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
*ino_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);
75 static inline bool is_valid_blkaddr(struct f2fs_sb_info
*sbi
,
76 block_t blkaddr
, int type
)
82 if (unlikely(blkaddr
>= SIT_BLK_CNT(sbi
)))
86 if (unlikely(blkaddr
>= MAIN_BLKADDR(sbi
) ||
87 blkaddr
< SM_I(sbi
)->ssa_blkaddr
))
91 if (unlikely(blkaddr
>= SIT_I(sbi
)->sit_base_addr
||
92 blkaddr
< __start_cp_addr(sbi
)))
96 if (unlikely(blkaddr
>= MAX_BLKADDR(sbi
) ||
97 blkaddr
< MAIN_BLKADDR(sbi
)))
108 * Readahead CP/NAT/SIT/SSA pages
110 int ra_meta_pages(struct f2fs_sb_info
*sbi
, block_t start
, int nrpages
, int type
)
112 block_t prev_blk_addr
= 0;
114 block_t blkno
= start
;
116 struct f2fs_io_info fio
= {
118 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
121 for (; nrpages
-- > 0; blkno
++) {
124 if (!is_valid_blkaddr(sbi
, blkno
, type
))
129 if (unlikely(blkno
>=
130 NAT_BLOCK_OFFSET(NM_I(sbi
)->max_nid
)))
132 /* get nat block addr */
133 blk_addr
= current_nat_addr(sbi
,
134 blkno
* NAT_ENTRY_PER_BLOCK
);
137 /* get sit block addr */
138 blk_addr
= current_sit_addr(sbi
,
139 blkno
* SIT_ENTRY_PER_BLOCK
);
140 if (blkno
!= start
&& prev_blk_addr
+ 1 != blk_addr
)
142 prev_blk_addr
= blk_addr
;
153 page
= grab_cache_page(META_MAPPING(sbi
), blk_addr
);
156 if (PageUptodate(page
)) {
157 f2fs_put_page(page
, 1);
161 f2fs_submit_page_mbio(sbi
, page
, blk_addr
, &fio
);
162 f2fs_put_page(page
, 0);
165 f2fs_submit_merged_bio(sbi
, META
, READ
);
166 return blkno
- start
;
169 void ra_meta_pages_cond(struct f2fs_sb_info
*sbi
, pgoff_t index
)
172 bool readahead
= false;
174 page
= find_get_page(META_MAPPING(sbi
), index
);
175 if (!page
|| (page
&& !PageUptodate(page
)))
177 f2fs_put_page(page
, 0);
180 ra_meta_pages(sbi
, index
, MAX_BIO_BLOCKS(sbi
), META_POR
);
183 static int f2fs_write_meta_page(struct page
*page
,
184 struct writeback_control
*wbc
)
186 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
188 trace_f2fs_writepage(page
, META
);
190 if (unlikely(sbi
->por_doing
))
192 if (wbc
->for_reclaim
&& page
->index
< GET_SUM_BLOCK(sbi
, 0))
194 if (unlikely(f2fs_cp_error(sbi
)))
197 f2fs_wait_on_page_writeback(page
, META
);
198 write_meta_page(sbi
, page
);
199 dec_page_count(sbi
, F2FS_DIRTY_META
);
202 if (wbc
->for_reclaim
)
203 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
207 redirty_page_for_writepage(wbc
, page
);
208 return AOP_WRITEPAGE_ACTIVATE
;
211 static int f2fs_write_meta_pages(struct address_space
*mapping
,
212 struct writeback_control
*wbc
)
214 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
217 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
219 /* collect a number of dirty meta pages and write together */
220 if (wbc
->for_kupdate
||
221 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
224 /* if mounting is failed, skip writing node pages */
225 mutex_lock(&sbi
->cp_mutex
);
226 diff
= nr_pages_to_write(sbi
, META
, wbc
);
227 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
228 mutex_unlock(&sbi
->cp_mutex
);
229 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
233 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
237 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
240 struct address_space
*mapping
= META_MAPPING(sbi
);
241 pgoff_t index
= 0, end
= LONG_MAX
;
244 struct writeback_control wbc
= {
248 pagevec_init(&pvec
, 0);
250 while (index
<= end
) {
252 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
254 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
255 if (unlikely(nr_pages
== 0))
258 for (i
= 0; i
< nr_pages
; i
++) {
259 struct page
*page
= pvec
.pages
[i
];
263 if (unlikely(page
->mapping
!= mapping
)) {
268 if (!PageDirty(page
)) {
269 /* someone wrote it for us */
270 goto continue_unlock
;
273 if (!clear_page_dirty_for_io(page
))
274 goto continue_unlock
;
276 if (f2fs_write_meta_page(page
, &wbc
)) {
281 if (unlikely(nwritten
>= nr_to_write
))
284 pagevec_release(&pvec
);
289 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
294 static int f2fs_set_meta_page_dirty(struct page
*page
)
296 trace_f2fs_set_page_dirty(page
, META
);
298 SetPageUptodate(page
);
299 if (!PageDirty(page
)) {
300 __set_page_dirty_nobuffers(page
);
301 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_META
);
307 const struct address_space_operations f2fs_meta_aops
= {
308 .writepage
= f2fs_write_meta_page
,
309 .writepages
= f2fs_write_meta_pages
,
310 .set_page_dirty
= f2fs_set_meta_page_dirty
,
313 static void __add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
315 struct inode_management
*im
= &sbi
->im
[type
];
318 if (radix_tree_preload(GFP_NOFS
)) {
323 spin_lock(&im
->ino_lock
);
325 e
= radix_tree_lookup(&im
->ino_root
, ino
);
327 e
= kmem_cache_alloc(ino_entry_slab
, GFP_ATOMIC
);
329 spin_unlock(&im
->ino_lock
);
330 radix_tree_preload_end();
333 if (radix_tree_insert(&im
->ino_root
, ino
, e
)) {
334 spin_unlock(&im
->ino_lock
);
335 kmem_cache_free(ino_entry_slab
, e
);
336 radix_tree_preload_end();
339 memset(e
, 0, sizeof(struct ino_entry
));
342 list_add_tail(&e
->list
, &im
->ino_list
);
343 if (type
!= ORPHAN_INO
)
346 spin_unlock(&im
->ino_lock
);
347 radix_tree_preload_end();
350 static void __remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
352 struct inode_management
*im
= &sbi
->im
[type
];
355 spin_lock(&im
->ino_lock
);
356 e
= radix_tree_lookup(&im
->ino_root
, ino
);
359 radix_tree_delete(&im
->ino_root
, ino
);
361 spin_unlock(&im
->ino_lock
);
362 kmem_cache_free(ino_entry_slab
, e
);
365 spin_unlock(&im
->ino_lock
);
368 void add_dirty_inode(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
370 /* add new dirty ino entry into list */
371 __add_ino_entry(sbi
, ino
, type
);
374 void remove_dirty_inode(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
376 /* remove dirty ino entry from list */
377 __remove_ino_entry(sbi
, ino
, type
);
380 /* mode should be APPEND_INO or UPDATE_INO */
381 bool exist_written_data(struct f2fs_sb_info
*sbi
, nid_t ino
, int mode
)
383 struct inode_management
*im
= &sbi
->im
[mode
];
386 spin_lock(&im
->ino_lock
);
387 e
= radix_tree_lookup(&im
->ino_root
, ino
);
388 spin_unlock(&im
->ino_lock
);
389 return e
? true : false;
392 void release_dirty_inode(struct f2fs_sb_info
*sbi
)
394 struct ino_entry
*e
, *tmp
;
397 for (i
= APPEND_INO
; i
<= UPDATE_INO
; i
++) {
398 struct inode_management
*im
= &sbi
->im
[i
];
400 spin_lock(&im
->ino_lock
);
401 list_for_each_entry_safe(e
, tmp
, &im
->ino_list
, list
) {
403 radix_tree_delete(&im
->ino_root
, e
->ino
);
404 kmem_cache_free(ino_entry_slab
, e
);
407 spin_unlock(&im
->ino_lock
);
411 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
413 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
416 spin_lock(&im
->ino_lock
);
417 if (unlikely(im
->ino_num
>= sbi
->max_orphans
))
421 spin_unlock(&im
->ino_lock
);
426 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
428 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
430 spin_lock(&im
->ino_lock
);
431 f2fs_bug_on(sbi
, im
->ino_num
== 0);
433 spin_unlock(&im
->ino_lock
);
436 void add_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
438 /* add new orphan ino entry into list */
439 __add_ino_entry(sbi
, ino
, ORPHAN_INO
);
442 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
444 /* remove orphan entry from orphan list */
445 __remove_ino_entry(sbi
, ino
, ORPHAN_INO
);
448 static void recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
450 struct inode
*inode
= f2fs_iget(sbi
->sb
, ino
);
451 f2fs_bug_on(sbi
, IS_ERR(inode
));
454 /* truncate all the data during iput */
458 void recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
460 block_t start_blk
, orphan_blkaddr
, i
, j
;
462 if (!is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
))
465 sbi
->por_doing
= true;
467 start_blk
= __start_cp_addr(sbi
) + 1 +
468 le32_to_cpu(F2FS_RAW_SUPER(sbi
)->cp_payload
);
469 orphan_blkaddr
= __start_sum_addr(sbi
) - 1;
471 ra_meta_pages(sbi
, start_blk
, orphan_blkaddr
, META_CP
);
473 for (i
= 0; i
< orphan_blkaddr
; i
++) {
474 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
475 struct f2fs_orphan_block
*orphan_blk
;
477 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
478 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
479 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
480 recover_orphan_inode(sbi
, ino
);
482 f2fs_put_page(page
, 1);
484 /* clear Orphan Flag */
485 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
486 sbi
->por_doing
= false;
490 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
492 struct list_head
*head
;
493 struct f2fs_orphan_block
*orphan_blk
= NULL
;
494 unsigned int nentries
= 0;
495 unsigned short index
;
496 unsigned short orphan_blocks
;
497 struct page
*page
= NULL
;
498 struct ino_entry
*orphan
= NULL
;
499 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
501 orphan_blocks
= GET_ORPHAN_BLOCKS(im
->ino_num
);
503 for (index
= 0; index
< orphan_blocks
; index
++)
504 grab_meta_page(sbi
, start_blk
+ index
);
507 spin_lock(&im
->ino_lock
);
508 head
= &im
->ino_list
;
510 /* loop for each orphan inode entry and write them in Jornal block */
511 list_for_each_entry(orphan
, head
, list
) {
513 page
= find_get_page(META_MAPPING(sbi
), start_blk
++);
514 f2fs_bug_on(sbi
, !page
);
516 (struct f2fs_orphan_block
*)page_address(page
);
517 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
518 f2fs_put_page(page
, 0);
521 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
523 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
525 * an orphan block is full of 1020 entries,
526 * then we need to flush current orphan blocks
527 * and bring another one in memory
529 orphan_blk
->blk_addr
= cpu_to_le16(index
);
530 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
531 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
532 set_page_dirty(page
);
533 f2fs_put_page(page
, 1);
541 orphan_blk
->blk_addr
= cpu_to_le16(index
);
542 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
543 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
544 set_page_dirty(page
);
545 f2fs_put_page(page
, 1);
548 spin_unlock(&im
->ino_lock
);
551 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
552 block_t cp_addr
, unsigned long long *version
)
554 struct page
*cp_page_1
, *cp_page_2
= NULL
;
555 unsigned long blk_size
= sbi
->blocksize
;
556 struct f2fs_checkpoint
*cp_block
;
557 unsigned long long cur_version
= 0, pre_version
= 0;
561 /* Read the 1st cp block in this CP pack */
562 cp_page_1
= get_meta_page(sbi
, cp_addr
);
564 /* get the version number */
565 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
566 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
567 if (crc_offset
>= blk_size
)
570 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
571 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
574 pre_version
= cur_cp_version(cp_block
);
576 /* Read the 2nd cp block in this CP pack */
577 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
578 cp_page_2
= get_meta_page(sbi
, cp_addr
);
580 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
581 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
582 if (crc_offset
>= blk_size
)
585 crc
= le32_to_cpu(*((__u32
*)((unsigned char *)cp_block
+ crc_offset
)));
586 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
589 cur_version
= cur_cp_version(cp_block
);
591 if (cur_version
== pre_version
) {
592 *version
= cur_version
;
593 f2fs_put_page(cp_page_2
, 1);
597 f2fs_put_page(cp_page_2
, 1);
599 f2fs_put_page(cp_page_1
, 1);
603 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
605 struct f2fs_checkpoint
*cp_block
;
606 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
607 struct page
*cp1
, *cp2
, *cur_page
;
608 unsigned long blk_size
= sbi
->blocksize
;
609 unsigned long long cp1_version
= 0, cp2_version
= 0;
610 unsigned long long cp_start_blk_no
;
611 unsigned int cp_blks
= 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi
)->cp_payload
);
615 sbi
->ckpt
= kzalloc(cp_blks
* blk_size
, GFP_KERNEL
);
619 * Finding out valid cp block involves read both
620 * sets( cp pack1 and cp pack 2)
622 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
623 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
625 /* The second checkpoint pack should start at the next segment */
626 cp_start_blk_no
+= ((unsigned long long)1) <<
627 le32_to_cpu(fsb
->log_blocks_per_seg
);
628 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
631 if (ver_after(cp2_version
, cp1_version
))
643 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
644 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
649 cp_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
651 cp_blk_no
+= 1 << le32_to_cpu(fsb
->log_blocks_per_seg
);
653 for (i
= 1; i
< cp_blks
; i
++) {
654 void *sit_bitmap_ptr
;
655 unsigned char *ckpt
= (unsigned char *)sbi
->ckpt
;
657 cur_page
= get_meta_page(sbi
, cp_blk_no
+ i
);
658 sit_bitmap_ptr
= page_address(cur_page
);
659 memcpy(ckpt
+ i
* blk_size
, sit_bitmap_ptr
, blk_size
);
660 f2fs_put_page(cur_page
, 1);
663 f2fs_put_page(cp1
, 1);
664 f2fs_put_page(cp2
, 1);
672 static int __add_dirty_inode(struct inode
*inode
, struct dir_inode_entry
*new)
674 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
676 if (is_inode_flag_set(F2FS_I(inode
), FI_DIRTY_DIR
))
679 set_inode_flag(F2FS_I(inode
), FI_DIRTY_DIR
);
680 F2FS_I(inode
)->dirty_dir
= new;
681 list_add_tail(&new->list
, &sbi
->dir_inode_list
);
682 stat_inc_dirty_dir(sbi
);
686 void update_dirty_page(struct inode
*inode
, struct page
*page
)
688 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
689 struct dir_inode_entry
*new;
692 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
))
695 if (!S_ISDIR(inode
->i_mode
)) {
696 inode_inc_dirty_pages(inode
);
700 new = f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
702 INIT_LIST_HEAD(&new->list
);
704 spin_lock(&sbi
->dir_inode_lock
);
705 ret
= __add_dirty_inode(inode
, new);
706 inode_inc_dirty_pages(inode
);
707 spin_unlock(&sbi
->dir_inode_lock
);
710 kmem_cache_free(inode_entry_slab
, new);
712 SetPagePrivate(page
);
715 void add_dirty_dir_inode(struct inode
*inode
)
717 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
718 struct dir_inode_entry
*new =
719 f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
723 INIT_LIST_HEAD(&new->list
);
725 spin_lock(&sbi
->dir_inode_lock
);
726 ret
= __add_dirty_inode(inode
, new);
727 spin_unlock(&sbi
->dir_inode_lock
);
730 kmem_cache_free(inode_entry_slab
, new);
733 void remove_dirty_dir_inode(struct inode
*inode
)
735 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
736 struct dir_inode_entry
*entry
;
738 if (!S_ISDIR(inode
->i_mode
))
741 spin_lock(&sbi
->dir_inode_lock
);
742 if (get_dirty_pages(inode
) ||
743 !is_inode_flag_set(F2FS_I(inode
), FI_DIRTY_DIR
)) {
744 spin_unlock(&sbi
->dir_inode_lock
);
748 entry
= F2FS_I(inode
)->dirty_dir
;
749 list_del(&entry
->list
);
750 F2FS_I(inode
)->dirty_dir
= NULL
;
751 clear_inode_flag(F2FS_I(inode
), FI_DIRTY_DIR
);
752 stat_dec_dirty_dir(sbi
);
753 spin_unlock(&sbi
->dir_inode_lock
);
754 kmem_cache_free(inode_entry_slab
, entry
);
756 /* Only from the recovery routine */
757 if (is_inode_flag_set(F2FS_I(inode
), FI_DELAY_IPUT
)) {
758 clear_inode_flag(F2FS_I(inode
), FI_DELAY_IPUT
);
763 void sync_dirty_dir_inodes(struct f2fs_sb_info
*sbi
)
765 struct list_head
*head
;
766 struct dir_inode_entry
*entry
;
769 if (unlikely(f2fs_cp_error(sbi
)))
772 spin_lock(&sbi
->dir_inode_lock
);
774 head
= &sbi
->dir_inode_list
;
775 if (list_empty(head
)) {
776 spin_unlock(&sbi
->dir_inode_lock
);
779 entry
= list_entry(head
->next
, struct dir_inode_entry
, list
);
780 inode
= igrab(entry
->inode
);
781 spin_unlock(&sbi
->dir_inode_lock
);
783 filemap_fdatawrite(inode
->i_mapping
);
787 * We should submit bio, since it exists several
788 * wribacking dentry pages in the freeing inode.
790 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
796 * Freeze all the FS-operations for checkpoint.
798 static int block_operations(struct f2fs_sb_info
*sbi
)
800 struct writeback_control wbc
= {
801 .sync_mode
= WB_SYNC_ALL
,
802 .nr_to_write
= LONG_MAX
,
805 struct blk_plug plug
;
808 blk_start_plug(&plug
);
812 /* write all the dirty dentry pages */
813 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
814 f2fs_unlock_all(sbi
);
815 sync_dirty_dir_inodes(sbi
);
816 if (unlikely(f2fs_cp_error(sbi
))) {
820 goto retry_flush_dents
;
824 * POR: we should ensure that there are no dirty node pages
825 * until finishing nat/sit flush.
828 down_write(&sbi
->node_write
);
830 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
831 up_write(&sbi
->node_write
);
832 sync_node_pages(sbi
, 0, &wbc
);
833 if (unlikely(f2fs_cp_error(sbi
))) {
834 f2fs_unlock_all(sbi
);
838 goto retry_flush_nodes
;
841 blk_finish_plug(&plug
);
845 static void unblock_operations(struct f2fs_sb_info
*sbi
)
847 up_write(&sbi
->node_write
);
848 f2fs_unlock_all(sbi
);
851 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
856 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
858 if (!get_pages(sbi
, F2FS_WRITEBACK
))
863 finish_wait(&sbi
->cp_wait
, &wait
);
866 static void do_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
868 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
869 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_WARM_NODE
);
870 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
871 unsigned long orphan_num
= sbi
->im
[ORPHAN_INO
].ino_num
;
872 nid_t last_nid
= nm_i
->next_scan_nid
;
874 struct page
*cp_page
;
875 unsigned int data_sum_blocks
, orphan_blocks
;
879 int cp_payload_blks
= le32_to_cpu(F2FS_RAW_SUPER(sbi
)->cp_payload
);
882 * This avoids to conduct wrong roll-forward operations and uses
883 * metapages, so should be called prior to sync_meta_pages below.
885 discard_next_dnode(sbi
, NEXT_FREE_BLKADDR(sbi
, curseg
));
887 /* Flush all the NAT/SIT pages */
888 while (get_pages(sbi
, F2FS_DIRTY_META
)) {
889 sync_meta_pages(sbi
, META
, LONG_MAX
);
890 if (unlikely(f2fs_cp_error(sbi
)))
894 next_free_nid(sbi
, &last_nid
);
898 * version number is already updated
900 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
901 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
902 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
903 for (i
= 0; i
< NR_CURSEG_NODE_TYPE
; i
++) {
904 ckpt
->cur_node_segno
[i
] =
905 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
906 ckpt
->cur_node_blkoff
[i
] =
907 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
908 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
909 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
911 for (i
= 0; i
< NR_CURSEG_DATA_TYPE
; i
++) {
912 ckpt
->cur_data_segno
[i
] =
913 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
914 ckpt
->cur_data_blkoff
[i
] =
915 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
916 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
917 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
920 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
921 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
922 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
924 /* 2 cp + n data seg summary + orphan inode blocks */
925 data_sum_blocks
= npages_for_summary_flush(sbi
);
926 if (data_sum_blocks
< NR_CURSEG_DATA_TYPE
)
927 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
929 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
931 orphan_blocks
= GET_ORPHAN_BLOCKS(orphan_num
);
932 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + cp_payload_blks
+
935 if (cpc
->reason
== CP_UMOUNT
) {
936 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
937 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
938 cp_payload_blks
+ data_sum_blocks
+
939 orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
941 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
942 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
943 cp_payload_blks
+ data_sum_blocks
+
948 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
950 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
953 set_ckpt_flags(ckpt
, CP_FSCK_FLAG
);
955 /* update SIT/NAT bitmap */
956 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
957 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
959 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
960 *((__le32
*)((unsigned char *)ckpt
+
961 le32_to_cpu(ckpt
->checksum_offset
)))
962 = cpu_to_le32(crc32
);
964 start_blk
= __start_cp_addr(sbi
);
966 /* write out checkpoint buffer at block 0 */
967 cp_page
= grab_meta_page(sbi
, start_blk
++);
968 kaddr
= page_address(cp_page
);
969 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
970 set_page_dirty(cp_page
);
971 f2fs_put_page(cp_page
, 1);
973 for (i
= 1; i
< 1 + cp_payload_blks
; i
++) {
974 cp_page
= grab_meta_page(sbi
, start_blk
++);
975 kaddr
= page_address(cp_page
);
976 memcpy(kaddr
, (char *)ckpt
+ i
* F2FS_BLKSIZE
,
977 (1 << sbi
->log_blocksize
));
978 set_page_dirty(cp_page
);
979 f2fs_put_page(cp_page
, 1);
983 write_orphan_inodes(sbi
, start_blk
);
984 start_blk
+= orphan_blocks
;
987 write_data_summaries(sbi
, start_blk
);
988 start_blk
+= data_sum_blocks
;
989 if (cpc
->reason
== CP_UMOUNT
) {
990 write_node_summaries(sbi
, start_blk
);
991 start_blk
+= NR_CURSEG_NODE_TYPE
;
994 /* writeout checkpoint block */
995 cp_page
= grab_meta_page(sbi
, start_blk
);
996 kaddr
= page_address(cp_page
);
997 memcpy(kaddr
, ckpt
, (1 << sbi
->log_blocksize
));
998 set_page_dirty(cp_page
);
999 f2fs_put_page(cp_page
, 1);
1001 /* wait for previous submitted node/meta pages writeback */
1002 wait_on_all_pages_writeback(sbi
);
1004 if (unlikely(f2fs_cp_error(sbi
)))
1007 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
1008 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
1010 /* update user_block_counts */
1011 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
1012 sbi
->alloc_valid_block_count
= 0;
1014 /* Here, we only have one bio having CP pack */
1015 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
1017 /* wait for previous submitted meta pages writeback */
1018 wait_on_all_pages_writeback(sbi
);
1020 release_dirty_inode(sbi
);
1022 if (unlikely(f2fs_cp_error(sbi
)))
1025 clear_prefree_segments(sbi
);
1026 F2FS_RESET_SB_DIRT(sbi
);
1030 * We guarantee that this checkpoint procedure will not fail.
1032 void write_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1034 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1035 unsigned long long ckpt_ver
;
1037 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "start block_ops");
1039 mutex_lock(&sbi
->cp_mutex
);
1041 if (!sbi
->s_dirty
&& cpc
->reason
!= CP_DISCARD
)
1043 if (unlikely(f2fs_cp_error(sbi
)))
1045 if (block_operations(sbi
))
1048 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish block_ops");
1050 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1051 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1052 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
1055 * update checkpoint pack index
1056 * Increase the version number so that
1057 * SIT entries and seg summaries are written at correct place
1059 ckpt_ver
= cur_cp_version(ckpt
);
1060 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
1062 /* write cached NAT/SIT entries to NAT/SIT area */
1063 flush_nat_entries(sbi
);
1064 flush_sit_entries(sbi
, cpc
);
1066 /* unlock all the fs_lock[] in do_checkpoint() */
1067 do_checkpoint(sbi
, cpc
);
1069 unblock_operations(sbi
);
1070 stat_inc_cp_count(sbi
->stat_info
);
1072 mutex_unlock(&sbi
->cp_mutex
);
1073 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish checkpoint");
1076 void init_ino_entry_info(struct f2fs_sb_info
*sbi
)
1080 for (i
= 0; i
< MAX_INO_ENTRY
; i
++) {
1081 struct inode_management
*im
= &sbi
->im
[i
];
1083 INIT_RADIX_TREE(&im
->ino_root
, GFP_ATOMIC
);
1084 spin_lock_init(&im
->ino_lock
);
1085 INIT_LIST_HEAD(&im
->ino_list
);
1090 * considering 512 blocks in a segment 8 blocks are needed for cp
1091 * and log segment summaries. Remaining blocks are used to keep
1092 * orphan entries with the limitation one reserved segment
1093 * for cp pack we can have max 1020*504 orphan entries
1095 sbi
->max_orphans
= (sbi
->blocks_per_seg
- F2FS_CP_PACKS
-
1096 NR_CURSEG_TYPE
) * F2FS_ORPHANS_PER_BLOCK
;
1099 int __init
create_checkpoint_caches(void)
1101 ino_entry_slab
= f2fs_kmem_cache_create("f2fs_ino_entry",
1102 sizeof(struct ino_entry
));
1103 if (!ino_entry_slab
)
1105 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1106 sizeof(struct dir_inode_entry
));
1107 if (!inode_entry_slab
) {
1108 kmem_cache_destroy(ino_entry_slab
);
1114 void destroy_checkpoint_caches(void)
1116 kmem_cache_destroy(ino_entry_slab
);
1117 kmem_cache_destroy(inode_entry_slab
);