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 trace_f2fs_writepage(page
, META
);
162 if (unlikely(sbi
->por_doing
))
164 if (wbc
->for_reclaim
)
167 /* Should not write any meta pages, if any IO error was occurred */
168 if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ERROR_FLAG
)))
171 f2fs_wait_on_page_writeback(page
, META
);
172 write_meta_page(sbi
, page
);
174 dec_page_count(sbi
, F2FS_DIRTY_META
);
179 redirty_page_for_writepage(wbc
, page
);
180 return AOP_WRITEPAGE_ACTIVATE
;
183 static int f2fs_write_meta_pages(struct address_space
*mapping
,
184 struct writeback_control
*wbc
)
186 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
189 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
191 /* collect a number of dirty meta pages and write together */
192 if (wbc
->for_kupdate
||
193 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
196 /* if mounting is failed, skip writing node pages */
197 mutex_lock(&sbi
->cp_mutex
);
198 diff
= nr_pages_to_write(sbi
, META
, wbc
);
199 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
200 mutex_unlock(&sbi
->cp_mutex
);
201 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
205 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
209 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
212 struct address_space
*mapping
= META_MAPPING(sbi
);
213 pgoff_t index
= 0, end
= LONG_MAX
;
216 struct writeback_control wbc
= {
220 pagevec_init(&pvec
, 0);
222 while (index
<= end
) {
224 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
226 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
227 if (unlikely(nr_pages
== 0))
230 for (i
= 0; i
< nr_pages
; i
++) {
231 struct page
*page
= pvec
.pages
[i
];
235 if (unlikely(page
->mapping
!= mapping
)) {
240 if (!PageDirty(page
)) {
241 /* someone wrote it for us */
242 goto continue_unlock
;
245 if (!clear_page_dirty_for_io(page
))
246 goto continue_unlock
;
248 if (f2fs_write_meta_page(page
, &wbc
)) {
253 if (unlikely(nwritten
>= nr_to_write
))
256 pagevec_release(&pvec
);
261 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
266 static int f2fs_set_meta_page_dirty(struct page
*page
)
268 struct address_space
*mapping
= page
->mapping
;
269 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
271 trace_f2fs_set_page_dirty(page
, META
);
273 SetPageUptodate(page
);
274 if (!PageDirty(page
)) {
275 __set_page_dirty_nobuffers(page
);
276 inc_page_count(sbi
, F2FS_DIRTY_META
);
282 const struct address_space_operations f2fs_meta_aops
= {
283 .writepage
= f2fs_write_meta_page
,
284 .writepages
= f2fs_write_meta_pages
,
285 .set_page_dirty
= f2fs_set_meta_page_dirty
,
288 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
292 spin_lock(&sbi
->orphan_inode_lock
);
293 if (unlikely(sbi
->n_orphans
>= sbi
->max_orphans
))
297 spin_unlock(&sbi
->orphan_inode_lock
);
302 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
304 spin_lock(&sbi
->orphan_inode_lock
);
305 f2fs_bug_on(sbi
->n_orphans
== 0);
307 spin_unlock(&sbi
->orphan_inode_lock
);
310 void add_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
312 struct list_head
*head
;
313 struct orphan_inode_entry
*new, *orphan
;
315 new = f2fs_kmem_cache_alloc(orphan_entry_slab
, GFP_ATOMIC
);
318 spin_lock(&sbi
->orphan_inode_lock
);
319 head
= &sbi
->orphan_inode_list
;
320 list_for_each_entry(orphan
, head
, list
) {
321 if (orphan
->ino
== ino
) {
322 spin_unlock(&sbi
->orphan_inode_lock
);
323 kmem_cache_free(orphan_entry_slab
, new);
327 if (orphan
->ino
> ino
)
331 /* add new orphan entry into list which is sorted by inode number */
332 list_add_tail(&new->list
, &orphan
->list
);
333 spin_unlock(&sbi
->orphan_inode_lock
);
336 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
338 struct list_head
*head
;
339 struct orphan_inode_entry
*orphan
;
341 spin_lock(&sbi
->orphan_inode_lock
);
342 head
= &sbi
->orphan_inode_list
;
343 list_for_each_entry(orphan
, head
, list
) {
344 if (orphan
->ino
== ino
) {
345 list_del(&orphan
->list
);
346 f2fs_bug_on(sbi
->n_orphans
== 0);
348 spin_unlock(&sbi
->orphan_inode_lock
);
349 kmem_cache_free(orphan_entry_slab
, orphan
);
353 spin_unlock(&sbi
->orphan_inode_lock
);
356 static void recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
358 struct inode
*inode
= f2fs_iget(sbi
->sb
, ino
);
359 f2fs_bug_on(IS_ERR(inode
));
362 /* truncate all the data during iput */
366 void recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
368 block_t start_blk
, orphan_blkaddr
, i
, j
;
370 if (!is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
))
373 sbi
->por_doing
= true;
375 start_blk
= __start_cp_addr(sbi
) + 1 +
376 le32_to_cpu(F2FS_RAW_SUPER(sbi
)->cp_payload
);
377 orphan_blkaddr
= __start_sum_addr(sbi
) - 1;
379 ra_meta_pages(sbi
, start_blk
, orphan_blkaddr
, META_CP
);
381 for (i
= 0; i
< orphan_blkaddr
; i
++) {
382 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
383 struct f2fs_orphan_block
*orphan_blk
;
385 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
386 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
387 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
388 recover_orphan_inode(sbi
, ino
);
390 f2fs_put_page(page
, 1);
392 /* clear Orphan Flag */
393 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
394 sbi
->por_doing
= false;
398 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
400 struct list_head
*head
;
401 struct f2fs_orphan_block
*orphan_blk
= NULL
;
402 unsigned int nentries
= 0;
403 unsigned short index
;
404 unsigned short orphan_blocks
= (unsigned short)((sbi
->n_orphans
+
405 (F2FS_ORPHANS_PER_BLOCK
- 1)) / F2FS_ORPHANS_PER_BLOCK
);
406 struct page
*page
= NULL
;
407 struct orphan_inode_entry
*orphan
= NULL
;
409 for (index
= 0; index
< orphan_blocks
; index
++)
410 grab_meta_page(sbi
, start_blk
+ index
);
413 spin_lock(&sbi
->orphan_inode_lock
);
414 head
= &sbi
->orphan_inode_list
;
416 /* loop for each orphan inode entry and write them in Jornal block */
417 list_for_each_entry(orphan
, head
, list
) {
419 page
= find_get_page(META_MAPPING(sbi
), start_blk
++);
422 (struct f2fs_orphan_block
*)page_address(page
);
423 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
424 f2fs_put_page(page
, 0);
427 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
429 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
431 * an orphan block is full of 1020 entries,
432 * then we need to flush current orphan blocks
433 * and bring another one in memory
435 orphan_blk
->blk_addr
= cpu_to_le16(index
);
436 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
437 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
438 set_page_dirty(page
);
439 f2fs_put_page(page
, 1);
447 orphan_blk
->blk_addr
= cpu_to_le16(index
);
448 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
449 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
450 set_page_dirty(page
);
451 f2fs_put_page(page
, 1);
454 spin_unlock(&sbi
->orphan_inode_lock
);
457 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
458 block_t cp_addr
, unsigned long long *version
)
460 struct page
*cp_page_1
, *cp_page_2
= NULL
;
461 unsigned long blk_size
= sbi
->blocksize
;
462 struct f2fs_checkpoint
*cp_block
;
463 unsigned long long cur_version
= 0, pre_version
= 0;
467 /* Read the 1st cp block in this CP pack */
468 cp_page_1
= get_meta_page(sbi
, cp_addr
);
470 /* get the version number */
471 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
472 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
473 if (crc_offset
>= blk_size
)
476 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
477 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
480 pre_version
= cur_cp_version(cp_block
);
482 /* Read the 2nd cp block in this CP pack */
483 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
484 cp_page_2
= get_meta_page(sbi
, cp_addr
);
486 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
487 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
488 if (crc_offset
>= blk_size
)
491 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
492 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
495 cur_version
= cur_cp_version(cp_block
);
497 if (cur_version
== pre_version
) {
498 *version
= cur_version
;
499 f2fs_put_page(cp_page_2
, 1);
503 f2fs_put_page(cp_page_2
, 1);
505 f2fs_put_page(cp_page_1
, 1);
509 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
511 struct f2fs_checkpoint
*cp_block
;
512 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
513 struct page
*cp1
, *cp2
, *cur_page
;
514 unsigned long blk_size
= sbi
->blocksize
;
515 unsigned long long cp1_version
= 0, cp2_version
= 0;
516 unsigned long long cp_start_blk_no
;
517 unsigned int cp_blks
= 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi
)->cp_payload
);
521 sbi
->ckpt
= kzalloc(cp_blks
* blk_size
, GFP_KERNEL
);
525 * Finding out valid cp block involves read both
526 * sets( cp pack1 and cp pack 2)
528 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
529 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
531 /* The second checkpoint pack should start at the next segment */
532 cp_start_blk_no
+= ((unsigned long long)1) <<
533 le32_to_cpu(fsb
->log_blocks_per_seg
);
534 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
537 if (ver_after(cp2_version
, cp1_version
))
549 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
550 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
555 cp_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
557 cp_blk_no
+= 1 << le32_to_cpu(fsb
->log_blocks_per_seg
);
559 for (i
= 1; i
< cp_blks
; i
++) {
560 void *sit_bitmap_ptr
;
561 unsigned char *ckpt
= (unsigned char *)sbi
->ckpt
;
563 cur_page
= get_meta_page(sbi
, cp_blk_no
+ i
);
564 sit_bitmap_ptr
= page_address(cur_page
);
565 memcpy(ckpt
+ i
* blk_size
, sit_bitmap_ptr
, blk_size
);
566 f2fs_put_page(cur_page
, 1);
569 f2fs_put_page(cp1
, 1);
570 f2fs_put_page(cp2
, 1);
578 static int __add_dirty_inode(struct inode
*inode
, struct dir_inode_entry
*new)
580 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
582 if (is_inode_flag_set(F2FS_I(inode
), FI_DIRTY_DIR
))
585 set_inode_flag(F2FS_I(inode
), FI_DIRTY_DIR
);
586 F2FS_I(inode
)->dirty_dir
= new;
587 list_add_tail(&new->list
, &sbi
->dir_inode_list
);
588 stat_inc_dirty_dir(sbi
);
592 void set_dirty_dir_page(struct inode
*inode
, struct page
*page
)
594 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
595 struct dir_inode_entry
*new;
598 if (!S_ISDIR(inode
->i_mode
))
601 new = f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
603 INIT_LIST_HEAD(&new->list
);
605 spin_lock(&sbi
->dir_inode_lock
);
606 ret
= __add_dirty_inode(inode
, new);
607 inode_inc_dirty_dents(inode
);
608 SetPagePrivate(page
);
609 spin_unlock(&sbi
->dir_inode_lock
);
612 kmem_cache_free(inode_entry_slab
, new);
615 void add_dirty_dir_inode(struct inode
*inode
)
617 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
618 struct dir_inode_entry
*new =
619 f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
623 INIT_LIST_HEAD(&new->list
);
625 spin_lock(&sbi
->dir_inode_lock
);
626 ret
= __add_dirty_inode(inode
, new);
627 spin_unlock(&sbi
->dir_inode_lock
);
630 kmem_cache_free(inode_entry_slab
, new);
633 void remove_dirty_dir_inode(struct inode
*inode
)
635 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
636 struct dir_inode_entry
*entry
;
638 if (!S_ISDIR(inode
->i_mode
))
641 spin_lock(&sbi
->dir_inode_lock
);
642 if (get_dirty_dents(inode
) ||
643 !is_inode_flag_set(F2FS_I(inode
), FI_DIRTY_DIR
)) {
644 spin_unlock(&sbi
->dir_inode_lock
);
648 entry
= F2FS_I(inode
)->dirty_dir
;
649 list_del(&entry
->list
);
650 F2FS_I(inode
)->dirty_dir
= NULL
;
651 clear_inode_flag(F2FS_I(inode
), FI_DIRTY_DIR
);
652 stat_dec_dirty_dir(sbi
);
653 spin_unlock(&sbi
->dir_inode_lock
);
654 kmem_cache_free(inode_entry_slab
, entry
);
656 /* Only from the recovery routine */
657 if (is_inode_flag_set(F2FS_I(inode
), FI_DELAY_IPUT
)) {
658 clear_inode_flag(F2FS_I(inode
), FI_DELAY_IPUT
);
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
;
761 int cp_payload_blks
= le32_to_cpu(F2FS_RAW_SUPER(sbi
)->cp_payload
);
764 * This avoids to conduct wrong roll-forward operations and uses
765 * metapages, so should be called prior to sync_meta_pages below.
767 discard_next_dnode(sbi
);
769 /* Flush all the NAT/SIT pages */
770 while (get_pages(sbi
, F2FS_DIRTY_META
))
771 sync_meta_pages(sbi
, META
, LONG_MAX
);
773 next_free_nid(sbi
, &last_nid
);
777 * version number is already updated
779 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
780 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
781 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
782 for (i
= 0; i
< 3; i
++) {
783 ckpt
->cur_node_segno
[i
] =
784 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
785 ckpt
->cur_node_blkoff
[i
] =
786 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
787 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
788 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
790 for (i
= 0; i
< 3; i
++) {
791 ckpt
->cur_data_segno
[i
] =
792 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
793 ckpt
->cur_data_blkoff
[i
] =
794 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
795 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
796 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
799 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
800 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
801 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
803 /* 2 cp + n data seg summary + orphan inode blocks */
804 data_sum_blocks
= npages_for_summary_flush(sbi
);
805 if (data_sum_blocks
< 3)
806 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
808 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
810 orphan_blocks
= (sbi
->n_orphans
+ F2FS_ORPHANS_PER_BLOCK
- 1)
811 / F2FS_ORPHANS_PER_BLOCK
;
812 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + cp_payload_blks
+
816 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
817 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
818 cp_payload_blks
+ data_sum_blocks
+
819 orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
821 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
822 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
823 cp_payload_blks
+ data_sum_blocks
+
828 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
830 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
832 /* update SIT/NAT bitmap */
833 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
834 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
836 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
837 *((__le32
*)((unsigned char *)ckpt
+
838 le32_to_cpu(ckpt
->checksum_offset
)))
839 = cpu_to_le32(crc32
);
841 start_blk
= __start_cp_addr(sbi
);
843 /* write out checkpoint buffer at block 0 */
844 cp_page
= grab_meta_page(sbi
, start_blk
++);
845 kaddr
= page_address(cp_page
);
846 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
847 set_page_dirty(cp_page
);
848 f2fs_put_page(cp_page
, 1);
850 for (i
= 1; i
< 1 + cp_payload_blks
; i
++) {
851 cp_page
= grab_meta_page(sbi
, start_blk
++);
852 kaddr
= page_address(cp_page
);
853 memcpy(kaddr
, (char *)ckpt
+ i
* F2FS_BLKSIZE
,
854 (1 << sbi
->log_blocksize
));
855 set_page_dirty(cp_page
);
856 f2fs_put_page(cp_page
, 1);
859 if (sbi
->n_orphans
) {
860 write_orphan_inodes(sbi
, start_blk
);
861 start_blk
+= orphan_blocks
;
864 write_data_summaries(sbi
, start_blk
);
865 start_blk
+= data_sum_blocks
;
867 write_node_summaries(sbi
, start_blk
);
868 start_blk
+= NR_CURSEG_NODE_TYPE
;
871 /* writeout checkpoint block */
872 cp_page
= grab_meta_page(sbi
, start_blk
);
873 kaddr
= page_address(cp_page
);
874 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
875 set_page_dirty(cp_page
);
876 f2fs_put_page(cp_page
, 1);
878 /* wait for previous submitted node/meta pages writeback */
879 wait_on_all_pages_writeback(sbi
);
881 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
882 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
884 /* update user_block_counts */
885 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
886 sbi
->alloc_valid_block_count
= 0;
888 /* Here, we only have one bio having CP pack */
889 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
891 if (unlikely(!is_set_ckpt_flags(ckpt
, CP_ERROR_FLAG
))) {
892 clear_prefree_segments(sbi
);
893 F2FS_RESET_SB_DIRT(sbi
);
898 * We guarantee that this checkpoint procedure should not fail.
900 void write_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
902 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
903 unsigned long long ckpt_ver
;
905 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "start block_ops");
907 mutex_lock(&sbi
->cp_mutex
);
908 block_operations(sbi
);
910 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish block_ops");
912 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
913 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
914 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
917 * update checkpoint pack index
918 * Increase the version number so that
919 * SIT entries and seg summaries are written at correct place
921 ckpt_ver
= cur_cp_version(ckpt
);
922 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
924 /* write cached NAT/SIT entries to NAT/SIT area */
925 flush_nat_entries(sbi
);
926 flush_sit_entries(sbi
);
928 /* unlock all the fs_lock[] in do_checkpoint() */
929 do_checkpoint(sbi
, is_umount
);
931 unblock_operations(sbi
);
932 mutex_unlock(&sbi
->cp_mutex
);
934 stat_inc_cp_count(sbi
->stat_info
);
935 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish checkpoint");
938 void init_orphan_info(struct f2fs_sb_info
*sbi
)
940 spin_lock_init(&sbi
->orphan_inode_lock
);
941 INIT_LIST_HEAD(&sbi
->orphan_inode_list
);
944 * considering 512 blocks in a segment 8 blocks are needed for cp
945 * and log segment summaries. Remaining blocks are used to keep
946 * orphan entries with the limitation one reserved segment
947 * for cp pack we can have max 1020*504 orphan entries
949 sbi
->max_orphans
= (sbi
->blocks_per_seg
- 2 - NR_CURSEG_TYPE
)
950 * F2FS_ORPHANS_PER_BLOCK
;
953 int __init
create_checkpoint_caches(void)
955 orphan_entry_slab
= f2fs_kmem_cache_create("f2fs_orphan_entry",
956 sizeof(struct orphan_inode_entry
));
957 if (!orphan_entry_slab
)
959 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
960 sizeof(struct dir_inode_entry
));
961 if (!inode_entry_slab
) {
962 kmem_cache_destroy(orphan_entry_slab
);
968 void destroy_checkpoint_caches(void)
970 kmem_cache_destroy(orphan_entry_slab
);
971 kmem_cache_destroy(inode_entry_slab
);