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"
28 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
29 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
31 static void recalculate_thresholds(struct btrfs_block_group_cache
33 static int link_free_space(struct btrfs_block_group_cache *block_group,
34 struct btrfs_free_space *info);
36 struct inode *lookup_free_space_inode(struct btrfs_root *root,
37 struct btrfs_block_group_cache
38 *block_group, 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 spin_lock(&block_group->lock);
49 if (block_group->inode)
50 inode = igrab(block_group->inode);
51 spin_unlock(&block_group->lock);
55 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
56 key.offset = block_group->key.objectid;
59 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
63 btrfs_release_path(root, path);
64 return ERR_PTR(-ENOENT);
67 leaf = path->nodes[0];
68 header = btrfs_item_ptr(leaf, path->slots[0],
69 struct btrfs_free_space_header);
70 btrfs_free_space_key(leaf, header, &disk_key);
71 btrfs_disk_key_to_cpu(&location, &disk_key);
72 btrfs_release_path(root, path);
74 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
76 return ERR_PTR(-ENOENT);
79 if (is_bad_inode(inode)) {
81 return ERR_PTR(-ENOENT);
84 spin_lock(&block_group->lock);
85 if (!root->fs_info->closing) {
86 block_group->inode = igrab(inode);
87 block_group->iref = 1;
89 spin_unlock(&block_group->lock);
94 int create_free_space_inode(struct btrfs_root *root,
95 struct btrfs_trans_handle *trans,
96 struct btrfs_block_group_cache *block_group,
97 struct btrfs_path *path)
100 struct btrfs_disk_key disk_key;
101 struct btrfs_free_space_header *header;
102 struct btrfs_inode_item *inode_item;
103 struct extent_buffer *leaf;
107 ret = btrfs_find_free_objectid(trans, root, 0, &objectid);
111 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
115 leaf = path->nodes[0];
116 inode_item = btrfs_item_ptr(leaf, path->slots[0],
117 struct btrfs_inode_item);
118 btrfs_item_key(leaf, &disk_key, path->slots[0]);
119 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
120 sizeof(*inode_item));
121 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
122 btrfs_set_inode_size(leaf, inode_item, 0);
123 btrfs_set_inode_nbytes(leaf, inode_item, 0);
124 btrfs_set_inode_uid(leaf, inode_item, 0);
125 btrfs_set_inode_gid(leaf, inode_item, 0);
126 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
127 btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS |
128 BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM);
129 btrfs_set_inode_nlink(leaf, inode_item, 1);
130 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
131 btrfs_set_inode_block_group(leaf, inode_item,
132 block_group->key.objectid);
133 btrfs_mark_buffer_dirty(leaf);
134 btrfs_release_path(root, path);
136 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
137 key.offset = block_group->key.objectid;
140 ret = btrfs_insert_empty_item(trans, root, path, &key,
141 sizeof(struct btrfs_free_space_header));
143 btrfs_release_path(root, path);
146 leaf = path->nodes[0];
147 header = btrfs_item_ptr(leaf, path->slots[0],
148 struct btrfs_free_space_header);
149 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
150 btrfs_set_free_space_key(leaf, header, &disk_key);
151 btrfs_mark_buffer_dirty(leaf);
152 btrfs_release_path(root, path);
157 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
158 struct btrfs_trans_handle *trans,
159 struct btrfs_path *path,
165 trans->block_rsv = root->orphan_block_rsv;
166 ret = btrfs_block_rsv_check(trans, root,
167 root->orphan_block_rsv,
172 oldsize = i_size_read(inode);
173 btrfs_i_size_write(inode, 0);
174 truncate_pagecache(inode, oldsize, 0);
177 * We don't need an orphan item because truncating the free space cache
178 * will never be split across transactions.
180 ret = btrfs_truncate_inode_items(trans, root, inode,
181 0, BTRFS_EXTENT_DATA_KEY);
187 return btrfs_update_inode(trans, root, inode);
190 static int readahead_cache(struct inode *inode)
192 struct file_ra_state *ra;
193 unsigned long last_index;
195 ra = kzalloc(sizeof(*ra), GFP_NOFS);
199 file_ra_state_init(ra, inode->i_mapping);
200 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
202 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
209 int load_free_space_cache(struct btrfs_fs_info *fs_info,
210 struct btrfs_block_group_cache *block_group)
212 struct btrfs_root *root = fs_info->tree_root;
214 struct btrfs_free_space_header *header;
215 struct extent_buffer *leaf;
217 struct btrfs_path *path;
218 u32 *checksums = NULL, *crc;
219 char *disk_crcs = NULL;
220 struct btrfs_key key;
221 struct list_head bitmaps;
225 u32 cur_crc = ~(u32)0;
227 unsigned long first_page_offset;
232 * If we're unmounting then just return, since this does a search on the
233 * normal root and not the commit root and we could deadlock.
236 if (fs_info->closing)
240 * If this block group has been marked to be cleared for one reason or
241 * another then we can't trust the on disk cache, so just return.
243 spin_lock(&block_group->lock);
244 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
245 spin_unlock(&block_group->lock);
248 spin_unlock(&block_group->lock);
250 INIT_LIST_HEAD(&bitmaps);
252 path = btrfs_alloc_path();
256 inode = lookup_free_space_inode(root, block_group, path);
258 btrfs_free_path(path);
262 /* Nothing in the space cache, goodbye */
263 if (!i_size_read(inode)) {
264 btrfs_free_path(path);
268 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
269 key.offset = block_group->key.objectid;
272 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
274 btrfs_free_path(path);
278 leaf = path->nodes[0];
279 header = btrfs_item_ptr(leaf, path->slots[0],
280 struct btrfs_free_space_header);
281 num_entries = btrfs_free_space_entries(leaf, header);
282 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
283 generation = btrfs_free_space_generation(leaf, header);
284 btrfs_free_path(path);
286 if (BTRFS_I(inode)->generation != generation) {
287 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
288 " not match free space cache generation (%llu) for "
289 "block group %llu\n",
290 (unsigned long long)BTRFS_I(inode)->generation,
291 (unsigned long long)generation,
292 (unsigned long long)block_group->key.objectid);
299 /* Setup everything for doing checksumming */
300 num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
301 checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
304 first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
305 disk_crcs = kzalloc(first_page_offset, GFP_NOFS);
309 ret = readahead_cache(inode);
316 struct btrfs_free_space_entry *entry;
317 struct btrfs_free_space *e;
319 unsigned long offset = 0;
320 unsigned long start_offset = 0;
323 if (!num_entries && !num_bitmaps)
327 start_offset = first_page_offset;
328 offset = start_offset;
331 page = grab_cache_page(inode->i_mapping, index);
337 if (!PageUptodate(page)) {
338 btrfs_readpage(NULL, page);
340 if (!PageUptodate(page)) {
342 page_cache_release(page);
343 printk(KERN_ERR "btrfs: error reading free "
344 "space cache: %llu\n",
346 block_group->key.objectid);
355 memcpy(disk_crcs, addr, first_page_offset);
356 gen = addr + (sizeof(u32) * num_checksums);
357 if (*gen != BTRFS_I(inode)->generation) {
358 printk(KERN_ERR "btrfs: space cache generation"
359 " (%llu) does not match inode (%llu) "
360 "for block group %llu\n",
361 (unsigned long long)*gen,
363 BTRFS_I(inode)->generation,
365 block_group->key.objectid);
368 page_cache_release(page);
371 crc = (u32 *)disk_crcs;
373 entry = addr + start_offset;
375 /* First lets check our crc before we do anything fun */
377 cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc,
378 PAGE_CACHE_SIZE - start_offset);
379 btrfs_csum_final(cur_crc, (char *)&cur_crc);
380 if (cur_crc != *crc) {
381 printk(KERN_ERR "btrfs: crc mismatch for page %lu in "
382 "block group %llu\n", index,
383 (unsigned long long)block_group->key.objectid);
386 page_cache_release(page);
396 e = kmem_cache_zalloc(btrfs_free_space_cachep,
401 page_cache_release(page);
405 e->offset = le64_to_cpu(entry->offset);
406 e->bytes = le64_to_cpu(entry->bytes);
409 kmem_cache_free(btrfs_free_space_cachep, e);
411 page_cache_release(page);
415 if (entry->type == BTRFS_FREE_SPACE_EXTENT) {
416 spin_lock(&block_group->tree_lock);
417 ret = link_free_space(block_group, e);
418 spin_unlock(&block_group->tree_lock);
421 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
425 btrfs_free_space_cachep, e);
427 page_cache_release(page);
430 spin_lock(&block_group->tree_lock);
431 ret = link_free_space(block_group, e);
432 block_group->total_bitmaps++;
433 recalculate_thresholds(block_group);
434 spin_unlock(&block_group->tree_lock);
435 list_add_tail(&e->list, &bitmaps);
439 offset += sizeof(struct btrfs_free_space_entry);
440 if (offset + sizeof(struct btrfs_free_space_entry) >=
447 * We read an entry out of this page, we need to move on to the
456 * We add the bitmaps at the end of the entries in order that
457 * the bitmap entries are added to the cache.
459 e = list_entry(bitmaps.next, struct btrfs_free_space, list);
460 list_del_init(&e->list);
461 memcpy(e->bitmap, addr, PAGE_CACHE_SIZE);
466 page_cache_release(page);
478 /* This cache is bogus, make sure it gets cleared */
479 spin_lock(&block_group->lock);
480 block_group->disk_cache_state = BTRFS_DC_CLEAR;
481 spin_unlock(&block_group->lock);
482 btrfs_remove_free_space_cache(block_group);
486 int btrfs_write_out_cache(struct btrfs_root *root,
487 struct btrfs_trans_handle *trans,
488 struct btrfs_block_group_cache *block_group,
489 struct btrfs_path *path)
491 struct btrfs_free_space_header *header;
492 struct extent_buffer *leaf;
494 struct rb_node *node;
495 struct list_head *pos, *n;
497 struct extent_state *cached_state = NULL;
498 struct list_head bitmap_list;
499 struct btrfs_key key;
501 u32 *crc, *checksums;
502 pgoff_t index = 0, last_index = 0;
503 unsigned long first_page_offset;
509 root = root->fs_info->tree_root;
511 INIT_LIST_HEAD(&bitmap_list);
513 spin_lock(&block_group->lock);
514 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
515 spin_unlock(&block_group->lock);
518 spin_unlock(&block_group->lock);
520 inode = lookup_free_space_inode(root, block_group, path);
524 if (!i_size_read(inode)) {
529 node = rb_first(&block_group->free_space_offset);
535 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
536 filemap_write_and_wait(inode->i_mapping);
537 btrfs_wait_ordered_range(inode, inode->i_size &
538 ~(root->sectorsize - 1), (u64)-1);
540 /* We need a checksum per page. */
541 num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
542 crc = checksums = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
548 /* Since the first page has all of our checksums and our generation we
549 * need to calculate the offset into the page that we can start writing
552 first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
555 * Lock all pages first so we can lock the extent safely.
557 * NOTE: Because we hold the ref the entire time we're going to write to
558 * the page find_get_page should never fail, so we don't do a check
559 * after find_get_page at this point. Just putting this here so people
560 * know and don't freak out.
562 while (index <= last_index) {
563 page = grab_cache_page(inode->i_mapping, index);
568 page = find_get_page(inode->i_mapping, i);
570 page_cache_release(page);
571 page_cache_release(page);
580 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
581 0, &cached_state, GFP_NOFS);
583 /* Write out the extent entries */
585 struct btrfs_free_space_entry *entry;
587 unsigned long offset = 0;
588 unsigned long start_offset = 0;
591 start_offset = first_page_offset;
592 offset = start_offset;
595 page = find_get_page(inode->i_mapping, index);
598 entry = addr + start_offset;
600 memset(addr, 0, PAGE_CACHE_SIZE);
602 struct btrfs_free_space *e;
604 e = rb_entry(node, struct btrfs_free_space, offset_index);
607 entry->offset = cpu_to_le64(e->offset);
608 entry->bytes = cpu_to_le64(e->bytes);
610 entry->type = BTRFS_FREE_SPACE_BITMAP;
611 list_add_tail(&e->list, &bitmap_list);
614 entry->type = BTRFS_FREE_SPACE_EXTENT;
616 node = rb_next(node);
619 offset += sizeof(struct btrfs_free_space_entry);
620 if (offset + sizeof(struct btrfs_free_space_entry) >=
626 *crc = btrfs_csum_data(root, addr + start_offset, *crc,
627 PAGE_CACHE_SIZE - start_offset);
630 btrfs_csum_final(*crc, (char *)crc);
633 bytes += PAGE_CACHE_SIZE;
635 ClearPageChecked(page);
636 set_page_extent_mapped(page);
637 SetPageUptodate(page);
638 set_page_dirty(page);
641 * We need to release our reference we got for grab_cache_page,
642 * except for the first page which will hold our checksums, we
647 page_cache_release(page);
650 page_cache_release(page);
655 /* Write out the bitmaps */
656 list_for_each_safe(pos, n, &bitmap_list) {
658 struct btrfs_free_space *entry =
659 list_entry(pos, struct btrfs_free_space, list);
661 page = find_get_page(inode->i_mapping, index);
664 memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE);
666 *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE);
668 btrfs_csum_final(*crc, (char *)crc);
670 bytes += PAGE_CACHE_SIZE;
672 ClearPageChecked(page);
673 set_page_extent_mapped(page);
674 SetPageUptodate(page);
675 set_page_dirty(page);
677 page_cache_release(page);
678 page_cache_release(page);
679 list_del_init(&entry->list);
683 /* Zero out the rest of the pages just to make sure */
684 while (index <= last_index) {
687 page = find_get_page(inode->i_mapping, index);
690 memset(addr, 0, PAGE_CACHE_SIZE);
692 ClearPageChecked(page);
693 set_page_extent_mapped(page);
694 SetPageUptodate(page);
695 set_page_dirty(page);
697 page_cache_release(page);
698 page_cache_release(page);
699 bytes += PAGE_CACHE_SIZE;
703 btrfs_set_extent_delalloc(inode, 0, bytes - 1, &cached_state);
705 /* Write the checksums and trans id to the first page */
710 page = find_get_page(inode->i_mapping, 0);
713 memcpy(addr, checksums, sizeof(u32) * num_checksums);
714 gen = addr + (sizeof(u32) * num_checksums);
715 *gen = trans->transid;
717 ClearPageChecked(page);
718 set_page_extent_mapped(page);
719 SetPageUptodate(page);
720 set_page_dirty(page);
722 page_cache_release(page);
723 page_cache_release(page);
725 BTRFS_I(inode)->generation = trans->transid;
727 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
728 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
730 filemap_write_and_wait(inode->i_mapping);
732 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
733 key.offset = block_group->key.objectid;
736 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
739 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
740 EXTENT_DIRTY | EXTENT_DELALLOC |
741 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
744 leaf = path->nodes[0];
746 struct btrfs_key found_key;
747 BUG_ON(!path->slots[0]);
749 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
750 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
751 found_key.offset != block_group->key.objectid) {
753 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
754 EXTENT_DIRTY | EXTENT_DELALLOC |
755 EXTENT_DO_ACCOUNTING, 0, 0, NULL,
757 btrfs_release_path(root, path);
761 header = btrfs_item_ptr(leaf, path->slots[0],
762 struct btrfs_free_space_header);
763 btrfs_set_free_space_entries(leaf, header, entries);
764 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
765 btrfs_set_free_space_generation(leaf, header, trans->transid);
766 btrfs_mark_buffer_dirty(leaf);
767 btrfs_release_path(root, path);
773 invalidate_inode_pages2_range(inode->i_mapping, 0, index);
774 spin_lock(&block_group->lock);
775 block_group->disk_cache_state = BTRFS_DC_ERROR;
776 spin_unlock(&block_group->lock);
777 BTRFS_I(inode)->generation = 0;
780 btrfs_update_inode(trans, root, inode);
785 static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize,
788 BUG_ON(offset < bitmap_start);
789 offset -= bitmap_start;
790 return (unsigned long)(div64_u64(offset, sectorsize));
793 static inline unsigned long bytes_to_bits(u64 bytes, u64 sectorsize)
795 return (unsigned long)(div64_u64(bytes, sectorsize));
798 static inline u64 offset_to_bitmap(struct btrfs_block_group_cache *block_group,
802 u64 bytes_per_bitmap;
804 bytes_per_bitmap = BITS_PER_BITMAP * block_group->sectorsize;
805 bitmap_start = offset - block_group->key.objectid;
806 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
807 bitmap_start *= bytes_per_bitmap;
808 bitmap_start += block_group->key.objectid;
813 static int tree_insert_offset(struct rb_root *root, u64 offset,
814 struct rb_node *node, int bitmap)
816 struct rb_node **p = &root->rb_node;
817 struct rb_node *parent = NULL;
818 struct btrfs_free_space *info;
822 info = rb_entry(parent, struct btrfs_free_space, offset_index);
824 if (offset < info->offset) {
826 } else if (offset > info->offset) {
830 * we could have a bitmap entry and an extent entry
831 * share the same offset. If this is the case, we want
832 * the extent entry to always be found first if we do a
833 * linear search through the tree, since we want to have
834 * the quickest allocation time, and allocating from an
835 * extent is faster than allocating from a bitmap. So
836 * if we're inserting a bitmap and we find an entry at
837 * this offset, we want to go right, or after this entry
838 * logically. If we are inserting an extent and we've
839 * found a bitmap, we want to go left, or before
843 WARN_ON(info->bitmap);
846 WARN_ON(!info->bitmap);
852 rb_link_node(node, parent, p);
853 rb_insert_color(node, root);
859 * searches the tree for the given offset.
861 * fuzzy - If this is set, then we are trying to make an allocation, and we just
862 * want a section that has at least bytes size and comes at or after the given
865 static struct btrfs_free_space *
866 tree_search_offset(struct btrfs_block_group_cache *block_group,
867 u64 offset, int bitmap_only, int fuzzy)
869 struct rb_node *n = block_group->free_space_offset.rb_node;
870 struct btrfs_free_space *entry, *prev = NULL;
872 /* find entry that is closest to the 'offset' */
879 entry = rb_entry(n, struct btrfs_free_space, offset_index);
882 if (offset < entry->offset)
884 else if (offset > entry->offset)
897 * bitmap entry and extent entry may share same offset,
898 * in that case, bitmap entry comes after extent entry.
903 entry = rb_entry(n, struct btrfs_free_space, offset_index);
904 if (entry->offset != offset)
907 WARN_ON(!entry->bitmap);
912 * if previous extent entry covers the offset,
913 * we should return it instead of the bitmap entry
915 n = &entry->offset_index;
920 prev = rb_entry(n, struct btrfs_free_space,
923 if (prev->offset + prev->bytes > offset)
935 /* find last entry before the 'offset' */
937 if (entry->offset > offset) {
938 n = rb_prev(&entry->offset_index);
940 entry = rb_entry(n, struct btrfs_free_space,
942 BUG_ON(entry->offset > offset);
952 n = &entry->offset_index;
957 prev = rb_entry(n, struct btrfs_free_space,
960 if (prev->offset + prev->bytes > offset)
965 if (entry->offset + BITS_PER_BITMAP *
966 block_group->sectorsize > offset)
968 } else if (entry->offset + entry->bytes > offset)
976 if (entry->offset + BITS_PER_BITMAP *
977 block_group->sectorsize > offset)
980 if (entry->offset + entry->bytes > offset)
984 n = rb_next(&entry->offset_index);
987 entry = rb_entry(n, struct btrfs_free_space, offset_index);
993 __unlink_free_space(struct btrfs_block_group_cache *block_group,
994 struct btrfs_free_space *info)
996 rb_erase(&info->offset_index, &block_group->free_space_offset);
997 block_group->free_extents--;
1000 static void unlink_free_space(struct btrfs_block_group_cache *block_group,
1001 struct btrfs_free_space *info)
1003 __unlink_free_space(block_group, info);
1004 block_group->free_space -= info->bytes;
1007 static int link_free_space(struct btrfs_block_group_cache *block_group,
1008 struct btrfs_free_space *info)
1012 BUG_ON(!info->bitmap && !info->bytes);
1013 ret = tree_insert_offset(&block_group->free_space_offset, info->offset,
1014 &info->offset_index, (info->bitmap != NULL));
1018 block_group->free_space += info->bytes;
1019 block_group->free_extents++;
1023 static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
1028 u64 size = block_group->key.offset;
1031 * The goal is to keep the total amount of memory used per 1gb of space
1032 * at or below 32k, so we need to adjust how much memory we allow to be
1033 * used by extent based free space tracking
1035 if (size < 1024 * 1024 * 1024)
1036 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1038 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1039 div64_u64(size, 1024 * 1024 * 1024);
1042 * we want to account for 1 more bitmap than what we have so we can make
1043 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1044 * we add more bitmaps.
1046 bitmap_bytes = (block_group->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1048 if (bitmap_bytes >= max_bytes) {
1049 block_group->extents_thresh = 0;
1054 * we want the extent entry threshold to always be at most 1/2 the maxw
1055 * bytes we can have, or whatever is less than that.
1057 extent_bytes = max_bytes - bitmap_bytes;
1058 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1060 block_group->extents_thresh =
1061 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1064 static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group,
1065 struct btrfs_free_space *info, u64 offset,
1068 unsigned long start, end;
1071 start = offset_to_bit(info->offset, block_group->sectorsize, offset);
1072 end = start + bytes_to_bits(bytes, block_group->sectorsize);
1073 BUG_ON(end > BITS_PER_BITMAP);
1075 for (i = start; i < end; i++)
1076 clear_bit(i, info->bitmap);
1078 info->bytes -= bytes;
1079 block_group->free_space -= bytes;
1082 static void bitmap_set_bits(struct btrfs_block_group_cache *block_group,
1083 struct btrfs_free_space *info, u64 offset,
1086 unsigned long start, end;
1089 start = offset_to_bit(info->offset, block_group->sectorsize, offset);
1090 end = start + bytes_to_bits(bytes, block_group->sectorsize);
1091 BUG_ON(end > BITS_PER_BITMAP);
1093 for (i = start; i < end; i++)
1094 set_bit(i, info->bitmap);
1096 info->bytes += bytes;
1097 block_group->free_space += bytes;
1100 static int search_bitmap(struct btrfs_block_group_cache *block_group,
1101 struct btrfs_free_space *bitmap_info, u64 *offset,
1104 unsigned long found_bits = 0;
1105 unsigned long bits, i;
1106 unsigned long next_zero;
1108 i = offset_to_bit(bitmap_info->offset, block_group->sectorsize,
1109 max_t(u64, *offset, bitmap_info->offset));
1110 bits = bytes_to_bits(*bytes, block_group->sectorsize);
1112 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1113 i < BITS_PER_BITMAP;
1114 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1115 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1116 BITS_PER_BITMAP, i);
1117 if ((next_zero - i) >= bits) {
1118 found_bits = next_zero - i;
1125 *offset = (u64)(i * block_group->sectorsize) +
1126 bitmap_info->offset;
1127 *bytes = (u64)(found_bits) * block_group->sectorsize;
1134 static struct btrfs_free_space *find_free_space(struct btrfs_block_group_cache
1135 *block_group, u64 *offset,
1136 u64 *bytes, int debug)
1138 struct btrfs_free_space *entry;
1139 struct rb_node *node;
1142 if (!block_group->free_space_offset.rb_node)
1145 entry = tree_search_offset(block_group,
1146 offset_to_bitmap(block_group, *offset),
1151 for (node = &entry->offset_index; node; node = rb_next(node)) {
1152 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1153 if (entry->bytes < *bytes)
1156 if (entry->bitmap) {
1157 ret = search_bitmap(block_group, entry, offset, bytes);
1163 *offset = entry->offset;
1164 *bytes = entry->bytes;
1171 static void add_new_bitmap(struct btrfs_block_group_cache *block_group,
1172 struct btrfs_free_space *info, u64 offset)
1174 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1175 int max_bitmaps = (int)div64_u64(block_group->key.offset +
1176 bytes_per_bg - 1, bytes_per_bg);
1177 BUG_ON(block_group->total_bitmaps >= max_bitmaps);
1179 info->offset = offset_to_bitmap(block_group, offset);
1181 link_free_space(block_group, info);
1182 block_group->total_bitmaps++;
1184 recalculate_thresholds(block_group);
1187 static void free_bitmap(struct btrfs_block_group_cache *block_group,
1188 struct btrfs_free_space *bitmap_info)
1190 unlink_free_space(block_group, bitmap_info);
1191 kfree(bitmap_info->bitmap);
1192 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1193 block_group->total_bitmaps--;
1194 recalculate_thresholds(block_group);
1197 static noinline int remove_from_bitmap(struct btrfs_block_group_cache *block_group,
1198 struct btrfs_free_space *bitmap_info,
1199 u64 *offset, u64 *bytes)
1202 u64 search_start, search_bytes;
1206 end = bitmap_info->offset +
1207 (u64)(BITS_PER_BITMAP * block_group->sectorsize) - 1;
1210 * XXX - this can go away after a few releases.
1212 * since the only user of btrfs_remove_free_space is the tree logging
1213 * stuff, and the only way to test that is under crash conditions, we
1214 * want to have this debug stuff here just in case somethings not
1215 * working. Search the bitmap for the space we are trying to use to
1216 * make sure its actually there. If its not there then we need to stop
1217 * because something has gone wrong.
1219 search_start = *offset;
1220 search_bytes = *bytes;
1221 search_bytes = min(search_bytes, end - search_start + 1);
1222 ret = search_bitmap(block_group, bitmap_info, &search_start,
1224 BUG_ON(ret < 0 || search_start != *offset);
1226 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1227 bitmap_clear_bits(block_group, bitmap_info, *offset,
1229 *bytes -= end - *offset + 1;
1231 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1232 bitmap_clear_bits(block_group, bitmap_info, *offset, *bytes);
1237 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1238 if (!bitmap_info->bytes)
1239 free_bitmap(block_group, bitmap_info);
1242 * no entry after this bitmap, but we still have bytes to
1243 * remove, so something has gone wrong.
1248 bitmap_info = rb_entry(next, struct btrfs_free_space,
1252 * if the next entry isn't a bitmap we need to return to let the
1253 * extent stuff do its work.
1255 if (!bitmap_info->bitmap)
1259 * Ok the next item is a bitmap, but it may not actually hold
1260 * the information for the rest of this free space stuff, so
1261 * look for it, and if we don't find it return so we can try
1262 * everything over again.
1264 search_start = *offset;
1265 search_bytes = *bytes;
1266 ret = search_bitmap(block_group, bitmap_info, &search_start,
1268 if (ret < 0 || search_start != *offset)
1272 } else if (!bitmap_info->bytes)
1273 free_bitmap(block_group, bitmap_info);
1278 static int insert_into_bitmap(struct btrfs_block_group_cache *block_group,
1279 struct btrfs_free_space *info)
1281 struct btrfs_free_space *bitmap_info;
1283 u64 bytes, offset, end;
1287 * If we are below the extents threshold then we can add this as an
1288 * extent, and don't have to deal with the bitmap
1290 if (block_group->free_extents < block_group->extents_thresh) {
1292 * If this block group has some small extents we don't want to
1293 * use up all of our free slots in the cache with them, we want
1294 * to reserve them to larger extents, however if we have plent
1295 * of cache left then go ahead an dadd them, no sense in adding
1296 * the overhead of a bitmap if we don't have to.
1298 if (info->bytes <= block_group->sectorsize * 4) {
1299 if (block_group->free_extents * 2 <=
1300 block_group->extents_thresh)
1308 * some block groups are so tiny they can't be enveloped by a bitmap, so
1309 * don't even bother to create a bitmap for this
1311 if (BITS_PER_BITMAP * block_group->sectorsize >
1312 block_group->key.offset)
1315 bytes = info->bytes;
1316 offset = info->offset;
1319 bitmap_info = tree_search_offset(block_group,
1320 offset_to_bitmap(block_group, offset),
1327 end = bitmap_info->offset +
1328 (u64)(BITS_PER_BITMAP * block_group->sectorsize);
1330 if (offset >= bitmap_info->offset && offset + bytes > end) {
1331 bitmap_set_bits(block_group, bitmap_info, offset,
1333 bytes -= end - offset;
1336 } else if (offset >= bitmap_info->offset && offset + bytes <= end) {
1337 bitmap_set_bits(block_group, bitmap_info, offset, bytes);
1350 if (info && info->bitmap) {
1351 add_new_bitmap(block_group, info, offset);
1356 spin_unlock(&block_group->tree_lock);
1358 /* no pre-allocated info, allocate a new one */
1360 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1363 spin_lock(&block_group->tree_lock);
1369 /* allocate the bitmap */
1370 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1371 spin_lock(&block_group->tree_lock);
1372 if (!info->bitmap) {
1382 kfree(info->bitmap);
1383 kmem_cache_free(btrfs_free_space_cachep, info);
1389 bool try_merge_free_space(struct btrfs_block_group_cache *block_group,
1390 struct btrfs_free_space *info, bool update_stat)
1392 struct btrfs_free_space *left_info;
1393 struct btrfs_free_space *right_info;
1394 bool merged = false;
1395 u64 offset = info->offset;
1396 u64 bytes = info->bytes;
1399 * first we want to see if there is free space adjacent to the range we
1400 * are adding, if there is remove that struct and add a new one to
1401 * cover the entire range
1403 right_info = tree_search_offset(block_group, offset + bytes, 0, 0);
1404 if (right_info && rb_prev(&right_info->offset_index))
1405 left_info = rb_entry(rb_prev(&right_info->offset_index),
1406 struct btrfs_free_space, offset_index);
1408 left_info = tree_search_offset(block_group, offset - 1, 0, 0);
1410 if (right_info && !right_info->bitmap) {
1412 unlink_free_space(block_group, right_info);
1414 __unlink_free_space(block_group, right_info);
1415 info->bytes += right_info->bytes;
1416 kmem_cache_free(btrfs_free_space_cachep, right_info);
1420 if (left_info && !left_info->bitmap &&
1421 left_info->offset + left_info->bytes == offset) {
1423 unlink_free_space(block_group, left_info);
1425 __unlink_free_space(block_group, left_info);
1426 info->offset = left_info->offset;
1427 info->bytes += left_info->bytes;
1428 kmem_cache_free(btrfs_free_space_cachep, left_info);
1435 int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
1436 u64 offset, u64 bytes)
1438 struct btrfs_free_space *info;
1441 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1445 info->offset = offset;
1446 info->bytes = bytes;
1448 spin_lock(&block_group->tree_lock);
1450 if (try_merge_free_space(block_group, info, true))
1454 * There was no extent directly to the left or right of this new
1455 * extent then we know we're going to have to allocate a new extent, so
1456 * before we do that see if we need to drop this into a bitmap
1458 ret = insert_into_bitmap(block_group, info);
1466 ret = link_free_space(block_group, info);
1468 kmem_cache_free(btrfs_free_space_cachep, info);
1470 spin_unlock(&block_group->tree_lock);
1473 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1474 BUG_ON(ret == -EEXIST);
1480 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1481 u64 offset, u64 bytes)
1483 struct btrfs_free_space *info;
1484 struct btrfs_free_space *next_info = NULL;
1487 spin_lock(&block_group->tree_lock);
1490 info = tree_search_offset(block_group, offset, 0, 0);
1493 * oops didn't find an extent that matched the space we wanted
1494 * to remove, look for a bitmap instead
1496 info = tree_search_offset(block_group,
1497 offset_to_bitmap(block_group, offset),
1505 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1507 next_info = rb_entry(rb_next(&info->offset_index),
1508 struct btrfs_free_space,
1511 if (next_info->bitmap)
1512 end = next_info->offset + BITS_PER_BITMAP *
1513 block_group->sectorsize - 1;
1515 end = next_info->offset + next_info->bytes;
1517 if (next_info->bytes < bytes ||
1518 next_info->offset > offset || offset > end) {
1519 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1520 " trying to use %llu\n",
1521 (unsigned long long)info->offset,
1522 (unsigned long long)info->bytes,
1523 (unsigned long long)bytes);
1532 if (info->bytes == bytes) {
1533 unlink_free_space(block_group, info);
1535 kfree(info->bitmap);
1536 block_group->total_bitmaps--;
1538 kmem_cache_free(btrfs_free_space_cachep, info);
1542 if (!info->bitmap && info->offset == offset) {
1543 unlink_free_space(block_group, info);
1544 info->offset += bytes;
1545 info->bytes -= bytes;
1546 link_free_space(block_group, info);
1550 if (!info->bitmap && info->offset <= offset &&
1551 info->offset + info->bytes >= offset + bytes) {
1552 u64 old_start = info->offset;
1554 * we're freeing space in the middle of the info,
1555 * this can happen during tree log replay
1557 * first unlink the old info and then
1558 * insert it again after the hole we're creating
1560 unlink_free_space(block_group, info);
1561 if (offset + bytes < info->offset + info->bytes) {
1562 u64 old_end = info->offset + info->bytes;
1564 info->offset = offset + bytes;
1565 info->bytes = old_end - info->offset;
1566 ret = link_free_space(block_group, info);
1571 /* the hole we're creating ends at the end
1572 * of the info struct, just free the info
1574 kmem_cache_free(btrfs_free_space_cachep, info);
1576 spin_unlock(&block_group->tree_lock);
1578 /* step two, insert a new info struct to cover
1579 * anything before the hole
1581 ret = btrfs_add_free_space(block_group, old_start,
1582 offset - old_start);
1587 ret = remove_from_bitmap(block_group, info, &offset, &bytes);
1592 spin_unlock(&block_group->tree_lock);
1597 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1600 struct btrfs_free_space *info;
1604 for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) {
1605 info = rb_entry(n, struct btrfs_free_space, offset_index);
1606 if (info->bytes >= bytes)
1608 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1609 (unsigned long long)info->offset,
1610 (unsigned long long)info->bytes,
1611 (info->bitmap) ? "yes" : "no");
1613 printk(KERN_INFO "block group has cluster?: %s\n",
1614 list_empty(&block_group->cluster_list) ? "no" : "yes");
1615 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1619 u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group)
1621 struct btrfs_free_space *info;
1625 for (n = rb_first(&block_group->free_space_offset); n;
1627 info = rb_entry(n, struct btrfs_free_space, offset_index);
1635 * for a given cluster, put all of its extents back into the free
1636 * space cache. If the block group passed doesn't match the block group
1637 * pointed to by the cluster, someone else raced in and freed the
1638 * cluster already. In that case, we just return without changing anything
1641 __btrfs_return_cluster_to_free_space(
1642 struct btrfs_block_group_cache *block_group,
1643 struct btrfs_free_cluster *cluster)
1645 struct btrfs_free_space *entry;
1646 struct rb_node *node;
1648 spin_lock(&cluster->lock);
1649 if (cluster->block_group != block_group)
1652 cluster->block_group = NULL;
1653 cluster->window_start = 0;
1654 list_del_init(&cluster->block_group_list);
1656 node = rb_first(&cluster->root);
1660 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1661 node = rb_next(&entry->offset_index);
1662 rb_erase(&entry->offset_index, &cluster->root);
1664 bitmap = (entry->bitmap != NULL);
1666 try_merge_free_space(block_group, entry, false);
1667 tree_insert_offset(&block_group->free_space_offset,
1668 entry->offset, &entry->offset_index, bitmap);
1670 cluster->root = RB_ROOT;
1673 spin_unlock(&cluster->lock);
1674 btrfs_put_block_group(block_group);
1678 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
1680 struct btrfs_free_space *info;
1681 struct rb_node *node;
1682 struct btrfs_free_cluster *cluster;
1683 struct list_head *head;
1685 spin_lock(&block_group->tree_lock);
1686 while ((head = block_group->cluster_list.next) !=
1687 &block_group->cluster_list) {
1688 cluster = list_entry(head, struct btrfs_free_cluster,
1691 WARN_ON(cluster->block_group != block_group);
1692 __btrfs_return_cluster_to_free_space(block_group, cluster);
1693 if (need_resched()) {
1694 spin_unlock(&block_group->tree_lock);
1696 spin_lock(&block_group->tree_lock);
1700 while ((node = rb_last(&block_group->free_space_offset)) != NULL) {
1701 info = rb_entry(node, struct btrfs_free_space, offset_index);
1702 unlink_free_space(block_group, info);
1704 kfree(info->bitmap);
1705 kmem_cache_free(btrfs_free_space_cachep, info);
1706 if (need_resched()) {
1707 spin_unlock(&block_group->tree_lock);
1709 spin_lock(&block_group->tree_lock);
1713 spin_unlock(&block_group->tree_lock);
1716 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
1717 u64 offset, u64 bytes, u64 empty_size)
1719 struct btrfs_free_space *entry = NULL;
1720 u64 bytes_search = bytes + empty_size;
1723 spin_lock(&block_group->tree_lock);
1724 entry = find_free_space(block_group, &offset, &bytes_search, 0);
1729 if (entry->bitmap) {
1730 bitmap_clear_bits(block_group, entry, offset, bytes);
1732 free_bitmap(block_group, entry);
1734 unlink_free_space(block_group, entry);
1735 entry->offset += bytes;
1736 entry->bytes -= bytes;
1738 kmem_cache_free(btrfs_free_space_cachep, entry);
1740 link_free_space(block_group, entry);
1744 spin_unlock(&block_group->tree_lock);
1750 * given a cluster, put all of its extents back into the free space
1751 * cache. If a block group is passed, this function will only free
1752 * a cluster that belongs to the passed block group.
1754 * Otherwise, it'll get a reference on the block group pointed to by the
1755 * cluster and remove the cluster from it.
1757 int btrfs_return_cluster_to_free_space(
1758 struct btrfs_block_group_cache *block_group,
1759 struct btrfs_free_cluster *cluster)
1763 /* first, get a safe pointer to the block group */
1764 spin_lock(&cluster->lock);
1766 block_group = cluster->block_group;
1768 spin_unlock(&cluster->lock);
1771 } else if (cluster->block_group != block_group) {
1772 /* someone else has already freed it don't redo their work */
1773 spin_unlock(&cluster->lock);
1776 atomic_inc(&block_group->count);
1777 spin_unlock(&cluster->lock);
1779 /* now return any extents the cluster had on it */
1780 spin_lock(&block_group->tree_lock);
1781 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
1782 spin_unlock(&block_group->tree_lock);
1784 /* finally drop our ref */
1785 btrfs_put_block_group(block_group);
1789 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
1790 struct btrfs_free_cluster *cluster,
1791 struct btrfs_free_space *entry,
1792 u64 bytes, u64 min_start)
1795 u64 search_start = cluster->window_start;
1796 u64 search_bytes = bytes;
1799 search_start = min_start;
1800 search_bytes = bytes;
1802 err = search_bitmap(block_group, entry, &search_start,
1808 bitmap_clear_bits(block_group, entry, ret, bytes);
1814 * given a cluster, try to allocate 'bytes' from it, returns 0
1815 * if it couldn't find anything suitably large, or a logical disk offset
1816 * if things worked out
1818 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
1819 struct btrfs_free_cluster *cluster, u64 bytes,
1822 struct btrfs_free_space *entry = NULL;
1823 struct rb_node *node;
1826 spin_lock(&cluster->lock);
1827 if (bytes > cluster->max_size)
1830 if (cluster->block_group != block_group)
1833 node = rb_first(&cluster->root);
1837 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1839 if (entry->bytes < bytes ||
1840 (!entry->bitmap && entry->offset < min_start)) {
1841 struct rb_node *node;
1843 node = rb_next(&entry->offset_index);
1846 entry = rb_entry(node, struct btrfs_free_space,
1851 if (entry->bitmap) {
1852 ret = btrfs_alloc_from_bitmap(block_group,
1853 cluster, entry, bytes,
1856 struct rb_node *node;
1857 node = rb_next(&entry->offset_index);
1860 entry = rb_entry(node, struct btrfs_free_space,
1866 ret = entry->offset;
1868 entry->offset += bytes;
1869 entry->bytes -= bytes;
1872 if (entry->bytes == 0)
1873 rb_erase(&entry->offset_index, &cluster->root);
1877 spin_unlock(&cluster->lock);
1882 spin_lock(&block_group->tree_lock);
1884 block_group->free_space -= bytes;
1885 if (entry->bytes == 0) {
1886 block_group->free_extents--;
1887 if (entry->bitmap) {
1888 kfree(entry->bitmap);
1889 block_group->total_bitmaps--;
1890 recalculate_thresholds(block_group);
1892 kmem_cache_free(btrfs_free_space_cachep, entry);
1895 spin_unlock(&block_group->tree_lock);
1900 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
1901 struct btrfs_free_space *entry,
1902 struct btrfs_free_cluster *cluster,
1903 u64 offset, u64 bytes, u64 min_bytes)
1905 unsigned long next_zero;
1907 unsigned long search_bits;
1908 unsigned long total_bits;
1909 unsigned long found_bits;
1910 unsigned long start = 0;
1911 unsigned long total_found = 0;
1915 i = offset_to_bit(entry->offset, block_group->sectorsize,
1916 max_t(u64, offset, entry->offset));
1917 search_bits = bytes_to_bits(bytes, block_group->sectorsize);
1918 total_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
1922 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
1923 i < BITS_PER_BITMAP;
1924 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
1925 next_zero = find_next_zero_bit(entry->bitmap,
1926 BITS_PER_BITMAP, i);
1927 if (next_zero - i >= search_bits) {
1928 found_bits = next_zero - i;
1942 total_found += found_bits;
1944 if (cluster->max_size < found_bits * block_group->sectorsize)
1945 cluster->max_size = found_bits * block_group->sectorsize;
1947 if (total_found < total_bits) {
1948 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
1949 if (i - start > total_bits * 2) {
1951 cluster->max_size = 0;
1957 cluster->window_start = start * block_group->sectorsize +
1959 rb_erase(&entry->offset_index, &block_group->free_space_offset);
1960 ret = tree_insert_offset(&cluster->root, entry->offset,
1961 &entry->offset_index, 1);
1968 * This searches the block group for just extents to fill the cluster with.
1970 static int setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
1971 struct btrfs_free_cluster *cluster,
1972 u64 offset, u64 bytes, u64 min_bytes)
1974 struct btrfs_free_space *first = NULL;
1975 struct btrfs_free_space *entry = NULL;
1976 struct btrfs_free_space *prev = NULL;
1977 struct btrfs_free_space *last;
1978 struct rb_node *node;
1982 u64 max_gap = 128 * 1024;
1984 entry = tree_search_offset(block_group, offset, 0, 1);
1989 * We don't want bitmaps, so just move along until we find a normal
1992 while (entry->bitmap) {
1993 node = rb_next(&entry->offset_index);
1996 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1999 window_start = entry->offset;
2000 window_free = entry->bytes;
2001 max_extent = entry->bytes;
2006 while (window_free <= min_bytes) {
2007 node = rb_next(&entry->offset_index);
2010 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2015 * we haven't filled the empty size and the window is
2016 * very large. reset and try again
2018 if (entry->offset - (prev->offset + prev->bytes) > max_gap ||
2019 entry->offset - window_start > (min_bytes * 2)) {
2021 window_start = entry->offset;
2022 window_free = entry->bytes;
2024 max_extent = entry->bytes;
2027 window_free += entry->bytes;
2028 if (entry->bytes > max_extent)
2029 max_extent = entry->bytes;
2034 cluster->window_start = first->offset;
2036 node = &first->offset_index;
2039 * now we've found our entries, pull them out of the free space
2040 * cache and put them into the cluster rbtree
2045 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2046 node = rb_next(&entry->offset_index);
2050 rb_erase(&entry->offset_index, &block_group->free_space_offset);
2051 ret = tree_insert_offset(&cluster->root, entry->offset,
2052 &entry->offset_index, 0);
2054 } while (node && entry != last);
2056 cluster->max_size = max_extent;
2062 * This specifically looks for bitmaps that may work in the cluster, we assume
2063 * that we have already failed to find extents that will work.
2065 static int setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2066 struct btrfs_free_cluster *cluster,
2067 u64 offset, u64 bytes, u64 min_bytes)
2069 struct btrfs_free_space *entry;
2070 struct rb_node *node;
2073 if (block_group->total_bitmaps == 0)
2076 entry = tree_search_offset(block_group,
2077 offset_to_bitmap(block_group, offset),
2082 node = &entry->offset_index;
2084 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2085 node = rb_next(&entry->offset_index);
2088 if (entry->bytes < min_bytes)
2090 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2092 } while (ret && node);
2098 * here we try to find a cluster of blocks in a block group. The goal
2099 * is to find at least bytes free and up to empty_size + bytes free.
2100 * We might not find them all in one contiguous area.
2102 * returns zero and sets up cluster if things worked out, otherwise
2103 * it returns -enospc
2105 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *root,
2107 struct btrfs_block_group_cache *block_group,
2108 struct btrfs_free_cluster *cluster,
2109 u64 offset, u64 bytes, u64 empty_size)
2114 /* for metadata, allow allocates with more holes */
2115 if (btrfs_test_opt(root, SSD_SPREAD)) {
2116 min_bytes = bytes + empty_size;
2117 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2119 * we want to do larger allocations when we are
2120 * flushing out the delayed refs, it helps prevent
2121 * making more work as we go along.
2123 if (trans->transaction->delayed_refs.flushing)
2124 min_bytes = max(bytes, (bytes + empty_size) >> 1);
2126 min_bytes = max(bytes, (bytes + empty_size) >> 4);
2128 min_bytes = max(bytes, (bytes + empty_size) >> 2);
2130 spin_lock(&block_group->tree_lock);
2133 * If we know we don't have enough space to make a cluster don't even
2134 * bother doing all the work to try and find one.
2136 if (block_group->free_space < min_bytes) {
2137 spin_unlock(&block_group->tree_lock);
2141 spin_lock(&cluster->lock);
2143 /* someone already found a cluster, hooray */
2144 if (cluster->block_group) {
2149 ret = setup_cluster_no_bitmap(block_group, cluster, offset, bytes,
2152 ret = setup_cluster_bitmap(block_group, cluster, offset,
2156 atomic_inc(&block_group->count);
2157 list_add_tail(&cluster->block_group_list,
2158 &block_group->cluster_list);
2159 cluster->block_group = block_group;
2162 spin_unlock(&cluster->lock);
2163 spin_unlock(&block_group->tree_lock);
2169 * simple code to zero out a cluster
2171 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2173 spin_lock_init(&cluster->lock);
2174 spin_lock_init(&cluster->refill_lock);
2175 cluster->root = RB_ROOT;
2176 cluster->max_size = 0;
2177 INIT_LIST_HEAD(&cluster->block_group_list);
2178 cluster->block_group = NULL;
2181 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2182 u64 *trimmed, u64 start, u64 end, u64 minlen)
2184 struct btrfs_free_space *entry = NULL;
2185 struct btrfs_fs_info *fs_info = block_group->fs_info;
2187 u64 actually_trimmed;
2192 while (start < end) {
2193 spin_lock(&block_group->tree_lock);
2195 if (block_group->free_space < minlen) {
2196 spin_unlock(&block_group->tree_lock);
2200 entry = tree_search_offset(block_group, start, 0, 1);
2202 entry = tree_search_offset(block_group,
2203 offset_to_bitmap(block_group,
2207 if (!entry || entry->offset >= end) {
2208 spin_unlock(&block_group->tree_lock);
2212 if (entry->bitmap) {
2213 ret = search_bitmap(block_group, entry, &start, &bytes);
2216 spin_unlock(&block_group->tree_lock);
2219 bytes = min(bytes, end - start);
2220 bitmap_clear_bits(block_group, entry,
2222 if (entry->bytes == 0)
2223 free_bitmap(block_group, entry);
2225 start = entry->offset + BITS_PER_BITMAP *
2226 block_group->sectorsize;
2227 spin_unlock(&block_group->tree_lock);
2232 start = entry->offset;
2233 bytes = min(entry->bytes, end - start);
2234 unlink_free_space(block_group, entry);
2238 spin_unlock(&block_group->tree_lock);
2240 if (bytes >= minlen) {
2242 update_ret = btrfs_update_reserved_bytes(block_group,
2245 ret = btrfs_error_discard_extent(fs_info->extent_root,
2250 btrfs_add_free_space(block_group,
2253 btrfs_update_reserved_bytes(block_group,
2258 *trimmed += actually_trimmed;
2263 if (fatal_signal_pending(current)) {