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);
37 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
38 struct btrfs_path *path,
42 struct btrfs_key location;
43 struct btrfs_disk_key disk_key;
44 struct btrfs_free_space_header *header;
45 struct extent_buffer *leaf;
46 struct inode *inode = NULL;
49 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
53 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
57 btrfs_release_path(path);
58 return ERR_PTR(-ENOENT);
61 leaf = path->nodes[0];
62 header = btrfs_item_ptr(leaf, path->slots[0],
63 struct btrfs_free_space_header);
64 btrfs_free_space_key(leaf, header, &disk_key);
65 btrfs_disk_key_to_cpu(&location, &disk_key);
66 btrfs_release_path(path);
68 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
70 return ERR_PTR(-ENOENT);
73 if (is_bad_inode(inode)) {
75 return ERR_PTR(-ENOENT);
78 inode->i_mapping->flags &= ~__GFP_FS;
83 struct inode *lookup_free_space_inode(struct btrfs_root *root,
84 struct btrfs_block_group_cache
85 *block_group, struct btrfs_path *path)
87 struct inode *inode = NULL;
88 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
90 spin_lock(&block_group->lock);
91 if (block_group->inode)
92 inode = igrab(block_group->inode);
93 spin_unlock(&block_group->lock);
97 inode = __lookup_free_space_inode(root, path,
98 block_group->key.objectid);
102 spin_lock(&block_group->lock);
103 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
104 printk(KERN_INFO "Old style space inode found, converting.\n");
105 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
106 BTRFS_INODE_NODATACOW;
107 block_group->disk_cache_state = BTRFS_DC_CLEAR;
110 if (!block_group->iref) {
111 block_group->inode = igrab(inode);
112 block_group->iref = 1;
114 spin_unlock(&block_group->lock);
119 int __create_free_space_inode(struct btrfs_root *root,
120 struct btrfs_trans_handle *trans,
121 struct btrfs_path *path, u64 ino, u64 offset)
123 struct btrfs_key key;
124 struct btrfs_disk_key disk_key;
125 struct btrfs_free_space_header *header;
126 struct btrfs_inode_item *inode_item;
127 struct extent_buffer *leaf;
128 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
131 ret = btrfs_insert_empty_inode(trans, root, path, ino);
135 /* We inline crc's for the free disk space cache */
136 if (ino != BTRFS_FREE_INO_OBJECTID)
137 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
139 leaf = path->nodes[0];
140 inode_item = btrfs_item_ptr(leaf, path->slots[0],
141 struct btrfs_inode_item);
142 btrfs_item_key(leaf, &disk_key, path->slots[0]);
143 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
144 sizeof(*inode_item));
145 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
146 btrfs_set_inode_size(leaf, inode_item, 0);
147 btrfs_set_inode_nbytes(leaf, inode_item, 0);
148 btrfs_set_inode_uid(leaf, inode_item, 0);
149 btrfs_set_inode_gid(leaf, inode_item, 0);
150 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
151 btrfs_set_inode_flags(leaf, inode_item, flags);
152 btrfs_set_inode_nlink(leaf, inode_item, 1);
153 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
154 btrfs_set_inode_block_group(leaf, inode_item, offset);
155 btrfs_mark_buffer_dirty(leaf);
156 btrfs_release_path(path);
158 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
163 sizeof(struct btrfs_free_space_header));
165 btrfs_release_path(path);
168 leaf = path->nodes[0];
169 header = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_free_space_header);
171 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
172 btrfs_set_free_space_key(leaf, header, &disk_key);
173 btrfs_mark_buffer_dirty(leaf);
174 btrfs_release_path(path);
179 int create_free_space_inode(struct btrfs_root *root,
180 struct btrfs_trans_handle *trans,
181 struct btrfs_block_group_cache *block_group,
182 struct btrfs_path *path)
187 ret = btrfs_find_free_objectid(root, &ino);
191 return __create_free_space_inode(root, trans, path, ino,
192 block_group->key.objectid);
195 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
196 struct btrfs_trans_handle *trans,
197 struct btrfs_path *path,
200 struct btrfs_block_rsv *rsv;
205 rsv = trans->block_rsv;
206 trans->block_rsv = &root->fs_info->global_block_rsv;
208 /* 1 for slack space, 1 for updating the inode */
209 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
210 btrfs_calc_trans_metadata_size(root, 1);
212 spin_lock(&trans->block_rsv->lock);
213 if (trans->block_rsv->reserved < needed_bytes) {
214 spin_unlock(&trans->block_rsv->lock);
215 trans->block_rsv = rsv;
218 spin_unlock(&trans->block_rsv->lock);
220 oldsize = i_size_read(inode);
221 btrfs_i_size_write(inode, 0);
222 truncate_pagecache(inode, oldsize, 0);
225 * We don't need an orphan item because truncating the free space cache
226 * will never be split across transactions.
228 ret = btrfs_truncate_inode_items(trans, root, inode,
229 0, BTRFS_EXTENT_DATA_KEY);
232 trans->block_rsv = rsv;
237 ret = btrfs_update_inode(trans, root, inode);
238 trans->block_rsv = rsv;
243 static int readahead_cache(struct inode *inode)
245 struct file_ra_state *ra;
246 unsigned long last_index;
248 ra = kzalloc(sizeof(*ra), GFP_NOFS);
252 file_ra_state_init(ra, inode->i_mapping);
253 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
255 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
266 struct btrfs_root *root;
270 unsigned check_crcs:1;
273 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
274 struct btrfs_root *root)
276 memset(io_ctl, 0, sizeof(struct io_ctl));
277 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
279 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
284 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
285 io_ctl->check_crcs = 1;
289 static void io_ctl_free(struct io_ctl *io_ctl)
291 kfree(io_ctl->pages);
294 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
297 kunmap(io_ctl->page);
303 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
305 WARN_ON(io_ctl->cur);
306 BUG_ON(io_ctl->index >= io_ctl->num_pages);
307 io_ctl->page = io_ctl->pages[io_ctl->index++];
308 io_ctl->cur = kmap(io_ctl->page);
309 io_ctl->orig = io_ctl->cur;
310 io_ctl->size = PAGE_CACHE_SIZE;
312 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
315 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
319 io_ctl_unmap_page(io_ctl);
321 for (i = 0; i < io_ctl->num_pages; i++) {
322 ClearPageChecked(io_ctl->pages[i]);
323 unlock_page(io_ctl->pages[i]);
324 page_cache_release(io_ctl->pages[i]);
328 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
332 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
335 for (i = 0; i < io_ctl->num_pages; i++) {
336 page = find_or_create_page(inode->i_mapping, i, mask);
338 io_ctl_drop_pages(io_ctl);
341 io_ctl->pages[i] = page;
342 if (uptodate && !PageUptodate(page)) {
343 btrfs_readpage(NULL, page);
345 if (!PageUptodate(page)) {
346 printk(KERN_ERR "btrfs: error reading free "
348 io_ctl_drop_pages(io_ctl);
354 for (i = 0; i < io_ctl->num_pages; i++) {
355 clear_page_dirty_for_io(io_ctl->pages[i]);
356 set_page_extent_mapped(io_ctl->pages[i]);
362 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
366 io_ctl_map_page(io_ctl, 1);
369 * Skip the csum areas. If we don't check crcs then we just have a
370 * 64bit chunk at the front of the first page.
372 if (io_ctl->check_crcs) {
373 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
374 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
376 io_ctl->cur += sizeof(u64);
377 io_ctl->size -= sizeof(u64) * 2;
381 *val = cpu_to_le64(generation);
382 io_ctl->cur += sizeof(u64);
385 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
390 * Skip the crc area. If we don't check crcs then we just have a 64bit
391 * chunk at the front of the first page.
393 if (io_ctl->check_crcs) {
394 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
395 io_ctl->size -= sizeof(u64) +
396 (sizeof(u32) * io_ctl->num_pages);
398 io_ctl->cur += sizeof(u64);
399 io_ctl->size -= sizeof(u64) * 2;
403 if (le64_to_cpu(*gen) != generation) {
404 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
405 "(%Lu) does not match inode (%Lu)\n", *gen,
407 io_ctl_unmap_page(io_ctl);
410 io_ctl->cur += sizeof(u64);
414 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
420 if (!io_ctl->check_crcs) {
421 io_ctl_unmap_page(io_ctl);
426 offset = sizeof(u32) * io_ctl->num_pages;;
428 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
429 PAGE_CACHE_SIZE - offset);
430 btrfs_csum_final(crc, (char *)&crc);
431 io_ctl_unmap_page(io_ctl);
432 tmp = kmap(io_ctl->pages[0]);
435 kunmap(io_ctl->pages[0]);
438 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
444 if (!io_ctl->check_crcs) {
445 io_ctl_map_page(io_ctl, 0);
450 offset = sizeof(u32) * io_ctl->num_pages;
452 tmp = kmap(io_ctl->pages[0]);
455 kunmap(io_ctl->pages[0]);
457 io_ctl_map_page(io_ctl, 0);
458 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
459 PAGE_CACHE_SIZE - offset);
460 btrfs_csum_final(crc, (char *)&crc);
462 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
464 io_ctl_unmap_page(io_ctl);
471 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
474 struct btrfs_free_space_entry *entry;
480 entry->offset = cpu_to_le64(offset);
481 entry->bytes = cpu_to_le64(bytes);
482 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
483 BTRFS_FREE_SPACE_EXTENT;
484 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
485 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
487 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
490 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
492 /* No more pages to map */
493 if (io_ctl->index >= io_ctl->num_pages)
496 /* map the next page */
497 io_ctl_map_page(io_ctl, 1);
501 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
507 * If we aren't at the start of the current page, unmap this one and
508 * map the next one if there is any left.
510 if (io_ctl->cur != io_ctl->orig) {
511 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
512 if (io_ctl->index >= io_ctl->num_pages)
514 io_ctl_map_page(io_ctl, 0);
517 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
518 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519 if (io_ctl->index < io_ctl->num_pages)
520 io_ctl_map_page(io_ctl, 0);
524 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
527 * If we're not on the boundary we know we've modified the page and we
528 * need to crc the page.
530 if (io_ctl->cur != io_ctl->orig)
531 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
533 io_ctl_unmap_page(io_ctl);
535 while (io_ctl->index < io_ctl->num_pages) {
536 io_ctl_map_page(io_ctl, 1);
537 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
541 static int io_ctl_read_entry(struct io_ctl *io_ctl,
542 struct btrfs_free_space *entry, u8 *type)
544 struct btrfs_free_space_entry *e;
548 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
554 entry->offset = le64_to_cpu(e->offset);
555 entry->bytes = le64_to_cpu(e->bytes);
557 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
558 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
560 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
563 io_ctl_unmap_page(io_ctl);
568 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
569 struct btrfs_free_space *entry)
573 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
577 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
578 io_ctl_unmap_page(io_ctl);
583 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
584 struct btrfs_free_space_ctl *ctl,
585 struct btrfs_path *path, u64 offset)
587 struct btrfs_free_space_header *header;
588 struct extent_buffer *leaf;
589 struct io_ctl io_ctl;
590 struct btrfs_key key;
591 struct btrfs_free_space *e, *n;
592 struct list_head bitmaps;
599 INIT_LIST_HEAD(&bitmaps);
601 /* Nothing in the space cache, goodbye */
602 if (!i_size_read(inode))
605 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
609 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
613 btrfs_release_path(path);
619 leaf = path->nodes[0];
620 header = btrfs_item_ptr(leaf, path->slots[0],
621 struct btrfs_free_space_header);
622 num_entries = btrfs_free_space_entries(leaf, header);
623 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
624 generation = btrfs_free_space_generation(leaf, header);
625 btrfs_release_path(path);
627 if (BTRFS_I(inode)->generation != generation) {
628 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
629 " not match free space cache generation (%llu)\n",
630 (unsigned long long)BTRFS_I(inode)->generation,
631 (unsigned long long)generation);
638 io_ctl_init(&io_ctl, inode, root);
639 ret = readahead_cache(inode);
643 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
647 ret = io_ctl_check_crc(&io_ctl, 0);
651 ret = io_ctl_check_generation(&io_ctl, generation);
655 while (num_entries) {
656 e = kmem_cache_zalloc(btrfs_free_space_cachep,
661 ret = io_ctl_read_entry(&io_ctl, e, &type);
663 kmem_cache_free(btrfs_free_space_cachep, e);
668 kmem_cache_free(btrfs_free_space_cachep, e);
672 if (type == BTRFS_FREE_SPACE_EXTENT) {
673 spin_lock(&ctl->tree_lock);
674 ret = link_free_space(ctl, e);
675 spin_unlock(&ctl->tree_lock);
677 printk(KERN_ERR "Duplicate entries in "
678 "free space cache, dumping\n");
679 kmem_cache_free(btrfs_free_space_cachep, e);
683 BUG_ON(!num_bitmaps);
685 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
688 btrfs_free_space_cachep, e);
691 spin_lock(&ctl->tree_lock);
692 ret = link_free_space(ctl, e);
693 ctl->total_bitmaps++;
694 ctl->op->recalc_thresholds(ctl);
695 spin_unlock(&ctl->tree_lock);
697 printk(KERN_ERR "Duplicate entries in "
698 "free space cache, dumping\n");
699 kmem_cache_free(btrfs_free_space_cachep, e);
702 list_add_tail(&e->list, &bitmaps);
708 io_ctl_unmap_page(&io_ctl);
711 * We add the bitmaps at the end of the entries in order that
712 * the bitmap entries are added to the cache.
714 list_for_each_entry_safe(e, n, &bitmaps, list) {
715 list_del_init(&e->list);
716 ret = io_ctl_read_bitmap(&io_ctl, e);
721 io_ctl_drop_pages(&io_ctl);
724 io_ctl_free(&io_ctl);
727 io_ctl_drop_pages(&io_ctl);
728 __btrfs_remove_free_space_cache(ctl);
732 int load_free_space_cache(struct btrfs_fs_info *fs_info,
733 struct btrfs_block_group_cache *block_group)
735 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
736 struct btrfs_root *root = fs_info->tree_root;
738 struct btrfs_path *path;
741 u64 used = btrfs_block_group_used(&block_group->item);
744 * If we're unmounting then just return, since this does a search on the
745 * normal root and not the commit root and we could deadlock.
747 if (btrfs_fs_closing(fs_info))
751 * If this block group has been marked to be cleared for one reason or
752 * another then we can't trust the on disk cache, so just return.
754 spin_lock(&block_group->lock);
755 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
756 spin_unlock(&block_group->lock);
759 spin_unlock(&block_group->lock);
761 path = btrfs_alloc_path();
765 inode = lookup_free_space_inode(root, block_group, path);
767 btrfs_free_path(path);
771 /* We may have converted the inode and made the cache invalid. */
772 spin_lock(&block_group->lock);
773 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
774 spin_unlock(&block_group->lock);
777 spin_unlock(&block_group->lock);
779 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
780 path, block_group->key.objectid);
781 btrfs_free_path(path);
785 spin_lock(&ctl->tree_lock);
786 matched = (ctl->free_space == (block_group->key.offset - used -
787 block_group->bytes_super));
788 spin_unlock(&ctl->tree_lock);
791 __btrfs_remove_free_space_cache(ctl);
792 printk(KERN_ERR "block group %llu has an wrong amount of free "
793 "space\n", block_group->key.objectid);
798 /* This cache is bogus, make sure it gets cleared */
799 spin_lock(&block_group->lock);
800 block_group->disk_cache_state = BTRFS_DC_CLEAR;
801 spin_unlock(&block_group->lock);
804 printk(KERN_ERR "btrfs: failed to load free space cache "
805 "for block group %llu\n", block_group->key.objectid);
813 * __btrfs_write_out_cache - write out cached info to an inode
814 * @root - the root the inode belongs to
815 * @ctl - the free space cache we are going to write out
816 * @block_group - the block_group for this cache if it belongs to a block_group
817 * @trans - the trans handle
818 * @path - the path to use
819 * @offset - the offset for the key we'll insert
821 * This function writes out a free space cache struct to disk for quick recovery
822 * on mount. This will return 0 if it was successfull in writing the cache out,
823 * and -1 if it was not.
825 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
826 struct btrfs_free_space_ctl *ctl,
827 struct btrfs_block_group_cache *block_group,
828 struct btrfs_trans_handle *trans,
829 struct btrfs_path *path, u64 offset)
831 struct btrfs_free_space_header *header;
832 struct extent_buffer *leaf;
833 struct rb_node *node;
834 struct list_head *pos, *n;
835 struct extent_state *cached_state = NULL;
836 struct btrfs_free_cluster *cluster = NULL;
837 struct extent_io_tree *unpin = NULL;
838 struct io_ctl io_ctl;
839 struct list_head bitmap_list;
840 struct btrfs_key key;
841 u64 start, extent_start, extent_end, len;
847 INIT_LIST_HEAD(&bitmap_list);
849 if (!i_size_read(inode))
852 io_ctl_init(&io_ctl, inode, root);
854 /* Get the cluster for this block_group if it exists */
855 if (block_group && !list_empty(&block_group->cluster_list))
856 cluster = list_entry(block_group->cluster_list.next,
857 struct btrfs_free_cluster,
860 /* Lock all pages first so we can lock the extent safely. */
861 io_ctl_prepare_pages(&io_ctl, inode, 0);
863 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
864 0, &cached_state, GFP_NOFS);
866 node = rb_first(&ctl->free_space_offset);
867 if (!node && cluster) {
868 node = rb_first(&cluster->root);
872 /* Make sure we can fit our crcs into the first page */
873 if (io_ctl.check_crcs &&
874 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
879 io_ctl_set_generation(&io_ctl, trans->transid);
881 /* Write out the extent entries */
883 struct btrfs_free_space *e;
885 e = rb_entry(node, struct btrfs_free_space, offset_index);
888 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
894 list_add_tail(&e->list, &bitmap_list);
897 node = rb_next(node);
898 if (!node && cluster) {
899 node = rb_first(&cluster->root);
905 * We want to add any pinned extents to our free space cache
906 * so we don't leak the space
910 * We shouldn't have switched the pinned extents yet so this is the
913 unpin = root->fs_info->pinned_extents;
916 start = block_group->key.objectid;
918 while (block_group && (start < block_group->key.objectid +
919 block_group->key.offset)) {
920 ret = find_first_extent_bit(unpin, start,
921 &extent_start, &extent_end,
928 /* This pinned extent is out of our range */
929 if (extent_start >= block_group->key.objectid +
930 block_group->key.offset)
933 extent_start = max(extent_start, start);
934 extent_end = min(block_group->key.objectid +
935 block_group->key.offset, extent_end + 1);
936 len = extent_end - extent_start;
939 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
946 /* Write out the bitmaps */
947 list_for_each_safe(pos, n, &bitmap_list) {
948 struct btrfs_free_space *entry =
949 list_entry(pos, struct btrfs_free_space, list);
951 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
954 list_del_init(&entry->list);
957 /* Zero out the rest of the pages just to make sure */
958 io_ctl_zero_remaining_pages(&io_ctl);
960 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
961 0, i_size_read(inode), &cached_state);
962 io_ctl_drop_pages(&io_ctl);
963 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
964 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
970 ret = filemap_write_and_wait(inode->i_mapping);
974 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
978 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
980 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
981 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
985 leaf = path->nodes[0];
987 struct btrfs_key found_key;
988 BUG_ON(!path->slots[0]);
990 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
991 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
992 found_key.offset != offset) {
993 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
995 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
997 btrfs_release_path(path);
1002 BTRFS_I(inode)->generation = trans->transid;
1003 header = btrfs_item_ptr(leaf, path->slots[0],
1004 struct btrfs_free_space_header);
1005 btrfs_set_free_space_entries(leaf, header, entries);
1006 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1007 btrfs_set_free_space_generation(leaf, header, trans->transid);
1008 btrfs_mark_buffer_dirty(leaf);
1009 btrfs_release_path(path);
1013 io_ctl_free(&io_ctl);
1015 invalidate_inode_pages2(inode->i_mapping);
1016 BTRFS_I(inode)->generation = 0;
1018 btrfs_update_inode(trans, root, inode);
1022 list_for_each_safe(pos, n, &bitmap_list) {
1023 struct btrfs_free_space *entry =
1024 list_entry(pos, struct btrfs_free_space, list);
1025 list_del_init(&entry->list);
1027 io_ctl_drop_pages(&io_ctl);
1028 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1029 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1033 int btrfs_write_out_cache(struct btrfs_root *root,
1034 struct btrfs_trans_handle *trans,
1035 struct btrfs_block_group_cache *block_group,
1036 struct btrfs_path *path)
1038 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1039 struct inode *inode;
1042 root = root->fs_info->tree_root;
1044 spin_lock(&block_group->lock);
1045 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1046 spin_unlock(&block_group->lock);
1049 spin_unlock(&block_group->lock);
1051 inode = lookup_free_space_inode(root, block_group, path);
1055 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1056 path, block_group->key.objectid);
1058 spin_lock(&block_group->lock);
1059 block_group->disk_cache_state = BTRFS_DC_ERROR;
1060 spin_unlock(&block_group->lock);
1063 printk(KERN_ERR "btrfs: failed to write free space cace "
1064 "for block group %llu\n", block_group->key.objectid);
1072 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1075 BUG_ON(offset < bitmap_start);
1076 offset -= bitmap_start;
1077 return (unsigned long)(div_u64(offset, unit));
1080 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1082 return (unsigned long)(div_u64(bytes, unit));
1085 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1089 u64 bytes_per_bitmap;
1091 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1092 bitmap_start = offset - ctl->start;
1093 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1094 bitmap_start *= bytes_per_bitmap;
1095 bitmap_start += ctl->start;
1097 return bitmap_start;
1100 static int tree_insert_offset(struct rb_root *root, u64 offset,
1101 struct rb_node *node, int bitmap)
1103 struct rb_node **p = &root->rb_node;
1104 struct rb_node *parent = NULL;
1105 struct btrfs_free_space *info;
1109 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1111 if (offset < info->offset) {
1113 } else if (offset > info->offset) {
1114 p = &(*p)->rb_right;
1117 * we could have a bitmap entry and an extent entry
1118 * share the same offset. If this is the case, we want
1119 * the extent entry to always be found first if we do a
1120 * linear search through the tree, since we want to have
1121 * the quickest allocation time, and allocating from an
1122 * extent is faster than allocating from a bitmap. So
1123 * if we're inserting a bitmap and we find an entry at
1124 * this offset, we want to go right, or after this entry
1125 * logically. If we are inserting an extent and we've
1126 * found a bitmap, we want to go left, or before
1134 p = &(*p)->rb_right;
1136 if (!info->bitmap) {
1145 rb_link_node(node, parent, p);
1146 rb_insert_color(node, root);
1152 * searches the tree for the given offset.
1154 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1155 * want a section that has at least bytes size and comes at or after the given
1158 static struct btrfs_free_space *
1159 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1160 u64 offset, int bitmap_only, int fuzzy)
1162 struct rb_node *n = ctl->free_space_offset.rb_node;
1163 struct btrfs_free_space *entry, *prev = NULL;
1165 /* find entry that is closest to the 'offset' */
1172 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1175 if (offset < entry->offset)
1177 else if (offset > entry->offset)
1190 * bitmap entry and extent entry may share same offset,
1191 * in that case, bitmap entry comes after extent entry.
1196 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1197 if (entry->offset != offset)
1200 WARN_ON(!entry->bitmap);
1203 if (entry->bitmap) {
1205 * if previous extent entry covers the offset,
1206 * we should return it instead of the bitmap entry
1208 n = &entry->offset_index;
1213 prev = rb_entry(n, struct btrfs_free_space,
1215 if (!prev->bitmap) {
1216 if (prev->offset + prev->bytes > offset)
1228 /* find last entry before the 'offset' */
1230 if (entry->offset > offset) {
1231 n = rb_prev(&entry->offset_index);
1233 entry = rb_entry(n, struct btrfs_free_space,
1235 BUG_ON(entry->offset > offset);
1244 if (entry->bitmap) {
1245 n = &entry->offset_index;
1250 prev = rb_entry(n, struct btrfs_free_space,
1252 if (!prev->bitmap) {
1253 if (prev->offset + prev->bytes > offset)
1258 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1260 } else if (entry->offset + entry->bytes > offset)
1267 if (entry->bitmap) {
1268 if (entry->offset + BITS_PER_BITMAP *
1272 if (entry->offset + entry->bytes > offset)
1276 n = rb_next(&entry->offset_index);
1279 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1285 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1286 struct btrfs_free_space *info)
1288 rb_erase(&info->offset_index, &ctl->free_space_offset);
1289 ctl->free_extents--;
1292 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1293 struct btrfs_free_space *info)
1295 __unlink_free_space(ctl, info);
1296 ctl->free_space -= info->bytes;
1299 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1300 struct btrfs_free_space *info)
1304 BUG_ON(!info->bitmap && !info->bytes);
1305 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1306 &info->offset_index, (info->bitmap != NULL));
1310 ctl->free_space += info->bytes;
1311 ctl->free_extents++;
1315 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1317 struct btrfs_block_group_cache *block_group = ctl->private;
1321 u64 size = block_group->key.offset;
1322 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1323 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1325 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1328 * The goal is to keep the total amount of memory used per 1gb of space
1329 * at or below 32k, so we need to adjust how much memory we allow to be
1330 * used by extent based free space tracking
1332 if (size < 1024 * 1024 * 1024)
1333 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1335 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1336 div64_u64(size, 1024 * 1024 * 1024);
1339 * we want to account for 1 more bitmap than what we have so we can make
1340 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1341 * we add more bitmaps.
1343 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1345 if (bitmap_bytes >= max_bytes) {
1346 ctl->extents_thresh = 0;
1351 * we want the extent entry threshold to always be at most 1/2 the maxw
1352 * bytes we can have, or whatever is less than that.
1354 extent_bytes = max_bytes - bitmap_bytes;
1355 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1357 ctl->extents_thresh =
1358 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1361 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1362 struct btrfs_free_space *info,
1363 u64 offset, u64 bytes)
1365 unsigned long start, count;
1367 start = offset_to_bit(info->offset, ctl->unit, offset);
1368 count = bytes_to_bits(bytes, ctl->unit);
1369 BUG_ON(start + count > BITS_PER_BITMAP);
1371 bitmap_clear(info->bitmap, start, count);
1373 info->bytes -= bytes;
1376 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1377 struct btrfs_free_space *info, u64 offset,
1380 __bitmap_clear_bits(ctl, info, offset, bytes);
1381 ctl->free_space -= bytes;
1384 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1385 struct btrfs_free_space *info, u64 offset,
1388 unsigned long start, count;
1390 start = offset_to_bit(info->offset, ctl->unit, offset);
1391 count = bytes_to_bits(bytes, ctl->unit);
1392 BUG_ON(start + count > BITS_PER_BITMAP);
1394 bitmap_set(info->bitmap, start, count);
1396 info->bytes += bytes;
1397 ctl->free_space += bytes;
1400 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1401 struct btrfs_free_space *bitmap_info, u64 *offset,
1404 unsigned long found_bits = 0;
1405 unsigned long bits, i;
1406 unsigned long next_zero;
1408 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1409 max_t(u64, *offset, bitmap_info->offset));
1410 bits = bytes_to_bits(*bytes, ctl->unit);
1412 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1413 i < BITS_PER_BITMAP;
1414 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1415 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1416 BITS_PER_BITMAP, i);
1417 if ((next_zero - i) >= bits) {
1418 found_bits = next_zero - i;
1425 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1426 *bytes = (u64)(found_bits) * ctl->unit;
1433 static struct btrfs_free_space *
1434 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1436 struct btrfs_free_space *entry;
1437 struct rb_node *node;
1440 if (!ctl->free_space_offset.rb_node)
1443 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1447 for (node = &entry->offset_index; node; node = rb_next(node)) {
1448 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1449 if (entry->bytes < *bytes)
1452 if (entry->bitmap) {
1453 ret = search_bitmap(ctl, entry, offset, bytes);
1459 *offset = entry->offset;
1460 *bytes = entry->bytes;
1467 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1468 struct btrfs_free_space *info, u64 offset)
1470 info->offset = offset_to_bitmap(ctl, offset);
1472 INIT_LIST_HEAD(&info->list);
1473 link_free_space(ctl, info);
1474 ctl->total_bitmaps++;
1476 ctl->op->recalc_thresholds(ctl);
1479 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1480 struct btrfs_free_space *bitmap_info)
1482 unlink_free_space(ctl, bitmap_info);
1483 kfree(bitmap_info->bitmap);
1484 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1485 ctl->total_bitmaps--;
1486 ctl->op->recalc_thresholds(ctl);
1489 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1490 struct btrfs_free_space *bitmap_info,
1491 u64 *offset, u64 *bytes)
1494 u64 search_start, search_bytes;
1498 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1501 * XXX - this can go away after a few releases.
1503 * since the only user of btrfs_remove_free_space is the tree logging
1504 * stuff, and the only way to test that is under crash conditions, we
1505 * want to have this debug stuff here just in case somethings not
1506 * working. Search the bitmap for the space we are trying to use to
1507 * make sure its actually there. If its not there then we need to stop
1508 * because something has gone wrong.
1510 search_start = *offset;
1511 search_bytes = *bytes;
1512 search_bytes = min(search_bytes, end - search_start + 1);
1513 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1514 BUG_ON(ret < 0 || search_start != *offset);
1516 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1517 bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
1518 *bytes -= end - *offset + 1;
1520 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1521 bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
1526 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1527 if (!bitmap_info->bytes)
1528 free_bitmap(ctl, bitmap_info);
1531 * no entry after this bitmap, but we still have bytes to
1532 * remove, so something has gone wrong.
1537 bitmap_info = rb_entry(next, struct btrfs_free_space,
1541 * if the next entry isn't a bitmap we need to return to let the
1542 * extent stuff do its work.
1544 if (!bitmap_info->bitmap)
1548 * Ok the next item is a bitmap, but it may not actually hold
1549 * the information for the rest of this free space stuff, so
1550 * look for it, and if we don't find it return so we can try
1551 * everything over again.
1553 search_start = *offset;
1554 search_bytes = *bytes;
1555 ret = search_bitmap(ctl, bitmap_info, &search_start,
1557 if (ret < 0 || search_start != *offset)
1561 } else if (!bitmap_info->bytes)
1562 free_bitmap(ctl, bitmap_info);
1567 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1568 struct btrfs_free_space *info, u64 offset,
1571 u64 bytes_to_set = 0;
1574 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1576 bytes_to_set = min(end - offset, bytes);
1578 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1580 return bytes_to_set;
1584 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1585 struct btrfs_free_space *info)
1587 struct btrfs_block_group_cache *block_group = ctl->private;
1590 * If we are below the extents threshold then we can add this as an
1591 * extent, and don't have to deal with the bitmap
1593 if (ctl->free_extents < ctl->extents_thresh) {
1595 * If this block group has some small extents we don't want to
1596 * use up all of our free slots in the cache with them, we want
1597 * to reserve them to larger extents, however if we have plent
1598 * of cache left then go ahead an dadd them, no sense in adding
1599 * the overhead of a bitmap if we don't have to.
1601 if (info->bytes <= block_group->sectorsize * 4) {
1602 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1610 * some block groups are so tiny they can't be enveloped by a bitmap, so
1611 * don't even bother to create a bitmap for this
1613 if (BITS_PER_BITMAP * block_group->sectorsize >
1614 block_group->key.offset)
1620 static struct btrfs_free_space_op free_space_op = {
1621 .recalc_thresholds = recalculate_thresholds,
1622 .use_bitmap = use_bitmap,
1625 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1626 struct btrfs_free_space *info)
1628 struct btrfs_free_space *bitmap_info;
1629 struct btrfs_block_group_cache *block_group = NULL;
1631 u64 bytes, offset, bytes_added;
1634 bytes = info->bytes;
1635 offset = info->offset;
1637 if (!ctl->op->use_bitmap(ctl, info))
1640 if (ctl->op == &free_space_op)
1641 block_group = ctl->private;
1644 * Since we link bitmaps right into the cluster we need to see if we
1645 * have a cluster here, and if so and it has our bitmap we need to add
1646 * the free space to that bitmap.
1648 if (block_group && !list_empty(&block_group->cluster_list)) {
1649 struct btrfs_free_cluster *cluster;
1650 struct rb_node *node;
1651 struct btrfs_free_space *entry;
1653 cluster = list_entry(block_group->cluster_list.next,
1654 struct btrfs_free_cluster,
1656 spin_lock(&cluster->lock);
1657 node = rb_first(&cluster->root);
1659 spin_unlock(&cluster->lock);
1660 goto no_cluster_bitmap;
1663 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1664 if (!entry->bitmap) {
1665 spin_unlock(&cluster->lock);
1666 goto no_cluster_bitmap;
1669 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1670 bytes_added = add_bytes_to_bitmap(ctl, entry,
1672 bytes -= bytes_added;
1673 offset += bytes_added;
1675 spin_unlock(&cluster->lock);
1683 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1690 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1691 bytes -= bytes_added;
1692 offset += bytes_added;
1702 if (info && info->bitmap) {
1703 add_new_bitmap(ctl, info, offset);
1708 spin_unlock(&ctl->tree_lock);
1710 /* no pre-allocated info, allocate a new one */
1712 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1715 spin_lock(&ctl->tree_lock);
1721 /* allocate the bitmap */
1722 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1723 spin_lock(&ctl->tree_lock);
1724 if (!info->bitmap) {
1734 kfree(info->bitmap);
1735 kmem_cache_free(btrfs_free_space_cachep, info);
1741 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1742 struct btrfs_free_space *info, bool update_stat)
1744 struct btrfs_free_space *left_info;
1745 struct btrfs_free_space *right_info;
1746 bool merged = false;
1747 u64 offset = info->offset;
1748 u64 bytes = info->bytes;
1751 * first we want to see if there is free space adjacent to the range we
1752 * are adding, if there is remove that struct and add a new one to
1753 * cover the entire range
1755 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1756 if (right_info && rb_prev(&right_info->offset_index))
1757 left_info = rb_entry(rb_prev(&right_info->offset_index),
1758 struct btrfs_free_space, offset_index);
1760 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1762 if (right_info && !right_info->bitmap) {
1764 unlink_free_space(ctl, right_info);
1766 __unlink_free_space(ctl, right_info);
1767 info->bytes += right_info->bytes;
1768 kmem_cache_free(btrfs_free_space_cachep, right_info);
1772 if (left_info && !left_info->bitmap &&
1773 left_info->offset + left_info->bytes == offset) {
1775 unlink_free_space(ctl, left_info);
1777 __unlink_free_space(ctl, left_info);
1778 info->offset = left_info->offset;
1779 info->bytes += left_info->bytes;
1780 kmem_cache_free(btrfs_free_space_cachep, left_info);
1787 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1788 u64 offset, u64 bytes)
1790 struct btrfs_free_space *info;
1793 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1797 info->offset = offset;
1798 info->bytes = bytes;
1800 spin_lock(&ctl->tree_lock);
1802 if (try_merge_free_space(ctl, info, true))
1806 * There was no extent directly to the left or right of this new
1807 * extent then we know we're going to have to allocate a new extent, so
1808 * before we do that see if we need to drop this into a bitmap
1810 ret = insert_into_bitmap(ctl, info);
1818 ret = link_free_space(ctl, info);
1820 kmem_cache_free(btrfs_free_space_cachep, info);
1822 spin_unlock(&ctl->tree_lock);
1825 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1826 BUG_ON(ret == -EEXIST);
1832 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1833 u64 offset, u64 bytes)
1835 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1836 struct btrfs_free_space *info;
1837 struct btrfs_free_space *next_info = NULL;
1840 spin_lock(&ctl->tree_lock);
1843 info = tree_search_offset(ctl, offset, 0, 0);
1846 * oops didn't find an extent that matched the space we wanted
1847 * to remove, look for a bitmap instead
1849 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1852 /* the tree logging code might be calling us before we
1853 * have fully loaded the free space rbtree for this
1854 * block group. So it is possible the entry won't
1855 * be in the rbtree yet at all. The caching code
1856 * will make sure not to put it in the rbtree if
1857 * the logging code has pinned it.
1863 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1865 next_info = rb_entry(rb_next(&info->offset_index),
1866 struct btrfs_free_space,
1869 if (next_info->bitmap)
1870 end = next_info->offset +
1871 BITS_PER_BITMAP * ctl->unit - 1;
1873 end = next_info->offset + next_info->bytes;
1875 if (next_info->bytes < bytes ||
1876 next_info->offset > offset || offset > end) {
1877 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1878 " trying to use %llu\n",
1879 (unsigned long long)info->offset,
1880 (unsigned long long)info->bytes,
1881 (unsigned long long)bytes);
1890 if (info->bytes == bytes) {
1891 unlink_free_space(ctl, info);
1893 kfree(info->bitmap);
1894 ctl->total_bitmaps--;
1896 kmem_cache_free(btrfs_free_space_cachep, info);
1901 if (!info->bitmap && info->offset == offset) {
1902 unlink_free_space(ctl, info);
1903 info->offset += bytes;
1904 info->bytes -= bytes;
1905 ret = link_free_space(ctl, info);
1910 if (!info->bitmap && info->offset <= offset &&
1911 info->offset + info->bytes >= offset + bytes) {
1912 u64 old_start = info->offset;
1914 * we're freeing space in the middle of the info,
1915 * this can happen during tree log replay
1917 * first unlink the old info and then
1918 * insert it again after the hole we're creating
1920 unlink_free_space(ctl, info);
1921 if (offset + bytes < info->offset + info->bytes) {
1922 u64 old_end = info->offset + info->bytes;
1924 info->offset = offset + bytes;
1925 info->bytes = old_end - info->offset;
1926 ret = link_free_space(ctl, info);
1931 /* the hole we're creating ends at the end
1932 * of the info struct, just free the info
1934 kmem_cache_free(btrfs_free_space_cachep, info);
1936 spin_unlock(&ctl->tree_lock);
1938 /* step two, insert a new info struct to cover
1939 * anything before the hole
1941 ret = btrfs_add_free_space(block_group, old_start,
1942 offset - old_start);
1947 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1952 spin_unlock(&ctl->tree_lock);
1957 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1960 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1961 struct btrfs_free_space *info;
1965 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1966 info = rb_entry(n, struct btrfs_free_space, offset_index);
1967 if (info->bytes >= bytes)
1969 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1970 (unsigned long long)info->offset,
1971 (unsigned long long)info->bytes,
1972 (info->bitmap) ? "yes" : "no");
1974 printk(KERN_INFO "block group has cluster?: %s\n",
1975 list_empty(&block_group->cluster_list) ? "no" : "yes");
1976 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1980 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1982 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1984 spin_lock_init(&ctl->tree_lock);
1985 ctl->unit = block_group->sectorsize;
1986 ctl->start = block_group->key.objectid;
1987 ctl->private = block_group;
1988 ctl->op = &free_space_op;
1991 * we only want to have 32k of ram per block group for keeping
1992 * track of free space, and if we pass 1/2 of that we want to
1993 * start converting things over to using bitmaps
1995 ctl->extents_thresh = ((1024 * 32) / 2) /
1996 sizeof(struct btrfs_free_space);
2000 * for a given cluster, put all of its extents back into the free
2001 * space cache. If the block group passed doesn't match the block group
2002 * pointed to by the cluster, someone else raced in and freed the
2003 * cluster already. In that case, we just return without changing anything
2006 __btrfs_return_cluster_to_free_space(
2007 struct btrfs_block_group_cache *block_group,
2008 struct btrfs_free_cluster *cluster)
2010 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2011 struct btrfs_free_space *entry;
2012 struct rb_node *node;
2014 spin_lock(&cluster->lock);
2015 if (cluster->block_group != block_group)
2018 cluster->block_group = NULL;
2019 cluster->window_start = 0;
2020 list_del_init(&cluster->block_group_list);
2022 node = rb_first(&cluster->root);
2026 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2027 node = rb_next(&entry->offset_index);
2028 rb_erase(&entry->offset_index, &cluster->root);
2030 bitmap = (entry->bitmap != NULL);
2032 try_merge_free_space(ctl, entry, false);
2033 tree_insert_offset(&ctl->free_space_offset,
2034 entry->offset, &entry->offset_index, bitmap);
2036 cluster->root = RB_ROOT;
2039 spin_unlock(&cluster->lock);
2040 btrfs_put_block_group(block_group);
2044 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
2046 struct btrfs_free_space *info;
2047 struct rb_node *node;
2049 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2050 info = rb_entry(node, struct btrfs_free_space, offset_index);
2051 if (!info->bitmap) {
2052 unlink_free_space(ctl, info);
2053 kmem_cache_free(btrfs_free_space_cachep, info);
2055 free_bitmap(ctl, info);
2057 if (need_resched()) {
2058 spin_unlock(&ctl->tree_lock);
2060 spin_lock(&ctl->tree_lock);
2065 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2067 spin_lock(&ctl->tree_lock);
2068 __btrfs_remove_free_space_cache_locked(ctl);
2069 spin_unlock(&ctl->tree_lock);
2072 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2074 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2075 struct btrfs_free_cluster *cluster;
2076 struct list_head *head;
2078 spin_lock(&ctl->tree_lock);
2079 while ((head = block_group->cluster_list.next) !=
2080 &block_group->cluster_list) {
2081 cluster = list_entry(head, struct btrfs_free_cluster,
2084 WARN_ON(cluster->block_group != block_group);
2085 __btrfs_return_cluster_to_free_space(block_group, cluster);
2086 if (need_resched()) {
2087 spin_unlock(&ctl->tree_lock);
2089 spin_lock(&ctl->tree_lock);
2092 __btrfs_remove_free_space_cache_locked(ctl);
2093 spin_unlock(&ctl->tree_lock);
2097 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2098 u64 offset, u64 bytes, u64 empty_size)
2100 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2101 struct btrfs_free_space *entry = NULL;
2102 u64 bytes_search = bytes + empty_size;
2105 spin_lock(&ctl->tree_lock);
2106 entry = find_free_space(ctl, &offset, &bytes_search);
2111 if (entry->bitmap) {
2112 bitmap_clear_bits(ctl, entry, offset, bytes);
2114 free_bitmap(ctl, entry);
2116 unlink_free_space(ctl, entry);
2117 entry->offset += bytes;
2118 entry->bytes -= bytes;
2120 kmem_cache_free(btrfs_free_space_cachep, entry);
2122 link_free_space(ctl, entry);
2126 spin_unlock(&ctl->tree_lock);
2132 * given a cluster, put all of its extents back into the free space
2133 * cache. If a block group is passed, this function will only free
2134 * a cluster that belongs to the passed block group.
2136 * Otherwise, it'll get a reference on the block group pointed to by the
2137 * cluster and remove the cluster from it.
2139 int btrfs_return_cluster_to_free_space(
2140 struct btrfs_block_group_cache *block_group,
2141 struct btrfs_free_cluster *cluster)
2143 struct btrfs_free_space_ctl *ctl;
2146 /* first, get a safe pointer to the block group */
2147 spin_lock(&cluster->lock);
2149 block_group = cluster->block_group;
2151 spin_unlock(&cluster->lock);
2154 } else if (cluster->block_group != block_group) {
2155 /* someone else has already freed it don't redo their work */
2156 spin_unlock(&cluster->lock);
2159 atomic_inc(&block_group->count);
2160 spin_unlock(&cluster->lock);
2162 ctl = block_group->free_space_ctl;
2164 /* now return any extents the cluster had on it */
2165 spin_lock(&ctl->tree_lock);
2166 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2167 spin_unlock(&ctl->tree_lock);
2169 /* finally drop our ref */
2170 btrfs_put_block_group(block_group);
2174 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2175 struct btrfs_free_cluster *cluster,
2176 struct btrfs_free_space *entry,
2177 u64 bytes, u64 min_start)
2179 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2181 u64 search_start = cluster->window_start;
2182 u64 search_bytes = bytes;
2185 search_start = min_start;
2186 search_bytes = bytes;
2188 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2193 __bitmap_clear_bits(ctl, entry, ret, bytes);
2199 * given a cluster, try to allocate 'bytes' from it, returns 0
2200 * if it couldn't find anything suitably large, or a logical disk offset
2201 * if things worked out
2203 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2204 struct btrfs_free_cluster *cluster, u64 bytes,
2207 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2208 struct btrfs_free_space *entry = NULL;
2209 struct rb_node *node;
2212 spin_lock(&cluster->lock);
2213 if (bytes > cluster->max_size)
2216 if (cluster->block_group != block_group)
2219 node = rb_first(&cluster->root);
2223 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2225 if (entry->bytes < bytes ||
2226 (!entry->bitmap && entry->offset < min_start)) {
2227 node = rb_next(&entry->offset_index);
2230 entry = rb_entry(node, struct btrfs_free_space,
2235 if (entry->bitmap) {
2236 ret = btrfs_alloc_from_bitmap(block_group,
2237 cluster, entry, bytes,
2240 node = rb_next(&entry->offset_index);
2243 entry = rb_entry(node, struct btrfs_free_space,
2248 ret = entry->offset;
2250 entry->offset += bytes;
2251 entry->bytes -= bytes;
2254 if (entry->bytes == 0)
2255 rb_erase(&entry->offset_index, &cluster->root);
2259 spin_unlock(&cluster->lock);
2264 spin_lock(&ctl->tree_lock);
2266 ctl->free_space -= bytes;
2267 if (entry->bytes == 0) {
2268 ctl->free_extents--;
2269 if (entry->bitmap) {
2270 kfree(entry->bitmap);
2271 ctl->total_bitmaps--;
2272 ctl->op->recalc_thresholds(ctl);
2274 kmem_cache_free(btrfs_free_space_cachep, entry);
2277 spin_unlock(&ctl->tree_lock);
2282 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2283 struct btrfs_free_space *entry,
2284 struct btrfs_free_cluster *cluster,
2285 u64 offset, u64 bytes, u64 min_bytes)
2287 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2288 unsigned long next_zero;
2290 unsigned long search_bits;
2291 unsigned long total_bits;
2292 unsigned long found_bits;
2293 unsigned long start = 0;
2294 unsigned long total_found = 0;
2298 i = offset_to_bit(entry->offset, block_group->sectorsize,
2299 max_t(u64, offset, entry->offset));
2300 search_bits = bytes_to_bits(bytes, block_group->sectorsize);
2301 total_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2305 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
2306 i < BITS_PER_BITMAP;
2307 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
2308 next_zero = find_next_zero_bit(entry->bitmap,
2309 BITS_PER_BITMAP, i);
2310 if (next_zero - i >= search_bits) {
2311 found_bits = next_zero - i;
2322 cluster->max_size = 0;
2326 total_found += found_bits;
2328 if (cluster->max_size < found_bits * block_group->sectorsize)
2329 cluster->max_size = found_bits * block_group->sectorsize;
2331 if (total_found < total_bits) {
2332 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
2333 if (i - start > total_bits * 2) {
2335 cluster->max_size = 0;
2341 cluster->window_start = start * block_group->sectorsize +
2343 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2344 ret = tree_insert_offset(&cluster->root, entry->offset,
2345 &entry->offset_index, 1);
2352 * This searches the block group for just extents to fill the cluster with.
2355 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2356 struct btrfs_free_cluster *cluster,
2357 struct list_head *bitmaps, u64 offset, u64 bytes,
2360 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2361 struct btrfs_free_space *first = NULL;
2362 struct btrfs_free_space *entry = NULL;
2363 struct btrfs_free_space *prev = NULL;
2364 struct btrfs_free_space *last;
2365 struct rb_node *node;
2369 u64 max_gap = 128 * 1024;
2371 entry = tree_search_offset(ctl, offset, 0, 1);
2376 * We don't want bitmaps, so just move along until we find a normal
2379 while (entry->bitmap) {
2380 if (list_empty(&entry->list))
2381 list_add_tail(&entry->list, bitmaps);
2382 node = rb_next(&entry->offset_index);
2385 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2388 window_start = entry->offset;
2389 window_free = entry->bytes;
2390 max_extent = entry->bytes;
2395 while (window_free <= min_bytes) {
2396 node = rb_next(&entry->offset_index);
2399 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2401 if (entry->bitmap) {
2402 if (list_empty(&entry->list))
2403 list_add_tail(&entry->list, bitmaps);
2408 * we haven't filled the empty size and the window is
2409 * very large. reset and try again
2411 if (entry->offset - (prev->offset + prev->bytes) > max_gap ||
2412 entry->offset - window_start > (min_bytes * 2)) {
2414 window_start = entry->offset;
2415 window_free = entry->bytes;
2417 max_extent = entry->bytes;
2420 window_free += entry->bytes;
2421 if (entry->bytes > max_extent)
2422 max_extent = entry->bytes;
2427 cluster->window_start = first->offset;
2429 node = &first->offset_index;
2432 * now we've found our entries, pull them out of the free space
2433 * cache and put them into the cluster rbtree
2438 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2439 node = rb_next(&entry->offset_index);
2443 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2444 ret = tree_insert_offset(&cluster->root, entry->offset,
2445 &entry->offset_index, 0);
2447 } while (node && entry != last);
2449 cluster->max_size = max_extent;
2455 * This specifically looks for bitmaps that may work in the cluster, we assume
2456 * that we have already failed to find extents that will work.
2459 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2460 struct btrfs_free_cluster *cluster,
2461 struct list_head *bitmaps, u64 offset, u64 bytes,
2464 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2465 struct btrfs_free_space *entry;
2467 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2469 if (ctl->total_bitmaps == 0)
2473 * The bitmap that covers offset won't be in the list unless offset
2474 * is just its start offset.
2476 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2477 if (entry->offset != bitmap_offset) {
2478 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2479 if (entry && list_empty(&entry->list))
2480 list_add(&entry->list, bitmaps);
2483 list_for_each_entry(entry, bitmaps, list) {
2484 if (entry->bytes < min_bytes)
2486 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2493 * The bitmaps list has all the bitmaps that record free space
2494 * starting after offset, so no more search is required.
2500 * here we try to find a cluster of blocks in a block group. The goal
2501 * is to find at least bytes free and up to empty_size + bytes free.
2502 * We might not find them all in one contiguous area.
2504 * returns zero and sets up cluster if things worked out, otherwise
2505 * it returns -enospc
2507 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2508 struct btrfs_root *root,
2509 struct btrfs_block_group_cache *block_group,
2510 struct btrfs_free_cluster *cluster,
2511 u64 offset, u64 bytes, u64 empty_size)
2513 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2514 struct btrfs_free_space *entry, *tmp;
2519 /* for metadata, allow allocates with more holes */
2520 if (btrfs_test_opt(root, SSD_SPREAD)) {
2521 min_bytes = bytes + empty_size;
2522 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2524 * we want to do larger allocations when we are
2525 * flushing out the delayed refs, it helps prevent
2526 * making more work as we go along.
2528 if (trans->transaction->delayed_refs.flushing)
2529 min_bytes = max(bytes, (bytes + empty_size) >> 1);
2531 min_bytes = max(bytes, (bytes + empty_size) >> 4);
2533 min_bytes = max(bytes, (bytes + empty_size) >> 2);
2535 spin_lock(&ctl->tree_lock);
2538 * If we know we don't have enough space to make a cluster don't even
2539 * bother doing all the work to try and find one.
2541 if (ctl->free_space < min_bytes) {
2542 spin_unlock(&ctl->tree_lock);
2546 spin_lock(&cluster->lock);
2548 /* someone already found a cluster, hooray */
2549 if (cluster->block_group) {
2554 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2557 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2558 offset, bytes, min_bytes);
2560 /* Clear our temporary list */
2561 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2562 list_del_init(&entry->list);
2565 atomic_inc(&block_group->count);
2566 list_add_tail(&cluster->block_group_list,
2567 &block_group->cluster_list);
2568 cluster->block_group = block_group;
2571 spin_unlock(&cluster->lock);
2572 spin_unlock(&ctl->tree_lock);
2578 * simple code to zero out a cluster
2580 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2582 spin_lock_init(&cluster->lock);
2583 spin_lock_init(&cluster->refill_lock);
2584 cluster->root = RB_ROOT;
2585 cluster->max_size = 0;
2586 INIT_LIST_HEAD(&cluster->block_group_list);
2587 cluster->block_group = NULL;
2590 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2591 u64 *trimmed, u64 start, u64 end, u64 minlen)
2593 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2594 struct btrfs_free_space *entry = NULL;
2595 struct btrfs_fs_info *fs_info = block_group->fs_info;
2597 u64 actually_trimmed;
2602 while (start < end) {
2603 spin_lock(&ctl->tree_lock);
2605 if (ctl->free_space < minlen) {
2606 spin_unlock(&ctl->tree_lock);
2610 entry = tree_search_offset(ctl, start, 0, 1);
2612 entry = tree_search_offset(ctl,
2613 offset_to_bitmap(ctl, start),
2616 if (!entry || entry->offset >= end) {
2617 spin_unlock(&ctl->tree_lock);
2621 if (entry->bitmap) {
2622 ret = search_bitmap(ctl, entry, &start, &bytes);
2625 spin_unlock(&ctl->tree_lock);
2628 bytes = min(bytes, end - start);
2629 bitmap_clear_bits(ctl, entry, start, bytes);
2630 if (entry->bytes == 0)
2631 free_bitmap(ctl, entry);
2633 start = entry->offset + BITS_PER_BITMAP *
2634 block_group->sectorsize;
2635 spin_unlock(&ctl->tree_lock);
2640 start = entry->offset;
2641 bytes = min(entry->bytes, end - start);
2642 unlink_free_space(ctl, entry);
2643 kmem_cache_free(btrfs_free_space_cachep, entry);
2646 spin_unlock(&ctl->tree_lock);
2648 if (bytes >= minlen) {
2649 struct btrfs_space_info *space_info;
2652 space_info = block_group->space_info;
2653 spin_lock(&space_info->lock);
2654 spin_lock(&block_group->lock);
2655 if (!block_group->ro) {
2656 block_group->reserved += bytes;
2657 space_info->bytes_reserved += bytes;
2660 spin_unlock(&block_group->lock);
2661 spin_unlock(&space_info->lock);
2663 ret = btrfs_error_discard_extent(fs_info->extent_root,
2668 btrfs_add_free_space(block_group, start, bytes);
2670 spin_lock(&space_info->lock);
2671 spin_lock(&block_group->lock);
2672 if (block_group->ro)
2673 space_info->bytes_readonly += bytes;
2674 block_group->reserved -= bytes;
2675 space_info->bytes_reserved -= bytes;
2676 spin_unlock(&space_info->lock);
2677 spin_unlock(&block_group->lock);
2682 *trimmed += actually_trimmed;
2687 if (fatal_signal_pending(current)) {
2699 * Find the left-most item in the cache tree, and then return the
2700 * smallest inode number in the item.
2702 * Note: the returned inode number may not be the smallest one in
2703 * the tree, if the left-most item is a bitmap.
2705 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2707 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2708 struct btrfs_free_space *entry = NULL;
2711 spin_lock(&ctl->tree_lock);
2713 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2716 entry = rb_entry(rb_first(&ctl->free_space_offset),
2717 struct btrfs_free_space, offset_index);
2719 if (!entry->bitmap) {
2720 ino = entry->offset;
2722 unlink_free_space(ctl, entry);
2726 kmem_cache_free(btrfs_free_space_cachep, entry);
2728 link_free_space(ctl, entry);
2734 ret = search_bitmap(ctl, entry, &offset, &count);
2738 bitmap_clear_bits(ctl, entry, offset, 1);
2739 if (entry->bytes == 0)
2740 free_bitmap(ctl, entry);
2743 spin_unlock(&ctl->tree_lock);
2748 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2749 struct btrfs_path *path)
2751 struct inode *inode = NULL;
2753 spin_lock(&root->cache_lock);
2754 if (root->cache_inode)
2755 inode = igrab(root->cache_inode);
2756 spin_unlock(&root->cache_lock);
2760 inode = __lookup_free_space_inode(root, path, 0);
2764 spin_lock(&root->cache_lock);
2765 if (!btrfs_fs_closing(root->fs_info))
2766 root->cache_inode = igrab(inode);
2767 spin_unlock(&root->cache_lock);
2772 int create_free_ino_inode(struct btrfs_root *root,
2773 struct btrfs_trans_handle *trans,
2774 struct btrfs_path *path)
2776 return __create_free_space_inode(root, trans, path,
2777 BTRFS_FREE_INO_OBJECTID, 0);
2780 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2782 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2783 struct btrfs_path *path;
2784 struct inode *inode;
2786 u64 root_gen = btrfs_root_generation(&root->root_item);
2788 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2792 * If we're unmounting then just return, since this does a search on the
2793 * normal root and not the commit root and we could deadlock.
2795 if (btrfs_fs_closing(fs_info))
2798 path = btrfs_alloc_path();
2802 inode = lookup_free_ino_inode(root, path);
2806 if (root_gen != BTRFS_I(inode)->generation)
2809 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2812 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2813 "root %llu\n", root->root_key.objectid);
2817 btrfs_free_path(path);
2821 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2822 struct btrfs_trans_handle *trans,
2823 struct btrfs_path *path)
2825 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2826 struct inode *inode;
2829 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2832 inode = lookup_free_ino_inode(root, path);
2836 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2838 btrfs_delalloc_release_metadata(inode, inode->i_size);
2840 printk(KERN_ERR "btrfs: failed to write free ino cache "
2841 "for root %llu\n", root->root_key.objectid);