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
29void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
30{
31 set_ckpt_flags(sbi, CP_ERROR_FLAG);
32 if (!end_io)
33 f2fs_flush_merged_writes(sbi);
34}
35
36/*
37 * We guarantee no failure on the returned page.
38 */
39struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
40{
41 struct address_space *mapping = META_MAPPING(sbi);
42 struct page *page = NULL;
43repeat:
44 page = f2fs_grab_cache_page(mapping, index, false);
45 if (!page) {
46 cond_resched();
47 goto repeat;
48 }
49 f2fs_wait_on_page_writeback(page, META, true);
50 if (!PageUptodate(page))
51 SetPageUptodate(page);
52 return page;
53}
54
55/*
56 * We guarantee no failure on the returned page.
57 */
58static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
59 bool is_meta)
60{
61 struct address_space *mapping = META_MAPPING(sbi);
62 struct page *page;
63 struct f2fs_io_info fio = {
64 .sbi = sbi,
65 .type = META,
66 .op = REQ_OP_READ,
67 .op_flags = REQ_META | REQ_PRIO,
68 .old_blkaddr = index,
69 .new_blkaddr = index,
70 .encrypted_page = NULL,
71 .is_meta = is_meta,
72 };
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 if (f2fs_submit_page_bio(&fio)) {
88 f2fs_put_page(page, 1);
89 goto repeat;
90 }
91
92 lock_page(page);
93 if (unlikely(page->mapping != mapping)) {
94 f2fs_put_page(page, 1);
95 goto repeat;
96 }
97
98 /*
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
101 * meta page.
102 */
103 if (unlikely(!PageUptodate(page)))
104 f2fs_stop_checkpoint(sbi, false);
105out:
106 return page;
107}
108
109struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
110{
111 return __get_meta_page(sbi, index, true);
112}
113
114/* for POR only */
115struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
116{
117 return __get_meta_page(sbi, index, false);
118}
119
120bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
121{
122 switch (type) {
123 case META_NAT:
124 break;
125 case META_SIT:
126 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
127 return false;
128 break;
129 case META_SSA:
130 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
131 blkaddr < SM_I(sbi)->ssa_blkaddr))
132 return false;
133 break;
134 case META_CP:
135 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
136 blkaddr < __start_cp_addr(sbi)))
137 return false;
138 break;
139 case META_POR:
140 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
141 blkaddr < MAIN_BLKADDR(sbi)))
142 return false;
143 break;
144 default:
145 BUG();
146 }
147
148 return true;
149}
150
151/*
152 * Readahead CP/NAT/SIT/SSA pages
153 */
154int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
155 int type, bool sync)
156{
157 struct page *page;
158 block_t blkno = start;
159 struct f2fs_io_info fio = {
160 .sbi = sbi,
161 .type = META,
162 .op = REQ_OP_READ,
163 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
164 .encrypted_page = NULL,
165 .in_list = false,
166 .is_meta = (type != META_POR),
167 };
168 struct blk_plug plug;
169
170 if (unlikely(type == META_POR))
171 fio.op_flags &= ~REQ_META;
172
173 blk_start_plug(&plug);
174 for (; nrpages-- > 0; blkno++) {
175
176 if (!is_valid_blkaddr(sbi, blkno, type))
177 goto out;
178
179 switch (type) {
180 case META_NAT:
181 if (unlikely(blkno >=
182 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
183 blkno = 0;
184 /* get nat block addr */
185 fio.new_blkaddr = current_nat_addr(sbi,
186 blkno * NAT_ENTRY_PER_BLOCK);
187 break;
188 case META_SIT:
189 /* get sit block addr */
190 fio.new_blkaddr = current_sit_addr(sbi,
191 blkno * SIT_ENTRY_PER_BLOCK);
192 break;
193 case META_SSA:
194 case META_CP:
195 case META_POR:
196 fio.new_blkaddr = blkno;
197 break;
198 default:
199 BUG();
200 }
201
202 page = f2fs_grab_cache_page(META_MAPPING(sbi),
203 fio.new_blkaddr, false);
204 if (!page)
205 continue;
206 if (PageUptodate(page)) {
207 f2fs_put_page(page, 1);
208 continue;
209 }
210
211 fio.page = page;
212 f2fs_submit_page_bio(&fio);
213 f2fs_put_page(page, 0);
214 }
215out:
216 blk_finish_plug(&plug);
217 return blkno - start;
218}
219
220void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
221{
222 struct page *page;
223 bool readahead = false;
224
225 page = find_get_page(META_MAPPING(sbi), index);
226 if (!page || !PageUptodate(page))
227 readahead = true;
228 f2fs_put_page(page, 0);
229
230 if (readahead)
231 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
232}
233
234static int __f2fs_write_meta_page(struct page *page,
235 struct writeback_control *wbc,
236 enum iostat_type io_type)
237{
238 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
239
240 trace_f2fs_writepage(page, META);
241
242 if (unlikely(f2fs_cp_error(sbi))) {
243 dec_page_count(sbi, F2FS_DIRTY_META);
244 unlock_page(page);
245 return 0;
246 }
247 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
248 goto redirty_out;
249 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
250 goto redirty_out;
251
252 write_meta_page(sbi, page, io_type);
253 dec_page_count(sbi, F2FS_DIRTY_META);
254
255 if (wbc->for_reclaim)
256 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
257 0, page->index, META);
258
259 unlock_page(page);
260
261 if (unlikely(f2fs_cp_error(sbi)))
262 f2fs_submit_merged_write(sbi, META);
263
264 return 0;
265
266redirty_out:
267 redirty_page_for_writepage(wbc, page);
268 return AOP_WRITEPAGE_ACTIVATE;
269}
270
271static int f2fs_write_meta_page(struct page *page,
272 struct writeback_control *wbc)
273{
274 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
275}
276
277static int f2fs_write_meta_pages(struct address_space *mapping,
278 struct writeback_control *wbc)
279{
280 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
281 long diff, written;
282
283 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
284 goto skip_write;
285
286 /* collect a number of dirty meta pages and write together */
287 if (wbc->for_kupdate ||
288 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
289 goto skip_write;
290
291 /* if locked failed, cp will flush dirty pages instead */
292 if (!mutex_trylock(&sbi->cp_mutex))
293 goto skip_write;
294
295 trace_f2fs_writepages(mapping->host, wbc, META);
296 diff = nr_pages_to_write(sbi, META, wbc);
297 written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
298 mutex_unlock(&sbi->cp_mutex);
299 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
300 return 0;
301
302skip_write:
303 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
304 trace_f2fs_writepages(mapping->host, wbc, META);
305 return 0;
306}
307
308long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
309 long nr_to_write, enum iostat_type io_type)
310{
311 struct address_space *mapping = META_MAPPING(sbi);
312 pgoff_t index = 0, prev = ULONG_MAX;
313 struct pagevec pvec;
314 long nwritten = 0;
315 int nr_pages;
316 struct writeback_control wbc = {
317 .for_reclaim = 0,
318 };
319 struct blk_plug plug;
320
321 pagevec_init(&pvec);
322
323 blk_start_plug(&plug);
324
325 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
326 PAGECACHE_TAG_DIRTY))) {
327 int i;
328
329 for (i = 0; i < nr_pages; i++) {
330 struct page *page = pvec.pages[i];
331
332 if (prev == ULONG_MAX)
333 prev = page->index - 1;
334 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
335 pagevec_release(&pvec);
336 goto stop;
337 }
338
339 lock_page(page);
340
341 if (unlikely(page->mapping != mapping)) {
342continue_unlock:
343 unlock_page(page);
344 continue;
345 }
346 if (!PageDirty(page)) {
347 /* someone wrote it for us */
348 goto continue_unlock;
349 }
350
351 f2fs_wait_on_page_writeback(page, META, true);
352
353 BUG_ON(PageWriteback(page));
354 if (!clear_page_dirty_for_io(page))
355 goto continue_unlock;
356
357 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
358 unlock_page(page);
359 break;
360 }
361 nwritten++;
362 prev = page->index;
363 if (unlikely(nwritten >= nr_to_write))
364 break;
365 }
366 pagevec_release(&pvec);
367 cond_resched();
368 }
369stop:
370 if (nwritten)
371 f2fs_submit_merged_write(sbi, type);
372
373 blk_finish_plug(&plug);
374
375 return nwritten;
376}
377
378static int f2fs_set_meta_page_dirty(struct page *page)
379{
380 trace_f2fs_set_page_dirty(page, META);
381
382 if (!PageUptodate(page))
383 SetPageUptodate(page);
384 if (!PageDirty(page)) {
385 f2fs_set_page_dirty_nobuffers(page);
386 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
387 SetPagePrivate(page);
388 f2fs_trace_pid(page);
389 return 1;
390 }
391 return 0;
392}
393
394const struct address_space_operations f2fs_meta_aops = {
395 .writepage = f2fs_write_meta_page,
396 .writepages = f2fs_write_meta_pages,
397 .set_page_dirty = f2fs_set_meta_page_dirty,
398 .invalidatepage = f2fs_invalidate_page,
399 .releasepage = f2fs_release_page,
400#ifdef CONFIG_MIGRATION
401 .migratepage = f2fs_migrate_page,
402#endif
403};
404
405static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
406 unsigned int devidx, int type)
407{
408 struct inode_management *im = &sbi->im[type];
409 struct ino_entry *e, *tmp;
410
411 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
412
413 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
414
415 spin_lock(&im->ino_lock);
416 e = radix_tree_lookup(&im->ino_root, ino);
417 if (!e) {
418 e = tmp;
419 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
420 f2fs_bug_on(sbi, 1);
421
422 memset(e, 0, sizeof(struct ino_entry));
423 e->ino = ino;
424
425 list_add_tail(&e->list, &im->ino_list);
426 if (type != ORPHAN_INO)
427 im->ino_num++;
428 }
429
430 if (type == FLUSH_INO)
431 f2fs_set_bit(devidx, (char *)&e->dirty_device);
432
433 spin_unlock(&im->ino_lock);
434 radix_tree_preload_end();
435
436 if (e != tmp)
437 kmem_cache_free(ino_entry_slab, tmp);
438}
439
440static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
441{
442 struct inode_management *im = &sbi->im[type];
443 struct ino_entry *e;
444
445 spin_lock(&im->ino_lock);
446 e = radix_tree_lookup(&im->ino_root, ino);
447 if (e) {
448 list_del(&e->list);
449 radix_tree_delete(&im->ino_root, ino);
450 im->ino_num--;
451 spin_unlock(&im->ino_lock);
452 kmem_cache_free(ino_entry_slab, e);
453 return;
454 }
455 spin_unlock(&im->ino_lock);
456}
457
458void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
459{
460 /* add new dirty ino entry into list */
461 __add_ino_entry(sbi, ino, 0, type);
462}
463
464void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
465{
466 /* remove dirty ino entry from list */
467 __remove_ino_entry(sbi, ino, type);
468}
469
470/* mode should be APPEND_INO or UPDATE_INO */
471bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
472{
473 struct inode_management *im = &sbi->im[mode];
474 struct ino_entry *e;
475
476 spin_lock(&im->ino_lock);
477 e = radix_tree_lookup(&im->ino_root, ino);
478 spin_unlock(&im->ino_lock);
479 return e ? true : false;
480}
481
482void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
483{
484 struct ino_entry *e, *tmp;
485 int i;
486
487 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
488 struct inode_management *im = &sbi->im[i];
489
490 spin_lock(&im->ino_lock);
491 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
492 list_del(&e->list);
493 radix_tree_delete(&im->ino_root, e->ino);
494 kmem_cache_free(ino_entry_slab, e);
495 im->ino_num--;
496 }
497 spin_unlock(&im->ino_lock);
498 }
499}
500
501void set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
502 unsigned int devidx, int type)
503{
504 __add_ino_entry(sbi, ino, devidx, type);
505}
506
507bool is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
508 unsigned int devidx, int type)
509{
510 struct inode_management *im = &sbi->im[type];
511 struct ino_entry *e;
512 bool is_dirty = false;
513
514 spin_lock(&im->ino_lock);
515 e = radix_tree_lookup(&im->ino_root, ino);
516 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
517 is_dirty = true;
518 spin_unlock(&im->ino_lock);
519 return is_dirty;
520}
521
522int acquire_orphan_inode(struct f2fs_sb_info *sbi)
523{
524 struct inode_management *im = &sbi->im[ORPHAN_INO];
525 int err = 0;
526
527 spin_lock(&im->ino_lock);
528
529#ifdef CONFIG_F2FS_FAULT_INJECTION
530 if (time_to_inject(sbi, FAULT_ORPHAN)) {
531 spin_unlock(&im->ino_lock);
532 f2fs_show_injection_info(FAULT_ORPHAN);
533 return -ENOSPC;
534 }
535#endif
536 if (unlikely(im->ino_num >= sbi->max_orphans))
537 err = -ENOSPC;
538 else
539 im->ino_num++;
540 spin_unlock(&im->ino_lock);
541
542 return err;
543}
544
545void release_orphan_inode(struct f2fs_sb_info *sbi)
546{
547 struct inode_management *im = &sbi->im[ORPHAN_INO];
548
549 spin_lock(&im->ino_lock);
550 f2fs_bug_on(sbi, im->ino_num == 0);
551 im->ino_num--;
552 spin_unlock(&im->ino_lock);
553}
554
555void add_orphan_inode(struct inode *inode)
556{
557 /* add new orphan ino entry into list */
558 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
559 update_inode_page(inode);
560}
561
562void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
563{
564 /* remove orphan entry from orphan list */
565 __remove_ino_entry(sbi, ino, ORPHAN_INO);
566}
567
568static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
569{
570 struct inode *inode;
571 struct node_info ni;
572 int err = acquire_orphan_inode(sbi);
573
574 if (err)
575 goto err_out;
576
577 __add_ino_entry(sbi, ino, 0, ORPHAN_INO);
578
579 inode = f2fs_iget_retry(sbi->sb, ino);
580 if (IS_ERR(inode)) {
581 /*
582 * there should be a bug that we can't find the entry
583 * to orphan inode.
584 */
585 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
586 return PTR_ERR(inode);
587 }
588
589 err = dquot_initialize(inode);
590 if (err)
591 goto err_out;
592
593 dquot_initialize(inode);
594 clear_nlink(inode);
595
596 /* truncate all the data during iput */
597 iput(inode);
598
599 get_node_info(sbi, ino, &ni);
600
601 /* ENOMEM was fully retried in f2fs_evict_inode. */
602 if (ni.blk_addr != NULL_ADDR) {
603 err = -EIO;
604 goto err_out;
605 }
606 __remove_ino_entry(sbi, ino, ORPHAN_INO);
607 return 0;
608
609err_out:
610 set_sbi_flag(sbi, SBI_NEED_FSCK);
611 f2fs_msg(sbi->sb, KERN_WARNING,
612 "%s: orphan failed (ino=%x), run fsck to fix.",
613 __func__, ino);
614 return err;
615}
616
617int recover_orphan_inodes(struct f2fs_sb_info *sbi)
618{
619 block_t start_blk, orphan_blocks, i, j;
620 unsigned int s_flags = sbi->sb->s_flags;
621 int err = 0;
622#ifdef CONFIG_QUOTA
623 int quota_enabled;
624#endif
625
626 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
627 return 0;
628
629 if (s_flags & SB_RDONLY) {
630 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
631 sbi->sb->s_flags &= ~SB_RDONLY;
632 }
633
634#ifdef CONFIG_QUOTA
635 /* Needed for iput() to work correctly and not trash data */
636 sbi->sb->s_flags |= SB_ACTIVE;
637
638 /* Turn on quotas so that they are updated correctly */
639 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
640#endif
641
642 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
643 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
644
645 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
646
647 for (i = 0; i < orphan_blocks; i++) {
648 struct page *page = get_meta_page(sbi, start_blk + i);
649 struct f2fs_orphan_block *orphan_blk;
650
651 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
652 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
653 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
654 err = recover_orphan_inode(sbi, ino);
655 if (err) {
656 f2fs_put_page(page, 1);
657 goto out;
658 }
659 }
660 f2fs_put_page(page, 1);
661 }
662 /* clear Orphan Flag */
663 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
664out:
665#ifdef CONFIG_QUOTA
666 /* Turn quotas off */
667 if (quota_enabled)
668 f2fs_quota_off_umount(sbi->sb);
669#endif
670 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
671
672 return err;
673}
674
675static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
676{
677 struct list_head *head;
678 struct f2fs_orphan_block *orphan_blk = NULL;
679 unsigned int nentries = 0;
680 unsigned short index = 1;
681 unsigned short orphan_blocks;
682 struct page *page = NULL;
683 struct ino_entry *orphan = NULL;
684 struct inode_management *im = &sbi->im[ORPHAN_INO];
685
686 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
687
688 /*
689 * we don't need to do spin_lock(&im->ino_lock) here, since all the
690 * orphan inode operations are covered under f2fs_lock_op().
691 * And, spin_lock should be avoided due to page operations below.
692 */
693 head = &im->ino_list;
694
695 /* loop for each orphan inode entry and write them in Jornal block */
696 list_for_each_entry(orphan, head, list) {
697 if (!page) {
698 page = grab_meta_page(sbi, start_blk++);
699 orphan_blk =
700 (struct f2fs_orphan_block *)page_address(page);
701 memset(orphan_blk, 0, sizeof(*orphan_blk));
702 }
703
704 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
705
706 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
707 /*
708 * an orphan block is full of 1020 entries,
709 * then we need to flush current orphan blocks
710 * and bring another one in memory
711 */
712 orphan_blk->blk_addr = cpu_to_le16(index);
713 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
714 orphan_blk->entry_count = cpu_to_le32(nentries);
715 set_page_dirty(page);
716 f2fs_put_page(page, 1);
717 index++;
718 nentries = 0;
719 page = NULL;
720 }
721 }
722
723 if (page) {
724 orphan_blk->blk_addr = cpu_to_le16(index);
725 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
726 orphan_blk->entry_count = cpu_to_le32(nentries);
727 set_page_dirty(page);
728 f2fs_put_page(page, 1);
729 }
730}
731
732static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
733 struct f2fs_checkpoint **cp_block, struct page **cp_page,
734 unsigned long long *version)
735{
736 unsigned long blk_size = sbi->blocksize;
737 size_t crc_offset = 0;
738 __u32 crc = 0;
739
740 *cp_page = get_meta_page(sbi, cp_addr);
741 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
742
743 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
744 if (crc_offset > (blk_size - sizeof(__le32))) {
745 f2fs_msg(sbi->sb, KERN_WARNING,
746 "invalid crc_offset: %zu", crc_offset);
747 return -EINVAL;
748 }
749
750 crc = cur_cp_crc(*cp_block);
751 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
752 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
753 return -EINVAL;
754 }
755
756 *version = cur_cp_version(*cp_block);
757 return 0;
758}
759
760static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
761 block_t cp_addr, unsigned long long *version)
762{
763 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
764 struct f2fs_checkpoint *cp_block = NULL;
765 unsigned long long cur_version = 0, pre_version = 0;
766 int err;
767
768 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
769 &cp_page_1, version);
770 if (err)
771 goto invalid_cp1;
772 pre_version = *version;
773
774 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
775 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
776 &cp_page_2, version);
777 if (err)
778 goto invalid_cp2;
779 cur_version = *version;
780
781 if (cur_version == pre_version) {
782 *version = cur_version;
783 f2fs_put_page(cp_page_2, 1);
784 return cp_page_1;
785 }
786invalid_cp2:
787 f2fs_put_page(cp_page_2, 1);
788invalid_cp1:
789 f2fs_put_page(cp_page_1, 1);
790 return NULL;
791}
792
793int get_valid_checkpoint(struct f2fs_sb_info *sbi)
794{
795 struct f2fs_checkpoint *cp_block;
796 struct f2fs_super_block *fsb = sbi->raw_super;
797 struct page *cp1, *cp2, *cur_page;
798 unsigned long blk_size = sbi->blocksize;
799 unsigned long long cp1_version = 0, cp2_version = 0;
800 unsigned long long cp_start_blk_no;
801 unsigned int cp_blks = 1 + __cp_payload(sbi);
802 block_t cp_blk_no;
803 int i;
804
805 sbi->ckpt = f2fs_kzalloc(sbi, cp_blks * blk_size, GFP_KERNEL);
806 if (!sbi->ckpt)
807 return -ENOMEM;
808 /*
809 * Finding out valid cp block involves read both
810 * sets( cp pack1 and cp pack 2)
811 */
812 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
813 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
814
815 /* The second checkpoint pack should start at the next segment */
816 cp_start_blk_no += ((unsigned long long)1) <<
817 le32_to_cpu(fsb->log_blocks_per_seg);
818 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
819
820 if (cp1 && cp2) {
821 if (ver_after(cp2_version, cp1_version))
822 cur_page = cp2;
823 else
824 cur_page = cp1;
825 } else if (cp1) {
826 cur_page = cp1;
827 } else if (cp2) {
828 cur_page = cp2;
829 } else {
830 goto fail_no_cp;
831 }
832
833 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
834 memcpy(sbi->ckpt, cp_block, blk_size);
835
836 /* Sanity checking of checkpoint */
837 if (sanity_check_ckpt(sbi))
838 goto free_fail_no_cp;
839
840 if (cur_page == cp1)
841 sbi->cur_cp_pack = 1;
842 else
843 sbi->cur_cp_pack = 2;
844
845 if (cp_blks <= 1)
846 goto done;
847
848 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
849 if (cur_page == cp2)
850 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
851
852 for (i = 1; i < cp_blks; i++) {
853 void *sit_bitmap_ptr;
854 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
855
856 cur_page = get_meta_page(sbi, cp_blk_no + i);
857 sit_bitmap_ptr = page_address(cur_page);
858 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
859 f2fs_put_page(cur_page, 1);
860 }
861done:
862 f2fs_put_page(cp1, 1);
863 f2fs_put_page(cp2, 1);
864 return 0;
865
866free_fail_no_cp:
867 f2fs_put_page(cp1, 1);
868 f2fs_put_page(cp2, 1);
869fail_no_cp:
870 kfree(sbi->ckpt);
871 return -EINVAL;
872}
873
874static void __add_dirty_inode(struct inode *inode, enum inode_type type)
875{
876 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
877 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
878
879 if (is_inode_flag_set(inode, flag))
880 return;
881
882 set_inode_flag(inode, flag);
883 if (!f2fs_is_volatile_file(inode))
884 list_add_tail(&F2FS_I(inode)->dirty_list,
885 &sbi->inode_list[type]);
886 stat_inc_dirty_inode(sbi, type);
887}
888
889static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
890{
891 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
892
893 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
894 return;
895
896 list_del_init(&F2FS_I(inode)->dirty_list);
897 clear_inode_flag(inode, flag);
898 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
899}
900
901void update_dirty_page(struct inode *inode, struct page *page)
902{
903 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
904 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
905
906 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
907 !S_ISLNK(inode->i_mode))
908 return;
909
910 spin_lock(&sbi->inode_lock[type]);
911 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
912 __add_dirty_inode(inode, type);
913 inode_inc_dirty_pages(inode);
914 spin_unlock(&sbi->inode_lock[type]);
915
916 SetPagePrivate(page);
917 f2fs_trace_pid(page);
918}
919
920void remove_dirty_inode(struct inode *inode)
921{
922 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
923 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
924
925 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
926 !S_ISLNK(inode->i_mode))
927 return;
928
929 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
930 return;
931
932 spin_lock(&sbi->inode_lock[type]);
933 __remove_dirty_inode(inode, type);
934 spin_unlock(&sbi->inode_lock[type]);
935}
936
937int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
938{
939 struct list_head *head;
940 struct inode *inode;
941 struct f2fs_inode_info *fi;
942 bool is_dir = (type == DIR_INODE);
943 unsigned long ino = 0;
944
945 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
946 get_pages(sbi, is_dir ?
947 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
948retry:
949 if (unlikely(f2fs_cp_error(sbi)))
950 return -EIO;
951
952 spin_lock(&sbi->inode_lock[type]);
953
954 head = &sbi->inode_list[type];
955 if (list_empty(head)) {
956 spin_unlock(&sbi->inode_lock[type]);
957 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
958 get_pages(sbi, is_dir ?
959 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
960 return 0;
961 }
962 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
963 inode = igrab(&fi->vfs_inode);
964 spin_unlock(&sbi->inode_lock[type]);
965 if (inode) {
966 unsigned long cur_ino = inode->i_ino;
967
968 if (is_dir)
969 F2FS_I(inode)->cp_task = current;
970
971 filemap_fdatawrite(inode->i_mapping);
972
973 if (is_dir)
974 F2FS_I(inode)->cp_task = NULL;
975
976 iput(inode);
977 /* We need to give cpu to another writers. */
978 if (ino == cur_ino) {
979 congestion_wait(BLK_RW_ASYNC, HZ/50);
980 cond_resched();
981 } else {
982 ino = cur_ino;
983 }
984 } else {
985 /*
986 * We should submit bio, since it exists several
987 * wribacking dentry pages in the freeing inode.
988 */
989 f2fs_submit_merged_write(sbi, DATA);
990 cond_resched();
991 }
992 goto retry;
993}
994
995int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
996{
997 struct list_head *head = &sbi->inode_list[DIRTY_META];
998 struct inode *inode;
999 struct f2fs_inode_info *fi;
1000 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1001
1002 while (total--) {
1003 if (unlikely(f2fs_cp_error(sbi)))
1004 return -EIO;
1005
1006 spin_lock(&sbi->inode_lock[DIRTY_META]);
1007 if (list_empty(head)) {
1008 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1009 return 0;
1010 }
1011 fi = list_first_entry(head, struct f2fs_inode_info,
1012 gdirty_list);
1013 inode = igrab(&fi->vfs_inode);
1014 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1015 if (inode) {
1016 sync_inode_metadata(inode, 0);
1017
1018 /* it's on eviction */
1019 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1020 update_inode_page(inode);
1021 iput(inode);
1022 }
1023 }
1024 return 0;
1025}
1026
1027static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1028{
1029 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1030 struct f2fs_nm_info *nm_i = NM_I(sbi);
1031 nid_t last_nid = nm_i->next_scan_nid;
1032
1033 next_free_nid(sbi, &last_nid);
1034 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1035 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1036 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1037 ckpt->next_free_nid = cpu_to_le32(last_nid);
1038}
1039
1040/*
1041 * Freeze all the FS-operations for checkpoint.
1042 */
1043static int block_operations(struct f2fs_sb_info *sbi)
1044{
1045 struct writeback_control wbc = {
1046 .sync_mode = WB_SYNC_ALL,
1047 .nr_to_write = LONG_MAX,
1048 .for_reclaim = 0,
1049 };
1050 struct blk_plug plug;
1051 int err = 0;
1052
1053 blk_start_plug(&plug);
1054
1055retry_flush_dents:
1056 f2fs_lock_all(sbi);
1057 /* write all the dirty dentry pages */
1058 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1059 f2fs_unlock_all(sbi);
1060 err = sync_dirty_inodes(sbi, DIR_INODE);
1061 if (err)
1062 goto out;
1063 cond_resched();
1064 goto retry_flush_dents;
1065 }
1066
1067 /*
1068 * POR: we should ensure that there are no dirty node pages
1069 * until finishing nat/sit flush. inode->i_blocks can be updated.
1070 */
1071 down_write(&sbi->node_change);
1072
1073 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1074 up_write(&sbi->node_change);
1075 f2fs_unlock_all(sbi);
1076 err = f2fs_sync_inode_meta(sbi);
1077 if (err)
1078 goto out;
1079 cond_resched();
1080 goto retry_flush_dents;
1081 }
1082
1083retry_flush_nodes:
1084 down_write(&sbi->node_write);
1085
1086 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1087 up_write(&sbi->node_write);
1088 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1089 if (err) {
1090 up_write(&sbi->node_change);
1091 f2fs_unlock_all(sbi);
1092 goto out;
1093 }
1094 cond_resched();
1095 goto retry_flush_nodes;
1096 }
1097
1098 /*
1099 * sbi->node_change is used only for AIO write_begin path which produces
1100 * dirty node blocks and some checkpoint values by block allocation.
1101 */
1102 __prepare_cp_block(sbi);
1103 up_write(&sbi->node_change);
1104out:
1105 blk_finish_plug(&plug);
1106 return err;
1107}
1108
1109static void unblock_operations(struct f2fs_sb_info *sbi)
1110{
1111 up_write(&sbi->node_write);
1112 f2fs_unlock_all(sbi);
1113}
1114
1115static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1116{
1117 DEFINE_WAIT(wait);
1118
1119 for (;;) {
1120 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1121
1122 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1123 break;
1124
1125 io_schedule_timeout(5*HZ);
1126 }
1127 finish_wait(&sbi->cp_wait, &wait);
1128}
1129
1130static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1131{
1132 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1133 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1134 unsigned long flags;
1135
1136 spin_lock_irqsave(&sbi->cp_lock, flags);
1137
1138 if ((cpc->reason & CP_UMOUNT) &&
1139 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1140 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1141 disable_nat_bits(sbi, false);
1142
1143 if (cpc->reason & CP_TRIMMED)
1144 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1145 else
1146 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1147
1148 if (cpc->reason & CP_UMOUNT)
1149 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1150 else
1151 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1152
1153 if (cpc->reason & CP_FASTBOOT)
1154 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1155 else
1156 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1157
1158 if (orphan_num)
1159 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1160 else
1161 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1162
1163 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1164 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1165
1166 /* set this flag to activate crc|cp_ver for recovery */
1167 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1168 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1169
1170 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1171}
1172
1173static void commit_checkpoint(struct f2fs_sb_info *sbi,
1174 void *src, block_t blk_addr)
1175{
1176 struct writeback_control wbc = {
1177 .for_reclaim = 0,
1178 };
1179
1180 /*
1181 * pagevec_lookup_tag and lock_page again will take
1182 * some extra time. Therefore, update_meta_pages and
1183 * sync_meta_pages are combined in this function.
1184 */
1185 struct page *page = grab_meta_page(sbi, blk_addr);
1186 int err;
1187
1188 memcpy(page_address(page), src, PAGE_SIZE);
1189 set_page_dirty(page);
1190
1191 f2fs_wait_on_page_writeback(page, META, true);
1192 f2fs_bug_on(sbi, PageWriteback(page));
1193 if (unlikely(!clear_page_dirty_for_io(page)))
1194 f2fs_bug_on(sbi, 1);
1195
1196 /* writeout cp pack 2 page */
1197 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1198 f2fs_bug_on(sbi, err);
1199
1200 f2fs_put_page(page, 0);
1201
1202 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1203 f2fs_submit_merged_write(sbi, META_FLUSH);
1204}
1205
1206static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1207{
1208 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1209 struct f2fs_nm_info *nm_i = NM_I(sbi);
1210 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1211 block_t start_blk;
1212 unsigned int data_sum_blocks, orphan_blocks;
1213 __u32 crc32 = 0;
1214 int i;
1215 int cp_payload_blks = __cp_payload(sbi);
1216 struct super_block *sb = sbi->sb;
1217 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1218 u64 kbytes_written;
1219 int err;
1220
1221 /* Flush all the NAT/SIT pages */
1222 while (get_pages(sbi, F2FS_DIRTY_META)) {
1223 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1224 if (unlikely(f2fs_cp_error(sbi)))
1225 return -EIO;
1226 }
1227
1228 /*
1229 * modify checkpoint
1230 * version number is already updated
1231 */
1232 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1233 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1234 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1235 ckpt->cur_node_segno[i] =
1236 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1237 ckpt->cur_node_blkoff[i] =
1238 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1239 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1240 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1241 }
1242 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1243 ckpt->cur_data_segno[i] =
1244 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1245 ckpt->cur_data_blkoff[i] =
1246 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1247 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1248 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1249 }
1250
1251 /* 2 cp + n data seg summary + orphan inode blocks */
1252 data_sum_blocks = npages_for_summary_flush(sbi, false);
1253 spin_lock_irqsave(&sbi->cp_lock, flags);
1254 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1255 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1256 else
1257 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1258 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1259
1260 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1261 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1262 orphan_blocks);
1263
1264 if (__remain_node_summaries(cpc->reason))
1265 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1266 cp_payload_blks + data_sum_blocks +
1267 orphan_blocks + NR_CURSEG_NODE_TYPE);
1268 else
1269 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1270 cp_payload_blks + data_sum_blocks +
1271 orphan_blocks);
1272
1273 /* update ckpt flag for checkpoint */
1274 update_ckpt_flags(sbi, cpc);
1275
1276 /* update SIT/NAT bitmap */
1277 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1278 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1279
1280 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1281 *((__le32 *)((unsigned char *)ckpt +
1282 le32_to_cpu(ckpt->checksum_offset)))
1283 = cpu_to_le32(crc32);
1284
1285 start_blk = __start_cp_next_addr(sbi);
1286
1287 /* write nat bits */
1288 if (enabled_nat_bits(sbi, cpc)) {
1289 __u64 cp_ver = cur_cp_version(ckpt);
1290 block_t blk;
1291
1292 cp_ver |= ((__u64)crc32 << 32);
1293 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1294
1295 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1296 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1297 update_meta_page(sbi, nm_i->nat_bits +
1298 (i << F2FS_BLKSIZE_BITS), blk + i);
1299
1300 /* Flush all the NAT BITS pages */
1301 while (get_pages(sbi, F2FS_DIRTY_META)) {
1302 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1303 if (unlikely(f2fs_cp_error(sbi)))
1304 return -EIO;
1305 }
1306 }
1307
1308 /* write out checkpoint buffer at block 0 */
1309 update_meta_page(sbi, ckpt, start_blk++);
1310
1311 for (i = 1; i < 1 + cp_payload_blks; i++)
1312 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1313 start_blk++);
1314
1315 if (orphan_num) {
1316 write_orphan_inodes(sbi, start_blk);
1317 start_blk += orphan_blocks;
1318 }
1319
1320 write_data_summaries(sbi, start_blk);
1321 start_blk += data_sum_blocks;
1322
1323 /* Record write statistics in the hot node summary */
1324 kbytes_written = sbi->kbytes_written;
1325 if (sb->s_bdev->bd_part)
1326 kbytes_written += BD_PART_WRITTEN(sbi);
1327
1328 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1329
1330 if (__remain_node_summaries(cpc->reason)) {
1331 write_node_summaries(sbi, start_blk);
1332 start_blk += NR_CURSEG_NODE_TYPE;
1333 }
1334
1335 /* update user_block_counts */
1336 sbi->last_valid_block_count = sbi->total_valid_block_count;
1337 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1338
1339 /* Here, we have one bio having CP pack except cp pack 2 page */
1340 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1341
1342 /* wait for previous submitted meta pages writeback */
1343 wait_on_all_pages_writeback(sbi);
1344
1345 if (unlikely(f2fs_cp_error(sbi)))
1346 return -EIO;
1347
1348 /* flush all device cache */
1349 err = f2fs_flush_device_cache(sbi);
1350 if (err)
1351 return err;
1352
1353 /* barrier and flush checkpoint cp pack 2 page if it can */
1354 commit_checkpoint(sbi, ckpt, start_blk);
1355 wait_on_all_pages_writeback(sbi);
1356
1357 release_ino_entry(sbi, false);
1358
1359 if (unlikely(f2fs_cp_error(sbi)))
1360 return -EIO;
1361
1362 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1363 clear_sbi_flag(sbi, SBI_NEED_CP);
1364 __set_cp_next_pack(sbi);
1365
1366 /*
1367 * redirty superblock if metadata like node page or inode cache is
1368 * updated during writing checkpoint.
1369 */
1370 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1371 get_pages(sbi, F2FS_DIRTY_IMETA))
1372 set_sbi_flag(sbi, SBI_IS_DIRTY);
1373
1374 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1375
1376 return 0;
1377}
1378
1379/*
1380 * We guarantee that this checkpoint procedure will not fail.
1381 */
1382int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1383{
1384 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1385 unsigned long long ckpt_ver;
1386 int err = 0;
1387
1388 mutex_lock(&sbi->cp_mutex);
1389
1390 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1391 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1392 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1393 goto out;
1394 if (unlikely(f2fs_cp_error(sbi))) {
1395 err = -EIO;
1396 goto out;
1397 }
1398 if (f2fs_readonly(sbi->sb)) {
1399 err = -EROFS;
1400 goto out;
1401 }
1402
1403 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1404
1405 err = block_operations(sbi);
1406 if (err)
1407 goto out;
1408
1409 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1410
1411 f2fs_flush_merged_writes(sbi);
1412
1413 /* this is the case of multiple fstrims without any changes */
1414 if (cpc->reason & CP_DISCARD) {
1415 if (!exist_trim_candidates(sbi, cpc)) {
1416 unblock_operations(sbi);
1417 goto out;
1418 }
1419
1420 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1421 SIT_I(sbi)->dirty_sentries == 0 &&
1422 prefree_segments(sbi) == 0) {
1423 flush_sit_entries(sbi, cpc);
1424 clear_prefree_segments(sbi, cpc);
1425 unblock_operations(sbi);
1426 goto out;
1427 }
1428 }
1429
1430 /*
1431 * update checkpoint pack index
1432 * Increase the version number so that
1433 * SIT entries and seg summaries are written at correct place
1434 */
1435 ckpt_ver = cur_cp_version(ckpt);
1436 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1437
1438 /* write cached NAT/SIT entries to NAT/SIT area */
1439 flush_nat_entries(sbi, cpc);
1440 flush_sit_entries(sbi, cpc);
1441
1442 /* unlock all the fs_lock[] in do_checkpoint() */
1443 err = do_checkpoint(sbi, cpc);
1444 if (err)
1445 release_discard_addrs(sbi);
1446 else
1447 clear_prefree_segments(sbi, cpc);
1448
1449 unblock_operations(sbi);
1450 stat_inc_cp_count(sbi->stat_info);
1451
1452 if (cpc->reason & CP_RECOVERY)
1453 f2fs_msg(sbi->sb, KERN_NOTICE,
1454 "checkpoint: version = %llx", ckpt_ver);
1455
1456 /* do checkpoint periodically */
1457 f2fs_update_time(sbi, CP_TIME);
1458 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1459out:
1460 mutex_unlock(&sbi->cp_mutex);
1461 return err;
1462}
1463
1464void init_ino_entry_info(struct f2fs_sb_info *sbi)
1465{
1466 int i;
1467
1468 for (i = 0; i < MAX_INO_ENTRY; i++) {
1469 struct inode_management *im = &sbi->im[i];
1470
1471 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1472 spin_lock_init(&im->ino_lock);
1473 INIT_LIST_HEAD(&im->ino_list);
1474 im->ino_num = 0;
1475 }
1476
1477 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1478 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1479 F2FS_ORPHANS_PER_BLOCK;
1480}
1481
1482int __init create_checkpoint_caches(void)
1483{
1484 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1485 sizeof(struct ino_entry));
1486 if (!ino_entry_slab)
1487 return -ENOMEM;
1488 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1489 sizeof(struct inode_entry));
1490 if (!inode_entry_slab) {
1491 kmem_cache_destroy(ino_entry_slab);
1492 return -ENOMEM;
1493 }
1494 return 0;
1495}
1496
1497void destroy_checkpoint_caches(void)
1498{
1499 kmem_cache_destroy(ino_entry_slab);
1500 kmem_cache_destroy(inode_entry_slab);
1501}
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 set_ckpt_flags(sbi, CP_ERROR_FLAG);
34 if (!end_io) {
35 f2fs_flush_merged_writes(sbi);
36
37 f2fs_handle_stop(sbi, reason);
38 }
39}
40
41/*
42 * We guarantee no failure on the returned page.
43 */
44struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
45{
46 struct address_space *mapping = META_MAPPING(sbi);
47 struct page *page;
48repeat:
49 page = f2fs_grab_cache_page(mapping, index, false);
50 if (!page) {
51 cond_resched();
52 goto repeat;
53 }
54 f2fs_wait_on_page_writeback(page, META, true, true);
55 if (!PageUptodate(page))
56 SetPageUptodate(page);
57 return page;
58}
59
60static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
61 bool is_meta)
62{
63 struct address_space *mapping = META_MAPPING(sbi);
64 struct page *page;
65 struct f2fs_io_info fio = {
66 .sbi = sbi,
67 .type = META,
68 .op = REQ_OP_READ,
69 .op_flags = REQ_META | REQ_PRIO,
70 .old_blkaddr = index,
71 .new_blkaddr = index,
72 .encrypted_page = NULL,
73 .is_por = !is_meta,
74 };
75 int err;
76
77 if (unlikely(!is_meta))
78 fio.op_flags &= ~REQ_META;
79repeat:
80 page = f2fs_grab_cache_page(mapping, index, false);
81 if (!page) {
82 cond_resched();
83 goto repeat;
84 }
85 if (PageUptodate(page))
86 goto out;
87
88 fio.page = page;
89
90 err = f2fs_submit_page_bio(&fio);
91 if (err) {
92 f2fs_put_page(page, 1);
93 return ERR_PTR(err);
94 }
95
96 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
97
98 lock_page(page);
99 if (unlikely(page->mapping != mapping)) {
100 f2fs_put_page(page, 1);
101 goto repeat;
102 }
103
104 if (unlikely(!PageUptodate(page))) {
105 f2fs_handle_page_eio(sbi, page->index, META);
106 f2fs_put_page(page, 1);
107 return ERR_PTR(-EIO);
108 }
109out:
110 return page;
111}
112
113struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
114{
115 return __get_meta_page(sbi, index, true);
116}
117
118struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
119{
120 struct page *page;
121 int count = 0;
122
123retry:
124 page = __get_meta_page(sbi, index, true);
125 if (IS_ERR(page)) {
126 if (PTR_ERR(page) == -EIO &&
127 ++count <= DEFAULT_RETRY_IO_COUNT)
128 goto retry;
129 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
130 }
131 return page;
132}
133
134/* for POR only */
135struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
136{
137 return __get_meta_page(sbi, index, false);
138}
139
140static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
141 int type)
142{
143 struct seg_entry *se;
144 unsigned int segno, offset;
145 bool exist;
146
147 if (type == DATA_GENERIC)
148 return true;
149
150 segno = GET_SEGNO(sbi, blkaddr);
151 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
152 se = get_seg_entry(sbi, segno);
153
154 exist = f2fs_test_bit(offset, se->cur_valid_map);
155 if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
156 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
157 blkaddr, exist);
158 set_sbi_flag(sbi, SBI_NEED_FSCK);
159 return exist;
160 }
161
162 if (!exist && type == DATA_GENERIC_ENHANCE) {
163 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
164 blkaddr, exist);
165 set_sbi_flag(sbi, SBI_NEED_FSCK);
166 dump_stack();
167 }
168 return exist;
169}
170
171bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
172 block_t blkaddr, int type)
173{
174 if (time_to_inject(sbi, FAULT_BLKADDR)) {
175 f2fs_show_injection_info(sbi, FAULT_BLKADDR);
176 return false;
177 }
178
179 switch (type) {
180 case META_NAT:
181 break;
182 case META_SIT:
183 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
184 return false;
185 break;
186 case META_SSA:
187 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
188 blkaddr < SM_I(sbi)->ssa_blkaddr))
189 return false;
190 break;
191 case META_CP:
192 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
193 blkaddr < __start_cp_addr(sbi)))
194 return false;
195 break;
196 case META_POR:
197 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
198 blkaddr < MAIN_BLKADDR(sbi)))
199 return false;
200 break;
201 case DATA_GENERIC:
202 case DATA_GENERIC_ENHANCE:
203 case DATA_GENERIC_ENHANCE_READ:
204 case DATA_GENERIC_ENHANCE_UPDATE:
205 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
206 blkaddr < MAIN_BLKADDR(sbi))) {
207 f2fs_warn(sbi, "access invalid blkaddr:%u",
208 blkaddr);
209 set_sbi_flag(sbi, SBI_NEED_FSCK);
210 dump_stack();
211 return false;
212 } else {
213 return __is_bitmap_valid(sbi, blkaddr, type);
214 }
215 break;
216 case META_GENERIC:
217 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
218 blkaddr >= MAIN_BLKADDR(sbi)))
219 return false;
220 break;
221 default:
222 BUG();
223 }
224
225 return true;
226}
227
228/*
229 * Readahead CP/NAT/SIT/SSA/POR pages
230 */
231int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
232 int type, bool sync)
233{
234 struct page *page;
235 block_t blkno = start;
236 struct f2fs_io_info fio = {
237 .sbi = sbi,
238 .type = META,
239 .op = REQ_OP_READ,
240 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
241 .encrypted_page = NULL,
242 .in_list = false,
243 .is_por = (type == META_POR),
244 };
245 struct blk_plug plug;
246 int err;
247
248 if (unlikely(type == META_POR))
249 fio.op_flags &= ~REQ_META;
250
251 blk_start_plug(&plug);
252 for (; nrpages-- > 0; blkno++) {
253
254 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
255 goto out;
256
257 switch (type) {
258 case META_NAT:
259 if (unlikely(blkno >=
260 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
261 blkno = 0;
262 /* get nat block addr */
263 fio.new_blkaddr = current_nat_addr(sbi,
264 blkno * NAT_ENTRY_PER_BLOCK);
265 break;
266 case META_SIT:
267 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
268 goto out;
269 /* get sit block addr */
270 fio.new_blkaddr = current_sit_addr(sbi,
271 blkno * SIT_ENTRY_PER_BLOCK);
272 break;
273 case META_SSA:
274 case META_CP:
275 case META_POR:
276 fio.new_blkaddr = blkno;
277 break;
278 default:
279 BUG();
280 }
281
282 page = f2fs_grab_cache_page(META_MAPPING(sbi),
283 fio.new_blkaddr, false);
284 if (!page)
285 continue;
286 if (PageUptodate(page)) {
287 f2fs_put_page(page, 1);
288 continue;
289 }
290
291 fio.page = page;
292 err = f2fs_submit_page_bio(&fio);
293 f2fs_put_page(page, err ? 1 : 0);
294
295 if (!err)
296 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
297 F2FS_BLKSIZE);
298 }
299out:
300 blk_finish_plug(&plug);
301 return blkno - start;
302}
303
304void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
305 unsigned int ra_blocks)
306{
307 struct page *page;
308 bool readahead = false;
309
310 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
311 return;
312
313 page = find_get_page(META_MAPPING(sbi), index);
314 if (!page || !PageUptodate(page))
315 readahead = true;
316 f2fs_put_page(page, 0);
317
318 if (readahead)
319 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
320}
321
322static int __f2fs_write_meta_page(struct page *page,
323 struct writeback_control *wbc,
324 enum iostat_type io_type)
325{
326 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
327
328 trace_f2fs_writepage(page, META);
329
330 if (unlikely(f2fs_cp_error(sbi)))
331 goto redirty_out;
332 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
333 goto redirty_out;
334 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
335 goto redirty_out;
336
337 f2fs_do_write_meta_page(sbi, page, io_type);
338 dec_page_count(sbi, F2FS_DIRTY_META);
339
340 if (wbc->for_reclaim)
341 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
342
343 unlock_page(page);
344
345 if (unlikely(f2fs_cp_error(sbi)))
346 f2fs_submit_merged_write(sbi, META);
347
348 return 0;
349
350redirty_out:
351 redirty_page_for_writepage(wbc, page);
352 return AOP_WRITEPAGE_ACTIVATE;
353}
354
355static int f2fs_write_meta_page(struct page *page,
356 struct writeback_control *wbc)
357{
358 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
359}
360
361static int f2fs_write_meta_pages(struct address_space *mapping,
362 struct writeback_control *wbc)
363{
364 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
365 long diff, written;
366
367 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
368 goto skip_write;
369
370 /* collect a number of dirty meta pages and write together */
371 if (wbc->sync_mode != WB_SYNC_ALL &&
372 get_pages(sbi, F2FS_DIRTY_META) <
373 nr_pages_to_skip(sbi, META))
374 goto skip_write;
375
376 /* if locked failed, cp will flush dirty pages instead */
377 if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
378 goto skip_write;
379
380 trace_f2fs_writepages(mapping->host, wbc, META);
381 diff = nr_pages_to_write(sbi, META, wbc);
382 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
383 f2fs_up_write(&sbi->cp_global_sem);
384 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
385 return 0;
386
387skip_write:
388 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
389 trace_f2fs_writepages(mapping->host, wbc, META);
390 return 0;
391}
392
393long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
394 long nr_to_write, enum iostat_type io_type)
395{
396 struct address_space *mapping = META_MAPPING(sbi);
397 pgoff_t index = 0, prev = ULONG_MAX;
398 struct pagevec pvec;
399 long nwritten = 0;
400 int nr_pages;
401 struct writeback_control wbc = {
402 .for_reclaim = 0,
403 };
404 struct blk_plug plug;
405
406 pagevec_init(&pvec);
407
408 blk_start_plug(&plug);
409
410 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
411 PAGECACHE_TAG_DIRTY))) {
412 int i;
413
414 for (i = 0; i < nr_pages; i++) {
415 struct page *page = pvec.pages[i];
416
417 if (prev == ULONG_MAX)
418 prev = page->index - 1;
419 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
420 pagevec_release(&pvec);
421 goto stop;
422 }
423
424 lock_page(page);
425
426 if (unlikely(page->mapping != mapping)) {
427continue_unlock:
428 unlock_page(page);
429 continue;
430 }
431 if (!PageDirty(page)) {
432 /* someone wrote it for us */
433 goto continue_unlock;
434 }
435
436 f2fs_wait_on_page_writeback(page, META, true, true);
437
438 if (!clear_page_dirty_for_io(page))
439 goto continue_unlock;
440
441 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
442 unlock_page(page);
443 break;
444 }
445 nwritten++;
446 prev = page->index;
447 if (unlikely(nwritten >= nr_to_write))
448 break;
449 }
450 pagevec_release(&pvec);
451 cond_resched();
452 }
453stop:
454 if (nwritten)
455 f2fs_submit_merged_write(sbi, type);
456
457 blk_finish_plug(&plug);
458
459 return nwritten;
460}
461
462static bool f2fs_dirty_meta_folio(struct address_space *mapping,
463 struct folio *folio)
464{
465 trace_f2fs_set_page_dirty(&folio->page, META);
466
467 if (!folio_test_uptodate(folio))
468 folio_mark_uptodate(folio);
469 if (filemap_dirty_folio(mapping, folio)) {
470 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
471 set_page_private_reference(&folio->page);
472 return true;
473 }
474 return false;
475}
476
477const struct address_space_operations f2fs_meta_aops = {
478 .writepage = f2fs_write_meta_page,
479 .writepages = f2fs_write_meta_pages,
480 .dirty_folio = f2fs_dirty_meta_folio,
481 .invalidate_folio = f2fs_invalidate_folio,
482 .release_folio = f2fs_release_folio,
483 .migrate_folio = filemap_migrate_folio,
484};
485
486static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
487 unsigned int devidx, int type)
488{
489 struct inode_management *im = &sbi->im[type];
490 struct ino_entry *e = NULL, *new = NULL;
491
492 if (type == FLUSH_INO) {
493 rcu_read_lock();
494 e = radix_tree_lookup(&im->ino_root, ino);
495 rcu_read_unlock();
496 }
497
498retry:
499 if (!e)
500 new = f2fs_kmem_cache_alloc(ino_entry_slab,
501 GFP_NOFS, true, NULL);
502
503 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
504
505 spin_lock(&im->ino_lock);
506 e = radix_tree_lookup(&im->ino_root, ino);
507 if (!e) {
508 if (!new) {
509 spin_unlock(&im->ino_lock);
510 goto retry;
511 }
512 e = new;
513 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
514 f2fs_bug_on(sbi, 1);
515
516 memset(e, 0, sizeof(struct ino_entry));
517 e->ino = ino;
518
519 list_add_tail(&e->list, &im->ino_list);
520 if (type != ORPHAN_INO)
521 im->ino_num++;
522 }
523
524 if (type == FLUSH_INO)
525 f2fs_set_bit(devidx, (char *)&e->dirty_device);
526
527 spin_unlock(&im->ino_lock);
528 radix_tree_preload_end();
529
530 if (new && e != new)
531 kmem_cache_free(ino_entry_slab, new);
532}
533
534static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
535{
536 struct inode_management *im = &sbi->im[type];
537 struct ino_entry *e;
538
539 spin_lock(&im->ino_lock);
540 e = radix_tree_lookup(&im->ino_root, ino);
541 if (e) {
542 list_del(&e->list);
543 radix_tree_delete(&im->ino_root, ino);
544 im->ino_num--;
545 spin_unlock(&im->ino_lock);
546 kmem_cache_free(ino_entry_slab, e);
547 return;
548 }
549 spin_unlock(&im->ino_lock);
550}
551
552void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
553{
554 /* add new dirty ino entry into list */
555 __add_ino_entry(sbi, ino, 0, type);
556}
557
558void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
559{
560 /* remove dirty ino entry from list */
561 __remove_ino_entry(sbi, ino, type);
562}
563
564/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
565bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
566{
567 struct inode_management *im = &sbi->im[mode];
568 struct ino_entry *e;
569
570 spin_lock(&im->ino_lock);
571 e = radix_tree_lookup(&im->ino_root, ino);
572 spin_unlock(&im->ino_lock);
573 return e ? true : false;
574}
575
576void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
577{
578 struct ino_entry *e, *tmp;
579 int i;
580
581 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
582 struct inode_management *im = &sbi->im[i];
583
584 spin_lock(&im->ino_lock);
585 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
586 list_del(&e->list);
587 radix_tree_delete(&im->ino_root, e->ino);
588 kmem_cache_free(ino_entry_slab, e);
589 im->ino_num--;
590 }
591 spin_unlock(&im->ino_lock);
592 }
593}
594
595void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
596 unsigned int devidx, int type)
597{
598 __add_ino_entry(sbi, ino, devidx, type);
599}
600
601bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
602 unsigned int devidx, int type)
603{
604 struct inode_management *im = &sbi->im[type];
605 struct ino_entry *e;
606 bool is_dirty = false;
607
608 spin_lock(&im->ino_lock);
609 e = radix_tree_lookup(&im->ino_root, ino);
610 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
611 is_dirty = true;
612 spin_unlock(&im->ino_lock);
613 return is_dirty;
614}
615
616int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
617{
618 struct inode_management *im = &sbi->im[ORPHAN_INO];
619 int err = 0;
620
621 spin_lock(&im->ino_lock);
622
623 if (time_to_inject(sbi, FAULT_ORPHAN)) {
624 spin_unlock(&im->ino_lock);
625 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
626 return -ENOSPC;
627 }
628
629 if (unlikely(im->ino_num >= sbi->max_orphans))
630 err = -ENOSPC;
631 else
632 im->ino_num++;
633 spin_unlock(&im->ino_lock);
634
635 return err;
636}
637
638void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
639{
640 struct inode_management *im = &sbi->im[ORPHAN_INO];
641
642 spin_lock(&im->ino_lock);
643 f2fs_bug_on(sbi, im->ino_num == 0);
644 im->ino_num--;
645 spin_unlock(&im->ino_lock);
646}
647
648void f2fs_add_orphan_inode(struct inode *inode)
649{
650 /* add new orphan ino entry into list */
651 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
652 f2fs_update_inode_page(inode);
653}
654
655void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
656{
657 /* remove orphan entry from orphan list */
658 __remove_ino_entry(sbi, ino, ORPHAN_INO);
659}
660
661static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
662{
663 struct inode *inode;
664 struct node_info ni;
665 int err;
666
667 inode = f2fs_iget_retry(sbi->sb, ino);
668 if (IS_ERR(inode)) {
669 /*
670 * there should be a bug that we can't find the entry
671 * to orphan inode.
672 */
673 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
674 return PTR_ERR(inode);
675 }
676
677 err = f2fs_dquot_initialize(inode);
678 if (err) {
679 iput(inode);
680 goto err_out;
681 }
682
683 clear_nlink(inode);
684
685 /* truncate all the data during iput */
686 iput(inode);
687
688 err = f2fs_get_node_info(sbi, ino, &ni, false);
689 if (err)
690 goto err_out;
691
692 /* ENOMEM was fully retried in f2fs_evict_inode. */
693 if (ni.blk_addr != NULL_ADDR) {
694 err = -EIO;
695 goto err_out;
696 }
697 return 0;
698
699err_out:
700 set_sbi_flag(sbi, SBI_NEED_FSCK);
701 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
702 __func__, ino);
703 return err;
704}
705
706int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
707{
708 block_t start_blk, orphan_blocks, i, j;
709 unsigned int s_flags = sbi->sb->s_flags;
710 int err = 0;
711#ifdef CONFIG_QUOTA
712 int quota_enabled;
713#endif
714
715 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
716 return 0;
717
718 if (bdev_read_only(sbi->sb->s_bdev)) {
719 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
720 return 0;
721 }
722
723 if (s_flags & SB_RDONLY) {
724 f2fs_info(sbi, "orphan cleanup on readonly fs");
725 sbi->sb->s_flags &= ~SB_RDONLY;
726 }
727
728#ifdef CONFIG_QUOTA
729 /*
730 * Turn on quotas which were not enabled for read-only mounts if
731 * filesystem has quota feature, so that they are updated correctly.
732 */
733 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
734#endif
735
736 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
737 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
738
739 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
740
741 for (i = 0; i < orphan_blocks; i++) {
742 struct page *page;
743 struct f2fs_orphan_block *orphan_blk;
744
745 page = f2fs_get_meta_page(sbi, start_blk + i);
746 if (IS_ERR(page)) {
747 err = PTR_ERR(page);
748 goto out;
749 }
750
751 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
752 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
753 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
754
755 err = recover_orphan_inode(sbi, ino);
756 if (err) {
757 f2fs_put_page(page, 1);
758 goto out;
759 }
760 }
761 f2fs_put_page(page, 1);
762 }
763 /* clear Orphan Flag */
764 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
765out:
766 set_sbi_flag(sbi, SBI_IS_RECOVERED);
767
768#ifdef CONFIG_QUOTA
769 /* Turn quotas off */
770 if (quota_enabled)
771 f2fs_quota_off_umount(sbi->sb);
772#endif
773 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
774
775 return err;
776}
777
778static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
779{
780 struct list_head *head;
781 struct f2fs_orphan_block *orphan_blk = NULL;
782 unsigned int nentries = 0;
783 unsigned short index = 1;
784 unsigned short orphan_blocks;
785 struct page *page = NULL;
786 struct ino_entry *orphan = NULL;
787 struct inode_management *im = &sbi->im[ORPHAN_INO];
788
789 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
790
791 /*
792 * we don't need to do spin_lock(&im->ino_lock) here, since all the
793 * orphan inode operations are covered under f2fs_lock_op().
794 * And, spin_lock should be avoided due to page operations below.
795 */
796 head = &im->ino_list;
797
798 /* loop for each orphan inode entry and write them in Jornal block */
799 list_for_each_entry(orphan, head, list) {
800 if (!page) {
801 page = f2fs_grab_meta_page(sbi, start_blk++);
802 orphan_blk =
803 (struct f2fs_orphan_block *)page_address(page);
804 memset(orphan_blk, 0, sizeof(*orphan_blk));
805 }
806
807 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
808
809 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
810 /*
811 * an orphan block is full of 1020 entries,
812 * then we need to flush current orphan blocks
813 * and bring another one in memory
814 */
815 orphan_blk->blk_addr = cpu_to_le16(index);
816 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
817 orphan_blk->entry_count = cpu_to_le32(nentries);
818 set_page_dirty(page);
819 f2fs_put_page(page, 1);
820 index++;
821 nentries = 0;
822 page = NULL;
823 }
824 }
825
826 if (page) {
827 orphan_blk->blk_addr = cpu_to_le16(index);
828 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
829 orphan_blk->entry_count = cpu_to_le32(nentries);
830 set_page_dirty(page);
831 f2fs_put_page(page, 1);
832 }
833}
834
835static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
836 struct f2fs_checkpoint *ckpt)
837{
838 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
839 __u32 chksum;
840
841 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
842 if (chksum_ofs < CP_CHKSUM_OFFSET) {
843 chksum_ofs += sizeof(chksum);
844 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
845 F2FS_BLKSIZE - chksum_ofs);
846 }
847 return chksum;
848}
849
850static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
851 struct f2fs_checkpoint **cp_block, struct page **cp_page,
852 unsigned long long *version)
853{
854 size_t crc_offset = 0;
855 __u32 crc;
856
857 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
858 if (IS_ERR(*cp_page))
859 return PTR_ERR(*cp_page);
860
861 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
862
863 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
864 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
865 crc_offset > CP_CHKSUM_OFFSET) {
866 f2fs_put_page(*cp_page, 1);
867 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
868 return -EINVAL;
869 }
870
871 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
872 if (crc != cur_cp_crc(*cp_block)) {
873 f2fs_put_page(*cp_page, 1);
874 f2fs_warn(sbi, "invalid crc value");
875 return -EINVAL;
876 }
877
878 *version = cur_cp_version(*cp_block);
879 return 0;
880}
881
882static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
883 block_t cp_addr, unsigned long long *version)
884{
885 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
886 struct f2fs_checkpoint *cp_block = NULL;
887 unsigned long long cur_version = 0, pre_version = 0;
888 unsigned int cp_blocks;
889 int err;
890
891 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
892 &cp_page_1, version);
893 if (err)
894 return NULL;
895
896 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
897
898 if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
899 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
900 le32_to_cpu(cp_block->cp_pack_total_block_count));
901 goto invalid_cp;
902 }
903 pre_version = *version;
904
905 cp_addr += cp_blocks - 1;
906 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
907 &cp_page_2, version);
908 if (err)
909 goto invalid_cp;
910 cur_version = *version;
911
912 if (cur_version == pre_version) {
913 *version = cur_version;
914 f2fs_put_page(cp_page_2, 1);
915 return cp_page_1;
916 }
917 f2fs_put_page(cp_page_2, 1);
918invalid_cp:
919 f2fs_put_page(cp_page_1, 1);
920 return NULL;
921}
922
923int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
924{
925 struct f2fs_checkpoint *cp_block;
926 struct f2fs_super_block *fsb = sbi->raw_super;
927 struct page *cp1, *cp2, *cur_page;
928 unsigned long blk_size = sbi->blocksize;
929 unsigned long long cp1_version = 0, cp2_version = 0;
930 unsigned long long cp_start_blk_no;
931 unsigned int cp_blks = 1 + __cp_payload(sbi);
932 block_t cp_blk_no;
933 int i;
934 int err;
935
936 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
937 GFP_KERNEL);
938 if (!sbi->ckpt)
939 return -ENOMEM;
940 /*
941 * Finding out valid cp block involves read both
942 * sets( cp pack 1 and cp pack 2)
943 */
944 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
945 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
946
947 /* The second checkpoint pack should start at the next segment */
948 cp_start_blk_no += ((unsigned long long)1) <<
949 le32_to_cpu(fsb->log_blocks_per_seg);
950 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
951
952 if (cp1 && cp2) {
953 if (ver_after(cp2_version, cp1_version))
954 cur_page = cp2;
955 else
956 cur_page = cp1;
957 } else if (cp1) {
958 cur_page = cp1;
959 } else if (cp2) {
960 cur_page = cp2;
961 } else {
962 err = -EFSCORRUPTED;
963 goto fail_no_cp;
964 }
965
966 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
967 memcpy(sbi->ckpt, cp_block, blk_size);
968
969 if (cur_page == cp1)
970 sbi->cur_cp_pack = 1;
971 else
972 sbi->cur_cp_pack = 2;
973
974 /* Sanity checking of checkpoint */
975 if (f2fs_sanity_check_ckpt(sbi)) {
976 err = -EFSCORRUPTED;
977 goto free_fail_no_cp;
978 }
979
980 if (cp_blks <= 1)
981 goto done;
982
983 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
984 if (cur_page == cp2)
985 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
986
987 for (i = 1; i < cp_blks; i++) {
988 void *sit_bitmap_ptr;
989 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
990
991 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
992 if (IS_ERR(cur_page)) {
993 err = PTR_ERR(cur_page);
994 goto free_fail_no_cp;
995 }
996 sit_bitmap_ptr = page_address(cur_page);
997 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
998 f2fs_put_page(cur_page, 1);
999 }
1000done:
1001 f2fs_put_page(cp1, 1);
1002 f2fs_put_page(cp2, 1);
1003 return 0;
1004
1005free_fail_no_cp:
1006 f2fs_put_page(cp1, 1);
1007 f2fs_put_page(cp2, 1);
1008fail_no_cp:
1009 kvfree(sbi->ckpt);
1010 return err;
1011}
1012
1013static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1014{
1015 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1016 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1017
1018 if (is_inode_flag_set(inode, flag))
1019 return;
1020
1021 set_inode_flag(inode, flag);
1022 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1023 stat_inc_dirty_inode(sbi, type);
1024}
1025
1026static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1027{
1028 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1029
1030 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1031 return;
1032
1033 list_del_init(&F2FS_I(inode)->dirty_list);
1034 clear_inode_flag(inode, flag);
1035 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1036}
1037
1038void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1039{
1040 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1041 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1042
1043 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1044 !S_ISLNK(inode->i_mode))
1045 return;
1046
1047 spin_lock(&sbi->inode_lock[type]);
1048 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1049 __add_dirty_inode(inode, type);
1050 inode_inc_dirty_pages(inode);
1051 spin_unlock(&sbi->inode_lock[type]);
1052
1053 set_page_private_reference(&folio->page);
1054}
1055
1056void f2fs_remove_dirty_inode(struct inode *inode)
1057{
1058 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1059 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1060
1061 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1062 !S_ISLNK(inode->i_mode))
1063 return;
1064
1065 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1066 return;
1067
1068 spin_lock(&sbi->inode_lock[type]);
1069 __remove_dirty_inode(inode, type);
1070 spin_unlock(&sbi->inode_lock[type]);
1071}
1072
1073int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1074 bool from_cp)
1075{
1076 struct list_head *head;
1077 struct inode *inode;
1078 struct f2fs_inode_info *fi;
1079 bool is_dir = (type == DIR_INODE);
1080 unsigned long ino = 0;
1081
1082 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1083 get_pages(sbi, is_dir ?
1084 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085retry:
1086 if (unlikely(f2fs_cp_error(sbi))) {
1087 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1088 get_pages(sbi, is_dir ?
1089 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1090 return -EIO;
1091 }
1092
1093 spin_lock(&sbi->inode_lock[type]);
1094
1095 head = &sbi->inode_list[type];
1096 if (list_empty(head)) {
1097 spin_unlock(&sbi->inode_lock[type]);
1098 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1099 get_pages(sbi, is_dir ?
1100 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1101 return 0;
1102 }
1103 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1104 inode = igrab(&fi->vfs_inode);
1105 spin_unlock(&sbi->inode_lock[type]);
1106 if (inode) {
1107 unsigned long cur_ino = inode->i_ino;
1108
1109 if (from_cp)
1110 F2FS_I(inode)->cp_task = current;
1111 F2FS_I(inode)->wb_task = current;
1112
1113 filemap_fdatawrite(inode->i_mapping);
1114
1115 F2FS_I(inode)->wb_task = NULL;
1116 if (from_cp)
1117 F2FS_I(inode)->cp_task = NULL;
1118
1119 iput(inode);
1120 /* We need to give cpu to another writers. */
1121 if (ino == cur_ino)
1122 cond_resched();
1123 else
1124 ino = cur_ino;
1125 } else {
1126 /*
1127 * We should submit bio, since it exists several
1128 * wribacking dentry pages in the freeing inode.
1129 */
1130 f2fs_submit_merged_write(sbi, DATA);
1131 cond_resched();
1132 }
1133 goto retry;
1134}
1135
1136int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1137{
1138 struct list_head *head = &sbi->inode_list[DIRTY_META];
1139 struct inode *inode;
1140 struct f2fs_inode_info *fi;
1141 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1142
1143 while (total--) {
1144 if (unlikely(f2fs_cp_error(sbi)))
1145 return -EIO;
1146
1147 spin_lock(&sbi->inode_lock[DIRTY_META]);
1148 if (list_empty(head)) {
1149 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1150 return 0;
1151 }
1152 fi = list_first_entry(head, struct f2fs_inode_info,
1153 gdirty_list);
1154 inode = igrab(&fi->vfs_inode);
1155 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1156 if (inode) {
1157 sync_inode_metadata(inode, 0);
1158
1159 /* it's on eviction */
1160 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1161 f2fs_update_inode_page(inode);
1162 iput(inode);
1163 }
1164 }
1165 return 0;
1166}
1167
1168static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1169{
1170 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1171 struct f2fs_nm_info *nm_i = NM_I(sbi);
1172 nid_t last_nid = nm_i->next_scan_nid;
1173
1174 next_free_nid(sbi, &last_nid);
1175 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1176 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1177 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1178 ckpt->next_free_nid = cpu_to_le32(last_nid);
1179}
1180
1181static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1182{
1183 bool ret = false;
1184
1185 if (!is_journalled_quota(sbi))
1186 return false;
1187
1188 if (!f2fs_down_write_trylock(&sbi->quota_sem))
1189 return true;
1190 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1191 ret = false;
1192 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1193 ret = false;
1194 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1195 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1196 ret = true;
1197 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1198 ret = true;
1199 }
1200 f2fs_up_write(&sbi->quota_sem);
1201 return ret;
1202}
1203
1204/*
1205 * Freeze all the FS-operations for checkpoint.
1206 */
1207static int block_operations(struct f2fs_sb_info *sbi)
1208{
1209 struct writeback_control wbc = {
1210 .sync_mode = WB_SYNC_ALL,
1211 .nr_to_write = LONG_MAX,
1212 .for_reclaim = 0,
1213 };
1214 int err = 0, cnt = 0;
1215
1216 /*
1217 * Let's flush inline_data in dirty node pages.
1218 */
1219 f2fs_flush_inline_data(sbi);
1220
1221retry_flush_quotas:
1222 f2fs_lock_all(sbi);
1223 if (__need_flush_quota(sbi)) {
1224 int locked;
1225
1226 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1227 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1228 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1229 goto retry_flush_dents;
1230 }
1231 f2fs_unlock_all(sbi);
1232
1233 /* only failed during mount/umount/freeze/quotactl */
1234 locked = down_read_trylock(&sbi->sb->s_umount);
1235 f2fs_quota_sync(sbi->sb, -1);
1236 if (locked)
1237 up_read(&sbi->sb->s_umount);
1238 cond_resched();
1239 goto retry_flush_quotas;
1240 }
1241
1242retry_flush_dents:
1243 /* write all the dirty dentry pages */
1244 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1245 f2fs_unlock_all(sbi);
1246 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1247 if (err)
1248 return err;
1249 cond_resched();
1250 goto retry_flush_quotas;
1251 }
1252
1253 /*
1254 * POR: we should ensure that there are no dirty node pages
1255 * until finishing nat/sit flush. inode->i_blocks can be updated.
1256 */
1257 f2fs_down_write(&sbi->node_change);
1258
1259 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1260 f2fs_up_write(&sbi->node_change);
1261 f2fs_unlock_all(sbi);
1262 err = f2fs_sync_inode_meta(sbi);
1263 if (err)
1264 return err;
1265 cond_resched();
1266 goto retry_flush_quotas;
1267 }
1268
1269retry_flush_nodes:
1270 f2fs_down_write(&sbi->node_write);
1271
1272 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1273 f2fs_up_write(&sbi->node_write);
1274 atomic_inc(&sbi->wb_sync_req[NODE]);
1275 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1276 atomic_dec(&sbi->wb_sync_req[NODE]);
1277 if (err) {
1278 f2fs_up_write(&sbi->node_change);
1279 f2fs_unlock_all(sbi);
1280 return err;
1281 }
1282 cond_resched();
1283 goto retry_flush_nodes;
1284 }
1285
1286 /*
1287 * sbi->node_change is used only for AIO write_begin path which produces
1288 * dirty node blocks and some checkpoint values by block allocation.
1289 */
1290 __prepare_cp_block(sbi);
1291 f2fs_up_write(&sbi->node_change);
1292 return err;
1293}
1294
1295static void unblock_operations(struct f2fs_sb_info *sbi)
1296{
1297 f2fs_up_write(&sbi->node_write);
1298 f2fs_unlock_all(sbi);
1299}
1300
1301void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1302{
1303 DEFINE_WAIT(wait);
1304
1305 for (;;) {
1306 if (!get_pages(sbi, type))
1307 break;
1308
1309 if (unlikely(f2fs_cp_error(sbi)))
1310 break;
1311
1312 if (type == F2FS_DIRTY_META)
1313 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1314 FS_CP_META_IO);
1315 else if (type == F2FS_WB_CP_DATA)
1316 f2fs_submit_merged_write(sbi, DATA);
1317
1318 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1319 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1320 }
1321 finish_wait(&sbi->cp_wait, &wait);
1322}
1323
1324static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1325{
1326 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1327 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1328 unsigned long flags;
1329
1330 if (cpc->reason & CP_UMOUNT) {
1331 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1332 NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1333 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1334 f2fs_notice(sbi, "Disable nat_bits due to no space");
1335 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1336 f2fs_nat_bitmap_enabled(sbi)) {
1337 f2fs_enable_nat_bits(sbi);
1338 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1339 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1340 }
1341 }
1342
1343 spin_lock_irqsave(&sbi->cp_lock, flags);
1344
1345 if (cpc->reason & CP_TRIMMED)
1346 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1347 else
1348 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1349
1350 if (cpc->reason & CP_UMOUNT)
1351 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1352 else
1353 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1354
1355 if (cpc->reason & CP_FASTBOOT)
1356 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1357 else
1358 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1359
1360 if (orphan_num)
1361 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1362 else
1363 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1364
1365 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1366 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1367
1368 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1369 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1370 else
1371 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1372
1373 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1374 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1375 else
1376 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1377
1378 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1379 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1380 else
1381 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1382
1383 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1384 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1385 else
1386 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1387
1388 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1389 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1390
1391 /* set this flag to activate crc|cp_ver for recovery */
1392 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1393 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1394
1395 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1396}
1397
1398static void commit_checkpoint(struct f2fs_sb_info *sbi,
1399 void *src, block_t blk_addr)
1400{
1401 struct writeback_control wbc = {
1402 .for_reclaim = 0,
1403 };
1404
1405 /*
1406 * pagevec_lookup_tag and lock_page again will take
1407 * some extra time. Therefore, f2fs_update_meta_pages and
1408 * f2fs_sync_meta_pages are combined in this function.
1409 */
1410 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1411 int err;
1412
1413 f2fs_wait_on_page_writeback(page, META, true, true);
1414
1415 memcpy(page_address(page), src, PAGE_SIZE);
1416
1417 set_page_dirty(page);
1418 if (unlikely(!clear_page_dirty_for_io(page)))
1419 f2fs_bug_on(sbi, 1);
1420
1421 /* writeout cp pack 2 page */
1422 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1423 if (unlikely(err && f2fs_cp_error(sbi))) {
1424 f2fs_put_page(page, 1);
1425 return;
1426 }
1427
1428 f2fs_bug_on(sbi, err);
1429 f2fs_put_page(page, 0);
1430
1431 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1432 f2fs_submit_merged_write(sbi, META_FLUSH);
1433}
1434
1435static inline u64 get_sectors_written(struct block_device *bdev)
1436{
1437 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1438}
1439
1440u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1441{
1442 if (f2fs_is_multi_device(sbi)) {
1443 u64 sectors = 0;
1444 int i;
1445
1446 for (i = 0; i < sbi->s_ndevs; i++)
1447 sectors += get_sectors_written(FDEV(i).bdev);
1448
1449 return sectors;
1450 }
1451
1452 return get_sectors_written(sbi->sb->s_bdev);
1453}
1454
1455static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1456{
1457 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1458 struct f2fs_nm_info *nm_i = NM_I(sbi);
1459 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1460 block_t start_blk;
1461 unsigned int data_sum_blocks, orphan_blocks;
1462 __u32 crc32 = 0;
1463 int i;
1464 int cp_payload_blks = __cp_payload(sbi);
1465 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1466 u64 kbytes_written;
1467 int err;
1468
1469 /* Flush all the NAT/SIT pages */
1470 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1471
1472 /* start to update checkpoint, cp ver is already updated previously */
1473 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1474 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1475 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1476 ckpt->cur_node_segno[i] =
1477 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1478 ckpt->cur_node_blkoff[i] =
1479 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1480 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1481 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1482 }
1483 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1484 ckpt->cur_data_segno[i] =
1485 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1486 ckpt->cur_data_blkoff[i] =
1487 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1488 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1489 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1490 }
1491
1492 /* 2 cp + n data seg summary + orphan inode blocks */
1493 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1494 spin_lock_irqsave(&sbi->cp_lock, flags);
1495 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1496 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1497 else
1498 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1499 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1500
1501 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1502 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1503 orphan_blocks);
1504
1505 if (__remain_node_summaries(cpc->reason))
1506 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1507 cp_payload_blks + data_sum_blocks +
1508 orphan_blocks + NR_CURSEG_NODE_TYPE);
1509 else
1510 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1511 cp_payload_blks + data_sum_blocks +
1512 orphan_blocks);
1513
1514 /* update ckpt flag for checkpoint */
1515 update_ckpt_flags(sbi, cpc);
1516
1517 /* update SIT/NAT bitmap */
1518 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1519 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1520
1521 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1522 *((__le32 *)((unsigned char *)ckpt +
1523 le32_to_cpu(ckpt->checksum_offset)))
1524 = cpu_to_le32(crc32);
1525
1526 start_blk = __start_cp_next_addr(sbi);
1527
1528 /* write nat bits */
1529 if ((cpc->reason & CP_UMOUNT) &&
1530 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1531 __u64 cp_ver = cur_cp_version(ckpt);
1532 block_t blk;
1533
1534 cp_ver |= ((__u64)crc32 << 32);
1535 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1536
1537 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1538 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1539 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1540 (i << F2FS_BLKSIZE_BITS), blk + i);
1541 }
1542
1543 /* write out checkpoint buffer at block 0 */
1544 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1545
1546 for (i = 1; i < 1 + cp_payload_blks; i++)
1547 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1548 start_blk++);
1549
1550 if (orphan_num) {
1551 write_orphan_inodes(sbi, start_blk);
1552 start_blk += orphan_blocks;
1553 }
1554
1555 f2fs_write_data_summaries(sbi, start_blk);
1556 start_blk += data_sum_blocks;
1557
1558 /* Record write statistics in the hot node summary */
1559 kbytes_written = sbi->kbytes_written;
1560 kbytes_written += (f2fs_get_sectors_written(sbi) -
1561 sbi->sectors_written_start) >> 1;
1562 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1563
1564 if (__remain_node_summaries(cpc->reason)) {
1565 f2fs_write_node_summaries(sbi, start_blk);
1566 start_blk += NR_CURSEG_NODE_TYPE;
1567 }
1568
1569 /* update user_block_counts */
1570 sbi->last_valid_block_count = sbi->total_valid_block_count;
1571 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1572 percpu_counter_set(&sbi->rf_node_block_count, 0);
1573
1574 /* Here, we have one bio having CP pack except cp pack 2 page */
1575 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1576 /* Wait for all dirty meta pages to be submitted for IO */
1577 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1578
1579 /* wait for previous submitted meta pages writeback */
1580 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1581
1582 /* flush all device cache */
1583 err = f2fs_flush_device_cache(sbi);
1584 if (err)
1585 return err;
1586
1587 /* barrier and flush checkpoint cp pack 2 page if it can */
1588 commit_checkpoint(sbi, ckpt, start_blk);
1589 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1590
1591 /*
1592 * invalidate intermediate page cache borrowed from meta inode which are
1593 * used for migration of encrypted, verity or compressed inode's blocks.
1594 */
1595 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1596 f2fs_sb_has_compression(sbi))
1597 invalidate_mapping_pages(META_MAPPING(sbi),
1598 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1599
1600 f2fs_release_ino_entry(sbi, false);
1601
1602 f2fs_reset_fsync_node_info(sbi);
1603
1604 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1605 clear_sbi_flag(sbi, SBI_NEED_CP);
1606 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1607
1608 spin_lock(&sbi->stat_lock);
1609 sbi->unusable_block_count = 0;
1610 spin_unlock(&sbi->stat_lock);
1611
1612 __set_cp_next_pack(sbi);
1613
1614 /*
1615 * redirty superblock if metadata like node page or inode cache is
1616 * updated during writing checkpoint.
1617 */
1618 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1619 get_pages(sbi, F2FS_DIRTY_IMETA))
1620 set_sbi_flag(sbi, SBI_IS_DIRTY);
1621
1622 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1623
1624 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1625}
1626
1627int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1628{
1629 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1630 unsigned long long ckpt_ver;
1631 int err = 0;
1632
1633 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1634 return -EROFS;
1635
1636 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1637 if (cpc->reason != CP_PAUSE)
1638 return 0;
1639 f2fs_warn(sbi, "Start checkpoint disabled!");
1640 }
1641 if (cpc->reason != CP_RESIZE)
1642 f2fs_down_write(&sbi->cp_global_sem);
1643
1644 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1645 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1646 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1647 goto out;
1648 if (unlikely(f2fs_cp_error(sbi))) {
1649 err = -EIO;
1650 goto out;
1651 }
1652
1653 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1654
1655 err = block_operations(sbi);
1656 if (err)
1657 goto out;
1658
1659 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1660
1661 f2fs_flush_merged_writes(sbi);
1662
1663 /* this is the case of multiple fstrims without any changes */
1664 if (cpc->reason & CP_DISCARD) {
1665 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1666 unblock_operations(sbi);
1667 goto out;
1668 }
1669
1670 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1671 SIT_I(sbi)->dirty_sentries == 0 &&
1672 prefree_segments(sbi) == 0) {
1673 f2fs_flush_sit_entries(sbi, cpc);
1674 f2fs_clear_prefree_segments(sbi, cpc);
1675 unblock_operations(sbi);
1676 goto out;
1677 }
1678 }
1679
1680 /*
1681 * update checkpoint pack index
1682 * Increase the version number so that
1683 * SIT entries and seg summaries are written at correct place
1684 */
1685 ckpt_ver = cur_cp_version(ckpt);
1686 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1687
1688 /* write cached NAT/SIT entries to NAT/SIT area */
1689 err = f2fs_flush_nat_entries(sbi, cpc);
1690 if (err) {
1691 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1692 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1693 goto stop;
1694 }
1695
1696 f2fs_flush_sit_entries(sbi, cpc);
1697
1698 /* save inmem log status */
1699 f2fs_save_inmem_curseg(sbi);
1700
1701 err = do_checkpoint(sbi, cpc);
1702 if (err) {
1703 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1704 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1705 f2fs_release_discard_addrs(sbi);
1706 } else {
1707 f2fs_clear_prefree_segments(sbi, cpc);
1708 }
1709
1710 f2fs_restore_inmem_curseg(sbi);
1711stop:
1712 unblock_operations(sbi);
1713 stat_inc_cp_count(sbi->stat_info);
1714
1715 if (cpc->reason & CP_RECOVERY)
1716 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1717
1718 /* update CP_TIME to trigger checkpoint periodically */
1719 f2fs_update_time(sbi, CP_TIME);
1720 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1721out:
1722 if (cpc->reason != CP_RESIZE)
1723 f2fs_up_write(&sbi->cp_global_sem);
1724 return err;
1725}
1726
1727void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1728{
1729 int i;
1730
1731 for (i = 0; i < MAX_INO_ENTRY; i++) {
1732 struct inode_management *im = &sbi->im[i];
1733
1734 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1735 spin_lock_init(&im->ino_lock);
1736 INIT_LIST_HEAD(&im->ino_list);
1737 im->ino_num = 0;
1738 }
1739
1740 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1741 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1742 F2FS_ORPHANS_PER_BLOCK;
1743}
1744
1745int __init f2fs_create_checkpoint_caches(void)
1746{
1747 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1748 sizeof(struct ino_entry));
1749 if (!ino_entry_slab)
1750 return -ENOMEM;
1751 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1752 sizeof(struct inode_entry));
1753 if (!f2fs_inode_entry_slab) {
1754 kmem_cache_destroy(ino_entry_slab);
1755 return -ENOMEM;
1756 }
1757 return 0;
1758}
1759
1760void f2fs_destroy_checkpoint_caches(void)
1761{
1762 kmem_cache_destroy(ino_entry_slab);
1763 kmem_cache_destroy(f2fs_inode_entry_slab);
1764}
1765
1766static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1767{
1768 struct cp_control cpc = { .reason = CP_SYNC, };
1769 int err;
1770
1771 f2fs_down_write(&sbi->gc_lock);
1772 err = f2fs_write_checkpoint(sbi, &cpc);
1773 f2fs_up_write(&sbi->gc_lock);
1774
1775 return err;
1776}
1777
1778static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1779{
1780 struct ckpt_req_control *cprc = &sbi->cprc_info;
1781 struct ckpt_req *req, *next;
1782 struct llist_node *dispatch_list;
1783 u64 sum_diff = 0, diff, count = 0;
1784 int ret;
1785
1786 dispatch_list = llist_del_all(&cprc->issue_list);
1787 if (!dispatch_list)
1788 return;
1789 dispatch_list = llist_reverse_order(dispatch_list);
1790
1791 ret = __write_checkpoint_sync(sbi);
1792 atomic_inc(&cprc->issued_ckpt);
1793
1794 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1795 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1796 req->ret = ret;
1797 complete(&req->wait);
1798
1799 sum_diff += diff;
1800 count++;
1801 }
1802 atomic_sub(count, &cprc->queued_ckpt);
1803 atomic_add(count, &cprc->total_ckpt);
1804
1805 spin_lock(&cprc->stat_lock);
1806 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1807 if (cprc->peak_time < cprc->cur_time)
1808 cprc->peak_time = cprc->cur_time;
1809 spin_unlock(&cprc->stat_lock);
1810}
1811
1812static int issue_checkpoint_thread(void *data)
1813{
1814 struct f2fs_sb_info *sbi = data;
1815 struct ckpt_req_control *cprc = &sbi->cprc_info;
1816 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1817repeat:
1818 if (kthread_should_stop())
1819 return 0;
1820
1821 if (!llist_empty(&cprc->issue_list))
1822 __checkpoint_and_complete_reqs(sbi);
1823
1824 wait_event_interruptible(*q,
1825 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1826 goto repeat;
1827}
1828
1829static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1830 struct ckpt_req *wait_req)
1831{
1832 struct ckpt_req_control *cprc = &sbi->cprc_info;
1833
1834 if (!llist_empty(&cprc->issue_list)) {
1835 __checkpoint_and_complete_reqs(sbi);
1836 } else {
1837 /* already dispatched by issue_checkpoint_thread */
1838 if (wait_req)
1839 wait_for_completion(&wait_req->wait);
1840 }
1841}
1842
1843static void init_ckpt_req(struct ckpt_req *req)
1844{
1845 memset(req, 0, sizeof(struct ckpt_req));
1846
1847 init_completion(&req->wait);
1848 req->queue_time = ktime_get();
1849}
1850
1851int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1852{
1853 struct ckpt_req_control *cprc = &sbi->cprc_info;
1854 struct ckpt_req req;
1855 struct cp_control cpc;
1856
1857 cpc.reason = __get_cp_reason(sbi);
1858 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1859 int ret;
1860
1861 f2fs_down_write(&sbi->gc_lock);
1862 ret = f2fs_write_checkpoint(sbi, &cpc);
1863 f2fs_up_write(&sbi->gc_lock);
1864
1865 return ret;
1866 }
1867
1868 if (!cprc->f2fs_issue_ckpt)
1869 return __write_checkpoint_sync(sbi);
1870
1871 init_ckpt_req(&req);
1872
1873 llist_add(&req.llnode, &cprc->issue_list);
1874 atomic_inc(&cprc->queued_ckpt);
1875
1876 /*
1877 * update issue_list before we wake up issue_checkpoint thread,
1878 * this smp_mb() pairs with another barrier in ___wait_event(),
1879 * see more details in comments of waitqueue_active().
1880 */
1881 smp_mb();
1882
1883 if (waitqueue_active(&cprc->ckpt_wait_queue))
1884 wake_up(&cprc->ckpt_wait_queue);
1885
1886 if (cprc->f2fs_issue_ckpt)
1887 wait_for_completion(&req.wait);
1888 else
1889 flush_remained_ckpt_reqs(sbi, &req);
1890
1891 return req.ret;
1892}
1893
1894int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1895{
1896 dev_t dev = sbi->sb->s_bdev->bd_dev;
1897 struct ckpt_req_control *cprc = &sbi->cprc_info;
1898
1899 if (cprc->f2fs_issue_ckpt)
1900 return 0;
1901
1902 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1903 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1904 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1905 int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1906
1907 cprc->f2fs_issue_ckpt = NULL;
1908 return err;
1909 }
1910
1911 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1912
1913 return 0;
1914}
1915
1916void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1917{
1918 struct ckpt_req_control *cprc = &sbi->cprc_info;
1919 struct task_struct *ckpt_task;
1920
1921 if (!cprc->f2fs_issue_ckpt)
1922 return;
1923
1924 ckpt_task = cprc->f2fs_issue_ckpt;
1925 cprc->f2fs_issue_ckpt = NULL;
1926 kthread_stop(ckpt_task);
1927
1928 f2fs_flush_ckpt_thread(sbi);
1929}
1930
1931void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1932{
1933 struct ckpt_req_control *cprc = &sbi->cprc_info;
1934
1935 flush_remained_ckpt_reqs(sbi, NULL);
1936
1937 /* Let's wait for the previous dispatched checkpoint. */
1938 while (atomic_read(&cprc->queued_ckpt))
1939 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1940}
1941
1942void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1943{
1944 struct ckpt_req_control *cprc = &sbi->cprc_info;
1945
1946 atomic_set(&cprc->issued_ckpt, 0);
1947 atomic_set(&cprc->total_ckpt, 0);
1948 atomic_set(&cprc->queued_ckpt, 0);
1949 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1950 init_waitqueue_head(&cprc->ckpt_wait_queue);
1951 init_llist_head(&cprc->issue_list);
1952 spin_lock_init(&cprc->stat_lock);
1953}