4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/bio.h>
13 #include <linux/mpage.h>
14 #include <linux/writeback.h>
15 #include <linux/blkdev.h>
16 #include <linux/f2fs_fs.h>
17 #include <linux/pagevec.h>
18 #include <linux/swap.h>
23 #include <trace/events/f2fs.h>
25 static struct kmem_cache
*orphan_entry_slab
;
26 static struct kmem_cache
*inode_entry_slab
;
29 * We guarantee no failure on the returned page.
31 struct page
*grab_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
33 struct address_space
*mapping
= META_MAPPING(sbi
);
34 struct page
*page
= NULL
;
36 page
= grab_cache_page(mapping
, index
);
41 f2fs_wait_on_page_writeback(page
, META
);
42 SetPageUptodate(page
);
47 * We guarantee no failure on the returned page.
49 struct page
*get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
51 struct address_space
*mapping
= META_MAPPING(sbi
);
54 page
= grab_cache_page(mapping
, index
);
59 if (PageUptodate(page
))
62 if (f2fs_submit_page_bio(sbi
, page
, index
,
63 READ_SYNC
| REQ_META
| REQ_PRIO
))
67 if (unlikely(page
->mapping
!= mapping
)) {
68 f2fs_put_page(page
, 1);
72 mark_page_accessed(page
);
76 static inline int get_max_meta_blks(struct f2fs_sb_info
*sbi
, int type
)
80 return NM_I(sbi
)->max_nid
/ NAT_ENTRY_PER_BLOCK
;
82 return SIT_BLK_CNT(sbi
);
92 * Readahead CP/NAT/SIT/SSA pages
94 int ra_meta_pages(struct f2fs_sb_info
*sbi
, int start
, int nrpages
, int type
)
96 block_t prev_blk_addr
= 0;
99 int max_blks
= get_max_meta_blks(sbi
, type
);
101 struct f2fs_io_info fio
= {
103 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
106 for (; nrpages
-- > 0; blkno
++) {
111 /* get nat block addr */
112 if (unlikely(blkno
>= max_blks
))
114 blk_addr
= current_nat_addr(sbi
,
115 blkno
* NAT_ENTRY_PER_BLOCK
);
118 /* get sit block addr */
119 if (unlikely(blkno
>= max_blks
))
121 blk_addr
= current_sit_addr(sbi
,
122 blkno
* SIT_ENTRY_PER_BLOCK
);
123 if (blkno
!= start
&& prev_blk_addr
+ 1 != blk_addr
)
125 prev_blk_addr
= blk_addr
;
129 /* get ssa/cp block addr */
136 page
= grab_cache_page(META_MAPPING(sbi
), blk_addr
);
139 if (PageUptodate(page
)) {
140 mark_page_accessed(page
);
141 f2fs_put_page(page
, 1);
145 f2fs_submit_page_mbio(sbi
, page
, blk_addr
, &fio
);
146 mark_page_accessed(page
);
147 f2fs_put_page(page
, 0);
150 f2fs_submit_merged_bio(sbi
, META
, READ
);
151 return blkno
- start
;
154 static int f2fs_write_meta_page(struct page
*page
,
155 struct writeback_control
*wbc
)
157 struct inode
*inode
= page
->mapping
->host
;
158 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
160 if (unlikely(sbi
->por_doing
))
162 if (wbc
->for_reclaim
)
165 /* Should not write any meta pages, if any IO error was occurred */
166 if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ERROR_FLAG
)))
169 f2fs_wait_on_page_writeback(page
, META
);
170 write_meta_page(sbi
, page
);
172 dec_page_count(sbi
, F2FS_DIRTY_META
);
177 redirty_page_for_writepage(wbc
, page
);
178 return AOP_WRITEPAGE_ACTIVATE
;
181 static int f2fs_write_meta_pages(struct address_space
*mapping
,
182 struct writeback_control
*wbc
)
184 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
187 /* collect a number of dirty meta pages and write together */
188 if (wbc
->for_kupdate
||
189 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
192 /* if mounting is failed, skip writing node pages */
193 mutex_lock(&sbi
->cp_mutex
);
194 diff
= nr_pages_to_write(sbi
, META
, wbc
);
195 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
196 mutex_unlock(&sbi
->cp_mutex
);
197 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
201 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
205 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
208 struct address_space
*mapping
= META_MAPPING(sbi
);
209 pgoff_t index
= 0, end
= LONG_MAX
;
212 struct writeback_control wbc
= {
216 pagevec_init(&pvec
, 0);
218 while (index
<= end
) {
220 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
222 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
223 if (unlikely(nr_pages
== 0))
226 for (i
= 0; i
< nr_pages
; i
++) {
227 struct page
*page
= pvec
.pages
[i
];
231 if (unlikely(page
->mapping
!= mapping
)) {
236 if (!PageDirty(page
)) {
237 /* someone wrote it for us */
238 goto continue_unlock
;
241 if (!clear_page_dirty_for_io(page
))
242 goto continue_unlock
;
244 if (f2fs_write_meta_page(page
, &wbc
)) {
249 if (unlikely(nwritten
>= nr_to_write
))
252 pagevec_release(&pvec
);
257 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
262 static int f2fs_set_meta_page_dirty(struct page
*page
)
264 struct address_space
*mapping
= page
->mapping
;
265 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
267 trace_f2fs_set_page_dirty(page
, META
);
269 SetPageUptodate(page
);
270 if (!PageDirty(page
)) {
271 __set_page_dirty_nobuffers(page
);
272 inc_page_count(sbi
, F2FS_DIRTY_META
);
278 const struct address_space_operations f2fs_meta_aops
= {
279 .writepage
= f2fs_write_meta_page
,
280 .writepages
= f2fs_write_meta_pages
,
281 .set_page_dirty
= f2fs_set_meta_page_dirty
,
284 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
288 spin_lock(&sbi
->orphan_inode_lock
);
289 if (unlikely(sbi
->n_orphans
>= sbi
->max_orphans
))
293 spin_unlock(&sbi
->orphan_inode_lock
);
298 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
300 spin_lock(&sbi
->orphan_inode_lock
);
301 f2fs_bug_on(sbi
->n_orphans
== 0);
303 spin_unlock(&sbi
->orphan_inode_lock
);
306 void add_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
308 struct list_head
*head
;
309 struct orphan_inode_entry
*new, *orphan
;
311 new = f2fs_kmem_cache_alloc(orphan_entry_slab
, GFP_ATOMIC
);
314 spin_lock(&sbi
->orphan_inode_lock
);
315 head
= &sbi
->orphan_inode_list
;
316 list_for_each_entry(orphan
, head
, list
) {
317 if (orphan
->ino
== ino
) {
318 spin_unlock(&sbi
->orphan_inode_lock
);
319 kmem_cache_free(orphan_entry_slab
, new);
323 if (orphan
->ino
> ino
)
327 /* add new orphan entry into list which is sorted by inode number */
328 list_add_tail(&new->list
, &orphan
->list
);
329 spin_unlock(&sbi
->orphan_inode_lock
);
332 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
334 struct list_head
*head
;
335 struct orphan_inode_entry
*orphan
;
337 spin_lock(&sbi
->orphan_inode_lock
);
338 head
= &sbi
->orphan_inode_list
;
339 list_for_each_entry(orphan
, head
, list
) {
340 if (orphan
->ino
== ino
) {
341 list_del(&orphan
->list
);
342 f2fs_bug_on(sbi
->n_orphans
== 0);
344 spin_unlock(&sbi
->orphan_inode_lock
);
345 kmem_cache_free(orphan_entry_slab
, orphan
);
349 spin_unlock(&sbi
->orphan_inode_lock
);
352 static void recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
354 struct inode
*inode
= f2fs_iget(sbi
->sb
, ino
);
355 f2fs_bug_on(IS_ERR(inode
));
358 /* truncate all the data during iput */
362 void recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
364 block_t start_blk
, orphan_blkaddr
, i
, j
;
366 if (!is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
))
369 sbi
->por_doing
= true;
370 start_blk
= __start_cp_addr(sbi
) + 1;
371 orphan_blkaddr
= __start_sum_addr(sbi
) - 1;
373 ra_meta_pages(sbi
, start_blk
, orphan_blkaddr
, META_CP
);
375 for (i
= 0; i
< orphan_blkaddr
; i
++) {
376 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
377 struct f2fs_orphan_block
*orphan_blk
;
379 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
380 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
381 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
382 recover_orphan_inode(sbi
, ino
);
384 f2fs_put_page(page
, 1);
386 /* clear Orphan Flag */
387 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
388 sbi
->por_doing
= false;
392 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
394 struct list_head
*head
;
395 struct f2fs_orphan_block
*orphan_blk
= NULL
;
396 unsigned int nentries
= 0;
397 unsigned short index
;
398 unsigned short orphan_blocks
= (unsigned short)((sbi
->n_orphans
+
399 (F2FS_ORPHANS_PER_BLOCK
- 1)) / F2FS_ORPHANS_PER_BLOCK
);
400 struct page
*page
= NULL
;
401 struct orphan_inode_entry
*orphan
= NULL
;
403 for (index
= 0; index
< orphan_blocks
; index
++)
404 grab_meta_page(sbi
, start_blk
+ index
);
407 spin_lock(&sbi
->orphan_inode_lock
);
408 head
= &sbi
->orphan_inode_list
;
410 /* loop for each orphan inode entry and write them in Jornal block */
411 list_for_each_entry(orphan
, head
, list
) {
413 page
= find_get_page(META_MAPPING(sbi
), start_blk
++);
416 (struct f2fs_orphan_block
*)page_address(page
);
417 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
418 f2fs_put_page(page
, 0);
421 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
423 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
425 * an orphan block is full of 1020 entries,
426 * then we need to flush current orphan blocks
427 * and bring another one in memory
429 orphan_blk
->blk_addr
= cpu_to_le16(index
);
430 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
431 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
432 set_page_dirty(page
);
433 f2fs_put_page(page
, 1);
441 orphan_blk
->blk_addr
= cpu_to_le16(index
);
442 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
443 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
444 set_page_dirty(page
);
445 f2fs_put_page(page
, 1);
448 spin_unlock(&sbi
->orphan_inode_lock
);
451 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
452 block_t cp_addr
, unsigned long long *version
)
454 struct page
*cp_page_1
, *cp_page_2
= NULL
;
455 unsigned long blk_size
= sbi
->blocksize
;
456 struct f2fs_checkpoint
*cp_block
;
457 unsigned long long cur_version
= 0, pre_version
= 0;
461 /* Read the 1st cp block in this CP pack */
462 cp_page_1
= get_meta_page(sbi
, cp_addr
);
464 /* get the version number */
465 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
466 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
467 if (crc_offset
>= blk_size
)
470 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
471 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
474 pre_version
= cur_cp_version(cp_block
);
476 /* Read the 2nd cp block in this CP pack */
477 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
478 cp_page_2
= get_meta_page(sbi
, cp_addr
);
480 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
481 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
482 if (crc_offset
>= blk_size
)
485 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
486 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
489 cur_version
= cur_cp_version(cp_block
);
491 if (cur_version
== pre_version
) {
492 *version
= cur_version
;
493 f2fs_put_page(cp_page_2
, 1);
497 f2fs_put_page(cp_page_2
, 1);
499 f2fs_put_page(cp_page_1
, 1);
503 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
505 struct f2fs_checkpoint
*cp_block
;
506 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
507 struct page
*cp1
, *cp2
, *cur_page
;
508 unsigned long blk_size
= sbi
->blocksize
;
509 unsigned long long cp1_version
= 0, cp2_version
= 0;
510 unsigned long long cp_start_blk_no
;
512 sbi
->ckpt
= kzalloc(blk_size
, GFP_KERNEL
);
516 * Finding out valid cp block involves read both
517 * sets( cp pack1 and cp pack 2)
519 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
520 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
522 /* The second checkpoint pack should start at the next segment */
523 cp_start_blk_no
+= ((unsigned long long)1) <<
524 le32_to_cpu(fsb
->log_blocks_per_seg
);
525 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
528 if (ver_after(cp2_version
, cp1_version
))
540 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
541 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
543 f2fs_put_page(cp1
, 1);
544 f2fs_put_page(cp2
, 1);
552 static int __add_dirty_inode(struct inode
*inode
, struct dir_inode_entry
*new)
554 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
556 if (is_inode_flag_set(F2FS_I(inode
), FI_DIRTY_DIR
))
559 set_inode_flag(F2FS_I(inode
), FI_DIRTY_DIR
);
560 F2FS_I(inode
)->dirty_dir
= new;
561 list_add_tail(&new->list
, &sbi
->dir_inode_list
);
562 stat_inc_dirty_dir(sbi
);
566 void set_dirty_dir_page(struct inode
*inode
, struct page
*page
)
568 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
569 struct dir_inode_entry
*new;
572 if (!S_ISDIR(inode
->i_mode
))
575 new = f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
577 INIT_LIST_HEAD(&new->list
);
579 spin_lock(&sbi
->dir_inode_lock
);
580 ret
= __add_dirty_inode(inode
, new);
581 inode_inc_dirty_dents(inode
);
582 SetPagePrivate(page
);
583 spin_unlock(&sbi
->dir_inode_lock
);
586 kmem_cache_free(inode_entry_slab
, new);
589 void add_dirty_dir_inode(struct inode
*inode
)
591 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
592 struct dir_inode_entry
*new =
593 f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
597 INIT_LIST_HEAD(&new->list
);
599 spin_lock(&sbi
->dir_inode_lock
);
600 ret
= __add_dirty_inode(inode
, new);
601 spin_unlock(&sbi
->dir_inode_lock
);
604 kmem_cache_free(inode_entry_slab
, new);
607 void remove_dirty_dir_inode(struct inode
*inode
)
609 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
610 struct dir_inode_entry
*entry
;
612 if (!S_ISDIR(inode
->i_mode
))
615 spin_lock(&sbi
->dir_inode_lock
);
616 if (get_dirty_dents(inode
) ||
617 !is_inode_flag_set(F2FS_I(inode
), FI_DIRTY_DIR
)) {
618 spin_unlock(&sbi
->dir_inode_lock
);
622 entry
= F2FS_I(inode
)->dirty_dir
;
623 list_del(&entry
->list
);
624 F2FS_I(inode
)->dirty_dir
= NULL
;
625 clear_inode_flag(F2FS_I(inode
), FI_DIRTY_DIR
);
626 stat_dec_dirty_dir(sbi
);
627 spin_unlock(&sbi
->dir_inode_lock
);
628 kmem_cache_free(inode_entry_slab
, entry
);
630 /* Only from the recovery routine */
631 if (is_inode_flag_set(F2FS_I(inode
), FI_DELAY_IPUT
)) {
632 clear_inode_flag(F2FS_I(inode
), FI_DELAY_IPUT
);
637 void sync_dirty_dir_inodes(struct f2fs_sb_info
*sbi
)
639 struct list_head
*head
;
640 struct dir_inode_entry
*entry
;
643 spin_lock(&sbi
->dir_inode_lock
);
645 head
= &sbi
->dir_inode_list
;
646 if (list_empty(head
)) {
647 spin_unlock(&sbi
->dir_inode_lock
);
650 entry
= list_entry(head
->next
, struct dir_inode_entry
, list
);
651 inode
= igrab(entry
->inode
);
652 spin_unlock(&sbi
->dir_inode_lock
);
654 filemap_fdatawrite(inode
->i_mapping
);
658 * We should submit bio, since it exists several
659 * wribacking dentry pages in the freeing inode.
661 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
667 * Freeze all the FS-operations for checkpoint.
669 static void block_operations(struct f2fs_sb_info
*sbi
)
671 struct writeback_control wbc
= {
672 .sync_mode
= WB_SYNC_ALL
,
673 .nr_to_write
= LONG_MAX
,
676 struct blk_plug plug
;
678 blk_start_plug(&plug
);
682 /* write all the dirty dentry pages */
683 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
684 f2fs_unlock_all(sbi
);
685 sync_dirty_dir_inodes(sbi
);
686 goto retry_flush_dents
;
690 * POR: we should ensure that there is no dirty node pages
691 * until finishing nat/sit flush.
694 mutex_lock(&sbi
->node_write
);
696 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
697 mutex_unlock(&sbi
->node_write
);
698 sync_node_pages(sbi
, 0, &wbc
);
699 goto retry_flush_nodes
;
701 blk_finish_plug(&plug
);
704 static void unblock_operations(struct f2fs_sb_info
*sbi
)
706 mutex_unlock(&sbi
->node_write
);
707 f2fs_unlock_all(sbi
);
710 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
715 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
717 if (!get_pages(sbi
, F2FS_WRITEBACK
))
722 finish_wait(&sbi
->cp_wait
, &wait
);
725 static void do_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
727 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
730 struct page
*cp_page
;
731 unsigned int data_sum_blocks
, orphan_blocks
;
737 * This avoids to conduct wrong roll-forward operations and uses
738 * metapages, so should be called prior to sync_meta_pages below.
740 discard_next_dnode(sbi
);
742 /* Flush all the NAT/SIT pages */
743 while (get_pages(sbi
, F2FS_DIRTY_META
))
744 sync_meta_pages(sbi
, META
, LONG_MAX
);
746 next_free_nid(sbi
, &last_nid
);
750 * version number is already updated
752 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
753 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
754 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
755 for (i
= 0; i
< 3; i
++) {
756 ckpt
->cur_node_segno
[i
] =
757 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
758 ckpt
->cur_node_blkoff
[i
] =
759 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
760 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
761 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
763 for (i
= 0; i
< 3; i
++) {
764 ckpt
->cur_data_segno
[i
] =
765 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
766 ckpt
->cur_data_blkoff
[i
] =
767 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
768 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
769 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
772 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
773 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
774 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
776 /* 2 cp + n data seg summary + orphan inode blocks */
777 data_sum_blocks
= npages_for_summary_flush(sbi
);
778 if (data_sum_blocks
< 3)
779 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
781 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
783 orphan_blocks
= (sbi
->n_orphans
+ F2FS_ORPHANS_PER_BLOCK
- 1)
784 / F2FS_ORPHANS_PER_BLOCK
;
785 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + orphan_blocks
);
788 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
789 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
790 data_sum_blocks
+ orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
792 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
793 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
794 data_sum_blocks
+ orphan_blocks
);
798 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
800 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
802 /* update SIT/NAT bitmap */
803 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
804 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
806 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
807 *((__le32
*)((unsigned char *)ckpt
+
808 le32_to_cpu(ckpt
->checksum_offset
)))
809 = cpu_to_le32(crc32
);
811 start_blk
= __start_cp_addr(sbi
);
813 /* write out checkpoint buffer at block 0 */
814 cp_page
= grab_meta_page(sbi
, start_blk
++);
815 kaddr
= page_address(cp_page
);
816 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
817 set_page_dirty(cp_page
);
818 f2fs_put_page(cp_page
, 1);
820 if (sbi
->n_orphans
) {
821 write_orphan_inodes(sbi
, start_blk
);
822 start_blk
+= orphan_blocks
;
825 write_data_summaries(sbi
, start_blk
);
826 start_blk
+= data_sum_blocks
;
828 write_node_summaries(sbi
, start_blk
);
829 start_blk
+= NR_CURSEG_NODE_TYPE
;
832 /* writeout checkpoint block */
833 cp_page
= grab_meta_page(sbi
, start_blk
);
834 kaddr
= page_address(cp_page
);
835 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
836 set_page_dirty(cp_page
);
837 f2fs_put_page(cp_page
, 1);
839 /* wait for previous submitted node/meta pages writeback */
840 wait_on_all_pages_writeback(sbi
);
842 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
843 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
845 /* update user_block_counts */
846 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
847 sbi
->alloc_valid_block_count
= 0;
849 /* Here, we only have one bio having CP pack */
850 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
852 if (unlikely(!is_set_ckpt_flags(ckpt
, CP_ERROR_FLAG
))) {
853 clear_prefree_segments(sbi
);
854 F2FS_RESET_SB_DIRT(sbi
);
859 * We guarantee that this checkpoint procedure should not fail.
861 void write_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
863 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
864 unsigned long long ckpt_ver
;
866 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "start block_ops");
868 mutex_lock(&sbi
->cp_mutex
);
869 block_operations(sbi
);
871 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish block_ops");
873 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
874 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
875 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
878 * update checkpoint pack index
879 * Increase the version number so that
880 * SIT entries and seg summaries are written at correct place
882 ckpt_ver
= cur_cp_version(ckpt
);
883 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
885 /* write cached NAT/SIT entries to NAT/SIT area */
886 flush_nat_entries(sbi
);
887 flush_sit_entries(sbi
);
889 /* unlock all the fs_lock[] in do_checkpoint() */
890 do_checkpoint(sbi
, is_umount
);
892 unblock_operations(sbi
);
893 mutex_unlock(&sbi
->cp_mutex
);
895 stat_inc_cp_count(sbi
->stat_info
);
896 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish checkpoint");
899 void init_orphan_info(struct f2fs_sb_info
*sbi
)
901 spin_lock_init(&sbi
->orphan_inode_lock
);
902 INIT_LIST_HEAD(&sbi
->orphan_inode_list
);
905 * considering 512 blocks in a segment 8 blocks are needed for cp
906 * and log segment summaries. Remaining blocks are used to keep
907 * orphan entries with the limitation one reserved segment
908 * for cp pack we can have max 1020*504 orphan entries
910 sbi
->max_orphans
= (sbi
->blocks_per_seg
- 2 - NR_CURSEG_TYPE
)
911 * F2FS_ORPHANS_PER_BLOCK
;
914 int __init
create_checkpoint_caches(void)
916 orphan_entry_slab
= f2fs_kmem_cache_create("f2fs_orphan_entry",
917 sizeof(struct orphan_inode_entry
));
918 if (!orphan_entry_slab
)
920 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
921 sizeof(struct dir_inode_entry
));
922 if (!inode_entry_slab
) {
923 kmem_cache_destroy(orphan_entry_slab
);
929 void destroy_checkpoint_caches(void)
931 kmem_cache_destroy(orphan_entry_slab
);
932 kmem_cache_destroy(inode_entry_slab
);