2 * Copyright (C) 2008 Red Hat. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
23 #include <linux/ratelimit.h>
25 #include "free-space-cache.h"
26 #include "transaction.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
31 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
34 static int link_free_space(struct btrfs_free_space_ctl *ctl,
35 struct btrfs_free_space *info);
36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
37 struct btrfs_free_space *info);
39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
40 struct btrfs_path *path,
44 struct btrfs_key location;
45 struct btrfs_disk_key disk_key;
46 struct btrfs_free_space_header *header;
47 struct extent_buffer *leaf;
48 struct inode *inode = NULL;
51 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
55 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
59 btrfs_release_path(path);
60 return ERR_PTR(-ENOENT);
63 leaf = path->nodes[0];
64 header = btrfs_item_ptr(leaf, path->slots[0],
65 struct btrfs_free_space_header);
66 btrfs_free_space_key(leaf, header, &disk_key);
67 btrfs_disk_key_to_cpu(&location, &disk_key);
68 btrfs_release_path(path);
70 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
72 return ERR_PTR(-ENOENT);
75 if (is_bad_inode(inode)) {
77 return ERR_PTR(-ENOENT);
80 mapping_set_gfp_mask(inode->i_mapping,
81 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
86 struct inode *lookup_free_space_inode(struct btrfs_root *root,
87 struct btrfs_block_group_cache
88 *block_group, struct btrfs_path *path)
90 struct inode *inode = NULL;
91 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
93 spin_lock(&block_group->lock);
94 if (block_group->inode)
95 inode = igrab(block_group->inode);
96 spin_unlock(&block_group->lock);
100 inode = __lookup_free_space_inode(root, path,
101 block_group->key.objectid);
105 spin_lock(&block_group->lock);
106 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
107 printk(KERN_INFO "Old style space inode found, converting.\n");
108 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
109 BTRFS_INODE_NODATACOW;
110 block_group->disk_cache_state = BTRFS_DC_CLEAR;
113 if (!block_group->iref) {
114 block_group->inode = igrab(inode);
115 block_group->iref = 1;
117 spin_unlock(&block_group->lock);
122 int __create_free_space_inode(struct btrfs_root *root,
123 struct btrfs_trans_handle *trans,
124 struct btrfs_path *path, u64 ino, u64 offset)
126 struct btrfs_key key;
127 struct btrfs_disk_key disk_key;
128 struct btrfs_free_space_header *header;
129 struct btrfs_inode_item *inode_item;
130 struct extent_buffer *leaf;
131 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
134 ret = btrfs_insert_empty_inode(trans, root, path, ino);
138 /* We inline crc's for the free disk space cache */
139 if (ino != BTRFS_FREE_INO_OBJECTID)
140 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
142 leaf = path->nodes[0];
143 inode_item = btrfs_item_ptr(leaf, path->slots[0],
144 struct btrfs_inode_item);
145 btrfs_item_key(leaf, &disk_key, path->slots[0]);
146 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
147 sizeof(*inode_item));
148 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
149 btrfs_set_inode_size(leaf, inode_item, 0);
150 btrfs_set_inode_nbytes(leaf, inode_item, 0);
151 btrfs_set_inode_uid(leaf, inode_item, 0);
152 btrfs_set_inode_gid(leaf, inode_item, 0);
153 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
154 btrfs_set_inode_flags(leaf, inode_item, flags);
155 btrfs_set_inode_nlink(leaf, inode_item, 1);
156 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
157 btrfs_set_inode_block_group(leaf, inode_item, offset);
158 btrfs_mark_buffer_dirty(leaf);
159 btrfs_release_path(path);
161 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
165 ret = btrfs_insert_empty_item(trans, root, path, &key,
166 sizeof(struct btrfs_free_space_header));
168 btrfs_release_path(path);
171 leaf = path->nodes[0];
172 header = btrfs_item_ptr(leaf, path->slots[0],
173 struct btrfs_free_space_header);
174 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
175 btrfs_set_free_space_key(leaf, header, &disk_key);
176 btrfs_mark_buffer_dirty(leaf);
177 btrfs_release_path(path);
182 int create_free_space_inode(struct btrfs_root *root,
183 struct btrfs_trans_handle *trans,
184 struct btrfs_block_group_cache *block_group,
185 struct btrfs_path *path)
190 ret = btrfs_find_free_objectid(root, &ino);
194 return __create_free_space_inode(root, trans, path, ino,
195 block_group->key.objectid);
198 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
199 struct btrfs_trans_handle *trans,
200 struct btrfs_path *path,
203 struct btrfs_block_rsv *rsv;
208 rsv = trans->block_rsv;
209 trans->block_rsv = &root->fs_info->global_block_rsv;
211 /* 1 for slack space, 1 for updating the inode */
212 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
213 btrfs_calc_trans_metadata_size(root, 1);
215 spin_lock(&trans->block_rsv->lock);
216 if (trans->block_rsv->reserved < needed_bytes) {
217 spin_unlock(&trans->block_rsv->lock);
218 trans->block_rsv = rsv;
221 spin_unlock(&trans->block_rsv->lock);
223 oldsize = i_size_read(inode);
224 btrfs_i_size_write(inode, 0);
225 truncate_pagecache(inode, oldsize, 0);
228 * We don't need an orphan item because truncating the free space cache
229 * will never be split across transactions.
231 ret = btrfs_truncate_inode_items(trans, root, inode,
232 0, BTRFS_EXTENT_DATA_KEY);
235 trans->block_rsv = rsv;
236 btrfs_abort_transaction(trans, root, ret);
240 ret = btrfs_update_inode(trans, root, inode);
242 btrfs_abort_transaction(trans, root, ret);
243 trans->block_rsv = rsv;
248 static int readahead_cache(struct inode *inode)
250 struct file_ra_state *ra;
251 unsigned long last_index;
253 ra = kzalloc(sizeof(*ra), GFP_NOFS);
257 file_ra_state_init(ra, inode->i_mapping);
258 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
260 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
271 struct btrfs_root *root;
275 unsigned check_crcs:1;
278 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
279 struct btrfs_root *root)
281 memset(io_ctl, 0, sizeof(struct io_ctl));
282 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
284 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
289 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
290 io_ctl->check_crcs = 1;
294 static void io_ctl_free(struct io_ctl *io_ctl)
296 kfree(io_ctl->pages);
299 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
302 kunmap(io_ctl->page);
308 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
310 BUG_ON(io_ctl->index >= io_ctl->num_pages);
311 io_ctl->page = io_ctl->pages[io_ctl->index++];
312 io_ctl->cur = kmap(io_ctl->page);
313 io_ctl->orig = io_ctl->cur;
314 io_ctl->size = PAGE_CACHE_SIZE;
316 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
319 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
323 io_ctl_unmap_page(io_ctl);
325 for (i = 0; i < io_ctl->num_pages; i++) {
326 if (io_ctl->pages[i]) {
327 ClearPageChecked(io_ctl->pages[i]);
328 unlock_page(io_ctl->pages[i]);
329 page_cache_release(io_ctl->pages[i]);
334 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
338 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
341 for (i = 0; i < io_ctl->num_pages; i++) {
342 page = find_or_create_page(inode->i_mapping, i, mask);
344 io_ctl_drop_pages(io_ctl);
347 io_ctl->pages[i] = page;
348 if (uptodate && !PageUptodate(page)) {
349 btrfs_readpage(NULL, page);
351 if (!PageUptodate(page)) {
352 printk(KERN_ERR "btrfs: error reading free "
354 io_ctl_drop_pages(io_ctl);
360 for (i = 0; i < io_ctl->num_pages; i++) {
361 clear_page_dirty_for_io(io_ctl->pages[i]);
362 set_page_extent_mapped(io_ctl->pages[i]);
368 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
372 io_ctl_map_page(io_ctl, 1);
375 * Skip the csum areas. If we don't check crcs then we just have a
376 * 64bit chunk at the front of the first page.
378 if (io_ctl->check_crcs) {
379 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
380 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
382 io_ctl->cur += sizeof(u64);
383 io_ctl->size -= sizeof(u64) * 2;
387 *val = cpu_to_le64(generation);
388 io_ctl->cur += sizeof(u64);
391 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
396 * Skip the crc area. If we don't check crcs then we just have a 64bit
397 * chunk at the front of the first page.
399 if (io_ctl->check_crcs) {
400 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
401 io_ctl->size -= sizeof(u64) +
402 (sizeof(u32) * io_ctl->num_pages);
404 io_ctl->cur += sizeof(u64);
405 io_ctl->size -= sizeof(u64) * 2;
409 if (le64_to_cpu(*gen) != generation) {
410 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
411 "(%Lu) does not match inode (%Lu)\n", *gen,
413 io_ctl_unmap_page(io_ctl);
416 io_ctl->cur += sizeof(u64);
420 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
426 if (!io_ctl->check_crcs) {
427 io_ctl_unmap_page(io_ctl);
432 offset = sizeof(u32) * io_ctl->num_pages;
434 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
435 PAGE_CACHE_SIZE - offset);
436 btrfs_csum_final(crc, (char *)&crc);
437 io_ctl_unmap_page(io_ctl);
438 tmp = kmap(io_ctl->pages[0]);
441 kunmap(io_ctl->pages[0]);
444 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
450 if (!io_ctl->check_crcs) {
451 io_ctl_map_page(io_ctl, 0);
456 offset = sizeof(u32) * io_ctl->num_pages;
458 tmp = kmap(io_ctl->pages[0]);
461 kunmap(io_ctl->pages[0]);
463 io_ctl_map_page(io_ctl, 0);
464 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
465 PAGE_CACHE_SIZE - offset);
466 btrfs_csum_final(crc, (char *)&crc);
468 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
470 io_ctl_unmap_page(io_ctl);
477 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
480 struct btrfs_free_space_entry *entry;
486 entry->offset = cpu_to_le64(offset);
487 entry->bytes = cpu_to_le64(bytes);
488 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
489 BTRFS_FREE_SPACE_EXTENT;
490 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
491 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
493 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
496 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
498 /* No more pages to map */
499 if (io_ctl->index >= io_ctl->num_pages)
502 /* map the next page */
503 io_ctl_map_page(io_ctl, 1);
507 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
513 * If we aren't at the start of the current page, unmap this one and
514 * map the next one if there is any left.
516 if (io_ctl->cur != io_ctl->orig) {
517 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
518 if (io_ctl->index >= io_ctl->num_pages)
520 io_ctl_map_page(io_ctl, 0);
523 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
524 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
525 if (io_ctl->index < io_ctl->num_pages)
526 io_ctl_map_page(io_ctl, 0);
530 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
533 * If we're not on the boundary we know we've modified the page and we
534 * need to crc the page.
536 if (io_ctl->cur != io_ctl->orig)
537 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
539 io_ctl_unmap_page(io_ctl);
541 while (io_ctl->index < io_ctl->num_pages) {
542 io_ctl_map_page(io_ctl, 1);
543 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
547 static int io_ctl_read_entry(struct io_ctl *io_ctl,
548 struct btrfs_free_space *entry, u8 *type)
550 struct btrfs_free_space_entry *e;
554 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
560 entry->offset = le64_to_cpu(e->offset);
561 entry->bytes = le64_to_cpu(e->bytes);
563 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
564 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
566 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
569 io_ctl_unmap_page(io_ctl);
574 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
575 struct btrfs_free_space *entry)
579 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
583 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
584 io_ctl_unmap_page(io_ctl);
590 * Since we attach pinned extents after the fact we can have contiguous sections
591 * of free space that are split up in entries. This poses a problem with the
592 * tree logging stuff since it could have allocated across what appears to be 2
593 * entries since we would have merged the entries when adding the pinned extents
594 * back to the free space cache. So run through the space cache that we just
595 * loaded and merge contiguous entries. This will make the log replay stuff not
596 * blow up and it will make for nicer allocator behavior.
598 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
600 struct btrfs_free_space *e, *prev = NULL;
604 spin_lock(&ctl->tree_lock);
605 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
606 e = rb_entry(n, struct btrfs_free_space, offset_index);
609 if (e->bitmap || prev->bitmap)
611 if (prev->offset + prev->bytes == e->offset) {
612 unlink_free_space(ctl, prev);
613 unlink_free_space(ctl, e);
614 prev->bytes += e->bytes;
615 kmem_cache_free(btrfs_free_space_cachep, e);
616 link_free_space(ctl, prev);
618 spin_unlock(&ctl->tree_lock);
624 spin_unlock(&ctl->tree_lock);
627 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
628 struct btrfs_free_space_ctl *ctl,
629 struct btrfs_path *path, u64 offset)
631 struct btrfs_free_space_header *header;
632 struct extent_buffer *leaf;
633 struct io_ctl io_ctl;
634 struct btrfs_key key;
635 struct btrfs_free_space *e, *n;
636 struct list_head bitmaps;
643 INIT_LIST_HEAD(&bitmaps);
645 /* Nothing in the space cache, goodbye */
646 if (!i_size_read(inode))
649 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
653 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
657 btrfs_release_path(path);
663 leaf = path->nodes[0];
664 header = btrfs_item_ptr(leaf, path->slots[0],
665 struct btrfs_free_space_header);
666 num_entries = btrfs_free_space_entries(leaf, header);
667 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
668 generation = btrfs_free_space_generation(leaf, header);
669 btrfs_release_path(path);
671 if (BTRFS_I(inode)->generation != generation) {
672 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
673 " not match free space cache generation (%llu)\n",
674 (unsigned long long)BTRFS_I(inode)->generation,
675 (unsigned long long)generation);
682 ret = io_ctl_init(&io_ctl, inode, root);
686 ret = readahead_cache(inode);
690 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
694 ret = io_ctl_check_crc(&io_ctl, 0);
698 ret = io_ctl_check_generation(&io_ctl, generation);
702 while (num_entries) {
703 e = kmem_cache_zalloc(btrfs_free_space_cachep,
708 ret = io_ctl_read_entry(&io_ctl, e, &type);
710 kmem_cache_free(btrfs_free_space_cachep, e);
715 kmem_cache_free(btrfs_free_space_cachep, e);
719 if (type == BTRFS_FREE_SPACE_EXTENT) {
720 spin_lock(&ctl->tree_lock);
721 ret = link_free_space(ctl, e);
722 spin_unlock(&ctl->tree_lock);
724 printk(KERN_ERR "Duplicate entries in "
725 "free space cache, dumping\n");
726 kmem_cache_free(btrfs_free_space_cachep, e);
730 BUG_ON(!num_bitmaps);
732 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
735 btrfs_free_space_cachep, e);
738 spin_lock(&ctl->tree_lock);
739 ret = link_free_space(ctl, e);
740 ctl->total_bitmaps++;
741 ctl->op->recalc_thresholds(ctl);
742 spin_unlock(&ctl->tree_lock);
744 printk(KERN_ERR "Duplicate entries in "
745 "free space cache, dumping\n");
746 kmem_cache_free(btrfs_free_space_cachep, e);
749 list_add_tail(&e->list, &bitmaps);
755 io_ctl_unmap_page(&io_ctl);
758 * We add the bitmaps at the end of the entries in order that
759 * the bitmap entries are added to the cache.
761 list_for_each_entry_safe(e, n, &bitmaps, list) {
762 list_del_init(&e->list);
763 ret = io_ctl_read_bitmap(&io_ctl, e);
768 io_ctl_drop_pages(&io_ctl);
769 merge_space_tree(ctl);
772 io_ctl_free(&io_ctl);
775 io_ctl_drop_pages(&io_ctl);
776 __btrfs_remove_free_space_cache(ctl);
780 int load_free_space_cache(struct btrfs_fs_info *fs_info,
781 struct btrfs_block_group_cache *block_group)
783 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
784 struct btrfs_root *root = fs_info->tree_root;
786 struct btrfs_path *path;
789 u64 used = btrfs_block_group_used(&block_group->item);
792 * If this block group has been marked to be cleared for one reason or
793 * another then we can't trust the on disk cache, so just return.
795 spin_lock(&block_group->lock);
796 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
797 spin_unlock(&block_group->lock);
800 spin_unlock(&block_group->lock);
802 path = btrfs_alloc_path();
805 path->search_commit_root = 1;
806 path->skip_locking = 1;
808 inode = lookup_free_space_inode(root, block_group, path);
810 btrfs_free_path(path);
814 /* We may have converted the inode and made the cache invalid. */
815 spin_lock(&block_group->lock);
816 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
817 spin_unlock(&block_group->lock);
818 btrfs_free_path(path);
821 spin_unlock(&block_group->lock);
823 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
824 path, block_group->key.objectid);
825 btrfs_free_path(path);
829 spin_lock(&ctl->tree_lock);
830 matched = (ctl->free_space == (block_group->key.offset - used -
831 block_group->bytes_super));
832 spin_unlock(&ctl->tree_lock);
835 __btrfs_remove_free_space_cache(ctl);
836 printk(KERN_ERR "block group %llu has an wrong amount of free "
837 "space\n", block_group->key.objectid);
842 /* This cache is bogus, make sure it gets cleared */
843 spin_lock(&block_group->lock);
844 block_group->disk_cache_state = BTRFS_DC_CLEAR;
845 spin_unlock(&block_group->lock);
848 printk(KERN_ERR "btrfs: failed to load free space cache "
849 "for block group %llu\n", block_group->key.objectid);
857 * __btrfs_write_out_cache - write out cached info to an inode
858 * @root - the root the inode belongs to
859 * @ctl - the free space cache we are going to write out
860 * @block_group - the block_group for this cache if it belongs to a block_group
861 * @trans - the trans handle
862 * @path - the path to use
863 * @offset - the offset for the key we'll insert
865 * This function writes out a free space cache struct to disk for quick recovery
866 * on mount. This will return 0 if it was successfull in writing the cache out,
867 * and -1 if it was not.
869 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
870 struct btrfs_free_space_ctl *ctl,
871 struct btrfs_block_group_cache *block_group,
872 struct btrfs_trans_handle *trans,
873 struct btrfs_path *path, u64 offset)
875 struct btrfs_free_space_header *header;
876 struct extent_buffer *leaf;
877 struct rb_node *node;
878 struct list_head *pos, *n;
879 struct extent_state *cached_state = NULL;
880 struct btrfs_free_cluster *cluster = NULL;
881 struct extent_io_tree *unpin = NULL;
882 struct io_ctl io_ctl;
883 struct list_head bitmap_list;
884 struct btrfs_key key;
885 u64 start, extent_start, extent_end, len;
891 INIT_LIST_HEAD(&bitmap_list);
893 if (!i_size_read(inode))
896 ret = io_ctl_init(&io_ctl, inode, root);
900 /* Get the cluster for this block_group if it exists */
901 if (block_group && !list_empty(&block_group->cluster_list))
902 cluster = list_entry(block_group->cluster_list.next,
903 struct btrfs_free_cluster,
906 /* Lock all pages first so we can lock the extent safely. */
907 io_ctl_prepare_pages(&io_ctl, inode, 0);
909 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
912 node = rb_first(&ctl->free_space_offset);
913 if (!node && cluster) {
914 node = rb_first(&cluster->root);
918 /* Make sure we can fit our crcs into the first page */
919 if (io_ctl.check_crcs &&
920 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
925 io_ctl_set_generation(&io_ctl, trans->transid);
927 /* Write out the extent entries */
929 struct btrfs_free_space *e;
931 e = rb_entry(node, struct btrfs_free_space, offset_index);
934 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
940 list_add_tail(&e->list, &bitmap_list);
943 node = rb_next(node);
944 if (!node && cluster) {
945 node = rb_first(&cluster->root);
951 * We want to add any pinned extents to our free space cache
952 * so we don't leak the space
956 * We shouldn't have switched the pinned extents yet so this is the
959 unpin = root->fs_info->pinned_extents;
962 start = block_group->key.objectid;
964 while (block_group && (start < block_group->key.objectid +
965 block_group->key.offset)) {
966 ret = find_first_extent_bit(unpin, start,
967 &extent_start, &extent_end,
974 /* This pinned extent is out of our range */
975 if (extent_start >= block_group->key.objectid +
976 block_group->key.offset)
979 extent_start = max(extent_start, start);
980 extent_end = min(block_group->key.objectid +
981 block_group->key.offset, extent_end + 1);
982 len = extent_end - extent_start;
985 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
992 /* Write out the bitmaps */
993 list_for_each_safe(pos, n, &bitmap_list) {
994 struct btrfs_free_space *entry =
995 list_entry(pos, struct btrfs_free_space, list);
997 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
1000 list_del_init(&entry->list);
1003 /* Zero out the rest of the pages just to make sure */
1004 io_ctl_zero_remaining_pages(&io_ctl);
1006 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1007 0, i_size_read(inode), &cached_state);
1008 io_ctl_drop_pages(&io_ctl);
1009 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1010 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1016 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1018 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1019 key.offset = offset;
1022 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1024 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1025 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1029 leaf = path->nodes[0];
1031 struct btrfs_key found_key;
1032 BUG_ON(!path->slots[0]);
1034 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1035 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1036 found_key.offset != offset) {
1037 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1039 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1041 btrfs_release_path(path);
1046 BTRFS_I(inode)->generation = trans->transid;
1047 header = btrfs_item_ptr(leaf, path->slots[0],
1048 struct btrfs_free_space_header);
1049 btrfs_set_free_space_entries(leaf, header, entries);
1050 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1051 btrfs_set_free_space_generation(leaf, header, trans->transid);
1052 btrfs_mark_buffer_dirty(leaf);
1053 btrfs_release_path(path);
1057 io_ctl_free(&io_ctl);
1059 invalidate_inode_pages2(inode->i_mapping);
1060 BTRFS_I(inode)->generation = 0;
1062 btrfs_update_inode(trans, root, inode);
1066 list_for_each_safe(pos, n, &bitmap_list) {
1067 struct btrfs_free_space *entry =
1068 list_entry(pos, struct btrfs_free_space, list);
1069 list_del_init(&entry->list);
1071 io_ctl_drop_pages(&io_ctl);
1072 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1073 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1077 int btrfs_write_out_cache(struct btrfs_root *root,
1078 struct btrfs_trans_handle *trans,
1079 struct btrfs_block_group_cache *block_group,
1080 struct btrfs_path *path)
1082 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1083 struct inode *inode;
1086 root = root->fs_info->tree_root;
1088 spin_lock(&block_group->lock);
1089 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1090 spin_unlock(&block_group->lock);
1093 spin_unlock(&block_group->lock);
1095 inode = lookup_free_space_inode(root, block_group, path);
1099 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1100 path, block_group->key.objectid);
1102 spin_lock(&block_group->lock);
1103 block_group->disk_cache_state = BTRFS_DC_ERROR;
1104 spin_unlock(&block_group->lock);
1107 printk(KERN_ERR "btrfs: failed to write free space cache "
1108 "for block group %llu\n", block_group->key.objectid);
1116 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1119 BUG_ON(offset < bitmap_start);
1120 offset -= bitmap_start;
1121 return (unsigned long)(div_u64(offset, unit));
1124 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1126 return (unsigned long)(div_u64(bytes, unit));
1129 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1133 u64 bytes_per_bitmap;
1135 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1136 bitmap_start = offset - ctl->start;
1137 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1138 bitmap_start *= bytes_per_bitmap;
1139 bitmap_start += ctl->start;
1141 return bitmap_start;
1144 static int tree_insert_offset(struct rb_root *root, u64 offset,
1145 struct rb_node *node, int bitmap)
1147 struct rb_node **p = &root->rb_node;
1148 struct rb_node *parent = NULL;
1149 struct btrfs_free_space *info;
1153 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1155 if (offset < info->offset) {
1157 } else if (offset > info->offset) {
1158 p = &(*p)->rb_right;
1161 * we could have a bitmap entry and an extent entry
1162 * share the same offset. If this is the case, we want
1163 * the extent entry to always be found first if we do a
1164 * linear search through the tree, since we want to have
1165 * the quickest allocation time, and allocating from an
1166 * extent is faster than allocating from a bitmap. So
1167 * if we're inserting a bitmap and we find an entry at
1168 * this offset, we want to go right, or after this entry
1169 * logically. If we are inserting an extent and we've
1170 * found a bitmap, we want to go left, or before
1178 p = &(*p)->rb_right;
1180 if (!info->bitmap) {
1189 rb_link_node(node, parent, p);
1190 rb_insert_color(node, root);
1196 * searches the tree for the given offset.
1198 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1199 * want a section that has at least bytes size and comes at or after the given
1202 static struct btrfs_free_space *
1203 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1204 u64 offset, int bitmap_only, int fuzzy)
1206 struct rb_node *n = ctl->free_space_offset.rb_node;
1207 struct btrfs_free_space *entry, *prev = NULL;
1209 /* find entry that is closest to the 'offset' */
1216 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1219 if (offset < entry->offset)
1221 else if (offset > entry->offset)
1234 * bitmap entry and extent entry may share same offset,
1235 * in that case, bitmap entry comes after extent entry.
1240 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1241 if (entry->offset != offset)
1244 WARN_ON(!entry->bitmap);
1247 if (entry->bitmap) {
1249 * if previous extent entry covers the offset,
1250 * we should return it instead of the bitmap entry
1252 n = rb_prev(&entry->offset_index);
1254 prev = rb_entry(n, struct btrfs_free_space,
1256 if (!prev->bitmap &&
1257 prev->offset + prev->bytes > offset)
1267 /* find last entry before the 'offset' */
1269 if (entry->offset > offset) {
1270 n = rb_prev(&entry->offset_index);
1272 entry = rb_entry(n, struct btrfs_free_space,
1274 BUG_ON(entry->offset > offset);
1283 if (entry->bitmap) {
1284 n = rb_prev(&entry->offset_index);
1286 prev = rb_entry(n, struct btrfs_free_space,
1288 if (!prev->bitmap &&
1289 prev->offset + prev->bytes > offset)
1292 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1294 } else if (entry->offset + entry->bytes > offset)
1301 if (entry->bitmap) {
1302 if (entry->offset + BITS_PER_BITMAP *
1306 if (entry->offset + entry->bytes > offset)
1310 n = rb_next(&entry->offset_index);
1313 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1319 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1320 struct btrfs_free_space *info)
1322 rb_erase(&info->offset_index, &ctl->free_space_offset);
1323 ctl->free_extents--;
1326 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1327 struct btrfs_free_space *info)
1329 __unlink_free_space(ctl, info);
1330 ctl->free_space -= info->bytes;
1333 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1334 struct btrfs_free_space *info)
1338 BUG_ON(!info->bitmap && !info->bytes);
1339 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1340 &info->offset_index, (info->bitmap != NULL));
1344 ctl->free_space += info->bytes;
1345 ctl->free_extents++;
1349 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1351 struct btrfs_block_group_cache *block_group = ctl->private;
1355 u64 size = block_group->key.offset;
1356 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1357 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1359 BUG_ON(ctl->total_bitmaps > max_bitmaps);
1362 * The goal is to keep the total amount of memory used per 1gb of space
1363 * at or below 32k, so we need to adjust how much memory we allow to be
1364 * used by extent based free space tracking
1366 if (size < 1024 * 1024 * 1024)
1367 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1369 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1370 div64_u64(size, 1024 * 1024 * 1024);
1373 * we want to account for 1 more bitmap than what we have so we can make
1374 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1375 * we add more bitmaps.
1377 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1379 if (bitmap_bytes >= max_bytes) {
1380 ctl->extents_thresh = 0;
1385 * we want the extent entry threshold to always be at most 1/2 the maxw
1386 * bytes we can have, or whatever is less than that.
1388 extent_bytes = max_bytes - bitmap_bytes;
1389 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1391 ctl->extents_thresh =
1392 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1395 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1396 struct btrfs_free_space *info,
1397 u64 offset, u64 bytes)
1399 unsigned long start, count;
1401 start = offset_to_bit(info->offset, ctl->unit, offset);
1402 count = bytes_to_bits(bytes, ctl->unit);
1403 BUG_ON(start + count > BITS_PER_BITMAP);
1405 bitmap_clear(info->bitmap, start, count);
1407 info->bytes -= bytes;
1410 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1411 struct btrfs_free_space *info, u64 offset,
1414 __bitmap_clear_bits(ctl, info, offset, bytes);
1415 ctl->free_space -= bytes;
1418 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1419 struct btrfs_free_space *info, u64 offset,
1422 unsigned long start, count;
1424 start = offset_to_bit(info->offset, ctl->unit, offset);
1425 count = bytes_to_bits(bytes, ctl->unit);
1426 BUG_ON(start + count > BITS_PER_BITMAP);
1428 bitmap_set(info->bitmap, start, count);
1430 info->bytes += bytes;
1431 ctl->free_space += bytes;
1434 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1435 struct btrfs_free_space *bitmap_info, u64 *offset,
1438 unsigned long found_bits = 0;
1439 unsigned long bits, i;
1440 unsigned long next_zero;
1442 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1443 max_t(u64, *offset, bitmap_info->offset));
1444 bits = bytes_to_bits(*bytes, ctl->unit);
1446 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1447 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1448 BITS_PER_BITMAP, i);
1449 if ((next_zero - i) >= bits) {
1450 found_bits = next_zero - i;
1457 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1458 *bytes = (u64)(found_bits) * ctl->unit;
1465 static struct btrfs_free_space *
1466 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1468 struct btrfs_free_space *entry;
1469 struct rb_node *node;
1472 if (!ctl->free_space_offset.rb_node)
1475 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1479 for (node = &entry->offset_index; node; node = rb_next(node)) {
1480 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1481 if (entry->bytes < *bytes)
1484 if (entry->bitmap) {
1485 ret = search_bitmap(ctl, entry, offset, bytes);
1491 *offset = entry->offset;
1492 *bytes = entry->bytes;
1499 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1500 struct btrfs_free_space *info, u64 offset)
1502 info->offset = offset_to_bitmap(ctl, offset);
1504 INIT_LIST_HEAD(&info->list);
1505 link_free_space(ctl, info);
1506 ctl->total_bitmaps++;
1508 ctl->op->recalc_thresholds(ctl);
1511 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1512 struct btrfs_free_space *bitmap_info)
1514 unlink_free_space(ctl, bitmap_info);
1515 kfree(bitmap_info->bitmap);
1516 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1517 ctl->total_bitmaps--;
1518 ctl->op->recalc_thresholds(ctl);
1521 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1522 struct btrfs_free_space *bitmap_info,
1523 u64 *offset, u64 *bytes)
1526 u64 search_start, search_bytes;
1530 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1533 * We need to search for bits in this bitmap. We could only cover some
1534 * of the extent in this bitmap thanks to how we add space, so we need
1535 * to search for as much as it as we can and clear that amount, and then
1536 * go searching for the next bit.
1538 search_start = *offset;
1539 search_bytes = ctl->unit;
1540 search_bytes = min(search_bytes, end - search_start + 1);
1541 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1542 BUG_ON(ret < 0 || search_start != *offset);
1544 /* We may have found more bits than what we need */
1545 search_bytes = min(search_bytes, *bytes);
1547 /* Cannot clear past the end of the bitmap */
1548 search_bytes = min(search_bytes, end - search_start + 1);
1550 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1551 *offset += search_bytes;
1552 *bytes -= search_bytes;
1555 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1556 if (!bitmap_info->bytes)
1557 free_bitmap(ctl, bitmap_info);
1560 * no entry after this bitmap, but we still have bytes to
1561 * remove, so something has gone wrong.
1566 bitmap_info = rb_entry(next, struct btrfs_free_space,
1570 * if the next entry isn't a bitmap we need to return to let the
1571 * extent stuff do its work.
1573 if (!bitmap_info->bitmap)
1577 * Ok the next item is a bitmap, but it may not actually hold
1578 * the information for the rest of this free space stuff, so
1579 * look for it, and if we don't find it return so we can try
1580 * everything over again.
1582 search_start = *offset;
1583 search_bytes = ctl->unit;
1584 ret = search_bitmap(ctl, bitmap_info, &search_start,
1586 if (ret < 0 || search_start != *offset)
1590 } else if (!bitmap_info->bytes)
1591 free_bitmap(ctl, bitmap_info);
1596 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1597 struct btrfs_free_space *info, u64 offset,
1600 u64 bytes_to_set = 0;
1603 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1605 bytes_to_set = min(end - offset, bytes);
1607 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1609 return bytes_to_set;
1613 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1614 struct btrfs_free_space *info)
1616 struct btrfs_block_group_cache *block_group = ctl->private;
1619 * If we are below the extents threshold then we can add this as an
1620 * extent, and don't have to deal with the bitmap
1622 if (ctl->free_extents < ctl->extents_thresh) {
1624 * If this block group has some small extents we don't want to
1625 * use up all of our free slots in the cache with them, we want
1626 * to reserve them to larger extents, however if we have plent
1627 * of cache left then go ahead an dadd them, no sense in adding
1628 * the overhead of a bitmap if we don't have to.
1630 if (info->bytes <= block_group->sectorsize * 4) {
1631 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1639 * some block groups are so tiny they can't be enveloped by a bitmap, so
1640 * don't even bother to create a bitmap for this
1642 if (BITS_PER_BITMAP * ctl->unit > block_group->key.offset)
1648 static struct btrfs_free_space_op free_space_op = {
1649 .recalc_thresholds = recalculate_thresholds,
1650 .use_bitmap = use_bitmap,
1653 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1654 struct btrfs_free_space *info)
1656 struct btrfs_free_space *bitmap_info;
1657 struct btrfs_block_group_cache *block_group = NULL;
1659 u64 bytes, offset, bytes_added;
1662 bytes = info->bytes;
1663 offset = info->offset;
1665 if (!ctl->op->use_bitmap(ctl, info))
1668 if (ctl->op == &free_space_op)
1669 block_group = ctl->private;
1672 * Since we link bitmaps right into the cluster we need to see if we
1673 * have a cluster here, and if so and it has our bitmap we need to add
1674 * the free space to that bitmap.
1676 if (block_group && !list_empty(&block_group->cluster_list)) {
1677 struct btrfs_free_cluster *cluster;
1678 struct rb_node *node;
1679 struct btrfs_free_space *entry;
1681 cluster = list_entry(block_group->cluster_list.next,
1682 struct btrfs_free_cluster,
1684 spin_lock(&cluster->lock);
1685 node = rb_first(&cluster->root);
1687 spin_unlock(&cluster->lock);
1688 goto no_cluster_bitmap;
1691 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1692 if (!entry->bitmap) {
1693 spin_unlock(&cluster->lock);
1694 goto no_cluster_bitmap;
1697 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1698 bytes_added = add_bytes_to_bitmap(ctl, entry,
1700 bytes -= bytes_added;
1701 offset += bytes_added;
1703 spin_unlock(&cluster->lock);
1711 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1718 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1719 bytes -= bytes_added;
1720 offset += bytes_added;
1730 if (info && info->bitmap) {
1731 add_new_bitmap(ctl, info, offset);
1736 spin_unlock(&ctl->tree_lock);
1738 /* no pre-allocated info, allocate a new one */
1740 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1743 spin_lock(&ctl->tree_lock);
1749 /* allocate the bitmap */
1750 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1751 spin_lock(&ctl->tree_lock);
1752 if (!info->bitmap) {
1762 kfree(info->bitmap);
1763 kmem_cache_free(btrfs_free_space_cachep, info);
1769 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1770 struct btrfs_free_space *info, bool update_stat)
1772 struct btrfs_free_space *left_info;
1773 struct btrfs_free_space *right_info;
1774 bool merged = false;
1775 u64 offset = info->offset;
1776 u64 bytes = info->bytes;
1779 * first we want to see if there is free space adjacent to the range we
1780 * are adding, if there is remove that struct and add a new one to
1781 * cover the entire range
1783 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1784 if (right_info && rb_prev(&right_info->offset_index))
1785 left_info = rb_entry(rb_prev(&right_info->offset_index),
1786 struct btrfs_free_space, offset_index);
1788 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1790 if (right_info && !right_info->bitmap) {
1792 unlink_free_space(ctl, right_info);
1794 __unlink_free_space(ctl, right_info);
1795 info->bytes += right_info->bytes;
1796 kmem_cache_free(btrfs_free_space_cachep, right_info);
1800 if (left_info && !left_info->bitmap &&
1801 left_info->offset + left_info->bytes == offset) {
1803 unlink_free_space(ctl, left_info);
1805 __unlink_free_space(ctl, left_info);
1806 info->offset = left_info->offset;
1807 info->bytes += left_info->bytes;
1808 kmem_cache_free(btrfs_free_space_cachep, left_info);
1815 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1816 u64 offset, u64 bytes)
1818 struct btrfs_free_space *info;
1821 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1825 info->offset = offset;
1826 info->bytes = bytes;
1828 spin_lock(&ctl->tree_lock);
1830 if (try_merge_free_space(ctl, info, true))
1834 * There was no extent directly to the left or right of this new
1835 * extent then we know we're going to have to allocate a new extent, so
1836 * before we do that see if we need to drop this into a bitmap
1838 ret = insert_into_bitmap(ctl, info);
1846 ret = link_free_space(ctl, info);
1848 kmem_cache_free(btrfs_free_space_cachep, info);
1850 spin_unlock(&ctl->tree_lock);
1853 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1854 BUG_ON(ret == -EEXIST);
1860 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1861 u64 offset, u64 bytes)
1863 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1864 struct btrfs_free_space *info;
1866 bool re_search = false;
1868 spin_lock(&ctl->tree_lock);
1875 info = tree_search_offset(ctl, offset, 0, 0);
1878 * oops didn't find an extent that matched the space we wanted
1879 * to remove, look for a bitmap instead
1881 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1885 * If we found a partial bit of our free space in a
1886 * bitmap but then couldn't find the other part this may
1887 * be a problem, so WARN about it.
1895 if (!info->bitmap) {
1896 unlink_free_space(ctl, info);
1897 if (offset == info->offset) {
1898 u64 to_free = min(bytes, info->bytes);
1900 info->bytes -= to_free;
1901 info->offset += to_free;
1903 ret = link_free_space(ctl, info);
1906 kmem_cache_free(btrfs_free_space_cachep, info);
1913 u64 old_end = info->bytes + info->offset;
1915 info->bytes = offset - info->offset;
1916 ret = link_free_space(ctl, info);
1921 /* Not enough bytes in this entry to satisfy us */
1922 if (old_end < offset + bytes) {
1923 bytes -= old_end - offset;
1926 } else if (old_end == offset + bytes) {
1930 spin_unlock(&ctl->tree_lock);
1932 ret = btrfs_add_free_space(block_group, offset + bytes,
1933 old_end - (offset + bytes));
1939 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1940 if (ret == -EAGAIN) {
1944 BUG_ON(ret); /* logic error */
1946 spin_unlock(&ctl->tree_lock);
1951 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1954 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1955 struct btrfs_free_space *info;
1959 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1960 info = rb_entry(n, struct btrfs_free_space, offset_index);
1961 if (info->bytes >= bytes && !block_group->ro)
1963 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1964 (unsigned long long)info->offset,
1965 (unsigned long long)info->bytes,
1966 (info->bitmap) ? "yes" : "no");
1968 printk(KERN_INFO "block group has cluster?: %s\n",
1969 list_empty(&block_group->cluster_list) ? "no" : "yes");
1970 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1974 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1976 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1978 spin_lock_init(&ctl->tree_lock);
1979 ctl->unit = block_group->sectorsize;
1980 ctl->start = block_group->key.objectid;
1981 ctl->private = block_group;
1982 ctl->op = &free_space_op;
1985 * we only want to have 32k of ram per block group for keeping
1986 * track of free space, and if we pass 1/2 of that we want to
1987 * start converting things over to using bitmaps
1989 ctl->extents_thresh = ((1024 * 32) / 2) /
1990 sizeof(struct btrfs_free_space);
1994 * for a given cluster, put all of its extents back into the free
1995 * space cache. If the block group passed doesn't match the block group
1996 * pointed to by the cluster, someone else raced in and freed the
1997 * cluster already. In that case, we just return without changing anything
2000 __btrfs_return_cluster_to_free_space(
2001 struct btrfs_block_group_cache *block_group,
2002 struct btrfs_free_cluster *cluster)
2004 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2005 struct btrfs_free_space *entry;
2006 struct rb_node *node;
2008 spin_lock(&cluster->lock);
2009 if (cluster->block_group != block_group)
2012 cluster->block_group = NULL;
2013 cluster->window_start = 0;
2014 list_del_init(&cluster->block_group_list);
2016 node = rb_first(&cluster->root);
2020 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2021 node = rb_next(&entry->offset_index);
2022 rb_erase(&entry->offset_index, &cluster->root);
2024 bitmap = (entry->bitmap != NULL);
2026 try_merge_free_space(ctl, entry, false);
2027 tree_insert_offset(&ctl->free_space_offset,
2028 entry->offset, &entry->offset_index, bitmap);
2030 cluster->root = RB_ROOT;
2033 spin_unlock(&cluster->lock);
2034 btrfs_put_block_group(block_group);
2038 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
2040 struct btrfs_free_space *info;
2041 struct rb_node *node;
2043 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2044 info = rb_entry(node, struct btrfs_free_space, offset_index);
2045 if (!info->bitmap) {
2046 unlink_free_space(ctl, info);
2047 kmem_cache_free(btrfs_free_space_cachep, info);
2049 free_bitmap(ctl, info);
2051 if (need_resched()) {
2052 spin_unlock(&ctl->tree_lock);
2054 spin_lock(&ctl->tree_lock);
2059 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2061 spin_lock(&ctl->tree_lock);
2062 __btrfs_remove_free_space_cache_locked(ctl);
2063 spin_unlock(&ctl->tree_lock);
2066 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2068 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2069 struct btrfs_free_cluster *cluster;
2070 struct list_head *head;
2072 spin_lock(&ctl->tree_lock);
2073 while ((head = block_group->cluster_list.next) !=
2074 &block_group->cluster_list) {
2075 cluster = list_entry(head, struct btrfs_free_cluster,
2078 WARN_ON(cluster->block_group != block_group);
2079 __btrfs_return_cluster_to_free_space(block_group, cluster);
2080 if (need_resched()) {
2081 spin_unlock(&ctl->tree_lock);
2083 spin_lock(&ctl->tree_lock);
2086 __btrfs_remove_free_space_cache_locked(ctl);
2087 spin_unlock(&ctl->tree_lock);
2091 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2092 u64 offset, u64 bytes, u64 empty_size)
2094 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2095 struct btrfs_free_space *entry = NULL;
2096 u64 bytes_search = bytes + empty_size;
2099 spin_lock(&ctl->tree_lock);
2100 entry = find_free_space(ctl, &offset, &bytes_search);
2105 if (entry->bitmap) {
2106 bitmap_clear_bits(ctl, entry, offset, bytes);
2108 free_bitmap(ctl, entry);
2110 unlink_free_space(ctl, entry);
2111 entry->offset += bytes;
2112 entry->bytes -= bytes;
2114 kmem_cache_free(btrfs_free_space_cachep, entry);
2116 link_free_space(ctl, entry);
2120 spin_unlock(&ctl->tree_lock);
2126 * given a cluster, put all of its extents back into the free space
2127 * cache. If a block group is passed, this function will only free
2128 * a cluster that belongs to the passed block group.
2130 * Otherwise, it'll get a reference on the block group pointed to by the
2131 * cluster and remove the cluster from it.
2133 int btrfs_return_cluster_to_free_space(
2134 struct btrfs_block_group_cache *block_group,
2135 struct btrfs_free_cluster *cluster)
2137 struct btrfs_free_space_ctl *ctl;
2140 /* first, get a safe pointer to the block group */
2141 spin_lock(&cluster->lock);
2143 block_group = cluster->block_group;
2145 spin_unlock(&cluster->lock);
2148 } else if (cluster->block_group != block_group) {
2149 /* someone else has already freed it don't redo their work */
2150 spin_unlock(&cluster->lock);
2153 atomic_inc(&block_group->count);
2154 spin_unlock(&cluster->lock);
2156 ctl = block_group->free_space_ctl;
2158 /* now return any extents the cluster had on it */
2159 spin_lock(&ctl->tree_lock);
2160 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2161 spin_unlock(&ctl->tree_lock);
2163 /* finally drop our ref */
2164 btrfs_put_block_group(block_group);
2168 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2169 struct btrfs_free_cluster *cluster,
2170 struct btrfs_free_space *entry,
2171 u64 bytes, u64 min_start)
2173 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2175 u64 search_start = cluster->window_start;
2176 u64 search_bytes = bytes;
2179 search_start = min_start;
2180 search_bytes = bytes;
2182 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2187 __bitmap_clear_bits(ctl, entry, ret, bytes);
2193 * given a cluster, try to allocate 'bytes' from it, returns 0
2194 * if it couldn't find anything suitably large, or a logical disk offset
2195 * if things worked out
2197 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2198 struct btrfs_free_cluster *cluster, u64 bytes,
2201 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2202 struct btrfs_free_space *entry = NULL;
2203 struct rb_node *node;
2206 spin_lock(&cluster->lock);
2207 if (bytes > cluster->max_size)
2210 if (cluster->block_group != block_group)
2213 node = rb_first(&cluster->root);
2217 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2219 if (entry->bytes < bytes ||
2220 (!entry->bitmap && entry->offset < min_start)) {
2221 node = rb_next(&entry->offset_index);
2224 entry = rb_entry(node, struct btrfs_free_space,
2229 if (entry->bitmap) {
2230 ret = btrfs_alloc_from_bitmap(block_group,
2231 cluster, entry, bytes,
2232 cluster->window_start);
2234 node = rb_next(&entry->offset_index);
2237 entry = rb_entry(node, struct btrfs_free_space,
2241 cluster->window_start += bytes;
2243 ret = entry->offset;
2245 entry->offset += bytes;
2246 entry->bytes -= bytes;
2249 if (entry->bytes == 0)
2250 rb_erase(&entry->offset_index, &cluster->root);
2254 spin_unlock(&cluster->lock);
2259 spin_lock(&ctl->tree_lock);
2261 ctl->free_space -= bytes;
2262 if (entry->bytes == 0) {
2263 ctl->free_extents--;
2264 if (entry->bitmap) {
2265 kfree(entry->bitmap);
2266 ctl->total_bitmaps--;
2267 ctl->op->recalc_thresholds(ctl);
2269 kmem_cache_free(btrfs_free_space_cachep, entry);
2272 spin_unlock(&ctl->tree_lock);
2277 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2278 struct btrfs_free_space *entry,
2279 struct btrfs_free_cluster *cluster,
2280 u64 offset, u64 bytes,
2281 u64 cont1_bytes, u64 min_bytes)
2283 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2284 unsigned long next_zero;
2286 unsigned long want_bits;
2287 unsigned long min_bits;
2288 unsigned long found_bits;
2289 unsigned long start = 0;
2290 unsigned long total_found = 0;
2293 i = offset_to_bit(entry->offset, ctl->unit,
2294 max_t(u64, offset, entry->offset));
2295 want_bits = bytes_to_bits(bytes, ctl->unit);
2296 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2300 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2301 next_zero = find_next_zero_bit(entry->bitmap,
2302 BITS_PER_BITMAP, i);
2303 if (next_zero - i >= min_bits) {
2304 found_bits = next_zero - i;
2315 cluster->max_size = 0;
2318 total_found += found_bits;
2320 if (cluster->max_size < found_bits * ctl->unit)
2321 cluster->max_size = found_bits * ctl->unit;
2323 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2328 cluster->window_start = start * ctl->unit + entry->offset;
2329 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2330 ret = tree_insert_offset(&cluster->root, entry->offset,
2331 &entry->offset_index, 1);
2332 BUG_ON(ret); /* -EEXIST; Logic error */
2334 trace_btrfs_setup_cluster(block_group, cluster,
2335 total_found * ctl->unit, 1);
2340 * This searches the block group for just extents to fill the cluster with.
2341 * Try to find a cluster with at least bytes total bytes, at least one
2342 * extent of cont1_bytes, and other clusters of at least min_bytes.
2345 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2346 struct btrfs_free_cluster *cluster,
2347 struct list_head *bitmaps, u64 offset, u64 bytes,
2348 u64 cont1_bytes, u64 min_bytes)
2350 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2351 struct btrfs_free_space *first = NULL;
2352 struct btrfs_free_space *entry = NULL;
2353 struct btrfs_free_space *last;
2354 struct rb_node *node;
2360 entry = tree_search_offset(ctl, offset, 0, 1);
2365 * We don't want bitmaps, so just move along until we find a normal
2368 while (entry->bitmap || entry->bytes < min_bytes) {
2369 if (entry->bitmap && list_empty(&entry->list))
2370 list_add_tail(&entry->list, bitmaps);
2371 node = rb_next(&entry->offset_index);
2374 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2377 window_start = entry->offset;
2378 window_free = entry->bytes;
2379 max_extent = entry->bytes;
2383 for (node = rb_next(&entry->offset_index); node;
2384 node = rb_next(&entry->offset_index)) {
2385 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2387 if (entry->bitmap) {
2388 if (list_empty(&entry->list))
2389 list_add_tail(&entry->list, bitmaps);
2393 if (entry->bytes < min_bytes)
2397 window_free += entry->bytes;
2398 if (entry->bytes > max_extent)
2399 max_extent = entry->bytes;
2402 if (window_free < bytes || max_extent < cont1_bytes)
2405 cluster->window_start = first->offset;
2407 node = &first->offset_index;
2410 * now we've found our entries, pull them out of the free space
2411 * cache and put them into the cluster rbtree
2416 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2417 node = rb_next(&entry->offset_index);
2418 if (entry->bitmap || entry->bytes < min_bytes)
2421 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2422 ret = tree_insert_offset(&cluster->root, entry->offset,
2423 &entry->offset_index, 0);
2424 total_size += entry->bytes;
2425 BUG_ON(ret); /* -EEXIST; Logic error */
2426 } while (node && entry != last);
2428 cluster->max_size = max_extent;
2429 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2434 * This specifically looks for bitmaps that may work in the cluster, we assume
2435 * that we have already failed to find extents that will work.
2438 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2439 struct btrfs_free_cluster *cluster,
2440 struct list_head *bitmaps, u64 offset, u64 bytes,
2441 u64 cont1_bytes, u64 min_bytes)
2443 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2444 struct btrfs_free_space *entry;
2446 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2448 if (ctl->total_bitmaps == 0)
2452 * The bitmap that covers offset won't be in the list unless offset
2453 * is just its start offset.
2455 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2456 if (entry->offset != bitmap_offset) {
2457 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2458 if (entry && list_empty(&entry->list))
2459 list_add(&entry->list, bitmaps);
2462 list_for_each_entry(entry, bitmaps, list) {
2463 if (entry->bytes < bytes)
2465 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2466 bytes, cont1_bytes, min_bytes);
2472 * The bitmaps list has all the bitmaps that record free space
2473 * starting after offset, so no more search is required.
2479 * here we try to find a cluster of blocks in a block group. The goal
2480 * is to find at least bytes+empty_size.
2481 * We might not find them all in one contiguous area.
2483 * returns zero and sets up cluster if things worked out, otherwise
2484 * it returns -enospc
2486 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2487 struct btrfs_root *root,
2488 struct btrfs_block_group_cache *block_group,
2489 struct btrfs_free_cluster *cluster,
2490 u64 offset, u64 bytes, u64 empty_size)
2492 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2493 struct btrfs_free_space *entry, *tmp;
2500 * Choose the minimum extent size we'll require for this
2501 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2502 * For metadata, allow allocates with smaller extents. For
2503 * data, keep it dense.
2505 if (btrfs_test_opt(root, SSD_SPREAD)) {
2506 cont1_bytes = min_bytes = bytes + empty_size;
2507 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2508 cont1_bytes = bytes;
2509 min_bytes = block_group->sectorsize;
2511 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2512 min_bytes = block_group->sectorsize;
2515 spin_lock(&ctl->tree_lock);
2518 * If we know we don't have enough space to make a cluster don't even
2519 * bother doing all the work to try and find one.
2521 if (ctl->free_space < bytes) {
2522 spin_unlock(&ctl->tree_lock);
2526 spin_lock(&cluster->lock);
2528 /* someone already found a cluster, hooray */
2529 if (cluster->block_group) {
2534 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2537 INIT_LIST_HEAD(&bitmaps);
2538 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2540 cont1_bytes, min_bytes);
2542 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2543 offset, bytes + empty_size,
2544 cont1_bytes, min_bytes);
2546 /* Clear our temporary list */
2547 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2548 list_del_init(&entry->list);
2551 atomic_inc(&block_group->count);
2552 list_add_tail(&cluster->block_group_list,
2553 &block_group->cluster_list);
2554 cluster->block_group = block_group;
2556 trace_btrfs_failed_cluster_setup(block_group);
2559 spin_unlock(&cluster->lock);
2560 spin_unlock(&ctl->tree_lock);
2566 * simple code to zero out a cluster
2568 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2570 spin_lock_init(&cluster->lock);
2571 spin_lock_init(&cluster->refill_lock);
2572 cluster->root = RB_ROOT;
2573 cluster->max_size = 0;
2574 INIT_LIST_HEAD(&cluster->block_group_list);
2575 cluster->block_group = NULL;
2578 static int do_trimming(struct btrfs_block_group_cache *block_group,
2579 u64 *total_trimmed, u64 start, u64 bytes,
2580 u64 reserved_start, u64 reserved_bytes)
2582 struct btrfs_space_info *space_info = block_group->space_info;
2583 struct btrfs_fs_info *fs_info = block_group->fs_info;
2588 spin_lock(&space_info->lock);
2589 spin_lock(&block_group->lock);
2590 if (!block_group->ro) {
2591 block_group->reserved += reserved_bytes;
2592 space_info->bytes_reserved += reserved_bytes;
2595 spin_unlock(&block_group->lock);
2596 spin_unlock(&space_info->lock);
2598 ret = btrfs_error_discard_extent(fs_info->extent_root,
2599 start, bytes, &trimmed);
2601 *total_trimmed += trimmed;
2603 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2606 spin_lock(&space_info->lock);
2607 spin_lock(&block_group->lock);
2608 if (block_group->ro)
2609 space_info->bytes_readonly += reserved_bytes;
2610 block_group->reserved -= reserved_bytes;
2611 space_info->bytes_reserved -= reserved_bytes;
2612 spin_unlock(&space_info->lock);
2613 spin_unlock(&block_group->lock);
2619 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2620 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2622 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2623 struct btrfs_free_space *entry;
2624 struct rb_node *node;
2630 while (start < end) {
2631 spin_lock(&ctl->tree_lock);
2633 if (ctl->free_space < minlen) {
2634 spin_unlock(&ctl->tree_lock);
2638 entry = tree_search_offset(ctl, start, 0, 1);
2640 spin_unlock(&ctl->tree_lock);
2645 while (entry->bitmap) {
2646 node = rb_next(&entry->offset_index);
2648 spin_unlock(&ctl->tree_lock);
2651 entry = rb_entry(node, struct btrfs_free_space,
2655 if (entry->offset >= end) {
2656 spin_unlock(&ctl->tree_lock);
2660 extent_start = entry->offset;
2661 extent_bytes = entry->bytes;
2662 start = max(start, extent_start);
2663 bytes = min(extent_start + extent_bytes, end) - start;
2664 if (bytes < minlen) {
2665 spin_unlock(&ctl->tree_lock);
2669 unlink_free_space(ctl, entry);
2670 kmem_cache_free(btrfs_free_space_cachep, entry);
2672 spin_unlock(&ctl->tree_lock);
2674 ret = do_trimming(block_group, total_trimmed, start, bytes,
2675 extent_start, extent_bytes);
2681 if (fatal_signal_pending(current)) {
2692 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2693 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2695 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2696 struct btrfs_free_space *entry;
2700 u64 offset = offset_to_bitmap(ctl, start);
2702 while (offset < end) {
2703 bool next_bitmap = false;
2705 spin_lock(&ctl->tree_lock);
2707 if (ctl->free_space < minlen) {
2708 spin_unlock(&ctl->tree_lock);
2712 entry = tree_search_offset(ctl, offset, 1, 0);
2714 spin_unlock(&ctl->tree_lock);
2720 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2721 if (ret2 || start >= end) {
2722 spin_unlock(&ctl->tree_lock);
2727 bytes = min(bytes, end - start);
2728 if (bytes < minlen) {
2729 spin_unlock(&ctl->tree_lock);
2733 bitmap_clear_bits(ctl, entry, start, bytes);
2734 if (entry->bytes == 0)
2735 free_bitmap(ctl, entry);
2737 spin_unlock(&ctl->tree_lock);
2739 ret = do_trimming(block_group, total_trimmed, start, bytes,
2745 offset += BITS_PER_BITMAP * ctl->unit;
2748 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2749 offset += BITS_PER_BITMAP * ctl->unit;
2752 if (fatal_signal_pending(current)) {
2763 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2764 u64 *trimmed, u64 start, u64 end, u64 minlen)
2770 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2774 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2780 * Find the left-most item in the cache tree, and then return the
2781 * smallest inode number in the item.
2783 * Note: the returned inode number may not be the smallest one in
2784 * the tree, if the left-most item is a bitmap.
2786 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2788 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2789 struct btrfs_free_space *entry = NULL;
2792 spin_lock(&ctl->tree_lock);
2794 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2797 entry = rb_entry(rb_first(&ctl->free_space_offset),
2798 struct btrfs_free_space, offset_index);
2800 if (!entry->bitmap) {
2801 ino = entry->offset;
2803 unlink_free_space(ctl, entry);
2807 kmem_cache_free(btrfs_free_space_cachep, entry);
2809 link_free_space(ctl, entry);
2815 ret = search_bitmap(ctl, entry, &offset, &count);
2816 /* Logic error; Should be empty if it can't find anything */
2820 bitmap_clear_bits(ctl, entry, offset, 1);
2821 if (entry->bytes == 0)
2822 free_bitmap(ctl, entry);
2825 spin_unlock(&ctl->tree_lock);
2830 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2831 struct btrfs_path *path)
2833 struct inode *inode = NULL;
2835 spin_lock(&root->cache_lock);
2836 if (root->cache_inode)
2837 inode = igrab(root->cache_inode);
2838 spin_unlock(&root->cache_lock);
2842 inode = __lookup_free_space_inode(root, path, 0);
2846 spin_lock(&root->cache_lock);
2847 if (!btrfs_fs_closing(root->fs_info))
2848 root->cache_inode = igrab(inode);
2849 spin_unlock(&root->cache_lock);
2854 int create_free_ino_inode(struct btrfs_root *root,
2855 struct btrfs_trans_handle *trans,
2856 struct btrfs_path *path)
2858 return __create_free_space_inode(root, trans, path,
2859 BTRFS_FREE_INO_OBJECTID, 0);
2862 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2864 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2865 struct btrfs_path *path;
2866 struct inode *inode;
2868 u64 root_gen = btrfs_root_generation(&root->root_item);
2870 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2874 * If we're unmounting then just return, since this does a search on the
2875 * normal root and not the commit root and we could deadlock.
2877 if (btrfs_fs_closing(fs_info))
2880 path = btrfs_alloc_path();
2884 inode = lookup_free_ino_inode(root, path);
2888 if (root_gen != BTRFS_I(inode)->generation)
2891 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2894 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2895 "root %llu\n", root->root_key.objectid);
2899 btrfs_free_path(path);
2903 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2904 struct btrfs_trans_handle *trans,
2905 struct btrfs_path *path)
2907 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2908 struct inode *inode;
2911 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2914 inode = lookup_free_ino_inode(root, path);
2918 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2920 btrfs_delalloc_release_metadata(inode, inode->i_size);
2922 printk(KERN_ERR "btrfs: failed to write free ino cache "
2923 "for root %llu\n", root->root_key.objectid);