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
, 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
= {
68 .op_flags
= REQ_META
| REQ_PRIO
,
71 .encrypted_page
= NULL
,
74 if (unlikely(!is_meta
))
75 fio
.op_flags
&= ~REQ_META
;
77 page
= f2fs_grab_cache_page(mapping
, index
, false);
82 if (PageUptodate(page
))
87 if (f2fs_submit_page_bio(&fio
)) {
88 f2fs_put_page(page
, 1);
93 if (unlikely(page
->mapping
!= mapping
)) {
94 f2fs_put_page(page
, 1);
99 * if there is any IO error when accessing device, make our filesystem
100 * readonly and make sure do not write checkpoint with non-uptodate
103 if (unlikely(!PageUptodate(page
)))
104 f2fs_stop_checkpoint(sbi
, false);
109 struct page
*get_meta_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
111 return __get_meta_page(sbi
, index
, true);
115 struct page
*get_tmp_page(struct f2fs_sb_info
*sbi
, pgoff_t index
)
117 return __get_meta_page(sbi
, index
, false);
120 bool is_valid_blkaddr(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int type
)
126 if (unlikely(blkaddr
>= SIT_BLK_CNT(sbi
)))
130 if (unlikely(blkaddr
>= MAIN_BLKADDR(sbi
) ||
131 blkaddr
< SM_I(sbi
)->ssa_blkaddr
))
135 if (unlikely(blkaddr
>= SIT_I(sbi
)->sit_base_addr
||
136 blkaddr
< __start_cp_addr(sbi
)))
140 if (unlikely(blkaddr
>= MAX_BLKADDR(sbi
) ||
141 blkaddr
< MAIN_BLKADDR(sbi
)))
152 * Readahead CP/NAT/SIT/SSA pages
154 int ra_meta_pages(struct f2fs_sb_info
*sbi
, block_t start
, int nrpages
,
158 block_t blkno
= start
;
159 struct f2fs_io_info fio
= {
163 .op_flags
= sync
? (REQ_META
| REQ_PRIO
) : REQ_RAHEAD
,
164 .encrypted_page
= NULL
,
166 struct blk_plug plug
;
168 if (unlikely(type
== META_POR
))
169 fio
.op_flags
&= ~REQ_META
;
171 blk_start_plug(&plug
);
172 for (; nrpages
-- > 0; blkno
++) {
174 if (!is_valid_blkaddr(sbi
, blkno
, type
))
179 if (unlikely(blkno
>=
180 NAT_BLOCK_OFFSET(NM_I(sbi
)->max_nid
)))
182 /* get nat block addr */
183 fio
.new_blkaddr
= current_nat_addr(sbi
,
184 blkno
* NAT_ENTRY_PER_BLOCK
);
187 /* get sit block addr */
188 fio
.new_blkaddr
= current_sit_addr(sbi
,
189 blkno
* SIT_ENTRY_PER_BLOCK
);
194 fio
.new_blkaddr
= blkno
;
200 page
= f2fs_grab_cache_page(META_MAPPING(sbi
),
201 fio
.new_blkaddr
, false);
204 if (PageUptodate(page
)) {
205 f2fs_put_page(page
, 1);
210 fio
.old_blkaddr
= fio
.new_blkaddr
;
211 f2fs_submit_page_mbio(&fio
);
212 f2fs_put_page(page
, 0);
215 f2fs_submit_merged_bio(sbi
, META
, READ
);
216 blk_finish_plug(&plug
);
217 return blkno
- start
;
220 void ra_meta_pages_cond(struct f2fs_sb_info
*sbi
, pgoff_t index
)
223 bool readahead
= false;
225 page
= find_get_page(META_MAPPING(sbi
), index
);
226 if (!page
|| !PageUptodate(page
))
228 f2fs_put_page(page
, 0);
231 ra_meta_pages(sbi
, index
, BIO_MAX_PAGES
, META_POR
, true);
234 static int f2fs_write_meta_page(struct page
*page
,
235 struct writeback_control
*wbc
)
237 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
239 trace_f2fs_writepage(page
, META
);
241 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
243 if (wbc
->for_reclaim
&& page
->index
< GET_SUM_BLOCK(sbi
, 0))
245 if (unlikely(f2fs_cp_error(sbi
)))
248 write_meta_page(sbi
, page
);
249 dec_page_count(sbi
, F2FS_DIRTY_META
);
251 if (wbc
->for_reclaim
)
252 f2fs_submit_merged_bio_cond(sbi
, page
->mapping
->host
,
253 0, page
->index
, META
, WRITE
);
257 if (unlikely(f2fs_cp_error(sbi
)))
258 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
263 redirty_page_for_writepage(wbc
, page
);
264 return AOP_WRITEPAGE_ACTIVATE
;
267 static int f2fs_write_meta_pages(struct address_space
*mapping
,
268 struct writeback_control
*wbc
)
270 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
273 /* collect a number of dirty meta pages and write together */
274 if (wbc
->for_kupdate
||
275 get_pages(sbi
, F2FS_DIRTY_META
) < nr_pages_to_skip(sbi
, META
))
278 /* if locked failed, cp will flush dirty pages instead */
279 if (!mutex_trylock(&sbi
->cp_mutex
))
282 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
283 diff
= nr_pages_to_write(sbi
, META
, wbc
);
284 written
= sync_meta_pages(sbi
, META
, wbc
->nr_to_write
);
285 mutex_unlock(&sbi
->cp_mutex
);
286 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- written
- diff
);
290 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_META
);
291 trace_f2fs_writepages(mapping
->host
, wbc
, META
);
295 long sync_meta_pages(struct f2fs_sb_info
*sbi
, enum page_type type
,
298 struct address_space
*mapping
= META_MAPPING(sbi
);
299 pgoff_t index
= 0, end
= ULONG_MAX
, prev
= ULONG_MAX
;
302 struct writeback_control wbc
= {
305 struct blk_plug plug
;
307 pagevec_init(&pvec
, 0);
309 blk_start_plug(&plug
);
311 while (index
<= end
) {
313 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
315 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
316 if (unlikely(nr_pages
== 0))
319 for (i
= 0; i
< nr_pages
; i
++) {
320 struct page
*page
= pvec
.pages
[i
];
322 if (prev
== ULONG_MAX
)
323 prev
= page
->index
- 1;
324 if (nr_to_write
!= LONG_MAX
&& page
->index
!= prev
+ 1) {
325 pagevec_release(&pvec
);
331 if (unlikely(page
->mapping
!= mapping
)) {
336 if (!PageDirty(page
)) {
337 /* someone wrote it for us */
338 goto continue_unlock
;
341 f2fs_wait_on_page_writeback(page
, META
, true);
343 BUG_ON(PageWriteback(page
));
344 if (!clear_page_dirty_for_io(page
))
345 goto continue_unlock
;
347 if (mapping
->a_ops
->writepage(page
, &wbc
)) {
353 if (unlikely(nwritten
>= nr_to_write
))
356 pagevec_release(&pvec
);
361 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
363 blk_finish_plug(&plug
);
368 static int f2fs_set_meta_page_dirty(struct page
*page
)
370 trace_f2fs_set_page_dirty(page
, META
);
372 if (!PageUptodate(page
))
373 SetPageUptodate(page
);
374 if (!PageDirty(page
)) {
375 f2fs_set_page_dirty_nobuffers(page
);
376 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_META
);
377 SetPagePrivate(page
);
378 f2fs_trace_pid(page
);
384 const struct address_space_operations f2fs_meta_aops
= {
385 .writepage
= f2fs_write_meta_page
,
386 .writepages
= f2fs_write_meta_pages
,
387 .set_page_dirty
= f2fs_set_meta_page_dirty
,
388 .invalidatepage
= f2fs_invalidate_page
,
389 .releasepage
= f2fs_release_page
,
390 #ifdef CONFIG_MIGRATION
391 .migratepage
= f2fs_migrate_page
,
395 static void __add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
397 struct inode_management
*im
= &sbi
->im
[type
];
398 struct ino_entry
*e
, *tmp
;
400 tmp
= f2fs_kmem_cache_alloc(ino_entry_slab
, GFP_NOFS
);
402 radix_tree_preload(GFP_NOFS
| __GFP_NOFAIL
);
404 spin_lock(&im
->ino_lock
);
405 e
= radix_tree_lookup(&im
->ino_root
, ino
);
408 if (radix_tree_insert(&im
->ino_root
, ino
, e
)) {
409 spin_unlock(&im
->ino_lock
);
410 radix_tree_preload_end();
413 memset(e
, 0, sizeof(struct ino_entry
));
416 list_add_tail(&e
->list
, &im
->ino_list
);
417 if (type
!= ORPHAN_INO
)
420 spin_unlock(&im
->ino_lock
);
421 radix_tree_preload_end();
424 kmem_cache_free(ino_entry_slab
, tmp
);
427 static void __remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
429 struct inode_management
*im
= &sbi
->im
[type
];
432 spin_lock(&im
->ino_lock
);
433 e
= radix_tree_lookup(&im
->ino_root
, ino
);
436 radix_tree_delete(&im
->ino_root
, ino
);
438 spin_unlock(&im
->ino_lock
);
439 kmem_cache_free(ino_entry_slab
, e
);
442 spin_unlock(&im
->ino_lock
);
445 void add_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
447 /* add new dirty ino entry into list */
448 __add_ino_entry(sbi
, ino
, type
);
451 void remove_ino_entry(struct f2fs_sb_info
*sbi
, nid_t ino
, int type
)
453 /* remove dirty ino entry from list */
454 __remove_ino_entry(sbi
, ino
, type
);
457 /* mode should be APPEND_INO or UPDATE_INO */
458 bool exist_written_data(struct f2fs_sb_info
*sbi
, nid_t ino
, int mode
)
460 struct inode_management
*im
= &sbi
->im
[mode
];
463 spin_lock(&im
->ino_lock
);
464 e
= radix_tree_lookup(&im
->ino_root
, ino
);
465 spin_unlock(&im
->ino_lock
);
466 return e
? true : false;
469 void release_ino_entry(struct f2fs_sb_info
*sbi
, bool all
)
471 struct ino_entry
*e
, *tmp
;
474 for (i
= all
? ORPHAN_INO
: APPEND_INO
; i
<= UPDATE_INO
; i
++) {
475 struct inode_management
*im
= &sbi
->im
[i
];
477 spin_lock(&im
->ino_lock
);
478 list_for_each_entry_safe(e
, tmp
, &im
->ino_list
, list
) {
480 radix_tree_delete(&im
->ino_root
, e
->ino
);
481 kmem_cache_free(ino_entry_slab
, e
);
484 spin_unlock(&im
->ino_lock
);
488 int acquire_orphan_inode(struct f2fs_sb_info
*sbi
)
490 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
493 spin_lock(&im
->ino_lock
);
495 #ifdef CONFIG_F2FS_FAULT_INJECTION
496 if (time_to_inject(sbi
, FAULT_ORPHAN
)) {
497 spin_unlock(&im
->ino_lock
);
498 f2fs_show_injection_info(FAULT_ORPHAN
);
502 if (unlikely(im
->ino_num
>= sbi
->max_orphans
))
506 spin_unlock(&im
->ino_lock
);
511 void release_orphan_inode(struct f2fs_sb_info
*sbi
)
513 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
515 spin_lock(&im
->ino_lock
);
516 f2fs_bug_on(sbi
, im
->ino_num
== 0);
518 spin_unlock(&im
->ino_lock
);
521 void add_orphan_inode(struct inode
*inode
)
523 /* add new orphan ino entry into list */
524 __add_ino_entry(F2FS_I_SB(inode
), inode
->i_ino
, ORPHAN_INO
);
525 update_inode_page(inode
);
528 void remove_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
530 /* remove orphan entry from orphan list */
531 __remove_ino_entry(sbi
, ino
, ORPHAN_INO
);
534 static int recover_orphan_inode(struct f2fs_sb_info
*sbi
, nid_t ino
)
538 int err
= acquire_orphan_inode(sbi
);
541 set_sbi_flag(sbi
, SBI_NEED_FSCK
);
542 f2fs_msg(sbi
->sb
, KERN_WARNING
,
543 "%s: orphan failed (ino=%x), run fsck to fix.",
548 __add_ino_entry(sbi
, ino
, ORPHAN_INO
);
550 inode
= f2fs_iget_retry(sbi
->sb
, ino
);
553 * there should be a bug that we can't find the entry
556 f2fs_bug_on(sbi
, PTR_ERR(inode
) == -ENOENT
);
557 return PTR_ERR(inode
);
562 /* truncate all the data during iput */
565 get_node_info(sbi
, ino
, &ni
);
567 /* ENOMEM was fully retried in f2fs_evict_inode. */
568 if (ni
.blk_addr
!= NULL_ADDR
) {
569 set_sbi_flag(sbi
, SBI_NEED_FSCK
);
570 f2fs_msg(sbi
->sb
, KERN_WARNING
,
571 "%s: orphan failed (ino=%x) by kernel, retry mount.",
575 __remove_ino_entry(sbi
, ino
, ORPHAN_INO
);
579 int recover_orphan_inodes(struct f2fs_sb_info
*sbi
)
581 block_t start_blk
, orphan_blocks
, i
, j
;
584 if (!is_set_ckpt_flags(sbi
, CP_ORPHAN_PRESENT_FLAG
))
587 start_blk
= __start_cp_addr(sbi
) + 1 + __cp_payload(sbi
);
588 orphan_blocks
= __start_sum_addr(sbi
) - 1 - __cp_payload(sbi
);
590 ra_meta_pages(sbi
, start_blk
, orphan_blocks
, META_CP
, true);
592 for (i
= 0; i
< orphan_blocks
; i
++) {
593 struct page
*page
= get_meta_page(sbi
, start_blk
+ i
);
594 struct f2fs_orphan_block
*orphan_blk
;
596 orphan_blk
= (struct f2fs_orphan_block
*)page_address(page
);
597 for (j
= 0; j
< le32_to_cpu(orphan_blk
->entry_count
); j
++) {
598 nid_t ino
= le32_to_cpu(orphan_blk
->ino
[j
]);
599 err
= recover_orphan_inode(sbi
, ino
);
601 f2fs_put_page(page
, 1);
605 f2fs_put_page(page
, 1);
607 /* clear Orphan Flag */
608 clear_ckpt_flags(sbi
, CP_ORPHAN_PRESENT_FLAG
);
612 static void write_orphan_inodes(struct f2fs_sb_info
*sbi
, block_t start_blk
)
614 struct list_head
*head
;
615 struct f2fs_orphan_block
*orphan_blk
= NULL
;
616 unsigned int nentries
= 0;
617 unsigned short index
= 1;
618 unsigned short orphan_blocks
;
619 struct page
*page
= NULL
;
620 struct ino_entry
*orphan
= NULL
;
621 struct inode_management
*im
= &sbi
->im
[ORPHAN_INO
];
623 orphan_blocks
= GET_ORPHAN_BLOCKS(im
->ino_num
);
626 * we don't need to do spin_lock(&im->ino_lock) here, since all the
627 * orphan inode operations are covered under f2fs_lock_op().
628 * And, spin_lock should be avoided due to page operations below.
630 head
= &im
->ino_list
;
632 /* loop for each orphan inode entry and write them in Jornal block */
633 list_for_each_entry(orphan
, head
, list
) {
635 page
= grab_meta_page(sbi
, start_blk
++);
637 (struct f2fs_orphan_block
*)page_address(page
);
638 memset(orphan_blk
, 0, sizeof(*orphan_blk
));
641 orphan_blk
->ino
[nentries
++] = cpu_to_le32(orphan
->ino
);
643 if (nentries
== F2FS_ORPHANS_PER_BLOCK
) {
645 * an orphan block is full of 1020 entries,
646 * then we need to flush current orphan blocks
647 * and bring another one in memory
649 orphan_blk
->blk_addr
= cpu_to_le16(index
);
650 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
651 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
652 set_page_dirty(page
);
653 f2fs_put_page(page
, 1);
661 orphan_blk
->blk_addr
= cpu_to_le16(index
);
662 orphan_blk
->blk_count
= cpu_to_le16(orphan_blocks
);
663 orphan_blk
->entry_count
= cpu_to_le32(nentries
);
664 set_page_dirty(page
);
665 f2fs_put_page(page
, 1);
669 static int get_checkpoint_version(struct f2fs_sb_info
*sbi
, block_t cp_addr
,
670 struct f2fs_checkpoint
**cp_block
, struct page
**cp_page
,
671 unsigned long long *version
)
673 unsigned long blk_size
= sbi
->blocksize
;
674 size_t crc_offset
= 0;
677 *cp_page
= get_meta_page(sbi
, cp_addr
);
678 *cp_block
= (struct f2fs_checkpoint
*)page_address(*cp_page
);
680 crc_offset
= le32_to_cpu((*cp_block
)->checksum_offset
);
681 if (crc_offset
> (blk_size
- sizeof(__le32
))) {
682 f2fs_msg(sbi
->sb
, KERN_WARNING
,
683 "invalid crc_offset: %zu", crc_offset
);
687 crc
= cur_cp_crc(*cp_block
);
688 if (!f2fs_crc_valid(sbi
, crc
, *cp_block
, crc_offset
)) {
689 f2fs_msg(sbi
->sb
, KERN_WARNING
, "invalid crc value");
693 *version
= cur_cp_version(*cp_block
);
697 static struct page
*validate_checkpoint(struct f2fs_sb_info
*sbi
,
698 block_t cp_addr
, unsigned long long *version
)
700 struct page
*cp_page_1
= NULL
, *cp_page_2
= NULL
;
701 struct f2fs_checkpoint
*cp_block
= NULL
;
702 unsigned long long cur_version
= 0, pre_version
= 0;
705 err
= get_checkpoint_version(sbi
, cp_addr
, &cp_block
,
706 &cp_page_1
, version
);
709 pre_version
= *version
;
711 cp_addr
+= le32_to_cpu(cp_block
->cp_pack_total_block_count
) - 1;
712 err
= get_checkpoint_version(sbi
, cp_addr
, &cp_block
,
713 &cp_page_2
, version
);
716 cur_version
= *version
;
718 if (cur_version
== pre_version
) {
719 *version
= cur_version
;
720 f2fs_put_page(cp_page_2
, 1);
724 f2fs_put_page(cp_page_2
, 1);
726 f2fs_put_page(cp_page_1
, 1);
730 int get_valid_checkpoint(struct f2fs_sb_info
*sbi
)
732 struct f2fs_checkpoint
*cp_block
;
733 struct f2fs_super_block
*fsb
= sbi
->raw_super
;
734 struct page
*cp1
, *cp2
, *cur_page
;
735 unsigned long blk_size
= sbi
->blocksize
;
736 unsigned long long cp1_version
= 0, cp2_version
= 0;
737 unsigned long long cp_start_blk_no
;
738 unsigned int cp_blks
= 1 + __cp_payload(sbi
);
742 sbi
->ckpt
= kzalloc(cp_blks
* blk_size
, GFP_KERNEL
);
746 * Finding out valid cp block involves read both
747 * sets( cp pack1 and cp pack 2)
749 cp_start_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
750 cp1
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp1_version
);
752 /* The second checkpoint pack should start at the next segment */
753 cp_start_blk_no
+= ((unsigned long long)1) <<
754 le32_to_cpu(fsb
->log_blocks_per_seg
);
755 cp2
= validate_checkpoint(sbi
, cp_start_blk_no
, &cp2_version
);
758 if (ver_after(cp2_version
, cp1_version
))
770 cp_block
= (struct f2fs_checkpoint
*)page_address(cur_page
);
771 memcpy(sbi
->ckpt
, cp_block
, blk_size
);
773 /* Sanity checking of checkpoint */
774 if (sanity_check_ckpt(sbi
))
775 goto free_fail_no_cp
;
778 sbi
->cur_cp_pack
= 1;
780 sbi
->cur_cp_pack
= 2;
785 cp_blk_no
= le32_to_cpu(fsb
->cp_blkaddr
);
787 cp_blk_no
+= 1 << le32_to_cpu(fsb
->log_blocks_per_seg
);
789 for (i
= 1; i
< cp_blks
; i
++) {
790 void *sit_bitmap_ptr
;
791 unsigned char *ckpt
= (unsigned char *)sbi
->ckpt
;
793 cur_page
= get_meta_page(sbi
, cp_blk_no
+ i
);
794 sit_bitmap_ptr
= page_address(cur_page
);
795 memcpy(ckpt
+ i
* blk_size
, sit_bitmap_ptr
, blk_size
);
796 f2fs_put_page(cur_page
, 1);
799 f2fs_put_page(cp1
, 1);
800 f2fs_put_page(cp2
, 1);
804 f2fs_put_page(cp1
, 1);
805 f2fs_put_page(cp2
, 1);
811 static void __add_dirty_inode(struct inode
*inode
, enum inode_type type
)
813 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
814 int flag
= (type
== DIR_INODE
) ? FI_DIRTY_DIR
: FI_DIRTY_FILE
;
816 if (is_inode_flag_set(inode
, flag
))
819 set_inode_flag(inode
, flag
);
820 if (!f2fs_is_volatile_file(inode
))
821 list_add_tail(&F2FS_I(inode
)->dirty_list
,
822 &sbi
->inode_list
[type
]);
823 stat_inc_dirty_inode(sbi
, type
);
826 static void __remove_dirty_inode(struct inode
*inode
, enum inode_type type
)
828 int flag
= (type
== DIR_INODE
) ? FI_DIRTY_DIR
: FI_DIRTY_FILE
;
830 if (get_dirty_pages(inode
) || !is_inode_flag_set(inode
, flag
))
833 list_del_init(&F2FS_I(inode
)->dirty_list
);
834 clear_inode_flag(inode
, flag
);
835 stat_dec_dirty_inode(F2FS_I_SB(inode
), type
);
838 void update_dirty_page(struct inode
*inode
, struct page
*page
)
840 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
841 enum inode_type type
= S_ISDIR(inode
->i_mode
) ? DIR_INODE
: FILE_INODE
;
843 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
) &&
844 !S_ISLNK(inode
->i_mode
))
847 spin_lock(&sbi
->inode_lock
[type
]);
848 if (type
!= FILE_INODE
|| test_opt(sbi
, DATA_FLUSH
))
849 __add_dirty_inode(inode
, type
);
850 inode_inc_dirty_pages(inode
);
851 spin_unlock(&sbi
->inode_lock
[type
]);
853 SetPagePrivate(page
);
854 f2fs_trace_pid(page
);
857 void remove_dirty_inode(struct inode
*inode
)
859 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
860 enum inode_type type
= S_ISDIR(inode
->i_mode
) ? DIR_INODE
: FILE_INODE
;
862 if (!S_ISDIR(inode
->i_mode
) && !S_ISREG(inode
->i_mode
) &&
863 !S_ISLNK(inode
->i_mode
))
866 if (type
== FILE_INODE
&& !test_opt(sbi
, DATA_FLUSH
))
869 spin_lock(&sbi
->inode_lock
[type
]);
870 __remove_dirty_inode(inode
, type
);
871 spin_unlock(&sbi
->inode_lock
[type
]);
874 int sync_dirty_inodes(struct f2fs_sb_info
*sbi
, enum inode_type type
)
876 struct list_head
*head
;
878 struct f2fs_inode_info
*fi
;
879 bool is_dir
= (type
== DIR_INODE
);
881 trace_f2fs_sync_dirty_inodes_enter(sbi
->sb
, is_dir
,
882 get_pages(sbi
, is_dir
?
883 F2FS_DIRTY_DENTS
: F2FS_DIRTY_DATA
));
885 if (unlikely(f2fs_cp_error(sbi
)))
888 spin_lock(&sbi
->inode_lock
[type
]);
890 head
= &sbi
->inode_list
[type
];
891 if (list_empty(head
)) {
892 spin_unlock(&sbi
->inode_lock
[type
]);
893 trace_f2fs_sync_dirty_inodes_exit(sbi
->sb
, is_dir
,
894 get_pages(sbi
, is_dir
?
895 F2FS_DIRTY_DENTS
: F2FS_DIRTY_DATA
));
898 fi
= list_first_entry(head
, struct f2fs_inode_info
, dirty_list
);
899 inode
= igrab(&fi
->vfs_inode
);
900 spin_unlock(&sbi
->inode_lock
[type
]);
902 filemap_fdatawrite(inode
->i_mapping
);
906 * We should submit bio, since it exists several
907 * wribacking dentry pages in the freeing inode.
909 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
915 int f2fs_sync_inode_meta(struct f2fs_sb_info
*sbi
)
917 struct list_head
*head
= &sbi
->inode_list
[DIRTY_META
];
919 struct f2fs_inode_info
*fi
;
920 s64 total
= get_pages(sbi
, F2FS_DIRTY_IMETA
);
923 if (unlikely(f2fs_cp_error(sbi
)))
926 spin_lock(&sbi
->inode_lock
[DIRTY_META
]);
927 if (list_empty(head
)) {
928 spin_unlock(&sbi
->inode_lock
[DIRTY_META
]);
931 fi
= list_first_entry(head
, struct f2fs_inode_info
,
933 inode
= igrab(&fi
->vfs_inode
);
934 spin_unlock(&sbi
->inode_lock
[DIRTY_META
]);
936 sync_inode_metadata(inode
, 0);
938 /* it's on eviction */
939 if (is_inode_flag_set(inode
, FI_DIRTY_INODE
))
940 update_inode_page(inode
);
947 static void __prepare_cp_block(struct f2fs_sb_info
*sbi
)
949 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
950 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
951 nid_t last_nid
= nm_i
->next_scan_nid
;
953 next_free_nid(sbi
, &last_nid
);
954 ckpt
->valid_block_count
= cpu_to_le64(valid_user_blocks(sbi
));
955 ckpt
->valid_node_count
= cpu_to_le32(valid_node_count(sbi
));
956 ckpt
->valid_inode_count
= cpu_to_le32(valid_inode_count(sbi
));
957 ckpt
->next_free_nid
= cpu_to_le32(last_nid
);
961 * Freeze all the FS-operations for checkpoint.
963 static int block_operations(struct f2fs_sb_info
*sbi
)
965 struct writeback_control wbc
= {
966 .sync_mode
= WB_SYNC_ALL
,
967 .nr_to_write
= LONG_MAX
,
970 struct blk_plug plug
;
973 blk_start_plug(&plug
);
977 /* write all the dirty dentry pages */
978 if (get_pages(sbi
, F2FS_DIRTY_DENTS
)) {
979 f2fs_unlock_all(sbi
);
980 err
= sync_dirty_inodes(sbi
, DIR_INODE
);
984 goto retry_flush_dents
;
988 * POR: we should ensure that there are no dirty node pages
989 * until finishing nat/sit flush. inode->i_blocks can be updated.
991 down_write(&sbi
->node_change
);
993 if (get_pages(sbi
, F2FS_DIRTY_IMETA
)) {
994 up_write(&sbi
->node_change
);
995 f2fs_unlock_all(sbi
);
996 err
= f2fs_sync_inode_meta(sbi
);
1000 goto retry_flush_dents
;
1004 down_write(&sbi
->node_write
);
1006 if (get_pages(sbi
, F2FS_DIRTY_NODES
)) {
1007 up_write(&sbi
->node_write
);
1008 err
= sync_node_pages(sbi
, &wbc
);
1010 up_write(&sbi
->node_change
);
1011 f2fs_unlock_all(sbi
);
1015 goto retry_flush_nodes
;
1019 * sbi->node_change is used only for AIO write_begin path which produces
1020 * dirty node blocks and some checkpoint values by block allocation.
1022 __prepare_cp_block(sbi
);
1023 up_write(&sbi
->node_change
);
1025 blk_finish_plug(&plug
);
1029 static void unblock_operations(struct f2fs_sb_info
*sbi
)
1031 up_write(&sbi
->node_write
);
1032 f2fs_unlock_all(sbi
);
1035 static void wait_on_all_pages_writeback(struct f2fs_sb_info
*sbi
)
1040 prepare_to_wait(&sbi
->cp_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
1042 if (!get_pages(sbi
, F2FS_WB_CP_DATA
))
1045 io_schedule_timeout(5*HZ
);
1047 finish_wait(&sbi
->cp_wait
, &wait
);
1050 static void update_ckpt_flags(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1052 unsigned long orphan_num
= sbi
->im
[ORPHAN_INO
].ino_num
;
1053 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1055 spin_lock(&sbi
->cp_lock
);
1057 if ((cpc
->reason
& CP_UMOUNT
) &&
1058 le32_to_cpu(ckpt
->cp_pack_total_block_count
) >
1059 sbi
->blocks_per_seg
- NM_I(sbi
)->nat_bits_blocks
)
1060 disable_nat_bits(sbi
, false);
1062 if (cpc
->reason
& CP_TRIMMED
)
1063 __set_ckpt_flags(ckpt
, CP_TRIMMED_FLAG
);
1065 if (cpc
->reason
& CP_UMOUNT
)
1066 __set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
1068 __clear_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
);
1070 if (cpc
->reason
& CP_FASTBOOT
)
1071 __set_ckpt_flags(ckpt
, CP_FASTBOOT_FLAG
);
1073 __clear_ckpt_flags(ckpt
, CP_FASTBOOT_FLAG
);
1076 __set_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
1078 __clear_ckpt_flags(ckpt
, CP_ORPHAN_PRESENT_FLAG
);
1080 if (is_sbi_flag_set(sbi
, SBI_NEED_FSCK
))
1081 __set_ckpt_flags(ckpt
, CP_FSCK_FLAG
);
1083 /* set this flag to activate crc|cp_ver for recovery */
1084 __set_ckpt_flags(ckpt
, CP_CRC_RECOVERY_FLAG
);
1086 spin_unlock(&sbi
->cp_lock
);
1089 static int do_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1091 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1092 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1093 unsigned long orphan_num
= sbi
->im
[ORPHAN_INO
].ino_num
;
1095 unsigned int data_sum_blocks
, orphan_blocks
;
1098 int cp_payload_blks
= __cp_payload(sbi
);
1099 struct super_block
*sb
= sbi
->sb
;
1100 struct curseg_info
*seg_i
= CURSEG_I(sbi
, CURSEG_HOT_NODE
);
1103 /* Flush all the NAT/SIT pages */
1104 while (get_pages(sbi
, F2FS_DIRTY_META
)) {
1105 sync_meta_pages(sbi
, META
, LONG_MAX
);
1106 if (unlikely(f2fs_cp_error(sbi
)))
1112 * version number is already updated
1114 ckpt
->elapsed_time
= cpu_to_le64(get_mtime(sbi
));
1115 ckpt
->free_segment_count
= cpu_to_le32(free_segments(sbi
));
1116 for (i
= 0; i
< NR_CURSEG_NODE_TYPE
; i
++) {
1117 ckpt
->cur_node_segno
[i
] =
1118 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_NODE
));
1119 ckpt
->cur_node_blkoff
[i
] =
1120 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_NODE
));
1121 ckpt
->alloc_type
[i
+ CURSEG_HOT_NODE
] =
1122 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_NODE
);
1124 for (i
= 0; i
< NR_CURSEG_DATA_TYPE
; i
++) {
1125 ckpt
->cur_data_segno
[i
] =
1126 cpu_to_le32(curseg_segno(sbi
, i
+ CURSEG_HOT_DATA
));
1127 ckpt
->cur_data_blkoff
[i
] =
1128 cpu_to_le16(curseg_blkoff(sbi
, i
+ CURSEG_HOT_DATA
));
1129 ckpt
->alloc_type
[i
+ CURSEG_HOT_DATA
] =
1130 curseg_alloc_type(sbi
, i
+ CURSEG_HOT_DATA
);
1133 /* 2 cp + n data seg summary + orphan inode blocks */
1134 data_sum_blocks
= npages_for_summary_flush(sbi
, false);
1135 spin_lock(&sbi
->cp_lock
);
1136 if (data_sum_blocks
< NR_CURSEG_DATA_TYPE
)
1137 __set_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
1139 __clear_ckpt_flags(ckpt
, CP_COMPACT_SUM_FLAG
);
1140 spin_unlock(&sbi
->cp_lock
);
1142 orphan_blocks
= GET_ORPHAN_BLOCKS(orphan_num
);
1143 ckpt
->cp_pack_start_sum
= cpu_to_le32(1 + cp_payload_blks
+
1146 if (__remain_node_summaries(cpc
->reason
))
1147 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
1148 cp_payload_blks
+ data_sum_blocks
+
1149 orphan_blocks
+ NR_CURSEG_NODE_TYPE
);
1151 ckpt
->cp_pack_total_block_count
= cpu_to_le32(F2FS_CP_PACKS
+
1152 cp_payload_blks
+ data_sum_blocks
+
1155 /* update ckpt flag for checkpoint */
1156 update_ckpt_flags(sbi
, cpc
);
1158 /* update SIT/NAT bitmap */
1159 get_sit_bitmap(sbi
, __bitmap_ptr(sbi
, SIT_BITMAP
));
1160 get_nat_bitmap(sbi
, __bitmap_ptr(sbi
, NAT_BITMAP
));
1162 crc32
= f2fs_crc32(sbi
, ckpt
, le32_to_cpu(ckpt
->checksum_offset
));
1163 *((__le32
*)((unsigned char *)ckpt
+
1164 le32_to_cpu(ckpt
->checksum_offset
)))
1165 = cpu_to_le32(crc32
);
1167 start_blk
= __start_cp_next_addr(sbi
);
1169 /* write nat bits */
1170 if (enabled_nat_bits(sbi
, cpc
)) {
1171 __u64 cp_ver
= cur_cp_version(ckpt
);
1174 cp_ver
|= ((__u64
)crc32
<< 32);
1175 *(__le64
*)nm_i
->nat_bits
= cpu_to_le64(cp_ver
);
1177 blk
= start_blk
+ sbi
->blocks_per_seg
- nm_i
->nat_bits_blocks
;
1178 for (i
= 0; i
< nm_i
->nat_bits_blocks
; i
++)
1179 update_meta_page(sbi
, nm_i
->nat_bits
+
1180 (i
<< F2FS_BLKSIZE_BITS
), blk
+ i
);
1182 /* Flush all the NAT BITS pages */
1183 while (get_pages(sbi
, F2FS_DIRTY_META
)) {
1184 sync_meta_pages(sbi
, META
, LONG_MAX
);
1185 if (unlikely(f2fs_cp_error(sbi
)))
1190 /* need to wait for end_io results */
1191 wait_on_all_pages_writeback(sbi
);
1192 if (unlikely(f2fs_cp_error(sbi
)))
1195 /* write out checkpoint buffer at block 0 */
1196 update_meta_page(sbi
, ckpt
, start_blk
++);
1198 for (i
= 1; i
< 1 + cp_payload_blks
; i
++)
1199 update_meta_page(sbi
, (char *)ckpt
+ i
* F2FS_BLKSIZE
,
1203 write_orphan_inodes(sbi
, start_blk
);
1204 start_blk
+= orphan_blocks
;
1207 write_data_summaries(sbi
, start_blk
);
1208 start_blk
+= data_sum_blocks
;
1210 /* Record write statistics in the hot node summary */
1211 kbytes_written
= sbi
->kbytes_written
;
1212 if (sb
->s_bdev
->bd_part
)
1213 kbytes_written
+= BD_PART_WRITTEN(sbi
);
1215 seg_i
->journal
->info
.kbytes_written
= cpu_to_le64(kbytes_written
);
1217 if (__remain_node_summaries(cpc
->reason
)) {
1218 write_node_summaries(sbi
, start_blk
);
1219 start_blk
+= NR_CURSEG_NODE_TYPE
;
1222 /* writeout checkpoint block */
1223 update_meta_page(sbi
, ckpt
, start_blk
);
1225 /* wait for previous submitted node/meta pages writeback */
1226 wait_on_all_pages_writeback(sbi
);
1228 if (unlikely(f2fs_cp_error(sbi
)))
1231 filemap_fdatawait_range(NODE_MAPPING(sbi
), 0, LLONG_MAX
);
1232 filemap_fdatawait_range(META_MAPPING(sbi
), 0, LLONG_MAX
);
1234 /* update user_block_counts */
1235 sbi
->last_valid_block_count
= sbi
->total_valid_block_count
;
1236 percpu_counter_set(&sbi
->alloc_valid_block_count
, 0);
1238 /* Here, we only have one bio having CP pack */
1239 sync_meta_pages(sbi
, META_FLUSH
, LONG_MAX
);
1241 /* wait for previous submitted meta pages writeback */
1242 wait_on_all_pages_writeback(sbi
);
1244 release_ino_entry(sbi
, false);
1246 if (unlikely(f2fs_cp_error(sbi
)))
1249 clear_sbi_flag(sbi
, SBI_IS_DIRTY
);
1250 clear_sbi_flag(sbi
, SBI_NEED_CP
);
1251 __set_cp_next_pack(sbi
);
1254 * redirty superblock if metadata like node page or inode cache is
1255 * updated during writing checkpoint.
1257 if (get_pages(sbi
, F2FS_DIRTY_NODES
) ||
1258 get_pages(sbi
, F2FS_DIRTY_IMETA
))
1259 set_sbi_flag(sbi
, SBI_IS_DIRTY
);
1261 f2fs_bug_on(sbi
, get_pages(sbi
, F2FS_DIRTY_DENTS
));
1267 * We guarantee that this checkpoint procedure will not fail.
1269 int write_checkpoint(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1271 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1272 unsigned long long ckpt_ver
;
1275 mutex_lock(&sbi
->cp_mutex
);
1277 if (!is_sbi_flag_set(sbi
, SBI_IS_DIRTY
) &&
1278 ((cpc
->reason
& CP_FASTBOOT
) || (cpc
->reason
& CP_SYNC
) ||
1279 ((cpc
->reason
& CP_DISCARD
) && !sbi
->discard_blks
)))
1281 if (unlikely(f2fs_cp_error(sbi
))) {
1285 if (f2fs_readonly(sbi
->sb
)) {
1290 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "start block_ops");
1292 err
= block_operations(sbi
);
1296 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish block_ops");
1298 f2fs_flush_merged_bios(sbi
);
1300 /* this is the case of multiple fstrims without any changes */
1301 if (cpc
->reason
& CP_DISCARD
) {
1302 if (!exist_trim_candidates(sbi
, cpc
)) {
1303 unblock_operations(sbi
);
1307 if (NM_I(sbi
)->dirty_nat_cnt
== 0 &&
1308 SIT_I(sbi
)->dirty_sentries
== 0 &&
1309 prefree_segments(sbi
) == 0) {
1310 flush_sit_entries(sbi
, cpc
);
1311 clear_prefree_segments(sbi
, cpc
);
1312 unblock_operations(sbi
);
1318 * update checkpoint pack index
1319 * Increase the version number so that
1320 * SIT entries and seg summaries are written at correct place
1322 ckpt_ver
= cur_cp_version(ckpt
);
1323 ckpt
->checkpoint_ver
= cpu_to_le64(++ckpt_ver
);
1325 /* write cached NAT/SIT entries to NAT/SIT area */
1326 flush_nat_entries(sbi
, cpc
);
1327 flush_sit_entries(sbi
, cpc
);
1329 /* unlock all the fs_lock[] in do_checkpoint() */
1330 err
= do_checkpoint(sbi
, cpc
);
1332 release_discard_addrs(sbi
);
1334 clear_prefree_segments(sbi
, cpc
);
1336 unblock_operations(sbi
);
1337 stat_inc_cp_count(sbi
->stat_info
);
1339 if (cpc
->reason
& CP_RECOVERY
)
1340 f2fs_msg(sbi
->sb
, KERN_NOTICE
,
1341 "checkpoint: version = %llx", ckpt_ver
);
1343 /* do checkpoint periodically */
1344 f2fs_update_time(sbi
, CP_TIME
);
1345 trace_f2fs_write_checkpoint(sbi
->sb
, cpc
->reason
, "finish checkpoint");
1347 mutex_unlock(&sbi
->cp_mutex
);
1351 void init_ino_entry_info(struct f2fs_sb_info
*sbi
)
1355 for (i
= 0; i
< MAX_INO_ENTRY
; i
++) {
1356 struct inode_management
*im
= &sbi
->im
[i
];
1358 INIT_RADIX_TREE(&im
->ino_root
, GFP_ATOMIC
);
1359 spin_lock_init(&im
->ino_lock
);
1360 INIT_LIST_HEAD(&im
->ino_list
);
1364 sbi
->max_orphans
= (sbi
->blocks_per_seg
- F2FS_CP_PACKS
-
1365 NR_CURSEG_TYPE
- __cp_payload(sbi
)) *
1366 F2FS_ORPHANS_PER_BLOCK
;
1369 int __init
create_checkpoint_caches(void)
1371 ino_entry_slab
= f2fs_kmem_cache_create("f2fs_ino_entry",
1372 sizeof(struct ino_entry
));
1373 if (!ino_entry_slab
)
1375 inode_entry_slab
= f2fs_kmem_cache_create("f2fs_inode_entry",
1376 sizeof(struct inode_entry
));
1377 if (!inode_entry_slab
) {
1378 kmem_cache_destroy(ino_entry_slab
);
1384 void destroy_checkpoint_caches(void)
1386 kmem_cache_destroy(ino_entry_slab
);
1387 kmem_cache_destroy(inode_entry_slab
);