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"
28 #include "inode-map.h"
30 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
31 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
33 static int link_free_space(struct btrfs_free_space_ctl *ctl,
34 struct btrfs_free_space *info);
36 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
37 struct btrfs_path *path,
41 struct btrfs_key location;
42 struct btrfs_disk_key disk_key;
43 struct btrfs_free_space_header *header;
44 struct extent_buffer *leaf;
45 struct inode *inode = NULL;
48 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
52 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
56 btrfs_release_path(path);
57 return ERR_PTR(-ENOENT);
60 leaf = path->nodes[0];
61 header = btrfs_item_ptr(leaf, path->slots[0],
62 struct btrfs_free_space_header);
63 btrfs_free_space_key(leaf, header, &disk_key);
64 btrfs_disk_key_to_cpu(&location, &disk_key);
65 btrfs_release_path(path);
67 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
69 return ERR_PTR(-ENOENT);
72 if (is_bad_inode(inode)) {
74 return ERR_PTR(-ENOENT);
77 inode->i_mapping->flags &= ~__GFP_FS;
82 struct inode *lookup_free_space_inode(struct btrfs_root *root,
83 struct btrfs_block_group_cache
84 *block_group, struct btrfs_path *path)
86 struct inode *inode = NULL;
88 spin_lock(&block_group->lock);
89 if (block_group->inode)
90 inode = igrab(block_group->inode);
91 spin_unlock(&block_group->lock);
95 inode = __lookup_free_space_inode(root, path,
96 block_group->key.objectid);
100 spin_lock(&block_group->lock);
101 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) {
102 printk(KERN_INFO "Old style space inode found, converting.\n");
103 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NODATASUM;
104 block_group->disk_cache_state = BTRFS_DC_CLEAR;
107 if (!btrfs_fs_closing(root->fs_info)) {
108 block_group->inode = igrab(inode);
109 block_group->iref = 1;
111 spin_unlock(&block_group->lock);
116 int __create_free_space_inode(struct btrfs_root *root,
117 struct btrfs_trans_handle *trans,
118 struct btrfs_path *path, u64 ino, u64 offset)
120 struct btrfs_key key;
121 struct btrfs_disk_key disk_key;
122 struct btrfs_free_space_header *header;
123 struct btrfs_inode_item *inode_item;
124 struct extent_buffer *leaf;
127 ret = btrfs_insert_empty_inode(trans, root, path, ino);
131 leaf = path->nodes[0];
132 inode_item = btrfs_item_ptr(leaf, path->slots[0],
133 struct btrfs_inode_item);
134 btrfs_item_key(leaf, &disk_key, path->slots[0]);
135 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
136 sizeof(*inode_item));
137 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
138 btrfs_set_inode_size(leaf, inode_item, 0);
139 btrfs_set_inode_nbytes(leaf, inode_item, 0);
140 btrfs_set_inode_uid(leaf, inode_item, 0);
141 btrfs_set_inode_gid(leaf, inode_item, 0);
142 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
143 btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS |
144 BTRFS_INODE_PREALLOC);
145 btrfs_set_inode_nlink(leaf, inode_item, 1);
146 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
147 btrfs_set_inode_block_group(leaf, inode_item, offset);
148 btrfs_mark_buffer_dirty(leaf);
149 btrfs_release_path(path);
151 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
155 ret = btrfs_insert_empty_item(trans, root, path, &key,
156 sizeof(struct btrfs_free_space_header));
158 btrfs_release_path(path);
161 leaf = path->nodes[0];
162 header = btrfs_item_ptr(leaf, path->slots[0],
163 struct btrfs_free_space_header);
164 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
165 btrfs_set_free_space_key(leaf, header, &disk_key);
166 btrfs_mark_buffer_dirty(leaf);
167 btrfs_release_path(path);
172 int create_free_space_inode(struct btrfs_root *root,
173 struct btrfs_trans_handle *trans,
174 struct btrfs_block_group_cache *block_group,
175 struct btrfs_path *path)
180 ret = btrfs_find_free_objectid(root, &ino);
184 return __create_free_space_inode(root, trans, path, ino,
185 block_group->key.objectid);
188 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
189 struct btrfs_trans_handle *trans,
190 struct btrfs_path *path,
193 struct btrfs_block_rsv *rsv;
197 rsv = trans->block_rsv;
198 trans->block_rsv = root->orphan_block_rsv;
199 ret = btrfs_block_rsv_check(trans, root,
200 root->orphan_block_rsv,
205 oldsize = i_size_read(inode);
206 btrfs_i_size_write(inode, 0);
207 truncate_pagecache(inode, oldsize, 0);
210 * We don't need an orphan item because truncating the free space cache
211 * will never be split across transactions.
213 ret = btrfs_truncate_inode_items(trans, root, inode,
214 0, BTRFS_EXTENT_DATA_KEY);
216 trans->block_rsv = rsv;
222 ret = btrfs_update_inode(trans, root, inode);
226 static int readahead_cache(struct inode *inode)
228 struct file_ra_state *ra;
229 unsigned long last_index;
231 ra = kzalloc(sizeof(*ra), GFP_NOFS);
235 file_ra_state_init(ra, inode->i_mapping);
236 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
238 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
245 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
246 struct btrfs_free_space_ctl *ctl,
247 struct btrfs_path *path, u64 offset)
249 struct btrfs_free_space_header *header;
250 struct extent_buffer *leaf;
252 struct btrfs_key key;
253 struct list_head bitmaps;
260 INIT_LIST_HEAD(&bitmaps);
262 /* Nothing in the space cache, goodbye */
263 if (!i_size_read(inode))
266 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
270 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
274 btrfs_release_path(path);
281 leaf = path->nodes[0];
282 header = btrfs_item_ptr(leaf, path->slots[0],
283 struct btrfs_free_space_header);
284 num_entries = btrfs_free_space_entries(leaf, header);
285 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
286 generation = btrfs_free_space_generation(leaf, header);
287 btrfs_release_path(path);
289 if (BTRFS_I(inode)->generation != generation) {
290 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
291 " not match free space cache generation (%llu)\n",
292 (unsigned long long)BTRFS_I(inode)->generation,
293 (unsigned long long)generation);
300 ret = readahead_cache(inode);
305 struct btrfs_free_space_entry *entry;
306 struct btrfs_free_space *e;
308 unsigned long offset = 0;
311 if (!num_entries && !num_bitmaps)
314 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
318 if (!PageUptodate(page)) {
319 btrfs_readpage(NULL, page);
321 if (!PageUptodate(page)) {
323 page_cache_release(page);
324 printk(KERN_ERR "btrfs: error reading free "
335 * We put a bogus crc in the front of the first page in
336 * case old kernels try to mount a fs with the new
337 * format to make sure they discard the cache.
340 offset += sizeof(u64);
343 if (*gen != BTRFS_I(inode)->generation) {
344 printk(KERN_ERR "btrfs: space cache generation"
345 " (%llu) does not match inode (%llu)\n",
346 (unsigned long long)*gen,
348 BTRFS_I(inode)->generation);
351 page_cache_release(page);
355 offset += sizeof(u64);
364 e = kmem_cache_zalloc(btrfs_free_space_cachep,
369 page_cache_release(page);
373 e->offset = le64_to_cpu(entry->offset);
374 e->bytes = le64_to_cpu(entry->bytes);
377 kmem_cache_free(btrfs_free_space_cachep, e);
379 page_cache_release(page);
383 if (entry->type == BTRFS_FREE_SPACE_EXTENT) {
384 spin_lock(&ctl->tree_lock);
385 ret = link_free_space(ctl, e);
386 spin_unlock(&ctl->tree_lock);
388 printk(KERN_ERR "Duplicate entries in "
389 "free space cache, dumping\n");
392 page_cache_release(page);
396 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
400 btrfs_free_space_cachep, e);
402 page_cache_release(page);
405 spin_lock(&ctl->tree_lock);
406 ret = link_free_space(ctl, e);
407 ctl->total_bitmaps++;
408 ctl->op->recalc_thresholds(ctl);
409 spin_unlock(&ctl->tree_lock);
411 printk(KERN_ERR "Duplicate entries in "
412 "free space cache, dumping\n");
415 page_cache_release(page);
418 list_add_tail(&e->list, &bitmaps);
422 offset += sizeof(struct btrfs_free_space_entry);
423 if (offset + sizeof(struct btrfs_free_space_entry) >=
430 * We read an entry out of this page, we need to move on to the
439 * We add the bitmaps at the end of the entries in order that
440 * the bitmap entries are added to the cache.
442 e = list_entry(bitmaps.next, struct btrfs_free_space, list);
443 list_del_init(&e->list);
444 memcpy(e->bitmap, addr, PAGE_CACHE_SIZE);
449 page_cache_release(page);
457 __btrfs_remove_free_space_cache(ctl);
461 int load_free_space_cache(struct btrfs_fs_info *fs_info,
462 struct btrfs_block_group_cache *block_group)
464 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
465 struct btrfs_root *root = fs_info->tree_root;
467 struct btrfs_path *path;
470 u64 used = btrfs_block_group_used(&block_group->item);
473 * If we're unmounting then just return, since this does a search on the
474 * normal root and not the commit root and we could deadlock.
476 if (btrfs_fs_closing(fs_info))
480 * If this block group has been marked to be cleared for one reason or
481 * another then we can't trust the on disk cache, so just return.
483 spin_lock(&block_group->lock);
484 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
485 spin_unlock(&block_group->lock);
488 spin_unlock(&block_group->lock);
490 path = btrfs_alloc_path();
494 inode = lookup_free_space_inode(root, block_group, path);
496 btrfs_free_path(path);
500 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
501 path, block_group->key.objectid);
502 btrfs_free_path(path);
506 spin_lock(&ctl->tree_lock);
507 matched = (ctl->free_space == (block_group->key.offset - used -
508 block_group->bytes_super));
509 spin_unlock(&ctl->tree_lock);
512 __btrfs_remove_free_space_cache(ctl);
513 printk(KERN_ERR "block group %llu has an wrong amount of free "
514 "space\n", block_group->key.objectid);
519 /* This cache is bogus, make sure it gets cleared */
520 spin_lock(&block_group->lock);
521 block_group->disk_cache_state = BTRFS_DC_CLEAR;
522 spin_unlock(&block_group->lock);
525 printk(KERN_ERR "btrfs: failed to load free space cache "
526 "for block group %llu\n", block_group->key.objectid);
533 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
534 struct btrfs_free_space_ctl *ctl,
535 struct btrfs_block_group_cache *block_group,
536 struct btrfs_trans_handle *trans,
537 struct btrfs_path *path, u64 offset)
539 struct btrfs_free_space_header *header;
540 struct extent_buffer *leaf;
541 struct rb_node *node;
542 struct list_head *pos, *n;
545 struct extent_state *cached_state = NULL;
546 struct btrfs_free_cluster *cluster = NULL;
547 struct extent_io_tree *unpin = NULL;
548 struct list_head bitmap_list;
549 struct btrfs_key key;
553 int index = 0, num_pages = 0;
557 bool next_page = false;
558 bool out_of_space = false;
560 INIT_LIST_HEAD(&bitmap_list);
562 node = rb_first(&ctl->free_space_offset);
566 if (!i_size_read(inode))
569 num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
572 filemap_write_and_wait(inode->i_mapping);
573 btrfs_wait_ordered_range(inode, inode->i_size &
574 ~(root->sectorsize - 1), (u64)-1);
576 pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS);
580 /* Get the cluster for this block_group if it exists */
581 if (block_group && !list_empty(&block_group->cluster_list))
582 cluster = list_entry(block_group->cluster_list.next,
583 struct btrfs_free_cluster,
587 * We shouldn't have switched the pinned extents yet so this is the
590 unpin = root->fs_info->pinned_extents;
593 * Lock all pages first so we can lock the extent safely.
595 * NOTE: Because we hold the ref the entire time we're going to write to
596 * the page find_get_page should never fail, so we don't do a check
597 * after find_get_page at this point. Just putting this here so people
598 * know and don't freak out.
600 while (index < num_pages) {
601 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
605 for (i = 0; i < num_pages; i++) {
606 unlock_page(pages[i]);
607 page_cache_release(pages[i]);
616 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
617 0, &cached_state, GFP_NOFS);
620 * 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;
634 if (index >= num_pages) {
641 orig = addr = kmap(page);
646 * We're going to put in a bogus crc for this page to
647 * make sure that old kernels who aren't aware of this
648 * format will be sure to discard the cache.
651 offset += sizeof(u64);
654 *gen = trans->transid;
656 offset += sizeof(u64);
660 memset(addr, 0, PAGE_CACHE_SIZE - offset);
661 while (node && !next_page) {
662 struct btrfs_free_space *e;
664 e = rb_entry(node, struct btrfs_free_space, offset_index);
667 entry->offset = cpu_to_le64(e->offset);
668 entry->bytes = cpu_to_le64(e->bytes);
670 entry->type = BTRFS_FREE_SPACE_BITMAP;
671 list_add_tail(&e->list, &bitmap_list);
674 entry->type = BTRFS_FREE_SPACE_EXTENT;
676 node = rb_next(node);
677 if (!node && cluster) {
678 node = rb_first(&cluster->root);
681 offset += sizeof(struct btrfs_free_space_entry);
682 if (offset + sizeof(struct btrfs_free_space_entry) >=
689 * We want to add any pinned extents to our free space cache
690 * so we don't leak the space
692 while (block_group && !next_page &&
693 (start < block_group->key.objectid +
694 block_group->key.offset)) {
695 ret = find_first_extent_bit(unpin, start, &start, &end,
702 /* This pinned extent is out of our range */
703 if (start >= block_group->key.objectid +
704 block_group->key.offset)
707 len = block_group->key.objectid +
708 block_group->key.offset - start;
709 len = min(len, end + 1 - start);
712 entry->offset = cpu_to_le64(start);
713 entry->bytes = cpu_to_le64(len);
714 entry->type = BTRFS_FREE_SPACE_EXTENT;
717 offset += sizeof(struct btrfs_free_space_entry);
718 if (offset + sizeof(struct btrfs_free_space_entry) >=
724 /* Generate bogus crc value */
727 crc = btrfs_csum_data(root, orig + sizeof(u64), crc,
728 PAGE_CACHE_SIZE - sizeof(u64));
729 btrfs_csum_final(crc, (char *)&crc);
737 bytes += PAGE_CACHE_SIZE;
740 } while (node || next_page);
742 /* Write out the bitmaps */
743 list_for_each_safe(pos, n, &bitmap_list) {
745 struct btrfs_free_space *entry =
746 list_entry(pos, struct btrfs_free_space, list);
748 if (index >= num_pages) {
755 memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE);
757 bytes += PAGE_CACHE_SIZE;
759 list_del_init(&entry->list);
764 btrfs_drop_pages(pages, num_pages);
765 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
766 i_size_read(inode) - 1, &cached_state,
772 /* Zero out the rest of the pages just to make sure */
773 while (index < num_pages) {
778 memset(addr, 0, PAGE_CACHE_SIZE);
780 bytes += PAGE_CACHE_SIZE;
784 ret = btrfs_dirty_pages(root, inode, pages, num_pages, 0,
785 bytes, &cached_state);
786 btrfs_drop_pages(pages, num_pages);
787 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
788 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
795 BTRFS_I(inode)->generation = trans->transid;
797 filemap_write_and_wait(inode->i_mapping);
799 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
803 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
806 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
807 EXTENT_DIRTY | EXTENT_DELALLOC |
808 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
811 leaf = path->nodes[0];
813 struct btrfs_key found_key;
814 BUG_ON(!path->slots[0]);
816 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
817 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
818 found_key.offset != offset) {
820 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
821 EXTENT_DIRTY | EXTENT_DELALLOC |
822 EXTENT_DO_ACCOUNTING, 0, 0, NULL,
824 btrfs_release_path(path);
828 header = btrfs_item_ptr(leaf, path->slots[0],
829 struct btrfs_free_space_header);
830 btrfs_set_free_space_entries(leaf, header, entries);
831 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
832 btrfs_set_free_space_generation(leaf, header, trans->transid);
833 btrfs_mark_buffer_dirty(leaf);
834 btrfs_release_path(path);
841 invalidate_inode_pages2_range(inode->i_mapping, 0, index);
842 BTRFS_I(inode)->generation = 0;
844 btrfs_update_inode(trans, root, inode);
848 int btrfs_write_out_cache(struct btrfs_root *root,
849 struct btrfs_trans_handle *trans,
850 struct btrfs_block_group_cache *block_group,
851 struct btrfs_path *path)
853 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
857 root = root->fs_info->tree_root;
859 spin_lock(&block_group->lock);
860 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
861 spin_unlock(&block_group->lock);
864 spin_unlock(&block_group->lock);
866 inode = lookup_free_space_inode(root, block_group, path);
870 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
871 path, block_group->key.objectid);
873 spin_lock(&block_group->lock);
874 block_group->disk_cache_state = BTRFS_DC_ERROR;
875 spin_unlock(&block_group->lock);
878 printk(KERN_ERR "btrfs: failed to write free space cace "
879 "for block group %llu\n", block_group->key.objectid);
886 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
889 BUG_ON(offset < bitmap_start);
890 offset -= bitmap_start;
891 return (unsigned long)(div_u64(offset, unit));
894 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
896 return (unsigned long)(div_u64(bytes, unit));
899 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
903 u64 bytes_per_bitmap;
905 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
906 bitmap_start = offset - ctl->start;
907 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
908 bitmap_start *= bytes_per_bitmap;
909 bitmap_start += ctl->start;
914 static int tree_insert_offset(struct rb_root *root, u64 offset,
915 struct rb_node *node, int bitmap)
917 struct rb_node **p = &root->rb_node;
918 struct rb_node *parent = NULL;
919 struct btrfs_free_space *info;
923 info = rb_entry(parent, struct btrfs_free_space, offset_index);
925 if (offset < info->offset) {
927 } else if (offset > info->offset) {
931 * we could have a bitmap entry and an extent entry
932 * share the same offset. If this is the case, we want
933 * the extent entry to always be found first if we do a
934 * linear search through the tree, since we want to have
935 * the quickest allocation time, and allocating from an
936 * extent is faster than allocating from a bitmap. So
937 * if we're inserting a bitmap and we find an entry at
938 * this offset, we want to go right, or after this entry
939 * logically. If we are inserting an extent and we've
940 * found a bitmap, we want to go left, or before
959 rb_link_node(node, parent, p);
960 rb_insert_color(node, root);
966 * searches the tree for the given offset.
968 * fuzzy - If this is set, then we are trying to make an allocation, and we just
969 * want a section that has at least bytes size and comes at or after the given
972 static struct btrfs_free_space *
973 tree_search_offset(struct btrfs_free_space_ctl *ctl,
974 u64 offset, int bitmap_only, int fuzzy)
976 struct rb_node *n = ctl->free_space_offset.rb_node;
977 struct btrfs_free_space *entry, *prev = NULL;
979 /* find entry that is closest to the 'offset' */
986 entry = rb_entry(n, struct btrfs_free_space, offset_index);
989 if (offset < entry->offset)
991 else if (offset > entry->offset)
1004 * bitmap entry and extent entry may share same offset,
1005 * in that case, bitmap entry comes after extent entry.
1010 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1011 if (entry->offset != offset)
1014 WARN_ON(!entry->bitmap);
1017 if (entry->bitmap) {
1019 * if previous extent entry covers the offset,
1020 * we should return it instead of the bitmap entry
1022 n = &entry->offset_index;
1027 prev = rb_entry(n, struct btrfs_free_space,
1029 if (!prev->bitmap) {
1030 if (prev->offset + prev->bytes > offset)
1042 /* find last entry before the 'offset' */
1044 if (entry->offset > offset) {
1045 n = rb_prev(&entry->offset_index);
1047 entry = rb_entry(n, struct btrfs_free_space,
1049 BUG_ON(entry->offset > offset);
1058 if (entry->bitmap) {
1059 n = &entry->offset_index;
1064 prev = rb_entry(n, struct btrfs_free_space,
1066 if (!prev->bitmap) {
1067 if (prev->offset + prev->bytes > offset)
1072 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1074 } else if (entry->offset + entry->bytes > offset)
1081 if (entry->bitmap) {
1082 if (entry->offset + BITS_PER_BITMAP *
1086 if (entry->offset + entry->bytes > offset)
1090 n = rb_next(&entry->offset_index);
1093 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1099 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1100 struct btrfs_free_space *info)
1102 rb_erase(&info->offset_index, &ctl->free_space_offset);
1103 ctl->free_extents--;
1106 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1107 struct btrfs_free_space *info)
1109 __unlink_free_space(ctl, info);
1110 ctl->free_space -= info->bytes;
1113 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1114 struct btrfs_free_space *info)
1118 BUG_ON(!info->bitmap && !info->bytes);
1119 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1120 &info->offset_index, (info->bitmap != NULL));
1124 ctl->free_space += info->bytes;
1125 ctl->free_extents++;
1129 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1131 struct btrfs_block_group_cache *block_group = ctl->private;
1135 u64 size = block_group->key.offset;
1136 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1137 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1139 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1142 * The goal is to keep the total amount of memory used per 1gb of space
1143 * at or below 32k, so we need to adjust how much memory we allow to be
1144 * used by extent based free space tracking
1146 if (size < 1024 * 1024 * 1024)
1147 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1149 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1150 div64_u64(size, 1024 * 1024 * 1024);
1153 * we want to account for 1 more bitmap than what we have so we can make
1154 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1155 * we add more bitmaps.
1157 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1159 if (bitmap_bytes >= max_bytes) {
1160 ctl->extents_thresh = 0;
1165 * we want the extent entry threshold to always be at most 1/2 the maxw
1166 * bytes we can have, or whatever is less than that.
1168 extent_bytes = max_bytes - bitmap_bytes;
1169 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1171 ctl->extents_thresh =
1172 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1175 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1176 struct btrfs_free_space *info,
1177 u64 offset, u64 bytes)
1179 unsigned long start, count;
1181 start = offset_to_bit(info->offset, ctl->unit, offset);
1182 count = bytes_to_bits(bytes, ctl->unit);
1183 BUG_ON(start + count > BITS_PER_BITMAP);
1185 bitmap_clear(info->bitmap, start, count);
1187 info->bytes -= bytes;
1190 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1191 struct btrfs_free_space *info, u64 offset,
1194 __bitmap_clear_bits(ctl, info, offset, bytes);
1195 ctl->free_space -= bytes;
1198 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1199 struct btrfs_free_space *info, u64 offset,
1202 unsigned long start, count;
1204 start = offset_to_bit(info->offset, ctl->unit, offset);
1205 count = bytes_to_bits(bytes, ctl->unit);
1206 BUG_ON(start + count > BITS_PER_BITMAP);
1208 bitmap_set(info->bitmap, start, count);
1210 info->bytes += bytes;
1211 ctl->free_space += bytes;
1214 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1215 struct btrfs_free_space *bitmap_info, u64 *offset,
1218 unsigned long found_bits = 0;
1219 unsigned long bits, i;
1220 unsigned long next_zero;
1222 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1223 max_t(u64, *offset, bitmap_info->offset));
1224 bits = bytes_to_bits(*bytes, ctl->unit);
1226 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1227 i < BITS_PER_BITMAP;
1228 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1229 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1230 BITS_PER_BITMAP, i);
1231 if ((next_zero - i) >= bits) {
1232 found_bits = next_zero - i;
1239 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1240 *bytes = (u64)(found_bits) * ctl->unit;
1247 static struct btrfs_free_space *
1248 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1250 struct btrfs_free_space *entry;
1251 struct rb_node *node;
1254 if (!ctl->free_space_offset.rb_node)
1257 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1261 for (node = &entry->offset_index; node; node = rb_next(node)) {
1262 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1263 if (entry->bytes < *bytes)
1266 if (entry->bitmap) {
1267 ret = search_bitmap(ctl, entry, offset, bytes);
1273 *offset = entry->offset;
1274 *bytes = entry->bytes;
1281 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1282 struct btrfs_free_space *info, u64 offset)
1284 info->offset = offset_to_bitmap(ctl, offset);
1286 link_free_space(ctl, info);
1287 ctl->total_bitmaps++;
1289 ctl->op->recalc_thresholds(ctl);
1292 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1293 struct btrfs_free_space *bitmap_info)
1295 unlink_free_space(ctl, bitmap_info);
1296 kfree(bitmap_info->bitmap);
1297 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1298 ctl->total_bitmaps--;
1299 ctl->op->recalc_thresholds(ctl);
1302 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1303 struct btrfs_free_space *bitmap_info,
1304 u64 *offset, u64 *bytes)
1307 u64 search_start, search_bytes;
1311 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1314 * XXX - this can go away after a few releases.
1316 * since the only user of btrfs_remove_free_space is the tree logging
1317 * stuff, and the only way to test that is under crash conditions, we
1318 * want to have this debug stuff here just in case somethings not
1319 * working. Search the bitmap for the space we are trying to use to
1320 * make sure its actually there. If its not there then we need to stop
1321 * because something has gone wrong.
1323 search_start = *offset;
1324 search_bytes = *bytes;
1325 search_bytes = min(search_bytes, end - search_start + 1);
1326 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1327 BUG_ON(ret < 0 || search_start != *offset);
1329 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1330 bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
1331 *bytes -= end - *offset + 1;
1333 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1334 bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
1339 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1340 if (!bitmap_info->bytes)
1341 free_bitmap(ctl, bitmap_info);
1344 * no entry after this bitmap, but we still have bytes to
1345 * remove, so something has gone wrong.
1350 bitmap_info = rb_entry(next, struct btrfs_free_space,
1354 * if the next entry isn't a bitmap we need to return to let the
1355 * extent stuff do its work.
1357 if (!bitmap_info->bitmap)
1361 * Ok the next item is a bitmap, but it may not actually hold
1362 * the information for the rest of this free space stuff, so
1363 * look for it, and if we don't find it return so we can try
1364 * everything over again.
1366 search_start = *offset;
1367 search_bytes = *bytes;
1368 ret = search_bitmap(ctl, bitmap_info, &search_start,
1370 if (ret < 0 || search_start != *offset)
1374 } else if (!bitmap_info->bytes)
1375 free_bitmap(ctl, bitmap_info);
1380 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1381 struct btrfs_free_space *info, u64 offset,
1384 u64 bytes_to_set = 0;
1387 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1389 bytes_to_set = min(end - offset, bytes);
1391 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1393 return bytes_to_set;
1397 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1398 struct btrfs_free_space *info)
1400 struct btrfs_block_group_cache *block_group = ctl->private;
1403 * If we are below the extents threshold then we can add this as an
1404 * extent, and don't have to deal with the bitmap
1406 if (ctl->free_extents < ctl->extents_thresh) {
1408 * If this block group has some small extents we don't want to
1409 * use up all of our free slots in the cache with them, we want
1410 * to reserve them to larger extents, however if we have plent
1411 * of cache left then go ahead an dadd them, no sense in adding
1412 * the overhead of a bitmap if we don't have to.
1414 if (info->bytes <= block_group->sectorsize * 4) {
1415 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1423 * some block groups are so tiny they can't be enveloped by a bitmap, so
1424 * don't even bother to create a bitmap for this
1426 if (BITS_PER_BITMAP * block_group->sectorsize >
1427 block_group->key.offset)
1433 static struct btrfs_free_space_op free_space_op = {
1434 .recalc_thresholds = recalculate_thresholds,
1435 .use_bitmap = use_bitmap,
1438 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1439 struct btrfs_free_space *info)
1441 struct btrfs_free_space *bitmap_info;
1442 struct btrfs_block_group_cache *block_group = NULL;
1444 u64 bytes, offset, bytes_added;
1447 bytes = info->bytes;
1448 offset = info->offset;
1450 if (!ctl->op->use_bitmap(ctl, info))
1453 if (ctl->op == &free_space_op)
1454 block_group = ctl->private;
1457 * Since we link bitmaps right into the cluster we need to see if we
1458 * have a cluster here, and if so and it has our bitmap we need to add
1459 * the free space to that bitmap.
1461 if (block_group && !list_empty(&block_group->cluster_list)) {
1462 struct btrfs_free_cluster *cluster;
1463 struct rb_node *node;
1464 struct btrfs_free_space *entry;
1466 cluster = list_entry(block_group->cluster_list.next,
1467 struct btrfs_free_cluster,
1469 spin_lock(&cluster->lock);
1470 node = rb_first(&cluster->root);
1472 spin_unlock(&cluster->lock);
1473 goto no_cluster_bitmap;
1476 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1477 if (!entry->bitmap) {
1478 spin_unlock(&cluster->lock);
1479 goto no_cluster_bitmap;
1482 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1483 bytes_added = add_bytes_to_bitmap(ctl, entry,
1485 bytes -= bytes_added;
1486 offset += bytes_added;
1488 spin_unlock(&cluster->lock);
1496 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1503 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1504 bytes -= bytes_added;
1505 offset += bytes_added;
1515 if (info && info->bitmap) {
1516 add_new_bitmap(ctl, info, offset);
1521 spin_unlock(&ctl->tree_lock);
1523 /* no pre-allocated info, allocate a new one */
1525 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1528 spin_lock(&ctl->tree_lock);
1534 /* allocate the bitmap */
1535 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1536 spin_lock(&ctl->tree_lock);
1537 if (!info->bitmap) {
1547 kfree(info->bitmap);
1548 kmem_cache_free(btrfs_free_space_cachep, info);
1554 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1555 struct btrfs_free_space *info, bool update_stat)
1557 struct btrfs_free_space *left_info;
1558 struct btrfs_free_space *right_info;
1559 bool merged = false;
1560 u64 offset = info->offset;
1561 u64 bytes = info->bytes;
1564 * first we want to see if there is free space adjacent to the range we
1565 * are adding, if there is remove that struct and add a new one to
1566 * cover the entire range
1568 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1569 if (right_info && rb_prev(&right_info->offset_index))
1570 left_info = rb_entry(rb_prev(&right_info->offset_index),
1571 struct btrfs_free_space, offset_index);
1573 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1575 if (right_info && !right_info->bitmap) {
1577 unlink_free_space(ctl, right_info);
1579 __unlink_free_space(ctl, right_info);
1580 info->bytes += right_info->bytes;
1581 kmem_cache_free(btrfs_free_space_cachep, right_info);
1585 if (left_info && !left_info->bitmap &&
1586 left_info->offset + left_info->bytes == offset) {
1588 unlink_free_space(ctl, left_info);
1590 __unlink_free_space(ctl, left_info);
1591 info->offset = left_info->offset;
1592 info->bytes += left_info->bytes;
1593 kmem_cache_free(btrfs_free_space_cachep, left_info);
1600 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1601 u64 offset, u64 bytes)
1603 struct btrfs_free_space *info;
1606 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1610 info->offset = offset;
1611 info->bytes = bytes;
1613 spin_lock(&ctl->tree_lock);
1615 if (try_merge_free_space(ctl, info, true))
1619 * There was no extent directly to the left or right of this new
1620 * extent then we know we're going to have to allocate a new extent, so
1621 * before we do that see if we need to drop this into a bitmap
1623 ret = insert_into_bitmap(ctl, info);
1631 ret = link_free_space(ctl, info);
1633 kmem_cache_free(btrfs_free_space_cachep, info);
1635 spin_unlock(&ctl->tree_lock);
1638 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1639 BUG_ON(ret == -EEXIST);
1645 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1646 u64 offset, u64 bytes)
1648 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1649 struct btrfs_free_space *info;
1650 struct btrfs_free_space *next_info = NULL;
1653 spin_lock(&ctl->tree_lock);
1656 info = tree_search_offset(ctl, offset, 0, 0);
1659 * oops didn't find an extent that matched the space we wanted
1660 * to remove, look for a bitmap instead
1662 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1670 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1672 next_info = rb_entry(rb_next(&info->offset_index),
1673 struct btrfs_free_space,
1676 if (next_info->bitmap)
1677 end = next_info->offset +
1678 BITS_PER_BITMAP * ctl->unit - 1;
1680 end = next_info->offset + next_info->bytes;
1682 if (next_info->bytes < bytes ||
1683 next_info->offset > offset || offset > end) {
1684 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1685 " trying to use %llu\n",
1686 (unsigned long long)info->offset,
1687 (unsigned long long)info->bytes,
1688 (unsigned long long)bytes);
1697 if (info->bytes == bytes) {
1698 unlink_free_space(ctl, info);
1700 kfree(info->bitmap);
1701 ctl->total_bitmaps--;
1703 kmem_cache_free(btrfs_free_space_cachep, info);
1707 if (!info->bitmap && info->offset == offset) {
1708 unlink_free_space(ctl, info);
1709 info->offset += bytes;
1710 info->bytes -= bytes;
1711 link_free_space(ctl, info);
1715 if (!info->bitmap && info->offset <= offset &&
1716 info->offset + info->bytes >= offset + bytes) {
1717 u64 old_start = info->offset;
1719 * we're freeing space in the middle of the info,
1720 * this can happen during tree log replay
1722 * first unlink the old info and then
1723 * insert it again after the hole we're creating
1725 unlink_free_space(ctl, info);
1726 if (offset + bytes < info->offset + info->bytes) {
1727 u64 old_end = info->offset + info->bytes;
1729 info->offset = offset + bytes;
1730 info->bytes = old_end - info->offset;
1731 ret = link_free_space(ctl, info);
1736 /* the hole we're creating ends at the end
1737 * of the info struct, just free the info
1739 kmem_cache_free(btrfs_free_space_cachep, info);
1741 spin_unlock(&ctl->tree_lock);
1743 /* step two, insert a new info struct to cover
1744 * anything before the hole
1746 ret = btrfs_add_free_space(block_group, old_start,
1747 offset - old_start);
1752 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1757 spin_unlock(&ctl->tree_lock);
1762 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1765 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1766 struct btrfs_free_space *info;
1770 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1771 info = rb_entry(n, struct btrfs_free_space, offset_index);
1772 if (info->bytes >= bytes)
1774 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1775 (unsigned long long)info->offset,
1776 (unsigned long long)info->bytes,
1777 (info->bitmap) ? "yes" : "no");
1779 printk(KERN_INFO "block group has cluster?: %s\n",
1780 list_empty(&block_group->cluster_list) ? "no" : "yes");
1781 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1785 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1787 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1789 spin_lock_init(&ctl->tree_lock);
1790 ctl->unit = block_group->sectorsize;
1791 ctl->start = block_group->key.objectid;
1792 ctl->private = block_group;
1793 ctl->op = &free_space_op;
1796 * we only want to have 32k of ram per block group for keeping
1797 * track of free space, and if we pass 1/2 of that we want to
1798 * start converting things over to using bitmaps
1800 ctl->extents_thresh = ((1024 * 32) / 2) /
1801 sizeof(struct btrfs_free_space);
1805 * for a given cluster, put all of its extents back into the free
1806 * space cache. If the block group passed doesn't match the block group
1807 * pointed to by the cluster, someone else raced in and freed the
1808 * cluster already. In that case, we just return without changing anything
1811 __btrfs_return_cluster_to_free_space(
1812 struct btrfs_block_group_cache *block_group,
1813 struct btrfs_free_cluster *cluster)
1815 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1816 struct btrfs_free_space *entry;
1817 struct rb_node *node;
1819 spin_lock(&cluster->lock);
1820 if (cluster->block_group != block_group)
1823 cluster->block_group = NULL;
1824 cluster->window_start = 0;
1825 list_del_init(&cluster->block_group_list);
1827 node = rb_first(&cluster->root);
1831 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1832 node = rb_next(&entry->offset_index);
1833 rb_erase(&entry->offset_index, &cluster->root);
1835 bitmap = (entry->bitmap != NULL);
1837 try_merge_free_space(ctl, entry, false);
1838 tree_insert_offset(&ctl->free_space_offset,
1839 entry->offset, &entry->offset_index, bitmap);
1841 cluster->root = RB_ROOT;
1844 spin_unlock(&cluster->lock);
1845 btrfs_put_block_group(block_group);
1849 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
1851 struct btrfs_free_space *info;
1852 struct rb_node *node;
1854 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
1855 info = rb_entry(node, struct btrfs_free_space, offset_index);
1856 if (!info->bitmap) {
1857 unlink_free_space(ctl, info);
1858 kmem_cache_free(btrfs_free_space_cachep, info);
1860 free_bitmap(ctl, info);
1862 if (need_resched()) {
1863 spin_unlock(&ctl->tree_lock);
1865 spin_lock(&ctl->tree_lock);
1870 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
1872 spin_lock(&ctl->tree_lock);
1873 __btrfs_remove_free_space_cache_locked(ctl);
1874 spin_unlock(&ctl->tree_lock);
1877 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
1879 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1880 struct btrfs_free_cluster *cluster;
1881 struct list_head *head;
1883 spin_lock(&ctl->tree_lock);
1884 while ((head = block_group->cluster_list.next) !=
1885 &block_group->cluster_list) {
1886 cluster = list_entry(head, struct btrfs_free_cluster,
1889 WARN_ON(cluster->block_group != block_group);
1890 __btrfs_return_cluster_to_free_space(block_group, cluster);
1891 if (need_resched()) {
1892 spin_unlock(&ctl->tree_lock);
1894 spin_lock(&ctl->tree_lock);
1897 __btrfs_remove_free_space_cache_locked(ctl);
1898 spin_unlock(&ctl->tree_lock);
1902 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
1903 u64 offset, u64 bytes, u64 empty_size)
1905 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1906 struct btrfs_free_space *entry = NULL;
1907 u64 bytes_search = bytes + empty_size;
1910 spin_lock(&ctl->tree_lock);
1911 entry = find_free_space(ctl, &offset, &bytes_search);
1916 if (entry->bitmap) {
1917 bitmap_clear_bits(ctl, entry, offset, bytes);
1919 free_bitmap(ctl, entry);
1921 unlink_free_space(ctl, entry);
1922 entry->offset += bytes;
1923 entry->bytes -= bytes;
1925 kmem_cache_free(btrfs_free_space_cachep, entry);
1927 link_free_space(ctl, entry);
1931 spin_unlock(&ctl->tree_lock);
1937 * given a cluster, put all of its extents back into the free space
1938 * cache. If a block group is passed, this function will only free
1939 * a cluster that belongs to the passed block group.
1941 * Otherwise, it'll get a reference on the block group pointed to by the
1942 * cluster and remove the cluster from it.
1944 int btrfs_return_cluster_to_free_space(
1945 struct btrfs_block_group_cache *block_group,
1946 struct btrfs_free_cluster *cluster)
1948 struct btrfs_free_space_ctl *ctl;
1951 /* first, get a safe pointer to the block group */
1952 spin_lock(&cluster->lock);
1954 block_group = cluster->block_group;
1956 spin_unlock(&cluster->lock);
1959 } else if (cluster->block_group != block_group) {
1960 /* someone else has already freed it don't redo their work */
1961 spin_unlock(&cluster->lock);
1964 atomic_inc(&block_group->count);
1965 spin_unlock(&cluster->lock);
1967 ctl = block_group->free_space_ctl;
1969 /* now return any extents the cluster had on it */
1970 spin_lock(&ctl->tree_lock);
1971 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
1972 spin_unlock(&ctl->tree_lock);
1974 /* finally drop our ref */
1975 btrfs_put_block_group(block_group);
1979 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
1980 struct btrfs_free_cluster *cluster,
1981 struct btrfs_free_space *entry,
1982 u64 bytes, u64 min_start)
1984 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1986 u64 search_start = cluster->window_start;
1987 u64 search_bytes = bytes;
1990 search_start = min_start;
1991 search_bytes = bytes;
1993 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
1998 __bitmap_clear_bits(ctl, entry, ret, bytes);
2004 * given a cluster, try to allocate 'bytes' from it, returns 0
2005 * if it couldn't find anything suitably large, or a logical disk offset
2006 * if things worked out
2008 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2009 struct btrfs_free_cluster *cluster, u64 bytes,
2012 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2013 struct btrfs_free_space *entry = NULL;
2014 struct rb_node *node;
2017 spin_lock(&cluster->lock);
2018 if (bytes > cluster->max_size)
2021 if (cluster->block_group != block_group)
2024 node = rb_first(&cluster->root);
2028 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2030 if (entry->bytes < bytes ||
2031 (!entry->bitmap && entry->offset < min_start)) {
2032 node = rb_next(&entry->offset_index);
2035 entry = rb_entry(node, struct btrfs_free_space,
2040 if (entry->bitmap) {
2041 ret = btrfs_alloc_from_bitmap(block_group,
2042 cluster, entry, bytes,
2045 node = rb_next(&entry->offset_index);
2048 entry = rb_entry(node, struct btrfs_free_space,
2053 ret = entry->offset;
2055 entry->offset += bytes;
2056 entry->bytes -= bytes;
2059 if (entry->bytes == 0)
2060 rb_erase(&entry->offset_index, &cluster->root);
2064 spin_unlock(&cluster->lock);
2069 spin_lock(&ctl->tree_lock);
2071 ctl->free_space -= bytes;
2072 if (entry->bytes == 0) {
2073 ctl->free_extents--;
2074 if (entry->bitmap) {
2075 kfree(entry->bitmap);
2076 ctl->total_bitmaps--;
2077 ctl->op->recalc_thresholds(ctl);
2079 kmem_cache_free(btrfs_free_space_cachep, entry);
2082 spin_unlock(&ctl->tree_lock);
2087 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2088 struct btrfs_free_space *entry,
2089 struct btrfs_free_cluster *cluster,
2090 u64 offset, u64 bytes, u64 min_bytes)
2092 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2093 unsigned long next_zero;
2095 unsigned long search_bits;
2096 unsigned long total_bits;
2097 unsigned long found_bits;
2098 unsigned long start = 0;
2099 unsigned long total_found = 0;
2103 i = offset_to_bit(entry->offset, block_group->sectorsize,
2104 max_t(u64, offset, entry->offset));
2105 search_bits = bytes_to_bits(bytes, block_group->sectorsize);
2106 total_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2110 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
2111 i < BITS_PER_BITMAP;
2112 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
2113 next_zero = find_next_zero_bit(entry->bitmap,
2114 BITS_PER_BITMAP, i);
2115 if (next_zero - i >= search_bits) {
2116 found_bits = next_zero - i;
2130 total_found += found_bits;
2132 if (cluster->max_size < found_bits * block_group->sectorsize)
2133 cluster->max_size = found_bits * block_group->sectorsize;
2135 if (total_found < total_bits) {
2136 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
2137 if (i - start > total_bits * 2) {
2139 cluster->max_size = 0;
2145 cluster->window_start = start * block_group->sectorsize +
2147 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2148 ret = tree_insert_offset(&cluster->root, entry->offset,
2149 &entry->offset_index, 1);
2156 * This searches the block group for just extents to fill the cluster with.
2159 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2160 struct btrfs_free_cluster *cluster,
2161 struct list_head *bitmaps, u64 offset, u64 bytes,
2164 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2165 struct btrfs_free_space *first = NULL;
2166 struct btrfs_free_space *entry = NULL;
2167 struct btrfs_free_space *prev = NULL;
2168 struct btrfs_free_space *last;
2169 struct rb_node *node;
2173 u64 max_gap = 128 * 1024;
2175 entry = tree_search_offset(ctl, offset, 0, 1);
2180 * We don't want bitmaps, so just move along until we find a normal
2183 while (entry->bitmap) {
2184 if (list_empty(&entry->list))
2185 list_add_tail(&entry->list, bitmaps);
2186 node = rb_next(&entry->offset_index);
2189 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2192 window_start = entry->offset;
2193 window_free = entry->bytes;
2194 max_extent = entry->bytes;
2199 while (window_free <= min_bytes) {
2200 node = rb_next(&entry->offset_index);
2203 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2205 if (entry->bitmap) {
2206 if (list_empty(&entry->list))
2207 list_add_tail(&entry->list, bitmaps);
2212 * we haven't filled the empty size and the window is
2213 * very large. reset and try again
2215 if (entry->offset - (prev->offset + prev->bytes) > max_gap ||
2216 entry->offset - window_start > (min_bytes * 2)) {
2218 window_start = entry->offset;
2219 window_free = entry->bytes;
2221 max_extent = entry->bytes;
2224 window_free += entry->bytes;
2225 if (entry->bytes > max_extent)
2226 max_extent = entry->bytes;
2231 cluster->window_start = first->offset;
2233 node = &first->offset_index;
2236 * now we've found our entries, pull them out of the free space
2237 * cache and put them into the cluster rbtree
2242 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2243 node = rb_next(&entry->offset_index);
2247 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2248 ret = tree_insert_offset(&cluster->root, entry->offset,
2249 &entry->offset_index, 0);
2251 } while (node && entry != last);
2253 cluster->max_size = max_extent;
2259 * This specifically looks for bitmaps that may work in the cluster, we assume
2260 * that we have already failed to find extents that will work.
2263 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2264 struct btrfs_free_cluster *cluster,
2265 struct list_head *bitmaps, u64 offset, u64 bytes,
2268 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2269 struct btrfs_free_space *entry;
2270 struct rb_node *node;
2273 if (ctl->total_bitmaps == 0)
2277 * First check our cached list of bitmaps and see if there is an entry
2278 * here that will work.
2280 list_for_each_entry(entry, bitmaps, list) {
2281 if (entry->bytes < min_bytes)
2283 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2290 * If we do have entries on our list and we are here then we didn't find
2291 * anything, so go ahead and get the next entry after the last entry in
2292 * this list and start the search from there.
2294 if (!list_empty(bitmaps)) {
2295 entry = list_entry(bitmaps->prev, struct btrfs_free_space,
2297 node = rb_next(&entry->offset_index);
2300 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2304 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 0, 1);
2309 node = &entry->offset_index;
2311 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2312 node = rb_next(&entry->offset_index);
2315 if (entry->bytes < min_bytes)
2317 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2319 } while (ret && node);
2325 * here we try to find a cluster of blocks in a block group. The goal
2326 * is to find at least bytes free and up to empty_size + bytes free.
2327 * We might not find them all in one contiguous area.
2329 * returns zero and sets up cluster if things worked out, otherwise
2330 * it returns -enospc
2332 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2333 struct btrfs_root *root,
2334 struct btrfs_block_group_cache *block_group,
2335 struct btrfs_free_cluster *cluster,
2336 u64 offset, u64 bytes, u64 empty_size)
2338 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2339 struct list_head bitmaps;
2340 struct btrfs_free_space *entry, *tmp;
2344 /* for metadata, allow allocates with more holes */
2345 if (btrfs_test_opt(root, SSD_SPREAD)) {
2346 min_bytes = bytes + empty_size;
2347 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2349 * we want to do larger allocations when we are
2350 * flushing out the delayed refs, it helps prevent
2351 * making more work as we go along.
2353 if (trans->transaction->delayed_refs.flushing)
2354 min_bytes = max(bytes, (bytes + empty_size) >> 1);
2356 min_bytes = max(bytes, (bytes + empty_size) >> 4);
2358 min_bytes = max(bytes, (bytes + empty_size) >> 2);
2360 spin_lock(&ctl->tree_lock);
2363 * If we know we don't have enough space to make a cluster don't even
2364 * bother doing all the work to try and find one.
2366 if (ctl->free_space < min_bytes) {
2367 spin_unlock(&ctl->tree_lock);
2371 spin_lock(&cluster->lock);
2373 /* someone already found a cluster, hooray */
2374 if (cluster->block_group) {
2379 INIT_LIST_HEAD(&bitmaps);
2380 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2383 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2384 offset, bytes, min_bytes);
2386 /* Clear our temporary list */
2387 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2388 list_del_init(&entry->list);
2391 atomic_inc(&block_group->count);
2392 list_add_tail(&cluster->block_group_list,
2393 &block_group->cluster_list);
2394 cluster->block_group = block_group;
2397 spin_unlock(&cluster->lock);
2398 spin_unlock(&ctl->tree_lock);
2404 * simple code to zero out a cluster
2406 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2408 spin_lock_init(&cluster->lock);
2409 spin_lock_init(&cluster->refill_lock);
2410 cluster->root = RB_ROOT;
2411 cluster->max_size = 0;
2412 INIT_LIST_HEAD(&cluster->block_group_list);
2413 cluster->block_group = NULL;
2416 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2417 u64 *trimmed, u64 start, u64 end, u64 minlen)
2419 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2420 struct btrfs_free_space *entry = NULL;
2421 struct btrfs_fs_info *fs_info = block_group->fs_info;
2423 u64 actually_trimmed;
2428 while (start < end) {
2429 spin_lock(&ctl->tree_lock);
2431 if (ctl->free_space < minlen) {
2432 spin_unlock(&ctl->tree_lock);
2436 entry = tree_search_offset(ctl, start, 0, 1);
2438 entry = tree_search_offset(ctl,
2439 offset_to_bitmap(ctl, start),
2442 if (!entry || entry->offset >= end) {
2443 spin_unlock(&ctl->tree_lock);
2447 if (entry->bitmap) {
2448 ret = search_bitmap(ctl, entry, &start, &bytes);
2451 spin_unlock(&ctl->tree_lock);
2454 bytes = min(bytes, end - start);
2455 bitmap_clear_bits(ctl, entry, start, bytes);
2456 if (entry->bytes == 0)
2457 free_bitmap(ctl, entry);
2459 start = entry->offset + BITS_PER_BITMAP *
2460 block_group->sectorsize;
2461 spin_unlock(&ctl->tree_lock);
2466 start = entry->offset;
2467 bytes = min(entry->bytes, end - start);
2468 unlink_free_space(ctl, entry);
2469 kmem_cache_free(btrfs_free_space_cachep, entry);
2472 spin_unlock(&ctl->tree_lock);
2474 if (bytes >= minlen) {
2476 update_ret = btrfs_update_reserved_bytes(block_group,
2479 ret = btrfs_error_discard_extent(fs_info->extent_root,
2484 btrfs_add_free_space(block_group, start, bytes);
2486 btrfs_update_reserved_bytes(block_group,
2491 *trimmed += actually_trimmed;
2496 if (fatal_signal_pending(current)) {
2508 * Find the left-most item in the cache tree, and then return the
2509 * smallest inode number in the item.
2511 * Note: the returned inode number may not be the smallest one in
2512 * the tree, if the left-most item is a bitmap.
2514 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2516 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2517 struct btrfs_free_space *entry = NULL;
2520 spin_lock(&ctl->tree_lock);
2522 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2525 entry = rb_entry(rb_first(&ctl->free_space_offset),
2526 struct btrfs_free_space, offset_index);
2528 if (!entry->bitmap) {
2529 ino = entry->offset;
2531 unlink_free_space(ctl, entry);
2535 kmem_cache_free(btrfs_free_space_cachep, entry);
2537 link_free_space(ctl, entry);
2543 ret = search_bitmap(ctl, entry, &offset, &count);
2547 bitmap_clear_bits(ctl, entry, offset, 1);
2548 if (entry->bytes == 0)
2549 free_bitmap(ctl, entry);
2552 spin_unlock(&ctl->tree_lock);
2557 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2558 struct btrfs_path *path)
2560 struct inode *inode = NULL;
2562 spin_lock(&root->cache_lock);
2563 if (root->cache_inode)
2564 inode = igrab(root->cache_inode);
2565 spin_unlock(&root->cache_lock);
2569 inode = __lookup_free_space_inode(root, path, 0);
2573 spin_lock(&root->cache_lock);
2574 if (!btrfs_fs_closing(root->fs_info))
2575 root->cache_inode = igrab(inode);
2576 spin_unlock(&root->cache_lock);
2581 int create_free_ino_inode(struct btrfs_root *root,
2582 struct btrfs_trans_handle *trans,
2583 struct btrfs_path *path)
2585 return __create_free_space_inode(root, trans, path,
2586 BTRFS_FREE_INO_OBJECTID, 0);
2589 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2591 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2592 struct btrfs_path *path;
2593 struct inode *inode;
2595 u64 root_gen = btrfs_root_generation(&root->root_item);
2597 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2601 * If we're unmounting then just return, since this does a search on the
2602 * normal root and not the commit root and we could deadlock.
2604 if (btrfs_fs_closing(fs_info))
2607 path = btrfs_alloc_path();
2611 inode = lookup_free_ino_inode(root, path);
2615 if (root_gen != BTRFS_I(inode)->generation)
2618 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2621 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2622 "root %llu\n", root->root_key.objectid);
2626 btrfs_free_path(path);
2630 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2631 struct btrfs_trans_handle *trans,
2632 struct btrfs_path *path)
2634 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2635 struct inode *inode;
2638 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2641 inode = lookup_free_ino_inode(root, path);
2645 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2647 printk(KERN_ERR "btrfs: failed to write free ino cache "
2648 "for root %llu\n", root->root_key.objectid);