xref: /linux/fs/f2fs/checkpoint.c (revision 07f0148aafe8c95a3a76cd59e9e75b4d78d1d31d)
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 
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *f2fs_inode_entry_slab;
28 
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io,
30 						unsigned char reason)
31 {
32 	f2fs_build_fault_attr(sbi, 0, 0);
33 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
34 	if (!end_io) {
35 		f2fs_flush_merged_writes(sbi);
36 
37 		f2fs_handle_stop(sbi, reason);
38 	}
39 }
40 
41 /*
42  * We guarantee no failure on the returned page.
43  */
44 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
45 {
46 	struct address_space *mapping = META_MAPPING(sbi);
47 	struct page *page;
48 repeat:
49 	page = f2fs_grab_cache_page(mapping, index, false);
50 	if (!page) {
51 		cond_resched();
52 		goto repeat;
53 	}
54 	f2fs_wait_on_page_writeback(page, META, true, true);
55 	if (!PageUptodate(page))
56 		SetPageUptodate(page);
57 	return page;
58 }
59 
60 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
61 							bool is_meta)
62 {
63 	struct address_space *mapping = META_MAPPING(sbi);
64 	struct page *page;
65 	struct f2fs_io_info fio = {
66 		.sbi = sbi,
67 		.type = META,
68 		.op = REQ_OP_READ,
69 		.op_flags = REQ_META | REQ_PRIO,
70 		.old_blkaddr = index,
71 		.new_blkaddr = index,
72 		.encrypted_page = NULL,
73 		.is_por = !is_meta,
74 	};
75 	int err;
76 
77 	if (unlikely(!is_meta))
78 		fio.op_flags &= ~REQ_META;
79 repeat:
80 	page = f2fs_grab_cache_page(mapping, index, false);
81 	if (!page) {
82 		cond_resched();
83 		goto repeat;
84 	}
85 	if (PageUptodate(page))
86 		goto out;
87 
88 	fio.page = page;
89 
90 	err = f2fs_submit_page_bio(&fio);
91 	if (err) {
92 		f2fs_put_page(page, 1);
93 		return ERR_PTR(err);
94 	}
95 
96 	f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
97 
98 	lock_page(page);
99 	if (unlikely(page->mapping != mapping)) {
100 		f2fs_put_page(page, 1);
101 		goto repeat;
102 	}
103 
104 	if (unlikely(!PageUptodate(page))) {
105 		f2fs_handle_page_eio(sbi, page->index, META);
106 		f2fs_put_page(page, 1);
107 		return ERR_PTR(-EIO);
108 	}
109 out:
110 	return page;
111 }
112 
113 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
114 {
115 	return __get_meta_page(sbi, index, true);
116 }
117 
118 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
119 {
120 	struct page *page;
121 	int count = 0;
122 
123 retry:
124 	page = __get_meta_page(sbi, index, true);
125 	if (IS_ERR(page)) {
126 		if (PTR_ERR(page) == -EIO &&
127 				++count <= DEFAULT_RETRY_IO_COUNT)
128 			goto retry;
129 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
130 	}
131 	return page;
132 }
133 
134 /* for POR only */
135 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
136 {
137 	return __get_meta_page(sbi, index, false);
138 }
139 
140 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
141 							int type)
142 {
143 	struct seg_entry *se;
144 	unsigned int segno, offset;
145 	bool exist;
146 
147 	if (type == DATA_GENERIC)
148 		return true;
149 
150 	segno = GET_SEGNO(sbi, blkaddr);
151 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
152 	se = get_seg_entry(sbi, segno);
153 
154 	exist = f2fs_test_bit(offset, se->cur_valid_map);
155 	if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
156 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
157 			 blkaddr, exist);
158 		set_sbi_flag(sbi, SBI_NEED_FSCK);
159 		return exist;
160 	}
161 
162 	if (!exist && type == DATA_GENERIC_ENHANCE) {
163 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
164 			 blkaddr, exist);
165 		set_sbi_flag(sbi, SBI_NEED_FSCK);
166 		dump_stack();
167 	}
168 	return exist;
169 }
170 
171 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
172 					block_t blkaddr, int type)
173 {
174 	if (time_to_inject(sbi, FAULT_BLKADDR)) {
175 		f2fs_show_injection_info(sbi, FAULT_BLKADDR);
176 		return false;
177 	}
178 
179 	switch (type) {
180 	case META_NAT:
181 		break;
182 	case META_SIT:
183 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
184 			return false;
185 		break;
186 	case META_SSA:
187 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
188 			blkaddr < SM_I(sbi)->ssa_blkaddr))
189 			return false;
190 		break;
191 	case META_CP:
192 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
193 			blkaddr < __start_cp_addr(sbi)))
194 			return false;
195 		break;
196 	case META_POR:
197 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
198 			blkaddr < MAIN_BLKADDR(sbi)))
199 			return false;
200 		break;
201 	case DATA_GENERIC:
202 	case DATA_GENERIC_ENHANCE:
203 	case DATA_GENERIC_ENHANCE_READ:
204 	case DATA_GENERIC_ENHANCE_UPDATE:
205 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
206 				blkaddr < MAIN_BLKADDR(sbi))) {
207 			f2fs_warn(sbi, "access invalid blkaddr:%u",
208 				  blkaddr);
209 			set_sbi_flag(sbi, SBI_NEED_FSCK);
210 			dump_stack();
211 			return false;
212 		} else {
213 			return __is_bitmap_valid(sbi, blkaddr, type);
214 		}
215 		break;
216 	case META_GENERIC:
217 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
218 			blkaddr >= MAIN_BLKADDR(sbi)))
219 			return false;
220 		break;
221 	default:
222 		BUG();
223 	}
224 
225 	return true;
226 }
227 
228 /*
229  * Readahead CP/NAT/SIT/SSA/POR pages
230  */
231 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
232 							int type, bool sync)
233 {
234 	struct page *page;
235 	block_t blkno = start;
236 	struct f2fs_io_info fio = {
237 		.sbi = sbi,
238 		.type = META,
239 		.op = REQ_OP_READ,
240 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
241 		.encrypted_page = NULL,
242 		.in_list = false,
243 		.is_por = (type == META_POR),
244 	};
245 	struct blk_plug plug;
246 	int err;
247 
248 	if (unlikely(type == META_POR))
249 		fio.op_flags &= ~REQ_META;
250 
251 	blk_start_plug(&plug);
252 	for (; nrpages-- > 0; blkno++) {
253 
254 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
255 			goto out;
256 
257 		switch (type) {
258 		case META_NAT:
259 			if (unlikely(blkno >=
260 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
261 				blkno = 0;
262 			/* get nat block addr */
263 			fio.new_blkaddr = current_nat_addr(sbi,
264 					blkno * NAT_ENTRY_PER_BLOCK);
265 			break;
266 		case META_SIT:
267 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
268 				goto out;
269 			/* get sit block addr */
270 			fio.new_blkaddr = current_sit_addr(sbi,
271 					blkno * SIT_ENTRY_PER_BLOCK);
272 			break;
273 		case META_SSA:
274 		case META_CP:
275 		case META_POR:
276 			fio.new_blkaddr = blkno;
277 			break;
278 		default:
279 			BUG();
280 		}
281 
282 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
283 						fio.new_blkaddr, false);
284 		if (!page)
285 			continue;
286 		if (PageUptodate(page)) {
287 			f2fs_put_page(page, 1);
288 			continue;
289 		}
290 
291 		fio.page = page;
292 		err = f2fs_submit_page_bio(&fio);
293 		f2fs_put_page(page, err ? 1 : 0);
294 
295 		if (!err)
296 			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
297 							F2FS_BLKSIZE);
298 	}
299 out:
300 	blk_finish_plug(&plug);
301 	return blkno - start;
302 }
303 
304 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
305 							unsigned int ra_blocks)
306 {
307 	struct page *page;
308 	bool readahead = false;
309 
310 	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
311 		return;
312 
313 	page = find_get_page(META_MAPPING(sbi), index);
314 	if (!page || !PageUptodate(page))
315 		readahead = true;
316 	f2fs_put_page(page, 0);
317 
318 	if (readahead)
319 		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
320 }
321 
322 static int __f2fs_write_meta_page(struct page *page,
323 				struct writeback_control *wbc,
324 				enum iostat_type io_type)
325 {
326 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
327 
328 	trace_f2fs_writepage(page, META);
329 
330 	if (unlikely(f2fs_cp_error(sbi)))
331 		goto redirty_out;
332 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
333 		goto redirty_out;
334 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
335 		goto redirty_out;
336 
337 	f2fs_do_write_meta_page(sbi, page, io_type);
338 	dec_page_count(sbi, F2FS_DIRTY_META);
339 
340 	if (wbc->for_reclaim)
341 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
342 
343 	unlock_page(page);
344 
345 	if (unlikely(f2fs_cp_error(sbi)))
346 		f2fs_submit_merged_write(sbi, META);
347 
348 	return 0;
349 
350 redirty_out:
351 	redirty_page_for_writepage(wbc, page);
352 	return AOP_WRITEPAGE_ACTIVATE;
353 }
354 
355 static int f2fs_write_meta_page(struct page *page,
356 				struct writeback_control *wbc)
357 {
358 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
359 }
360 
361 static int f2fs_write_meta_pages(struct address_space *mapping,
362 				struct writeback_control *wbc)
363 {
364 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
365 	long diff, written;
366 
367 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
368 		goto skip_write;
369 
370 	/* collect a number of dirty meta pages and write together */
371 	if (wbc->sync_mode != WB_SYNC_ALL &&
372 			get_pages(sbi, F2FS_DIRTY_META) <
373 					nr_pages_to_skip(sbi, META))
374 		goto skip_write;
375 
376 	/* if locked failed, cp will flush dirty pages instead */
377 	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
378 		goto skip_write;
379 
380 	trace_f2fs_writepages(mapping->host, wbc, META);
381 	diff = nr_pages_to_write(sbi, META, wbc);
382 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
383 	f2fs_up_write(&sbi->cp_global_sem);
384 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
385 	return 0;
386 
387 skip_write:
388 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
389 	trace_f2fs_writepages(mapping->host, wbc, META);
390 	return 0;
391 }
392 
393 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
394 				long nr_to_write, enum iostat_type io_type)
395 {
396 	struct address_space *mapping = META_MAPPING(sbi);
397 	pgoff_t index = 0, prev = ULONG_MAX;
398 	struct pagevec pvec;
399 	long nwritten = 0;
400 	int nr_pages;
401 	struct writeback_control wbc = {
402 		.for_reclaim = 0,
403 	};
404 	struct blk_plug plug;
405 
406 	pagevec_init(&pvec);
407 
408 	blk_start_plug(&plug);
409 
410 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
411 				PAGECACHE_TAG_DIRTY))) {
412 		int i;
413 
414 		for (i = 0; i < nr_pages; i++) {
415 			struct page *page = pvec.pages[i];
416 
417 			if (prev == ULONG_MAX)
418 				prev = page->index - 1;
419 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
420 				pagevec_release(&pvec);
421 				goto stop;
422 			}
423 
424 			lock_page(page);
425 
426 			if (unlikely(page->mapping != mapping)) {
427 continue_unlock:
428 				unlock_page(page);
429 				continue;
430 			}
431 			if (!PageDirty(page)) {
432 				/* someone wrote it for us */
433 				goto continue_unlock;
434 			}
435 
436 			f2fs_wait_on_page_writeback(page, META, true, true);
437 
438 			if (!clear_page_dirty_for_io(page))
439 				goto continue_unlock;
440 
441 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
442 				unlock_page(page);
443 				break;
444 			}
445 			nwritten++;
446 			prev = page->index;
447 			if (unlikely(nwritten >= nr_to_write))
448 				break;
449 		}
450 		pagevec_release(&pvec);
451 		cond_resched();
452 	}
453 stop:
454 	if (nwritten)
455 		f2fs_submit_merged_write(sbi, type);
456 
457 	blk_finish_plug(&plug);
458 
459 	return nwritten;
460 }
461 
462 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
463 		struct folio *folio)
464 {
465 	trace_f2fs_set_page_dirty(&folio->page, META);
466 
467 	if (!folio_test_uptodate(folio))
468 		folio_mark_uptodate(folio);
469 	if (filemap_dirty_folio(mapping, folio)) {
470 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
471 		set_page_private_reference(&folio->page);
472 		return true;
473 	}
474 	return false;
475 }
476 
477 const struct address_space_operations f2fs_meta_aops = {
478 	.writepage	= f2fs_write_meta_page,
479 	.writepages	= f2fs_write_meta_pages,
480 	.dirty_folio	= f2fs_dirty_meta_folio,
481 	.invalidate_folio = f2fs_invalidate_folio,
482 	.release_folio	= f2fs_release_folio,
483 	.migrate_folio	= filemap_migrate_folio,
484 };
485 
486 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
487 						unsigned int devidx, int type)
488 {
489 	struct inode_management *im = &sbi->im[type];
490 	struct ino_entry *e = NULL, *new = NULL;
491 
492 	if (type == FLUSH_INO) {
493 		rcu_read_lock();
494 		e = radix_tree_lookup(&im->ino_root, ino);
495 		rcu_read_unlock();
496 	}
497 
498 retry:
499 	if (!e)
500 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
501 						GFP_NOFS, true, NULL);
502 
503 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
504 
505 	spin_lock(&im->ino_lock);
506 	e = radix_tree_lookup(&im->ino_root, ino);
507 	if (!e) {
508 		if (!new) {
509 			spin_unlock(&im->ino_lock);
510 			goto retry;
511 		}
512 		e = new;
513 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
514 			f2fs_bug_on(sbi, 1);
515 
516 		memset(e, 0, sizeof(struct ino_entry));
517 		e->ino = ino;
518 
519 		list_add_tail(&e->list, &im->ino_list);
520 		if (type != ORPHAN_INO)
521 			im->ino_num++;
522 	}
523 
524 	if (type == FLUSH_INO)
525 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
526 
527 	spin_unlock(&im->ino_lock);
528 	radix_tree_preload_end();
529 
530 	if (new && e != new)
531 		kmem_cache_free(ino_entry_slab, new);
532 }
533 
534 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
535 {
536 	struct inode_management *im = &sbi->im[type];
537 	struct ino_entry *e;
538 
539 	spin_lock(&im->ino_lock);
540 	e = radix_tree_lookup(&im->ino_root, ino);
541 	if (e) {
542 		list_del(&e->list);
543 		radix_tree_delete(&im->ino_root, ino);
544 		im->ino_num--;
545 		spin_unlock(&im->ino_lock);
546 		kmem_cache_free(ino_entry_slab, e);
547 		return;
548 	}
549 	spin_unlock(&im->ino_lock);
550 }
551 
552 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
553 {
554 	/* add new dirty ino entry into list */
555 	__add_ino_entry(sbi, ino, 0, type);
556 }
557 
558 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
559 {
560 	/* remove dirty ino entry from list */
561 	__remove_ino_entry(sbi, ino, type);
562 }
563 
564 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
565 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
566 {
567 	struct inode_management *im = &sbi->im[mode];
568 	struct ino_entry *e;
569 
570 	spin_lock(&im->ino_lock);
571 	e = radix_tree_lookup(&im->ino_root, ino);
572 	spin_unlock(&im->ino_lock);
573 	return e ? true : false;
574 }
575 
576 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
577 {
578 	struct ino_entry *e, *tmp;
579 	int i;
580 
581 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
582 		struct inode_management *im = &sbi->im[i];
583 
584 		spin_lock(&im->ino_lock);
585 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
586 			list_del(&e->list);
587 			radix_tree_delete(&im->ino_root, e->ino);
588 			kmem_cache_free(ino_entry_slab, e);
589 			im->ino_num--;
590 		}
591 		spin_unlock(&im->ino_lock);
592 	}
593 }
594 
595 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
596 					unsigned int devidx, int type)
597 {
598 	__add_ino_entry(sbi, ino, devidx, type);
599 }
600 
601 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
602 					unsigned int devidx, int type)
603 {
604 	struct inode_management *im = &sbi->im[type];
605 	struct ino_entry *e;
606 	bool is_dirty = false;
607 
608 	spin_lock(&im->ino_lock);
609 	e = radix_tree_lookup(&im->ino_root, ino);
610 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
611 		is_dirty = true;
612 	spin_unlock(&im->ino_lock);
613 	return is_dirty;
614 }
615 
616 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
617 {
618 	struct inode_management *im = &sbi->im[ORPHAN_INO];
619 	int err = 0;
620 
621 	spin_lock(&im->ino_lock);
622 
623 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
624 		spin_unlock(&im->ino_lock);
625 		f2fs_show_injection_info(sbi, FAULT_ORPHAN);
626 		return -ENOSPC;
627 	}
628 
629 	if (unlikely(im->ino_num >= sbi->max_orphans))
630 		err = -ENOSPC;
631 	else
632 		im->ino_num++;
633 	spin_unlock(&im->ino_lock);
634 
635 	return err;
636 }
637 
638 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
639 {
640 	struct inode_management *im = &sbi->im[ORPHAN_INO];
641 
642 	spin_lock(&im->ino_lock);
643 	f2fs_bug_on(sbi, im->ino_num == 0);
644 	im->ino_num--;
645 	spin_unlock(&im->ino_lock);
646 }
647 
648 void f2fs_add_orphan_inode(struct inode *inode)
649 {
650 	/* add new orphan ino entry into list */
651 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
652 	f2fs_update_inode_page(inode);
653 }
654 
655 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
656 {
657 	/* remove orphan entry from orphan list */
658 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
659 }
660 
661 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
662 {
663 	struct inode *inode;
664 	struct node_info ni;
665 	int err;
666 
667 	inode = f2fs_iget_retry(sbi->sb, ino);
668 	if (IS_ERR(inode)) {
669 		/*
670 		 * there should be a bug that we can't find the entry
671 		 * to orphan inode.
672 		 */
673 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
674 		return PTR_ERR(inode);
675 	}
676 
677 	err = f2fs_dquot_initialize(inode);
678 	if (err) {
679 		iput(inode);
680 		goto err_out;
681 	}
682 
683 	clear_nlink(inode);
684 
685 	/* truncate all the data during iput */
686 	iput(inode);
687 
688 	err = f2fs_get_node_info(sbi, ino, &ni, false);
689 	if (err)
690 		goto err_out;
691 
692 	/* ENOMEM was fully retried in f2fs_evict_inode. */
693 	if (ni.blk_addr != NULL_ADDR) {
694 		err = -EIO;
695 		goto err_out;
696 	}
697 	return 0;
698 
699 err_out:
700 	set_sbi_flag(sbi, SBI_NEED_FSCK);
701 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
702 		  __func__, ino);
703 	return err;
704 }
705 
706 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
707 {
708 	block_t start_blk, orphan_blocks, i, j;
709 	unsigned int s_flags = sbi->sb->s_flags;
710 	int err = 0;
711 #ifdef CONFIG_QUOTA
712 	int quota_enabled;
713 #endif
714 
715 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
716 		return 0;
717 
718 	if (bdev_read_only(sbi->sb->s_bdev)) {
719 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
720 		return 0;
721 	}
722 
723 	if (s_flags & SB_RDONLY) {
724 		f2fs_info(sbi, "orphan cleanup on readonly fs");
725 		sbi->sb->s_flags &= ~SB_RDONLY;
726 	}
727 
728 #ifdef CONFIG_QUOTA
729 	/*
730 	 * Turn on quotas which were not enabled for read-only mounts if
731 	 * filesystem has quota feature, so that they are updated correctly.
732 	 */
733 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
734 #endif
735 
736 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
737 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
738 
739 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
740 
741 	for (i = 0; i < orphan_blocks; i++) {
742 		struct page *page;
743 		struct f2fs_orphan_block *orphan_blk;
744 
745 		page = f2fs_get_meta_page(sbi, start_blk + i);
746 		if (IS_ERR(page)) {
747 			err = PTR_ERR(page);
748 			goto out;
749 		}
750 
751 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
752 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
753 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
754 
755 			err = recover_orphan_inode(sbi, ino);
756 			if (err) {
757 				f2fs_put_page(page, 1);
758 				goto out;
759 			}
760 		}
761 		f2fs_put_page(page, 1);
762 	}
763 	/* clear Orphan Flag */
764 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
765 out:
766 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
767 
768 #ifdef CONFIG_QUOTA
769 	/* Turn quotas off */
770 	if (quota_enabled)
771 		f2fs_quota_off_umount(sbi->sb);
772 #endif
773 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
774 
775 	return err;
776 }
777 
778 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
779 {
780 	struct list_head *head;
781 	struct f2fs_orphan_block *orphan_blk = NULL;
782 	unsigned int nentries = 0;
783 	unsigned short index = 1;
784 	unsigned short orphan_blocks;
785 	struct page *page = NULL;
786 	struct ino_entry *orphan = NULL;
787 	struct inode_management *im = &sbi->im[ORPHAN_INO];
788 
789 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
790 
791 	/*
792 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
793 	 * orphan inode operations are covered under f2fs_lock_op().
794 	 * And, spin_lock should be avoided due to page operations below.
795 	 */
796 	head = &im->ino_list;
797 
798 	/* loop for each orphan inode entry and write them in Jornal block */
799 	list_for_each_entry(orphan, head, list) {
800 		if (!page) {
801 			page = f2fs_grab_meta_page(sbi, start_blk++);
802 			orphan_blk =
803 				(struct f2fs_orphan_block *)page_address(page);
804 			memset(orphan_blk, 0, sizeof(*orphan_blk));
805 		}
806 
807 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
808 
809 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
810 			/*
811 			 * an orphan block is full of 1020 entries,
812 			 * then we need to flush current orphan blocks
813 			 * and bring another one in memory
814 			 */
815 			orphan_blk->blk_addr = cpu_to_le16(index);
816 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
817 			orphan_blk->entry_count = cpu_to_le32(nentries);
818 			set_page_dirty(page);
819 			f2fs_put_page(page, 1);
820 			index++;
821 			nentries = 0;
822 			page = NULL;
823 		}
824 	}
825 
826 	if (page) {
827 		orphan_blk->blk_addr = cpu_to_le16(index);
828 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
829 		orphan_blk->entry_count = cpu_to_le32(nentries);
830 		set_page_dirty(page);
831 		f2fs_put_page(page, 1);
832 	}
833 }
834 
835 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
836 						struct f2fs_checkpoint *ckpt)
837 {
838 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
839 	__u32 chksum;
840 
841 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
842 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
843 		chksum_ofs += sizeof(chksum);
844 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
845 						F2FS_BLKSIZE - chksum_ofs);
846 	}
847 	return chksum;
848 }
849 
850 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
851 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
852 		unsigned long long *version)
853 {
854 	size_t crc_offset = 0;
855 	__u32 crc;
856 
857 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
858 	if (IS_ERR(*cp_page))
859 		return PTR_ERR(*cp_page);
860 
861 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
862 
863 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
864 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
865 			crc_offset > CP_CHKSUM_OFFSET) {
866 		f2fs_put_page(*cp_page, 1);
867 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
868 		return -EINVAL;
869 	}
870 
871 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
872 	if (crc != cur_cp_crc(*cp_block)) {
873 		f2fs_put_page(*cp_page, 1);
874 		f2fs_warn(sbi, "invalid crc value");
875 		return -EINVAL;
876 	}
877 
878 	*version = cur_cp_version(*cp_block);
879 	return 0;
880 }
881 
882 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
883 				block_t cp_addr, unsigned long long *version)
884 {
885 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
886 	struct f2fs_checkpoint *cp_block = NULL;
887 	unsigned long long cur_version = 0, pre_version = 0;
888 	unsigned int cp_blocks;
889 	int err;
890 
891 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
892 					&cp_page_1, version);
893 	if (err)
894 		return NULL;
895 
896 	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
897 
898 	if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
899 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
900 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
901 		goto invalid_cp;
902 	}
903 	pre_version = *version;
904 
905 	cp_addr += cp_blocks - 1;
906 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
907 					&cp_page_2, version);
908 	if (err)
909 		goto invalid_cp;
910 	cur_version = *version;
911 
912 	if (cur_version == pre_version) {
913 		*version = cur_version;
914 		f2fs_put_page(cp_page_2, 1);
915 		return cp_page_1;
916 	}
917 	f2fs_put_page(cp_page_2, 1);
918 invalid_cp:
919 	f2fs_put_page(cp_page_1, 1);
920 	return NULL;
921 }
922 
923 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
924 {
925 	struct f2fs_checkpoint *cp_block;
926 	struct f2fs_super_block *fsb = sbi->raw_super;
927 	struct page *cp1, *cp2, *cur_page;
928 	unsigned long blk_size = sbi->blocksize;
929 	unsigned long long cp1_version = 0, cp2_version = 0;
930 	unsigned long long cp_start_blk_no;
931 	unsigned int cp_blks = 1 + __cp_payload(sbi);
932 	block_t cp_blk_no;
933 	int i;
934 	int err;
935 
936 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
937 				  GFP_KERNEL);
938 	if (!sbi->ckpt)
939 		return -ENOMEM;
940 	/*
941 	 * Finding out valid cp block involves read both
942 	 * sets( cp pack 1 and cp pack 2)
943 	 */
944 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
945 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
946 
947 	/* The second checkpoint pack should start at the next segment */
948 	cp_start_blk_no += ((unsigned long long)1) <<
949 				le32_to_cpu(fsb->log_blocks_per_seg);
950 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
951 
952 	if (cp1 && cp2) {
953 		if (ver_after(cp2_version, cp1_version))
954 			cur_page = cp2;
955 		else
956 			cur_page = cp1;
957 	} else if (cp1) {
958 		cur_page = cp1;
959 	} else if (cp2) {
960 		cur_page = cp2;
961 	} else {
962 		err = -EFSCORRUPTED;
963 		goto fail_no_cp;
964 	}
965 
966 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
967 	memcpy(sbi->ckpt, cp_block, blk_size);
968 
969 	if (cur_page == cp1)
970 		sbi->cur_cp_pack = 1;
971 	else
972 		sbi->cur_cp_pack = 2;
973 
974 	/* Sanity checking of checkpoint */
975 	if (f2fs_sanity_check_ckpt(sbi)) {
976 		err = -EFSCORRUPTED;
977 		goto free_fail_no_cp;
978 	}
979 
980 	if (cp_blks <= 1)
981 		goto done;
982 
983 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
984 	if (cur_page == cp2)
985 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
986 
987 	for (i = 1; i < cp_blks; i++) {
988 		void *sit_bitmap_ptr;
989 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
990 
991 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
992 		if (IS_ERR(cur_page)) {
993 			err = PTR_ERR(cur_page);
994 			goto free_fail_no_cp;
995 		}
996 		sit_bitmap_ptr = page_address(cur_page);
997 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
998 		f2fs_put_page(cur_page, 1);
999 	}
1000 done:
1001 	f2fs_put_page(cp1, 1);
1002 	f2fs_put_page(cp2, 1);
1003 	return 0;
1004 
1005 free_fail_no_cp:
1006 	f2fs_put_page(cp1, 1);
1007 	f2fs_put_page(cp2, 1);
1008 fail_no_cp:
1009 	kvfree(sbi->ckpt);
1010 	return err;
1011 }
1012 
1013 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1014 {
1015 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1016 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1017 
1018 	if (is_inode_flag_set(inode, flag))
1019 		return;
1020 
1021 	set_inode_flag(inode, flag);
1022 	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1023 	stat_inc_dirty_inode(sbi, type);
1024 }
1025 
1026 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1027 {
1028 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1029 
1030 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1031 		return;
1032 
1033 	list_del_init(&F2FS_I(inode)->dirty_list);
1034 	clear_inode_flag(inode, flag);
1035 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1036 }
1037 
1038 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1039 {
1040 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1041 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1042 
1043 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1044 			!S_ISLNK(inode->i_mode))
1045 		return;
1046 
1047 	spin_lock(&sbi->inode_lock[type]);
1048 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1049 		__add_dirty_inode(inode, type);
1050 	inode_inc_dirty_pages(inode);
1051 	spin_unlock(&sbi->inode_lock[type]);
1052 
1053 	set_page_private_reference(&folio->page);
1054 }
1055 
1056 void f2fs_remove_dirty_inode(struct inode *inode)
1057 {
1058 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1059 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1060 
1061 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1062 			!S_ISLNK(inode->i_mode))
1063 		return;
1064 
1065 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1066 		return;
1067 
1068 	spin_lock(&sbi->inode_lock[type]);
1069 	__remove_dirty_inode(inode, type);
1070 	spin_unlock(&sbi->inode_lock[type]);
1071 }
1072 
1073 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1074 						bool from_cp)
1075 {
1076 	struct list_head *head;
1077 	struct inode *inode;
1078 	struct f2fs_inode_info *fi;
1079 	bool is_dir = (type == DIR_INODE);
1080 	unsigned long ino = 0;
1081 
1082 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1083 				get_pages(sbi, is_dir ?
1084 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085 retry:
1086 	if (unlikely(f2fs_cp_error(sbi))) {
1087 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1088 				get_pages(sbi, is_dir ?
1089 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1090 		return -EIO;
1091 	}
1092 
1093 	spin_lock(&sbi->inode_lock[type]);
1094 
1095 	head = &sbi->inode_list[type];
1096 	if (list_empty(head)) {
1097 		spin_unlock(&sbi->inode_lock[type]);
1098 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1099 				get_pages(sbi, is_dir ?
1100 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1101 		return 0;
1102 	}
1103 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1104 	inode = igrab(&fi->vfs_inode);
1105 	spin_unlock(&sbi->inode_lock[type]);
1106 	if (inode) {
1107 		unsigned long cur_ino = inode->i_ino;
1108 
1109 		if (from_cp)
1110 			F2FS_I(inode)->cp_task = current;
1111 		F2FS_I(inode)->wb_task = current;
1112 
1113 		filemap_fdatawrite(inode->i_mapping);
1114 
1115 		F2FS_I(inode)->wb_task = NULL;
1116 		if (from_cp)
1117 			F2FS_I(inode)->cp_task = NULL;
1118 
1119 		iput(inode);
1120 		/* We need to give cpu to another writers. */
1121 		if (ino == cur_ino)
1122 			cond_resched();
1123 		else
1124 			ino = cur_ino;
1125 	} else {
1126 		/*
1127 		 * We should submit bio, since it exists several
1128 		 * wribacking dentry pages in the freeing inode.
1129 		 */
1130 		f2fs_submit_merged_write(sbi, DATA);
1131 		cond_resched();
1132 	}
1133 	goto retry;
1134 }
1135 
1136 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1137 {
1138 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1139 	struct inode *inode;
1140 	struct f2fs_inode_info *fi;
1141 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1142 
1143 	while (total--) {
1144 		if (unlikely(f2fs_cp_error(sbi)))
1145 			return -EIO;
1146 
1147 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1148 		if (list_empty(head)) {
1149 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1150 			return 0;
1151 		}
1152 		fi = list_first_entry(head, struct f2fs_inode_info,
1153 							gdirty_list);
1154 		inode = igrab(&fi->vfs_inode);
1155 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1156 		if (inode) {
1157 			sync_inode_metadata(inode, 0);
1158 
1159 			/* it's on eviction */
1160 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1161 				f2fs_update_inode_page(inode);
1162 			iput(inode);
1163 		}
1164 	}
1165 	return 0;
1166 }
1167 
1168 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1169 {
1170 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1171 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1172 	nid_t last_nid = nm_i->next_scan_nid;
1173 
1174 	next_free_nid(sbi, &last_nid);
1175 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1176 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1177 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1178 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1179 }
1180 
1181 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1182 {
1183 	bool ret = false;
1184 
1185 	if (!is_journalled_quota(sbi))
1186 		return false;
1187 
1188 	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1189 		return true;
1190 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1191 		ret = false;
1192 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1193 		ret = false;
1194 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1195 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1196 		ret = true;
1197 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1198 		ret = true;
1199 	}
1200 	f2fs_up_write(&sbi->quota_sem);
1201 	return ret;
1202 }
1203 
1204 /*
1205  * Freeze all the FS-operations for checkpoint.
1206  */
1207 static int block_operations(struct f2fs_sb_info *sbi)
1208 {
1209 	struct writeback_control wbc = {
1210 		.sync_mode = WB_SYNC_ALL,
1211 		.nr_to_write = LONG_MAX,
1212 		.for_reclaim = 0,
1213 	};
1214 	int err = 0, cnt = 0;
1215 
1216 	/*
1217 	 * Let's flush inline_data in dirty node pages.
1218 	 */
1219 	f2fs_flush_inline_data(sbi);
1220 
1221 retry_flush_quotas:
1222 	f2fs_lock_all(sbi);
1223 	if (__need_flush_quota(sbi)) {
1224 		int locked;
1225 
1226 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1227 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1228 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1229 			goto retry_flush_dents;
1230 		}
1231 		f2fs_unlock_all(sbi);
1232 
1233 		/* only failed during mount/umount/freeze/quotactl */
1234 		locked = down_read_trylock(&sbi->sb->s_umount);
1235 		f2fs_quota_sync(sbi->sb, -1);
1236 		if (locked)
1237 			up_read(&sbi->sb->s_umount);
1238 		cond_resched();
1239 		goto retry_flush_quotas;
1240 	}
1241 
1242 retry_flush_dents:
1243 	/* write all the dirty dentry pages */
1244 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1245 		f2fs_unlock_all(sbi);
1246 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1247 		if (err)
1248 			return err;
1249 		cond_resched();
1250 		goto retry_flush_quotas;
1251 	}
1252 
1253 	/*
1254 	 * POR: we should ensure that there are no dirty node pages
1255 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1256 	 */
1257 	f2fs_down_write(&sbi->node_change);
1258 
1259 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1260 		f2fs_up_write(&sbi->node_change);
1261 		f2fs_unlock_all(sbi);
1262 		err = f2fs_sync_inode_meta(sbi);
1263 		if (err)
1264 			return err;
1265 		cond_resched();
1266 		goto retry_flush_quotas;
1267 	}
1268 
1269 retry_flush_nodes:
1270 	f2fs_down_write(&sbi->node_write);
1271 
1272 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1273 		f2fs_up_write(&sbi->node_write);
1274 		atomic_inc(&sbi->wb_sync_req[NODE]);
1275 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1276 		atomic_dec(&sbi->wb_sync_req[NODE]);
1277 		if (err) {
1278 			f2fs_up_write(&sbi->node_change);
1279 			f2fs_unlock_all(sbi);
1280 			return err;
1281 		}
1282 		cond_resched();
1283 		goto retry_flush_nodes;
1284 	}
1285 
1286 	/*
1287 	 * sbi->node_change is used only for AIO write_begin path which produces
1288 	 * dirty node blocks and some checkpoint values by block allocation.
1289 	 */
1290 	__prepare_cp_block(sbi);
1291 	f2fs_up_write(&sbi->node_change);
1292 	return err;
1293 }
1294 
1295 static void unblock_operations(struct f2fs_sb_info *sbi)
1296 {
1297 	f2fs_up_write(&sbi->node_write);
1298 	f2fs_unlock_all(sbi);
1299 }
1300 
1301 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1302 {
1303 	DEFINE_WAIT(wait);
1304 
1305 	for (;;) {
1306 		if (!get_pages(sbi, type))
1307 			break;
1308 
1309 		if (unlikely(f2fs_cp_error(sbi)))
1310 			break;
1311 
1312 		if (type == F2FS_DIRTY_META)
1313 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1314 							FS_CP_META_IO);
1315 		else if (type == F2FS_WB_CP_DATA)
1316 			f2fs_submit_merged_write(sbi, DATA);
1317 
1318 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1319 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1320 	}
1321 	finish_wait(&sbi->cp_wait, &wait);
1322 }
1323 
1324 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1325 {
1326 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1327 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1328 	unsigned long flags;
1329 
1330 	if (cpc->reason & CP_UMOUNT) {
1331 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1332 			NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1333 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1334 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1335 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1336 						f2fs_nat_bitmap_enabled(sbi)) {
1337 			f2fs_enable_nat_bits(sbi);
1338 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1339 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1340 		}
1341 	}
1342 
1343 	spin_lock_irqsave(&sbi->cp_lock, flags);
1344 
1345 	if (cpc->reason & CP_TRIMMED)
1346 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1347 	else
1348 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1349 
1350 	if (cpc->reason & CP_UMOUNT)
1351 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1352 	else
1353 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1354 
1355 	if (cpc->reason & CP_FASTBOOT)
1356 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1357 	else
1358 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1359 
1360 	if (orphan_num)
1361 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1362 	else
1363 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1364 
1365 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1366 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1367 
1368 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1369 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1370 	else
1371 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1372 
1373 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1374 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1375 	else
1376 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1377 
1378 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1379 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1380 	else
1381 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1382 
1383 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1384 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1385 	else
1386 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1387 
1388 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1389 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1390 
1391 	/* set this flag to activate crc|cp_ver for recovery */
1392 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1393 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1394 
1395 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1396 }
1397 
1398 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1399 	void *src, block_t blk_addr)
1400 {
1401 	struct writeback_control wbc = {
1402 		.for_reclaim = 0,
1403 	};
1404 
1405 	/*
1406 	 * pagevec_lookup_tag and lock_page again will take
1407 	 * some extra time. Therefore, f2fs_update_meta_pages and
1408 	 * f2fs_sync_meta_pages are combined in this function.
1409 	 */
1410 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1411 	int err;
1412 
1413 	f2fs_wait_on_page_writeback(page, META, true, true);
1414 
1415 	memcpy(page_address(page), src, PAGE_SIZE);
1416 
1417 	set_page_dirty(page);
1418 	if (unlikely(!clear_page_dirty_for_io(page)))
1419 		f2fs_bug_on(sbi, 1);
1420 
1421 	/* writeout cp pack 2 page */
1422 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1423 	if (unlikely(err && f2fs_cp_error(sbi))) {
1424 		f2fs_put_page(page, 1);
1425 		return;
1426 	}
1427 
1428 	f2fs_bug_on(sbi, err);
1429 	f2fs_put_page(page, 0);
1430 
1431 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1432 	f2fs_submit_merged_write(sbi, META_FLUSH);
1433 }
1434 
1435 static inline u64 get_sectors_written(struct block_device *bdev)
1436 {
1437 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1438 }
1439 
1440 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1441 {
1442 	if (f2fs_is_multi_device(sbi)) {
1443 		u64 sectors = 0;
1444 		int i;
1445 
1446 		for (i = 0; i < sbi->s_ndevs; i++)
1447 			sectors += get_sectors_written(FDEV(i).bdev);
1448 
1449 		return sectors;
1450 	}
1451 
1452 	return get_sectors_written(sbi->sb->s_bdev);
1453 }
1454 
1455 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1456 {
1457 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1458 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1459 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1460 	block_t start_blk;
1461 	unsigned int data_sum_blocks, orphan_blocks;
1462 	__u32 crc32 = 0;
1463 	int i;
1464 	int cp_payload_blks = __cp_payload(sbi);
1465 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1466 	u64 kbytes_written;
1467 	int err;
1468 
1469 	/* Flush all the NAT/SIT pages */
1470 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1471 
1472 	/* start to update checkpoint, cp ver is already updated previously */
1473 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1474 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1475 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1476 		ckpt->cur_node_segno[i] =
1477 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1478 		ckpt->cur_node_blkoff[i] =
1479 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1480 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1481 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1482 	}
1483 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1484 		ckpt->cur_data_segno[i] =
1485 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1486 		ckpt->cur_data_blkoff[i] =
1487 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1488 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1489 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1490 	}
1491 
1492 	/* 2 cp + n data seg summary + orphan inode blocks */
1493 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1494 	spin_lock_irqsave(&sbi->cp_lock, flags);
1495 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1496 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1497 	else
1498 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1499 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1500 
1501 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1502 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1503 			orphan_blocks);
1504 
1505 	if (__remain_node_summaries(cpc->reason))
1506 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1507 				cp_payload_blks + data_sum_blocks +
1508 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1509 	else
1510 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1511 				cp_payload_blks + data_sum_blocks +
1512 				orphan_blocks);
1513 
1514 	/* update ckpt flag for checkpoint */
1515 	update_ckpt_flags(sbi, cpc);
1516 
1517 	/* update SIT/NAT bitmap */
1518 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1519 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1520 
1521 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1522 	*((__le32 *)((unsigned char *)ckpt +
1523 				le32_to_cpu(ckpt->checksum_offset)))
1524 				= cpu_to_le32(crc32);
1525 
1526 	start_blk = __start_cp_next_addr(sbi);
1527 
1528 	/* write nat bits */
1529 	if ((cpc->reason & CP_UMOUNT) &&
1530 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1531 		__u64 cp_ver = cur_cp_version(ckpt);
1532 		block_t blk;
1533 
1534 		cp_ver |= ((__u64)crc32 << 32);
1535 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1536 
1537 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1538 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1539 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1540 					(i << F2FS_BLKSIZE_BITS), blk + i);
1541 	}
1542 
1543 	/* write out checkpoint buffer at block 0 */
1544 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1545 
1546 	for (i = 1; i < 1 + cp_payload_blks; i++)
1547 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1548 							start_blk++);
1549 
1550 	if (orphan_num) {
1551 		write_orphan_inodes(sbi, start_blk);
1552 		start_blk += orphan_blocks;
1553 	}
1554 
1555 	f2fs_write_data_summaries(sbi, start_blk);
1556 	start_blk += data_sum_blocks;
1557 
1558 	/* Record write statistics in the hot node summary */
1559 	kbytes_written = sbi->kbytes_written;
1560 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1561 				sbi->sectors_written_start) >> 1;
1562 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1563 
1564 	if (__remain_node_summaries(cpc->reason)) {
1565 		f2fs_write_node_summaries(sbi, start_blk);
1566 		start_blk += NR_CURSEG_NODE_TYPE;
1567 	}
1568 
1569 	/* update user_block_counts */
1570 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1571 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1572 	percpu_counter_set(&sbi->rf_node_block_count, 0);
1573 
1574 	/* Here, we have one bio having CP pack except cp pack 2 page */
1575 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1576 	/* Wait for all dirty meta pages to be submitted for IO */
1577 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1578 
1579 	/* wait for previous submitted meta pages writeback */
1580 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1581 
1582 	/* flush all device cache */
1583 	err = f2fs_flush_device_cache(sbi);
1584 	if (err)
1585 		return err;
1586 
1587 	/* barrier and flush checkpoint cp pack 2 page if it can */
1588 	commit_checkpoint(sbi, ckpt, start_blk);
1589 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1590 
1591 	/*
1592 	 * invalidate intermediate page cache borrowed from meta inode which are
1593 	 * used for migration of encrypted, verity or compressed inode's blocks.
1594 	 */
1595 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1596 		f2fs_sb_has_compression(sbi))
1597 		invalidate_mapping_pages(META_MAPPING(sbi),
1598 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1599 
1600 	f2fs_release_ino_entry(sbi, false);
1601 
1602 	f2fs_reset_fsync_node_info(sbi);
1603 
1604 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1605 	clear_sbi_flag(sbi, SBI_NEED_CP);
1606 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1607 
1608 	spin_lock(&sbi->stat_lock);
1609 	sbi->unusable_block_count = 0;
1610 	spin_unlock(&sbi->stat_lock);
1611 
1612 	__set_cp_next_pack(sbi);
1613 
1614 	/*
1615 	 * redirty superblock if metadata like node page or inode cache is
1616 	 * updated during writing checkpoint.
1617 	 */
1618 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1619 			get_pages(sbi, F2FS_DIRTY_IMETA))
1620 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1621 
1622 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1623 
1624 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1625 }
1626 
1627 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1628 {
1629 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1630 	unsigned long long ckpt_ver;
1631 	int err = 0;
1632 
1633 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1634 		return -EROFS;
1635 
1636 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1637 		if (cpc->reason != CP_PAUSE)
1638 			return 0;
1639 		f2fs_warn(sbi, "Start checkpoint disabled!");
1640 	}
1641 	if (cpc->reason != CP_RESIZE)
1642 		f2fs_down_write(&sbi->cp_global_sem);
1643 
1644 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1645 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1646 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1647 		goto out;
1648 	if (unlikely(f2fs_cp_error(sbi))) {
1649 		err = -EIO;
1650 		goto out;
1651 	}
1652 
1653 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1654 
1655 	err = block_operations(sbi);
1656 	if (err)
1657 		goto out;
1658 
1659 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1660 
1661 	f2fs_flush_merged_writes(sbi);
1662 
1663 	/* this is the case of multiple fstrims without any changes */
1664 	if (cpc->reason & CP_DISCARD) {
1665 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1666 			unblock_operations(sbi);
1667 			goto out;
1668 		}
1669 
1670 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1671 				SIT_I(sbi)->dirty_sentries == 0 &&
1672 				prefree_segments(sbi) == 0) {
1673 			f2fs_flush_sit_entries(sbi, cpc);
1674 			f2fs_clear_prefree_segments(sbi, cpc);
1675 			unblock_operations(sbi);
1676 			goto out;
1677 		}
1678 	}
1679 
1680 	/*
1681 	 * update checkpoint pack index
1682 	 * Increase the version number so that
1683 	 * SIT entries and seg summaries are written at correct place
1684 	 */
1685 	ckpt_ver = cur_cp_version(ckpt);
1686 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1687 
1688 	/* write cached NAT/SIT entries to NAT/SIT area */
1689 	err = f2fs_flush_nat_entries(sbi, cpc);
1690 	if (err) {
1691 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1692 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1693 		goto stop;
1694 	}
1695 
1696 	f2fs_flush_sit_entries(sbi, cpc);
1697 
1698 	/* save inmem log status */
1699 	f2fs_save_inmem_curseg(sbi);
1700 
1701 	err = do_checkpoint(sbi, cpc);
1702 	if (err) {
1703 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1704 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1705 		f2fs_release_discard_addrs(sbi);
1706 	} else {
1707 		f2fs_clear_prefree_segments(sbi, cpc);
1708 	}
1709 
1710 	f2fs_restore_inmem_curseg(sbi);
1711 stop:
1712 	unblock_operations(sbi);
1713 	stat_inc_cp_count(sbi->stat_info);
1714 
1715 	if (cpc->reason & CP_RECOVERY)
1716 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1717 
1718 	/* update CP_TIME to trigger checkpoint periodically */
1719 	f2fs_update_time(sbi, CP_TIME);
1720 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1721 out:
1722 	if (cpc->reason != CP_RESIZE)
1723 		f2fs_up_write(&sbi->cp_global_sem);
1724 	return err;
1725 }
1726 
1727 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1728 {
1729 	int i;
1730 
1731 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1732 		struct inode_management *im = &sbi->im[i];
1733 
1734 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1735 		spin_lock_init(&im->ino_lock);
1736 		INIT_LIST_HEAD(&im->ino_list);
1737 		im->ino_num = 0;
1738 	}
1739 
1740 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1741 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1742 				F2FS_ORPHANS_PER_BLOCK;
1743 }
1744 
1745 int __init f2fs_create_checkpoint_caches(void)
1746 {
1747 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1748 			sizeof(struct ino_entry));
1749 	if (!ino_entry_slab)
1750 		return -ENOMEM;
1751 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1752 			sizeof(struct inode_entry));
1753 	if (!f2fs_inode_entry_slab) {
1754 		kmem_cache_destroy(ino_entry_slab);
1755 		return -ENOMEM;
1756 	}
1757 	return 0;
1758 }
1759 
1760 void f2fs_destroy_checkpoint_caches(void)
1761 {
1762 	kmem_cache_destroy(ino_entry_slab);
1763 	kmem_cache_destroy(f2fs_inode_entry_slab);
1764 }
1765 
1766 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1767 {
1768 	struct cp_control cpc = { .reason = CP_SYNC, };
1769 	int err;
1770 
1771 	f2fs_down_write(&sbi->gc_lock);
1772 	err = f2fs_write_checkpoint(sbi, &cpc);
1773 	f2fs_up_write(&sbi->gc_lock);
1774 
1775 	return err;
1776 }
1777 
1778 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1779 {
1780 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1781 	struct ckpt_req *req, *next;
1782 	struct llist_node *dispatch_list;
1783 	u64 sum_diff = 0, diff, count = 0;
1784 	int ret;
1785 
1786 	dispatch_list = llist_del_all(&cprc->issue_list);
1787 	if (!dispatch_list)
1788 		return;
1789 	dispatch_list = llist_reverse_order(dispatch_list);
1790 
1791 	ret = __write_checkpoint_sync(sbi);
1792 	atomic_inc(&cprc->issued_ckpt);
1793 
1794 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1795 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1796 		req->ret = ret;
1797 		complete(&req->wait);
1798 
1799 		sum_diff += diff;
1800 		count++;
1801 	}
1802 	atomic_sub(count, &cprc->queued_ckpt);
1803 	atomic_add(count, &cprc->total_ckpt);
1804 
1805 	spin_lock(&cprc->stat_lock);
1806 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1807 	if (cprc->peak_time < cprc->cur_time)
1808 		cprc->peak_time = cprc->cur_time;
1809 	spin_unlock(&cprc->stat_lock);
1810 }
1811 
1812 static int issue_checkpoint_thread(void *data)
1813 {
1814 	struct f2fs_sb_info *sbi = data;
1815 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1816 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1817 repeat:
1818 	if (kthread_should_stop())
1819 		return 0;
1820 
1821 	if (!llist_empty(&cprc->issue_list))
1822 		__checkpoint_and_complete_reqs(sbi);
1823 
1824 	wait_event_interruptible(*q,
1825 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1826 	goto repeat;
1827 }
1828 
1829 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1830 		struct ckpt_req *wait_req)
1831 {
1832 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1833 
1834 	if (!llist_empty(&cprc->issue_list)) {
1835 		__checkpoint_and_complete_reqs(sbi);
1836 	} else {
1837 		/* already dispatched by issue_checkpoint_thread */
1838 		if (wait_req)
1839 			wait_for_completion(&wait_req->wait);
1840 	}
1841 }
1842 
1843 static void init_ckpt_req(struct ckpt_req *req)
1844 {
1845 	memset(req, 0, sizeof(struct ckpt_req));
1846 
1847 	init_completion(&req->wait);
1848 	req->queue_time = ktime_get();
1849 }
1850 
1851 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1852 {
1853 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1854 	struct ckpt_req req;
1855 	struct cp_control cpc;
1856 
1857 	cpc.reason = __get_cp_reason(sbi);
1858 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1859 		int ret;
1860 
1861 		f2fs_down_write(&sbi->gc_lock);
1862 		ret = f2fs_write_checkpoint(sbi, &cpc);
1863 		f2fs_up_write(&sbi->gc_lock);
1864 
1865 		return ret;
1866 	}
1867 
1868 	if (!cprc->f2fs_issue_ckpt)
1869 		return __write_checkpoint_sync(sbi);
1870 
1871 	init_ckpt_req(&req);
1872 
1873 	llist_add(&req.llnode, &cprc->issue_list);
1874 	atomic_inc(&cprc->queued_ckpt);
1875 
1876 	/*
1877 	 * update issue_list before we wake up issue_checkpoint thread,
1878 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1879 	 * see more details in comments of waitqueue_active().
1880 	 */
1881 	smp_mb();
1882 
1883 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1884 		wake_up(&cprc->ckpt_wait_queue);
1885 
1886 	if (cprc->f2fs_issue_ckpt)
1887 		wait_for_completion(&req.wait);
1888 	else
1889 		flush_remained_ckpt_reqs(sbi, &req);
1890 
1891 	return req.ret;
1892 }
1893 
1894 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1895 {
1896 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1897 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1898 
1899 	if (cprc->f2fs_issue_ckpt)
1900 		return 0;
1901 
1902 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1903 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1904 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1905 		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1906 
1907 		cprc->f2fs_issue_ckpt = NULL;
1908 		return err;
1909 	}
1910 
1911 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1912 
1913 	return 0;
1914 }
1915 
1916 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1917 {
1918 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1919 	struct task_struct *ckpt_task;
1920 
1921 	if (!cprc->f2fs_issue_ckpt)
1922 		return;
1923 
1924 	ckpt_task = cprc->f2fs_issue_ckpt;
1925 	cprc->f2fs_issue_ckpt = NULL;
1926 	kthread_stop(ckpt_task);
1927 
1928 	f2fs_flush_ckpt_thread(sbi);
1929 }
1930 
1931 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1932 {
1933 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1934 
1935 	flush_remained_ckpt_reqs(sbi, NULL);
1936 
1937 	/* Let's wait for the previous dispatched checkpoint. */
1938 	while (atomic_read(&cprc->queued_ckpt))
1939 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1940 }
1941 
1942 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1943 {
1944 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1945 
1946 	atomic_set(&cprc->issued_ckpt, 0);
1947 	atomic_set(&cprc->total_ckpt, 0);
1948 	atomic_set(&cprc->queued_ckpt, 0);
1949 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1950 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1951 	init_llist_head(&cprc->issue_list);
1952 	spin_lock_init(&cprc->stat_lock);
1953 }
1954