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 (!root->fs_info->closing) {
102 block_group->inode = igrab(inode);
103 block_group->iref = 1;
105 spin_unlock(&block_group->lock);
110 int __create_free_space_inode(struct btrfs_root *root,
111 struct btrfs_trans_handle *trans,
112 struct btrfs_path *path, u64 ino, u64 offset)
114 struct btrfs_key key;
115 struct btrfs_disk_key disk_key;
116 struct btrfs_free_space_header *header;
117 struct btrfs_inode_item *inode_item;
118 struct extent_buffer *leaf;
121 ret = btrfs_insert_empty_inode(trans, root, path, ino);
125 leaf = path->nodes[0];
126 inode_item = btrfs_item_ptr(leaf, path->slots[0],
127 struct btrfs_inode_item);
128 btrfs_item_key(leaf, &disk_key, path->slots[0]);
129 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
130 sizeof(*inode_item));
131 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
132 btrfs_set_inode_size(leaf, inode_item, 0);
133 btrfs_set_inode_nbytes(leaf, inode_item, 0);
134 btrfs_set_inode_uid(leaf, inode_item, 0);
135 btrfs_set_inode_gid(leaf, inode_item, 0);
136 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
137 btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS |
138 BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM);
139 btrfs_set_inode_nlink(leaf, inode_item, 1);
140 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
141 btrfs_set_inode_block_group(leaf, inode_item, offset);
142 btrfs_mark_buffer_dirty(leaf);
143 btrfs_release_path(path);
145 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
149 ret = btrfs_insert_empty_item(trans, root, path, &key,
150 sizeof(struct btrfs_free_space_header));
152 btrfs_release_path(path);
155 leaf = path->nodes[0];
156 header = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_free_space_header);
158 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
159 btrfs_set_free_space_key(leaf, header, &disk_key);
160 btrfs_mark_buffer_dirty(leaf);
161 btrfs_release_path(path);
166 int create_free_space_inode(struct btrfs_root *root,
167 struct btrfs_trans_handle *trans,
168 struct btrfs_block_group_cache *block_group,
169 struct btrfs_path *path)
174 ret = btrfs_find_free_objectid(root, &ino);
178 return __create_free_space_inode(root, trans, path, ino,
179 block_group->key.objectid);
182 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
183 struct btrfs_trans_handle *trans,
184 struct btrfs_path *path,
190 trans->block_rsv = root->orphan_block_rsv;
191 ret = btrfs_block_rsv_check(trans, root,
192 root->orphan_block_rsv,
197 oldsize = i_size_read(inode);
198 btrfs_i_size_write(inode, 0);
199 truncate_pagecache(inode, oldsize, 0);
202 * We don't need an orphan item because truncating the free space cache
203 * will never be split across transactions.
205 ret = btrfs_truncate_inode_items(trans, root, inode,
206 0, BTRFS_EXTENT_DATA_KEY);
212 ret = btrfs_update_inode(trans, root, inode);
216 static int readahead_cache(struct inode *inode)
218 struct file_ra_state *ra;
219 unsigned long last_index;
221 ra = kzalloc(sizeof(*ra), GFP_NOFS);
225 file_ra_state_init(ra, inode->i_mapping);
226 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
228 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
235 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
236 struct btrfs_free_space_ctl *ctl,
237 struct btrfs_path *path, u64 offset)
239 struct btrfs_free_space_header *header;
240 struct extent_buffer *leaf;
242 u32 *checksums = NULL, *crc;
243 char *disk_crcs = NULL;
244 struct btrfs_key key;
245 struct list_head bitmaps;
249 u32 cur_crc = ~(u32)0;
251 unsigned long first_page_offset;
255 INIT_LIST_HEAD(&bitmaps);
257 /* Nothing in the space cache, goodbye */
258 if (!i_size_read(inode))
261 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
265 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
269 btrfs_release_path(path);
276 leaf = path->nodes[0];
277 header = btrfs_item_ptr(leaf, path->slots[0],
278 struct btrfs_free_space_header);
279 num_entries = btrfs_free_space_entries(leaf, header);
280 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
281 generation = btrfs_free_space_generation(leaf, header);
282 btrfs_release_path(path);
284 if (BTRFS_I(inode)->generation != generation) {
285 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
286 " not match free space cache generation (%llu)\n",
287 (unsigned long long)BTRFS_I(inode)->generation,
288 (unsigned long long)generation);
295 /* Setup everything for doing checksumming */
296 num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
297 checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
300 first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
301 disk_crcs = kzalloc(first_page_offset, GFP_NOFS);
305 ret = readahead_cache(inode);
310 struct btrfs_free_space_entry *entry;
311 struct btrfs_free_space *e;
313 unsigned long offset = 0;
314 unsigned long start_offset = 0;
317 if (!num_entries && !num_bitmaps)
321 start_offset = first_page_offset;
322 offset = start_offset;
325 page = grab_cache_page(inode->i_mapping, index);
329 if (!PageUptodate(page)) {
330 btrfs_readpage(NULL, page);
332 if (!PageUptodate(page)) {
334 page_cache_release(page);
335 printk(KERN_ERR "btrfs: error reading free "
345 memcpy(disk_crcs, addr, first_page_offset);
346 gen = addr + (sizeof(u32) * num_checksums);
347 if (*gen != BTRFS_I(inode)->generation) {
348 printk(KERN_ERR "btrfs: space cache generation"
349 " (%llu) does not match inode (%llu)\n",
350 (unsigned long long)*gen,
352 BTRFS_I(inode)->generation);
355 page_cache_release(page);
358 crc = (u32 *)disk_crcs;
360 entry = addr + start_offset;
362 /* First lets check our crc before we do anything fun */
364 cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc,
365 PAGE_CACHE_SIZE - start_offset);
366 btrfs_csum_final(cur_crc, (char *)&cur_crc);
367 if (cur_crc != *crc) {
368 printk(KERN_ERR "btrfs: crc mismatch for page %lu\n",
372 page_cache_release(page);
382 e = kmem_cache_zalloc(btrfs_free_space_cachep,
387 page_cache_release(page);
391 e->offset = le64_to_cpu(entry->offset);
392 e->bytes = le64_to_cpu(entry->bytes);
395 kmem_cache_free(btrfs_free_space_cachep, e);
397 page_cache_release(page);
401 if (entry->type == BTRFS_FREE_SPACE_EXTENT) {
402 spin_lock(&ctl->tree_lock);
403 ret = link_free_space(ctl, e);
404 spin_unlock(&ctl->tree_lock);
407 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
411 btrfs_free_space_cachep, e);
413 page_cache_release(page);
416 spin_lock(&ctl->tree_lock);
417 ret2 = link_free_space(ctl, e);
418 ctl->total_bitmaps++;
419 ctl->op->recalc_thresholds(ctl);
420 spin_unlock(&ctl->tree_lock);
421 list_add_tail(&e->list, &bitmaps);
425 offset += sizeof(struct btrfs_free_space_entry);
426 if (offset + sizeof(struct btrfs_free_space_entry) >=
433 * We read an entry out of this page, we need to move on to the
442 * We add the bitmaps at the end of the entries in order that
443 * the bitmap entries are added to the cache.
445 e = list_entry(bitmaps.next, struct btrfs_free_space, list);
446 list_del_init(&e->list);
447 memcpy(e->bitmap, addr, PAGE_CACHE_SIZE);
452 page_cache_release(page);
462 __btrfs_remove_free_space_cache(ctl);
466 int load_free_space_cache(struct btrfs_fs_info *fs_info,
467 struct btrfs_block_group_cache *block_group)
469 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
470 struct btrfs_root *root = fs_info->tree_root;
472 struct btrfs_path *path;
475 u64 used = btrfs_block_group_used(&block_group->item);
478 * If we're unmounting then just return, since this does a search on the
479 * normal root and not the commit root and we could deadlock.
482 if (fs_info->closing)
486 * If this block group has been marked to be cleared for one reason or
487 * another then we can't trust the on disk cache, so just return.
489 spin_lock(&block_group->lock);
490 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
491 spin_unlock(&block_group->lock);
494 spin_unlock(&block_group->lock);
496 path = btrfs_alloc_path();
500 inode = lookup_free_space_inode(root, block_group, path);
502 btrfs_free_path(path);
506 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
507 path, block_group->key.objectid);
508 btrfs_free_path(path);
512 spin_lock(&ctl->tree_lock);
513 matched = (ctl->free_space == (block_group->key.offset - used -
514 block_group->bytes_super));
515 spin_unlock(&ctl->tree_lock);
518 __btrfs_remove_free_space_cache(ctl);
519 printk(KERN_ERR "block group %llu has an wrong amount of free "
520 "space\n", block_group->key.objectid);
525 /* This cache is bogus, make sure it gets cleared */
526 spin_lock(&block_group->lock);
527 block_group->disk_cache_state = BTRFS_DC_CLEAR;
528 spin_unlock(&block_group->lock);
531 printk(KERN_ERR "btrfs: failed to load free space cache "
532 "for block group %llu\n", block_group->key.objectid);
539 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
540 struct btrfs_free_space_ctl *ctl,
541 struct btrfs_block_group_cache *block_group,
542 struct btrfs_trans_handle *trans,
543 struct btrfs_path *path, u64 offset)
545 struct btrfs_free_space_header *header;
546 struct extent_buffer *leaf;
547 struct rb_node *node;
548 struct list_head *pos, *n;
551 struct extent_state *cached_state = NULL;
552 struct btrfs_free_cluster *cluster = NULL;
553 struct extent_io_tree *unpin = NULL;
554 struct list_head bitmap_list;
555 struct btrfs_key key;
558 u32 *crc, *checksums;
559 unsigned long first_page_offset;
560 int index = 0, num_pages = 0;
564 bool next_page = false;
565 bool out_of_space = false;
567 INIT_LIST_HEAD(&bitmap_list);
569 node = rb_first(&ctl->free_space_offset);
573 if (!i_size_read(inode))
576 num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
578 filemap_write_and_wait(inode->i_mapping);
579 btrfs_wait_ordered_range(inode, inode->i_size &
580 ~(root->sectorsize - 1), (u64)-1);
582 /* We need a checksum per page. */
583 crc = checksums = kzalloc(sizeof(u32) * num_pages, GFP_NOFS);
587 pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS);
593 /* Since the first page has all of our checksums and our generation we
594 * need to calculate the offset into the page that we can start writing
597 first_page_offset = (sizeof(u32) * num_pages) + sizeof(u64);
599 /* Get the cluster for this block_group if it exists */
600 if (block_group && !list_empty(&block_group->cluster_list))
601 cluster = list_entry(block_group->cluster_list.next,
602 struct btrfs_free_cluster,
606 * We shouldn't have switched the pinned extents yet so this is the
609 unpin = root->fs_info->pinned_extents;
612 * Lock all pages first so we can lock the extent safely.
614 * NOTE: Because we hold the ref the entire time we're going to write to
615 * the page find_get_page should never fail, so we don't do a check
616 * after find_get_page at this point. Just putting this here so people
617 * know and don't freak out.
619 while (index < num_pages) {
620 page = grab_cache_page(inode->i_mapping, index);
624 for (i = 0; i < num_pages; i++) {
625 unlock_page(pages[i]);
626 page_cache_release(pages[i]);
635 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
636 0, &cached_state, GFP_NOFS);
639 * When searching for pinned extents, we need to start at our start
643 start = block_group->key.objectid;
645 /* Write out the extent entries */
647 struct btrfs_free_space_entry *entry;
649 unsigned long offset = 0;
650 unsigned long start_offset = 0;
655 start_offset = first_page_offset;
656 offset = start_offset;
659 if (index >= num_pages) {
667 entry = addr + start_offset;
669 memset(addr, 0, PAGE_CACHE_SIZE);
670 while (node && !next_page) {
671 struct btrfs_free_space *e;
673 e = rb_entry(node, struct btrfs_free_space, offset_index);
676 entry->offset = cpu_to_le64(e->offset);
677 entry->bytes = cpu_to_le64(e->bytes);
679 entry->type = BTRFS_FREE_SPACE_BITMAP;
680 list_add_tail(&e->list, &bitmap_list);
683 entry->type = BTRFS_FREE_SPACE_EXTENT;
685 node = rb_next(node);
686 if (!node && cluster) {
687 node = rb_first(&cluster->root);
690 offset += sizeof(struct btrfs_free_space_entry);
691 if (offset + sizeof(struct btrfs_free_space_entry) >=
698 * We want to add any pinned extents to our free space cache
699 * so we don't leak the space
701 while (block_group && !next_page &&
702 (start < block_group->key.objectid +
703 block_group->key.offset)) {
704 ret = find_first_extent_bit(unpin, start, &start, &end,
711 /* This pinned extent is out of our range */
712 if (start >= block_group->key.objectid +
713 block_group->key.offset)
716 len = block_group->key.objectid +
717 block_group->key.offset - start;
718 len = min(len, end + 1 - start);
721 entry->offset = cpu_to_le64(start);
722 entry->bytes = cpu_to_le64(len);
723 entry->type = BTRFS_FREE_SPACE_EXTENT;
726 offset += sizeof(struct btrfs_free_space_entry);
727 if (offset + sizeof(struct btrfs_free_space_entry) >=
733 *crc = btrfs_csum_data(root, addr + start_offset, *crc,
734 PAGE_CACHE_SIZE - start_offset);
737 btrfs_csum_final(*crc, (char *)crc);
740 bytes += PAGE_CACHE_SIZE;
743 } while (node || next_page);
745 /* Write out the bitmaps */
746 list_for_each_safe(pos, n, &bitmap_list) {
748 struct btrfs_free_space *entry =
749 list_entry(pos, struct btrfs_free_space, list);
751 if (index >= num_pages) {
758 memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE);
760 *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE);
762 btrfs_csum_final(*crc, (char *)crc);
764 bytes += PAGE_CACHE_SIZE;
766 list_del_init(&entry->list);
771 btrfs_drop_pages(pages, num_pages);
772 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
773 i_size_read(inode) - 1, &cached_state,
779 /* Zero out the rest of the pages just to make sure */
780 while (index < num_pages) {
785 memset(addr, 0, PAGE_CACHE_SIZE);
787 bytes += PAGE_CACHE_SIZE;
791 /* Write the checksums and trans id to the first page */
799 memcpy(addr, checksums, sizeof(u32) * num_pages);
800 gen = addr + (sizeof(u32) * num_pages);
801 *gen = trans->transid;
805 ret = btrfs_dirty_pages(root, inode, pages, num_pages, 0,
806 bytes, &cached_state);
807 btrfs_drop_pages(pages, num_pages);
808 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
809 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
816 BTRFS_I(inode)->generation = trans->transid;
818 filemap_write_and_wait(inode->i_mapping);
820 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
824 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
827 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
828 EXTENT_DIRTY | EXTENT_DELALLOC |
829 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
832 leaf = path->nodes[0];
834 struct btrfs_key found_key;
835 BUG_ON(!path->slots[0]);
837 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
838 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
839 found_key.offset != offset) {
841 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
842 EXTENT_DIRTY | EXTENT_DELALLOC |
843 EXTENT_DO_ACCOUNTING, 0, 0, NULL,
845 btrfs_release_path(path);
849 header = btrfs_item_ptr(leaf, path->slots[0],
850 struct btrfs_free_space_header);
851 btrfs_set_free_space_entries(leaf, header, entries);
852 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
853 btrfs_set_free_space_generation(leaf, header, trans->transid);
854 btrfs_mark_buffer_dirty(leaf);
855 btrfs_release_path(path);
861 invalidate_inode_pages2_range(inode->i_mapping, 0, index);
862 BTRFS_I(inode)->generation = 0;
866 btrfs_update_inode(trans, root, inode);
870 int btrfs_write_out_cache(struct btrfs_root *root,
871 struct btrfs_trans_handle *trans,
872 struct btrfs_block_group_cache *block_group,
873 struct btrfs_path *path)
875 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
879 root = root->fs_info->tree_root;
881 spin_lock(&block_group->lock);
882 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
883 spin_unlock(&block_group->lock);
886 spin_unlock(&block_group->lock);
888 inode = lookup_free_space_inode(root, block_group, path);
892 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
893 path, block_group->key.objectid);
895 spin_lock(&block_group->lock);
896 block_group->disk_cache_state = BTRFS_DC_ERROR;
897 spin_unlock(&block_group->lock);
900 printk(KERN_ERR "btrfs: failed to write free space cace "
901 "for block group %llu\n", block_group->key.objectid);
908 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
911 BUG_ON(offset < bitmap_start);
912 offset -= bitmap_start;
913 return (unsigned long)(div_u64(offset, unit));
916 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
918 return (unsigned long)(div_u64(bytes, unit));
921 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
925 u64 bytes_per_bitmap;
927 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
928 bitmap_start = offset - ctl->start;
929 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
930 bitmap_start *= bytes_per_bitmap;
931 bitmap_start += ctl->start;
936 static int tree_insert_offset(struct rb_root *root, u64 offset,
937 struct rb_node *node, int bitmap)
939 struct rb_node **p = &root->rb_node;
940 struct rb_node *parent = NULL;
941 struct btrfs_free_space *info;
945 info = rb_entry(parent, struct btrfs_free_space, offset_index);
947 if (offset < info->offset) {
949 } else if (offset > info->offset) {
953 * we could have a bitmap entry and an extent entry
954 * share the same offset. If this is the case, we want
955 * the extent entry to always be found first if we do a
956 * linear search through the tree, since we want to have
957 * the quickest allocation time, and allocating from an
958 * extent is faster than allocating from a bitmap. So
959 * if we're inserting a bitmap and we find an entry at
960 * this offset, we want to go right, or after this entry
961 * logically. If we are inserting an extent and we've
962 * found a bitmap, we want to go left, or before
966 WARN_ON(info->bitmap);
969 WARN_ON(!info->bitmap);
975 rb_link_node(node, parent, p);
976 rb_insert_color(node, root);
982 * searches the tree for the given offset.
984 * fuzzy - If this is set, then we are trying to make an allocation, and we just
985 * want a section that has at least bytes size and comes at or after the given
988 static struct btrfs_free_space *
989 tree_search_offset(struct btrfs_free_space_ctl *ctl,
990 u64 offset, int bitmap_only, int fuzzy)
992 struct rb_node *n = ctl->free_space_offset.rb_node;
993 struct btrfs_free_space *entry, *prev = NULL;
995 /* find entry that is closest to the 'offset' */
1002 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1005 if (offset < entry->offset)
1007 else if (offset > entry->offset)
1020 * bitmap entry and extent entry may share same offset,
1021 * in that case, bitmap entry comes after extent entry.
1026 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1027 if (entry->offset != offset)
1030 WARN_ON(!entry->bitmap);
1033 if (entry->bitmap) {
1035 * if previous extent entry covers the offset,
1036 * we should return it instead of the bitmap entry
1038 n = &entry->offset_index;
1043 prev = rb_entry(n, struct btrfs_free_space,
1045 if (!prev->bitmap) {
1046 if (prev->offset + prev->bytes > offset)
1058 /* find last entry before the 'offset' */
1060 if (entry->offset > offset) {
1061 n = rb_prev(&entry->offset_index);
1063 entry = rb_entry(n, struct btrfs_free_space,
1065 BUG_ON(entry->offset > offset);
1074 if (entry->bitmap) {
1075 n = &entry->offset_index;
1080 prev = rb_entry(n, struct btrfs_free_space,
1082 if (!prev->bitmap) {
1083 if (prev->offset + prev->bytes > offset)
1088 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1090 } else if (entry->offset + entry->bytes > offset)
1097 if (entry->bitmap) {
1098 if (entry->offset + BITS_PER_BITMAP *
1102 if (entry->offset + entry->bytes > offset)
1106 n = rb_next(&entry->offset_index);
1109 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1115 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1116 struct btrfs_free_space *info)
1118 rb_erase(&info->offset_index, &ctl->free_space_offset);
1119 ctl->free_extents--;
1122 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1123 struct btrfs_free_space *info)
1125 __unlink_free_space(ctl, info);
1126 ctl->free_space -= info->bytes;
1129 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1130 struct btrfs_free_space *info)
1134 BUG_ON(!info->bitmap && !info->bytes);
1135 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1136 &info->offset_index, (info->bitmap != NULL));
1140 ctl->free_space += info->bytes;
1141 ctl->free_extents++;
1145 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1147 struct btrfs_block_group_cache *block_group = ctl->private;
1151 u64 size = block_group->key.offset;
1152 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1153 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1155 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1158 * The goal is to keep the total amount of memory used per 1gb of space
1159 * at or below 32k, so we need to adjust how much memory we allow to be
1160 * used by extent based free space tracking
1162 if (size < 1024 * 1024 * 1024)
1163 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1165 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1166 div64_u64(size, 1024 * 1024 * 1024);
1169 * we want to account for 1 more bitmap than what we have so we can make
1170 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1171 * we add more bitmaps.
1173 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1175 if (bitmap_bytes >= max_bytes) {
1176 ctl->extents_thresh = 0;
1181 * we want the extent entry threshold to always be at most 1/2 the maxw
1182 * bytes we can have, or whatever is less than that.
1184 extent_bytes = max_bytes - bitmap_bytes;
1185 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1187 ctl->extents_thresh =
1188 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1191 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1192 struct btrfs_free_space *info, u64 offset,
1195 unsigned long start, count;
1197 start = offset_to_bit(info->offset, ctl->unit, offset);
1198 count = bytes_to_bits(bytes, ctl->unit);
1199 BUG_ON(start + count > BITS_PER_BITMAP);
1201 bitmap_clear(info->bitmap, start, count);
1203 info->bytes -= bytes;
1204 ctl->free_space -= bytes;
1207 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1208 struct btrfs_free_space *info, u64 offset,
1211 unsigned long start, count;
1213 start = offset_to_bit(info->offset, ctl->unit, offset);
1214 count = bytes_to_bits(bytes, ctl->unit);
1215 BUG_ON(start + count > BITS_PER_BITMAP);
1217 bitmap_set(info->bitmap, start, count);
1219 info->bytes += bytes;
1220 ctl->free_space += bytes;
1223 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1224 struct btrfs_free_space *bitmap_info, u64 *offset,
1227 unsigned long found_bits = 0;
1228 unsigned long bits, i;
1229 unsigned long next_zero;
1231 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1232 max_t(u64, *offset, bitmap_info->offset));
1233 bits = bytes_to_bits(*bytes, ctl->unit);
1235 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1236 i < BITS_PER_BITMAP;
1237 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1238 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1239 BITS_PER_BITMAP, i);
1240 if ((next_zero - i) >= bits) {
1241 found_bits = next_zero - i;
1248 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1249 *bytes = (u64)(found_bits) * ctl->unit;
1256 static struct btrfs_free_space *
1257 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1259 struct btrfs_free_space *entry;
1260 struct rb_node *node;
1263 if (!ctl->free_space_offset.rb_node)
1266 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1270 for (node = &entry->offset_index; node; node = rb_next(node)) {
1271 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1272 if (entry->bytes < *bytes)
1275 if (entry->bitmap) {
1276 ret = search_bitmap(ctl, entry, offset, bytes);
1282 *offset = entry->offset;
1283 *bytes = entry->bytes;
1290 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1291 struct btrfs_free_space *info, u64 offset)
1293 info->offset = offset_to_bitmap(ctl, offset);
1295 link_free_space(ctl, info);
1296 ctl->total_bitmaps++;
1298 ctl->op->recalc_thresholds(ctl);
1301 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1302 struct btrfs_free_space *bitmap_info)
1304 unlink_free_space(ctl, bitmap_info);
1305 kfree(bitmap_info->bitmap);
1306 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1307 ctl->total_bitmaps--;
1308 ctl->op->recalc_thresholds(ctl);
1311 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1312 struct btrfs_free_space *bitmap_info,
1313 u64 *offset, u64 *bytes)
1316 u64 search_start, search_bytes;
1320 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1323 * XXX - this can go away after a few releases.
1325 * since the only user of btrfs_remove_free_space is the tree logging
1326 * stuff, and the only way to test that is under crash conditions, we
1327 * want to have this debug stuff here just in case somethings not
1328 * working. Search the bitmap for the space we are trying to use to
1329 * make sure its actually there. If its not there then we need to stop
1330 * because something has gone wrong.
1332 search_start = *offset;
1333 search_bytes = *bytes;
1334 search_bytes = min(search_bytes, end - search_start + 1);
1335 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1336 BUG_ON(ret < 0 || search_start != *offset);
1338 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1339 bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
1340 *bytes -= end - *offset + 1;
1342 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1343 bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
1348 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1349 if (!bitmap_info->bytes)
1350 free_bitmap(ctl, bitmap_info);
1353 * no entry after this bitmap, but we still have bytes to
1354 * remove, so something has gone wrong.
1359 bitmap_info = rb_entry(next, struct btrfs_free_space,
1363 * if the next entry isn't a bitmap we need to return to let the
1364 * extent stuff do its work.
1366 if (!bitmap_info->bitmap)
1370 * Ok the next item is a bitmap, but it may not actually hold
1371 * the information for the rest of this free space stuff, so
1372 * look for it, and if we don't find it return so we can try
1373 * everything over again.
1375 search_start = *offset;
1376 search_bytes = *bytes;
1377 ret = search_bitmap(ctl, bitmap_info, &search_start,
1379 if (ret < 0 || search_start != *offset)
1383 } else if (!bitmap_info->bytes)
1384 free_bitmap(ctl, bitmap_info);
1389 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1390 struct btrfs_free_space *info)
1392 struct btrfs_block_group_cache *block_group = ctl->private;
1395 * If we are below the extents threshold then we can add this as an
1396 * extent, and don't have to deal with the bitmap
1398 if (ctl->free_extents < ctl->extents_thresh) {
1400 * If this block group has some small extents we don't want to
1401 * use up all of our free slots in the cache with them, we want
1402 * to reserve them to larger extents, however if we have plent
1403 * of cache left then go ahead an dadd them, no sense in adding
1404 * the overhead of a bitmap if we don't have to.
1406 if (info->bytes <= block_group->sectorsize * 4) {
1407 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1415 * some block groups are so tiny they can't be enveloped by a bitmap, so
1416 * don't even bother to create a bitmap for this
1418 if (BITS_PER_BITMAP * block_group->sectorsize >
1419 block_group->key.offset)
1425 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1426 struct btrfs_free_space *info)
1428 struct btrfs_free_space *bitmap_info;
1430 u64 bytes, offset, end;
1433 bytes = info->bytes;
1434 offset = info->offset;
1436 if (!ctl->op->use_bitmap(ctl, info))
1440 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1447 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1449 if (offset >= bitmap_info->offset && offset + bytes > end) {
1450 bitmap_set_bits(ctl, bitmap_info, offset, end - offset);
1451 bytes -= end - offset;
1454 } else if (offset >= bitmap_info->offset && offset + bytes <= end) {
1455 bitmap_set_bits(ctl, bitmap_info, offset, bytes);
1468 if (info && info->bitmap) {
1469 add_new_bitmap(ctl, info, offset);
1474 spin_unlock(&ctl->tree_lock);
1476 /* no pre-allocated info, allocate a new one */
1478 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1481 spin_lock(&ctl->tree_lock);
1487 /* allocate the bitmap */
1488 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1489 spin_lock(&ctl->tree_lock);
1490 if (!info->bitmap) {
1500 kfree(info->bitmap);
1501 kmem_cache_free(btrfs_free_space_cachep, info);
1507 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1508 struct btrfs_free_space *info, bool update_stat)
1510 struct btrfs_free_space *left_info;
1511 struct btrfs_free_space *right_info;
1512 bool merged = false;
1513 u64 offset = info->offset;
1514 u64 bytes = info->bytes;
1517 * first we want to see if there is free space adjacent to the range we
1518 * are adding, if there is remove that struct and add a new one to
1519 * cover the entire range
1521 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1522 if (right_info && rb_prev(&right_info->offset_index))
1523 left_info = rb_entry(rb_prev(&right_info->offset_index),
1524 struct btrfs_free_space, offset_index);
1526 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1528 if (right_info && !right_info->bitmap) {
1530 unlink_free_space(ctl, right_info);
1532 __unlink_free_space(ctl, right_info);
1533 info->bytes += right_info->bytes;
1534 kmem_cache_free(btrfs_free_space_cachep, right_info);
1538 if (left_info && !left_info->bitmap &&
1539 left_info->offset + left_info->bytes == offset) {
1541 unlink_free_space(ctl, left_info);
1543 __unlink_free_space(ctl, left_info);
1544 info->offset = left_info->offset;
1545 info->bytes += left_info->bytes;
1546 kmem_cache_free(btrfs_free_space_cachep, left_info);
1553 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1554 u64 offset, u64 bytes)
1556 struct btrfs_free_space *info;
1559 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1563 info->offset = offset;
1564 info->bytes = bytes;
1566 spin_lock(&ctl->tree_lock);
1568 if (try_merge_free_space(ctl, info, true))
1572 * There was no extent directly to the left or right of this new
1573 * extent then we know we're going to have to allocate a new extent, so
1574 * before we do that see if we need to drop this into a bitmap
1576 ret = insert_into_bitmap(ctl, info);
1584 ret = link_free_space(ctl, info);
1586 kmem_cache_free(btrfs_free_space_cachep, info);
1588 spin_unlock(&ctl->tree_lock);
1591 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1592 BUG_ON(ret == -EEXIST);
1598 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1599 u64 offset, u64 bytes)
1601 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1602 struct btrfs_free_space *info;
1603 struct btrfs_free_space *next_info = NULL;
1606 spin_lock(&ctl->tree_lock);
1609 info = tree_search_offset(ctl, offset, 0, 0);
1612 * oops didn't find an extent that matched the space we wanted
1613 * to remove, look for a bitmap instead
1615 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1623 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1625 next_info = rb_entry(rb_next(&info->offset_index),
1626 struct btrfs_free_space,
1629 if (next_info->bitmap)
1630 end = next_info->offset +
1631 BITS_PER_BITMAP * ctl->unit - 1;
1633 end = next_info->offset + next_info->bytes;
1635 if (next_info->bytes < bytes ||
1636 next_info->offset > offset || offset > end) {
1637 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1638 " trying to use %llu\n",
1639 (unsigned long long)info->offset,
1640 (unsigned long long)info->bytes,
1641 (unsigned long long)bytes);
1650 if (info->bytes == bytes) {
1651 unlink_free_space(ctl, info);
1653 kfree(info->bitmap);
1654 ctl->total_bitmaps--;
1656 kmem_cache_free(btrfs_free_space_cachep, info);
1660 if (!info->bitmap && info->offset == offset) {
1661 unlink_free_space(ctl, info);
1662 info->offset += bytes;
1663 info->bytes -= bytes;
1664 link_free_space(ctl, info);
1668 if (!info->bitmap && info->offset <= offset &&
1669 info->offset + info->bytes >= offset + bytes) {
1670 u64 old_start = info->offset;
1672 * we're freeing space in the middle of the info,
1673 * this can happen during tree log replay
1675 * first unlink the old info and then
1676 * insert it again after the hole we're creating
1678 unlink_free_space(ctl, info);
1679 if (offset + bytes < info->offset + info->bytes) {
1680 u64 old_end = info->offset + info->bytes;
1682 info->offset = offset + bytes;
1683 info->bytes = old_end - info->offset;
1684 ret = link_free_space(ctl, info);
1689 /* the hole we're creating ends at the end
1690 * of the info struct, just free the info
1692 kmem_cache_free(btrfs_free_space_cachep, info);
1694 spin_unlock(&ctl->tree_lock);
1696 /* step two, insert a new info struct to cover
1697 * anything before the hole
1699 ret = btrfs_add_free_space(block_group, old_start,
1700 offset - old_start);
1705 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1710 spin_unlock(&ctl->tree_lock);
1715 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1718 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1719 struct btrfs_free_space *info;
1723 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1724 info = rb_entry(n, struct btrfs_free_space, offset_index);
1725 if (info->bytes >= bytes)
1727 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1728 (unsigned long long)info->offset,
1729 (unsigned long long)info->bytes,
1730 (info->bitmap) ? "yes" : "no");
1732 printk(KERN_INFO "block group has cluster?: %s\n",
1733 list_empty(&block_group->cluster_list) ? "no" : "yes");
1734 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1738 static struct btrfs_free_space_op free_space_op = {
1739 .recalc_thresholds = recalculate_thresholds,
1740 .use_bitmap = use_bitmap,
1743 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1745 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1747 spin_lock_init(&ctl->tree_lock);
1748 ctl->unit = block_group->sectorsize;
1749 ctl->start = block_group->key.objectid;
1750 ctl->private = block_group;
1751 ctl->op = &free_space_op;
1754 * we only want to have 32k of ram per block group for keeping
1755 * track of free space, and if we pass 1/2 of that we want to
1756 * start converting things over to using bitmaps
1758 ctl->extents_thresh = ((1024 * 32) / 2) /
1759 sizeof(struct btrfs_free_space);
1763 * for a given cluster, put all of its extents back into the free
1764 * space cache. If the block group passed doesn't match the block group
1765 * pointed to by the cluster, someone else raced in and freed the
1766 * cluster already. In that case, we just return without changing anything
1769 __btrfs_return_cluster_to_free_space(
1770 struct btrfs_block_group_cache *block_group,
1771 struct btrfs_free_cluster *cluster)
1773 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1774 struct btrfs_free_space *entry;
1775 struct rb_node *node;
1777 spin_lock(&cluster->lock);
1778 if (cluster->block_group != block_group)
1781 cluster->block_group = NULL;
1782 cluster->window_start = 0;
1783 list_del_init(&cluster->block_group_list);
1785 node = rb_first(&cluster->root);
1789 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1790 node = rb_next(&entry->offset_index);
1791 rb_erase(&entry->offset_index, &cluster->root);
1793 bitmap = (entry->bitmap != NULL);
1795 try_merge_free_space(ctl, entry, false);
1796 tree_insert_offset(&ctl->free_space_offset,
1797 entry->offset, &entry->offset_index, bitmap);
1799 cluster->root = RB_ROOT;
1802 spin_unlock(&cluster->lock);
1803 btrfs_put_block_group(block_group);
1807 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
1809 struct btrfs_free_space *info;
1810 struct rb_node *node;
1812 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
1813 info = rb_entry(node, struct btrfs_free_space, offset_index);
1814 unlink_free_space(ctl, info);
1815 kfree(info->bitmap);
1816 kmem_cache_free(btrfs_free_space_cachep, info);
1817 if (need_resched()) {
1818 spin_unlock(&ctl->tree_lock);
1820 spin_lock(&ctl->tree_lock);
1825 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
1827 spin_lock(&ctl->tree_lock);
1828 __btrfs_remove_free_space_cache_locked(ctl);
1829 spin_unlock(&ctl->tree_lock);
1832 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
1834 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1835 struct btrfs_free_cluster *cluster;
1836 struct list_head *head;
1838 spin_lock(&ctl->tree_lock);
1839 while ((head = block_group->cluster_list.next) !=
1840 &block_group->cluster_list) {
1841 cluster = list_entry(head, struct btrfs_free_cluster,
1844 WARN_ON(cluster->block_group != block_group);
1845 __btrfs_return_cluster_to_free_space(block_group, cluster);
1846 if (need_resched()) {
1847 spin_unlock(&ctl->tree_lock);
1849 spin_lock(&ctl->tree_lock);
1852 __btrfs_remove_free_space_cache_locked(ctl);
1853 spin_unlock(&ctl->tree_lock);
1857 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
1858 u64 offset, u64 bytes, u64 empty_size)
1860 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1861 struct btrfs_free_space *entry = NULL;
1862 u64 bytes_search = bytes + empty_size;
1865 spin_lock(&ctl->tree_lock);
1866 entry = find_free_space(ctl, &offset, &bytes_search);
1871 if (entry->bitmap) {
1872 bitmap_clear_bits(ctl, entry, offset, bytes);
1874 free_bitmap(ctl, entry);
1876 unlink_free_space(ctl, entry);
1877 entry->offset += bytes;
1878 entry->bytes -= bytes;
1880 kmem_cache_free(btrfs_free_space_cachep, entry);
1882 link_free_space(ctl, entry);
1886 spin_unlock(&ctl->tree_lock);
1892 * given a cluster, put all of its extents back into the free space
1893 * cache. If a block group is passed, this function will only free
1894 * a cluster that belongs to the passed block group.
1896 * Otherwise, it'll get a reference on the block group pointed to by the
1897 * cluster and remove the cluster from it.
1899 int btrfs_return_cluster_to_free_space(
1900 struct btrfs_block_group_cache *block_group,
1901 struct btrfs_free_cluster *cluster)
1903 struct btrfs_free_space_ctl *ctl;
1906 /* first, get a safe pointer to the block group */
1907 spin_lock(&cluster->lock);
1909 block_group = cluster->block_group;
1911 spin_unlock(&cluster->lock);
1914 } else if (cluster->block_group != block_group) {
1915 /* someone else has already freed it don't redo their work */
1916 spin_unlock(&cluster->lock);
1919 atomic_inc(&block_group->count);
1920 spin_unlock(&cluster->lock);
1922 ctl = block_group->free_space_ctl;
1924 /* now return any extents the cluster had on it */
1925 spin_lock(&ctl->tree_lock);
1926 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
1927 spin_unlock(&ctl->tree_lock);
1929 /* finally drop our ref */
1930 btrfs_put_block_group(block_group);
1934 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
1935 struct btrfs_free_cluster *cluster,
1936 struct btrfs_free_space *entry,
1937 u64 bytes, u64 min_start)
1939 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1941 u64 search_start = cluster->window_start;
1942 u64 search_bytes = bytes;
1945 search_start = min_start;
1946 search_bytes = bytes;
1948 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
1953 bitmap_clear_bits(ctl, entry, ret, bytes);
1959 * given a cluster, try to allocate 'bytes' from it, returns 0
1960 * if it couldn't find anything suitably large, or a logical disk offset
1961 * if things worked out
1963 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
1964 struct btrfs_free_cluster *cluster, u64 bytes,
1967 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1968 struct btrfs_free_space *entry = NULL;
1969 struct rb_node *node;
1972 spin_lock(&cluster->lock);
1973 if (bytes > cluster->max_size)
1976 if (cluster->block_group != block_group)
1979 node = rb_first(&cluster->root);
1983 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1985 if (entry->bytes < bytes ||
1986 (!entry->bitmap && entry->offset < min_start)) {
1987 node = rb_next(&entry->offset_index);
1990 entry = rb_entry(node, struct btrfs_free_space,
1995 if (entry->bitmap) {
1996 ret = btrfs_alloc_from_bitmap(block_group,
1997 cluster, entry, bytes,
2000 node = rb_next(&entry->offset_index);
2003 entry = rb_entry(node, struct btrfs_free_space,
2009 ret = entry->offset;
2011 entry->offset += bytes;
2012 entry->bytes -= bytes;
2015 if (entry->bytes == 0)
2016 rb_erase(&entry->offset_index, &cluster->root);
2020 spin_unlock(&cluster->lock);
2025 spin_lock(&ctl->tree_lock);
2027 ctl->free_space -= bytes;
2028 if (entry->bytes == 0) {
2029 ctl->free_extents--;
2030 if (entry->bitmap) {
2031 kfree(entry->bitmap);
2032 ctl->total_bitmaps--;
2033 ctl->op->recalc_thresholds(ctl);
2035 kmem_cache_free(btrfs_free_space_cachep, entry);
2038 spin_unlock(&ctl->tree_lock);
2043 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2044 struct btrfs_free_space *entry,
2045 struct btrfs_free_cluster *cluster,
2046 u64 offset, u64 bytes, u64 min_bytes)
2048 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2049 unsigned long next_zero;
2051 unsigned long search_bits;
2052 unsigned long total_bits;
2053 unsigned long found_bits;
2054 unsigned long start = 0;
2055 unsigned long total_found = 0;
2059 i = offset_to_bit(entry->offset, block_group->sectorsize,
2060 max_t(u64, offset, entry->offset));
2061 search_bits = bytes_to_bits(bytes, block_group->sectorsize);
2062 total_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2066 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
2067 i < BITS_PER_BITMAP;
2068 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
2069 next_zero = find_next_zero_bit(entry->bitmap,
2070 BITS_PER_BITMAP, i);
2071 if (next_zero - i >= search_bits) {
2072 found_bits = next_zero - i;
2086 total_found += found_bits;
2088 if (cluster->max_size < found_bits * block_group->sectorsize)
2089 cluster->max_size = found_bits * block_group->sectorsize;
2091 if (total_found < total_bits) {
2092 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
2093 if (i - start > total_bits * 2) {
2095 cluster->max_size = 0;
2101 cluster->window_start = start * block_group->sectorsize +
2103 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2104 ret = tree_insert_offset(&cluster->root, entry->offset,
2105 &entry->offset_index, 1);
2112 * This searches the block group for just extents to fill the cluster with.
2114 static int setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2115 struct btrfs_free_cluster *cluster,
2116 u64 offset, u64 bytes, u64 min_bytes)
2118 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2119 struct btrfs_free_space *first = NULL;
2120 struct btrfs_free_space *entry = NULL;
2121 struct btrfs_free_space *prev = NULL;
2122 struct btrfs_free_space *last;
2123 struct rb_node *node;
2127 u64 max_gap = 128 * 1024;
2129 entry = tree_search_offset(ctl, offset, 0, 1);
2134 * We don't want bitmaps, so just move along until we find a normal
2137 while (entry->bitmap) {
2138 node = rb_next(&entry->offset_index);
2141 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2144 window_start = entry->offset;
2145 window_free = entry->bytes;
2146 max_extent = entry->bytes;
2151 while (window_free <= min_bytes) {
2152 node = rb_next(&entry->offset_index);
2155 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2160 * we haven't filled the empty size and the window is
2161 * very large. reset and try again
2163 if (entry->offset - (prev->offset + prev->bytes) > max_gap ||
2164 entry->offset - window_start > (min_bytes * 2)) {
2166 window_start = entry->offset;
2167 window_free = entry->bytes;
2169 max_extent = entry->bytes;
2172 window_free += entry->bytes;
2173 if (entry->bytes > max_extent)
2174 max_extent = entry->bytes;
2179 cluster->window_start = first->offset;
2181 node = &first->offset_index;
2184 * now we've found our entries, pull them out of the free space
2185 * cache and put them into the cluster rbtree
2190 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2191 node = rb_next(&entry->offset_index);
2195 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2196 ret = tree_insert_offset(&cluster->root, entry->offset,
2197 &entry->offset_index, 0);
2199 } while (node && entry != last);
2201 cluster->max_size = max_extent;
2207 * This specifically looks for bitmaps that may work in the cluster, we assume
2208 * that we have already failed to find extents that will work.
2210 static int setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2211 struct btrfs_free_cluster *cluster,
2212 u64 offset, u64 bytes, u64 min_bytes)
2214 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2215 struct btrfs_free_space *entry;
2216 struct rb_node *node;
2219 if (ctl->total_bitmaps == 0)
2222 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 0, 1);
2226 node = &entry->offset_index;
2228 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2229 node = rb_next(&entry->offset_index);
2232 if (entry->bytes < min_bytes)
2234 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2236 } while (ret && node);
2242 * here we try to find a cluster of blocks in a block group. The goal
2243 * is to find at least bytes free and up to empty_size + bytes free.
2244 * We might not find them all in one contiguous area.
2246 * returns zero and sets up cluster if things worked out, otherwise
2247 * it returns -enospc
2249 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2250 struct btrfs_root *root,
2251 struct btrfs_block_group_cache *block_group,
2252 struct btrfs_free_cluster *cluster,
2253 u64 offset, u64 bytes, u64 empty_size)
2255 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2259 /* for metadata, allow allocates with more holes */
2260 if (btrfs_test_opt(root, SSD_SPREAD)) {
2261 min_bytes = bytes + empty_size;
2262 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2264 * we want to do larger allocations when we are
2265 * flushing out the delayed refs, it helps prevent
2266 * making more work as we go along.
2268 if (trans->transaction->delayed_refs.flushing)
2269 min_bytes = max(bytes, (bytes + empty_size) >> 1);
2271 min_bytes = max(bytes, (bytes + empty_size) >> 4);
2273 min_bytes = max(bytes, (bytes + empty_size) >> 2);
2275 spin_lock(&ctl->tree_lock);
2278 * If we know we don't have enough space to make a cluster don't even
2279 * bother doing all the work to try and find one.
2281 if (ctl->free_space < min_bytes) {
2282 spin_unlock(&ctl->tree_lock);
2286 spin_lock(&cluster->lock);
2288 /* someone already found a cluster, hooray */
2289 if (cluster->block_group) {
2294 ret = setup_cluster_no_bitmap(block_group, cluster, offset, bytes,
2297 ret = setup_cluster_bitmap(block_group, cluster, offset,
2301 atomic_inc(&block_group->count);
2302 list_add_tail(&cluster->block_group_list,
2303 &block_group->cluster_list);
2304 cluster->block_group = block_group;
2307 spin_unlock(&cluster->lock);
2308 spin_unlock(&ctl->tree_lock);
2314 * simple code to zero out a cluster
2316 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2318 spin_lock_init(&cluster->lock);
2319 spin_lock_init(&cluster->refill_lock);
2320 cluster->root = RB_ROOT;
2321 cluster->max_size = 0;
2322 INIT_LIST_HEAD(&cluster->block_group_list);
2323 cluster->block_group = NULL;
2326 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2327 u64 *trimmed, u64 start, u64 end, u64 minlen)
2329 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2330 struct btrfs_free_space *entry = NULL;
2331 struct btrfs_fs_info *fs_info = block_group->fs_info;
2333 u64 actually_trimmed;
2338 while (start < end) {
2339 spin_lock(&ctl->tree_lock);
2341 if (ctl->free_space < minlen) {
2342 spin_unlock(&ctl->tree_lock);
2346 entry = tree_search_offset(ctl, start, 0, 1);
2348 entry = tree_search_offset(ctl,
2349 offset_to_bitmap(ctl, start),
2352 if (!entry || entry->offset >= end) {
2353 spin_unlock(&ctl->tree_lock);
2357 if (entry->bitmap) {
2358 ret = search_bitmap(ctl, entry, &start, &bytes);
2361 spin_unlock(&ctl->tree_lock);
2364 bytes = min(bytes, end - start);
2365 bitmap_clear_bits(ctl, entry, start, bytes);
2366 if (entry->bytes == 0)
2367 free_bitmap(ctl, entry);
2369 start = entry->offset + BITS_PER_BITMAP *
2370 block_group->sectorsize;
2371 spin_unlock(&ctl->tree_lock);
2376 start = entry->offset;
2377 bytes = min(entry->bytes, end - start);
2378 unlink_free_space(ctl, entry);
2379 kmem_cache_free(btrfs_free_space_cachep, entry);
2382 spin_unlock(&ctl->tree_lock);
2384 if (bytes >= minlen) {
2386 update_ret = btrfs_update_reserved_bytes(block_group,
2389 ret = btrfs_error_discard_extent(fs_info->extent_root,
2394 btrfs_add_free_space(block_group, start, bytes);
2396 btrfs_update_reserved_bytes(block_group,
2401 *trimmed += actually_trimmed;
2406 if (fatal_signal_pending(current)) {
2418 * Find the left-most item in the cache tree, and then return the
2419 * smallest inode number in the item.
2421 * Note: the returned inode number may not be the smallest one in
2422 * the tree, if the left-most item is a bitmap.
2424 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2426 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2427 struct btrfs_free_space *entry = NULL;
2430 spin_lock(&ctl->tree_lock);
2432 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2435 entry = rb_entry(rb_first(&ctl->free_space_offset),
2436 struct btrfs_free_space, offset_index);
2438 if (!entry->bitmap) {
2439 ino = entry->offset;
2441 unlink_free_space(ctl, entry);
2445 kmem_cache_free(btrfs_free_space_cachep, entry);
2447 link_free_space(ctl, entry);
2453 ret = search_bitmap(ctl, entry, &offset, &count);
2457 bitmap_clear_bits(ctl, entry, offset, 1);
2458 if (entry->bytes == 0)
2459 free_bitmap(ctl, entry);
2462 spin_unlock(&ctl->tree_lock);
2467 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2468 struct btrfs_path *path)
2470 struct inode *inode = NULL;
2472 spin_lock(&root->cache_lock);
2473 if (root->cache_inode)
2474 inode = igrab(root->cache_inode);
2475 spin_unlock(&root->cache_lock);
2479 inode = __lookup_free_space_inode(root, path, 0);
2483 spin_lock(&root->cache_lock);
2484 if (!root->fs_info->closing)
2485 root->cache_inode = igrab(inode);
2486 spin_unlock(&root->cache_lock);
2491 int create_free_ino_inode(struct btrfs_root *root,
2492 struct btrfs_trans_handle *trans,
2493 struct btrfs_path *path)
2495 return __create_free_space_inode(root, trans, path,
2496 BTRFS_FREE_INO_OBJECTID, 0);
2499 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2501 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2502 struct btrfs_path *path;
2503 struct inode *inode;
2505 u64 root_gen = btrfs_root_generation(&root->root_item);
2508 * If we're unmounting then just return, since this does a search on the
2509 * normal root and not the commit root and we could deadlock.
2512 if (fs_info->closing)
2515 path = btrfs_alloc_path();
2519 inode = lookup_free_ino_inode(root, path);
2523 if (root_gen != BTRFS_I(inode)->generation)
2526 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2529 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2530 "root %llu\n", root->root_key.objectid);
2534 btrfs_free_path(path);
2538 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2539 struct btrfs_trans_handle *trans,
2540 struct btrfs_path *path)
2542 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2543 struct inode *inode;
2546 inode = lookup_free_ino_inode(root, path);
2550 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2552 printk(KERN_ERR "btrfs: failed to write free ino cache "
2553 "for root %llu\n", root->root_key.objectid);