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>
24 #include <trace/events/f2fs.h>
26 static struct kmem_cache
*ino_entry_slab
;
27 struct kmem_cache
*inode_entry_slab
;
29 void f2fs_stop_checkpoint(struct f2fs_sb_info
*sbi
, bool end_io
)
31 set_ckpt_flags(sbi
->ckpt
, CP_ERROR_FLAG
);
32 sbi
->sb
->s_flags
|= MS_RDONLY
;
34 f2fs_flush_merged_bios(sbi
);
38 * We guarantee no failure on the returned page.
40 struct page
*grab_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
42 struct address_space
*mapping
= META_MAPPING(sbi
);
43 struct page
*page
= NULL
;
45 page
= f2fs_grab_cache_page(mapping
, index
, false);
50 f2fs_wait_on_page_writeback(page
, META
, true);
51 if (!PageUptodate(page
))
52 SetPageUptodate(page
);
57 * We guarantee no failure on the returned page.
59 static struct page
*__get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
,
62 struct address_space
*mapping
= META_MAPPING(sbi
);
64 struct f2fs_io_info fio
= {
67 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
,
70 .encrypted_page
= NULL
,
73 if (unlikely(!is_meta
))
76 page
= f2fs_grab_cache_page(mapping
, index
, false);
81 if (PageUptodate(page
))
86 if (f2fs_submit_page_bio(&fio
)) {
87 f2fs_put_page(page
, 1);
92 if (unlikely(page
->mapping
!= mapping
)) {
93 f2fs_put_page(page
, 1);
98 * if there is any IO error when accessing device, make our filesystem
99 * readonly and make sure do not write checkpoint with non-uptodate
102 if (unlikely(!PageUptodate(page
)))
103 f2fs_stop_checkpoint(sbi
, false);
108 struct page
*get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
110 return __get_meta_page(sbi
, index
, true);
114 struct page
*get_tmp_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
116 return __get_meta_page(sbi
, index
, false);
119 bool is_valid_blkaddr(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int type
)
125 if (unlikely(blkaddr
>= SIT_BLK_CNT(sbi
)))
129 if (unlikely(blkaddr
>= MAIN_BLKADDR(sbi
) ||
130 blkaddr
< SM_I(sbi
)->ssa_blkaddr
))
134 if (unlikely(blkaddr
>= SIT_I(sbi
)->sit_base_addr
||
135 blkaddr
< __start_cp_addr(sbi
)))
139 if (unlikely(blkaddr
>= MAX_BLKADDR(sbi
) ||
140 blkaddr
< MAIN_BLKADDR(sbi
)))
151 * Readahead CP/NAT/SIT/SSA pages
153 int ra_meta_pages(struct f2fs_sb_info
*sbi
, block_t start
, int nrpages
,
157 block_t blkno
= start
;
158 struct f2fs_io_info fio
= {
161 .rw
= sync
? (READ_SYNC
| REQ_META
| REQ_PRIO
) : READA
,
162 .encrypted_page
= NULL
,
164 struct blk_plug plug
;
166 if (unlikely(type
== META_POR
))
169 blk_start_plug(&plug
);
170 for (; nrpages
-- > 0; blkno
++) {
172 if (!is_valid_blkaddr(sbi
, blkno
, type
))
177 if (unlikely(blkno
>=
178 NAT_BLOCK_OFFSET(NM_I(sbi
)->max_nid
)))
180 /* get nat block addr */
181 fio
.new_blkaddr
= current_nat_addr(sbi
,
182 blkno
* NAT_ENTRY_PER_BLOCK
);
185 /* get sit block addr */
186 fio
.new_blkaddr
= current_sit_addr(sbi
,
187 blkno
* SIT_ENTRY_PER_BLOCK
);
192 fio
.new_blkaddr
= blkno
;
198 page
= f2fs_grab_cache_page(META_MAPPING(sbi
),
199 fio
.new_blkaddr
, false);
202 if (PageUptodate(page
)) {
203 f2fs_put_page(page
, 1);
208 fio
.old_blkaddr
= fio
.new_blkaddr
;
209 f2fs_submit_page_mbio(&fio
);
210 f2fs_put_page(page
, 0);
213 f2fs_submit_merged_bio(sbi
, META
, READ
);
214 blk_finish_plug(&plug
);
215 return blkno
- start
;
218 void ra_meta_pages_cond(struct f2fs_sb_info
*sbi
, pgoff_t index
)
221 bool readahead
= false;
223 page
= find_get_page(META_MAPPING(sbi
), index
);
224 if (!page
|| !PageUptodate(page
))
226 f2fs_put_page(page
, 0);
229 ra_meta_pages(sbi
, index
, MAX_BIO_BLOCKS(sbi
), META_POR
, true);
232 static int f2fs_write_meta_page(struct page
*page
,
233 struct writeback_control
*wbc
)
235 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
237 trace_f2fs_writepage(page
, META
);
239 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
241 if (wbc
->for_reclaim
&& page
->index
< GET_SUM_BLOCK(sbi
, 0))
243 if (unlikely(f2fs_cp_error(sbi
)))
246 write_meta_page(sbi
, page
);
247 dec_page_count(sbi
, F2FS_DIRTY_META
);
249 if (wbc
->for_reclaim
)
250 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, META
, WRITE
);
254 if (unlikely(f2fs_cp_error(sbi
)))
255 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
260 redirty_page_for_writepage(wbc
, page
);
261 return AOP_WRITEPAGE_ACTIVATE
;
264 static int f2fs_write_meta_pages(struct address_space
*mapping
,
265 struct writeback_control
*wbc
)
267 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
270 /* collect a number of dirty meta pages and write together */
271 if (wbc
->for_kupdate
||
272 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
275 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
277 /* if mounting is failed, skip writing node pages */
278 mutex_lock(&sbi
->cp_mutex
);
279 diff
= nr_pages_to_write(sbi
, META
, wbc
);
280 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
281 mutex_unlock(&sbi
->cp_mutex
);
282 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
286 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
287 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
291 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
294 struct address_space
*mapping
= META_MAPPING(sbi
);
295 pgoff_t index
= 0, end
= ULONG_MAX
, prev
= ULONG_MAX
;
298 struct writeback_control wbc
= {
301 struct blk_plug plug
;
303 pagevec_init(&pvec
, 0);
305 blk_start_plug(&plug
);
307 while (index
<= end
) {
309 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
311 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
312 if (unlikely(nr_pages
== 0))
315 for (i
= 0; i
< nr_pages
; i
++) {
316 struct page
*page
= pvec
.pages
[i
];
318 if (prev
== ULONG_MAX
)
319 prev
= page
->index
- 1;
320 if (nr_to_write
!= LONG_MAX
&& page
->index
!= prev
+ 1) {
321 pagevec_release(&pvec
);
327 if (unlikely(page
->mapping
!= mapping
)) {
332 if (!PageDirty(page
)) {
333 /* someone wrote it for us */
334 goto continue_unlock
;
337 f2fs_wait_on_page_writeback(page
, META
, true);
339 BUG_ON(PageWriteback(page
));
340 if (!clear_page_dirty_for_io(page
))
341 goto continue_unlock
;
343 if (mapping
->a_ops
->writepage(page
, &wbc
)) {
349 if (unlikely(nwritten
>= nr_to_write
))
352 pagevec_release(&pvec
);
357 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
359 blk_finish_plug(&plug
);
364 static int f2fs_set_meta_page_dirty(struct page
*page
)
366 trace_f2fs_set_page_dirty(page
, META
);
368 if (!PageUptodate(page
))
369 SetPageUptodate(page
);
370 if (!PageDirty(page
)) {
371 f2fs_set_page_dirty_nobuffers(page
);
372 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_META
);
373 SetPagePrivate(page
);
374 f2fs_trace_pid(page
);
380 const struct address_space_operations f2fs_meta_aops
= {
381 .writepage
= f2fs_write_meta_page
,
382 .writepages
= f2fs_write_meta_pages
,
383 .set_page_dirty
= f2fs_set_meta_page_dirty
,
384 .invalidatepage
= f2fs_invalidate_page
,
385 .releasepage
= f2fs_release_page
,
388 static void __add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
390 struct inode_management
*im
= &sbi
->im
[type
];
391 struct ino_entry
*e
, *tmp
;
393 tmp
= f2fs_kmem_cache_alloc(ino_entry_slab
, GFP_NOFS
);
395 radix_tree_preload(GFP_NOFS
| __GFP_NOFAIL
);
397 spin_lock(&im
->ino_lock
);
398 e
= radix_tree_lookup(&im
->ino_root
, ino
);
401 if (radix_tree_insert(&im
->ino_root
, ino
, e
)) {
402 spin_unlock(&im
->ino_lock
);
403 radix_tree_preload_end();
406 memset(e
, 0, sizeof(struct ino_entry
));
409 list_add_tail(&e
->list
, &im
->ino_list
);
410 if (type
!= ORPHAN_INO
)
413 spin_unlock(&im
->ino_lock
);
414 radix_tree_preload_end();
417 kmem_cache_free(ino_entry_slab
, tmp
);
420 static void __remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
422 struct inode_management
*im
= &sbi
->im
[type
];
425 spin_lock(&im
->ino_lock
);
426 e
= radix_tree_lookup(&im
->ino_root
, ino
);
429 radix_tree_delete(&im
->ino_root
, ino
);
431 spin_unlock(&im
->ino_lock
);
432 kmem_cache_free(ino_entry_slab
, e
);
435 spin_unlock(&im
->ino_lock
);
438 void add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
440 /* add new dirty ino entry into list */
441 __add_ino_entry(sbi
, ino
, type
);
444 void remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
446 /* remove dirty ino entry from list */
447 __remove_ino_entry(sbi
, ino
, type
);
450 /* mode should be APPEND_INO or UPDATE_INO */
451 bool exist_written_data(struct f2fs_sb_info
*sbi
, nid_t ino
, int mode
)
453 struct inode_management
*im
= &sbi
->im
[mode
];
456 spin_lock(&im
->ino_lock
);
457 e
= radix_tree_lookup(&im
->ino_root
, ino
);
458 spin_unlock(&im
->ino_lock
);
459 return e
? true : false;
462 void release_ino_entry(struct f2fs_sb_info
*sbi
, bool all
)
464 struct ino_entry
*e
, *tmp
;
467 for (i
= all
? ORPHAN_INO
: APPEND_INO
; i
<= UPDATE_INO
; i
++) {
468 struct inode_management
*im
= &sbi
->im
[i
];
470 spin_lock(&im
->ino_lock
);
471 list_for_each_entry_safe(e
, tmp
, &im
->ino_list
, list
) {
473 radix_tree_delete(&im
->ino_root
, e
->ino
);
474 kmem_cache_free(ino_entry_slab
, e
);
477 spin_unlock(&im
->ino_lock
);
481 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
483 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
486 spin_lock(&im
->ino_lock
);
488 #ifdef CONFIG_F2FS_FAULT_INJECTION
489 if (time_to_inject(FAULT_ORPHAN
)) {
490 spin_unlock(&im
->ino_lock
);
494 if (unlikely(im
->ino_num
>= sbi
->max_orphans
))
498 spin_unlock(&im
->ino_lock
);
503 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
505 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
507 spin_lock(&im
->ino_lock
);
508 f2fs_bug_on(sbi
, im
->ino_num
== 0);
510 spin_unlock(&im
->ino_lock
);
513 void add_orphan_inode(struct inode
*inode
)
515 /* add new orphan ino entry into list */
516 __add_ino_entry(F2FS_I_SB(inode
), inode
->i_ino
, ORPHAN_INO
);
517 update_inode_page(inode
);
520 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
522 /* remove orphan entry from orphan list */
523 __remove_ino_entry(sbi
, ino
, ORPHAN_INO
);
526 static int recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
530 inode
= f2fs_iget(sbi
->sb
, ino
);
533 * there should be a bug that we can't find the entry
536 f2fs_bug_on(sbi
, PTR_ERR(inode
) == -ENOENT
);
537 return PTR_ERR(inode
);
542 /* truncate all the data during iput */
547 int recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
549 block_t start_blk
, orphan_blocks
, i
, j
;
552 if (!is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
))
555 start_blk
= __start_cp_addr(sbi
) + 1 + __cp_payload(sbi
);
556 orphan_blocks
= __start_sum_addr(sbi
) - 1 - __cp_payload(sbi
);
558 ra_meta_pages(sbi
, start_blk
, orphan_blocks
, META_CP
, true);
560 for (i
= 0; i
< orphan_blocks
; i
++) {
561 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
562 struct f2fs_orphan_block
*orphan_blk
;
564 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
565 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
566 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
567 err
= recover_orphan_inode(sbi
, ino
);
569 f2fs_put_page(page
, 1);
573 f2fs_put_page(page
, 1);
575 /* clear Orphan Flag */
576 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
580 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
582 struct list_head
*head
;
583 struct f2fs_orphan_block
*orphan_blk
= NULL
;
584 unsigned int nentries
= 0;
585 unsigned short index
= 1;
586 unsigned short orphan_blocks
;
587 struct page
*page
= NULL
;
588 struct ino_entry
*orphan
= NULL
;
589 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
591 orphan_blocks
= GET_ORPHAN_BLOCKS(im
->ino_num
);
594 * we don't need to do spin_lock(&im->ino_lock) here, since all the
595 * orphan inode operations are covered under f2fs_lock_op().
596 * And, spin_lock should be avoided due to page operations below.
598 head
= &im
->ino_list
;
600 /* loop for each orphan inode entry and write them in Jornal block */
601 list_for_each_entry(orphan
, head
, list
) {
603 page
= grab_meta_page(sbi
, start_blk
++);
605 (struct f2fs_orphan_block
*)page_address(page
);
606 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
609 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
611 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
613 * an orphan block is full of 1020 entries,
614 * then we need to flush current orphan blocks
615 * and bring another one in memory
617 orphan_blk
->blk_addr
= cpu_to_le16(index
);
618 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
619 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
620 set_page_dirty(page
);
621 f2fs_put_page(page
, 1);
629 orphan_blk
->blk_addr
= cpu_to_le16(index
);
630 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
631 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
632 set_page_dirty(page
);
633 f2fs_put_page(page
, 1);
637 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
638 block_t cp_addr
, unsigned long long *version
)
640 struct page
*cp_page_1
, *cp_page_2
= NULL
;
641 unsigned long blk_size
= sbi
->blocksize
;
642 struct f2fs_checkpoint
*cp_block
;
643 unsigned long long cur_version
= 0, pre_version
= 0;
647 /* Read the 1st cp block in this CP pack */
648 cp_page_1
= get_meta_page(sbi
, cp_addr
);
650 /* get the version number */
651 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
652 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
653 if (crc_offset
>= blk_size
)
656 crc
= le32_to_cpu(*((__le32
*)((unsigned char *)cp_block
+ crc_offset
)));
657 if (!f2fs_crc_valid(sbi
, crc
, cp_block
, crc_offset
))
660 pre_version
= cur_cp_version(cp_block
);
662 /* Read the 2nd cp block in this CP pack */
663 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
664 cp_page_2
= get_meta_page(sbi
, cp_addr
);
666 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
667 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
668 if (crc_offset
>= blk_size
)
671 crc
= le32_to_cpu(*((__le32
*)((unsigned char *)cp_block
+ crc_offset
)));
672 if (!f2fs_crc_valid(sbi
, crc
, cp_block
, crc_offset
))
675 cur_version
= cur_cp_version(cp_block
);
677 if (cur_version
== pre_version
) {
678 *version
= cur_version
;
679 f2fs_put_page(cp_page_2
, 1);
683 f2fs_put_page(cp_page_2
, 1);
685 f2fs_put_page(cp_page_1
, 1);
689 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
691 struct f2fs_checkpoint
*cp_block
;
692 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
693 struct page
*cp1
, *cp2
, *cur_page
;
694 unsigned long blk_size
= sbi
->blocksize
;
695 unsigned long long cp1_version
= 0, cp2_version
= 0;
696 unsigned long long cp_start_blk_no
;
697 unsigned int cp_blks
= 1 + __cp_payload(sbi
);
701 sbi
->ckpt
= kzalloc(cp_blks
* blk_size
, GFP_KERNEL
);
705 * Finding out valid cp block involves read both
706 * sets( cp pack1 and cp pack 2)
708 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
709 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
711 /* The second checkpoint pack should start at the next segment */
712 cp_start_blk_no
+= ((unsigned long long)1) <<
713 le32_to_cpu(fsb
->log_blocks_per_seg
);
714 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
717 if (ver_after(cp2_version
, cp1_version
))
729 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
730 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
732 /* Sanity checking of checkpoint */
733 if (sanity_check_ckpt(sbi
))
739 cp_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
741 cp_blk_no
+= 1 << le32_to_cpu(fsb
->log_blocks_per_seg
);
743 for (i
= 1; i
< cp_blks
; i
++) {
744 void *sit_bitmap_ptr
;
745 unsigned char *ckpt
= (unsigned char *)sbi
->ckpt
;
747 cur_page
= get_meta_page(sbi
, cp_blk_no
+ i
);
748 sit_bitmap_ptr
= page_address(cur_page
);
749 memcpy(ckpt
+ i
* blk_size
, sit_bitmap_ptr
, blk_size
);
750 f2fs_put_page(cur_page
, 1);
753 f2fs_put_page(cp1
, 1);
754 f2fs_put_page(cp2
, 1);
762 static void __add_dirty_inode(struct inode
*inode
, enum inode_type type
)
764 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
765 int flag
= (type
== DIR_INODE
) ? FI_DIRTY_DIR
: FI_DIRTY_FILE
;
767 if (is_inode_flag_set(inode
, flag
))
770 set_inode_flag(inode
, flag
);
771 list_add_tail(&F2FS_I(inode
)->dirty_list
, &sbi
->inode_list
[type
]);
772 stat_inc_dirty_inode(sbi
, type
);
775 static void __remove_dirty_inode(struct inode
*inode
, enum inode_type type
)
777 int flag
= (type
== DIR_INODE
) ? FI_DIRTY_DIR
: FI_DIRTY_FILE
;
779 if (get_dirty_pages(inode
) || !is_inode_flag_set(inode
, flag
))
782 list_del_init(&F2FS_I(inode
)->dirty_list
);
783 clear_inode_flag(inode
, flag
);
784 stat_dec_dirty_inode(F2FS_I_SB(inode
), type
);
787 void update_dirty_page(struct inode
*inode
, struct page
*page
)
789 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
790 enum inode_type type
= S_ISDIR(inode
->i_mode
) ? DIR_INODE
: FILE_INODE
;
792 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
) &&
793 !S_ISLNK(inode
->i_mode
))
796 spin_lock(&sbi
->inode_lock
[type
]);
797 if (type
!= FILE_INODE
|| test_opt(sbi
, DATA_FLUSH
))
798 __add_dirty_inode(inode
, type
);
799 inode_inc_dirty_pages(inode
);
800 spin_unlock(&sbi
->inode_lock
[type
]);
802 SetPagePrivate(page
);
803 f2fs_trace_pid(page
);
806 void remove_dirty_inode(struct inode
*inode
)
808 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
809 enum inode_type type
= S_ISDIR(inode
->i_mode
) ? DIR_INODE
: FILE_INODE
;
811 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
) &&
812 !S_ISLNK(inode
->i_mode
))
815 if (type
== FILE_INODE
&& !test_opt(sbi
, DATA_FLUSH
))
818 spin_lock(&sbi
->inode_lock
[type
]);
819 __remove_dirty_inode(inode
, type
);
820 spin_unlock(&sbi
->inode_lock
[type
]);
823 int sync_dirty_inodes(struct f2fs_sb_info
*sbi
, enum inode_type type
)
825 struct list_head
*head
;
827 struct f2fs_inode_info
*fi
;
828 bool is_dir
= (type
== DIR_INODE
);
830 trace_f2fs_sync_dirty_inodes_enter(sbi
->sb
, is_dir
,
831 get_pages(sbi
, is_dir
?
832 F2FS_DIRTY_DENTS
: F2FS_DIRTY_DATA
));
834 if (unlikely(f2fs_cp_error(sbi
)))
837 spin_lock(&sbi
->inode_lock
[type
]);
839 head
= &sbi
->inode_list
[type
];
840 if (list_empty(head
)) {
841 spin_unlock(&sbi
->inode_lock
[type
]);
842 trace_f2fs_sync_dirty_inodes_exit(sbi
->sb
, is_dir
,
843 get_pages(sbi
, is_dir
?
844 F2FS_DIRTY_DENTS
: F2FS_DIRTY_DATA
));
847 fi
= list_entry(head
->next
, struct f2fs_inode_info
, dirty_list
);
848 inode
= igrab(&fi
->vfs_inode
);
849 spin_unlock(&sbi
->inode_lock
[type
]);
851 filemap_fdatawrite(inode
->i_mapping
);
855 * We should submit bio, since it exists several
856 * wribacking dentry pages in the freeing inode.
858 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
864 int f2fs_sync_inode_meta(struct f2fs_sb_info
*sbi
)
866 struct list_head
*head
= &sbi
->inode_list
[DIRTY_META
];
868 struct f2fs_inode_info
*fi
;
869 s64 total
= get_pages(sbi
, F2FS_DIRTY_IMETA
);
872 if (unlikely(f2fs_cp_error(sbi
)))
875 spin_lock(&sbi
->inode_lock
[DIRTY_META
]);
876 if (list_empty(head
)) {
877 spin_unlock(&sbi
->inode_lock
[DIRTY_META
]);
880 fi
= list_entry(head
->next
, struct f2fs_inode_info
,
882 inode
= igrab(&fi
->vfs_inode
);
883 spin_unlock(&sbi
->inode_lock
[DIRTY_META
]);
885 update_inode_page(inode
);
893 * Freeze all the FS-operations for checkpoint.
895 static int block_operations(struct f2fs_sb_info
*sbi
)
897 struct writeback_control wbc
= {
898 .sync_mode
= WB_SYNC_ALL
,
899 .nr_to_write
= LONG_MAX
,
906 /* write all the dirty dentry pages */
907 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
908 f2fs_unlock_all(sbi
);
909 err
= sync_dirty_inodes(sbi
, DIR_INODE
);
912 goto retry_flush_dents
;
915 if (get_pages(sbi
, F2FS_DIRTY_IMETA
)) {
916 f2fs_unlock_all(sbi
);
917 err
= f2fs_sync_inode_meta(sbi
);
920 goto retry_flush_dents
;
924 * POR: we should ensure that there are no dirty node pages
925 * until finishing nat/sit flush.
928 down_write(&sbi
->node_write
);
930 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
931 up_write(&sbi
->node_write
);
932 err
= sync_node_pages(sbi
, &wbc
);
934 f2fs_unlock_all(sbi
);
937 goto retry_flush_nodes
;
943 static void unblock_operations(struct f2fs_sb_info
*sbi
)
945 up_write(&sbi
->node_write
);
947 build_free_nids(sbi
);
948 f2fs_unlock_all(sbi
);
951 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
956 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
958 if (!atomic_read(&sbi
->nr_wb_bios
))
961 io_schedule_timeout(5*HZ
);
963 finish_wait(&sbi
->cp_wait
, &wait
);
966 static int do_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
968 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
969 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_WARM_NODE
);
970 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
971 unsigned long orphan_num
= sbi
->im
[ORPHAN_INO
].ino_num
;
972 nid_t last_nid
= nm_i
->next_scan_nid
;
974 unsigned int data_sum_blocks
, orphan_blocks
;
977 int cp_payload_blks
= __cp_payload(sbi
);
978 block_t discard_blk
= NEXT_FREE_BLKADDR(sbi
, curseg
);
979 bool invalidate
= false;
980 struct super_block
*sb
= sbi
->sb
;
981 struct curseg_info
*seg_i
= CURSEG_I(sbi
, CURSEG_HOT_NODE
);
985 * This avoids to conduct wrong roll-forward operations and uses
986 * metapages, so should be called prior to sync_meta_pages below.
988 if (!test_opt(sbi
, LFS
) && discard_next_dnode(sbi
, discard_blk
))
991 /* Flush all the NAT/SIT pages */
992 while (get_pages(sbi
, F2FS_DIRTY_META
)) {
993 sync_meta_pages(sbi
, META
, LONG_MAX
);
994 if (unlikely(f2fs_cp_error(sbi
)))
998 next_free_nid(sbi
, &last_nid
);
1002 * version number is already updated
1004 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
1005 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
1006 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
1007 for (i
= 0; i
< NR_CURSEG_NODE_TYPE
; i
++) {
1008 ckpt
->cur_node_segno
[i
] =
1009 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
1010 ckpt
->cur_node_blkoff
[i
] =
1011 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
1012 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
1013 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
1015 for (i
= 0; i
< NR_CURSEG_DATA_TYPE
; i
++) {
1016 ckpt
->cur_data_segno
[i
] =
1017 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
1018 ckpt
->cur_data_blkoff
[i
] =
1019 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
1020 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
1021 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
1024 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
1025 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
1026 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
1028 /* 2 cp + n data seg summary + orphan inode blocks */
1029 data_sum_blocks
= npages_for_summary_flush(sbi
, false);
1030 if (data_sum_blocks
< NR_CURSEG_DATA_TYPE
)
1031 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
1033 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
1035 orphan_blocks
= GET_ORPHAN_BLOCKS(orphan_num
);
1036 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + cp_payload_blks
+
1039 if (__remain_node_summaries(cpc
->reason
))
1040 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
1041 cp_payload_blks
+ data_sum_blocks
+
1042 orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
1044 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
1045 cp_payload_blks
+ data_sum_blocks
+
1048 if (cpc
->reason
== CP_UMOUNT
)
1049 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
1051 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
1053 if (cpc
->reason
== CP_FASTBOOT
)
1054 set_ckpt_flags(ckpt
, CP_FASTBOOT_FLAG
);
1056 clear_ckpt_flags(ckpt
, CP_FASTBOOT_FLAG
);
1059 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
1061 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
1063 if (is_sbi_flag_set(sbi
, SBI_NEED_FSCK
))
1064 set_ckpt_flags(ckpt
, CP_FSCK_FLAG
);
1066 /* update SIT/NAT bitmap */
1067 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
1068 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
1070 crc32
= f2fs_crc32(sbi
, ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
1071 *((__le32
*)((unsigned char *)ckpt
+
1072 le32_to_cpu(ckpt
->checksum_offset
)))
1073 = cpu_to_le32(crc32
);
1075 start_blk
= __start_cp_addr(sbi
);
1077 /* need to wait for end_io results */
1078 wait_on_all_pages_writeback(sbi
);
1079 if (unlikely(f2fs_cp_error(sbi
)))
1082 /* write out checkpoint buffer at block 0 */
1083 update_meta_page(sbi
, ckpt
, start_blk
++);
1085 for (i
= 1; i
< 1 + cp_payload_blks
; i
++)
1086 update_meta_page(sbi
, (char *)ckpt
+ i
* F2FS_BLKSIZE
,
1090 write_orphan_inodes(sbi
, start_blk
);
1091 start_blk
+= orphan_blocks
;
1094 write_data_summaries(sbi
, start_blk
);
1095 start_blk
+= data_sum_blocks
;
1097 /* Record write statistics in the hot node summary */
1098 kbytes_written
= sbi
->kbytes_written
;
1099 if (sb
->s_bdev
->bd_part
)
1100 kbytes_written
+= BD_PART_WRITTEN(sbi
);
1102 seg_i
->journal
->info
.kbytes_written
= cpu_to_le64(kbytes_written
);
1104 if (__remain_node_summaries(cpc
->reason
)) {
1105 write_node_summaries(sbi
, start_blk
);
1106 start_blk
+= NR_CURSEG_NODE_TYPE
;
1109 /* writeout checkpoint block */
1110 update_meta_page(sbi
, ckpt
, start_blk
);
1112 /* wait for previous submitted node/meta pages writeback */
1113 wait_on_all_pages_writeback(sbi
);
1115 if (unlikely(f2fs_cp_error(sbi
)))
1118 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LLONG_MAX
);
1119 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LLONG_MAX
);
1121 /* update user_block_counts */
1122 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
1123 percpu_counter_set(&sbi
->alloc_valid_block_count
, 0);
1125 /* Here, we only have one bio having CP pack */
1126 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
1128 /* wait for previous submitted meta pages writeback */
1129 wait_on_all_pages_writeback(sbi
);
1132 * invalidate meta page which is used temporarily for zeroing out
1133 * block at the end of warm node chain.
1136 invalidate_mapping_pages(META_MAPPING(sbi
), discard_blk
,
1139 release_ino_entry(sbi
, false);
1141 if (unlikely(f2fs_cp_error(sbi
)))
1144 clear_prefree_segments(sbi
, cpc
);
1145 clear_sbi_flag(sbi
, SBI_IS_DIRTY
);
1151 * We guarantee that this checkpoint procedure will not fail.
1153 int write_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1155 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1156 unsigned long long ckpt_ver
;
1159 mutex_lock(&sbi
->cp_mutex
);
1161 if (!is_sbi_flag_set(sbi
, SBI_IS_DIRTY
) &&
1162 (cpc
->reason
== CP_FASTBOOT
|| cpc
->reason
== CP_SYNC
||
1163 (cpc
->reason
== CP_DISCARD
&& !sbi
->discard_blks
)))
1165 if (unlikely(f2fs_cp_error(sbi
))) {
1169 if (f2fs_readonly(sbi
->sb
)) {
1174 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "start block_ops");
1176 err
= block_operations(sbi
);
1180 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish block_ops");
1182 f2fs_flush_merged_bios(sbi
);
1185 * update checkpoint pack index
1186 * Increase the version number so that
1187 * SIT entries and seg summaries are written at correct place
1189 ckpt_ver
= cur_cp_version(ckpt
);
1190 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
1192 /* write cached NAT/SIT entries to NAT/SIT area */
1193 flush_nat_entries(sbi
);
1194 flush_sit_entries(sbi
, cpc
);
1196 /* unlock all the fs_lock[] in do_checkpoint() */
1197 err
= do_checkpoint(sbi
, cpc
);
1199 unblock_operations(sbi
);
1200 stat_inc_cp_count(sbi
->stat_info
);
1202 if (cpc
->reason
== CP_RECOVERY
)
1203 f2fs_msg(sbi
->sb
, KERN_NOTICE
,
1204 "checkpoint: version = %llx", ckpt_ver
);
1206 /* do checkpoint periodically */
1207 f2fs_update_time(sbi
, CP_TIME
);
1208 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish checkpoint");
1210 mutex_unlock(&sbi
->cp_mutex
);
1214 void init_ino_entry_info(struct f2fs_sb_info
*sbi
)
1218 for (i
= 0; i
< MAX_INO_ENTRY
; i
++) {
1219 struct inode_management
*im
= &sbi
->im
[i
];
1221 INIT_RADIX_TREE(&im
->ino_root
, GFP_ATOMIC
);
1222 spin_lock_init(&im
->ino_lock
);
1223 INIT_LIST_HEAD(&im
->ino_list
);
1227 sbi
->max_orphans
= (sbi
->blocks_per_seg
- F2FS_CP_PACKS
-
1228 NR_CURSEG_TYPE
- __cp_payload(sbi
)) *
1229 F2FS_ORPHANS_PER_BLOCK
;
1232 int __init
create_checkpoint_caches(void)
1234 ino_entry_slab
= f2fs_kmem_cache_create("f2fs_ino_entry",
1235 sizeof(struct ino_entry
));
1236 if (!ino_entry_slab
)
1238 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_inode_entry",
1239 sizeof(struct inode_entry
));
1240 if (!inode_entry_slab
) {
1241 kmem_cache_destroy(ino_entry_slab
);
1247 void destroy_checkpoint_caches(void)
1249 kmem_cache_destroy(ino_entry_slab
);
1250 kmem_cache_destroy(inode_entry_slab
);