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,
227 oldsize = i_size_read(inode);
228 btrfs_i_size_write(inode, 0);
229 truncate_pagecache(inode, oldsize, 0);
232 * We don't need an orphan item because truncating the free space cache
233 * will never be split across transactions.
235 ret = btrfs_truncate_inode_items(trans, root, inode,
236 0, BTRFS_EXTENT_DATA_KEY);
238 btrfs_abort_transaction(trans, root, ret);
242 ret = btrfs_update_inode(trans, root, inode);
244 btrfs_abort_transaction(trans, root, ret);
249 static int readahead_cache(struct inode *inode)
251 struct file_ra_state *ra;
252 unsigned long last_index;
254 ra = kzalloc(sizeof(*ra), GFP_NOFS);
258 file_ra_state_init(ra, inode->i_mapping);
259 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
261 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
272 struct btrfs_root *root;
276 unsigned check_crcs:1;
279 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
280 struct btrfs_root *root)
282 memset(io_ctl, 0, sizeof(struct io_ctl));
283 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
285 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
290 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
291 io_ctl->check_crcs = 1;
295 static void io_ctl_free(struct io_ctl *io_ctl)
297 kfree(io_ctl->pages);
300 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
303 kunmap(io_ctl->page);
309 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
311 BUG_ON(io_ctl->index >= io_ctl->num_pages);
312 io_ctl->page = io_ctl->pages[io_ctl->index++];
313 io_ctl->cur = kmap(io_ctl->page);
314 io_ctl->orig = io_ctl->cur;
315 io_ctl->size = PAGE_CACHE_SIZE;
317 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
320 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
324 io_ctl_unmap_page(io_ctl);
326 for (i = 0; i < io_ctl->num_pages; i++) {
327 if (io_ctl->pages[i]) {
328 ClearPageChecked(io_ctl->pages[i]);
329 unlock_page(io_ctl->pages[i]);
330 page_cache_release(io_ctl->pages[i]);
335 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
339 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
342 for (i = 0; i < io_ctl->num_pages; i++) {
343 page = find_or_create_page(inode->i_mapping, i, mask);
345 io_ctl_drop_pages(io_ctl);
348 io_ctl->pages[i] = page;
349 if (uptodate && !PageUptodate(page)) {
350 btrfs_readpage(NULL, page);
352 if (!PageUptodate(page)) {
353 printk(KERN_ERR "btrfs: error reading free "
355 io_ctl_drop_pages(io_ctl);
361 for (i = 0; i < io_ctl->num_pages; i++) {
362 clear_page_dirty_for_io(io_ctl->pages[i]);
363 set_page_extent_mapped(io_ctl->pages[i]);
369 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
373 io_ctl_map_page(io_ctl, 1);
376 * Skip the csum areas. If we don't check crcs then we just have a
377 * 64bit chunk at the front of the first page.
379 if (io_ctl->check_crcs) {
380 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
381 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
383 io_ctl->cur += sizeof(u64);
384 io_ctl->size -= sizeof(u64) * 2;
388 *val = cpu_to_le64(generation);
389 io_ctl->cur += sizeof(u64);
392 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
397 * Skip the crc area. If we don't check crcs then we just have a 64bit
398 * chunk at the front of the first page.
400 if (io_ctl->check_crcs) {
401 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
402 io_ctl->size -= sizeof(u64) +
403 (sizeof(u32) * io_ctl->num_pages);
405 io_ctl->cur += sizeof(u64);
406 io_ctl->size -= sizeof(u64) * 2;
410 if (le64_to_cpu(*gen) != generation) {
411 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
412 "(%Lu) does not match inode (%Lu)\n", *gen,
414 io_ctl_unmap_page(io_ctl);
417 io_ctl->cur += sizeof(u64);
421 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
427 if (!io_ctl->check_crcs) {
428 io_ctl_unmap_page(io_ctl);
433 offset = sizeof(u32) * io_ctl->num_pages;
435 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
436 PAGE_CACHE_SIZE - offset);
437 btrfs_csum_final(crc, (char *)&crc);
438 io_ctl_unmap_page(io_ctl);
439 tmp = kmap(io_ctl->pages[0]);
442 kunmap(io_ctl->pages[0]);
445 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
451 if (!io_ctl->check_crcs) {
452 io_ctl_map_page(io_ctl, 0);
457 offset = sizeof(u32) * io_ctl->num_pages;
459 tmp = kmap(io_ctl->pages[0]);
462 kunmap(io_ctl->pages[0]);
464 io_ctl_map_page(io_ctl, 0);
465 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
466 PAGE_CACHE_SIZE - offset);
467 btrfs_csum_final(crc, (char *)&crc);
469 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
471 io_ctl_unmap_page(io_ctl);
478 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
481 struct btrfs_free_space_entry *entry;
487 entry->offset = cpu_to_le64(offset);
488 entry->bytes = cpu_to_le64(bytes);
489 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
490 BTRFS_FREE_SPACE_EXTENT;
491 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
492 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
494 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
497 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
499 /* No more pages to map */
500 if (io_ctl->index >= io_ctl->num_pages)
503 /* map the next page */
504 io_ctl_map_page(io_ctl, 1);
508 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
514 * If we aren't at the start of the current page, unmap this one and
515 * map the next one if there is any left.
517 if (io_ctl->cur != io_ctl->orig) {
518 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519 if (io_ctl->index >= io_ctl->num_pages)
521 io_ctl_map_page(io_ctl, 0);
524 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
525 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
526 if (io_ctl->index < io_ctl->num_pages)
527 io_ctl_map_page(io_ctl, 0);
531 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
534 * If we're not on the boundary we know we've modified the page and we
535 * need to crc the page.
537 if (io_ctl->cur != io_ctl->orig)
538 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
540 io_ctl_unmap_page(io_ctl);
542 while (io_ctl->index < io_ctl->num_pages) {
543 io_ctl_map_page(io_ctl, 1);
544 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
548 static int io_ctl_read_entry(struct io_ctl *io_ctl,
549 struct btrfs_free_space *entry, u8 *type)
551 struct btrfs_free_space_entry *e;
555 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
561 entry->offset = le64_to_cpu(e->offset);
562 entry->bytes = le64_to_cpu(e->bytes);
564 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
565 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
567 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
570 io_ctl_unmap_page(io_ctl);
575 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
576 struct btrfs_free_space *entry)
580 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
584 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
585 io_ctl_unmap_page(io_ctl);
591 * Since we attach pinned extents after the fact we can have contiguous sections
592 * of free space that are split up in entries. This poses a problem with the
593 * tree logging stuff since it could have allocated across what appears to be 2
594 * entries since we would have merged the entries when adding the pinned extents
595 * back to the free space cache. So run through the space cache that we just
596 * loaded and merge contiguous entries. This will make the log replay stuff not
597 * blow up and it will make for nicer allocator behavior.
599 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
601 struct btrfs_free_space *e, *prev = NULL;
605 spin_lock(&ctl->tree_lock);
606 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
607 e = rb_entry(n, struct btrfs_free_space, offset_index);
610 if (e->bitmap || prev->bitmap)
612 if (prev->offset + prev->bytes == e->offset) {
613 unlink_free_space(ctl, prev);
614 unlink_free_space(ctl, e);
615 prev->bytes += e->bytes;
616 kmem_cache_free(btrfs_free_space_cachep, e);
617 link_free_space(ctl, prev);
619 spin_unlock(&ctl->tree_lock);
625 spin_unlock(&ctl->tree_lock);
628 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
629 struct btrfs_free_space_ctl *ctl,
630 struct btrfs_path *path, u64 offset)
632 struct btrfs_free_space_header *header;
633 struct extent_buffer *leaf;
634 struct io_ctl io_ctl;
635 struct btrfs_key key;
636 struct btrfs_free_space *e, *n;
637 struct list_head bitmaps;
644 INIT_LIST_HEAD(&bitmaps);
646 /* Nothing in the space cache, goodbye */
647 if (!i_size_read(inode))
650 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
654 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
658 btrfs_release_path(path);
664 leaf = path->nodes[0];
665 header = btrfs_item_ptr(leaf, path->slots[0],
666 struct btrfs_free_space_header);
667 num_entries = btrfs_free_space_entries(leaf, header);
668 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
669 generation = btrfs_free_space_generation(leaf, header);
670 btrfs_release_path(path);
672 if (BTRFS_I(inode)->generation != generation) {
673 btrfs_err(root->fs_info,
674 "free space inode generation (%llu) "
675 "did not match free space cache generation (%llu)",
676 (unsigned long long)BTRFS_I(inode)->generation,
677 (unsigned long long)generation);
684 ret = io_ctl_init(&io_ctl, inode, root);
688 ret = readahead_cache(inode);
692 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
696 ret = io_ctl_check_crc(&io_ctl, 0);
700 ret = io_ctl_check_generation(&io_ctl, generation);
704 while (num_entries) {
705 e = kmem_cache_zalloc(btrfs_free_space_cachep,
710 ret = io_ctl_read_entry(&io_ctl, e, &type);
712 kmem_cache_free(btrfs_free_space_cachep, e);
717 kmem_cache_free(btrfs_free_space_cachep, e);
721 if (type == BTRFS_FREE_SPACE_EXTENT) {
722 spin_lock(&ctl->tree_lock);
723 ret = link_free_space(ctl, e);
724 spin_unlock(&ctl->tree_lock);
726 btrfs_err(root->fs_info,
727 "Duplicate entries in free space cache, dumping");
728 kmem_cache_free(btrfs_free_space_cachep, e);
732 BUG_ON(!num_bitmaps);
734 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
737 btrfs_free_space_cachep, e);
740 spin_lock(&ctl->tree_lock);
741 ret = link_free_space(ctl, e);
742 ctl->total_bitmaps++;
743 ctl->op->recalc_thresholds(ctl);
744 spin_unlock(&ctl->tree_lock);
746 btrfs_err(root->fs_info,
747 "Duplicate entries in free space cache, dumping");
748 kmem_cache_free(btrfs_free_space_cachep, e);
751 list_add_tail(&e->list, &bitmaps);
757 io_ctl_unmap_page(&io_ctl);
760 * We add the bitmaps at the end of the entries in order that
761 * the bitmap entries are added to the cache.
763 list_for_each_entry_safe(e, n, &bitmaps, list) {
764 list_del_init(&e->list);
765 ret = io_ctl_read_bitmap(&io_ctl, e);
770 io_ctl_drop_pages(&io_ctl);
771 merge_space_tree(ctl);
774 io_ctl_free(&io_ctl);
777 io_ctl_drop_pages(&io_ctl);
778 __btrfs_remove_free_space_cache(ctl);
782 int load_free_space_cache(struct btrfs_fs_info *fs_info,
783 struct btrfs_block_group_cache *block_group)
785 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
786 struct btrfs_root *root = fs_info->tree_root;
788 struct btrfs_path *path;
791 u64 used = btrfs_block_group_used(&block_group->item);
794 * If this block group has been marked to be cleared for one reason or
795 * another then we can't trust the on disk cache, so just return.
797 spin_lock(&block_group->lock);
798 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
799 spin_unlock(&block_group->lock);
802 spin_unlock(&block_group->lock);
804 path = btrfs_alloc_path();
807 path->search_commit_root = 1;
808 path->skip_locking = 1;
810 inode = lookup_free_space_inode(root, block_group, path);
812 btrfs_free_path(path);
816 /* We may have converted the inode and made the cache invalid. */
817 spin_lock(&block_group->lock);
818 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
819 spin_unlock(&block_group->lock);
820 btrfs_free_path(path);
823 spin_unlock(&block_group->lock);
825 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
826 path, block_group->key.objectid);
827 btrfs_free_path(path);
831 spin_lock(&ctl->tree_lock);
832 matched = (ctl->free_space == (block_group->key.offset - used -
833 block_group->bytes_super));
834 spin_unlock(&ctl->tree_lock);
837 __btrfs_remove_free_space_cache(ctl);
838 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
839 block_group->key.objectid);
844 /* This cache is bogus, make sure it gets cleared */
845 spin_lock(&block_group->lock);
846 block_group->disk_cache_state = BTRFS_DC_CLEAR;
847 spin_unlock(&block_group->lock);
850 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
851 block_group->key.objectid);
859 * __btrfs_write_out_cache - write out cached info to an inode
860 * @root - the root the inode belongs to
861 * @ctl - the free space cache we are going to write out
862 * @block_group - the block_group for this cache if it belongs to a block_group
863 * @trans - the trans handle
864 * @path - the path to use
865 * @offset - the offset for the key we'll insert
867 * This function writes out a free space cache struct to disk for quick recovery
868 * on mount. This will return 0 if it was successfull in writing the cache out,
869 * and -1 if it was not.
871 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
872 struct btrfs_free_space_ctl *ctl,
873 struct btrfs_block_group_cache *block_group,
874 struct btrfs_trans_handle *trans,
875 struct btrfs_path *path, u64 offset)
877 struct btrfs_free_space_header *header;
878 struct extent_buffer *leaf;
879 struct rb_node *node;
880 struct list_head *pos, *n;
881 struct extent_state *cached_state = NULL;
882 struct btrfs_free_cluster *cluster = NULL;
883 struct extent_io_tree *unpin = NULL;
884 struct io_ctl io_ctl;
885 struct list_head bitmap_list;
886 struct btrfs_key key;
887 u64 start, extent_start, extent_end, len;
893 INIT_LIST_HEAD(&bitmap_list);
895 if (!i_size_read(inode))
898 ret = io_ctl_init(&io_ctl, inode, root);
902 /* Get the cluster for this block_group if it exists */
903 if (block_group && !list_empty(&block_group->cluster_list))
904 cluster = list_entry(block_group->cluster_list.next,
905 struct btrfs_free_cluster,
908 /* Lock all pages first so we can lock the extent safely. */
909 io_ctl_prepare_pages(&io_ctl, inode, 0);
911 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
914 node = rb_first(&ctl->free_space_offset);
915 if (!node && cluster) {
916 node = rb_first(&cluster->root);
920 /* Make sure we can fit our crcs into the first page */
921 if (io_ctl.check_crcs &&
922 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
925 io_ctl_set_generation(&io_ctl, trans->transid);
927 /* Write out the extent entries */
929 struct btrfs_free_space *e;
931 e = rb_entry(node, struct btrfs_free_space, offset_index);
934 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
940 list_add_tail(&e->list, &bitmap_list);
943 node = rb_next(node);
944 if (!node && cluster) {
945 node = rb_first(&cluster->root);
951 * We want to add any pinned extents to our free space cache
952 * so we don't leak the space
956 * We shouldn't have switched the pinned extents yet so this is the
959 unpin = root->fs_info->pinned_extents;
962 start = block_group->key.objectid;
964 while (block_group && (start < block_group->key.objectid +
965 block_group->key.offset)) {
966 ret = find_first_extent_bit(unpin, start,
967 &extent_start, &extent_end,
974 /* This pinned extent is out of our range */
975 if (extent_start >= block_group->key.objectid +
976 block_group->key.offset)
979 extent_start = max(extent_start, start);
980 extent_end = min(block_group->key.objectid +
981 block_group->key.offset, extent_end + 1);
982 len = extent_end - extent_start;
985 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
992 /* Write out the bitmaps */
993 list_for_each_safe(pos, n, &bitmap_list) {
994 struct btrfs_free_space *entry =
995 list_entry(pos, struct btrfs_free_space, list);
997 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
1000 list_del_init(&entry->list);
1003 /* Zero out the rest of the pages just to make sure */
1004 io_ctl_zero_remaining_pages(&io_ctl);
1006 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1007 0, i_size_read(inode), &cached_state);
1008 io_ctl_drop_pages(&io_ctl);
1009 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1010 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1016 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1018 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1019 key.offset = offset;
1022 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1024 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1025 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1029 leaf = path->nodes[0];
1031 struct btrfs_key found_key;
1032 BUG_ON(!path->slots[0]);
1034 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1035 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1036 found_key.offset != offset) {
1037 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1039 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1041 btrfs_release_path(path);
1046 BTRFS_I(inode)->generation = trans->transid;
1047 header = btrfs_item_ptr(leaf, path->slots[0],
1048 struct btrfs_free_space_header);
1049 btrfs_set_free_space_entries(leaf, header, entries);
1050 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1051 btrfs_set_free_space_generation(leaf, header, trans->transid);
1052 btrfs_mark_buffer_dirty(leaf);
1053 btrfs_release_path(path);
1057 io_ctl_free(&io_ctl);
1059 invalidate_inode_pages2(inode->i_mapping);
1060 BTRFS_I(inode)->generation = 0;
1062 btrfs_update_inode(trans, root, inode);
1066 list_for_each_safe(pos, n, &bitmap_list) {
1067 struct btrfs_free_space *entry =
1068 list_entry(pos, struct btrfs_free_space, list);
1069 list_del_init(&entry->list);
1071 io_ctl_drop_pages(&io_ctl);
1072 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1073 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1077 int btrfs_write_out_cache(struct btrfs_root *root,
1078 struct btrfs_trans_handle *trans,
1079 struct btrfs_block_group_cache *block_group,
1080 struct btrfs_path *path)
1082 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1083 struct inode *inode;
1086 root = root->fs_info->tree_root;
1088 spin_lock(&block_group->lock);
1089 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1090 spin_unlock(&block_group->lock);
1093 spin_unlock(&block_group->lock);
1095 inode = lookup_free_space_inode(root, block_group, path);
1099 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1100 path, block_group->key.objectid);
1102 spin_lock(&block_group->lock);
1103 block_group->disk_cache_state = BTRFS_DC_ERROR;
1104 spin_unlock(&block_group->lock);
1107 btrfs_err(root->fs_info,
1108 "failed to write free space cache for block group %llu",
1109 block_group->key.objectid);
1117 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1120 BUG_ON(offset < bitmap_start);
1121 offset -= bitmap_start;
1122 return (unsigned long)(div_u64(offset, unit));
1125 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1127 return (unsigned long)(div_u64(bytes, unit));
1130 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1134 u64 bytes_per_bitmap;
1136 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1137 bitmap_start = offset - ctl->start;
1138 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1139 bitmap_start *= bytes_per_bitmap;
1140 bitmap_start += ctl->start;
1142 return bitmap_start;
1145 static int tree_insert_offset(struct rb_root *root, u64 offset,
1146 struct rb_node *node, int bitmap)
1148 struct rb_node **p = &root->rb_node;
1149 struct rb_node *parent = NULL;
1150 struct btrfs_free_space *info;
1154 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1156 if (offset < info->offset) {
1158 } else if (offset > info->offset) {
1159 p = &(*p)->rb_right;
1162 * we could have a bitmap entry and an extent entry
1163 * share the same offset. If this is the case, we want
1164 * the extent entry to always be found first if we do a
1165 * linear search through the tree, since we want to have
1166 * the quickest allocation time, and allocating from an
1167 * extent is faster than allocating from a bitmap. So
1168 * if we're inserting a bitmap and we find an entry at
1169 * this offset, we want to go right, or after this entry
1170 * logically. If we are inserting an extent and we've
1171 * found a bitmap, we want to go left, or before
1179 p = &(*p)->rb_right;
1181 if (!info->bitmap) {
1190 rb_link_node(node, parent, p);
1191 rb_insert_color(node, root);
1197 * searches the tree for the given offset.
1199 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1200 * want a section that has at least bytes size and comes at or after the given
1203 static struct btrfs_free_space *
1204 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1205 u64 offset, int bitmap_only, int fuzzy)
1207 struct rb_node *n = ctl->free_space_offset.rb_node;
1208 struct btrfs_free_space *entry, *prev = NULL;
1210 /* find entry that is closest to the 'offset' */
1217 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1220 if (offset < entry->offset)
1222 else if (offset > entry->offset)
1235 * bitmap entry and extent entry may share same offset,
1236 * in that case, bitmap entry comes after extent entry.
1241 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1242 if (entry->offset != offset)
1245 WARN_ON(!entry->bitmap);
1248 if (entry->bitmap) {
1250 * if previous extent entry covers the offset,
1251 * we should return it instead of the bitmap entry
1253 n = rb_prev(&entry->offset_index);
1255 prev = rb_entry(n, struct btrfs_free_space,
1257 if (!prev->bitmap &&
1258 prev->offset + prev->bytes > offset)
1268 /* find last entry before the 'offset' */
1270 if (entry->offset > offset) {
1271 n = rb_prev(&entry->offset_index);
1273 entry = rb_entry(n, struct btrfs_free_space,
1275 BUG_ON(entry->offset > offset);
1284 if (entry->bitmap) {
1285 n = rb_prev(&entry->offset_index);
1287 prev = rb_entry(n, struct btrfs_free_space,
1289 if (!prev->bitmap &&
1290 prev->offset + prev->bytes > offset)
1293 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1295 } else if (entry->offset + entry->bytes > offset)
1302 if (entry->bitmap) {
1303 if (entry->offset + BITS_PER_BITMAP *
1307 if (entry->offset + entry->bytes > offset)
1311 n = rb_next(&entry->offset_index);
1314 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1320 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1321 struct btrfs_free_space *info)
1323 rb_erase(&info->offset_index, &ctl->free_space_offset);
1324 ctl->free_extents--;
1327 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1328 struct btrfs_free_space *info)
1330 __unlink_free_space(ctl, info);
1331 ctl->free_space -= info->bytes;
1334 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1335 struct btrfs_free_space *info)
1339 BUG_ON(!info->bitmap && !info->bytes);
1340 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1341 &info->offset_index, (info->bitmap != NULL));
1345 ctl->free_space += info->bytes;
1346 ctl->free_extents++;
1350 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1352 struct btrfs_block_group_cache *block_group = ctl->private;
1356 u64 size = block_group->key.offset;
1357 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1358 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1360 max_bitmaps = max(max_bitmaps, 1);
1362 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1365 * The goal is to keep the total amount of memory used per 1gb of space
1366 * at or below 32k, so we need to adjust how much memory we allow to be
1367 * used by extent based free space tracking
1369 if (size < 1024 * 1024 * 1024)
1370 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1372 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1373 div64_u64(size, 1024 * 1024 * 1024);
1376 * we want to account for 1 more bitmap than what we have so we can make
1377 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1378 * we add more bitmaps.
1380 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1382 if (bitmap_bytes >= max_bytes) {
1383 ctl->extents_thresh = 0;
1388 * we want the extent entry threshold to always be at most 1/2 the maxw
1389 * bytes we can have, or whatever is less than that.
1391 extent_bytes = max_bytes - bitmap_bytes;
1392 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1394 ctl->extents_thresh =
1395 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1398 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1399 struct btrfs_free_space *info,
1400 u64 offset, u64 bytes)
1402 unsigned long start, count;
1404 start = offset_to_bit(info->offset, ctl->unit, offset);
1405 count = bytes_to_bits(bytes, ctl->unit);
1406 BUG_ON(start + count > BITS_PER_BITMAP);
1408 bitmap_clear(info->bitmap, start, count);
1410 info->bytes -= bytes;
1413 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1414 struct btrfs_free_space *info, u64 offset,
1417 __bitmap_clear_bits(ctl, info, offset, bytes);
1418 ctl->free_space -= bytes;
1421 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1422 struct btrfs_free_space *info, u64 offset,
1425 unsigned long start, count;
1427 start = offset_to_bit(info->offset, ctl->unit, offset);
1428 count = bytes_to_bits(bytes, ctl->unit);
1429 BUG_ON(start + count > BITS_PER_BITMAP);
1431 bitmap_set(info->bitmap, start, count);
1433 info->bytes += bytes;
1434 ctl->free_space += bytes;
1437 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1438 struct btrfs_free_space *bitmap_info, u64 *offset,
1441 unsigned long found_bits = 0;
1442 unsigned long bits, i;
1443 unsigned long next_zero;
1445 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1446 max_t(u64, *offset, bitmap_info->offset));
1447 bits = bytes_to_bits(*bytes, ctl->unit);
1449 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1450 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1451 BITS_PER_BITMAP, i);
1452 if ((next_zero - i) >= bits) {
1453 found_bits = next_zero - i;
1460 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1461 *bytes = (u64)(found_bits) * ctl->unit;
1468 static struct btrfs_free_space *
1469 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1470 unsigned long align)
1472 struct btrfs_free_space *entry;
1473 struct rb_node *node;
1479 if (!ctl->free_space_offset.rb_node)
1482 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1486 for (node = &entry->offset_index; node; node = rb_next(node)) {
1487 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1488 if (entry->bytes < *bytes)
1491 /* make sure the space returned is big enough
1492 * to match our requested alignment
1494 if (*bytes >= align) {
1495 ctl_off = entry->offset - ctl->start;
1496 tmp = ctl_off + align - 1;;
1498 tmp = tmp * align + ctl->start;
1499 align_off = tmp - entry->offset;
1502 tmp = entry->offset;
1505 if (entry->bytes < *bytes + align_off)
1508 if (entry->bitmap) {
1509 ret = search_bitmap(ctl, entry, &tmp, bytes);
1518 *bytes = entry->bytes - align_off;
1525 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1526 struct btrfs_free_space *info, u64 offset)
1528 info->offset = offset_to_bitmap(ctl, offset);
1530 INIT_LIST_HEAD(&info->list);
1531 link_free_space(ctl, info);
1532 ctl->total_bitmaps++;
1534 ctl->op->recalc_thresholds(ctl);
1537 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1538 struct btrfs_free_space *bitmap_info)
1540 unlink_free_space(ctl, bitmap_info);
1541 kfree(bitmap_info->bitmap);
1542 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1543 ctl->total_bitmaps--;
1544 ctl->op->recalc_thresholds(ctl);
1547 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1548 struct btrfs_free_space *bitmap_info,
1549 u64 *offset, u64 *bytes)
1552 u64 search_start, search_bytes;
1556 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1559 * We need to search for bits in this bitmap. We could only cover some
1560 * of the extent in this bitmap thanks to how we add space, so we need
1561 * to search for as much as it as we can and clear that amount, and then
1562 * go searching for the next bit.
1564 search_start = *offset;
1565 search_bytes = ctl->unit;
1566 search_bytes = min(search_bytes, end - search_start + 1);
1567 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1568 if (ret < 0 || search_start != *offset)
1571 /* We may have found more bits than what we need */
1572 search_bytes = min(search_bytes, *bytes);
1574 /* Cannot clear past the end of the bitmap */
1575 search_bytes = min(search_bytes, end - search_start + 1);
1577 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1578 *offset += search_bytes;
1579 *bytes -= search_bytes;
1582 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1583 if (!bitmap_info->bytes)
1584 free_bitmap(ctl, bitmap_info);
1587 * no entry after this bitmap, but we still have bytes to
1588 * remove, so something has gone wrong.
1593 bitmap_info = rb_entry(next, struct btrfs_free_space,
1597 * if the next entry isn't a bitmap we need to return to let the
1598 * extent stuff do its work.
1600 if (!bitmap_info->bitmap)
1604 * Ok the next item is a bitmap, but it may not actually hold
1605 * the information for the rest of this free space stuff, so
1606 * look for it, and if we don't find it return so we can try
1607 * everything over again.
1609 search_start = *offset;
1610 search_bytes = ctl->unit;
1611 ret = search_bitmap(ctl, bitmap_info, &search_start,
1613 if (ret < 0 || search_start != *offset)
1617 } else if (!bitmap_info->bytes)
1618 free_bitmap(ctl, bitmap_info);
1623 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1624 struct btrfs_free_space *info, u64 offset,
1627 u64 bytes_to_set = 0;
1630 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1632 bytes_to_set = min(end - offset, bytes);
1634 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1636 return bytes_to_set;
1640 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1641 struct btrfs_free_space *info)
1643 struct btrfs_block_group_cache *block_group = ctl->private;
1646 * If we are below the extents threshold then we can add this as an
1647 * extent, and don't have to deal with the bitmap
1649 if (ctl->free_extents < ctl->extents_thresh) {
1651 * If this block group has some small extents we don't want to
1652 * use up all of our free slots in the cache with them, we want
1653 * to reserve them to larger extents, however if we have plent
1654 * of cache left then go ahead an dadd them, no sense in adding
1655 * the overhead of a bitmap if we don't have to.
1657 if (info->bytes <= block_group->sectorsize * 4) {
1658 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1666 * The original block groups from mkfs can be really small, like 8
1667 * megabytes, so don't bother with a bitmap for those entries. However
1668 * some block groups can be smaller than what a bitmap would cover but
1669 * are still large enough that they could overflow the 32k memory limit,
1670 * so allow those block groups to still be allowed to have a bitmap
1673 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1679 static struct btrfs_free_space_op free_space_op = {
1680 .recalc_thresholds = recalculate_thresholds,
1681 .use_bitmap = use_bitmap,
1684 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1685 struct btrfs_free_space *info)
1687 struct btrfs_free_space *bitmap_info;
1688 struct btrfs_block_group_cache *block_group = NULL;
1690 u64 bytes, offset, bytes_added;
1693 bytes = info->bytes;
1694 offset = info->offset;
1696 if (!ctl->op->use_bitmap(ctl, info))
1699 if (ctl->op == &free_space_op)
1700 block_group = ctl->private;
1703 * Since we link bitmaps right into the cluster we need to see if we
1704 * have a cluster here, and if so and it has our bitmap we need to add
1705 * the free space to that bitmap.
1707 if (block_group && !list_empty(&block_group->cluster_list)) {
1708 struct btrfs_free_cluster *cluster;
1709 struct rb_node *node;
1710 struct btrfs_free_space *entry;
1712 cluster = list_entry(block_group->cluster_list.next,
1713 struct btrfs_free_cluster,
1715 spin_lock(&cluster->lock);
1716 node = rb_first(&cluster->root);
1718 spin_unlock(&cluster->lock);
1719 goto no_cluster_bitmap;
1722 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1723 if (!entry->bitmap) {
1724 spin_unlock(&cluster->lock);
1725 goto no_cluster_bitmap;
1728 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1729 bytes_added = add_bytes_to_bitmap(ctl, entry,
1731 bytes -= bytes_added;
1732 offset += bytes_added;
1734 spin_unlock(&cluster->lock);
1742 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1749 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1750 bytes -= bytes_added;
1751 offset += bytes_added;
1761 if (info && info->bitmap) {
1762 add_new_bitmap(ctl, info, offset);
1767 spin_unlock(&ctl->tree_lock);
1769 /* no pre-allocated info, allocate a new one */
1771 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1774 spin_lock(&ctl->tree_lock);
1780 /* allocate the bitmap */
1781 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1782 spin_lock(&ctl->tree_lock);
1783 if (!info->bitmap) {
1793 kfree(info->bitmap);
1794 kmem_cache_free(btrfs_free_space_cachep, info);
1800 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1801 struct btrfs_free_space *info, bool update_stat)
1803 struct btrfs_free_space *left_info;
1804 struct btrfs_free_space *right_info;
1805 bool merged = false;
1806 u64 offset = info->offset;
1807 u64 bytes = info->bytes;
1810 * first we want to see if there is free space adjacent to the range we
1811 * are adding, if there is remove that struct and add a new one to
1812 * cover the entire range
1814 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1815 if (right_info && rb_prev(&right_info->offset_index))
1816 left_info = rb_entry(rb_prev(&right_info->offset_index),
1817 struct btrfs_free_space, offset_index);
1819 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1821 if (right_info && !right_info->bitmap) {
1823 unlink_free_space(ctl, right_info);
1825 __unlink_free_space(ctl, right_info);
1826 info->bytes += right_info->bytes;
1827 kmem_cache_free(btrfs_free_space_cachep, right_info);
1831 if (left_info && !left_info->bitmap &&
1832 left_info->offset + left_info->bytes == offset) {
1834 unlink_free_space(ctl, left_info);
1836 __unlink_free_space(ctl, left_info);
1837 info->offset = left_info->offset;
1838 info->bytes += left_info->bytes;
1839 kmem_cache_free(btrfs_free_space_cachep, left_info);
1846 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1847 u64 offset, u64 bytes)
1849 struct btrfs_free_space *info;
1852 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1856 info->offset = offset;
1857 info->bytes = bytes;
1859 spin_lock(&ctl->tree_lock);
1861 if (try_merge_free_space(ctl, info, true))
1865 * There was no extent directly to the left or right of this new
1866 * extent then we know we're going to have to allocate a new extent, so
1867 * before we do that see if we need to drop this into a bitmap
1869 ret = insert_into_bitmap(ctl, info);
1877 ret = link_free_space(ctl, info);
1879 kmem_cache_free(btrfs_free_space_cachep, info);
1881 spin_unlock(&ctl->tree_lock);
1884 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1885 BUG_ON(ret == -EEXIST);
1891 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1892 u64 offset, u64 bytes)
1894 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1895 struct btrfs_free_space *info;
1897 bool re_search = false;
1899 spin_lock(&ctl->tree_lock);
1906 info = tree_search_offset(ctl, offset, 0, 0);
1909 * oops didn't find an extent that matched the space we wanted
1910 * to remove, look for a bitmap instead
1912 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1916 * If we found a partial bit of our free space in a
1917 * bitmap but then couldn't find the other part this may
1918 * be a problem, so WARN about it.
1926 if (!info->bitmap) {
1927 unlink_free_space(ctl, info);
1928 if (offset == info->offset) {
1929 u64 to_free = min(bytes, info->bytes);
1931 info->bytes -= to_free;
1932 info->offset += to_free;
1934 ret = link_free_space(ctl, info);
1937 kmem_cache_free(btrfs_free_space_cachep, info);
1944 u64 old_end = info->bytes + info->offset;
1946 info->bytes = offset - info->offset;
1947 ret = link_free_space(ctl, info);
1952 /* Not enough bytes in this entry to satisfy us */
1953 if (old_end < offset + bytes) {
1954 bytes -= old_end - offset;
1957 } else if (old_end == offset + bytes) {
1961 spin_unlock(&ctl->tree_lock);
1963 ret = btrfs_add_free_space(block_group, offset + bytes,
1964 old_end - (offset + bytes));
1970 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1971 if (ret == -EAGAIN) {
1976 spin_unlock(&ctl->tree_lock);
1981 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1984 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1985 struct btrfs_free_space *info;
1989 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1990 info = rb_entry(n, struct btrfs_free_space, offset_index);
1991 if (info->bytes >= bytes && !block_group->ro)
1993 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1994 (unsigned long long)info->offset,
1995 (unsigned long long)info->bytes,
1996 (info->bitmap) ? "yes" : "no");
1998 printk(KERN_INFO "block group has cluster?: %s\n",
1999 list_empty(&block_group->cluster_list) ? "no" : "yes");
2000 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2004 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2006 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2008 spin_lock_init(&ctl->tree_lock);
2009 ctl->unit = block_group->sectorsize;
2010 ctl->start = block_group->key.objectid;
2011 ctl->private = block_group;
2012 ctl->op = &free_space_op;
2015 * we only want to have 32k of ram per block group for keeping
2016 * track of free space, and if we pass 1/2 of that we want to
2017 * start converting things over to using bitmaps
2019 ctl->extents_thresh = ((1024 * 32) / 2) /
2020 sizeof(struct btrfs_free_space);
2024 * for a given cluster, put all of its extents back into the free
2025 * space cache. If the block group passed doesn't match the block group
2026 * pointed to by the cluster, someone else raced in and freed the
2027 * cluster already. In that case, we just return without changing anything
2030 __btrfs_return_cluster_to_free_space(
2031 struct btrfs_block_group_cache *block_group,
2032 struct btrfs_free_cluster *cluster)
2034 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2035 struct btrfs_free_space *entry;
2036 struct rb_node *node;
2038 spin_lock(&cluster->lock);
2039 if (cluster->block_group != block_group)
2042 cluster->block_group = NULL;
2043 cluster->window_start = 0;
2044 list_del_init(&cluster->block_group_list);
2046 node = rb_first(&cluster->root);
2050 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2051 node = rb_next(&entry->offset_index);
2052 rb_erase(&entry->offset_index, &cluster->root);
2054 bitmap = (entry->bitmap != NULL);
2056 try_merge_free_space(ctl, entry, false);
2057 tree_insert_offset(&ctl->free_space_offset,
2058 entry->offset, &entry->offset_index, bitmap);
2060 cluster->root = RB_ROOT;
2063 spin_unlock(&cluster->lock);
2064 btrfs_put_block_group(block_group);
2068 static void __btrfs_remove_free_space_cache_locked(
2069 struct btrfs_free_space_ctl *ctl)
2071 struct btrfs_free_space *info;
2072 struct rb_node *node;
2074 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2075 info = rb_entry(node, struct btrfs_free_space, offset_index);
2076 if (!info->bitmap) {
2077 unlink_free_space(ctl, info);
2078 kmem_cache_free(btrfs_free_space_cachep, info);
2080 free_bitmap(ctl, info);
2082 if (need_resched()) {
2083 spin_unlock(&ctl->tree_lock);
2085 spin_lock(&ctl->tree_lock);
2090 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2092 spin_lock(&ctl->tree_lock);
2093 __btrfs_remove_free_space_cache_locked(ctl);
2094 spin_unlock(&ctl->tree_lock);
2097 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2099 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2100 struct btrfs_free_cluster *cluster;
2101 struct list_head *head;
2103 spin_lock(&ctl->tree_lock);
2104 while ((head = block_group->cluster_list.next) !=
2105 &block_group->cluster_list) {
2106 cluster = list_entry(head, struct btrfs_free_cluster,
2109 WARN_ON(cluster->block_group != block_group);
2110 __btrfs_return_cluster_to_free_space(block_group, cluster);
2111 if (need_resched()) {
2112 spin_unlock(&ctl->tree_lock);
2114 spin_lock(&ctl->tree_lock);
2117 __btrfs_remove_free_space_cache_locked(ctl);
2118 spin_unlock(&ctl->tree_lock);
2122 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2123 u64 offset, u64 bytes, u64 empty_size)
2125 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2126 struct btrfs_free_space *entry = NULL;
2127 u64 bytes_search = bytes + empty_size;
2130 u64 align_gap_len = 0;
2132 spin_lock(&ctl->tree_lock);
2133 entry = find_free_space(ctl, &offset, &bytes_search,
2134 block_group->full_stripe_len);
2139 if (entry->bitmap) {
2140 bitmap_clear_bits(ctl, entry, offset, bytes);
2142 free_bitmap(ctl, entry);
2145 unlink_free_space(ctl, entry);
2146 align_gap_len = offset - entry->offset;
2147 align_gap = entry->offset;
2149 entry->offset = offset + bytes;
2150 WARN_ON(entry->bytes < bytes + align_gap_len);
2152 entry->bytes -= bytes + align_gap_len;
2154 kmem_cache_free(btrfs_free_space_cachep, entry);
2156 link_free_space(ctl, entry);
2160 spin_unlock(&ctl->tree_lock);
2163 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2168 * given a cluster, put all of its extents back into the free space
2169 * cache. If a block group is passed, this function will only free
2170 * a cluster that belongs to the passed block group.
2172 * Otherwise, it'll get a reference on the block group pointed to by the
2173 * cluster and remove the cluster from it.
2175 int btrfs_return_cluster_to_free_space(
2176 struct btrfs_block_group_cache *block_group,
2177 struct btrfs_free_cluster *cluster)
2179 struct btrfs_free_space_ctl *ctl;
2182 /* first, get a safe pointer to the block group */
2183 spin_lock(&cluster->lock);
2185 block_group = cluster->block_group;
2187 spin_unlock(&cluster->lock);
2190 } else if (cluster->block_group != block_group) {
2191 /* someone else has already freed it don't redo their work */
2192 spin_unlock(&cluster->lock);
2195 atomic_inc(&block_group->count);
2196 spin_unlock(&cluster->lock);
2198 ctl = block_group->free_space_ctl;
2200 /* now return any extents the cluster had on it */
2201 spin_lock(&ctl->tree_lock);
2202 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2203 spin_unlock(&ctl->tree_lock);
2205 /* finally drop our ref */
2206 btrfs_put_block_group(block_group);
2210 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2211 struct btrfs_free_cluster *cluster,
2212 struct btrfs_free_space *entry,
2213 u64 bytes, u64 min_start)
2215 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2217 u64 search_start = cluster->window_start;
2218 u64 search_bytes = bytes;
2221 search_start = min_start;
2222 search_bytes = bytes;
2224 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2229 __bitmap_clear_bits(ctl, entry, ret, bytes);
2235 * given a cluster, try to allocate 'bytes' from it, returns 0
2236 * if it couldn't find anything suitably large, or a logical disk offset
2237 * if things worked out
2239 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2240 struct btrfs_free_cluster *cluster, u64 bytes,
2243 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2244 struct btrfs_free_space *entry = NULL;
2245 struct rb_node *node;
2248 spin_lock(&cluster->lock);
2249 if (bytes > cluster->max_size)
2252 if (cluster->block_group != block_group)
2255 node = rb_first(&cluster->root);
2259 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2261 if (entry->bytes < bytes ||
2262 (!entry->bitmap && entry->offset < min_start)) {
2263 node = rb_next(&entry->offset_index);
2266 entry = rb_entry(node, struct btrfs_free_space,
2271 if (entry->bitmap) {
2272 ret = btrfs_alloc_from_bitmap(block_group,
2273 cluster, entry, bytes,
2274 cluster->window_start);
2276 node = rb_next(&entry->offset_index);
2279 entry = rb_entry(node, struct btrfs_free_space,
2283 cluster->window_start += bytes;
2285 ret = entry->offset;
2287 entry->offset += bytes;
2288 entry->bytes -= bytes;
2291 if (entry->bytes == 0)
2292 rb_erase(&entry->offset_index, &cluster->root);
2296 spin_unlock(&cluster->lock);
2301 spin_lock(&ctl->tree_lock);
2303 ctl->free_space -= bytes;
2304 if (entry->bytes == 0) {
2305 ctl->free_extents--;
2306 if (entry->bitmap) {
2307 kfree(entry->bitmap);
2308 ctl->total_bitmaps--;
2309 ctl->op->recalc_thresholds(ctl);
2311 kmem_cache_free(btrfs_free_space_cachep, entry);
2314 spin_unlock(&ctl->tree_lock);
2319 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2320 struct btrfs_free_space *entry,
2321 struct btrfs_free_cluster *cluster,
2322 u64 offset, u64 bytes,
2323 u64 cont1_bytes, u64 min_bytes)
2325 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2326 unsigned long next_zero;
2328 unsigned long want_bits;
2329 unsigned long min_bits;
2330 unsigned long found_bits;
2331 unsigned long start = 0;
2332 unsigned long total_found = 0;
2335 i = offset_to_bit(entry->offset, ctl->unit,
2336 max_t(u64, offset, entry->offset));
2337 want_bits = bytes_to_bits(bytes, ctl->unit);
2338 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2342 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2343 next_zero = find_next_zero_bit(entry->bitmap,
2344 BITS_PER_BITMAP, i);
2345 if (next_zero - i >= min_bits) {
2346 found_bits = next_zero - i;
2357 cluster->max_size = 0;
2360 total_found += found_bits;
2362 if (cluster->max_size < found_bits * ctl->unit)
2363 cluster->max_size = found_bits * ctl->unit;
2365 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2370 cluster->window_start = start * ctl->unit + entry->offset;
2371 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2372 ret = tree_insert_offset(&cluster->root, entry->offset,
2373 &entry->offset_index, 1);
2374 BUG_ON(ret); /* -EEXIST; Logic error */
2376 trace_btrfs_setup_cluster(block_group, cluster,
2377 total_found * ctl->unit, 1);
2382 * This searches the block group for just extents to fill the cluster with.
2383 * Try to find a cluster with at least bytes total bytes, at least one
2384 * extent of cont1_bytes, and other clusters of at least min_bytes.
2387 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2388 struct btrfs_free_cluster *cluster,
2389 struct list_head *bitmaps, u64 offset, u64 bytes,
2390 u64 cont1_bytes, u64 min_bytes)
2392 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2393 struct btrfs_free_space *first = NULL;
2394 struct btrfs_free_space *entry = NULL;
2395 struct btrfs_free_space *last;
2396 struct rb_node *node;
2402 entry = tree_search_offset(ctl, offset, 0, 1);
2407 * We don't want bitmaps, so just move along until we find a normal
2410 while (entry->bitmap || entry->bytes < min_bytes) {
2411 if (entry->bitmap && list_empty(&entry->list))
2412 list_add_tail(&entry->list, bitmaps);
2413 node = rb_next(&entry->offset_index);
2416 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2419 window_start = entry->offset;
2420 window_free = entry->bytes;
2421 max_extent = entry->bytes;
2425 for (node = rb_next(&entry->offset_index); node;
2426 node = rb_next(&entry->offset_index)) {
2427 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2429 if (entry->bitmap) {
2430 if (list_empty(&entry->list))
2431 list_add_tail(&entry->list, bitmaps);
2435 if (entry->bytes < min_bytes)
2439 window_free += entry->bytes;
2440 if (entry->bytes > max_extent)
2441 max_extent = entry->bytes;
2444 if (window_free < bytes || max_extent < cont1_bytes)
2447 cluster->window_start = first->offset;
2449 node = &first->offset_index;
2452 * now we've found our entries, pull them out of the free space
2453 * cache and put them into the cluster rbtree
2458 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2459 node = rb_next(&entry->offset_index);
2460 if (entry->bitmap || entry->bytes < min_bytes)
2463 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2464 ret = tree_insert_offset(&cluster->root, entry->offset,
2465 &entry->offset_index, 0);
2466 total_size += entry->bytes;
2467 BUG_ON(ret); /* -EEXIST; Logic error */
2468 } while (node && entry != last);
2470 cluster->max_size = max_extent;
2471 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2476 * This specifically looks for bitmaps that may work in the cluster, we assume
2477 * that we have already failed to find extents that will work.
2480 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2481 struct btrfs_free_cluster *cluster,
2482 struct list_head *bitmaps, u64 offset, u64 bytes,
2483 u64 cont1_bytes, u64 min_bytes)
2485 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2486 struct btrfs_free_space *entry;
2488 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2490 if (ctl->total_bitmaps == 0)
2494 * The bitmap that covers offset won't be in the list unless offset
2495 * is just its start offset.
2497 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2498 if (entry->offset != bitmap_offset) {
2499 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2500 if (entry && list_empty(&entry->list))
2501 list_add(&entry->list, bitmaps);
2504 list_for_each_entry(entry, bitmaps, list) {
2505 if (entry->bytes < bytes)
2507 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2508 bytes, cont1_bytes, min_bytes);
2514 * The bitmaps list has all the bitmaps that record free space
2515 * starting after offset, so no more search is required.
2521 * here we try to find a cluster of blocks in a block group. The goal
2522 * is to find at least bytes+empty_size.
2523 * We might not find them all in one contiguous area.
2525 * returns zero and sets up cluster if things worked out, otherwise
2526 * it returns -enospc
2528 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2529 struct btrfs_root *root,
2530 struct btrfs_block_group_cache *block_group,
2531 struct btrfs_free_cluster *cluster,
2532 u64 offset, u64 bytes, u64 empty_size)
2534 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2535 struct btrfs_free_space *entry, *tmp;
2542 * Choose the minimum extent size we'll require for this
2543 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2544 * For metadata, allow allocates with smaller extents. For
2545 * data, keep it dense.
2547 if (btrfs_test_opt(root, SSD_SPREAD)) {
2548 cont1_bytes = min_bytes = bytes + empty_size;
2549 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2550 cont1_bytes = bytes;
2551 min_bytes = block_group->sectorsize;
2553 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2554 min_bytes = block_group->sectorsize;
2557 spin_lock(&ctl->tree_lock);
2560 * If we know we don't have enough space to make a cluster don't even
2561 * bother doing all the work to try and find one.
2563 if (ctl->free_space < bytes) {
2564 spin_unlock(&ctl->tree_lock);
2568 spin_lock(&cluster->lock);
2570 /* someone already found a cluster, hooray */
2571 if (cluster->block_group) {
2576 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2579 INIT_LIST_HEAD(&bitmaps);
2580 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2582 cont1_bytes, min_bytes);
2584 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2585 offset, bytes + empty_size,
2586 cont1_bytes, min_bytes);
2588 /* Clear our temporary list */
2589 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2590 list_del_init(&entry->list);
2593 atomic_inc(&block_group->count);
2594 list_add_tail(&cluster->block_group_list,
2595 &block_group->cluster_list);
2596 cluster->block_group = block_group;
2598 trace_btrfs_failed_cluster_setup(block_group);
2601 spin_unlock(&cluster->lock);
2602 spin_unlock(&ctl->tree_lock);
2608 * simple code to zero out a cluster
2610 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2612 spin_lock_init(&cluster->lock);
2613 spin_lock_init(&cluster->refill_lock);
2614 cluster->root = RB_ROOT;
2615 cluster->max_size = 0;
2616 INIT_LIST_HEAD(&cluster->block_group_list);
2617 cluster->block_group = NULL;
2620 static int do_trimming(struct btrfs_block_group_cache *block_group,
2621 u64 *total_trimmed, u64 start, u64 bytes,
2622 u64 reserved_start, u64 reserved_bytes)
2624 struct btrfs_space_info *space_info = block_group->space_info;
2625 struct btrfs_fs_info *fs_info = block_group->fs_info;
2630 spin_lock(&space_info->lock);
2631 spin_lock(&block_group->lock);
2632 if (!block_group->ro) {
2633 block_group->reserved += reserved_bytes;
2634 space_info->bytes_reserved += reserved_bytes;
2637 spin_unlock(&block_group->lock);
2638 spin_unlock(&space_info->lock);
2640 ret = btrfs_error_discard_extent(fs_info->extent_root,
2641 start, bytes, &trimmed);
2643 *total_trimmed += trimmed;
2645 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2648 spin_lock(&space_info->lock);
2649 spin_lock(&block_group->lock);
2650 if (block_group->ro)
2651 space_info->bytes_readonly += reserved_bytes;
2652 block_group->reserved -= reserved_bytes;
2653 space_info->bytes_reserved -= reserved_bytes;
2654 spin_unlock(&space_info->lock);
2655 spin_unlock(&block_group->lock);
2661 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2662 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2664 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2665 struct btrfs_free_space *entry;
2666 struct rb_node *node;
2672 while (start < end) {
2673 spin_lock(&ctl->tree_lock);
2675 if (ctl->free_space < minlen) {
2676 spin_unlock(&ctl->tree_lock);
2680 entry = tree_search_offset(ctl, start, 0, 1);
2682 spin_unlock(&ctl->tree_lock);
2687 while (entry->bitmap) {
2688 node = rb_next(&entry->offset_index);
2690 spin_unlock(&ctl->tree_lock);
2693 entry = rb_entry(node, struct btrfs_free_space,
2697 if (entry->offset >= end) {
2698 spin_unlock(&ctl->tree_lock);
2702 extent_start = entry->offset;
2703 extent_bytes = entry->bytes;
2704 start = max(start, extent_start);
2705 bytes = min(extent_start + extent_bytes, end) - start;
2706 if (bytes < minlen) {
2707 spin_unlock(&ctl->tree_lock);
2711 unlink_free_space(ctl, entry);
2712 kmem_cache_free(btrfs_free_space_cachep, entry);
2714 spin_unlock(&ctl->tree_lock);
2716 ret = do_trimming(block_group, total_trimmed, start, bytes,
2717 extent_start, extent_bytes);
2723 if (fatal_signal_pending(current)) {
2734 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2735 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2737 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2738 struct btrfs_free_space *entry;
2742 u64 offset = offset_to_bitmap(ctl, start);
2744 while (offset < end) {
2745 bool next_bitmap = false;
2747 spin_lock(&ctl->tree_lock);
2749 if (ctl->free_space < minlen) {
2750 spin_unlock(&ctl->tree_lock);
2754 entry = tree_search_offset(ctl, offset, 1, 0);
2756 spin_unlock(&ctl->tree_lock);
2762 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2763 if (ret2 || start >= end) {
2764 spin_unlock(&ctl->tree_lock);
2769 bytes = min(bytes, end - start);
2770 if (bytes < minlen) {
2771 spin_unlock(&ctl->tree_lock);
2775 bitmap_clear_bits(ctl, entry, start, bytes);
2776 if (entry->bytes == 0)
2777 free_bitmap(ctl, entry);
2779 spin_unlock(&ctl->tree_lock);
2781 ret = do_trimming(block_group, total_trimmed, start, bytes,
2787 offset += BITS_PER_BITMAP * ctl->unit;
2790 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2791 offset += BITS_PER_BITMAP * ctl->unit;
2794 if (fatal_signal_pending(current)) {
2805 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2806 u64 *trimmed, u64 start, u64 end, u64 minlen)
2812 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2816 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2822 * Find the left-most item in the cache tree, and then return the
2823 * smallest inode number in the item.
2825 * Note: the returned inode number may not be the smallest one in
2826 * the tree, if the left-most item is a bitmap.
2828 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2830 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2831 struct btrfs_free_space *entry = NULL;
2834 spin_lock(&ctl->tree_lock);
2836 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2839 entry = rb_entry(rb_first(&ctl->free_space_offset),
2840 struct btrfs_free_space, offset_index);
2842 if (!entry->bitmap) {
2843 ino = entry->offset;
2845 unlink_free_space(ctl, entry);
2849 kmem_cache_free(btrfs_free_space_cachep, entry);
2851 link_free_space(ctl, entry);
2857 ret = search_bitmap(ctl, entry, &offset, &count);
2858 /* Logic error; Should be empty if it can't find anything */
2862 bitmap_clear_bits(ctl, entry, offset, 1);
2863 if (entry->bytes == 0)
2864 free_bitmap(ctl, entry);
2867 spin_unlock(&ctl->tree_lock);
2872 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2873 struct btrfs_path *path)
2875 struct inode *inode = NULL;
2877 spin_lock(&root->cache_lock);
2878 if (root->cache_inode)
2879 inode = igrab(root->cache_inode);
2880 spin_unlock(&root->cache_lock);
2884 inode = __lookup_free_space_inode(root, path, 0);
2888 spin_lock(&root->cache_lock);
2889 if (!btrfs_fs_closing(root->fs_info))
2890 root->cache_inode = igrab(inode);
2891 spin_unlock(&root->cache_lock);
2896 int create_free_ino_inode(struct btrfs_root *root,
2897 struct btrfs_trans_handle *trans,
2898 struct btrfs_path *path)
2900 return __create_free_space_inode(root, trans, path,
2901 BTRFS_FREE_INO_OBJECTID, 0);
2904 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2906 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2907 struct btrfs_path *path;
2908 struct inode *inode;
2910 u64 root_gen = btrfs_root_generation(&root->root_item);
2912 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2916 * If we're unmounting then just return, since this does a search on the
2917 * normal root and not the commit root and we could deadlock.
2919 if (btrfs_fs_closing(fs_info))
2922 path = btrfs_alloc_path();
2926 inode = lookup_free_ino_inode(root, path);
2930 if (root_gen != BTRFS_I(inode)->generation)
2933 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2937 "failed to load free ino cache for root %llu",
2938 root->root_key.objectid);
2942 btrfs_free_path(path);
2946 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2947 struct btrfs_trans_handle *trans,
2948 struct btrfs_path *path)
2950 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2951 struct inode *inode;
2954 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2957 inode = lookup_free_ino_inode(root, path);
2961 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2963 btrfs_delalloc_release_metadata(inode, inode->i_size);
2965 btrfs_err(root->fs_info,
2966 "failed to write free ino cache for root %llu",
2967 root->root_key.objectid);
2975 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2976 static struct btrfs_block_group_cache *init_test_block_group(void)
2978 struct btrfs_block_group_cache *cache;
2980 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2983 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
2985 if (!cache->free_space_ctl) {
2990 cache->key.objectid = 0;
2991 cache->key.offset = 1024 * 1024 * 1024;
2992 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2993 cache->sectorsize = 4096;
2995 spin_lock_init(&cache->lock);
2996 INIT_LIST_HEAD(&cache->list);
2997 INIT_LIST_HEAD(&cache->cluster_list);
2998 INIT_LIST_HEAD(&cache->new_bg_list);
3000 btrfs_init_free_space_ctl(cache);
3006 * Checks to see if the given range is in the free space cache. This is really
3007 * just used to check the absence of space, so if there is free space in the
3008 * range at all we will return 1.
3010 static int check_exists(struct btrfs_block_group_cache *cache, u64 offset,
3013 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3014 struct btrfs_free_space *info;
3017 spin_lock(&ctl->tree_lock);
3018 info = tree_search_offset(ctl, offset, 0, 0);
3020 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3028 u64 bit_off, bit_bytes;
3030 struct btrfs_free_space *tmp;
3033 bit_bytes = ctl->unit;
3034 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3036 if (bit_off == offset) {
3039 } else if (bit_off > offset &&
3040 offset + bytes > bit_off) {
3046 n = rb_prev(&info->offset_index);
3048 tmp = rb_entry(n, struct btrfs_free_space,
3050 if (tmp->offset + tmp->bytes < offset)
3052 if (offset + bytes < tmp->offset) {
3053 n = rb_prev(&info->offset_index);
3060 n = rb_next(&info->offset_index);
3062 tmp = rb_entry(n, struct btrfs_free_space,
3064 if (offset + bytes < tmp->offset)
3066 if (tmp->offset + tmp->bytes < offset) {
3067 n = rb_next(&info->offset_index);
3077 if (info->offset == offset) {
3082 if (offset > info->offset && offset < info->offset + info->bytes)
3085 spin_unlock(&ctl->tree_lock);
3090 * Use this if you need to make a bitmap or extent entry specifically, it
3091 * doesn't do any of the merging that add_free_space does, this acts a lot like
3092 * how the free space cache loading stuff works, so you can get really weird
3095 static int add_free_space_entry(struct btrfs_block_group_cache *cache,
3096 u64 offset, u64 bytes, bool bitmap)
3098 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3099 struct btrfs_free_space *info = NULL, *bitmap_info;
3106 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3112 spin_lock(&ctl->tree_lock);
3113 info->offset = offset;
3114 info->bytes = bytes;
3115 ret = link_free_space(ctl, info);
3116 spin_unlock(&ctl->tree_lock);
3118 kmem_cache_free(btrfs_free_space_cachep, info);
3123 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3125 kmem_cache_free(btrfs_free_space_cachep, info);
3130 spin_lock(&ctl->tree_lock);
3131 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3136 add_new_bitmap(ctl, info, offset);
3140 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3141 bytes -= bytes_added;
3142 offset += bytes_added;
3143 spin_unlock(&ctl->tree_lock);
3154 * This test just does basic sanity checking, making sure we can add an exten
3155 * entry and remove space from either end and the middle, and make sure we can
3156 * remove space that covers adjacent extent entries.
3158 static int test_extents(struct btrfs_block_group_cache *cache)
3162 printk(KERN_ERR "Running extent only tests\n");
3164 /* First just make sure we can remove an entire entry */
3165 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3167 printk(KERN_ERR "Error adding initial extents %d\n", ret);
3171 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3173 printk(KERN_ERR "Error removing extent %d\n", ret);
3177 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3178 printk(KERN_ERR "Full remove left some lingering space\n");
3182 /* Ok edge and middle cases now */
3183 ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024);
3185 printk(KERN_ERR "Error adding half extent %d\n", ret);
3189 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 1 * 1024 * 1024);
3191 printk(KERN_ERR "Error removing tail end %d\n", ret);
3195 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3197 printk(KERN_ERR "Error removing front end %d\n", ret);
3201 ret = btrfs_remove_free_space(cache, 2 * 1024 * 1024, 4096);
3203 printk(KERN_ERR "Error removing middle peice %d\n", ret);
3207 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3208 printk(KERN_ERR "Still have space at the front\n");
3212 if (check_exists(cache, 2 * 1024 * 1024, 4096)) {
3213 printk(KERN_ERR "Still have space in the middle\n");
3217 if (check_exists(cache, 3 * 1024 * 1024, 1 * 1024 * 1024)) {
3218 printk(KERN_ERR "Still have space at the end\n");
3223 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3228 static int test_bitmaps(struct btrfs_block_group_cache *cache)
3230 u64 next_bitmap_offset;
3233 printk(KERN_ERR "Running bitmap only tests\n");
3235 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3237 printk(KERN_ERR "Couldn't create a bitmap entry %d\n", ret);
3241 ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024);
3243 printk(KERN_ERR "Error removing bitmap full range %d\n", ret);
3247 if (check_exists(cache, 0, 4 * 1024 * 1024)) {
3248 printk(KERN_ERR "Left some space in bitmap\n");
3252 ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1);
3254 printk(KERN_ERR "Couldn't add to our bitmap entry %d\n", ret);
3258 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 2 * 1024 * 1024);
3260 printk(KERN_ERR "Couldn't remove middle chunk %d\n", ret);
3265 * The first bitmap we have starts at offset 0 so the next one is just
3266 * at the end of the first bitmap.
3268 next_bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3270 /* Test a bit straddling two bitmaps */
3271 ret = add_free_space_entry(cache, next_bitmap_offset -
3272 (2 * 1024 * 1024), 4 * 1024 * 1024, 1);
3274 printk(KERN_ERR "Couldn't add space that straddles two bitmaps"
3279 ret = btrfs_remove_free_space(cache, next_bitmap_offset -
3280 (1 * 1024 * 1024), 2 * 1024 * 1024);
3282 printk(KERN_ERR "Couldn't remove overlapping space %d\n", ret);
3286 if (check_exists(cache, next_bitmap_offset - (1 * 1024 * 1024),
3288 printk(KERN_ERR "Left some space when removing overlapping\n");
3292 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3297 /* This is the high grade jackassery */
3298 static int test_bitmaps_and_extents(struct btrfs_block_group_cache *cache)
3300 u64 bitmap_offset = (u64)(BITS_PER_BITMAP * 4096);
3303 printk(KERN_ERR "Running bitmap and extent tests\n");
3306 * First let's do something simple, an extent at the same offset as the
3307 * bitmap, but the free space completely in the extent and then
3308 * completely in the bitmap.
3310 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 1 * 1024 * 1024, 1);
3312 printk(KERN_ERR "Couldn't create bitmap entry %d\n", ret);
3316 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3318 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3322 ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024);
3324 printk(KERN_ERR "Couldn't remove extent entry %d\n", ret);
3328 if (check_exists(cache, 0, 1 * 1024 * 1024)) {
3329 printk(KERN_ERR "Left remnants after our remove\n");
3333 /* Now to add back the extent entry and remove from the bitmap */
3334 ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0);
3336 printk(KERN_ERR "Couldn't re-add extent entry %d\n", ret);
3340 ret = btrfs_remove_free_space(cache, 4 * 1024 * 1024, 1 * 1024 * 1024);
3342 printk(KERN_ERR "Couldn't remove from bitmap %d\n", ret);
3346 if (check_exists(cache, 4 * 1024 * 1024, 1 * 1024 * 1024)) {
3347 printk(KERN_ERR "Left remnants in the bitmap\n");
3352 * Ok so a little more evil, extent entry and bitmap at the same offset,
3353 * removing an overlapping chunk.
3355 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 4 * 1024 * 1024, 1);
3357 printk(KERN_ERR "Couldn't add to a bitmap %d\n", ret);
3361 ret = btrfs_remove_free_space(cache, 512 * 1024, 3 * 1024 * 1024);
3363 printk(KERN_ERR "Couldn't remove overlapping space %d\n", ret);
3367 if (check_exists(cache, 512 * 1024, 3 * 1024 * 1024)) {
3368 printk(KERN_ERR "Left over peices after removing "
3373 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3375 /* Now with the extent entry offset into the bitmap */
3376 ret = add_free_space_entry(cache, 4 * 1024 * 1024, 4 * 1024 * 1024, 1);
3378 printk(KERN_ERR "Couldn't add space to the bitmap %d\n", ret);
3382 ret = add_free_space_entry(cache, 2 * 1024 * 1024, 2 * 1024 * 1024, 0);
3384 printk(KERN_ERR "Couldn't add extent to the cache %d\n", ret);
3388 ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 4 * 1024 * 1024);
3390 printk(KERN_ERR "Problem removing overlapping space %d\n", ret);
3394 if (check_exists(cache, 3 * 1024 * 1024, 4 * 1024 * 1024)) {
3395 printk(KERN_ERR "Left something behind when removing space");
3400 * This has blown up in the past, the extent entry starts before the
3401 * bitmap entry, but we're trying to remove an offset that falls
3402 * completely within the bitmap range and is in both the extent entry
3403 * and the bitmap entry, looks like this
3409 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3410 ret = add_free_space_entry(cache, bitmap_offset + 4 * 1024 * 1024,
3411 4 * 1024 * 1024, 1);
3413 printk(KERN_ERR "Couldn't add bitmap %d\n", ret);
3417 ret = add_free_space_entry(cache, bitmap_offset - 1 * 1024 * 1024,
3418 5 * 1024 * 1024, 0);
3420 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3424 ret = btrfs_remove_free_space(cache, bitmap_offset + 1 * 1024 * 1024,
3427 printk(KERN_ERR "Failed to free our space %d\n", ret);
3431 if (check_exists(cache, bitmap_offset + 1 * 1024 * 1024,
3433 printk(KERN_ERR "Left stuff over\n");
3437 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3440 * This blew up before, we have part of the free space in a bitmap and
3441 * then the entirety of the rest of the space in an extent. This used
3442 * to return -EAGAIN back from btrfs_remove_extent, make sure this
3445 ret = add_free_space_entry(cache, 1 * 1024 * 1024, 2 * 1024 * 1024, 1);
3447 printk(KERN_ERR "Couldn't add bitmap entry %d\n", ret);
3451 ret = add_free_space_entry(cache, 3 * 1024 * 1024, 1 * 1024 * 1024, 0);
3453 printk(KERN_ERR "Couldn't add extent entry %d\n", ret);
3457 ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 3 * 1024 * 1024);
3459 printk(KERN_ERR "Error removing bitmap and extent "
3460 "overlapping %d\n", ret);
3464 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3468 void btrfs_test_free_space_cache(void)
3470 struct btrfs_block_group_cache *cache;
3472 printk(KERN_ERR "Running btrfs free space cache tests\n");
3474 cache = init_test_block_group();
3476 printk(KERN_ERR "Couldn't run the tests\n");
3480 if (test_extents(cache))
3482 if (test_bitmaps(cache))
3484 if (test_bitmaps_and_extents(cache))
3487 __btrfs_remove_free_space_cache(cache->free_space_ctl);
3488 kfree(cache->free_space_ctl);
3490 printk(KERN_ERR "Free space cache tests finished\n");
3492 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */