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 ASSERT(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 BTRFS_I(inode)->generation, generation);
681 ret = io_ctl_init(&io_ctl, inode, root);
685 ret = readahead_cache(inode);
689 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
693 ret = io_ctl_check_crc(&io_ctl, 0);
697 ret = io_ctl_check_generation(&io_ctl, generation);
701 while (num_entries) {
702 e = kmem_cache_zalloc(btrfs_free_space_cachep,
707 ret = io_ctl_read_entry(&io_ctl, e, &type);
709 kmem_cache_free(btrfs_free_space_cachep, e);
714 kmem_cache_free(btrfs_free_space_cachep, e);
718 if (type == BTRFS_FREE_SPACE_EXTENT) {
719 spin_lock(&ctl->tree_lock);
720 ret = link_free_space(ctl, e);
721 spin_unlock(&ctl->tree_lock);
723 btrfs_err(root->fs_info,
724 "Duplicate entries in free space cache, dumping");
725 kmem_cache_free(btrfs_free_space_cachep, e);
731 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
734 btrfs_free_space_cachep, e);
737 spin_lock(&ctl->tree_lock);
738 ret = link_free_space(ctl, e);
739 ctl->total_bitmaps++;
740 ctl->op->recalc_thresholds(ctl);
741 spin_unlock(&ctl->tree_lock);
743 btrfs_err(root->fs_info,
744 "Duplicate entries in free space cache, dumping");
745 kmem_cache_free(btrfs_free_space_cachep, e);
748 list_add_tail(&e->list, &bitmaps);
754 io_ctl_unmap_page(&io_ctl);
757 * We add the bitmaps at the end of the entries in order that
758 * the bitmap entries are added to the cache.
760 list_for_each_entry_safe(e, n, &bitmaps, list) {
761 list_del_init(&e->list);
762 ret = io_ctl_read_bitmap(&io_ctl, e);
767 io_ctl_drop_pages(&io_ctl);
768 merge_space_tree(ctl);
771 io_ctl_free(&io_ctl);
774 io_ctl_drop_pages(&io_ctl);
775 __btrfs_remove_free_space_cache(ctl);
779 int load_free_space_cache(struct btrfs_fs_info *fs_info,
780 struct btrfs_block_group_cache *block_group)
782 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
783 struct btrfs_root *root = fs_info->tree_root;
785 struct btrfs_path *path;
788 u64 used = btrfs_block_group_used(&block_group->item);
791 * If this block group has been marked to be cleared for one reason or
792 * another then we can't trust the on disk cache, so just return.
794 spin_lock(&block_group->lock);
795 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
796 spin_unlock(&block_group->lock);
799 spin_unlock(&block_group->lock);
801 path = btrfs_alloc_path();
804 path->search_commit_root = 1;
805 path->skip_locking = 1;
807 inode = lookup_free_space_inode(root, block_group, path);
809 btrfs_free_path(path);
813 /* We may have converted the inode and made the cache invalid. */
814 spin_lock(&block_group->lock);
815 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
816 spin_unlock(&block_group->lock);
817 btrfs_free_path(path);
820 spin_unlock(&block_group->lock);
822 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
823 path, block_group->key.objectid);
824 btrfs_free_path(path);
828 spin_lock(&ctl->tree_lock);
829 matched = (ctl->free_space == (block_group->key.offset - used -
830 block_group->bytes_super));
831 spin_unlock(&ctl->tree_lock);
834 __btrfs_remove_free_space_cache(ctl);
835 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
836 block_group->key.objectid);
841 /* This cache is bogus, make sure it gets cleared */
842 spin_lock(&block_group->lock);
843 block_group->disk_cache_state = BTRFS_DC_CLEAR;
844 spin_unlock(&block_group->lock);
847 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
848 block_group->key.objectid);
856 * __btrfs_write_out_cache - write out cached info to an inode
857 * @root - the root the inode belongs to
858 * @ctl - the free space cache we are going to write out
859 * @block_group - the block_group for this cache if it belongs to a block_group
860 * @trans - the trans handle
861 * @path - the path to use
862 * @offset - the offset for the key we'll insert
864 * This function writes out a free space cache struct to disk for quick recovery
865 * on mount. This will return 0 if it was successfull in writing the cache out,
866 * and -1 if it was not.
868 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
869 struct btrfs_free_space_ctl *ctl,
870 struct btrfs_block_group_cache *block_group,
871 struct btrfs_trans_handle *trans,
872 struct btrfs_path *path, u64 offset)
874 struct btrfs_free_space_header *header;
875 struct extent_buffer *leaf;
876 struct rb_node *node;
877 struct list_head *pos, *n;
878 struct extent_state *cached_state = NULL;
879 struct btrfs_free_cluster *cluster = NULL;
880 struct extent_io_tree *unpin = NULL;
881 struct io_ctl io_ctl;
882 struct list_head bitmap_list;
883 struct btrfs_key key;
884 u64 start, extent_start, extent_end, len;
890 INIT_LIST_HEAD(&bitmap_list);
892 if (!i_size_read(inode))
895 ret = io_ctl_init(&io_ctl, inode, root);
899 /* Get the cluster for this block_group if it exists */
900 if (block_group && !list_empty(&block_group->cluster_list))
901 cluster = list_entry(block_group->cluster_list.next,
902 struct btrfs_free_cluster,
905 /* Lock all pages first so we can lock the extent safely. */
906 io_ctl_prepare_pages(&io_ctl, inode, 0);
908 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
911 node = rb_first(&ctl->free_space_offset);
912 if (!node && cluster) {
913 node = rb_first(&cluster->root);
917 /* Make sure we can fit our crcs into the first page */
918 if (io_ctl.check_crcs &&
919 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
922 io_ctl_set_generation(&io_ctl, trans->transid);
924 /* Write out the extent entries */
926 struct btrfs_free_space *e;
928 e = rb_entry(node, struct btrfs_free_space, offset_index);
931 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
937 list_add_tail(&e->list, &bitmap_list);
940 node = rb_next(node);
941 if (!node && cluster) {
942 node = rb_first(&cluster->root);
948 * We want to add any pinned extents to our free space cache
949 * so we don't leak the space
953 * We shouldn't have switched the pinned extents yet so this is the
956 unpin = root->fs_info->pinned_extents;
959 start = block_group->key.objectid;
961 while (block_group && (start < block_group->key.objectid +
962 block_group->key.offset)) {
963 ret = find_first_extent_bit(unpin, start,
964 &extent_start, &extent_end,
971 /* This pinned extent is out of our range */
972 if (extent_start >= block_group->key.objectid +
973 block_group->key.offset)
976 extent_start = max(extent_start, start);
977 extent_end = min(block_group->key.objectid +
978 block_group->key.offset, extent_end + 1);
979 len = extent_end - extent_start;
982 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
989 /* Write out the bitmaps */
990 list_for_each_safe(pos, n, &bitmap_list) {
991 struct btrfs_free_space *entry =
992 list_entry(pos, struct btrfs_free_space, list);
994 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
997 list_del_init(&entry->list);
1000 /* Zero out the rest of the pages just to make sure */
1001 io_ctl_zero_remaining_pages(&io_ctl);
1003 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1004 0, i_size_read(inode), &cached_state);
1005 io_ctl_drop_pages(&io_ctl);
1006 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1007 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1013 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1015 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1016 key.offset = offset;
1019 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1021 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1022 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1026 leaf = path->nodes[0];
1028 struct btrfs_key found_key;
1029 ASSERT(path->slots[0]);
1031 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1032 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1033 found_key.offset != offset) {
1034 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1036 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1038 btrfs_release_path(path);
1043 BTRFS_I(inode)->generation = trans->transid;
1044 header = btrfs_item_ptr(leaf, path->slots[0],
1045 struct btrfs_free_space_header);
1046 btrfs_set_free_space_entries(leaf, header, entries);
1047 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1048 btrfs_set_free_space_generation(leaf, header, trans->transid);
1049 btrfs_mark_buffer_dirty(leaf);
1050 btrfs_release_path(path);
1054 io_ctl_free(&io_ctl);
1056 invalidate_inode_pages2(inode->i_mapping);
1057 BTRFS_I(inode)->generation = 0;
1059 btrfs_update_inode(trans, root, inode);
1063 list_for_each_safe(pos, n, &bitmap_list) {
1064 struct btrfs_free_space *entry =
1065 list_entry(pos, struct btrfs_free_space, list);
1066 list_del_init(&entry->list);
1068 io_ctl_drop_pages(&io_ctl);
1069 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1070 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1074 int btrfs_write_out_cache(struct btrfs_root *root,
1075 struct btrfs_trans_handle *trans,
1076 struct btrfs_block_group_cache *block_group,
1077 struct btrfs_path *path)
1079 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1080 struct inode *inode;
1083 root = root->fs_info->tree_root;
1085 spin_lock(&block_group->lock);
1086 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1087 spin_unlock(&block_group->lock);
1090 spin_unlock(&block_group->lock);
1092 inode = lookup_free_space_inode(root, block_group, path);
1096 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1097 path, block_group->key.objectid);
1099 spin_lock(&block_group->lock);
1100 block_group->disk_cache_state = BTRFS_DC_ERROR;
1101 spin_unlock(&block_group->lock);
1104 btrfs_err(root->fs_info,
1105 "failed to write free space cache for block group %llu",
1106 block_group->key.objectid);
1114 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1117 ASSERT(offset >= bitmap_start);
1118 offset -= bitmap_start;
1119 return (unsigned long)(div_u64(offset, unit));
1122 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1124 return (unsigned long)(div_u64(bytes, unit));
1127 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1131 u64 bytes_per_bitmap;
1133 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1134 bitmap_start = offset - ctl->start;
1135 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1136 bitmap_start *= bytes_per_bitmap;
1137 bitmap_start += ctl->start;
1139 return bitmap_start;
1142 static int tree_insert_offset(struct rb_root *root, u64 offset,
1143 struct rb_node *node, int bitmap)
1145 struct rb_node **p = &root->rb_node;
1146 struct rb_node *parent = NULL;
1147 struct btrfs_free_space *info;
1151 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1153 if (offset < info->offset) {
1155 } else if (offset > info->offset) {
1156 p = &(*p)->rb_right;
1159 * we could have a bitmap entry and an extent entry
1160 * share the same offset. If this is the case, we want
1161 * the extent entry to always be found first if we do a
1162 * linear search through the tree, since we want to have
1163 * the quickest allocation time, and allocating from an
1164 * extent is faster than allocating from a bitmap. So
1165 * if we're inserting a bitmap and we find an entry at
1166 * this offset, we want to go right, or after this entry
1167 * logically. If we are inserting an extent and we've
1168 * found a bitmap, we want to go left, or before
1176 p = &(*p)->rb_right;
1178 if (!info->bitmap) {
1187 rb_link_node(node, parent, p);
1188 rb_insert_color(node, root);
1194 * searches the tree for the given offset.
1196 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1197 * want a section that has at least bytes size and comes at or after the given
1200 static struct btrfs_free_space *
1201 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1202 u64 offset, int bitmap_only, int fuzzy)
1204 struct rb_node *n = ctl->free_space_offset.rb_node;
1205 struct btrfs_free_space *entry, *prev = NULL;
1207 /* find entry that is closest to the 'offset' */
1214 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1217 if (offset < entry->offset)
1219 else if (offset > entry->offset)
1232 * bitmap entry and extent entry may share same offset,
1233 * in that case, bitmap entry comes after extent entry.
1238 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1239 if (entry->offset != offset)
1242 WARN_ON(!entry->bitmap);
1245 if (entry->bitmap) {
1247 * if previous extent entry covers the offset,
1248 * we should return it instead of the bitmap entry
1250 n = rb_prev(&entry->offset_index);
1252 prev = rb_entry(n, struct btrfs_free_space,
1254 if (!prev->bitmap &&
1255 prev->offset + prev->bytes > offset)
1265 /* find last entry before the 'offset' */
1267 if (entry->offset > offset) {
1268 n = rb_prev(&entry->offset_index);
1270 entry = rb_entry(n, struct btrfs_free_space,
1272 ASSERT(entry->offset <= offset);
1281 if (entry->bitmap) {
1282 n = rb_prev(&entry->offset_index);
1284 prev = rb_entry(n, struct btrfs_free_space,
1286 if (!prev->bitmap &&
1287 prev->offset + prev->bytes > offset)
1290 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1292 } else if (entry->offset + entry->bytes > offset)
1299 if (entry->bitmap) {
1300 if (entry->offset + BITS_PER_BITMAP *
1304 if (entry->offset + entry->bytes > offset)
1308 n = rb_next(&entry->offset_index);
1311 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1317 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1318 struct btrfs_free_space *info)
1320 rb_erase(&info->offset_index, &ctl->free_space_offset);
1321 ctl->free_extents--;
1324 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1325 struct btrfs_free_space *info)
1327 __unlink_free_space(ctl, info);
1328 ctl->free_space -= info->bytes;
1331 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1332 struct btrfs_free_space *info)
1336 ASSERT(info->bytes || info->bitmap);
1337 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1338 &info->offset_index, (info->bitmap != NULL));
1342 ctl->free_space += info->bytes;
1343 ctl->free_extents++;
1347 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1349 struct btrfs_block_group_cache *block_group = ctl->private;
1353 u64 size = block_group->key.offset;
1354 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1355 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1357 max_bitmaps = max(max_bitmaps, 1);
1359 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1362 * The goal is to keep the total amount of memory used per 1gb of space
1363 * at or below 32k, so we need to adjust how much memory we allow to be
1364 * used by extent based free space tracking
1366 if (size < 1024 * 1024 * 1024)
1367 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1369 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1370 div64_u64(size, 1024 * 1024 * 1024);
1373 * we want to account for 1 more bitmap than what we have so we can make
1374 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1375 * we add more bitmaps.
1377 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1379 if (bitmap_bytes >= max_bytes) {
1380 ctl->extents_thresh = 0;
1385 * we want the extent entry threshold to always be at most 1/2 the maxw
1386 * bytes we can have, or whatever is less than that.
1388 extent_bytes = max_bytes - bitmap_bytes;
1389 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1391 ctl->extents_thresh =
1392 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1395 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1396 struct btrfs_free_space *info,
1397 u64 offset, u64 bytes)
1399 unsigned long start, count;
1401 start = offset_to_bit(info->offset, ctl->unit, offset);
1402 count = bytes_to_bits(bytes, ctl->unit);
1403 ASSERT(start + count <= BITS_PER_BITMAP);
1405 bitmap_clear(info->bitmap, start, count);
1407 info->bytes -= bytes;
1410 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1411 struct btrfs_free_space *info, u64 offset,
1414 __bitmap_clear_bits(ctl, info, offset, bytes);
1415 ctl->free_space -= bytes;
1418 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1419 struct btrfs_free_space *info, u64 offset,
1422 unsigned long start, count;
1424 start = offset_to_bit(info->offset, ctl->unit, offset);
1425 count = bytes_to_bits(bytes, ctl->unit);
1426 ASSERT(start + count <= BITS_PER_BITMAP);
1428 bitmap_set(info->bitmap, start, count);
1430 info->bytes += bytes;
1431 ctl->free_space += bytes;
1434 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1435 struct btrfs_free_space *bitmap_info, u64 *offset,
1438 unsigned long found_bits = 0;
1439 unsigned long bits, i;
1440 unsigned long next_zero;
1442 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1443 max_t(u64, *offset, bitmap_info->offset));
1444 bits = bytes_to_bits(*bytes, ctl->unit);
1446 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1447 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1448 BITS_PER_BITMAP, i);
1449 if ((next_zero - i) >= bits) {
1450 found_bits = next_zero - i;
1457 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1458 *bytes = (u64)(found_bits) * ctl->unit;
1465 static struct btrfs_free_space *
1466 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1467 unsigned long align)
1469 struct btrfs_free_space *entry;
1470 struct rb_node *node;
1476 if (!ctl->free_space_offset.rb_node)
1479 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1483 for (node = &entry->offset_index; node; node = rb_next(node)) {
1484 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1485 if (entry->bytes < *bytes)
1488 /* make sure the space returned is big enough
1489 * to match our requested alignment
1491 if (*bytes >= align) {
1492 ctl_off = entry->offset - ctl->start;
1493 tmp = ctl_off + align - 1;;
1495 tmp = tmp * align + ctl->start;
1496 align_off = tmp - entry->offset;
1499 tmp = entry->offset;
1502 if (entry->bytes < *bytes + align_off)
1505 if (entry->bitmap) {
1506 ret = search_bitmap(ctl, entry, &tmp, bytes);
1515 *bytes = entry->bytes - align_off;
1522 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1523 struct btrfs_free_space *info, u64 offset)
1525 info->offset = offset_to_bitmap(ctl, offset);
1527 INIT_LIST_HEAD(&info->list);
1528 link_free_space(ctl, info);
1529 ctl->total_bitmaps++;
1531 ctl->op->recalc_thresholds(ctl);
1534 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1535 struct btrfs_free_space *bitmap_info)
1537 unlink_free_space(ctl, bitmap_info);
1538 kfree(bitmap_info->bitmap);
1539 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1540 ctl->total_bitmaps--;
1541 ctl->op->recalc_thresholds(ctl);
1544 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1545 struct btrfs_free_space *bitmap_info,
1546 u64 *offset, u64 *bytes)
1549 u64 search_start, search_bytes;
1553 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1556 * We need to search for bits in this bitmap. We could only cover some
1557 * of the extent in this bitmap thanks to how we add space, so we need
1558 * to search for as much as it as we can and clear that amount, and then
1559 * go searching for the next bit.
1561 search_start = *offset;
1562 search_bytes = ctl->unit;
1563 search_bytes = min(search_bytes, end - search_start + 1);
1564 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1565 if (ret < 0 || search_start != *offset)
1568 /* We may have found more bits than what we need */
1569 search_bytes = min(search_bytes, *bytes);
1571 /* Cannot clear past the end of the bitmap */
1572 search_bytes = min(search_bytes, end - search_start + 1);
1574 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1575 *offset += search_bytes;
1576 *bytes -= search_bytes;
1579 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1580 if (!bitmap_info->bytes)
1581 free_bitmap(ctl, bitmap_info);
1584 * no entry after this bitmap, but we still have bytes to
1585 * remove, so something has gone wrong.
1590 bitmap_info = rb_entry(next, struct btrfs_free_space,
1594 * if the next entry isn't a bitmap we need to return to let the
1595 * extent stuff do its work.
1597 if (!bitmap_info->bitmap)
1601 * Ok the next item is a bitmap, but it may not actually hold
1602 * the information for the rest of this free space stuff, so
1603 * look for it, and if we don't find it return so we can try
1604 * everything over again.
1606 search_start = *offset;
1607 search_bytes = ctl->unit;
1608 ret = search_bitmap(ctl, bitmap_info, &search_start,
1610 if (ret < 0 || search_start != *offset)
1614 } else if (!bitmap_info->bytes)
1615 free_bitmap(ctl, bitmap_info);
1620 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1621 struct btrfs_free_space *info, u64 offset,
1624 u64 bytes_to_set = 0;
1627 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1629 bytes_to_set = min(end - offset, bytes);
1631 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1633 return bytes_to_set;
1637 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1638 struct btrfs_free_space *info)
1640 struct btrfs_block_group_cache *block_group = ctl->private;
1643 * If we are below the extents threshold then we can add this as an
1644 * extent, and don't have to deal with the bitmap
1646 if (ctl->free_extents < ctl->extents_thresh) {
1648 * If this block group has some small extents we don't want to
1649 * use up all of our free slots in the cache with them, we want
1650 * to reserve them to larger extents, however if we have plent
1651 * of cache left then go ahead an dadd them, no sense in adding
1652 * the overhead of a bitmap if we don't have to.
1654 if (info->bytes <= block_group->sectorsize * 4) {
1655 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1663 * The original block groups from mkfs can be really small, like 8
1664 * megabytes, so don't bother with a bitmap for those entries. However
1665 * some block groups can be smaller than what a bitmap would cover but
1666 * are still large enough that they could overflow the 32k memory limit,
1667 * so allow those block groups to still be allowed to have a bitmap
1670 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1676 static struct btrfs_free_space_op free_space_op = {
1677 .recalc_thresholds = recalculate_thresholds,
1678 .use_bitmap = use_bitmap,
1681 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1682 struct btrfs_free_space *info)
1684 struct btrfs_free_space *bitmap_info;
1685 struct btrfs_block_group_cache *block_group = NULL;
1687 u64 bytes, offset, bytes_added;
1690 bytes = info->bytes;
1691 offset = info->offset;
1693 if (!ctl->op->use_bitmap(ctl, info))
1696 if (ctl->op == &free_space_op)
1697 block_group = ctl->private;
1700 * Since we link bitmaps right into the cluster we need to see if we
1701 * have a cluster here, and if so and it has our bitmap we need to add
1702 * the free space to that bitmap.
1704 if (block_group && !list_empty(&block_group->cluster_list)) {
1705 struct btrfs_free_cluster *cluster;
1706 struct rb_node *node;
1707 struct btrfs_free_space *entry;
1709 cluster = list_entry(block_group->cluster_list.next,
1710 struct btrfs_free_cluster,
1712 spin_lock(&cluster->lock);
1713 node = rb_first(&cluster->root);
1715 spin_unlock(&cluster->lock);
1716 goto no_cluster_bitmap;
1719 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1720 if (!entry->bitmap) {
1721 spin_unlock(&cluster->lock);
1722 goto no_cluster_bitmap;
1725 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1726 bytes_added = add_bytes_to_bitmap(ctl, entry,
1728 bytes -= bytes_added;
1729 offset += bytes_added;
1731 spin_unlock(&cluster->lock);
1739 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1746 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1747 bytes -= bytes_added;
1748 offset += bytes_added;
1758 if (info && info->bitmap) {
1759 add_new_bitmap(ctl, info, offset);
1764 spin_unlock(&ctl->tree_lock);
1766 /* no pre-allocated info, allocate a new one */
1768 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1771 spin_lock(&ctl->tree_lock);
1777 /* allocate the bitmap */
1778 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1779 spin_lock(&ctl->tree_lock);
1780 if (!info->bitmap) {
1790 kfree(info->bitmap);
1791 kmem_cache_free(btrfs_free_space_cachep, info);
1797 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1798 struct btrfs_free_space *info, bool update_stat)
1800 struct btrfs_free_space *left_info;
1801 struct btrfs_free_space *right_info;
1802 bool merged = false;
1803 u64 offset = info->offset;
1804 u64 bytes = info->bytes;
1807 * first we want to see if there is free space adjacent to the range we
1808 * are adding, if there is remove that struct and add a new one to
1809 * cover the entire range
1811 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1812 if (right_info && rb_prev(&right_info->offset_index))
1813 left_info = rb_entry(rb_prev(&right_info->offset_index),
1814 struct btrfs_free_space, offset_index);
1816 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1818 if (right_info && !right_info->bitmap) {
1820 unlink_free_space(ctl, right_info);
1822 __unlink_free_space(ctl, right_info);
1823 info->bytes += right_info->bytes;
1824 kmem_cache_free(btrfs_free_space_cachep, right_info);
1828 if (left_info && !left_info->bitmap &&
1829 left_info->offset + left_info->bytes == offset) {
1831 unlink_free_space(ctl, left_info);
1833 __unlink_free_space(ctl, left_info);
1834 info->offset = left_info->offset;
1835 info->bytes += left_info->bytes;
1836 kmem_cache_free(btrfs_free_space_cachep, left_info);
1843 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1844 u64 offset, u64 bytes)
1846 struct btrfs_free_space *info;
1849 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1853 info->offset = offset;
1854 info->bytes = bytes;
1856 spin_lock(&ctl->tree_lock);
1858 if (try_merge_free_space(ctl, info, true))
1862 * There was no extent directly to the left or right of this new
1863 * extent then we know we're going to have to allocate a new extent, so
1864 * before we do that see if we need to drop this into a bitmap
1866 ret = insert_into_bitmap(ctl, info);
1874 ret = link_free_space(ctl, info);
1876 kmem_cache_free(btrfs_free_space_cachep, info);
1878 spin_unlock(&ctl->tree_lock);
1881 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1882 ASSERT(ret != -EEXIST);
1888 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1889 u64 offset, u64 bytes)
1891 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1892 struct btrfs_free_space *info;
1894 bool re_search = false;
1896 spin_lock(&ctl->tree_lock);
1903 info = tree_search_offset(ctl, offset, 0, 0);
1906 * oops didn't find an extent that matched the space we wanted
1907 * to remove, look for a bitmap instead
1909 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1913 * If we found a partial bit of our free space in a
1914 * bitmap but then couldn't find the other part this may
1915 * be a problem, so WARN about it.
1923 if (!info->bitmap) {
1924 unlink_free_space(ctl, info);
1925 if (offset == info->offset) {
1926 u64 to_free = min(bytes, info->bytes);
1928 info->bytes -= to_free;
1929 info->offset += to_free;
1931 ret = link_free_space(ctl, info);
1934 kmem_cache_free(btrfs_free_space_cachep, info);
1941 u64 old_end = info->bytes + info->offset;
1943 info->bytes = offset - info->offset;
1944 ret = link_free_space(ctl, info);
1949 /* Not enough bytes in this entry to satisfy us */
1950 if (old_end < offset + bytes) {
1951 bytes -= old_end - offset;
1954 } else if (old_end == offset + bytes) {
1958 spin_unlock(&ctl->tree_lock);
1960 ret = btrfs_add_free_space(block_group, offset + bytes,
1961 old_end - (offset + bytes));
1967 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1968 if (ret == -EAGAIN) {
1973 spin_unlock(&ctl->tree_lock);
1978 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1981 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1982 struct btrfs_free_space *info;
1986 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1987 info = rb_entry(n, struct btrfs_free_space, offset_index);
1988 if (info->bytes >= bytes && !block_group->ro)
1990 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1991 info->offset, info->bytes,
1992 (info->bitmap) ? "yes" : "no");
1994 printk(KERN_INFO "block group has cluster?: %s\n",
1995 list_empty(&block_group->cluster_list) ? "no" : "yes");
1996 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2000 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2002 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2004 spin_lock_init(&ctl->tree_lock);
2005 ctl->unit = block_group->sectorsize;
2006 ctl->start = block_group->key.objectid;
2007 ctl->private = block_group;
2008 ctl->op = &free_space_op;
2011 * we only want to have 32k of ram per block group for keeping
2012 * track of free space, and if we pass 1/2 of that we want to
2013 * start converting things over to using bitmaps
2015 ctl->extents_thresh = ((1024 * 32) / 2) /
2016 sizeof(struct btrfs_free_space);
2020 * for a given cluster, put all of its extents back into the free
2021 * space cache. If the block group passed doesn't match the block group
2022 * pointed to by the cluster, someone else raced in and freed the
2023 * cluster already. In that case, we just return without changing anything
2026 __btrfs_return_cluster_to_free_space(
2027 struct btrfs_block_group_cache *block_group,
2028 struct btrfs_free_cluster *cluster)
2030 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2031 struct btrfs_free_space *entry;
2032 struct rb_node *node;
2034 spin_lock(&cluster->lock);
2035 if (cluster->block_group != block_group)
2038 cluster->block_group = NULL;
2039 cluster->window_start = 0;
2040 list_del_init(&cluster->block_group_list);
2042 node = rb_first(&cluster->root);
2046 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2047 node = rb_next(&entry->offset_index);
2048 rb_erase(&entry->offset_index, &cluster->root);
2050 bitmap = (entry->bitmap != NULL);
2052 try_merge_free_space(ctl, entry, false);
2053 tree_insert_offset(&ctl->free_space_offset,
2054 entry->offset, &entry->offset_index, bitmap);
2056 cluster->root = RB_ROOT;
2059 spin_unlock(&cluster->lock);
2060 btrfs_put_block_group(block_group);
2064 static void __btrfs_remove_free_space_cache_locked(
2065 struct btrfs_free_space_ctl *ctl)
2067 struct btrfs_free_space *info;
2068 struct rb_node *node;
2070 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2071 info = rb_entry(node, struct btrfs_free_space, offset_index);
2072 if (!info->bitmap) {
2073 unlink_free_space(ctl, info);
2074 kmem_cache_free(btrfs_free_space_cachep, info);
2076 free_bitmap(ctl, info);
2078 if (need_resched()) {
2079 spin_unlock(&ctl->tree_lock);
2081 spin_lock(&ctl->tree_lock);
2086 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2088 spin_lock(&ctl->tree_lock);
2089 __btrfs_remove_free_space_cache_locked(ctl);
2090 spin_unlock(&ctl->tree_lock);
2093 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2095 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2096 struct btrfs_free_cluster *cluster;
2097 struct list_head *head;
2099 spin_lock(&ctl->tree_lock);
2100 while ((head = block_group->cluster_list.next) !=
2101 &block_group->cluster_list) {
2102 cluster = list_entry(head, struct btrfs_free_cluster,
2105 WARN_ON(cluster->block_group != block_group);
2106 __btrfs_return_cluster_to_free_space(block_group, cluster);
2107 if (need_resched()) {
2108 spin_unlock(&ctl->tree_lock);
2110 spin_lock(&ctl->tree_lock);
2113 __btrfs_remove_free_space_cache_locked(ctl);
2114 spin_unlock(&ctl->tree_lock);
2118 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2119 u64 offset, u64 bytes, u64 empty_size)
2121 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2122 struct btrfs_free_space *entry = NULL;
2123 u64 bytes_search = bytes + empty_size;
2126 u64 align_gap_len = 0;
2128 spin_lock(&ctl->tree_lock);
2129 entry = find_free_space(ctl, &offset, &bytes_search,
2130 block_group->full_stripe_len);
2135 if (entry->bitmap) {
2136 bitmap_clear_bits(ctl, entry, offset, bytes);
2138 free_bitmap(ctl, entry);
2141 unlink_free_space(ctl, entry);
2142 align_gap_len = offset - entry->offset;
2143 align_gap = entry->offset;
2145 entry->offset = offset + bytes;
2146 WARN_ON(entry->bytes < bytes + align_gap_len);
2148 entry->bytes -= bytes + align_gap_len;
2150 kmem_cache_free(btrfs_free_space_cachep, entry);
2152 link_free_space(ctl, entry);
2156 spin_unlock(&ctl->tree_lock);
2159 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2164 * given a cluster, put all of its extents back into the free space
2165 * cache. If a block group is passed, this function will only free
2166 * a cluster that belongs to the passed block group.
2168 * Otherwise, it'll get a reference on the block group pointed to by the
2169 * cluster and remove the cluster from it.
2171 int btrfs_return_cluster_to_free_space(
2172 struct btrfs_block_group_cache *block_group,
2173 struct btrfs_free_cluster *cluster)
2175 struct btrfs_free_space_ctl *ctl;
2178 /* first, get a safe pointer to the block group */
2179 spin_lock(&cluster->lock);
2181 block_group = cluster->block_group;
2183 spin_unlock(&cluster->lock);
2186 } else if (cluster->block_group != block_group) {
2187 /* someone else has already freed it don't redo their work */
2188 spin_unlock(&cluster->lock);
2191 atomic_inc(&block_group->count);
2192 spin_unlock(&cluster->lock);
2194 ctl = block_group->free_space_ctl;
2196 /* now return any extents the cluster had on it */
2197 spin_lock(&ctl->tree_lock);
2198 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2199 spin_unlock(&ctl->tree_lock);
2201 /* finally drop our ref */
2202 btrfs_put_block_group(block_group);
2206 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2207 struct btrfs_free_cluster *cluster,
2208 struct btrfs_free_space *entry,
2209 u64 bytes, u64 min_start)
2211 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2213 u64 search_start = cluster->window_start;
2214 u64 search_bytes = bytes;
2217 search_start = min_start;
2218 search_bytes = bytes;
2220 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2225 __bitmap_clear_bits(ctl, entry, ret, bytes);
2231 * given a cluster, try to allocate 'bytes' from it, returns 0
2232 * if it couldn't find anything suitably large, or a logical disk offset
2233 * if things worked out
2235 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2236 struct btrfs_free_cluster *cluster, u64 bytes,
2239 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2240 struct btrfs_free_space *entry = NULL;
2241 struct rb_node *node;
2244 spin_lock(&cluster->lock);
2245 if (bytes > cluster->max_size)
2248 if (cluster->block_group != block_group)
2251 node = rb_first(&cluster->root);
2255 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2257 if (entry->bytes < bytes ||
2258 (!entry->bitmap && entry->offset < min_start)) {
2259 node = rb_next(&entry->offset_index);
2262 entry = rb_entry(node, struct btrfs_free_space,
2267 if (entry->bitmap) {
2268 ret = btrfs_alloc_from_bitmap(block_group,
2269 cluster, entry, bytes,
2270 cluster->window_start);
2272 node = rb_next(&entry->offset_index);
2275 entry = rb_entry(node, struct btrfs_free_space,
2279 cluster->window_start += bytes;
2281 ret = entry->offset;
2283 entry->offset += bytes;
2284 entry->bytes -= bytes;
2287 if (entry->bytes == 0)
2288 rb_erase(&entry->offset_index, &cluster->root);
2292 spin_unlock(&cluster->lock);
2297 spin_lock(&ctl->tree_lock);
2299 ctl->free_space -= bytes;
2300 if (entry->bytes == 0) {
2301 ctl->free_extents--;
2302 if (entry->bitmap) {
2303 kfree(entry->bitmap);
2304 ctl->total_bitmaps--;
2305 ctl->op->recalc_thresholds(ctl);
2307 kmem_cache_free(btrfs_free_space_cachep, entry);
2310 spin_unlock(&ctl->tree_lock);
2315 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2316 struct btrfs_free_space *entry,
2317 struct btrfs_free_cluster *cluster,
2318 u64 offset, u64 bytes,
2319 u64 cont1_bytes, u64 min_bytes)
2321 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2322 unsigned long next_zero;
2324 unsigned long want_bits;
2325 unsigned long min_bits;
2326 unsigned long found_bits;
2327 unsigned long start = 0;
2328 unsigned long total_found = 0;
2331 i = offset_to_bit(entry->offset, ctl->unit,
2332 max_t(u64, offset, entry->offset));
2333 want_bits = bytes_to_bits(bytes, ctl->unit);
2334 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2338 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2339 next_zero = find_next_zero_bit(entry->bitmap,
2340 BITS_PER_BITMAP, i);
2341 if (next_zero - i >= min_bits) {
2342 found_bits = next_zero - i;
2353 cluster->max_size = 0;
2356 total_found += found_bits;
2358 if (cluster->max_size < found_bits * ctl->unit)
2359 cluster->max_size = found_bits * ctl->unit;
2361 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2366 cluster->window_start = start * ctl->unit + entry->offset;
2367 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2368 ret = tree_insert_offset(&cluster->root, entry->offset,
2369 &entry->offset_index, 1);
2370 ASSERT(!ret); /* -EEXIST; Logic error */
2372 trace_btrfs_setup_cluster(block_group, cluster,
2373 total_found * ctl->unit, 1);
2378 * This searches the block group for just extents to fill the cluster with.
2379 * Try to find a cluster with at least bytes total bytes, at least one
2380 * extent of cont1_bytes, and other clusters of at least min_bytes.
2383 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2384 struct btrfs_free_cluster *cluster,
2385 struct list_head *bitmaps, u64 offset, u64 bytes,
2386 u64 cont1_bytes, u64 min_bytes)
2388 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2389 struct btrfs_free_space *first = NULL;
2390 struct btrfs_free_space *entry = NULL;
2391 struct btrfs_free_space *last;
2392 struct rb_node *node;
2398 entry = tree_search_offset(ctl, offset, 0, 1);
2403 * We don't want bitmaps, so just move along until we find a normal
2406 while (entry->bitmap || entry->bytes < min_bytes) {
2407 if (entry->bitmap && list_empty(&entry->list))
2408 list_add_tail(&entry->list, bitmaps);
2409 node = rb_next(&entry->offset_index);
2412 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2415 window_start = entry->offset;
2416 window_free = entry->bytes;
2417 max_extent = entry->bytes;
2421 for (node = rb_next(&entry->offset_index); node;
2422 node = rb_next(&entry->offset_index)) {
2423 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2425 if (entry->bitmap) {
2426 if (list_empty(&entry->list))
2427 list_add_tail(&entry->list, bitmaps);
2431 if (entry->bytes < min_bytes)
2435 window_free += entry->bytes;
2436 if (entry->bytes > max_extent)
2437 max_extent = entry->bytes;
2440 if (window_free < bytes || max_extent < cont1_bytes)
2443 cluster->window_start = first->offset;
2445 node = &first->offset_index;
2448 * now we've found our entries, pull them out of the free space
2449 * cache and put them into the cluster rbtree
2454 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2455 node = rb_next(&entry->offset_index);
2456 if (entry->bitmap || entry->bytes < min_bytes)
2459 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2460 ret = tree_insert_offset(&cluster->root, entry->offset,
2461 &entry->offset_index, 0);
2462 total_size += entry->bytes;
2463 ASSERT(!ret); /* -EEXIST; Logic error */
2464 } while (node && entry != last);
2466 cluster->max_size = max_extent;
2467 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2472 * This specifically looks for bitmaps that may work in the cluster, we assume
2473 * that we have already failed to find extents that will work.
2476 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2477 struct btrfs_free_cluster *cluster,
2478 struct list_head *bitmaps, u64 offset, u64 bytes,
2479 u64 cont1_bytes, u64 min_bytes)
2481 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2482 struct btrfs_free_space *entry;
2484 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2486 if (ctl->total_bitmaps == 0)
2490 * The bitmap that covers offset won't be in the list unless offset
2491 * is just its start offset.
2493 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2494 if (entry->offset != bitmap_offset) {
2495 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2496 if (entry && list_empty(&entry->list))
2497 list_add(&entry->list, bitmaps);
2500 list_for_each_entry(entry, bitmaps, list) {
2501 if (entry->bytes < bytes)
2503 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2504 bytes, cont1_bytes, min_bytes);
2510 * The bitmaps list has all the bitmaps that record free space
2511 * starting after offset, so no more search is required.
2517 * here we try to find a cluster of blocks in a block group. The goal
2518 * is to find at least bytes+empty_size.
2519 * We might not find them all in one contiguous area.
2521 * returns zero and sets up cluster if things worked out, otherwise
2522 * it returns -enospc
2524 int btrfs_find_space_cluster(struct btrfs_root *root,
2525 struct btrfs_block_group_cache *block_group,
2526 struct btrfs_free_cluster *cluster,
2527 u64 offset, u64 bytes, u64 empty_size)
2529 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2530 struct btrfs_free_space *entry, *tmp;
2537 * Choose the minimum extent size we'll require for this
2538 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2539 * For metadata, allow allocates with smaller extents. For
2540 * data, keep it dense.
2542 if (btrfs_test_opt(root, SSD_SPREAD)) {
2543 cont1_bytes = min_bytes = bytes + empty_size;
2544 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2545 cont1_bytes = bytes;
2546 min_bytes = block_group->sectorsize;
2548 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2549 min_bytes = block_group->sectorsize;
2552 spin_lock(&ctl->tree_lock);
2555 * If we know we don't have enough space to make a cluster don't even
2556 * bother doing all the work to try and find one.
2558 if (ctl->free_space < bytes) {
2559 spin_unlock(&ctl->tree_lock);
2563 spin_lock(&cluster->lock);
2565 /* someone already found a cluster, hooray */
2566 if (cluster->block_group) {
2571 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2574 INIT_LIST_HEAD(&bitmaps);
2575 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2577 cont1_bytes, min_bytes);
2579 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2580 offset, bytes + empty_size,
2581 cont1_bytes, min_bytes);
2583 /* Clear our temporary list */
2584 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2585 list_del_init(&entry->list);
2588 atomic_inc(&block_group->count);
2589 list_add_tail(&cluster->block_group_list,
2590 &block_group->cluster_list);
2591 cluster->block_group = block_group;
2593 trace_btrfs_failed_cluster_setup(block_group);
2596 spin_unlock(&cluster->lock);
2597 spin_unlock(&ctl->tree_lock);
2603 * simple code to zero out a cluster
2605 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2607 spin_lock_init(&cluster->lock);
2608 spin_lock_init(&cluster->refill_lock);
2609 cluster->root = RB_ROOT;
2610 cluster->max_size = 0;
2611 INIT_LIST_HEAD(&cluster->block_group_list);
2612 cluster->block_group = NULL;
2615 static int do_trimming(struct btrfs_block_group_cache *block_group,
2616 u64 *total_trimmed, u64 start, u64 bytes,
2617 u64 reserved_start, u64 reserved_bytes)
2619 struct btrfs_space_info *space_info = block_group->space_info;
2620 struct btrfs_fs_info *fs_info = block_group->fs_info;
2625 spin_lock(&space_info->lock);
2626 spin_lock(&block_group->lock);
2627 if (!block_group->ro) {
2628 block_group->reserved += reserved_bytes;
2629 space_info->bytes_reserved += reserved_bytes;
2632 spin_unlock(&block_group->lock);
2633 spin_unlock(&space_info->lock);
2635 ret = btrfs_error_discard_extent(fs_info->extent_root,
2636 start, bytes, &trimmed);
2638 *total_trimmed += trimmed;
2640 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2643 spin_lock(&space_info->lock);
2644 spin_lock(&block_group->lock);
2645 if (block_group->ro)
2646 space_info->bytes_readonly += reserved_bytes;
2647 block_group->reserved -= reserved_bytes;
2648 space_info->bytes_reserved -= reserved_bytes;
2649 spin_unlock(&space_info->lock);
2650 spin_unlock(&block_group->lock);
2656 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2657 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2659 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2660 struct btrfs_free_space *entry;
2661 struct rb_node *node;
2667 while (start < end) {
2668 spin_lock(&ctl->tree_lock);
2670 if (ctl->free_space < minlen) {
2671 spin_unlock(&ctl->tree_lock);
2675 entry = tree_search_offset(ctl, start, 0, 1);
2677 spin_unlock(&ctl->tree_lock);
2682 while (entry->bitmap) {
2683 node = rb_next(&entry->offset_index);
2685 spin_unlock(&ctl->tree_lock);
2688 entry = rb_entry(node, struct btrfs_free_space,
2692 if (entry->offset >= end) {
2693 spin_unlock(&ctl->tree_lock);
2697 extent_start = entry->offset;
2698 extent_bytes = entry->bytes;
2699 start = max(start, extent_start);
2700 bytes = min(extent_start + extent_bytes, end) - start;
2701 if (bytes < minlen) {
2702 spin_unlock(&ctl->tree_lock);
2706 unlink_free_space(ctl, entry);
2707 kmem_cache_free(btrfs_free_space_cachep, entry);
2709 spin_unlock(&ctl->tree_lock);
2711 ret = do_trimming(block_group, total_trimmed, start, bytes,
2712 extent_start, extent_bytes);
2718 if (fatal_signal_pending(current)) {
2729 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2730 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2732 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2733 struct btrfs_free_space *entry;
2737 u64 offset = offset_to_bitmap(ctl, start);
2739 while (offset < end) {
2740 bool next_bitmap = false;
2742 spin_lock(&ctl->tree_lock);
2744 if (ctl->free_space < minlen) {
2745 spin_unlock(&ctl->tree_lock);
2749 entry = tree_search_offset(ctl, offset, 1, 0);
2751 spin_unlock(&ctl->tree_lock);
2757 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2758 if (ret2 || start >= end) {
2759 spin_unlock(&ctl->tree_lock);
2764 bytes = min(bytes, end - start);
2765 if (bytes < minlen) {
2766 spin_unlock(&ctl->tree_lock);
2770 bitmap_clear_bits(ctl, entry, start, bytes);
2771 if (entry->bytes == 0)
2772 free_bitmap(ctl, entry);
2774 spin_unlock(&ctl->tree_lock);
2776 ret = do_trimming(block_group, total_trimmed, start, bytes,
2782 offset += BITS_PER_BITMAP * ctl->unit;
2785 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2786 offset += BITS_PER_BITMAP * ctl->unit;
2789 if (fatal_signal_pending(current)) {
2800 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2801 u64 *trimmed, u64 start, u64 end, u64 minlen)
2807 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2811 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2817 * Find the left-most item in the cache tree, and then return the
2818 * smallest inode number in the item.
2820 * Note: the returned inode number may not be the smallest one in
2821 * the tree, if the left-most item is a bitmap.
2823 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2825 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2826 struct btrfs_free_space *entry = NULL;
2829 spin_lock(&ctl->tree_lock);
2831 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2834 entry = rb_entry(rb_first(&ctl->free_space_offset),
2835 struct btrfs_free_space, offset_index);
2837 if (!entry->bitmap) {
2838 ino = entry->offset;
2840 unlink_free_space(ctl, entry);
2844 kmem_cache_free(btrfs_free_space_cachep, entry);
2846 link_free_space(ctl, entry);
2852 ret = search_bitmap(ctl, entry, &offset, &count);
2853 /* Logic error; Should be empty if it can't find anything */
2857 bitmap_clear_bits(ctl, entry, offset, 1);
2858 if (entry->bytes == 0)
2859 free_bitmap(ctl, entry);
2862 spin_unlock(&ctl->tree_lock);
2867 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2868 struct btrfs_path *path)
2870 struct inode *inode = NULL;
2872 spin_lock(&root->cache_lock);
2873 if (root->cache_inode)
2874 inode = igrab(root->cache_inode);
2875 spin_unlock(&root->cache_lock);
2879 inode = __lookup_free_space_inode(root, path, 0);
2883 spin_lock(&root->cache_lock);
2884 if (!btrfs_fs_closing(root->fs_info))
2885 root->cache_inode = igrab(inode);
2886 spin_unlock(&root->cache_lock);
2891 int create_free_ino_inode(struct btrfs_root *root,
2892 struct btrfs_trans_handle *trans,
2893 struct btrfs_path *path)
2895 return __create_free_space_inode(root, trans, path,
2896 BTRFS_FREE_INO_OBJECTID, 0);
2899 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2901 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2902 struct btrfs_path *path;
2903 struct inode *inode;
2905 u64 root_gen = btrfs_root_generation(&root->root_item);
2907 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2911 * If we're unmounting then just return, since this does a search on the
2912 * normal root and not the commit root and we could deadlock.
2914 if (btrfs_fs_closing(fs_info))
2917 path = btrfs_alloc_path();
2921 inode = lookup_free_ino_inode(root, path);
2925 if (root_gen != BTRFS_I(inode)->generation)
2928 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2932 "failed to load free ino cache for root %llu",
2933 root->root_key.objectid);
2937 btrfs_free_path(path);
2941 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2942 struct btrfs_trans_handle *trans,
2943 struct btrfs_path *path)
2945 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2946 struct inode *inode;
2949 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2952 inode = lookup_free_ino_inode(root, path);
2956 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2958 btrfs_delalloc_release_metadata(inode, inode->i_size);
2960 btrfs_err(root->fs_info,
2961 "failed to write free ino cache for root %llu",
2962 root->root_key.objectid);
2970 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2972 * Use this if you need to make a bitmap or extent entry specifically, it
2973 * doesn't do any of the merging that add_free_space does, this acts a lot like
2974 * how the free space cache loading stuff works, so you can get really weird
2977 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
2978 u64 offset, u64 bytes, bool bitmap)
2980 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
2981 struct btrfs_free_space *info = NULL, *bitmap_info;
2988 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2994 spin_lock(&ctl->tree_lock);
2995 info->offset = offset;
2996 info->bytes = bytes;
2997 ret = link_free_space(ctl, info);
2998 spin_unlock(&ctl->tree_lock);
3000 kmem_cache_free(btrfs_free_space_cachep, info);
3005 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3007 kmem_cache_free(btrfs_free_space_cachep, info);
3012 spin_lock(&ctl->tree_lock);
3013 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3018 add_new_bitmap(ctl, info, offset);
3022 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3023 bytes -= bytes_added;
3024 offset += bytes_added;
3025 spin_unlock(&ctl->tree_lock);
3036 * Checks to see if the given range is in the free space cache. This is really
3037 * just used to check the absence of space, so if there is free space in the
3038 * range at all we will return 1.
3040 int test_check_exists(struct btrfs_block_group_cache *cache,
3041 u64 offset, u64 bytes)
3043 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3044 struct btrfs_free_space *info;
3047 spin_lock(&ctl->tree_lock);
3048 info = tree_search_offset(ctl, offset, 0, 0);
3050 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3058 u64 bit_off, bit_bytes;
3060 struct btrfs_free_space *tmp;
3063 bit_bytes = ctl->unit;
3064 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3066 if (bit_off == offset) {
3069 } else if (bit_off > offset &&
3070 offset + bytes > bit_off) {
3076 n = rb_prev(&info->offset_index);
3078 tmp = rb_entry(n, struct btrfs_free_space,
3080 if (tmp->offset + tmp->bytes < offset)
3082 if (offset + bytes < tmp->offset) {
3083 n = rb_prev(&info->offset_index);
3090 n = rb_next(&info->offset_index);
3092 tmp = rb_entry(n, struct btrfs_free_space,
3094 if (offset + bytes < tmp->offset)
3096 if (tmp->offset + tmp->bytes < offset) {
3097 n = rb_next(&info->offset_index);
3107 if (info->offset == offset) {
3112 if (offset > info->offset && offset < info->offset + info->bytes)
3115 spin_unlock(&ctl->tree_lock);
3118 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */