145 if (block_group->total_bitmaps == 0)
2148 entry = tree_search_offset(block_group,
2149 offset_to_bitmap(block_group, offset),
2154 node = &entry->offset_index;
2156 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2157 node = rb_next(&entry->offset_index);
2160 if (entry->bytes < min_bytes)
2162 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2164 } while (ret && node);
2170 * here we try to find a cluster of blocks in a block group. The goal
2171 * is to find at least bytes free and up to empty_size + bytes free.
2172 * We might not find them all in one contiguous area.
2174 * returns zero and sets up cluster if things worked out, otherwise
2175 * it returns -enospc
2177 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2178 struct btrfs_root *root,
2179 struct btrfs_block_group_cache *block_group,
2180 struct btrfs_free_cluster *cluster,
2181 u64 offset, u64 bytes, u64 empty_size)
2186 /* for metadata, allow allocates with more holes */
2187 if (btrfs_test_opt(root, SSD_SPREAD)) {
2188 min_bytes = bytes + empty_size;
2189 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2191 * we want to do larger allocations when we are
2192 * flushing out the delayed refs, it helps prevent
2193 * making more work as we go along.
2195 if (trans->transaction->delayed_refs.flushing)
2196 min_bytes = max(bytes, (bytes + empty_size) >> 1);
2198 min_bytes = max(bytes, (bytes + empty_size) >> 4);
2200 min_bytes = max(bytes, (bytes + empty_size) >> 2);
2202 spin_lock(&block_group->tree_lock);
2205 * If we know we don't have enough space to make a cluster don't even
2206 * bother doing all the work to try and find one.
2208 if (block_group->free_space < min_bytes) {
2209 spin_unlock(&block_group->tree_lock);
2213 spin_lock(&cluster->lock);
2215 /* someone already found a cluster, hooray */
2216 if (cluster->block_group) {
2221 ret = setup_cluster_no_bitmap(block_group, cluster, offset, bytes,
2224 ret = setup_cluster_bitmap(block_group, cluster, offset,
2228 atomic_inc(&block_group->count);
2229 list_add_tail(&cluster->block_group_list,
2230 &block_group->cluster_list);
2231 cluster->block_group = block_group;
2234 spin_unlock(&cluster->lock);
2235 spin_unlock(&block_group->tree_lock);
2241 * simple code to zero out a cluster
2243 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2245 spin_lock_init(&cluster->lock);
2246 spin_lock_init(&cluster->refill_lock);
2247 cluster->root = RB_ROOT;
2248 cluster->max_size = 0;
2249 INIT_LIST_HEAD(&cluster->block_group_list);
2250 cluster->block_group = NULL;
2253 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2254 u64 *trimmed, u64 start, u64 end, u64 minlen)
2256 struct btrfs_free_space *entry = NULL;
2257 struct btrfs_fs_info *fs_info = block_group->fs_info;
2259 u64 actually_trimmed;
2264 while (start < end) {
2265 spin_lock(&block_group->tree_lock);
2267 if (block_group->free_space < minlen) {
2268 spin_unlock(&block_group->tree_lock);
2272 entry = tree_search_offset(block_group, start, 0, 1);
2274 entry = tree_search_offset(block_group,
2275 offset_to_bitmap(block_group,
2279 if (!entry || entry->offset >= end) {
2280 spin_unlock(&block_group->tree_lock);
2284 if (entry->bitmap) {
2285 ret = search_bitmap(block_group, entry, &start, &bytes);
2288 spin_unlock(&block_group->tree_lock);
2291 bytes = min(bytes, end - start);
2292 bitmap_clear_bits(block_group, entry,
2294 if (entry->bytes == 0)
2295 free_bitmap(block_group, entry);
2297 start = entry->offset + BITS_PER_BITMAP *
2298 block_group->sectorsize;
2299 spin_unlock(&block_group->tree_lock);
2304 start = entry->offset;
2305 bytes = min(entry->bytes, end - start);
2306 unlink_free_space(block_group, entry);
2307 kmem_cache_free(btrfs_free_space_cachep, entry);
2310 spin_unlock(&block_group->tree_lock);
2312 if (bytes >= minlen) {
2314 update_ret = btrfs_update_reserved_bytes(block_group,
2317 ret = btrfs_error_discard_extent(fs_info->extent_root,
2322 btrfs_add_free_space(block_group,
2325 btrfs_update_reserved_bytes(block_group,
2330 *trimmed += actually_trimmed;
2335 if (fatal_signal_pending(current)) {