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"
32 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
33 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
35 struct btrfs_trim_range
{
38 struct list_head list
;
41 static int link_free_space(struct btrfs_free_space_ctl
*ctl
,
42 struct btrfs_free_space
*info
);
43 static void unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
44 struct btrfs_free_space
*info
);
46 static struct inode
*__lookup_free_space_inode(struct btrfs_root
*root
,
47 struct btrfs_path
*path
,
51 struct btrfs_key location
;
52 struct btrfs_disk_key disk_key
;
53 struct btrfs_free_space_header
*header
;
54 struct extent_buffer
*leaf
;
55 struct inode
*inode
= NULL
;
58 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
62 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
66 btrfs_release_path(path
);
67 return ERR_PTR(-ENOENT
);
70 leaf
= path
->nodes
[0];
71 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
72 struct btrfs_free_space_header
);
73 btrfs_free_space_key(leaf
, header
, &disk_key
);
74 btrfs_disk_key_to_cpu(&location
, &disk_key
);
75 btrfs_release_path(path
);
77 inode
= btrfs_iget(root
->fs_info
->sb
, &location
, root
, NULL
);
79 return ERR_PTR(-ENOENT
);
82 if (is_bad_inode(inode
)) {
84 return ERR_PTR(-ENOENT
);
87 mapping_set_gfp_mask(inode
->i_mapping
,
88 mapping_gfp_mask(inode
->i_mapping
) &
89 ~(__GFP_FS
| __GFP_HIGHMEM
));
94 struct inode
*lookup_free_space_inode(struct btrfs_root
*root
,
95 struct btrfs_block_group_cache
96 *block_group
, struct btrfs_path
*path
)
98 struct inode
*inode
= NULL
;
99 u32 flags
= BTRFS_INODE_NODATASUM
| BTRFS_INODE_NODATACOW
;
101 spin_lock(&block_group
->lock
);
102 if (block_group
->inode
)
103 inode
= igrab(block_group
->inode
);
104 spin_unlock(&block_group
->lock
);
108 inode
= __lookup_free_space_inode(root
, path
,
109 block_group
->key
.objectid
);
113 spin_lock(&block_group
->lock
);
114 if (!((BTRFS_I(inode
)->flags
& flags
) == flags
)) {
115 btrfs_info(root
->fs_info
,
116 "Old style space inode found, converting.");
117 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
|
118 BTRFS_INODE_NODATACOW
;
119 block_group
->disk_cache_state
= BTRFS_DC_CLEAR
;
122 if (!block_group
->iref
) {
123 block_group
->inode
= igrab(inode
);
124 block_group
->iref
= 1;
126 spin_unlock(&block_group
->lock
);
131 static int __create_free_space_inode(struct btrfs_root
*root
,
132 struct btrfs_trans_handle
*trans
,
133 struct btrfs_path
*path
,
136 struct btrfs_key key
;
137 struct btrfs_disk_key disk_key
;
138 struct btrfs_free_space_header
*header
;
139 struct btrfs_inode_item
*inode_item
;
140 struct extent_buffer
*leaf
;
141 u64 flags
= BTRFS_INODE_NOCOMPRESS
| BTRFS_INODE_PREALLOC
;
144 ret
= btrfs_insert_empty_inode(trans
, root
, path
, ino
);
148 /* We inline crc's for the free disk space cache */
149 if (ino
!= BTRFS_FREE_INO_OBJECTID
)
150 flags
|= BTRFS_INODE_NODATASUM
| BTRFS_INODE_NODATACOW
;
152 leaf
= path
->nodes
[0];
153 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
154 struct btrfs_inode_item
);
155 btrfs_item_key(leaf
, &disk_key
, path
->slots
[0]);
156 memset_extent_buffer(leaf
, 0, (unsigned long)inode_item
,
157 sizeof(*inode_item
));
158 btrfs_set_inode_generation(leaf
, inode_item
, trans
->transid
);
159 btrfs_set_inode_size(leaf
, inode_item
, 0);
160 btrfs_set_inode_nbytes(leaf
, inode_item
, 0);
161 btrfs_set_inode_uid(leaf
, inode_item
, 0);
162 btrfs_set_inode_gid(leaf
, inode_item
, 0);
163 btrfs_set_inode_mode(leaf
, inode_item
, S_IFREG
| 0600);
164 btrfs_set_inode_flags(leaf
, inode_item
, flags
);
165 btrfs_set_inode_nlink(leaf
, inode_item
, 1);
166 btrfs_set_inode_transid(leaf
, inode_item
, trans
->transid
);
167 btrfs_set_inode_block_group(leaf
, inode_item
, offset
);
168 btrfs_mark_buffer_dirty(leaf
);
169 btrfs_release_path(path
);
171 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
174 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
175 sizeof(struct btrfs_free_space_header
));
177 btrfs_release_path(path
);
181 leaf
= path
->nodes
[0];
182 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
183 struct btrfs_free_space_header
);
184 memset_extent_buffer(leaf
, 0, (unsigned long)header
, sizeof(*header
));
185 btrfs_set_free_space_key(leaf
, header
, &disk_key
);
186 btrfs_mark_buffer_dirty(leaf
);
187 btrfs_release_path(path
);
192 int create_free_space_inode(struct btrfs_root
*root
,
193 struct btrfs_trans_handle
*trans
,
194 struct btrfs_block_group_cache
*block_group
,
195 struct btrfs_path
*path
)
200 ret
= btrfs_find_free_objectid(root
, &ino
);
204 return __create_free_space_inode(root
, trans
, path
, ino
,
205 block_group
->key
.objectid
);
208 int btrfs_check_trunc_cache_free_space(struct btrfs_root
*root
,
209 struct btrfs_block_rsv
*rsv
)
214 /* 1 for slack space, 1 for updating the inode */
215 needed_bytes
= btrfs_calc_trunc_metadata_size(root
, 1) +
216 btrfs_calc_trans_metadata_size(root
, 1);
218 spin_lock(&rsv
->lock
);
219 if (rsv
->reserved
< needed_bytes
)
223 spin_unlock(&rsv
->lock
);
227 int btrfs_truncate_free_space_cache(struct btrfs_root
*root
,
228 struct btrfs_trans_handle
*trans
,
229 struct btrfs_block_group_cache
*block_group
,
233 struct btrfs_path
*path
= btrfs_alloc_path();
243 mutex_lock(&trans
->transaction
->cache_write_mutex
);
244 if (!list_empty(&block_group
->io_list
)) {
245 list_del_init(&block_group
->io_list
);
247 btrfs_wait_cache_io(root
, trans
, block_group
,
248 &block_group
->io_ctl
, path
,
249 block_group
->key
.objectid
);
250 btrfs_put_block_group(block_group
);
254 * now that we've truncated the cache away, its no longer
257 spin_lock(&block_group
->lock
);
258 block_group
->disk_cache_state
= BTRFS_DC_CLEAR
;
259 spin_unlock(&block_group
->lock
);
261 btrfs_free_path(path
);
263 btrfs_i_size_write(inode
, 0);
264 truncate_pagecache(inode
, 0);
267 * We don't need an orphan item because truncating the free space cache
268 * will never be split across transactions.
269 * We don't need to check for -EAGAIN because we're a free space
272 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
273 0, BTRFS_EXTENT_DATA_KEY
);
277 ret
= btrfs_update_inode(trans
, root
, inode
);
281 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
283 btrfs_abort_transaction(trans
, root
, ret
);
288 static int readahead_cache(struct inode
*inode
)
290 struct file_ra_state
*ra
;
291 unsigned long last_index
;
293 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
297 file_ra_state_init(ra
, inode
->i_mapping
);
298 last_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
300 page_cache_sync_readahead(inode
->i_mapping
, ra
, NULL
, 0, last_index
);
307 static int io_ctl_init(struct btrfs_io_ctl
*io_ctl
, struct inode
*inode
,
308 struct btrfs_root
*root
, int write
)
313 num_pages
= DIV_ROUND_UP(i_size_read(inode
), PAGE_CACHE_SIZE
);
315 if (btrfs_ino(inode
) != BTRFS_FREE_INO_OBJECTID
)
318 /* Make sure we can fit our crcs into the first page */
319 if (write
&& check_crcs
&&
320 (num_pages
* sizeof(u32
)) >= PAGE_CACHE_SIZE
)
323 memset(io_ctl
, 0, sizeof(struct btrfs_io_ctl
));
325 io_ctl
->pages
= kcalloc(num_pages
, sizeof(struct page
*), GFP_NOFS
);
329 io_ctl
->num_pages
= num_pages
;
331 io_ctl
->check_crcs
= check_crcs
;
332 io_ctl
->inode
= inode
;
337 static void io_ctl_free(struct btrfs_io_ctl
*io_ctl
)
339 kfree(io_ctl
->pages
);
340 io_ctl
->pages
= NULL
;
343 static void io_ctl_unmap_page(struct btrfs_io_ctl
*io_ctl
)
351 static void io_ctl_map_page(struct btrfs_io_ctl
*io_ctl
, int clear
)
353 ASSERT(io_ctl
->index
< io_ctl
->num_pages
);
354 io_ctl
->page
= io_ctl
->pages
[io_ctl
->index
++];
355 io_ctl
->cur
= page_address(io_ctl
->page
);
356 io_ctl
->orig
= io_ctl
->cur
;
357 io_ctl
->size
= PAGE_CACHE_SIZE
;
359 memset(io_ctl
->cur
, 0, PAGE_CACHE_SIZE
);
362 static void io_ctl_drop_pages(struct btrfs_io_ctl
*io_ctl
)
366 io_ctl_unmap_page(io_ctl
);
368 for (i
= 0; i
< io_ctl
->num_pages
; i
++) {
369 if (io_ctl
->pages
[i
]) {
370 ClearPageChecked(io_ctl
->pages
[i
]);
371 unlock_page(io_ctl
->pages
[i
]);
372 page_cache_release(io_ctl
->pages
[i
]);
377 static int io_ctl_prepare_pages(struct btrfs_io_ctl
*io_ctl
, struct inode
*inode
,
381 gfp_t mask
= btrfs_alloc_write_mask(inode
->i_mapping
);
384 for (i
= 0; i
< io_ctl
->num_pages
; i
++) {
385 page
= find_or_create_page(inode
->i_mapping
, i
, mask
);
387 io_ctl_drop_pages(io_ctl
);
390 io_ctl
->pages
[i
] = page
;
391 if (uptodate
&& !PageUptodate(page
)) {
392 btrfs_readpage(NULL
, page
);
394 if (!PageUptodate(page
)) {
395 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
396 "error reading free space cache");
397 io_ctl_drop_pages(io_ctl
);
403 for (i
= 0; i
< io_ctl
->num_pages
; i
++) {
404 clear_page_dirty_for_io(io_ctl
->pages
[i
]);
405 set_page_extent_mapped(io_ctl
->pages
[i
]);
411 static void io_ctl_set_generation(struct btrfs_io_ctl
*io_ctl
, u64 generation
)
415 io_ctl_map_page(io_ctl
, 1);
418 * Skip the csum areas. If we don't check crcs then we just have a
419 * 64bit chunk at the front of the first page.
421 if (io_ctl
->check_crcs
) {
422 io_ctl
->cur
+= (sizeof(u32
) * io_ctl
->num_pages
);
423 io_ctl
->size
-= sizeof(u64
) + (sizeof(u32
) * io_ctl
->num_pages
);
425 io_ctl
->cur
+= sizeof(u64
);
426 io_ctl
->size
-= sizeof(u64
) * 2;
430 *val
= cpu_to_le64(generation
);
431 io_ctl
->cur
+= sizeof(u64
);
434 static int io_ctl_check_generation(struct btrfs_io_ctl
*io_ctl
, u64 generation
)
439 * Skip the crc area. If we don't check crcs then we just have a 64bit
440 * chunk at the front of the first page.
442 if (io_ctl
->check_crcs
) {
443 io_ctl
->cur
+= sizeof(u32
) * io_ctl
->num_pages
;
444 io_ctl
->size
-= sizeof(u64
) +
445 (sizeof(u32
) * io_ctl
->num_pages
);
447 io_ctl
->cur
+= sizeof(u64
);
448 io_ctl
->size
-= sizeof(u64
) * 2;
452 if (le64_to_cpu(*gen
) != generation
) {
453 btrfs_err_rl(io_ctl
->root
->fs_info
,
454 "space cache generation (%llu) does not match inode (%llu)",
456 io_ctl_unmap_page(io_ctl
);
459 io_ctl
->cur
+= sizeof(u64
);
463 static void io_ctl_set_crc(struct btrfs_io_ctl
*io_ctl
, int index
)
469 if (!io_ctl
->check_crcs
) {
470 io_ctl_unmap_page(io_ctl
);
475 offset
= sizeof(u32
) * io_ctl
->num_pages
;
477 crc
= btrfs_csum_data(io_ctl
->orig
+ offset
, crc
,
478 PAGE_CACHE_SIZE
- offset
);
479 btrfs_csum_final(crc
, (char *)&crc
);
480 io_ctl_unmap_page(io_ctl
);
481 tmp
= page_address(io_ctl
->pages
[0]);
486 static int io_ctl_check_crc(struct btrfs_io_ctl
*io_ctl
, int index
)
492 if (!io_ctl
->check_crcs
) {
493 io_ctl_map_page(io_ctl
, 0);
498 offset
= sizeof(u32
) * io_ctl
->num_pages
;
500 tmp
= page_address(io_ctl
->pages
[0]);
504 io_ctl_map_page(io_ctl
, 0);
505 crc
= btrfs_csum_data(io_ctl
->orig
+ offset
, crc
,
506 PAGE_CACHE_SIZE
- offset
);
507 btrfs_csum_final(crc
, (char *)&crc
);
509 btrfs_err_rl(io_ctl
->root
->fs_info
,
510 "csum mismatch on free space cache");
511 io_ctl_unmap_page(io_ctl
);
518 static int io_ctl_add_entry(struct btrfs_io_ctl
*io_ctl
, u64 offset
, u64 bytes
,
521 struct btrfs_free_space_entry
*entry
;
527 entry
->offset
= cpu_to_le64(offset
);
528 entry
->bytes
= cpu_to_le64(bytes
);
529 entry
->type
= (bitmap
) ? BTRFS_FREE_SPACE_BITMAP
:
530 BTRFS_FREE_SPACE_EXTENT
;
531 io_ctl
->cur
+= sizeof(struct btrfs_free_space_entry
);
532 io_ctl
->size
-= sizeof(struct btrfs_free_space_entry
);
534 if (io_ctl
->size
>= sizeof(struct btrfs_free_space_entry
))
537 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
539 /* No more pages to map */
540 if (io_ctl
->index
>= io_ctl
->num_pages
)
543 /* map the next page */
544 io_ctl_map_page(io_ctl
, 1);
548 static int io_ctl_add_bitmap(struct btrfs_io_ctl
*io_ctl
, void *bitmap
)
554 * If we aren't at the start of the current page, unmap this one and
555 * map the next one if there is any left.
557 if (io_ctl
->cur
!= io_ctl
->orig
) {
558 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
559 if (io_ctl
->index
>= io_ctl
->num_pages
)
561 io_ctl_map_page(io_ctl
, 0);
564 memcpy(io_ctl
->cur
, bitmap
, PAGE_CACHE_SIZE
);
565 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
566 if (io_ctl
->index
< io_ctl
->num_pages
)
567 io_ctl_map_page(io_ctl
, 0);
571 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl
*io_ctl
)
574 * If we're not on the boundary we know we've modified the page and we
575 * need to crc the page.
577 if (io_ctl
->cur
!= io_ctl
->orig
)
578 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
580 io_ctl_unmap_page(io_ctl
);
582 while (io_ctl
->index
< io_ctl
->num_pages
) {
583 io_ctl_map_page(io_ctl
, 1);
584 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
588 static int io_ctl_read_entry(struct btrfs_io_ctl
*io_ctl
,
589 struct btrfs_free_space
*entry
, u8
*type
)
591 struct btrfs_free_space_entry
*e
;
595 ret
= io_ctl_check_crc(io_ctl
, io_ctl
->index
);
601 entry
->offset
= le64_to_cpu(e
->offset
);
602 entry
->bytes
= le64_to_cpu(e
->bytes
);
604 io_ctl
->cur
+= sizeof(struct btrfs_free_space_entry
);
605 io_ctl
->size
-= sizeof(struct btrfs_free_space_entry
);
607 if (io_ctl
->size
>= sizeof(struct btrfs_free_space_entry
))
610 io_ctl_unmap_page(io_ctl
);
615 static int io_ctl_read_bitmap(struct btrfs_io_ctl
*io_ctl
,
616 struct btrfs_free_space
*entry
)
620 ret
= io_ctl_check_crc(io_ctl
, io_ctl
->index
);
624 memcpy(entry
->bitmap
, io_ctl
->cur
, PAGE_CACHE_SIZE
);
625 io_ctl_unmap_page(io_ctl
);
631 * Since we attach pinned extents after the fact we can have contiguous sections
632 * of free space that are split up in entries. This poses a problem with the
633 * tree logging stuff since it could have allocated across what appears to be 2
634 * entries since we would have merged the entries when adding the pinned extents
635 * back to the free space cache. So run through the space cache that we just
636 * loaded and merge contiguous entries. This will make the log replay stuff not
637 * blow up and it will make for nicer allocator behavior.
639 static void merge_space_tree(struct btrfs_free_space_ctl
*ctl
)
641 struct btrfs_free_space
*e
, *prev
= NULL
;
645 spin_lock(&ctl
->tree_lock
);
646 for (n
= rb_first(&ctl
->free_space_offset
); n
; n
= rb_next(n
)) {
647 e
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
650 if (e
->bitmap
|| prev
->bitmap
)
652 if (prev
->offset
+ prev
->bytes
== e
->offset
) {
653 unlink_free_space(ctl
, prev
);
654 unlink_free_space(ctl
, e
);
655 prev
->bytes
+= e
->bytes
;
656 kmem_cache_free(btrfs_free_space_cachep
, e
);
657 link_free_space(ctl
, prev
);
659 spin_unlock(&ctl
->tree_lock
);
665 spin_unlock(&ctl
->tree_lock
);
668 static int __load_free_space_cache(struct btrfs_root
*root
, struct inode
*inode
,
669 struct btrfs_free_space_ctl
*ctl
,
670 struct btrfs_path
*path
, u64 offset
)
672 struct btrfs_free_space_header
*header
;
673 struct extent_buffer
*leaf
;
674 struct btrfs_io_ctl io_ctl
;
675 struct btrfs_key key
;
676 struct btrfs_free_space
*e
, *n
;
684 /* Nothing in the space cache, goodbye */
685 if (!i_size_read(inode
))
688 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
692 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
696 btrfs_release_path(path
);
702 leaf
= path
->nodes
[0];
703 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
704 struct btrfs_free_space_header
);
705 num_entries
= btrfs_free_space_entries(leaf
, header
);
706 num_bitmaps
= btrfs_free_space_bitmaps(leaf
, header
);
707 generation
= btrfs_free_space_generation(leaf
, header
);
708 btrfs_release_path(path
);
710 if (!BTRFS_I(inode
)->generation
) {
711 btrfs_info(root
->fs_info
,
712 "The free space cache file (%llu) is invalid. skip it\n",
717 if (BTRFS_I(inode
)->generation
!= generation
) {
718 btrfs_err(root
->fs_info
,
719 "free space inode generation (%llu) "
720 "did not match free space cache generation (%llu)",
721 BTRFS_I(inode
)->generation
, generation
);
728 ret
= io_ctl_init(&io_ctl
, inode
, root
, 0);
732 ret
= readahead_cache(inode
);
736 ret
= io_ctl_prepare_pages(&io_ctl
, inode
, 1);
740 ret
= io_ctl_check_crc(&io_ctl
, 0);
744 ret
= io_ctl_check_generation(&io_ctl
, generation
);
748 while (num_entries
) {
749 e
= kmem_cache_zalloc(btrfs_free_space_cachep
,
754 ret
= io_ctl_read_entry(&io_ctl
, e
, &type
);
756 kmem_cache_free(btrfs_free_space_cachep
, e
);
761 kmem_cache_free(btrfs_free_space_cachep
, e
);
765 if (type
== BTRFS_FREE_SPACE_EXTENT
) {
766 spin_lock(&ctl
->tree_lock
);
767 ret
= link_free_space(ctl
, e
);
768 spin_unlock(&ctl
->tree_lock
);
770 btrfs_err(root
->fs_info
,
771 "Duplicate entries in free space cache, dumping");
772 kmem_cache_free(btrfs_free_space_cachep
, e
);
778 e
->bitmap
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
781 btrfs_free_space_cachep
, e
);
784 spin_lock(&ctl
->tree_lock
);
785 ret
= link_free_space(ctl
, e
);
786 ctl
->total_bitmaps
++;
787 ctl
->op
->recalc_thresholds(ctl
);
788 spin_unlock(&ctl
->tree_lock
);
790 btrfs_err(root
->fs_info
,
791 "Duplicate entries in free space cache, dumping");
792 kmem_cache_free(btrfs_free_space_cachep
, e
);
795 list_add_tail(&e
->list
, &bitmaps
);
801 io_ctl_unmap_page(&io_ctl
);
804 * We add the bitmaps at the end of the entries in order that
805 * the bitmap entries are added to the cache.
807 list_for_each_entry_safe(e
, n
, &bitmaps
, list
) {
808 list_del_init(&e
->list
);
809 ret
= io_ctl_read_bitmap(&io_ctl
, e
);
814 io_ctl_drop_pages(&io_ctl
);
815 merge_space_tree(ctl
);
818 io_ctl_free(&io_ctl
);
821 io_ctl_drop_pages(&io_ctl
);
822 __btrfs_remove_free_space_cache(ctl
);
826 int load_free_space_cache(struct btrfs_fs_info
*fs_info
,
827 struct btrfs_block_group_cache
*block_group
)
829 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
830 struct btrfs_root
*root
= fs_info
->tree_root
;
832 struct btrfs_path
*path
;
835 u64 used
= btrfs_block_group_used(&block_group
->item
);
838 * If this block group has been marked to be cleared for one reason or
839 * another then we can't trust the on disk cache, so just return.
841 spin_lock(&block_group
->lock
);
842 if (block_group
->disk_cache_state
!= BTRFS_DC_WRITTEN
) {
843 spin_unlock(&block_group
->lock
);
846 spin_unlock(&block_group
->lock
);
848 path
= btrfs_alloc_path();
851 path
->search_commit_root
= 1;
852 path
->skip_locking
= 1;
854 inode
= lookup_free_space_inode(root
, block_group
, path
);
856 btrfs_free_path(path
);
860 /* We may have converted the inode and made the cache invalid. */
861 spin_lock(&block_group
->lock
);
862 if (block_group
->disk_cache_state
!= BTRFS_DC_WRITTEN
) {
863 spin_unlock(&block_group
->lock
);
864 btrfs_free_path(path
);
867 spin_unlock(&block_group
->lock
);
869 ret
= __load_free_space_cache(fs_info
->tree_root
, inode
, ctl
,
870 path
, block_group
->key
.objectid
);
871 btrfs_free_path(path
);
875 spin_lock(&ctl
->tree_lock
);
876 matched
= (ctl
->free_space
== (block_group
->key
.offset
- used
-
877 block_group
->bytes_super
));
878 spin_unlock(&ctl
->tree_lock
);
881 __btrfs_remove_free_space_cache(ctl
);
882 btrfs_warn(fs_info
, "block group %llu has wrong amount of free space",
883 block_group
->key
.objectid
);
888 /* This cache is bogus, make sure it gets cleared */
889 spin_lock(&block_group
->lock
);
890 block_group
->disk_cache_state
= BTRFS_DC_CLEAR
;
891 spin_unlock(&block_group
->lock
);
894 btrfs_warn(fs_info
, "failed to load free space cache for block group %llu, rebuild it now",
895 block_group
->key
.objectid
);
902 static noinline_for_stack
903 int write_cache_extent_entries(struct btrfs_io_ctl
*io_ctl
,
904 struct btrfs_free_space_ctl
*ctl
,
905 struct btrfs_block_group_cache
*block_group
,
906 int *entries
, int *bitmaps
,
907 struct list_head
*bitmap_list
)
910 struct btrfs_free_cluster
*cluster
= NULL
;
911 struct btrfs_free_cluster
*cluster_locked
= NULL
;
912 struct rb_node
*node
= rb_first(&ctl
->free_space_offset
);
913 struct btrfs_trim_range
*trim_entry
;
915 /* Get the cluster for this block_group if it exists */
916 if (block_group
&& !list_empty(&block_group
->cluster_list
)) {
917 cluster
= list_entry(block_group
->cluster_list
.next
,
918 struct btrfs_free_cluster
,
922 if (!node
&& cluster
) {
923 cluster_locked
= cluster
;
924 spin_lock(&cluster_locked
->lock
);
925 node
= rb_first(&cluster
->root
);
929 /* Write out the extent entries */
931 struct btrfs_free_space
*e
;
933 e
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
936 ret
= io_ctl_add_entry(io_ctl
, e
->offset
, e
->bytes
,
942 list_add_tail(&e
->list
, bitmap_list
);
945 node
= rb_next(node
);
946 if (!node
&& cluster
) {
947 node
= rb_first(&cluster
->root
);
948 cluster_locked
= cluster
;
949 spin_lock(&cluster_locked
->lock
);
953 if (cluster_locked
) {
954 spin_unlock(&cluster_locked
->lock
);
955 cluster_locked
= NULL
;
959 * Make sure we don't miss any range that was removed from our rbtree
960 * because trimming is running. Otherwise after a umount+mount (or crash
961 * after committing the transaction) we would leak free space and get
962 * an inconsistent free space cache report from fsck.
964 list_for_each_entry(trim_entry
, &ctl
->trimming_ranges
, list
) {
965 ret
= io_ctl_add_entry(io_ctl
, trim_entry
->start
,
966 trim_entry
->bytes
, NULL
);
975 spin_unlock(&cluster_locked
->lock
);
979 static noinline_for_stack
int
980 update_cache_item(struct btrfs_trans_handle
*trans
,
981 struct btrfs_root
*root
,
983 struct btrfs_path
*path
, u64 offset
,
984 int entries
, int bitmaps
)
986 struct btrfs_key key
;
987 struct btrfs_free_space_header
*header
;
988 struct extent_buffer
*leaf
;
991 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
995 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
997 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0, inode
->i_size
- 1,
998 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, NULL
,
1002 leaf
= path
->nodes
[0];
1004 struct btrfs_key found_key
;
1005 ASSERT(path
->slots
[0]);
1007 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1008 if (found_key
.objectid
!= BTRFS_FREE_SPACE_OBJECTID
||
1009 found_key
.offset
!= offset
) {
1010 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0,
1012 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0,
1014 btrfs_release_path(path
);
1019 BTRFS_I(inode
)->generation
= trans
->transid
;
1020 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
1021 struct btrfs_free_space_header
);
1022 btrfs_set_free_space_entries(leaf
, header
, entries
);
1023 btrfs_set_free_space_bitmaps(leaf
, header
, bitmaps
);
1024 btrfs_set_free_space_generation(leaf
, header
, trans
->transid
);
1025 btrfs_mark_buffer_dirty(leaf
);
1026 btrfs_release_path(path
);
1034 static noinline_for_stack
int
1035 write_pinned_extent_entries(struct btrfs_root
*root
,
1036 struct btrfs_block_group_cache
*block_group
,
1037 struct btrfs_io_ctl
*io_ctl
,
1040 u64 start
, extent_start
, extent_end
, len
;
1041 struct extent_io_tree
*unpin
= NULL
;
1048 * We want to add any pinned extents to our free space cache
1049 * so we don't leak the space
1051 * We shouldn't have switched the pinned extents yet so this is the
1054 unpin
= root
->fs_info
->pinned_extents
;
1056 start
= block_group
->key
.objectid
;
1058 while (start
< block_group
->key
.objectid
+ block_group
->key
.offset
) {
1059 ret
= find_first_extent_bit(unpin
, start
,
1060 &extent_start
, &extent_end
,
1061 EXTENT_DIRTY
, NULL
);
1065 /* This pinned extent is out of our range */
1066 if (extent_start
>= block_group
->key
.objectid
+
1067 block_group
->key
.offset
)
1070 extent_start
= max(extent_start
, start
);
1071 extent_end
= min(block_group
->key
.objectid
+
1072 block_group
->key
.offset
, extent_end
+ 1);
1073 len
= extent_end
- extent_start
;
1076 ret
= io_ctl_add_entry(io_ctl
, extent_start
, len
, NULL
);
1086 static noinline_for_stack
int
1087 write_bitmap_entries(struct btrfs_io_ctl
*io_ctl
, struct list_head
*bitmap_list
)
1089 struct list_head
*pos
, *n
;
1092 /* Write out the bitmaps */
1093 list_for_each_safe(pos
, n
, bitmap_list
) {
1094 struct btrfs_free_space
*entry
=
1095 list_entry(pos
, struct btrfs_free_space
, list
);
1097 ret
= io_ctl_add_bitmap(io_ctl
, entry
->bitmap
);
1100 list_del_init(&entry
->list
);
1106 static int flush_dirty_cache(struct inode
*inode
)
1110 ret
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
1112 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0, inode
->i_size
- 1,
1113 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, NULL
,
1119 static void noinline_for_stack
1120 cleanup_bitmap_list(struct list_head
*bitmap_list
)
1122 struct list_head
*pos
, *n
;
1124 list_for_each_safe(pos
, n
, bitmap_list
) {
1125 struct btrfs_free_space
*entry
=
1126 list_entry(pos
, struct btrfs_free_space
, list
);
1127 list_del_init(&entry
->list
);
1131 static void noinline_for_stack
1132 cleanup_write_cache_enospc(struct inode
*inode
,
1133 struct btrfs_io_ctl
*io_ctl
,
1134 struct extent_state
**cached_state
,
1135 struct list_head
*bitmap_list
)
1137 io_ctl_drop_pages(io_ctl
);
1138 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, 0,
1139 i_size_read(inode
) - 1, cached_state
,
1143 int btrfs_wait_cache_io(struct btrfs_root
*root
,
1144 struct btrfs_trans_handle
*trans
,
1145 struct btrfs_block_group_cache
*block_group
,
1146 struct btrfs_io_ctl
*io_ctl
,
1147 struct btrfs_path
*path
, u64 offset
)
1150 struct inode
*inode
= io_ctl
->inode
;
1156 root
= root
->fs_info
->tree_root
;
1158 /* Flush the dirty pages in the cache file. */
1159 ret
= flush_dirty_cache(inode
);
1163 /* Update the cache item to tell everyone this cache file is valid. */
1164 ret
= update_cache_item(trans
, root
, inode
, path
, offset
,
1165 io_ctl
->entries
, io_ctl
->bitmaps
);
1167 io_ctl_free(io_ctl
);
1169 invalidate_inode_pages2(inode
->i_mapping
);
1170 BTRFS_I(inode
)->generation
= 0;
1173 btrfs_err(root
->fs_info
,
1174 "failed to write free space cache for block group %llu",
1175 block_group
->key
.objectid
);
1179 btrfs_update_inode(trans
, root
, inode
);
1182 /* the dirty list is protected by the dirty_bgs_lock */
1183 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
1185 /* the disk_cache_state is protected by the block group lock */
1186 spin_lock(&block_group
->lock
);
1189 * only mark this as written if we didn't get put back on
1190 * the dirty list while waiting for IO. Otherwise our
1191 * cache state won't be right, and we won't get written again
1193 if (!ret
&& list_empty(&block_group
->dirty_list
))
1194 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
1196 block_group
->disk_cache_state
= BTRFS_DC_ERROR
;
1198 spin_unlock(&block_group
->lock
);
1199 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
1200 io_ctl
->inode
= NULL
;
1209 * __btrfs_write_out_cache - write out cached info to an inode
1210 * @root - the root the inode belongs to
1211 * @ctl - the free space cache we are going to write out
1212 * @block_group - the block_group for this cache if it belongs to a block_group
1213 * @trans - the trans handle
1214 * @path - the path to use
1215 * @offset - the offset for the key we'll insert
1217 * This function writes out a free space cache struct to disk for quick recovery
1218 * on mount. This will return 0 if it was successful in writing the cache out,
1219 * or an errno if it was not.
1221 static int __btrfs_write_out_cache(struct btrfs_root
*root
, struct inode
*inode
,
1222 struct btrfs_free_space_ctl
*ctl
,
1223 struct btrfs_block_group_cache
*block_group
,
1224 struct btrfs_io_ctl
*io_ctl
,
1225 struct btrfs_trans_handle
*trans
,
1226 struct btrfs_path
*path
, u64 offset
)
1228 struct extent_state
*cached_state
= NULL
;
1229 LIST_HEAD(bitmap_list
);
1235 if (!i_size_read(inode
))
1238 WARN_ON(io_ctl
->pages
);
1239 ret
= io_ctl_init(io_ctl
, inode
, root
, 1);
1243 if (block_group
&& (block_group
->flags
& BTRFS_BLOCK_GROUP_DATA
)) {
1244 down_write(&block_group
->data_rwsem
);
1245 spin_lock(&block_group
->lock
);
1246 if (block_group
->delalloc_bytes
) {
1247 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
1248 spin_unlock(&block_group
->lock
);
1249 up_write(&block_group
->data_rwsem
);
1250 BTRFS_I(inode
)->generation
= 0;
1255 spin_unlock(&block_group
->lock
);
1258 /* Lock all pages first so we can lock the extent safely. */
1259 ret
= io_ctl_prepare_pages(io_ctl
, inode
, 0);
1263 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, 0, i_size_read(inode
) - 1,
1266 io_ctl_set_generation(io_ctl
, trans
->transid
);
1268 mutex_lock(&ctl
->cache_writeout_mutex
);
1269 /* Write out the extent entries in the free space cache */
1270 spin_lock(&ctl
->tree_lock
);
1271 ret
= write_cache_extent_entries(io_ctl
, ctl
,
1272 block_group
, &entries
, &bitmaps
,
1275 goto out_nospc_locked
;
1278 * Some spaces that are freed in the current transaction are pinned,
1279 * they will be added into free space cache after the transaction is
1280 * committed, we shouldn't lose them.
1282 * If this changes while we are working we'll get added back to
1283 * the dirty list and redo it. No locking needed
1285 ret
= write_pinned_extent_entries(root
, block_group
, io_ctl
, &entries
);
1287 goto out_nospc_locked
;
1290 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1291 * locked while doing it because a concurrent trim can be manipulating
1292 * or freeing the bitmap.
1294 ret
= write_bitmap_entries(io_ctl
, &bitmap_list
);
1295 spin_unlock(&ctl
->tree_lock
);
1296 mutex_unlock(&ctl
->cache_writeout_mutex
);
1300 /* Zero out the rest of the pages just to make sure */
1301 io_ctl_zero_remaining_pages(io_ctl
);
1303 /* Everything is written out, now we dirty the pages in the file. */
1304 ret
= btrfs_dirty_pages(root
, inode
, io_ctl
->pages
, io_ctl
->num_pages
,
1305 0, i_size_read(inode
), &cached_state
);
1309 if (block_group
&& (block_group
->flags
& BTRFS_BLOCK_GROUP_DATA
))
1310 up_write(&block_group
->data_rwsem
);
1312 * Release the pages and unlock the extent, we will flush
1315 io_ctl_drop_pages(io_ctl
);
1317 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, 0,
1318 i_size_read(inode
) - 1, &cached_state
, GFP_NOFS
);
1321 * at this point the pages are under IO and we're happy,
1322 * The caller is responsible for waiting on them and updating the
1323 * the cache and the inode
1325 io_ctl
->entries
= entries
;
1326 io_ctl
->bitmaps
= bitmaps
;
1328 ret
= btrfs_fdatawrite_range(inode
, 0, (u64
)-1);
1335 io_ctl
->inode
= NULL
;
1336 io_ctl_free(io_ctl
);
1338 invalidate_inode_pages2(inode
->i_mapping
);
1339 BTRFS_I(inode
)->generation
= 0;
1341 btrfs_update_inode(trans
, root
, inode
);
1347 cleanup_bitmap_list(&bitmap_list
);
1348 spin_unlock(&ctl
->tree_lock
);
1349 mutex_unlock(&ctl
->cache_writeout_mutex
);
1352 cleanup_write_cache_enospc(inode
, io_ctl
, &cached_state
, &bitmap_list
);
1354 if (block_group
&& (block_group
->flags
& BTRFS_BLOCK_GROUP_DATA
))
1355 up_write(&block_group
->data_rwsem
);
1360 int btrfs_write_out_cache(struct btrfs_root
*root
,
1361 struct btrfs_trans_handle
*trans
,
1362 struct btrfs_block_group_cache
*block_group
,
1363 struct btrfs_path
*path
)
1365 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1366 struct inode
*inode
;
1369 root
= root
->fs_info
->tree_root
;
1371 spin_lock(&block_group
->lock
);
1372 if (block_group
->disk_cache_state
< BTRFS_DC_SETUP
) {
1373 spin_unlock(&block_group
->lock
);
1376 spin_unlock(&block_group
->lock
);
1378 inode
= lookup_free_space_inode(root
, block_group
, path
);
1382 ret
= __btrfs_write_out_cache(root
, inode
, ctl
, block_group
,
1383 &block_group
->io_ctl
, trans
,
1384 path
, block_group
->key
.objectid
);
1387 btrfs_err(root
->fs_info
,
1388 "failed to write free space cache for block group %llu",
1389 block_group
->key
.objectid
);
1391 spin_lock(&block_group
->lock
);
1392 block_group
->disk_cache_state
= BTRFS_DC_ERROR
;
1393 spin_unlock(&block_group
->lock
);
1395 block_group
->io_ctl
.inode
= NULL
;
1400 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1401 * to wait for IO and put the inode
1407 static inline unsigned long offset_to_bit(u64 bitmap_start
, u32 unit
,
1410 ASSERT(offset
>= bitmap_start
);
1411 offset
-= bitmap_start
;
1412 return (unsigned long)(div_u64(offset
, unit
));
1415 static inline unsigned long bytes_to_bits(u64 bytes
, u32 unit
)
1417 return (unsigned long)(div_u64(bytes
, unit
));
1420 static inline u64
offset_to_bitmap(struct btrfs_free_space_ctl
*ctl
,
1424 u32 bytes_per_bitmap
;
1426 bytes_per_bitmap
= BITS_PER_BITMAP
* ctl
->unit
;
1427 bitmap_start
= offset
- ctl
->start
;
1428 bitmap_start
= div_u64(bitmap_start
, bytes_per_bitmap
);
1429 bitmap_start
*= bytes_per_bitmap
;
1430 bitmap_start
+= ctl
->start
;
1432 return bitmap_start
;
1435 static int tree_insert_offset(struct rb_root
*root
, u64 offset
,
1436 struct rb_node
*node
, int bitmap
)
1438 struct rb_node
**p
= &root
->rb_node
;
1439 struct rb_node
*parent
= NULL
;
1440 struct btrfs_free_space
*info
;
1444 info
= rb_entry(parent
, struct btrfs_free_space
, offset_index
);
1446 if (offset
< info
->offset
) {
1448 } else if (offset
> info
->offset
) {
1449 p
= &(*p
)->rb_right
;
1452 * we could have a bitmap entry and an extent entry
1453 * share the same offset. If this is the case, we want
1454 * the extent entry to always be found first if we do a
1455 * linear search through the tree, since we want to have
1456 * the quickest allocation time, and allocating from an
1457 * extent is faster than allocating from a bitmap. So
1458 * if we're inserting a bitmap and we find an entry at
1459 * this offset, we want to go right, or after this entry
1460 * logically. If we are inserting an extent and we've
1461 * found a bitmap, we want to go left, or before
1469 p
= &(*p
)->rb_right
;
1471 if (!info
->bitmap
) {
1480 rb_link_node(node
, parent
, p
);
1481 rb_insert_color(node
, root
);
1487 * searches the tree for the given offset.
1489 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1490 * want a section that has at least bytes size and comes at or after the given
1493 static struct btrfs_free_space
*
1494 tree_search_offset(struct btrfs_free_space_ctl
*ctl
,
1495 u64 offset
, int bitmap_only
, int fuzzy
)
1497 struct rb_node
*n
= ctl
->free_space_offset
.rb_node
;
1498 struct btrfs_free_space
*entry
, *prev
= NULL
;
1500 /* find entry that is closest to the 'offset' */
1507 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1510 if (offset
< entry
->offset
)
1512 else if (offset
> entry
->offset
)
1525 * bitmap entry and extent entry may share same offset,
1526 * in that case, bitmap entry comes after extent entry.
1531 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1532 if (entry
->offset
!= offset
)
1535 WARN_ON(!entry
->bitmap
);
1538 if (entry
->bitmap
) {
1540 * if previous extent entry covers the offset,
1541 * we should return it instead of the bitmap entry
1543 n
= rb_prev(&entry
->offset_index
);
1545 prev
= rb_entry(n
, struct btrfs_free_space
,
1547 if (!prev
->bitmap
&&
1548 prev
->offset
+ prev
->bytes
> offset
)
1558 /* find last entry before the 'offset' */
1560 if (entry
->offset
> offset
) {
1561 n
= rb_prev(&entry
->offset_index
);
1563 entry
= rb_entry(n
, struct btrfs_free_space
,
1565 ASSERT(entry
->offset
<= offset
);
1574 if (entry
->bitmap
) {
1575 n
= rb_prev(&entry
->offset_index
);
1577 prev
= rb_entry(n
, struct btrfs_free_space
,
1579 if (!prev
->bitmap
&&
1580 prev
->offset
+ prev
->bytes
> offset
)
1583 if (entry
->offset
+ BITS_PER_BITMAP
* ctl
->unit
> offset
)
1585 } else if (entry
->offset
+ entry
->bytes
> offset
)
1592 if (entry
->bitmap
) {
1593 if (entry
->offset
+ BITS_PER_BITMAP
*
1597 if (entry
->offset
+ entry
->bytes
> offset
)
1601 n
= rb_next(&entry
->offset_index
);
1604 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1610 __unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
1611 struct btrfs_free_space
*info
)
1613 rb_erase(&info
->offset_index
, &ctl
->free_space_offset
);
1614 ctl
->free_extents
--;
1617 static void unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
1618 struct btrfs_free_space
*info
)
1620 __unlink_free_space(ctl
, info
);
1621 ctl
->free_space
-= info
->bytes
;
1624 static int link_free_space(struct btrfs_free_space_ctl
*ctl
,
1625 struct btrfs_free_space
*info
)
1629 ASSERT(info
->bytes
|| info
->bitmap
);
1630 ret
= tree_insert_offset(&ctl
->free_space_offset
, info
->offset
,
1631 &info
->offset_index
, (info
->bitmap
!= NULL
));
1635 ctl
->free_space
+= info
->bytes
;
1636 ctl
->free_extents
++;
1640 static void recalculate_thresholds(struct btrfs_free_space_ctl
*ctl
)
1642 struct btrfs_block_group_cache
*block_group
= ctl
->private;
1646 u64 size
= block_group
->key
.offset
;
1647 u32 bytes_per_bg
= BITS_PER_BITMAP
* ctl
->unit
;
1648 u32 max_bitmaps
= div_u64(size
+ bytes_per_bg
- 1, bytes_per_bg
);
1650 max_bitmaps
= max_t(u32
, max_bitmaps
, 1);
1652 ASSERT(ctl
->total_bitmaps
<= max_bitmaps
);
1655 * The goal is to keep the total amount of memory used per 1gb of space
1656 * at or below 32k, so we need to adjust how much memory we allow to be
1657 * used by extent based free space tracking
1659 if (size
< 1024 * 1024 * 1024)
1660 max_bytes
= MAX_CACHE_BYTES_PER_GIG
;
1662 max_bytes
= MAX_CACHE_BYTES_PER_GIG
*
1663 div_u64(size
, 1024 * 1024 * 1024);
1666 * we want to account for 1 more bitmap than what we have so we can make
1667 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1668 * we add more bitmaps.
1670 bitmap_bytes
= (ctl
->total_bitmaps
+ 1) * PAGE_CACHE_SIZE
;
1672 if (bitmap_bytes
>= max_bytes
) {
1673 ctl
->extents_thresh
= 0;
1678 * we want the extent entry threshold to always be at most 1/2 the max
1679 * bytes we can have, or whatever is less than that.
1681 extent_bytes
= max_bytes
- bitmap_bytes
;
1682 extent_bytes
= min_t(u64
, extent_bytes
, max_bytes
>> 1);
1684 ctl
->extents_thresh
=
1685 div_u64(extent_bytes
, sizeof(struct btrfs_free_space
));
1688 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl
*ctl
,
1689 struct btrfs_free_space
*info
,
1690 u64 offset
, u64 bytes
)
1692 unsigned long start
, count
;
1694 start
= offset_to_bit(info
->offset
, ctl
->unit
, offset
);
1695 count
= bytes_to_bits(bytes
, ctl
->unit
);
1696 ASSERT(start
+ count
<= BITS_PER_BITMAP
);
1698 bitmap_clear(info
->bitmap
, start
, count
);
1700 info
->bytes
-= bytes
;
1703 static void bitmap_clear_bits(struct btrfs_free_space_ctl
*ctl
,
1704 struct btrfs_free_space
*info
, u64 offset
,
1707 __bitmap_clear_bits(ctl
, info
, offset
, bytes
);
1708 ctl
->free_space
-= bytes
;
1711 static void bitmap_set_bits(struct btrfs_free_space_ctl
*ctl
,
1712 struct btrfs_free_space
*info
, u64 offset
,
1715 unsigned long start
, count
;
1717 start
= offset_to_bit(info
->offset
, ctl
->unit
, offset
);
1718 count
= bytes_to_bits(bytes
, ctl
->unit
);
1719 ASSERT(start
+ count
<= BITS_PER_BITMAP
);
1721 bitmap_set(info
->bitmap
, start
, count
);
1723 info
->bytes
+= bytes
;
1724 ctl
->free_space
+= bytes
;
1728 * If we can not find suitable extent, we will use bytes to record
1729 * the size of the max extent.
1731 static int search_bitmap(struct btrfs_free_space_ctl
*ctl
,
1732 struct btrfs_free_space
*bitmap_info
, u64
*offset
,
1735 unsigned long found_bits
= 0;
1736 unsigned long max_bits
= 0;
1737 unsigned long bits
, i
;
1738 unsigned long next_zero
;
1739 unsigned long extent_bits
;
1742 * Skip searching the bitmap if we don't have a contiguous section that
1743 * is large enough for this allocation.
1745 if (bitmap_info
->max_extent_size
&&
1746 bitmap_info
->max_extent_size
< *bytes
) {
1747 *bytes
= bitmap_info
->max_extent_size
;
1751 i
= offset_to_bit(bitmap_info
->offset
, ctl
->unit
,
1752 max_t(u64
, *offset
, bitmap_info
->offset
));
1753 bits
= bytes_to_bits(*bytes
, ctl
->unit
);
1755 for_each_set_bit_from(i
, bitmap_info
->bitmap
, BITS_PER_BITMAP
) {
1756 next_zero
= find_next_zero_bit(bitmap_info
->bitmap
,
1757 BITS_PER_BITMAP
, i
);
1758 extent_bits
= next_zero
- i
;
1759 if (extent_bits
>= bits
) {
1760 found_bits
= extent_bits
;
1762 } else if (extent_bits
> max_bits
) {
1763 max_bits
= extent_bits
;
1769 *offset
= (u64
)(i
* ctl
->unit
) + bitmap_info
->offset
;
1770 *bytes
= (u64
)(found_bits
) * ctl
->unit
;
1774 *bytes
= (u64
)(max_bits
) * ctl
->unit
;
1775 bitmap_info
->max_extent_size
= *bytes
;
1779 /* Cache the size of the max extent in bytes */
1780 static struct btrfs_free_space
*
1781 find_free_space(struct btrfs_free_space_ctl
*ctl
, u64
*offset
, u64
*bytes
,
1782 unsigned long align
, u64
*max_extent_size
)
1784 struct btrfs_free_space
*entry
;
1785 struct rb_node
*node
;
1790 if (!ctl
->free_space_offset
.rb_node
)
1793 entry
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, *offset
), 0, 1);
1797 for (node
= &entry
->offset_index
; node
; node
= rb_next(node
)) {
1798 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
1799 if (entry
->bytes
< *bytes
) {
1800 if (entry
->bytes
> *max_extent_size
)
1801 *max_extent_size
= entry
->bytes
;
1805 /* make sure the space returned is big enough
1806 * to match our requested alignment
1808 if (*bytes
>= align
) {
1809 tmp
= entry
->offset
- ctl
->start
+ align
- 1;
1810 tmp
= div64_u64(tmp
, align
);
1811 tmp
= tmp
* align
+ ctl
->start
;
1812 align_off
= tmp
- entry
->offset
;
1815 tmp
= entry
->offset
;
1818 if (entry
->bytes
< *bytes
+ align_off
) {
1819 if (entry
->bytes
> *max_extent_size
)
1820 *max_extent_size
= entry
->bytes
;
1824 if (entry
->bitmap
) {
1827 ret
= search_bitmap(ctl
, entry
, &tmp
, &size
);
1832 } else if (size
> *max_extent_size
) {
1833 *max_extent_size
= size
;
1839 *bytes
= entry
->bytes
- align_off
;
1846 static void add_new_bitmap(struct btrfs_free_space_ctl
*ctl
,
1847 struct btrfs_free_space
*info
, u64 offset
)
1849 info
->offset
= offset_to_bitmap(ctl
, offset
);
1851 INIT_LIST_HEAD(&info
->list
);
1852 link_free_space(ctl
, info
);
1853 ctl
->total_bitmaps
++;
1855 ctl
->op
->recalc_thresholds(ctl
);
1858 static void free_bitmap(struct btrfs_free_space_ctl
*ctl
,
1859 struct btrfs_free_space
*bitmap_info
)
1861 unlink_free_space(ctl
, bitmap_info
);
1862 kfree(bitmap_info
->bitmap
);
1863 kmem_cache_free(btrfs_free_space_cachep
, bitmap_info
);
1864 ctl
->total_bitmaps
--;
1865 ctl
->op
->recalc_thresholds(ctl
);
1868 static noinline
int remove_from_bitmap(struct btrfs_free_space_ctl
*ctl
,
1869 struct btrfs_free_space
*bitmap_info
,
1870 u64
*offset
, u64
*bytes
)
1873 u64 search_start
, search_bytes
;
1877 end
= bitmap_info
->offset
+ (u64
)(BITS_PER_BITMAP
* ctl
->unit
) - 1;
1880 * We need to search for bits in this bitmap. We could only cover some
1881 * of the extent in this bitmap thanks to how we add space, so we need
1882 * to search for as much as it as we can and clear that amount, and then
1883 * go searching for the next bit.
1885 search_start
= *offset
;
1886 search_bytes
= ctl
->unit
;
1887 search_bytes
= min(search_bytes
, end
- search_start
+ 1);
1888 ret
= search_bitmap(ctl
, bitmap_info
, &search_start
, &search_bytes
);
1889 if (ret
< 0 || search_start
!= *offset
)
1892 /* We may have found more bits than what we need */
1893 search_bytes
= min(search_bytes
, *bytes
);
1895 /* Cannot clear past the end of the bitmap */
1896 search_bytes
= min(search_bytes
, end
- search_start
+ 1);
1898 bitmap_clear_bits(ctl
, bitmap_info
, search_start
, search_bytes
);
1899 *offset
+= search_bytes
;
1900 *bytes
-= search_bytes
;
1903 struct rb_node
*next
= rb_next(&bitmap_info
->offset_index
);
1904 if (!bitmap_info
->bytes
)
1905 free_bitmap(ctl
, bitmap_info
);
1908 * no entry after this bitmap, but we still have bytes to
1909 * remove, so something has gone wrong.
1914 bitmap_info
= rb_entry(next
, struct btrfs_free_space
,
1918 * if the next entry isn't a bitmap we need to return to let the
1919 * extent stuff do its work.
1921 if (!bitmap_info
->bitmap
)
1925 * Ok the next item is a bitmap, but it may not actually hold
1926 * the information for the rest of this free space stuff, so
1927 * look for it, and if we don't find it return so we can try
1928 * everything over again.
1930 search_start
= *offset
;
1931 search_bytes
= ctl
->unit
;
1932 ret
= search_bitmap(ctl
, bitmap_info
, &search_start
,
1934 if (ret
< 0 || search_start
!= *offset
)
1938 } else if (!bitmap_info
->bytes
)
1939 free_bitmap(ctl
, bitmap_info
);
1944 static u64
add_bytes_to_bitmap(struct btrfs_free_space_ctl
*ctl
,
1945 struct btrfs_free_space
*info
, u64 offset
,
1948 u64 bytes_to_set
= 0;
1951 end
= info
->offset
+ (u64
)(BITS_PER_BITMAP
* ctl
->unit
);
1953 bytes_to_set
= min(end
- offset
, bytes
);
1955 bitmap_set_bits(ctl
, info
, offset
, bytes_to_set
);
1958 * We set some bytes, we have no idea what the max extent size is
1961 info
->max_extent_size
= 0;
1963 return bytes_to_set
;
1967 static bool use_bitmap(struct btrfs_free_space_ctl
*ctl
,
1968 struct btrfs_free_space
*info
)
1970 struct btrfs_block_group_cache
*block_group
= ctl
->private;
1971 bool forced
= false;
1973 #ifdef CONFIG_BTRFS_DEBUG
1974 if (btrfs_should_fragment_free_space(block_group
->fs_info
->extent_root
,
1980 * If we are below the extents threshold then we can add this as an
1981 * extent, and don't have to deal with the bitmap
1983 if (!forced
&& ctl
->free_extents
< ctl
->extents_thresh
) {
1985 * If this block group has some small extents we don't want to
1986 * use up all of our free slots in the cache with them, we want
1987 * to reserve them to larger extents, however if we have plent
1988 * of cache left then go ahead an dadd them, no sense in adding
1989 * the overhead of a bitmap if we don't have to.
1991 if (info
->bytes
<= block_group
->sectorsize
* 4) {
1992 if (ctl
->free_extents
* 2 <= ctl
->extents_thresh
)
2000 * The original block groups from mkfs can be really small, like 8
2001 * megabytes, so don't bother with a bitmap for those entries. However
2002 * some block groups can be smaller than what a bitmap would cover but
2003 * are still large enough that they could overflow the 32k memory limit,
2004 * so allow those block groups to still be allowed to have a bitmap
2007 if (((BITS_PER_BITMAP
* ctl
->unit
) >> 1) > block_group
->key
.offset
)
2013 static struct btrfs_free_space_op free_space_op
= {
2014 .recalc_thresholds
= recalculate_thresholds
,
2015 .use_bitmap
= use_bitmap
,
2018 static int insert_into_bitmap(struct btrfs_free_space_ctl
*ctl
,
2019 struct btrfs_free_space
*info
)
2021 struct btrfs_free_space
*bitmap_info
;
2022 struct btrfs_block_group_cache
*block_group
= NULL
;
2024 u64 bytes
, offset
, bytes_added
;
2027 bytes
= info
->bytes
;
2028 offset
= info
->offset
;
2030 if (!ctl
->op
->use_bitmap(ctl
, info
))
2033 if (ctl
->op
== &free_space_op
)
2034 block_group
= ctl
->private;
2037 * Since we link bitmaps right into the cluster we need to see if we
2038 * have a cluster here, and if so and it has our bitmap we need to add
2039 * the free space to that bitmap.
2041 if (block_group
&& !list_empty(&block_group
->cluster_list
)) {
2042 struct btrfs_free_cluster
*cluster
;
2043 struct rb_node
*node
;
2044 struct btrfs_free_space
*entry
;
2046 cluster
= list_entry(block_group
->cluster_list
.next
,
2047 struct btrfs_free_cluster
,
2049 spin_lock(&cluster
->lock
);
2050 node
= rb_first(&cluster
->root
);
2052 spin_unlock(&cluster
->lock
);
2053 goto no_cluster_bitmap
;
2056 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2057 if (!entry
->bitmap
) {
2058 spin_unlock(&cluster
->lock
);
2059 goto no_cluster_bitmap
;
2062 if (entry
->offset
== offset_to_bitmap(ctl
, offset
)) {
2063 bytes_added
= add_bytes_to_bitmap(ctl
, entry
,
2065 bytes
-= bytes_added
;
2066 offset
+= bytes_added
;
2068 spin_unlock(&cluster
->lock
);
2076 bitmap_info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
2083 bytes_added
= add_bytes_to_bitmap(ctl
, bitmap_info
, offset
, bytes
);
2084 bytes
-= bytes_added
;
2085 offset
+= bytes_added
;
2095 if (info
&& info
->bitmap
) {
2096 add_new_bitmap(ctl
, info
, offset
);
2101 spin_unlock(&ctl
->tree_lock
);
2103 /* no pre-allocated info, allocate a new one */
2105 info
= kmem_cache_zalloc(btrfs_free_space_cachep
,
2108 spin_lock(&ctl
->tree_lock
);
2114 /* allocate the bitmap */
2115 info
->bitmap
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
2116 spin_lock(&ctl
->tree_lock
);
2117 if (!info
->bitmap
) {
2127 kfree(info
->bitmap
);
2128 kmem_cache_free(btrfs_free_space_cachep
, info
);
2134 static bool try_merge_free_space(struct btrfs_free_space_ctl
*ctl
,
2135 struct btrfs_free_space
*info
, bool update_stat
)
2137 struct btrfs_free_space
*left_info
;
2138 struct btrfs_free_space
*right_info
;
2139 bool merged
= false;
2140 u64 offset
= info
->offset
;
2141 u64 bytes
= info
->bytes
;
2144 * first we want to see if there is free space adjacent to the range we
2145 * are adding, if there is remove that struct and add a new one to
2146 * cover the entire range
2148 right_info
= tree_search_offset(ctl
, offset
+ bytes
, 0, 0);
2149 if (right_info
&& rb_prev(&right_info
->offset_index
))
2150 left_info
= rb_entry(rb_prev(&right_info
->offset_index
),
2151 struct btrfs_free_space
, offset_index
);
2153 left_info
= tree_search_offset(ctl
, offset
- 1, 0, 0);
2155 if (right_info
&& !right_info
->bitmap
) {
2157 unlink_free_space(ctl
, right_info
);
2159 __unlink_free_space(ctl
, right_info
);
2160 info
->bytes
+= right_info
->bytes
;
2161 kmem_cache_free(btrfs_free_space_cachep
, right_info
);
2165 if (left_info
&& !left_info
->bitmap
&&
2166 left_info
->offset
+ left_info
->bytes
== offset
) {
2168 unlink_free_space(ctl
, left_info
);
2170 __unlink_free_space(ctl
, left_info
);
2171 info
->offset
= left_info
->offset
;
2172 info
->bytes
+= left_info
->bytes
;
2173 kmem_cache_free(btrfs_free_space_cachep
, left_info
);
2180 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl
*ctl
,
2181 struct btrfs_free_space
*info
,
2184 struct btrfs_free_space
*bitmap
;
2187 const u64 end
= info
->offset
+ info
->bytes
;
2188 const u64 bitmap_offset
= offset_to_bitmap(ctl
, end
);
2191 bitmap
= tree_search_offset(ctl
, bitmap_offset
, 1, 0);
2195 i
= offset_to_bit(bitmap
->offset
, ctl
->unit
, end
);
2196 j
= find_next_zero_bit(bitmap
->bitmap
, BITS_PER_BITMAP
, i
);
2199 bytes
= (j
- i
) * ctl
->unit
;
2200 info
->bytes
+= bytes
;
2203 bitmap_clear_bits(ctl
, bitmap
, end
, bytes
);
2205 __bitmap_clear_bits(ctl
, bitmap
, end
, bytes
);
2208 free_bitmap(ctl
, bitmap
);
2213 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl
*ctl
,
2214 struct btrfs_free_space
*info
,
2217 struct btrfs_free_space
*bitmap
;
2221 unsigned long prev_j
;
2224 bitmap_offset
= offset_to_bitmap(ctl
, info
->offset
);
2225 /* If we're on a boundary, try the previous logical bitmap. */
2226 if (bitmap_offset
== info
->offset
) {
2227 if (info
->offset
== 0)
2229 bitmap_offset
= offset_to_bitmap(ctl
, info
->offset
- 1);
2232 bitmap
= tree_search_offset(ctl
, bitmap_offset
, 1, 0);
2236 i
= offset_to_bit(bitmap
->offset
, ctl
->unit
, info
->offset
) - 1;
2238 prev_j
= (unsigned long)-1;
2239 for_each_clear_bit_from(j
, bitmap
->bitmap
, BITS_PER_BITMAP
) {
2247 if (prev_j
== (unsigned long)-1)
2248 bytes
= (i
+ 1) * ctl
->unit
;
2250 bytes
= (i
- prev_j
) * ctl
->unit
;
2252 info
->offset
-= bytes
;
2253 info
->bytes
+= bytes
;
2256 bitmap_clear_bits(ctl
, bitmap
, info
->offset
, bytes
);
2258 __bitmap_clear_bits(ctl
, bitmap
, info
->offset
, bytes
);
2261 free_bitmap(ctl
, bitmap
);
2267 * We prefer always to allocate from extent entries, both for clustered and
2268 * non-clustered allocation requests. So when attempting to add a new extent
2269 * entry, try to see if there's adjacent free space in bitmap entries, and if
2270 * there is, migrate that space from the bitmaps to the extent.
2271 * Like this we get better chances of satisfying space allocation requests
2272 * because we attempt to satisfy them based on a single cache entry, and never
2273 * on 2 or more entries - even if the entries represent a contiguous free space
2274 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2277 static void steal_from_bitmap(struct btrfs_free_space_ctl
*ctl
,
2278 struct btrfs_free_space
*info
,
2282 * Only work with disconnected entries, as we can change their offset,
2283 * and must be extent entries.
2285 ASSERT(!info
->bitmap
);
2286 ASSERT(RB_EMPTY_NODE(&info
->offset_index
));
2288 if (ctl
->total_bitmaps
> 0) {
2290 bool stole_front
= false;
2292 stole_end
= steal_from_bitmap_to_end(ctl
, info
, update_stat
);
2293 if (ctl
->total_bitmaps
> 0)
2294 stole_front
= steal_from_bitmap_to_front(ctl
, info
,
2297 if (stole_end
|| stole_front
)
2298 try_merge_free_space(ctl
, info
, update_stat
);
2302 int __btrfs_add_free_space(struct btrfs_free_space_ctl
*ctl
,
2303 u64 offset
, u64 bytes
)
2305 struct btrfs_free_space
*info
;
2308 info
= kmem_cache_zalloc(btrfs_free_space_cachep
, GFP_NOFS
);
2312 info
->offset
= offset
;
2313 info
->bytes
= bytes
;
2314 RB_CLEAR_NODE(&info
->offset_index
);
2316 spin_lock(&ctl
->tree_lock
);
2318 if (try_merge_free_space(ctl
, info
, true))
2322 * There was no extent directly to the left or right of this new
2323 * extent then we know we're going to have to allocate a new extent, so
2324 * before we do that see if we need to drop this into a bitmap
2326 ret
= insert_into_bitmap(ctl
, info
);
2335 * Only steal free space from adjacent bitmaps if we're sure we're not
2336 * going to add the new free space to existing bitmap entries - because
2337 * that would mean unnecessary work that would be reverted. Therefore
2338 * attempt to steal space from bitmaps if we're adding an extent entry.
2340 steal_from_bitmap(ctl
, info
, true);
2342 ret
= link_free_space(ctl
, info
);
2344 kmem_cache_free(btrfs_free_space_cachep
, info
);
2346 spin_unlock(&ctl
->tree_lock
);
2349 printk(KERN_CRIT
"BTRFS: unable to add free space :%d\n", ret
);
2350 ASSERT(ret
!= -EEXIST
);
2356 int btrfs_remove_free_space(struct btrfs_block_group_cache
*block_group
,
2357 u64 offset
, u64 bytes
)
2359 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2360 struct btrfs_free_space
*info
;
2362 bool re_search
= false;
2364 spin_lock(&ctl
->tree_lock
);
2371 info
= tree_search_offset(ctl
, offset
, 0, 0);
2374 * oops didn't find an extent that matched the space we wanted
2375 * to remove, look for a bitmap instead
2377 info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
2381 * If we found a partial bit of our free space in a
2382 * bitmap but then couldn't find the other part this may
2383 * be a problem, so WARN about it.
2391 if (!info
->bitmap
) {
2392 unlink_free_space(ctl
, info
);
2393 if (offset
== info
->offset
) {
2394 u64 to_free
= min(bytes
, info
->bytes
);
2396 info
->bytes
-= to_free
;
2397 info
->offset
+= to_free
;
2399 ret
= link_free_space(ctl
, info
);
2402 kmem_cache_free(btrfs_free_space_cachep
, info
);
2409 u64 old_end
= info
->bytes
+ info
->offset
;
2411 info
->bytes
= offset
- info
->offset
;
2412 ret
= link_free_space(ctl
, info
);
2417 /* Not enough bytes in this entry to satisfy us */
2418 if (old_end
< offset
+ bytes
) {
2419 bytes
-= old_end
- offset
;
2422 } else if (old_end
== offset
+ bytes
) {
2426 spin_unlock(&ctl
->tree_lock
);
2428 ret
= btrfs_add_free_space(block_group
, offset
+ bytes
,
2429 old_end
- (offset
+ bytes
));
2435 ret
= remove_from_bitmap(ctl
, info
, &offset
, &bytes
);
2436 if (ret
== -EAGAIN
) {
2441 spin_unlock(&ctl
->tree_lock
);
2446 void btrfs_dump_free_space(struct btrfs_block_group_cache
*block_group
,
2449 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2450 struct btrfs_free_space
*info
;
2454 for (n
= rb_first(&ctl
->free_space_offset
); n
; n
= rb_next(n
)) {
2455 info
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
2456 if (info
->bytes
>= bytes
&& !block_group
->ro
)
2458 btrfs_crit(block_group
->fs_info
,
2459 "entry offset %llu, bytes %llu, bitmap %s",
2460 info
->offset
, info
->bytes
,
2461 (info
->bitmap
) ? "yes" : "no");
2463 btrfs_info(block_group
->fs_info
, "block group has cluster?: %s",
2464 list_empty(&block_group
->cluster_list
) ? "no" : "yes");
2465 btrfs_info(block_group
->fs_info
,
2466 "%d blocks of free space at or bigger than bytes is", count
);
2469 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache
*block_group
)
2471 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2473 spin_lock_init(&ctl
->tree_lock
);
2474 ctl
->unit
= block_group
->sectorsize
;
2475 ctl
->start
= block_group
->key
.objectid
;
2476 ctl
->private = block_group
;
2477 ctl
->op
= &free_space_op
;
2478 INIT_LIST_HEAD(&ctl
->trimming_ranges
);
2479 mutex_init(&ctl
->cache_writeout_mutex
);
2482 * we only want to have 32k of ram per block group for keeping
2483 * track of free space, and if we pass 1/2 of that we want to
2484 * start converting things over to using bitmaps
2486 ctl
->extents_thresh
= ((1024 * 32) / 2) /
2487 sizeof(struct btrfs_free_space
);
2491 * for a given cluster, put all of its extents back into the free
2492 * space cache. If the block group passed doesn't match the block group
2493 * pointed to by the cluster, someone else raced in and freed the
2494 * cluster already. In that case, we just return without changing anything
2497 __btrfs_return_cluster_to_free_space(
2498 struct btrfs_block_group_cache
*block_group
,
2499 struct btrfs_free_cluster
*cluster
)
2501 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2502 struct btrfs_free_space
*entry
;
2503 struct rb_node
*node
;
2505 spin_lock(&cluster
->lock
);
2506 if (cluster
->block_group
!= block_group
)
2509 cluster
->block_group
= NULL
;
2510 cluster
->window_start
= 0;
2511 list_del_init(&cluster
->block_group_list
);
2513 node
= rb_first(&cluster
->root
);
2517 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2518 node
= rb_next(&entry
->offset_index
);
2519 rb_erase(&entry
->offset_index
, &cluster
->root
);
2520 RB_CLEAR_NODE(&entry
->offset_index
);
2522 bitmap
= (entry
->bitmap
!= NULL
);
2524 try_merge_free_space(ctl
, entry
, false);
2525 steal_from_bitmap(ctl
, entry
, false);
2527 tree_insert_offset(&ctl
->free_space_offset
,
2528 entry
->offset
, &entry
->offset_index
, bitmap
);
2530 cluster
->root
= RB_ROOT
;
2533 spin_unlock(&cluster
->lock
);
2534 btrfs_put_block_group(block_group
);
2538 static void __btrfs_remove_free_space_cache_locked(
2539 struct btrfs_free_space_ctl
*ctl
)
2541 struct btrfs_free_space
*info
;
2542 struct rb_node
*node
;
2544 while ((node
= rb_last(&ctl
->free_space_offset
)) != NULL
) {
2545 info
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2546 if (!info
->bitmap
) {
2547 unlink_free_space(ctl
, info
);
2548 kmem_cache_free(btrfs_free_space_cachep
, info
);
2550 free_bitmap(ctl
, info
);
2553 cond_resched_lock(&ctl
->tree_lock
);
2557 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl
*ctl
)
2559 spin_lock(&ctl
->tree_lock
);
2560 __btrfs_remove_free_space_cache_locked(ctl
);
2561 spin_unlock(&ctl
->tree_lock
);
2564 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache
*block_group
)
2566 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2567 struct btrfs_free_cluster
*cluster
;
2568 struct list_head
*head
;
2570 spin_lock(&ctl
->tree_lock
);
2571 while ((head
= block_group
->cluster_list
.next
) !=
2572 &block_group
->cluster_list
) {
2573 cluster
= list_entry(head
, struct btrfs_free_cluster
,
2576 WARN_ON(cluster
->block_group
!= block_group
);
2577 __btrfs_return_cluster_to_free_space(block_group
, cluster
);
2579 cond_resched_lock(&ctl
->tree_lock
);
2581 __btrfs_remove_free_space_cache_locked(ctl
);
2582 spin_unlock(&ctl
->tree_lock
);
2586 u64
btrfs_find_space_for_alloc(struct btrfs_block_group_cache
*block_group
,
2587 u64 offset
, u64 bytes
, u64 empty_size
,
2588 u64
*max_extent_size
)
2590 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2591 struct btrfs_free_space
*entry
= NULL
;
2592 u64 bytes_search
= bytes
+ empty_size
;
2595 u64 align_gap_len
= 0;
2597 spin_lock(&ctl
->tree_lock
);
2598 entry
= find_free_space(ctl
, &offset
, &bytes_search
,
2599 block_group
->full_stripe_len
, max_extent_size
);
2604 if (entry
->bitmap
) {
2605 bitmap_clear_bits(ctl
, entry
, offset
, bytes
);
2607 free_bitmap(ctl
, entry
);
2609 unlink_free_space(ctl
, entry
);
2610 align_gap_len
= offset
- entry
->offset
;
2611 align_gap
= entry
->offset
;
2613 entry
->offset
= offset
+ bytes
;
2614 WARN_ON(entry
->bytes
< bytes
+ align_gap_len
);
2616 entry
->bytes
-= bytes
+ align_gap_len
;
2618 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2620 link_free_space(ctl
, entry
);
2623 spin_unlock(&ctl
->tree_lock
);
2626 __btrfs_add_free_space(ctl
, align_gap
, align_gap_len
);
2631 * given a cluster, put all of its extents back into the free space
2632 * cache. If a block group is passed, this function will only free
2633 * a cluster that belongs to the passed block group.
2635 * Otherwise, it'll get a reference on the block group pointed to by the
2636 * cluster and remove the cluster from it.
2638 int btrfs_return_cluster_to_free_space(
2639 struct btrfs_block_group_cache
*block_group
,
2640 struct btrfs_free_cluster
*cluster
)
2642 struct btrfs_free_space_ctl
*ctl
;
2645 /* first, get a safe pointer to the block group */
2646 spin_lock(&cluster
->lock
);
2648 block_group
= cluster
->block_group
;
2650 spin_unlock(&cluster
->lock
);
2653 } else if (cluster
->block_group
!= block_group
) {
2654 /* someone else has already freed it don't redo their work */
2655 spin_unlock(&cluster
->lock
);
2658 atomic_inc(&block_group
->count
);
2659 spin_unlock(&cluster
->lock
);
2661 ctl
= block_group
->free_space_ctl
;
2663 /* now return any extents the cluster had on it */
2664 spin_lock(&ctl
->tree_lock
);
2665 ret
= __btrfs_return_cluster_to_free_space(block_group
, cluster
);
2666 spin_unlock(&ctl
->tree_lock
);
2668 /* finally drop our ref */
2669 btrfs_put_block_group(block_group
);
2673 static u64
btrfs_alloc_from_bitmap(struct btrfs_block_group_cache
*block_group
,
2674 struct btrfs_free_cluster
*cluster
,
2675 struct btrfs_free_space
*entry
,
2676 u64 bytes
, u64 min_start
,
2677 u64
*max_extent_size
)
2679 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2681 u64 search_start
= cluster
->window_start
;
2682 u64 search_bytes
= bytes
;
2685 search_start
= min_start
;
2686 search_bytes
= bytes
;
2688 err
= search_bitmap(ctl
, entry
, &search_start
, &search_bytes
);
2690 if (search_bytes
> *max_extent_size
)
2691 *max_extent_size
= search_bytes
;
2696 __bitmap_clear_bits(ctl
, entry
, ret
, bytes
);
2702 * given a cluster, try to allocate 'bytes' from it, returns 0
2703 * if it couldn't find anything suitably large, or a logical disk offset
2704 * if things worked out
2706 u64
btrfs_alloc_from_cluster(struct btrfs_block_group_cache
*block_group
,
2707 struct btrfs_free_cluster
*cluster
, u64 bytes
,
2708 u64 min_start
, u64
*max_extent_size
)
2710 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2711 struct btrfs_free_space
*entry
= NULL
;
2712 struct rb_node
*node
;
2715 spin_lock(&cluster
->lock
);
2716 if (bytes
> cluster
->max_size
)
2719 if (cluster
->block_group
!= block_group
)
2722 node
= rb_first(&cluster
->root
);
2726 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2728 if (entry
->bytes
< bytes
&& entry
->bytes
> *max_extent_size
)
2729 *max_extent_size
= entry
->bytes
;
2731 if (entry
->bytes
< bytes
||
2732 (!entry
->bitmap
&& entry
->offset
< min_start
)) {
2733 node
= rb_next(&entry
->offset_index
);
2736 entry
= rb_entry(node
, struct btrfs_free_space
,
2741 if (entry
->bitmap
) {
2742 ret
= btrfs_alloc_from_bitmap(block_group
,
2743 cluster
, entry
, bytes
,
2744 cluster
->window_start
,
2747 node
= rb_next(&entry
->offset_index
);
2750 entry
= rb_entry(node
, struct btrfs_free_space
,
2754 cluster
->window_start
+= bytes
;
2756 ret
= entry
->offset
;
2758 entry
->offset
+= bytes
;
2759 entry
->bytes
-= bytes
;
2762 if (entry
->bytes
== 0)
2763 rb_erase(&entry
->offset_index
, &cluster
->root
);
2767 spin_unlock(&cluster
->lock
);
2772 spin_lock(&ctl
->tree_lock
);
2774 ctl
->free_space
-= bytes
;
2775 if (entry
->bytes
== 0) {
2776 ctl
->free_extents
--;
2777 if (entry
->bitmap
) {
2778 kfree(entry
->bitmap
);
2779 ctl
->total_bitmaps
--;
2780 ctl
->op
->recalc_thresholds(ctl
);
2782 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2785 spin_unlock(&ctl
->tree_lock
);
2790 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache
*block_group
,
2791 struct btrfs_free_space
*entry
,
2792 struct btrfs_free_cluster
*cluster
,
2793 u64 offset
, u64 bytes
,
2794 u64 cont1_bytes
, u64 min_bytes
)
2796 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2797 unsigned long next_zero
;
2799 unsigned long want_bits
;
2800 unsigned long min_bits
;
2801 unsigned long found_bits
;
2802 unsigned long max_bits
= 0;
2803 unsigned long start
= 0;
2804 unsigned long total_found
= 0;
2807 i
= offset_to_bit(entry
->offset
, ctl
->unit
,
2808 max_t(u64
, offset
, entry
->offset
));
2809 want_bits
= bytes_to_bits(bytes
, ctl
->unit
);
2810 min_bits
= bytes_to_bits(min_bytes
, ctl
->unit
);
2813 * Don't bother looking for a cluster in this bitmap if it's heavily
2816 if (entry
->max_extent_size
&&
2817 entry
->max_extent_size
< cont1_bytes
)
2821 for_each_set_bit_from(i
, entry
->bitmap
, BITS_PER_BITMAP
) {
2822 next_zero
= find_next_zero_bit(entry
->bitmap
,
2823 BITS_PER_BITMAP
, i
);
2824 if (next_zero
- i
>= min_bits
) {
2825 found_bits
= next_zero
- i
;
2826 if (found_bits
> max_bits
)
2827 max_bits
= found_bits
;
2830 if (next_zero
- i
> max_bits
)
2831 max_bits
= next_zero
- i
;
2836 entry
->max_extent_size
= (u64
)max_bits
* ctl
->unit
;
2842 cluster
->max_size
= 0;
2845 total_found
+= found_bits
;
2847 if (cluster
->max_size
< found_bits
* ctl
->unit
)
2848 cluster
->max_size
= found_bits
* ctl
->unit
;
2850 if (total_found
< want_bits
|| cluster
->max_size
< cont1_bytes
) {
2855 cluster
->window_start
= start
* ctl
->unit
+ entry
->offset
;
2856 rb_erase(&entry
->offset_index
, &ctl
->free_space_offset
);
2857 ret
= tree_insert_offset(&cluster
->root
, entry
->offset
,
2858 &entry
->offset_index
, 1);
2859 ASSERT(!ret
); /* -EEXIST; Logic error */
2861 trace_btrfs_setup_cluster(block_group
, cluster
,
2862 total_found
* ctl
->unit
, 1);
2867 * This searches the block group for just extents to fill the cluster with.
2868 * Try to find a cluster with at least bytes total bytes, at least one
2869 * extent of cont1_bytes, and other clusters of at least min_bytes.
2872 setup_cluster_no_bitmap(struct btrfs_block_group_cache
*block_group
,
2873 struct btrfs_free_cluster
*cluster
,
2874 struct list_head
*bitmaps
, u64 offset
, u64 bytes
,
2875 u64 cont1_bytes
, u64 min_bytes
)
2877 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2878 struct btrfs_free_space
*first
= NULL
;
2879 struct btrfs_free_space
*entry
= NULL
;
2880 struct btrfs_free_space
*last
;
2881 struct rb_node
*node
;
2886 entry
= tree_search_offset(ctl
, offset
, 0, 1);
2891 * We don't want bitmaps, so just move along until we find a normal
2894 while (entry
->bitmap
|| entry
->bytes
< min_bytes
) {
2895 if (entry
->bitmap
&& list_empty(&entry
->list
))
2896 list_add_tail(&entry
->list
, bitmaps
);
2897 node
= rb_next(&entry
->offset_index
);
2900 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2903 window_free
= entry
->bytes
;
2904 max_extent
= entry
->bytes
;
2908 for (node
= rb_next(&entry
->offset_index
); node
;
2909 node
= rb_next(&entry
->offset_index
)) {
2910 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2912 if (entry
->bitmap
) {
2913 if (list_empty(&entry
->list
))
2914 list_add_tail(&entry
->list
, bitmaps
);
2918 if (entry
->bytes
< min_bytes
)
2922 window_free
+= entry
->bytes
;
2923 if (entry
->bytes
> max_extent
)
2924 max_extent
= entry
->bytes
;
2927 if (window_free
< bytes
|| max_extent
< cont1_bytes
)
2930 cluster
->window_start
= first
->offset
;
2932 node
= &first
->offset_index
;
2935 * now we've found our entries, pull them out of the free space
2936 * cache and put them into the cluster rbtree
2941 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2942 node
= rb_next(&entry
->offset_index
);
2943 if (entry
->bitmap
|| entry
->bytes
< min_bytes
)
2946 rb_erase(&entry
->offset_index
, &ctl
->free_space_offset
);
2947 ret
= tree_insert_offset(&cluster
->root
, entry
->offset
,
2948 &entry
->offset_index
, 0);
2949 total_size
+= entry
->bytes
;
2950 ASSERT(!ret
); /* -EEXIST; Logic error */
2951 } while (node
&& entry
!= last
);
2953 cluster
->max_size
= max_extent
;
2954 trace_btrfs_setup_cluster(block_group
, cluster
, total_size
, 0);
2959 * This specifically looks for bitmaps that may work in the cluster, we assume
2960 * that we have already failed to find extents that will work.
2963 setup_cluster_bitmap(struct btrfs_block_group_cache
*block_group
,
2964 struct btrfs_free_cluster
*cluster
,
2965 struct list_head
*bitmaps
, u64 offset
, u64 bytes
,
2966 u64 cont1_bytes
, u64 min_bytes
)
2968 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2969 struct btrfs_free_space
*entry
;
2971 u64 bitmap_offset
= offset_to_bitmap(ctl
, offset
);
2973 if (ctl
->total_bitmaps
== 0)
2977 * The bitmap that covers offset won't be in the list unless offset
2978 * is just its start offset.
2980 entry
= list_first_entry(bitmaps
, struct btrfs_free_space
, list
);
2981 if (entry
->offset
!= bitmap_offset
) {
2982 entry
= tree_search_offset(ctl
, bitmap_offset
, 1, 0);
2983 if (entry
&& list_empty(&entry
->list
))
2984 list_add(&entry
->list
, bitmaps
);
2987 list_for_each_entry(entry
, bitmaps
, list
) {
2988 if (entry
->bytes
< bytes
)
2990 ret
= btrfs_bitmap_cluster(block_group
, entry
, cluster
, offset
,
2991 bytes
, cont1_bytes
, min_bytes
);
2997 * The bitmaps list has all the bitmaps that record free space
2998 * starting after offset, so no more search is required.
3004 * here we try to find a cluster of blocks in a block group. The goal
3005 * is to find at least bytes+empty_size.
3006 * We might not find them all in one contiguous area.
3008 * returns zero and sets up cluster if things worked out, otherwise
3009 * it returns -enospc
3011 int btrfs_find_space_cluster(struct btrfs_root
*root
,
3012 struct btrfs_block_group_cache
*block_group
,
3013 struct btrfs_free_cluster
*cluster
,
3014 u64 offset
, u64 bytes
, u64 empty_size
)
3016 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
3017 struct btrfs_free_space
*entry
, *tmp
;
3024 * Choose the minimum extent size we'll require for this
3025 * cluster. For SSD_SPREAD, don't allow any fragmentation.
3026 * For metadata, allow allocates with smaller extents. For
3027 * data, keep it dense.
3029 if (btrfs_test_opt(root
, SSD_SPREAD
)) {
3030 cont1_bytes
= min_bytes
= bytes
+ empty_size
;
3031 } else if (block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
3032 cont1_bytes
= bytes
;
3033 min_bytes
= block_group
->sectorsize
;
3035 cont1_bytes
= max(bytes
, (bytes
+ empty_size
) >> 2);
3036 min_bytes
= block_group
->sectorsize
;
3039 spin_lock(&ctl
->tree_lock
);
3042 * If we know we don't have enough space to make a cluster don't even
3043 * bother doing all the work to try and find one.
3045 if (ctl
->free_space
< bytes
) {
3046 spin_unlock(&ctl
->tree_lock
);
3050 spin_lock(&cluster
->lock
);
3052 /* someone already found a cluster, hooray */
3053 if (cluster
->block_group
) {
3058 trace_btrfs_find_cluster(block_group
, offset
, bytes
, empty_size
,
3061 ret
= setup_cluster_no_bitmap(block_group
, cluster
, &bitmaps
, offset
,
3063 cont1_bytes
, min_bytes
);
3065 ret
= setup_cluster_bitmap(block_group
, cluster
, &bitmaps
,
3066 offset
, bytes
+ empty_size
,
3067 cont1_bytes
, min_bytes
);
3069 /* Clear our temporary list */
3070 list_for_each_entry_safe(entry
, tmp
, &bitmaps
, list
)
3071 list_del_init(&entry
->list
);
3074 atomic_inc(&block_group
->count
);
3075 list_add_tail(&cluster
->block_group_list
,
3076 &block_group
->cluster_list
);
3077 cluster
->block_group
= block_group
;
3079 trace_btrfs_failed_cluster_setup(block_group
);
3082 spin_unlock(&cluster
->lock
);
3083 spin_unlock(&ctl
->tree_lock
);
3089 * simple code to zero out a cluster
3091 void btrfs_init_free_cluster(struct btrfs_free_cluster
*cluster
)
3093 spin_lock_init(&cluster
->lock
);
3094 spin_lock_init(&cluster
->refill_lock
);
3095 cluster
->root
= RB_ROOT
;
3096 cluster
->max_size
= 0;
3097 cluster
->fragmented
= false;
3098 INIT_LIST_HEAD(&cluster
->block_group_list
);
3099 cluster
->block_group
= NULL
;
3102 static int do_trimming(struct btrfs_block_group_cache
*block_group
,
3103 u64
*total_trimmed
, u64 start
, u64 bytes
,
3104 u64 reserved_start
, u64 reserved_bytes
,
3105 struct btrfs_trim_range
*trim_entry
)
3107 struct btrfs_space_info
*space_info
= block_group
->space_info
;
3108 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3109 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
3114 spin_lock(&space_info
->lock
);
3115 spin_lock(&block_group
->lock
);
3116 if (!block_group
->ro
) {
3117 block_group
->reserved
+= reserved_bytes
;
3118 space_info
->bytes_reserved
+= reserved_bytes
;
3121 spin_unlock(&block_group
->lock
);
3122 spin_unlock(&space_info
->lock
);
3124 ret
= btrfs_discard_extent(fs_info
->extent_root
,
3125 start
, bytes
, &trimmed
);
3127 *total_trimmed
+= trimmed
;
3129 mutex_lock(&ctl
->cache_writeout_mutex
);
3130 btrfs_add_free_space(block_group
, reserved_start
, reserved_bytes
);
3131 list_del(&trim_entry
->list
);
3132 mutex_unlock(&ctl
->cache_writeout_mutex
);
3135 spin_lock(&space_info
->lock
);
3136 spin_lock(&block_group
->lock
);
3137 if (block_group
->ro
)
3138 space_info
->bytes_readonly
+= reserved_bytes
;
3139 block_group
->reserved
-= reserved_bytes
;
3140 space_info
->bytes_reserved
-= reserved_bytes
;
3141 spin_unlock(&space_info
->lock
);
3142 spin_unlock(&block_group
->lock
);
3148 static int trim_no_bitmap(struct btrfs_block_group_cache
*block_group
,
3149 u64
*total_trimmed
, u64 start
, u64 end
, u64 minlen
)
3151 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
3152 struct btrfs_free_space
*entry
;
3153 struct rb_node
*node
;
3159 while (start
< end
) {
3160 struct btrfs_trim_range trim_entry
;
3162 mutex_lock(&ctl
->cache_writeout_mutex
);
3163 spin_lock(&ctl
->tree_lock
);
3165 if (ctl
->free_space
< minlen
) {
3166 spin_unlock(&ctl
->tree_lock
);
3167 mutex_unlock(&ctl
->cache_writeout_mutex
);
3171 entry
= tree_search_offset(ctl
, start
, 0, 1);
3173 spin_unlock(&ctl
->tree_lock
);
3174 mutex_unlock(&ctl
->cache_writeout_mutex
);
3179 while (entry
->bitmap
) {
3180 node
= rb_next(&entry
->offset_index
);
3182 spin_unlock(&ctl
->tree_lock
);
3183 mutex_unlock(&ctl
->cache_writeout_mutex
);
3186 entry
= rb_entry(node
, struct btrfs_free_space
,
3190 if (entry
->offset
>= end
) {
3191 spin_unlock(&ctl
->tree_lock
);
3192 mutex_unlock(&ctl
->cache_writeout_mutex
);
3196 extent_start
= entry
->offset
;
3197 extent_bytes
= entry
->bytes
;
3198 start
= max(start
, extent_start
);
3199 bytes
= min(extent_start
+ extent_bytes
, end
) - start
;
3200 if (bytes
< minlen
) {
3201 spin_unlock(&ctl
->tree_lock
);
3202 mutex_unlock(&ctl
->cache_writeout_mutex
);
3206 unlink_free_space(ctl
, entry
);
3207 kmem_cache_free(btrfs_free_space_cachep
, entry
);
3209 spin_unlock(&ctl
->tree_lock
);
3210 trim_entry
.start
= extent_start
;
3211 trim_entry
.bytes
= extent_bytes
;
3212 list_add_tail(&trim_entry
.list
, &ctl
->trimming_ranges
);
3213 mutex_unlock(&ctl
->cache_writeout_mutex
);
3215 ret
= do_trimming(block_group
, total_trimmed
, start
, bytes
,
3216 extent_start
, extent_bytes
, &trim_entry
);
3222 if (fatal_signal_pending(current
)) {
3233 static int trim_bitmaps(struct btrfs_block_group_cache
*block_group
,
3234 u64
*total_trimmed
, u64 start
, u64 end
, u64 minlen
)
3236 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
3237 struct btrfs_free_space
*entry
;
3241 u64 offset
= offset_to_bitmap(ctl
, start
);
3243 while (offset
< end
) {
3244 bool next_bitmap
= false;
3245 struct btrfs_trim_range trim_entry
;
3247 mutex_lock(&ctl
->cache_writeout_mutex
);
3248 spin_lock(&ctl
->tree_lock
);
3250 if (ctl
->free_space
< minlen
) {
3251 spin_unlock(&ctl
->tree_lock
);
3252 mutex_unlock(&ctl
->cache_writeout_mutex
);
3256 entry
= tree_search_offset(ctl
, offset
, 1, 0);
3258 spin_unlock(&ctl
->tree_lock
);
3259 mutex_unlock(&ctl
->cache_writeout_mutex
);
3265 ret2
= search_bitmap(ctl
, entry
, &start
, &bytes
);
3266 if (ret2
|| start
>= end
) {
3267 spin_unlock(&ctl
->tree_lock
);
3268 mutex_unlock(&ctl
->cache_writeout_mutex
);
3273 bytes
= min(bytes
, end
- start
);
3274 if (bytes
< minlen
) {
3275 spin_unlock(&ctl
->tree_lock
);
3276 mutex_unlock(&ctl
->cache_writeout_mutex
);
3280 bitmap_clear_bits(ctl
, entry
, start
, bytes
);
3281 if (entry
->bytes
== 0)
3282 free_bitmap(ctl
, entry
);
3284 spin_unlock(&ctl
->tree_lock
);
3285 trim_entry
.start
= start
;
3286 trim_entry
.bytes
= bytes
;
3287 list_add_tail(&trim_entry
.list
, &ctl
->trimming_ranges
);
3288 mutex_unlock(&ctl
->cache_writeout_mutex
);
3290 ret
= do_trimming(block_group
, total_trimmed
, start
, bytes
,
3291 start
, bytes
, &trim_entry
);
3296 offset
+= BITS_PER_BITMAP
* ctl
->unit
;
3299 if (start
>= offset
+ BITS_PER_BITMAP
* ctl
->unit
)
3300 offset
+= BITS_PER_BITMAP
* ctl
->unit
;
3303 if (fatal_signal_pending(current
)) {
3314 void btrfs_get_block_group_trimming(struct btrfs_block_group_cache
*cache
)
3316 atomic_inc(&cache
->trimming
);
3319 void btrfs_put_block_group_trimming(struct btrfs_block_group_cache
*block_group
)
3321 struct extent_map_tree
*em_tree
;
3322 struct extent_map
*em
;
3325 spin_lock(&block_group
->lock
);
3326 cleanup
= (atomic_dec_and_test(&block_group
->trimming
) &&
3327 block_group
->removed
);
3328 spin_unlock(&block_group
->lock
);
3331 lock_chunks(block_group
->fs_info
->chunk_root
);
3332 em_tree
= &block_group
->fs_info
->mapping_tree
.map_tree
;
3333 write_lock(&em_tree
->lock
);
3334 em
= lookup_extent_mapping(em_tree
, block_group
->key
.objectid
,
3336 BUG_ON(!em
); /* logic error, can't happen */
3338 * remove_extent_mapping() will delete us from the pinned_chunks
3339 * list, which is protected by the chunk mutex.
3341 remove_extent_mapping(em_tree
, em
);
3342 write_unlock(&em_tree
->lock
);
3343 unlock_chunks(block_group
->fs_info
->chunk_root
);
3345 /* once for us and once for the tree */
3346 free_extent_map(em
);
3347 free_extent_map(em
);
3350 * We've left one free space entry and other tasks trimming
3351 * this block group have left 1 entry each one. Free them.
3353 __btrfs_remove_free_space_cache(block_group
->free_space_ctl
);
3357 int btrfs_trim_block_group(struct btrfs_block_group_cache
*block_group
,
3358 u64
*trimmed
, u64 start
, u64 end
, u64 minlen
)
3364 spin_lock(&block_group
->lock
);
3365 if (block_group
->removed
) {
3366 spin_unlock(&block_group
->lock
);
3369 btrfs_get_block_group_trimming(block_group
);
3370 spin_unlock(&block_group
->lock
);
3372 ret
= trim_no_bitmap(block_group
, trimmed
, start
, end
, minlen
);
3376 ret
= trim_bitmaps(block_group
, trimmed
, start
, end
, minlen
);
3378 btrfs_put_block_group_trimming(block_group
);
3383 * Find the left-most item in the cache tree, and then return the
3384 * smallest inode number in the item.
3386 * Note: the returned inode number may not be the smallest one in
3387 * the tree, if the left-most item is a bitmap.
3389 u64
btrfs_find_ino_for_alloc(struct btrfs_root
*fs_root
)
3391 struct btrfs_free_space_ctl
*ctl
= fs_root
->free_ino_ctl
;
3392 struct btrfs_free_space
*entry
= NULL
;
3395 spin_lock(&ctl
->tree_lock
);
3397 if (RB_EMPTY_ROOT(&ctl
->free_space_offset
))
3400 entry
= rb_entry(rb_first(&ctl
->free_space_offset
),
3401 struct btrfs_free_space
, offset_index
);
3403 if (!entry
->bitmap
) {
3404 ino
= entry
->offset
;
3406 unlink_free_space(ctl
, entry
);
3410 kmem_cache_free(btrfs_free_space_cachep
, entry
);
3412 link_free_space(ctl
, entry
);
3418 ret
= search_bitmap(ctl
, entry
, &offset
, &count
);
3419 /* Logic error; Should be empty if it can't find anything */
3423 bitmap_clear_bits(ctl
, entry
, offset
, 1);
3424 if (entry
->bytes
== 0)
3425 free_bitmap(ctl
, entry
);
3428 spin_unlock(&ctl
->tree_lock
);
3433 struct inode
*lookup_free_ino_inode(struct btrfs_root
*root
,
3434 struct btrfs_path
*path
)
3436 struct inode
*inode
= NULL
;
3438 spin_lock(&root
->ino_cache_lock
);
3439 if (root
->ino_cache_inode
)
3440 inode
= igrab(root
->ino_cache_inode
);
3441 spin_unlock(&root
->ino_cache_lock
);
3445 inode
= __lookup_free_space_inode(root
, path
, 0);
3449 spin_lock(&root
->ino_cache_lock
);
3450 if (!btrfs_fs_closing(root
->fs_info
))
3451 root
->ino_cache_inode
= igrab(inode
);
3452 spin_unlock(&root
->ino_cache_lock
);
3457 int create_free_ino_inode(struct btrfs_root
*root
,
3458 struct btrfs_trans_handle
*trans
,
3459 struct btrfs_path
*path
)
3461 return __create_free_space_inode(root
, trans
, path
,
3462 BTRFS_FREE_INO_OBJECTID
, 0);
3465 int load_free_ino_cache(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3467 struct btrfs_free_space_ctl
*ctl
= root
->free_ino_ctl
;
3468 struct btrfs_path
*path
;
3469 struct inode
*inode
;
3471 u64 root_gen
= btrfs_root_generation(&root
->root_item
);
3473 if (!btrfs_test_opt(root
, INODE_MAP_CACHE
))
3477 * If we're unmounting then just return, since this does a search on the
3478 * normal root and not the commit root and we could deadlock.
3480 if (btrfs_fs_closing(fs_info
))
3483 path
= btrfs_alloc_path();
3487 inode
= lookup_free_ino_inode(root
, path
);
3491 if (root_gen
!= BTRFS_I(inode
)->generation
)
3494 ret
= __load_free_space_cache(root
, inode
, ctl
, path
, 0);
3498 "failed to load free ino cache for root %llu",
3499 root
->root_key
.objectid
);
3503 btrfs_free_path(path
);
3507 int btrfs_write_out_ino_cache(struct btrfs_root
*root
,
3508 struct btrfs_trans_handle
*trans
,
3509 struct btrfs_path
*path
,
3510 struct inode
*inode
)
3512 struct btrfs_free_space_ctl
*ctl
= root
->free_ino_ctl
;
3514 struct btrfs_io_ctl io_ctl
;
3515 bool release_metadata
= true;
3517 if (!btrfs_test_opt(root
, INODE_MAP_CACHE
))
3520 memset(&io_ctl
, 0, sizeof(io_ctl
));
3521 ret
= __btrfs_write_out_cache(root
, inode
, ctl
, NULL
, &io_ctl
,
3525 * At this point writepages() didn't error out, so our metadata
3526 * reservation is released when the writeback finishes, at
3527 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3528 * with or without an error.
3530 release_metadata
= false;
3531 ret
= btrfs_wait_cache_io(root
, trans
, NULL
, &io_ctl
, path
, 0);
3535 if (release_metadata
)
3536 btrfs_delalloc_release_metadata(inode
, inode
->i_size
);
3538 btrfs_err(root
->fs_info
,
3539 "failed to write free ino cache for root %llu",
3540 root
->root_key
.objectid
);
3547 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3549 * Use this if you need to make a bitmap or extent entry specifically, it
3550 * doesn't do any of the merging that add_free_space does, this acts a lot like
3551 * how the free space cache loading stuff works, so you can get really weird
3554 int test_add_free_space_entry(struct btrfs_block_group_cache
*cache
,
3555 u64 offset
, u64 bytes
, bool bitmap
)
3557 struct btrfs_free_space_ctl
*ctl
= cache
->free_space_ctl
;
3558 struct btrfs_free_space
*info
= NULL
, *bitmap_info
;
3565 info
= kmem_cache_zalloc(btrfs_free_space_cachep
, GFP_NOFS
);
3571 spin_lock(&ctl
->tree_lock
);
3572 info
->offset
= offset
;
3573 info
->bytes
= bytes
;
3574 info
->max_extent_size
= 0;
3575 ret
= link_free_space(ctl
, info
);
3576 spin_unlock(&ctl
->tree_lock
);
3578 kmem_cache_free(btrfs_free_space_cachep
, info
);
3583 map
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
3585 kmem_cache_free(btrfs_free_space_cachep
, info
);
3590 spin_lock(&ctl
->tree_lock
);
3591 bitmap_info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
3596 add_new_bitmap(ctl
, info
, offset
);
3601 bytes_added
= add_bytes_to_bitmap(ctl
, bitmap_info
, offset
, bytes
);
3603 /* We used the newly allocated info, set the max_extent_size to bytes */
3605 bitmap_info
->max_extent_size
= bytes_added
;
3607 bytes
-= bytes_added
;
3608 offset
+= bytes_added
;
3609 spin_unlock(&ctl
->tree_lock
);
3615 kmem_cache_free(btrfs_free_space_cachep
, info
);
3622 * Checks to see if the given range is in the free space cache. This is really
3623 * just used to check the absence of space, so if there is free space in the
3624 * range at all we will return 1.
3626 int test_check_exists(struct btrfs_block_group_cache
*cache
,
3627 u64 offset
, u64 bytes
)
3629 struct btrfs_free_space_ctl
*ctl
= cache
->free_space_ctl
;
3630 struct btrfs_free_space
*info
;
3633 spin_lock(&ctl
->tree_lock
);
3634 info
= tree_search_offset(ctl
, offset
, 0, 0);
3636 info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
3644 u64 bit_off
, bit_bytes
;
3646 struct btrfs_free_space
*tmp
;
3649 bit_bytes
= ctl
->unit
;
3650 ret
= search_bitmap(ctl
, info
, &bit_off
, &bit_bytes
);
3652 if (bit_off
== offset
) {
3655 } else if (bit_off
> offset
&&
3656 offset
+ bytes
> bit_off
) {
3662 n
= rb_prev(&info
->offset_index
);
3664 tmp
= rb_entry(n
, struct btrfs_free_space
,
3666 if (tmp
->offset
+ tmp
->bytes
< offset
)
3668 if (offset
+ bytes
< tmp
->offset
) {
3669 n
= rb_prev(&info
->offset_index
);
3676 n
= rb_next(&info
->offset_index
);
3678 tmp
= rb_entry(n
, struct btrfs_free_space
,
3680 if (offset
+ bytes
< tmp
->offset
)
3682 if (tmp
->offset
+ tmp
->bytes
< offset
) {
3683 n
= rb_next(&info
->offset_index
);
3694 if (info
->offset
== offset
) {
3699 if (offset
> info
->offset
&& offset
< info
->offset
+ info
->bytes
)
3702 spin_unlock(&ctl
->tree_lock
);
3705 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */