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 io_ctl_init(&io_ctl
, inode
, root
);
641 ret
= readahead_cache(inode
);
645 ret
= io_ctl_prepare_pages(&io_ctl
, inode
, 1);
649 ret
= io_ctl_check_crc(&io_ctl
, 0);
653 ret
= io_ctl_check_generation(&io_ctl
, generation
);
657 while (num_entries
) {
658 e
= kmem_cache_zalloc(btrfs_free_space_cachep
,
663 ret
= io_ctl_read_entry(&io_ctl
, e
, &type
);
665 kmem_cache_free(btrfs_free_space_cachep
, e
);
670 kmem_cache_free(btrfs_free_space_cachep
, e
);
674 if (type
== BTRFS_FREE_SPACE_EXTENT
) {
675 spin_lock(&ctl
->tree_lock
);
676 ret
= link_free_space(ctl
, e
);
677 spin_unlock(&ctl
->tree_lock
);
679 printk(KERN_ERR
"Duplicate entries in "
680 "free space cache, dumping\n");
681 kmem_cache_free(btrfs_free_space_cachep
, e
);
685 BUG_ON(!num_bitmaps
);
687 e
->bitmap
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
690 btrfs_free_space_cachep
, e
);
693 spin_lock(&ctl
->tree_lock
);
694 ret
= link_free_space(ctl
, e
);
695 ctl
->total_bitmaps
++;
696 ctl
->op
->recalc_thresholds(ctl
);
697 spin_unlock(&ctl
->tree_lock
);
699 printk(KERN_ERR
"Duplicate entries in "
700 "free space cache, dumping\n");
701 kmem_cache_free(btrfs_free_space_cachep
, e
);
704 list_add_tail(&e
->list
, &bitmaps
);
710 io_ctl_unmap_page(&io_ctl
);
713 * We add the bitmaps at the end of the entries in order that
714 * the bitmap entries are added to the cache.
716 list_for_each_entry_safe(e
, n
, &bitmaps
, list
) {
717 list_del_init(&e
->list
);
718 ret
= io_ctl_read_bitmap(&io_ctl
, e
);
723 io_ctl_drop_pages(&io_ctl
);
726 io_ctl_free(&io_ctl
);
729 io_ctl_drop_pages(&io_ctl
);
730 __btrfs_remove_free_space_cache(ctl
);
734 int load_free_space_cache(struct btrfs_fs_info
*fs_info
,
735 struct btrfs_block_group_cache
*block_group
)
737 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
738 struct btrfs_root
*root
= fs_info
->tree_root
;
740 struct btrfs_path
*path
;
743 u64 used
= btrfs_block_group_used(&block_group
->item
);
746 * If we're unmounting then just return, since this does a search on the
747 * normal root and not the commit root and we could deadlock.
749 if (btrfs_fs_closing(fs_info
))
753 * If this block group has been marked to be cleared for one reason or
754 * another then we can't trust the on disk cache, so just return.
756 spin_lock(&block_group
->lock
);
757 if (block_group
->disk_cache_state
!= BTRFS_DC_WRITTEN
) {
758 spin_unlock(&block_group
->lock
);
761 spin_unlock(&block_group
->lock
);
763 path
= btrfs_alloc_path();
767 inode
= lookup_free_space_inode(root
, block_group
, path
);
769 btrfs_free_path(path
);
773 /* We may have converted the inode and made the cache invalid. */
774 spin_lock(&block_group
->lock
);
775 if (block_group
->disk_cache_state
!= BTRFS_DC_WRITTEN
) {
776 spin_unlock(&block_group
->lock
);
779 spin_unlock(&block_group
->lock
);
781 ret
= __load_free_space_cache(fs_info
->tree_root
, inode
, ctl
,
782 path
, block_group
->key
.objectid
);
783 btrfs_free_path(path
);
787 spin_lock(&ctl
->tree_lock
);
788 matched
= (ctl
->free_space
== (block_group
->key
.offset
- used
-
789 block_group
->bytes_super
));
790 spin_unlock(&ctl
->tree_lock
);
793 __btrfs_remove_free_space_cache(ctl
);
794 printk(KERN_ERR
"block group %llu has an wrong amount of free "
795 "space\n", block_group
->key
.objectid
);
800 /* This cache is bogus, make sure it gets cleared */
801 spin_lock(&block_group
->lock
);
802 block_group
->disk_cache_state
= BTRFS_DC_CLEAR
;
803 spin_unlock(&block_group
->lock
);
806 printk(KERN_ERR
"btrfs: failed to load free space cache "
807 "for block group %llu\n", block_group
->key
.objectid
);
815 * __btrfs_write_out_cache - write out cached info to an inode
816 * @root - the root the inode belongs to
817 * @ctl - the free space cache we are going to write out
818 * @block_group - the block_group for this cache if it belongs to a block_group
819 * @trans - the trans handle
820 * @path - the path to use
821 * @offset - the offset for the key we'll insert
823 * This function writes out a free space cache struct to disk for quick recovery
824 * on mount. This will return 0 if it was successfull in writing the cache out,
825 * and -1 if it was not.
827 int __btrfs_write_out_cache(struct btrfs_root
*root
, struct inode
*inode
,
828 struct btrfs_free_space_ctl
*ctl
,
829 struct btrfs_block_group_cache
*block_group
,
830 struct btrfs_trans_handle
*trans
,
831 struct btrfs_path
*path
, u64 offset
)
833 struct btrfs_free_space_header
*header
;
834 struct extent_buffer
*leaf
;
835 struct rb_node
*node
;
836 struct list_head
*pos
, *n
;
837 struct extent_state
*cached_state
= NULL
;
838 struct btrfs_free_cluster
*cluster
= NULL
;
839 struct extent_io_tree
*unpin
= NULL
;
840 struct io_ctl io_ctl
;
841 struct list_head bitmap_list
;
842 struct btrfs_key key
;
843 u64 start
, extent_start
, extent_end
, len
;
849 INIT_LIST_HEAD(&bitmap_list
);
851 if (!i_size_read(inode
))
854 io_ctl_init(&io_ctl
, inode
, root
);
856 /* Get the cluster for this block_group if it exists */
857 if (block_group
&& !list_empty(&block_group
->cluster_list
))
858 cluster
= list_entry(block_group
->cluster_list
.next
,
859 struct btrfs_free_cluster
,
862 /* Lock all pages first so we can lock the extent safely. */
863 io_ctl_prepare_pages(&io_ctl
, inode
, 0);
865 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, 0, i_size_read(inode
) - 1,
866 0, &cached_state
, GFP_NOFS
);
868 node
= rb_first(&ctl
->free_space_offset
);
869 if (!node
&& cluster
) {
870 node
= rb_first(&cluster
->root
);
874 /* Make sure we can fit our crcs into the first page */
875 if (io_ctl
.check_crcs
&&
876 (io_ctl
.num_pages
* sizeof(u32
)) >= PAGE_CACHE_SIZE
) {
881 io_ctl_set_generation(&io_ctl
, trans
->transid
);
883 /* Write out the extent entries */
885 struct btrfs_free_space
*e
;
887 e
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
890 ret
= io_ctl_add_entry(&io_ctl
, e
->offset
, e
->bytes
,
896 list_add_tail(&e
->list
, &bitmap_list
);
899 node
= rb_next(node
);
900 if (!node
&& cluster
) {
901 node
= rb_first(&cluster
->root
);
907 * We want to add any pinned extents to our free space cache
908 * so we don't leak the space
912 * We shouldn't have switched the pinned extents yet so this is the
915 unpin
= root
->fs_info
->pinned_extents
;
918 start
= block_group
->key
.objectid
;
920 while (block_group
&& (start
< block_group
->key
.objectid
+
921 block_group
->key
.offset
)) {
922 ret
= find_first_extent_bit(unpin
, start
,
923 &extent_start
, &extent_end
,
930 /* This pinned extent is out of our range */
931 if (extent_start
>= block_group
->key
.objectid
+
932 block_group
->key
.offset
)
935 extent_start
= max(extent_start
, start
);
936 extent_end
= min(block_group
->key
.objectid
+
937 block_group
->key
.offset
, extent_end
+ 1);
938 len
= extent_end
- extent_start
;
941 ret
= io_ctl_add_entry(&io_ctl
, extent_start
, len
, NULL
);
948 /* Write out the bitmaps */
949 list_for_each_safe(pos
, n
, &bitmap_list
) {
950 struct btrfs_free_space
*entry
=
951 list_entry(pos
, struct btrfs_free_space
, list
);
953 ret
= io_ctl_add_bitmap(&io_ctl
, entry
->bitmap
);
956 list_del_init(&entry
->list
);
959 /* Zero out the rest of the pages just to make sure */
960 io_ctl_zero_remaining_pages(&io_ctl
);
962 ret
= btrfs_dirty_pages(root
, inode
, io_ctl
.pages
, io_ctl
.num_pages
,
963 0, i_size_read(inode
), &cached_state
);
964 io_ctl_drop_pages(&io_ctl
);
965 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, 0,
966 i_size_read(inode
) - 1, &cached_state
, GFP_NOFS
);
972 ret
= filemap_write_and_wait(inode
->i_mapping
);
976 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
980 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
982 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0, inode
->i_size
- 1,
983 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, NULL
,
987 leaf
= path
->nodes
[0];
989 struct btrfs_key found_key
;
990 BUG_ON(!path
->slots
[0]);
992 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
993 if (found_key
.objectid
!= BTRFS_FREE_SPACE_OBJECTID
||
994 found_key
.offset
!= offset
) {
995 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0,
997 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0,
999 btrfs_release_path(path
);
1004 BTRFS_I(inode
)->generation
= trans
->transid
;
1005 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
1006 struct btrfs_free_space_header
);
1007 btrfs_set_free_space_entries(leaf
, header
, entries
);
1008 btrfs_set_free_space_bitmaps(leaf
, header
, bitmaps
);
1009 btrfs_set_free_space_generation(leaf
, header
, trans
->transid
);
1010 btrfs_mark_buffer_dirty(leaf
);
1011 btrfs_release_path(path
);
1015 io_ctl_free(&io_ctl
);
1017 invalidate_inode_pages2(inode
->i_mapping
);
1018 BTRFS_I(inode
)->generation
= 0;
1020 btrfs_update_inode(trans
, root
, inode
);
1024 list_for_each_safe(pos
, n
, &bitmap_list
) {
1025 struct btrfs_free_space
*entry
=
1026 list_entry(pos
, struct btrfs_free_space
, list
);
1027 list_del_init(&entry
->list
);
1029 io_ctl_drop_pages(&io_ctl
);
1030 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, 0,
1031 i_size_read(inode
) - 1, &cached_state
, GFP_NOFS
);
1035 int btrfs_write_out_cache(struct btrfs_root
*root
,
1036 struct btrfs_trans_handle
*trans
,
1037 struct btrfs_block_group_cache
*block_group
,
1038 struct btrfs_path
*path
)
1040 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1041 struct inode
*inode
;
1044 root
= root
->fs_info
->tree_root
;
1046 spin_lock(&block_group
->lock
);
1047 if (block_group
->disk_cache_state
< BTRFS_DC_SETUP
) {
1048 spin_unlock(&block_group
->lock
);
1051 spin_unlock(&block_group
->lock
);
1053 inode
= lookup_free_space_inode(root
, block_group
, path
);
1057 ret
= __btrfs_write_out_cache(root
, inode
, ctl
, block_group
, trans
,
1058 path
, block_group
->key
.objectid
);
1060 spin_lock(&block_group
->lock
);
1061 block_group
->disk_cache_state
= BTRFS_DC_ERROR
;
1062 spin_unlock(&block_group
->lock
);
1065 printk(KERN_ERR
"btrfs: failed to write free space cace "
1066 "for block group %llu\n", block_group
->key
.objectid
);
1074 static inline unsigned long offset_to_bit(u64 bitmap_start
, u32 unit
,
1077 BUG_ON(offset
< bitmap_start
);
1078 offset
-= bitmap_start
;
1079 return (unsigned long)(div_u64(offset
, unit
));
1082 static inline unsigned long bytes_to_bits(u64 bytes
, u32 unit
)
1084 return (unsigned long)(div_u64(bytes
, unit
));
1087 static inline u64
offset_to_bitmap(struct btrfs_free_space_ctl
*ctl
,
1091 u64 bytes_per_bitmap
;
1093 bytes_per_bitmap
= BITS_PER_BITMAP
* ctl
->unit
;
1094 bitmap_start
= offset
- ctl
->start
;
1095 bitmap_start
= div64_u64(bitmap_start
, bytes_per_bitmap
);
1096 bitmap_start
*= bytes_per_bitmap
;
1097 bitmap_start
+= ctl
->start
;
1099 return bitmap_start
;
1102 static int tree_insert_offset(struct rb_root
*root
, u64 offset
,
1103 struct rb_node
*node
, int bitmap
)
1105 struct rb_node
**p
= &root
->rb_node
;
1106 struct rb_node
*parent
= NULL
;
1107 struct btrfs_free_space
*info
;
1111 info
= rb_entry(parent
, struct btrfs_free_space
, offset_index
);
1113 if (offset
< info
->offset
) {
1115 } else if (offset
> info
->offset
) {
1116 p
= &(*p
)->rb_right
;
1119 * we could have a bitmap entry and an extent entry
1120 * share the same offset. If this is the case, we want
1121 * the extent entry to always be found first if we do a
1122 * linear search through the tree, since we want to have
1123 * the quickest allocation time, and allocating from an
1124 * extent is faster than allocating from a bitmap. So
1125 * if we're inserting a bitmap and we find an entry at
1126 * this offset, we want to go right, or after this entry
1127 * logically. If we are inserting an extent and we've
1128 * found a bitmap, we want to go left, or before
1136 p
= &(*p
)->rb_right
;
1138 if (!info
->bitmap
) {
1147 rb_link_node(node
, parent
, p
);
1148 rb_insert_color(node
, root
);
1154 * searches the tree for the given offset.
1156 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1157 * want a section that has at least bytes size and comes at or after the given
1160 static struct btrfs_free_space
*
1161 tree_search_offset(struct btrfs_free_space_ctl
*ctl
,
1162 u64 offset
, int bitmap_only
, int fuzzy
)
1164 struct rb_node
*n
= ctl
->free_space_offset
.rb_node
;
1165 struct btrfs_free_space
*entry
, *prev
= NULL
;
1167 /* find entry that is closest to the 'offset' */
1174 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1177 if (offset
< entry
->offset
)
1179 else if (offset
> entry
->offset
)
1192 * bitmap entry and extent entry may share same offset,
1193 * in that case, bitmap entry comes after extent entry.
1198 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1199 if (entry
->offset
!= offset
)
1202 WARN_ON(!entry
->bitmap
);
1205 if (entry
->bitmap
) {
1207 * if previous extent entry covers the offset,
1208 * we should return it instead of the bitmap entry
1210 n
= &entry
->offset_index
;
1215 prev
= rb_entry(n
, struct btrfs_free_space
,
1217 if (!prev
->bitmap
) {
1218 if (prev
->offset
+ prev
->bytes
> offset
)
1230 /* find last entry before the 'offset' */
1232 if (entry
->offset
> offset
) {
1233 n
= rb_prev(&entry
->offset_index
);
1235 entry
= rb_entry(n
, struct btrfs_free_space
,
1237 BUG_ON(entry
->offset
> offset
);
1246 if (entry
->bitmap
) {
1247 n
= &entry
->offset_index
;
1252 prev
= rb_entry(n
, struct btrfs_free_space
,
1254 if (!prev
->bitmap
) {
1255 if (prev
->offset
+ prev
->bytes
> offset
)
1260 if (entry
->offset
+ BITS_PER_BITMAP
* ctl
->unit
> offset
)
1262 } else if (entry
->offset
+ entry
->bytes
> offset
)
1269 if (entry
->bitmap
) {
1270 if (entry
->offset
+ BITS_PER_BITMAP
*
1274 if (entry
->offset
+ entry
->bytes
> offset
)
1278 n
= rb_next(&entry
->offset_index
);
1281 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1287 __unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
1288 struct btrfs_free_space
*info
)
1290 rb_erase(&info
->offset_index
, &ctl
->free_space_offset
);
1291 ctl
->free_extents
--;
1294 static void unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
1295 struct btrfs_free_space
*info
)
1297 __unlink_free_space(ctl
, info
);
1298 ctl
->free_space
-= info
->bytes
;
1301 static int link_free_space(struct btrfs_free_space_ctl
*ctl
,
1302 struct btrfs_free_space
*info
)
1306 BUG_ON(!info
->bitmap
&& !info
->bytes
);
1307 ret
= tree_insert_offset(&ctl
->free_space_offset
, info
->offset
,
1308 &info
->offset_index
, (info
->bitmap
!= NULL
));
1312 ctl
->free_space
+= info
->bytes
;
1313 ctl
->free_extents
++;
1317 static void recalculate_thresholds(struct btrfs_free_space_ctl
*ctl
)
1319 struct btrfs_block_group_cache
*block_group
= ctl
->private;
1323 u64 size
= block_group
->key
.offset
;
1324 u64 bytes_per_bg
= BITS_PER_BITMAP
* block_group
->sectorsize
;
1325 int max_bitmaps
= div64_u64(size
+ bytes_per_bg
- 1, bytes_per_bg
);
1327 BUG_ON(ctl
->total_bitmaps
> max_bitmaps
);
1330 * The goal is to keep the total amount of memory used per 1gb of space
1331 * at or below 32k, so we need to adjust how much memory we allow to be
1332 * used by extent based free space tracking
1334 if (size
< 1024 * 1024 * 1024)
1335 max_bytes
= MAX_CACHE_BYTES_PER_GIG
;
1337 max_bytes
= MAX_CACHE_BYTES_PER_GIG
*
1338 div64_u64(size
, 1024 * 1024 * 1024);
1341 * we want to account for 1 more bitmap than what we have so we can make
1342 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1343 * we add more bitmaps.
1345 bitmap_bytes
= (ctl
->total_bitmaps
+ 1) * PAGE_CACHE_SIZE
;
1347 if (bitmap_bytes
>= max_bytes
) {
1348 ctl
->extents_thresh
= 0;
1353 * we want the extent entry threshold to always be at most 1/2 the maxw
1354 * bytes we can have, or whatever is less than that.
1356 extent_bytes
= max_bytes
- bitmap_bytes
;
1357 extent_bytes
= min_t(u64
, extent_bytes
, div64_u64(max_bytes
, 2));
1359 ctl
->extents_thresh
=
1360 div64_u64(extent_bytes
, (sizeof(struct btrfs_free_space
)));
1363 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl
*ctl
,
1364 struct btrfs_free_space
*info
,
1365 u64 offset
, u64 bytes
)
1367 unsigned long start
, count
;
1369 start
= offset_to_bit(info
->offset
, ctl
->unit
, offset
);
1370 count
= bytes_to_bits(bytes
, ctl
->unit
);
1371 BUG_ON(start
+ count
> BITS_PER_BITMAP
);
1373 bitmap_clear(info
->bitmap
, start
, count
);
1375 info
->bytes
-= bytes
;
1378 static void bitmap_clear_bits(struct btrfs_free_space_ctl
*ctl
,
1379 struct btrfs_free_space
*info
, u64 offset
,
1382 __bitmap_clear_bits(ctl
, info
, offset
, bytes
);
1383 ctl
->free_space
-= bytes
;
1386 static void bitmap_set_bits(struct btrfs_free_space_ctl
*ctl
,
1387 struct btrfs_free_space
*info
, u64 offset
,
1390 unsigned long start
, count
;
1392 start
= offset_to_bit(info
->offset
, ctl
->unit
, offset
);
1393 count
= bytes_to_bits(bytes
, ctl
->unit
);
1394 BUG_ON(start
+ count
> BITS_PER_BITMAP
);
1396 bitmap_set(info
->bitmap
, start
, count
);
1398 info
->bytes
+= bytes
;
1399 ctl
->free_space
+= bytes
;
1402 static int search_bitmap(struct btrfs_free_space_ctl
*ctl
,
1403 struct btrfs_free_space
*bitmap_info
, u64
*offset
,
1406 unsigned long found_bits
= 0;
1407 unsigned long bits
, i
;
1408 unsigned long next_zero
;
1410 i
= offset_to_bit(bitmap_info
->offset
, ctl
->unit
,
1411 max_t(u64
, *offset
, bitmap_info
->offset
));
1412 bits
= bytes_to_bits(*bytes
, ctl
->unit
);
1414 for (i
= find_next_bit(bitmap_info
->bitmap
, BITS_PER_BITMAP
, i
);
1415 i
< BITS_PER_BITMAP
;
1416 i
= find_next_bit(bitmap_info
->bitmap
, BITS_PER_BITMAP
, i
+ 1)) {
1417 next_zero
= find_next_zero_bit(bitmap_info
->bitmap
,
1418 BITS_PER_BITMAP
, i
);
1419 if ((next_zero
- i
) >= bits
) {
1420 found_bits
= next_zero
- i
;
1427 *offset
= (u64
)(i
* ctl
->unit
) + bitmap_info
->offset
;
1428 *bytes
= (u64
)(found_bits
) * ctl
->unit
;
1435 static struct btrfs_free_space
*
1436 find_free_space(struct btrfs_free_space_ctl
*ctl
, u64
*offset
, u64
*bytes
)
1438 struct btrfs_free_space
*entry
;
1439 struct rb_node
*node
;
1442 if (!ctl
->free_space_offset
.rb_node
)
1445 entry
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, *offset
), 0, 1);
1449 for (node
= &entry
->offset_index
; node
; node
= rb_next(node
)) {
1450 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
1451 if (entry
->bytes
< *bytes
)
1454 if (entry
->bitmap
) {
1455 ret
= search_bitmap(ctl
, entry
, offset
, bytes
);
1461 *offset
= entry
->offset
;
1462 *bytes
= entry
->bytes
;
1469 static void add_new_bitmap(struct btrfs_free_space_ctl
*ctl
,
1470 struct btrfs_free_space
*info
, u64 offset
)
1472 info
->offset
= offset_to_bitmap(ctl
, offset
);
1474 INIT_LIST_HEAD(&info
->list
);
1475 link_free_space(ctl
, info
);
1476 ctl
->total_bitmaps
++;
1478 ctl
->op
->recalc_thresholds(ctl
);
1481 static void free_bitmap(struct btrfs_free_space_ctl
*ctl
,
1482 struct btrfs_free_space
*bitmap_info
)
1484 unlink_free_space(ctl
, bitmap_info
);
1485 kfree(bitmap_info
->bitmap
);
1486 kmem_cache_free(btrfs_free_space_cachep
, bitmap_info
);
1487 ctl
->total_bitmaps
--;
1488 ctl
->op
->recalc_thresholds(ctl
);
1491 static noinline
int remove_from_bitmap(struct btrfs_free_space_ctl
*ctl
,
1492 struct btrfs_free_space
*bitmap_info
,
1493 u64
*offset
, u64
*bytes
)
1496 u64 search_start
, search_bytes
;
1500 end
= bitmap_info
->offset
+ (u64
)(BITS_PER_BITMAP
* ctl
->unit
) - 1;
1503 * XXX - this can go away after a few releases.
1505 * since the only user of btrfs_remove_free_space is the tree logging
1506 * stuff, and the only way to test that is under crash conditions, we
1507 * want to have this debug stuff here just in case somethings not
1508 * working. Search the bitmap for the space we are trying to use to
1509 * make sure its actually there. If its not there then we need to stop
1510 * because something has gone wrong.
1512 search_start
= *offset
;
1513 search_bytes
= *bytes
;
1514 search_bytes
= min(search_bytes
, end
- search_start
+ 1);
1515 ret
= search_bitmap(ctl
, bitmap_info
, &search_start
, &search_bytes
);
1516 BUG_ON(ret
< 0 || search_start
!= *offset
);
1518 if (*offset
> bitmap_info
->offset
&& *offset
+ *bytes
> end
) {
1519 bitmap_clear_bits(ctl
, bitmap_info
, *offset
, end
- *offset
+ 1);
1520 *bytes
-= end
- *offset
+ 1;
1522 } else if (*offset
>= bitmap_info
->offset
&& *offset
+ *bytes
<= end
) {
1523 bitmap_clear_bits(ctl
, bitmap_info
, *offset
, *bytes
);
1528 struct rb_node
*next
= rb_next(&bitmap_info
->offset_index
);
1529 if (!bitmap_info
->bytes
)
1530 free_bitmap(ctl
, bitmap_info
);
1533 * no entry after this bitmap, but we still have bytes to
1534 * remove, so something has gone wrong.
1539 bitmap_info
= rb_entry(next
, struct btrfs_free_space
,
1543 * if the next entry isn't a bitmap we need to return to let the
1544 * extent stuff do its work.
1546 if (!bitmap_info
->bitmap
)
1550 * Ok the next item is a bitmap, but it may not actually hold
1551 * the information for the rest of this free space stuff, so
1552 * look for it, and if we don't find it return so we can try
1553 * everything over again.
1555 search_start
= *offset
;
1556 search_bytes
= *bytes
;
1557 ret
= search_bitmap(ctl
, bitmap_info
, &search_start
,
1559 if (ret
< 0 || search_start
!= *offset
)
1563 } else if (!bitmap_info
->bytes
)
1564 free_bitmap(ctl
, bitmap_info
);
1569 static u64
add_bytes_to_bitmap(struct btrfs_free_space_ctl
*ctl
,
1570 struct btrfs_free_space
*info
, u64 offset
,
1573 u64 bytes_to_set
= 0;
1576 end
= info
->offset
+ (u64
)(BITS_PER_BITMAP
* ctl
->unit
);
1578 bytes_to_set
= min(end
- offset
, bytes
);
1580 bitmap_set_bits(ctl
, info
, offset
, bytes_to_set
);
1582 return bytes_to_set
;
1586 static bool use_bitmap(struct btrfs_free_space_ctl
*ctl
,
1587 struct btrfs_free_space
*info
)
1589 struct btrfs_block_group_cache
*block_group
= ctl
->private;
1592 * If we are below the extents threshold then we can add this as an
1593 * extent, and don't have to deal with the bitmap
1595 if (ctl
->free_extents
< ctl
->extents_thresh
) {
1597 * If this block group has some small extents we don't want to
1598 * use up all of our free slots in the cache with them, we want
1599 * to reserve them to larger extents, however if we have plent
1600 * of cache left then go ahead an dadd them, no sense in adding
1601 * the overhead of a bitmap if we don't have to.
1603 if (info
->bytes
<= block_group
->sectorsize
* 4) {
1604 if (ctl
->free_extents
* 2 <= ctl
->extents_thresh
)
1612 * some block groups are so tiny they can't be enveloped by a bitmap, so
1613 * don't even bother to create a bitmap for this
1615 if (BITS_PER_BITMAP
* block_group
->sectorsize
>
1616 block_group
->key
.offset
)
1622 static struct btrfs_free_space_op free_space_op
= {
1623 .recalc_thresholds
= recalculate_thresholds
,
1624 .use_bitmap
= use_bitmap
,
1627 static int insert_into_bitmap(struct btrfs_free_space_ctl
*ctl
,
1628 struct btrfs_free_space
*info
)
1630 struct btrfs_free_space
*bitmap_info
;
1631 struct btrfs_block_group_cache
*block_group
= NULL
;
1633 u64 bytes
, offset
, bytes_added
;
1636 bytes
= info
->bytes
;
1637 offset
= info
->offset
;
1639 if (!ctl
->op
->use_bitmap(ctl
, info
))
1642 if (ctl
->op
== &free_space_op
)
1643 block_group
= ctl
->private;
1646 * Since we link bitmaps right into the cluster we need to see if we
1647 * have a cluster here, and if so and it has our bitmap we need to add
1648 * the free space to that bitmap.
1650 if (block_group
&& !list_empty(&block_group
->cluster_list
)) {
1651 struct btrfs_free_cluster
*cluster
;
1652 struct rb_node
*node
;
1653 struct btrfs_free_space
*entry
;
1655 cluster
= list_entry(block_group
->cluster_list
.next
,
1656 struct btrfs_free_cluster
,
1658 spin_lock(&cluster
->lock
);
1659 node
= rb_first(&cluster
->root
);
1661 spin_unlock(&cluster
->lock
);
1662 goto no_cluster_bitmap
;
1665 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
1666 if (!entry
->bitmap
) {
1667 spin_unlock(&cluster
->lock
);
1668 goto no_cluster_bitmap
;
1671 if (entry
->offset
== offset_to_bitmap(ctl
, offset
)) {
1672 bytes_added
= add_bytes_to_bitmap(ctl
, entry
,
1674 bytes
-= bytes_added
;
1675 offset
+= bytes_added
;
1677 spin_unlock(&cluster
->lock
);
1685 bitmap_info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
1692 bytes_added
= add_bytes_to_bitmap(ctl
, bitmap_info
, offset
, bytes
);
1693 bytes
-= bytes_added
;
1694 offset
+= bytes_added
;
1704 if (info
&& info
->bitmap
) {
1705 add_new_bitmap(ctl
, info
, offset
);
1710 spin_unlock(&ctl
->tree_lock
);
1712 /* no pre-allocated info, allocate a new one */
1714 info
= kmem_cache_zalloc(btrfs_free_space_cachep
,
1717 spin_lock(&ctl
->tree_lock
);
1723 /* allocate the bitmap */
1724 info
->bitmap
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
1725 spin_lock(&ctl
->tree_lock
);
1726 if (!info
->bitmap
) {
1736 kfree(info
->bitmap
);
1737 kmem_cache_free(btrfs_free_space_cachep
, info
);
1743 static bool try_merge_free_space(struct btrfs_free_space_ctl
*ctl
,
1744 struct btrfs_free_space
*info
, bool update_stat
)
1746 struct btrfs_free_space
*left_info
;
1747 struct btrfs_free_space
*right_info
;
1748 bool merged
= false;
1749 u64 offset
= info
->offset
;
1750 u64 bytes
= info
->bytes
;
1753 * first we want to see if there is free space adjacent to the range we
1754 * are adding, if there is remove that struct and add a new one to
1755 * cover the entire range
1757 right_info
= tree_search_offset(ctl
, offset
+ bytes
, 0, 0);
1758 if (right_info
&& rb_prev(&right_info
->offset_index
))
1759 left_info
= rb_entry(rb_prev(&right_info
->offset_index
),
1760 struct btrfs_free_space
, offset_index
);
1762 left_info
= tree_search_offset(ctl
, offset
- 1, 0, 0);
1764 if (right_info
&& !right_info
->bitmap
) {
1766 unlink_free_space(ctl
, right_info
);
1768 __unlink_free_space(ctl
, right_info
);
1769 info
->bytes
+= right_info
->bytes
;
1770 kmem_cache_free(btrfs_free_space_cachep
, right_info
);
1774 if (left_info
&& !left_info
->bitmap
&&
1775 left_info
->offset
+ left_info
->bytes
== offset
) {
1777 unlink_free_space(ctl
, left_info
);
1779 __unlink_free_space(ctl
, left_info
);
1780 info
->offset
= left_info
->offset
;
1781 info
->bytes
+= left_info
->bytes
;
1782 kmem_cache_free(btrfs_free_space_cachep
, left_info
);
1789 int __btrfs_add_free_space(struct btrfs_free_space_ctl
*ctl
,
1790 u64 offset
, u64 bytes
)
1792 struct btrfs_free_space
*info
;
1795 info
= kmem_cache_zalloc(btrfs_free_space_cachep
, GFP_NOFS
);
1799 info
->offset
= offset
;
1800 info
->bytes
= bytes
;
1802 spin_lock(&ctl
->tree_lock
);
1804 if (try_merge_free_space(ctl
, info
, true))
1808 * There was no extent directly to the left or right of this new
1809 * extent then we know we're going to have to allocate a new extent, so
1810 * before we do that see if we need to drop this into a bitmap
1812 ret
= insert_into_bitmap(ctl
, info
);
1820 ret
= link_free_space(ctl
, info
);
1822 kmem_cache_free(btrfs_free_space_cachep
, info
);
1824 spin_unlock(&ctl
->tree_lock
);
1827 printk(KERN_CRIT
"btrfs: unable to add free space :%d\n", ret
);
1828 BUG_ON(ret
== -EEXIST
);
1834 int btrfs_remove_free_space(struct btrfs_block_group_cache
*block_group
,
1835 u64 offset
, u64 bytes
)
1837 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1838 struct btrfs_free_space
*info
;
1839 struct btrfs_free_space
*next_info
= NULL
;
1842 spin_lock(&ctl
->tree_lock
);
1845 info
= tree_search_offset(ctl
, offset
, 0, 0);
1848 * oops didn't find an extent that matched the space we wanted
1849 * to remove, look for a bitmap instead
1851 info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
1854 /* the tree logging code might be calling us before we
1855 * have fully loaded the free space rbtree for this
1856 * block group. So it is possible the entry won't
1857 * be in the rbtree yet at all. The caching code
1858 * will make sure not to put it in the rbtree if
1859 * the logging code has pinned it.
1865 if (info
->bytes
< bytes
&& rb_next(&info
->offset_index
)) {
1867 next_info
= rb_entry(rb_next(&info
->offset_index
),
1868 struct btrfs_free_space
,
1871 if (next_info
->bitmap
)
1872 end
= next_info
->offset
+
1873 BITS_PER_BITMAP
* ctl
->unit
- 1;
1875 end
= next_info
->offset
+ next_info
->bytes
;
1877 if (next_info
->bytes
< bytes
||
1878 next_info
->offset
> offset
|| offset
> end
) {
1879 printk(KERN_CRIT
"Found free space at %llu, size %llu,"
1880 " trying to use %llu\n",
1881 (unsigned long long)info
->offset
,
1882 (unsigned long long)info
->bytes
,
1883 (unsigned long long)bytes
);
1892 if (info
->bytes
== bytes
) {
1893 unlink_free_space(ctl
, info
);
1895 kfree(info
->bitmap
);
1896 ctl
->total_bitmaps
--;
1898 kmem_cache_free(btrfs_free_space_cachep
, info
);
1903 if (!info
->bitmap
&& info
->offset
== offset
) {
1904 unlink_free_space(ctl
, info
);
1905 info
->offset
+= bytes
;
1906 info
->bytes
-= bytes
;
1907 ret
= link_free_space(ctl
, info
);
1912 if (!info
->bitmap
&& info
->offset
<= offset
&&
1913 info
->offset
+ info
->bytes
>= offset
+ bytes
) {
1914 u64 old_start
= info
->offset
;
1916 * we're freeing space in the middle of the info,
1917 * this can happen during tree log replay
1919 * first unlink the old info and then
1920 * insert it again after the hole we're creating
1922 unlink_free_space(ctl
, info
);
1923 if (offset
+ bytes
< info
->offset
+ info
->bytes
) {
1924 u64 old_end
= info
->offset
+ info
->bytes
;
1926 info
->offset
= offset
+ bytes
;
1927 info
->bytes
= old_end
- info
->offset
;
1928 ret
= link_free_space(ctl
, info
);
1933 /* the hole we're creating ends at the end
1934 * of the info struct, just free the info
1936 kmem_cache_free(btrfs_free_space_cachep
, info
);
1938 spin_unlock(&ctl
->tree_lock
);
1940 /* step two, insert a new info struct to cover
1941 * anything before the hole
1943 ret
= btrfs_add_free_space(block_group
, old_start
,
1944 offset
- old_start
);
1949 ret
= remove_from_bitmap(ctl
, info
, &offset
, &bytes
);
1954 spin_unlock(&ctl
->tree_lock
);
1959 void btrfs_dump_free_space(struct btrfs_block_group_cache
*block_group
,
1962 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1963 struct btrfs_free_space
*info
;
1967 for (n
= rb_first(&ctl
->free_space_offset
); n
; n
= rb_next(n
)) {
1968 info
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1969 if (info
->bytes
>= bytes
)
1971 printk(KERN_CRIT
"entry offset %llu, bytes %llu, bitmap %s\n",
1972 (unsigned long long)info
->offset
,
1973 (unsigned long long)info
->bytes
,
1974 (info
->bitmap
) ? "yes" : "no");
1976 printk(KERN_INFO
"block group has cluster?: %s\n",
1977 list_empty(&block_group
->cluster_list
) ? "no" : "yes");
1978 printk(KERN_INFO
"%d blocks of free space at or bigger than bytes is"
1982 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache
*block_group
)
1984 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1986 spin_lock_init(&ctl
->tree_lock
);
1987 ctl
->unit
= block_group
->sectorsize
;
1988 ctl
->start
= block_group
->key
.objectid
;
1989 ctl
->private = block_group
;
1990 ctl
->op
= &free_space_op
;
1993 * we only want to have 32k of ram per block group for keeping
1994 * track of free space, and if we pass 1/2 of that we want to
1995 * start converting things over to using bitmaps
1997 ctl
->extents_thresh
= ((1024 * 32) / 2) /
1998 sizeof(struct btrfs_free_space
);
2002 * for a given cluster, put all of its extents back into the free
2003 * space cache. If the block group passed doesn't match the block group
2004 * pointed to by the cluster, someone else raced in and freed the
2005 * cluster already. In that case, we just return without changing anything
2008 __btrfs_return_cluster_to_free_space(
2009 struct btrfs_block_group_cache
*block_group
,
2010 struct btrfs_free_cluster
*cluster
)
2012 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2013 struct btrfs_free_space
*entry
;
2014 struct rb_node
*node
;
2016 spin_lock(&cluster
->lock
);
2017 if (cluster
->block_group
!= block_group
)
2020 cluster
->block_group
= NULL
;
2021 cluster
->window_start
= 0;
2022 list_del_init(&cluster
->block_group_list
);
2024 node
= rb_first(&cluster
->root
);
2028 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2029 node
= rb_next(&entry
->offset_index
);
2030 rb_erase(&entry
->offset_index
, &cluster
->root
);
2032 bitmap
= (entry
->bitmap
!= NULL
);
2034 try_merge_free_space(ctl
, entry
, false);
2035 tree_insert_offset(&ctl
->free_space_offset
,
2036 entry
->offset
, &entry
->offset_index
, bitmap
);
2038 cluster
->root
= RB_ROOT
;
2041 spin_unlock(&cluster
->lock
);
2042 btrfs_put_block_group(block_group
);
2046 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl
*ctl
)
2048 struct btrfs_free_space
*info
;
2049 struct rb_node
*node
;
2051 while ((node
= rb_last(&ctl
->free_space_offset
)) != NULL
) {
2052 info
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2053 if (!info
->bitmap
) {
2054 unlink_free_space(ctl
, info
);
2055 kmem_cache_free(btrfs_free_space_cachep
, info
);
2057 free_bitmap(ctl
, info
);
2059 if (need_resched()) {
2060 spin_unlock(&ctl
->tree_lock
);
2062 spin_lock(&ctl
->tree_lock
);
2067 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl
*ctl
)
2069 spin_lock(&ctl
->tree_lock
);
2070 __btrfs_remove_free_space_cache_locked(ctl
);
2071 spin_unlock(&ctl
->tree_lock
);
2074 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache
*block_group
)
2076 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2077 struct btrfs_free_cluster
*cluster
;
2078 struct list_head
*head
;
2080 spin_lock(&ctl
->tree_lock
);
2081 while ((head
= block_group
->cluster_list
.next
) !=
2082 &block_group
->cluster_list
) {
2083 cluster
= list_entry(head
, struct btrfs_free_cluster
,
2086 WARN_ON(cluster
->block_group
!= block_group
);
2087 __btrfs_return_cluster_to_free_space(block_group
, cluster
);
2088 if (need_resched()) {
2089 spin_unlock(&ctl
->tree_lock
);
2091 spin_lock(&ctl
->tree_lock
);
2094 __btrfs_remove_free_space_cache_locked(ctl
);
2095 spin_unlock(&ctl
->tree_lock
);
2099 u64
btrfs_find_space_for_alloc(struct btrfs_block_group_cache
*block_group
,
2100 u64 offset
, u64 bytes
, u64 empty_size
)
2102 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2103 struct btrfs_free_space
*entry
= NULL
;
2104 u64 bytes_search
= bytes
+ empty_size
;
2107 spin_lock(&ctl
->tree_lock
);
2108 entry
= find_free_space(ctl
, &offset
, &bytes_search
);
2113 if (entry
->bitmap
) {
2114 bitmap_clear_bits(ctl
, entry
, offset
, bytes
);
2116 free_bitmap(ctl
, entry
);
2118 unlink_free_space(ctl
, entry
);
2119 entry
->offset
+= bytes
;
2120 entry
->bytes
-= bytes
;
2122 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2124 link_free_space(ctl
, entry
);
2128 spin_unlock(&ctl
->tree_lock
);
2134 * given a cluster, put all of its extents back into the free space
2135 * cache. If a block group is passed, this function will only free
2136 * a cluster that belongs to the passed block group.
2138 * Otherwise, it'll get a reference on the block group pointed to by the
2139 * cluster and remove the cluster from it.
2141 int btrfs_return_cluster_to_free_space(
2142 struct btrfs_block_group_cache
*block_group
,
2143 struct btrfs_free_cluster
*cluster
)
2145 struct btrfs_free_space_ctl
*ctl
;
2148 /* first, get a safe pointer to the block group */
2149 spin_lock(&cluster
->lock
);
2151 block_group
= cluster
->block_group
;
2153 spin_unlock(&cluster
->lock
);
2156 } else if (cluster
->block_group
!= block_group
) {
2157 /* someone else has already freed it don't redo their work */
2158 spin_unlock(&cluster
->lock
);
2161 atomic_inc(&block_group
->count
);
2162 spin_unlock(&cluster
->lock
);
2164 ctl
= block_group
->free_space_ctl
;
2166 /* now return any extents the cluster had on it */
2167 spin_lock(&ctl
->tree_lock
);
2168 ret
= __btrfs_return_cluster_to_free_space(block_group
, cluster
);
2169 spin_unlock(&ctl
->tree_lock
);
2171 /* finally drop our ref */
2172 btrfs_put_block_group(block_group
);
2176 static u64
btrfs_alloc_from_bitmap(struct btrfs_block_group_cache
*block_group
,
2177 struct btrfs_free_cluster
*cluster
,
2178 struct btrfs_free_space
*entry
,
2179 u64 bytes
, u64 min_start
)
2181 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2183 u64 search_start
= cluster
->window_start
;
2184 u64 search_bytes
= bytes
;
2187 search_start
= min_start
;
2188 search_bytes
= bytes
;
2190 err
= search_bitmap(ctl
, entry
, &search_start
, &search_bytes
);
2195 __bitmap_clear_bits(ctl
, entry
, ret
, bytes
);
2201 * given a cluster, try to allocate 'bytes' from it, returns 0
2202 * if it couldn't find anything suitably large, or a logical disk offset
2203 * if things worked out
2205 u64
btrfs_alloc_from_cluster(struct btrfs_block_group_cache
*block_group
,
2206 struct btrfs_free_cluster
*cluster
, u64 bytes
,
2209 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2210 struct btrfs_free_space
*entry
= NULL
;
2211 struct rb_node
*node
;
2214 spin_lock(&cluster
->lock
);
2215 if (bytes
> cluster
->max_size
)
2218 if (cluster
->block_group
!= block_group
)
2221 node
= rb_first(&cluster
->root
);
2225 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2227 if (entry
->bytes
< bytes
||
2228 (!entry
->bitmap
&& entry
->offset
< min_start
)) {
2229 node
= rb_next(&entry
->offset_index
);
2232 entry
= rb_entry(node
, struct btrfs_free_space
,
2237 if (entry
->bitmap
) {
2238 ret
= btrfs_alloc_from_bitmap(block_group
,
2239 cluster
, entry
, bytes
,
2242 node
= rb_next(&entry
->offset_index
);
2245 entry
= rb_entry(node
, struct btrfs_free_space
,
2250 ret
= entry
->offset
;
2252 entry
->offset
+= bytes
;
2253 entry
->bytes
-= bytes
;
2256 if (entry
->bytes
== 0)
2257 rb_erase(&entry
->offset_index
, &cluster
->root
);
2261 spin_unlock(&cluster
->lock
);
2266 spin_lock(&ctl
->tree_lock
);
2268 ctl
->free_space
-= bytes
;
2269 if (entry
->bytes
== 0) {
2270 ctl
->free_extents
--;
2271 if (entry
->bitmap
) {
2272 kfree(entry
->bitmap
);
2273 ctl
->total_bitmaps
--;
2274 ctl
->op
->recalc_thresholds(ctl
);
2276 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2279 spin_unlock(&ctl
->tree_lock
);
2284 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache
*block_group
,
2285 struct btrfs_free_space
*entry
,
2286 struct btrfs_free_cluster
*cluster
,
2287 u64 offset
, u64 bytes
, u64 min_bytes
)
2289 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2290 unsigned long next_zero
;
2292 unsigned long search_bits
;
2293 unsigned long total_bits
;
2294 unsigned long found_bits
;
2295 unsigned long start
= 0;
2296 unsigned long total_found
= 0;
2300 i
= offset_to_bit(entry
->offset
, block_group
->sectorsize
,
2301 max_t(u64
, offset
, entry
->offset
));
2302 search_bits
= bytes_to_bits(bytes
, block_group
->sectorsize
);
2303 total_bits
= bytes_to_bits(min_bytes
, block_group
->sectorsize
);
2307 for (i
= find_next_bit(entry
->bitmap
, BITS_PER_BITMAP
, i
);
2308 i
< BITS_PER_BITMAP
;
2309 i
= find_next_bit(entry
->bitmap
, BITS_PER_BITMAP
, i
+ 1)) {
2310 next_zero
= find_next_zero_bit(entry
->bitmap
,
2311 BITS_PER_BITMAP
, i
);
2312 if (next_zero
- i
>= search_bits
) {
2313 found_bits
= next_zero
- i
;
2324 cluster
->max_size
= 0;
2328 total_found
+= found_bits
;
2330 if (cluster
->max_size
< found_bits
* block_group
->sectorsize
)
2331 cluster
->max_size
= found_bits
* block_group
->sectorsize
;
2333 if (total_found
< total_bits
) {
2334 i
= find_next_bit(entry
->bitmap
, BITS_PER_BITMAP
, next_zero
);
2335 if (i
- start
> total_bits
* 2) {
2337 cluster
->max_size
= 0;
2343 cluster
->window_start
= start
* block_group
->sectorsize
+
2345 rb_erase(&entry
->offset_index
, &ctl
->free_space_offset
);
2346 ret
= tree_insert_offset(&cluster
->root
, entry
->offset
,
2347 &entry
->offset_index
, 1);
2354 * This searches the block group for just extents to fill the cluster with.
2357 setup_cluster_no_bitmap(struct btrfs_block_group_cache
*block_group
,
2358 struct btrfs_free_cluster
*cluster
,
2359 struct list_head
*bitmaps
, u64 offset
, u64 bytes
,
2362 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2363 struct btrfs_free_space
*first
= NULL
;
2364 struct btrfs_free_space
*entry
= NULL
;
2365 struct btrfs_free_space
*prev
= NULL
;
2366 struct btrfs_free_space
*last
;
2367 struct rb_node
*node
;
2371 u64 max_gap
= 128 * 1024;
2373 entry
= tree_search_offset(ctl
, offset
, 0, 1);
2378 * We don't want bitmaps, so just move along until we find a normal
2381 while (entry
->bitmap
) {
2382 if (list_empty(&entry
->list
))
2383 list_add_tail(&entry
->list
, bitmaps
);
2384 node
= rb_next(&entry
->offset_index
);
2387 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2390 window_start
= entry
->offset
;
2391 window_free
= entry
->bytes
;
2392 max_extent
= entry
->bytes
;
2397 while (window_free
<= min_bytes
) {
2398 node
= rb_next(&entry
->offset_index
);
2401 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2403 if (entry
->bitmap
) {
2404 if (list_empty(&entry
->list
))
2405 list_add_tail(&entry
->list
, bitmaps
);
2410 * we haven't filled the empty size and the window is
2411 * very large. reset and try again
2413 if (entry
->offset
- (prev
->offset
+ prev
->bytes
) > max_gap
||
2414 entry
->offset
- window_start
> (min_bytes
* 2)) {
2416 window_start
= entry
->offset
;
2417 window_free
= entry
->bytes
;
2419 max_extent
= entry
->bytes
;
2422 window_free
+= entry
->bytes
;
2423 if (entry
->bytes
> max_extent
)
2424 max_extent
= entry
->bytes
;
2429 cluster
->window_start
= first
->offset
;
2431 node
= &first
->offset_index
;
2434 * now we've found our entries, pull them out of the free space
2435 * cache and put them into the cluster rbtree
2440 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2441 node
= rb_next(&entry
->offset_index
);
2445 rb_erase(&entry
->offset_index
, &ctl
->free_space_offset
);
2446 ret
= tree_insert_offset(&cluster
->root
, entry
->offset
,
2447 &entry
->offset_index
, 0);
2449 } while (node
&& entry
!= last
);
2451 cluster
->max_size
= max_extent
;
2457 * This specifically looks for bitmaps that may work in the cluster, we assume
2458 * that we have already failed to find extents that will work.
2461 setup_cluster_bitmap(struct btrfs_block_group_cache
*block_group
,
2462 struct btrfs_free_cluster
*cluster
,
2463 struct list_head
*bitmaps
, u64 offset
, u64 bytes
,
2466 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2467 struct btrfs_free_space
*entry
;
2469 u64 bitmap_offset
= offset_to_bitmap(ctl
, offset
);
2471 if (ctl
->total_bitmaps
== 0)
2475 * The bitmap that covers offset won't be in the list unless offset
2476 * is just its start offset.
2478 entry
= list_first_entry(bitmaps
, struct btrfs_free_space
, list
);
2479 if (entry
->offset
!= bitmap_offset
) {
2480 entry
= tree_search_offset(ctl
, bitmap_offset
, 1, 0);
2481 if (entry
&& list_empty(&entry
->list
))
2482 list_add(&entry
->list
, bitmaps
);
2485 list_for_each_entry(entry
, bitmaps
, list
) {
2486 if (entry
->bytes
< min_bytes
)
2488 ret
= btrfs_bitmap_cluster(block_group
, entry
, cluster
, offset
,
2495 * The bitmaps list has all the bitmaps that record free space
2496 * starting after offset, so no more search is required.
2502 * here we try to find a cluster of blocks in a block group. The goal
2503 * is to find at least bytes free and up to empty_size + bytes free.
2504 * We might not find them all in one contiguous area.
2506 * returns zero and sets up cluster if things worked out, otherwise
2507 * it returns -enospc
2509 int btrfs_find_space_cluster(struct btrfs_trans_handle
*trans
,
2510 struct btrfs_root
*root
,
2511 struct btrfs_block_group_cache
*block_group
,
2512 struct btrfs_free_cluster
*cluster
,
2513 u64 offset
, u64 bytes
, u64 empty_size
)
2515 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2516 struct btrfs_free_space
*entry
, *tmp
;
2521 /* for metadata, allow allocates with more holes */
2522 if (btrfs_test_opt(root
, SSD_SPREAD
)) {
2523 min_bytes
= bytes
+ empty_size
;
2524 } else if (block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
2526 * we want to do larger allocations when we are
2527 * flushing out the delayed refs, it helps prevent
2528 * making more work as we go along.
2530 if (trans
->transaction
->delayed_refs
.flushing
)
2531 min_bytes
= max(bytes
, (bytes
+ empty_size
) >> 1);
2533 min_bytes
= max(bytes
, (bytes
+ empty_size
) >> 4);
2535 min_bytes
= max(bytes
, (bytes
+ empty_size
) >> 2);
2537 spin_lock(&ctl
->tree_lock
);
2540 * If we know we don't have enough space to make a cluster don't even
2541 * bother doing all the work to try and find one.
2543 if (ctl
->free_space
< min_bytes
) {
2544 spin_unlock(&ctl
->tree_lock
);
2548 spin_lock(&cluster
->lock
);
2550 /* someone already found a cluster, hooray */
2551 if (cluster
->block_group
) {
2556 ret
= setup_cluster_no_bitmap(block_group
, cluster
, &bitmaps
, offset
,
2559 ret
= setup_cluster_bitmap(block_group
, cluster
, &bitmaps
,
2560 offset
, bytes
, min_bytes
);
2562 /* Clear our temporary list */
2563 list_for_each_entry_safe(entry
, tmp
, &bitmaps
, list
)
2564 list_del_init(&entry
->list
);
2567 atomic_inc(&block_group
->count
);
2568 list_add_tail(&cluster
->block_group_list
,
2569 &block_group
->cluster_list
);
2570 cluster
->block_group
= block_group
;
2573 spin_unlock(&cluster
->lock
);
2574 spin_unlock(&ctl
->tree_lock
);
2580 * simple code to zero out a cluster
2582 void btrfs_init_free_cluster(struct btrfs_free_cluster
*cluster
)
2584 spin_lock_init(&cluster
->lock
);
2585 spin_lock_init(&cluster
->refill_lock
);
2586 cluster
->root
= RB_ROOT
;
2587 cluster
->max_size
= 0;
2588 INIT_LIST_HEAD(&cluster
->block_group_list
);
2589 cluster
->block_group
= NULL
;
2592 int btrfs_trim_block_group(struct btrfs_block_group_cache
*block_group
,
2593 u64
*trimmed
, u64 start
, u64 end
, u64 minlen
)
2595 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2596 struct btrfs_free_space
*entry
= NULL
;
2597 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
2599 u64 actually_trimmed
;
2604 while (start
< end
) {
2605 spin_lock(&ctl
->tree_lock
);
2607 if (ctl
->free_space
< minlen
) {
2608 spin_unlock(&ctl
->tree_lock
);
2612 entry
= tree_search_offset(ctl
, start
, 0, 1);
2614 entry
= tree_search_offset(ctl
,
2615 offset_to_bitmap(ctl
, start
),
2618 if (!entry
|| entry
->offset
>= end
) {
2619 spin_unlock(&ctl
->tree_lock
);
2623 if (entry
->bitmap
) {
2624 ret
= search_bitmap(ctl
, entry
, &start
, &bytes
);
2627 spin_unlock(&ctl
->tree_lock
);
2630 bytes
= min(bytes
, end
- start
);
2631 bitmap_clear_bits(ctl
, entry
, start
, bytes
);
2632 if (entry
->bytes
== 0)
2633 free_bitmap(ctl
, entry
);
2635 start
= entry
->offset
+ BITS_PER_BITMAP
*
2636 block_group
->sectorsize
;
2637 spin_unlock(&ctl
->tree_lock
);
2642 start
= entry
->offset
;
2643 bytes
= min(entry
->bytes
, end
- start
);
2644 unlink_free_space(ctl
, entry
);
2645 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2648 spin_unlock(&ctl
->tree_lock
);
2650 if (bytes
>= minlen
) {
2651 struct btrfs_space_info
*space_info
;
2654 space_info
= block_group
->space_info
;
2655 spin_lock(&space_info
->lock
);
2656 spin_lock(&block_group
->lock
);
2657 if (!block_group
->ro
) {
2658 block_group
->reserved
+= bytes
;
2659 space_info
->bytes_reserved
+= bytes
;
2662 spin_unlock(&block_group
->lock
);
2663 spin_unlock(&space_info
->lock
);
2665 ret
= btrfs_error_discard_extent(fs_info
->extent_root
,
2670 btrfs_add_free_space(block_group
, start
, bytes
);
2672 spin_lock(&space_info
->lock
);
2673 spin_lock(&block_group
->lock
);
2674 if (block_group
->ro
)
2675 space_info
->bytes_readonly
+= bytes
;
2676 block_group
->reserved
-= bytes
;
2677 space_info
->bytes_reserved
-= bytes
;
2678 spin_unlock(&space_info
->lock
);
2679 spin_unlock(&block_group
->lock
);
2684 *trimmed
+= actually_trimmed
;
2689 if (fatal_signal_pending(current
)) {
2701 * Find the left-most item in the cache tree, and then return the
2702 * smallest inode number in the item.
2704 * Note: the returned inode number may not be the smallest one in
2705 * the tree, if the left-most item is a bitmap.
2707 u64
btrfs_find_ino_for_alloc(struct btrfs_root
*fs_root
)
2709 struct btrfs_free_space_ctl
*ctl
= fs_root
->free_ino_ctl
;
2710 struct btrfs_free_space
*entry
= NULL
;
2713 spin_lock(&ctl
->tree_lock
);
2715 if (RB_EMPTY_ROOT(&ctl
->free_space_offset
))
2718 entry
= rb_entry(rb_first(&ctl
->free_space_offset
),
2719 struct btrfs_free_space
, offset_index
);
2721 if (!entry
->bitmap
) {
2722 ino
= entry
->offset
;
2724 unlink_free_space(ctl
, entry
);
2728 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2730 link_free_space(ctl
, entry
);
2736 ret
= search_bitmap(ctl
, entry
, &offset
, &count
);
2740 bitmap_clear_bits(ctl
, entry
, offset
, 1);
2741 if (entry
->bytes
== 0)
2742 free_bitmap(ctl
, entry
);
2745 spin_unlock(&ctl
->tree_lock
);
2750 struct inode
*lookup_free_ino_inode(struct btrfs_root
*root
,
2751 struct btrfs_path
*path
)
2753 struct inode
*inode
= NULL
;
2755 spin_lock(&root
->cache_lock
);
2756 if (root
->cache_inode
)
2757 inode
= igrab(root
->cache_inode
);
2758 spin_unlock(&root
->cache_lock
);
2762 inode
= __lookup_free_space_inode(root
, path
, 0);
2766 spin_lock(&root
->cache_lock
);
2767 if (!btrfs_fs_closing(root
->fs_info
))
2768 root
->cache_inode
= igrab(inode
);
2769 spin_unlock(&root
->cache_lock
);
2774 int create_free_ino_inode(struct btrfs_root
*root
,
2775 struct btrfs_trans_handle
*trans
,
2776 struct btrfs_path
*path
)
2778 return __create_free_space_inode(root
, trans
, path
,
2779 BTRFS_FREE_INO_OBJECTID
, 0);
2782 int load_free_ino_cache(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
2784 struct btrfs_free_space_ctl
*ctl
= root
->free_ino_ctl
;
2785 struct btrfs_path
*path
;
2786 struct inode
*inode
;
2788 u64 root_gen
= btrfs_root_generation(&root
->root_item
);
2790 if (!btrfs_test_opt(root
, INODE_MAP_CACHE
))
2794 * If we're unmounting then just return, since this does a search on the
2795 * normal root and not the commit root and we could deadlock.
2797 if (btrfs_fs_closing(fs_info
))
2800 path
= btrfs_alloc_path();
2804 inode
= lookup_free_ino_inode(root
, path
);
2808 if (root_gen
!= BTRFS_I(inode
)->generation
)
2811 ret
= __load_free_space_cache(root
, inode
, ctl
, path
, 0);
2814 printk(KERN_ERR
"btrfs: failed to load free ino cache for "
2815 "root %llu\n", root
->root_key
.objectid
);
2819 btrfs_free_path(path
);
2823 int btrfs_write_out_ino_cache(struct btrfs_root
*root
,
2824 struct btrfs_trans_handle
*trans
,
2825 struct btrfs_path
*path
)
2827 struct btrfs_free_space_ctl
*ctl
= root
->free_ino_ctl
;
2828 struct inode
*inode
;
2831 if (!btrfs_test_opt(root
, INODE_MAP_CACHE
))
2834 inode
= lookup_free_ino_inode(root
, path
);
2838 ret
= __btrfs_write_out_cache(root
, inode
, ctl
, NULL
, trans
, path
, 0);
2840 btrfs_delalloc_release_metadata(inode
, inode
->i_size
);
2842 printk(KERN_ERR
"btrfs: failed to write free ino cache "
2843 "for root %llu\n", root
->root_key
.objectid
);