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);
75 inline int get_max_meta_blks(struct f2fs_sb_info
*sbi
, int type
)
79 return NM_I(sbi
)->max_nid
/ NAT_ENTRY_PER_BLOCK
;
81 return SIT_BLK_CNT(sbi
);
91 * Readahead CP/NAT/SIT/SSA pages
93 int ra_meta_pages(struct f2fs_sb_info
*sbi
, int start
, int nrpages
, int type
)
95 block_t prev_blk_addr
= 0;
98 int max_blks
= get_max_meta_blks(sbi
, type
);
100 struct f2fs_io_info fio
= {
102 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
105 for (; nrpages
-- > 0; blkno
++) {
110 /* get nat block addr */
111 if (unlikely(blkno
>= max_blks
))
113 blk_addr
= current_nat_addr(sbi
,
114 blkno
* NAT_ENTRY_PER_BLOCK
);
117 /* get sit block addr */
118 if (unlikely(blkno
>= max_blks
))
120 blk_addr
= current_sit_addr(sbi
,
121 blkno
* SIT_ENTRY_PER_BLOCK
);
122 if (blkno
!= start
&& prev_blk_addr
+ 1 != blk_addr
)
124 prev_blk_addr
= blk_addr
;
128 /* get ssa/cp block addr */
135 page
= grab_cache_page(META_MAPPING(sbi
), blk_addr
);
138 if (PageUptodate(page
)) {
139 f2fs_put_page(page
, 1);
143 f2fs_submit_page_mbio(sbi
, page
, blk_addr
, &fio
);
144 f2fs_put_page(page
, 0);
147 f2fs_submit_merged_bio(sbi
, META
, READ
);
148 return blkno
- start
;
151 static int f2fs_write_meta_page(struct page
*page
,
152 struct writeback_control
*wbc
)
154 struct inode
*inode
= page
->mapping
->host
;
155 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
157 if (unlikely(sbi
->por_doing
))
159 if (wbc
->for_reclaim
)
162 /* Should not write any meta pages, if any IO error was occurred */
163 if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ERROR_FLAG
)))
166 f2fs_wait_on_page_writeback(page
, META
);
167 write_meta_page(sbi
, page
);
169 dec_page_count(sbi
, F2FS_DIRTY_META
);
174 dec_page_count(sbi
, F2FS_DIRTY_META
);
175 wbc
->pages_skipped
++;
176 account_page_redirty(page
);
177 set_page_dirty(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
);
555 struct list_head
*head
= &sbi
->dir_inode_list
;
556 struct dir_inode_entry
*entry
;
558 list_for_each_entry(entry
, head
, list
)
559 if (unlikely(entry
->inode
== inode
))
562 list_add_tail(&new->list
, head
);
563 stat_inc_dirty_dir(sbi
);
567 void set_dirty_dir_page(struct inode
*inode
, struct page
*page
)
569 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
570 struct dir_inode_entry
*new;
573 if (!S_ISDIR(inode
->i_mode
))
576 new = f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
578 INIT_LIST_HEAD(&new->list
);
580 spin_lock(&sbi
->dir_inode_lock
);
581 ret
= __add_dirty_inode(inode
, new);
582 inode_inc_dirty_dents(inode
);
583 SetPagePrivate(page
);
584 spin_unlock(&sbi
->dir_inode_lock
);
587 kmem_cache_free(inode_entry_slab
, new);
590 void add_dirty_dir_inode(struct inode
*inode
)
592 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
593 struct dir_inode_entry
*new =
594 f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
598 INIT_LIST_HEAD(&new->list
);
600 spin_lock(&sbi
->dir_inode_lock
);
601 ret
= __add_dirty_inode(inode
, new);
602 spin_unlock(&sbi
->dir_inode_lock
);
605 kmem_cache_free(inode_entry_slab
, new);
608 void remove_dirty_dir_inode(struct inode
*inode
)
610 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
611 struct list_head
*head
;
612 struct dir_inode_entry
*entry
;
614 if (!S_ISDIR(inode
->i_mode
))
617 spin_lock(&sbi
->dir_inode_lock
);
618 if (get_dirty_dents(inode
)) {
619 spin_unlock(&sbi
->dir_inode_lock
);
623 head
= &sbi
->dir_inode_list
;
624 list_for_each_entry(entry
, head
, list
) {
625 if (entry
->inode
== inode
) {
626 list_del(&entry
->list
);
627 stat_dec_dirty_dir(sbi
);
628 spin_unlock(&sbi
->dir_inode_lock
);
629 kmem_cache_free(inode_entry_slab
, entry
);
633 spin_unlock(&sbi
->dir_inode_lock
);
636 /* Only from the recovery routine */
637 if (is_inode_flag_set(F2FS_I(inode
), FI_DELAY_IPUT
)) {
638 clear_inode_flag(F2FS_I(inode
), FI_DELAY_IPUT
);
643 struct inode
*check_dirty_dir_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
646 struct list_head
*head
;
647 struct inode
*inode
= NULL
;
648 struct dir_inode_entry
*entry
;
650 spin_lock(&sbi
->dir_inode_lock
);
652 head
= &sbi
->dir_inode_list
;
653 list_for_each_entry(entry
, head
, list
) {
654 if (entry
->inode
->i_ino
== ino
) {
655 inode
= entry
->inode
;
659 spin_unlock(&sbi
->dir_inode_lock
);
663 void sync_dirty_dir_inodes(struct f2fs_sb_info
*sbi
)
665 struct list_head
*head
;
666 struct dir_inode_entry
*entry
;
669 spin_lock(&sbi
->dir_inode_lock
);
671 head
= &sbi
->dir_inode_list
;
672 if (list_empty(head
)) {
673 spin_unlock(&sbi
->dir_inode_lock
);
676 entry
= list_entry(head
->next
, struct dir_inode_entry
, list
);
677 inode
= igrab(entry
->inode
);
678 spin_unlock(&sbi
->dir_inode_lock
);
680 filemap_fdatawrite(inode
->i_mapping
);
684 * We should submit bio, since it exists several
685 * wribacking dentry pages in the freeing inode.
687 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
693 * Freeze all the FS-operations for checkpoint.
695 static void block_operations(struct f2fs_sb_info
*sbi
)
697 struct writeback_control wbc
= {
698 .sync_mode
= WB_SYNC_ALL
,
699 .nr_to_write
= LONG_MAX
,
702 struct blk_plug plug
;
704 blk_start_plug(&plug
);
708 /* write all the dirty dentry pages */
709 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
710 f2fs_unlock_all(sbi
);
711 sync_dirty_dir_inodes(sbi
);
712 goto retry_flush_dents
;
716 * POR: we should ensure that there is no dirty node pages
717 * until finishing nat/sit flush.
720 mutex_lock(&sbi
->node_write
);
722 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
723 mutex_unlock(&sbi
->node_write
);
724 sync_node_pages(sbi
, 0, &wbc
);
725 goto retry_flush_nodes
;
727 blk_finish_plug(&plug
);
730 static void unblock_operations(struct f2fs_sb_info
*sbi
)
732 mutex_unlock(&sbi
->node_write
);
733 f2fs_unlock_all(sbi
);
736 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
741 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
743 if (!get_pages(sbi
, F2FS_WRITEBACK
))
748 finish_wait(&sbi
->cp_wait
, &wait
);
751 static void do_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
753 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
756 struct page
*cp_page
;
757 unsigned int data_sum_blocks
, orphan_blocks
;
762 /* Flush all the NAT/SIT pages */
763 while (get_pages(sbi
, F2FS_DIRTY_META
))
764 sync_meta_pages(sbi
, META
, LONG_MAX
);
766 next_free_nid(sbi
, &last_nid
);
770 * version number is already updated
772 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
773 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
774 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
775 for (i
= 0; i
< 3; i
++) {
776 ckpt
->cur_node_segno
[i
] =
777 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
778 ckpt
->cur_node_blkoff
[i
] =
779 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
780 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
781 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
783 for (i
= 0; i
< 3; i
++) {
784 ckpt
->cur_data_segno
[i
] =
785 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
786 ckpt
->cur_data_blkoff
[i
] =
787 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
788 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
789 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
792 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
793 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
794 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
796 /* 2 cp + n data seg summary + orphan inode blocks */
797 data_sum_blocks
= npages_for_summary_flush(sbi
);
798 if (data_sum_blocks
< 3)
799 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
801 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
803 orphan_blocks
= (sbi
->n_orphans
+ F2FS_ORPHANS_PER_BLOCK
- 1)
804 / F2FS_ORPHANS_PER_BLOCK
;
805 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + orphan_blocks
);
808 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
809 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
810 data_sum_blocks
+ orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
812 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
813 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
814 data_sum_blocks
+ orphan_blocks
);
818 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
820 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
822 /* update SIT/NAT bitmap */
823 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
824 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
826 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
827 *((__le32
*)((unsigned char *)ckpt
+
828 le32_to_cpu(ckpt
->checksum_offset
)))
829 = cpu_to_le32(crc32
);
831 start_blk
= __start_cp_addr(sbi
);
833 /* write out checkpoint buffer at block 0 */
834 cp_page
= grab_meta_page(sbi
, start_blk
++);
835 kaddr
= page_address(cp_page
);
836 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
837 set_page_dirty(cp_page
);
838 f2fs_put_page(cp_page
, 1);
840 if (sbi
->n_orphans
) {
841 write_orphan_inodes(sbi
, start_blk
);
842 start_blk
+= orphan_blocks
;
845 write_data_summaries(sbi
, start_blk
);
846 start_blk
+= data_sum_blocks
;
848 write_node_summaries(sbi
, start_blk
);
849 start_blk
+= NR_CURSEG_NODE_TYPE
;
852 /* writeout checkpoint block */
853 cp_page
= grab_meta_page(sbi
, start_blk
);
854 kaddr
= page_address(cp_page
);
855 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
856 set_page_dirty(cp_page
);
857 f2fs_put_page(cp_page
, 1);
859 /* wait for previous submitted node/meta pages writeback */
860 wait_on_all_pages_writeback(sbi
);
862 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
863 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
865 /* update user_block_counts */
866 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
867 sbi
->alloc_valid_block_count
= 0;
869 /* Here, we only have one bio having CP pack */
870 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
872 if (unlikely(!is_set_ckpt_flags(ckpt
, CP_ERROR_FLAG
))) {
873 clear_prefree_segments(sbi
);
874 F2FS_RESET_SB_DIRT(sbi
);
879 * We guarantee that this checkpoint procedure should not fail.
881 void write_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
883 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
884 unsigned long long ckpt_ver
;
886 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "start block_ops");
888 mutex_lock(&sbi
->cp_mutex
);
889 block_operations(sbi
);
891 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish block_ops");
893 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
894 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
895 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
898 * update checkpoint pack index
899 * Increase the version number so that
900 * SIT entries and seg summaries are written at correct place
902 ckpt_ver
= cur_cp_version(ckpt
);
903 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
905 /* write cached NAT/SIT entries to NAT/SIT area */
906 flush_nat_entries(sbi
);
907 flush_sit_entries(sbi
);
909 /* unlock all the fs_lock[] in do_checkpoint() */
910 do_checkpoint(sbi
, is_umount
);
912 unblock_operations(sbi
);
913 mutex_unlock(&sbi
->cp_mutex
);
915 stat_inc_cp_count(sbi
->stat_info
);
916 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish checkpoint");
919 void init_orphan_info(struct f2fs_sb_info
*sbi
)
921 spin_lock_init(&sbi
->orphan_inode_lock
);
922 INIT_LIST_HEAD(&sbi
->orphan_inode_list
);
925 * considering 512 blocks in a segment 8 blocks are needed for cp
926 * and log segment summaries. Remaining blocks are used to keep
927 * orphan entries with the limitation one reserved segment
928 * for cp pack we can have max 1020*504 orphan entries
930 sbi
->max_orphans
= (sbi
->blocks_per_seg
- 2 - NR_CURSEG_TYPE
)
931 * F2FS_ORPHANS_PER_BLOCK
;
934 int __init
create_checkpoint_caches(void)
936 orphan_entry_slab
= f2fs_kmem_cache_create("f2fs_orphan_entry",
937 sizeof(struct orphan_inode_entry
));
938 if (!orphan_entry_slab
)
940 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
941 sizeof(struct dir_inode_entry
));
942 if (!inode_entry_slab
) {
943 kmem_cache_destroy(orphan_entry_slab
);
949 void destroy_checkpoint_caches(void)
951 kmem_cache_destroy(orphan_entry_slab
);
952 kmem_cache_destroy(inode_entry_slab
);