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_truncate_free_space_cache(struct btrfs_root *root,
201 struct btrfs_trans_handle *trans,
202 struct btrfs_path *path,
205 struct btrfs_block_rsv *rsv;
210 rsv = trans->block_rsv;
211 trans->block_rsv = &root->fs_info->global_block_rsv;
213 /* 1 for slack space, 1 for updating the inode */
214 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
215 btrfs_calc_trans_metadata_size(root, 1);
217 spin_lock(&trans->block_rsv->lock);
218 if (trans->block_rsv->reserved < needed_bytes) {
219 spin_unlock(&trans->block_rsv->lock);
220 trans->block_rsv = rsv;
223 spin_unlock(&trans->block_rsv->lock);
225 oldsize = i_size_read(inode);
226 btrfs_i_size_write(inode, 0);
227 truncate_pagecache(inode, oldsize, 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);
237 trans->block_rsv = rsv;
238 btrfs_abort_transaction(trans, root, ret);
242 ret = btrfs_update_inode(trans, root, inode);
244 btrfs_abort_transaction(trans, root, ret);
245 trans->block_rsv = rsv;
250 static int readahead_cache(struct inode *inode)
252 struct file_ra_state *ra;
253 unsigned long last_index;
255 ra = kzalloc(sizeof(*ra), GFP_NOFS);
259 file_ra_state_init(ra, inode->i_mapping);
260 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
262 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
273 struct btrfs_root *root;
277 unsigned check_crcs:1;
280 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
281 struct btrfs_root *root)
283 memset(io_ctl, 0, sizeof(struct io_ctl));
284 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
286 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
291 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
292 io_ctl->check_crcs = 1;
296 static void io_ctl_free(struct io_ctl *io_ctl)
298 kfree(io_ctl->pages);
301 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
304 kunmap(io_ctl->page);
310 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
312 BUG_ON(io_ctl->index >= io_ctl->num_pages);
313 io_ctl->page = io_ctl->pages[io_ctl->index++];
314 io_ctl->cur = kmap(io_ctl->page);
315 io_ctl->orig = io_ctl->cur;
316 io_ctl->size = PAGE_CACHE_SIZE;
318 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
321 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
325 io_ctl_unmap_page(io_ctl);
327 for (i = 0; i < io_ctl->num_pages; i++) {
328 if (io_ctl->pages[i]) {
329 ClearPageChecked(io_ctl->pages[i]);
330 unlock_page(io_ctl->pages[i]);
331 page_cache_release(io_ctl->pages[i]);
336 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
340 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
343 for (i = 0; i < io_ctl->num_pages; i++) {
344 page = find_or_create_page(inode->i_mapping, i, mask);
346 io_ctl_drop_pages(io_ctl);
349 io_ctl->pages[i] = page;
350 if (uptodate && !PageUptodate(page)) {
351 btrfs_readpage(NULL, page);
353 if (!PageUptodate(page)) {
354 printk(KERN_ERR "btrfs: error reading free "
356 io_ctl_drop_pages(io_ctl);
362 for (i = 0; i < io_ctl->num_pages; i++) {
363 clear_page_dirty_for_io(io_ctl->pages[i]);
364 set_page_extent_mapped(io_ctl->pages[i]);
370 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
374 io_ctl_map_page(io_ctl, 1);
377 * Skip the csum areas. If we don't check crcs then we just have a
378 * 64bit chunk at the front of the first page.
380 if (io_ctl->check_crcs) {
381 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
382 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
384 io_ctl->cur += sizeof(u64);
385 io_ctl->size -= sizeof(u64) * 2;
389 *val = cpu_to_le64(generation);
390 io_ctl->cur += sizeof(u64);
393 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
398 * Skip the crc area. If we don't check crcs then we just have a 64bit
399 * chunk at the front of the first page.
401 if (io_ctl->check_crcs) {
402 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
403 io_ctl->size -= sizeof(u64) +
404 (sizeof(u32) * io_ctl->num_pages);
406 io_ctl->cur += sizeof(u64);
407 io_ctl->size -= sizeof(u64) * 2;
411 if (le64_to_cpu(*gen) != generation) {
412 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
413 "(%Lu) does not match inode (%Lu)\n", *gen,
415 io_ctl_unmap_page(io_ctl);
418 io_ctl->cur += sizeof(u64);
422 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
428 if (!io_ctl->check_crcs) {
429 io_ctl_unmap_page(io_ctl);
434 offset = sizeof(u32) * io_ctl->num_pages;
436 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
437 PAGE_CACHE_SIZE - offset);
438 btrfs_csum_final(crc, (char *)&crc);
439 io_ctl_unmap_page(io_ctl);
440 tmp = kmap(io_ctl->pages[0]);
443 kunmap(io_ctl->pages[0]);
446 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
452 if (!io_ctl->check_crcs) {
453 io_ctl_map_page(io_ctl, 0);
458 offset = sizeof(u32) * io_ctl->num_pages;
460 tmp = kmap(io_ctl->pages[0]);
463 kunmap(io_ctl->pages[0]);
465 io_ctl_map_page(io_ctl, 0);
466 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
467 PAGE_CACHE_SIZE - offset);
468 btrfs_csum_final(crc, (char *)&crc);
470 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
472 io_ctl_unmap_page(io_ctl);
479 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
482 struct btrfs_free_space_entry *entry;
488 entry->offset = cpu_to_le64(offset);
489 entry->bytes = cpu_to_le64(bytes);
490 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
491 BTRFS_FREE_SPACE_EXTENT;
492 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
493 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
495 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
498 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
500 /* No more pages to map */
501 if (io_ctl->index >= io_ctl->num_pages)
504 /* map the next page */
505 io_ctl_map_page(io_ctl, 1);
509 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
515 * If we aren't at the start of the current page, unmap this one and
516 * map the next one if there is any left.
518 if (io_ctl->cur != io_ctl->orig) {
519 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
520 if (io_ctl->index >= io_ctl->num_pages)
522 io_ctl_map_page(io_ctl, 0);
525 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
526 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
527 if (io_ctl->index < io_ctl->num_pages)
528 io_ctl_map_page(io_ctl, 0);
532 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
535 * If we're not on the boundary we know we've modified the page and we
536 * need to crc the page.
538 if (io_ctl->cur != io_ctl->orig)
539 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
541 io_ctl_unmap_page(io_ctl);
543 while (io_ctl->index < io_ctl->num_pages) {
544 io_ctl_map_page(io_ctl, 1);
545 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
549 static int io_ctl_read_entry(struct io_ctl *io_ctl,
550 struct btrfs_free_space *entry, u8 *type)
552 struct btrfs_free_space_entry *e;
556 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
562 entry->offset = le64_to_cpu(e->offset);
563 entry->bytes = le64_to_cpu(e->bytes);
565 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
566 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
568 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
571 io_ctl_unmap_page(io_ctl);
576 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
577 struct btrfs_free_space *entry)
581 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
585 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
586 io_ctl_unmap_page(io_ctl);
592 * Since we attach pinned extents after the fact we can have contiguous sections
593 * of free space that are split up in entries. This poses a problem with the
594 * tree logging stuff since it could have allocated across what appears to be 2
595 * entries since we would have merged the entries when adding the pinned extents
596 * back to the free space cache. So run through the space cache that we just
597 * loaded and merge contiguous entries. This will make the log replay stuff not
598 * blow up and it will make for nicer allocator behavior.
600 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
602 struct btrfs_free_space *e, *prev = NULL;
606 spin_lock(&ctl->tree_lock);
607 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
608 e = rb_entry(n, struct btrfs_free_space, offset_index);
611 if (e->bitmap || prev->bitmap)
613 if (prev->offset + prev->bytes == e->offset) {
614 unlink_free_space(ctl, prev);
615 unlink_free_space(ctl, e);
616 prev->bytes += e->bytes;
617 kmem_cache_free(btrfs_free_space_cachep, e);
618 link_free_space(ctl, prev);
620 spin_unlock(&ctl->tree_lock);
626 spin_unlock(&ctl->tree_lock);
629 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
630 struct btrfs_free_space_ctl *ctl,
631 struct btrfs_path *path, u64 offset)
633 struct btrfs_free_space_header *header;
634 struct extent_buffer *leaf;
635 struct io_ctl io_ctl;
636 struct btrfs_key key;
637 struct btrfs_free_space *e, *n;
638 struct list_head bitmaps;
645 INIT_LIST_HEAD(&bitmaps);
647 /* Nothing in the space cache, goodbye */
648 if (!i_size_read(inode))
651 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
655 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
659 btrfs_release_path(path);
665 leaf = path->nodes[0];
666 header = btrfs_item_ptr(leaf, path->slots[0],
667 struct btrfs_free_space_header);
668 num_entries = btrfs_free_space_entries(leaf, header);
669 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
670 generation = btrfs_free_space_generation(leaf, header);
671 btrfs_release_path(path);
673 if (BTRFS_I(inode)->generation != generation) {
674 btrfs_err(root->fs_info,
675 "free space inode generation (%llu) "
676 "did not match free space cache generation (%llu)",
677 (unsigned long long)BTRFS_I(inode)->generation,
678 (unsigned long long)generation);
685 ret = io_ctl_init(&io_ctl, inode, root);
689 ret = readahead_cache(inode);
693 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
697 ret = io_ctl_check_crc(&io_ctl, 0);
701 ret = io_ctl_check_generation(&io_ctl, generation);
705 while (num_entries) {
706 e = kmem_cache_zalloc(btrfs_free_space_cachep,
711 ret = io_ctl_read_entry(&io_ctl, e, &type);
713 kmem_cache_free(btrfs_free_space_cachep, e);
718 kmem_cache_free(btrfs_free_space_cachep, e);
722 if (type == BTRFS_FREE_SPACE_EXTENT) {
723 spin_lock(&ctl->tree_lock);
724 ret = link_free_space(ctl, e);
725 spin_unlock(&ctl->tree_lock);
727 btrfs_err(root->fs_info,
728 "Duplicate entries in free space cache, dumping");
729 kmem_cache_free(btrfs_free_space_cachep, e);
733 BUG_ON(!num_bitmaps);
735 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
738 btrfs_free_space_cachep, e);
741 spin_lock(&ctl->tree_lock);
742 ret = link_free_space(ctl, e);
743 ctl->total_bitmaps++;
744 ctl->op->recalc_thresholds(ctl);
745 spin_unlock(&ctl->tree_lock);
747 btrfs_err(root->fs_info,
748 "Duplicate entries in free space cache, dumping");
749 kmem_cache_free(btrfs_free_space_cachep, e);
752 list_add_tail(&e->list, &bitmaps);
758 io_ctl_unmap_page(&io_ctl);
761 * We add the bitmaps at the end of the entries in order that
762 * the bitmap entries are added to the cache.
764 list_for_each_entry_safe(e, n, &bitmaps, list) {
765 list_del_init(&e->list);
766 ret = io_ctl_read_bitmap(&io_ctl, e);
771 io_ctl_drop_pages(&io_ctl);
772 merge_space_tree(ctl);
775 io_ctl_free(&io_ctl);
778 io_ctl_drop_pages(&io_ctl);
779 __btrfs_remove_free_space_cache(ctl);
783 int load_free_space_cache(struct btrfs_fs_info *fs_info,
784 struct btrfs_block_group_cache *block_group)
786 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
787 struct btrfs_root *root = fs_info->tree_root;
789 struct btrfs_path *path;
792 u64 used = btrfs_block_group_used(&block_group->item);
795 * If this block group has been marked to be cleared for one reason or
796 * another then we can't trust the on disk cache, so just return.
798 spin_lock(&block_group->lock);
799 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
800 spin_unlock(&block_group->lock);
803 spin_unlock(&block_group->lock);
805 path = btrfs_alloc_path();
808 path->search_commit_root = 1;
809 path->skip_locking = 1;
811 inode = lookup_free_space_inode(root, block_group, path);
813 btrfs_free_path(path);
817 /* We may have converted the inode and made the cache invalid. */
818 spin_lock(&block_group->lock);
819 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
820 spin_unlock(&block_group->lock);
821 btrfs_free_path(path);
824 spin_unlock(&block_group->lock);
826 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
827 path, block_group->key.objectid);
828 btrfs_free_path(path);
832 spin_lock(&ctl->tree_lock);
833 matched = (ctl->free_space == (block_group->key.offset - used -
834 block_group->bytes_super));
835 spin_unlock(&ctl->tree_lock);
838 __btrfs_remove_free_space_cache(ctl);
839 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
840 block_group->key.objectid);
845 /* This cache is bogus, make sure it gets cleared */
846 spin_lock(&block_group->lock);
847 block_group->disk_cache_state = BTRFS_DC_CLEAR;
848 spin_unlock(&block_group->lock);
851 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
852 block_group->key.objectid);
860 * __btrfs_write_out_cache - write out cached info to an inode
861 * @root - the root the inode belongs to
862 * @ctl - the free space cache we are going to write out
863 * @block_group - the block_group for this cache if it belongs to a block_group
864 * @trans - the trans handle
865 * @path - the path to use
866 * @offset - the offset for the key we'll insert
868 * This function writes out a free space cache struct to disk for quick recovery
869 * on mount. This will return 0 if it was successfull in writing the cache out,
870 * and -1 if it was not.
872 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
873 struct btrfs_free_space_ctl *ctl,
874 struct btrfs_block_group_cache *block_group,
875 struct btrfs_trans_handle *trans,
876 struct btrfs_path *path, u64 offset)
878 struct btrfs_free_space_header *header;
879 struct extent_buffer *leaf;
880 struct rb_node *node;
881 struct list_head *pos, *n;
882 struct extent_state *cached_state = NULL;
883 struct btrfs_free_cluster *cluster = NULL;
884 struct extent_io_tree *unpin = NULL;
885 struct io_ctl io_ctl;
886 struct list_head bitmap_list;
887 struct btrfs_key key;
888 u64 start, extent_start, extent_end, len;
894 INIT_LIST_HEAD(&bitmap_list);
896 if (!i_size_read(inode))
899 ret = io_ctl_init(&io_ctl, inode, root);
903 /* Get the cluster for this block_group if it exists */
904 if (block_group && !list_empty(&block_group->cluster_list))
905 cluster = list_entry(block_group->cluster_list.next,
906 struct btrfs_free_cluster,
909 /* Lock all pages first so we can lock the extent safely. */
910 io_ctl_prepare_pages(&io_ctl, inode, 0);
912 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
915 node = rb_first(&ctl->free_space_offset);
916 if (!node && cluster) {
917 node = rb_first(&cluster->root);
921 /* Make sure we can fit our crcs into the first page */
922 if (io_ctl.check_crcs &&
923 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
928 io_ctl_set_generation(&io_ctl, trans->transid);
930 /* Write out the extent entries */
932 struct btrfs_free_space *e;
934 e = rb_entry(node, struct btrfs_free_space, offset_index);
937 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
943 list_add_tail(&e->list, &bitmap_list);
946 node = rb_next(node);
947 if (!node && cluster) {
948 node = rb_first(&cluster->root);
954 * We want to add any pinned extents to our free space cache
955 * so we don't leak the space
959 * We shouldn't have switched the pinned extents yet so this is the
962 unpin = root->fs_info->pinned_extents;
965 start = block_group->key.objectid;
967 while (block_group && (start < block_group->key.objectid +
968 block_group->key.offset)) {
969 ret = find_first_extent_bit(unpin, start,
970 &extent_start, &extent_end,
977 /* This pinned extent is out of our range */
978 if (extent_start >= block_group->key.objectid +
979 block_group->key.offset)
982 extent_start = max(extent_start, start);
983 extent_end = min(block_group->key.objectid +
984 block_group->key.offset, extent_end + 1);
985 len = extent_end - extent_start;
988 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
995 /* Write out the bitmaps */
996 list_for_each_safe(pos, n, &bitmap_list) {
997 struct btrfs_free_space *entry =
998 list_entry(pos, struct btrfs_free_space, list);
1000 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
1003 list_del_init(&entry->list);
1006 /* Zero out the rest of the pages just to make sure */
1007 io_ctl_zero_remaining_pages(&io_ctl);
1009 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1010 0, i_size_read(inode), &cached_state);
1011 io_ctl_drop_pages(&io_ctl);
1012 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1013 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1019 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1021 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1022 key.offset = offset;
1025 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1027 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1028 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1032 leaf = path->nodes[0];
1034 struct btrfs_key found_key;
1035 BUG_ON(!path->slots[0]);
1037 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1038 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1039 found_key.offset != offset) {
1040 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1042 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1044 btrfs_release_path(path);
1049 BTRFS_I(inode)->generation = trans->transid;
1050 header = btrfs_item_ptr(leaf, path->slots[0],
1051 struct btrfs_free_space_header);
1052 btrfs_set_free_space_entries(leaf, header, entries);
1053 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1054 btrfs_set_free_space_generation(leaf, header, trans->transid);
1055 btrfs_mark_buffer_dirty(leaf);
1056 btrfs_release_path(path);
1060 io_ctl_free(&io_ctl);
1062 invalidate_inode_pages2(inode->i_mapping);
1063 BTRFS_I(inode)->generation = 0;
1065 btrfs_update_inode(trans, root, inode);
1069 list_for_each_safe(pos, n, &bitmap_list) {
1070 struct btrfs_free_space *entry =
1071 list_entry(pos, struct btrfs_free_space, list);
1072 list_del_init(&entry->list);
1074 io_ctl_drop_pages(&io_ctl);
1075 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1076 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1080 int btrfs_write_out_cache(struct btrfs_root *root,
1081 struct btrfs_trans_handle *trans,
1082 struct btrfs_block_group_cache *block_group,
1083 struct btrfs_path *path)
1085 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1086 struct inode *inode;
1089 root = root->fs_info->tree_root;
1091 spin_lock(&block_group->lock);
1092 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1093 spin_unlock(&block_group->lock);
1096 spin_unlock(&block_group->lock);
1098 inode = lookup_free_space_inode(root, block_group, path);
1102 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1103 path, block_group->key.objectid);
1105 spin_lock(&block_group->lock);
1106 block_group->disk_cache_state = BTRFS_DC_ERROR;
1107 spin_unlock(&block_group->lock);
1110 btrfs_err(root->fs_info,
1111 "failed to write free space cache for block group %llu",
1112 block_group->key.objectid);
1120 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1123 BUG_ON(offset < bitmap_start);
1124 offset -= bitmap_start;
1125 return (unsigned long)(div_u64(offset, unit));
1128 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1130 return (unsigned long)(div_u64(bytes, unit));
1133 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1137 u64 bytes_per_bitmap;
1139 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1140 bitmap_start = offset - ctl->start;
1141 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1142 bitmap_start *= bytes_per_bitmap;
1143 bitmap_start += ctl->start;
1145 return bitmap_start;
1148 static int tree_insert_offset(struct rb_root *root, u64 offset,
1149 struct rb_node *node, int bitmap)
1151 struct rb_node **p = &root->rb_node;
1152 struct rb_node *parent = NULL;
1153 struct btrfs_free_space *info;
1157 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1159 if (offset < info->offset) {
1161 } else if (offset > info->offset) {
1162 p = &(*p)->rb_right;
1165 * we could have a bitmap entry and an extent entry
1166 * share the same offset. If this is the case, we want
1167 * the extent entry to always be found first if we do a
1168 * linear search through the tree, since we want to have
1169 * the quickest allocation time, and allocating from an
1170 * extent is faster than allocating from a bitmap. So
1171 * if we're inserting a bitmap and we find an entry at
1172 * this offset, we want to go right, or after this entry
1173 * logically. If we are inserting an extent and we've
1174 * found a bitmap, we want to go left, or before
1182 p = &(*p)->rb_right;
1184 if (!info->bitmap) {
1193 rb_link_node(node, parent, p);
1194 rb_insert_color(node, root);
1200 * searches the tree for the given offset.
1202 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1203 * want a section that has at least bytes size and comes at or after the given
1206 static struct btrfs_free_space *
1207 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1208 u64 offset, int bitmap_only, int fuzzy)
1210 struct rb_node *n = ctl->free_space_offset.rb_node;
1211 struct btrfs_free_space *entry, *prev = NULL;
1213 /* find entry that is closest to the 'offset' */
1220 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1223 if (offset < entry->offset)
1225 else if (offset > entry->offset)
1238 * bitmap entry and extent entry may share same offset,
1239 * in that case, bitmap entry comes after extent entry.
1244 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1245 if (entry->offset != offset)
1248 WARN_ON(!entry->bitmap);
1251 if (entry->bitmap) {
1253 * if previous extent entry covers the offset,
1254 * we should return it instead of the bitmap entry
1256 n = rb_prev(&entry->offset_index);
1258 prev = rb_entry(n, struct btrfs_free_space,
1260 if (!prev->bitmap &&
1261 prev->offset + prev->bytes > offset)
1271 /* find last entry before the 'offset' */
1273 if (entry->offset > offset) {
1274 n = rb_prev(&entry->offset_index);
1276 entry = rb_entry(n, struct btrfs_free_space,
1278 BUG_ON(entry->offset > offset);
1287 if (entry->bitmap) {
1288 n = rb_prev(&entry->offset_index);
1290 prev = rb_entry(n, struct btrfs_free_space,
1292 if (!prev->bitmap &&
1293 prev->offset + prev->bytes > offset)
1296 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1298 } else if (entry->offset + entry->bytes > offset)
1305 if (entry->bitmap) {
1306 if (entry->offset + BITS_PER_BITMAP *
1310 if (entry->offset + entry->bytes > offset)
1314 n = rb_next(&entry->offset_index);
1317 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1323 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1324 struct btrfs_free_space *info)
1326 rb_erase(&info->offset_index, &ctl->free_space_offset);
1327 ctl->free_extents--;
1330 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1331 struct btrfs_free_space *info)
1333 __unlink_free_space(ctl, info);
1334 ctl->free_space -= info->bytes;
1337 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1338 struct btrfs_free_space *info)
1342 BUG_ON(!info->bitmap && !info->bytes);
1343 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1344 &info->offset_index, (info->bitmap != NULL));
1348 ctl->free_space += info->bytes;
1349 ctl->free_extents++;
1353 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1355 struct btrfs_block_group_cache *block_group = ctl->private;
1359 u64 size = block_group->key.offset;
1360 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1361 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1363 max_bitmaps = max(max_bitmaps, 1);
1365 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1368 * The goal is to keep the total amount of memory used per 1gb of space
1369 * at or below 32k, so we need to adjust how much memory we allow to be
1370 * used by extent based free space tracking
1372 if (size < 1024 * 1024 * 1024)
1373 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1375 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1376 div64_u64(size, 1024 * 1024 * 1024);
1379 * we want to account for 1 more bitmap than what we have so we can make
1380 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1381 * we add more bitmaps.
1383 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1385 if (bitmap_bytes >= max_bytes) {
1386 ctl->extents_thresh = 0;
1391 * we want the extent entry threshold to always be at most 1/2 the maxw
1392 * bytes we can have, or whatever is less than that.
1394 extent_bytes = max_bytes - bitmap_bytes;
1395 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1397 ctl->extents_thresh =
1398 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1401 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1402 struct btrfs_free_space *info,
1403 u64 offset, u64 bytes)
1405 unsigned long start, count;
1407 start = offset_to_bit(info->offset, ctl->unit, offset);
1408 count = bytes_to_bits(bytes, ctl->unit);
1409 BUG_ON(start + count > BITS_PER_BITMAP);
1411 bitmap_clear(info->bitmap, start, count);
1413 info->bytes -= bytes;
1416 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1417 struct btrfs_free_space *info, u64 offset,
1420 __bitmap_clear_bits(ctl, info, offset, bytes);
1421 ctl->free_space -= bytes;
1424 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1425 struct btrfs_free_space *info, u64 offset,
1428 unsigned long start, count;
1430 start = offset_to_bit(info->offset, ctl->unit, offset);
1431 count = bytes_to_bits(bytes, ctl->unit);
1432 BUG_ON(start + count > BITS_PER_BITMAP);
1434 bitmap_set(info->bitmap, start, count);
1436 info->bytes += bytes;
1437 ctl->free_space += bytes;
1440 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1441 struct btrfs_free_space *bitmap_info, u64 *offset,
1444 unsigned long found_bits = 0;
1445 unsigned long bits, i;
1446 unsigned long next_zero;
1448 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1449 max_t(u64, *offset, bitmap_info->offset));
1450 bits = bytes_to_bits(*bytes, ctl->unit);
1452 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1453 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1454 BITS_PER_BITMAP, i);
1455 if ((next_zero - i) >= bits) {
1456 found_bits = next_zero - i;
1463 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1464 *bytes = (u64)(found_bits) * ctl->unit;
1471 static struct btrfs_free_space *
1472 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1473 unsigned long align)
1475 struct btrfs_free_space *entry;
1476 struct rb_node *node;
1482 if (!ctl->free_space_offset.rb_node)
1485 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1489 for (node = &entry->offset_index; node; node = rb_next(node)) {
1490 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1491 if (entry->bytes < *bytes)
1494 /* make sure the space returned is big enough
1495 * to match our requested alignment
1497 if (*bytes >= align) {
1498 ctl_off = entry->offset - ctl->start;
1499 tmp = ctl_off + align - 1;;
1501 tmp = tmp * align + ctl->start;
1502 align_off = tmp - entry->offset;
1505 tmp = entry->offset;
1508 if (entry->bytes < *bytes + align_off)
1511 if (entry->bitmap) {
1512 ret = search_bitmap(ctl, entry, &tmp, bytes);
1521 *bytes = entry->bytes - align_off;
1528 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1529 struct btrfs_free_space *info, u64 offset)
1531 info->offset = offset_to_bitmap(ctl, offset);
1533 INIT_LIST_HEAD(&info->list);
1534 link_free_space(ctl, info);
1535 ctl->total_bitmaps++;
1537 ctl->op->recalc_thresholds(ctl);
1540 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1541 struct btrfs_free_space *bitmap_info)
1543 unlink_free_space(ctl, bitmap_info);
1544 kfree(bitmap_info->bitmap);
1545 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1546 ctl->total_bitmaps--;
1547 ctl->op->recalc_thresholds(ctl);
1550 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1551 struct btrfs_free_space *bitmap_info,
1552 u64 *offset, u64 *bytes)
1555 u64 search_start, search_bytes;
1559 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1562 * We need to search for bits in this bitmap. We could only cover some
1563 * of the extent in this bitmap thanks to how we add space, so we need
1564 * to search for as much as it as we can and clear that amount, and then
1565 * go searching for the next bit.
1567 search_start = *offset;
1568 search_bytes = ctl->unit;
1569 search_bytes = min(search_bytes, end - search_start + 1);
1570 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1571 if (ret < 0 || search_start != *offset)
1574 /* We may have found more bits than what we need */
1575 search_bytes = min(search_bytes, *bytes);
1577 /* Cannot clear past the end of the bitmap */
1578 search_bytes = min(search_bytes, end - search_start + 1);
1580 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1581 *offset += search_bytes;
1582 *bytes -= search_bytes;
1585 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1586 if (!bitmap_info->bytes)
1587 free_bitmap(ctl, bitmap_info);
1590 * no entry after this bitmap, but we still have bytes to
1591 * remove, so something has gone wrong.
1596 bitmap_info = rb_entry(next, struct btrfs_free_space,
1600 * if the next entry isn't a bitmap we need to return to let the
1601 * extent stuff do its work.
1603 if (!bitmap_info->bitmap)
1607 * Ok the next item is a bitmap, but it may not actually hold
1608 * the information for the rest of this free space stuff, so
1609 * look for it, and if we don't find it return so we can try
1610 * everything over again.
1612 search_start = *offset;
1613 search_bytes = ctl->unit;
1614 ret = search_bitmap(ctl, bitmap_info, &search_start,
1616 if (ret < 0 || search_start != *offset)
1620 } else if (!bitmap_info->bytes)
1621 free_bitmap(ctl, bitmap_info);
1626 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1627 struct btrfs_free_space *info, u64 offset,
1630 u64 bytes_to_set = 0;
1633 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1635 bytes_to_set = min(end - offset, bytes);
1637 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1639 return bytes_to_set;
1643 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1644 struct btrfs_free_space *info)
1646 struct btrfs_block_group_cache *block_group = ctl->private;
1649 * If we are below the extents threshold then we can add this as an
1650 * extent, and don't have to deal with the bitmap
1652 if (ctl->free_extents < ctl->extents_thresh) {
1654 * If this block group has some small extents we don't want to
1655 * use up all of our free slots in the cache with them, we want
1656 * to reserve them to larger extents, however if we have plent
1657 * of cache left then go ahead an dadd them, no sense in adding
1658 * the overhead of a bitmap if we don't have to.
1660 if (info->bytes <= block_group->sectorsize * 4) {
1661 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1669 * The original block groups from mkfs can be really small, like 8
1670 * megabytes, so don't bother with a bitmap for those entries. However
1671 * some block groups can be smaller than what a bitmap would cover but
1672 * are still large enough that they could overflow the 32k memory limit,
1673 * so allow those block groups to still be allowed to have a bitmap
1676 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1682 static struct btrfs_free_space_op free_space_op = {
1683 .recalc_thresholds = recalculate_thresholds,
1684 .use_bitmap = use_bitmap,
1687 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1688 struct btrfs_free_space *info)
1690 struct btrfs_free_space *bitmap_info;
1691 struct btrfs_block_group_cache *block_group = NULL;
1693 u64 bytes, offset, bytes_added;
1696 bytes = info->bytes;
1697 offset = info->offset;
1699 if (!ctl->op->use_bitmap(ctl, info))
1702 if (ctl->op == &free_space_op)
1703 block_group = ctl->private;
1706 * Since we link bitmaps right into the cluster we need to see if we
1707 * have a cluster here, and if so and it has our bitmap we need to add
1708 * the free space to that bitmap.
1710 if (block_group && !list_empty(&block_group->cluster_list)) {
1711 struct btrfs_free_cluster *cluster;
1712 struct rb_node *node;
1713 struct btrfs_free_space *entry;
1715 cluster = list_entry(block_group->cluster_list.next,
1716 struct btrfs_free_cluster,
1718 spin_lock(&cluster->lock);
1719 node = rb_first(&cluster->root);
1721 spin_unlock(&cluster->lock);
1722 goto no_cluster_bitmap;
1725 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1726 if (!entry->bitmap) {
1727 spin_unlock(&cluster->lock);
1728 goto no_cluster_bitmap;
1731 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1732 bytes_added = add_bytes_to_bitmap(ctl, entry,
1734 bytes -= bytes_added;
1735 offset += bytes_added;
1737 spin_unlock(&cluster->lock);
1745 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1752 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1753 bytes -= bytes_added;
1754 offset += bytes_added;
1764 if (info && info->bitmap) {
1765 add_new_bitmap(ctl, info, offset);
1770 spin_unlock(&ctl->tree_lock);
1772 /* no pre-allocated info, allocate a new one */
1774 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1777 spin_lock(&ctl->tree_lock);
1783 /* allocate the bitmap */
1784 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1785 spin_lock(&ctl->tree_lock);
1786 if (!info->bitmap) {
1796 kfree(info->bitmap);
1797 kmem_cache_free(btrfs_free_space_cachep, info);
1803 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1804 struct btrfs_free_space *info, bool update_stat)
1806 struct btrfs_free_space *left_info;
1807 struct btrfs_free_space *right_info;
1808 bool merged = false;
1809 u64 offset = info->offset;
1810 u64 bytes = info->bytes;
1813 * first we want to see if there is free space adjacent to the range we
1814 * are adding, if there is remove that struct and add a new one to
1815 * cover the entire range
1817 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1818 if (right_info && rb_prev(&right_info->offset_index))
1819 left_info = rb_entry(rb_prev(&right_info->offset_index),
1820 struct btrfs_free_space, offset_index);
1822 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1824 if (right_info && !right_info->bitmap) {
1826 unlink_free_space(ctl, right_info);
1828 __unlink_free_space(ctl, right_info);
1829 info->bytes += right_info->bytes;
1830 kmem_cache_free(btrfs_free_space_cachep, right_info);
1834 if (left_info && !left_info->bitmap &&
1835 left_info->offset + left_info->bytes == offset) {
1837 unlink_free_space(ctl, left_info);
1839 __unlink_free_space(ctl, left_info);
1840 info->offset = left_info->offset;
1841 info->bytes += left_info->bytes;
1842 kmem_cache_free(btrfs_free_space_cachep, left_info);
1849 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1850 u64 offset, u64 bytes)
1852 struct btrfs_free_space *info;
1855 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1859 info->offset = offset;
1860 info->bytes = bytes;
1862 spin_lock(&ctl->tree_lock);
1864 if (try_merge_free_space(ctl, info, true))
1868 * There was no extent directly to the left or right of this new
1869 * extent then we know we're going to have to allocate a new extent, so
1870 * before we do that see if we need to drop this into a bitmap
1872 ret = insert_into_bitmap(ctl, info);
1880 ret = link_free_space(ctl, info);
1882 kmem_cache_free(btrfs_free_space_cachep, info);
1884 spin_unlock(&ctl->tree_lock);
1887 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1888 BUG_ON(ret == -EEXIST);
1894 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1895 u64 offset, u64 bytes)
1897 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1898 struct btrfs_free_space *info;
1900 bool re_search = false;
1902 spin_lock(&ctl->tree_lock);
1909 info = tree_search_offset(ctl, offset, 0, 0);
1912 * oops didn't find an extent that matched the space we wanted
1913 * to remove, look for a bitmap instead
1915 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1919 * If we found a partial bit of our free space in a
1920 * bitmap but then couldn't find the other part this may
1921 * be a problem, so WARN about it.
1929 if (!info->bitmap) {
1930 unlink_free_space(ctl, info);
1931 if (offset == info->offset) {
1932 u64 to_free = min(bytes, info->bytes);
1934 info->bytes -= to_free;
1935 info->offset += to_free;
1937 ret = link_free_space(ctl, info);
1940 kmem_cache_free(btrfs_free_space_cachep, info);
1947 u64 old_end = info->bytes + info->offset;
1949 info->bytes = offset - info->offset;
1950 ret = link_free_space(ctl, info);
1955 /* Not enough bytes in this entry to satisfy us */
1956 if (old_end < offset + bytes) {
1957 bytes -= old_end - offset;
1960 } else if (old_end == offset + bytes) {
1964 spin_unlock(&ctl->tree_lock);
1966 ret = btrfs_add_free_space(block_group, offset + bytes,
1967 old_end - (offset + bytes));
1973 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1974 if (ret == -EAGAIN) {
1979 spin_unlock(&ctl->tree_lock);
1984 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1987 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1988 struct btrfs_free_space *info;
1992 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1993 info = rb_entry(n, struct btrfs_free_space, offset_index);
1994 if (info->bytes >= bytes && !block_group->ro)
1996 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1997 (unsigned long long)info->offset,
1998 (unsigned long long)info->bytes,
1999 (info->bitmap) ? "yes" : "no");
2001 printk(KERN_INFO "block group has cluster?: %s\n",
2002 list_empty(&block_group->cluster_list) ? "no" : "yes");
2003 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2007 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2009 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2011 spin_lock_init(&ctl->tree_lock);
2012 ctl->unit = block_group->sectorsize;
2013 ctl->start = block_group->key.objectid;
2014 ctl->private = block_group;
2015 ctl->op = &free_space_op;
2018 * we only want to have 32k of ram per block group for keeping
2019 * track of free space, and if we pass 1/2 of that we want to
2020 * start converting things over to using bitmaps
2022 ctl->extents_thresh = ((1024 * 32) / 2) /
2023 sizeof(struct btrfs_free_space);
2027 * for a given cluster, put all of its extents back into the free
2028 * space cache. If the block group passed doesn't match the block group
2029 * pointed to by the cluster, someone else raced in and freed the
2030 * cluster already. In that case, we just return without changing anything
2033 __btrfs_return_cluster_to_free_space(
2034 struct btrfs_block_group_cache *block_group,
2035 struct btrfs_free_cluster *cluster)
2037 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2038 struct btrfs_free_space *entry;
2039 struct rb_node *node;
2041 spin_lock(&cluster->lock);
2042 if (cluster->block_group != block_group)
2045 cluster->block_group = NULL;
2046 cluster->window_start = 0;
2047 list_del_init(&cluster->block_group_list);
2049 node = rb_first(&cluster->root);
2053 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2054 node = rb_next(&entry->offset_index);
2055 rb_erase(&entry->offset_index, &cluster->root);
2057 bitmap = (entry->bitmap != NULL);
2059 try_merge_free_space(ctl, entry, false);
2060 tree_insert_offset(&ctl->free_space_offset,
2061 entry->offset, &entry->offset_index, bitmap);
2063 cluster->root = RB_ROOT;
2066 spin_unlock(&cluster->lock);
2067 btrfs_put_block_group(block_group);
2071 static void __btrfs_remove_free_space_cache_locked(
2072 struct btrfs_free_space_ctl *ctl)
2074 struct btrfs_free_space *info;
2075 struct rb_node *node;
2077 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2078 info = rb_entry(node, struct btrfs_free_space, offset_index);
2079 if (!info->bitmap) {
2080 unlink_free_space(ctl, info);
2081 kmem_cache_free(btrfs_free_space_cachep, info);
2083 free_bitmap(ctl, info);
2085 if (need_resched()) {
2086 spin_unlock(&ctl->tree_lock);
2088 spin_lock(&ctl->tree_lock);
2093 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2095 spin_lock(&ctl->tree_lock);
2096 __btrfs_remove_free_space_cache_locked(ctl);
2097 spin_unlock(&ctl->tree_lock);
2100 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2102 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2103 struct btrfs_free_cluster *cluster;
2104 struct list_head *head;
2106 spin_lock(&ctl->tree_lock);
2107 while ((head = block_group->cluster_list.next) !=
2108 &block_group->cluster_list) {
2109 cluster = list_entry(head, struct btrfs_free_cluster,
2112 WARN_ON(cluster->block_group != block_group);
2113 __btrfs_return_cluster_to_free_space(block_group, cluster);
2114 if (need_resched()) {
2115 spin_unlock(&ctl->tree_lock);
2117 spin_lock(&ctl->tree_lock);
2120 __btrfs_remove_free_space_cache_locked(ctl);
2121 spin_unlock(&ctl->tree_lock);
2125 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2126 u64 offset, u64 bytes, u64 empty_size)
2128 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2129 struct btrfs_free_space *entry = NULL;
2130 u64 bytes_search = bytes + empty_size;
2133 u64 align_gap_len = 0;
2135 spin_lock(&ctl->tree_lock);
2136 entry = find_free_space(ctl, &offset, &bytes_search,
2137 block_group->full_stripe_len);
2142 if (entry->bitmap) {
2143 bitmap_clear_bits(ctl, entry, offset, bytes);
2145 free_bitmap(ctl, entry);
2148 unlink_free_space(ctl, entry);
2149 align_gap_len = offset - entry->offset;
2150 align_gap = entry->offset;
2152 entry->offset = offset + bytes;
2153 WARN_ON(entry->bytes < bytes + align_gap_len);
2155 entry->bytes -= bytes + align_gap_len;
2157 kmem_cache_free(btrfs_free_space_cachep, entry);
2159 link_free_space(ctl, entry);
2163 spin_unlock(&ctl->tree_lock);
2166 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2171 * given a cluster, put all of its extents back into the free space
2172 * cache. If a block group is passed, this function will only free
2173 * a cluster that belongs to the passed block group.
2175 * Otherwise, it'll get a reference on the block group pointed to by the
2176 * cluster and remove the cluster from it.
2178 int btrfs_return_cluster_to_free_space(
2179 struct btrfs_block_group_cache *block_group,
2180 struct btrfs_free_cluster *cluster)
2182 struct btrfs_free_space_ctl *ctl;
2185 /* first, get a safe pointer to the block group */
2186 spin_lock(&cluster->lock);
2188 block_group = cluster->block_group;
2190 spin_unlock(&cluster->lock);
2193 } else if (cluster->block_group != block_group) {
2194 /* someone else has already freed it don't redo their work */
2195 spin_unlock(&cluster->lock);
2198 atomic_inc(&block_group->count);
2199 spin_unlock(&cluster->lock);
2201 ctl = block_group->free_space_ctl;
2203 /* now return any extents the cluster had on it */
2204 spin_lock(&ctl->tree_lock);
2205 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2206 spin_unlock(&ctl->tree_lock);
2208 /* finally drop our ref */
2209 btrfs_put_block_group(block_group);
2213 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2214 struct btrfs_free_cluster *cluster,
2215 struct btrfs_free_space *entry,
2216 u64 bytes, u64 min_start)
2218 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2220 u64 search_start = cluster->window_start;
2221 u64 search_bytes = bytes;
2224 search_start = min_start;
2225 search_bytes = bytes;
2227 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2232 __bitmap_clear_bits(ctl, entry, ret, bytes);
2238 * given a cluster, try to allocate 'bytes' from it, returns 0
2239 * if it couldn't find anything suitably large, or a logical disk offset
2240 * if things worked out
2242 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2243 struct btrfs_free_cluster *cluster, u64 bytes,
2246 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2247 struct btrfs_free_space *entry = NULL;
2248 struct rb_node *node;
2251 spin_lock(&cluster->lock);
2252 if (bytes > cluster->max_size)
2255 if (cluster->block_group != block_group)
2258 node = rb_first(&cluster->root);
2262 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2264 if (entry->bytes < bytes ||
2265 (!entry->bitmap && entry->offset < min_start)) {
2266 node = rb_next(&entry->offset_index);
2269 entry = rb_entry(node, struct btrfs_free_space,
2274 if (entry->bitmap) {
2275 ret = btrfs_alloc_from_bitmap(block_group,
2276 cluster, entry, bytes,
2277 cluster->window_start);
2279 node = rb_next(&entry->offset_index);
2282 entry = rb_entry(node, struct btrfs_free_space,
2286 cluster->window_start += bytes;
2288 ret = entry->offset;
2290 entry->offset += bytes;
2291 entry->bytes -= bytes;
2294 if (entry->bytes == 0)
2295 rb_erase(&entry->offset_index, &cluster->root);
2299 spin_unlock(&cluster->lock);
2304 spin_lock(&ctl->tree_lock);
2306 ctl->free_space -= bytes;
2307 if (entry->bytes == 0) {
2308 ctl->free_extents--;
2309 if (entry->bitmap) {
2310 kfree(entry->bitmap);
2311 ctl->total_bitmaps--;
2312 ctl->op->recalc_thresholds(ctl);
2314 kmem_cache_free(btrfs_free_space_cachep, entry);
2317 spin_unlock(&ctl->tree_lock);
2322 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2323 struct btrfs_free_space *entry,
2324 struct btrfs_free_cluster *cluster,
2325 u64 offset, u64 bytes,
2326 u64 cont1_bytes, u64 min_bytes)
2328 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2329 unsigned long next_zero;
2331 unsigned long want_bits;
2332 unsigned long min_bits;
2333 unsigned long found_bits;
2334 unsigned long start = 0;
2335 unsigned long total_found = 0;
2338 i = offset_to_bit(entry->offset, ctl->unit,
2339 max_t(u64, offset, entry->offset));
2340 want_bits = bytes_to_bits(bytes, ctl->unit);
2341 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2345 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2346 next_zero = find_next_zero_bit(entry->bitmap,
2347 BITS_PER_BITMAP, i);
2348 if (next_zero - i >= min_bits) {
2349 found_bits = next_zero - i;
2360 cluster->max_size = 0;
2363 total_found += found_bits;
2365 if (cluster->max_size < found_bits * ctl->unit)
2366 cluster->max_size = found_bits * ctl->unit;
2368 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2373 cluster->window_start = start * ctl->unit + entry->offset;
2374 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2375 ret = tree_insert_offset(&cluster->root, entry->offset,
2376 &entry->offset_index, 1);
2377 BUG_ON(ret); /* -EEXIST; Logic error */
2379 trace_btrfs_setup_cluster(block_group, cluster,
2380 total_found * ctl->unit, 1);
2385 * This searches the block group for just extents to fill the cluster with.
2386 * Try to find a cluster with at least bytes total bytes, at least one
2387 * extent of cont1_bytes, and other clusters of at least min_bytes.
2390 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2391 struct btrfs_free_cluster *cluster,
2392 struct list_head *bitmaps, u64 offset, u64 bytes,
2393 u64 cont1_bytes, u64 min_bytes)
2395 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2396 struct btrfs_free_space *first = NULL;
2397 struct btrfs_free_space *entry = NULL;
2398 struct btrfs_free_space *last;
2399 struct rb_node *node;
2405 entry = tree_search_offset(ctl, offset, 0, 1);
2410 * We don't want bitmaps, so just move along until we find a normal
2413 while (entry->bitmap || entry->bytes < min_bytes) {
2414 if (entry->bitmap && list_empty(&entry->list))
2415 list_add_tail(&entry->list, bitmaps);
2416 node = rb_next(&entry->offset_index);
2419 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2422 window_start = entry->offset;
2423 window_free = entry->bytes;
2424 max_extent = entry->bytes;
2428 for (node = rb_next(&entry->offset_index); node;
2429 node = rb_next(&entry->offset_index)) {
2430 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2432 if (entry->bitmap) {
2433 if (list_empty(&entry->list))
2434 list_add_tail(&entry->list, bitmaps);
2438 if (entry->bytes < min_bytes)
2442 window_free += entry->bytes;
2443 if (entry->bytes > max_extent)
2444 max_extent = entry->bytes;
2447 if (window_free < bytes || max_extent < cont1_bytes)
2450 cluster->window_start = first->offset;
2452 node = &first->offset_index;
2455 * now we've found our entries, pull them out of the free space
2456 * cache and put them into the cluster rbtree
2461 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2462 node = rb_next(&entry->offset_index);
2463 if (entry->bitmap || entry->bytes < min_bytes)
2466 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2467 ret = tree_insert_offset(&cluster->root, entry->offset,
2468 &entry->offset_index, 0);
2469 total_size += entry->bytes;
2470 BUG_ON(ret); /* -EEXIST; Logic error */
2471 } while (node && entry != last);
2473 cluster->max_size = max_extent;
2474 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2479 * This specifically looks for bitmaps that may work in the cluster, we assume
2480 * that we have already failed to find extents that will work.
2483 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2484 struct btrfs_free_cluster *cluster,
2485 struct list_head *bitmaps, u64 offset, u64 bytes,
2486 u64 cont1_bytes, u64 min_bytes)
2488 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2489 struct btrfs_free_space *entry;
2491 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2493 if (ctl->total_bitmaps == 0)
2497 * The bitmap that covers offset won't be in the list unless offset
2498 * is just its start offset.
2500 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2501 if (entry->offset != bitmap_offset) {
2502 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2503 if (entry && list_empty(&entry->list))
2504 list_add(&entry->list, bitmaps);
2507 list_for_each_entry(entry, bitmaps, list) {
2508 if (entry->bytes < bytes)
2510 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2511 bytes, cont1_bytes, min_bytes);
2517 * The bitmaps list has all the bitmaps that record free space
2518 * starting after offset, so no more search is required.
2524 * here we try to find a cluster of blocks in a block group. The goal
2525 * is to find at least bytes+empty_size.
2526 * We might not find them all in one contiguous area.
2528 * returns zero and sets up cluster if things worked out, otherwise
2529 * it returns -enospc
2531 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2532 struct btrfs_root *root,
2533 struct btrfs_block_group_cache *block_group,
2534 struct btrfs_free_cluster *cluster,
2535 u64 offset, u64 bytes, u64 empty_size)
2537 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2538 struct btrfs_free_space *entry, *tmp;
2545 * Choose the minimum extent size we'll require for this
2546 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2547 * For metadata, allow allocates with smaller extents. For
2548 * data, keep it dense.
2550 if (btrfs_test_opt(root, SSD_SPREAD)) {
2551 cont1_bytes = min_bytes = bytes + empty_size;
2552 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2553 cont1_bytes = bytes;
2554 min_bytes = block_group->sectorsize;
2556 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2557 min_bytes = block_group->sectorsize;
2560 spin_lock(&ctl->tree_lock);
2563 * If we know we don't have enough space to make a cluster don't even
2564 * bother doing all the work to try and find one.
2566 if (ctl->free_space < bytes) {
2567 spin_unlock(&ctl->tree_lock);
2571 spin_lock(&cluster->lock);
2573 /* someone already found a cluster, hooray */
2574 if (cluster->block_group) {
2579 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2582 INIT_LIST_HEAD(&bitmaps);
2583 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2585 cont1_bytes, min_bytes);
2587 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2588 offset, bytes + empty_size,
2589 cont1_bytes, min_bytes);
2591 /* Clear our temporary list */
2592 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2593 list_del_init(&entry->list);
2596 atomic_inc(&block_group->count);
2597 list_add_tail(&cluster->block_group_list,
2598 &block_group->cluster_list);
2599 cluster->block_group = block_group;
2601 trace_btrfs_failed_cluster_setup(block_group);
2604 spin_unlock(&cluster->lock);
2605 spin_unlock(&ctl->tree_lock);
2611 * simple code to zero out a cluster
2613 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2615 spin_lock_init(&cluster->lock);
2616 spin_lock_init(&cluster->refill_lock);
2617 cluster->root = RB_ROOT;
2618 cluster->max_size = 0;
2619 INIT_LIST_HEAD(&cluster->block_group_list);
2620 cluster->block_group = NULL;
2623 static int do_trimming(struct btrfs_block_group_cache *block_group,
2624 u64 *total_trimmed, u64 start, u64 bytes,
2625 u64 reserved_start, u64 reserved_bytes)
2627 struct btrfs_space_info *space_info = block_group->space_info;
2628 struct btrfs_fs_info *fs_info = block_group->fs_info;
2633 spin_lock(&space_info->lock);
2634 spin_lock(&block_group->lock);
2635 if (!block_group->ro) {
2636 block_group->reserved += reserved_bytes;
2637 space_info->bytes_reserved += reserved_bytes;
2640 spin_unlock(&block_group->lock);
2641 spin_unlock(&space_info->lock);
2643 ret = btrfs_error_discard_extent(fs_info->extent_root,
2644 start, bytes, &trimmed);
2646 *total_trimmed += trimmed;
2648 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2651 spin_lock(&space_info->lock);
2652 spin_lock(&block_group->lock);
2653 if (block_group->ro)
2654 space_info->bytes_readonly += reserved_bytes;
2655 block_group->reserved -= reserved_bytes;
2656 space_info->bytes_reserved -= reserved_bytes;
2657 spin_unlock(&space_info->lock);
2658 spin_unlock(&block_group->lock);
2664 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2665 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2667 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2668 struct btrfs_free_space *entry;
2669 struct rb_node *node;
2675 while (start < end) {
2676 spin_lock(&ctl->tree_lock);
2678 if (ctl->free_space < minlen) {
2679 spin_unlock(&ctl->tree_lock);
2683 entry = tree_search_offset(ctl, start, 0, 1);
2685 spin_unlock(&ctl->tree_lock);
2690 while (entry->bitmap) {
2691 node = rb_next(&entry->offset_index);
2693 spin_unlock(&ctl->tree_lock);
2696 entry = rb_entry(node, struct btrfs_free_space,
2700 if (entry->offset >= end) {
2701 spin_unlock(&ctl->tree_lock);
2705 extent_start = entry->offset;
2706 extent_bytes = entry->bytes;
2707 start = max(start, extent_start);
2708 bytes = min(extent_start + extent_bytes, end) - start;
2709 if (bytes < minlen) {
2710 spin_unlock(&ctl->tree_lock);
2714 unlink_free_space(ctl, entry);
2715 kmem_cache_free(btrfs_free_space_cachep, entry);
2717 spin_unlock(&ctl->tree_lock);
2719 ret = do_trimming(block_group, total_trimmed, start, bytes,
2720 extent_start, extent_bytes);
2726 if (fatal_signal_pending(current)) {
2737 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2738 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2740 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2741 struct btrfs_free_space *entry;
2745 u64 offset = offset_to_bitmap(ctl, start);
2747 while (offset < end) {
2748 bool next_bitmap = false;
2750 spin_lock(&ctl->tree_lock);
2752 if (ctl->free_space < minlen) {
2753 spin_unlock(&ctl->tree_lock);
2757 entry = tree_search_offset(ctl, offset, 1, 0);
2759 spin_unlock(&ctl->tree_lock);
2765 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2766 if (ret2 || start >= end) {
2767 spin_unlock(&ctl->tree_lock);
2772 bytes = min(bytes, end - start);
2773 if (bytes < minlen) {
2774 spin_unlock(&ctl->tree_lock);
2778 bitmap_clear_bits(ctl, entry, start, bytes);
2779 if (entry->bytes == 0)
2780 free_bitmap(ctl, entry);
2782 spin_unlock(&ctl->tree_lock);
2784 ret = do_trimming(block_group, total_trimmed, start, bytes,
2790 offset += BITS_PER_BITMAP * ctl->unit;
2793 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2794 offset += BITS_PER_BITMAP * ctl->unit;
2797 if (fatal_signal_pending(current)) {
2808 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2809 u64 *trimmed, u64 start, u64 end, u64 minlen)
2815 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2819 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2825 * Find the left-most item in the cache tree, and then return the
2826 * smallest inode number in the item.
2828 * Note: the returned inode number may not be the smallest one in
2829 * the tree, if the left-most item is a bitmap.
2831 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2833 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2834 struct btrfs_free_space *entry = NULL;
2837 spin_lock(&ctl->tree_lock);
2839 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2842 entry = rb_entry(rb_first(&ctl->free_space_offset),
2843 struct btrfs_free_space, offset_index);
2845 if (!entry->bitmap) {
2846 ino = entry->offset;
2848 unlink_free_space(ctl, entry);
2852 kmem_cache_free(btrfs_free_space_cachep, entry);
2854 link_free_space(ctl, entry);
2860 ret = search_bitmap(ctl, entry, &offset, &count);
2861 /* Logic error; Should be empty if it can't find anything */
2865 bitmap_clear_bits(ctl, entry, offset, 1);
2866 if (entry->bytes == 0)
2867 free_bitmap(ctl, entry);
2870 spin_unlock(&ctl->tree_lock);
2875 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2876 struct btrfs_path *path)
2878 struct inode *inode = NULL;
2880 spin_lock(&root->cache_lock);
2881 if (root->cache_inode)
2882 inode = igrab(root->cache_inode);
2883 spin_unlock(&root->cache_lock);
2887 inode = __lookup_free_space_inode(root, path, 0);
2891 spin_lock(&root->cache_lock);
2892 if (!btrfs_fs_closing(root->fs_info))
2893 root->cache_inode = igrab(inode);
2894 spin_unlock(&root->cache_lock);
2899 int create_free_ino_inode(struct btrfs_root *root,
2900 struct btrfs_trans_handle *trans,
2901 struct btrfs_path *path)
2903 return __create_free_space_inode(root, trans, path,
2904 BTRFS_FREE_INO_OBJECTID, 0);
2907 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2909 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2910 struct btrfs_path *path;
2911 struct inode *inode;
2913 u64 root_gen = btrfs_root_generation(&root->root_item);
2915 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2919 * If we're unmounting then just return, since this does a search on the
2920 * normal root and not the commit root and we could deadlock.
2922 if (btrfs_fs_closing(fs_info))
2925 path = btrfs_alloc_path();
2929 inode = lookup_free_ino_inode(root, path);
2933 if (root_gen != BTRFS_I(inode)->generation)
2936 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2940 "failed to load free ino cache for root %llu",
2941 root->root_key.objectid);
2945 btrfs_free_path(path);
2949 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2950 struct btrfs_trans_handle *trans,
2951 struct btrfs_path *path)
2953 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2954 struct inode *inode;
2957 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2960 inode = lookup_free_ino_inode(root, path);
2964 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2966 btrfs_delalloc_release_metadata(inode, inode->i_size);
2968 btrfs_err(root->fs_info,
2969 "failed to write free ino cache for root %llu",
2970 root->root_key.objectid);
2978 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2979 static struct btrfs_block_group_cache *init_test_block_group(void)
2981 struct btrfs_block_group_cache *cache;
2983 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2986 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
2988 if (!cache->free_space_ctl) {
2993 cache->key.objectid = 0;
2994 cache->key.offset = 1024 * 1024 * 1024;
2995 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2996 cache->sectorsize = 4096;
2998 spin_lock_init(&cache->lock);
2999 INIT_LIST_HEAD(&cache->list);
3000 INIT_LIST_HEAD(&cache->cluster_list);
3001 INIT_LIST_HEAD(&cache->new_bg_list);
3003 btrfs_init_free_space_ctl(cache);
3009 * Checks to see if the given range is in the free space cache. This is really
3010 * just used to check the absence of space, so if there is free space in the
3011 * range at all we will return 1.
3013 static int check_exists(struct btrfs_block_group_cache *cache, u64 offset,
3016 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3017 struct btrfs_free_space *info;
3020 spin_lock(&ctl->tree_lock);
3021 info = tree_search_offset(ctl, offset, 0, 0);
3023 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3031 u64 bit_off, bit_bytes;
3033 struct btrfs_free_space *tmp;
3036 bit_bytes = ctl->unit;
3037 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3039 if (bit_off == offset) {
3042 } else if (bit_off > offset &&
3043 offset + bytes > bit_off) {
3049 n = rb_prev(&info->offset_index);
3051 tmp = rb_entry(n, struct btrfs_free_space,
3053 if (tmp->offset + tmp->bytes < offset)
3055 if (offset + bytes < tmp->offset) {
3056 n = rb_prev(&info->offset_index);
3063 n = rb_next(&info->offset_index);
3065 tmp = rb_entry(n, struct btrfs_free_space,
3067 if (offset + bytes < tmp->offset)
3069 if (tmp->offset + tmp->bytes < offset) {
3070 n = rb_next(&info->offset_index);
3080 if (info->offset == offset) {
3085 if (offset > info->offset && offset < info->offset + info->bytes)
3088 spin_unlock(&ctl->tree_lock);
3093 * Use this if you need to make a bitmap or extent entry specifically, it
3094 * doesn't do any of the merging that add_free_space does, this acts a lot like
3095 * how the free space cache loading stuff works, so you can get really weird
3098 static int add_free_space_entry(struct btrfs_block_group_cache *cache,
3099 u64 offset, u64 bytes, bool bitmap)
3101 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3102 struct btrfs_free_space *info = NULL, *bitmap_info;
3109 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3115 spin_lock(&ctl->tree_lock);
3116 info->offset = offset;
3117 info->bytes = bytes;
3118 ret = link_free_space(ctl, info);
3119 spin_unlock(&ctl->tree_lock);
3121 kmem_cache_free(btrfs_free_space_cachep, info);
3126 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3128 kmem_cache_free(btrfs_free_space_cachep, info);
3133 spin_lock(&ctl->tree_lock);
3134 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3139 add_new_bitmap(ctl, info, offset);
3143 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3144 bytes -= bytes_added;
3145 offset += bytes_added;
3146 spin_unlock(&ctl->tree_lock);
3157 * This test just does basic sanity checking, making sure we can add an exten
3158 * entry and remove space from either end and the middle, and make sure we can
3159 * remove space that covers adjacent extent entries.
3161 static int test_extents(struct btrfs_block_group_cache *cache)
3165 printk(KERN_ERR "Running extent only tests\n");
3167 /* First just make sure we can remove an entire entry */
3168 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3170 printk(KERN_ERR "Error adding initial extents %d\n", ret);
3174 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3176 printk(KERN_ERR "Error removing extent %d\n", ret);
3180 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3181 printk(KERN_ERR "Full remove left some lingering space\n");
3185 /* Ok edge and middle cases now */
3186 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3188 printk(KERN_ERR "Error adding half extent %d\n", ret);
3192 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 1 * 1024 * 1024);
3194 printk(KERN_ERR "Error removing tail end %d\n", ret);
3198 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3200 printk(KERN_ERR "Error removing front end %d\n", ret);
3204 ret = btrfs_remove_free_space(cache, 2 * 1024 * 1024, 4096);
3206 printk(KERN_ERR "Error removing middle peice %d\n", ret);
3210 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3211 printk(KERN_ERR "Still have space at the front\n");
3215 if (check_exists(cache, 2 * 1024 * 1024, 4096)) {
3216 printk(KERN_ERR "Still have space in the middle\n");
3220 if (check_exists(cache, 3 * 1024 * 1024, 1 * 1024 * 1024)) {
3221 printk(KERN_ERR "Still have space at the end\n");
3226 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3231 static int test_bitmaps(struct btrfs_block_group_cache *cache)
3233 u64 next_bitmap_offset;
3236 printk(KERN_ERR "Running bitmap only tests\n");
3238 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3240 printk(KERN_ERR "Couldn't create a bitmap entry %d\n", ret);
3244 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3246 printk(KERN_ERR "Error removing bitmap full range %d\n", ret);
3250 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3251 printk(KERN_ERR "Left some space in bitmap\n");
3255 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3257 printk(KERN_ERR "Couldn't add to our bitmap entry %d\n", ret);
3261 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 2 * 1024 * 1024);
3263 printk(KERN_ERR "Couldn't remove middle chunk %d\n", ret);
3268 * The first bitmap we have starts at offset 0 so the next one is just
3269 * at the end of the first bitmap.
3271 next_bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3273 /* Test a bit straddling two bitmaps */
3274 ret = add_free_space_entry(cache, next_bitmap_offset -
3275 (2 * 1024 * 1024), 4 * 1024 * 1024, 1);
3277 printk(KERN_ERR "Couldn't add space that straddles two bitmaps"
3282 ret = btrfs_remove_free_space(cache, next_bitmap_offset -
3283 (1 * 1024 * 1024), 2 * 1024 * 1024);
3285 printk(KERN_ERR "Couldn't remove overlapping space %d\n", ret);
3289 if (check_exists(cache, next_bitmap_offset - (1 * 1024 * 1024),
3291 printk(KERN_ERR "Left some space when removing overlapping\n");
3295 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3300 /* This is the high grade jackassery */
3301 static int test_bitmaps_and_extents(struct btrfs_block_group_cache *cache)
3303 u64 bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3306 printk(KERN_ERR "Running bitmap and extent tests\n");
3309 * First let's do something simple, an extent at the same offset as the
3310 * bitmap, but the free space completely in the extent and then
3311 * completely in the bitmap.
3313 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 1 * 1024 * 1024, 1);
3315 printk(KERN_ERR "Couldn't create bitmap entry %d\n", ret);
3319 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3321 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3325 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3327 printk(KERN_ERR "Couldn't remove extent entry %d\n", ret);
3331 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3332 printk(KERN_ERR "Left remnants after our remove\n");
3336 /* Now to add back the extent entry and remove from the bitmap */
3337 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3339 printk(KERN_ERR "Couldn't re-add extent entry %d\n", ret);
3343 ret = btrfs_remove_free_space(cache, 4 * 1024 * 1024, 1 * 1024 * 1024);
3345 printk(KERN_ERR "Couldn't remove from bitmap %d\n", ret);
3349 if (check_exists(cache, 4 * 1024 * 1024, 1 * 1024 * 1024)) {
3350 printk(KERN_ERR "Left remnants in the bitmap\n");
3355 * Ok so a little more evil, extent entry and bitmap at the same offset,
3356 * removing an overlapping chunk.
3358 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 4 * 1024 * 1024, 1);
3360 printk(KERN_ERR "Couldn't add to a bitmap %d\n", ret);
3364 ret = btrfs_remove_free_space(cache, 512 * 1024, 3 * 1024 * 1024);
3366 printk(KERN_ERR "Couldn't remove overlapping space %d\n", ret);
3370 if (check_exists(cache, 512 * 1024, 3 * 1024 * 1024)) {
3371 printk(KERN_ERR "Left over peices after removing "
3376 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3378 /* Now with the extent entry offset into the bitmap */
3379 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 4 * 1024 * 1024, 1);
3381 printk(KERN_ERR "Couldn't add space to the bitmap %d\n", ret);
3385 ret = add_free_space_entry(cache, 2 * 1024 * 1024, 2 * 1024 * 1024, 0);
3387 printk(KERN_ERR "Couldn't add extent to the cache %d\n", ret);
3391 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 4 * 1024 * 1024);
3393 printk(KERN_ERR "Problem removing overlapping space %d\n", ret);
3397 if (check_exists(cache, 3 * 1024 * 1024, 4 * 1024 * 1024)) {
3398 printk(KERN_ERR "Left something behind when removing space");
3403 * This has blown up in the past, the extent entry starts before the
3404 * bitmap entry, but we're trying to remove an offset that falls
3405 * completely within the bitmap range and is in both the extent entry
3406 * and the bitmap entry, looks like this
3412 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3413 ret = add_free_space_entry(cache, bitmap_offset + 4 * 1024 * 1024,
3414 4 * 1024 * 1024, 1);
3416 printk(KERN_ERR "Couldn't add bitmap %d\n", ret);
3420 ret = add_free_space_entry(cache, bitmap_offset - 1 * 1024 * 1024,
3421 5 * 1024 * 1024, 0);
3423 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3427 ret = btrfs_remove_free_space(cache, bitmap_offset + 1 * 1024 * 1024,
3430 printk(KERN_ERR "Failed to free our space %d\n", ret);
3434 if (check_exists(cache, bitmap_offset + 1 * 1024 * 1024,
3436 printk(KERN_ERR "Left stuff over\n");
3440 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3443 * This blew up before, we have part of the free space in a bitmap and
3444 * then the entirety of the rest of the space in an extent. This used
3445 * to return -EAGAIN back from btrfs_remove_extent, make sure this
3448 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 2 * 1024 * 1024, 1);
3450 printk(KERN_ERR "Couldn't add bitmap entry %d\n", ret);
3454 ret = add_free_space_entry(cache, 3 * 1024 * 1024, 1 * 1024 * 1024, 0);
3456 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3460 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 3 * 1024 * 1024);
3462 printk(KERN_ERR "Error removing bitmap and extent "
3463 "overlapping %d\n", ret);
3467 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3471 void btrfs_test_free_space_cache(void)
3473 struct btrfs_block_group_cache *cache;
3475 printk(KERN_ERR "Running btrfs free space cache tests\n");
3477 cache = init_test_block_group();
3479 printk(KERN_ERR "Couldn't run the tests\n");
3483 if (test_extents(cache))
3485 if (test_bitmaps(cache))
3487 if (test_bitmaps_and_extents(cache))
3490 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3491 kfree(cache->free_space_ctl);
3493 printk(KERN_ERR "Free space cache tests finished\n");
3495 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */