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 if (io_ctl->pages[i]) {
323 ClearPageChecked(io_ctl->pages[i]);
324 unlock_page(io_ctl->pages[i]);
325 page_cache_release(io_ctl->pages[i]);
330 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
334 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
337 for (i = 0; i < io_ctl->num_pages; i++) {
338 page = find_or_create_page(inode->i_mapping, i, mask);
340 io_ctl_drop_pages(io_ctl);
343 io_ctl->pages[i] = page;
344 if (uptodate && !PageUptodate(page)) {
345 btrfs_readpage(NULL, page);
347 if (!PageUptodate(page)) {
348 printk(KERN_ERR "btrfs: error reading free "
350 io_ctl_drop_pages(io_ctl);
356 for (i = 0; i < io_ctl->num_pages; i++) {
357 clear_page_dirty_for_io(io_ctl->pages[i]);
358 set_page_extent_mapped(io_ctl->pages[i]);
364 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
368 io_ctl_map_page(io_ctl, 1);
371 * Skip the csum areas. If we don't check crcs then we just have a
372 * 64bit chunk at the front of the first page.
374 if (io_ctl->check_crcs) {
375 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
376 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
378 io_ctl->cur += sizeof(u64);
379 io_ctl->size -= sizeof(u64) * 2;
383 *val = cpu_to_le64(generation);
384 io_ctl->cur += sizeof(u64);
387 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
392 * Skip the crc area. If we don't check crcs then we just have a 64bit
393 * chunk at the front of the first page.
395 if (io_ctl->check_crcs) {
396 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
397 io_ctl->size -= sizeof(u64) +
398 (sizeof(u32) * io_ctl->num_pages);
400 io_ctl->cur += sizeof(u64);
401 io_ctl->size -= sizeof(u64) * 2;
405 if (le64_to_cpu(*gen) != generation) {
406 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
407 "(%Lu) does not match inode (%Lu)\n", *gen,
409 io_ctl_unmap_page(io_ctl);
412 io_ctl->cur += sizeof(u64);
416 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
422 if (!io_ctl->check_crcs) {
423 io_ctl_unmap_page(io_ctl);
428 offset = sizeof(u32) * io_ctl->num_pages;;
430 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
431 PAGE_CACHE_SIZE - offset);
432 btrfs_csum_final(crc, (char *)&crc);
433 io_ctl_unmap_page(io_ctl);
434 tmp = kmap(io_ctl->pages[0]);
437 kunmap(io_ctl->pages[0]);
440 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
446 if (!io_ctl->check_crcs) {
447 io_ctl_map_page(io_ctl, 0);
452 offset = sizeof(u32) * io_ctl->num_pages;
454 tmp = kmap(io_ctl->pages[0]);
457 kunmap(io_ctl->pages[0]);
459 io_ctl_map_page(io_ctl, 0);
460 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
461 PAGE_CACHE_SIZE - offset);
462 btrfs_csum_final(crc, (char *)&crc);
464 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
466 io_ctl_unmap_page(io_ctl);
473 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
476 struct btrfs_free_space_entry *entry;
482 entry->offset = cpu_to_le64(offset);
483 entry->bytes = cpu_to_le64(bytes);
484 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
485 BTRFS_FREE_SPACE_EXTENT;
486 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
487 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
489 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
492 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
494 /* No more pages to map */
495 if (io_ctl->index >= io_ctl->num_pages)
498 /* map the next page */
499 io_ctl_map_page(io_ctl, 1);
503 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
509 * If we aren't at the start of the current page, unmap this one and
510 * map the next one if there is any left.
512 if (io_ctl->cur != io_ctl->orig) {
513 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
514 if (io_ctl->index >= io_ctl->num_pages)
516 io_ctl_map_page(io_ctl, 0);
519 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
520 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
521 if (io_ctl->index < io_ctl->num_pages)
522 io_ctl_map_page(io_ctl, 0);
526 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
529 * If we're not on the boundary we know we've modified the page and we
530 * need to crc the page.
532 if (io_ctl->cur != io_ctl->orig)
533 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
535 io_ctl_unmap_page(io_ctl);
537 while (io_ctl->index < io_ctl->num_pages) {
538 io_ctl_map_page(io_ctl, 1);
539 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
543 static int io_ctl_read_entry(struct io_ctl *io_ctl,
544 struct btrfs_free_space *entry, u8 *type)
546 struct btrfs_free_space_entry *e;
550 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
556 entry->offset = le64_to_cpu(e->offset);
557 entry->bytes = le64_to_cpu(e->bytes);
559 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
560 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
562 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
565 io_ctl_unmap_page(io_ctl);
570 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
571 struct btrfs_free_space *entry)
575 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
579 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
580 io_ctl_unmap_page(io_ctl);
585 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
586 struct btrfs_free_space_ctl *ctl,
587 struct btrfs_path *path, u64 offset)
589 struct btrfs_free_space_header *header;
590 struct extent_buffer *leaf;
591 struct io_ctl io_ctl;
592 struct btrfs_key key;
593 struct btrfs_free_space *e, *n;
594 struct list_head bitmaps;
601 INIT_LIST_HEAD(&bitmaps);
603 /* Nothing in the space cache, goodbye */
604 if (!i_size_read(inode))
607 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
611 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
615 btrfs_release_path(path);
621 leaf = path->nodes[0];
622 header = btrfs_item_ptr(leaf, path->slots[0],
623 struct btrfs_free_space_header);
624 num_entries = btrfs_free_space_entries(leaf, header);
625 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
626 generation = btrfs_free_space_generation(leaf, header);
627 btrfs_release_path(path);
629 if (BTRFS_I(inode)->generation != generation) {
630 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
631 " not match free space cache generation (%llu)\n",
632 (unsigned long long)BTRFS_I(inode)->generation,
633 (unsigned long long)generation);
640 ret = io_ctl_init(&io_ctl, inode, root);
644 ret = readahead_cache(inode);
648 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
652 ret = io_ctl_check_crc(&io_ctl, 0);
656 ret = io_ctl_check_generation(&io_ctl, generation);
660 while (num_entries) {
661 e = kmem_cache_zalloc(btrfs_free_space_cachep,
666 ret = io_ctl_read_entry(&io_ctl, e, &type);
668 kmem_cache_free(btrfs_free_space_cachep, e);
673 kmem_cache_free(btrfs_free_space_cachep, e);
677 if (type == BTRFS_FREE_SPACE_EXTENT) {
678 spin_lock(&ctl->tree_lock);
679 ret = link_free_space(ctl, e);
680 spin_unlock(&ctl->tree_lock);
682 printk(KERN_ERR "Duplicate entries in "
683 "free space cache, dumping\n");
684 kmem_cache_free(btrfs_free_space_cachep, e);
688 BUG_ON(!num_bitmaps);
690 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
693 btrfs_free_space_cachep, e);
696 spin_lock(&ctl->tree_lock);
697 ret = link_free_space(ctl, e);
698 ctl->total_bitmaps++;
699 ctl->op->recalc_thresholds(ctl);
700 spin_unlock(&ctl->tree_lock);
702 printk(KERN_ERR "Duplicate entries in "
703 "free space cache, dumping\n");
704 kmem_cache_free(btrfs_free_space_cachep, e);
707 list_add_tail(&e->list, &bitmaps);
713 io_ctl_unmap_page(&io_ctl);
716 * We add the bitmaps at the end of the entries in order that
717 * the bitmap entries are added to the cache.
719 list_for_each_entry_safe(e, n, &bitmaps, list) {
720 list_del_init(&e->list);
721 ret = io_ctl_read_bitmap(&io_ctl, e);
726 io_ctl_drop_pages(&io_ctl);
729 io_ctl_free(&io_ctl);
732 io_ctl_drop_pages(&io_ctl);
733 __btrfs_remove_free_space_cache(ctl);
737 int load_free_space_cache(struct btrfs_fs_info *fs_info,
738 struct btrfs_block_group_cache *block_group)
740 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
741 struct btrfs_root *root = fs_info->tree_root;
743 struct btrfs_path *path;
746 u64 used = btrfs_block_group_used(&block_group->item);
749 * If we're unmounting then just return, since this does a search on the
750 * normal root and not the commit root and we could deadlock.
752 if (btrfs_fs_closing(fs_info))
756 * If this block group has been marked to be cleared for one reason or
757 * another then we can't trust the on disk cache, so just return.
759 spin_lock(&block_group->lock);
760 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
761 spin_unlock(&block_group->lock);
764 spin_unlock(&block_group->lock);
766 path = btrfs_alloc_path();
770 inode = lookup_free_space_inode(root, block_group, path);
772 btrfs_free_path(path);
776 /* We may have converted the inode and made the cache invalid. */
777 spin_lock(&block_group->lock);
778 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
779 spin_unlock(&block_group->lock);
782 spin_unlock(&block_group->lock);
784 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
785 path, block_group->key.objectid);
786 btrfs_free_path(path);
790 spin_lock(&ctl->tree_lock);
791 matched = (ctl->free_space == (block_group->key.offset - used -
792 block_group->bytes_super));
793 spin_unlock(&ctl->tree_lock);
796 __btrfs_remove_free_space_cache(ctl);
797 printk(KERN_ERR "block group %llu has an wrong amount of free "
798 "space\n", block_group->key.objectid);
803 /* This cache is bogus, make sure it gets cleared */
804 spin_lock(&block_group->lock);
805 block_group->disk_cache_state = BTRFS_DC_CLEAR;
806 spin_unlock(&block_group->lock);
809 printk(KERN_ERR "btrfs: failed to load free space cache "
810 "for block group %llu\n", block_group->key.objectid);
818 * __btrfs_write_out_cache - write out cached info to an inode
819 * @root - the root the inode belongs to
820 * @ctl - the free space cache we are going to write out
821 * @block_group - the block_group for this cache if it belongs to a block_group
822 * @trans - the trans handle
823 * @path - the path to use
824 * @offset - the offset for the key we'll insert
826 * This function writes out a free space cache struct to disk for quick recovery
827 * on mount. This will return 0 if it was successfull in writing the cache out,
828 * and -1 if it was not.
830 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
831 struct btrfs_free_space_ctl *ctl,
832 struct btrfs_block_group_cache *block_group,
833 struct btrfs_trans_handle *trans,
834 struct btrfs_path *path, u64 offset)
836 struct btrfs_free_space_header *header;
837 struct extent_buffer *leaf;
838 struct rb_node *node;
839 struct list_head *pos, *n;
840 struct extent_state *cached_state = NULL;
841 struct btrfs_free_cluster *cluster = NULL;
842 struct extent_io_tree *unpin = NULL;
843 struct io_ctl io_ctl;
844 struct list_head bitmap_list;
845 struct btrfs_key key;
846 u64 start, extent_start, extent_end, len;
852 INIT_LIST_HEAD(&bitmap_list);
854 if (!i_size_read(inode))
857 ret = io_ctl_init(&io_ctl, inode, root);
861 /* Get the cluster for this block_group if it exists */
862 if (block_group && !list_empty(&block_group->cluster_list))
863 cluster = list_entry(block_group->cluster_list.next,
864 struct btrfs_free_cluster,
867 /* Lock all pages first so we can lock the extent safely. */
868 io_ctl_prepare_pages(&io_ctl, inode, 0);
870 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
871 0, &cached_state, GFP_NOFS);
873 node = rb_first(&ctl->free_space_offset);
874 if (!node && cluster) {
875 node = rb_first(&cluster->root);
879 /* Make sure we can fit our crcs into the first page */
880 if (io_ctl.check_crcs &&
881 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
886 io_ctl_set_generation(&io_ctl, trans->transid);
888 /* Write out the extent entries */
890 struct btrfs_free_space *e;
892 e = rb_entry(node, struct btrfs_free_space, offset_index);
895 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
901 list_add_tail(&e->list, &bitmap_list);
904 node = rb_next(node);
905 if (!node && cluster) {
906 node = rb_first(&cluster->root);
912 * We want to add any pinned extents to our free space cache
913 * so we don't leak the space
917 * We shouldn't have switched the pinned extents yet so this is the
920 unpin = root->fs_info->pinned_extents;
923 start = block_group->key.objectid;
925 while (block_group && (start < block_group->key.objectid +
926 block_group->key.offset)) {
927 ret = find_first_extent_bit(unpin, start,
928 &extent_start, &extent_end,
935 /* This pinned extent is out of our range */
936 if (extent_start >= block_group->key.objectid +
937 block_group->key.offset)
940 extent_start = max(extent_start, start);
941 extent_end = min(block_group->key.objectid +
942 block_group->key.offset, extent_end + 1);
943 len = extent_end - extent_start;
946 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
953 /* Write out the bitmaps */
954 list_for_each_safe(pos, n, &bitmap_list) {
955 struct btrfs_free_space *entry =
956 list_entry(pos, struct btrfs_free_space, list);
958 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
961 list_del_init(&entry->list);
964 /* Zero out the rest of the pages just to make sure */
965 io_ctl_zero_remaining_pages(&io_ctl);
967 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
968 0, i_size_read(inode), &cached_state);
969 io_ctl_drop_pages(&io_ctl);
970 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
971 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
977 ret = filemap_write_and_wait(inode->i_mapping);
981 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
985 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
987 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
988 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
992 leaf = path->nodes[0];
994 struct btrfs_key found_key;
995 BUG_ON(!path->slots[0]);
997 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
998 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
999 found_key.offset != offset) {
1000 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1002 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1004 btrfs_release_path(path);
1009 BTRFS_I(inode)->generation = trans->transid;
1010 header = btrfs_item_ptr(leaf, path->slots[0],
1011 struct btrfs_free_space_header);
1012 btrfs_set_free_space_entries(leaf, header, entries);
1013 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1014 btrfs_set_free_space_generation(leaf, header, trans->transid);
1015 btrfs_mark_buffer_dirty(leaf);
1016 btrfs_release_path(path);
1020 io_ctl_free(&io_ctl);
1022 invalidate_inode_pages2(inode->i_mapping);
1023 BTRFS_I(inode)->generation = 0;
1025 btrfs_update_inode(trans, root, inode);
1029 list_for_each_safe(pos, n, &bitmap_list) {
1030 struct btrfs_free_space *entry =
1031 list_entry(pos, struct btrfs_free_space, list);
1032 list_del_init(&entry->list);
1034 io_ctl_drop_pages(&io_ctl);
1035 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1036 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1040 int btrfs_write_out_cache(struct btrfs_root *root,
1041 struct btrfs_trans_handle *trans,
1042 struct btrfs_block_group_cache *block_group,
1043 struct btrfs_path *path)
1045 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1046 struct inode *inode;
1049 root = root->fs_info->tree_root;
1051 spin_lock(&block_group->lock);
1052 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1053 spin_unlock(&block_group->lock);
1056 spin_unlock(&block_group->lock);
1058 inode = lookup_free_space_inode(root, block_group, path);
1062 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1063 path, block_group->key.objectid);
1065 spin_lock(&block_group->lock);
1066 block_group->disk_cache_state = BTRFS_DC_ERROR;
1067 spin_unlock(&block_group->lock);
1070 printk(KERN_ERR "btrfs: failed to write free space cace "
1071 "for block group %llu\n", block_group->key.objectid);
1079 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1082 BUG_ON(offset < bitmap_start);
1083 offset -= bitmap_start;
1084 return (unsigned long)(div_u64(offset, unit));
1087 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1089 return (unsigned long)(div_u64(bytes, unit));
1092 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1096 u64 bytes_per_bitmap;
1098 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1099 bitmap_start = offset - ctl->start;
1100 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1101 bitmap_start *= bytes_per_bitmap;
1102 bitmap_start += ctl->start;
1104 return bitmap_start;
1107 static int tree_insert_offset(struct rb_root *root, u64 offset,
1108 struct rb_node *node, int bitmap)
1110 struct rb_node **p = &root->rb_node;
1111 struct rb_node *parent = NULL;
1112 struct btrfs_free_space *info;
1116 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1118 if (offset < info->offset) {
1120 } else if (offset > info->offset) {
1121 p = &(*p)->rb_right;
1124 * we could have a bitmap entry and an extent entry
1125 * share the same offset. If this is the case, we want
1126 * the extent entry to always be found first if we do a
1127 * linear search through the tree, since we want to have
1128 * the quickest allocation time, and allocating from an
1129 * extent is faster than allocating from a bitmap. So
1130 * if we're inserting a bitmap and we find an entry at
1131 * this offset, we want to go right, or after this entry
1132 * logically. If we are inserting an extent and we've
1133 * found a bitmap, we want to go left, or before
1141 p = &(*p)->rb_right;
1143 if (!info->bitmap) {
1152 rb_link_node(node, parent, p);
1153 rb_insert_color(node, root);
1159 * searches the tree for the given offset.
1161 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1162 * want a section that has at least bytes size and comes at or after the given
1165 static struct btrfs_free_space *
1166 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1167 u64 offset, int bitmap_only, int fuzzy)
1169 struct rb_node *n = ctl->free_space_offset.rb_node;
1170 struct btrfs_free_space *entry, *prev = NULL;
1172 /* find entry that is closest to the 'offset' */
1179 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1182 if (offset < entry->offset)
1184 else if (offset > entry->offset)
1197 * bitmap entry and extent entry may share same offset,
1198 * in that case, bitmap entry comes after extent entry.
1203 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1204 if (entry->offset != offset)
1207 WARN_ON(!entry->bitmap);
1210 if (entry->bitmap) {
1212 * if previous extent entry covers the offset,
1213 * we should return it instead of the bitmap entry
1215 n = &entry->offset_index;
1220 prev = rb_entry(n, struct btrfs_free_space,
1222 if (!prev->bitmap) {
1223 if (prev->offset + prev->bytes > offset)
1235 /* find last entry before the 'offset' */
1237 if (entry->offset > offset) {
1238 n = rb_prev(&entry->offset_index);
1240 entry = rb_entry(n, struct btrfs_free_space,
1242 BUG_ON(entry->offset > offset);
1251 if (entry->bitmap) {
1252 n = &entry->offset_index;
1257 prev = rb_entry(n, struct btrfs_free_space,
1259 if (!prev->bitmap) {
1260 if (prev->offset + prev->bytes > offset)
1265 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1267 } else if (entry->offset + entry->bytes > offset)
1274 if (entry->bitmap) {
1275 if (entry->offset + BITS_PER_BITMAP *
1279 if (entry->offset + entry->bytes > offset)
1283 n = rb_next(&entry->offset_index);
1286 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1292 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1293 struct btrfs_free_space *info)
1295 rb_erase(&info->offset_index, &ctl->free_space_offset);
1296 ctl->free_extents--;
1299 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1300 struct btrfs_free_space *info)
1302 __unlink_free_space(ctl, info);
1303 ctl->free_space -= info->bytes;
1306 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1307 struct btrfs_free_space *info)
1311 BUG_ON(!info->bitmap && !info->bytes);
1312 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1313 &info->offset_index, (info->bitmap != NULL));
1317 ctl->free_space += info->bytes;
1318 ctl->free_extents++;
1322 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1324 struct btrfs_block_group_cache *block_group = ctl->private;
1328 u64 size = block_group->key.offset;
1329 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1330 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1332 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1335 * The goal is to keep the total amount of memory used per 1gb of space
1336 * at or below 32k, so we need to adjust how much memory we allow to be
1337 * used by extent based free space tracking
1339 if (size < 1024 * 1024 * 1024)
1340 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1342 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1343 div64_u64(size, 1024 * 1024 * 1024);
1346 * we want to account for 1 more bitmap than what we have so we can make
1347 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1348 * we add more bitmaps.
1350 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1352 if (bitmap_bytes >= max_bytes) {
1353 ctl->extents_thresh = 0;
1358 * we want the extent entry threshold to always be at most 1/2 the maxw
1359 * bytes we can have, or whatever is less than that.
1361 extent_bytes = max_bytes - bitmap_bytes;
1362 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1364 ctl->extents_thresh =
1365 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1368 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1369 struct btrfs_free_space *info,
1370 u64 offset, u64 bytes)
1372 unsigned long start, count;
1374 start = offset_to_bit(info->offset, ctl->unit, offset);
1375 count = bytes_to_bits(bytes, ctl->unit);
1376 BUG_ON(start + count > BITS_PER_BITMAP);
1378 bitmap_clear(info->bitmap, start, count);
1380 info->bytes -= bytes;
1383 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1384 struct btrfs_free_space *info, u64 offset,
1387 __bitmap_clear_bits(ctl, info, offset, bytes);
1388 ctl->free_space -= bytes;
1391 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1392 struct btrfs_free_space *info, u64 offset,
1395 unsigned long start, count;
1397 start = offset_to_bit(info->offset, ctl->unit, offset);
1398 count = bytes_to_bits(bytes, ctl->unit);
1399 BUG_ON(start + count > BITS_PER_BITMAP);
1401 bitmap_set(info->bitmap, start, count);
1403 info->bytes += bytes;
1404 ctl->free_space += bytes;
1407 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1408 struct btrfs_free_space *bitmap_info, u64 *offset,
1411 unsigned long found_bits = 0;
1412 unsigned long bits, i;
1413 unsigned long next_zero;
1415 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1416 max_t(u64, *offset, bitmap_info->offset));
1417 bits = bytes_to_bits(*bytes, ctl->unit);
1419 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1420 i < BITS_PER_BITMAP;
1421 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1422 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1423 BITS_PER_BITMAP, i);
1424 if ((next_zero - i) >= bits) {
1425 found_bits = next_zero - i;
1432 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1433 *bytes = (u64)(found_bits) * ctl->unit;
1440 static struct btrfs_free_space *
1441 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1443 struct btrfs_free_space *entry;
1444 struct rb_node *node;
1447 if (!ctl->free_space_offset.rb_node)
1450 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1454 for (node = &entry->offset_index; node; node = rb_next(node)) {
1455 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1456 if (entry->bytes < *bytes)
1459 if (entry->bitmap) {
1460 ret = search_bitmap(ctl, entry, offset, bytes);
1466 *offset = entry->offset;
1467 *bytes = entry->bytes;
1474 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1475 struct btrfs_free_space *info, u64 offset)
1477 info->offset = offset_to_bitmap(ctl, offset);
1479 INIT_LIST_HEAD(&info->list);
1480 link_free_space(ctl, info);
1481 ctl->total_bitmaps++;
1483 ctl->op->recalc_thresholds(ctl);
1486 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1487 struct btrfs_free_space *bitmap_info)
1489 unlink_free_space(ctl, bitmap_info);
1490 kfree(bitmap_info->bitmap);
1491 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1492 ctl->total_bitmaps--;
1493 ctl->op->recalc_thresholds(ctl);
1496 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1497 struct btrfs_free_space *bitmap_info,
1498 u64 *offset, u64 *bytes)
1501 u64 search_start, search_bytes;
1505 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1508 * XXX - this can go away after a few releases.
1510 * since the only user of btrfs_remove_free_space is the tree logging
1511 * stuff, and the only way to test that is under crash conditions, we
1512 * want to have this debug stuff here just in case somethings not
1513 * working. Search the bitmap for the space we are trying to use to
1514 * make sure its actually there. If its not there then we need to stop
1515 * because something has gone wrong.
1517 search_start = *offset;
1518 search_bytes = *bytes;
1519 search_bytes = min(search_bytes, end - search_start + 1);
1520 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1521 BUG_ON(ret < 0 || search_start != *offset);
1523 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1524 bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
1525 *bytes -= end - *offset + 1;
1527 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1528 bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
1533 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1534 if (!bitmap_info->bytes)
1535 free_bitmap(ctl, bitmap_info);
1538 * no entry after this bitmap, but we still have bytes to
1539 * remove, so something has gone wrong.
1544 bitmap_info = rb_entry(next, struct btrfs_free_space,
1548 * if the next entry isn't a bitmap we need to return to let the
1549 * extent stuff do its work.
1551 if (!bitmap_info->bitmap)
1555 * Ok the next item is a bitmap, but it may not actually hold
1556 * the information for the rest of this free space stuff, so
1557 * look for it, and if we don't find it return so we can try
1558 * everything over again.
1560 search_start = *offset;
1561 search_bytes = *bytes;
1562 ret = search_bitmap(ctl, bitmap_info, &search_start,
1564 if (ret < 0 || search_start != *offset)
1568 } else if (!bitmap_info->bytes)
1569 free_bitmap(ctl, bitmap_info);
1574 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1575 struct btrfs_free_space *info, u64 offset,
1578 u64 bytes_to_set = 0;
1581 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1583 bytes_to_set = min(end - offset, bytes);
1585 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1587 return bytes_to_set;
1591 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1592 struct btrfs_free_space *info)
1594 struct btrfs_block_group_cache *block_group = ctl->private;
1597 * If we are below the extents threshold then we can add this as an
1598 * extent, and don't have to deal with the bitmap
1600 if (ctl->free_extents < ctl->extents_thresh) {
1602 * If this block group has some small extents we don't want to
1603 * use up all of our free slots in the cache with them, we want
1604 * to reserve them to larger extents, however if we have plent
1605 * of cache left then go ahead an dadd them, no sense in adding
1606 * the overhead of a bitmap if we don't have to.
1608 if (info->bytes <= block_group->sectorsize * 4) {
1609 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1617 * some block groups are so tiny they can't be enveloped by a bitmap, so
1618 * don't even bother to create a bitmap for this
1620 if (BITS_PER_BITMAP * block_group->sectorsize >
1621 block_group->key.offset)
1627 static struct btrfs_free_space_op free_space_op = {
1628 .recalc_thresholds = recalculate_thresholds,
1629 .use_bitmap = use_bitmap,
1632 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1633 struct btrfs_free_space *info)
1635 struct btrfs_free_space *bitmap_info;
1636 struct btrfs_block_group_cache *block_group = NULL;
1638 u64 bytes, offset, bytes_added;
1641 bytes = info->bytes;
1642 offset = info->offset;
1644 if (!ctl->op->use_bitmap(ctl, info))
1647 if (ctl->op == &free_space_op)
1648 block_group = ctl->private;
1651 * Since we link bitmaps right into the cluster we need to see if we
1652 * have a cluster here, and if so and it has our bitmap we need to add
1653 * the free space to that bitmap.
1655 if (block_group && !list_empty(&block_group->cluster_list)) {
1656 struct btrfs_free_cluster *cluster;
1657 struct rb_node *node;
1658 struct btrfs_free_space *entry;
1660 cluster = list_entry(block_group->cluster_list.next,
1661 struct btrfs_free_cluster,
1663 spin_lock(&cluster->lock);
1664 node = rb_first(&cluster->root);
1666 spin_unlock(&cluster->lock);
1667 goto no_cluster_bitmap;
1670 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1671 if (!entry->bitmap) {
1672 spin_unlock(&cluster->lock);
1673 goto no_cluster_bitmap;
1676 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1677 bytes_added = add_bytes_to_bitmap(ctl, entry,
1679 bytes -= bytes_added;
1680 offset += bytes_added;
1682 spin_unlock(&cluster->lock);
1690 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1697 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1698 bytes -= bytes_added;
1699 offset += bytes_added;
1709 if (info && info->bitmap) {
1710 add_new_bitmap(ctl, info, offset);
1715 spin_unlock(&ctl->tree_lock);
1717 /* no pre-allocated info, allocate a new one */
1719 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1722 spin_lock(&ctl->tree_lock);
1728 /* allocate the bitmap */
1729 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1730 spin_lock(&ctl->tree_lock);
1731 if (!info->bitmap) {
1741 kfree(info->bitmap);
1742 kmem_cache_free(btrfs_free_space_cachep, info);
1748 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1749 struct btrfs_free_space *info, bool update_stat)
1751 struct btrfs_free_space *left_info;
1752 struct btrfs_free_space *right_info;
1753 bool merged = false;
1754 u64 offset = info->offset;
1755 u64 bytes = info->bytes;
1758 * first we want to see if there is free space adjacent to the range we
1759 * are adding, if there is remove that struct and add a new one to
1760 * cover the entire range
1762 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1763 if (right_info && rb_prev(&right_info->offset_index))
1764 left_info = rb_entry(rb_prev(&right_info->offset_index),
1765 struct btrfs_free_space, offset_index);
1767 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1769 if (right_info && !right_info->bitmap) {
1771 unlink_free_space(ctl, right_info);
1773 __unlink_free_space(ctl, right_info);
1774 info->bytes += right_info->bytes;
1775 kmem_cache_free(btrfs_free_space_cachep, right_info);
1779 if (left_info && !left_info->bitmap &&
1780 left_info->offset + left_info->bytes == offset) {
1782 unlink_free_space(ctl, left_info);
1784 __unlink_free_space(ctl, left_info);
1785 info->offset = left_info->offset;
1786 info->bytes += left_info->bytes;
1787 kmem_cache_free(btrfs_free_space_cachep, left_info);
1794 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1795 u64 offset, u64 bytes)
1797 struct btrfs_free_space *info;
1800 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1804 info->offset = offset;
1805 info->bytes = bytes;
1807 spin_lock(&ctl->tree_lock);
1809 if (try_merge_free_space(ctl, info, true))
1813 * There was no extent directly to the left or right of this new
1814 * extent then we know we're going to have to allocate a new extent, so
1815 * before we do that see if we need to drop this into a bitmap
1817 ret = insert_into_bitmap(ctl, info);
1825 ret = link_free_space(ctl, info);
1827 kmem_cache_free(btrfs_free_space_cachep, info);
1829 spin_unlock(&ctl->tree_lock);
1832 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1833 BUG_ON(ret == -EEXIST);
1839 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1840 u64 offset, u64 bytes)
1842 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1843 struct btrfs_free_space *info;
1844 struct btrfs_free_space *next_info = NULL;
1847 spin_lock(&ctl->tree_lock);
1850 info = tree_search_offset(ctl, offset, 0, 0);
1853 * oops didn't find an extent that matched the space we wanted
1854 * to remove, look for a bitmap instead
1856 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1859 /* the tree logging code might be calling us before we
1860 * have fully loaded the free space rbtree for this
1861 * block group. So it is possible the entry won't
1862 * be in the rbtree yet at all. The caching code
1863 * will make sure not to put it in the rbtree if
1864 * the logging code has pinned it.
1870 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1872 next_info = rb_entry(rb_next(&info->offset_index),
1873 struct btrfs_free_space,
1876 if (next_info->bitmap)
1877 end = next_info->offset +
1878 BITS_PER_BITMAP * ctl->unit - 1;
1880 end = next_info->offset + next_info->bytes;
1882 if (next_info->bytes < bytes ||
1883 next_info->offset > offset || offset > end) {
1884 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1885 " trying to use %llu\n",
1886 (unsigned long long)info->offset,
1887 (unsigned long long)info->bytes,
1888 (unsigned long long)bytes);
1897 if (info->bytes == bytes) {
1898 unlink_free_space(ctl, info);
1900 kfree(info->bitmap);
1901 ctl->total_bitmaps--;
1903 kmem_cache_free(btrfs_free_space_cachep, info);
1908 if (!info->bitmap && info->offset == offset) {
1909 unlink_free_space(ctl, info);
1910 info->offset += bytes;
1911 info->bytes -= bytes;
1912 ret = link_free_space(ctl, info);
1917 if (!info->bitmap && info->offset <= offset &&
1918 info->offset + info->bytes >= offset + bytes) {
1919 u64 old_start = info->offset;
1921 * we're freeing space in the middle of the info,
1922 * this can happen during tree log replay
1924 * first unlink the old info and then
1925 * insert it again after the hole we're creating
1927 unlink_free_space(ctl, info);
1928 if (offset + bytes < info->offset + info->bytes) {
1929 u64 old_end = info->offset + info->bytes;
1931 info->offset = offset + bytes;
1932 info->bytes = old_end - info->offset;
1933 ret = link_free_space(ctl, info);
1938 /* the hole we're creating ends at the end
1939 * of the info struct, just free the info
1941 kmem_cache_free(btrfs_free_space_cachep, info);
1943 spin_unlock(&ctl->tree_lock);
1945 /* step two, insert a new info struct to cover
1946 * anything before the hole
1948 ret = btrfs_add_free_space(block_group, old_start,
1949 offset - old_start);
1954 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1959 spin_unlock(&ctl->tree_lock);
1964 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1967 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1968 struct btrfs_free_space *info;
1972 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1973 info = rb_entry(n, struct btrfs_free_space, offset_index);
1974 if (info->bytes >= bytes)
1976 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1977 (unsigned long long)info->offset,
1978 (unsigned long long)info->bytes,
1979 (info->bitmap) ? "yes" : "no");
1981 printk(KERN_INFO "block group has cluster?: %s\n",
1982 list_empty(&block_group->cluster_list) ? "no" : "yes");
1983 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1987 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1989 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1991 spin_lock_init(&ctl->tree_lock);
1992 ctl->unit = block_group->sectorsize;
1993 ctl->start = block_group->key.objectid;
1994 ctl->private = block_group;
1995 ctl->op = &free_space_op;
1998 * we only want to have 32k of ram per block group for keeping
1999 * track of free space, and if we pass 1/2 of that we want to
2000 * start converting things over to using bitmaps
2002 ctl->extents_thresh = ((1024 * 32) / 2) /
2003 sizeof(struct btrfs_free_space);
2007 * for a given cluster, put all of its extents back into the free
2008 * space cache. If the block group passed doesn't match the block group
2009 * pointed to by the cluster, someone else raced in and freed the
2010 * cluster already. In that case, we just return without changing anything
2013 __btrfs_return_cluster_to_free_space(
2014 struct btrfs_block_group_cache *block_group,
2015 struct btrfs_free_cluster *cluster)
2017 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2018 struct btrfs_free_space *entry;
2019 struct rb_node *node;
2021 spin_lock(&cluster->lock);
2022 if (cluster->block_group != block_group)
2025 cluster->block_group = NULL;
2026 cluster->window_start = 0;
2027 list_del_init(&cluster->block_group_list);
2029 node = rb_first(&cluster->root);
2033 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2034 node = rb_next(&entry->offset_index);
2035 rb_erase(&entry->offset_index, &cluster->root);
2037 bitmap = (entry->bitmap != NULL);
2039 try_merge_free_space(ctl, entry, false);
2040 tree_insert_offset(&ctl->free_space_offset,
2041 entry->offset, &entry->offset_index, bitmap);
2043 cluster->root = RB_ROOT;
2046 spin_unlock(&cluster->lock);
2047 btrfs_put_block_group(block_group);
2051 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
2053 struct btrfs_free_space *info;
2054 struct rb_node *node;
2056 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2057 info = rb_entry(node, struct btrfs_free_space, offset_index);
2058 if (!info->bitmap) {
2059 unlink_free_space(ctl, info);
2060 kmem_cache_free(btrfs_free_space_cachep, info);
2062 free_bitmap(ctl, info);
2064 if (need_resched()) {
2065 spin_unlock(&ctl->tree_lock);
2067 spin_lock(&ctl->tree_lock);
2072 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2074 spin_lock(&ctl->tree_lock);
2075 __btrfs_remove_free_space_cache_locked(ctl);
2076 spin_unlock(&ctl->tree_lock);
2079 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2081 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2082 struct btrfs_free_cluster *cluster;
2083 struct list_head *head;
2085 spin_lock(&ctl->tree_lock);
2086 while ((head = block_group->cluster_list.next) !=
2087 &block_group->cluster_list) {
2088 cluster = list_entry(head, struct btrfs_free_cluster,
2091 WARN_ON(cluster->block_group != block_group);
2092 __btrfs_return_cluster_to_free_space(block_group, cluster);
2093 if (need_resched()) {
2094 spin_unlock(&ctl->tree_lock);
2096 spin_lock(&ctl->tree_lock);
2099 __btrfs_remove_free_space_cache_locked(ctl);
2100 spin_unlock(&ctl->tree_lock);
2104 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2105 u64 offset, u64 bytes, u64 empty_size)
2107 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2108 struct btrfs_free_space *entry = NULL;
2109 u64 bytes_search = bytes + empty_size;
2112 spin_lock(&ctl->tree_lock);
2113 entry = find_free_space(ctl, &offset, &bytes_search);
2118 if (entry->bitmap) {
2119 bitmap_clear_bits(ctl, entry, offset, bytes);
2121 free_bitmap(ctl, entry);
2123 unlink_free_space(ctl, entry);
2124 entry->offset += bytes;
2125 entry->bytes -= bytes;
2127 kmem_cache_free(btrfs_free_space_cachep, entry);
2129 link_free_space(ctl, entry);
2133 spin_unlock(&ctl->tree_lock);
2139 * given a cluster, put all of its extents back into the free space
2140 * cache. If a block group is passed, this function will only free
2141 * a cluster that belongs to the passed block group.
2143 * Otherwise, it'll get a reference on the block group pointed to by the
2144 * cluster and remove the cluster from it.
2146 int btrfs_return_cluster_to_free_space(
2147 struct btrfs_block_group_cache *block_group,
2148 struct btrfs_free_cluster *cluster)
2150 struct btrfs_free_space_ctl *ctl;
2153 /* first, get a safe pointer to the block group */
2154 spin_lock(&cluster->lock);
2156 block_group = cluster->block_group;
2158 spin_unlock(&cluster->lock);
2161 } else if (cluster->block_group != block_group) {
2162 /* someone else has already freed it don't redo their work */
2163 spin_unlock(&cluster->lock);
2166 atomic_inc(&block_group->count);
2167 spin_unlock(&cluster->lock);
2169 ctl = block_group->free_space_ctl;
2171 /* now return any extents the cluster had on it */
2172 spin_lock(&ctl->tree_lock);
2173 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2174 spin_unlock(&ctl->tree_lock);
2176 /* finally drop our ref */
2177 btrfs_put_block_group(block_group);
2181 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2182 struct btrfs_free_cluster *cluster,
2183 struct btrfs_free_space *entry,
2184 u64 bytes, u64 min_start)
2186 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2188 u64 search_start = cluster->window_start;
2189 u64 search_bytes = bytes;
2192 search_start = min_start;
2193 search_bytes = bytes;
2195 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2200 __bitmap_clear_bits(ctl, entry, ret, bytes);
2206 * given a cluster, try to allocate 'bytes' from it, returns 0
2207 * if it couldn't find anything suitably large, or a logical disk offset
2208 * if things worked out
2210 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2211 struct btrfs_free_cluster *cluster, u64 bytes,
2214 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2215 struct btrfs_free_space *entry = NULL;
2216 struct rb_node *node;
2219 spin_lock(&cluster->lock);
2220 if (bytes > cluster->max_size)
2223 if (cluster->block_group != block_group)
2226 node = rb_first(&cluster->root);
2230 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2232 if (entry->bytes < bytes ||
2233 (!entry->bitmap && entry->offset < min_start)) {
2234 node = rb_next(&entry->offset_index);
2237 entry = rb_entry(node, struct btrfs_free_space,
2242 if (entry->bitmap) {
2243 ret = btrfs_alloc_from_bitmap(block_group,
2244 cluster, entry, bytes,
2247 node = rb_next(&entry->offset_index);
2250 entry = rb_entry(node, struct btrfs_free_space,
2255 ret = entry->offset;
2257 entry->offset += bytes;
2258 entry->bytes -= bytes;
2261 if (entry->bytes == 0)
2262 rb_erase(&entry->offset_index, &cluster->root);
2266 spin_unlock(&cluster->lock);
2271 spin_lock(&ctl->tree_lock);
2273 ctl->free_space -= bytes;
2274 if (entry->bytes == 0) {
2275 ctl->free_extents--;
2276 if (entry->bitmap) {
2277 kfree(entry->bitmap);
2278 ctl->total_bitmaps--;
2279 ctl->op->recalc_thresholds(ctl);
2281 kmem_cache_free(btrfs_free_space_cachep, entry);
2284 spin_unlock(&ctl->tree_lock);
2289 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2290 struct btrfs_free_space *entry,
2291 struct btrfs_free_cluster *cluster,
2292 u64 offset, u64 bytes,
2293 u64 cont1_bytes, u64 min_bytes)
2295 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2296 unsigned long next_zero;
2298 unsigned long want_bits;
2299 unsigned long min_bits;
2300 unsigned long found_bits;
2301 unsigned long start = 0;
2302 unsigned long total_found = 0;
2305 i = offset_to_bit(entry->offset, block_group->sectorsize,
2306 max_t(u64, offset, entry->offset));
2307 want_bits = bytes_to_bits(bytes, block_group->sectorsize);
2308 min_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2312 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
2313 i < BITS_PER_BITMAP;
2314 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
2315 next_zero = find_next_zero_bit(entry->bitmap,
2316 BITS_PER_BITMAP, i);
2317 if (next_zero - i >= min_bits) {
2318 found_bits = next_zero - i;
2329 cluster->max_size = 0;
2332 total_found += found_bits;
2334 if (cluster->max_size < found_bits * block_group->sectorsize)
2335 cluster->max_size = found_bits * block_group->sectorsize;
2337 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2342 cluster->window_start = start * block_group->sectorsize +
2344 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2345 ret = tree_insert_offset(&cluster->root, entry->offset,
2346 &entry->offset_index, 1);
2353 * This searches the block group for just extents to fill the cluster with.
2354 * Try to find a cluster with at least bytes total bytes, at least one
2355 * extent of cont1_bytes, and other clusters of at least min_bytes.
2358 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2359 struct btrfs_free_cluster *cluster,
2360 struct list_head *bitmaps, u64 offset, u64 bytes,
2361 u64 cont1_bytes, u64 min_bytes)
2363 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2364 struct btrfs_free_space *first = NULL;
2365 struct btrfs_free_space *entry = NULL;
2366 struct btrfs_free_space *last;
2367 struct rb_node *node;
2372 entry = tree_search_offset(ctl, offset, 0, 1);
2377 * We don't want bitmaps, so just move along until we find a normal
2380 while (entry->bitmap || entry->bytes < min_bytes) {
2381 if (entry->bitmap && list_empty(&entry->list))
2382 list_add_tail(&entry->list, bitmaps);
2383 node = rb_next(&entry->offset_index);
2386 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2389 window_start = entry->offset;
2390 window_free = entry->bytes;
2391 max_extent = entry->bytes;
2395 for (node = rb_next(&entry->offset_index); node;
2396 node = rb_next(&entry->offset_index)) {
2397 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2399 if (entry->bitmap) {
2400 if (list_empty(&entry->list))
2401 list_add_tail(&entry->list, bitmaps);
2405 if (entry->bytes < min_bytes)
2409 window_free += entry->bytes;
2410 if (entry->bytes > max_extent)
2411 max_extent = entry->bytes;
2414 if (window_free < bytes || max_extent < cont1_bytes)
2417 cluster->window_start = first->offset;
2419 node = &first->offset_index;
2422 * now we've found our entries, pull them out of the free space
2423 * cache and put them into the cluster rbtree
2428 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2429 node = rb_next(&entry->offset_index);
2430 if (entry->bitmap || entry->bytes < min_bytes)
2433 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2434 ret = tree_insert_offset(&cluster->root, entry->offset,
2435 &entry->offset_index, 0);
2437 } while (node && entry != last);
2439 cluster->max_size = max_extent;
2445 * This specifically looks for bitmaps that may work in the cluster, we assume
2446 * that we have already failed to find extents that will work.
2449 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2450 struct btrfs_free_cluster *cluster,
2451 struct list_head *bitmaps, u64 offset, u64 bytes,
2452 u64 cont1_bytes, u64 min_bytes)
2454 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2455 struct btrfs_free_space *entry;
2457 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2459 if (ctl->total_bitmaps == 0)
2463 * The bitmap that covers offset won't be in the list unless offset
2464 * is just its start offset.
2466 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2467 if (entry->offset != bitmap_offset) {
2468 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2469 if (entry && list_empty(&entry->list))
2470 list_add(&entry->list, bitmaps);
2473 list_for_each_entry(entry, bitmaps, list) {
2474 if (entry->bytes < min_bytes)
2476 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2477 bytes, cont1_bytes, min_bytes);
2483 * The bitmaps list has all the bitmaps that record free space
2484 * starting after offset, so no more search is required.
2490 * here we try to find a cluster of blocks in a block group. The goal
2491 * is to find at least bytes+empty_size.
2492 * We might not find them all in one contiguous area.
2494 * returns zero and sets up cluster if things worked out, otherwise
2495 * it returns -enospc
2497 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2498 struct btrfs_root *root,
2499 struct btrfs_block_group_cache *block_group,
2500 struct btrfs_free_cluster *cluster,
2501 u64 offset, u64 bytes, u64 empty_size)
2503 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2504 struct btrfs_free_space *entry, *tmp;
2511 * Choose the minimum extent size we'll require for this
2512 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2513 * For metadata, allow allocates with smaller extents. For
2514 * data, keep it dense.
2516 if (btrfs_test_opt(root, SSD_SPREAD)) {
2517 cont1_bytes = min_bytes = bytes + empty_size;
2518 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2519 cont1_bytes = bytes;
2520 min_bytes = block_group->sectorsize;
2522 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2523 min_bytes = block_group->sectorsize;
2526 spin_lock(&ctl->tree_lock);
2529 * If we know we don't have enough space to make a cluster don't even
2530 * bother doing all the work to try and find one.
2532 if (ctl->free_space < bytes) {
2533 spin_unlock(&ctl->tree_lock);
2537 spin_lock(&cluster->lock);
2539 /* someone already found a cluster, hooray */
2540 if (cluster->block_group) {
2545 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2547 cont1_bytes, min_bytes);
2549 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2550 offset, bytes + empty_size,
2551 cont1_bytes, min_bytes);
2553 /* Clear our temporary list */
2554 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2555 list_del_init(&entry->list);
2558 atomic_inc(&block_group->count);
2559 list_add_tail(&cluster->block_group_list,
2560 &block_group->cluster_list);
2561 cluster->block_group = block_group;
2564 spin_unlock(&cluster->lock);
2565 spin_unlock(&ctl->tree_lock);
2571 * simple code to zero out a cluster
2573 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2575 spin_lock_init(&cluster->lock);
2576 spin_lock_init(&cluster->refill_lock);
2577 cluster->root = RB_ROOT;
2578 cluster->max_size = 0;
2579 INIT_LIST_HEAD(&cluster->block_group_list);
2580 cluster->block_group = NULL;
2583 static int do_trimming(struct btrfs_block_group_cache *block_group,
2584 u64 *total_trimmed, u64 start, u64 bytes,
2585 u64 reserved_start, u64 reserved_bytes)
2587 struct btrfs_space_info *space_info = block_group->space_info;
2588 struct btrfs_fs_info *fs_info = block_group->fs_info;
2593 spin_lock(&space_info->lock);
2594 spin_lock(&block_group->lock);
2595 if (!block_group->ro) {
2596 block_group->reserved += reserved_bytes;
2597 space_info->bytes_reserved += reserved_bytes;
2600 spin_unlock(&block_group->lock);
2601 spin_unlock(&space_info->lock);
2603 ret = btrfs_error_discard_extent(fs_info->extent_root,
2604 start, bytes, &trimmed);
2606 *total_trimmed += trimmed;
2608 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2611 spin_lock(&space_info->lock);
2612 spin_lock(&block_group->lock);
2613 if (block_group->ro)
2614 space_info->bytes_readonly += reserved_bytes;
2615 block_group->reserved -= reserved_bytes;
2616 space_info->bytes_reserved -= reserved_bytes;
2617 spin_unlock(&space_info->lock);
2618 spin_unlock(&block_group->lock);
2624 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2625 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2627 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2628 struct btrfs_free_space *entry;
2629 struct rb_node *node;
2635 while (start < end) {
2636 spin_lock(&ctl->tree_lock);
2638 if (ctl->free_space < minlen) {
2639 spin_unlock(&ctl->tree_lock);
2643 entry = tree_search_offset(ctl, start, 0, 1);
2645 spin_unlock(&ctl->tree_lock);
2650 while (entry->bitmap) {
2651 node = rb_next(&entry->offset_index);
2653 spin_unlock(&ctl->tree_lock);
2656 entry = rb_entry(node, struct btrfs_free_space,
2660 if (entry->offset >= end) {
2661 spin_unlock(&ctl->tree_lock);
2665 extent_start = entry->offset;
2666 extent_bytes = entry->bytes;
2667 start = max(start, extent_start);
2668 bytes = min(extent_start + extent_bytes, end) - start;
2669 if (bytes < minlen) {
2670 spin_unlock(&ctl->tree_lock);
2674 unlink_free_space(ctl, entry);
2675 kmem_cache_free(btrfs_free_space_cachep, entry);
2677 spin_unlock(&ctl->tree_lock);
2679 ret = do_trimming(block_group, total_trimmed, start, bytes,
2680 extent_start, extent_bytes);
2686 if (fatal_signal_pending(current)) {
2697 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2698 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2700 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2701 struct btrfs_free_space *entry;
2705 u64 offset = offset_to_bitmap(ctl, start);
2707 while (offset < end) {
2708 bool next_bitmap = false;
2710 spin_lock(&ctl->tree_lock);
2712 if (ctl->free_space < minlen) {
2713 spin_unlock(&ctl->tree_lock);
2717 entry = tree_search_offset(ctl, offset, 1, 0);
2719 spin_unlock(&ctl->tree_lock);
2725 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2726 if (ret2 || start >= end) {
2727 spin_unlock(&ctl->tree_lock);
2732 bytes = min(bytes, end - start);
2733 if (bytes < minlen) {
2734 spin_unlock(&ctl->tree_lock);
2738 bitmap_clear_bits(ctl, entry, start, bytes);
2739 if (entry->bytes == 0)
2740 free_bitmap(ctl, entry);
2742 spin_unlock(&ctl->tree_lock);
2744 ret = do_trimming(block_group, total_trimmed, start, bytes,
2750 offset += BITS_PER_BITMAP * ctl->unit;
2753 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2754 offset += BITS_PER_BITMAP * ctl->unit;
2757 if (fatal_signal_pending(current)) {
2768 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2769 u64 *trimmed, u64 start, u64 end, u64 minlen)
2775 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2779 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2785 * Find the left-most item in the cache tree, and then return the
2786 * smallest inode number in the item.
2788 * Note: the returned inode number may not be the smallest one in
2789 * the tree, if the left-most item is a bitmap.
2791 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2793 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2794 struct btrfs_free_space *entry = NULL;
2797 spin_lock(&ctl->tree_lock);
2799 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2802 entry = rb_entry(rb_first(&ctl->free_space_offset),
2803 struct btrfs_free_space, offset_index);
2805 if (!entry->bitmap) {
2806 ino = entry->offset;
2808 unlink_free_space(ctl, entry);
2812 kmem_cache_free(btrfs_free_space_cachep, entry);
2814 link_free_space(ctl, entry);
2820 ret = search_bitmap(ctl, entry, &offset, &count);
2824 bitmap_clear_bits(ctl, entry, offset, 1);
2825 if (entry->bytes == 0)
2826 free_bitmap(ctl, entry);
2829 spin_unlock(&ctl->tree_lock);
2834 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2835 struct btrfs_path *path)
2837 struct inode *inode = NULL;
2839 spin_lock(&root->cache_lock);
2840 if (root->cache_inode)
2841 inode = igrab(root->cache_inode);
2842 spin_unlock(&root->cache_lock);
2846 inode = __lookup_free_space_inode(root, path, 0);
2850 spin_lock(&root->cache_lock);
2851 if (!btrfs_fs_closing(root->fs_info))
2852 root->cache_inode = igrab(inode);
2853 spin_unlock(&root->cache_lock);
2858 int create_free_ino_inode(struct btrfs_root *root,
2859 struct btrfs_trans_handle *trans,
2860 struct btrfs_path *path)
2862 return __create_free_space_inode(root, trans, path,
2863 BTRFS_FREE_INO_OBJECTID, 0);
2866 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2868 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2869 struct btrfs_path *path;
2870 struct inode *inode;
2872 u64 root_gen = btrfs_root_generation(&root->root_item);
2874 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2878 * If we're unmounting then just return, since this does a search on the
2879 * normal root and not the commit root and we could deadlock.
2881 if (btrfs_fs_closing(fs_info))
2884 path = btrfs_alloc_path();
2888 inode = lookup_free_ino_inode(root, path);
2892 if (root_gen != BTRFS_I(inode)->generation)
2895 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2898 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2899 "root %llu\n", root->root_key.objectid);
2903 btrfs_free_path(path);
2907 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2908 struct btrfs_trans_handle *trans,
2909 struct btrfs_path *path)
2911 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2912 struct inode *inode;
2915 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2918 inode = lookup_free_ino_inode(root, path);
2922 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2924 btrfs_delalloc_release_metadata(inode, inode->i_size);
2926 printk(KERN_ERR "btrfs: failed to write free ino cache "
2927 "for root %llu\n", root->root_key.objectid);