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 ASSERT(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 BTRFS_I(inode)->generation, generation);
683 ret = io_ctl_init(&io_ctl, inode, root);
687 ret = readahead_cache(inode);
691 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
695 ret = io_ctl_check_crc(&io_ctl, 0);
699 ret = io_ctl_check_generation(&io_ctl, generation);
703 while (num_entries) {
704 e = kmem_cache_zalloc(btrfs_free_space_cachep,
709 ret = io_ctl_read_entry(&io_ctl, e, &type);
711 kmem_cache_free(btrfs_free_space_cachep, e);
716 kmem_cache_free(btrfs_free_space_cachep, e);
720 if (type == BTRFS_FREE_SPACE_EXTENT) {
721 spin_lock(&ctl->tree_lock);
722 ret = link_free_space(ctl, e);
723 spin_unlock(&ctl->tree_lock);
725 btrfs_err(root->fs_info,
726 "Duplicate entries in free space cache, dumping");
727 kmem_cache_free(btrfs_free_space_cachep, e);
733 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
736 btrfs_free_space_cachep, e);
739 spin_lock(&ctl->tree_lock);
740 ret = link_free_space(ctl, e);
741 ctl->total_bitmaps++;
742 ctl->op->recalc_thresholds(ctl);
743 spin_unlock(&ctl->tree_lock);
745 btrfs_err(root->fs_info,
746 "Duplicate entries in free space cache, dumping");
747 kmem_cache_free(btrfs_free_space_cachep, e);
750 list_add_tail(&e->list, &bitmaps);
756 io_ctl_unmap_page(&io_ctl);
759 * We add the bitmaps at the end of the entries in order that
760 * the bitmap entries are added to the cache.
762 list_for_each_entry_safe(e, n, &bitmaps, list) {
763 list_del_init(&e->list);
764 ret = io_ctl_read_bitmap(&io_ctl, e);
769 io_ctl_drop_pages(&io_ctl);
770 merge_space_tree(ctl);
773 io_ctl_free(&io_ctl);
776 io_ctl_drop_pages(&io_ctl);
777 __btrfs_remove_free_space_cache(ctl);
781 int load_free_space_cache(struct btrfs_fs_info *fs_info,
782 struct btrfs_block_group_cache *block_group)
784 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
785 struct btrfs_root *root = fs_info->tree_root;
787 struct btrfs_path *path;
790 u64 used = btrfs_block_group_used(&block_group->item);
793 * If this block group has been marked to be cleared for one reason or
794 * another then we can't trust the on disk cache, so just return.
796 spin_lock(&block_group->lock);
797 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
798 spin_unlock(&block_group->lock);
801 spin_unlock(&block_group->lock);
803 path = btrfs_alloc_path();
806 path->search_commit_root = 1;
807 path->skip_locking = 1;
809 inode = lookup_free_space_inode(root, block_group, path);
811 btrfs_free_path(path);
815 /* We may have converted the inode and made the cache invalid. */
816 spin_lock(&block_group->lock);
817 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
818 spin_unlock(&block_group->lock);
819 btrfs_free_path(path);
822 spin_unlock(&block_group->lock);
824 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
825 path, block_group->key.objectid);
826 btrfs_free_path(path);
830 spin_lock(&ctl->tree_lock);
831 matched = (ctl->free_space == (block_group->key.offset - used -
832 block_group->bytes_super));
833 spin_unlock(&ctl->tree_lock);
836 __btrfs_remove_free_space_cache(ctl);
837 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
838 block_group->key.objectid);
843 /* This cache is bogus, make sure it gets cleared */
844 spin_lock(&block_group->lock);
845 block_group->disk_cache_state = BTRFS_DC_CLEAR;
846 spin_unlock(&block_group->lock);
849 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
850 block_group->key.objectid);
858 * __btrfs_write_out_cache - write out cached info to an inode
859 * @root - the root the inode belongs to
860 * @ctl - the free space cache we are going to write out
861 * @block_group - the block_group for this cache if it belongs to a block_group
862 * @trans - the trans handle
863 * @path - the path to use
864 * @offset - the offset for the key we'll insert
866 * This function writes out a free space cache struct to disk for quick recovery
867 * on mount. This will return 0 if it was successfull in writing the cache out,
868 * and -1 if it was not.
870 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
871 struct btrfs_free_space_ctl *ctl,
872 struct btrfs_block_group_cache *block_group,
873 struct btrfs_trans_handle *trans,
874 struct btrfs_path *path, u64 offset)
876 struct btrfs_free_space_header *header;
877 struct extent_buffer *leaf;
878 struct rb_node *node;
879 struct list_head *pos, *n;
880 struct extent_state *cached_state = NULL;
881 struct btrfs_free_cluster *cluster = NULL;
882 struct extent_io_tree *unpin = NULL;
883 struct io_ctl io_ctl;
884 struct list_head bitmap_list;
885 struct btrfs_key key;
886 u64 start, extent_start, extent_end, len;
892 INIT_LIST_HEAD(&bitmap_list);
894 if (!i_size_read(inode))
897 ret = io_ctl_init(&io_ctl, inode, root);
901 /* Get the cluster for this block_group if it exists */
902 if (block_group && !list_empty(&block_group->cluster_list))
903 cluster = list_entry(block_group->cluster_list.next,
904 struct btrfs_free_cluster,
907 /* Lock all pages first so we can lock the extent safely. */
908 io_ctl_prepare_pages(&io_ctl, inode, 0);
910 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
913 node = rb_first(&ctl->free_space_offset);
914 if (!node && cluster) {
915 node = rb_first(&cluster->root);
919 /* Make sure we can fit our crcs into the first page */
920 if (io_ctl.check_crcs &&
921 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
924 io_ctl_set_generation(&io_ctl, trans->transid);
926 /* Write out the extent entries */
928 struct btrfs_free_space *e;
930 e = rb_entry(node, struct btrfs_free_space, offset_index);
933 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
939 list_add_tail(&e->list, &bitmap_list);
942 node = rb_next(node);
943 if (!node && cluster) {
944 node = rb_first(&cluster->root);
950 * We want to add any pinned extents to our free space cache
951 * so we don't leak the space
955 * We shouldn't have switched the pinned extents yet so this is the
958 unpin = root->fs_info->pinned_extents;
961 start = block_group->key.objectid;
963 while (block_group && (start < block_group->key.objectid +
964 block_group->key.offset)) {
965 ret = find_first_extent_bit(unpin, start,
966 &extent_start, &extent_end,
973 /* This pinned extent is out of our range */
974 if (extent_start >= block_group->key.objectid +
975 block_group->key.offset)
978 extent_start = max(extent_start, start);
979 extent_end = min(block_group->key.objectid +
980 block_group->key.offset, extent_end + 1);
981 len = extent_end - extent_start;
984 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
991 /* Write out the bitmaps */
992 list_for_each_safe(pos, n, &bitmap_list) {
993 struct btrfs_free_space *entry =
994 list_entry(pos, struct btrfs_free_space, list);
996 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
999 list_del_init(&entry->list);
1002 /* Zero out the rest of the pages just to make sure */
1003 io_ctl_zero_remaining_pages(&io_ctl);
1005 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1006 0, i_size_read(inode), &cached_state);
1007 io_ctl_drop_pages(&io_ctl);
1008 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1009 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1015 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1017 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1018 key.offset = offset;
1021 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1023 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1024 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1028 leaf = path->nodes[0];
1030 struct btrfs_key found_key;
1031 ASSERT(path->slots[0]);
1033 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1034 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1035 found_key.offset != offset) {
1036 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1038 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1040 btrfs_release_path(path);
1045 BTRFS_I(inode)->generation = trans->transid;
1046 header = btrfs_item_ptr(leaf, path->slots[0],
1047 struct btrfs_free_space_header);
1048 btrfs_set_free_space_entries(leaf, header, entries);
1049 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1050 btrfs_set_free_space_generation(leaf, header, trans->transid);
1051 btrfs_mark_buffer_dirty(leaf);
1052 btrfs_release_path(path);
1056 io_ctl_free(&io_ctl);
1058 invalidate_inode_pages2(inode->i_mapping);
1059 BTRFS_I(inode)->generation = 0;
1061 btrfs_update_inode(trans, root, inode);
1065 list_for_each_safe(pos, n, &bitmap_list) {
1066 struct btrfs_free_space *entry =
1067 list_entry(pos, struct btrfs_free_space, list);
1068 list_del_init(&entry->list);
1070 io_ctl_drop_pages(&io_ctl);
1071 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1072 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1076 int btrfs_write_out_cache(struct btrfs_root *root,
1077 struct btrfs_trans_handle *trans,
1078 struct btrfs_block_group_cache *block_group,
1079 struct btrfs_path *path)
1081 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1082 struct inode *inode;
1085 root = root->fs_info->tree_root;
1087 spin_lock(&block_group->lock);
1088 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1089 spin_unlock(&block_group->lock);
1092 spin_unlock(&block_group->lock);
1094 inode = lookup_free_space_inode(root, block_group, path);
1098 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1099 path, block_group->key.objectid);
1101 spin_lock(&block_group->lock);
1102 block_group->disk_cache_state = BTRFS_DC_ERROR;
1103 spin_unlock(&block_group->lock);
1106 btrfs_err(root->fs_info,
1107 "failed to write free space cache for block group %llu",
1108 block_group->key.objectid);
1116 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1119 ASSERT(offset >= bitmap_start);
1120 offset -= bitmap_start;
1121 return (unsigned long)(div_u64(offset, unit));
1124 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1126 return (unsigned long)(div_u64(bytes, unit));
1129 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1133 u64 bytes_per_bitmap;
1135 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1136 bitmap_start = offset - ctl->start;
1137 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1138 bitmap_start *= bytes_per_bitmap;
1139 bitmap_start += ctl->start;
1141 return bitmap_start;
1144 static int tree_insert_offset(struct rb_root *root, u64 offset,
1145 struct rb_node *node, int bitmap)
1147 struct rb_node **p = &root->rb_node;
1148 struct rb_node *parent = NULL;
1149 struct btrfs_free_space *info;
1153 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1155 if (offset < info->offset) {
1157 } else if (offset > info->offset) {
1158 p = &(*p)->rb_right;
1161 * we could have a bitmap entry and an extent entry
1162 * share the same offset. If this is the case, we want
1163 * the extent entry to always be found first if we do a
1164 * linear search through the tree, since we want to have
1165 * the quickest allocation time, and allocating from an
1166 * extent is faster than allocating from a bitmap. So
1167 * if we're inserting a bitmap and we find an entry at
1168 * this offset, we want to go right, or after this entry
1169 * logically. If we are inserting an extent and we've
1170 * found a bitmap, we want to go left, or before
1178 p = &(*p)->rb_right;
1180 if (!info->bitmap) {
1189 rb_link_node(node, parent, p);
1190 rb_insert_color(node, root);
1196 * searches the tree for the given offset.
1198 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1199 * want a section that has at least bytes size and comes at or after the given
1202 static struct btrfs_free_space *
1203 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1204 u64 offset, int bitmap_only, int fuzzy)
1206 struct rb_node *n = ctl->free_space_offset.rb_node;
1207 struct btrfs_free_space *entry, *prev = NULL;
1209 /* find entry that is closest to the 'offset' */
1216 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1219 if (offset < entry->offset)
1221 else if (offset > entry->offset)
1234 * bitmap entry and extent entry may share same offset,
1235 * in that case, bitmap entry comes after extent entry.
1240 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1241 if (entry->offset != offset)
1244 WARN_ON(!entry->bitmap);
1247 if (entry->bitmap) {
1249 * if previous extent entry covers the offset,
1250 * we should return it instead of the bitmap entry
1252 n = rb_prev(&entry->offset_index);
1254 prev = rb_entry(n, struct btrfs_free_space,
1256 if (!prev->bitmap &&
1257 prev->offset + prev->bytes > offset)
1267 /* find last entry before the 'offset' */
1269 if (entry->offset > offset) {
1270 n = rb_prev(&entry->offset_index);
1272 entry = rb_entry(n, struct btrfs_free_space,
1274 ASSERT(entry->offset <= offset);
1283 if (entry->bitmap) {
1284 n = rb_prev(&entry->offset_index);
1286 prev = rb_entry(n, struct btrfs_free_space,
1288 if (!prev->bitmap &&
1289 prev->offset + prev->bytes > offset)
1292 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1294 } else if (entry->offset + entry->bytes > offset)
1301 if (entry->bitmap) {
1302 if (entry->offset + BITS_PER_BITMAP *
1306 if (entry->offset + entry->bytes > offset)
1310 n = rb_next(&entry->offset_index);
1313 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1319 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1320 struct btrfs_free_space *info)
1322 rb_erase(&info->offset_index, &ctl->free_space_offset);
1323 ctl->free_extents--;
1326 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1327 struct btrfs_free_space *info)
1329 __unlink_free_space(ctl, info);
1330 ctl->free_space -= info->bytes;
1333 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1334 struct btrfs_free_space *info)
1338 ASSERT(info->bytes || info->bitmap);
1339 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1340 &info->offset_index, (info->bitmap != NULL));
1344 ctl->free_space += info->bytes;
1345 ctl->free_extents++;
1349 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1351 struct btrfs_block_group_cache *block_group = ctl->private;
1355 u64 size = block_group->key.offset;
1356 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1357 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1359 max_bitmaps = max(max_bitmaps, 1);
1361 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1364 * The goal is to keep the total amount of memory used per 1gb of space
1365 * at or below 32k, so we need to adjust how much memory we allow to be
1366 * used by extent based free space tracking
1368 if (size < 1024 * 1024 * 1024)
1369 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1371 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1372 div64_u64(size, 1024 * 1024 * 1024);
1375 * we want to account for 1 more bitmap than what we have so we can make
1376 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1377 * we add more bitmaps.
1379 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1381 if (bitmap_bytes >= max_bytes) {
1382 ctl->extents_thresh = 0;
1387 * we want the extent entry threshold to always be at most 1/2 the maxw
1388 * bytes we can have, or whatever is less than that.
1390 extent_bytes = max_bytes - bitmap_bytes;
1391 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1393 ctl->extents_thresh =
1394 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1397 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1398 struct btrfs_free_space *info,
1399 u64 offset, u64 bytes)
1401 unsigned long start, count;
1403 start = offset_to_bit(info->offset, ctl->unit, offset);
1404 count = bytes_to_bits(bytes, ctl->unit);
1405 ASSERT(start + count <= BITS_PER_BITMAP);
1407 bitmap_clear(info->bitmap, start, count);
1409 info->bytes -= bytes;
1412 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1413 struct btrfs_free_space *info, u64 offset,
1416 __bitmap_clear_bits(ctl, info, offset, bytes);
1417 ctl->free_space -= bytes;
1420 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1421 struct btrfs_free_space *info, u64 offset,
1424 unsigned long start, count;
1426 start = offset_to_bit(info->offset, ctl->unit, offset);
1427 count = bytes_to_bits(bytes, ctl->unit);
1428 ASSERT(start + count <= BITS_PER_BITMAP);
1430 bitmap_set(info->bitmap, start, count);
1432 info->bytes += bytes;
1433 ctl->free_space += bytes;
1436 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1437 struct btrfs_free_space *bitmap_info, u64 *offset,
1440 unsigned long found_bits = 0;
1441 unsigned long bits, i;
1442 unsigned long next_zero;
1444 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1445 max_t(u64, *offset, bitmap_info->offset));
1446 bits = bytes_to_bits(*bytes, ctl->unit);
1448 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1449 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1450 BITS_PER_BITMAP, i);
1451 if ((next_zero - i) >= bits) {
1452 found_bits = next_zero - i;
1459 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1460 *bytes = (u64)(found_bits) * ctl->unit;
1467 static struct btrfs_free_space *
1468 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1469 unsigned long align)
1471 struct btrfs_free_space *entry;
1472 struct rb_node *node;
1478 if (!ctl->free_space_offset.rb_node)
1481 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1485 for (node = &entry->offset_index; node; node = rb_next(node)) {
1486 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1487 if (entry->bytes < *bytes)
1490 /* make sure the space returned is big enough
1491 * to match our requested alignment
1493 if (*bytes >= align) {
1494 ctl_off = entry->offset - ctl->start;
1495 tmp = ctl_off + align - 1;;
1497 tmp = tmp * align + ctl->start;
1498 align_off = tmp - entry->offset;
1501 tmp = entry->offset;
1504 if (entry->bytes < *bytes + align_off)
1507 if (entry->bitmap) {
1508 ret = search_bitmap(ctl, entry, &tmp, bytes);
1517 *bytes = entry->bytes - align_off;
1524 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1525 struct btrfs_free_space *info, u64 offset)
1527 info->offset = offset_to_bitmap(ctl, offset);
1529 INIT_LIST_HEAD(&info->list);
1530 link_free_space(ctl, info);
1531 ctl->total_bitmaps++;
1533 ctl->op->recalc_thresholds(ctl);
1536 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1537 struct btrfs_free_space *bitmap_info)
1539 unlink_free_space(ctl, bitmap_info);
1540 kfree(bitmap_info->bitmap);
1541 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1542 ctl->total_bitmaps--;
1543 ctl->op->recalc_thresholds(ctl);
1546 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1547 struct btrfs_free_space *bitmap_info,
1548 u64 *offset, u64 *bytes)
1551 u64 search_start, search_bytes;
1555 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1558 * We need to search for bits in this bitmap. We could only cover some
1559 * of the extent in this bitmap thanks to how we add space, so we need
1560 * to search for as much as it as we can and clear that amount, and then
1561 * go searching for the next bit.
1563 search_start = *offset;
1564 search_bytes = ctl->unit;
1565 search_bytes = min(search_bytes, end - search_start + 1);
1566 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1567 if (ret < 0 || search_start != *offset)
1570 /* We may have found more bits than what we need */
1571 search_bytes = min(search_bytes, *bytes);
1573 /* Cannot clear past the end of the bitmap */
1574 search_bytes = min(search_bytes, end - search_start + 1);
1576 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1577 *offset += search_bytes;
1578 *bytes -= search_bytes;
1581 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1582 if (!bitmap_info->bytes)
1583 free_bitmap(ctl, bitmap_info);
1586 * no entry after this bitmap, but we still have bytes to
1587 * remove, so something has gone wrong.
1592 bitmap_info = rb_entry(next, struct btrfs_free_space,
1596 * if the next entry isn't a bitmap we need to return to let the
1597 * extent stuff do its work.
1599 if (!bitmap_info->bitmap)
1603 * Ok the next item is a bitmap, but it may not actually hold
1604 * the information for the rest of this free space stuff, so
1605 * look for it, and if we don't find it return so we can try
1606 * everything over again.
1608 search_start = *offset;
1609 search_bytes = ctl->unit;
1610 ret = search_bitmap(ctl, bitmap_info, &search_start,
1612 if (ret < 0 || search_start != *offset)
1616 } else if (!bitmap_info->bytes)
1617 free_bitmap(ctl, bitmap_info);
1622 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1623 struct btrfs_free_space *info, u64 offset,
1626 u64 bytes_to_set = 0;
1629 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1631 bytes_to_set = min(end - offset, bytes);
1633 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1635 return bytes_to_set;
1639 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1640 struct btrfs_free_space *info)
1642 struct btrfs_block_group_cache *block_group = ctl->private;
1645 * If we are below the extents threshold then we can add this as an
1646 * extent, and don't have to deal with the bitmap
1648 if (ctl->free_extents < ctl->extents_thresh) {
1650 * If this block group has some small extents we don't want to
1651 * use up all of our free slots in the cache with them, we want
1652 * to reserve them to larger extents, however if we have plent
1653 * of cache left then go ahead an dadd them, no sense in adding
1654 * the overhead of a bitmap if we don't have to.
1656 if (info->bytes <= block_group->sectorsize * 4) {
1657 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1665 * The original block groups from mkfs can be really small, like 8
1666 * megabytes, so don't bother with a bitmap for those entries. However
1667 * some block groups can be smaller than what a bitmap would cover but
1668 * are still large enough that they could overflow the 32k memory limit,
1669 * so allow those block groups to still be allowed to have a bitmap
1672 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1678 static struct btrfs_free_space_op free_space_op = {
1679 .recalc_thresholds = recalculate_thresholds,
1680 .use_bitmap = use_bitmap,
1683 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1684 struct btrfs_free_space *info)
1686 struct btrfs_free_space *bitmap_info;
1687 struct btrfs_block_group_cache *block_group = NULL;
1689 u64 bytes, offset, bytes_added;
1692 bytes = info->bytes;
1693 offset = info->offset;
1695 if (!ctl->op->use_bitmap(ctl, info))
1698 if (ctl->op == &free_space_op)
1699 block_group = ctl->private;
1702 * Since we link bitmaps right into the cluster we need to see if we
1703 * have a cluster here, and if so and it has our bitmap we need to add
1704 * the free space to that bitmap.
1706 if (block_group && !list_empty(&block_group->cluster_list)) {
1707 struct btrfs_free_cluster *cluster;
1708 struct rb_node *node;
1709 struct btrfs_free_space *entry;
1711 cluster = list_entry(block_group->cluster_list.next,
1712 struct btrfs_free_cluster,
1714 spin_lock(&cluster->lock);
1715 node = rb_first(&cluster->root);
1717 spin_unlock(&cluster->lock);
1718 goto no_cluster_bitmap;
1721 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1722 if (!entry->bitmap) {
1723 spin_unlock(&cluster->lock);
1724 goto no_cluster_bitmap;
1727 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1728 bytes_added = add_bytes_to_bitmap(ctl, entry,
1730 bytes -= bytes_added;
1731 offset += bytes_added;
1733 spin_unlock(&cluster->lock);
1741 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1748 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1749 bytes -= bytes_added;
1750 offset += bytes_added;
1760 if (info && info->bitmap) {
1761 add_new_bitmap(ctl, info, offset);
1766 spin_unlock(&ctl->tree_lock);
1768 /* no pre-allocated info, allocate a new one */
1770 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1773 spin_lock(&ctl->tree_lock);
1779 /* allocate the bitmap */
1780 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1781 spin_lock(&ctl->tree_lock);
1782 if (!info->bitmap) {
1792 kfree(info->bitmap);
1793 kmem_cache_free(btrfs_free_space_cachep, info);
1799 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1800 struct btrfs_free_space *info, bool update_stat)
1802 struct btrfs_free_space *left_info;
1803 struct btrfs_free_space *right_info;
1804 bool merged = false;
1805 u64 offset = info->offset;
1806 u64 bytes = info->bytes;
1809 * first we want to see if there is free space adjacent to the range we
1810 * are adding, if there is remove that struct and add a new one to
1811 * cover the entire range
1813 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1814 if (right_info && rb_prev(&right_info->offset_index))
1815 left_info = rb_entry(rb_prev(&right_info->offset_index),
1816 struct btrfs_free_space, offset_index);
1818 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1820 if (right_info && !right_info->bitmap) {
1822 unlink_free_space(ctl, right_info);
1824 __unlink_free_space(ctl, right_info);
1825 info->bytes += right_info->bytes;
1826 kmem_cache_free(btrfs_free_space_cachep, right_info);
1830 if (left_info && !left_info->bitmap &&
1831 left_info->offset + left_info->bytes == offset) {
1833 unlink_free_space(ctl, left_info);
1835 __unlink_free_space(ctl, left_info);
1836 info->offset = left_info->offset;
1837 info->bytes += left_info->bytes;
1838 kmem_cache_free(btrfs_free_space_cachep, left_info);
1845 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1846 u64 offset, u64 bytes)
1848 struct btrfs_free_space *info;
1851 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1855 info->offset = offset;
1856 info->bytes = bytes;
1858 spin_lock(&ctl->tree_lock);
1860 if (try_merge_free_space(ctl, info, true))
1864 * There was no extent directly to the left or right of this new
1865 * extent then we know we're going to have to allocate a new extent, so
1866 * before we do that see if we need to drop this into a bitmap
1868 ret = insert_into_bitmap(ctl, info);
1876 ret = link_free_space(ctl, info);
1878 kmem_cache_free(btrfs_free_space_cachep, info);
1880 spin_unlock(&ctl->tree_lock);
1883 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1884 ASSERT(ret != -EEXIST);
1890 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1891 u64 offset, u64 bytes)
1893 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1894 struct btrfs_free_space *info;
1896 bool re_search = false;
1898 spin_lock(&ctl->tree_lock);
1905 info = tree_search_offset(ctl, offset, 0, 0);
1908 * oops didn't find an extent that matched the space we wanted
1909 * to remove, look for a bitmap instead
1911 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1915 * If we found a partial bit of our free space in a
1916 * bitmap but then couldn't find the other part this may
1917 * be a problem, so WARN about it.
1925 if (!info->bitmap) {
1926 unlink_free_space(ctl, info);
1927 if (offset == info->offset) {
1928 u64 to_free = min(bytes, info->bytes);
1930 info->bytes -= to_free;
1931 info->offset += to_free;
1933 ret = link_free_space(ctl, info);
1936 kmem_cache_free(btrfs_free_space_cachep, info);
1943 u64 old_end = info->bytes + info->offset;
1945 info->bytes = offset - info->offset;
1946 ret = link_free_space(ctl, info);
1951 /* Not enough bytes in this entry to satisfy us */
1952 if (old_end < offset + bytes) {
1953 bytes -= old_end - offset;
1956 } else if (old_end == offset + bytes) {
1960 spin_unlock(&ctl->tree_lock);
1962 ret = btrfs_add_free_space(block_group, offset + bytes,
1963 old_end - (offset + bytes));
1969 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1970 if (ret == -EAGAIN) {
1975 spin_unlock(&ctl->tree_lock);
1980 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1983 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1984 struct btrfs_free_space *info;
1988 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1989 info = rb_entry(n, struct btrfs_free_space, offset_index);
1990 if (info->bytes >= bytes && !block_group->ro)
1992 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1993 info->offset, info->bytes,
1994 (info->bitmap) ? "yes" : "no");
1996 printk(KERN_INFO "block group has cluster?: %s\n",
1997 list_empty(&block_group->cluster_list) ? "no" : "yes");
1998 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2002 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2004 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2006 spin_lock_init(&ctl->tree_lock);
2007 ctl->unit = block_group->sectorsize;
2008 ctl->start = block_group->key.objectid;
2009 ctl->private = block_group;
2010 ctl->op = &free_space_op;
2013 * we only want to have 32k of ram per block group for keeping
2014 * track of free space, and if we pass 1/2 of that we want to
2015 * start converting things over to using bitmaps
2017 ctl->extents_thresh = ((1024 * 32) / 2) /
2018 sizeof(struct btrfs_free_space);
2022 * for a given cluster, put all of its extents back into the free
2023 * space cache. If the block group passed doesn't match the block group
2024 * pointed to by the cluster, someone else raced in and freed the
2025 * cluster already. In that case, we just return without changing anything
2028 __btrfs_return_cluster_to_free_space(
2029 struct btrfs_block_group_cache *block_group,
2030 struct btrfs_free_cluster *cluster)
2032 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2033 struct btrfs_free_space *entry;
2034 struct rb_node *node;
2036 spin_lock(&cluster->lock);
2037 if (cluster->block_group != block_group)
2040 cluster->block_group = NULL;
2041 cluster->window_start = 0;
2042 list_del_init(&cluster->block_group_list);
2044 node = rb_first(&cluster->root);
2048 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2049 node = rb_next(&entry->offset_index);
2050 rb_erase(&entry->offset_index, &cluster->root);
2052 bitmap = (entry->bitmap != NULL);
2054 try_merge_free_space(ctl, entry, false);
2055 tree_insert_offset(&ctl->free_space_offset,
2056 entry->offset, &entry->offset_index, bitmap);
2058 cluster->root = RB_ROOT;
2061 spin_unlock(&cluster->lock);
2062 btrfs_put_block_group(block_group);
2066 static void __btrfs_remove_free_space_cache_locked(
2067 struct btrfs_free_space_ctl *ctl)
2069 struct btrfs_free_space *info;
2070 struct rb_node *node;
2072 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2073 info = rb_entry(node, struct btrfs_free_space, offset_index);
2074 if (!info->bitmap) {
2075 unlink_free_space(ctl, info);
2076 kmem_cache_free(btrfs_free_space_cachep, info);
2078 free_bitmap(ctl, info);
2080 if (need_resched()) {
2081 spin_unlock(&ctl->tree_lock);
2083 spin_lock(&ctl->tree_lock);
2088 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2090 spin_lock(&ctl->tree_lock);
2091 __btrfs_remove_free_space_cache_locked(ctl);
2092 spin_unlock(&ctl->tree_lock);
2095 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2097 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2098 struct btrfs_free_cluster *cluster;
2099 struct list_head *head;
2101 spin_lock(&ctl->tree_lock);
2102 while ((head = block_group->cluster_list.next) !=
2103 &block_group->cluster_list) {
2104 cluster = list_entry(head, struct btrfs_free_cluster,
2107 WARN_ON(cluster->block_group != block_group);
2108 __btrfs_return_cluster_to_free_space(block_group, cluster);
2109 if (need_resched()) {
2110 spin_unlock(&ctl->tree_lock);
2112 spin_lock(&ctl->tree_lock);
2115 __btrfs_remove_free_space_cache_locked(ctl);
2116 spin_unlock(&ctl->tree_lock);
2120 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2121 u64 offset, u64 bytes, u64 empty_size)
2123 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2124 struct btrfs_free_space *entry = NULL;
2125 u64 bytes_search = bytes + empty_size;
2128 u64 align_gap_len = 0;
2130 spin_lock(&ctl->tree_lock);
2131 entry = find_free_space(ctl, &offset, &bytes_search,
2132 block_group->full_stripe_len);
2137 if (entry->bitmap) {
2138 bitmap_clear_bits(ctl, entry, offset, bytes);
2140 free_bitmap(ctl, entry);
2143 unlink_free_space(ctl, entry);
2144 align_gap_len = offset - entry->offset;
2145 align_gap = entry->offset;
2147 entry->offset = offset + bytes;
2148 WARN_ON(entry->bytes < bytes + align_gap_len);
2150 entry->bytes -= bytes + align_gap_len;
2152 kmem_cache_free(btrfs_free_space_cachep, entry);
2154 link_free_space(ctl, entry);
2158 spin_unlock(&ctl->tree_lock);
2161 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2166 * given a cluster, put all of its extents back into the free space
2167 * cache. If a block group is passed, this function will only free
2168 * a cluster that belongs to the passed block group.
2170 * Otherwise, it'll get a reference on the block group pointed to by the
2171 * cluster and remove the cluster from it.
2173 int btrfs_return_cluster_to_free_space(
2174 struct btrfs_block_group_cache *block_group,
2175 struct btrfs_free_cluster *cluster)
2177 struct btrfs_free_space_ctl *ctl;
2180 /* first, get a safe pointer to the block group */
2181 spin_lock(&cluster->lock);
2183 block_group = cluster->block_group;
2185 spin_unlock(&cluster->lock);
2188 } else if (cluster->block_group != block_group) {
2189 /* someone else has already freed it don't redo their work */
2190 spin_unlock(&cluster->lock);
2193 atomic_inc(&block_group->count);
2194 spin_unlock(&cluster->lock);
2196 ctl = block_group->free_space_ctl;
2198 /* now return any extents the cluster had on it */
2199 spin_lock(&ctl->tree_lock);
2200 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2201 spin_unlock(&ctl->tree_lock);
2203 /* finally drop our ref */
2204 btrfs_put_block_group(block_group);
2208 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2209 struct btrfs_free_cluster *cluster,
2210 struct btrfs_free_space *entry,
2211 u64 bytes, u64 min_start)
2213 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2215 u64 search_start = cluster->window_start;
2216 u64 search_bytes = bytes;
2219 search_start = min_start;
2220 search_bytes = bytes;
2222 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2227 __bitmap_clear_bits(ctl, entry, ret, bytes);
2233 * given a cluster, try to allocate 'bytes' from it, returns 0
2234 * if it couldn't find anything suitably large, or a logical disk offset
2235 * if things worked out
2237 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2238 struct btrfs_free_cluster *cluster, u64 bytes,
2241 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2242 struct btrfs_free_space *entry = NULL;
2243 struct rb_node *node;
2246 spin_lock(&cluster->lock);
2247 if (bytes > cluster->max_size)
2250 if (cluster->block_group != block_group)
2253 node = rb_first(&cluster->root);
2257 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2259 if (entry->bytes < bytes ||
2260 (!entry->bitmap && entry->offset < min_start)) {
2261 node = rb_next(&entry->offset_index);
2264 entry = rb_entry(node, struct btrfs_free_space,
2269 if (entry->bitmap) {
2270 ret = btrfs_alloc_from_bitmap(block_group,
2271 cluster, entry, bytes,
2272 cluster->window_start);
2274 node = rb_next(&entry->offset_index);
2277 entry = rb_entry(node, struct btrfs_free_space,
2281 cluster->window_start += bytes;
2283 ret = entry->offset;
2285 entry->offset += bytes;
2286 entry->bytes -= bytes;
2289 if (entry->bytes == 0)
2290 rb_erase(&entry->offset_index, &cluster->root);
2294 spin_unlock(&cluster->lock);
2299 spin_lock(&ctl->tree_lock);
2301 ctl->free_space -= bytes;
2302 if (entry->bytes == 0) {
2303 ctl->free_extents--;
2304 if (entry->bitmap) {
2305 kfree(entry->bitmap);
2306 ctl->total_bitmaps--;
2307 ctl->op->recalc_thresholds(ctl);
2309 kmem_cache_free(btrfs_free_space_cachep, entry);
2312 spin_unlock(&ctl->tree_lock);
2317 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2318 struct btrfs_free_space *entry,
2319 struct btrfs_free_cluster *cluster,
2320 u64 offset, u64 bytes,
2321 u64 cont1_bytes, u64 min_bytes)
2323 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2324 unsigned long next_zero;
2326 unsigned long want_bits;
2327 unsigned long min_bits;
2328 unsigned long found_bits;
2329 unsigned long start = 0;
2330 unsigned long total_found = 0;
2333 i = offset_to_bit(entry->offset, ctl->unit,
2334 max_t(u64, offset, entry->offset));
2335 want_bits = bytes_to_bits(bytes, ctl->unit);
2336 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2340 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2341 next_zero = find_next_zero_bit(entry->bitmap,
2342 BITS_PER_BITMAP, i);
2343 if (next_zero - i >= min_bits) {
2344 found_bits = next_zero - i;
2355 cluster->max_size = 0;
2358 total_found += found_bits;
2360 if (cluster->max_size < found_bits * ctl->unit)
2361 cluster->max_size = found_bits * ctl->unit;
2363 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2368 cluster->window_start = start * ctl->unit + entry->offset;
2369 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2370 ret = tree_insert_offset(&cluster->root, entry->offset,
2371 &entry->offset_index, 1);
2372 ASSERT(!ret); /* -EEXIST; Logic error */
2374 trace_btrfs_setup_cluster(block_group, cluster,
2375 total_found * ctl->unit, 1);
2380 * This searches the block group for just extents to fill the cluster with.
2381 * Try to find a cluster with at least bytes total bytes, at least one
2382 * extent of cont1_bytes, and other clusters of at least min_bytes.
2385 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2386 struct btrfs_free_cluster *cluster,
2387 struct list_head *bitmaps, u64 offset, u64 bytes,
2388 u64 cont1_bytes, u64 min_bytes)
2390 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2391 struct btrfs_free_space *first = NULL;
2392 struct btrfs_free_space *entry = NULL;
2393 struct btrfs_free_space *last;
2394 struct rb_node *node;
2400 entry = tree_search_offset(ctl, offset, 0, 1);
2405 * We don't want bitmaps, so just move along until we find a normal
2408 while (entry->bitmap || entry->bytes < min_bytes) {
2409 if (entry->bitmap && list_empty(&entry->list))
2410 list_add_tail(&entry->list, bitmaps);
2411 node = rb_next(&entry->offset_index);
2414 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2417 window_start = entry->offset;
2418 window_free = entry->bytes;
2419 max_extent = entry->bytes;
2423 for (node = rb_next(&entry->offset_index); node;
2424 node = rb_next(&entry->offset_index)) {
2425 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2427 if (entry->bitmap) {
2428 if (list_empty(&entry->list))
2429 list_add_tail(&entry->list, bitmaps);
2433 if (entry->bytes < min_bytes)
2437 window_free += entry->bytes;
2438 if (entry->bytes > max_extent)
2439 max_extent = entry->bytes;
2442 if (window_free < bytes || max_extent < cont1_bytes)
2445 cluster->window_start = first->offset;
2447 node = &first->offset_index;
2450 * now we've found our entries, pull them out of the free space
2451 * cache and put them into the cluster rbtree
2456 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2457 node = rb_next(&entry->offset_index);
2458 if (entry->bitmap || entry->bytes < min_bytes)
2461 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2462 ret = tree_insert_offset(&cluster->root, entry->offset,
2463 &entry->offset_index, 0);
2464 total_size += entry->bytes;
2465 ASSERT(!ret); /* -EEXIST; Logic error */
2466 } while (node && entry != last);
2468 cluster->max_size = max_extent;
2469 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2474 * This specifically looks for bitmaps that may work in the cluster, we assume
2475 * that we have already failed to find extents that will work.
2478 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2479 struct btrfs_free_cluster *cluster,
2480 struct list_head *bitmaps, u64 offset, u64 bytes,
2481 u64 cont1_bytes, u64 min_bytes)
2483 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2484 struct btrfs_free_space *entry;
2486 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2488 if (ctl->total_bitmaps == 0)
2492 * The bitmap that covers offset won't be in the list unless offset
2493 * is just its start offset.
2495 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2496 if (entry->offset != bitmap_offset) {
2497 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2498 if (entry && list_empty(&entry->list))
2499 list_add(&entry->list, bitmaps);
2502 list_for_each_entry(entry, bitmaps, list) {
2503 if (entry->bytes < bytes)
2505 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2506 bytes, cont1_bytes, min_bytes);
2512 * The bitmaps list has all the bitmaps that record free space
2513 * starting after offset, so no more search is required.
2519 * here we try to find a cluster of blocks in a block group. The goal
2520 * is to find at least bytes+empty_size.
2521 * We might not find them all in one contiguous area.
2523 * returns zero and sets up cluster if things worked out, otherwise
2524 * it returns -enospc
2526 int btrfs_find_space_cluster(struct btrfs_root *root,
2527 struct btrfs_block_group_cache *block_group,
2528 struct btrfs_free_cluster *cluster,
2529 u64 offset, u64 bytes, u64 empty_size)
2531 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2532 struct btrfs_free_space *entry, *tmp;
2539 * Choose the minimum extent size we'll require for this
2540 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2541 * For metadata, allow allocates with smaller extents. For
2542 * data, keep it dense.
2544 if (btrfs_test_opt(root, SSD_SPREAD)) {
2545 cont1_bytes = min_bytes = bytes + empty_size;
2546 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2547 cont1_bytes = bytes;
2548 min_bytes = block_group->sectorsize;
2550 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2551 min_bytes = block_group->sectorsize;
2554 spin_lock(&ctl->tree_lock);
2557 * If we know we don't have enough space to make a cluster don't even
2558 * bother doing all the work to try and find one.
2560 if (ctl->free_space < bytes) {
2561 spin_unlock(&ctl->tree_lock);
2565 spin_lock(&cluster->lock);
2567 /* someone already found a cluster, hooray */
2568 if (cluster->block_group) {
2573 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2576 INIT_LIST_HEAD(&bitmaps);
2577 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2579 cont1_bytes, min_bytes);
2581 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2582 offset, bytes + empty_size,
2583 cont1_bytes, min_bytes);
2585 /* Clear our temporary list */
2586 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2587 list_del_init(&entry->list);
2590 atomic_inc(&block_group->count);
2591 list_add_tail(&cluster->block_group_list,
2592 &block_group->cluster_list);
2593 cluster->block_group = block_group;
2595 trace_btrfs_failed_cluster_setup(block_group);
2598 spin_unlock(&cluster->lock);
2599 spin_unlock(&ctl->tree_lock);
2605 * simple code to zero out a cluster
2607 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2609 spin_lock_init(&cluster->lock);
2610 spin_lock_init(&cluster->refill_lock);
2611 cluster->root = RB_ROOT;
2612 cluster->max_size = 0;
2613 INIT_LIST_HEAD(&cluster->block_group_list);
2614 cluster->block_group = NULL;
2617 static int do_trimming(struct btrfs_block_group_cache *block_group,
2618 u64 *total_trimmed, u64 start, u64 bytes,
2619 u64 reserved_start, u64 reserved_bytes)
2621 struct btrfs_space_info *space_info = block_group->space_info;
2622 struct btrfs_fs_info *fs_info = block_group->fs_info;
2627 spin_lock(&space_info->lock);
2628 spin_lock(&block_group->lock);
2629 if (!block_group->ro) {
2630 block_group->reserved += reserved_bytes;
2631 space_info->bytes_reserved += reserved_bytes;
2634 spin_unlock(&block_group->lock);
2635 spin_unlock(&space_info->lock);
2637 ret = btrfs_error_discard_extent(fs_info->extent_root,
2638 start, bytes, &trimmed);
2640 *total_trimmed += trimmed;
2642 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2645 spin_lock(&space_info->lock);
2646 spin_lock(&block_group->lock);
2647 if (block_group->ro)
2648 space_info->bytes_readonly += reserved_bytes;
2649 block_group->reserved -= reserved_bytes;
2650 space_info->bytes_reserved -= reserved_bytes;
2651 spin_unlock(&space_info->lock);
2652 spin_unlock(&block_group->lock);
2658 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2659 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2661 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2662 struct btrfs_free_space *entry;
2663 struct rb_node *node;
2669 while (start < end) {
2670 spin_lock(&ctl->tree_lock);
2672 if (ctl->free_space < minlen) {
2673 spin_unlock(&ctl->tree_lock);
2677 entry = tree_search_offset(ctl, start, 0, 1);
2679 spin_unlock(&ctl->tree_lock);
2684 while (entry->bitmap) {
2685 node = rb_next(&entry->offset_index);
2687 spin_unlock(&ctl->tree_lock);
2690 entry = rb_entry(node, struct btrfs_free_space,
2694 if (entry->offset >= end) {
2695 spin_unlock(&ctl->tree_lock);
2699 extent_start = entry->offset;
2700 extent_bytes = entry->bytes;
2701 start = max(start, extent_start);
2702 bytes = min(extent_start + extent_bytes, end) - start;
2703 if (bytes < minlen) {
2704 spin_unlock(&ctl->tree_lock);
2708 unlink_free_space(ctl, entry);
2709 kmem_cache_free(btrfs_free_space_cachep, entry);
2711 spin_unlock(&ctl->tree_lock);
2713 ret = do_trimming(block_group, total_trimmed, start, bytes,
2714 extent_start, extent_bytes);
2720 if (fatal_signal_pending(current)) {
2731 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2732 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2734 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2735 struct btrfs_free_space *entry;
2739 u64 offset = offset_to_bitmap(ctl, start);
2741 while (offset < end) {
2742 bool next_bitmap = false;
2744 spin_lock(&ctl->tree_lock);
2746 if (ctl->free_space < minlen) {
2747 spin_unlock(&ctl->tree_lock);
2751 entry = tree_search_offset(ctl, offset, 1, 0);
2753 spin_unlock(&ctl->tree_lock);
2759 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2760 if (ret2 || start >= end) {
2761 spin_unlock(&ctl->tree_lock);
2766 bytes = min(bytes, end - start);
2767 if (bytes < minlen) {
2768 spin_unlock(&ctl->tree_lock);
2772 bitmap_clear_bits(ctl, entry, start, bytes);
2773 if (entry->bytes == 0)
2774 free_bitmap(ctl, entry);
2776 spin_unlock(&ctl->tree_lock);
2778 ret = do_trimming(block_group, total_trimmed, start, bytes,
2784 offset += BITS_PER_BITMAP * ctl->unit;
2787 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2788 offset += BITS_PER_BITMAP * ctl->unit;
2791 if (fatal_signal_pending(current)) {
2802 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2803 u64 *trimmed, u64 start, u64 end, u64 minlen)
2809 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2813 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2819 * Find the left-most item in the cache tree, and then return the
2820 * smallest inode number in the item.
2822 * Note: the returned inode number may not be the smallest one in
2823 * the tree, if the left-most item is a bitmap.
2825 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2827 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2828 struct btrfs_free_space *entry = NULL;
2831 spin_lock(&ctl->tree_lock);
2833 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2836 entry = rb_entry(rb_first(&ctl->free_space_offset),
2837 struct btrfs_free_space, offset_index);
2839 if (!entry->bitmap) {
2840 ino = entry->offset;
2842 unlink_free_space(ctl, entry);
2846 kmem_cache_free(btrfs_free_space_cachep, entry);
2848 link_free_space(ctl, entry);
2854 ret = search_bitmap(ctl, entry, &offset, &count);
2855 /* Logic error; Should be empty if it can't find anything */
2859 bitmap_clear_bits(ctl, entry, offset, 1);
2860 if (entry->bytes == 0)
2861 free_bitmap(ctl, entry);
2864 spin_unlock(&ctl->tree_lock);
2869 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2870 struct btrfs_path *path)
2872 struct inode *inode = NULL;
2874 spin_lock(&root->cache_lock);
2875 if (root->cache_inode)
2876 inode = igrab(root->cache_inode);
2877 spin_unlock(&root->cache_lock);
2881 inode = __lookup_free_space_inode(root, path, 0);
2885 spin_lock(&root->cache_lock);
2886 if (!btrfs_fs_closing(root->fs_info))
2887 root->cache_inode = igrab(inode);
2888 spin_unlock(&root->cache_lock);
2893 int create_free_ino_inode(struct btrfs_root *root,
2894 struct btrfs_trans_handle *trans,
2895 struct btrfs_path *path)
2897 return __create_free_space_inode(root, trans, path,
2898 BTRFS_FREE_INO_OBJECTID, 0);
2901 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2903 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2904 struct btrfs_path *path;
2905 struct inode *inode;
2907 u64 root_gen = btrfs_root_generation(&root->root_item);
2909 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2913 * If we're unmounting then just return, since this does a search on the
2914 * normal root and not the commit root and we could deadlock.
2916 if (btrfs_fs_closing(fs_info))
2919 path = btrfs_alloc_path();
2923 inode = lookup_free_ino_inode(root, path);
2927 if (root_gen != BTRFS_I(inode)->generation)
2930 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2934 "failed to load free ino cache for root %llu",
2935 root->root_key.objectid);
2939 btrfs_free_path(path);
2943 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2944 struct btrfs_trans_handle *trans,
2945 struct btrfs_path *path)
2947 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2948 struct inode *inode;
2951 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2954 inode = lookup_free_ino_inode(root, path);
2958 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2960 btrfs_delalloc_release_metadata(inode, inode->i_size);
2962 btrfs_err(root->fs_info,
2963 "failed to write free ino cache for root %llu",
2964 root->root_key.objectid);
2972 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2974 * Use this if you need to make a bitmap or extent entry specifically, it
2975 * doesn't do any of the merging that add_free_space does, this acts a lot like
2976 * how the free space cache loading stuff works, so you can get really weird
2979 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
2980 u64 offset, u64 bytes, bool bitmap)
2982 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
2983 struct btrfs_free_space *info = NULL, *bitmap_info;
2990 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2996 spin_lock(&ctl->tree_lock);
2997 info->offset = offset;
2998 info->bytes = bytes;
2999 ret = link_free_space(ctl, info);
3000 spin_unlock(&ctl->tree_lock);
3002 kmem_cache_free(btrfs_free_space_cachep, info);
3007 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3009 kmem_cache_free(btrfs_free_space_cachep, info);
3014 spin_lock(&ctl->tree_lock);
3015 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3020 add_new_bitmap(ctl, info, offset);
3024 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3025 bytes -= bytes_added;
3026 offset += bytes_added;
3027 spin_unlock(&ctl->tree_lock);
3038 * Checks to see if the given range is in the free space cache. This is really
3039 * just used to check the absence of space, so if there is free space in the
3040 * range at all we will return 1.
3042 int test_check_exists(struct btrfs_block_group_cache *cache,
3043 u64 offset, u64 bytes)
3045 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3046 struct btrfs_free_space *info;
3049 spin_lock(&ctl->tree_lock);
3050 info = tree_search_offset(ctl, offset, 0, 0);
3052 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3060 u64 bit_off, bit_bytes;
3062 struct btrfs_free_space *tmp;
3065 bit_bytes = ctl->unit;
3066 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3068 if (bit_off == offset) {
3071 } else if (bit_off > offset &&
3072 offset + bytes > bit_off) {
3078 n = rb_prev(&info->offset_index);
3080 tmp = rb_entry(n, struct btrfs_free_space,
3082 if (tmp->offset + tmp->bytes < offset)
3084 if (offset + bytes < tmp->offset) {
3085 n = rb_prev(&info->offset_index);
3092 n = rb_next(&info->offset_index);
3094 tmp = rb_entry(n, struct btrfs_free_space,
3096 if (offset + bytes < tmp->offset)
3098 if (tmp->offset + tmp->bytes < offset) {
3099 n = rb_next(&info->offset_index);
3109 if (info->offset == offset) {
3114 if (offset > info->offset && offset < info->offset + info->bytes)
3117 spin_unlock(&ctl->tree_lock);
3120 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */