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
;
30 * We guarantee no failure on the returned page.
32 struct page
*grab_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
34 struct address_space
*mapping
= META_MAPPING(sbi
);
35 struct page
*page
= NULL
;
37 page
= grab_cache_page(mapping
, index
);
42 f2fs_wait_on_page_writeback(page
, META
, true);
43 SetPageUptodate(page
);
48 * We guarantee no failure on the returned page.
50 static struct page
*__get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
,
53 struct address_space
*mapping
= META_MAPPING(sbi
);
55 struct f2fs_io_info fio
= {
58 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
,
60 .encrypted_page
= NULL
,
63 if (unlikely(!is_meta
))
66 page
= grab_cache_page(mapping
, index
);
71 if (PageUptodate(page
))
76 if (f2fs_submit_page_bio(&fio
)) {
77 f2fs_put_page(page
, 1);
82 if (unlikely(page
->mapping
!= mapping
)) {
83 f2fs_put_page(page
, 1);
88 * if there is any IO error when accessing device, make our filesystem
89 * readonly and make sure do not write checkpoint with non-uptodate
92 if (unlikely(!PageUptodate(page
)))
93 f2fs_stop_checkpoint(sbi
);
98 struct page
*get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
100 return __get_meta_page(sbi
, index
, true);
104 struct page
*get_tmp_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
106 return __get_meta_page(sbi
, index
, false);
109 bool is_valid_blkaddr(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int type
)
115 if (unlikely(blkaddr
>= SIT_BLK_CNT(sbi
)))
119 if (unlikely(blkaddr
>= MAIN_BLKADDR(sbi
) ||
120 blkaddr
< SM_I(sbi
)->ssa_blkaddr
))
124 if (unlikely(blkaddr
>= SIT_I(sbi
)->sit_base_addr
||
125 blkaddr
< __start_cp_addr(sbi
)))
129 if (unlikely(blkaddr
>= MAX_BLKADDR(sbi
) ||
130 blkaddr
< MAIN_BLKADDR(sbi
)))
141 * Readahead CP/NAT/SIT/SSA pages
143 int ra_meta_pages(struct f2fs_sb_info
*sbi
, block_t start
, int nrpages
,
146 block_t prev_blk_addr
= 0;
148 block_t blkno
= start
;
149 struct f2fs_io_info fio
= {
152 .rw
= sync
? (READ_SYNC
| REQ_META
| REQ_PRIO
) : READA
,
153 .encrypted_page
= NULL
,
156 if (unlikely(type
== META_POR
))
159 for (; nrpages
-- > 0; blkno
++) {
161 if (!is_valid_blkaddr(sbi
, blkno
, type
))
166 if (unlikely(blkno
>=
167 NAT_BLOCK_OFFSET(NM_I(sbi
)->max_nid
)))
169 /* get nat block addr */
170 fio
.blk_addr
= current_nat_addr(sbi
,
171 blkno
* NAT_ENTRY_PER_BLOCK
);
174 /* get sit block addr */
175 fio
.blk_addr
= current_sit_addr(sbi
,
176 blkno
* SIT_ENTRY_PER_BLOCK
);
177 if (blkno
!= start
&& prev_blk_addr
+ 1 != fio
.blk_addr
)
179 prev_blk_addr
= fio
.blk_addr
;
184 fio
.blk_addr
= blkno
;
190 page
= grab_cache_page(META_MAPPING(sbi
), fio
.blk_addr
);
193 if (PageUptodate(page
)) {
194 f2fs_put_page(page
, 1);
199 f2fs_submit_page_mbio(&fio
);
200 f2fs_put_page(page
, 0);
203 f2fs_submit_merged_bio(sbi
, META
, READ
);
204 return blkno
- start
;
207 void ra_meta_pages_cond(struct f2fs_sb_info
*sbi
, pgoff_t index
)
210 bool readahead
= false;
212 page
= find_get_page(META_MAPPING(sbi
), index
);
213 if (!page
|| (page
&& !PageUptodate(page
)))
215 f2fs_put_page(page
, 0);
218 ra_meta_pages(sbi
, index
, MAX_BIO_BLOCKS(sbi
), META_POR
, true);
221 static int f2fs_write_meta_page(struct page
*page
,
222 struct writeback_control
*wbc
)
224 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
226 trace_f2fs_writepage(page
, META
);
228 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
230 if (wbc
->for_reclaim
&& page
->index
< GET_SUM_BLOCK(sbi
, 0))
232 if (unlikely(f2fs_cp_error(sbi
)))
235 write_meta_page(sbi
, page
);
236 dec_page_count(sbi
, F2FS_DIRTY_META
);
238 if (wbc
->for_reclaim
)
239 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, META
, WRITE
);
243 if (unlikely(f2fs_cp_error(sbi
)))
244 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
249 redirty_page_for_writepage(wbc
, page
);
250 return AOP_WRITEPAGE_ACTIVATE
;
253 static int f2fs_write_meta_pages(struct address_space
*mapping
,
254 struct writeback_control
*wbc
)
256 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
259 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
261 /* collect a number of dirty meta pages and write together */
262 if (wbc
->for_kupdate
||
263 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
266 /* if mounting is failed, skip writing node pages */
267 mutex_lock(&sbi
->cp_mutex
);
268 diff
= nr_pages_to_write(sbi
, META
, wbc
);
269 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
270 mutex_unlock(&sbi
->cp_mutex
);
271 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
275 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
279 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
282 struct address_space
*mapping
= META_MAPPING(sbi
);
283 pgoff_t index
= 0, end
= LONG_MAX
, prev
= LONG_MAX
;
286 struct writeback_control wbc
= {
290 pagevec_init(&pvec
, 0);
292 while (index
<= end
) {
294 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
296 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
297 if (unlikely(nr_pages
== 0))
300 for (i
= 0; i
< nr_pages
; i
++) {
301 struct page
*page
= pvec
.pages
[i
];
303 if (prev
== LONG_MAX
)
304 prev
= page
->index
- 1;
305 if (nr_to_write
!= LONG_MAX
&& page
->index
!= prev
+ 1) {
306 pagevec_release(&pvec
);
312 if (unlikely(page
->mapping
!= mapping
)) {
317 if (!PageDirty(page
)) {
318 /* someone wrote it for us */
319 goto continue_unlock
;
322 f2fs_wait_on_page_writeback(page
, META
, true);
324 BUG_ON(PageWriteback(page
));
325 if (!clear_page_dirty_for_io(page
))
326 goto continue_unlock
;
328 if (mapping
->a_ops
->writepage(page
, &wbc
)) {
334 if (unlikely(nwritten
>= nr_to_write
))
337 pagevec_release(&pvec
);
342 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
347 static int f2fs_set_meta_page_dirty(struct page
*page
)
349 trace_f2fs_set_page_dirty(page
, META
);
351 SetPageUptodate(page
);
352 if (!PageDirty(page
)) {
353 __set_page_dirty_nobuffers(page
);
354 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_META
);
355 SetPagePrivate(page
);
356 f2fs_trace_pid(page
);
362 const struct address_space_operations f2fs_meta_aops
= {
363 .writepage
= f2fs_write_meta_page
,
364 .writepages
= f2fs_write_meta_pages
,
365 .set_page_dirty
= f2fs_set_meta_page_dirty
,
366 .invalidatepage
= f2fs_invalidate_page
,
367 .releasepage
= f2fs_release_page
,
370 static void __add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
372 struct inode_management
*im
= &sbi
->im
[type
];
373 struct ino_entry
*e
, *tmp
;
375 tmp
= f2fs_kmem_cache_alloc(ino_entry_slab
, GFP_NOFS
);
377 radix_tree_preload(GFP_NOFS
| __GFP_NOFAIL
);
379 spin_lock(&im
->ino_lock
);
380 e
= radix_tree_lookup(&im
->ino_root
, ino
);
383 if (radix_tree_insert(&im
->ino_root
, ino
, e
)) {
384 spin_unlock(&im
->ino_lock
);
385 radix_tree_preload_end();
388 memset(e
, 0, sizeof(struct ino_entry
));
391 list_add_tail(&e
->list
, &im
->ino_list
);
392 if (type
!= ORPHAN_INO
)
395 spin_unlock(&im
->ino_lock
);
396 radix_tree_preload_end();
399 kmem_cache_free(ino_entry_slab
, tmp
);
402 static void __remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
404 struct inode_management
*im
= &sbi
->im
[type
];
407 spin_lock(&im
->ino_lock
);
408 e
= radix_tree_lookup(&im
->ino_root
, ino
);
411 radix_tree_delete(&im
->ino_root
, ino
);
413 spin_unlock(&im
->ino_lock
);
414 kmem_cache_free(ino_entry_slab
, e
);
417 spin_unlock(&im
->ino_lock
);
420 void add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
422 /* add new dirty ino entry into list */
423 __add_ino_entry(sbi
, ino
, type
);
426 void remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
428 /* remove dirty ino entry from list */
429 __remove_ino_entry(sbi
, ino
, type
);
432 /* mode should be APPEND_INO or UPDATE_INO */
433 bool exist_written_data(struct f2fs_sb_info
*sbi
, nid_t ino
, int mode
)
435 struct inode_management
*im
= &sbi
->im
[mode
];
438 spin_lock(&im
->ino_lock
);
439 e
= radix_tree_lookup(&im
->ino_root
, ino
);
440 spin_unlock(&im
->ino_lock
);
441 return e
? true : false;
444 void release_ino_entry(struct f2fs_sb_info
*sbi
)
446 struct ino_entry
*e
, *tmp
;
449 for (i
= APPEND_INO
; i
<= UPDATE_INO
; i
++) {
450 struct inode_management
*im
= &sbi
->im
[i
];
452 spin_lock(&im
->ino_lock
);
453 list_for_each_entry_safe(e
, tmp
, &im
->ino_list
, list
) {
455 radix_tree_delete(&im
->ino_root
, e
->ino
);
456 kmem_cache_free(ino_entry_slab
, e
);
459 spin_unlock(&im
->ino_lock
);
463 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
465 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
468 spin_lock(&im
->ino_lock
);
469 if (unlikely(im
->ino_num
>= sbi
->max_orphans
))
473 spin_unlock(&im
->ino_lock
);
478 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
480 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
482 spin_lock(&im
->ino_lock
);
483 f2fs_bug_on(sbi
, im
->ino_num
== 0);
485 spin_unlock(&im
->ino_lock
);
488 void add_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
490 /* add new orphan ino entry into list */
491 __add_ino_entry(sbi
, ino
, ORPHAN_INO
);
494 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
496 /* remove orphan entry from orphan list */
497 __remove_ino_entry(sbi
, ino
, ORPHAN_INO
);
500 static int recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
504 inode
= f2fs_iget(sbi
->sb
, ino
);
507 * there should be a bug that we can't find the entry
510 f2fs_bug_on(sbi
, PTR_ERR(inode
) == -ENOENT
);
511 return PTR_ERR(inode
);
516 /* truncate all the data during iput */
521 int recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
523 block_t start_blk
, orphan_blocks
, i
, j
;
526 if (!is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
))
529 start_blk
= __start_cp_addr(sbi
) + 1 + __cp_payload(sbi
);
530 orphan_blocks
= __start_sum_addr(sbi
) - 1 - __cp_payload(sbi
);
532 ra_meta_pages(sbi
, start_blk
, orphan_blocks
, META_CP
, true);
534 for (i
= 0; i
< orphan_blocks
; i
++) {
535 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
536 struct f2fs_orphan_block
*orphan_blk
;
538 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
539 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
540 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
541 err
= recover_orphan_inode(sbi
, ino
);
543 f2fs_put_page(page
, 1);
547 f2fs_put_page(page
, 1);
549 /* clear Orphan Flag */
550 clear_ckpt_flags(F2FS_CKPT(sbi
), CP_ORPHAN_PRESENT_FLAG
);
554 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
556 struct list_head
*head
;
557 struct f2fs_orphan_block
*orphan_blk
= NULL
;
558 unsigned int nentries
= 0;
559 unsigned short index
= 1;
560 unsigned short orphan_blocks
;
561 struct page
*page
= NULL
;
562 struct ino_entry
*orphan
= NULL
;
563 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
565 orphan_blocks
= GET_ORPHAN_BLOCKS(im
->ino_num
);
568 * we don't need to do spin_lock(&im->ino_lock) here, since all the
569 * orphan inode operations are covered under f2fs_lock_op().
570 * And, spin_lock should be avoided due to page operations below.
572 head
= &im
->ino_list
;
574 /* loop for each orphan inode entry and write them in Jornal block */
575 list_for_each_entry(orphan
, head
, list
) {
577 page
= grab_meta_page(sbi
, start_blk
++);
579 (struct f2fs_orphan_block
*)page_address(page
);
580 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
583 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
585 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
587 * an orphan block is full of 1020 entries,
588 * then we need to flush current orphan blocks
589 * and bring another one in memory
591 orphan_blk
->blk_addr
= cpu_to_le16(index
);
592 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
593 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
594 set_page_dirty(page
);
595 f2fs_put_page(page
, 1);
603 orphan_blk
->blk_addr
= cpu_to_le16(index
);
604 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
605 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
606 set_page_dirty(page
);
607 f2fs_put_page(page
, 1);
611 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
612 block_t cp_addr
, unsigned long long *version
)
614 struct page
*cp_page_1
, *cp_page_2
= NULL
;
615 unsigned long blk_size
= sbi
->blocksize
;
616 struct f2fs_checkpoint
*cp_block
;
617 unsigned long long cur_version
= 0, pre_version
= 0;
621 /* Read the 1st cp block in this CP pack */
622 cp_page_1
= get_meta_page(sbi
, cp_addr
);
624 /* get the version number */
625 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_1
);
626 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
627 if (crc_offset
>= blk_size
)
630 crc
= le32_to_cpu(*((__le32
*)((unsigned char *)cp_block
+ crc_offset
)));
631 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
634 pre_version
= cur_cp_version(cp_block
);
636 /* Read the 2nd cp block in this CP pack */
637 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
638 cp_page_2
= get_meta_page(sbi
, cp_addr
);
640 cp_block
= (struct f2fs_checkpoint
*)page_address(cp_page_2
);
641 crc_offset
= le32_to_cpu(cp_block
->checksum_offset
);
642 if (crc_offset
>= blk_size
)
645 crc
= le32_to_cpu(*((__le32
*)((unsigned char *)cp_block
+ crc_offset
)));
646 if (!f2fs_crc_valid(crc
, cp_block
, crc_offset
))
649 cur_version
= cur_cp_version(cp_block
);
651 if (cur_version
== pre_version
) {
652 *version
= cur_version
;
653 f2fs_put_page(cp_page_2
, 1);
657 f2fs_put_page(cp_page_2
, 1);
659 f2fs_put_page(cp_page_1
, 1);
663 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
665 struct f2fs_checkpoint
*cp_block
;
666 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
667 struct page
*cp1
, *cp2
, *cur_page
;
668 unsigned long blk_size
= sbi
->blocksize
;
669 unsigned long long cp1_version
= 0, cp2_version
= 0;
670 unsigned long long cp_start_blk_no
;
671 unsigned int cp_blks
= 1 + __cp_payload(sbi
);
675 sbi
->ckpt
= kzalloc(cp_blks
* blk_size
, GFP_KERNEL
);
679 * Finding out valid cp block involves read both
680 * sets( cp pack1 and cp pack 2)
682 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
683 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
685 /* The second checkpoint pack should start at the next segment */
686 cp_start_blk_no
+= ((unsigned long long)1) <<
687 le32_to_cpu(fsb
->log_blocks_per_seg
);
688 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
691 if (ver_after(cp2_version
, cp1_version
))
703 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
704 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
709 cp_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
711 cp_blk_no
+= 1 << le32_to_cpu(fsb
->log_blocks_per_seg
);
713 for (i
= 1; i
< cp_blks
; i
++) {
714 void *sit_bitmap_ptr
;
715 unsigned char *ckpt
= (unsigned char *)sbi
->ckpt
;
717 cur_page
= get_meta_page(sbi
, cp_blk_no
+ i
);
718 sit_bitmap_ptr
= page_address(cur_page
);
719 memcpy(ckpt
+ i
* blk_size
, sit_bitmap_ptr
, blk_size
);
720 f2fs_put_page(cur_page
, 1);
723 f2fs_put_page(cp1
, 1);
724 f2fs_put_page(cp2
, 1);
732 static void __add_dirty_inode(struct inode
*inode
, enum inode_type type
)
734 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
735 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
736 int flag
= (type
== DIR_INODE
) ? FI_DIRTY_DIR
: FI_DIRTY_FILE
;
738 if (is_inode_flag_set(fi
, flag
))
741 set_inode_flag(fi
, flag
);
742 list_add_tail(&fi
->dirty_list
, &sbi
->inode_list
[type
]);
743 stat_inc_dirty_inode(sbi
, type
);
746 static void __remove_dirty_inode(struct inode
*inode
, enum inode_type type
)
748 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
749 int flag
= (type
== DIR_INODE
) ? FI_DIRTY_DIR
: FI_DIRTY_FILE
;
751 if (get_dirty_pages(inode
) ||
752 !is_inode_flag_set(F2FS_I(inode
), flag
))
755 list_del_init(&fi
->dirty_list
);
756 clear_inode_flag(fi
, flag
);
757 stat_dec_dirty_inode(F2FS_I_SB(inode
), type
);
760 void update_dirty_page(struct inode
*inode
, struct page
*page
)
762 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
763 enum inode_type type
= S_ISDIR(inode
->i_mode
) ? DIR_INODE
: FILE_INODE
;
765 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
) &&
766 !S_ISLNK(inode
->i_mode
))
769 spin_lock(&sbi
->inode_lock
[type
]);
770 __add_dirty_inode(inode
, type
);
771 inode_inc_dirty_pages(inode
);
772 spin_unlock(&sbi
->inode_lock
[type
]);
774 SetPagePrivate(page
);
775 f2fs_trace_pid(page
);
778 void add_dirty_dir_inode(struct inode
*inode
)
780 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
782 spin_lock(&sbi
->inode_lock
[DIR_INODE
]);
783 __add_dirty_inode(inode
, DIR_INODE
);
784 spin_unlock(&sbi
->inode_lock
[DIR_INODE
]);
787 void remove_dirty_inode(struct inode
*inode
)
789 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
790 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
791 enum inode_type type
= S_ISDIR(inode
->i_mode
) ? DIR_INODE
: FILE_INODE
;
793 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
) &&
794 !S_ISLNK(inode
->i_mode
))
797 spin_lock(&sbi
->inode_lock
[type
]);
798 __remove_dirty_inode(inode
, type
);
799 spin_unlock(&sbi
->inode_lock
[type
]);
801 /* Only from the recovery routine */
802 if (is_inode_flag_set(fi
, FI_DELAY_IPUT
)) {
803 clear_inode_flag(fi
, FI_DELAY_IPUT
);
808 int sync_dirty_inodes(struct f2fs_sb_info
*sbi
, enum inode_type type
)
810 struct list_head
*head
;
812 struct f2fs_inode_info
*fi
;
813 bool is_dir
= (type
== DIR_INODE
);
815 trace_f2fs_sync_dirty_inodes_enter(sbi
->sb
, is_dir
,
816 get_pages(sbi
, is_dir
?
817 F2FS_DIRTY_DENTS
: F2FS_DIRTY_DATA
));
819 if (unlikely(f2fs_cp_error(sbi
)))
822 spin_lock(&sbi
->inode_lock
[type
]);
824 head
= &sbi
->inode_list
[type
];
825 if (list_empty(head
)) {
826 spin_unlock(&sbi
->inode_lock
[type
]);
827 trace_f2fs_sync_dirty_inodes_exit(sbi
->sb
, is_dir
,
828 get_pages(sbi
, is_dir
?
829 F2FS_DIRTY_DENTS
: F2FS_DIRTY_DATA
));
832 fi
= list_entry(head
->next
, struct f2fs_inode_info
, dirty_list
);
833 inode
= igrab(&fi
->vfs_inode
);
834 spin_unlock(&sbi
->inode_lock
[type
]);
836 filemap_fdatawrite(inode
->i_mapping
);
840 * We should submit bio, since it exists several
841 * wribacking dentry pages in the freeing inode.
843 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
850 * Freeze all the FS-operations for checkpoint.
852 static int block_operations(struct f2fs_sb_info
*sbi
)
854 struct writeback_control wbc
= {
855 .sync_mode
= WB_SYNC_ALL
,
856 .nr_to_write
= LONG_MAX
,
859 struct blk_plug plug
;
862 blk_start_plug(&plug
);
866 /* write all the dirty dentry pages */
867 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
868 f2fs_unlock_all(sbi
);
869 err
= sync_dirty_inodes(sbi
, DIR_INODE
);
872 goto retry_flush_dents
;
876 * POR: we should ensure that there are no dirty node pages
877 * until finishing nat/sit flush.
880 down_write(&sbi
->node_write
);
882 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
883 up_write(&sbi
->node_write
);
884 err
= sync_node_pages(sbi
, 0, &wbc
);
886 f2fs_unlock_all(sbi
);
889 goto retry_flush_nodes
;
892 blk_finish_plug(&plug
);
896 static void unblock_operations(struct f2fs_sb_info
*sbi
)
898 up_write(&sbi
->node_write
);
899 f2fs_unlock_all(sbi
);
902 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
907 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
909 if (!get_pages(sbi
, F2FS_WRITEBACK
))
914 finish_wait(&sbi
->cp_wait
, &wait
);
917 static int do_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
919 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
920 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_WARM_NODE
);
921 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
922 unsigned long orphan_num
= sbi
->im
[ORPHAN_INO
].ino_num
;
923 nid_t last_nid
= nm_i
->next_scan_nid
;
925 unsigned int data_sum_blocks
, orphan_blocks
;
928 int cp_payload_blks
= __cp_payload(sbi
);
929 block_t discard_blk
= NEXT_FREE_BLKADDR(sbi
, curseg
);
930 bool invalidate
= false;
931 struct super_block
*sb
= sbi
->sb
;
932 struct curseg_info
*seg_i
= CURSEG_I(sbi
, CURSEG_HOT_NODE
);
936 * This avoids to conduct wrong roll-forward operations and uses
937 * metapages, so should be called prior to sync_meta_pages below.
939 if (discard_next_dnode(sbi
, discard_blk
))
942 /* Flush all the NAT/SIT pages */
943 while (get_pages(sbi
, F2FS_DIRTY_META
)) {
944 sync_meta_pages(sbi
, META
, LONG_MAX
);
945 if (unlikely(f2fs_cp_error(sbi
)))
949 next_free_nid(sbi
, &last_nid
);
953 * version number is already updated
955 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
956 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
957 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
958 for (i
= 0; i
< NR_CURSEG_NODE_TYPE
; i
++) {
959 ckpt
->cur_node_segno
[i
] =
960 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
961 ckpt
->cur_node_blkoff
[i
] =
962 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
963 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
964 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
966 for (i
= 0; i
< NR_CURSEG_DATA_TYPE
; i
++) {
967 ckpt
->cur_data_segno
[i
] =
968 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
969 ckpt
->cur_data_blkoff
[i
] =
970 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
971 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
972 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
975 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
976 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
977 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
979 /* 2 cp + n data seg summary + orphan inode blocks */
980 data_sum_blocks
= npages_for_summary_flush(sbi
, false);
981 if (data_sum_blocks
< NR_CURSEG_DATA_TYPE
)
982 set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
984 clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
986 orphan_blocks
= GET_ORPHAN_BLOCKS(orphan_num
);
987 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + cp_payload_blks
+
990 if (__remain_node_summaries(cpc
->reason
))
991 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
992 cp_payload_blks
+ data_sum_blocks
+
993 orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
995 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
996 cp_payload_blks
+ data_sum_blocks
+
999 if (cpc
->reason
== CP_UMOUNT
)
1000 set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
1002 clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
1004 if (cpc
->reason
== CP_FASTBOOT
)
1005 set_ckpt_flags(ckpt
, CP_FASTBOOT_FLAG
);
1007 clear_ckpt_flags(ckpt
, CP_FASTBOOT_FLAG
);
1010 set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
1012 clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
1014 if (is_sbi_flag_set(sbi
, SBI_NEED_FSCK
))
1015 set_ckpt_flags(ckpt
, CP_FSCK_FLAG
);
1017 /* update SIT/NAT bitmap */
1018 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
1019 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
1021 crc32
= f2fs_crc32(ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
1022 *((__le32
*)((unsigned char *)ckpt
+
1023 le32_to_cpu(ckpt
->checksum_offset
)))
1024 = cpu_to_le32(crc32
);
1026 start_blk
= __start_cp_addr(sbi
);
1028 /* need to wait for end_io results */
1029 wait_on_all_pages_writeback(sbi
);
1030 if (unlikely(f2fs_cp_error(sbi
)))
1033 /* write out checkpoint buffer at block 0 */
1034 update_meta_page(sbi
, ckpt
, start_blk
++);
1036 for (i
= 1; i
< 1 + cp_payload_blks
; i
++)
1037 update_meta_page(sbi
, (char *)ckpt
+ i
* F2FS_BLKSIZE
,
1041 write_orphan_inodes(sbi
, start_blk
);
1042 start_blk
+= orphan_blocks
;
1045 write_data_summaries(sbi
, start_blk
);
1046 start_blk
+= data_sum_blocks
;
1048 /* Record write statistics in the hot node summary */
1049 kbytes_written
= sbi
->kbytes_written
;
1050 if (sb
->s_bdev
->bd_part
)
1051 kbytes_written
+= BD_PART_WRITTEN(sbi
);
1053 seg_i
->sum_blk
->info
.kbytes_written
= cpu_to_le64(kbytes_written
);
1055 if (__remain_node_summaries(cpc
->reason
)) {
1056 write_node_summaries(sbi
, start_blk
);
1057 start_blk
+= NR_CURSEG_NODE_TYPE
;
1060 /* writeout checkpoint block */
1061 update_meta_page(sbi
, ckpt
, start_blk
);
1063 /* wait for previous submitted node/meta pages writeback */
1064 wait_on_all_pages_writeback(sbi
);
1066 if (unlikely(f2fs_cp_error(sbi
)))
1069 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LONG_MAX
);
1070 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LONG_MAX
);
1072 /* update user_block_counts */
1073 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
1074 sbi
->alloc_valid_block_count
= 0;
1076 /* Here, we only have one bio having CP pack */
1077 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
1079 /* wait for previous submitted meta pages writeback */
1080 wait_on_all_pages_writeback(sbi
);
1083 * invalidate meta page which is used temporarily for zeroing out
1084 * block at the end of warm node chain.
1087 invalidate_mapping_pages(META_MAPPING(sbi
), discard_blk
,
1090 release_ino_entry(sbi
);
1092 if (unlikely(f2fs_cp_error(sbi
)))
1095 clear_prefree_segments(sbi
, cpc
);
1096 clear_sbi_flag(sbi
, SBI_IS_DIRTY
);
1102 * We guarantee that this checkpoint procedure will not fail.
1104 int write_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1106 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1107 unsigned long long ckpt_ver
;
1110 mutex_lock(&sbi
->cp_mutex
);
1112 if (!is_sbi_flag_set(sbi
, SBI_IS_DIRTY
) &&
1113 (cpc
->reason
== CP_FASTBOOT
|| cpc
->reason
== CP_SYNC
||
1114 (cpc
->reason
== CP_DISCARD
&& !sbi
->discard_blks
)))
1116 if (unlikely(f2fs_cp_error(sbi
))) {
1120 if (f2fs_readonly(sbi
->sb
)) {
1125 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "start block_ops");
1127 err
= block_operations(sbi
);
1131 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish block_ops");
1133 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1134 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1135 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
1138 * update checkpoint pack index
1139 * Increase the version number so that
1140 * SIT entries and seg summaries are written at correct place
1142 ckpt_ver
= cur_cp_version(ckpt
);
1143 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
1145 /* write cached NAT/SIT entries to NAT/SIT area */
1146 flush_nat_entries(sbi
);
1147 flush_sit_entries(sbi
, cpc
);
1149 /* unlock all the fs_lock[] in do_checkpoint() */
1150 err
= do_checkpoint(sbi
, cpc
);
1152 unblock_operations(sbi
);
1153 stat_inc_cp_count(sbi
->stat_info
);
1155 if (cpc
->reason
== CP_RECOVERY
)
1156 f2fs_msg(sbi
->sb
, KERN_NOTICE
,
1157 "checkpoint: version = %llx", ckpt_ver
);
1159 /* do checkpoint periodically */
1160 f2fs_update_time(sbi
, CP_TIME
);
1161 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish checkpoint");
1163 mutex_unlock(&sbi
->cp_mutex
);
1167 void init_ino_entry_info(struct f2fs_sb_info
*sbi
)
1171 for (i
= 0; i
< MAX_INO_ENTRY
; i
++) {
1172 struct inode_management
*im
= &sbi
->im
[i
];
1174 INIT_RADIX_TREE(&im
->ino_root
, GFP_ATOMIC
);
1175 spin_lock_init(&im
->ino_lock
);
1176 INIT_LIST_HEAD(&im
->ino_list
);
1180 sbi
->max_orphans
= (sbi
->blocks_per_seg
- F2FS_CP_PACKS
-
1181 NR_CURSEG_TYPE
- __cp_payload(sbi
)) *
1182 F2FS_ORPHANS_PER_BLOCK
;
1185 int __init
create_checkpoint_caches(void)
1187 ino_entry_slab
= f2fs_kmem_cache_create("f2fs_ino_entry",
1188 sizeof(struct ino_entry
));
1189 if (!ino_entry_slab
)
1191 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_inode_entry",
1192 sizeof(struct inode_entry
));
1193 if (!inode_entry_slab
) {
1194 kmem_cache_destroy(ino_entry_slab
);
1200 void destroy_checkpoint_caches(void)
1202 kmem_cache_destroy(ino_entry_slab
);
1203 kmem_cache_destroy(inode_entry_slab
);