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 if (!end_io)
34 f2fs_flush_merged_writes(sbi);
35 f2fs_handle_critical_error(sbi, reason);
36}
37
38/*
39 * We guarantee no failure on the returned page.
40 */
41struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
42{
43 struct address_space *mapping = META_MAPPING(sbi);
44 struct page *page;
45repeat:
46 page = f2fs_grab_cache_page(mapping, index, false);
47 if (!page) {
48 cond_resched();
49 goto repeat;
50 }
51 f2fs_wait_on_page_writeback(page, META, true, true);
52 if (!PageUptodate(page))
53 SetPageUptodate(page);
54 return page;
55}
56
57static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
58 bool is_meta)
59{
60 struct address_space *mapping = META_MAPPING(sbi);
61 struct page *page;
62 struct f2fs_io_info fio = {
63 .sbi = sbi,
64 .type = META,
65 .op = REQ_OP_READ,
66 .op_flags = REQ_META | REQ_PRIO,
67 .old_blkaddr = index,
68 .new_blkaddr = index,
69 .encrypted_page = NULL,
70 .is_por = !is_meta ? 1 : 0,
71 };
72 int err;
73
74 if (unlikely(!is_meta))
75 fio.op_flags &= ~REQ_META;
76repeat:
77 page = f2fs_grab_cache_page(mapping, index, false);
78 if (!page) {
79 cond_resched();
80 goto repeat;
81 }
82 if (PageUptodate(page))
83 goto out;
84
85 fio.page = page;
86
87 err = f2fs_submit_page_bio(&fio);
88 if (err) {
89 f2fs_put_page(page, 1);
90 return ERR_PTR(err);
91 }
92
93 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
94
95 lock_page(page);
96 if (unlikely(page->mapping != mapping)) {
97 f2fs_put_page(page, 1);
98 goto repeat;
99 }
100
101 if (unlikely(!PageUptodate(page))) {
102 f2fs_handle_page_eio(sbi, page_folio(page), META);
103 f2fs_put_page(page, 1);
104 return ERR_PTR(-EIO);
105 }
106out:
107 return page;
108}
109
110struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
111{
112 return __get_meta_page(sbi, index, true);
113}
114
115struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
116{
117 struct page *page;
118 int count = 0;
119
120retry:
121 page = __get_meta_page(sbi, index, true);
122 if (IS_ERR(page)) {
123 if (PTR_ERR(page) == -EIO &&
124 ++count <= DEFAULT_RETRY_IO_COUNT)
125 goto retry;
126 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
127 }
128 return page;
129}
130
131/* for POR only */
132struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
133{
134 return __get_meta_page(sbi, index, false);
135}
136
137static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
138 int type)
139{
140 struct seg_entry *se;
141 unsigned int segno, offset;
142 bool exist;
143
144 if (type == DATA_GENERIC)
145 return true;
146
147 segno = GET_SEGNO(sbi, blkaddr);
148 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
149 se = get_seg_entry(sbi, segno);
150
151 exist = f2fs_test_bit(offset, se->cur_valid_map);
152
153 /* skip data, if we already have an error in checkpoint. */
154 if (unlikely(f2fs_cp_error(sbi)))
155 return exist;
156
157 if ((exist && type == DATA_GENERIC_ENHANCE_UPDATE) ||
158 (!exist && type == DATA_GENERIC_ENHANCE))
159 goto out_err;
160 if (!exist && type != DATA_GENERIC_ENHANCE_UPDATE)
161 goto out_handle;
162 return exist;
163
164out_err:
165 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
166 blkaddr, exist);
167 set_sbi_flag(sbi, SBI_NEED_FSCK);
168 dump_stack();
169out_handle:
170 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
171 return exist;
172}
173
174static bool __f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
175 block_t blkaddr, int type)
176{
177 switch (type) {
178 case META_NAT:
179 break;
180 case META_SIT:
181 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
182 goto check_only;
183 break;
184 case META_SSA:
185 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
186 blkaddr < SM_I(sbi)->ssa_blkaddr))
187 goto check_only;
188 break;
189 case META_CP:
190 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
191 blkaddr < __start_cp_addr(sbi)))
192 goto check_only;
193 break;
194 case META_POR:
195 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
196 blkaddr < MAIN_BLKADDR(sbi)))
197 goto check_only;
198 break;
199 case DATA_GENERIC:
200 case DATA_GENERIC_ENHANCE:
201 case DATA_GENERIC_ENHANCE_READ:
202 case DATA_GENERIC_ENHANCE_UPDATE:
203 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
204 blkaddr < MAIN_BLKADDR(sbi))) {
205
206 /* Skip to emit an error message. */
207 if (unlikely(f2fs_cp_error(sbi)))
208 return false;
209
210 f2fs_warn(sbi, "access invalid blkaddr:%u",
211 blkaddr);
212 set_sbi_flag(sbi, SBI_NEED_FSCK);
213 dump_stack();
214 goto err;
215 } else {
216 return __is_bitmap_valid(sbi, blkaddr, type);
217 }
218 break;
219 case META_GENERIC:
220 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
221 blkaddr >= MAIN_BLKADDR(sbi)))
222 goto err;
223 break;
224 default:
225 BUG();
226 }
227
228 return true;
229err:
230 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
231check_only:
232 return false;
233}
234
235bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
236 block_t blkaddr, int type)
237{
238 if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY))
239 return false;
240 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
241}
242
243bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi,
244 block_t blkaddr, int type)
245{
246 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
247}
248
249/*
250 * Readahead CP/NAT/SIT/SSA/POR pages
251 */
252int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
253 int type, bool sync)
254{
255 struct page *page;
256 block_t blkno = start;
257 struct f2fs_io_info fio = {
258 .sbi = sbi,
259 .type = META,
260 .op = REQ_OP_READ,
261 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
262 .encrypted_page = NULL,
263 .in_list = 0,
264 .is_por = (type == META_POR) ? 1 : 0,
265 };
266 struct blk_plug plug;
267 int err;
268
269 if (unlikely(type == META_POR))
270 fio.op_flags &= ~REQ_META;
271
272 blk_start_plug(&plug);
273 for (; nrpages-- > 0; blkno++) {
274
275 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
276 goto out;
277
278 switch (type) {
279 case META_NAT:
280 if (unlikely(blkno >=
281 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
282 blkno = 0;
283 /* get nat block addr */
284 fio.new_blkaddr = current_nat_addr(sbi,
285 blkno * NAT_ENTRY_PER_BLOCK);
286 break;
287 case META_SIT:
288 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
289 goto out;
290 /* get sit block addr */
291 fio.new_blkaddr = current_sit_addr(sbi,
292 blkno * SIT_ENTRY_PER_BLOCK);
293 break;
294 case META_SSA:
295 case META_CP:
296 case META_POR:
297 fio.new_blkaddr = blkno;
298 break;
299 default:
300 BUG();
301 }
302
303 page = f2fs_grab_cache_page(META_MAPPING(sbi),
304 fio.new_blkaddr, false);
305 if (!page)
306 continue;
307 if (PageUptodate(page)) {
308 f2fs_put_page(page, 1);
309 continue;
310 }
311
312 fio.page = page;
313 err = f2fs_submit_page_bio(&fio);
314 f2fs_put_page(page, err ? 1 : 0);
315
316 if (!err)
317 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
318 F2FS_BLKSIZE);
319 }
320out:
321 blk_finish_plug(&plug);
322 return blkno - start;
323}
324
325void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
326 unsigned int ra_blocks)
327{
328 struct page *page;
329 bool readahead = false;
330
331 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
332 return;
333
334 page = find_get_page(META_MAPPING(sbi), index);
335 if (!page || !PageUptodate(page))
336 readahead = true;
337 f2fs_put_page(page, 0);
338
339 if (readahead)
340 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
341}
342
343static int __f2fs_write_meta_page(struct page *page,
344 struct writeback_control *wbc,
345 enum iostat_type io_type)
346{
347 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
348 struct folio *folio = page_folio(page);
349
350 trace_f2fs_writepage(folio, META);
351
352 if (unlikely(f2fs_cp_error(sbi))) {
353 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
354 folio_clear_uptodate(folio);
355 dec_page_count(sbi, F2FS_DIRTY_META);
356 folio_unlock(folio);
357 return 0;
358 }
359 goto redirty_out;
360 }
361 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
362 goto redirty_out;
363 if (wbc->for_reclaim && folio->index < GET_SUM_BLOCK(sbi, 0))
364 goto redirty_out;
365
366 f2fs_do_write_meta_page(sbi, folio, io_type);
367 dec_page_count(sbi, F2FS_DIRTY_META);
368
369 if (wbc->for_reclaim)
370 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
371
372 folio_unlock(folio);
373
374 if (unlikely(f2fs_cp_error(sbi)))
375 f2fs_submit_merged_write(sbi, META);
376
377 return 0;
378
379redirty_out:
380 redirty_page_for_writepage(wbc, page);
381 return AOP_WRITEPAGE_ACTIVATE;
382}
383
384static int f2fs_write_meta_page(struct page *page,
385 struct writeback_control *wbc)
386{
387 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
388}
389
390static int f2fs_write_meta_pages(struct address_space *mapping,
391 struct writeback_control *wbc)
392{
393 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
394 long diff, written;
395
396 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
397 goto skip_write;
398
399 /* collect a number of dirty meta pages and write together */
400 if (wbc->sync_mode != WB_SYNC_ALL &&
401 get_pages(sbi, F2FS_DIRTY_META) <
402 nr_pages_to_skip(sbi, META))
403 goto skip_write;
404
405 /* if locked failed, cp will flush dirty pages instead */
406 if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
407 goto skip_write;
408
409 trace_f2fs_writepages(mapping->host, wbc, META);
410 diff = nr_pages_to_write(sbi, META, wbc);
411 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
412 f2fs_up_write(&sbi->cp_global_sem);
413 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
414 return 0;
415
416skip_write:
417 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
418 trace_f2fs_writepages(mapping->host, wbc, META);
419 return 0;
420}
421
422long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
423 long nr_to_write, enum iostat_type io_type)
424{
425 struct address_space *mapping = META_MAPPING(sbi);
426 pgoff_t index = 0, prev = ULONG_MAX;
427 struct folio_batch fbatch;
428 long nwritten = 0;
429 int nr_folios;
430 struct writeback_control wbc = {
431 .for_reclaim = 0,
432 };
433 struct blk_plug plug;
434
435 folio_batch_init(&fbatch);
436
437 blk_start_plug(&plug);
438
439 while ((nr_folios = filemap_get_folios_tag(mapping, &index,
440 (pgoff_t)-1,
441 PAGECACHE_TAG_DIRTY, &fbatch))) {
442 int i;
443
444 for (i = 0; i < nr_folios; i++) {
445 struct folio *folio = fbatch.folios[i];
446
447 if (nr_to_write != LONG_MAX && i != 0 &&
448 folio->index != prev +
449 folio_nr_pages(fbatch.folios[i-1])) {
450 folio_batch_release(&fbatch);
451 goto stop;
452 }
453
454 folio_lock(folio);
455
456 if (unlikely(folio->mapping != mapping)) {
457continue_unlock:
458 folio_unlock(folio);
459 continue;
460 }
461 if (!folio_test_dirty(folio)) {
462 /* someone wrote it for us */
463 goto continue_unlock;
464 }
465
466 f2fs_wait_on_page_writeback(&folio->page, META,
467 true, true);
468
469 if (!folio_clear_dirty_for_io(folio))
470 goto continue_unlock;
471
472 if (__f2fs_write_meta_page(&folio->page, &wbc,
473 io_type)) {
474 folio_unlock(folio);
475 break;
476 }
477 nwritten += folio_nr_pages(folio);
478 prev = folio->index;
479 if (unlikely(nwritten >= nr_to_write))
480 break;
481 }
482 folio_batch_release(&fbatch);
483 cond_resched();
484 }
485stop:
486 if (nwritten)
487 f2fs_submit_merged_write(sbi, type);
488
489 blk_finish_plug(&plug);
490
491 return nwritten;
492}
493
494static bool f2fs_dirty_meta_folio(struct address_space *mapping,
495 struct folio *folio)
496{
497 trace_f2fs_set_page_dirty(folio, META);
498
499 if (!folio_test_uptodate(folio))
500 folio_mark_uptodate(folio);
501 if (filemap_dirty_folio(mapping, folio)) {
502 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
503 set_page_private_reference(&folio->page);
504 return true;
505 }
506 return false;
507}
508
509const struct address_space_operations f2fs_meta_aops = {
510 .writepage = f2fs_write_meta_page,
511 .writepages = f2fs_write_meta_pages,
512 .dirty_folio = f2fs_dirty_meta_folio,
513 .invalidate_folio = f2fs_invalidate_folio,
514 .release_folio = f2fs_release_folio,
515 .migrate_folio = filemap_migrate_folio,
516};
517
518static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
519 unsigned int devidx, int type)
520{
521 struct inode_management *im = &sbi->im[type];
522 struct ino_entry *e = NULL, *new = NULL;
523
524 if (type == FLUSH_INO) {
525 rcu_read_lock();
526 e = radix_tree_lookup(&im->ino_root, ino);
527 rcu_read_unlock();
528 }
529
530retry:
531 if (!e)
532 new = f2fs_kmem_cache_alloc(ino_entry_slab,
533 GFP_NOFS, true, NULL);
534
535 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
536
537 spin_lock(&im->ino_lock);
538 e = radix_tree_lookup(&im->ino_root, ino);
539 if (!e) {
540 if (!new) {
541 spin_unlock(&im->ino_lock);
542 radix_tree_preload_end();
543 goto retry;
544 }
545 e = new;
546 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
547 f2fs_bug_on(sbi, 1);
548
549 memset(e, 0, sizeof(struct ino_entry));
550 e->ino = ino;
551
552 list_add_tail(&e->list, &im->ino_list);
553 if (type != ORPHAN_INO)
554 im->ino_num++;
555 }
556
557 if (type == FLUSH_INO)
558 f2fs_set_bit(devidx, (char *)&e->dirty_device);
559
560 spin_unlock(&im->ino_lock);
561 radix_tree_preload_end();
562
563 if (new && e != new)
564 kmem_cache_free(ino_entry_slab, new);
565}
566
567static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
568{
569 struct inode_management *im = &sbi->im[type];
570 struct ino_entry *e;
571
572 spin_lock(&im->ino_lock);
573 e = radix_tree_lookup(&im->ino_root, ino);
574 if (e) {
575 list_del(&e->list);
576 radix_tree_delete(&im->ino_root, ino);
577 im->ino_num--;
578 spin_unlock(&im->ino_lock);
579 kmem_cache_free(ino_entry_slab, e);
580 return;
581 }
582 spin_unlock(&im->ino_lock);
583}
584
585void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
586{
587 /* add new dirty ino entry into list */
588 __add_ino_entry(sbi, ino, 0, type);
589}
590
591void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
592{
593 /* remove dirty ino entry from list */
594 __remove_ino_entry(sbi, ino, type);
595}
596
597/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
598bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
599{
600 struct inode_management *im = &sbi->im[mode];
601 struct ino_entry *e;
602
603 spin_lock(&im->ino_lock);
604 e = radix_tree_lookup(&im->ino_root, ino);
605 spin_unlock(&im->ino_lock);
606 return e ? true : false;
607}
608
609void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
610{
611 struct ino_entry *e, *tmp;
612 int i;
613
614 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
615 struct inode_management *im = &sbi->im[i];
616
617 spin_lock(&im->ino_lock);
618 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
619 list_del(&e->list);
620 radix_tree_delete(&im->ino_root, e->ino);
621 kmem_cache_free(ino_entry_slab, e);
622 im->ino_num--;
623 }
624 spin_unlock(&im->ino_lock);
625 }
626}
627
628void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
629 unsigned int devidx, int type)
630{
631 __add_ino_entry(sbi, ino, devidx, type);
632}
633
634bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
635 unsigned int devidx, int type)
636{
637 struct inode_management *im = &sbi->im[type];
638 struct ino_entry *e;
639 bool is_dirty = false;
640
641 spin_lock(&im->ino_lock);
642 e = radix_tree_lookup(&im->ino_root, ino);
643 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
644 is_dirty = true;
645 spin_unlock(&im->ino_lock);
646 return is_dirty;
647}
648
649int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
650{
651 struct inode_management *im = &sbi->im[ORPHAN_INO];
652 int err = 0;
653
654 spin_lock(&im->ino_lock);
655
656 if (time_to_inject(sbi, FAULT_ORPHAN)) {
657 spin_unlock(&im->ino_lock);
658 return -ENOSPC;
659 }
660
661 if (unlikely(im->ino_num >= sbi->max_orphans))
662 err = -ENOSPC;
663 else
664 im->ino_num++;
665 spin_unlock(&im->ino_lock);
666
667 return err;
668}
669
670void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
671{
672 struct inode_management *im = &sbi->im[ORPHAN_INO];
673
674 spin_lock(&im->ino_lock);
675 f2fs_bug_on(sbi, im->ino_num == 0);
676 im->ino_num--;
677 spin_unlock(&im->ino_lock);
678}
679
680void f2fs_add_orphan_inode(struct inode *inode)
681{
682 /* add new orphan ino entry into list */
683 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
684 f2fs_update_inode_page(inode);
685}
686
687void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
688{
689 /* remove orphan entry from orphan list */
690 __remove_ino_entry(sbi, ino, ORPHAN_INO);
691}
692
693static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
694{
695 struct inode *inode;
696 struct node_info ni;
697 int err;
698
699 inode = f2fs_iget_retry(sbi->sb, ino);
700 if (IS_ERR(inode)) {
701 /*
702 * there should be a bug that we can't find the entry
703 * to orphan inode.
704 */
705 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
706 return PTR_ERR(inode);
707 }
708
709 err = f2fs_dquot_initialize(inode);
710 if (err) {
711 iput(inode);
712 goto err_out;
713 }
714
715 clear_nlink(inode);
716
717 /* truncate all the data during iput */
718 iput(inode);
719
720 err = f2fs_get_node_info(sbi, ino, &ni, false);
721 if (err)
722 goto err_out;
723
724 /* ENOMEM was fully retried in f2fs_evict_inode. */
725 if (ni.blk_addr != NULL_ADDR) {
726 err = -EIO;
727 goto err_out;
728 }
729 return 0;
730
731err_out:
732 set_sbi_flag(sbi, SBI_NEED_FSCK);
733 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
734 __func__, ino);
735 return err;
736}
737
738int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
739{
740 block_t start_blk, orphan_blocks, i, j;
741 int err = 0;
742
743 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
744 return 0;
745
746 if (f2fs_hw_is_readonly(sbi)) {
747 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
748 return 0;
749 }
750
751 if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
752 f2fs_info(sbi, "orphan cleanup on readonly fs");
753
754 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
755 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
756
757 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
758
759 for (i = 0; i < orphan_blocks; i++) {
760 struct page *page;
761 struct f2fs_orphan_block *orphan_blk;
762
763 page = f2fs_get_meta_page(sbi, start_blk + i);
764 if (IS_ERR(page)) {
765 err = PTR_ERR(page);
766 goto out;
767 }
768
769 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
770 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
771 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
772
773 err = recover_orphan_inode(sbi, ino);
774 if (err) {
775 f2fs_put_page(page, 1);
776 goto out;
777 }
778 }
779 f2fs_put_page(page, 1);
780 }
781 /* clear Orphan Flag */
782 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
783out:
784 set_sbi_flag(sbi, SBI_IS_RECOVERED);
785
786 return err;
787}
788
789static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
790{
791 struct list_head *head;
792 struct f2fs_orphan_block *orphan_blk = NULL;
793 unsigned int nentries = 0;
794 unsigned short index = 1;
795 unsigned short orphan_blocks;
796 struct page *page = NULL;
797 struct ino_entry *orphan = NULL;
798 struct inode_management *im = &sbi->im[ORPHAN_INO];
799
800 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
801
802 /*
803 * we don't need to do spin_lock(&im->ino_lock) here, since all the
804 * orphan inode operations are covered under f2fs_lock_op().
805 * And, spin_lock should be avoided due to page operations below.
806 */
807 head = &im->ino_list;
808
809 /* loop for each orphan inode entry and write them in journal block */
810 list_for_each_entry(orphan, head, list) {
811 if (!page) {
812 page = f2fs_grab_meta_page(sbi, start_blk++);
813 orphan_blk =
814 (struct f2fs_orphan_block *)page_address(page);
815 memset(orphan_blk, 0, sizeof(*orphan_blk));
816 }
817
818 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
819
820 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
821 /*
822 * an orphan block is full of 1020 entries,
823 * then we need to flush current orphan blocks
824 * and bring another one in memory
825 */
826 orphan_blk->blk_addr = cpu_to_le16(index);
827 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
828 orphan_blk->entry_count = cpu_to_le32(nentries);
829 set_page_dirty(page);
830 f2fs_put_page(page, 1);
831 index++;
832 nentries = 0;
833 page = NULL;
834 }
835 }
836
837 if (page) {
838 orphan_blk->blk_addr = cpu_to_le16(index);
839 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
840 orphan_blk->entry_count = cpu_to_le32(nentries);
841 set_page_dirty(page);
842 f2fs_put_page(page, 1);
843 }
844}
845
846static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
847 struct f2fs_checkpoint *ckpt)
848{
849 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
850 __u32 chksum;
851
852 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
853 if (chksum_ofs < CP_CHKSUM_OFFSET) {
854 chksum_ofs += sizeof(chksum);
855 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
856 F2FS_BLKSIZE - chksum_ofs);
857 }
858 return chksum;
859}
860
861static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
862 struct f2fs_checkpoint **cp_block, struct page **cp_page,
863 unsigned long long *version)
864{
865 size_t crc_offset = 0;
866 __u32 crc;
867
868 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
869 if (IS_ERR(*cp_page))
870 return PTR_ERR(*cp_page);
871
872 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
873
874 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
875 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
876 crc_offset > CP_CHKSUM_OFFSET) {
877 f2fs_put_page(*cp_page, 1);
878 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
879 return -EINVAL;
880 }
881
882 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
883 if (crc != cur_cp_crc(*cp_block)) {
884 f2fs_put_page(*cp_page, 1);
885 f2fs_warn(sbi, "invalid crc value");
886 return -EINVAL;
887 }
888
889 *version = cur_cp_version(*cp_block);
890 return 0;
891}
892
893static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
894 block_t cp_addr, unsigned long long *version)
895{
896 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
897 struct f2fs_checkpoint *cp_block = NULL;
898 unsigned long long cur_version = 0, pre_version = 0;
899 unsigned int cp_blocks;
900 int err;
901
902 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
903 &cp_page_1, version);
904 if (err)
905 return NULL;
906
907 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
908
909 if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
910 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
911 le32_to_cpu(cp_block->cp_pack_total_block_count));
912 goto invalid_cp;
913 }
914 pre_version = *version;
915
916 cp_addr += cp_blocks - 1;
917 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
918 &cp_page_2, version);
919 if (err)
920 goto invalid_cp;
921 cur_version = *version;
922
923 if (cur_version == pre_version) {
924 *version = cur_version;
925 f2fs_put_page(cp_page_2, 1);
926 return cp_page_1;
927 }
928 f2fs_put_page(cp_page_2, 1);
929invalid_cp:
930 f2fs_put_page(cp_page_1, 1);
931 return NULL;
932}
933
934int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
935{
936 struct f2fs_checkpoint *cp_block;
937 struct f2fs_super_block *fsb = sbi->raw_super;
938 struct page *cp1, *cp2, *cur_page;
939 unsigned long blk_size = sbi->blocksize;
940 unsigned long long cp1_version = 0, cp2_version = 0;
941 unsigned long long cp_start_blk_no;
942 unsigned int cp_blks = 1 + __cp_payload(sbi);
943 block_t cp_blk_no;
944 int i;
945 int err;
946
947 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
948 GFP_KERNEL);
949 if (!sbi->ckpt)
950 return -ENOMEM;
951 /*
952 * Finding out valid cp block involves read both
953 * sets( cp pack 1 and cp pack 2)
954 */
955 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
956 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
957
958 /* The second checkpoint pack should start at the next segment */
959 cp_start_blk_no += ((unsigned long long)1) <<
960 le32_to_cpu(fsb->log_blocks_per_seg);
961 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
962
963 if (cp1 && cp2) {
964 if (ver_after(cp2_version, cp1_version))
965 cur_page = cp2;
966 else
967 cur_page = cp1;
968 } else if (cp1) {
969 cur_page = cp1;
970 } else if (cp2) {
971 cur_page = cp2;
972 } else {
973 err = -EFSCORRUPTED;
974 goto fail_no_cp;
975 }
976
977 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
978 memcpy(sbi->ckpt, cp_block, blk_size);
979
980 if (cur_page == cp1)
981 sbi->cur_cp_pack = 1;
982 else
983 sbi->cur_cp_pack = 2;
984
985 /* Sanity checking of checkpoint */
986 if (f2fs_sanity_check_ckpt(sbi)) {
987 err = -EFSCORRUPTED;
988 goto free_fail_no_cp;
989 }
990
991 if (cp_blks <= 1)
992 goto done;
993
994 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
995 if (cur_page == cp2)
996 cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
997
998 for (i = 1; i < cp_blks; i++) {
999 void *sit_bitmap_ptr;
1000 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
1001
1002 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
1003 if (IS_ERR(cur_page)) {
1004 err = PTR_ERR(cur_page);
1005 goto free_fail_no_cp;
1006 }
1007 sit_bitmap_ptr = page_address(cur_page);
1008 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1009 f2fs_put_page(cur_page, 1);
1010 }
1011done:
1012 f2fs_put_page(cp1, 1);
1013 f2fs_put_page(cp2, 1);
1014 return 0;
1015
1016free_fail_no_cp:
1017 f2fs_put_page(cp1, 1);
1018 f2fs_put_page(cp2, 1);
1019fail_no_cp:
1020 kvfree(sbi->ckpt);
1021 return err;
1022}
1023
1024static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1025{
1026 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1027 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1028
1029 if (is_inode_flag_set(inode, flag))
1030 return;
1031
1032 set_inode_flag(inode, flag);
1033 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1034 stat_inc_dirty_inode(sbi, type);
1035}
1036
1037static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1038{
1039 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1040
1041 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1042 return;
1043
1044 list_del_init(&F2FS_I(inode)->dirty_list);
1045 clear_inode_flag(inode, flag);
1046 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1047}
1048
1049void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1050{
1051 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1052 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1053
1054 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1055 !S_ISLNK(inode->i_mode))
1056 return;
1057
1058 spin_lock(&sbi->inode_lock[type]);
1059 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1060 __add_dirty_inode(inode, type);
1061 inode_inc_dirty_pages(inode);
1062 spin_unlock(&sbi->inode_lock[type]);
1063
1064 set_page_private_reference(&folio->page);
1065}
1066
1067void f2fs_remove_dirty_inode(struct inode *inode)
1068{
1069 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1070 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1071
1072 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1073 !S_ISLNK(inode->i_mode))
1074 return;
1075
1076 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1077 return;
1078
1079 spin_lock(&sbi->inode_lock[type]);
1080 __remove_dirty_inode(inode, type);
1081 spin_unlock(&sbi->inode_lock[type]);
1082}
1083
1084int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1085 bool from_cp)
1086{
1087 struct list_head *head;
1088 struct inode *inode;
1089 struct f2fs_inode_info *fi;
1090 bool is_dir = (type == DIR_INODE);
1091 unsigned long ino = 0;
1092
1093 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1094 get_pages(sbi, is_dir ?
1095 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1096retry:
1097 if (unlikely(f2fs_cp_error(sbi))) {
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 -EIO;
1102 }
1103
1104 spin_lock(&sbi->inode_lock[type]);
1105
1106 head = &sbi->inode_list[type];
1107 if (list_empty(head)) {
1108 spin_unlock(&sbi->inode_lock[type]);
1109 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1110 get_pages(sbi, is_dir ?
1111 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1112 return 0;
1113 }
1114 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1115 inode = igrab(&fi->vfs_inode);
1116 spin_unlock(&sbi->inode_lock[type]);
1117 if (inode) {
1118 unsigned long cur_ino = inode->i_ino;
1119
1120 if (from_cp)
1121 F2FS_I(inode)->cp_task = current;
1122 F2FS_I(inode)->wb_task = current;
1123
1124 filemap_fdatawrite(inode->i_mapping);
1125
1126 F2FS_I(inode)->wb_task = NULL;
1127 if (from_cp)
1128 F2FS_I(inode)->cp_task = NULL;
1129
1130 iput(inode);
1131 /* We need to give cpu to another writers. */
1132 if (ino == cur_ino)
1133 cond_resched();
1134 else
1135 ino = cur_ino;
1136 } else {
1137 /*
1138 * We should submit bio, since it exists several
1139 * writebacking dentry pages in the freeing inode.
1140 */
1141 f2fs_submit_merged_write(sbi, DATA);
1142 cond_resched();
1143 }
1144 goto retry;
1145}
1146
1147static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1148{
1149 struct list_head *head = &sbi->inode_list[DIRTY_META];
1150 struct inode *inode;
1151 struct f2fs_inode_info *fi;
1152 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1153
1154 while (total--) {
1155 if (unlikely(f2fs_cp_error(sbi)))
1156 return -EIO;
1157
1158 spin_lock(&sbi->inode_lock[DIRTY_META]);
1159 if (list_empty(head)) {
1160 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1161 return 0;
1162 }
1163 fi = list_first_entry(head, struct f2fs_inode_info,
1164 gdirty_list);
1165 inode = igrab(&fi->vfs_inode);
1166 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1167 if (inode) {
1168 sync_inode_metadata(inode, 0);
1169
1170 /* it's on eviction */
1171 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1172 f2fs_update_inode_page(inode);
1173 iput(inode);
1174 }
1175 }
1176 return 0;
1177}
1178
1179static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1180{
1181 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1182 struct f2fs_nm_info *nm_i = NM_I(sbi);
1183 nid_t last_nid = nm_i->next_scan_nid;
1184
1185 next_free_nid(sbi, &last_nid);
1186 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1187 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1188 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1189 ckpt->next_free_nid = cpu_to_le32(last_nid);
1190
1191 /* update user_block_counts */
1192 sbi->last_valid_block_count = sbi->total_valid_block_count;
1193 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1194 percpu_counter_set(&sbi->rf_node_block_count, 0);
1195}
1196
1197static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1198{
1199 bool ret = false;
1200
1201 if (!is_journalled_quota(sbi))
1202 return false;
1203
1204 if (!f2fs_down_write_trylock(&sbi->quota_sem))
1205 return true;
1206 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1207 ret = false;
1208 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1209 ret = false;
1210 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1211 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1212 ret = true;
1213 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1214 ret = true;
1215 }
1216 f2fs_up_write(&sbi->quota_sem);
1217 return ret;
1218}
1219
1220/*
1221 * Freeze all the FS-operations for checkpoint.
1222 */
1223static int block_operations(struct f2fs_sb_info *sbi)
1224{
1225 struct writeback_control wbc = {
1226 .sync_mode = WB_SYNC_ALL,
1227 .nr_to_write = LONG_MAX,
1228 .for_reclaim = 0,
1229 };
1230 int err = 0, cnt = 0;
1231
1232 /*
1233 * Let's flush inline_data in dirty node pages.
1234 */
1235 f2fs_flush_inline_data(sbi);
1236
1237retry_flush_quotas:
1238 f2fs_lock_all(sbi);
1239 if (__need_flush_quota(sbi)) {
1240 int locked;
1241
1242 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1243 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1244 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1245 goto retry_flush_dents;
1246 }
1247 f2fs_unlock_all(sbi);
1248
1249 /* only failed during mount/umount/freeze/quotactl */
1250 locked = down_read_trylock(&sbi->sb->s_umount);
1251 f2fs_quota_sync(sbi->sb, -1);
1252 if (locked)
1253 up_read(&sbi->sb->s_umount);
1254 cond_resched();
1255 goto retry_flush_quotas;
1256 }
1257
1258retry_flush_dents:
1259 /* write all the dirty dentry pages */
1260 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1261 f2fs_unlock_all(sbi);
1262 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1263 if (err)
1264 return err;
1265 cond_resched();
1266 goto retry_flush_quotas;
1267 }
1268
1269 /*
1270 * POR: we should ensure that there are no dirty node pages
1271 * until finishing nat/sit flush. inode->i_blocks can be updated.
1272 */
1273 f2fs_down_write(&sbi->node_change);
1274
1275 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1276 f2fs_up_write(&sbi->node_change);
1277 f2fs_unlock_all(sbi);
1278 err = f2fs_sync_inode_meta(sbi);
1279 if (err)
1280 return err;
1281 cond_resched();
1282 goto retry_flush_quotas;
1283 }
1284
1285retry_flush_nodes:
1286 f2fs_down_write(&sbi->node_write);
1287
1288 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1289 f2fs_up_write(&sbi->node_write);
1290 atomic_inc(&sbi->wb_sync_req[NODE]);
1291 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1292 atomic_dec(&sbi->wb_sync_req[NODE]);
1293 if (err) {
1294 f2fs_up_write(&sbi->node_change);
1295 f2fs_unlock_all(sbi);
1296 return err;
1297 }
1298 cond_resched();
1299 goto retry_flush_nodes;
1300 }
1301
1302 /*
1303 * sbi->node_change is used only for AIO write_begin path which produces
1304 * dirty node blocks and some checkpoint values by block allocation.
1305 */
1306 __prepare_cp_block(sbi);
1307 f2fs_up_write(&sbi->node_change);
1308 return err;
1309}
1310
1311static void unblock_operations(struct f2fs_sb_info *sbi)
1312{
1313 f2fs_up_write(&sbi->node_write);
1314 f2fs_unlock_all(sbi);
1315}
1316
1317void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1318{
1319 DEFINE_WAIT(wait);
1320
1321 for (;;) {
1322 if (!get_pages(sbi, type))
1323 break;
1324
1325 if (unlikely(f2fs_cp_error(sbi) &&
1326 !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1327 break;
1328
1329 if (type == F2FS_DIRTY_META)
1330 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1331 FS_CP_META_IO);
1332 else if (type == F2FS_WB_CP_DATA)
1333 f2fs_submit_merged_write(sbi, DATA);
1334
1335 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1336 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1337 }
1338 finish_wait(&sbi->cp_wait, &wait);
1339}
1340
1341static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1342{
1343 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1344 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1345 unsigned long flags;
1346
1347 if (cpc->reason & CP_UMOUNT) {
1348 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1349 NM_I(sbi)->nat_bits_blocks > BLKS_PER_SEG(sbi)) {
1350 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1351 f2fs_notice(sbi, "Disable nat_bits due to no space");
1352 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1353 f2fs_nat_bitmap_enabled(sbi)) {
1354 f2fs_enable_nat_bits(sbi);
1355 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1356 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1357 }
1358 }
1359
1360 spin_lock_irqsave(&sbi->cp_lock, flags);
1361
1362 if (cpc->reason & CP_TRIMMED)
1363 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1364 else
1365 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1366
1367 if (cpc->reason & CP_UMOUNT)
1368 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1369 else
1370 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1371
1372 if (cpc->reason & CP_FASTBOOT)
1373 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1374 else
1375 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1376
1377 if (orphan_num)
1378 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1379 else
1380 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1381
1382 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1383 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1384
1385 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1386 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1387 else
1388 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1389
1390 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1391 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1392 else
1393 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1394
1395 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1396 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1397 else
1398 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1399
1400 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1401 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1402 else
1403 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1404
1405 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1406 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1407
1408 /* set this flag to activate crc|cp_ver for recovery */
1409 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1410 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1411
1412 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1413}
1414
1415static void commit_checkpoint(struct f2fs_sb_info *sbi,
1416 void *src, block_t blk_addr)
1417{
1418 struct writeback_control wbc = {
1419 .for_reclaim = 0,
1420 };
1421
1422 /*
1423 * filemap_get_folios_tag and lock_page again will take
1424 * some extra time. Therefore, f2fs_update_meta_pages and
1425 * f2fs_sync_meta_pages are combined in this function.
1426 */
1427 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1428 int err;
1429
1430 f2fs_wait_on_page_writeback(page, META, true, true);
1431
1432 memcpy(page_address(page), src, PAGE_SIZE);
1433
1434 set_page_dirty(page);
1435 if (unlikely(!clear_page_dirty_for_io(page)))
1436 f2fs_bug_on(sbi, 1);
1437
1438 /* writeout cp pack 2 page */
1439 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1440 if (unlikely(err && f2fs_cp_error(sbi))) {
1441 f2fs_put_page(page, 1);
1442 return;
1443 }
1444
1445 f2fs_bug_on(sbi, err);
1446 f2fs_put_page(page, 0);
1447
1448 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1449 f2fs_submit_merged_write(sbi, META_FLUSH);
1450}
1451
1452static inline u64 get_sectors_written(struct block_device *bdev)
1453{
1454 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1455}
1456
1457u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1458{
1459 if (f2fs_is_multi_device(sbi)) {
1460 u64 sectors = 0;
1461 int i;
1462
1463 for (i = 0; i < sbi->s_ndevs; i++)
1464 sectors += get_sectors_written(FDEV(i).bdev);
1465
1466 return sectors;
1467 }
1468
1469 return get_sectors_written(sbi->sb->s_bdev);
1470}
1471
1472static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1473{
1474 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1475 struct f2fs_nm_info *nm_i = NM_I(sbi);
1476 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1477 block_t start_blk;
1478 unsigned int data_sum_blocks, orphan_blocks;
1479 __u32 crc32 = 0;
1480 int i;
1481 int cp_payload_blks = __cp_payload(sbi);
1482 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1483 u64 kbytes_written;
1484 int err;
1485
1486 /* Flush all the NAT/SIT pages */
1487 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1488
1489 /* start to update checkpoint, cp ver is already updated previously */
1490 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1491 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1492 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1493 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1494
1495 ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1496 ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1497 ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1498 }
1499 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1500 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1501
1502 ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1503 ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1504 ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1505 }
1506
1507 /* 2 cp + n data seg summary + orphan inode blocks */
1508 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1509 spin_lock_irqsave(&sbi->cp_lock, flags);
1510 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1511 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1512 else
1513 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1514 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1515
1516 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1517 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1518 orphan_blocks);
1519
1520 if (__remain_node_summaries(cpc->reason))
1521 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1522 cp_payload_blks + data_sum_blocks +
1523 orphan_blocks + NR_CURSEG_NODE_TYPE);
1524 else
1525 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1526 cp_payload_blks + data_sum_blocks +
1527 orphan_blocks);
1528
1529 /* update ckpt flag for checkpoint */
1530 update_ckpt_flags(sbi, cpc);
1531
1532 /* update SIT/NAT bitmap */
1533 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1534 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1535
1536 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1537 *((__le32 *)((unsigned char *)ckpt +
1538 le32_to_cpu(ckpt->checksum_offset)))
1539 = cpu_to_le32(crc32);
1540
1541 start_blk = __start_cp_next_addr(sbi);
1542
1543 /* write nat bits */
1544 if ((cpc->reason & CP_UMOUNT) &&
1545 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1546 __u64 cp_ver = cur_cp_version(ckpt);
1547 block_t blk;
1548
1549 cp_ver |= ((__u64)crc32 << 32);
1550 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1551
1552 blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1553 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1554 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1555 F2FS_BLK_TO_BYTES(i), blk + i);
1556 }
1557
1558 /* write out checkpoint buffer at block 0 */
1559 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1560
1561 for (i = 1; i < 1 + cp_payload_blks; i++)
1562 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1563 start_blk++);
1564
1565 if (orphan_num) {
1566 write_orphan_inodes(sbi, start_blk);
1567 start_blk += orphan_blocks;
1568 }
1569
1570 f2fs_write_data_summaries(sbi, start_blk);
1571 start_blk += data_sum_blocks;
1572
1573 /* Record write statistics in the hot node summary */
1574 kbytes_written = sbi->kbytes_written;
1575 kbytes_written += (f2fs_get_sectors_written(sbi) -
1576 sbi->sectors_written_start) >> 1;
1577 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1578
1579 if (__remain_node_summaries(cpc->reason)) {
1580 f2fs_write_node_summaries(sbi, start_blk);
1581 start_blk += NR_CURSEG_NODE_TYPE;
1582 }
1583
1584 /* Here, we have one bio having CP pack except cp pack 2 page */
1585 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1586 /* Wait for all dirty meta pages to be submitted for IO */
1587 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1588
1589 /* wait for previous submitted meta pages writeback */
1590 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1591
1592 /* flush all device cache */
1593 err = f2fs_flush_device_cache(sbi);
1594 if (err)
1595 return err;
1596
1597 /* barrier and flush checkpoint cp pack 2 page if it can */
1598 commit_checkpoint(sbi, ckpt, start_blk);
1599 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1600
1601 /*
1602 * invalidate intermediate page cache borrowed from meta inode which are
1603 * used for migration of encrypted, verity or compressed inode's blocks.
1604 */
1605 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1606 f2fs_sb_has_compression(sbi))
1607 f2fs_bug_on(sbi,
1608 invalidate_inode_pages2_range(META_MAPPING(sbi),
1609 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1610
1611 f2fs_release_ino_entry(sbi, false);
1612
1613 f2fs_reset_fsync_node_info(sbi);
1614
1615 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1616 clear_sbi_flag(sbi, SBI_NEED_CP);
1617 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1618
1619 spin_lock(&sbi->stat_lock);
1620 sbi->unusable_block_count = 0;
1621 spin_unlock(&sbi->stat_lock);
1622
1623 __set_cp_next_pack(sbi);
1624
1625 /*
1626 * redirty superblock if metadata like node page or inode cache is
1627 * updated during writing checkpoint.
1628 */
1629 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1630 get_pages(sbi, F2FS_DIRTY_IMETA))
1631 set_sbi_flag(sbi, SBI_IS_DIRTY);
1632
1633 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1634
1635 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1636}
1637
1638int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1639{
1640 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1641 unsigned long long ckpt_ver;
1642 int err = 0;
1643
1644 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1645 return -EROFS;
1646
1647 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1648 if (cpc->reason != CP_PAUSE)
1649 return 0;
1650 f2fs_warn(sbi, "Start checkpoint disabled!");
1651 }
1652 if (cpc->reason != CP_RESIZE)
1653 f2fs_down_write(&sbi->cp_global_sem);
1654
1655 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1656 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1657 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1658 goto out;
1659 if (unlikely(f2fs_cp_error(sbi))) {
1660 err = -EIO;
1661 goto out;
1662 }
1663
1664 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1665
1666 err = block_operations(sbi);
1667 if (err)
1668 goto out;
1669
1670 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1671
1672 f2fs_flush_merged_writes(sbi);
1673
1674 /* this is the case of multiple fstrims without any changes */
1675 if (cpc->reason & CP_DISCARD) {
1676 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1677 unblock_operations(sbi);
1678 goto out;
1679 }
1680
1681 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1682 SIT_I(sbi)->dirty_sentries == 0 &&
1683 prefree_segments(sbi) == 0) {
1684 f2fs_flush_sit_entries(sbi, cpc);
1685 f2fs_clear_prefree_segments(sbi, cpc);
1686 unblock_operations(sbi);
1687 goto out;
1688 }
1689 }
1690
1691 /*
1692 * update checkpoint pack index
1693 * Increase the version number so that
1694 * SIT entries and seg summaries are written at correct place
1695 */
1696 ckpt_ver = cur_cp_version(ckpt);
1697 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1698
1699 /* write cached NAT/SIT entries to NAT/SIT area */
1700 err = f2fs_flush_nat_entries(sbi, cpc);
1701 if (err) {
1702 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1703 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1704 goto stop;
1705 }
1706
1707 f2fs_flush_sit_entries(sbi, cpc);
1708
1709 /* save inmem log status */
1710 f2fs_save_inmem_curseg(sbi);
1711
1712 err = do_checkpoint(sbi, cpc);
1713 if (err) {
1714 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1715 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1716 f2fs_release_discard_addrs(sbi);
1717 } else {
1718 f2fs_clear_prefree_segments(sbi, cpc);
1719 }
1720
1721 f2fs_restore_inmem_curseg(sbi);
1722 f2fs_reinit_atgc_curseg(sbi);
1723 stat_inc_cp_count(sbi);
1724stop:
1725 unblock_operations(sbi);
1726
1727 if (cpc->reason & CP_RECOVERY)
1728 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1729
1730 /* update CP_TIME to trigger checkpoint periodically */
1731 f2fs_update_time(sbi, CP_TIME);
1732 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1733out:
1734 if (cpc->reason != CP_RESIZE)
1735 f2fs_up_write(&sbi->cp_global_sem);
1736 return err;
1737}
1738
1739void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1740{
1741 int i;
1742
1743 for (i = 0; i < MAX_INO_ENTRY; i++) {
1744 struct inode_management *im = &sbi->im[i];
1745
1746 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1747 spin_lock_init(&im->ino_lock);
1748 INIT_LIST_HEAD(&im->ino_list);
1749 im->ino_num = 0;
1750 }
1751
1752 sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1753 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1754 F2FS_ORPHANS_PER_BLOCK;
1755}
1756
1757int __init f2fs_create_checkpoint_caches(void)
1758{
1759 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1760 sizeof(struct ino_entry));
1761 if (!ino_entry_slab)
1762 return -ENOMEM;
1763 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1764 sizeof(struct inode_entry));
1765 if (!f2fs_inode_entry_slab) {
1766 kmem_cache_destroy(ino_entry_slab);
1767 return -ENOMEM;
1768 }
1769 return 0;
1770}
1771
1772void f2fs_destroy_checkpoint_caches(void)
1773{
1774 kmem_cache_destroy(ino_entry_slab);
1775 kmem_cache_destroy(f2fs_inode_entry_slab);
1776}
1777
1778static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1779{
1780 struct cp_control cpc = { .reason = CP_SYNC, };
1781 int err;
1782
1783 f2fs_down_write(&sbi->gc_lock);
1784 err = f2fs_write_checkpoint(sbi, &cpc);
1785 f2fs_up_write(&sbi->gc_lock);
1786
1787 return err;
1788}
1789
1790static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1791{
1792 struct ckpt_req_control *cprc = &sbi->cprc_info;
1793 struct ckpt_req *req, *next;
1794 struct llist_node *dispatch_list;
1795 u64 sum_diff = 0, diff, count = 0;
1796 int ret;
1797
1798 dispatch_list = llist_del_all(&cprc->issue_list);
1799 if (!dispatch_list)
1800 return;
1801 dispatch_list = llist_reverse_order(dispatch_list);
1802
1803 ret = __write_checkpoint_sync(sbi);
1804 atomic_inc(&cprc->issued_ckpt);
1805
1806 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1807 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1808 req->ret = ret;
1809 complete(&req->wait);
1810
1811 sum_diff += diff;
1812 count++;
1813 }
1814 atomic_sub(count, &cprc->queued_ckpt);
1815 atomic_add(count, &cprc->total_ckpt);
1816
1817 spin_lock(&cprc->stat_lock);
1818 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1819 if (cprc->peak_time < cprc->cur_time)
1820 cprc->peak_time = cprc->cur_time;
1821 spin_unlock(&cprc->stat_lock);
1822}
1823
1824static int issue_checkpoint_thread(void *data)
1825{
1826 struct f2fs_sb_info *sbi = data;
1827 struct ckpt_req_control *cprc = &sbi->cprc_info;
1828 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1829repeat:
1830 if (kthread_should_stop())
1831 return 0;
1832
1833 if (!llist_empty(&cprc->issue_list))
1834 __checkpoint_and_complete_reqs(sbi);
1835
1836 wait_event_interruptible(*q,
1837 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1838 goto repeat;
1839}
1840
1841static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1842 struct ckpt_req *wait_req)
1843{
1844 struct ckpt_req_control *cprc = &sbi->cprc_info;
1845
1846 if (!llist_empty(&cprc->issue_list)) {
1847 __checkpoint_and_complete_reqs(sbi);
1848 } else {
1849 /* already dispatched by issue_checkpoint_thread */
1850 if (wait_req)
1851 wait_for_completion(&wait_req->wait);
1852 }
1853}
1854
1855static void init_ckpt_req(struct ckpt_req *req)
1856{
1857 memset(req, 0, sizeof(struct ckpt_req));
1858
1859 init_completion(&req->wait);
1860 req->queue_time = ktime_get();
1861}
1862
1863int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1864{
1865 struct ckpt_req_control *cprc = &sbi->cprc_info;
1866 struct ckpt_req req;
1867 struct cp_control cpc;
1868
1869 cpc.reason = __get_cp_reason(sbi);
1870 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1871 int ret;
1872
1873 f2fs_down_write(&sbi->gc_lock);
1874 ret = f2fs_write_checkpoint(sbi, &cpc);
1875 f2fs_up_write(&sbi->gc_lock);
1876
1877 return ret;
1878 }
1879
1880 if (!cprc->f2fs_issue_ckpt)
1881 return __write_checkpoint_sync(sbi);
1882
1883 init_ckpt_req(&req);
1884
1885 llist_add(&req.llnode, &cprc->issue_list);
1886 atomic_inc(&cprc->queued_ckpt);
1887
1888 /*
1889 * update issue_list before we wake up issue_checkpoint thread,
1890 * this smp_mb() pairs with another barrier in ___wait_event(),
1891 * see more details in comments of waitqueue_active().
1892 */
1893 smp_mb();
1894
1895 if (waitqueue_active(&cprc->ckpt_wait_queue))
1896 wake_up(&cprc->ckpt_wait_queue);
1897
1898 if (cprc->f2fs_issue_ckpt)
1899 wait_for_completion(&req.wait);
1900 else
1901 flush_remained_ckpt_reqs(sbi, &req);
1902
1903 return req.ret;
1904}
1905
1906int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1907{
1908 dev_t dev = sbi->sb->s_bdev->bd_dev;
1909 struct ckpt_req_control *cprc = &sbi->cprc_info;
1910
1911 if (cprc->f2fs_issue_ckpt)
1912 return 0;
1913
1914 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1915 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1916 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1917 int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1918
1919 cprc->f2fs_issue_ckpt = NULL;
1920 return err;
1921 }
1922
1923 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1924
1925 return 0;
1926}
1927
1928void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1929{
1930 struct ckpt_req_control *cprc = &sbi->cprc_info;
1931 struct task_struct *ckpt_task;
1932
1933 if (!cprc->f2fs_issue_ckpt)
1934 return;
1935
1936 ckpt_task = cprc->f2fs_issue_ckpt;
1937 cprc->f2fs_issue_ckpt = NULL;
1938 kthread_stop(ckpt_task);
1939
1940 f2fs_flush_ckpt_thread(sbi);
1941}
1942
1943void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1944{
1945 struct ckpt_req_control *cprc = &sbi->cprc_info;
1946
1947 flush_remained_ckpt_reqs(sbi, NULL);
1948
1949 /* Let's wait for the previous dispatched checkpoint. */
1950 while (atomic_read(&cprc->queued_ckpt))
1951 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1952}
1953
1954void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1955{
1956 struct ckpt_req_control *cprc = &sbi->cprc_info;
1957
1958 atomic_set(&cprc->issued_ckpt, 0);
1959 atomic_set(&cprc->total_ckpt, 0);
1960 atomic_set(&cprc->queued_ckpt, 0);
1961 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1962 init_waitqueue_head(&cprc->ckpt_wait_queue);
1963 init_llist_head(&cprc->issue_list);
1964 spin_lock_init(&cprc->stat_lock);
1965}