Loading...
1/*
2 * fs/f2fs/checkpoint.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
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.
10 */
11#include <linux/fs.h>
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>
19
20#include "f2fs.h"
21#include "node.h"
22#include "segment.h"
23#include "trace.h"
24#include <trace/events/f2fs.h>
25
26static struct kmem_cache *ino_entry_slab;
27struct kmem_cache *inode_entry_slab;
28
29/*
30 * We guarantee no failure on the returned page.
31 */
32struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
33{
34 struct address_space *mapping = META_MAPPING(sbi);
35 struct page *page = NULL;
36repeat:
37 page = grab_cache_page(mapping, index);
38 if (!page) {
39 cond_resched();
40 goto repeat;
41 }
42 f2fs_wait_on_page_writeback(page, META, true);
43 SetPageUptodate(page);
44 return page;
45}
46
47/*
48 * We guarantee no failure on the returned page.
49 */
50static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
51 bool is_meta)
52{
53 struct address_space *mapping = META_MAPPING(sbi);
54 struct page *page;
55 struct f2fs_io_info fio = {
56 .sbi = sbi,
57 .type = META,
58 .rw = READ_SYNC | REQ_META | REQ_PRIO,
59 .old_blkaddr = index,
60 .new_blkaddr = index,
61 .encrypted_page = NULL,
62 };
63
64 if (unlikely(!is_meta))
65 fio.rw &= ~REQ_META;
66repeat:
67 page = grab_cache_page(mapping, index);
68 if (!page) {
69 cond_resched();
70 goto repeat;
71 }
72 if (PageUptodate(page))
73 goto out;
74
75 fio.page = page;
76
77 if (f2fs_submit_page_bio(&fio)) {
78 f2fs_put_page(page, 1);
79 goto repeat;
80 }
81
82 lock_page(page);
83 if (unlikely(page->mapping != mapping)) {
84 f2fs_put_page(page, 1);
85 goto repeat;
86 }
87
88 /*
89 * if there is any IO error when accessing device, make our filesystem
90 * readonly and make sure do not write checkpoint with non-uptodate
91 * meta page.
92 */
93 if (unlikely(!PageUptodate(page)))
94 f2fs_stop_checkpoint(sbi);
95out:
96 return page;
97}
98
99struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
100{
101 return __get_meta_page(sbi, index, true);
102}
103
104/* for POR only */
105struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
106{
107 return __get_meta_page(sbi, index, false);
108}
109
110bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
111{
112 switch (type) {
113 case META_NAT:
114 break;
115 case META_SIT:
116 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
117 return false;
118 break;
119 case META_SSA:
120 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
121 blkaddr < SM_I(sbi)->ssa_blkaddr))
122 return false;
123 break;
124 case META_CP:
125 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
126 blkaddr < __start_cp_addr(sbi)))
127 return false;
128 break;
129 case META_POR:
130 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
131 blkaddr < MAIN_BLKADDR(sbi)))
132 return false;
133 break;
134 default:
135 BUG();
136 }
137
138 return true;
139}
140
141/*
142 * Readahead CP/NAT/SIT/SSA pages
143 */
144int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
145 int type, bool sync)
146{
147 struct page *page;
148 block_t blkno = start;
149 struct f2fs_io_info fio = {
150 .sbi = sbi,
151 .type = META,
152 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
153 .encrypted_page = NULL,
154 };
155 struct blk_plug plug;
156
157 if (unlikely(type == META_POR))
158 fio.rw &= ~REQ_META;
159
160 blk_start_plug(&plug);
161 for (; nrpages-- > 0; blkno++) {
162
163 if (!is_valid_blkaddr(sbi, blkno, type))
164 goto out;
165
166 switch (type) {
167 case META_NAT:
168 if (unlikely(blkno >=
169 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
170 blkno = 0;
171 /* get nat block addr */
172 fio.new_blkaddr = current_nat_addr(sbi,
173 blkno * NAT_ENTRY_PER_BLOCK);
174 break;
175 case META_SIT:
176 /* get sit block addr */
177 fio.new_blkaddr = current_sit_addr(sbi,
178 blkno * SIT_ENTRY_PER_BLOCK);
179 break;
180 case META_SSA:
181 case META_CP:
182 case META_POR:
183 fio.new_blkaddr = blkno;
184 break;
185 default:
186 BUG();
187 }
188
189 page = grab_cache_page(META_MAPPING(sbi), fio.new_blkaddr);
190 if (!page)
191 continue;
192 if (PageUptodate(page)) {
193 f2fs_put_page(page, 1);
194 continue;
195 }
196
197 fio.page = page;
198 fio.old_blkaddr = fio.new_blkaddr;
199 f2fs_submit_page_mbio(&fio);
200 f2fs_put_page(page, 0);
201 }
202out:
203 f2fs_submit_merged_bio(sbi, META, READ);
204 blk_finish_plug(&plug);
205 return blkno - start;
206}
207
208void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
209{
210 struct page *page;
211 bool readahead = false;
212
213 page = find_get_page(META_MAPPING(sbi), index);
214 if (!page || (page && !PageUptodate(page)))
215 readahead = true;
216 f2fs_put_page(page, 0);
217
218 if (readahead)
219 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
220}
221
222static int f2fs_write_meta_page(struct page *page,
223 struct writeback_control *wbc)
224{
225 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
226
227 trace_f2fs_writepage(page, META);
228
229 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
230 goto redirty_out;
231 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
232 goto redirty_out;
233 if (unlikely(f2fs_cp_error(sbi)))
234 goto redirty_out;
235
236 write_meta_page(sbi, page);
237 dec_page_count(sbi, F2FS_DIRTY_META);
238
239 if (wbc->for_reclaim)
240 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
241
242 unlock_page(page);
243
244 if (unlikely(f2fs_cp_error(sbi)))
245 f2fs_submit_merged_bio(sbi, META, WRITE);
246
247 return 0;
248
249redirty_out:
250 redirty_page_for_writepage(wbc, page);
251 return AOP_WRITEPAGE_ACTIVATE;
252}
253
254static int f2fs_write_meta_pages(struct address_space *mapping,
255 struct writeback_control *wbc)
256{
257 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
258 long diff, written;
259
260 /* collect a number of dirty meta pages and write together */
261 if (wbc->for_kupdate ||
262 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
263 goto skip_write;
264
265 trace_f2fs_writepages(mapping->host, wbc, META);
266
267 /* if mounting is failed, skip writing node pages */
268 mutex_lock(&sbi->cp_mutex);
269 diff = nr_pages_to_write(sbi, META, wbc);
270 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
271 mutex_unlock(&sbi->cp_mutex);
272 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
273 return 0;
274
275skip_write:
276 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
277 trace_f2fs_writepages(mapping->host, wbc, META);
278 return 0;
279}
280
281long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
282 long nr_to_write)
283{
284 struct address_space *mapping = META_MAPPING(sbi);
285 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
286 struct pagevec pvec;
287 long nwritten = 0;
288 struct writeback_control wbc = {
289 .for_reclaim = 0,
290 };
291 struct blk_plug plug;
292
293 pagevec_init(&pvec, 0);
294
295 blk_start_plug(&plug);
296
297 while (index <= end) {
298 int i, nr_pages;
299 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
300 PAGECACHE_TAG_DIRTY,
301 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
302 if (unlikely(nr_pages == 0))
303 break;
304
305 for (i = 0; i < nr_pages; i++) {
306 struct page *page = pvec.pages[i];
307
308 if (prev == ULONG_MAX)
309 prev = page->index - 1;
310 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
311 pagevec_release(&pvec);
312 goto stop;
313 }
314
315 lock_page(page);
316
317 if (unlikely(page->mapping != mapping)) {
318continue_unlock:
319 unlock_page(page);
320 continue;
321 }
322 if (!PageDirty(page)) {
323 /* someone wrote it for us */
324 goto continue_unlock;
325 }
326
327 f2fs_wait_on_page_writeback(page, META, true);
328
329 BUG_ON(PageWriteback(page));
330 if (!clear_page_dirty_for_io(page))
331 goto continue_unlock;
332
333 if (mapping->a_ops->writepage(page, &wbc)) {
334 unlock_page(page);
335 break;
336 }
337 nwritten++;
338 prev = page->index;
339 if (unlikely(nwritten >= nr_to_write))
340 break;
341 }
342 pagevec_release(&pvec);
343 cond_resched();
344 }
345stop:
346 if (nwritten)
347 f2fs_submit_merged_bio(sbi, type, WRITE);
348
349 blk_finish_plug(&plug);
350
351 return nwritten;
352}
353
354static int f2fs_set_meta_page_dirty(struct page *page)
355{
356 trace_f2fs_set_page_dirty(page, META);
357
358 SetPageUptodate(page);
359 if (!PageDirty(page)) {
360 __set_page_dirty_nobuffers(page);
361 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
362 SetPagePrivate(page);
363 f2fs_trace_pid(page);
364 return 1;
365 }
366 return 0;
367}
368
369const struct address_space_operations f2fs_meta_aops = {
370 .writepage = f2fs_write_meta_page,
371 .writepages = f2fs_write_meta_pages,
372 .set_page_dirty = f2fs_set_meta_page_dirty,
373 .invalidatepage = f2fs_invalidate_page,
374 .releasepage = f2fs_release_page,
375};
376
377static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
378{
379 struct inode_management *im = &sbi->im[type];
380 struct ino_entry *e, *tmp;
381
382 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
383retry:
384 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
385
386 spin_lock(&im->ino_lock);
387 e = radix_tree_lookup(&im->ino_root, ino);
388 if (!e) {
389 e = tmp;
390 if (radix_tree_insert(&im->ino_root, ino, e)) {
391 spin_unlock(&im->ino_lock);
392 radix_tree_preload_end();
393 goto retry;
394 }
395 memset(e, 0, sizeof(struct ino_entry));
396 e->ino = ino;
397
398 list_add_tail(&e->list, &im->ino_list);
399 if (type != ORPHAN_INO)
400 im->ino_num++;
401 }
402 spin_unlock(&im->ino_lock);
403 radix_tree_preload_end();
404
405 if (e != tmp)
406 kmem_cache_free(ino_entry_slab, tmp);
407}
408
409static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
410{
411 struct inode_management *im = &sbi->im[type];
412 struct ino_entry *e;
413
414 spin_lock(&im->ino_lock);
415 e = radix_tree_lookup(&im->ino_root, ino);
416 if (e) {
417 list_del(&e->list);
418 radix_tree_delete(&im->ino_root, ino);
419 im->ino_num--;
420 spin_unlock(&im->ino_lock);
421 kmem_cache_free(ino_entry_slab, e);
422 return;
423 }
424 spin_unlock(&im->ino_lock);
425}
426
427void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
428{
429 /* add new dirty ino entry into list */
430 __add_ino_entry(sbi, ino, type);
431}
432
433void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
434{
435 /* remove dirty ino entry from list */
436 __remove_ino_entry(sbi, ino, type);
437}
438
439/* mode should be APPEND_INO or UPDATE_INO */
440bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
441{
442 struct inode_management *im = &sbi->im[mode];
443 struct ino_entry *e;
444
445 spin_lock(&im->ino_lock);
446 e = radix_tree_lookup(&im->ino_root, ino);
447 spin_unlock(&im->ino_lock);
448 return e ? true : false;
449}
450
451void release_ino_entry(struct f2fs_sb_info *sbi)
452{
453 struct ino_entry *e, *tmp;
454 int i;
455
456 for (i = APPEND_INO; i <= UPDATE_INO; i++) {
457 struct inode_management *im = &sbi->im[i];
458
459 spin_lock(&im->ino_lock);
460 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
461 list_del(&e->list);
462 radix_tree_delete(&im->ino_root, e->ino);
463 kmem_cache_free(ino_entry_slab, e);
464 im->ino_num--;
465 }
466 spin_unlock(&im->ino_lock);
467 }
468}
469
470int acquire_orphan_inode(struct f2fs_sb_info *sbi)
471{
472 struct inode_management *im = &sbi->im[ORPHAN_INO];
473 int err = 0;
474
475 spin_lock(&im->ino_lock);
476 if (unlikely(im->ino_num >= sbi->max_orphans))
477 err = -ENOSPC;
478 else
479 im->ino_num++;
480 spin_unlock(&im->ino_lock);
481
482 return err;
483}
484
485void release_orphan_inode(struct f2fs_sb_info *sbi)
486{
487 struct inode_management *im = &sbi->im[ORPHAN_INO];
488
489 spin_lock(&im->ino_lock);
490 f2fs_bug_on(sbi, im->ino_num == 0);
491 im->ino_num--;
492 spin_unlock(&im->ino_lock);
493}
494
495void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
496{
497 /* add new orphan ino entry into list */
498 __add_ino_entry(sbi, ino, ORPHAN_INO);
499}
500
501void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
502{
503 /* remove orphan entry from orphan list */
504 __remove_ino_entry(sbi, ino, ORPHAN_INO);
505}
506
507static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
508{
509 struct inode *inode;
510
511 inode = f2fs_iget(sbi->sb, ino);
512 if (IS_ERR(inode)) {
513 /*
514 * there should be a bug that we can't find the entry
515 * to orphan inode.
516 */
517 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
518 return PTR_ERR(inode);
519 }
520
521 clear_nlink(inode);
522
523 /* truncate all the data during iput */
524 iput(inode);
525 return 0;
526}
527
528int recover_orphan_inodes(struct f2fs_sb_info *sbi)
529{
530 block_t start_blk, orphan_blocks, i, j;
531 int err;
532
533 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
534 return 0;
535
536 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
537 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
538
539 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
540
541 for (i = 0; i < orphan_blocks; i++) {
542 struct page *page = get_meta_page(sbi, start_blk + i);
543 struct f2fs_orphan_block *orphan_blk;
544
545 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
546 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
547 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
548 err = recover_orphan_inode(sbi, ino);
549 if (err) {
550 f2fs_put_page(page, 1);
551 return err;
552 }
553 }
554 f2fs_put_page(page, 1);
555 }
556 /* clear Orphan Flag */
557 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
558 return 0;
559}
560
561static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
562{
563 struct list_head *head;
564 struct f2fs_orphan_block *orphan_blk = NULL;
565 unsigned int nentries = 0;
566 unsigned short index = 1;
567 unsigned short orphan_blocks;
568 struct page *page = NULL;
569 struct ino_entry *orphan = NULL;
570 struct inode_management *im = &sbi->im[ORPHAN_INO];
571
572 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
573
574 /*
575 * we don't need to do spin_lock(&im->ino_lock) here, since all the
576 * orphan inode operations are covered under f2fs_lock_op().
577 * And, spin_lock should be avoided due to page operations below.
578 */
579 head = &im->ino_list;
580
581 /* loop for each orphan inode entry and write them in Jornal block */
582 list_for_each_entry(orphan, head, list) {
583 if (!page) {
584 page = grab_meta_page(sbi, start_blk++);
585 orphan_blk =
586 (struct f2fs_orphan_block *)page_address(page);
587 memset(orphan_blk, 0, sizeof(*orphan_blk));
588 }
589
590 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
591
592 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
593 /*
594 * an orphan block is full of 1020 entries,
595 * then we need to flush current orphan blocks
596 * and bring another one in memory
597 */
598 orphan_blk->blk_addr = cpu_to_le16(index);
599 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
600 orphan_blk->entry_count = cpu_to_le32(nentries);
601 set_page_dirty(page);
602 f2fs_put_page(page, 1);
603 index++;
604 nentries = 0;
605 page = NULL;
606 }
607 }
608
609 if (page) {
610 orphan_blk->blk_addr = cpu_to_le16(index);
611 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
612 orphan_blk->entry_count = cpu_to_le32(nentries);
613 set_page_dirty(page);
614 f2fs_put_page(page, 1);
615 }
616}
617
618static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
619 block_t cp_addr, unsigned long long *version)
620{
621 struct page *cp_page_1, *cp_page_2 = NULL;
622 unsigned long blk_size = sbi->blocksize;
623 struct f2fs_checkpoint *cp_block;
624 unsigned long long cur_version = 0, pre_version = 0;
625 size_t crc_offset;
626 __u32 crc = 0;
627
628 /* Read the 1st cp block in this CP pack */
629 cp_page_1 = get_meta_page(sbi, cp_addr);
630
631 /* get the version number */
632 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
633 crc_offset = le32_to_cpu(cp_block->checksum_offset);
634 if (crc_offset >= blk_size)
635 goto invalid_cp1;
636
637 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
638 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
639 goto invalid_cp1;
640
641 pre_version = cur_cp_version(cp_block);
642
643 /* Read the 2nd cp block in this CP pack */
644 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
645 cp_page_2 = get_meta_page(sbi, cp_addr);
646
647 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
648 crc_offset = le32_to_cpu(cp_block->checksum_offset);
649 if (crc_offset >= blk_size)
650 goto invalid_cp2;
651
652 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
653 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
654 goto invalid_cp2;
655
656 cur_version = cur_cp_version(cp_block);
657
658 if (cur_version == pre_version) {
659 *version = cur_version;
660 f2fs_put_page(cp_page_2, 1);
661 return cp_page_1;
662 }
663invalid_cp2:
664 f2fs_put_page(cp_page_2, 1);
665invalid_cp1:
666 f2fs_put_page(cp_page_1, 1);
667 return NULL;
668}
669
670int get_valid_checkpoint(struct f2fs_sb_info *sbi)
671{
672 struct f2fs_checkpoint *cp_block;
673 struct f2fs_super_block *fsb = sbi->raw_super;
674 struct page *cp1, *cp2, *cur_page;
675 unsigned long blk_size = sbi->blocksize;
676 unsigned long long cp1_version = 0, cp2_version = 0;
677 unsigned long long cp_start_blk_no;
678 unsigned int cp_blks = 1 + __cp_payload(sbi);
679 block_t cp_blk_no;
680 int i;
681
682 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
683 if (!sbi->ckpt)
684 return -ENOMEM;
685 /*
686 * Finding out valid cp block involves read both
687 * sets( cp pack1 and cp pack 2)
688 */
689 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
690 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
691
692 /* The second checkpoint pack should start at the next segment */
693 cp_start_blk_no += ((unsigned long long)1) <<
694 le32_to_cpu(fsb->log_blocks_per_seg);
695 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
696
697 if (cp1 && cp2) {
698 if (ver_after(cp2_version, cp1_version))
699 cur_page = cp2;
700 else
701 cur_page = cp1;
702 } else if (cp1) {
703 cur_page = cp1;
704 } else if (cp2) {
705 cur_page = cp2;
706 } else {
707 goto fail_no_cp;
708 }
709
710 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
711 memcpy(sbi->ckpt, cp_block, blk_size);
712
713 /* Sanity checking of checkpoint */
714 if (sanity_check_ckpt(sbi))
715 goto fail_no_cp;
716
717 if (cp_blks <= 1)
718 goto done;
719
720 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
721 if (cur_page == cp2)
722 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
723
724 for (i = 1; i < cp_blks; i++) {
725 void *sit_bitmap_ptr;
726 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
727
728 cur_page = get_meta_page(sbi, cp_blk_no + i);
729 sit_bitmap_ptr = page_address(cur_page);
730 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
731 f2fs_put_page(cur_page, 1);
732 }
733done:
734 f2fs_put_page(cp1, 1);
735 f2fs_put_page(cp2, 1);
736 return 0;
737
738fail_no_cp:
739 kfree(sbi->ckpt);
740 return -EINVAL;
741}
742
743static void __add_dirty_inode(struct inode *inode, enum inode_type type)
744{
745 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
746 struct f2fs_inode_info *fi = F2FS_I(inode);
747 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
748
749 if (is_inode_flag_set(fi, flag))
750 return;
751
752 set_inode_flag(fi, flag);
753 list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
754 stat_inc_dirty_inode(sbi, type);
755}
756
757static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
758{
759 struct f2fs_inode_info *fi = F2FS_I(inode);
760 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
761
762 if (get_dirty_pages(inode) ||
763 !is_inode_flag_set(F2FS_I(inode), flag))
764 return;
765
766 list_del_init(&fi->dirty_list);
767 clear_inode_flag(fi, flag);
768 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
769}
770
771void update_dirty_page(struct inode *inode, struct page *page)
772{
773 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
774 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
775
776 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
777 !S_ISLNK(inode->i_mode))
778 return;
779
780 spin_lock(&sbi->inode_lock[type]);
781 __add_dirty_inode(inode, type);
782 inode_inc_dirty_pages(inode);
783 spin_unlock(&sbi->inode_lock[type]);
784
785 SetPagePrivate(page);
786 f2fs_trace_pid(page);
787}
788
789void add_dirty_dir_inode(struct inode *inode)
790{
791 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
792
793 spin_lock(&sbi->inode_lock[DIR_INODE]);
794 __add_dirty_inode(inode, DIR_INODE);
795 spin_unlock(&sbi->inode_lock[DIR_INODE]);
796}
797
798void remove_dirty_inode(struct inode *inode)
799{
800 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
801 struct f2fs_inode_info *fi = F2FS_I(inode);
802 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
803
804 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
805 !S_ISLNK(inode->i_mode))
806 return;
807
808 spin_lock(&sbi->inode_lock[type]);
809 __remove_dirty_inode(inode, type);
810 spin_unlock(&sbi->inode_lock[type]);
811
812 /* Only from the recovery routine */
813 if (is_inode_flag_set(fi, FI_DELAY_IPUT)) {
814 clear_inode_flag(fi, FI_DELAY_IPUT);
815 iput(inode);
816 }
817}
818
819int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
820{
821 struct list_head *head;
822 struct inode *inode;
823 struct f2fs_inode_info *fi;
824 bool is_dir = (type == DIR_INODE);
825
826 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
827 get_pages(sbi, is_dir ?
828 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
829retry:
830 if (unlikely(f2fs_cp_error(sbi)))
831 return -EIO;
832
833 spin_lock(&sbi->inode_lock[type]);
834
835 head = &sbi->inode_list[type];
836 if (list_empty(head)) {
837 spin_unlock(&sbi->inode_lock[type]);
838 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
839 get_pages(sbi, is_dir ?
840 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
841 return 0;
842 }
843 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
844 inode = igrab(&fi->vfs_inode);
845 spin_unlock(&sbi->inode_lock[type]);
846 if (inode) {
847 filemap_fdatawrite(inode->i_mapping);
848 iput(inode);
849 } else {
850 /*
851 * We should submit bio, since it exists several
852 * wribacking dentry pages in the freeing inode.
853 */
854 f2fs_submit_merged_bio(sbi, DATA, WRITE);
855 cond_resched();
856 }
857 goto retry;
858}
859
860/*
861 * Freeze all the FS-operations for checkpoint.
862 */
863static int block_operations(struct f2fs_sb_info *sbi)
864{
865 struct writeback_control wbc = {
866 .sync_mode = WB_SYNC_ALL,
867 .nr_to_write = LONG_MAX,
868 .for_reclaim = 0,
869 };
870 struct blk_plug plug;
871 int err = 0;
872
873 blk_start_plug(&plug);
874
875retry_flush_dents:
876 f2fs_lock_all(sbi);
877 /* write all the dirty dentry pages */
878 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
879 f2fs_unlock_all(sbi);
880 err = sync_dirty_inodes(sbi, DIR_INODE);
881 if (err)
882 goto out;
883 goto retry_flush_dents;
884 }
885
886 /*
887 * POR: we should ensure that there are no dirty node pages
888 * until finishing nat/sit flush.
889 */
890retry_flush_nodes:
891 down_write(&sbi->node_write);
892
893 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
894 up_write(&sbi->node_write);
895 err = sync_node_pages(sbi, 0, &wbc);
896 if (err) {
897 f2fs_unlock_all(sbi);
898 goto out;
899 }
900 goto retry_flush_nodes;
901 }
902out:
903 blk_finish_plug(&plug);
904 return err;
905}
906
907static void unblock_operations(struct f2fs_sb_info *sbi)
908{
909 up_write(&sbi->node_write);
910 f2fs_unlock_all(sbi);
911}
912
913static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
914{
915 DEFINE_WAIT(wait);
916
917 for (;;) {
918 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
919
920 if (!get_pages(sbi, F2FS_WRITEBACK))
921 break;
922
923 io_schedule_timeout(5*HZ);
924 }
925 finish_wait(&sbi->cp_wait, &wait);
926}
927
928static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
929{
930 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
931 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
932 struct f2fs_nm_info *nm_i = NM_I(sbi);
933 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
934 nid_t last_nid = nm_i->next_scan_nid;
935 block_t start_blk;
936 unsigned int data_sum_blocks, orphan_blocks;
937 __u32 crc32 = 0;
938 int i;
939 int cp_payload_blks = __cp_payload(sbi);
940 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
941 bool invalidate = false;
942 struct super_block *sb = sbi->sb;
943 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
944 u64 kbytes_written;
945
946 /*
947 * This avoids to conduct wrong roll-forward operations and uses
948 * metapages, so should be called prior to sync_meta_pages below.
949 */
950 if (discard_next_dnode(sbi, discard_blk))
951 invalidate = true;
952
953 /* Flush all the NAT/SIT pages */
954 while (get_pages(sbi, F2FS_DIRTY_META)) {
955 sync_meta_pages(sbi, META, LONG_MAX);
956 if (unlikely(f2fs_cp_error(sbi)))
957 return -EIO;
958 }
959
960 next_free_nid(sbi, &last_nid);
961
962 /*
963 * modify checkpoint
964 * version number is already updated
965 */
966 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
967 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
968 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
969 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
970 ckpt->cur_node_segno[i] =
971 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
972 ckpt->cur_node_blkoff[i] =
973 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
974 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
975 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
976 }
977 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
978 ckpt->cur_data_segno[i] =
979 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
980 ckpt->cur_data_blkoff[i] =
981 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
982 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
983 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
984 }
985
986 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
987 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
988 ckpt->next_free_nid = cpu_to_le32(last_nid);
989
990 /* 2 cp + n data seg summary + orphan inode blocks */
991 data_sum_blocks = npages_for_summary_flush(sbi, false);
992 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
993 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
994 else
995 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
996
997 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
998 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
999 orphan_blocks);
1000
1001 if (__remain_node_summaries(cpc->reason))
1002 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1003 cp_payload_blks + data_sum_blocks +
1004 orphan_blocks + NR_CURSEG_NODE_TYPE);
1005 else
1006 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1007 cp_payload_blks + data_sum_blocks +
1008 orphan_blocks);
1009
1010 if (cpc->reason == CP_UMOUNT)
1011 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1012 else
1013 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1014
1015 if (cpc->reason == CP_FASTBOOT)
1016 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1017 else
1018 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1019
1020 if (orphan_num)
1021 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1022 else
1023 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1024
1025 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1026 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1027
1028 /* update SIT/NAT bitmap */
1029 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1030 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1031
1032 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1033 *((__le32 *)((unsigned char *)ckpt +
1034 le32_to_cpu(ckpt->checksum_offset)))
1035 = cpu_to_le32(crc32);
1036
1037 start_blk = __start_cp_addr(sbi);
1038
1039 /* need to wait for end_io results */
1040 wait_on_all_pages_writeback(sbi);
1041 if (unlikely(f2fs_cp_error(sbi)))
1042 return -EIO;
1043
1044 /* write out checkpoint buffer at block 0 */
1045 update_meta_page(sbi, ckpt, start_blk++);
1046
1047 for (i = 1; i < 1 + cp_payload_blks; i++)
1048 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1049 start_blk++);
1050
1051 if (orphan_num) {
1052 write_orphan_inodes(sbi, start_blk);
1053 start_blk += orphan_blocks;
1054 }
1055
1056 write_data_summaries(sbi, start_blk);
1057 start_blk += data_sum_blocks;
1058
1059 /* Record write statistics in the hot node summary */
1060 kbytes_written = sbi->kbytes_written;
1061 if (sb->s_bdev->bd_part)
1062 kbytes_written += BD_PART_WRITTEN(sbi);
1063
1064 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1065
1066 if (__remain_node_summaries(cpc->reason)) {
1067 write_node_summaries(sbi, start_blk);
1068 start_blk += NR_CURSEG_NODE_TYPE;
1069 }
1070
1071 /* writeout checkpoint block */
1072 update_meta_page(sbi, ckpt, start_blk);
1073
1074 /* wait for previous submitted node/meta pages writeback */
1075 wait_on_all_pages_writeback(sbi);
1076
1077 if (unlikely(f2fs_cp_error(sbi)))
1078 return -EIO;
1079
1080 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1081 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1082
1083 /* update user_block_counts */
1084 sbi->last_valid_block_count = sbi->total_valid_block_count;
1085 sbi->alloc_valid_block_count = 0;
1086
1087 /* Here, we only have one bio having CP pack */
1088 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1089
1090 /* wait for previous submitted meta pages writeback */
1091 wait_on_all_pages_writeback(sbi);
1092
1093 /*
1094 * invalidate meta page which is used temporarily for zeroing out
1095 * block at the end of warm node chain.
1096 */
1097 if (invalidate)
1098 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1099 discard_blk);
1100
1101 release_ino_entry(sbi);
1102
1103 if (unlikely(f2fs_cp_error(sbi)))
1104 return -EIO;
1105
1106 clear_prefree_segments(sbi, cpc);
1107 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1108
1109 return 0;
1110}
1111
1112/*
1113 * We guarantee that this checkpoint procedure will not fail.
1114 */
1115int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1116{
1117 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1118 unsigned long long ckpt_ver;
1119 int err = 0;
1120
1121 mutex_lock(&sbi->cp_mutex);
1122
1123 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1124 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1125 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1126 goto out;
1127 if (unlikely(f2fs_cp_error(sbi))) {
1128 err = -EIO;
1129 goto out;
1130 }
1131 if (f2fs_readonly(sbi->sb)) {
1132 err = -EROFS;
1133 goto out;
1134 }
1135
1136 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1137
1138 err = block_operations(sbi);
1139 if (err)
1140 goto out;
1141
1142 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1143
1144 f2fs_flush_merged_bios(sbi);
1145
1146 /*
1147 * update checkpoint pack index
1148 * Increase the version number so that
1149 * SIT entries and seg summaries are written at correct place
1150 */
1151 ckpt_ver = cur_cp_version(ckpt);
1152 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1153
1154 /* write cached NAT/SIT entries to NAT/SIT area */
1155 flush_nat_entries(sbi);
1156 flush_sit_entries(sbi, cpc);
1157
1158 /* unlock all the fs_lock[] in do_checkpoint() */
1159 err = do_checkpoint(sbi, cpc);
1160
1161 unblock_operations(sbi);
1162 stat_inc_cp_count(sbi->stat_info);
1163
1164 if (cpc->reason == CP_RECOVERY)
1165 f2fs_msg(sbi->sb, KERN_NOTICE,
1166 "checkpoint: version = %llx", ckpt_ver);
1167
1168 /* do checkpoint periodically */
1169 f2fs_update_time(sbi, CP_TIME);
1170 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1171out:
1172 mutex_unlock(&sbi->cp_mutex);
1173 return err;
1174}
1175
1176void init_ino_entry_info(struct f2fs_sb_info *sbi)
1177{
1178 int i;
1179
1180 for (i = 0; i < MAX_INO_ENTRY; i++) {
1181 struct inode_management *im = &sbi->im[i];
1182
1183 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1184 spin_lock_init(&im->ino_lock);
1185 INIT_LIST_HEAD(&im->ino_list);
1186 im->ino_num = 0;
1187 }
1188
1189 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1190 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1191 F2FS_ORPHANS_PER_BLOCK;
1192}
1193
1194int __init create_checkpoint_caches(void)
1195{
1196 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1197 sizeof(struct ino_entry));
1198 if (!ino_entry_slab)
1199 return -ENOMEM;
1200 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1201 sizeof(struct inode_entry));
1202 if (!inode_entry_slab) {
1203 kmem_cache_destroy(ino_entry_slab);
1204 return -ENOMEM;
1205 }
1206 return 0;
1207}
1208
1209void destroy_checkpoint_caches(void)
1210{
1211 kmem_cache_destroy(ino_entry_slab);
1212 kmem_cache_destroy(inode_entry_slab);
1213}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/f2fs/checkpoint.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8#include <linux/fs.h>
9#include <linux/bio.h>
10#include <linux/mpage.h>
11#include <linux/writeback.h>
12#include <linux/blkdev.h>
13#include <linux/f2fs_fs.h>
14#include <linux/pagevec.h>
15#include <linux/swap.h>
16#include <linux/kthread.h>
17
18#include "f2fs.h"
19#include "node.h"
20#include "segment.h"
21#include "iostat.h"
22#include <trace/events/f2fs.h>
23
24#define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
25
26static struct kmem_cache *ino_entry_slab;
27struct kmem_cache *f2fs_inode_entry_slab;
28
29void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io,
30 unsigned char reason)
31{
32 f2fs_build_fault_attr(sbi, 0, 0);
33 if (!end_io)
34 f2fs_flush_merged_writes(sbi);
35 f2fs_handle_critical_error(sbi, reason, end_io);
36}
37
38/*
39 * We guarantee no failure on the returned page.
40 */
41struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
42{
43 struct address_space *mapping = META_MAPPING(sbi);
44 struct page *page;
45repeat:
46 page = f2fs_grab_cache_page(mapping, index, false);
47 if (!page) {
48 cond_resched();
49 goto repeat;
50 }
51 f2fs_wait_on_page_writeback(page, META, true, true);
52 if (!PageUptodate(page))
53 SetPageUptodate(page);
54 return page;
55}
56
57static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
58 bool is_meta)
59{
60 struct address_space *mapping = META_MAPPING(sbi);
61 struct page *page;
62 struct f2fs_io_info fio = {
63 .sbi = sbi,
64 .type = META,
65 .op = REQ_OP_READ,
66 .op_flags = REQ_META | REQ_PRIO,
67 .old_blkaddr = index,
68 .new_blkaddr = index,
69 .encrypted_page = NULL,
70 .is_por = !is_meta ? 1 : 0,
71 };
72 int err;
73
74 if (unlikely(!is_meta))
75 fio.op_flags &= ~REQ_META;
76repeat:
77 page = f2fs_grab_cache_page(mapping, index, false);
78 if (!page) {
79 cond_resched();
80 goto repeat;
81 }
82 if (PageUptodate(page))
83 goto out;
84
85 fio.page = page;
86
87 err = f2fs_submit_page_bio(&fio);
88 if (err) {
89 f2fs_put_page(page, 1);
90 return ERR_PTR(err);
91 }
92
93 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
94
95 lock_page(page);
96 if (unlikely(page->mapping != mapping)) {
97 f2fs_put_page(page, 1);
98 goto repeat;
99 }
100
101 if (unlikely(!PageUptodate(page))) {
102 f2fs_handle_page_eio(sbi, page->index, META);
103 f2fs_put_page(page, 1);
104 return ERR_PTR(-EIO);
105 }
106out:
107 return page;
108}
109
110struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
111{
112 return __get_meta_page(sbi, index, true);
113}
114
115struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
116{
117 struct page *page;
118 int count = 0;
119
120retry:
121 page = __get_meta_page(sbi, index, true);
122 if (IS_ERR(page)) {
123 if (PTR_ERR(page) == -EIO &&
124 ++count <= DEFAULT_RETRY_IO_COUNT)
125 goto retry;
126 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
127 }
128 return page;
129}
130
131/* for POR only */
132struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
133{
134 return __get_meta_page(sbi, index, false);
135}
136
137static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
138 int type)
139{
140 struct seg_entry *se;
141 unsigned int segno, offset;
142 bool exist;
143
144 if (type == DATA_GENERIC)
145 return true;
146
147 segno = GET_SEGNO(sbi, blkaddr);
148 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
149 se = get_seg_entry(sbi, segno);
150
151 exist = f2fs_test_bit(offset, se->cur_valid_map);
152
153 /* skip data, if we already have an error in checkpoint. */
154 if (unlikely(f2fs_cp_error(sbi)))
155 return exist;
156
157 if ((exist && type == DATA_GENERIC_ENHANCE_UPDATE) ||
158 (!exist && type == DATA_GENERIC_ENHANCE))
159 goto out_err;
160 if (!exist && type != DATA_GENERIC_ENHANCE_UPDATE)
161 goto out_handle;
162 return exist;
163
164out_err:
165 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
166 blkaddr, exist);
167 set_sbi_flag(sbi, SBI_NEED_FSCK);
168 dump_stack();
169out_handle:
170 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
171 return exist;
172}
173
174static bool __f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
175 block_t blkaddr, int type)
176{
177 switch (type) {
178 case META_NAT:
179 break;
180 case META_SIT:
181 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
182 goto check_only;
183 break;
184 case META_SSA:
185 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
186 blkaddr < SM_I(sbi)->ssa_blkaddr))
187 goto check_only;
188 break;
189 case META_CP:
190 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
191 blkaddr < __start_cp_addr(sbi)))
192 goto check_only;
193 break;
194 case META_POR:
195 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
196 blkaddr < MAIN_BLKADDR(sbi)))
197 goto check_only;
198 break;
199 case DATA_GENERIC:
200 case DATA_GENERIC_ENHANCE:
201 case DATA_GENERIC_ENHANCE_READ:
202 case DATA_GENERIC_ENHANCE_UPDATE:
203 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
204 blkaddr < MAIN_BLKADDR(sbi))) {
205
206 /* Skip to emit an error message. */
207 if (unlikely(f2fs_cp_error(sbi)))
208 return false;
209
210 f2fs_warn(sbi, "access invalid blkaddr:%u",
211 blkaddr);
212 set_sbi_flag(sbi, SBI_NEED_FSCK);
213 dump_stack();
214 goto err;
215 } else {
216 return __is_bitmap_valid(sbi, blkaddr, type);
217 }
218 break;
219 case META_GENERIC:
220 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
221 blkaddr >= MAIN_BLKADDR(sbi)))
222 goto err;
223 break;
224 default:
225 BUG();
226 }
227
228 return true;
229err:
230 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
231check_only:
232 return false;
233}
234
235bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
236 block_t blkaddr, int type)
237{
238 if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY))
239 return false;
240 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
241}
242
243bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi,
244 block_t blkaddr, int type)
245{
246 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
247}
248
249/*
250 * Readahead CP/NAT/SIT/SSA/POR pages
251 */
252int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
253 int type, bool sync)
254{
255 struct page *page;
256 block_t blkno = start;
257 struct f2fs_io_info fio = {
258 .sbi = sbi,
259 .type = META,
260 .op = REQ_OP_READ,
261 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
262 .encrypted_page = NULL,
263 .in_list = 0,
264 .is_por = (type == META_POR) ? 1 : 0,
265 };
266 struct blk_plug plug;
267 int err;
268
269 if (unlikely(type == META_POR))
270 fio.op_flags &= ~REQ_META;
271
272 blk_start_plug(&plug);
273 for (; nrpages-- > 0; blkno++) {
274
275 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
276 goto out;
277
278 switch (type) {
279 case META_NAT:
280 if (unlikely(blkno >=
281 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
282 blkno = 0;
283 /* get nat block addr */
284 fio.new_blkaddr = current_nat_addr(sbi,
285 blkno * NAT_ENTRY_PER_BLOCK);
286 break;
287 case META_SIT:
288 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
289 goto out;
290 /* get sit block addr */
291 fio.new_blkaddr = current_sit_addr(sbi,
292 blkno * SIT_ENTRY_PER_BLOCK);
293 break;
294 case META_SSA:
295 case META_CP:
296 case META_POR:
297 fio.new_blkaddr = blkno;
298 break;
299 default:
300 BUG();
301 }
302
303 page = f2fs_grab_cache_page(META_MAPPING(sbi),
304 fio.new_blkaddr, false);
305 if (!page)
306 continue;
307 if (PageUptodate(page)) {
308 f2fs_put_page(page, 1);
309 continue;
310 }
311
312 fio.page = page;
313 err = f2fs_submit_page_bio(&fio);
314 f2fs_put_page(page, err ? 1 : 0);
315
316 if (!err)
317 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
318 F2FS_BLKSIZE);
319 }
320out:
321 blk_finish_plug(&plug);
322 return blkno - start;
323}
324
325void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
326 unsigned int ra_blocks)
327{
328 struct page *page;
329 bool readahead = false;
330
331 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
332 return;
333
334 page = find_get_page(META_MAPPING(sbi), index);
335 if (!page || !PageUptodate(page))
336 readahead = true;
337 f2fs_put_page(page, 0);
338
339 if (readahead)
340 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
341}
342
343static int __f2fs_write_meta_page(struct page *page,
344 struct writeback_control *wbc,
345 enum iostat_type io_type)
346{
347 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
348
349 trace_f2fs_writepage(page, META);
350
351 if (unlikely(f2fs_cp_error(sbi))) {
352 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
353 ClearPageUptodate(page);
354 dec_page_count(sbi, F2FS_DIRTY_META);
355 unlock_page(page);
356 return 0;
357 }
358 goto redirty_out;
359 }
360 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
361 goto redirty_out;
362 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
363 goto redirty_out;
364
365 f2fs_do_write_meta_page(sbi, page, io_type);
366 dec_page_count(sbi, F2FS_DIRTY_META);
367
368 if (wbc->for_reclaim)
369 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
370
371 unlock_page(page);
372
373 if (unlikely(f2fs_cp_error(sbi)))
374 f2fs_submit_merged_write(sbi, META);
375
376 return 0;
377
378redirty_out:
379 redirty_page_for_writepage(wbc, page);
380 return AOP_WRITEPAGE_ACTIVATE;
381}
382
383static int f2fs_write_meta_page(struct page *page,
384 struct writeback_control *wbc)
385{
386 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
387}
388
389static int f2fs_write_meta_pages(struct address_space *mapping,
390 struct writeback_control *wbc)
391{
392 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
393 long diff, written;
394
395 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
396 goto skip_write;
397
398 /* collect a number of dirty meta pages and write together */
399 if (wbc->sync_mode != WB_SYNC_ALL &&
400 get_pages(sbi, F2FS_DIRTY_META) <
401 nr_pages_to_skip(sbi, META))
402 goto skip_write;
403
404 /* if locked failed, cp will flush dirty pages instead */
405 if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
406 goto skip_write;
407
408 trace_f2fs_writepages(mapping->host, wbc, META);
409 diff = nr_pages_to_write(sbi, META, wbc);
410 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
411 f2fs_up_write(&sbi->cp_global_sem);
412 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
413 return 0;
414
415skip_write:
416 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
417 trace_f2fs_writepages(mapping->host, wbc, META);
418 return 0;
419}
420
421long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
422 long nr_to_write, enum iostat_type io_type)
423{
424 struct address_space *mapping = META_MAPPING(sbi);
425 pgoff_t index = 0, prev = ULONG_MAX;
426 struct folio_batch fbatch;
427 long nwritten = 0;
428 int nr_folios;
429 struct writeback_control wbc = {
430 .for_reclaim = 0,
431 };
432 struct blk_plug plug;
433
434 folio_batch_init(&fbatch);
435
436 blk_start_plug(&plug);
437
438 while ((nr_folios = filemap_get_folios_tag(mapping, &index,
439 (pgoff_t)-1,
440 PAGECACHE_TAG_DIRTY, &fbatch))) {
441 int i;
442
443 for (i = 0; i < nr_folios; i++) {
444 struct folio *folio = fbatch.folios[i];
445
446 if (nr_to_write != LONG_MAX && i != 0 &&
447 folio->index != prev +
448 folio_nr_pages(fbatch.folios[i-1])) {
449 folio_batch_release(&fbatch);
450 goto stop;
451 }
452
453 folio_lock(folio);
454
455 if (unlikely(folio->mapping != mapping)) {
456continue_unlock:
457 folio_unlock(folio);
458 continue;
459 }
460 if (!folio_test_dirty(folio)) {
461 /* someone wrote it for us */
462 goto continue_unlock;
463 }
464
465 f2fs_wait_on_page_writeback(&folio->page, META,
466 true, true);
467
468 if (!folio_clear_dirty_for_io(folio))
469 goto continue_unlock;
470
471 if (__f2fs_write_meta_page(&folio->page, &wbc,
472 io_type)) {
473 folio_unlock(folio);
474 break;
475 }
476 nwritten += folio_nr_pages(folio);
477 prev = folio->index;
478 if (unlikely(nwritten >= nr_to_write))
479 break;
480 }
481 folio_batch_release(&fbatch);
482 cond_resched();
483 }
484stop:
485 if (nwritten)
486 f2fs_submit_merged_write(sbi, type);
487
488 blk_finish_plug(&plug);
489
490 return nwritten;
491}
492
493static bool f2fs_dirty_meta_folio(struct address_space *mapping,
494 struct folio *folio)
495{
496 trace_f2fs_set_page_dirty(&folio->page, META);
497
498 if (!folio_test_uptodate(folio))
499 folio_mark_uptodate(folio);
500 if (filemap_dirty_folio(mapping, folio)) {
501 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
502 set_page_private_reference(&folio->page);
503 return true;
504 }
505 return false;
506}
507
508const struct address_space_operations f2fs_meta_aops = {
509 .writepage = f2fs_write_meta_page,
510 .writepages = f2fs_write_meta_pages,
511 .dirty_folio = f2fs_dirty_meta_folio,
512 .invalidate_folio = f2fs_invalidate_folio,
513 .release_folio = f2fs_release_folio,
514 .migrate_folio = filemap_migrate_folio,
515};
516
517static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
518 unsigned int devidx, int type)
519{
520 struct inode_management *im = &sbi->im[type];
521 struct ino_entry *e = NULL, *new = NULL;
522
523 if (type == FLUSH_INO) {
524 rcu_read_lock();
525 e = radix_tree_lookup(&im->ino_root, ino);
526 rcu_read_unlock();
527 }
528
529retry:
530 if (!e)
531 new = f2fs_kmem_cache_alloc(ino_entry_slab,
532 GFP_NOFS, true, NULL);
533
534 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
535
536 spin_lock(&im->ino_lock);
537 e = radix_tree_lookup(&im->ino_root, ino);
538 if (!e) {
539 if (!new) {
540 spin_unlock(&im->ino_lock);
541 radix_tree_preload_end();
542 goto retry;
543 }
544 e = new;
545 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
546 f2fs_bug_on(sbi, 1);
547
548 memset(e, 0, sizeof(struct ino_entry));
549 e->ino = ino;
550
551 list_add_tail(&e->list, &im->ino_list);
552 if (type != ORPHAN_INO)
553 im->ino_num++;
554 }
555
556 if (type == FLUSH_INO)
557 f2fs_set_bit(devidx, (char *)&e->dirty_device);
558
559 spin_unlock(&im->ino_lock);
560 radix_tree_preload_end();
561
562 if (new && e != new)
563 kmem_cache_free(ino_entry_slab, new);
564}
565
566static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
567{
568 struct inode_management *im = &sbi->im[type];
569 struct ino_entry *e;
570
571 spin_lock(&im->ino_lock);
572 e = radix_tree_lookup(&im->ino_root, ino);
573 if (e) {
574 list_del(&e->list);
575 radix_tree_delete(&im->ino_root, ino);
576 im->ino_num--;
577 spin_unlock(&im->ino_lock);
578 kmem_cache_free(ino_entry_slab, e);
579 return;
580 }
581 spin_unlock(&im->ino_lock);
582}
583
584void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
585{
586 /* add new dirty ino entry into list */
587 __add_ino_entry(sbi, ino, 0, type);
588}
589
590void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
591{
592 /* remove dirty ino entry from list */
593 __remove_ino_entry(sbi, ino, type);
594}
595
596/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
597bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
598{
599 struct inode_management *im = &sbi->im[mode];
600 struct ino_entry *e;
601
602 spin_lock(&im->ino_lock);
603 e = radix_tree_lookup(&im->ino_root, ino);
604 spin_unlock(&im->ino_lock);
605 return e ? true : false;
606}
607
608void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
609{
610 struct ino_entry *e, *tmp;
611 int i;
612
613 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
614 struct inode_management *im = &sbi->im[i];
615
616 spin_lock(&im->ino_lock);
617 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
618 list_del(&e->list);
619 radix_tree_delete(&im->ino_root, e->ino);
620 kmem_cache_free(ino_entry_slab, e);
621 im->ino_num--;
622 }
623 spin_unlock(&im->ino_lock);
624 }
625}
626
627void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
628 unsigned int devidx, int type)
629{
630 __add_ino_entry(sbi, ino, devidx, type);
631}
632
633bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
634 unsigned int devidx, int type)
635{
636 struct inode_management *im = &sbi->im[type];
637 struct ino_entry *e;
638 bool is_dirty = false;
639
640 spin_lock(&im->ino_lock);
641 e = radix_tree_lookup(&im->ino_root, ino);
642 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
643 is_dirty = true;
644 spin_unlock(&im->ino_lock);
645 return is_dirty;
646}
647
648int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
649{
650 struct inode_management *im = &sbi->im[ORPHAN_INO];
651 int err = 0;
652
653 spin_lock(&im->ino_lock);
654
655 if (time_to_inject(sbi, FAULT_ORPHAN)) {
656 spin_unlock(&im->ino_lock);
657 return -ENOSPC;
658 }
659
660 if (unlikely(im->ino_num >= sbi->max_orphans))
661 err = -ENOSPC;
662 else
663 im->ino_num++;
664 spin_unlock(&im->ino_lock);
665
666 return err;
667}
668
669void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
670{
671 struct inode_management *im = &sbi->im[ORPHAN_INO];
672
673 spin_lock(&im->ino_lock);
674 f2fs_bug_on(sbi, im->ino_num == 0);
675 im->ino_num--;
676 spin_unlock(&im->ino_lock);
677}
678
679void f2fs_add_orphan_inode(struct inode *inode)
680{
681 /* add new orphan ino entry into list */
682 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
683 f2fs_update_inode_page(inode);
684}
685
686void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
687{
688 /* remove orphan entry from orphan list */
689 __remove_ino_entry(sbi, ino, ORPHAN_INO);
690}
691
692static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
693{
694 struct inode *inode;
695 struct node_info ni;
696 int err;
697
698 inode = f2fs_iget_retry(sbi->sb, ino);
699 if (IS_ERR(inode)) {
700 /*
701 * there should be a bug that we can't find the entry
702 * to orphan inode.
703 */
704 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
705 return PTR_ERR(inode);
706 }
707
708 err = f2fs_dquot_initialize(inode);
709 if (err) {
710 iput(inode);
711 goto err_out;
712 }
713
714 clear_nlink(inode);
715
716 /* truncate all the data during iput */
717 iput(inode);
718
719 err = f2fs_get_node_info(sbi, ino, &ni, false);
720 if (err)
721 goto err_out;
722
723 /* ENOMEM was fully retried in f2fs_evict_inode. */
724 if (ni.blk_addr != NULL_ADDR) {
725 err = -EIO;
726 goto err_out;
727 }
728 return 0;
729
730err_out:
731 set_sbi_flag(sbi, SBI_NEED_FSCK);
732 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
733 __func__, ino);
734 return err;
735}
736
737int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
738{
739 block_t start_blk, orphan_blocks, i, j;
740 int err = 0;
741
742 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
743 return 0;
744
745 if (f2fs_hw_is_readonly(sbi)) {
746 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
747 return 0;
748 }
749
750 if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
751 f2fs_info(sbi, "orphan cleanup on readonly fs");
752
753 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
754 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
755
756 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
757
758 for (i = 0; i < orphan_blocks; i++) {
759 struct page *page;
760 struct f2fs_orphan_block *orphan_blk;
761
762 page = f2fs_get_meta_page(sbi, start_blk + i);
763 if (IS_ERR(page)) {
764 err = PTR_ERR(page);
765 goto out;
766 }
767
768 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
769 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
770 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
771
772 err = recover_orphan_inode(sbi, ino);
773 if (err) {
774 f2fs_put_page(page, 1);
775 goto out;
776 }
777 }
778 f2fs_put_page(page, 1);
779 }
780 /* clear Orphan Flag */
781 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
782out:
783 set_sbi_flag(sbi, SBI_IS_RECOVERED);
784
785 return err;
786}
787
788static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
789{
790 struct list_head *head;
791 struct f2fs_orphan_block *orphan_blk = NULL;
792 unsigned int nentries = 0;
793 unsigned short index = 1;
794 unsigned short orphan_blocks;
795 struct page *page = NULL;
796 struct ino_entry *orphan = NULL;
797 struct inode_management *im = &sbi->im[ORPHAN_INO];
798
799 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
800
801 /*
802 * we don't need to do spin_lock(&im->ino_lock) here, since all the
803 * orphan inode operations are covered under f2fs_lock_op().
804 * And, spin_lock should be avoided due to page operations below.
805 */
806 head = &im->ino_list;
807
808 /* loop for each orphan inode entry and write them in journal block */
809 list_for_each_entry(orphan, head, list) {
810 if (!page) {
811 page = f2fs_grab_meta_page(sbi, start_blk++);
812 orphan_blk =
813 (struct f2fs_orphan_block *)page_address(page);
814 memset(orphan_blk, 0, sizeof(*orphan_blk));
815 }
816
817 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
818
819 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
820 /*
821 * an orphan block is full of 1020 entries,
822 * then we need to flush current orphan blocks
823 * and bring another one in memory
824 */
825 orphan_blk->blk_addr = cpu_to_le16(index);
826 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
827 orphan_blk->entry_count = cpu_to_le32(nentries);
828 set_page_dirty(page);
829 f2fs_put_page(page, 1);
830 index++;
831 nentries = 0;
832 page = NULL;
833 }
834 }
835
836 if (page) {
837 orphan_blk->blk_addr = cpu_to_le16(index);
838 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
839 orphan_blk->entry_count = cpu_to_le32(nentries);
840 set_page_dirty(page);
841 f2fs_put_page(page, 1);
842 }
843}
844
845static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
846 struct f2fs_checkpoint *ckpt)
847{
848 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
849 __u32 chksum;
850
851 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
852 if (chksum_ofs < CP_CHKSUM_OFFSET) {
853 chksum_ofs += sizeof(chksum);
854 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
855 F2FS_BLKSIZE - chksum_ofs);
856 }
857 return chksum;
858}
859
860static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
861 struct f2fs_checkpoint **cp_block, struct page **cp_page,
862 unsigned long long *version)
863{
864 size_t crc_offset = 0;
865 __u32 crc;
866
867 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
868 if (IS_ERR(*cp_page))
869 return PTR_ERR(*cp_page);
870
871 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
872
873 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
874 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
875 crc_offset > CP_CHKSUM_OFFSET) {
876 f2fs_put_page(*cp_page, 1);
877 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
878 return -EINVAL;
879 }
880
881 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
882 if (crc != cur_cp_crc(*cp_block)) {
883 f2fs_put_page(*cp_page, 1);
884 f2fs_warn(sbi, "invalid crc value");
885 return -EINVAL;
886 }
887
888 *version = cur_cp_version(*cp_block);
889 return 0;
890}
891
892static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
893 block_t cp_addr, unsigned long long *version)
894{
895 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
896 struct f2fs_checkpoint *cp_block = NULL;
897 unsigned long long cur_version = 0, pre_version = 0;
898 unsigned int cp_blocks;
899 int err;
900
901 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
902 &cp_page_1, version);
903 if (err)
904 return NULL;
905
906 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
907
908 if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
909 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
910 le32_to_cpu(cp_block->cp_pack_total_block_count));
911 goto invalid_cp;
912 }
913 pre_version = *version;
914
915 cp_addr += cp_blocks - 1;
916 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
917 &cp_page_2, version);
918 if (err)
919 goto invalid_cp;
920 cur_version = *version;
921
922 if (cur_version == pre_version) {
923 *version = cur_version;
924 f2fs_put_page(cp_page_2, 1);
925 return cp_page_1;
926 }
927 f2fs_put_page(cp_page_2, 1);
928invalid_cp:
929 f2fs_put_page(cp_page_1, 1);
930 return NULL;
931}
932
933int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
934{
935 struct f2fs_checkpoint *cp_block;
936 struct f2fs_super_block *fsb = sbi->raw_super;
937 struct page *cp1, *cp2, *cur_page;
938 unsigned long blk_size = sbi->blocksize;
939 unsigned long long cp1_version = 0, cp2_version = 0;
940 unsigned long long cp_start_blk_no;
941 unsigned int cp_blks = 1 + __cp_payload(sbi);
942 block_t cp_blk_no;
943 int i;
944 int err;
945
946 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
947 GFP_KERNEL);
948 if (!sbi->ckpt)
949 return -ENOMEM;
950 /*
951 * Finding out valid cp block involves read both
952 * sets( cp pack 1 and cp pack 2)
953 */
954 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
955 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
956
957 /* The second checkpoint pack should start at the next segment */
958 cp_start_blk_no += ((unsigned long long)1) <<
959 le32_to_cpu(fsb->log_blocks_per_seg);
960 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
961
962 if (cp1 && cp2) {
963 if (ver_after(cp2_version, cp1_version))
964 cur_page = cp2;
965 else
966 cur_page = cp1;
967 } else if (cp1) {
968 cur_page = cp1;
969 } else if (cp2) {
970 cur_page = cp2;
971 } else {
972 err = -EFSCORRUPTED;
973 goto fail_no_cp;
974 }
975
976 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
977 memcpy(sbi->ckpt, cp_block, blk_size);
978
979 if (cur_page == cp1)
980 sbi->cur_cp_pack = 1;
981 else
982 sbi->cur_cp_pack = 2;
983
984 /* Sanity checking of checkpoint */
985 if (f2fs_sanity_check_ckpt(sbi)) {
986 err = -EFSCORRUPTED;
987 goto free_fail_no_cp;
988 }
989
990 if (cp_blks <= 1)
991 goto done;
992
993 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
994 if (cur_page == cp2)
995 cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
996
997 for (i = 1; i < cp_blks; i++) {
998 void *sit_bitmap_ptr;
999 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
1000
1001 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
1002 if (IS_ERR(cur_page)) {
1003 err = PTR_ERR(cur_page);
1004 goto free_fail_no_cp;
1005 }
1006 sit_bitmap_ptr = page_address(cur_page);
1007 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1008 f2fs_put_page(cur_page, 1);
1009 }
1010done:
1011 f2fs_put_page(cp1, 1);
1012 f2fs_put_page(cp2, 1);
1013 return 0;
1014
1015free_fail_no_cp:
1016 f2fs_put_page(cp1, 1);
1017 f2fs_put_page(cp2, 1);
1018fail_no_cp:
1019 kvfree(sbi->ckpt);
1020 return err;
1021}
1022
1023static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1024{
1025 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1026 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1027
1028 if (is_inode_flag_set(inode, flag))
1029 return;
1030
1031 set_inode_flag(inode, flag);
1032 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1033 stat_inc_dirty_inode(sbi, type);
1034}
1035
1036static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1037{
1038 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1039
1040 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1041 return;
1042
1043 list_del_init(&F2FS_I(inode)->dirty_list);
1044 clear_inode_flag(inode, flag);
1045 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1046}
1047
1048void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1049{
1050 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1051 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1052
1053 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1054 !S_ISLNK(inode->i_mode))
1055 return;
1056
1057 spin_lock(&sbi->inode_lock[type]);
1058 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1059 __add_dirty_inode(inode, type);
1060 inode_inc_dirty_pages(inode);
1061 spin_unlock(&sbi->inode_lock[type]);
1062
1063 set_page_private_reference(&folio->page);
1064}
1065
1066void f2fs_remove_dirty_inode(struct inode *inode)
1067{
1068 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1069 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1070
1071 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1072 !S_ISLNK(inode->i_mode))
1073 return;
1074
1075 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1076 return;
1077
1078 spin_lock(&sbi->inode_lock[type]);
1079 __remove_dirty_inode(inode, type);
1080 spin_unlock(&sbi->inode_lock[type]);
1081}
1082
1083int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1084 bool from_cp)
1085{
1086 struct list_head *head;
1087 struct inode *inode;
1088 struct f2fs_inode_info *fi;
1089 bool is_dir = (type == DIR_INODE);
1090 unsigned long ino = 0;
1091
1092 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1093 get_pages(sbi, is_dir ?
1094 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1095retry:
1096 if (unlikely(f2fs_cp_error(sbi))) {
1097 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1098 get_pages(sbi, is_dir ?
1099 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1100 return -EIO;
1101 }
1102
1103 spin_lock(&sbi->inode_lock[type]);
1104
1105 head = &sbi->inode_list[type];
1106 if (list_empty(head)) {
1107 spin_unlock(&sbi->inode_lock[type]);
1108 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1109 get_pages(sbi, is_dir ?
1110 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1111 return 0;
1112 }
1113 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1114 inode = igrab(&fi->vfs_inode);
1115 spin_unlock(&sbi->inode_lock[type]);
1116 if (inode) {
1117 unsigned long cur_ino = inode->i_ino;
1118
1119 if (from_cp)
1120 F2FS_I(inode)->cp_task = current;
1121 F2FS_I(inode)->wb_task = current;
1122
1123 filemap_fdatawrite(inode->i_mapping);
1124
1125 F2FS_I(inode)->wb_task = NULL;
1126 if (from_cp)
1127 F2FS_I(inode)->cp_task = NULL;
1128
1129 iput(inode);
1130 /* We need to give cpu to another writers. */
1131 if (ino == cur_ino)
1132 cond_resched();
1133 else
1134 ino = cur_ino;
1135 } else {
1136 /*
1137 * We should submit bio, since it exists several
1138 * writebacking dentry pages in the freeing inode.
1139 */
1140 f2fs_submit_merged_write(sbi, DATA);
1141 cond_resched();
1142 }
1143 goto retry;
1144}
1145
1146static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1147{
1148 struct list_head *head = &sbi->inode_list[DIRTY_META];
1149 struct inode *inode;
1150 struct f2fs_inode_info *fi;
1151 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1152
1153 while (total--) {
1154 if (unlikely(f2fs_cp_error(sbi)))
1155 return -EIO;
1156
1157 spin_lock(&sbi->inode_lock[DIRTY_META]);
1158 if (list_empty(head)) {
1159 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1160 return 0;
1161 }
1162 fi = list_first_entry(head, struct f2fs_inode_info,
1163 gdirty_list);
1164 inode = igrab(&fi->vfs_inode);
1165 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1166 if (inode) {
1167 sync_inode_metadata(inode, 0);
1168
1169 /* it's on eviction */
1170 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1171 f2fs_update_inode_page(inode);
1172 iput(inode);
1173 }
1174 }
1175 return 0;
1176}
1177
1178static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1179{
1180 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1181 struct f2fs_nm_info *nm_i = NM_I(sbi);
1182 nid_t last_nid = nm_i->next_scan_nid;
1183
1184 next_free_nid(sbi, &last_nid);
1185 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1186 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1187 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1188 ckpt->next_free_nid = cpu_to_le32(last_nid);
1189}
1190
1191static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1192{
1193 bool ret = false;
1194
1195 if (!is_journalled_quota(sbi))
1196 return false;
1197
1198 if (!f2fs_down_write_trylock(&sbi->quota_sem))
1199 return true;
1200 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1201 ret = false;
1202 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1203 ret = false;
1204 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1205 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1206 ret = true;
1207 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1208 ret = true;
1209 }
1210 f2fs_up_write(&sbi->quota_sem);
1211 return ret;
1212}
1213
1214/*
1215 * Freeze all the FS-operations for checkpoint.
1216 */
1217static int block_operations(struct f2fs_sb_info *sbi)
1218{
1219 struct writeback_control wbc = {
1220 .sync_mode = WB_SYNC_ALL,
1221 .nr_to_write = LONG_MAX,
1222 .for_reclaim = 0,
1223 };
1224 int err = 0, cnt = 0;
1225
1226 /*
1227 * Let's flush inline_data in dirty node pages.
1228 */
1229 f2fs_flush_inline_data(sbi);
1230
1231retry_flush_quotas:
1232 f2fs_lock_all(sbi);
1233 if (__need_flush_quota(sbi)) {
1234 int locked;
1235
1236 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1237 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1238 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1239 goto retry_flush_dents;
1240 }
1241 f2fs_unlock_all(sbi);
1242
1243 /* only failed during mount/umount/freeze/quotactl */
1244 locked = down_read_trylock(&sbi->sb->s_umount);
1245 f2fs_quota_sync(sbi->sb, -1);
1246 if (locked)
1247 up_read(&sbi->sb->s_umount);
1248 cond_resched();
1249 goto retry_flush_quotas;
1250 }
1251
1252retry_flush_dents:
1253 /* write all the dirty dentry pages */
1254 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1255 f2fs_unlock_all(sbi);
1256 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1257 if (err)
1258 return err;
1259 cond_resched();
1260 goto retry_flush_quotas;
1261 }
1262
1263 /*
1264 * POR: we should ensure that there are no dirty node pages
1265 * until finishing nat/sit flush. inode->i_blocks can be updated.
1266 */
1267 f2fs_down_write(&sbi->node_change);
1268
1269 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1270 f2fs_up_write(&sbi->node_change);
1271 f2fs_unlock_all(sbi);
1272 err = f2fs_sync_inode_meta(sbi);
1273 if (err)
1274 return err;
1275 cond_resched();
1276 goto retry_flush_quotas;
1277 }
1278
1279retry_flush_nodes:
1280 f2fs_down_write(&sbi->node_write);
1281
1282 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1283 f2fs_up_write(&sbi->node_write);
1284 atomic_inc(&sbi->wb_sync_req[NODE]);
1285 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1286 atomic_dec(&sbi->wb_sync_req[NODE]);
1287 if (err) {
1288 f2fs_up_write(&sbi->node_change);
1289 f2fs_unlock_all(sbi);
1290 return err;
1291 }
1292 cond_resched();
1293 goto retry_flush_nodes;
1294 }
1295
1296 /*
1297 * sbi->node_change is used only for AIO write_begin path which produces
1298 * dirty node blocks and some checkpoint values by block allocation.
1299 */
1300 __prepare_cp_block(sbi);
1301 f2fs_up_write(&sbi->node_change);
1302 return err;
1303}
1304
1305static void unblock_operations(struct f2fs_sb_info *sbi)
1306{
1307 f2fs_up_write(&sbi->node_write);
1308 f2fs_unlock_all(sbi);
1309}
1310
1311void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1312{
1313 DEFINE_WAIT(wait);
1314
1315 for (;;) {
1316 if (!get_pages(sbi, type))
1317 break;
1318
1319 if (unlikely(f2fs_cp_error(sbi) &&
1320 !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1321 break;
1322
1323 if (type == F2FS_DIRTY_META)
1324 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1325 FS_CP_META_IO);
1326 else if (type == F2FS_WB_CP_DATA)
1327 f2fs_submit_merged_write(sbi, DATA);
1328
1329 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1330 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1331 }
1332 finish_wait(&sbi->cp_wait, &wait);
1333}
1334
1335static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1336{
1337 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1338 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1339 unsigned long flags;
1340
1341 if (cpc->reason & CP_UMOUNT) {
1342 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1343 NM_I(sbi)->nat_bits_blocks > BLKS_PER_SEG(sbi)) {
1344 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1345 f2fs_notice(sbi, "Disable nat_bits due to no space");
1346 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1347 f2fs_nat_bitmap_enabled(sbi)) {
1348 f2fs_enable_nat_bits(sbi);
1349 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1350 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1351 }
1352 }
1353
1354 spin_lock_irqsave(&sbi->cp_lock, flags);
1355
1356 if (cpc->reason & CP_TRIMMED)
1357 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1358 else
1359 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1360
1361 if (cpc->reason & CP_UMOUNT)
1362 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1363 else
1364 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1365
1366 if (cpc->reason & CP_FASTBOOT)
1367 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1368 else
1369 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1370
1371 if (orphan_num)
1372 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1373 else
1374 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1375
1376 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1377 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1378
1379 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1380 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1381 else
1382 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1383
1384 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1385 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1386 else
1387 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1388
1389 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1390 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1391 else
1392 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1393
1394 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1395 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1396 else
1397 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1398
1399 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1400 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1401
1402 /* set this flag to activate crc|cp_ver for recovery */
1403 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1404 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1405
1406 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1407}
1408
1409static void commit_checkpoint(struct f2fs_sb_info *sbi,
1410 void *src, block_t blk_addr)
1411{
1412 struct writeback_control wbc = {
1413 .for_reclaim = 0,
1414 };
1415
1416 /*
1417 * filemap_get_folios_tag and lock_page again will take
1418 * some extra time. Therefore, f2fs_update_meta_pages and
1419 * f2fs_sync_meta_pages are combined in this function.
1420 */
1421 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1422 int err;
1423
1424 f2fs_wait_on_page_writeback(page, META, true, true);
1425
1426 memcpy(page_address(page), src, PAGE_SIZE);
1427
1428 set_page_dirty(page);
1429 if (unlikely(!clear_page_dirty_for_io(page)))
1430 f2fs_bug_on(sbi, 1);
1431
1432 /* writeout cp pack 2 page */
1433 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1434 if (unlikely(err && f2fs_cp_error(sbi))) {
1435 f2fs_put_page(page, 1);
1436 return;
1437 }
1438
1439 f2fs_bug_on(sbi, err);
1440 f2fs_put_page(page, 0);
1441
1442 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1443 f2fs_submit_merged_write(sbi, META_FLUSH);
1444}
1445
1446static inline u64 get_sectors_written(struct block_device *bdev)
1447{
1448 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1449}
1450
1451u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1452{
1453 if (f2fs_is_multi_device(sbi)) {
1454 u64 sectors = 0;
1455 int i;
1456
1457 for (i = 0; i < sbi->s_ndevs; i++)
1458 sectors += get_sectors_written(FDEV(i).bdev);
1459
1460 return sectors;
1461 }
1462
1463 return get_sectors_written(sbi->sb->s_bdev);
1464}
1465
1466static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1467{
1468 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1469 struct f2fs_nm_info *nm_i = NM_I(sbi);
1470 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1471 block_t start_blk;
1472 unsigned int data_sum_blocks, orphan_blocks;
1473 __u32 crc32 = 0;
1474 int i;
1475 int cp_payload_blks = __cp_payload(sbi);
1476 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1477 u64 kbytes_written;
1478 int err;
1479
1480 /* Flush all the NAT/SIT pages */
1481 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1482
1483 /* start to update checkpoint, cp ver is already updated previously */
1484 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1485 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1486 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1487 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1488
1489 ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1490 ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1491 ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1492 }
1493 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1494 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1495
1496 ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1497 ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1498 ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1499 }
1500
1501 /* 2 cp + n data seg summary + orphan inode blocks */
1502 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1503 spin_lock_irqsave(&sbi->cp_lock, flags);
1504 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1505 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1506 else
1507 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1508 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1509
1510 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1511 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1512 orphan_blocks);
1513
1514 if (__remain_node_summaries(cpc->reason))
1515 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1516 cp_payload_blks + data_sum_blocks +
1517 orphan_blocks + NR_CURSEG_NODE_TYPE);
1518 else
1519 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1520 cp_payload_blks + data_sum_blocks +
1521 orphan_blocks);
1522
1523 /* update ckpt flag for checkpoint */
1524 update_ckpt_flags(sbi, cpc);
1525
1526 /* update SIT/NAT bitmap */
1527 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1528 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1529
1530 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1531 *((__le32 *)((unsigned char *)ckpt +
1532 le32_to_cpu(ckpt->checksum_offset)))
1533 = cpu_to_le32(crc32);
1534
1535 start_blk = __start_cp_next_addr(sbi);
1536
1537 /* write nat bits */
1538 if ((cpc->reason & CP_UMOUNT) &&
1539 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1540 __u64 cp_ver = cur_cp_version(ckpt);
1541 block_t blk;
1542
1543 cp_ver |= ((__u64)crc32 << 32);
1544 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1545
1546 blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1547 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1548 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1549 (i << F2FS_BLKSIZE_BITS), blk + i);
1550 }
1551
1552 /* write out checkpoint buffer at block 0 */
1553 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1554
1555 for (i = 1; i < 1 + cp_payload_blks; i++)
1556 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1557 start_blk++);
1558
1559 if (orphan_num) {
1560 write_orphan_inodes(sbi, start_blk);
1561 start_blk += orphan_blocks;
1562 }
1563
1564 f2fs_write_data_summaries(sbi, start_blk);
1565 start_blk += data_sum_blocks;
1566
1567 /* Record write statistics in the hot node summary */
1568 kbytes_written = sbi->kbytes_written;
1569 kbytes_written += (f2fs_get_sectors_written(sbi) -
1570 sbi->sectors_written_start) >> 1;
1571 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1572
1573 if (__remain_node_summaries(cpc->reason)) {
1574 f2fs_write_node_summaries(sbi, start_blk);
1575 start_blk += NR_CURSEG_NODE_TYPE;
1576 }
1577
1578 /* update user_block_counts */
1579 sbi->last_valid_block_count = sbi->total_valid_block_count;
1580 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1581 percpu_counter_set(&sbi->rf_node_block_count, 0);
1582
1583 /* Here, we have one bio having CP pack except cp pack 2 page */
1584 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1585 /* Wait for all dirty meta pages to be submitted for IO */
1586 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1587
1588 /* wait for previous submitted meta pages writeback */
1589 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1590
1591 /* flush all device cache */
1592 err = f2fs_flush_device_cache(sbi);
1593 if (err)
1594 return err;
1595
1596 /* barrier and flush checkpoint cp pack 2 page if it can */
1597 commit_checkpoint(sbi, ckpt, start_blk);
1598 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1599
1600 /*
1601 * invalidate intermediate page cache borrowed from meta inode which are
1602 * used for migration of encrypted, verity or compressed inode's blocks.
1603 */
1604 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1605 f2fs_sb_has_compression(sbi))
1606 f2fs_bug_on(sbi,
1607 invalidate_inode_pages2_range(META_MAPPING(sbi),
1608 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1609
1610 f2fs_release_ino_entry(sbi, false);
1611
1612 f2fs_reset_fsync_node_info(sbi);
1613
1614 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1615 clear_sbi_flag(sbi, SBI_NEED_CP);
1616 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1617
1618 spin_lock(&sbi->stat_lock);
1619 sbi->unusable_block_count = 0;
1620 spin_unlock(&sbi->stat_lock);
1621
1622 __set_cp_next_pack(sbi);
1623
1624 /*
1625 * redirty superblock if metadata like node page or inode cache is
1626 * updated during writing checkpoint.
1627 */
1628 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1629 get_pages(sbi, F2FS_DIRTY_IMETA))
1630 set_sbi_flag(sbi, SBI_IS_DIRTY);
1631
1632 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1633
1634 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1635}
1636
1637int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1638{
1639 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1640 unsigned long long ckpt_ver;
1641 int err = 0;
1642
1643 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1644 return -EROFS;
1645
1646 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1647 if (cpc->reason != CP_PAUSE)
1648 return 0;
1649 f2fs_warn(sbi, "Start checkpoint disabled!");
1650 }
1651 if (cpc->reason != CP_RESIZE)
1652 f2fs_down_write(&sbi->cp_global_sem);
1653
1654 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1655 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1656 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1657 goto out;
1658 if (unlikely(f2fs_cp_error(sbi))) {
1659 err = -EIO;
1660 goto out;
1661 }
1662
1663 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1664
1665 err = block_operations(sbi);
1666 if (err)
1667 goto out;
1668
1669 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1670
1671 f2fs_flush_merged_writes(sbi);
1672
1673 /* this is the case of multiple fstrims without any changes */
1674 if (cpc->reason & CP_DISCARD) {
1675 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1676 unblock_operations(sbi);
1677 goto out;
1678 }
1679
1680 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1681 SIT_I(sbi)->dirty_sentries == 0 &&
1682 prefree_segments(sbi) == 0) {
1683 f2fs_flush_sit_entries(sbi, cpc);
1684 f2fs_clear_prefree_segments(sbi, cpc);
1685 unblock_operations(sbi);
1686 goto out;
1687 }
1688 }
1689
1690 /*
1691 * update checkpoint pack index
1692 * Increase the version number so that
1693 * SIT entries and seg summaries are written at correct place
1694 */
1695 ckpt_ver = cur_cp_version(ckpt);
1696 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1697
1698 /* write cached NAT/SIT entries to NAT/SIT area */
1699 err = f2fs_flush_nat_entries(sbi, cpc);
1700 if (err) {
1701 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1702 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1703 goto stop;
1704 }
1705
1706 f2fs_flush_sit_entries(sbi, cpc);
1707
1708 /* save inmem log status */
1709 f2fs_save_inmem_curseg(sbi);
1710
1711 err = do_checkpoint(sbi, cpc);
1712 if (err) {
1713 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1714 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1715 f2fs_release_discard_addrs(sbi);
1716 } else {
1717 f2fs_clear_prefree_segments(sbi, cpc);
1718 }
1719
1720 f2fs_restore_inmem_curseg(sbi);
1721 stat_inc_cp_count(sbi);
1722stop:
1723 unblock_operations(sbi);
1724
1725 if (cpc->reason & CP_RECOVERY)
1726 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1727
1728 /* update CP_TIME to trigger checkpoint periodically */
1729 f2fs_update_time(sbi, CP_TIME);
1730 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1731out:
1732 if (cpc->reason != CP_RESIZE)
1733 f2fs_up_write(&sbi->cp_global_sem);
1734 return err;
1735}
1736
1737void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1738{
1739 int i;
1740
1741 for (i = 0; i < MAX_INO_ENTRY; i++) {
1742 struct inode_management *im = &sbi->im[i];
1743
1744 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1745 spin_lock_init(&im->ino_lock);
1746 INIT_LIST_HEAD(&im->ino_list);
1747 im->ino_num = 0;
1748 }
1749
1750 sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1751 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1752 F2FS_ORPHANS_PER_BLOCK;
1753}
1754
1755int __init f2fs_create_checkpoint_caches(void)
1756{
1757 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1758 sizeof(struct ino_entry));
1759 if (!ino_entry_slab)
1760 return -ENOMEM;
1761 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1762 sizeof(struct inode_entry));
1763 if (!f2fs_inode_entry_slab) {
1764 kmem_cache_destroy(ino_entry_slab);
1765 return -ENOMEM;
1766 }
1767 return 0;
1768}
1769
1770void f2fs_destroy_checkpoint_caches(void)
1771{
1772 kmem_cache_destroy(ino_entry_slab);
1773 kmem_cache_destroy(f2fs_inode_entry_slab);
1774}
1775
1776static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1777{
1778 struct cp_control cpc = { .reason = CP_SYNC, };
1779 int err;
1780
1781 f2fs_down_write(&sbi->gc_lock);
1782 err = f2fs_write_checkpoint(sbi, &cpc);
1783 f2fs_up_write(&sbi->gc_lock);
1784
1785 return err;
1786}
1787
1788static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1789{
1790 struct ckpt_req_control *cprc = &sbi->cprc_info;
1791 struct ckpt_req *req, *next;
1792 struct llist_node *dispatch_list;
1793 u64 sum_diff = 0, diff, count = 0;
1794 int ret;
1795
1796 dispatch_list = llist_del_all(&cprc->issue_list);
1797 if (!dispatch_list)
1798 return;
1799 dispatch_list = llist_reverse_order(dispatch_list);
1800
1801 ret = __write_checkpoint_sync(sbi);
1802 atomic_inc(&cprc->issued_ckpt);
1803
1804 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1805 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1806 req->ret = ret;
1807 complete(&req->wait);
1808
1809 sum_diff += diff;
1810 count++;
1811 }
1812 atomic_sub(count, &cprc->queued_ckpt);
1813 atomic_add(count, &cprc->total_ckpt);
1814
1815 spin_lock(&cprc->stat_lock);
1816 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1817 if (cprc->peak_time < cprc->cur_time)
1818 cprc->peak_time = cprc->cur_time;
1819 spin_unlock(&cprc->stat_lock);
1820}
1821
1822static int issue_checkpoint_thread(void *data)
1823{
1824 struct f2fs_sb_info *sbi = data;
1825 struct ckpt_req_control *cprc = &sbi->cprc_info;
1826 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1827repeat:
1828 if (kthread_should_stop())
1829 return 0;
1830
1831 if (!llist_empty(&cprc->issue_list))
1832 __checkpoint_and_complete_reqs(sbi);
1833
1834 wait_event_interruptible(*q,
1835 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1836 goto repeat;
1837}
1838
1839static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1840 struct ckpt_req *wait_req)
1841{
1842 struct ckpt_req_control *cprc = &sbi->cprc_info;
1843
1844 if (!llist_empty(&cprc->issue_list)) {
1845 __checkpoint_and_complete_reqs(sbi);
1846 } else {
1847 /* already dispatched by issue_checkpoint_thread */
1848 if (wait_req)
1849 wait_for_completion(&wait_req->wait);
1850 }
1851}
1852
1853static void init_ckpt_req(struct ckpt_req *req)
1854{
1855 memset(req, 0, sizeof(struct ckpt_req));
1856
1857 init_completion(&req->wait);
1858 req->queue_time = ktime_get();
1859}
1860
1861int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1862{
1863 struct ckpt_req_control *cprc = &sbi->cprc_info;
1864 struct ckpt_req req;
1865 struct cp_control cpc;
1866
1867 cpc.reason = __get_cp_reason(sbi);
1868 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1869 int ret;
1870
1871 f2fs_down_write(&sbi->gc_lock);
1872 ret = f2fs_write_checkpoint(sbi, &cpc);
1873 f2fs_up_write(&sbi->gc_lock);
1874
1875 return ret;
1876 }
1877
1878 if (!cprc->f2fs_issue_ckpt)
1879 return __write_checkpoint_sync(sbi);
1880
1881 init_ckpt_req(&req);
1882
1883 llist_add(&req.llnode, &cprc->issue_list);
1884 atomic_inc(&cprc->queued_ckpt);
1885
1886 /*
1887 * update issue_list before we wake up issue_checkpoint thread,
1888 * this smp_mb() pairs with another barrier in ___wait_event(),
1889 * see more details in comments of waitqueue_active().
1890 */
1891 smp_mb();
1892
1893 if (waitqueue_active(&cprc->ckpt_wait_queue))
1894 wake_up(&cprc->ckpt_wait_queue);
1895
1896 if (cprc->f2fs_issue_ckpt)
1897 wait_for_completion(&req.wait);
1898 else
1899 flush_remained_ckpt_reqs(sbi, &req);
1900
1901 return req.ret;
1902}
1903
1904int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1905{
1906 dev_t dev = sbi->sb->s_bdev->bd_dev;
1907 struct ckpt_req_control *cprc = &sbi->cprc_info;
1908
1909 if (cprc->f2fs_issue_ckpt)
1910 return 0;
1911
1912 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1913 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1914 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1915 int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1916
1917 cprc->f2fs_issue_ckpt = NULL;
1918 return err;
1919 }
1920
1921 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1922
1923 return 0;
1924}
1925
1926void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1927{
1928 struct ckpt_req_control *cprc = &sbi->cprc_info;
1929 struct task_struct *ckpt_task;
1930
1931 if (!cprc->f2fs_issue_ckpt)
1932 return;
1933
1934 ckpt_task = cprc->f2fs_issue_ckpt;
1935 cprc->f2fs_issue_ckpt = NULL;
1936 kthread_stop(ckpt_task);
1937
1938 f2fs_flush_ckpt_thread(sbi);
1939}
1940
1941void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1942{
1943 struct ckpt_req_control *cprc = &sbi->cprc_info;
1944
1945 flush_remained_ckpt_reqs(sbi, NULL);
1946
1947 /* Let's wait for the previous dispatched checkpoint. */
1948 while (atomic_read(&cprc->queued_ckpt))
1949 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1950}
1951
1952void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1953{
1954 struct ckpt_req_control *cprc = &sbi->cprc_info;
1955
1956 atomic_set(&cprc->issued_ckpt, 0);
1957 atomic_set(&cprc->total_ckpt, 0);
1958 atomic_set(&cprc->queued_ckpt, 0);
1959 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1960 init_waitqueue_head(&cprc->ckpt_wait_queue);
1961 init_llist_head(&cprc->issue_list);
1962 spin_lock_init(&cprc->stat_lock);
1963}