2 * Copyright (C) 2008 Red Hat. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
23 #include <linux/ratelimit.h>
25 #include "free-space-cache.h"
26 #include "transaction.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
31 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
34 static int link_free_space(struct btrfs_free_space_ctl
*ctl
,
35 struct btrfs_free_space
*info
);
36 static void unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
37 struct btrfs_free_space
*info
);
39 static struct inode
*__lookup_free_space_inode(struct btrfs_root
*root
,
40 struct btrfs_path
*path
,
44 struct btrfs_key location
;
45 struct btrfs_disk_key disk_key
;
46 struct btrfs_free_space_header
*header
;
47 struct extent_buffer
*leaf
;
48 struct inode
*inode
= NULL
;
51 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
55 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
59 btrfs_release_path(path
);
60 return ERR_PTR(-ENOENT
);
63 leaf
= path
->nodes
[0];
64 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
65 struct btrfs_free_space_header
);
66 btrfs_free_space_key(leaf
, header
, &disk_key
);
67 btrfs_disk_key_to_cpu(&location
, &disk_key
);
68 btrfs_release_path(path
);
70 inode
= btrfs_iget(root
->fs_info
->sb
, &location
, root
, NULL
);
72 return ERR_PTR(-ENOENT
);
75 if (is_bad_inode(inode
)) {
77 return ERR_PTR(-ENOENT
);
80 mapping_set_gfp_mask(inode
->i_mapping
,
81 mapping_gfp_mask(inode
->i_mapping
) & ~__GFP_FS
);
86 struct inode
*lookup_free_space_inode(struct btrfs_root
*root
,
87 struct btrfs_block_group_cache
88 *block_group
, struct btrfs_path
*path
)
90 struct inode
*inode
= NULL
;
91 u32 flags
= BTRFS_INODE_NODATASUM
| BTRFS_INODE_NODATACOW
;
93 spin_lock(&block_group
->lock
);
94 if (block_group
->inode
)
95 inode
= igrab(block_group
->inode
);
96 spin_unlock(&block_group
->lock
);
100 inode
= __lookup_free_space_inode(root
, path
,
101 block_group
->key
.objectid
);
105 spin_lock(&block_group
->lock
);
106 if (!((BTRFS_I(inode
)->flags
& flags
) == flags
)) {
107 printk(KERN_INFO
"Old style space inode found, converting.\n");
108 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
|
109 BTRFS_INODE_NODATACOW
;
110 block_group
->disk_cache_state
= BTRFS_DC_CLEAR
;
113 if (!block_group
->iref
) {
114 block_group
->inode
= igrab(inode
);
115 block_group
->iref
= 1;
117 spin_unlock(&block_group
->lock
);
122 int __create_free_space_inode(struct btrfs_root
*root
,
123 struct btrfs_trans_handle
*trans
,
124 struct btrfs_path
*path
, u64 ino
, u64 offset
)
126 struct btrfs_key key
;
127 struct btrfs_disk_key disk_key
;
128 struct btrfs_free_space_header
*header
;
129 struct btrfs_inode_item
*inode_item
;
130 struct extent_buffer
*leaf
;
131 u64 flags
= BTRFS_INODE_NOCOMPRESS
| BTRFS_INODE_PREALLOC
;
134 ret
= btrfs_insert_empty_inode(trans
, root
, path
, ino
);
138 /* We inline crc's for the free disk space cache */
139 if (ino
!= BTRFS_FREE_INO_OBJECTID
)
140 flags
|= BTRFS_INODE_NODATASUM
| BTRFS_INODE_NODATACOW
;
142 leaf
= path
->nodes
[0];
143 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
144 struct btrfs_inode_item
);
145 btrfs_item_key(leaf
, &disk_key
, path
->slots
[0]);
146 memset_extent_buffer(leaf
, 0, (unsigned long)inode_item
,
147 sizeof(*inode_item
));
148 btrfs_set_inode_generation(leaf
, inode_item
, trans
->transid
);
149 btrfs_set_inode_size(leaf
, inode_item
, 0);
150 btrfs_set_inode_nbytes(leaf
, inode_item
, 0);
151 btrfs_set_inode_uid(leaf
, inode_item
, 0);
152 btrfs_set_inode_gid(leaf
, inode_item
, 0);
153 btrfs_set_inode_mode(leaf
, inode_item
, S_IFREG
| 0600);
154 btrfs_set_inode_flags(leaf
, inode_item
, flags
);
155 btrfs_set_inode_nlink(leaf
, inode_item
, 1);
156 btrfs_set_inode_transid(leaf
, inode_item
, trans
->transid
);
157 btrfs_set_inode_block_group(leaf
, inode_item
, offset
);
158 btrfs_mark_buffer_dirty(leaf
);
159 btrfs_release_path(path
);
161 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
165 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
166 sizeof(struct btrfs_free_space_header
));
168 btrfs_release_path(path
);
171 leaf
= path
->nodes
[0];
172 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
173 struct btrfs_free_space_header
);
174 memset_extent_buffer(leaf
, 0, (unsigned long)header
, sizeof(*header
));
175 btrfs_set_free_space_key(leaf
, header
, &disk_key
);
176 btrfs_mark_buffer_dirty(leaf
);
177 btrfs_release_path(path
);
182 int create_free_space_inode(struct btrfs_root
*root
,
183 struct btrfs_trans_handle
*trans
,
184 struct btrfs_block_group_cache
*block_group
,
185 struct btrfs_path
*path
)
190 ret
= btrfs_find_free_objectid(root
, &ino
);
194 return __create_free_space_inode(root
, trans
, path
, ino
,
195 block_group
->key
.objectid
);
198 int btrfs_truncate_free_space_cache(struct btrfs_root
*root
,
199 struct btrfs_trans_handle
*trans
,
200 struct btrfs_path
*path
,
203 struct btrfs_block_rsv
*rsv
;
208 rsv
= trans
->block_rsv
;
209 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
211 /* 1 for slack space, 1 for updating the inode */
212 needed_bytes
= btrfs_calc_trunc_metadata_size(root
, 1) +
213 btrfs_calc_trans_metadata_size(root
, 1);
215 spin_lock(&trans
->block_rsv
->lock
);
216 if (trans
->block_rsv
->reserved
< needed_bytes
) {
217 spin_unlock(&trans
->block_rsv
->lock
);
218 trans
->block_rsv
= rsv
;
221 spin_unlock(&trans
->block_rsv
->lock
);
223 oldsize
= i_size_read(inode
);
224 btrfs_i_size_write(inode
, 0);
225 truncate_pagecache(inode
, oldsize
, 0);
228 * We don't need an orphan item because truncating the free space cache
229 * will never be split across transactions.
231 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
232 0, BTRFS_EXTENT_DATA_KEY
);
235 trans
->block_rsv
= rsv
;
236 btrfs_abort_transaction(trans
, root
, ret
);
240 ret
= btrfs_update_inode(trans
, root
, inode
);
242 btrfs_abort_transaction(trans
, root
, ret
);
243 trans
->block_rsv
= rsv
;
248 static int readahead_cache(struct inode
*inode
)
250 struct file_ra_state
*ra
;
251 unsigned long last_index
;
253 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
257 file_ra_state_init(ra
, inode
->i_mapping
);
258 last_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
260 page_cache_sync_readahead(inode
->i_mapping
, ra
, NULL
, 0, last_index
);
271 struct btrfs_root
*root
;
275 unsigned check_crcs
:1;
278 static int io_ctl_init(struct io_ctl
*io_ctl
, struct inode
*inode
,
279 struct btrfs_root
*root
)
281 memset(io_ctl
, 0, sizeof(struct io_ctl
));
282 io_ctl
->num_pages
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >>
284 io_ctl
->pages
= kzalloc(sizeof(struct page
*) * io_ctl
->num_pages
,
289 if (btrfs_ino(inode
) != BTRFS_FREE_INO_OBJECTID
)
290 io_ctl
->check_crcs
= 1;
294 static void io_ctl_free(struct io_ctl
*io_ctl
)
296 kfree(io_ctl
->pages
);
299 static void io_ctl_unmap_page(struct io_ctl
*io_ctl
)
302 kunmap(io_ctl
->page
);
308 static void io_ctl_map_page(struct io_ctl
*io_ctl
, int clear
)
310 WARN_ON(io_ctl
->cur
);
311 BUG_ON(io_ctl
->index
>= io_ctl
->num_pages
);
312 io_ctl
->page
= io_ctl
->pages
[io_ctl
->index
++];
313 io_ctl
->cur
= kmap(io_ctl
->page
);
314 io_ctl
->orig
= io_ctl
->cur
;
315 io_ctl
->size
= PAGE_CACHE_SIZE
;
317 memset(io_ctl
->cur
, 0, PAGE_CACHE_SIZE
);
320 static void io_ctl_drop_pages(struct io_ctl
*io_ctl
)
324 io_ctl_unmap_page(io_ctl
);
326 for (i
= 0; i
< io_ctl
->num_pages
; i
++) {
327 if (io_ctl
->pages
[i
]) {
328 ClearPageChecked(io_ctl
->pages
[i
]);
329 unlock_page(io_ctl
->pages
[i
]);
330 page_cache_release(io_ctl
->pages
[i
]);
335 static int io_ctl_prepare_pages(struct io_ctl
*io_ctl
, struct inode
*inode
,
339 gfp_t mask
= btrfs_alloc_write_mask(inode
->i_mapping
);
342 for (i
= 0; i
< io_ctl
->num_pages
; i
++) {
343 page
= find_or_create_page(inode
->i_mapping
, i
, mask
);
345 io_ctl_drop_pages(io_ctl
);
348 io_ctl
->pages
[i
] = page
;
349 if (uptodate
&& !PageUptodate(page
)) {
350 btrfs_readpage(NULL
, page
);
352 if (!PageUptodate(page
)) {
353 printk(KERN_ERR
"btrfs: error reading free "
355 io_ctl_drop_pages(io_ctl
);
361 for (i
= 0; i
< io_ctl
->num_pages
; i
++) {
362 clear_page_dirty_for_io(io_ctl
->pages
[i
]);
363 set_page_extent_mapped(io_ctl
->pages
[i
]);
369 static void io_ctl_set_generation(struct io_ctl
*io_ctl
, u64 generation
)
373 io_ctl_map_page(io_ctl
, 1);
376 * Skip the csum areas. If we don't check crcs then we just have a
377 * 64bit chunk at the front of the first page.
379 if (io_ctl
->check_crcs
) {
380 io_ctl
->cur
+= (sizeof(u32
) * io_ctl
->num_pages
);
381 io_ctl
->size
-= sizeof(u64
) + (sizeof(u32
) * io_ctl
->num_pages
);
383 io_ctl
->cur
+= sizeof(u64
);
384 io_ctl
->size
-= sizeof(u64
) * 2;
388 *val
= cpu_to_le64(generation
);
389 io_ctl
->cur
+= sizeof(u64
);
392 static int io_ctl_check_generation(struct io_ctl
*io_ctl
, u64 generation
)
397 * Skip the crc area. If we don't check crcs then we just have a 64bit
398 * chunk at the front of the first page.
400 if (io_ctl
->check_crcs
) {
401 io_ctl
->cur
+= sizeof(u32
) * io_ctl
->num_pages
;
402 io_ctl
->size
-= sizeof(u64
) +
403 (sizeof(u32
) * io_ctl
->num_pages
);
405 io_ctl
->cur
+= sizeof(u64
);
406 io_ctl
->size
-= sizeof(u64
) * 2;
410 if (le64_to_cpu(*gen
) != generation
) {
411 printk_ratelimited(KERN_ERR
"btrfs: space cache generation "
412 "(%Lu) does not match inode (%Lu)\n", *gen
,
414 io_ctl_unmap_page(io_ctl
);
417 io_ctl
->cur
+= sizeof(u64
);
421 static void io_ctl_set_crc(struct io_ctl
*io_ctl
, int index
)
427 if (!io_ctl
->check_crcs
) {
428 io_ctl_unmap_page(io_ctl
);
433 offset
= sizeof(u32
) * io_ctl
->num_pages
;
435 crc
= btrfs_csum_data(io_ctl
->root
, io_ctl
->orig
+ offset
, crc
,
436 PAGE_CACHE_SIZE
- offset
);
437 btrfs_csum_final(crc
, (char *)&crc
);
438 io_ctl_unmap_page(io_ctl
);
439 tmp
= kmap(io_ctl
->pages
[0]);
442 kunmap(io_ctl
->pages
[0]);
445 static int io_ctl_check_crc(struct io_ctl
*io_ctl
, int index
)
451 if (!io_ctl
->check_crcs
) {
452 io_ctl_map_page(io_ctl
, 0);
457 offset
= sizeof(u32
) * io_ctl
->num_pages
;
459 tmp
= kmap(io_ctl
->pages
[0]);
462 kunmap(io_ctl
->pages
[0]);
464 io_ctl_map_page(io_ctl
, 0);
465 crc
= btrfs_csum_data(io_ctl
->root
, io_ctl
->orig
+ offset
, crc
,
466 PAGE_CACHE_SIZE
- offset
);
467 btrfs_csum_final(crc
, (char *)&crc
);
469 printk_ratelimited(KERN_ERR
"btrfs: csum mismatch on free "
471 io_ctl_unmap_page(io_ctl
);
478 static int io_ctl_add_entry(struct io_ctl
*io_ctl
, u64 offset
, u64 bytes
,
481 struct btrfs_free_space_entry
*entry
;
487 entry
->offset
= cpu_to_le64(offset
);
488 entry
->bytes
= cpu_to_le64(bytes
);
489 entry
->type
= (bitmap
) ? BTRFS_FREE_SPACE_BITMAP
:
490 BTRFS_FREE_SPACE_EXTENT
;
491 io_ctl
->cur
+= sizeof(struct btrfs_free_space_entry
);
492 io_ctl
->size
-= sizeof(struct btrfs_free_space_entry
);
494 if (io_ctl
->size
>= sizeof(struct btrfs_free_space_entry
))
497 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
499 /* No more pages to map */
500 if (io_ctl
->index
>= io_ctl
->num_pages
)
503 /* map the next page */
504 io_ctl_map_page(io_ctl
, 1);
508 static int io_ctl_add_bitmap(struct io_ctl
*io_ctl
, void *bitmap
)
514 * If we aren't at the start of the current page, unmap this one and
515 * map the next one if there is any left.
517 if (io_ctl
->cur
!= io_ctl
->orig
) {
518 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
519 if (io_ctl
->index
>= io_ctl
->num_pages
)
521 io_ctl_map_page(io_ctl
, 0);
524 memcpy(io_ctl
->cur
, bitmap
, PAGE_CACHE_SIZE
);
525 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
526 if (io_ctl
->index
< io_ctl
->num_pages
)
527 io_ctl_map_page(io_ctl
, 0);
531 static void io_ctl_zero_remaining_pages(struct io_ctl
*io_ctl
)
534 * If we're not on the boundary we know we've modified the page and we
535 * need to crc the page.
537 if (io_ctl
->cur
!= io_ctl
->orig
)
538 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
540 io_ctl_unmap_page(io_ctl
);
542 while (io_ctl
->index
< io_ctl
->num_pages
) {
543 io_ctl_map_page(io_ctl
, 1);
544 io_ctl_set_crc(io_ctl
, io_ctl
->index
- 1);
548 static int io_ctl_read_entry(struct io_ctl
*io_ctl
,
549 struct btrfs_free_space
*entry
, u8
*type
)
551 struct btrfs_free_space_entry
*e
;
555 ret
= io_ctl_check_crc(io_ctl
, io_ctl
->index
);
561 entry
->offset
= le64_to_cpu(e
->offset
);
562 entry
->bytes
= le64_to_cpu(e
->bytes
);
564 io_ctl
->cur
+= sizeof(struct btrfs_free_space_entry
);
565 io_ctl
->size
-= sizeof(struct btrfs_free_space_entry
);
567 if (io_ctl
->size
>= sizeof(struct btrfs_free_space_entry
))
570 io_ctl_unmap_page(io_ctl
);
575 static int io_ctl_read_bitmap(struct io_ctl
*io_ctl
,
576 struct btrfs_free_space
*entry
)
580 ret
= io_ctl_check_crc(io_ctl
, io_ctl
->index
);
584 memcpy(entry
->bitmap
, io_ctl
->cur
, PAGE_CACHE_SIZE
);
585 io_ctl_unmap_page(io_ctl
);
591 * Since we attach pinned extents after the fact we can have contiguous sections
592 * of free space that are split up in entries. This poses a problem with the
593 * tree logging stuff since it could have allocated across what appears to be 2
594 * entries since we would have merged the entries when adding the pinned extents
595 * back to the free space cache. So run through the space cache that we just
596 * loaded and merge contiguous entries. This will make the log replay stuff not
597 * blow up and it will make for nicer allocator behavior.
599 static void merge_space_tree(struct btrfs_free_space_ctl
*ctl
)
601 struct btrfs_free_space
*e
, *prev
= NULL
;
605 spin_lock(&ctl
->tree_lock
);
606 for (n
= rb_first(&ctl
->free_space_offset
); n
; n
= rb_next(n
)) {
607 e
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
610 if (e
->bitmap
|| prev
->bitmap
)
612 if (prev
->offset
+ prev
->bytes
== e
->offset
) {
613 unlink_free_space(ctl
, prev
);
614 unlink_free_space(ctl
, e
);
615 prev
->bytes
+= e
->bytes
;
616 kmem_cache_free(btrfs_free_space_cachep
, e
);
617 link_free_space(ctl
, prev
);
619 spin_unlock(&ctl
->tree_lock
);
625 spin_unlock(&ctl
->tree_lock
);
628 int __load_free_space_cache(struct btrfs_root
*root
, struct inode
*inode
,
629 struct btrfs_free_space_ctl
*ctl
,
630 struct btrfs_path
*path
, u64 offset
)
632 struct btrfs_free_space_header
*header
;
633 struct extent_buffer
*leaf
;
634 struct io_ctl io_ctl
;
635 struct btrfs_key key
;
636 struct btrfs_free_space
*e
, *n
;
637 struct list_head bitmaps
;
644 INIT_LIST_HEAD(&bitmaps
);
646 /* Nothing in the space cache, goodbye */
647 if (!i_size_read(inode
))
650 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
654 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
658 btrfs_release_path(path
);
664 leaf
= path
->nodes
[0];
665 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
666 struct btrfs_free_space_header
);
667 num_entries
= btrfs_free_space_entries(leaf
, header
);
668 num_bitmaps
= btrfs_free_space_bitmaps(leaf
, header
);
669 generation
= btrfs_free_space_generation(leaf
, header
);
670 btrfs_release_path(path
);
672 if (BTRFS_I(inode
)->generation
!= generation
) {
673 printk(KERN_ERR
"btrfs: free space inode generation (%llu) did"
674 " not match free space cache generation (%llu)\n",
675 (unsigned long long)BTRFS_I(inode
)->generation
,
676 (unsigned long long)generation
);
683 ret
= io_ctl_init(&io_ctl
, inode
, root
);
687 ret
= readahead_cache(inode
);
691 ret
= io_ctl_prepare_pages(&io_ctl
, inode
, 1);
695 ret
= io_ctl_check_crc(&io_ctl
, 0);
699 ret
= io_ctl_check_generation(&io_ctl
, generation
);
703 while (num_entries
) {
704 e
= kmem_cache_zalloc(btrfs_free_space_cachep
,
709 ret
= io_ctl_read_entry(&io_ctl
, e
, &type
);
711 kmem_cache_free(btrfs_free_space_cachep
, e
);
716 kmem_cache_free(btrfs_free_space_cachep
, e
);
720 if (type
== BTRFS_FREE_SPACE_EXTENT
) {
721 spin_lock(&ctl
->tree_lock
);
722 ret
= link_free_space(ctl
, e
);
723 spin_unlock(&ctl
->tree_lock
);
725 printk(KERN_ERR
"Duplicate entries in "
726 "free space cache, dumping\n");
727 kmem_cache_free(btrfs_free_space_cachep
, e
);
731 BUG_ON(!num_bitmaps
);
733 e
->bitmap
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
736 btrfs_free_space_cachep
, e
);
739 spin_lock(&ctl
->tree_lock
);
740 ret
= link_free_space(ctl
, e
);
741 ctl
->total_bitmaps
++;
742 ctl
->op
->recalc_thresholds(ctl
);
743 spin_unlock(&ctl
->tree_lock
);
745 printk(KERN_ERR
"Duplicate entries in "
746 "free space cache, dumping\n");
747 kmem_cache_free(btrfs_free_space_cachep
, e
);
750 list_add_tail(&e
->list
, &bitmaps
);
756 io_ctl_unmap_page(&io_ctl
);
759 * We add the bitmaps at the end of the entries in order that
760 * the bitmap entries are added to the cache.
762 list_for_each_entry_safe(e
, n
, &bitmaps
, list
) {
763 list_del_init(&e
->list
);
764 ret
= io_ctl_read_bitmap(&io_ctl
, e
);
769 io_ctl_drop_pages(&io_ctl
);
770 merge_space_tree(ctl
);
773 io_ctl_free(&io_ctl
);
776 io_ctl_drop_pages(&io_ctl
);
777 __btrfs_remove_free_space_cache(ctl
);
781 int load_free_space_cache(struct btrfs_fs_info
*fs_info
,
782 struct btrfs_block_group_cache
*block_group
)
784 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
785 struct btrfs_root
*root
= fs_info
->tree_root
;
787 struct btrfs_path
*path
;
790 u64 used
= btrfs_block_group_used(&block_group
->item
);
793 * If this block group has been marked to be cleared for one reason or
794 * another then we can't trust the on disk cache, so just return.
796 spin_lock(&block_group
->lock
);
797 if (block_group
->disk_cache_state
!= BTRFS_DC_WRITTEN
) {
798 spin_unlock(&block_group
->lock
);
801 spin_unlock(&block_group
->lock
);
803 path
= btrfs_alloc_path();
806 path
->search_commit_root
= 1;
807 path
->skip_locking
= 1;
809 inode
= lookup_free_space_inode(root
, block_group
, path
);
811 btrfs_free_path(path
);
815 /* We may have converted the inode and made the cache invalid. */
816 spin_lock(&block_group
->lock
);
817 if (block_group
->disk_cache_state
!= BTRFS_DC_WRITTEN
) {
818 spin_unlock(&block_group
->lock
);
819 btrfs_free_path(path
);
822 spin_unlock(&block_group
->lock
);
824 ret
= __load_free_space_cache(fs_info
->tree_root
, inode
, ctl
,
825 path
, block_group
->key
.objectid
);
826 btrfs_free_path(path
);
830 spin_lock(&ctl
->tree_lock
);
831 matched
= (ctl
->free_space
== (block_group
->key
.offset
- used
-
832 block_group
->bytes_super
));
833 spin_unlock(&ctl
->tree_lock
);
836 __btrfs_remove_free_space_cache(ctl
);
837 printk(KERN_ERR
"block group %llu has an wrong amount of free "
838 "space\n", block_group
->key
.objectid
);
843 /* This cache is bogus, make sure it gets cleared */
844 spin_lock(&block_group
->lock
);
845 block_group
->disk_cache_state
= BTRFS_DC_CLEAR
;
846 spin_unlock(&block_group
->lock
);
849 printk(KERN_ERR
"btrfs: failed to load free space cache "
850 "for block group %llu\n", block_group
->key
.objectid
);
858 * __btrfs_write_out_cache - write out cached info to an inode
859 * @root - the root the inode belongs to
860 * @ctl - the free space cache we are going to write out
861 * @block_group - the block_group for this cache if it belongs to a block_group
862 * @trans - the trans handle
863 * @path - the path to use
864 * @offset - the offset for the key we'll insert
866 * This function writes out a free space cache struct to disk for quick recovery
867 * on mount. This will return 0 if it was successfull in writing the cache out,
868 * and -1 if it was not.
870 int __btrfs_write_out_cache(struct btrfs_root
*root
, struct inode
*inode
,
871 struct btrfs_free_space_ctl
*ctl
,
872 struct btrfs_block_group_cache
*block_group
,
873 struct btrfs_trans_handle
*trans
,
874 struct btrfs_path
*path
, u64 offset
)
876 struct btrfs_free_space_header
*header
;
877 struct extent_buffer
*leaf
;
878 struct rb_node
*node
;
879 struct list_head
*pos
, *n
;
880 struct extent_state
*cached_state
= NULL
;
881 struct btrfs_free_cluster
*cluster
= NULL
;
882 struct extent_io_tree
*unpin
= NULL
;
883 struct io_ctl io_ctl
;
884 struct list_head bitmap_list
;
885 struct btrfs_key key
;
886 u64 start
, extent_start
, extent_end
, len
;
892 INIT_LIST_HEAD(&bitmap_list
);
894 if (!i_size_read(inode
))
897 ret
= io_ctl_init(&io_ctl
, inode
, root
);
901 /* Get the cluster for this block_group if it exists */
902 if (block_group
&& !list_empty(&block_group
->cluster_list
))
903 cluster
= list_entry(block_group
->cluster_list
.next
,
904 struct btrfs_free_cluster
,
907 /* Lock all pages first so we can lock the extent safely. */
908 io_ctl_prepare_pages(&io_ctl
, inode
, 0);
910 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, 0, i_size_read(inode
) - 1,
913 node
= rb_first(&ctl
->free_space_offset
);
914 if (!node
&& cluster
) {
915 node
= rb_first(&cluster
->root
);
919 /* Make sure we can fit our crcs into the first page */
920 if (io_ctl
.check_crcs
&&
921 (io_ctl
.num_pages
* sizeof(u32
)) >= PAGE_CACHE_SIZE
) {
926 io_ctl_set_generation(&io_ctl
, trans
->transid
);
928 /* Write out the extent entries */
930 struct btrfs_free_space
*e
;
932 e
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
935 ret
= io_ctl_add_entry(&io_ctl
, e
->offset
, e
->bytes
,
941 list_add_tail(&e
->list
, &bitmap_list
);
944 node
= rb_next(node
);
945 if (!node
&& cluster
) {
946 node
= rb_first(&cluster
->root
);
952 * We want to add any pinned extents to our free space cache
953 * so we don't leak the space
957 * We shouldn't have switched the pinned extents yet so this is the
960 unpin
= root
->fs_info
->pinned_extents
;
963 start
= block_group
->key
.objectid
;
965 while (block_group
&& (start
< block_group
->key
.objectid
+
966 block_group
->key
.offset
)) {
967 ret
= find_first_extent_bit(unpin
, start
,
968 &extent_start
, &extent_end
,
975 /* This pinned extent is out of our range */
976 if (extent_start
>= block_group
->key
.objectid
+
977 block_group
->key
.offset
)
980 extent_start
= max(extent_start
, start
);
981 extent_end
= min(block_group
->key
.objectid
+
982 block_group
->key
.offset
, extent_end
+ 1);
983 len
= extent_end
- extent_start
;
986 ret
= io_ctl_add_entry(&io_ctl
, extent_start
, len
, NULL
);
993 /* Write out the bitmaps */
994 list_for_each_safe(pos
, n
, &bitmap_list
) {
995 struct btrfs_free_space
*entry
=
996 list_entry(pos
, struct btrfs_free_space
, list
);
998 ret
= io_ctl_add_bitmap(&io_ctl
, entry
->bitmap
);
1001 list_del_init(&entry
->list
);
1004 /* Zero out the rest of the pages just to make sure */
1005 io_ctl_zero_remaining_pages(&io_ctl
);
1007 ret
= btrfs_dirty_pages(root
, inode
, io_ctl
.pages
, io_ctl
.num_pages
,
1008 0, i_size_read(inode
), &cached_state
);
1009 io_ctl_drop_pages(&io_ctl
);
1010 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, 0,
1011 i_size_read(inode
) - 1, &cached_state
, GFP_NOFS
);
1017 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
1019 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
1020 key
.offset
= offset
;
1023 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1025 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0, inode
->i_size
- 1,
1026 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, NULL
,
1030 leaf
= path
->nodes
[0];
1032 struct btrfs_key found_key
;
1033 BUG_ON(!path
->slots
[0]);
1035 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1036 if (found_key
.objectid
!= BTRFS_FREE_SPACE_OBJECTID
||
1037 found_key
.offset
!= offset
) {
1038 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, 0,
1040 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0,
1042 btrfs_release_path(path
);
1047 BTRFS_I(inode
)->generation
= trans
->transid
;
1048 header
= btrfs_item_ptr(leaf
, path
->slots
[0],
1049 struct btrfs_free_space_header
);
1050 btrfs_set_free_space_entries(leaf
, header
, entries
);
1051 btrfs_set_free_space_bitmaps(leaf
, header
, bitmaps
);
1052 btrfs_set_free_space_generation(leaf
, header
, trans
->transid
);
1053 btrfs_mark_buffer_dirty(leaf
);
1054 btrfs_release_path(path
);
1058 io_ctl_free(&io_ctl
);
1060 invalidate_inode_pages2(inode
->i_mapping
);
1061 BTRFS_I(inode
)->generation
= 0;
1063 btrfs_update_inode(trans
, root
, inode
);
1067 list_for_each_safe(pos
, n
, &bitmap_list
) {
1068 struct btrfs_free_space
*entry
=
1069 list_entry(pos
, struct btrfs_free_space
, list
);
1070 list_del_init(&entry
->list
);
1072 io_ctl_drop_pages(&io_ctl
);
1073 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, 0,
1074 i_size_read(inode
) - 1, &cached_state
, GFP_NOFS
);
1078 int btrfs_write_out_cache(struct btrfs_root
*root
,
1079 struct btrfs_trans_handle
*trans
,
1080 struct btrfs_block_group_cache
*block_group
,
1081 struct btrfs_path
*path
)
1083 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1084 struct inode
*inode
;
1087 root
= root
->fs_info
->tree_root
;
1089 spin_lock(&block_group
->lock
);
1090 if (block_group
->disk_cache_state
< BTRFS_DC_SETUP
) {
1091 spin_unlock(&block_group
->lock
);
1094 spin_unlock(&block_group
->lock
);
1096 inode
= lookup_free_space_inode(root
, block_group
, path
);
1100 ret
= __btrfs_write_out_cache(root
, inode
, ctl
, block_group
, trans
,
1101 path
, block_group
->key
.objectid
);
1103 spin_lock(&block_group
->lock
);
1104 block_group
->disk_cache_state
= BTRFS_DC_ERROR
;
1105 spin_unlock(&block_group
->lock
);
1108 printk(KERN_ERR
"btrfs: failed to write free space cache "
1109 "for block group %llu\n", block_group
->key
.objectid
);
1117 static inline unsigned long offset_to_bit(u64 bitmap_start
, u32 unit
,
1120 BUG_ON(offset
< bitmap_start
);
1121 offset
-= bitmap_start
;
1122 return (unsigned long)(div_u64(offset
, unit
));
1125 static inline unsigned long bytes_to_bits(u64 bytes
, u32 unit
)
1127 return (unsigned long)(div_u64(bytes
, unit
));
1130 static inline u64
offset_to_bitmap(struct btrfs_free_space_ctl
*ctl
,
1134 u64 bytes_per_bitmap
;
1136 bytes_per_bitmap
= BITS_PER_BITMAP
* ctl
->unit
;
1137 bitmap_start
= offset
- ctl
->start
;
1138 bitmap_start
= div64_u64(bitmap_start
, bytes_per_bitmap
);
1139 bitmap_start
*= bytes_per_bitmap
;
1140 bitmap_start
+= ctl
->start
;
1142 return bitmap_start
;
1145 static int tree_insert_offset(struct rb_root
*root
, u64 offset
,
1146 struct rb_node
*node
, int bitmap
)
1148 struct rb_node
**p
= &root
->rb_node
;
1149 struct rb_node
*parent
= NULL
;
1150 struct btrfs_free_space
*info
;
1154 info
= rb_entry(parent
, struct btrfs_free_space
, offset_index
);
1156 if (offset
< info
->offset
) {
1158 } else if (offset
> info
->offset
) {
1159 p
= &(*p
)->rb_right
;
1162 * we could have a bitmap entry and an extent entry
1163 * share the same offset. If this is the case, we want
1164 * the extent entry to always be found first if we do a
1165 * linear search through the tree, since we want to have
1166 * the quickest allocation time, and allocating from an
1167 * extent is faster than allocating from a bitmap. So
1168 * if we're inserting a bitmap and we find an entry at
1169 * this offset, we want to go right, or after this entry
1170 * logically. If we are inserting an extent and we've
1171 * found a bitmap, we want to go left, or before
1179 p
= &(*p
)->rb_right
;
1181 if (!info
->bitmap
) {
1190 rb_link_node(node
, parent
, p
);
1191 rb_insert_color(node
, root
);
1197 * searches the tree for the given offset.
1199 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1200 * want a section that has at least bytes size and comes at or after the given
1203 static struct btrfs_free_space
*
1204 tree_search_offset(struct btrfs_free_space_ctl
*ctl
,
1205 u64 offset
, int bitmap_only
, int fuzzy
)
1207 struct rb_node
*n
= ctl
->free_space_offset
.rb_node
;
1208 struct btrfs_free_space
*entry
, *prev
= NULL
;
1210 /* find entry that is closest to the 'offset' */
1217 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1220 if (offset
< entry
->offset
)
1222 else if (offset
> entry
->offset
)
1235 * bitmap entry and extent entry may share same offset,
1236 * in that case, bitmap entry comes after extent entry.
1241 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1242 if (entry
->offset
!= offset
)
1245 WARN_ON(!entry
->bitmap
);
1248 if (entry
->bitmap
) {
1250 * if previous extent entry covers the offset,
1251 * we should return it instead of the bitmap entry
1253 n
= &entry
->offset_index
;
1258 prev
= rb_entry(n
, struct btrfs_free_space
,
1260 if (!prev
->bitmap
) {
1261 if (prev
->offset
+ prev
->bytes
> offset
)
1273 /* find last entry before the 'offset' */
1275 if (entry
->offset
> offset
) {
1276 n
= rb_prev(&entry
->offset_index
);
1278 entry
= rb_entry(n
, struct btrfs_free_space
,
1280 BUG_ON(entry
->offset
> offset
);
1289 if (entry
->bitmap
) {
1290 n
= &entry
->offset_index
;
1295 prev
= rb_entry(n
, struct btrfs_free_space
,
1297 if (!prev
->bitmap
) {
1298 if (prev
->offset
+ prev
->bytes
> offset
)
1303 if (entry
->offset
+ BITS_PER_BITMAP
* ctl
->unit
> offset
)
1305 } else if (entry
->offset
+ entry
->bytes
> offset
)
1312 if (entry
->bitmap
) {
1313 if (entry
->offset
+ BITS_PER_BITMAP
*
1317 if (entry
->offset
+ entry
->bytes
> offset
)
1321 n
= rb_next(&entry
->offset_index
);
1324 entry
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
1330 __unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
1331 struct btrfs_free_space
*info
)
1333 rb_erase(&info
->offset_index
, &ctl
->free_space_offset
);
1334 ctl
->free_extents
--;
1337 static void unlink_free_space(struct btrfs_free_space_ctl
*ctl
,
1338 struct btrfs_free_space
*info
)
1340 __unlink_free_space(ctl
, info
);
1341 ctl
->free_space
-= info
->bytes
;
1344 static int link_free_space(struct btrfs_free_space_ctl
*ctl
,
1345 struct btrfs_free_space
*info
)
1349 BUG_ON(!info
->bitmap
&& !info
->bytes
);
1350 ret
= tree_insert_offset(&ctl
->free_space_offset
, info
->offset
,
1351 &info
->offset_index
, (info
->bitmap
!= NULL
));
1355 ctl
->free_space
+= info
->bytes
;
1356 ctl
->free_extents
++;
1360 static void recalculate_thresholds(struct btrfs_free_space_ctl
*ctl
)
1362 struct btrfs_block_group_cache
*block_group
= ctl
->private;
1366 u64 size
= block_group
->key
.offset
;
1367 u64 bytes_per_bg
= BITS_PER_BITMAP
* block_group
->sectorsize
;
1368 int max_bitmaps
= div64_u64(size
+ bytes_per_bg
- 1, bytes_per_bg
);
1370 BUG_ON(ctl
->total_bitmaps
> max_bitmaps
);
1373 * The goal is to keep the total amount of memory used per 1gb of space
1374 * at or below 32k, so we need to adjust how much memory we allow to be
1375 * used by extent based free space tracking
1377 if (size
< 1024 * 1024 * 1024)
1378 max_bytes
= MAX_CACHE_BYTES_PER_GIG
;
1380 max_bytes
= MAX_CACHE_BYTES_PER_GIG
*
1381 div64_u64(size
, 1024 * 1024 * 1024);
1384 * we want to account for 1 more bitmap than what we have so we can make
1385 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1386 * we add more bitmaps.
1388 bitmap_bytes
= (ctl
->total_bitmaps
+ 1) * PAGE_CACHE_SIZE
;
1390 if (bitmap_bytes
>= max_bytes
) {
1391 ctl
->extents_thresh
= 0;
1396 * we want the extent entry threshold to always be at most 1/2 the maxw
1397 * bytes we can have, or whatever is less than that.
1399 extent_bytes
= max_bytes
- bitmap_bytes
;
1400 extent_bytes
= min_t(u64
, extent_bytes
, div64_u64(max_bytes
, 2));
1402 ctl
->extents_thresh
=
1403 div64_u64(extent_bytes
, (sizeof(struct btrfs_free_space
)));
1406 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl
*ctl
,
1407 struct btrfs_free_space
*info
,
1408 u64 offset
, u64 bytes
)
1410 unsigned long start
, count
;
1412 start
= offset_to_bit(info
->offset
, ctl
->unit
, offset
);
1413 count
= bytes_to_bits(bytes
, ctl
->unit
);
1414 BUG_ON(start
+ count
> BITS_PER_BITMAP
);
1416 bitmap_clear(info
->bitmap
, start
, count
);
1418 info
->bytes
-= bytes
;
1421 static void bitmap_clear_bits(struct btrfs_free_space_ctl
*ctl
,
1422 struct btrfs_free_space
*info
, u64 offset
,
1425 __bitmap_clear_bits(ctl
, info
, offset
, bytes
);
1426 ctl
->free_space
-= bytes
;
1429 static void bitmap_set_bits(struct btrfs_free_space_ctl
*ctl
,
1430 struct btrfs_free_space
*info
, u64 offset
,
1433 unsigned long start
, count
;
1435 start
= offset_to_bit(info
->offset
, ctl
->unit
, offset
);
1436 count
= bytes_to_bits(bytes
, ctl
->unit
);
1437 BUG_ON(start
+ count
> BITS_PER_BITMAP
);
1439 bitmap_set(info
->bitmap
, start
, count
);
1441 info
->bytes
+= bytes
;
1442 ctl
->free_space
+= bytes
;
1445 static int search_bitmap(struct btrfs_free_space_ctl
*ctl
,
1446 struct btrfs_free_space
*bitmap_info
, u64
*offset
,
1449 unsigned long found_bits
= 0;
1450 unsigned long bits
, i
;
1451 unsigned long next_zero
;
1453 i
= offset_to_bit(bitmap_info
->offset
, ctl
->unit
,
1454 max_t(u64
, *offset
, bitmap_info
->offset
));
1455 bits
= bytes_to_bits(*bytes
, ctl
->unit
);
1457 for (i
= find_next_bit(bitmap_info
->bitmap
, BITS_PER_BITMAP
, i
);
1458 i
< BITS_PER_BITMAP
;
1459 i
= find_next_bit(bitmap_info
->bitmap
, BITS_PER_BITMAP
, i
+ 1)) {
1460 next_zero
= find_next_zero_bit(bitmap_info
->bitmap
,
1461 BITS_PER_BITMAP
, i
);
1462 if ((next_zero
- i
) >= bits
) {
1463 found_bits
= next_zero
- i
;
1470 *offset
= (u64
)(i
* ctl
->unit
) + bitmap_info
->offset
;
1471 *bytes
= (u64
)(found_bits
) * ctl
->unit
;
1478 static struct btrfs_free_space
*
1479 find_free_space(struct btrfs_free_space_ctl
*ctl
, u64
*offset
, u64
*bytes
)
1481 struct btrfs_free_space
*entry
;
1482 struct rb_node
*node
;
1485 if (!ctl
->free_space_offset
.rb_node
)
1488 entry
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, *offset
), 0, 1);
1492 for (node
= &entry
->offset_index
; node
; node
= rb_next(node
)) {
1493 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
1494 if (entry
->bytes
< *bytes
)
1497 if (entry
->bitmap
) {
1498 ret
= search_bitmap(ctl
, entry
, offset
, bytes
);
1504 *offset
= entry
->offset
;
1505 *bytes
= entry
->bytes
;
1512 static void add_new_bitmap(struct btrfs_free_space_ctl
*ctl
,
1513 struct btrfs_free_space
*info
, u64 offset
)
1515 info
->offset
= offset_to_bitmap(ctl
, offset
);
1517 INIT_LIST_HEAD(&info
->list
);
1518 link_free_space(ctl
, info
);
1519 ctl
->total_bitmaps
++;
1521 ctl
->op
->recalc_thresholds(ctl
);
1524 static void free_bitmap(struct btrfs_free_space_ctl
*ctl
,
1525 struct btrfs_free_space
*bitmap_info
)
1527 unlink_free_space(ctl
, bitmap_info
);
1528 kfree(bitmap_info
->bitmap
);
1529 kmem_cache_free(btrfs_free_space_cachep
, bitmap_info
);
1530 ctl
->total_bitmaps
--;
1531 ctl
->op
->recalc_thresholds(ctl
);
1534 static noinline
int remove_from_bitmap(struct btrfs_free_space_ctl
*ctl
,
1535 struct btrfs_free_space
*bitmap_info
,
1536 u64
*offset
, u64
*bytes
)
1539 u64 search_start
, search_bytes
;
1543 end
= bitmap_info
->offset
+ (u64
)(BITS_PER_BITMAP
* ctl
->unit
) - 1;
1546 * XXX - this can go away after a few releases.
1548 * since the only user of btrfs_remove_free_space is the tree logging
1549 * stuff, and the only way to test that is under crash conditions, we
1550 * want to have this debug stuff here just in case somethings not
1551 * working. Search the bitmap for the space we are trying to use to
1552 * make sure its actually there. If its not there then we need to stop
1553 * because something has gone wrong.
1555 search_start
= *offset
;
1556 search_bytes
= *bytes
;
1557 search_bytes
= min(search_bytes
, end
- search_start
+ 1);
1558 ret
= search_bitmap(ctl
, bitmap_info
, &search_start
, &search_bytes
);
1559 BUG_ON(ret
< 0 || search_start
!= *offset
);
1561 if (*offset
> bitmap_info
->offset
&& *offset
+ *bytes
> end
) {
1562 bitmap_clear_bits(ctl
, bitmap_info
, *offset
, end
- *offset
+ 1);
1563 *bytes
-= end
- *offset
+ 1;
1565 } else if (*offset
>= bitmap_info
->offset
&& *offset
+ *bytes
<= end
) {
1566 bitmap_clear_bits(ctl
, bitmap_info
, *offset
, *bytes
);
1571 struct rb_node
*next
= rb_next(&bitmap_info
->offset_index
);
1572 if (!bitmap_info
->bytes
)
1573 free_bitmap(ctl
, bitmap_info
);
1576 * no entry after this bitmap, but we still have bytes to
1577 * remove, so something has gone wrong.
1582 bitmap_info
= rb_entry(next
, struct btrfs_free_space
,
1586 * if the next entry isn't a bitmap we need to return to let the
1587 * extent stuff do its work.
1589 if (!bitmap_info
->bitmap
)
1593 * Ok the next item is a bitmap, but it may not actually hold
1594 * the information for the rest of this free space stuff, so
1595 * look for it, and if we don't find it return so we can try
1596 * everything over again.
1598 search_start
= *offset
;
1599 search_bytes
= *bytes
;
1600 ret
= search_bitmap(ctl
, bitmap_info
, &search_start
,
1602 if (ret
< 0 || search_start
!= *offset
)
1606 } else if (!bitmap_info
->bytes
)
1607 free_bitmap(ctl
, bitmap_info
);
1612 static u64
add_bytes_to_bitmap(struct btrfs_free_space_ctl
*ctl
,
1613 struct btrfs_free_space
*info
, u64 offset
,
1616 u64 bytes_to_set
= 0;
1619 end
= info
->offset
+ (u64
)(BITS_PER_BITMAP
* ctl
->unit
);
1621 bytes_to_set
= min(end
- offset
, bytes
);
1623 bitmap_set_bits(ctl
, info
, offset
, bytes_to_set
);
1625 return bytes_to_set
;
1629 static bool use_bitmap(struct btrfs_free_space_ctl
*ctl
,
1630 struct btrfs_free_space
*info
)
1632 struct btrfs_block_group_cache
*block_group
= ctl
->private;
1635 * If we are below the extents threshold then we can add this as an
1636 * extent, and don't have to deal with the bitmap
1638 if (ctl
->free_extents
< ctl
->extents_thresh
) {
1640 * If this block group has some small extents we don't want to
1641 * use up all of our free slots in the cache with them, we want
1642 * to reserve them to larger extents, however if we have plent
1643 * of cache left then go ahead an dadd them, no sense in adding
1644 * the overhead of a bitmap if we don't have to.
1646 if (info
->bytes
<= block_group
->sectorsize
* 4) {
1647 if (ctl
->free_extents
* 2 <= ctl
->extents_thresh
)
1655 * some block groups are so tiny they can't be enveloped by a bitmap, so
1656 * don't even bother to create a bitmap for this
1658 if (BITS_PER_BITMAP
* block_group
->sectorsize
>
1659 block_group
->key
.offset
)
1665 static struct btrfs_free_space_op free_space_op
= {
1666 .recalc_thresholds
= recalculate_thresholds
,
1667 .use_bitmap
= use_bitmap
,
1670 static int insert_into_bitmap(struct btrfs_free_space_ctl
*ctl
,
1671 struct btrfs_free_space
*info
)
1673 struct btrfs_free_space
*bitmap_info
;
1674 struct btrfs_block_group_cache
*block_group
= NULL
;
1676 u64 bytes
, offset
, bytes_added
;
1679 bytes
= info
->bytes
;
1680 offset
= info
->offset
;
1682 if (!ctl
->op
->use_bitmap(ctl
, info
))
1685 if (ctl
->op
== &free_space_op
)
1686 block_group
= ctl
->private;
1689 * Since we link bitmaps right into the cluster we need to see if we
1690 * have a cluster here, and if so and it has our bitmap we need to add
1691 * the free space to that bitmap.
1693 if (block_group
&& !list_empty(&block_group
->cluster_list
)) {
1694 struct btrfs_free_cluster
*cluster
;
1695 struct rb_node
*node
;
1696 struct btrfs_free_space
*entry
;
1698 cluster
= list_entry(block_group
->cluster_list
.next
,
1699 struct btrfs_free_cluster
,
1701 spin_lock(&cluster
->lock
);
1702 node
= rb_first(&cluster
->root
);
1704 spin_unlock(&cluster
->lock
);
1705 goto no_cluster_bitmap
;
1708 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
1709 if (!entry
->bitmap
) {
1710 spin_unlock(&cluster
->lock
);
1711 goto no_cluster_bitmap
;
1714 if (entry
->offset
== offset_to_bitmap(ctl
, offset
)) {
1715 bytes_added
= add_bytes_to_bitmap(ctl
, entry
,
1717 bytes
-= bytes_added
;
1718 offset
+= bytes_added
;
1720 spin_unlock(&cluster
->lock
);
1728 bitmap_info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
1735 bytes_added
= add_bytes_to_bitmap(ctl
, bitmap_info
, offset
, bytes
);
1736 bytes
-= bytes_added
;
1737 offset
+= bytes_added
;
1747 if (info
&& info
->bitmap
) {
1748 add_new_bitmap(ctl
, info
, offset
);
1753 spin_unlock(&ctl
->tree_lock
);
1755 /* no pre-allocated info, allocate a new one */
1757 info
= kmem_cache_zalloc(btrfs_free_space_cachep
,
1760 spin_lock(&ctl
->tree_lock
);
1766 /* allocate the bitmap */
1767 info
->bitmap
= kzalloc(PAGE_CACHE_SIZE
, GFP_NOFS
);
1768 spin_lock(&ctl
->tree_lock
);
1769 if (!info
->bitmap
) {
1779 kfree(info
->bitmap
);
1780 kmem_cache_free(btrfs_free_space_cachep
, info
);
1786 static bool try_merge_free_space(struct btrfs_free_space_ctl
*ctl
,
1787 struct btrfs_free_space
*info
, bool update_stat
)
1789 struct btrfs_free_space
*left_info
;
1790 struct btrfs_free_space
*right_info
;
1791 bool merged
= false;
1792 u64 offset
= info
->offset
;
1793 u64 bytes
= info
->bytes
;
1796 * first we want to see if there is free space adjacent to the range we
1797 * are adding, if there is remove that struct and add a new one to
1798 * cover the entire range
1800 right_info
= tree_search_offset(ctl
, offset
+ bytes
, 0, 0);
1801 if (right_info
&& rb_prev(&right_info
->offset_index
))
1802 left_info
= rb_entry(rb_prev(&right_info
->offset_index
),
1803 struct btrfs_free_space
, offset_index
);
1805 left_info
= tree_search_offset(ctl
, offset
- 1, 0, 0);
1807 if (right_info
&& !right_info
->bitmap
) {
1809 unlink_free_space(ctl
, right_info
);
1811 __unlink_free_space(ctl
, right_info
);
1812 info
->bytes
+= right_info
->bytes
;
1813 kmem_cache_free(btrfs_free_space_cachep
, right_info
);
1817 if (left_info
&& !left_info
->bitmap
&&
1818 left_info
->offset
+ left_info
->bytes
== offset
) {
1820 unlink_free_space(ctl
, left_info
);
1822 __unlink_free_space(ctl
, left_info
);
1823 info
->offset
= left_info
->offset
;
1824 info
->bytes
+= left_info
->bytes
;
1825 kmem_cache_free(btrfs_free_space_cachep
, left_info
);
1832 int __btrfs_add_free_space(struct btrfs_free_space_ctl
*ctl
,
1833 u64 offset
, u64 bytes
)
1835 struct btrfs_free_space
*info
;
1838 info
= kmem_cache_zalloc(btrfs_free_space_cachep
, GFP_NOFS
);
1842 info
->offset
= offset
;
1843 info
->bytes
= bytes
;
1845 spin_lock(&ctl
->tree_lock
);
1847 if (try_merge_free_space(ctl
, info
, true))
1851 * There was no extent directly to the left or right of this new
1852 * extent then we know we're going to have to allocate a new extent, so
1853 * before we do that see if we need to drop this into a bitmap
1855 ret
= insert_into_bitmap(ctl
, info
);
1863 ret
= link_free_space(ctl
, info
);
1865 kmem_cache_free(btrfs_free_space_cachep
, info
);
1867 spin_unlock(&ctl
->tree_lock
);
1870 printk(KERN_CRIT
"btrfs: unable to add free space :%d\n", ret
);
1871 BUG_ON(ret
== -EEXIST
);
1877 int btrfs_remove_free_space(struct btrfs_block_group_cache
*block_group
,
1878 u64 offset
, u64 bytes
)
1880 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
1881 struct btrfs_free_space
*info
;
1882 struct btrfs_free_space
*next_info
= NULL
;
1885 spin_lock(&ctl
->tree_lock
);
1888 info
= tree_search_offset(ctl
, offset
, 0, 0);
1891 * oops didn't find an extent that matched the space we wanted
1892 * to remove, look for a bitmap instead
1894 info
= tree_search_offset(ctl
, offset_to_bitmap(ctl
, offset
),
1897 /* the tree logging code might be calling us before we
1898 * have fully loaded the free space rbtree for this
1899 * block group. So it is possible the entry won't
1900 * be in the rbtree yet at all. The caching code
1901 * will make sure not to put it in the rbtree if
1902 * the logging code has pinned it.
1908 if (info
->bytes
< bytes
&& rb_next(&info
->offset_index
)) {
1910 next_info
= rb_entry(rb_next(&info
->offset_index
),
1911 struct btrfs_free_space
,
1914 if (next_info
->bitmap
)
1915 end
= next_info
->offset
+
1916 BITS_PER_BITMAP
* ctl
->unit
- 1;
1918 end
= next_info
->offset
+ next_info
->bytes
;
1920 if (next_info
->bytes
< bytes
||
1921 next_info
->offset
> offset
|| offset
> end
) {
1922 printk(KERN_CRIT
"Found free space at %llu, size %llu,"
1923 " trying to use %llu\n",
1924 (unsigned long long)info
->offset
,
1925 (unsigned long long)info
->bytes
,
1926 (unsigned long long)bytes
);
1935 if (info
->bytes
== bytes
) {
1936 unlink_free_space(ctl
, info
);
1938 kfree(info
->bitmap
);
1939 ctl
->total_bitmaps
--;
1941 kmem_cache_free(btrfs_free_space_cachep
, info
);
1946 if (!info
->bitmap
&& info
->offset
== offset
) {
1947 unlink_free_space(ctl
, info
);
1948 info
->offset
+= bytes
;
1949 info
->bytes
-= bytes
;
1950 ret
= link_free_space(ctl
, info
);
1955 if (!info
->bitmap
&& info
->offset
<= offset
&&
1956 info
->offset
+ info
->bytes
>= offset
+ bytes
) {
1957 u64 old_start
= info
->offset
;
1959 * we're freeing space in the middle of the info,
1960 * this can happen during tree log replay
1962 * first unlink the old info and then
1963 * insert it again after the hole we're creating
1965 unlink_free_space(ctl
, info
);
1966 if (offset
+ bytes
< info
->offset
+ info
->bytes
) {
1967 u64 old_end
= info
->offset
+ info
->bytes
;
1969 info
->offset
= offset
+ bytes
;
1970 info
->bytes
= old_end
- info
->offset
;
1971 ret
= link_free_space(ctl
, info
);
1976 /* the hole we're creating ends at the end
1977 * of the info struct, just free the info
1979 kmem_cache_free(btrfs_free_space_cachep
, info
);
1981 spin_unlock(&ctl
->tree_lock
);
1983 /* step two, insert a new info struct to cover
1984 * anything before the hole
1986 ret
= btrfs_add_free_space(block_group
, old_start
,
1987 offset
- old_start
);
1988 WARN_ON(ret
); /* -ENOMEM */
1992 ret
= remove_from_bitmap(ctl
, info
, &offset
, &bytes
);
1995 BUG_ON(ret
); /* logic error */
1997 spin_unlock(&ctl
->tree_lock
);
2002 void btrfs_dump_free_space(struct btrfs_block_group_cache
*block_group
,
2005 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2006 struct btrfs_free_space
*info
;
2010 for (n
= rb_first(&ctl
->free_space_offset
); n
; n
= rb_next(n
)) {
2011 info
= rb_entry(n
, struct btrfs_free_space
, offset_index
);
2012 if (info
->bytes
>= bytes
)
2014 printk(KERN_CRIT
"entry offset %llu, bytes %llu, bitmap %s\n",
2015 (unsigned long long)info
->offset
,
2016 (unsigned long long)info
->bytes
,
2017 (info
->bitmap
) ? "yes" : "no");
2019 printk(KERN_INFO
"block group has cluster?: %s\n",
2020 list_empty(&block_group
->cluster_list
) ? "no" : "yes");
2021 printk(KERN_INFO
"%d blocks of free space at or bigger than bytes is"
2025 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache
*block_group
)
2027 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2029 spin_lock_init(&ctl
->tree_lock
);
2030 ctl
->unit
= block_group
->sectorsize
;
2031 ctl
->start
= block_group
->key
.objectid
;
2032 ctl
->private = block_group
;
2033 ctl
->op
= &free_space_op
;
2036 * we only want to have 32k of ram per block group for keeping
2037 * track of free space, and if we pass 1/2 of that we want to
2038 * start converting things over to using bitmaps
2040 ctl
->extents_thresh
= ((1024 * 32) / 2) /
2041 sizeof(struct btrfs_free_space
);
2045 * for a given cluster, put all of its extents back into the free
2046 * space cache. If the block group passed doesn't match the block group
2047 * pointed to by the cluster, someone else raced in and freed the
2048 * cluster already. In that case, we just return without changing anything
2051 __btrfs_return_cluster_to_free_space(
2052 struct btrfs_block_group_cache
*block_group
,
2053 struct btrfs_free_cluster
*cluster
)
2055 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2056 struct btrfs_free_space
*entry
;
2057 struct rb_node
*node
;
2059 spin_lock(&cluster
->lock
);
2060 if (cluster
->block_group
!= block_group
)
2063 cluster
->block_group
= NULL
;
2064 cluster
->window_start
= 0;
2065 list_del_init(&cluster
->block_group_list
);
2067 node
= rb_first(&cluster
->root
);
2071 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2072 node
= rb_next(&entry
->offset_index
);
2073 rb_erase(&entry
->offset_index
, &cluster
->root
);
2075 bitmap
= (entry
->bitmap
!= NULL
);
2077 try_merge_free_space(ctl
, entry
, false);
2078 tree_insert_offset(&ctl
->free_space_offset
,
2079 entry
->offset
, &entry
->offset_index
, bitmap
);
2081 cluster
->root
= RB_ROOT
;
2084 spin_unlock(&cluster
->lock
);
2085 btrfs_put_block_group(block_group
);
2089 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl
*ctl
)
2091 struct btrfs_free_space
*info
;
2092 struct rb_node
*node
;
2094 while ((node
= rb_last(&ctl
->free_space_offset
)) != NULL
) {
2095 info
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2096 if (!info
->bitmap
) {
2097 unlink_free_space(ctl
, info
);
2098 kmem_cache_free(btrfs_free_space_cachep
, info
);
2100 free_bitmap(ctl
, info
);
2102 if (need_resched()) {
2103 spin_unlock(&ctl
->tree_lock
);
2105 spin_lock(&ctl
->tree_lock
);
2110 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl
*ctl
)
2112 spin_lock(&ctl
->tree_lock
);
2113 __btrfs_remove_free_space_cache_locked(ctl
);
2114 spin_unlock(&ctl
->tree_lock
);
2117 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache
*block_group
)
2119 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2120 struct btrfs_free_cluster
*cluster
;
2121 struct list_head
*head
;
2123 spin_lock(&ctl
->tree_lock
);
2124 while ((head
= block_group
->cluster_list
.next
) !=
2125 &block_group
->cluster_list
) {
2126 cluster
= list_entry(head
, struct btrfs_free_cluster
,
2129 WARN_ON(cluster
->block_group
!= block_group
);
2130 __btrfs_return_cluster_to_free_space(block_group
, cluster
);
2131 if (need_resched()) {
2132 spin_unlock(&ctl
->tree_lock
);
2134 spin_lock(&ctl
->tree_lock
);
2137 __btrfs_remove_free_space_cache_locked(ctl
);
2138 spin_unlock(&ctl
->tree_lock
);
2142 u64
btrfs_find_space_for_alloc(struct btrfs_block_group_cache
*block_group
,
2143 u64 offset
, u64 bytes
, u64 empty_size
)
2145 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2146 struct btrfs_free_space
*entry
= NULL
;
2147 u64 bytes_search
= bytes
+ empty_size
;
2150 spin_lock(&ctl
->tree_lock
);
2151 entry
= find_free_space(ctl
, &offset
, &bytes_search
);
2156 if (entry
->bitmap
) {
2157 bitmap_clear_bits(ctl
, entry
, offset
, bytes
);
2159 free_bitmap(ctl
, entry
);
2161 unlink_free_space(ctl
, entry
);
2162 entry
->offset
+= bytes
;
2163 entry
->bytes
-= bytes
;
2165 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2167 link_free_space(ctl
, entry
);
2171 spin_unlock(&ctl
->tree_lock
);
2177 * given a cluster, put all of its extents back into the free space
2178 * cache. If a block group is passed, this function will only free
2179 * a cluster that belongs to the passed block group.
2181 * Otherwise, it'll get a reference on the block group pointed to by the
2182 * cluster and remove the cluster from it.
2184 int btrfs_return_cluster_to_free_space(
2185 struct btrfs_block_group_cache
*block_group
,
2186 struct btrfs_free_cluster
*cluster
)
2188 struct btrfs_free_space_ctl
*ctl
;
2191 /* first, get a safe pointer to the block group */
2192 spin_lock(&cluster
->lock
);
2194 block_group
= cluster
->block_group
;
2196 spin_unlock(&cluster
->lock
);
2199 } else if (cluster
->block_group
!= block_group
) {
2200 /* someone else has already freed it don't redo their work */
2201 spin_unlock(&cluster
->lock
);
2204 atomic_inc(&block_group
->count
);
2205 spin_unlock(&cluster
->lock
);
2207 ctl
= block_group
->free_space_ctl
;
2209 /* now return any extents the cluster had on it */
2210 spin_lock(&ctl
->tree_lock
);
2211 ret
= __btrfs_return_cluster_to_free_space(block_group
, cluster
);
2212 spin_unlock(&ctl
->tree_lock
);
2214 /* finally drop our ref */
2215 btrfs_put_block_group(block_group
);
2219 static u64
btrfs_alloc_from_bitmap(struct btrfs_block_group_cache
*block_group
,
2220 struct btrfs_free_cluster
*cluster
,
2221 struct btrfs_free_space
*entry
,
2222 u64 bytes
, u64 min_start
)
2224 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2226 u64 search_start
= cluster
->window_start
;
2227 u64 search_bytes
= bytes
;
2230 search_start
= min_start
;
2231 search_bytes
= bytes
;
2233 err
= search_bitmap(ctl
, entry
, &search_start
, &search_bytes
);
2238 __bitmap_clear_bits(ctl
, entry
, ret
, bytes
);
2244 * given a cluster, try to allocate 'bytes' from it, returns 0
2245 * if it couldn't find anything suitably large, or a logical disk offset
2246 * if things worked out
2248 u64
btrfs_alloc_from_cluster(struct btrfs_block_group_cache
*block_group
,
2249 struct btrfs_free_cluster
*cluster
, u64 bytes
,
2252 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2253 struct btrfs_free_space
*entry
= NULL
;
2254 struct rb_node
*node
;
2257 spin_lock(&cluster
->lock
);
2258 if (bytes
> cluster
->max_size
)
2261 if (cluster
->block_group
!= block_group
)
2264 node
= rb_first(&cluster
->root
);
2268 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2270 if (entry
->bytes
< bytes
||
2271 (!entry
->bitmap
&& entry
->offset
< min_start
)) {
2272 node
= rb_next(&entry
->offset_index
);
2275 entry
= rb_entry(node
, struct btrfs_free_space
,
2280 if (entry
->bitmap
) {
2281 ret
= btrfs_alloc_from_bitmap(block_group
,
2282 cluster
, entry
, bytes
,
2283 cluster
->window_start
);
2285 node
= rb_next(&entry
->offset_index
);
2288 entry
= rb_entry(node
, struct btrfs_free_space
,
2292 cluster
->window_start
+= bytes
;
2294 ret
= entry
->offset
;
2296 entry
->offset
+= bytes
;
2297 entry
->bytes
-= bytes
;
2300 if (entry
->bytes
== 0)
2301 rb_erase(&entry
->offset_index
, &cluster
->root
);
2305 spin_unlock(&cluster
->lock
);
2310 spin_lock(&ctl
->tree_lock
);
2312 ctl
->free_space
-= bytes
;
2313 if (entry
->bytes
== 0) {
2314 ctl
->free_extents
--;
2315 if (entry
->bitmap
) {
2316 kfree(entry
->bitmap
);
2317 ctl
->total_bitmaps
--;
2318 ctl
->op
->recalc_thresholds(ctl
);
2320 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2323 spin_unlock(&ctl
->tree_lock
);
2328 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache
*block_group
,
2329 struct btrfs_free_space
*entry
,
2330 struct btrfs_free_cluster
*cluster
,
2331 u64 offset
, u64 bytes
,
2332 u64 cont1_bytes
, u64 min_bytes
)
2334 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2335 unsigned long next_zero
;
2337 unsigned long want_bits
;
2338 unsigned long min_bits
;
2339 unsigned long found_bits
;
2340 unsigned long start
= 0;
2341 unsigned long total_found
= 0;
2344 i
= offset_to_bit(entry
->offset
, block_group
->sectorsize
,
2345 max_t(u64
, offset
, entry
->offset
));
2346 want_bits
= bytes_to_bits(bytes
, block_group
->sectorsize
);
2347 min_bits
= bytes_to_bits(min_bytes
, block_group
->sectorsize
);
2351 for (i
= find_next_bit(entry
->bitmap
, BITS_PER_BITMAP
, i
);
2352 i
< BITS_PER_BITMAP
;
2353 i
= find_next_bit(entry
->bitmap
, BITS_PER_BITMAP
, i
+ 1)) {
2354 next_zero
= find_next_zero_bit(entry
->bitmap
,
2355 BITS_PER_BITMAP
, i
);
2356 if (next_zero
- i
>= min_bits
) {
2357 found_bits
= next_zero
- i
;
2368 cluster
->max_size
= 0;
2371 total_found
+= found_bits
;
2373 if (cluster
->max_size
< found_bits
* block_group
->sectorsize
)
2374 cluster
->max_size
= found_bits
* block_group
->sectorsize
;
2376 if (total_found
< want_bits
|| cluster
->max_size
< cont1_bytes
) {
2381 cluster
->window_start
= start
* block_group
->sectorsize
+
2383 rb_erase(&entry
->offset_index
, &ctl
->free_space_offset
);
2384 ret
= tree_insert_offset(&cluster
->root
, entry
->offset
,
2385 &entry
->offset_index
, 1);
2386 BUG_ON(ret
); /* -EEXIST; Logic error */
2388 trace_btrfs_setup_cluster(block_group
, cluster
,
2389 total_found
* block_group
->sectorsize
, 1);
2394 * This searches the block group for just extents to fill the cluster with.
2395 * Try to find a cluster with at least bytes total bytes, at least one
2396 * extent of cont1_bytes, and other clusters of at least min_bytes.
2399 setup_cluster_no_bitmap(struct btrfs_block_group_cache
*block_group
,
2400 struct btrfs_free_cluster
*cluster
,
2401 struct list_head
*bitmaps
, u64 offset
, u64 bytes
,
2402 u64 cont1_bytes
, u64 min_bytes
)
2404 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2405 struct btrfs_free_space
*first
= NULL
;
2406 struct btrfs_free_space
*entry
= NULL
;
2407 struct btrfs_free_space
*last
;
2408 struct rb_node
*node
;
2414 entry
= tree_search_offset(ctl
, offset
, 0, 1);
2419 * We don't want bitmaps, so just move along until we find a normal
2422 while (entry
->bitmap
|| entry
->bytes
< min_bytes
) {
2423 if (entry
->bitmap
&& list_empty(&entry
->list
))
2424 list_add_tail(&entry
->list
, bitmaps
);
2425 node
= rb_next(&entry
->offset_index
);
2428 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2431 window_start
= entry
->offset
;
2432 window_free
= entry
->bytes
;
2433 max_extent
= entry
->bytes
;
2437 for (node
= rb_next(&entry
->offset_index
); node
;
2438 node
= rb_next(&entry
->offset_index
)) {
2439 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2441 if (entry
->bitmap
) {
2442 if (list_empty(&entry
->list
))
2443 list_add_tail(&entry
->list
, bitmaps
);
2447 if (entry
->bytes
< min_bytes
)
2451 window_free
+= entry
->bytes
;
2452 if (entry
->bytes
> max_extent
)
2453 max_extent
= entry
->bytes
;
2456 if (window_free
< bytes
|| max_extent
< cont1_bytes
)
2459 cluster
->window_start
= first
->offset
;
2461 node
= &first
->offset_index
;
2464 * now we've found our entries, pull them out of the free space
2465 * cache and put them into the cluster rbtree
2470 entry
= rb_entry(node
, struct btrfs_free_space
, offset_index
);
2471 node
= rb_next(&entry
->offset_index
);
2472 if (entry
->bitmap
|| entry
->bytes
< min_bytes
)
2475 rb_erase(&entry
->offset_index
, &ctl
->free_space_offset
);
2476 ret
= tree_insert_offset(&cluster
->root
, entry
->offset
,
2477 &entry
->offset_index
, 0);
2478 total_size
+= entry
->bytes
;
2479 BUG_ON(ret
); /* -EEXIST; Logic error */
2480 } while (node
&& entry
!= last
);
2482 cluster
->max_size
= max_extent
;
2483 trace_btrfs_setup_cluster(block_group
, cluster
, total_size
, 0);
2488 * This specifically looks for bitmaps that may work in the cluster, we assume
2489 * that we have already failed to find extents that will work.
2492 setup_cluster_bitmap(struct btrfs_block_group_cache
*block_group
,
2493 struct btrfs_free_cluster
*cluster
,
2494 struct list_head
*bitmaps
, u64 offset
, u64 bytes
,
2495 u64 cont1_bytes
, u64 min_bytes
)
2497 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2498 struct btrfs_free_space
*entry
;
2500 u64 bitmap_offset
= offset_to_bitmap(ctl
, offset
);
2502 if (ctl
->total_bitmaps
== 0)
2506 * The bitmap that covers offset won't be in the list unless offset
2507 * is just its start offset.
2509 entry
= list_first_entry(bitmaps
, struct btrfs_free_space
, list
);
2510 if (entry
->offset
!= bitmap_offset
) {
2511 entry
= tree_search_offset(ctl
, bitmap_offset
, 1, 0);
2512 if (entry
&& list_empty(&entry
->list
))
2513 list_add(&entry
->list
, bitmaps
);
2516 list_for_each_entry(entry
, bitmaps
, list
) {
2517 if (entry
->bytes
< bytes
)
2519 ret
= btrfs_bitmap_cluster(block_group
, entry
, cluster
, offset
,
2520 bytes
, cont1_bytes
, min_bytes
);
2526 * The bitmaps list has all the bitmaps that record free space
2527 * starting after offset, so no more search is required.
2533 * here we try to find a cluster of blocks in a block group. The goal
2534 * is to find at least bytes+empty_size.
2535 * We might not find them all in one contiguous area.
2537 * returns zero and sets up cluster if things worked out, otherwise
2538 * it returns -enospc
2540 int btrfs_find_space_cluster(struct btrfs_trans_handle
*trans
,
2541 struct btrfs_root
*root
,
2542 struct btrfs_block_group_cache
*block_group
,
2543 struct btrfs_free_cluster
*cluster
,
2544 u64 offset
, u64 bytes
, u64 empty_size
)
2546 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2547 struct btrfs_free_space
*entry
, *tmp
;
2554 * Choose the minimum extent size we'll require for this
2555 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2556 * For metadata, allow allocates with smaller extents. For
2557 * data, keep it dense.
2559 if (btrfs_test_opt(root
, SSD_SPREAD
)) {
2560 cont1_bytes
= min_bytes
= bytes
+ empty_size
;
2561 } else if (block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
2562 cont1_bytes
= bytes
;
2563 min_bytes
= block_group
->sectorsize
;
2565 cont1_bytes
= max(bytes
, (bytes
+ empty_size
) >> 2);
2566 min_bytes
= block_group
->sectorsize
;
2569 spin_lock(&ctl
->tree_lock
);
2572 * If we know we don't have enough space to make a cluster don't even
2573 * bother doing all the work to try and find one.
2575 if (ctl
->free_space
< bytes
) {
2576 spin_unlock(&ctl
->tree_lock
);
2580 spin_lock(&cluster
->lock
);
2582 /* someone already found a cluster, hooray */
2583 if (cluster
->block_group
) {
2588 trace_btrfs_find_cluster(block_group
, offset
, bytes
, empty_size
,
2591 INIT_LIST_HEAD(&bitmaps
);
2592 ret
= setup_cluster_no_bitmap(block_group
, cluster
, &bitmaps
, offset
,
2594 cont1_bytes
, min_bytes
);
2596 ret
= setup_cluster_bitmap(block_group
, cluster
, &bitmaps
,
2597 offset
, bytes
+ empty_size
,
2598 cont1_bytes
, min_bytes
);
2600 /* Clear our temporary list */
2601 list_for_each_entry_safe(entry
, tmp
, &bitmaps
, list
)
2602 list_del_init(&entry
->list
);
2605 atomic_inc(&block_group
->count
);
2606 list_add_tail(&cluster
->block_group_list
,
2607 &block_group
->cluster_list
);
2608 cluster
->block_group
= block_group
;
2610 trace_btrfs_failed_cluster_setup(block_group
);
2613 spin_unlock(&cluster
->lock
);
2614 spin_unlock(&ctl
->tree_lock
);
2620 * simple code to zero out a cluster
2622 void btrfs_init_free_cluster(struct btrfs_free_cluster
*cluster
)
2624 spin_lock_init(&cluster
->lock
);
2625 spin_lock_init(&cluster
->refill_lock
);
2626 cluster
->root
= RB_ROOT
;
2627 cluster
->max_size
= 0;
2628 INIT_LIST_HEAD(&cluster
->block_group_list
);
2629 cluster
->block_group
= NULL
;
2632 static int do_trimming(struct btrfs_block_group_cache
*block_group
,
2633 u64
*total_trimmed
, u64 start
, u64 bytes
,
2634 u64 reserved_start
, u64 reserved_bytes
)
2636 struct btrfs_space_info
*space_info
= block_group
->space_info
;
2637 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
2642 spin_lock(&space_info
->lock
);
2643 spin_lock(&block_group
->lock
);
2644 if (!block_group
->ro
) {
2645 block_group
->reserved
+= reserved_bytes
;
2646 space_info
->bytes_reserved
+= reserved_bytes
;
2649 spin_unlock(&block_group
->lock
);
2650 spin_unlock(&space_info
->lock
);
2652 ret
= btrfs_error_discard_extent(fs_info
->extent_root
,
2653 start
, bytes
, &trimmed
);
2655 *total_trimmed
+= trimmed
;
2657 btrfs_add_free_space(block_group
, reserved_start
, reserved_bytes
);
2660 spin_lock(&space_info
->lock
);
2661 spin_lock(&block_group
->lock
);
2662 if (block_group
->ro
)
2663 space_info
->bytes_readonly
+= reserved_bytes
;
2664 block_group
->reserved
-= reserved_bytes
;
2665 space_info
->bytes_reserved
-= reserved_bytes
;
2666 spin_unlock(&space_info
->lock
);
2667 spin_unlock(&block_group
->lock
);
2673 static int trim_no_bitmap(struct btrfs_block_group_cache
*block_group
,
2674 u64
*total_trimmed
, u64 start
, u64 end
, u64 minlen
)
2676 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2677 struct btrfs_free_space
*entry
;
2678 struct rb_node
*node
;
2684 while (start
< end
) {
2685 spin_lock(&ctl
->tree_lock
);
2687 if (ctl
->free_space
< minlen
) {
2688 spin_unlock(&ctl
->tree_lock
);
2692 entry
= tree_search_offset(ctl
, start
, 0, 1);
2694 spin_unlock(&ctl
->tree_lock
);
2699 while (entry
->bitmap
) {
2700 node
= rb_next(&entry
->offset_index
);
2702 spin_unlock(&ctl
->tree_lock
);
2705 entry
= rb_entry(node
, struct btrfs_free_space
,
2709 if (entry
->offset
>= end
) {
2710 spin_unlock(&ctl
->tree_lock
);
2714 extent_start
= entry
->offset
;
2715 extent_bytes
= entry
->bytes
;
2716 start
= max(start
, extent_start
);
2717 bytes
= min(extent_start
+ extent_bytes
, end
) - start
;
2718 if (bytes
< minlen
) {
2719 spin_unlock(&ctl
->tree_lock
);
2723 unlink_free_space(ctl
, entry
);
2724 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2726 spin_unlock(&ctl
->tree_lock
);
2728 ret
= do_trimming(block_group
, total_trimmed
, start
, bytes
,
2729 extent_start
, extent_bytes
);
2735 if (fatal_signal_pending(current
)) {
2746 static int trim_bitmaps(struct btrfs_block_group_cache
*block_group
,
2747 u64
*total_trimmed
, u64 start
, u64 end
, u64 minlen
)
2749 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
2750 struct btrfs_free_space
*entry
;
2754 u64 offset
= offset_to_bitmap(ctl
, start
);
2756 while (offset
< end
) {
2757 bool next_bitmap
= false;
2759 spin_lock(&ctl
->tree_lock
);
2761 if (ctl
->free_space
< minlen
) {
2762 spin_unlock(&ctl
->tree_lock
);
2766 entry
= tree_search_offset(ctl
, offset
, 1, 0);
2768 spin_unlock(&ctl
->tree_lock
);
2774 ret2
= search_bitmap(ctl
, entry
, &start
, &bytes
);
2775 if (ret2
|| start
>= end
) {
2776 spin_unlock(&ctl
->tree_lock
);
2781 bytes
= min(bytes
, end
- start
);
2782 if (bytes
< minlen
) {
2783 spin_unlock(&ctl
->tree_lock
);
2787 bitmap_clear_bits(ctl
, entry
, start
, bytes
);
2788 if (entry
->bytes
== 0)
2789 free_bitmap(ctl
, entry
);
2791 spin_unlock(&ctl
->tree_lock
);
2793 ret
= do_trimming(block_group
, total_trimmed
, start
, bytes
,
2799 offset
+= BITS_PER_BITMAP
* ctl
->unit
;
2802 if (start
>= offset
+ BITS_PER_BITMAP
* ctl
->unit
)
2803 offset
+= BITS_PER_BITMAP
* ctl
->unit
;
2806 if (fatal_signal_pending(current
)) {
2817 int btrfs_trim_block_group(struct btrfs_block_group_cache
*block_group
,
2818 u64
*trimmed
, u64 start
, u64 end
, u64 minlen
)
2824 ret
= trim_no_bitmap(block_group
, trimmed
, start
, end
, minlen
);
2828 ret
= trim_bitmaps(block_group
, trimmed
, start
, end
, minlen
);
2834 * Find the left-most item in the cache tree, and then return the
2835 * smallest inode number in the item.
2837 * Note: the returned inode number may not be the smallest one in
2838 * the tree, if the left-most item is a bitmap.
2840 u64
btrfs_find_ino_for_alloc(struct btrfs_root
*fs_root
)
2842 struct btrfs_free_space_ctl
*ctl
= fs_root
->free_ino_ctl
;
2843 struct btrfs_free_space
*entry
= NULL
;
2846 spin_lock(&ctl
->tree_lock
);
2848 if (RB_EMPTY_ROOT(&ctl
->free_space_offset
))
2851 entry
= rb_entry(rb_first(&ctl
->free_space_offset
),
2852 struct btrfs_free_space
, offset_index
);
2854 if (!entry
->bitmap
) {
2855 ino
= entry
->offset
;
2857 unlink_free_space(ctl
, entry
);
2861 kmem_cache_free(btrfs_free_space_cachep
, entry
);
2863 link_free_space(ctl
, entry
);
2869 ret
= search_bitmap(ctl
, entry
, &offset
, &count
);
2870 /* Logic error; Should be empty if it can't find anything */
2874 bitmap_clear_bits(ctl
, entry
, offset
, 1);
2875 if (entry
->bytes
== 0)
2876 free_bitmap(ctl
, entry
);
2879 spin_unlock(&ctl
->tree_lock
);
2884 struct inode
*lookup_free_ino_inode(struct btrfs_root
*root
,
2885 struct btrfs_path
*path
)
2887 struct inode
*inode
= NULL
;
2889 spin_lock(&root
->cache_lock
);
2890 if (root
->cache_inode
)
2891 inode
= igrab(root
->cache_inode
);
2892 spin_unlock(&root
->cache_lock
);
2896 inode
= __lookup_free_space_inode(root
, path
, 0);
2900 spin_lock(&root
->cache_lock
);
2901 if (!btrfs_fs_closing(root
->fs_info
))
2902 root
->cache_inode
= igrab(inode
);
2903 spin_unlock(&root
->cache_lock
);
2908 int create_free_ino_inode(struct btrfs_root
*root
,
2909 struct btrfs_trans_handle
*trans
,
2910 struct btrfs_path
*path
)
2912 return __create_free_space_inode(root
, trans
, path
,
2913 BTRFS_FREE_INO_OBJECTID
, 0);
2916 int load_free_ino_cache(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
2918 struct btrfs_free_space_ctl
*ctl
= root
->free_ino_ctl
;
2919 struct btrfs_path
*path
;
2920 struct inode
*inode
;
2922 u64 root_gen
= btrfs_root_generation(&root
->root_item
);
2924 if (!btrfs_test_opt(root
, INODE_MAP_CACHE
))
2928 * If we're unmounting then just return, since this does a search on the
2929 * normal root and not the commit root and we could deadlock.
2931 if (btrfs_fs_closing(fs_info
))
2934 path
= btrfs_alloc_path();
2938 inode
= lookup_free_ino_inode(root
, path
);
2942 if (root_gen
!= BTRFS_I(inode
)->generation
)
2945 ret
= __load_free_space_cache(root
, inode
, ctl
, path
, 0);
2948 printk(KERN_ERR
"btrfs: failed to load free ino cache for "
2949 "root %llu\n", root
->root_key
.objectid
);
2953 btrfs_free_path(path
);
2957 int btrfs_write_out_ino_cache(struct btrfs_root
*root
,
2958 struct btrfs_trans_handle
*trans
,
2959 struct btrfs_path
*path
)
2961 struct btrfs_free_space_ctl
*ctl
= root
->free_ino_ctl
;
2962 struct inode
*inode
;
2965 if (!btrfs_test_opt(root
, INODE_MAP_CACHE
))
2968 inode
= lookup_free_ino_inode(root
, path
);
2972 ret
= __btrfs_write_out_cache(root
, inode
, ctl
, NULL
, trans
, path
, 0);
2974 btrfs_delalloc_release_metadata(inode
, inode
->i_size
);
2976 printk(KERN_ERR
"btrfs: failed to write free ino cache "
2977 "for root %llu\n", root
->root_key
.objectid
);