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>
23 #include <linux/ratelimit.h>
25 #include "free-space-cache.h"
26 #include "transaction.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
31 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
34 static int link_free_space(struct btrfs_free_space_ctl *ctl,
35 struct btrfs_free_space *info);
36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
37 struct btrfs_free_space *info);
39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
40 struct btrfs_path *path,
44 struct btrfs_key location;
45 struct btrfs_disk_key disk_key;
46 struct btrfs_free_space_header *header;
47 struct extent_buffer *leaf;
48 struct inode *inode = NULL;
51 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
55 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
59 btrfs_release_path(path);
60 return ERR_PTR(-ENOENT);
63 leaf = path->nodes[0];
64 header = btrfs_item_ptr(leaf, path->slots[0],
65 struct btrfs_free_space_header);
66 btrfs_free_space_key(leaf, header, &disk_key);
67 btrfs_disk_key_to_cpu(&location, &disk_key);
68 btrfs_release_path(path);
70 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
72 return ERR_PTR(-ENOENT);
75 if (is_bad_inode(inode)) {
77 return ERR_PTR(-ENOENT);
80 mapping_set_gfp_mask(inode->i_mapping,
81 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
86 struct inode *lookup_free_space_inode(struct btrfs_root *root,
87 struct btrfs_block_group_cache
88 *block_group, struct btrfs_path *path)
90 struct inode *inode = NULL;
91 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
93 spin_lock(&block_group->lock);
94 if (block_group->inode)
95 inode = igrab(block_group->inode);
96 spin_unlock(&block_group->lock);
100 inode = __lookup_free_space_inode(root, path,
101 block_group->key.objectid);
105 spin_lock(&block_group->lock);
106 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
107 btrfs_info(root->fs_info,
108 "Old style space inode found, converting.");
109 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
110 BTRFS_INODE_NODATACOW;
111 block_group->disk_cache_state = BTRFS_DC_CLEAR;
114 if (!block_group->iref) {
115 block_group->inode = igrab(inode);
116 block_group->iref = 1;
118 spin_unlock(&block_group->lock);
123 static int __create_free_space_inode(struct btrfs_root *root,
124 struct btrfs_trans_handle *trans,
125 struct btrfs_path *path,
128 struct btrfs_key key;
129 struct btrfs_disk_key disk_key;
130 struct btrfs_free_space_header *header;
131 struct btrfs_inode_item *inode_item;
132 struct extent_buffer *leaf;
133 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
136 ret = btrfs_insert_empty_inode(trans, root, path, ino);
140 /* We inline crc's for the free disk space cache */
141 if (ino != BTRFS_FREE_INO_OBJECTID)
142 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
144 leaf = path->nodes[0];
145 inode_item = btrfs_item_ptr(leaf, path->slots[0],
146 struct btrfs_inode_item);
147 btrfs_item_key(leaf, &disk_key, path->slots[0]);
148 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
149 sizeof(*inode_item));
150 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
151 btrfs_set_inode_size(leaf, inode_item, 0);
152 btrfs_set_inode_nbytes(leaf, inode_item, 0);
153 btrfs_set_inode_uid(leaf, inode_item, 0);
154 btrfs_set_inode_gid(leaf, inode_item, 0);
155 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
156 btrfs_set_inode_flags(leaf, inode_item, flags);
157 btrfs_set_inode_nlink(leaf, inode_item, 1);
158 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
159 btrfs_set_inode_block_group(leaf, inode_item, offset);
160 btrfs_mark_buffer_dirty(leaf);
161 btrfs_release_path(path);
163 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
167 ret = btrfs_insert_empty_item(trans, root, path, &key,
168 sizeof(struct btrfs_free_space_header));
170 btrfs_release_path(path);
173 leaf = path->nodes[0];
174 header = btrfs_item_ptr(leaf, path->slots[0],
175 struct btrfs_free_space_header);
176 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
177 btrfs_set_free_space_key(leaf, header, &disk_key);
178 btrfs_mark_buffer_dirty(leaf);
179 btrfs_release_path(path);
184 int create_free_space_inode(struct btrfs_root *root,
185 struct btrfs_trans_handle *trans,
186 struct btrfs_block_group_cache *block_group,
187 struct btrfs_path *path)
192 ret = btrfs_find_free_objectid(root, &ino);
196 return __create_free_space_inode(root, trans, path, ino,
197 block_group->key.objectid);
200 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
201 struct btrfs_block_rsv *rsv)
206 /* 1 for slack space, 1 for updating the inode */
207 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
208 btrfs_calc_trans_metadata_size(root, 1);
210 spin_lock(&rsv->lock);
211 if (rsv->reserved < needed_bytes)
215 spin_unlock(&rsv->lock);
219 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
220 struct btrfs_trans_handle *trans,
221 struct btrfs_path *path,
226 btrfs_i_size_write(inode, 0);
227 truncate_pagecache(inode, 0);
230 * We don't need an orphan item because truncating the free space cache
231 * will never be split across transactions.
233 ret = btrfs_truncate_inode_items(trans, root, inode,
234 0, BTRFS_EXTENT_DATA_KEY);
236 btrfs_abort_transaction(trans, root, ret);
240 ret = btrfs_update_inode(trans, root, inode);
242 btrfs_abort_transaction(trans, root, ret);
247 static int readahead_cache(struct inode *inode)
249 struct file_ra_state *ra;
250 unsigned long last_index;
252 ra = kzalloc(sizeof(*ra), GFP_NOFS);
256 file_ra_state_init(ra, inode->i_mapping);
257 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
259 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
270 struct btrfs_root *root;
274 unsigned check_crcs:1;
277 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
278 struct btrfs_root *root)
280 memset(io_ctl, 0, sizeof(struct io_ctl));
281 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
283 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
288 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
289 io_ctl->check_crcs = 1;
293 static void io_ctl_free(struct io_ctl *io_ctl)
295 kfree(io_ctl->pages);
298 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
301 kunmap(io_ctl->page);
307 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
309 BUG_ON(io_ctl->index >= io_ctl->num_pages);
310 io_ctl->page = io_ctl->pages[io_ctl->index++];
311 io_ctl->cur = kmap(io_ctl->page);
312 io_ctl->orig = io_ctl->cur;
313 io_ctl->size = PAGE_CACHE_SIZE;
315 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
318 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
322 io_ctl_unmap_page(io_ctl);
324 for (i = 0; i < io_ctl->num_pages; i++) {
325 if (io_ctl->pages[i]) {
326 ClearPageChecked(io_ctl->pages[i]);
327 unlock_page(io_ctl->pages[i]);
328 page_cache_release(io_ctl->pages[i]);
333 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
337 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
340 for (i = 0; i < io_ctl->num_pages; i++) {
341 page = find_or_create_page(inode->i_mapping, i, mask);
343 io_ctl_drop_pages(io_ctl);
346 io_ctl->pages[i] = page;
347 if (uptodate && !PageUptodate(page)) {
348 btrfs_readpage(NULL, page);
350 if (!PageUptodate(page)) {
351 printk(KERN_ERR "btrfs: error reading free "
353 io_ctl_drop_pages(io_ctl);
359 for (i = 0; i < io_ctl->num_pages; i++) {
360 clear_page_dirty_for_io(io_ctl->pages[i]);
361 set_page_extent_mapped(io_ctl->pages[i]);
367 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
371 io_ctl_map_page(io_ctl, 1);
374 * Skip the csum areas. If we don't check crcs then we just have a
375 * 64bit chunk at the front of the first page.
377 if (io_ctl->check_crcs) {
378 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
379 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
381 io_ctl->cur += sizeof(u64);
382 io_ctl->size -= sizeof(u64) * 2;
386 *val = cpu_to_le64(generation);
387 io_ctl->cur += sizeof(u64);
390 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
395 * Skip the crc area. If we don't check crcs then we just have a 64bit
396 * chunk at the front of the first page.
398 if (io_ctl->check_crcs) {
399 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
400 io_ctl->size -= sizeof(u64) +
401 (sizeof(u32) * io_ctl->num_pages);
403 io_ctl->cur += sizeof(u64);
404 io_ctl->size -= sizeof(u64) * 2;
408 if (le64_to_cpu(*gen) != generation) {
409 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
410 "(%Lu) does not match inode (%Lu)\n", *gen,
412 io_ctl_unmap_page(io_ctl);
415 io_ctl->cur += sizeof(u64);
419 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
425 if (!io_ctl->check_crcs) {
426 io_ctl_unmap_page(io_ctl);
431 offset = sizeof(u32) * io_ctl->num_pages;
433 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
434 PAGE_CACHE_SIZE - offset);
435 btrfs_csum_final(crc, (char *)&crc);
436 io_ctl_unmap_page(io_ctl);
437 tmp = kmap(io_ctl->pages[0]);
440 kunmap(io_ctl->pages[0]);
443 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
449 if (!io_ctl->check_crcs) {
450 io_ctl_map_page(io_ctl, 0);
455 offset = sizeof(u32) * io_ctl->num_pages;
457 tmp = kmap(io_ctl->pages[0]);
460 kunmap(io_ctl->pages[0]);
462 io_ctl_map_page(io_ctl, 0);
463 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
464 PAGE_CACHE_SIZE - offset);
465 btrfs_csum_final(crc, (char *)&crc);
467 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
469 io_ctl_unmap_page(io_ctl);
476 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
479 struct btrfs_free_space_entry *entry;
485 entry->offset = cpu_to_le64(offset);
486 entry->bytes = cpu_to_le64(bytes);
487 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
488 BTRFS_FREE_SPACE_EXTENT;
489 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
490 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
492 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
495 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
497 /* No more pages to map */
498 if (io_ctl->index >= io_ctl->num_pages)
501 /* map the next page */
502 io_ctl_map_page(io_ctl, 1);
506 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
512 * If we aren't at the start of the current page, unmap this one and
513 * map the next one if there is any left.
515 if (io_ctl->cur != io_ctl->orig) {
516 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
517 if (io_ctl->index >= io_ctl->num_pages)
519 io_ctl_map_page(io_ctl, 0);
522 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
523 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
524 if (io_ctl->index < io_ctl->num_pages)
525 io_ctl_map_page(io_ctl, 0);
529 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
532 * If we're not on the boundary we know we've modified the page and we
533 * need to crc the page.
535 if (io_ctl->cur != io_ctl->orig)
536 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
538 io_ctl_unmap_page(io_ctl);
540 while (io_ctl->index < io_ctl->num_pages) {
541 io_ctl_map_page(io_ctl, 1);
542 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
546 static int io_ctl_read_entry(struct io_ctl *io_ctl,
547 struct btrfs_free_space *entry, u8 *type)
549 struct btrfs_free_space_entry *e;
553 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
559 entry->offset = le64_to_cpu(e->offset);
560 entry->bytes = le64_to_cpu(e->bytes);
562 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
563 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
565 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
568 io_ctl_unmap_page(io_ctl);
573 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
574 struct btrfs_free_space *entry)
578 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
582 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
583 io_ctl_unmap_page(io_ctl);
589 * Since we attach pinned extents after the fact we can have contiguous sections
590 * of free space that are split up in entries. This poses a problem with the
591 * tree logging stuff since it could have allocated across what appears to be 2
592 * entries since we would have merged the entries when adding the pinned extents
593 * back to the free space cache. So run through the space cache that we just
594 * loaded and merge contiguous entries. This will make the log replay stuff not
595 * blow up and it will make for nicer allocator behavior.
597 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
599 struct btrfs_free_space *e, *prev = NULL;
603 spin_lock(&ctl->tree_lock);
604 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
605 e = rb_entry(n, struct btrfs_free_space, offset_index);
608 if (e->bitmap || prev->bitmap)
610 if (prev->offset + prev->bytes == e->offset) {
611 unlink_free_space(ctl, prev);
612 unlink_free_space(ctl, e);
613 prev->bytes += e->bytes;
614 kmem_cache_free(btrfs_free_space_cachep, e);
615 link_free_space(ctl, prev);
617 spin_unlock(&ctl->tree_lock);
623 spin_unlock(&ctl->tree_lock);
626 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
627 struct btrfs_free_space_ctl *ctl,
628 struct btrfs_path *path, u64 offset)
630 struct btrfs_free_space_header *header;
631 struct extent_buffer *leaf;
632 struct io_ctl io_ctl;
633 struct btrfs_key key;
634 struct btrfs_free_space *e, *n;
635 struct list_head bitmaps;
642 INIT_LIST_HEAD(&bitmaps);
644 /* Nothing in the space cache, goodbye */
645 if (!i_size_read(inode))
648 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
652 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
656 btrfs_release_path(path);
662 leaf = path->nodes[0];
663 header = btrfs_item_ptr(leaf, path->slots[0],
664 struct btrfs_free_space_header);
665 num_entries = btrfs_free_space_entries(leaf, header);
666 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
667 generation = btrfs_free_space_generation(leaf, header);
668 btrfs_release_path(path);
670 if (BTRFS_I(inode)->generation != generation) {
671 btrfs_err(root->fs_info,
672 "free space inode generation (%llu) "
673 "did not match free space cache generation (%llu)",
674 (unsigned long long)BTRFS_I(inode)->generation,
675 (unsigned long long)generation);
682 ret = io_ctl_init(&io_ctl, inode, root);
686 ret = readahead_cache(inode);
690 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
694 ret = io_ctl_check_crc(&io_ctl, 0);
698 ret = io_ctl_check_generation(&io_ctl, generation);
702 while (num_entries) {
703 e = kmem_cache_zalloc(btrfs_free_space_cachep,
708 ret = io_ctl_read_entry(&io_ctl, e, &type);
710 kmem_cache_free(btrfs_free_space_cachep, e);
715 kmem_cache_free(btrfs_free_space_cachep, e);
719 if (type == BTRFS_FREE_SPACE_EXTENT) {
720 spin_lock(&ctl->tree_lock);
721 ret = link_free_space(ctl, e);
722 spin_unlock(&ctl->tree_lock);
724 btrfs_err(root->fs_info,
725 "Duplicate entries in free space cache, dumping");
726 kmem_cache_free(btrfs_free_space_cachep, e);
730 BUG_ON(!num_bitmaps);
732 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
735 btrfs_free_space_cachep, e);
738 spin_lock(&ctl->tree_lock);
739 ret = link_free_space(ctl, e);
740 ctl->total_bitmaps++;
741 ctl->op->recalc_thresholds(ctl);
742 spin_unlock(&ctl->tree_lock);
744 btrfs_err(root->fs_info,
745 "Duplicate entries in free space cache, dumping");
746 kmem_cache_free(btrfs_free_space_cachep, e);
749 list_add_tail(&e->list, &bitmaps);
755 io_ctl_unmap_page(&io_ctl);
758 * We add the bitmaps at the end of the entries in order that
759 * the bitmap entries are added to the cache.
761 list_for_each_entry_safe(e, n, &bitmaps, list) {
762 list_del_init(&e->list);
763 ret = io_ctl_read_bitmap(&io_ctl, e);
768 io_ctl_drop_pages(&io_ctl);
769 merge_space_tree(ctl);
772 io_ctl_free(&io_ctl);
775 io_ctl_drop_pages(&io_ctl);
776 __btrfs_remove_free_space_cache(ctl);
780 int load_free_space_cache(struct btrfs_fs_info *fs_info,
781 struct btrfs_block_group_cache *block_group)
783 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
784 struct btrfs_root *root = fs_info->tree_root;
786 struct btrfs_path *path;
789 u64 used = btrfs_block_group_used(&block_group->item);
792 * If this block group has been marked to be cleared for one reason or
793 * another then we can't trust the on disk cache, so just return.
795 spin_lock(&block_group->lock);
796 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
797 spin_unlock(&block_group->lock);
800 spin_unlock(&block_group->lock);
802 path = btrfs_alloc_path();
805 path->search_commit_root = 1;
806 path->skip_locking = 1;
808 inode = lookup_free_space_inode(root, block_group, path);
810 btrfs_free_path(path);
814 /* We may have converted the inode and made the cache invalid. */
815 spin_lock(&block_group->lock);
816 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
817 spin_unlock(&block_group->lock);
818 btrfs_free_path(path);
821 spin_unlock(&block_group->lock);
823 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
824 path, block_group->key.objectid);
825 btrfs_free_path(path);
829 spin_lock(&ctl->tree_lock);
830 matched = (ctl->free_space == (block_group->key.offset - used -
831 block_group->bytes_super));
832 spin_unlock(&ctl->tree_lock);
835 __btrfs_remove_free_space_cache(ctl);
836 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
837 block_group->key.objectid);
842 /* This cache is bogus, make sure it gets cleared */
843 spin_lock(&block_group->lock);
844 block_group->disk_cache_state = BTRFS_DC_CLEAR;
845 spin_unlock(&block_group->lock);
848 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
849 block_group->key.objectid);
857 * __btrfs_write_out_cache - write out cached info to an inode
858 * @root - the root the inode belongs to
859 * @ctl - the free space cache we are going to write out
860 * @block_group - the block_group for this cache if it belongs to a block_group
861 * @trans - the trans handle
862 * @path - the path to use
863 * @offset - the offset for the key we'll insert
865 * This function writes out a free space cache struct to disk for quick recovery
866 * on mount. This will return 0 if it was successfull in writing the cache out,
867 * and -1 if it was not.
869 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
870 struct btrfs_free_space_ctl *ctl,
871 struct btrfs_block_group_cache *block_group,
872 struct btrfs_trans_handle *trans,
873 struct btrfs_path *path, u64 offset)
875 struct btrfs_free_space_header *header;
876 struct extent_buffer *leaf;
877 struct rb_node *node;
878 struct list_head *pos, *n;
879 struct extent_state *cached_state = NULL;
880 struct btrfs_free_cluster *cluster = NULL;
881 struct extent_io_tree *unpin = NULL;
882 struct io_ctl io_ctl;
883 struct list_head bitmap_list;
884 struct btrfs_key key;
885 u64 start, extent_start, extent_end, len;
891 INIT_LIST_HEAD(&bitmap_list);
893 if (!i_size_read(inode))
896 ret = io_ctl_init(&io_ctl, inode, root);
900 /* Get the cluster for this block_group if it exists */
901 if (block_group && !list_empty(&block_group->cluster_list))
902 cluster = list_entry(block_group->cluster_list.next,
903 struct btrfs_free_cluster,
906 /* Lock all pages first so we can lock the extent safely. */
907 io_ctl_prepare_pages(&io_ctl, inode, 0);
909 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
912 node = rb_first(&ctl->free_space_offset);
913 if (!node && cluster) {
914 node = rb_first(&cluster->root);
918 /* Make sure we can fit our crcs into the first page */
919 if (io_ctl.check_crcs &&
920 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
923 io_ctl_set_generation(&io_ctl, trans->transid);
925 /* Write out the extent entries */
927 struct btrfs_free_space *e;
929 e = rb_entry(node, struct btrfs_free_space, offset_index);
932 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
938 list_add_tail(&e->list, &bitmap_list);
941 node = rb_next(node);
942 if (!node && cluster) {
943 node = rb_first(&cluster->root);
949 * We want to add any pinned extents to our free space cache
950 * so we don't leak the space
954 * We shouldn't have switched the pinned extents yet so this is the
957 unpin = root->fs_info->pinned_extents;
960 start = block_group->key.objectid;
962 while (block_group && (start < block_group->key.objectid +
963 block_group->key.offset)) {
964 ret = find_first_extent_bit(unpin, start,
965 &extent_start, &extent_end,
972 /* This pinned extent is out of our range */
973 if (extent_start >= block_group->key.objectid +
974 block_group->key.offset)
977 extent_start = max(extent_start, start);
978 extent_end = min(block_group->key.objectid +
979 block_group->key.offset, extent_end + 1);
980 len = extent_end - extent_start;
983 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
990 /* Write out the bitmaps */
991 list_for_each_safe(pos, n, &bitmap_list) {
992 struct btrfs_free_space *entry =
993 list_entry(pos, struct btrfs_free_space, list);
995 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
998 list_del_init(&entry->list);
1001 /* Zero out the rest of the pages just to make sure */
1002 io_ctl_zero_remaining_pages(&io_ctl);
1004 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1005 0, i_size_read(inode), &cached_state);
1006 io_ctl_drop_pages(&io_ctl);
1007 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1008 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1014 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1016 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1017 key.offset = offset;
1020 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1022 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1023 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1027 leaf = path->nodes[0];
1029 struct btrfs_key found_key;
1030 BUG_ON(!path->slots[0]);
1032 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1033 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1034 found_key.offset != offset) {
1035 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1037 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1039 btrfs_release_path(path);
1044 BTRFS_I(inode)->generation = trans->transid;
1045 header = btrfs_item_ptr(leaf, path->slots[0],
1046 struct btrfs_free_space_header);
1047 btrfs_set_free_space_entries(leaf, header, entries);
1048 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1049 btrfs_set_free_space_generation(leaf, header, trans->transid);
1050 btrfs_mark_buffer_dirty(leaf);
1051 btrfs_release_path(path);
1055 io_ctl_free(&io_ctl);
1057 invalidate_inode_pages2(inode->i_mapping);
1058 BTRFS_I(inode)->generation = 0;
1060 btrfs_update_inode(trans, root, inode);
1064 list_for_each_safe(pos, n, &bitmap_list) {
1065 struct btrfs_free_space *entry =
1066 list_entry(pos, struct btrfs_free_space, list);
1067 list_del_init(&entry->list);
1069 io_ctl_drop_pages(&io_ctl);
1070 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1071 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1075 int btrfs_write_out_cache(struct btrfs_root *root,
1076 struct btrfs_trans_handle *trans,
1077 struct btrfs_block_group_cache *block_group,
1078 struct btrfs_path *path)
1080 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1081 struct inode *inode;
1084 root = root->fs_info->tree_root;
1086 spin_lock(&block_group->lock);
1087 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1088 spin_unlock(&block_group->lock);
1091 spin_unlock(&block_group->lock);
1093 inode = lookup_free_space_inode(root, block_group, path);
1097 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1098 path, block_group->key.objectid);
1100 spin_lock(&block_group->lock);
1101 block_group->disk_cache_state = BTRFS_DC_ERROR;
1102 spin_unlock(&block_group->lock);
1105 btrfs_err(root->fs_info,
1106 "failed to write free space cache for block group %llu",
1107 block_group->key.objectid);
1115 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1118 BUG_ON(offset < bitmap_start);
1119 offset -= bitmap_start;
1120 return (unsigned long)(div_u64(offset, unit));
1123 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1125 return (unsigned long)(div_u64(bytes, unit));
1128 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1132 u64 bytes_per_bitmap;
1134 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1135 bitmap_start = offset - ctl->start;
1136 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1137 bitmap_start *= bytes_per_bitmap;
1138 bitmap_start += ctl->start;
1140 return bitmap_start;
1143 static int tree_insert_offset(struct rb_root *root, u64 offset,
1144 struct rb_node *node, int bitmap)
1146 struct rb_node **p = &root->rb_node;
1147 struct rb_node *parent = NULL;
1148 struct btrfs_free_space *info;
1152 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1154 if (offset < info->offset) {
1156 } else if (offset > info->offset) {
1157 p = &(*p)->rb_right;
1160 * we could have a bitmap entry and an extent entry
1161 * share the same offset. If this is the case, we want
1162 * the extent entry to always be found first if we do a
1163 * linear search through the tree, since we want to have
1164 * the quickest allocation time, and allocating from an
1165 * extent is faster than allocating from a bitmap. So
1166 * if we're inserting a bitmap and we find an entry at
1167 * this offset, we want to go right, or after this entry
1168 * logically. If we are inserting an extent and we've
1169 * found a bitmap, we want to go left, or before
1177 p = &(*p)->rb_right;
1179 if (!info->bitmap) {
1188 rb_link_node(node, parent, p);
1189 rb_insert_color(node, root);
1195 * searches the tree for the given offset.
1197 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1198 * want a section that has at least bytes size and comes at or after the given
1201 static struct btrfs_free_space *
1202 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1203 u64 offset, int bitmap_only, int fuzzy)
1205 struct rb_node *n = ctl->free_space_offset.rb_node;
1206 struct btrfs_free_space *entry, *prev = NULL;
1208 /* find entry that is closest to the 'offset' */
1215 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1218 if (offset < entry->offset)
1220 else if (offset > entry->offset)
1233 * bitmap entry and extent entry may share same offset,
1234 * in that case, bitmap entry comes after extent entry.
1239 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1240 if (entry->offset != offset)
1243 WARN_ON(!entry->bitmap);
1246 if (entry->bitmap) {
1248 * if previous extent entry covers the offset,
1249 * we should return it instead of the bitmap entry
1251 n = rb_prev(&entry->offset_index);
1253 prev = rb_entry(n, struct btrfs_free_space,
1255 if (!prev->bitmap &&
1256 prev->offset + prev->bytes > offset)
1266 /* find last entry before the 'offset' */
1268 if (entry->offset > offset) {
1269 n = rb_prev(&entry->offset_index);
1271 entry = rb_entry(n, struct btrfs_free_space,
1273 BUG_ON(entry->offset > offset);
1282 if (entry->bitmap) {
1283 n = rb_prev(&entry->offset_index);
1285 prev = rb_entry(n, struct btrfs_free_space,
1287 if (!prev->bitmap &&
1288 prev->offset + prev->bytes > offset)
1291 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1293 } else if (entry->offset + entry->bytes > offset)
1300 if (entry->bitmap) {
1301 if (entry->offset + BITS_PER_BITMAP *
1305 if (entry->offset + entry->bytes > offset)
1309 n = rb_next(&entry->offset_index);
1312 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1318 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1319 struct btrfs_free_space *info)
1321 rb_erase(&info->offset_index, &ctl->free_space_offset);
1322 ctl->free_extents--;
1325 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1326 struct btrfs_free_space *info)
1328 __unlink_free_space(ctl, info);
1329 ctl->free_space -= info->bytes;
1332 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1333 struct btrfs_free_space *info)
1337 BUG_ON(!info->bitmap && !info->bytes);
1338 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1339 &info->offset_index, (info->bitmap != NULL));
1343 ctl->free_space += info->bytes;
1344 ctl->free_extents++;
1348 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1350 struct btrfs_block_group_cache *block_group = ctl->private;
1354 u64 size = block_group->key.offset;
1355 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1356 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1358 max_bitmaps = max(max_bitmaps, 1);
1360 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1363 * The goal is to keep the total amount of memory used per 1gb of space
1364 * at or below 32k, so we need to adjust how much memory we allow to be
1365 * used by extent based free space tracking
1367 if (size < 1024 * 1024 * 1024)
1368 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1370 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1371 div64_u64(size, 1024 * 1024 * 1024);
1374 * we want to account for 1 more bitmap than what we have so we can make
1375 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1376 * we add more bitmaps.
1378 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1380 if (bitmap_bytes >= max_bytes) {
1381 ctl->extents_thresh = 0;
1386 * we want the extent entry threshold to always be at most 1/2 the maxw
1387 * bytes we can have, or whatever is less than that.
1389 extent_bytes = max_bytes - bitmap_bytes;
1390 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1392 ctl->extents_thresh =
1393 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1396 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1397 struct btrfs_free_space *info,
1398 u64 offset, u64 bytes)
1400 unsigned long start, count;
1402 start = offset_to_bit(info->offset, ctl->unit, offset);
1403 count = bytes_to_bits(bytes, ctl->unit);
1404 BUG_ON(start + count > BITS_PER_BITMAP);
1406 bitmap_clear(info->bitmap, start, count);
1408 info->bytes -= bytes;
1411 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1412 struct btrfs_free_space *info, u64 offset,
1415 __bitmap_clear_bits(ctl, info, offset, bytes);
1416 ctl->free_space -= bytes;
1419 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1420 struct btrfs_free_space *info, u64 offset,
1423 unsigned long start, count;
1425 start = offset_to_bit(info->offset, ctl->unit, offset);
1426 count = bytes_to_bits(bytes, ctl->unit);
1427 BUG_ON(start + count > BITS_PER_BITMAP);
1429 bitmap_set(info->bitmap, start, count);
1431 info->bytes += bytes;
1432 ctl->free_space += bytes;
1435 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1436 struct btrfs_free_space *bitmap_info, u64 *offset,
1439 unsigned long found_bits = 0;
1440 unsigned long bits, i;
1441 unsigned long next_zero;
1443 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1444 max_t(u64, *offset, bitmap_info->offset));
1445 bits = bytes_to_bits(*bytes, ctl->unit);
1447 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1448 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1449 BITS_PER_BITMAP, i);
1450 if ((next_zero - i) >= bits) {
1451 found_bits = next_zero - i;
1458 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1459 *bytes = (u64)(found_bits) * ctl->unit;
1466 static struct btrfs_free_space *
1467 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1468 unsigned long align)
1470 struct btrfs_free_space *entry;
1471 struct rb_node *node;
1477 if (!ctl->free_space_offset.rb_node)
1480 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1484 for (node = &entry->offset_index; node; node = rb_next(node)) {
1485 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1486 if (entry->bytes < *bytes)
1489 /* make sure the space returned is big enough
1490 * to match our requested alignment
1492 if (*bytes >= align) {
1493 ctl_off = entry->offset - ctl->start;
1494 tmp = ctl_off + align - 1;;
1496 tmp = tmp * align + ctl->start;
1497 align_off = tmp - entry->offset;
1500 tmp = entry->offset;
1503 if (entry->bytes < *bytes + align_off)
1506 if (entry->bitmap) {
1507 ret = search_bitmap(ctl, entry, &tmp, bytes);
1516 *bytes = entry->bytes - align_off;
1523 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1524 struct btrfs_free_space *info, u64 offset)
1526 info->offset = offset_to_bitmap(ctl, offset);
1528 INIT_LIST_HEAD(&info->list);
1529 link_free_space(ctl, info);
1530 ctl->total_bitmaps++;
1532 ctl->op->recalc_thresholds(ctl);
1535 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1536 struct btrfs_free_space *bitmap_info)
1538 unlink_free_space(ctl, bitmap_info);
1539 kfree(bitmap_info->bitmap);
1540 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1541 ctl->total_bitmaps--;
1542 ctl->op->recalc_thresholds(ctl);
1545 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1546 struct btrfs_free_space *bitmap_info,
1547 u64 *offset, u64 *bytes)
1550 u64 search_start, search_bytes;
1554 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1557 * We need to search for bits in this bitmap. We could only cover some
1558 * of the extent in this bitmap thanks to how we add space, so we need
1559 * to search for as much as it as we can and clear that amount, and then
1560 * go searching for the next bit.
1562 search_start = *offset;
1563 search_bytes = ctl->unit;
1564 search_bytes = min(search_bytes, end - search_start + 1);
1565 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1566 if (ret < 0 || search_start != *offset)
1569 /* We may have found more bits than what we need */
1570 search_bytes = min(search_bytes, *bytes);
1572 /* Cannot clear past the end of the bitmap */
1573 search_bytes = min(search_bytes, end - search_start + 1);
1575 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1576 *offset += search_bytes;
1577 *bytes -= search_bytes;
1580 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1581 if (!bitmap_info->bytes)
1582 free_bitmap(ctl, bitmap_info);
1585 * no entry after this bitmap, but we still have bytes to
1586 * remove, so something has gone wrong.
1591 bitmap_info = rb_entry(next, struct btrfs_free_space,
1595 * if the next entry isn't a bitmap we need to return to let the
1596 * extent stuff do its work.
1598 if (!bitmap_info->bitmap)
1602 * Ok the next item is a bitmap, but it may not actually hold
1603 * the information for the rest of this free space stuff, so
1604 * look for it, and if we don't find it return so we can try
1605 * everything over again.
1607 search_start = *offset;
1608 search_bytes = ctl->unit;
1609 ret = search_bitmap(ctl, bitmap_info, &search_start,
1611 if (ret < 0 || search_start != *offset)
1615 } else if (!bitmap_info->bytes)
1616 free_bitmap(ctl, bitmap_info);
1621 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1622 struct btrfs_free_space *info, u64 offset,
1625 u64 bytes_to_set = 0;
1628 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1630 bytes_to_set = min(end - offset, bytes);
1632 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1634 return bytes_to_set;
1638 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1639 struct btrfs_free_space *info)
1641 struct btrfs_block_group_cache *block_group = ctl->private;
1644 * If we are below the extents threshold then we can add this as an
1645 * extent, and don't have to deal with the bitmap
1647 if (ctl->free_extents < ctl->extents_thresh) {
1649 * If this block group has some small extents we don't want to
1650 * use up all of our free slots in the cache with them, we want
1651 * to reserve them to larger extents, however if we have plent
1652 * of cache left then go ahead an dadd them, no sense in adding
1653 * the overhead of a bitmap if we don't have to.
1655 if (info->bytes <= block_group->sectorsize * 4) {
1656 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1664 * The original block groups from mkfs can be really small, like 8
1665 * megabytes, so don't bother with a bitmap for those entries. However
1666 * some block groups can be smaller than what a bitmap would cover but
1667 * are still large enough that they could overflow the 32k memory limit,
1668 * so allow those block groups to still be allowed to have a bitmap
1671 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1677 static struct btrfs_free_space_op free_space_op = {
1678 .recalc_thresholds = recalculate_thresholds,
1679 .use_bitmap = use_bitmap,
1682 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1683 struct btrfs_free_space *info)
1685 struct btrfs_free_space *bitmap_info;
1686 struct btrfs_block_group_cache *block_group = NULL;
1688 u64 bytes, offset, bytes_added;
1691 bytes = info->bytes;
1692 offset = info->offset;
1694 if (!ctl->op->use_bitmap(ctl, info))
1697 if (ctl->op == &free_space_op)
1698 block_group = ctl->private;
1701 * Since we link bitmaps right into the cluster we need to see if we
1702 * have a cluster here, and if so and it has our bitmap we need to add
1703 * the free space to that bitmap.
1705 if (block_group && !list_empty(&block_group->cluster_list)) {
1706 struct btrfs_free_cluster *cluster;
1707 struct rb_node *node;
1708 struct btrfs_free_space *entry;
1710 cluster = list_entry(block_group->cluster_list.next,
1711 struct btrfs_free_cluster,
1713 spin_lock(&cluster->lock);
1714 node = rb_first(&cluster->root);
1716 spin_unlock(&cluster->lock);
1717 goto no_cluster_bitmap;
1720 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1721 if (!entry->bitmap) {
1722 spin_unlock(&cluster->lock);
1723 goto no_cluster_bitmap;
1726 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1727 bytes_added = add_bytes_to_bitmap(ctl, entry,
1729 bytes -= bytes_added;
1730 offset += bytes_added;
1732 spin_unlock(&cluster->lock);
1740 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1747 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1748 bytes -= bytes_added;
1749 offset += bytes_added;
1759 if (info && info->bitmap) {
1760 add_new_bitmap(ctl, info, offset);
1765 spin_unlock(&ctl->tree_lock);
1767 /* no pre-allocated info, allocate a new one */
1769 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1772 spin_lock(&ctl->tree_lock);
1778 /* allocate the bitmap */
1779 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1780 spin_lock(&ctl->tree_lock);
1781 if (!info->bitmap) {
1791 kfree(info->bitmap);
1792 kmem_cache_free(btrfs_free_space_cachep, info);
1798 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1799 struct btrfs_free_space *info, bool update_stat)
1801 struct btrfs_free_space *left_info;
1802 struct btrfs_free_space *right_info;
1803 bool merged = false;
1804 u64 offset = info->offset;
1805 u64 bytes = info->bytes;
1808 * first we want to see if there is free space adjacent to the range we
1809 * are adding, if there is remove that struct and add a new one to
1810 * cover the entire range
1812 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1813 if (right_info && rb_prev(&right_info->offset_index))
1814 left_info = rb_entry(rb_prev(&right_info->offset_index),
1815 struct btrfs_free_space, offset_index);
1817 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1819 if (right_info && !right_info->bitmap) {
1821 unlink_free_space(ctl, right_info);
1823 __unlink_free_space(ctl, right_info);
1824 info->bytes += right_info->bytes;
1825 kmem_cache_free(btrfs_free_space_cachep, right_info);
1829 if (left_info && !left_info->bitmap &&
1830 left_info->offset + left_info->bytes == offset) {
1832 unlink_free_space(ctl, left_info);
1834 __unlink_free_space(ctl, left_info);
1835 info->offset = left_info->offset;
1836 info->bytes += left_info->bytes;
1837 kmem_cache_free(btrfs_free_space_cachep, left_info);
1844 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1845 u64 offset, u64 bytes)
1847 struct btrfs_free_space *info;
1850 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1854 info->offset = offset;
1855 info->bytes = bytes;
1857 spin_lock(&ctl->tree_lock);
1859 if (try_merge_free_space(ctl, info, true))
1863 * There was no extent directly to the left or right of this new
1864 * extent then we know we're going to have to allocate a new extent, so
1865 * before we do that see if we need to drop this into a bitmap
1867 ret = insert_into_bitmap(ctl, info);
1875 ret = link_free_space(ctl, info);
1877 kmem_cache_free(btrfs_free_space_cachep, info);
1879 spin_unlock(&ctl->tree_lock);
1882 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1883 BUG_ON(ret == -EEXIST);
1889 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1890 u64 offset, u64 bytes)
1892 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1893 struct btrfs_free_space *info;
1895 bool re_search = false;
1897 spin_lock(&ctl->tree_lock);
1904 info = tree_search_offset(ctl, offset, 0, 0);
1907 * oops didn't find an extent that matched the space we wanted
1908 * to remove, look for a bitmap instead
1910 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1914 * If we found a partial bit of our free space in a
1915 * bitmap but then couldn't find the other part this may
1916 * be a problem, so WARN about it.
1924 if (!info->bitmap) {
1925 unlink_free_space(ctl, info);
1926 if (offset == info->offset) {
1927 u64 to_free = min(bytes, info->bytes);
1929 info->bytes -= to_free;
1930 info->offset += to_free;
1932 ret = link_free_space(ctl, info);
1935 kmem_cache_free(btrfs_free_space_cachep, info);
1942 u64 old_end = info->bytes + info->offset;
1944 info->bytes = offset - info->offset;
1945 ret = link_free_space(ctl, info);
1950 /* Not enough bytes in this entry to satisfy us */
1951 if (old_end < offset + bytes) {
1952 bytes -= old_end - offset;
1955 } else if (old_end == offset + bytes) {
1959 spin_unlock(&ctl->tree_lock);
1961 ret = btrfs_add_free_space(block_group, offset + bytes,
1962 old_end - (offset + bytes));
1968 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1969 if (ret == -EAGAIN) {
1974 spin_unlock(&ctl->tree_lock);
1979 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1982 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1983 struct btrfs_free_space *info;
1987 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1988 info = rb_entry(n, struct btrfs_free_space, offset_index);
1989 if (info->bytes >= bytes && !block_group->ro)
1991 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1992 (unsigned long long)info->offset,
1993 (unsigned long long)info->bytes,
1994 (info->bitmap) ? "yes" : "no");
1996 printk(KERN_INFO "block group has cluster?: %s\n",
1997 list_empty(&block_group->cluster_list) ? "no" : "yes");
1998 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2002 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2004 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2006 spin_lock_init(&ctl->tree_lock);
2007 ctl->unit = block_group->sectorsize;
2008 ctl->start = block_group->key.objectid;
2009 ctl->private = block_group;
2010 ctl->op = &free_space_op;
2013 * we only want to have 32k of ram per block group for keeping
2014 * track of free space, and if we pass 1/2 of that we want to
2015 * start converting things over to using bitmaps
2017 ctl->extents_thresh = ((1024 * 32) / 2) /
2018 sizeof(struct btrfs_free_space);
2022 * for a given cluster, put all of its extents back into the free
2023 * space cache. If the block group passed doesn't match the block group
2024 * pointed to by the cluster, someone else raced in and freed the
2025 * cluster already. In that case, we just return without changing anything
2028 __btrfs_return_cluster_to_free_space(
2029 struct btrfs_block_group_cache *block_group,
2030 struct btrfs_free_cluster *cluster)
2032 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2033 struct btrfs_free_space *entry;
2034 struct rb_node *node;
2036 spin_lock(&cluster->lock);
2037 if (cluster->block_group != block_group)
2040 cluster->block_group = NULL;
2041 cluster->window_start = 0;
2042 list_del_init(&cluster->block_group_list);
2044 node = rb_first(&cluster->root);
2048 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2049 node = rb_next(&entry->offset_index);
2050 rb_erase(&entry->offset_index, &cluster->root);
2052 bitmap = (entry->bitmap != NULL);
2054 try_merge_free_space(ctl, entry, false);
2055 tree_insert_offset(&ctl->free_space_offset,
2056 entry->offset, &entry->offset_index, bitmap);
2058 cluster->root = RB_ROOT;
2061 spin_unlock(&cluster->lock);
2062 btrfs_put_block_group(block_group);
2066 static void __btrfs_remove_free_space_cache_locked(
2067 struct btrfs_free_space_ctl *ctl)
2069 struct btrfs_free_space *info;
2070 struct rb_node *node;
2072 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2073 info = rb_entry(node, struct btrfs_free_space, offset_index);
2074 if (!info->bitmap) {
2075 unlink_free_space(ctl, info);
2076 kmem_cache_free(btrfs_free_space_cachep, info);
2078 free_bitmap(ctl, info);
2080 if (need_resched()) {
2081 spin_unlock(&ctl->tree_lock);
2083 spin_lock(&ctl->tree_lock);
2088 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2090 spin_lock(&ctl->tree_lock);
2091 __btrfs_remove_free_space_cache_locked(ctl);
2092 spin_unlock(&ctl->tree_lock);
2095 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2097 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2098 struct btrfs_free_cluster *cluster;
2099 struct list_head *head;
2101 spin_lock(&ctl->tree_lock);
2102 while ((head = block_group->cluster_list.next) !=
2103 &block_group->cluster_list) {
2104 cluster = list_entry(head, struct btrfs_free_cluster,
2107 WARN_ON(cluster->block_group != block_group);
2108 __btrfs_return_cluster_to_free_space(block_group, cluster);
2109 if (need_resched()) {
2110 spin_unlock(&ctl->tree_lock);
2112 spin_lock(&ctl->tree_lock);
2115 __btrfs_remove_free_space_cache_locked(ctl);
2116 spin_unlock(&ctl->tree_lock);
2120 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2121 u64 offset, u64 bytes, u64 empty_size)
2123 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2124 struct btrfs_free_space *entry = NULL;
2125 u64 bytes_search = bytes + empty_size;
2128 u64 align_gap_len = 0;
2130 spin_lock(&ctl->tree_lock);
2131 entry = find_free_space(ctl, &offset, &bytes_search,
2132 block_group->full_stripe_len);
2137 if (entry->bitmap) {
2138 bitmap_clear_bits(ctl, entry, offset, bytes);
2140 free_bitmap(ctl, entry);
2143 unlink_free_space(ctl, entry);
2144 align_gap_len = offset - entry->offset;
2145 align_gap = entry->offset;
2147 entry->offset = offset + bytes;
2148 WARN_ON(entry->bytes < bytes + align_gap_len);
2150 entry->bytes -= bytes + align_gap_len;
2152 kmem_cache_free(btrfs_free_space_cachep, entry);
2154 link_free_space(ctl, entry);
2158 spin_unlock(&ctl->tree_lock);
2161 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2166 * given a cluster, put all of its extents back into the free space
2167 * cache. If a block group is passed, this function will only free
2168 * a cluster that belongs to the passed block group.
2170 * Otherwise, it'll get a reference on the block group pointed to by the
2171 * cluster and remove the cluster from it.
2173 int btrfs_return_cluster_to_free_space(
2174 struct btrfs_block_group_cache *block_group,
2175 struct btrfs_free_cluster *cluster)
2177 struct btrfs_free_space_ctl *ctl;
2180 /* first, get a safe pointer to the block group */
2181 spin_lock(&cluster->lock);
2183 block_group = cluster->block_group;
2185 spin_unlock(&cluster->lock);
2188 } else if (cluster->block_group != block_group) {
2189 /* someone else has already freed it don't redo their work */
2190 spin_unlock(&cluster->lock);
2193 atomic_inc(&block_group->count);
2194 spin_unlock(&cluster->lock);
2196 ctl = block_group->free_space_ctl;
2198 /* now return any extents the cluster had on it */
2199 spin_lock(&ctl->tree_lock);
2200 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2201 spin_unlock(&ctl->tree_lock);
2203 /* finally drop our ref */
2204 btrfs_put_block_group(block_group);
2208 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2209 struct btrfs_free_cluster *cluster,
2210 struct btrfs_free_space *entry,
2211 u64 bytes, u64 min_start)
2213 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2215 u64 search_start = cluster->window_start;
2216 u64 search_bytes = bytes;
2219 search_start = min_start;
2220 search_bytes = bytes;
2222 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2227 __bitmap_clear_bits(ctl, entry, ret, bytes);
2233 * given a cluster, try to allocate 'bytes' from it, returns 0
2234 * if it couldn't find anything suitably large, or a logical disk offset
2235 * if things worked out
2237 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2238 struct btrfs_free_cluster *cluster, u64 bytes,
2241 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2242 struct btrfs_free_space *entry = NULL;
2243 struct rb_node *node;
2246 spin_lock(&cluster->lock);
2247 if (bytes > cluster->max_size)
2250 if (cluster->block_group != block_group)
2253 node = rb_first(&cluster->root);
2257 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2259 if (entry->bytes < bytes ||
2260 (!entry->bitmap && entry->offset < min_start)) {
2261 node = rb_next(&entry->offset_index);
2264 entry = rb_entry(node, struct btrfs_free_space,
2269 if (entry->bitmap) {
2270 ret = btrfs_alloc_from_bitmap(block_group,
2271 cluster, entry, bytes,
2272 cluster->window_start);
2274 node = rb_next(&entry->offset_index);
2277 entry = rb_entry(node, struct btrfs_free_space,
2281 cluster->window_start += bytes;
2283 ret = entry->offset;
2285 entry->offset += bytes;
2286 entry->bytes -= bytes;
2289 if (entry->bytes == 0)
2290 rb_erase(&entry->offset_index, &cluster->root);
2294 spin_unlock(&cluster->lock);
2299 spin_lock(&ctl->tree_lock);
2301 ctl->free_space -= bytes;
2302 if (entry->bytes == 0) {
2303 ctl->free_extents--;
2304 if (entry->bitmap) {
2305 kfree(entry->bitmap);
2306 ctl->total_bitmaps--;
2307 ctl->op->recalc_thresholds(ctl);
2309 kmem_cache_free(btrfs_free_space_cachep, entry);
2312 spin_unlock(&ctl->tree_lock);
2317 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2318 struct btrfs_free_space *entry,
2319 struct btrfs_free_cluster *cluster,
2320 u64 offset, u64 bytes,
2321 u64 cont1_bytes, u64 min_bytes)
2323 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2324 unsigned long next_zero;
2326 unsigned long want_bits;
2327 unsigned long min_bits;
2328 unsigned long found_bits;
2329 unsigned long start = 0;
2330 unsigned long total_found = 0;
2333 i = offset_to_bit(entry->offset, ctl->unit,
2334 max_t(u64, offset, entry->offset));
2335 want_bits = bytes_to_bits(bytes, ctl->unit);
2336 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2340 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2341 next_zero = find_next_zero_bit(entry->bitmap,
2342 BITS_PER_BITMAP, i);
2343 if (next_zero - i >= min_bits) {
2344 found_bits = next_zero - i;
2355 cluster->max_size = 0;
2358 total_found += found_bits;
2360 if (cluster->max_size < found_bits * ctl->unit)
2361 cluster->max_size = found_bits * ctl->unit;
2363 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2368 cluster->window_start = start * ctl->unit + entry->offset;
2369 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2370 ret = tree_insert_offset(&cluster->root, entry->offset,
2371 &entry->offset_index, 1);
2372 BUG_ON(ret); /* -EEXIST; Logic error */
2374 trace_btrfs_setup_cluster(block_group, cluster,
2375 total_found * ctl->unit, 1);
2380 * This searches the block group for just extents to fill the cluster with.
2381 * Try to find a cluster with at least bytes total bytes, at least one
2382 * extent of cont1_bytes, and other clusters of at least min_bytes.
2385 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2386 struct btrfs_free_cluster *cluster,
2387 struct list_head *bitmaps, u64 offset, u64 bytes,
2388 u64 cont1_bytes, u64 min_bytes)
2390 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2391 struct btrfs_free_space *first = NULL;
2392 struct btrfs_free_space *entry = NULL;
2393 struct btrfs_free_space *last;
2394 struct rb_node *node;
2400 entry = tree_search_offset(ctl, offset, 0, 1);
2405 * We don't want bitmaps, so just move along until we find a normal
2408 while (entry->bitmap || entry->bytes < min_bytes) {
2409 if (entry->bitmap && list_empty(&entry->list))
2410 list_add_tail(&entry->list, bitmaps);
2411 node = rb_next(&entry->offset_index);
2414 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2417 window_start = entry->offset;
2418 window_free = entry->bytes;
2419 max_extent = entry->bytes;
2423 for (node = rb_next(&entry->offset_index); node;
2424 node = rb_next(&entry->offset_index)) {
2425 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2427 if (entry->bitmap) {
2428 if (list_empty(&entry->list))
2429 list_add_tail(&entry->list, bitmaps);
2433 if (entry->bytes < min_bytes)
2437 window_free += entry->bytes;
2438 if (entry->bytes > max_extent)
2439 max_extent = entry->bytes;
2442 if (window_free < bytes || max_extent < cont1_bytes)
2445 cluster->window_start = first->offset;
2447 node = &first->offset_index;
2450 * now we've found our entries, pull them out of the free space
2451 * cache and put them into the cluster rbtree
2456 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2457 node = rb_next(&entry->offset_index);
2458 if (entry->bitmap || entry->bytes < min_bytes)
2461 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2462 ret = tree_insert_offset(&cluster->root, entry->offset,
2463 &entry->offset_index, 0);
2464 total_size += entry->bytes;
2465 BUG_ON(ret); /* -EEXIST; Logic error */
2466 } while (node && entry != last);
2468 cluster->max_size = max_extent;
2469 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2474 * This specifically looks for bitmaps that may work in the cluster, we assume
2475 * that we have already failed to find extents that will work.
2478 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2479 struct btrfs_free_cluster *cluster,
2480 struct list_head *bitmaps, u64 offset, u64 bytes,
2481 u64 cont1_bytes, u64 min_bytes)
2483 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2484 struct btrfs_free_space *entry;
2486 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2488 if (ctl->total_bitmaps == 0)
2492 * The bitmap that covers offset won't be in the list unless offset
2493 * is just its start offset.
2495 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2496 if (entry->offset != bitmap_offset) {
2497 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2498 if (entry && list_empty(&entry->list))
2499 list_add(&entry->list, bitmaps);
2502 list_for_each_entry(entry, bitmaps, list) {
2503 if (entry->bytes < bytes)
2505 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2506 bytes, cont1_bytes, min_bytes);
2512 * The bitmaps list has all the bitmaps that record free space
2513 * starting after offset, so no more search is required.
2519 * here we try to find a cluster of blocks in a block group. The goal
2520 * is to find at least bytes+empty_size.
2521 * We might not find them all in one contiguous area.
2523 * returns zero and sets up cluster if things worked out, otherwise
2524 * it returns -enospc
2526 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2527 struct btrfs_root *root,
2528 struct btrfs_block_group_cache *block_group,
2529 struct btrfs_free_cluster *cluster,
2530 u64 offset, u64 bytes, u64 empty_size)
2532 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2533 struct btrfs_free_space *entry, *tmp;
2540 * Choose the minimum extent size we'll require for this
2541 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2542 * For metadata, allow allocates with smaller extents. For
2543 * data, keep it dense.
2545 if (btrfs_test_opt(root, SSD_SPREAD)) {
2546 cont1_bytes = min_bytes = bytes + empty_size;
2547 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2548 cont1_bytes = bytes;
2549 min_bytes = block_group->sectorsize;
2551 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2552 min_bytes = block_group->sectorsize;
2555 spin_lock(&ctl->tree_lock);
2558 * If we know we don't have enough space to make a cluster don't even
2559 * bother doing all the work to try and find one.
2561 if (ctl->free_space < bytes) {
2562 spin_unlock(&ctl->tree_lock);
2566 spin_lock(&cluster->lock);
2568 /* someone already found a cluster, hooray */
2569 if (cluster->block_group) {
2574 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2577 INIT_LIST_HEAD(&bitmaps);
2578 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2580 cont1_bytes, min_bytes);
2582 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2583 offset, bytes + empty_size,
2584 cont1_bytes, min_bytes);
2586 /* Clear our temporary list */
2587 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2588 list_del_init(&entry->list);
2591 atomic_inc(&block_group->count);
2592 list_add_tail(&cluster->block_group_list,
2593 &block_group->cluster_list);
2594 cluster->block_group = block_group;
2596 trace_btrfs_failed_cluster_setup(block_group);
2599 spin_unlock(&cluster->lock);
2600 spin_unlock(&ctl->tree_lock);
2606 * simple code to zero out a cluster
2608 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2610 spin_lock_init(&cluster->lock);
2611 spin_lock_init(&cluster->refill_lock);
2612 cluster->root = RB_ROOT;
2613 cluster->max_size = 0;
2614 INIT_LIST_HEAD(&cluster->block_group_list);
2615 cluster->block_group = NULL;
2618 static int do_trimming(struct btrfs_block_group_cache *block_group,
2619 u64 *total_trimmed, u64 start, u64 bytes,
2620 u64 reserved_start, u64 reserved_bytes)
2622 struct btrfs_space_info *space_info = block_group->space_info;
2623 struct btrfs_fs_info *fs_info = block_group->fs_info;
2628 spin_lock(&space_info->lock);
2629 spin_lock(&block_group->lock);
2630 if (!block_group->ro) {
2631 block_group->reserved += reserved_bytes;
2632 space_info->bytes_reserved += reserved_bytes;
2635 spin_unlock(&block_group->lock);
2636 spin_unlock(&space_info->lock);
2638 ret = btrfs_error_discard_extent(fs_info->extent_root,
2639 start, bytes, &trimmed);
2641 *total_trimmed += trimmed;
2643 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2646 spin_lock(&space_info->lock);
2647 spin_lock(&block_group->lock);
2648 if (block_group->ro)
2649 space_info->bytes_readonly += reserved_bytes;
2650 block_group->reserved -= reserved_bytes;
2651 space_info->bytes_reserved -= reserved_bytes;
2652 spin_unlock(&space_info->lock);
2653 spin_unlock(&block_group->lock);
2659 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2660 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2662 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2663 struct btrfs_free_space *entry;
2664 struct rb_node *node;
2670 while (start < end) {
2671 spin_lock(&ctl->tree_lock);
2673 if (ctl->free_space < minlen) {
2674 spin_unlock(&ctl->tree_lock);
2678 entry = tree_search_offset(ctl, start, 0, 1);
2680 spin_unlock(&ctl->tree_lock);
2685 while (entry->bitmap) {
2686 node = rb_next(&entry->offset_index);
2688 spin_unlock(&ctl->tree_lock);
2691 entry = rb_entry(node, struct btrfs_free_space,
2695 if (entry->offset >= end) {
2696 spin_unlock(&ctl->tree_lock);
2700 extent_start = entry->offset;
2701 extent_bytes = entry->bytes;
2702 start = max(start, extent_start);
2703 bytes = min(extent_start + extent_bytes, end) - start;
2704 if (bytes < minlen) {
2705 spin_unlock(&ctl->tree_lock);
2709 unlink_free_space(ctl, entry);
2710 kmem_cache_free(btrfs_free_space_cachep, entry);
2712 spin_unlock(&ctl->tree_lock);
2714 ret = do_trimming(block_group, total_trimmed, start, bytes,
2715 extent_start, extent_bytes);
2721 if (fatal_signal_pending(current)) {
2732 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2733 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2735 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2736 struct btrfs_free_space *entry;
2740 u64 offset = offset_to_bitmap(ctl, start);
2742 while (offset < end) {
2743 bool next_bitmap = false;
2745 spin_lock(&ctl->tree_lock);
2747 if (ctl->free_space < minlen) {
2748 spin_unlock(&ctl->tree_lock);
2752 entry = tree_search_offset(ctl, offset, 1, 0);
2754 spin_unlock(&ctl->tree_lock);
2760 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2761 if (ret2 || start >= end) {
2762 spin_unlock(&ctl->tree_lock);
2767 bytes = min(bytes, end - start);
2768 if (bytes < minlen) {
2769 spin_unlock(&ctl->tree_lock);
2773 bitmap_clear_bits(ctl, entry, start, bytes);
2774 if (entry->bytes == 0)
2775 free_bitmap(ctl, entry);
2777 spin_unlock(&ctl->tree_lock);
2779 ret = do_trimming(block_group, total_trimmed, start, bytes,
2785 offset += BITS_PER_BITMAP * ctl->unit;
2788 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2789 offset += BITS_PER_BITMAP * ctl->unit;
2792 if (fatal_signal_pending(current)) {
2803 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2804 u64 *trimmed, u64 start, u64 end, u64 minlen)
2810 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2814 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2820 * Find the left-most item in the cache tree, and then return the
2821 * smallest inode number in the item.
2823 * Note: the returned inode number may not be the smallest one in
2824 * the tree, if the left-most item is a bitmap.
2826 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2828 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2829 struct btrfs_free_space *entry = NULL;
2832 spin_lock(&ctl->tree_lock);
2834 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2837 entry = rb_entry(rb_first(&ctl->free_space_offset),
2838 struct btrfs_free_space, offset_index);
2840 if (!entry->bitmap) {
2841 ino = entry->offset;
2843 unlink_free_space(ctl, entry);
2847 kmem_cache_free(btrfs_free_space_cachep, entry);
2849 link_free_space(ctl, entry);
2855 ret = search_bitmap(ctl, entry, &offset, &count);
2856 /* Logic error; Should be empty if it can't find anything */
2860 bitmap_clear_bits(ctl, entry, offset, 1);
2861 if (entry->bytes == 0)
2862 free_bitmap(ctl, entry);
2865 spin_unlock(&ctl->tree_lock);
2870 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2871 struct btrfs_path *path)
2873 struct inode *inode = NULL;
2875 spin_lock(&root->cache_lock);
2876 if (root->cache_inode)
2877 inode = igrab(root->cache_inode);
2878 spin_unlock(&root->cache_lock);
2882 inode = __lookup_free_space_inode(root, path, 0);
2886 spin_lock(&root->cache_lock);
2887 if (!btrfs_fs_closing(root->fs_info))
2888 root->cache_inode = igrab(inode);
2889 spin_unlock(&root->cache_lock);
2894 int create_free_ino_inode(struct btrfs_root *root,
2895 struct btrfs_trans_handle *trans,
2896 struct btrfs_path *path)
2898 return __create_free_space_inode(root, trans, path,
2899 BTRFS_FREE_INO_OBJECTID, 0);
2902 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2904 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2905 struct btrfs_path *path;
2906 struct inode *inode;
2908 u64 root_gen = btrfs_root_generation(&root->root_item);
2910 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2914 * If we're unmounting then just return, since this does a search on the
2915 * normal root and not the commit root and we could deadlock.
2917 if (btrfs_fs_closing(fs_info))
2920 path = btrfs_alloc_path();
2924 inode = lookup_free_ino_inode(root, path);
2928 if (root_gen != BTRFS_I(inode)->generation)
2931 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2935 "failed to load free ino cache for root %llu",
2936 root->root_key.objectid);
2940 btrfs_free_path(path);
2944 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2945 struct btrfs_trans_handle *trans,
2946 struct btrfs_path *path)
2948 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2949 struct inode *inode;
2952 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2955 inode = lookup_free_ino_inode(root, path);
2959 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2961 btrfs_delalloc_release_metadata(inode, inode->i_size);
2963 btrfs_err(root->fs_info,
2964 "failed to write free ino cache for root %llu",
2965 root->root_key.objectid);
2973 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2974 static struct btrfs_block_group_cache *init_test_block_group(void)
2976 struct btrfs_block_group_cache *cache;
2978 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2981 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
2983 if (!cache->free_space_ctl) {
2988 cache->key.objectid = 0;
2989 cache->key.offset = 1024 * 1024 * 1024;
2990 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2991 cache->sectorsize = 4096;
2993 spin_lock_init(&cache->lock);
2994 INIT_LIST_HEAD(&cache->list);
2995 INIT_LIST_HEAD(&cache->cluster_list);
2996 INIT_LIST_HEAD(&cache->new_bg_list);
2998 btrfs_init_free_space_ctl(cache);
3004 * Checks to see if the given range is in the free space cache. This is really
3005 * just used to check the absence of space, so if there is free space in the
3006 * range at all we will return 1.
3008 static int check_exists(struct btrfs_block_group_cache *cache, u64 offset,
3011 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3012 struct btrfs_free_space *info;
3015 spin_lock(&ctl->tree_lock);
3016 info = tree_search_offset(ctl, offset, 0, 0);
3018 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3026 u64 bit_off, bit_bytes;
3028 struct btrfs_free_space *tmp;
3031 bit_bytes = ctl->unit;
3032 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3034 if (bit_off == offset) {
3037 } else if (bit_off > offset &&
3038 offset + bytes > bit_off) {
3044 n = rb_prev(&info->offset_index);
3046 tmp = rb_entry(n, struct btrfs_free_space,
3048 if (tmp->offset + tmp->bytes < offset)
3050 if (offset + bytes < tmp->offset) {
3051 n = rb_prev(&info->offset_index);
3058 n = rb_next(&info->offset_index);
3060 tmp = rb_entry(n, struct btrfs_free_space,
3062 if (offset + bytes < tmp->offset)
3064 if (tmp->offset + tmp->bytes < offset) {
3065 n = rb_next(&info->offset_index);
3075 if (info->offset == offset) {
3080 if (offset > info->offset && offset < info->offset + info->bytes)
3083 spin_unlock(&ctl->tree_lock);
3088 * Use this if you need to make a bitmap or extent entry specifically, it
3089 * doesn't do any of the merging that add_free_space does, this acts a lot like
3090 * how the free space cache loading stuff works, so you can get really weird
3093 static int add_free_space_entry(struct btrfs_block_group_cache *cache,
3094 u64 offset, u64 bytes, bool bitmap)
3096 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3097 struct btrfs_free_space *info = NULL, *bitmap_info;
3104 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3110 spin_lock(&ctl->tree_lock);
3111 info->offset = offset;
3112 info->bytes = bytes;
3113 ret = link_free_space(ctl, info);
3114 spin_unlock(&ctl->tree_lock);
3116 kmem_cache_free(btrfs_free_space_cachep, info);
3121 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3123 kmem_cache_free(btrfs_free_space_cachep, info);
3128 spin_lock(&ctl->tree_lock);
3129 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3134 add_new_bitmap(ctl, info, offset);
3138 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3139 bytes -= bytes_added;
3140 offset += bytes_added;
3141 spin_unlock(&ctl->tree_lock);
3151 #define test_msg(fmt, ...) printk(KERN_INFO "btrfs: selftest: " fmt, ##__VA_ARGS__)
3154 * This test just does basic sanity checking, making sure we can add an exten
3155 * entry and remove space from either end and the middle, and make sure we can
3156 * remove space that covers adjacent extent entries.
3158 static int test_extents(struct btrfs_block_group_cache *cache)
3162 test_msg("Running extent only tests\n");
3164 /* First just make sure we can remove an entire entry */
3165 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3167 test_msg("Error adding initial extents %d\n", ret);
3171 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3173 test_msg("Error removing extent %d\n", ret);
3177 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3178 test_msg("Full remove left some lingering space\n");
3182 /* Ok edge and middle cases now */
3183 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3185 test_msg("Error adding half extent %d\n", ret);
3189 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 1 * 1024 * 1024);
3191 test_msg("Error removing tail end %d\n", ret);
3195 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3197 test_msg("Error removing front end %d\n", ret);
3201 ret = btrfs_remove_free_space(cache, 2 * 1024 * 1024, 4096);
3203 test_msg("Error removing middle piece %d\n", ret);
3207 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3208 test_msg("Still have space at the front\n");
3212 if (check_exists(cache, 2 * 1024 * 1024, 4096)) {
3213 test_msg("Still have space in the middle\n");
3217 if (check_exists(cache, 3 * 1024 * 1024, 1 * 1024 * 1024)) {
3218 test_msg("Still have space at the end\n");
3223 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3228 static int test_bitmaps(struct btrfs_block_group_cache *cache)
3230 u64 next_bitmap_offset;
3233 test_msg("Running bitmap only tests\n");
3235 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3237 test_msg("Couldn't create a bitmap entry %d\n", ret);
3241 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3243 test_msg("Error removing bitmap full range %d\n", ret);
3247 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3248 test_msg("Left some space in bitmap\n");
3252 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3254 test_msg("Couldn't add to our bitmap entry %d\n", ret);
3258 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 2 * 1024 * 1024);
3260 test_msg("Couldn't remove middle chunk %d\n", ret);
3265 * The first bitmap we have starts at offset 0 so the next one is just
3266 * at the end of the first bitmap.
3268 next_bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3270 /* Test a bit straddling two bitmaps */
3271 ret = add_free_space_entry(cache, next_bitmap_offset -
3272 (2 * 1024 * 1024), 4 * 1024 * 1024, 1);
3274 test_msg("Couldn't add space that straddles two bitmaps %d\n",
3279 ret = btrfs_remove_free_space(cache, next_bitmap_offset -
3280 (1 * 1024 * 1024), 2 * 1024 * 1024);
3282 test_msg("Couldn't remove overlapping space %d\n", ret);
3286 if (check_exists(cache, next_bitmap_offset - (1 * 1024 * 1024),
3288 test_msg("Left some space when removing overlapping\n");
3292 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3297 /* This is the high grade jackassery */
3298 static int test_bitmaps_and_extents(struct btrfs_block_group_cache *cache)
3300 u64 bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3303 test_msg("Running bitmap and extent tests\n");
3306 * First let's do something simple, an extent at the same offset as the
3307 * bitmap, but the free space completely in the extent and then
3308 * completely in the bitmap.
3310 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 1 * 1024 * 1024, 1);
3312 test_msg("Couldn't create bitmap entry %d\n", ret);
3316 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3318 test_msg("Couldn't add extent entry %d\n", ret);
3322 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3324 test_msg("Couldn't remove extent entry %d\n", ret);
3328 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3329 test_msg("Left remnants after our remove\n");
3333 /* Now to add back the extent entry and remove from the bitmap */
3334 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3336 test_msg("Couldn't re-add extent entry %d\n", ret);
3340 ret = btrfs_remove_free_space(cache, 4 * 1024 * 1024, 1 * 1024 * 1024);
3342 test_msg("Couldn't remove from bitmap %d\n", ret);
3346 if (check_exists(cache, 4 * 1024 * 1024, 1 * 1024 * 1024)) {
3347 test_msg("Left remnants in the bitmap\n");
3352 * Ok so a little more evil, extent entry and bitmap at the same offset,
3353 * removing an overlapping chunk.
3355 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 4 * 1024 * 1024, 1);
3357 test_msg("Couldn't add to a bitmap %d\n", ret);
3361 ret = btrfs_remove_free_space(cache, 512 * 1024, 3 * 1024 * 1024);
3363 test_msg("Couldn't remove overlapping space %d\n", ret);
3367 if (check_exists(cache, 512 * 1024, 3 * 1024 * 1024)) {
3368 test_msg("Left over peices after removing overlapping\n");
3372 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3374 /* Now with the extent entry offset into the bitmap */
3375 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 4 * 1024 * 1024, 1);
3377 test_msg("Couldn't add space to the bitmap %d\n", ret);
3381 ret = add_free_space_entry(cache, 2 * 1024 * 1024, 2 * 1024 * 1024, 0);
3383 test_msg("Couldn't add extent to the cache %d\n", ret);
3387 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 4 * 1024 * 1024);
3389 test_msg("Problem removing overlapping space %d\n", ret);
3393 if (check_exists(cache, 3 * 1024 * 1024, 4 * 1024 * 1024)) {
3394 test_msg("Left something behind when removing space");
3399 * This has blown up in the past, the extent entry starts before the
3400 * bitmap entry, but we're trying to remove an offset that falls
3401 * completely within the bitmap range and is in both the extent entry
3402 * and the bitmap entry, looks like this
3408 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3409 ret = add_free_space_entry(cache, bitmap_offset + 4 * 1024 * 1024,
3410 4 * 1024 * 1024, 1);
3412 test_msg("Couldn't add bitmap %d\n", ret);
3416 ret = add_free_space_entry(cache, bitmap_offset - 1 * 1024 * 1024,
3417 5 * 1024 * 1024, 0);
3419 test_msg("Couldn't add extent entry %d\n", ret);
3423 ret = btrfs_remove_free_space(cache, bitmap_offset + 1 * 1024 * 1024,
3426 test_msg("Failed to free our space %d\n", ret);
3430 if (check_exists(cache, bitmap_offset + 1 * 1024 * 1024,
3432 test_msg("Left stuff over\n");
3436 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3439 * This blew up before, we have part of the free space in a bitmap and
3440 * then the entirety of the rest of the space in an extent. This used
3441 * to return -EAGAIN back from btrfs_remove_extent, make sure this
3444 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 2 * 1024 * 1024, 1);
3446 test_msg("Couldn't add bitmap entry %d\n", ret);
3450 ret = add_free_space_entry(cache, 3 * 1024 * 1024, 1 * 1024 * 1024, 0);
3452 test_msg("Couldn't add extent entry %d\n", ret);
3456 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 3 * 1024 * 1024);
3458 test_msg("Error removing bitmap and extent overlapping %d\n", ret);
3462 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3466 void btrfs_test_free_space_cache(void)
3468 struct btrfs_block_group_cache *cache;
3470 test_msg("Running btrfs free space cache tests\n");
3472 cache = init_test_block_group();
3474 test_msg("Couldn't run the tests\n");
3478 if (test_extents(cache))
3480 if (test_bitmaps(cache))
3482 if (test_bitmaps_and_extents(cache))
3485 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3486 kfree(cache->free_space_ctl);
3488 test_msg("Free space cache tests finished\n");
3491 #else /* !CONFIG_BTRFS_FS_RUN_SANITY_TESTS */
3492 void btrfs_test_free_space_cache(void) {}
3493 #endif /* !CONFIG_BTRFS_FS_RUN_SANITY_TESTS */