xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision 3db55767)
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 ? 1 : 0,
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 
156 	/* skip data, if we already have an error in checkpoint. */
157 	if (unlikely(f2fs_cp_error(sbi)))
158 		return exist;
159 
160 	if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
161 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
162 			 blkaddr, exist);
163 		set_sbi_flag(sbi, SBI_NEED_FSCK);
164 		return exist;
165 	}
166 
167 	if (!exist && type == DATA_GENERIC_ENHANCE) {
168 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
169 			 blkaddr, exist);
170 		set_sbi_flag(sbi, SBI_NEED_FSCK);
171 		dump_stack();
172 	}
173 	return exist;
174 }
175 
176 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
177 					block_t blkaddr, int type)
178 {
179 	if (time_to_inject(sbi, FAULT_BLKADDR))
180 		return false;
181 
182 	switch (type) {
183 	case META_NAT:
184 		break;
185 	case META_SIT:
186 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
187 			return false;
188 		break;
189 	case META_SSA:
190 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
191 			blkaddr < SM_I(sbi)->ssa_blkaddr))
192 			return false;
193 		break;
194 	case META_CP:
195 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
196 			blkaddr < __start_cp_addr(sbi)))
197 			return false;
198 		break;
199 	case META_POR:
200 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
201 			blkaddr < MAIN_BLKADDR(sbi)))
202 			return false;
203 		break;
204 	case DATA_GENERIC:
205 	case DATA_GENERIC_ENHANCE:
206 	case DATA_GENERIC_ENHANCE_READ:
207 	case DATA_GENERIC_ENHANCE_UPDATE:
208 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
209 				blkaddr < MAIN_BLKADDR(sbi))) {
210 
211 			/* Skip to emit an error message. */
212 			if (unlikely(f2fs_cp_error(sbi)))
213 				return false;
214 
215 			f2fs_warn(sbi, "access invalid blkaddr:%u",
216 				  blkaddr);
217 			set_sbi_flag(sbi, SBI_NEED_FSCK);
218 			dump_stack();
219 			return false;
220 		} else {
221 			return __is_bitmap_valid(sbi, blkaddr, type);
222 		}
223 		break;
224 	case META_GENERIC:
225 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
226 			blkaddr >= MAIN_BLKADDR(sbi)))
227 			return false;
228 		break;
229 	default:
230 		BUG();
231 	}
232 
233 	return true;
234 }
235 
236 /*
237  * Readahead CP/NAT/SIT/SSA/POR pages
238  */
239 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
240 							int type, bool sync)
241 {
242 	struct page *page;
243 	block_t blkno = start;
244 	struct f2fs_io_info fio = {
245 		.sbi = sbi,
246 		.type = META,
247 		.op = REQ_OP_READ,
248 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
249 		.encrypted_page = NULL,
250 		.in_list = 0,
251 		.is_por = (type == META_POR) ? 1 : 0,
252 	};
253 	struct blk_plug plug;
254 	int err;
255 
256 	if (unlikely(type == META_POR))
257 		fio.op_flags &= ~REQ_META;
258 
259 	blk_start_plug(&plug);
260 	for (; nrpages-- > 0; blkno++) {
261 
262 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
263 			goto out;
264 
265 		switch (type) {
266 		case META_NAT:
267 			if (unlikely(blkno >=
268 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
269 				blkno = 0;
270 			/* get nat block addr */
271 			fio.new_blkaddr = current_nat_addr(sbi,
272 					blkno * NAT_ENTRY_PER_BLOCK);
273 			break;
274 		case META_SIT:
275 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
276 				goto out;
277 			/* get sit block addr */
278 			fio.new_blkaddr = current_sit_addr(sbi,
279 					blkno * SIT_ENTRY_PER_BLOCK);
280 			break;
281 		case META_SSA:
282 		case META_CP:
283 		case META_POR:
284 			fio.new_blkaddr = blkno;
285 			break;
286 		default:
287 			BUG();
288 		}
289 
290 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
291 						fio.new_blkaddr, false);
292 		if (!page)
293 			continue;
294 		if (PageUptodate(page)) {
295 			f2fs_put_page(page, 1);
296 			continue;
297 		}
298 
299 		fio.page = page;
300 		err = f2fs_submit_page_bio(&fio);
301 		f2fs_put_page(page, err ? 1 : 0);
302 
303 		if (!err)
304 			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
305 							F2FS_BLKSIZE);
306 	}
307 out:
308 	blk_finish_plug(&plug);
309 	return blkno - start;
310 }
311 
312 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
313 							unsigned int ra_blocks)
314 {
315 	struct page *page;
316 	bool readahead = false;
317 
318 	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
319 		return;
320 
321 	page = find_get_page(META_MAPPING(sbi), index);
322 	if (!page || !PageUptodate(page))
323 		readahead = true;
324 	f2fs_put_page(page, 0);
325 
326 	if (readahead)
327 		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
328 }
329 
330 static int __f2fs_write_meta_page(struct page *page,
331 				struct writeback_control *wbc,
332 				enum iostat_type io_type)
333 {
334 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
335 
336 	trace_f2fs_writepage(page, META);
337 
338 	if (unlikely(f2fs_cp_error(sbi))) {
339 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
340 			ClearPageUptodate(page);
341 			dec_page_count(sbi, F2FS_DIRTY_META);
342 			unlock_page(page);
343 			return 0;
344 		}
345 		goto redirty_out;
346 	}
347 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
348 		goto redirty_out;
349 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
350 		goto redirty_out;
351 
352 	f2fs_do_write_meta_page(sbi, page, io_type);
353 	dec_page_count(sbi, F2FS_DIRTY_META);
354 
355 	if (wbc->for_reclaim)
356 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
357 
358 	unlock_page(page);
359 
360 	if (unlikely(f2fs_cp_error(sbi)))
361 		f2fs_submit_merged_write(sbi, META);
362 
363 	return 0;
364 
365 redirty_out:
366 	redirty_page_for_writepage(wbc, page);
367 	return AOP_WRITEPAGE_ACTIVATE;
368 }
369 
370 static int f2fs_write_meta_page(struct page *page,
371 				struct writeback_control *wbc)
372 {
373 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
374 }
375 
376 static int f2fs_write_meta_pages(struct address_space *mapping,
377 				struct writeback_control *wbc)
378 {
379 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
380 	long diff, written;
381 
382 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
383 		goto skip_write;
384 
385 	/* collect a number of dirty meta pages and write together */
386 	if (wbc->sync_mode != WB_SYNC_ALL &&
387 			get_pages(sbi, F2FS_DIRTY_META) <
388 					nr_pages_to_skip(sbi, META))
389 		goto skip_write;
390 
391 	/* if locked failed, cp will flush dirty pages instead */
392 	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
393 		goto skip_write;
394 
395 	trace_f2fs_writepages(mapping->host, wbc, META);
396 	diff = nr_pages_to_write(sbi, META, wbc);
397 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
398 	f2fs_up_write(&sbi->cp_global_sem);
399 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
400 	return 0;
401 
402 skip_write:
403 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
404 	trace_f2fs_writepages(mapping->host, wbc, META);
405 	return 0;
406 }
407 
408 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
409 				long nr_to_write, enum iostat_type io_type)
410 {
411 	struct address_space *mapping = META_MAPPING(sbi);
412 	pgoff_t index = 0, prev = ULONG_MAX;
413 	struct folio_batch fbatch;
414 	long nwritten = 0;
415 	int nr_folios;
416 	struct writeback_control wbc = {
417 		.for_reclaim = 0,
418 	};
419 	struct blk_plug plug;
420 
421 	folio_batch_init(&fbatch);
422 
423 	blk_start_plug(&plug);
424 
425 	while ((nr_folios = filemap_get_folios_tag(mapping, &index,
426 					(pgoff_t)-1,
427 					PAGECACHE_TAG_DIRTY, &fbatch))) {
428 		int i;
429 
430 		for (i = 0; i < nr_folios; i++) {
431 			struct folio *folio = fbatch.folios[i];
432 
433 			if (nr_to_write != LONG_MAX && i != 0 &&
434 					folio->index != prev +
435 					folio_nr_pages(fbatch.folios[i-1])) {
436 				folio_batch_release(&fbatch);
437 				goto stop;
438 			}
439 
440 			folio_lock(folio);
441 
442 			if (unlikely(folio->mapping != mapping)) {
443 continue_unlock:
444 				folio_unlock(folio);
445 				continue;
446 			}
447 			if (!folio_test_dirty(folio)) {
448 				/* someone wrote it for us */
449 				goto continue_unlock;
450 			}
451 
452 			f2fs_wait_on_page_writeback(&folio->page, META,
453 					true, true);
454 
455 			if (!folio_clear_dirty_for_io(folio))
456 				goto continue_unlock;
457 
458 			if (__f2fs_write_meta_page(&folio->page, &wbc,
459 						io_type)) {
460 				folio_unlock(folio);
461 				break;
462 			}
463 			nwritten += folio_nr_pages(folio);
464 			prev = folio->index;
465 			if (unlikely(nwritten >= nr_to_write))
466 				break;
467 		}
468 		folio_batch_release(&fbatch);
469 		cond_resched();
470 	}
471 stop:
472 	if (nwritten)
473 		f2fs_submit_merged_write(sbi, type);
474 
475 	blk_finish_plug(&plug);
476 
477 	return nwritten;
478 }
479 
480 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
481 		struct folio *folio)
482 {
483 	trace_f2fs_set_page_dirty(&folio->page, META);
484 
485 	if (!folio_test_uptodate(folio))
486 		folio_mark_uptodate(folio);
487 	if (filemap_dirty_folio(mapping, folio)) {
488 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
489 		set_page_private_reference(&folio->page);
490 		return true;
491 	}
492 	return false;
493 }
494 
495 const struct address_space_operations f2fs_meta_aops = {
496 	.writepage	= f2fs_write_meta_page,
497 	.writepages	= f2fs_write_meta_pages,
498 	.dirty_folio	= f2fs_dirty_meta_folio,
499 	.invalidate_folio = f2fs_invalidate_folio,
500 	.release_folio	= f2fs_release_folio,
501 	.migrate_folio	= filemap_migrate_folio,
502 };
503 
504 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
505 						unsigned int devidx, int type)
506 {
507 	struct inode_management *im = &sbi->im[type];
508 	struct ino_entry *e = NULL, *new = NULL;
509 
510 	if (type == FLUSH_INO) {
511 		rcu_read_lock();
512 		e = radix_tree_lookup(&im->ino_root, ino);
513 		rcu_read_unlock();
514 	}
515 
516 retry:
517 	if (!e)
518 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
519 						GFP_NOFS, true, NULL);
520 
521 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
522 
523 	spin_lock(&im->ino_lock);
524 	e = radix_tree_lookup(&im->ino_root, ino);
525 	if (!e) {
526 		if (!new) {
527 			spin_unlock(&im->ino_lock);
528 			radix_tree_preload_end();
529 			goto retry;
530 		}
531 		e = new;
532 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
533 			f2fs_bug_on(sbi, 1);
534 
535 		memset(e, 0, sizeof(struct ino_entry));
536 		e->ino = ino;
537 
538 		list_add_tail(&e->list, &im->ino_list);
539 		if (type != ORPHAN_INO)
540 			im->ino_num++;
541 	}
542 
543 	if (type == FLUSH_INO)
544 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
545 
546 	spin_unlock(&im->ino_lock);
547 	radix_tree_preload_end();
548 
549 	if (new && e != new)
550 		kmem_cache_free(ino_entry_slab, new);
551 }
552 
553 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
554 {
555 	struct inode_management *im = &sbi->im[type];
556 	struct ino_entry *e;
557 
558 	spin_lock(&im->ino_lock);
559 	e = radix_tree_lookup(&im->ino_root, ino);
560 	if (e) {
561 		list_del(&e->list);
562 		radix_tree_delete(&im->ino_root, ino);
563 		im->ino_num--;
564 		spin_unlock(&im->ino_lock);
565 		kmem_cache_free(ino_entry_slab, e);
566 		return;
567 	}
568 	spin_unlock(&im->ino_lock);
569 }
570 
571 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
572 {
573 	/* add new dirty ino entry into list */
574 	__add_ino_entry(sbi, ino, 0, type);
575 }
576 
577 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
578 {
579 	/* remove dirty ino entry from list */
580 	__remove_ino_entry(sbi, ino, type);
581 }
582 
583 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
584 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
585 {
586 	struct inode_management *im = &sbi->im[mode];
587 	struct ino_entry *e;
588 
589 	spin_lock(&im->ino_lock);
590 	e = radix_tree_lookup(&im->ino_root, ino);
591 	spin_unlock(&im->ino_lock);
592 	return e ? true : false;
593 }
594 
595 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
596 {
597 	struct ino_entry *e, *tmp;
598 	int i;
599 
600 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
601 		struct inode_management *im = &sbi->im[i];
602 
603 		spin_lock(&im->ino_lock);
604 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
605 			list_del(&e->list);
606 			radix_tree_delete(&im->ino_root, e->ino);
607 			kmem_cache_free(ino_entry_slab, e);
608 			im->ino_num--;
609 		}
610 		spin_unlock(&im->ino_lock);
611 	}
612 }
613 
614 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
615 					unsigned int devidx, int type)
616 {
617 	__add_ino_entry(sbi, ino, devidx, type);
618 }
619 
620 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
621 					unsigned int devidx, int type)
622 {
623 	struct inode_management *im = &sbi->im[type];
624 	struct ino_entry *e;
625 	bool is_dirty = false;
626 
627 	spin_lock(&im->ino_lock);
628 	e = radix_tree_lookup(&im->ino_root, ino);
629 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
630 		is_dirty = true;
631 	spin_unlock(&im->ino_lock);
632 	return is_dirty;
633 }
634 
635 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
636 {
637 	struct inode_management *im = &sbi->im[ORPHAN_INO];
638 	int err = 0;
639 
640 	spin_lock(&im->ino_lock);
641 
642 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
643 		spin_unlock(&im->ino_lock);
644 		return -ENOSPC;
645 	}
646 
647 	if (unlikely(im->ino_num >= sbi->max_orphans))
648 		err = -ENOSPC;
649 	else
650 		im->ino_num++;
651 	spin_unlock(&im->ino_lock);
652 
653 	return err;
654 }
655 
656 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
657 {
658 	struct inode_management *im = &sbi->im[ORPHAN_INO];
659 
660 	spin_lock(&im->ino_lock);
661 	f2fs_bug_on(sbi, im->ino_num == 0);
662 	im->ino_num--;
663 	spin_unlock(&im->ino_lock);
664 }
665 
666 void f2fs_add_orphan_inode(struct inode *inode)
667 {
668 	/* add new orphan ino entry into list */
669 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
670 	f2fs_update_inode_page(inode);
671 }
672 
673 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
674 {
675 	/* remove orphan entry from orphan list */
676 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
677 }
678 
679 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
680 {
681 	struct inode *inode;
682 	struct node_info ni;
683 	int err;
684 
685 	inode = f2fs_iget_retry(sbi->sb, ino);
686 	if (IS_ERR(inode)) {
687 		/*
688 		 * there should be a bug that we can't find the entry
689 		 * to orphan inode.
690 		 */
691 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
692 		return PTR_ERR(inode);
693 	}
694 
695 	err = f2fs_dquot_initialize(inode);
696 	if (err) {
697 		iput(inode);
698 		goto err_out;
699 	}
700 
701 	clear_nlink(inode);
702 
703 	/* truncate all the data during iput */
704 	iput(inode);
705 
706 	err = f2fs_get_node_info(sbi, ino, &ni, false);
707 	if (err)
708 		goto err_out;
709 
710 	/* ENOMEM was fully retried in f2fs_evict_inode. */
711 	if (ni.blk_addr != NULL_ADDR) {
712 		err = -EIO;
713 		goto err_out;
714 	}
715 	return 0;
716 
717 err_out:
718 	set_sbi_flag(sbi, SBI_NEED_FSCK);
719 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
720 		  __func__, ino);
721 	return err;
722 }
723 
724 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
725 {
726 	block_t start_blk, orphan_blocks, i, j;
727 	int err = 0;
728 
729 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
730 		return 0;
731 
732 	if (f2fs_hw_is_readonly(sbi)) {
733 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
734 		return 0;
735 	}
736 
737 	if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
738 		f2fs_info(sbi, "orphan cleanup on readonly fs");
739 
740 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
741 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
742 
743 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
744 
745 	for (i = 0; i < orphan_blocks; i++) {
746 		struct page *page;
747 		struct f2fs_orphan_block *orphan_blk;
748 
749 		page = f2fs_get_meta_page(sbi, start_blk + i);
750 		if (IS_ERR(page)) {
751 			err = PTR_ERR(page);
752 			goto out;
753 		}
754 
755 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
756 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
757 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
758 
759 			err = recover_orphan_inode(sbi, ino);
760 			if (err) {
761 				f2fs_put_page(page, 1);
762 				goto out;
763 			}
764 		}
765 		f2fs_put_page(page, 1);
766 	}
767 	/* clear Orphan Flag */
768 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
769 out:
770 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
771 
772 	return err;
773 }
774 
775 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
776 {
777 	struct list_head *head;
778 	struct f2fs_orphan_block *orphan_blk = NULL;
779 	unsigned int nentries = 0;
780 	unsigned short index = 1;
781 	unsigned short orphan_blocks;
782 	struct page *page = NULL;
783 	struct ino_entry *orphan = NULL;
784 	struct inode_management *im = &sbi->im[ORPHAN_INO];
785 
786 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
787 
788 	/*
789 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
790 	 * orphan inode operations are covered under f2fs_lock_op().
791 	 * And, spin_lock should be avoided due to page operations below.
792 	 */
793 	head = &im->ino_list;
794 
795 	/* loop for each orphan inode entry and write them in journal block */
796 	list_for_each_entry(orphan, head, list) {
797 		if (!page) {
798 			page = f2fs_grab_meta_page(sbi, start_blk++);
799 			orphan_blk =
800 				(struct f2fs_orphan_block *)page_address(page);
801 			memset(orphan_blk, 0, sizeof(*orphan_blk));
802 		}
803 
804 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
805 
806 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
807 			/*
808 			 * an orphan block is full of 1020 entries,
809 			 * then we need to flush current orphan blocks
810 			 * and bring another one in memory
811 			 */
812 			orphan_blk->blk_addr = cpu_to_le16(index);
813 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
814 			orphan_blk->entry_count = cpu_to_le32(nentries);
815 			set_page_dirty(page);
816 			f2fs_put_page(page, 1);
817 			index++;
818 			nentries = 0;
819 			page = NULL;
820 		}
821 	}
822 
823 	if (page) {
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 	}
830 }
831 
832 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
833 						struct f2fs_checkpoint *ckpt)
834 {
835 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
836 	__u32 chksum;
837 
838 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
839 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
840 		chksum_ofs += sizeof(chksum);
841 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
842 						F2FS_BLKSIZE - chksum_ofs);
843 	}
844 	return chksum;
845 }
846 
847 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
848 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
849 		unsigned long long *version)
850 {
851 	size_t crc_offset = 0;
852 	__u32 crc;
853 
854 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
855 	if (IS_ERR(*cp_page))
856 		return PTR_ERR(*cp_page);
857 
858 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
859 
860 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
861 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
862 			crc_offset > CP_CHKSUM_OFFSET) {
863 		f2fs_put_page(*cp_page, 1);
864 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
865 		return -EINVAL;
866 	}
867 
868 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
869 	if (crc != cur_cp_crc(*cp_block)) {
870 		f2fs_put_page(*cp_page, 1);
871 		f2fs_warn(sbi, "invalid crc value");
872 		return -EINVAL;
873 	}
874 
875 	*version = cur_cp_version(*cp_block);
876 	return 0;
877 }
878 
879 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
880 				block_t cp_addr, unsigned long long *version)
881 {
882 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
883 	struct f2fs_checkpoint *cp_block = NULL;
884 	unsigned long long cur_version = 0, pre_version = 0;
885 	unsigned int cp_blocks;
886 	int err;
887 
888 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
889 					&cp_page_1, version);
890 	if (err)
891 		return NULL;
892 
893 	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
894 
895 	if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
896 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
897 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
898 		goto invalid_cp;
899 	}
900 	pre_version = *version;
901 
902 	cp_addr += cp_blocks - 1;
903 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
904 					&cp_page_2, version);
905 	if (err)
906 		goto invalid_cp;
907 	cur_version = *version;
908 
909 	if (cur_version == pre_version) {
910 		*version = cur_version;
911 		f2fs_put_page(cp_page_2, 1);
912 		return cp_page_1;
913 	}
914 	f2fs_put_page(cp_page_2, 1);
915 invalid_cp:
916 	f2fs_put_page(cp_page_1, 1);
917 	return NULL;
918 }
919 
920 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
921 {
922 	struct f2fs_checkpoint *cp_block;
923 	struct f2fs_super_block *fsb = sbi->raw_super;
924 	struct page *cp1, *cp2, *cur_page;
925 	unsigned long blk_size = sbi->blocksize;
926 	unsigned long long cp1_version = 0, cp2_version = 0;
927 	unsigned long long cp_start_blk_no;
928 	unsigned int cp_blks = 1 + __cp_payload(sbi);
929 	block_t cp_blk_no;
930 	int i;
931 	int err;
932 
933 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
934 				  GFP_KERNEL);
935 	if (!sbi->ckpt)
936 		return -ENOMEM;
937 	/*
938 	 * Finding out valid cp block involves read both
939 	 * sets( cp pack 1 and cp pack 2)
940 	 */
941 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
942 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
943 
944 	/* The second checkpoint pack should start at the next segment */
945 	cp_start_blk_no += ((unsigned long long)1) <<
946 				le32_to_cpu(fsb->log_blocks_per_seg);
947 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
948 
949 	if (cp1 && cp2) {
950 		if (ver_after(cp2_version, cp1_version))
951 			cur_page = cp2;
952 		else
953 			cur_page = cp1;
954 	} else if (cp1) {
955 		cur_page = cp1;
956 	} else if (cp2) {
957 		cur_page = cp2;
958 	} else {
959 		err = -EFSCORRUPTED;
960 		goto fail_no_cp;
961 	}
962 
963 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
964 	memcpy(sbi->ckpt, cp_block, blk_size);
965 
966 	if (cur_page == cp1)
967 		sbi->cur_cp_pack = 1;
968 	else
969 		sbi->cur_cp_pack = 2;
970 
971 	/* Sanity checking of checkpoint */
972 	if (f2fs_sanity_check_ckpt(sbi)) {
973 		err = -EFSCORRUPTED;
974 		goto free_fail_no_cp;
975 	}
976 
977 	if (cp_blks <= 1)
978 		goto done;
979 
980 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
981 	if (cur_page == cp2)
982 		cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
983 
984 	for (i = 1; i < cp_blks; i++) {
985 		void *sit_bitmap_ptr;
986 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
987 
988 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
989 		if (IS_ERR(cur_page)) {
990 			err = PTR_ERR(cur_page);
991 			goto free_fail_no_cp;
992 		}
993 		sit_bitmap_ptr = page_address(cur_page);
994 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
995 		f2fs_put_page(cur_page, 1);
996 	}
997 done:
998 	f2fs_put_page(cp1, 1);
999 	f2fs_put_page(cp2, 1);
1000 	return 0;
1001 
1002 free_fail_no_cp:
1003 	f2fs_put_page(cp1, 1);
1004 	f2fs_put_page(cp2, 1);
1005 fail_no_cp:
1006 	kvfree(sbi->ckpt);
1007 	return err;
1008 }
1009 
1010 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1011 {
1012 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1013 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1014 
1015 	if (is_inode_flag_set(inode, flag))
1016 		return;
1017 
1018 	set_inode_flag(inode, flag);
1019 	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1020 	stat_inc_dirty_inode(sbi, type);
1021 }
1022 
1023 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1024 {
1025 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1026 
1027 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1028 		return;
1029 
1030 	list_del_init(&F2FS_I(inode)->dirty_list);
1031 	clear_inode_flag(inode, flag);
1032 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1033 }
1034 
1035 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1036 {
1037 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1038 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1039 
1040 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1041 			!S_ISLNK(inode->i_mode))
1042 		return;
1043 
1044 	spin_lock(&sbi->inode_lock[type]);
1045 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1046 		__add_dirty_inode(inode, type);
1047 	inode_inc_dirty_pages(inode);
1048 	spin_unlock(&sbi->inode_lock[type]);
1049 
1050 	set_page_private_reference(&folio->page);
1051 }
1052 
1053 void f2fs_remove_dirty_inode(struct inode *inode)
1054 {
1055 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1056 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1057 
1058 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1059 			!S_ISLNK(inode->i_mode))
1060 		return;
1061 
1062 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1063 		return;
1064 
1065 	spin_lock(&sbi->inode_lock[type]);
1066 	__remove_dirty_inode(inode, type);
1067 	spin_unlock(&sbi->inode_lock[type]);
1068 }
1069 
1070 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1071 						bool from_cp)
1072 {
1073 	struct list_head *head;
1074 	struct inode *inode;
1075 	struct f2fs_inode_info *fi;
1076 	bool is_dir = (type == DIR_INODE);
1077 	unsigned long ino = 0;
1078 
1079 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1080 				get_pages(sbi, is_dir ?
1081 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1082 retry:
1083 	if (unlikely(f2fs_cp_error(sbi))) {
1084 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1085 				get_pages(sbi, is_dir ?
1086 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1087 		return -EIO;
1088 	}
1089 
1090 	spin_lock(&sbi->inode_lock[type]);
1091 
1092 	head = &sbi->inode_list[type];
1093 	if (list_empty(head)) {
1094 		spin_unlock(&sbi->inode_lock[type]);
1095 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1096 				get_pages(sbi, is_dir ?
1097 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1098 		return 0;
1099 	}
1100 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1101 	inode = igrab(&fi->vfs_inode);
1102 	spin_unlock(&sbi->inode_lock[type]);
1103 	if (inode) {
1104 		unsigned long cur_ino = inode->i_ino;
1105 
1106 		if (from_cp)
1107 			F2FS_I(inode)->cp_task = current;
1108 		F2FS_I(inode)->wb_task = current;
1109 
1110 		filemap_fdatawrite(inode->i_mapping);
1111 
1112 		F2FS_I(inode)->wb_task = NULL;
1113 		if (from_cp)
1114 			F2FS_I(inode)->cp_task = NULL;
1115 
1116 		iput(inode);
1117 		/* We need to give cpu to another writers. */
1118 		if (ino == cur_ino)
1119 			cond_resched();
1120 		else
1121 			ino = cur_ino;
1122 	} else {
1123 		/*
1124 		 * We should submit bio, since it exists several
1125 		 * writebacking dentry pages in the freeing inode.
1126 		 */
1127 		f2fs_submit_merged_write(sbi, DATA);
1128 		cond_resched();
1129 	}
1130 	goto retry;
1131 }
1132 
1133 static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1134 {
1135 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1136 	struct inode *inode;
1137 	struct f2fs_inode_info *fi;
1138 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1139 
1140 	while (total--) {
1141 		if (unlikely(f2fs_cp_error(sbi)))
1142 			return -EIO;
1143 
1144 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1145 		if (list_empty(head)) {
1146 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1147 			return 0;
1148 		}
1149 		fi = list_first_entry(head, struct f2fs_inode_info,
1150 							gdirty_list);
1151 		inode = igrab(&fi->vfs_inode);
1152 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1153 		if (inode) {
1154 			sync_inode_metadata(inode, 0);
1155 
1156 			/* it's on eviction */
1157 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1158 				f2fs_update_inode_page(inode);
1159 			iput(inode);
1160 		}
1161 	}
1162 	return 0;
1163 }
1164 
1165 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1166 {
1167 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1168 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1169 	nid_t last_nid = nm_i->next_scan_nid;
1170 
1171 	next_free_nid(sbi, &last_nid);
1172 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1173 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1174 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1175 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1176 }
1177 
1178 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1179 {
1180 	bool ret = false;
1181 
1182 	if (!is_journalled_quota(sbi))
1183 		return false;
1184 
1185 	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1186 		return true;
1187 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1188 		ret = false;
1189 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1190 		ret = false;
1191 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1192 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1193 		ret = true;
1194 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1195 		ret = true;
1196 	}
1197 	f2fs_up_write(&sbi->quota_sem);
1198 	return ret;
1199 }
1200 
1201 /*
1202  * Freeze all the FS-operations for checkpoint.
1203  */
1204 static int block_operations(struct f2fs_sb_info *sbi)
1205 {
1206 	struct writeback_control wbc = {
1207 		.sync_mode = WB_SYNC_ALL,
1208 		.nr_to_write = LONG_MAX,
1209 		.for_reclaim = 0,
1210 	};
1211 	int err = 0, cnt = 0;
1212 
1213 	/*
1214 	 * Let's flush inline_data in dirty node pages.
1215 	 */
1216 	f2fs_flush_inline_data(sbi);
1217 
1218 retry_flush_quotas:
1219 	f2fs_lock_all(sbi);
1220 	if (__need_flush_quota(sbi)) {
1221 		int locked;
1222 
1223 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1224 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1225 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1226 			goto retry_flush_dents;
1227 		}
1228 		f2fs_unlock_all(sbi);
1229 
1230 		/* only failed during mount/umount/freeze/quotactl */
1231 		locked = down_read_trylock(&sbi->sb->s_umount);
1232 		f2fs_quota_sync(sbi->sb, -1);
1233 		if (locked)
1234 			up_read(&sbi->sb->s_umount);
1235 		cond_resched();
1236 		goto retry_flush_quotas;
1237 	}
1238 
1239 retry_flush_dents:
1240 	/* write all the dirty dentry pages */
1241 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1242 		f2fs_unlock_all(sbi);
1243 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1244 		if (err)
1245 			return err;
1246 		cond_resched();
1247 		goto retry_flush_quotas;
1248 	}
1249 
1250 	/*
1251 	 * POR: we should ensure that there are no dirty node pages
1252 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1253 	 */
1254 	f2fs_down_write(&sbi->node_change);
1255 
1256 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1257 		f2fs_up_write(&sbi->node_change);
1258 		f2fs_unlock_all(sbi);
1259 		err = f2fs_sync_inode_meta(sbi);
1260 		if (err)
1261 			return err;
1262 		cond_resched();
1263 		goto retry_flush_quotas;
1264 	}
1265 
1266 retry_flush_nodes:
1267 	f2fs_down_write(&sbi->node_write);
1268 
1269 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1270 		f2fs_up_write(&sbi->node_write);
1271 		atomic_inc(&sbi->wb_sync_req[NODE]);
1272 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1273 		atomic_dec(&sbi->wb_sync_req[NODE]);
1274 		if (err) {
1275 			f2fs_up_write(&sbi->node_change);
1276 			f2fs_unlock_all(sbi);
1277 			return err;
1278 		}
1279 		cond_resched();
1280 		goto retry_flush_nodes;
1281 	}
1282 
1283 	/*
1284 	 * sbi->node_change is used only for AIO write_begin path which produces
1285 	 * dirty node blocks and some checkpoint values by block allocation.
1286 	 */
1287 	__prepare_cp_block(sbi);
1288 	f2fs_up_write(&sbi->node_change);
1289 	return err;
1290 }
1291 
1292 static void unblock_operations(struct f2fs_sb_info *sbi)
1293 {
1294 	f2fs_up_write(&sbi->node_write);
1295 	f2fs_unlock_all(sbi);
1296 }
1297 
1298 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1299 {
1300 	DEFINE_WAIT(wait);
1301 
1302 	for (;;) {
1303 		if (!get_pages(sbi, type))
1304 			break;
1305 
1306 		if (unlikely(f2fs_cp_error(sbi) &&
1307 			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1308 			break;
1309 
1310 		if (type == F2FS_DIRTY_META)
1311 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1312 							FS_CP_META_IO);
1313 		else if (type == F2FS_WB_CP_DATA)
1314 			f2fs_submit_merged_write(sbi, DATA);
1315 
1316 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1317 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1318 	}
1319 	finish_wait(&sbi->cp_wait, &wait);
1320 }
1321 
1322 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1323 {
1324 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1325 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1326 	unsigned long flags;
1327 
1328 	if (cpc->reason & CP_UMOUNT) {
1329 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1330 			NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1331 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1332 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1333 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1334 						f2fs_nat_bitmap_enabled(sbi)) {
1335 			f2fs_enable_nat_bits(sbi);
1336 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1337 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1338 		}
1339 	}
1340 
1341 	spin_lock_irqsave(&sbi->cp_lock, flags);
1342 
1343 	if (cpc->reason & CP_TRIMMED)
1344 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1345 	else
1346 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1347 
1348 	if (cpc->reason & CP_UMOUNT)
1349 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1350 	else
1351 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1352 
1353 	if (cpc->reason & CP_FASTBOOT)
1354 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1355 	else
1356 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1357 
1358 	if (orphan_num)
1359 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1360 	else
1361 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1362 
1363 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1364 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1365 
1366 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1367 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1368 	else
1369 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1370 
1371 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1372 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1373 	else
1374 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1375 
1376 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1377 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1378 	else
1379 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1380 
1381 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1382 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1383 	else
1384 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1385 
1386 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1387 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1388 
1389 	/* set this flag to activate crc|cp_ver for recovery */
1390 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1391 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1392 
1393 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1394 }
1395 
1396 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1397 	void *src, block_t blk_addr)
1398 {
1399 	struct writeback_control wbc = {
1400 		.for_reclaim = 0,
1401 	};
1402 
1403 	/*
1404 	 * filemap_get_folios_tag and lock_page again will take
1405 	 * some extra time. Therefore, f2fs_update_meta_pages and
1406 	 * f2fs_sync_meta_pages are combined in this function.
1407 	 */
1408 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1409 	int err;
1410 
1411 	f2fs_wait_on_page_writeback(page, META, true, true);
1412 
1413 	memcpy(page_address(page), src, PAGE_SIZE);
1414 
1415 	set_page_dirty(page);
1416 	if (unlikely(!clear_page_dirty_for_io(page)))
1417 		f2fs_bug_on(sbi, 1);
1418 
1419 	/* writeout cp pack 2 page */
1420 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1421 	if (unlikely(err && f2fs_cp_error(sbi))) {
1422 		f2fs_put_page(page, 1);
1423 		return;
1424 	}
1425 
1426 	f2fs_bug_on(sbi, err);
1427 	f2fs_put_page(page, 0);
1428 
1429 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1430 	f2fs_submit_merged_write(sbi, META_FLUSH);
1431 }
1432 
1433 static inline u64 get_sectors_written(struct block_device *bdev)
1434 {
1435 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1436 }
1437 
1438 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1439 {
1440 	if (f2fs_is_multi_device(sbi)) {
1441 		u64 sectors = 0;
1442 		int i;
1443 
1444 		for (i = 0; i < sbi->s_ndevs; i++)
1445 			sectors += get_sectors_written(FDEV(i).bdev);
1446 
1447 		return sectors;
1448 	}
1449 
1450 	return get_sectors_written(sbi->sb->s_bdev);
1451 }
1452 
1453 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1454 {
1455 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1456 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1457 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1458 	block_t start_blk;
1459 	unsigned int data_sum_blocks, orphan_blocks;
1460 	__u32 crc32 = 0;
1461 	int i;
1462 	int cp_payload_blks = __cp_payload(sbi);
1463 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1464 	u64 kbytes_written;
1465 	int err;
1466 
1467 	/* Flush all the NAT/SIT pages */
1468 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1469 
1470 	/* start to update checkpoint, cp ver is already updated previously */
1471 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1472 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1473 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1474 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1475 
1476 		ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1477 		ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1478 		ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1479 	}
1480 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1481 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1482 
1483 		ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1484 		ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1485 		ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1486 	}
1487 
1488 	/* 2 cp + n data seg summary + orphan inode blocks */
1489 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1490 	spin_lock_irqsave(&sbi->cp_lock, flags);
1491 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1492 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1493 	else
1494 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1495 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1496 
1497 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1498 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1499 			orphan_blocks);
1500 
1501 	if (__remain_node_summaries(cpc->reason))
1502 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1503 				cp_payload_blks + data_sum_blocks +
1504 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1505 	else
1506 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1507 				cp_payload_blks + data_sum_blocks +
1508 				orphan_blocks);
1509 
1510 	/* update ckpt flag for checkpoint */
1511 	update_ckpt_flags(sbi, cpc);
1512 
1513 	/* update SIT/NAT bitmap */
1514 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1515 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1516 
1517 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1518 	*((__le32 *)((unsigned char *)ckpt +
1519 				le32_to_cpu(ckpt->checksum_offset)))
1520 				= cpu_to_le32(crc32);
1521 
1522 	start_blk = __start_cp_next_addr(sbi);
1523 
1524 	/* write nat bits */
1525 	if ((cpc->reason & CP_UMOUNT) &&
1526 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1527 		__u64 cp_ver = cur_cp_version(ckpt);
1528 		block_t blk;
1529 
1530 		cp_ver |= ((__u64)crc32 << 32);
1531 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1532 
1533 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1534 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1535 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1536 					(i << F2FS_BLKSIZE_BITS), blk + i);
1537 	}
1538 
1539 	/* write out checkpoint buffer at block 0 */
1540 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1541 
1542 	for (i = 1; i < 1 + cp_payload_blks; i++)
1543 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1544 							start_blk++);
1545 
1546 	if (orphan_num) {
1547 		write_orphan_inodes(sbi, start_blk);
1548 		start_blk += orphan_blocks;
1549 	}
1550 
1551 	f2fs_write_data_summaries(sbi, start_blk);
1552 	start_blk += data_sum_blocks;
1553 
1554 	/* Record write statistics in the hot node summary */
1555 	kbytes_written = sbi->kbytes_written;
1556 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1557 				sbi->sectors_written_start) >> 1;
1558 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1559 
1560 	if (__remain_node_summaries(cpc->reason)) {
1561 		f2fs_write_node_summaries(sbi, start_blk);
1562 		start_blk += NR_CURSEG_NODE_TYPE;
1563 	}
1564 
1565 	/* update user_block_counts */
1566 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1567 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1568 	percpu_counter_set(&sbi->rf_node_block_count, 0);
1569 
1570 	/* Here, we have one bio having CP pack except cp pack 2 page */
1571 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1572 	/* Wait for all dirty meta pages to be submitted for IO */
1573 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1574 
1575 	/* wait for previous submitted meta pages writeback */
1576 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1577 
1578 	/* flush all device cache */
1579 	err = f2fs_flush_device_cache(sbi);
1580 	if (err)
1581 		return err;
1582 
1583 	/* barrier and flush checkpoint cp pack 2 page if it can */
1584 	commit_checkpoint(sbi, ckpt, start_blk);
1585 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1586 
1587 	/*
1588 	 * invalidate intermediate page cache borrowed from meta inode which are
1589 	 * used for migration of encrypted, verity or compressed inode's blocks.
1590 	 */
1591 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1592 		f2fs_sb_has_compression(sbi))
1593 		invalidate_mapping_pages(META_MAPPING(sbi),
1594 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1595 
1596 	f2fs_release_ino_entry(sbi, false);
1597 
1598 	f2fs_reset_fsync_node_info(sbi);
1599 
1600 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1601 	clear_sbi_flag(sbi, SBI_NEED_CP);
1602 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1603 
1604 	spin_lock(&sbi->stat_lock);
1605 	sbi->unusable_block_count = 0;
1606 	spin_unlock(&sbi->stat_lock);
1607 
1608 	__set_cp_next_pack(sbi);
1609 
1610 	/*
1611 	 * redirty superblock if metadata like node page or inode cache is
1612 	 * updated during writing checkpoint.
1613 	 */
1614 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1615 			get_pages(sbi, F2FS_DIRTY_IMETA))
1616 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1617 
1618 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1619 
1620 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1621 }
1622 
1623 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1624 {
1625 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1626 	unsigned long long ckpt_ver;
1627 	int err = 0;
1628 
1629 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1630 		return -EROFS;
1631 
1632 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1633 		if (cpc->reason != CP_PAUSE)
1634 			return 0;
1635 		f2fs_warn(sbi, "Start checkpoint disabled!");
1636 	}
1637 	if (cpc->reason != CP_RESIZE)
1638 		f2fs_down_write(&sbi->cp_global_sem);
1639 
1640 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1641 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1642 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1643 		goto out;
1644 	if (unlikely(f2fs_cp_error(sbi))) {
1645 		err = -EIO;
1646 		goto out;
1647 	}
1648 
1649 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1650 
1651 	err = block_operations(sbi);
1652 	if (err)
1653 		goto out;
1654 
1655 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1656 
1657 	f2fs_flush_merged_writes(sbi);
1658 
1659 	/* this is the case of multiple fstrims without any changes */
1660 	if (cpc->reason & CP_DISCARD) {
1661 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1662 			unblock_operations(sbi);
1663 			goto out;
1664 		}
1665 
1666 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1667 				SIT_I(sbi)->dirty_sentries == 0 &&
1668 				prefree_segments(sbi) == 0) {
1669 			f2fs_flush_sit_entries(sbi, cpc);
1670 			f2fs_clear_prefree_segments(sbi, cpc);
1671 			unblock_operations(sbi);
1672 			goto out;
1673 		}
1674 	}
1675 
1676 	/*
1677 	 * update checkpoint pack index
1678 	 * Increase the version number so that
1679 	 * SIT entries and seg summaries are written at correct place
1680 	 */
1681 	ckpt_ver = cur_cp_version(ckpt);
1682 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1683 
1684 	/* write cached NAT/SIT entries to NAT/SIT area */
1685 	err = f2fs_flush_nat_entries(sbi, cpc);
1686 	if (err) {
1687 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1688 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1689 		goto stop;
1690 	}
1691 
1692 	f2fs_flush_sit_entries(sbi, cpc);
1693 
1694 	/* save inmem log status */
1695 	f2fs_save_inmem_curseg(sbi);
1696 
1697 	err = do_checkpoint(sbi, cpc);
1698 	if (err) {
1699 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1700 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1701 		f2fs_release_discard_addrs(sbi);
1702 	} else {
1703 		f2fs_clear_prefree_segments(sbi, cpc);
1704 	}
1705 
1706 	f2fs_restore_inmem_curseg(sbi);
1707 stop:
1708 	unblock_operations(sbi);
1709 	stat_inc_cp_count(sbi->stat_info);
1710 
1711 	if (cpc->reason & CP_RECOVERY)
1712 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1713 
1714 	/* update CP_TIME to trigger checkpoint periodically */
1715 	f2fs_update_time(sbi, CP_TIME);
1716 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1717 out:
1718 	if (cpc->reason != CP_RESIZE)
1719 		f2fs_up_write(&sbi->cp_global_sem);
1720 	return err;
1721 }
1722 
1723 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1724 {
1725 	int i;
1726 
1727 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1728 		struct inode_management *im = &sbi->im[i];
1729 
1730 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1731 		spin_lock_init(&im->ino_lock);
1732 		INIT_LIST_HEAD(&im->ino_list);
1733 		im->ino_num = 0;
1734 	}
1735 
1736 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1737 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1738 				F2FS_ORPHANS_PER_BLOCK;
1739 }
1740 
1741 int __init f2fs_create_checkpoint_caches(void)
1742 {
1743 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1744 			sizeof(struct ino_entry));
1745 	if (!ino_entry_slab)
1746 		return -ENOMEM;
1747 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1748 			sizeof(struct inode_entry));
1749 	if (!f2fs_inode_entry_slab) {
1750 		kmem_cache_destroy(ino_entry_slab);
1751 		return -ENOMEM;
1752 	}
1753 	return 0;
1754 }
1755 
1756 void f2fs_destroy_checkpoint_caches(void)
1757 {
1758 	kmem_cache_destroy(ino_entry_slab);
1759 	kmem_cache_destroy(f2fs_inode_entry_slab);
1760 }
1761 
1762 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1763 {
1764 	struct cp_control cpc = { .reason = CP_SYNC, };
1765 	int err;
1766 
1767 	f2fs_down_write(&sbi->gc_lock);
1768 	err = f2fs_write_checkpoint(sbi, &cpc);
1769 	f2fs_up_write(&sbi->gc_lock);
1770 
1771 	return err;
1772 }
1773 
1774 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1775 {
1776 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1777 	struct ckpt_req *req, *next;
1778 	struct llist_node *dispatch_list;
1779 	u64 sum_diff = 0, diff, count = 0;
1780 	int ret;
1781 
1782 	dispatch_list = llist_del_all(&cprc->issue_list);
1783 	if (!dispatch_list)
1784 		return;
1785 	dispatch_list = llist_reverse_order(dispatch_list);
1786 
1787 	ret = __write_checkpoint_sync(sbi);
1788 	atomic_inc(&cprc->issued_ckpt);
1789 
1790 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1791 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1792 		req->ret = ret;
1793 		complete(&req->wait);
1794 
1795 		sum_diff += diff;
1796 		count++;
1797 	}
1798 	atomic_sub(count, &cprc->queued_ckpt);
1799 	atomic_add(count, &cprc->total_ckpt);
1800 
1801 	spin_lock(&cprc->stat_lock);
1802 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1803 	if (cprc->peak_time < cprc->cur_time)
1804 		cprc->peak_time = cprc->cur_time;
1805 	spin_unlock(&cprc->stat_lock);
1806 }
1807 
1808 static int issue_checkpoint_thread(void *data)
1809 {
1810 	struct f2fs_sb_info *sbi = data;
1811 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1812 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1813 repeat:
1814 	if (kthread_should_stop())
1815 		return 0;
1816 
1817 	if (!llist_empty(&cprc->issue_list))
1818 		__checkpoint_and_complete_reqs(sbi);
1819 
1820 	wait_event_interruptible(*q,
1821 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1822 	goto repeat;
1823 }
1824 
1825 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1826 		struct ckpt_req *wait_req)
1827 {
1828 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1829 
1830 	if (!llist_empty(&cprc->issue_list)) {
1831 		__checkpoint_and_complete_reqs(sbi);
1832 	} else {
1833 		/* already dispatched by issue_checkpoint_thread */
1834 		if (wait_req)
1835 			wait_for_completion(&wait_req->wait);
1836 	}
1837 }
1838 
1839 static void init_ckpt_req(struct ckpt_req *req)
1840 {
1841 	memset(req, 0, sizeof(struct ckpt_req));
1842 
1843 	init_completion(&req->wait);
1844 	req->queue_time = ktime_get();
1845 }
1846 
1847 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1848 {
1849 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1850 	struct ckpt_req req;
1851 	struct cp_control cpc;
1852 
1853 	cpc.reason = __get_cp_reason(sbi);
1854 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1855 		int ret;
1856 
1857 		f2fs_down_write(&sbi->gc_lock);
1858 		ret = f2fs_write_checkpoint(sbi, &cpc);
1859 		f2fs_up_write(&sbi->gc_lock);
1860 
1861 		return ret;
1862 	}
1863 
1864 	if (!cprc->f2fs_issue_ckpt)
1865 		return __write_checkpoint_sync(sbi);
1866 
1867 	init_ckpt_req(&req);
1868 
1869 	llist_add(&req.llnode, &cprc->issue_list);
1870 	atomic_inc(&cprc->queued_ckpt);
1871 
1872 	/*
1873 	 * update issue_list before we wake up issue_checkpoint thread,
1874 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1875 	 * see more details in comments of waitqueue_active().
1876 	 */
1877 	smp_mb();
1878 
1879 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1880 		wake_up(&cprc->ckpt_wait_queue);
1881 
1882 	if (cprc->f2fs_issue_ckpt)
1883 		wait_for_completion(&req.wait);
1884 	else
1885 		flush_remained_ckpt_reqs(sbi, &req);
1886 
1887 	return req.ret;
1888 }
1889 
1890 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1891 {
1892 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1893 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1894 
1895 	if (cprc->f2fs_issue_ckpt)
1896 		return 0;
1897 
1898 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1899 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1900 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1901 		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1902 
1903 		cprc->f2fs_issue_ckpt = NULL;
1904 		return err;
1905 	}
1906 
1907 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1908 
1909 	return 0;
1910 }
1911 
1912 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1913 {
1914 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1915 	struct task_struct *ckpt_task;
1916 
1917 	if (!cprc->f2fs_issue_ckpt)
1918 		return;
1919 
1920 	ckpt_task = cprc->f2fs_issue_ckpt;
1921 	cprc->f2fs_issue_ckpt = NULL;
1922 	kthread_stop(ckpt_task);
1923 
1924 	f2fs_flush_ckpt_thread(sbi);
1925 }
1926 
1927 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1928 {
1929 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1930 
1931 	flush_remained_ckpt_reqs(sbi, NULL);
1932 
1933 	/* Let's wait for the previous dispatched checkpoint. */
1934 	while (atomic_read(&cprc->queued_ckpt))
1935 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1936 }
1937 
1938 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1939 {
1940 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1941 
1942 	atomic_set(&cprc->issued_ckpt, 0);
1943 	atomic_set(&cprc->total_ckpt, 0);
1944 	atomic_set(&cprc->queued_ckpt, 0);
1945 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1946 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1947 	init_llist_head(&cprc->issue_list);
1948 	spin_lock_init(&cprc->stat_lock);
1949 }
1950