xref: /linux/fs/f2fs/checkpoint.c (revision 40d269c000bda9fcd276a0412a9cebd3f6e344c5)
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 	if (!end_io)
34 		f2fs_flush_merged_writes(sbi);
35 	f2fs_handle_critical_error(sbi, reason, end_io);
36 }
37 
38 /*
39  * We guarantee no failure on the returned page.
40  */
41 struct 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;
45 repeat:
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 
57 static 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;
76 repeat:
77 	page = f2fs_grab_cache_page(mapping, index, false);
78 	if (!page) {
79 		cond_resched();
80 		goto repeat;
81 	}
82 	if (PageUptodate(page))
83 		goto out;
84 
85 	fio.page = page;
86 
87 	err = f2fs_submit_page_bio(&fio);
88 	if (err) {
89 		f2fs_put_page(page, 1);
90 		return ERR_PTR(err);
91 	}
92 
93 	f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
94 
95 	lock_page(page);
96 	if (unlikely(page->mapping != mapping)) {
97 		f2fs_put_page(page, 1);
98 		goto repeat;
99 	}
100 
101 	if (unlikely(!PageUptodate(page))) {
102 		f2fs_handle_page_eio(sbi, page->index, META);
103 		f2fs_put_page(page, 1);
104 		return ERR_PTR(-EIO);
105 	}
106 out:
107 	return page;
108 }
109 
110 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
111 {
112 	return __get_meta_page(sbi, index, true);
113 }
114 
115 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
116 {
117 	struct page *page;
118 	int count = 0;
119 
120 retry:
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 */
132 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
133 {
134 	return __get_meta_page(sbi, index, false);
135 }
136 
137 static 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 
164 out_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();
169 out_handle:
170 	f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
171 	return exist;
172 }
173 
174 static 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 err;
183 		break;
184 	case META_SSA:
185 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
186 			blkaddr < SM_I(sbi)->ssa_blkaddr))
187 			goto err;
188 		break;
189 	case META_CP:
190 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
191 			blkaddr < __start_cp_addr(sbi)))
192 			goto err;
193 		break;
194 	case META_POR:
195 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
196 			blkaddr < MAIN_BLKADDR(sbi)))
197 			goto err;
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;
229 err:
230 	f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
231 	return false;
232 }
233 
234 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
235 					block_t blkaddr, int type)
236 {
237 	if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY))
238 		return false;
239 	return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
240 }
241 
242 bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi,
243 					block_t blkaddr, int type)
244 {
245 	return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
246 }
247 
248 /*
249  * Readahead CP/NAT/SIT/SSA/POR pages
250  */
251 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
252 							int type, bool sync)
253 {
254 	struct page *page;
255 	block_t blkno = start;
256 	struct f2fs_io_info fio = {
257 		.sbi = sbi,
258 		.type = META,
259 		.op = REQ_OP_READ,
260 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
261 		.encrypted_page = NULL,
262 		.in_list = 0,
263 		.is_por = (type == META_POR) ? 1 : 0,
264 	};
265 	struct blk_plug plug;
266 	int err;
267 
268 	if (unlikely(type == META_POR))
269 		fio.op_flags &= ~REQ_META;
270 
271 	blk_start_plug(&plug);
272 	for (; nrpages-- > 0; blkno++) {
273 
274 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
275 			goto out;
276 
277 		switch (type) {
278 		case META_NAT:
279 			if (unlikely(blkno >=
280 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
281 				blkno = 0;
282 			/* get nat block addr */
283 			fio.new_blkaddr = current_nat_addr(sbi,
284 					blkno * NAT_ENTRY_PER_BLOCK);
285 			break;
286 		case META_SIT:
287 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
288 				goto out;
289 			/* get sit block addr */
290 			fio.new_blkaddr = current_sit_addr(sbi,
291 					blkno * SIT_ENTRY_PER_BLOCK);
292 			break;
293 		case META_SSA:
294 		case META_CP:
295 		case META_POR:
296 			fio.new_blkaddr = blkno;
297 			break;
298 		default:
299 			BUG();
300 		}
301 
302 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
303 						fio.new_blkaddr, false);
304 		if (!page)
305 			continue;
306 		if (PageUptodate(page)) {
307 			f2fs_put_page(page, 1);
308 			continue;
309 		}
310 
311 		fio.page = page;
312 		err = f2fs_submit_page_bio(&fio);
313 		f2fs_put_page(page, err ? 1 : 0);
314 
315 		if (!err)
316 			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
317 							F2FS_BLKSIZE);
318 	}
319 out:
320 	blk_finish_plug(&plug);
321 	return blkno - start;
322 }
323 
324 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
325 							unsigned int ra_blocks)
326 {
327 	struct page *page;
328 	bool readahead = false;
329 
330 	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
331 		return;
332 
333 	page = find_get_page(META_MAPPING(sbi), index);
334 	if (!page || !PageUptodate(page))
335 		readahead = true;
336 	f2fs_put_page(page, 0);
337 
338 	if (readahead)
339 		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
340 }
341 
342 static int __f2fs_write_meta_page(struct page *page,
343 				struct writeback_control *wbc,
344 				enum iostat_type io_type)
345 {
346 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
347 
348 	trace_f2fs_writepage(page, META);
349 
350 	if (unlikely(f2fs_cp_error(sbi))) {
351 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
352 			ClearPageUptodate(page);
353 			dec_page_count(sbi, F2FS_DIRTY_META);
354 			unlock_page(page);
355 			return 0;
356 		}
357 		goto redirty_out;
358 	}
359 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
360 		goto redirty_out;
361 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
362 		goto redirty_out;
363 
364 	f2fs_do_write_meta_page(sbi, page, io_type);
365 	dec_page_count(sbi, F2FS_DIRTY_META);
366 
367 	if (wbc->for_reclaim)
368 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
369 
370 	unlock_page(page);
371 
372 	if (unlikely(f2fs_cp_error(sbi)))
373 		f2fs_submit_merged_write(sbi, META);
374 
375 	return 0;
376 
377 redirty_out:
378 	redirty_page_for_writepage(wbc, page);
379 	return AOP_WRITEPAGE_ACTIVATE;
380 }
381 
382 static int f2fs_write_meta_page(struct page *page,
383 				struct writeback_control *wbc)
384 {
385 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
386 }
387 
388 static int f2fs_write_meta_pages(struct address_space *mapping,
389 				struct writeback_control *wbc)
390 {
391 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
392 	long diff, written;
393 
394 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
395 		goto skip_write;
396 
397 	/* collect a number of dirty meta pages and write together */
398 	if (wbc->sync_mode != WB_SYNC_ALL &&
399 			get_pages(sbi, F2FS_DIRTY_META) <
400 					nr_pages_to_skip(sbi, META))
401 		goto skip_write;
402 
403 	/* if locked failed, cp will flush dirty pages instead */
404 	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
405 		goto skip_write;
406 
407 	trace_f2fs_writepages(mapping->host, wbc, META);
408 	diff = nr_pages_to_write(sbi, META, wbc);
409 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
410 	f2fs_up_write(&sbi->cp_global_sem);
411 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
412 	return 0;
413 
414 skip_write:
415 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
416 	trace_f2fs_writepages(mapping->host, wbc, META);
417 	return 0;
418 }
419 
420 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
421 				long nr_to_write, enum iostat_type io_type)
422 {
423 	struct address_space *mapping = META_MAPPING(sbi);
424 	pgoff_t index = 0, prev = ULONG_MAX;
425 	struct folio_batch fbatch;
426 	long nwritten = 0;
427 	int nr_folios;
428 	struct writeback_control wbc = {
429 		.for_reclaim = 0,
430 	};
431 	struct blk_plug plug;
432 
433 	folio_batch_init(&fbatch);
434 
435 	blk_start_plug(&plug);
436 
437 	while ((nr_folios = filemap_get_folios_tag(mapping, &index,
438 					(pgoff_t)-1,
439 					PAGECACHE_TAG_DIRTY, &fbatch))) {
440 		int i;
441 
442 		for (i = 0; i < nr_folios; i++) {
443 			struct folio *folio = fbatch.folios[i];
444 
445 			if (nr_to_write != LONG_MAX && i != 0 &&
446 					folio->index != prev +
447 					folio_nr_pages(fbatch.folios[i-1])) {
448 				folio_batch_release(&fbatch);
449 				goto stop;
450 			}
451 
452 			folio_lock(folio);
453 
454 			if (unlikely(folio->mapping != mapping)) {
455 continue_unlock:
456 				folio_unlock(folio);
457 				continue;
458 			}
459 			if (!folio_test_dirty(folio)) {
460 				/* someone wrote it for us */
461 				goto continue_unlock;
462 			}
463 
464 			f2fs_wait_on_page_writeback(&folio->page, META,
465 					true, true);
466 
467 			if (!folio_clear_dirty_for_io(folio))
468 				goto continue_unlock;
469 
470 			if (__f2fs_write_meta_page(&folio->page, &wbc,
471 						io_type)) {
472 				folio_unlock(folio);
473 				break;
474 			}
475 			nwritten += folio_nr_pages(folio);
476 			prev = folio->index;
477 			if (unlikely(nwritten >= nr_to_write))
478 				break;
479 		}
480 		folio_batch_release(&fbatch);
481 		cond_resched();
482 	}
483 stop:
484 	if (nwritten)
485 		f2fs_submit_merged_write(sbi, type);
486 
487 	blk_finish_plug(&plug);
488 
489 	return nwritten;
490 }
491 
492 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
493 		struct folio *folio)
494 {
495 	trace_f2fs_set_page_dirty(&folio->page, META);
496 
497 	if (!folio_test_uptodate(folio))
498 		folio_mark_uptodate(folio);
499 	if (filemap_dirty_folio(mapping, folio)) {
500 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
501 		set_page_private_reference(&folio->page);
502 		return true;
503 	}
504 	return false;
505 }
506 
507 const struct address_space_operations f2fs_meta_aops = {
508 	.writepage	= f2fs_write_meta_page,
509 	.writepages	= f2fs_write_meta_pages,
510 	.dirty_folio	= f2fs_dirty_meta_folio,
511 	.invalidate_folio = f2fs_invalidate_folio,
512 	.release_folio	= f2fs_release_folio,
513 	.migrate_folio	= filemap_migrate_folio,
514 };
515 
516 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
517 						unsigned int devidx, int type)
518 {
519 	struct inode_management *im = &sbi->im[type];
520 	struct ino_entry *e = NULL, *new = NULL;
521 
522 	if (type == FLUSH_INO) {
523 		rcu_read_lock();
524 		e = radix_tree_lookup(&im->ino_root, ino);
525 		rcu_read_unlock();
526 	}
527 
528 retry:
529 	if (!e)
530 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
531 						GFP_NOFS, true, NULL);
532 
533 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
534 
535 	spin_lock(&im->ino_lock);
536 	e = radix_tree_lookup(&im->ino_root, ino);
537 	if (!e) {
538 		if (!new) {
539 			spin_unlock(&im->ino_lock);
540 			radix_tree_preload_end();
541 			goto retry;
542 		}
543 		e = new;
544 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
545 			f2fs_bug_on(sbi, 1);
546 
547 		memset(e, 0, sizeof(struct ino_entry));
548 		e->ino = ino;
549 
550 		list_add_tail(&e->list, &im->ino_list);
551 		if (type != ORPHAN_INO)
552 			im->ino_num++;
553 	}
554 
555 	if (type == FLUSH_INO)
556 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
557 
558 	spin_unlock(&im->ino_lock);
559 	radix_tree_preload_end();
560 
561 	if (new && e != new)
562 		kmem_cache_free(ino_entry_slab, new);
563 }
564 
565 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
566 {
567 	struct inode_management *im = &sbi->im[type];
568 	struct ino_entry *e;
569 
570 	spin_lock(&im->ino_lock);
571 	e = radix_tree_lookup(&im->ino_root, ino);
572 	if (e) {
573 		list_del(&e->list);
574 		radix_tree_delete(&im->ino_root, ino);
575 		im->ino_num--;
576 		spin_unlock(&im->ino_lock);
577 		kmem_cache_free(ino_entry_slab, e);
578 		return;
579 	}
580 	spin_unlock(&im->ino_lock);
581 }
582 
583 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
584 {
585 	/* add new dirty ino entry into list */
586 	__add_ino_entry(sbi, ino, 0, type);
587 }
588 
589 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
590 {
591 	/* remove dirty ino entry from list */
592 	__remove_ino_entry(sbi, ino, type);
593 }
594 
595 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
596 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
597 {
598 	struct inode_management *im = &sbi->im[mode];
599 	struct ino_entry *e;
600 
601 	spin_lock(&im->ino_lock);
602 	e = radix_tree_lookup(&im->ino_root, ino);
603 	spin_unlock(&im->ino_lock);
604 	return e ? true : false;
605 }
606 
607 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
608 {
609 	struct ino_entry *e, *tmp;
610 	int i;
611 
612 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
613 		struct inode_management *im = &sbi->im[i];
614 
615 		spin_lock(&im->ino_lock);
616 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
617 			list_del(&e->list);
618 			radix_tree_delete(&im->ino_root, e->ino);
619 			kmem_cache_free(ino_entry_slab, e);
620 			im->ino_num--;
621 		}
622 		spin_unlock(&im->ino_lock);
623 	}
624 }
625 
626 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
627 					unsigned int devidx, int type)
628 {
629 	__add_ino_entry(sbi, ino, devidx, type);
630 }
631 
632 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
633 					unsigned int devidx, int type)
634 {
635 	struct inode_management *im = &sbi->im[type];
636 	struct ino_entry *e;
637 	bool is_dirty = false;
638 
639 	spin_lock(&im->ino_lock);
640 	e = radix_tree_lookup(&im->ino_root, ino);
641 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
642 		is_dirty = true;
643 	spin_unlock(&im->ino_lock);
644 	return is_dirty;
645 }
646 
647 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
648 {
649 	struct inode_management *im = &sbi->im[ORPHAN_INO];
650 	int err = 0;
651 
652 	spin_lock(&im->ino_lock);
653 
654 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
655 		spin_unlock(&im->ino_lock);
656 		return -ENOSPC;
657 	}
658 
659 	if (unlikely(im->ino_num >= sbi->max_orphans))
660 		err = -ENOSPC;
661 	else
662 		im->ino_num++;
663 	spin_unlock(&im->ino_lock);
664 
665 	return err;
666 }
667 
668 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
669 {
670 	struct inode_management *im = &sbi->im[ORPHAN_INO];
671 
672 	spin_lock(&im->ino_lock);
673 	f2fs_bug_on(sbi, im->ino_num == 0);
674 	im->ino_num--;
675 	spin_unlock(&im->ino_lock);
676 }
677 
678 void f2fs_add_orphan_inode(struct inode *inode)
679 {
680 	/* add new orphan ino entry into list */
681 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
682 	f2fs_update_inode_page(inode);
683 }
684 
685 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
686 {
687 	/* remove orphan entry from orphan list */
688 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
689 }
690 
691 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
692 {
693 	struct inode *inode;
694 	struct node_info ni;
695 	int err;
696 
697 	inode = f2fs_iget_retry(sbi->sb, ino);
698 	if (IS_ERR(inode)) {
699 		/*
700 		 * there should be a bug that we can't find the entry
701 		 * to orphan inode.
702 		 */
703 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
704 		return PTR_ERR(inode);
705 	}
706 
707 	err = f2fs_dquot_initialize(inode);
708 	if (err) {
709 		iput(inode);
710 		goto err_out;
711 	}
712 
713 	clear_nlink(inode);
714 
715 	/* truncate all the data during iput */
716 	iput(inode);
717 
718 	err = f2fs_get_node_info(sbi, ino, &ni, false);
719 	if (err)
720 		goto err_out;
721 
722 	/* ENOMEM was fully retried in f2fs_evict_inode. */
723 	if (ni.blk_addr != NULL_ADDR) {
724 		err = -EIO;
725 		goto err_out;
726 	}
727 	return 0;
728 
729 err_out:
730 	set_sbi_flag(sbi, SBI_NEED_FSCK);
731 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
732 		  __func__, ino);
733 	return err;
734 }
735 
736 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
737 {
738 	block_t start_blk, orphan_blocks, i, j;
739 	int err = 0;
740 
741 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
742 		return 0;
743 
744 	if (f2fs_hw_is_readonly(sbi)) {
745 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
746 		return 0;
747 	}
748 
749 	if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
750 		f2fs_info(sbi, "orphan cleanup on readonly fs");
751 
752 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
753 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
754 
755 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
756 
757 	for (i = 0; i < orphan_blocks; i++) {
758 		struct page *page;
759 		struct f2fs_orphan_block *orphan_blk;
760 
761 		page = f2fs_get_meta_page(sbi, start_blk + i);
762 		if (IS_ERR(page)) {
763 			err = PTR_ERR(page);
764 			goto out;
765 		}
766 
767 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
768 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
769 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
770 
771 			err = recover_orphan_inode(sbi, ino);
772 			if (err) {
773 				f2fs_put_page(page, 1);
774 				goto out;
775 			}
776 		}
777 		f2fs_put_page(page, 1);
778 	}
779 	/* clear Orphan Flag */
780 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
781 out:
782 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
783 
784 	return err;
785 }
786 
787 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
788 {
789 	struct list_head *head;
790 	struct f2fs_orphan_block *orphan_blk = NULL;
791 	unsigned int nentries = 0;
792 	unsigned short index = 1;
793 	unsigned short orphan_blocks;
794 	struct page *page = NULL;
795 	struct ino_entry *orphan = NULL;
796 	struct inode_management *im = &sbi->im[ORPHAN_INO];
797 
798 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
799 
800 	/*
801 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
802 	 * orphan inode operations are covered under f2fs_lock_op().
803 	 * And, spin_lock should be avoided due to page operations below.
804 	 */
805 	head = &im->ino_list;
806 
807 	/* loop for each orphan inode entry and write them in journal block */
808 	list_for_each_entry(orphan, head, list) {
809 		if (!page) {
810 			page = f2fs_grab_meta_page(sbi, start_blk++);
811 			orphan_blk =
812 				(struct f2fs_orphan_block *)page_address(page);
813 			memset(orphan_blk, 0, sizeof(*orphan_blk));
814 		}
815 
816 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
817 
818 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
819 			/*
820 			 * an orphan block is full of 1020 entries,
821 			 * then we need to flush current orphan blocks
822 			 * and bring another one in memory
823 			 */
824 			orphan_blk->blk_addr = cpu_to_le16(index);
825 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
826 			orphan_blk->entry_count = cpu_to_le32(nentries);
827 			set_page_dirty(page);
828 			f2fs_put_page(page, 1);
829 			index++;
830 			nentries = 0;
831 			page = NULL;
832 		}
833 	}
834 
835 	if (page) {
836 		orphan_blk->blk_addr = cpu_to_le16(index);
837 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
838 		orphan_blk->entry_count = cpu_to_le32(nentries);
839 		set_page_dirty(page);
840 		f2fs_put_page(page, 1);
841 	}
842 }
843 
844 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
845 						struct f2fs_checkpoint *ckpt)
846 {
847 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
848 	__u32 chksum;
849 
850 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
851 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
852 		chksum_ofs += sizeof(chksum);
853 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
854 						F2FS_BLKSIZE - chksum_ofs);
855 	}
856 	return chksum;
857 }
858 
859 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
860 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
861 		unsigned long long *version)
862 {
863 	size_t crc_offset = 0;
864 	__u32 crc;
865 
866 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
867 	if (IS_ERR(*cp_page))
868 		return PTR_ERR(*cp_page);
869 
870 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
871 
872 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
873 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
874 			crc_offset > CP_CHKSUM_OFFSET) {
875 		f2fs_put_page(*cp_page, 1);
876 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
877 		return -EINVAL;
878 	}
879 
880 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
881 	if (crc != cur_cp_crc(*cp_block)) {
882 		f2fs_put_page(*cp_page, 1);
883 		f2fs_warn(sbi, "invalid crc value");
884 		return -EINVAL;
885 	}
886 
887 	*version = cur_cp_version(*cp_block);
888 	return 0;
889 }
890 
891 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
892 				block_t cp_addr, unsigned long long *version)
893 {
894 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
895 	struct f2fs_checkpoint *cp_block = NULL;
896 	unsigned long long cur_version = 0, pre_version = 0;
897 	unsigned int cp_blocks;
898 	int err;
899 
900 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
901 					&cp_page_1, version);
902 	if (err)
903 		return NULL;
904 
905 	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
906 
907 	if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
908 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
909 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
910 		goto invalid_cp;
911 	}
912 	pre_version = *version;
913 
914 	cp_addr += cp_blocks - 1;
915 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
916 					&cp_page_2, version);
917 	if (err)
918 		goto invalid_cp;
919 	cur_version = *version;
920 
921 	if (cur_version == pre_version) {
922 		*version = cur_version;
923 		f2fs_put_page(cp_page_2, 1);
924 		return cp_page_1;
925 	}
926 	f2fs_put_page(cp_page_2, 1);
927 invalid_cp:
928 	f2fs_put_page(cp_page_1, 1);
929 	return NULL;
930 }
931 
932 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
933 {
934 	struct f2fs_checkpoint *cp_block;
935 	struct f2fs_super_block *fsb = sbi->raw_super;
936 	struct page *cp1, *cp2, *cur_page;
937 	unsigned long blk_size = sbi->blocksize;
938 	unsigned long long cp1_version = 0, cp2_version = 0;
939 	unsigned long long cp_start_blk_no;
940 	unsigned int cp_blks = 1 + __cp_payload(sbi);
941 	block_t cp_blk_no;
942 	int i;
943 	int err;
944 
945 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
946 				  GFP_KERNEL);
947 	if (!sbi->ckpt)
948 		return -ENOMEM;
949 	/*
950 	 * Finding out valid cp block involves read both
951 	 * sets( cp pack 1 and cp pack 2)
952 	 */
953 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
954 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
955 
956 	/* The second checkpoint pack should start at the next segment */
957 	cp_start_blk_no += ((unsigned long long)1) <<
958 				le32_to_cpu(fsb->log_blocks_per_seg);
959 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
960 
961 	if (cp1 && cp2) {
962 		if (ver_after(cp2_version, cp1_version))
963 			cur_page = cp2;
964 		else
965 			cur_page = cp1;
966 	} else if (cp1) {
967 		cur_page = cp1;
968 	} else if (cp2) {
969 		cur_page = cp2;
970 	} else {
971 		err = -EFSCORRUPTED;
972 		goto fail_no_cp;
973 	}
974 
975 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
976 	memcpy(sbi->ckpt, cp_block, blk_size);
977 
978 	if (cur_page == cp1)
979 		sbi->cur_cp_pack = 1;
980 	else
981 		sbi->cur_cp_pack = 2;
982 
983 	/* Sanity checking of checkpoint */
984 	if (f2fs_sanity_check_ckpt(sbi)) {
985 		err = -EFSCORRUPTED;
986 		goto free_fail_no_cp;
987 	}
988 
989 	if (cp_blks <= 1)
990 		goto done;
991 
992 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
993 	if (cur_page == cp2)
994 		cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
995 
996 	for (i = 1; i < cp_blks; i++) {
997 		void *sit_bitmap_ptr;
998 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
999 
1000 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
1001 		if (IS_ERR(cur_page)) {
1002 			err = PTR_ERR(cur_page);
1003 			goto free_fail_no_cp;
1004 		}
1005 		sit_bitmap_ptr = page_address(cur_page);
1006 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1007 		f2fs_put_page(cur_page, 1);
1008 	}
1009 done:
1010 	f2fs_put_page(cp1, 1);
1011 	f2fs_put_page(cp2, 1);
1012 	return 0;
1013 
1014 free_fail_no_cp:
1015 	f2fs_put_page(cp1, 1);
1016 	f2fs_put_page(cp2, 1);
1017 fail_no_cp:
1018 	kvfree(sbi->ckpt);
1019 	return err;
1020 }
1021 
1022 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1023 {
1024 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1025 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1026 
1027 	if (is_inode_flag_set(inode, flag))
1028 		return;
1029 
1030 	set_inode_flag(inode, flag);
1031 	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1032 	stat_inc_dirty_inode(sbi, type);
1033 }
1034 
1035 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1036 {
1037 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1038 
1039 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1040 		return;
1041 
1042 	list_del_init(&F2FS_I(inode)->dirty_list);
1043 	clear_inode_flag(inode, flag);
1044 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1045 }
1046 
1047 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1048 {
1049 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1050 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1051 
1052 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1053 			!S_ISLNK(inode->i_mode))
1054 		return;
1055 
1056 	spin_lock(&sbi->inode_lock[type]);
1057 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1058 		__add_dirty_inode(inode, type);
1059 	inode_inc_dirty_pages(inode);
1060 	spin_unlock(&sbi->inode_lock[type]);
1061 
1062 	set_page_private_reference(&folio->page);
1063 }
1064 
1065 void f2fs_remove_dirty_inode(struct inode *inode)
1066 {
1067 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1068 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1069 
1070 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1071 			!S_ISLNK(inode->i_mode))
1072 		return;
1073 
1074 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1075 		return;
1076 
1077 	spin_lock(&sbi->inode_lock[type]);
1078 	__remove_dirty_inode(inode, type);
1079 	spin_unlock(&sbi->inode_lock[type]);
1080 }
1081 
1082 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1083 						bool from_cp)
1084 {
1085 	struct list_head *head;
1086 	struct inode *inode;
1087 	struct f2fs_inode_info *fi;
1088 	bool is_dir = (type == DIR_INODE);
1089 	unsigned long ino = 0;
1090 
1091 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1092 				get_pages(sbi, is_dir ?
1093 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1094 retry:
1095 	if (unlikely(f2fs_cp_error(sbi))) {
1096 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1097 				get_pages(sbi, is_dir ?
1098 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1099 		return -EIO;
1100 	}
1101 
1102 	spin_lock(&sbi->inode_lock[type]);
1103 
1104 	head = &sbi->inode_list[type];
1105 	if (list_empty(head)) {
1106 		spin_unlock(&sbi->inode_lock[type]);
1107 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1108 				get_pages(sbi, is_dir ?
1109 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1110 		return 0;
1111 	}
1112 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1113 	inode = igrab(&fi->vfs_inode);
1114 	spin_unlock(&sbi->inode_lock[type]);
1115 	if (inode) {
1116 		unsigned long cur_ino = inode->i_ino;
1117 
1118 		if (from_cp)
1119 			F2FS_I(inode)->cp_task = current;
1120 		F2FS_I(inode)->wb_task = current;
1121 
1122 		filemap_fdatawrite(inode->i_mapping);
1123 
1124 		F2FS_I(inode)->wb_task = NULL;
1125 		if (from_cp)
1126 			F2FS_I(inode)->cp_task = NULL;
1127 
1128 		iput(inode);
1129 		/* We need to give cpu to another writers. */
1130 		if (ino == cur_ino)
1131 			cond_resched();
1132 		else
1133 			ino = cur_ino;
1134 	} else {
1135 		/*
1136 		 * We should submit bio, since it exists several
1137 		 * writebacking dentry pages in the freeing inode.
1138 		 */
1139 		f2fs_submit_merged_write(sbi, DATA);
1140 		cond_resched();
1141 	}
1142 	goto retry;
1143 }
1144 
1145 static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1146 {
1147 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1148 	struct inode *inode;
1149 	struct f2fs_inode_info *fi;
1150 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1151 
1152 	while (total--) {
1153 		if (unlikely(f2fs_cp_error(sbi)))
1154 			return -EIO;
1155 
1156 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1157 		if (list_empty(head)) {
1158 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1159 			return 0;
1160 		}
1161 		fi = list_first_entry(head, struct f2fs_inode_info,
1162 							gdirty_list);
1163 		inode = igrab(&fi->vfs_inode);
1164 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1165 		if (inode) {
1166 			sync_inode_metadata(inode, 0);
1167 
1168 			/* it's on eviction */
1169 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1170 				f2fs_update_inode_page(inode);
1171 			iput(inode);
1172 		}
1173 	}
1174 	return 0;
1175 }
1176 
1177 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1178 {
1179 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1180 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1181 	nid_t last_nid = nm_i->next_scan_nid;
1182 
1183 	next_free_nid(sbi, &last_nid);
1184 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1185 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1186 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1187 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1188 }
1189 
1190 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1191 {
1192 	bool ret = false;
1193 
1194 	if (!is_journalled_quota(sbi))
1195 		return false;
1196 
1197 	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1198 		return true;
1199 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1200 		ret = false;
1201 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1202 		ret = false;
1203 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1204 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1205 		ret = true;
1206 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1207 		ret = true;
1208 	}
1209 	f2fs_up_write(&sbi->quota_sem);
1210 	return ret;
1211 }
1212 
1213 /*
1214  * Freeze all the FS-operations for checkpoint.
1215  */
1216 static int block_operations(struct f2fs_sb_info *sbi)
1217 {
1218 	struct writeback_control wbc = {
1219 		.sync_mode = WB_SYNC_ALL,
1220 		.nr_to_write = LONG_MAX,
1221 		.for_reclaim = 0,
1222 	};
1223 	int err = 0, cnt = 0;
1224 
1225 	/*
1226 	 * Let's flush inline_data in dirty node pages.
1227 	 */
1228 	f2fs_flush_inline_data(sbi);
1229 
1230 retry_flush_quotas:
1231 	f2fs_lock_all(sbi);
1232 	if (__need_flush_quota(sbi)) {
1233 		int locked;
1234 
1235 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1236 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1237 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1238 			goto retry_flush_dents;
1239 		}
1240 		f2fs_unlock_all(sbi);
1241 
1242 		/* only failed during mount/umount/freeze/quotactl */
1243 		locked = down_read_trylock(&sbi->sb->s_umount);
1244 		f2fs_quota_sync(sbi->sb, -1);
1245 		if (locked)
1246 			up_read(&sbi->sb->s_umount);
1247 		cond_resched();
1248 		goto retry_flush_quotas;
1249 	}
1250 
1251 retry_flush_dents:
1252 	/* write all the dirty dentry pages */
1253 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1254 		f2fs_unlock_all(sbi);
1255 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1256 		if (err)
1257 			return err;
1258 		cond_resched();
1259 		goto retry_flush_quotas;
1260 	}
1261 
1262 	/*
1263 	 * POR: we should ensure that there are no dirty node pages
1264 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1265 	 */
1266 	f2fs_down_write(&sbi->node_change);
1267 
1268 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1269 		f2fs_up_write(&sbi->node_change);
1270 		f2fs_unlock_all(sbi);
1271 		err = f2fs_sync_inode_meta(sbi);
1272 		if (err)
1273 			return err;
1274 		cond_resched();
1275 		goto retry_flush_quotas;
1276 	}
1277 
1278 retry_flush_nodes:
1279 	f2fs_down_write(&sbi->node_write);
1280 
1281 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1282 		f2fs_up_write(&sbi->node_write);
1283 		atomic_inc(&sbi->wb_sync_req[NODE]);
1284 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1285 		atomic_dec(&sbi->wb_sync_req[NODE]);
1286 		if (err) {
1287 			f2fs_up_write(&sbi->node_change);
1288 			f2fs_unlock_all(sbi);
1289 			return err;
1290 		}
1291 		cond_resched();
1292 		goto retry_flush_nodes;
1293 	}
1294 
1295 	/*
1296 	 * sbi->node_change is used only for AIO write_begin path which produces
1297 	 * dirty node blocks and some checkpoint values by block allocation.
1298 	 */
1299 	__prepare_cp_block(sbi);
1300 	f2fs_up_write(&sbi->node_change);
1301 	return err;
1302 }
1303 
1304 static void unblock_operations(struct f2fs_sb_info *sbi)
1305 {
1306 	f2fs_up_write(&sbi->node_write);
1307 	f2fs_unlock_all(sbi);
1308 }
1309 
1310 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1311 {
1312 	DEFINE_WAIT(wait);
1313 
1314 	for (;;) {
1315 		if (!get_pages(sbi, type))
1316 			break;
1317 
1318 		if (unlikely(f2fs_cp_error(sbi) &&
1319 			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1320 			break;
1321 
1322 		if (type == F2FS_DIRTY_META)
1323 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1324 							FS_CP_META_IO);
1325 		else if (type == F2FS_WB_CP_DATA)
1326 			f2fs_submit_merged_write(sbi, DATA);
1327 
1328 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1329 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1330 	}
1331 	finish_wait(&sbi->cp_wait, &wait);
1332 }
1333 
1334 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1335 {
1336 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1337 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1338 	unsigned long flags;
1339 
1340 	if (cpc->reason & CP_UMOUNT) {
1341 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1342 			NM_I(sbi)->nat_bits_blocks > BLKS_PER_SEG(sbi)) {
1343 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1344 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1345 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1346 						f2fs_nat_bitmap_enabled(sbi)) {
1347 			f2fs_enable_nat_bits(sbi);
1348 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1349 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1350 		}
1351 	}
1352 
1353 	spin_lock_irqsave(&sbi->cp_lock, flags);
1354 
1355 	if (cpc->reason & CP_TRIMMED)
1356 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1357 	else
1358 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1359 
1360 	if (cpc->reason & CP_UMOUNT)
1361 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1362 	else
1363 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1364 
1365 	if (cpc->reason & CP_FASTBOOT)
1366 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1367 	else
1368 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1369 
1370 	if (orphan_num)
1371 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1372 	else
1373 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1374 
1375 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1376 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1377 
1378 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1379 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1380 	else
1381 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1382 
1383 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1384 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1385 	else
1386 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1387 
1388 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1389 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1390 	else
1391 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1392 
1393 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1394 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1395 	else
1396 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1397 
1398 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1399 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1400 
1401 	/* set this flag to activate crc|cp_ver for recovery */
1402 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1403 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1404 
1405 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1406 }
1407 
1408 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1409 	void *src, block_t blk_addr)
1410 {
1411 	struct writeback_control wbc = {
1412 		.for_reclaim = 0,
1413 	};
1414 
1415 	/*
1416 	 * filemap_get_folios_tag and lock_page again will take
1417 	 * some extra time. Therefore, f2fs_update_meta_pages and
1418 	 * f2fs_sync_meta_pages are combined in this function.
1419 	 */
1420 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1421 	int err;
1422 
1423 	f2fs_wait_on_page_writeback(page, META, true, true);
1424 
1425 	memcpy(page_address(page), src, PAGE_SIZE);
1426 
1427 	set_page_dirty(page);
1428 	if (unlikely(!clear_page_dirty_for_io(page)))
1429 		f2fs_bug_on(sbi, 1);
1430 
1431 	/* writeout cp pack 2 page */
1432 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1433 	if (unlikely(err && f2fs_cp_error(sbi))) {
1434 		f2fs_put_page(page, 1);
1435 		return;
1436 	}
1437 
1438 	f2fs_bug_on(sbi, err);
1439 	f2fs_put_page(page, 0);
1440 
1441 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1442 	f2fs_submit_merged_write(sbi, META_FLUSH);
1443 }
1444 
1445 static inline u64 get_sectors_written(struct block_device *bdev)
1446 {
1447 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1448 }
1449 
1450 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1451 {
1452 	if (f2fs_is_multi_device(sbi)) {
1453 		u64 sectors = 0;
1454 		int i;
1455 
1456 		for (i = 0; i < sbi->s_ndevs; i++)
1457 			sectors += get_sectors_written(FDEV(i).bdev);
1458 
1459 		return sectors;
1460 	}
1461 
1462 	return get_sectors_written(sbi->sb->s_bdev);
1463 }
1464 
1465 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1466 {
1467 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1468 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1469 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1470 	block_t start_blk;
1471 	unsigned int data_sum_blocks, orphan_blocks;
1472 	__u32 crc32 = 0;
1473 	int i;
1474 	int cp_payload_blks = __cp_payload(sbi);
1475 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1476 	u64 kbytes_written;
1477 	int err;
1478 
1479 	/* Flush all the NAT/SIT pages */
1480 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1481 
1482 	/* start to update checkpoint, cp ver is already updated previously */
1483 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1484 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1485 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1486 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1487 
1488 		ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1489 		ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1490 		ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1491 	}
1492 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1493 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1494 
1495 		ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1496 		ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1497 		ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1498 	}
1499 
1500 	/* 2 cp + n data seg summary + orphan inode blocks */
1501 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1502 	spin_lock_irqsave(&sbi->cp_lock, flags);
1503 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1504 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1505 	else
1506 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1507 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1508 
1509 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1510 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1511 			orphan_blocks);
1512 
1513 	if (__remain_node_summaries(cpc->reason))
1514 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1515 				cp_payload_blks + data_sum_blocks +
1516 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1517 	else
1518 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1519 				cp_payload_blks + data_sum_blocks +
1520 				orphan_blocks);
1521 
1522 	/* update ckpt flag for checkpoint */
1523 	update_ckpt_flags(sbi, cpc);
1524 
1525 	/* update SIT/NAT bitmap */
1526 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1527 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1528 
1529 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1530 	*((__le32 *)((unsigned char *)ckpt +
1531 				le32_to_cpu(ckpt->checksum_offset)))
1532 				= cpu_to_le32(crc32);
1533 
1534 	start_blk = __start_cp_next_addr(sbi);
1535 
1536 	/* write nat bits */
1537 	if ((cpc->reason & CP_UMOUNT) &&
1538 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1539 		__u64 cp_ver = cur_cp_version(ckpt);
1540 		block_t blk;
1541 
1542 		cp_ver |= ((__u64)crc32 << 32);
1543 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1544 
1545 		blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1546 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1547 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1548 					(i << F2FS_BLKSIZE_BITS), blk + i);
1549 	}
1550 
1551 	/* write out checkpoint buffer at block 0 */
1552 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1553 
1554 	for (i = 1; i < 1 + cp_payload_blks; i++)
1555 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1556 							start_blk++);
1557 
1558 	if (orphan_num) {
1559 		write_orphan_inodes(sbi, start_blk);
1560 		start_blk += orphan_blocks;
1561 	}
1562 
1563 	f2fs_write_data_summaries(sbi, start_blk);
1564 	start_blk += data_sum_blocks;
1565 
1566 	/* Record write statistics in the hot node summary */
1567 	kbytes_written = sbi->kbytes_written;
1568 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1569 				sbi->sectors_written_start) >> 1;
1570 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1571 
1572 	if (__remain_node_summaries(cpc->reason)) {
1573 		f2fs_write_node_summaries(sbi, start_blk);
1574 		start_blk += NR_CURSEG_NODE_TYPE;
1575 	}
1576 
1577 	/* update user_block_counts */
1578 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1579 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1580 	percpu_counter_set(&sbi->rf_node_block_count, 0);
1581 
1582 	/* Here, we have one bio having CP pack except cp pack 2 page */
1583 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1584 	/* Wait for all dirty meta pages to be submitted for IO */
1585 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1586 
1587 	/* wait for previous submitted meta pages writeback */
1588 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1589 
1590 	/* flush all device cache */
1591 	err = f2fs_flush_device_cache(sbi);
1592 	if (err)
1593 		return err;
1594 
1595 	/* barrier and flush checkpoint cp pack 2 page if it can */
1596 	commit_checkpoint(sbi, ckpt, start_blk);
1597 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1598 
1599 	/*
1600 	 * invalidate intermediate page cache borrowed from meta inode which are
1601 	 * used for migration of encrypted, verity or compressed inode's blocks.
1602 	 */
1603 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1604 		f2fs_sb_has_compression(sbi))
1605 		f2fs_bug_on(sbi,
1606 			invalidate_inode_pages2_range(META_MAPPING(sbi),
1607 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1608 
1609 	f2fs_release_ino_entry(sbi, false);
1610 
1611 	f2fs_reset_fsync_node_info(sbi);
1612 
1613 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1614 	clear_sbi_flag(sbi, SBI_NEED_CP);
1615 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1616 
1617 	spin_lock(&sbi->stat_lock);
1618 	sbi->unusable_block_count = 0;
1619 	spin_unlock(&sbi->stat_lock);
1620 
1621 	__set_cp_next_pack(sbi);
1622 
1623 	/*
1624 	 * redirty superblock if metadata like node page or inode cache is
1625 	 * updated during writing checkpoint.
1626 	 */
1627 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1628 			get_pages(sbi, F2FS_DIRTY_IMETA))
1629 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1630 
1631 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1632 
1633 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1634 }
1635 
1636 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1637 {
1638 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1639 	unsigned long long ckpt_ver;
1640 	int err = 0;
1641 
1642 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1643 		return -EROFS;
1644 
1645 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1646 		if (cpc->reason != CP_PAUSE)
1647 			return 0;
1648 		f2fs_warn(sbi, "Start checkpoint disabled!");
1649 	}
1650 	if (cpc->reason != CP_RESIZE)
1651 		f2fs_down_write(&sbi->cp_global_sem);
1652 
1653 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1654 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1655 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1656 		goto out;
1657 	if (unlikely(f2fs_cp_error(sbi))) {
1658 		err = -EIO;
1659 		goto out;
1660 	}
1661 
1662 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1663 
1664 	err = block_operations(sbi);
1665 	if (err)
1666 		goto out;
1667 
1668 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1669 
1670 	f2fs_flush_merged_writes(sbi);
1671 
1672 	/* this is the case of multiple fstrims without any changes */
1673 	if (cpc->reason & CP_DISCARD) {
1674 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1675 			unblock_operations(sbi);
1676 			goto out;
1677 		}
1678 
1679 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1680 				SIT_I(sbi)->dirty_sentries == 0 &&
1681 				prefree_segments(sbi) == 0) {
1682 			f2fs_flush_sit_entries(sbi, cpc);
1683 			f2fs_clear_prefree_segments(sbi, cpc);
1684 			unblock_operations(sbi);
1685 			goto out;
1686 		}
1687 	}
1688 
1689 	/*
1690 	 * update checkpoint pack index
1691 	 * Increase the version number so that
1692 	 * SIT entries and seg summaries are written at correct place
1693 	 */
1694 	ckpt_ver = cur_cp_version(ckpt);
1695 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1696 
1697 	/* write cached NAT/SIT entries to NAT/SIT area */
1698 	err = f2fs_flush_nat_entries(sbi, cpc);
1699 	if (err) {
1700 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1701 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1702 		goto stop;
1703 	}
1704 
1705 	f2fs_flush_sit_entries(sbi, cpc);
1706 
1707 	/* save inmem log status */
1708 	f2fs_save_inmem_curseg(sbi);
1709 
1710 	err = do_checkpoint(sbi, cpc);
1711 	if (err) {
1712 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1713 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1714 		f2fs_release_discard_addrs(sbi);
1715 	} else {
1716 		f2fs_clear_prefree_segments(sbi, cpc);
1717 	}
1718 
1719 	f2fs_restore_inmem_curseg(sbi);
1720 	stat_inc_cp_count(sbi);
1721 stop:
1722 	unblock_operations(sbi);
1723 
1724 	if (cpc->reason & CP_RECOVERY)
1725 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1726 
1727 	/* update CP_TIME to trigger checkpoint periodically */
1728 	f2fs_update_time(sbi, CP_TIME);
1729 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1730 out:
1731 	if (cpc->reason != CP_RESIZE)
1732 		f2fs_up_write(&sbi->cp_global_sem);
1733 	return err;
1734 }
1735 
1736 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1737 {
1738 	int i;
1739 
1740 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1741 		struct inode_management *im = &sbi->im[i];
1742 
1743 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1744 		spin_lock_init(&im->ino_lock);
1745 		INIT_LIST_HEAD(&im->ino_list);
1746 		im->ino_num = 0;
1747 	}
1748 
1749 	sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1750 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1751 			F2FS_ORPHANS_PER_BLOCK;
1752 }
1753 
1754 int __init f2fs_create_checkpoint_caches(void)
1755 {
1756 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1757 			sizeof(struct ino_entry));
1758 	if (!ino_entry_slab)
1759 		return -ENOMEM;
1760 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1761 			sizeof(struct inode_entry));
1762 	if (!f2fs_inode_entry_slab) {
1763 		kmem_cache_destroy(ino_entry_slab);
1764 		return -ENOMEM;
1765 	}
1766 	return 0;
1767 }
1768 
1769 void f2fs_destroy_checkpoint_caches(void)
1770 {
1771 	kmem_cache_destroy(ino_entry_slab);
1772 	kmem_cache_destroy(f2fs_inode_entry_slab);
1773 }
1774 
1775 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1776 {
1777 	struct cp_control cpc = { .reason = CP_SYNC, };
1778 	int err;
1779 
1780 	f2fs_down_write(&sbi->gc_lock);
1781 	err = f2fs_write_checkpoint(sbi, &cpc);
1782 	f2fs_up_write(&sbi->gc_lock);
1783 
1784 	return err;
1785 }
1786 
1787 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1788 {
1789 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1790 	struct ckpt_req *req, *next;
1791 	struct llist_node *dispatch_list;
1792 	u64 sum_diff = 0, diff, count = 0;
1793 	int ret;
1794 
1795 	dispatch_list = llist_del_all(&cprc->issue_list);
1796 	if (!dispatch_list)
1797 		return;
1798 	dispatch_list = llist_reverse_order(dispatch_list);
1799 
1800 	ret = __write_checkpoint_sync(sbi);
1801 	atomic_inc(&cprc->issued_ckpt);
1802 
1803 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1804 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1805 		req->ret = ret;
1806 		complete(&req->wait);
1807 
1808 		sum_diff += diff;
1809 		count++;
1810 	}
1811 	atomic_sub(count, &cprc->queued_ckpt);
1812 	atomic_add(count, &cprc->total_ckpt);
1813 
1814 	spin_lock(&cprc->stat_lock);
1815 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1816 	if (cprc->peak_time < cprc->cur_time)
1817 		cprc->peak_time = cprc->cur_time;
1818 	spin_unlock(&cprc->stat_lock);
1819 }
1820 
1821 static int issue_checkpoint_thread(void *data)
1822 {
1823 	struct f2fs_sb_info *sbi = data;
1824 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1825 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1826 repeat:
1827 	if (kthread_should_stop())
1828 		return 0;
1829 
1830 	if (!llist_empty(&cprc->issue_list))
1831 		__checkpoint_and_complete_reqs(sbi);
1832 
1833 	wait_event_interruptible(*q,
1834 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1835 	goto repeat;
1836 }
1837 
1838 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1839 		struct ckpt_req *wait_req)
1840 {
1841 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1842 
1843 	if (!llist_empty(&cprc->issue_list)) {
1844 		__checkpoint_and_complete_reqs(sbi);
1845 	} else {
1846 		/* already dispatched by issue_checkpoint_thread */
1847 		if (wait_req)
1848 			wait_for_completion(&wait_req->wait);
1849 	}
1850 }
1851 
1852 static void init_ckpt_req(struct ckpt_req *req)
1853 {
1854 	memset(req, 0, sizeof(struct ckpt_req));
1855 
1856 	init_completion(&req->wait);
1857 	req->queue_time = ktime_get();
1858 }
1859 
1860 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1861 {
1862 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1863 	struct ckpt_req req;
1864 	struct cp_control cpc;
1865 
1866 	cpc.reason = __get_cp_reason(sbi);
1867 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1868 		int ret;
1869 
1870 		f2fs_down_write(&sbi->gc_lock);
1871 		ret = f2fs_write_checkpoint(sbi, &cpc);
1872 		f2fs_up_write(&sbi->gc_lock);
1873 
1874 		return ret;
1875 	}
1876 
1877 	if (!cprc->f2fs_issue_ckpt)
1878 		return __write_checkpoint_sync(sbi);
1879 
1880 	init_ckpt_req(&req);
1881 
1882 	llist_add(&req.llnode, &cprc->issue_list);
1883 	atomic_inc(&cprc->queued_ckpt);
1884 
1885 	/*
1886 	 * update issue_list before we wake up issue_checkpoint thread,
1887 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1888 	 * see more details in comments of waitqueue_active().
1889 	 */
1890 	smp_mb();
1891 
1892 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1893 		wake_up(&cprc->ckpt_wait_queue);
1894 
1895 	if (cprc->f2fs_issue_ckpt)
1896 		wait_for_completion(&req.wait);
1897 	else
1898 		flush_remained_ckpt_reqs(sbi, &req);
1899 
1900 	return req.ret;
1901 }
1902 
1903 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1904 {
1905 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1906 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1907 
1908 	if (cprc->f2fs_issue_ckpt)
1909 		return 0;
1910 
1911 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1912 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1913 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1914 		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1915 
1916 		cprc->f2fs_issue_ckpt = NULL;
1917 		return err;
1918 	}
1919 
1920 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1921 
1922 	return 0;
1923 }
1924 
1925 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1926 {
1927 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1928 	struct task_struct *ckpt_task;
1929 
1930 	if (!cprc->f2fs_issue_ckpt)
1931 		return;
1932 
1933 	ckpt_task = cprc->f2fs_issue_ckpt;
1934 	cprc->f2fs_issue_ckpt = NULL;
1935 	kthread_stop(ckpt_task);
1936 
1937 	f2fs_flush_ckpt_thread(sbi);
1938 }
1939 
1940 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1941 {
1942 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1943 
1944 	flush_remained_ckpt_reqs(sbi, NULL);
1945 
1946 	/* Let's wait for the previous dispatched checkpoint. */
1947 	while (atomic_read(&cprc->queued_ckpt))
1948 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1949 }
1950 
1951 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1952 {
1953 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1954 
1955 	atomic_set(&cprc->issued_ckpt, 0);
1956 	atomic_set(&cprc->total_ckpt, 0);
1957 	atomic_set(&cprc->queued_ckpt, 0);
1958 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1959 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1960 	init_llist_head(&cprc->issue_list);
1961 	spin_lock_init(&cprc->stat_lock);
1962 }
1963