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_write_begin(mapping
, index
, AOP_FLAG_NOFS
);
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 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 dec_page_count(sbi
, F2FS_DIRTY_META
);
178 wbc
->pages_skipped
++;
179 account_page_redirty(page
);
180 set_page_dirty(page
);
181 return AOP_WRITEPAGE_ACTIVATE
;
184 static int f2fs_write_meta_pages(struct address_space
*mapping
,
185 struct writeback_control
*wbc
)
187 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
190 /* collect a number of dirty meta pages and write together */
191 if (wbc
->for_kupdate
||
192 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
195 /* if mounting is failed, skip writing node pages */
196 mutex_lock(&sbi
->cp_mutex
);
197 diff
= nr_pages_to_write(sbi
, META
, wbc
);
198 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
199 mutex_unlock(&sbi
->cp_mutex
);
200 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
204 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
208 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
211 struct address_space
*mapping
= META_MAPPING(sbi
);
212 pgoff_t index
= 0, end
= LONG_MAX
;
215 struct writeback_control wbc
= {
219 pagevec_init(&pvec
, 0);
221 while (index
<= end
) {
223 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
225 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
226 if (unlikely(nr_pages
== 0))
229 for (i
= 0; i
< nr_pages
; i
++) {
230 struct page
*page
= pvec
.pages
[i
];
234 if (unlikely(page
->mapping
!= mapping
)) {
239 if (!PageDirty(page
)) {
240 /* someone wrote it for us */
241 goto continue_unlock
;
244 if (!clear_page_dirty_for_io(page
))
245 goto continue_unlock
;
247 if (f2fs_write_meta_page(page
, &wbc
)) {
252 if (unlikely(nwritten
>= nr_to_write
))
255 pagevec_release(&pvec
);
260 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
265 static int f2fs_set_meta_page_dirty(struct page
*page
)
267 struct address_space
*mapping
= page
->mapping
;
268 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
270 trace_f2fs_set_page_dirty(page
, META
);
272 SetPageUptodate(page
);
273 if (!PageDirty(page
)) {
274 __set_page_dirty_nobuffers(page
);
275 inc_page_count(sbi
, F2FS_DIRTY_META
);
281 const struct address_space_operations f2fs_meta_aops
= {
282 .writepage
= f2fs_write_meta_page
,
283 .writepages
= f2fs_write_meta_pages
,
284 .set_page_dirty
= f2fs_set_meta_page_dirty
,
287 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
291 spin_lock(&sbi
->orphan_inode_lock
);
292 if (unlikely(sbi
->n_orphans
>= sbi
->max_orphans
))
296 spin_unlock(&sbi
->orphan_inode_lock
);
301 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
303 spin_lock(&sbi
->orphan_inode_lock
);
304 f2fs_bug_on(sbi
->n_orphans
== 0);
306 spin_unlock(&sbi
->orphan_inode_lock
);
309 void add_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
311 struct list_head
*head
;
312 struct orphan_inode_entry
*new, *orphan
;
314 new = f2fs_kmem_cache_alloc(orphan_entry_slab
, GFP_ATOMIC
);
317 spin_lock(&sbi
->orphan_inode_lock
);
318 head
= &sbi
->orphan_inode_list
;
319 list_for_each_entry(orphan
, head
, list
) {
320 if (orphan
->ino
== ino
) {
321 spin_unlock(&sbi
->orphan_inode_lock
);
322 kmem_cache_free(orphan_entry_slab
, new);
326 if (orphan
->ino
> ino
)
330 /* add new orphan entry into list which is sorted by inode number */
331 list_add_tail(&new->list
, &orphan
->list
);
332 spin_unlock(&sbi
->orphan_inode_lock
);
335 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
337 struct list_head
*head
;
338 struct orphan_inode_entry
*orphan
;
340 spin_lock(&sbi
->orphan_inode_lock
);
341 head
= &sbi
->orphan_inode_list
;
342 list_for_each_entry(orphan
, head
, list
) {
343 if (orphan
->ino
== ino
) {
344 list_del(&orphan
->list
);
345 f2fs_bug_on(sbi
->n_orphans
== 0);
347 spin_unlock(&sbi
->orphan_inode_lock
);
348 kmem_cache_free(orphan_entry_slab
, orphan
);
352 spin_unlock(&sbi
->orphan_inode_lock
);
355 static void recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
357 struct inode
*inode
= f2fs_iget(sbi
->sb
, ino
);
358 f2fs_bug_on(IS_ERR(inode
));
361 /* truncate all the data during iput */
365 void recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
367 block_t start_blk
, orphan_blkaddr
, i
, j
;
369 if (!is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
))
372 sbi
->por_doing
= true;
373 start_blk
= __start_cp_addr(sbi
) + 1;
374 orphan_blkaddr
= __start_sum_addr(sbi
) - 1;
376 ra_meta_pages(sbi
, start_blk
, orphan_blkaddr
, META_CP
);
378 for (i
= 0; i
< orphan_blkaddr
; i
++) {
379 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
380 struct f2fs_orphan_block
*orphan_blk
;
382 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
383 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
384 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
385 recover_orphan_inode(sbi
, ino
);
387 f2fs_put_page(page
, 1);
389 /* clear Orphan Flag */
390 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
391 sbi
->por_doing
= false;
395 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
397 struct list_head
*head
;
398 struct f2fs_orphan_block
*orphan_blk
= NULL
;
399 unsigned int nentries
= 0;
400 unsigned short index
;
401 unsigned short orphan_blocks
= (unsigned short)((sbi
->n_orphans
+
402 (F2FS_ORPHANS_PER_BLOCK
- 1)) / F2FS_ORPHANS_PER_BLOCK
);
403 struct page
*page
= NULL
;
404 struct orphan_inode_entry
*orphan
= NULL
;
406 for (index
= 0; index
< orphan_blocks
; index
++)
407 grab_meta_page(sbi
, start_blk
+ index
);
410 spin_lock(&sbi
->orphan_inode_lock
);
411 head
= &sbi
->orphan_inode_list
;
413 /* loop for each orphan inode entry and write them in Jornal block */
414 list_for_each_entry(orphan
, head
, list
) {
416 page
= find_get_page(META_MAPPING(sbi
), start_blk
++);
419 (struct f2fs_orphan_block
*)page_address(page
);
420 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
421 f2fs_put_page(page
, 0);
424 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
426 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
428 * an orphan block is full of 1020 entries,
429 * then we need to flush current orphan blocks
430 * and bring another one in memory
432 orphan_blk
->blk_addr
= cpu_to_le16(index
);
433 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
434 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
435 set_page_dirty(page
);
436 f2fs_put_page(page
, 1);
444 orphan_blk
->blk_addr
= cpu_to_le16(index
);
445 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
446 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
447 set_page_dirty(page
);
448 f2fs_put_page(page
, 1);
451 spin_unlock(&sbi
->orphan_inode_lock
);
454 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
455 block_t cp_addr
, unsigned long long *version
)
457 struct page
*cp_page_1
, *cp_page_2
= NULL
;
458 unsigned long blk_size
= sbi
->blocksize
;
459 struct f2fs_checkpoint
*cp_block
;
460 unsigned long long cur_version
= 0, pre_version
= 0;
464 /* Read the 1st cp block in this CP pack */
465 cp_page_1
= get_meta_page(sbi
, cp_addr
);
467 /* get the version number */
468 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
469 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
470 if (crc_offset
>= blk_size
)
473 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
474 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
477 pre_version
= cur_cp_version(cp_block
);
479 /* Read the 2nd cp block in this CP pack */
480 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
481 cp_page_2
= get_meta_page(sbi
, cp_addr
);
483 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
484 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
485 if (crc_offset
>= blk_size
)
488 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
489 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
492 cur_version
= cur_cp_version(cp_block
);
494 if (cur_version
== pre_version
) {
495 *version
= cur_version
;
496 f2fs_put_page(cp_page_2
, 1);
500 f2fs_put_page(cp_page_2
, 1);
502 f2fs_put_page(cp_page_1
, 1);
506 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
508 struct f2fs_checkpoint
*cp_block
;
509 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
510 struct page
*cp1
, *cp2
, *cur_page
;
511 unsigned long blk_size
= sbi
->blocksize
;
512 unsigned long long cp1_version
= 0, cp2_version
= 0;
513 unsigned long long cp_start_blk_no
;
515 sbi
->ckpt
= kzalloc(blk_size
, GFP_KERNEL
);
519 * Finding out valid cp block involves read both
520 * sets( cp pack1 and cp pack 2)
522 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
523 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
525 /* The second checkpoint pack should start at the next segment */
526 cp_start_blk_no
+= ((unsigned long long)1) <<
527 le32_to_cpu(fsb
->log_blocks_per_seg
);
528 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
531 if (ver_after(cp2_version
, cp1_version
))
543 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
544 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
546 f2fs_put_page(cp1
, 1);
547 f2fs_put_page(cp2
, 1);
555 static int __add_dirty_inode(struct inode
*inode
, struct dir_inode_entry
*new)
557 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
558 struct list_head
*head
= &sbi
->dir_inode_list
;
559 struct dir_inode_entry
*entry
;
561 list_for_each_entry(entry
, head
, list
)
562 if (unlikely(entry
->inode
== inode
))
565 list_add_tail(&new->list
, head
);
566 stat_inc_dirty_dir(sbi
);
570 void set_dirty_dir_page(struct inode
*inode
, struct page
*page
)
572 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
573 struct dir_inode_entry
*new;
576 if (!S_ISDIR(inode
->i_mode
))
579 new = f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
581 INIT_LIST_HEAD(&new->list
);
583 spin_lock(&sbi
->dir_inode_lock
);
584 ret
= __add_dirty_inode(inode
, new);
585 inode_inc_dirty_dents(inode
);
586 SetPagePrivate(page
);
587 spin_unlock(&sbi
->dir_inode_lock
);
590 kmem_cache_free(inode_entry_slab
, new);
593 void add_dirty_dir_inode(struct inode
*inode
)
595 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
596 struct dir_inode_entry
*new =
597 f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
601 INIT_LIST_HEAD(&new->list
);
603 spin_lock(&sbi
->dir_inode_lock
);
604 ret
= __add_dirty_inode(inode
, new);
605 spin_unlock(&sbi
->dir_inode_lock
);
608 kmem_cache_free(inode_entry_slab
, new);
611 void remove_dirty_dir_inode(struct inode
*inode
)
613 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
614 struct list_head
*head
;
615 struct dir_inode_entry
*entry
;
617 if (!S_ISDIR(inode
->i_mode
))
620 spin_lock(&sbi
->dir_inode_lock
);
621 if (get_dirty_dents(inode
)) {
622 spin_unlock(&sbi
->dir_inode_lock
);
626 head
= &sbi
->dir_inode_list
;
627 list_for_each_entry(entry
, head
, list
) {
628 if (entry
->inode
== inode
) {
629 list_del(&entry
->list
);
630 stat_dec_dirty_dir(sbi
);
631 spin_unlock(&sbi
->dir_inode_lock
);
632 kmem_cache_free(inode_entry_slab
, entry
);
636 spin_unlock(&sbi
->dir_inode_lock
);
639 /* Only from the recovery routine */
640 if (is_inode_flag_set(F2FS_I(inode
), FI_DELAY_IPUT
)) {
641 clear_inode_flag(F2FS_I(inode
), FI_DELAY_IPUT
);
646 struct inode
*check_dirty_dir_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
649 struct list_head
*head
;
650 struct inode
*inode
= NULL
;
651 struct dir_inode_entry
*entry
;
653 spin_lock(&sbi
->dir_inode_lock
);
655 head
= &sbi
->dir_inode_list
;
656 list_for_each_entry(entry
, head
, list
) {
657 if (entry
->inode
->i_ino
== ino
) {
658 inode
= entry
->inode
;
662 spin_unlock(&sbi
->dir_inode_lock
);
666 void sync_dirty_dir_inodes(struct f2fs_sb_info
*sbi
)
668 struct list_head
*head
;
669 struct dir_inode_entry
*entry
;
672 spin_lock(&sbi
->dir_inode_lock
);
674 head
= &sbi
->dir_inode_list
;
675 if (list_empty(head
)) {
676 spin_unlock(&sbi
->dir_inode_lock
);
679 entry
= list_entry(head
->next
, struct dir_inode_entry
, list
);
680 inode
= igrab(entry
->inode
);
681 spin_unlock(&sbi
->dir_inode_lock
);
683 filemap_fdatawrite(inode
->i_mapping
);
687 * We should submit bio, since it exists several
688 * wribacking dentry pages in the freeing inode.
690 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
696 * Freeze all the FS-operations for checkpoint.
698 static void block_operations(struct f2fs_sb_info
*sbi
)
700 struct writeback_control wbc
= {
701 .sync_mode
= WB_SYNC_ALL
,
702 .nr_to_write
= LONG_MAX
,
705 struct blk_plug plug
;
707 blk_start_plug(&plug
);
711 /* write all the dirty dentry pages */
712 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
713 f2fs_unlock_all(sbi
);
714 sync_dirty_dir_inodes(sbi
);
715 goto retry_flush_dents
;
719 * POR: we should ensure that there is no dirty node pages
720 * until finishing nat/sit flush.
723 mutex_lock(&sbi
->node_write
);
725 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
726 mutex_unlock(&sbi
->node_write
);
727 sync_node_pages(sbi
, 0, &wbc
);
728 goto retry_flush_nodes
;
730 blk_finish_plug(&plug
);
733 static void unblock_operations(struct f2fs_sb_info
*sbi
)
735 mutex_unlock(&sbi
->node_write
);
736 f2fs_unlock_all(sbi
);
739 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
744 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
746 if (!get_pages(sbi
, F2FS_WRITEBACK
))
751 finish_wait(&sbi
->cp_wait
, &wait
);
754 static void do_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
756 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
759 struct page
*cp_page
;
760 unsigned int data_sum_blocks
, orphan_blocks
;
765 /* Flush all the NAT/SIT pages */
766 while (get_pages(sbi
, F2FS_DIRTY_META
))
767 sync_meta_pages(sbi
, META
, LONG_MAX
);
769 next_free_nid(sbi
, &last_nid
);
773 * version number is already updated
775 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
776 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
777 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
778 for (i
= 0; i
< 3; i
++) {
779 ckpt
->cur_node_segno
[i
] =
780 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
781 ckpt
->cur_node_blkoff
[i
] =
782 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
783 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
784 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
786 for (i
= 0; i
< 3; i
++) {
787 ckpt
->cur_data_segno
[i
] =
788 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
789 ckpt
->cur_data_blkoff
[i
] =
790 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
791 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
792 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
795 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
796 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
797 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
799 /* 2 cp + n data seg summary + orphan inode blocks */
800 data_sum_blocks
= npages_for_summary_flush(sbi
);
801 if (data_sum_blocks
< 3)
802 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
804 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
806 orphan_blocks
= (sbi
->n_orphans
+ F2FS_ORPHANS_PER_BLOCK
- 1)
807 / F2FS_ORPHANS_PER_BLOCK
;
808 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + orphan_blocks
);
811 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
812 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
813 data_sum_blocks
+ orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
815 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
816 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
817 data_sum_blocks
+ orphan_blocks
);
821 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
823 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
825 /* update SIT/NAT bitmap */
826 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
827 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
829 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
830 *((__le32
*)((unsigned char *)ckpt
+
831 le32_to_cpu(ckpt
->checksum_offset
)))
832 = cpu_to_le32(crc32
);
834 start_blk
= __start_cp_addr(sbi
);
836 /* write out checkpoint buffer at block 0 */
837 cp_page
= grab_meta_page(sbi
, start_blk
++);
838 kaddr
= page_address(cp_page
);
839 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
840 set_page_dirty(cp_page
);
841 f2fs_put_page(cp_page
, 1);
843 if (sbi
->n_orphans
) {
844 write_orphan_inodes(sbi
, start_blk
);
845 start_blk
+= orphan_blocks
;
848 write_data_summaries(sbi
, start_blk
);
849 start_blk
+= data_sum_blocks
;
851 write_node_summaries(sbi
, start_blk
);
852 start_blk
+= NR_CURSEG_NODE_TYPE
;
855 /* writeout checkpoint block */
856 cp_page
= grab_meta_page(sbi
, start_blk
);
857 kaddr
= page_address(cp_page
);
858 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
859 set_page_dirty(cp_page
);
860 f2fs_put_page(cp_page
, 1);
862 /* wait for previous submitted node/meta pages writeback */
863 wait_on_all_pages_writeback(sbi
);
865 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
866 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
868 /* update user_block_counts */
869 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
870 sbi
->alloc_valid_block_count
= 0;
872 /* Here, we only have one bio having CP pack */
873 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
875 if (unlikely(!is_set_ckpt_flags(ckpt
, CP_ERROR_FLAG
))) {
876 clear_prefree_segments(sbi
);
877 F2FS_RESET_SB_DIRT(sbi
);
882 * We guarantee that this checkpoint procedure should not fail.
884 void write_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
886 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
887 unsigned long long ckpt_ver
;
889 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "start block_ops");
891 mutex_lock(&sbi
->cp_mutex
);
892 block_operations(sbi
);
894 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish block_ops");
896 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
897 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
898 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
901 * update checkpoint pack index
902 * Increase the version number so that
903 * SIT entries and seg summaries are written at correct place
905 ckpt_ver
= cur_cp_version(ckpt
);
906 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
908 /* write cached NAT/SIT entries to NAT/SIT area */
909 flush_nat_entries(sbi
);
910 flush_sit_entries(sbi
);
912 /* unlock all the fs_lock[] in do_checkpoint() */
913 do_checkpoint(sbi
, is_umount
);
915 unblock_operations(sbi
);
916 mutex_unlock(&sbi
->cp_mutex
);
918 stat_inc_cp_count(sbi
->stat_info
);
919 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish checkpoint");
922 void init_orphan_info(struct f2fs_sb_info
*sbi
)
924 spin_lock_init(&sbi
->orphan_inode_lock
);
925 INIT_LIST_HEAD(&sbi
->orphan_inode_list
);
928 * considering 512 blocks in a segment 8 blocks are needed for cp
929 * and log segment summaries. Remaining blocks are used to keep
930 * orphan entries with the limitation one reserved segment
931 * for cp pack we can have max 1020*504 orphan entries
933 sbi
->max_orphans
= (sbi
->blocks_per_seg
- 2 - NR_CURSEG_TYPE
)
934 * F2FS_ORPHANS_PER_BLOCK
;
937 int __init
create_checkpoint_caches(void)
939 orphan_entry_slab
= f2fs_kmem_cache_create("f2fs_orphan_entry",
940 sizeof(struct orphan_inode_entry
));
941 if (!orphan_entry_slab
)
943 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
944 sizeof(struct dir_inode_entry
));
945 if (!inode_entry_slab
) {
946 kmem_cache_destroy(orphan_entry_slab
);
952 void destroy_checkpoint_caches(void)
954 kmem_cache_destroy(orphan_entry_slab
);
955 kmem_cache_destroy(inode_entry_slab
);