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
);
42 /* We wait writeback only inside grab_meta_page() */
43 wait_on_page_writeback(page
);
44 SetPageUptodate(page
);
49 * We guarantee no failure on the returned page.
51 struct page
*get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
53 struct address_space
*mapping
= META_MAPPING(sbi
);
56 page
= grab_cache_page(mapping
, index
);
61 if (PageUptodate(page
))
64 if (f2fs_submit_page_bio(sbi
, page
, index
,
65 READ_SYNC
| REQ_META
| REQ_PRIO
))
69 if (unlikely(page
->mapping
!= mapping
)) {
70 f2fs_put_page(page
, 1);
74 mark_page_accessed(page
);
78 inline int get_max_meta_blks(struct f2fs_sb_info
*sbi
, int type
)
82 return NM_I(sbi
)->max_nid
/ NAT_ENTRY_PER_BLOCK
;
84 return SIT_BLK_CNT(sbi
);
93 * Readahead CP/NAT/SIT pages
95 int ra_meta_pages(struct f2fs_sb_info
*sbi
, int start
, int nrpages
, int type
)
97 block_t prev_blk_addr
= 0;
100 int max_blks
= get_max_meta_blks(sbi
, type
);
102 struct f2fs_io_info fio
= {
104 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
107 for (; nrpages
-- > 0; blkno
++) {
112 /* get nat block addr */
113 if (unlikely(blkno
>= max_blks
))
115 blk_addr
= current_nat_addr(sbi
,
116 blkno
* NAT_ENTRY_PER_BLOCK
);
119 /* get sit block addr */
120 if (unlikely(blkno
>= max_blks
))
122 blk_addr
= current_sit_addr(sbi
,
123 blkno
* SIT_ENTRY_PER_BLOCK
);
124 if (blkno
!= start
&& prev_blk_addr
+ 1 != blk_addr
)
126 prev_blk_addr
= blk_addr
;
129 /* get 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 wait_on_page_writeback(page
);
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 set_page_dirty(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
);
187 int nrpages
= MAX_BIO_BLOCKS(max_hw_blocks(sbi
));
190 if (wbc
->for_kupdate
)
193 /* collect a number of dirty meta pages and write together */
194 if (get_pages(sbi
, F2FS_DIRTY_META
) < nrpages
)
197 /* if mounting is failed, skip writing node pages */
198 mutex_lock(&sbi
->cp_mutex
);
199 written
= sync_meta_pages(sbi
, META
, nrpages
);
200 mutex_unlock(&sbi
->cp_mutex
);
201 wbc
->nr_to_write
-= written
;
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
, *this;
309 struct orphan_inode_entry
*new = NULL
, *orphan
= NULL
;
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(this, head
) {
317 orphan
= list_entry(this, struct orphan_inode_entry
, list
);
318 if (orphan
->ino
== ino
) {
319 spin_unlock(&sbi
->orphan_inode_lock
);
320 kmem_cache_free(orphan_entry_slab
, new);
324 if (orphan
->ino
> ino
)
329 /* add new_oentry into list which is sorted by inode number */
331 list_add(&new->list
, this->prev
);
333 list_add_tail(&new->list
, head
);
334 spin_unlock(&sbi
->orphan_inode_lock
);
337 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
339 struct list_head
*head
;
340 struct orphan_inode_entry
*orphan
;
342 spin_lock(&sbi
->orphan_inode_lock
);
343 head
= &sbi
->orphan_inode_list
;
344 list_for_each_entry(orphan
, head
, list
) {
345 if (orphan
->ino
== ino
) {
346 list_del(&orphan
->list
);
347 kmem_cache_free(orphan_entry_slab
, orphan
);
348 f2fs_bug_on(sbi
->n_orphans
== 0);
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;
374 start_blk
= __start_cp_addr(sbi
) + 1;
375 orphan_blkaddr
= __start_sum_addr(sbi
) - 1;
377 ra_meta_pages(sbi
, start_blk
, orphan_blkaddr
, META_CP
);
379 for (i
= 0; i
< orphan_blkaddr
; i
++) {
380 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
381 struct f2fs_orphan_block
*orphan_blk
;
383 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
384 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
385 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
386 recover_orphan_inode(sbi
, ino
);
388 f2fs_put_page(page
, 1);
390 /* clear Orphan Flag */
391 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
392 sbi
->por_doing
= false;
396 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
398 struct list_head
*head
;
399 struct f2fs_orphan_block
*orphan_blk
= NULL
;
400 unsigned int nentries
= 0;
401 unsigned short index
;
402 unsigned short orphan_blocks
= (unsigned short)((sbi
->n_orphans
+
403 (F2FS_ORPHANS_PER_BLOCK
- 1)) / F2FS_ORPHANS_PER_BLOCK
);
404 struct page
*page
= NULL
;
405 struct orphan_inode_entry
*orphan
= NULL
;
407 for (index
= 0; index
< orphan_blocks
; index
++)
408 grab_meta_page(sbi
, start_blk
+ index
);
411 spin_lock(&sbi
->orphan_inode_lock
);
412 head
= &sbi
->orphan_inode_list
;
414 /* loop for each orphan inode entry and write them in Jornal block */
415 list_for_each_entry(orphan
, head
, list
) {
417 page
= find_get_page(META_MAPPING(sbi
), start_blk
++);
420 (struct f2fs_orphan_block
*)page_address(page
);
421 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
422 f2fs_put_page(page
, 0);
425 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
427 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
429 * an orphan block is full of 1020 entries,
430 * then we need to flush current orphan blocks
431 * and bring another one in memory
433 orphan_blk
->blk_addr
= cpu_to_le16(index
);
434 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
435 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
436 set_page_dirty(page
);
437 f2fs_put_page(page
, 1);
445 orphan_blk
->blk_addr
= cpu_to_le16(index
);
446 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
447 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
448 set_page_dirty(page
);
449 f2fs_put_page(page
, 1);
452 spin_unlock(&sbi
->orphan_inode_lock
);
455 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
456 block_t cp_addr
, unsigned long long *version
)
458 struct page
*cp_page_1
, *cp_page_2
= NULL
;
459 unsigned long blk_size
= sbi
->blocksize
;
460 struct f2fs_checkpoint
*cp_block
;
461 unsigned long long cur_version
= 0, pre_version
= 0;
465 /* Read the 1st cp block in this CP pack */
466 cp_page_1
= get_meta_page(sbi
, cp_addr
);
468 /* get the version number */
469 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
470 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
471 if (crc_offset
>= blk_size
)
474 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
475 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
478 pre_version
= cur_cp_version(cp_block
);
480 /* Read the 2nd cp block in this CP pack */
481 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
482 cp_page_2
= get_meta_page(sbi
, cp_addr
);
484 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
485 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
486 if (crc_offset
>= blk_size
)
489 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
490 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
493 cur_version
= cur_cp_version(cp_block
);
495 if (cur_version
== pre_version
) {
496 *version
= cur_version
;
497 f2fs_put_page(cp_page_2
, 1);
501 f2fs_put_page(cp_page_2
, 1);
503 f2fs_put_page(cp_page_1
, 1);
507 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
509 struct f2fs_checkpoint
*cp_block
;
510 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
511 struct page
*cp1
, *cp2
, *cur_page
;
512 unsigned long blk_size
= sbi
->blocksize
;
513 unsigned long long cp1_version
= 0, cp2_version
= 0;
514 unsigned long long cp_start_blk_no
;
516 sbi
->ckpt
= kzalloc(blk_size
, GFP_KERNEL
);
520 * Finding out valid cp block involves read both
521 * sets( cp pack1 and cp pack 2)
523 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
524 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
526 /* The second checkpoint pack should start at the next segment */
527 cp_start_blk_no
+= ((unsigned long long)1) <<
528 le32_to_cpu(fsb
->log_blocks_per_seg
);
529 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
532 if (ver_after(cp2_version
, cp1_version
))
544 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
545 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
547 f2fs_put_page(cp1
, 1);
548 f2fs_put_page(cp2
, 1);
556 static int __add_dirty_inode(struct inode
*inode
, struct dir_inode_entry
*new)
558 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
559 struct list_head
*head
= &sbi
->dir_inode_list
;
560 struct list_head
*this;
562 list_for_each(this, head
) {
563 struct dir_inode_entry
*entry
;
564 entry
= list_entry(this, struct dir_inode_entry
, list
);
565 if (unlikely(entry
->inode
== inode
))
568 list_add_tail(&new->list
, head
);
569 stat_inc_dirty_dir(sbi
);
573 void set_dirty_dir_page(struct inode
*inode
, struct page
*page
)
575 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
576 struct dir_inode_entry
*new;
578 if (!S_ISDIR(inode
->i_mode
))
581 new = f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
583 INIT_LIST_HEAD(&new->list
);
585 spin_lock(&sbi
->dir_inode_lock
);
586 if (__add_dirty_inode(inode
, new))
587 kmem_cache_free(inode_entry_slab
, new);
589 inode_inc_dirty_dents(inode
);
590 SetPagePrivate(page
);
591 spin_unlock(&sbi
->dir_inode_lock
);
594 void add_dirty_dir_inode(struct inode
*inode
)
596 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
597 struct dir_inode_entry
*new =
598 f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
601 INIT_LIST_HEAD(&new->list
);
603 spin_lock(&sbi
->dir_inode_lock
);
604 if (__add_dirty_inode(inode
, new))
605 kmem_cache_free(inode_entry_slab
, new);
606 spin_unlock(&sbi
->dir_inode_lock
);
609 void remove_dirty_dir_inode(struct inode
*inode
)
611 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
613 struct list_head
*this, *head
;
615 if (!S_ISDIR(inode
->i_mode
))
618 spin_lock(&sbi
->dir_inode_lock
);
619 if (atomic_read(&F2FS_I(inode
)->dirty_dents
)) {
620 spin_unlock(&sbi
->dir_inode_lock
);
624 head
= &sbi
->dir_inode_list
;
625 list_for_each(this, head
) {
626 struct dir_inode_entry
*entry
;
627 entry
= list_entry(this, struct dir_inode_entry
, list
);
628 if (entry
->inode
== inode
) {
629 list_del(&entry
->list
);
630 kmem_cache_free(inode_entry_slab
, entry
);
631 stat_dec_dirty_dir(sbi
);
635 spin_unlock(&sbi
->dir_inode_lock
);
637 /* Only from the recovery routine */
638 if (is_inode_flag_set(F2FS_I(inode
), FI_DELAY_IPUT
)) {
639 clear_inode_flag(F2FS_I(inode
), FI_DELAY_IPUT
);
644 struct inode
*check_dirty_dir_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
647 struct list_head
*this, *head
;
648 struct inode
*inode
= NULL
;
650 spin_lock(&sbi
->dir_inode_lock
);
652 head
= &sbi
->dir_inode_list
;
653 list_for_each(this, head
) {
654 struct dir_inode_entry
*entry
;
655 entry
= list_entry(this, struct dir_inode_entry
, list
);
656 if (entry
->inode
->i_ino
== ino
) {
657 inode
= entry
->inode
;
661 spin_unlock(&sbi
->dir_inode_lock
);
665 void sync_dirty_dir_inodes(struct f2fs_sb_info
*sbi
)
667 struct list_head
*head
;
668 struct dir_inode_entry
*entry
;
671 spin_lock(&sbi
->dir_inode_lock
);
673 head
= &sbi
->dir_inode_list
;
674 if (list_empty(head
)) {
675 spin_unlock(&sbi
->dir_inode_lock
);
678 entry
= list_entry(head
->next
, struct dir_inode_entry
, list
);
679 inode
= igrab(entry
->inode
);
680 spin_unlock(&sbi
->dir_inode_lock
);
682 filemap_flush(inode
->i_mapping
);
686 * We should submit bio, since it exists several
687 * wribacking dentry pages in the freeing inode.
689 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
695 * Freeze all the FS-operations for checkpoint.
697 static void block_operations(struct f2fs_sb_info
*sbi
)
699 struct writeback_control wbc
= {
700 .sync_mode
= WB_SYNC_ALL
,
701 .nr_to_write
= LONG_MAX
,
704 struct blk_plug plug
;
706 blk_start_plug(&plug
);
710 /* write all the dirty dentry pages */
711 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
712 f2fs_unlock_all(sbi
);
713 sync_dirty_dir_inodes(sbi
);
714 goto retry_flush_dents
;
718 * POR: we should ensure that there is no dirty node pages
719 * until finishing nat/sit flush.
722 mutex_lock(&sbi
->node_write
);
724 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
725 mutex_unlock(&sbi
->node_write
);
726 sync_node_pages(sbi
, 0, &wbc
);
727 goto retry_flush_nodes
;
729 blk_finish_plug(&plug
);
732 static void unblock_operations(struct f2fs_sb_info
*sbi
)
734 mutex_unlock(&sbi
->node_write
);
735 f2fs_unlock_all(sbi
);
738 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
743 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
745 if (!get_pages(sbi
, F2FS_WRITEBACK
))
750 finish_wait(&sbi
->cp_wait
, &wait
);
753 static void do_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
755 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
758 struct page
*cp_page
;
759 unsigned int data_sum_blocks
, orphan_blocks
;
764 /* Flush all the NAT/SIT pages */
765 while (get_pages(sbi
, F2FS_DIRTY_META
))
766 sync_meta_pages(sbi
, META
, LONG_MAX
);
768 next_free_nid(sbi
, &last_nid
);
772 * version number is already updated
774 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
775 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
776 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
777 for (i
= 0; i
< 3; i
++) {
778 ckpt
->cur_node_segno
[i
] =
779 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
780 ckpt
->cur_node_blkoff
[i
] =
781 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
782 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
783 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
785 for (i
= 0; i
< 3; i
++) {
786 ckpt
->cur_data_segno
[i
] =
787 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
788 ckpt
->cur_data_blkoff
[i
] =
789 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
790 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
791 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
794 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
795 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
796 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
798 /* 2 cp + n data seg summary + orphan inode blocks */
799 data_sum_blocks
= npages_for_summary_flush(sbi
);
800 if (data_sum_blocks
< 3)
801 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
803 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
805 orphan_blocks
= (sbi
->n_orphans
+ F2FS_ORPHANS_PER_BLOCK
- 1)
806 / F2FS_ORPHANS_PER_BLOCK
;
807 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + orphan_blocks
);
810 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
811 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
812 data_sum_blocks
+ orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
814 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
815 ckpt
->cp_pack_total_block_count
= cpu_to_le32(2 +
816 data_sum_blocks
+ orphan_blocks
);
820 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
822 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
824 /* update SIT/NAT bitmap */
825 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
826 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
828 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
829 *((__le32
*)((unsigned char *)ckpt
+
830 le32_to_cpu(ckpt
->checksum_offset
)))
831 = cpu_to_le32(crc32
);
833 start_blk
= __start_cp_addr(sbi
);
835 /* write out checkpoint buffer at block 0 */
836 cp_page
= grab_meta_page(sbi
, start_blk
++);
837 kaddr
= page_address(cp_page
);
838 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
839 set_page_dirty(cp_page
);
840 f2fs_put_page(cp_page
, 1);
842 if (sbi
->n_orphans
) {
843 write_orphan_inodes(sbi
, start_blk
);
844 start_blk
+= orphan_blocks
;
847 write_data_summaries(sbi
, start_blk
);
848 start_blk
+= data_sum_blocks
;
850 write_node_summaries(sbi
, start_blk
);
851 start_blk
+= NR_CURSEG_NODE_TYPE
;
854 /* writeout checkpoint block */
855 cp_page
= grab_meta_page(sbi
, start_blk
);
856 kaddr
= page_address(cp_page
);
857 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
858 set_page_dirty(cp_page
);
859 f2fs_put_page(cp_page
, 1);
861 /* wait for previous submitted node/meta pages writeback */
862 wait_on_all_pages_writeback(sbi
);
864 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
865 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
867 /* update user_block_counts */
868 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
869 sbi
->alloc_valid_block_count
= 0;
871 /* Here, we only have one bio having CP pack */
872 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
874 if (unlikely(!is_set_ckpt_flags(ckpt
, CP_ERROR_FLAG
))) {
875 clear_prefree_segments(sbi
);
876 F2FS_RESET_SB_DIRT(sbi
);
881 * We guarantee that this checkpoint procedure should not fail.
883 void write_checkpoint(struct f2fs_sb_info
*sbi
, bool is_umount
)
885 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
886 unsigned long long ckpt_ver
;
888 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "start block_ops");
890 mutex_lock(&sbi
->cp_mutex
);
891 block_operations(sbi
);
893 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish block_ops");
895 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
896 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
897 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
900 * update checkpoint pack index
901 * Increase the version number so that
902 * SIT entries and seg summaries are written at correct place
904 ckpt_ver
= cur_cp_version(ckpt
);
905 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
907 /* write cached NAT/SIT entries to NAT/SIT area */
908 flush_nat_entries(sbi
);
909 flush_sit_entries(sbi
);
911 /* unlock all the fs_lock[] in do_checkpoint() */
912 do_checkpoint(sbi
, is_umount
);
914 unblock_operations(sbi
);
915 mutex_unlock(&sbi
->cp_mutex
);
917 trace_f2fs_write_checkpoint(sbi
->sb
, is_umount
, "finish checkpoint");
920 void init_orphan_info(struct f2fs_sb_info
*sbi
)
922 spin_lock_init(&sbi
->orphan_inode_lock
);
923 INIT_LIST_HEAD(&sbi
->orphan_inode_list
);
926 * considering 512 blocks in a segment 8 blocks are needed for cp
927 * and log segment summaries. Remaining blocks are used to keep
928 * orphan entries with the limitation one reserved segment
929 * for cp pack we can have max 1020*504 orphan entries
931 sbi
->max_orphans
= (sbi
->blocks_per_seg
- 2 - NR_CURSEG_TYPE
)
932 * F2FS_ORPHANS_PER_BLOCK
;
935 int __init
create_checkpoint_caches(void)
937 orphan_entry_slab
= f2fs_kmem_cache_create("f2fs_orphan_entry",
938 sizeof(struct orphan_inode_entry
), NULL
);
939 if (!orphan_entry_slab
)
941 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
942 sizeof(struct dir_inode_entry
), NULL
);
943 if (!inode_entry_slab
) {
944 kmem_cache_destroy(orphan_entry_slab
);
950 void destroy_checkpoint_caches(void)
952 kmem_cache_destroy(orphan_entry_slab
);
953 kmem_cache_destroy(inode_entry_slab
);