xref: /linux/fs/f2fs/checkpoint.c (revision a9aaf1ff88a8cb99a1335c9eb76de637f0cf8c10)
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 check_only;
183 		break;
184 	case META_SSA:
185 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
186 			blkaddr < SM_I(sbi)->ssa_blkaddr))
187 			goto check_only;
188 		break;
189 	case META_CP:
190 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
191 			blkaddr < __start_cp_addr(sbi)))
192 			goto check_only;
193 		break;
194 	case META_POR:
195 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
196 			blkaddr < MAIN_BLKADDR(sbi)))
197 			goto check_only;
198 		break;
199 	case DATA_GENERIC:
200 	case DATA_GENERIC_ENHANCE:
201 	case DATA_GENERIC_ENHANCE_READ:
202 	case DATA_GENERIC_ENHANCE_UPDATE:
203 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
204 				blkaddr < MAIN_BLKADDR(sbi))) {
205 
206 			/* Skip to emit an error message. */
207 			if (unlikely(f2fs_cp_error(sbi)))
208 				return false;
209 
210 			f2fs_warn(sbi, "access invalid blkaddr:%u",
211 				  blkaddr);
212 			set_sbi_flag(sbi, SBI_NEED_FSCK);
213 			dump_stack();
214 			goto err;
215 		} else {
216 			return __is_bitmap_valid(sbi, blkaddr, type);
217 		}
218 		break;
219 	case META_GENERIC:
220 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
221 			blkaddr >= MAIN_BLKADDR(sbi)))
222 			goto err;
223 		break;
224 	default:
225 		BUG();
226 	}
227 
228 	return true;
229 err:
230 	f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
231 check_only:
232 	return false;
233 }
234 
235 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
236 					block_t blkaddr, int type)
237 {
238 	if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY))
239 		return false;
240 	return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
241 }
242 
243 bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi,
244 					block_t blkaddr, int type)
245 {
246 	return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
247 }
248 
249 /*
250  * Readahead CP/NAT/SIT/SSA/POR pages
251  */
252 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
253 							int type, bool sync)
254 {
255 	struct page *page;
256 	block_t blkno = start;
257 	struct f2fs_io_info fio = {
258 		.sbi = sbi,
259 		.type = META,
260 		.op = REQ_OP_READ,
261 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
262 		.encrypted_page = NULL,
263 		.in_list = 0,
264 		.is_por = (type == META_POR) ? 1 : 0,
265 	};
266 	struct blk_plug plug;
267 	int err;
268 
269 	if (unlikely(type == META_POR))
270 		fio.op_flags &= ~REQ_META;
271 
272 	blk_start_plug(&plug);
273 	for (; nrpages-- > 0; blkno++) {
274 
275 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
276 			goto out;
277 
278 		switch (type) {
279 		case META_NAT:
280 			if (unlikely(blkno >=
281 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
282 				blkno = 0;
283 			/* get nat block addr */
284 			fio.new_blkaddr = current_nat_addr(sbi,
285 					blkno * NAT_ENTRY_PER_BLOCK);
286 			break;
287 		case META_SIT:
288 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
289 				goto out;
290 			/* get sit block addr */
291 			fio.new_blkaddr = current_sit_addr(sbi,
292 					blkno * SIT_ENTRY_PER_BLOCK);
293 			break;
294 		case META_SSA:
295 		case META_CP:
296 		case META_POR:
297 			fio.new_blkaddr = blkno;
298 			break;
299 		default:
300 			BUG();
301 		}
302 
303 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
304 						fio.new_blkaddr, false);
305 		if (!page)
306 			continue;
307 		if (PageUptodate(page)) {
308 			f2fs_put_page(page, 1);
309 			continue;
310 		}
311 
312 		fio.page = page;
313 		err = f2fs_submit_page_bio(&fio);
314 		f2fs_put_page(page, err ? 1 : 0);
315 
316 		if (!err)
317 			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
318 							F2FS_BLKSIZE);
319 	}
320 out:
321 	blk_finish_plug(&plug);
322 	return blkno - start;
323 }
324 
325 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
326 							unsigned int ra_blocks)
327 {
328 	struct page *page;
329 	bool readahead = false;
330 
331 	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
332 		return;
333 
334 	page = find_get_page(META_MAPPING(sbi), index);
335 	if (!page || !PageUptodate(page))
336 		readahead = true;
337 	f2fs_put_page(page, 0);
338 
339 	if (readahead)
340 		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
341 }
342 
343 static int __f2fs_write_meta_page(struct page *page,
344 				struct writeback_control *wbc,
345 				enum iostat_type io_type)
346 {
347 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
348 
349 	trace_f2fs_writepage(page_folio(page), META);
350 
351 	if (unlikely(f2fs_cp_error(sbi))) {
352 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
353 			ClearPageUptodate(page);
354 			dec_page_count(sbi, F2FS_DIRTY_META);
355 			unlock_page(page);
356 			return 0;
357 		}
358 		goto redirty_out;
359 	}
360 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
361 		goto redirty_out;
362 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
363 		goto redirty_out;
364 
365 	f2fs_do_write_meta_page(sbi, page, io_type);
366 	dec_page_count(sbi, F2FS_DIRTY_META);
367 
368 	if (wbc->for_reclaim)
369 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
370 
371 	unlock_page(page);
372 
373 	if (unlikely(f2fs_cp_error(sbi)))
374 		f2fs_submit_merged_write(sbi, META);
375 
376 	return 0;
377 
378 redirty_out:
379 	redirty_page_for_writepage(wbc, page);
380 	return AOP_WRITEPAGE_ACTIVATE;
381 }
382 
383 static int f2fs_write_meta_page(struct page *page,
384 				struct writeback_control *wbc)
385 {
386 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
387 }
388 
389 static int f2fs_write_meta_pages(struct address_space *mapping,
390 				struct writeback_control *wbc)
391 {
392 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
393 	long diff, written;
394 
395 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
396 		goto skip_write;
397 
398 	/* collect a number of dirty meta pages and write together */
399 	if (wbc->sync_mode != WB_SYNC_ALL &&
400 			get_pages(sbi, F2FS_DIRTY_META) <
401 					nr_pages_to_skip(sbi, META))
402 		goto skip_write;
403 
404 	/* if locked failed, cp will flush dirty pages instead */
405 	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
406 		goto skip_write;
407 
408 	trace_f2fs_writepages(mapping->host, wbc, META);
409 	diff = nr_pages_to_write(sbi, META, wbc);
410 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
411 	f2fs_up_write(&sbi->cp_global_sem);
412 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
413 	return 0;
414 
415 skip_write:
416 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
417 	trace_f2fs_writepages(mapping->host, wbc, META);
418 	return 0;
419 }
420 
421 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
422 				long nr_to_write, enum iostat_type io_type)
423 {
424 	struct address_space *mapping = META_MAPPING(sbi);
425 	pgoff_t index = 0, prev = ULONG_MAX;
426 	struct folio_batch fbatch;
427 	long nwritten = 0;
428 	int nr_folios;
429 	struct writeback_control wbc = {
430 		.for_reclaim = 0,
431 	};
432 	struct blk_plug plug;
433 
434 	folio_batch_init(&fbatch);
435 
436 	blk_start_plug(&plug);
437 
438 	while ((nr_folios = filemap_get_folios_tag(mapping, &index,
439 					(pgoff_t)-1,
440 					PAGECACHE_TAG_DIRTY, &fbatch))) {
441 		int i;
442 
443 		for (i = 0; i < nr_folios; i++) {
444 			struct folio *folio = fbatch.folios[i];
445 
446 			if (nr_to_write != LONG_MAX && i != 0 &&
447 					folio->index != prev +
448 					folio_nr_pages(fbatch.folios[i-1])) {
449 				folio_batch_release(&fbatch);
450 				goto stop;
451 			}
452 
453 			folio_lock(folio);
454 
455 			if (unlikely(folio->mapping != mapping)) {
456 continue_unlock:
457 				folio_unlock(folio);
458 				continue;
459 			}
460 			if (!folio_test_dirty(folio)) {
461 				/* someone wrote it for us */
462 				goto continue_unlock;
463 			}
464 
465 			f2fs_wait_on_page_writeback(&folio->page, META,
466 					true, true);
467 
468 			if (!folio_clear_dirty_for_io(folio))
469 				goto continue_unlock;
470 
471 			if (__f2fs_write_meta_page(&folio->page, &wbc,
472 						io_type)) {
473 				folio_unlock(folio);
474 				break;
475 			}
476 			nwritten += folio_nr_pages(folio);
477 			prev = folio->index;
478 			if (unlikely(nwritten >= nr_to_write))
479 				break;
480 		}
481 		folio_batch_release(&fbatch);
482 		cond_resched();
483 	}
484 stop:
485 	if (nwritten)
486 		f2fs_submit_merged_write(sbi, type);
487 
488 	blk_finish_plug(&plug);
489 
490 	return nwritten;
491 }
492 
493 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
494 		struct folio *folio)
495 {
496 	trace_f2fs_set_page_dirty(folio, META);
497 
498 	if (!folio_test_uptodate(folio))
499 		folio_mark_uptodate(folio);
500 	if (filemap_dirty_folio(mapping, folio)) {
501 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
502 		set_page_private_reference(&folio->page);
503 		return true;
504 	}
505 	return false;
506 }
507 
508 const struct address_space_operations f2fs_meta_aops = {
509 	.writepage	= f2fs_write_meta_page,
510 	.writepages	= f2fs_write_meta_pages,
511 	.dirty_folio	= f2fs_dirty_meta_folio,
512 	.invalidate_folio = f2fs_invalidate_folio,
513 	.release_folio	= f2fs_release_folio,
514 	.migrate_folio	= filemap_migrate_folio,
515 };
516 
517 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
518 						unsigned int devidx, int type)
519 {
520 	struct inode_management *im = &sbi->im[type];
521 	struct ino_entry *e = NULL, *new = NULL;
522 
523 	if (type == FLUSH_INO) {
524 		rcu_read_lock();
525 		e = radix_tree_lookup(&im->ino_root, ino);
526 		rcu_read_unlock();
527 	}
528 
529 retry:
530 	if (!e)
531 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
532 						GFP_NOFS, true, NULL);
533 
534 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
535 
536 	spin_lock(&im->ino_lock);
537 	e = radix_tree_lookup(&im->ino_root, ino);
538 	if (!e) {
539 		if (!new) {
540 			spin_unlock(&im->ino_lock);
541 			radix_tree_preload_end();
542 			goto retry;
543 		}
544 		e = new;
545 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
546 			f2fs_bug_on(sbi, 1);
547 
548 		memset(e, 0, sizeof(struct ino_entry));
549 		e->ino = ino;
550 
551 		list_add_tail(&e->list, &im->ino_list);
552 		if (type != ORPHAN_INO)
553 			im->ino_num++;
554 	}
555 
556 	if (type == FLUSH_INO)
557 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
558 
559 	spin_unlock(&im->ino_lock);
560 	radix_tree_preload_end();
561 
562 	if (new && e != new)
563 		kmem_cache_free(ino_entry_slab, new);
564 }
565 
566 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
567 {
568 	struct inode_management *im = &sbi->im[type];
569 	struct ino_entry *e;
570 
571 	spin_lock(&im->ino_lock);
572 	e = radix_tree_lookup(&im->ino_root, ino);
573 	if (e) {
574 		list_del(&e->list);
575 		radix_tree_delete(&im->ino_root, ino);
576 		im->ino_num--;
577 		spin_unlock(&im->ino_lock);
578 		kmem_cache_free(ino_entry_slab, e);
579 		return;
580 	}
581 	spin_unlock(&im->ino_lock);
582 }
583 
584 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
585 {
586 	/* add new dirty ino entry into list */
587 	__add_ino_entry(sbi, ino, 0, type);
588 }
589 
590 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
591 {
592 	/* remove dirty ino entry from list */
593 	__remove_ino_entry(sbi, ino, type);
594 }
595 
596 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
597 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
598 {
599 	struct inode_management *im = &sbi->im[mode];
600 	struct ino_entry *e;
601 
602 	spin_lock(&im->ino_lock);
603 	e = radix_tree_lookup(&im->ino_root, ino);
604 	spin_unlock(&im->ino_lock);
605 	return e ? true : false;
606 }
607 
608 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
609 {
610 	struct ino_entry *e, *tmp;
611 	int i;
612 
613 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
614 		struct inode_management *im = &sbi->im[i];
615 
616 		spin_lock(&im->ino_lock);
617 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
618 			list_del(&e->list);
619 			radix_tree_delete(&im->ino_root, e->ino);
620 			kmem_cache_free(ino_entry_slab, e);
621 			im->ino_num--;
622 		}
623 		spin_unlock(&im->ino_lock);
624 	}
625 }
626 
627 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
628 					unsigned int devidx, int type)
629 {
630 	__add_ino_entry(sbi, ino, devidx, type);
631 }
632 
633 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
634 					unsigned int devidx, int type)
635 {
636 	struct inode_management *im = &sbi->im[type];
637 	struct ino_entry *e;
638 	bool is_dirty = false;
639 
640 	spin_lock(&im->ino_lock);
641 	e = radix_tree_lookup(&im->ino_root, ino);
642 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
643 		is_dirty = true;
644 	spin_unlock(&im->ino_lock);
645 	return is_dirty;
646 }
647 
648 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
649 {
650 	struct inode_management *im = &sbi->im[ORPHAN_INO];
651 	int err = 0;
652 
653 	spin_lock(&im->ino_lock);
654 
655 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
656 		spin_unlock(&im->ino_lock);
657 		return -ENOSPC;
658 	}
659 
660 	if (unlikely(im->ino_num >= sbi->max_orphans))
661 		err = -ENOSPC;
662 	else
663 		im->ino_num++;
664 	spin_unlock(&im->ino_lock);
665 
666 	return err;
667 }
668 
669 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
670 {
671 	struct inode_management *im = &sbi->im[ORPHAN_INO];
672 
673 	spin_lock(&im->ino_lock);
674 	f2fs_bug_on(sbi, im->ino_num == 0);
675 	im->ino_num--;
676 	spin_unlock(&im->ino_lock);
677 }
678 
679 void f2fs_add_orphan_inode(struct inode *inode)
680 {
681 	/* add new orphan ino entry into list */
682 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
683 	f2fs_update_inode_page(inode);
684 }
685 
686 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
687 {
688 	/* remove orphan entry from orphan list */
689 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
690 }
691 
692 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
693 {
694 	struct inode *inode;
695 	struct node_info ni;
696 	int err;
697 
698 	inode = f2fs_iget_retry(sbi->sb, ino);
699 	if (IS_ERR(inode)) {
700 		/*
701 		 * there should be a bug that we can't find the entry
702 		 * to orphan inode.
703 		 */
704 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
705 		return PTR_ERR(inode);
706 	}
707 
708 	err = f2fs_dquot_initialize(inode);
709 	if (err) {
710 		iput(inode);
711 		goto err_out;
712 	}
713 
714 	clear_nlink(inode);
715 
716 	/* truncate all the data during iput */
717 	iput(inode);
718 
719 	err = f2fs_get_node_info(sbi, ino, &ni, false);
720 	if (err)
721 		goto err_out;
722 
723 	/* ENOMEM was fully retried in f2fs_evict_inode. */
724 	if (ni.blk_addr != NULL_ADDR) {
725 		err = -EIO;
726 		goto err_out;
727 	}
728 	return 0;
729 
730 err_out:
731 	set_sbi_flag(sbi, SBI_NEED_FSCK);
732 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
733 		  __func__, ino);
734 	return err;
735 }
736 
737 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
738 {
739 	block_t start_blk, orphan_blocks, i, j;
740 	int err = 0;
741 
742 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
743 		return 0;
744 
745 	if (f2fs_hw_is_readonly(sbi)) {
746 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
747 		return 0;
748 	}
749 
750 	if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
751 		f2fs_info(sbi, "orphan cleanup on readonly fs");
752 
753 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
754 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
755 
756 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
757 
758 	for (i = 0; i < orphan_blocks; i++) {
759 		struct page *page;
760 		struct f2fs_orphan_block *orphan_blk;
761 
762 		page = f2fs_get_meta_page(sbi, start_blk + i);
763 		if (IS_ERR(page)) {
764 			err = PTR_ERR(page);
765 			goto out;
766 		}
767 
768 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
769 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
770 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
771 
772 			err = recover_orphan_inode(sbi, ino);
773 			if (err) {
774 				f2fs_put_page(page, 1);
775 				goto out;
776 			}
777 		}
778 		f2fs_put_page(page, 1);
779 	}
780 	/* clear Orphan Flag */
781 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
782 out:
783 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
784 
785 	return err;
786 }
787 
788 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
789 {
790 	struct list_head *head;
791 	struct f2fs_orphan_block *orphan_blk = NULL;
792 	unsigned int nentries = 0;
793 	unsigned short index = 1;
794 	unsigned short orphan_blocks;
795 	struct page *page = NULL;
796 	struct ino_entry *orphan = NULL;
797 	struct inode_management *im = &sbi->im[ORPHAN_INO];
798 
799 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
800 
801 	/*
802 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
803 	 * orphan inode operations are covered under f2fs_lock_op().
804 	 * And, spin_lock should be avoided due to page operations below.
805 	 */
806 	head = &im->ino_list;
807 
808 	/* loop for each orphan inode entry and write them in journal block */
809 	list_for_each_entry(orphan, head, list) {
810 		if (!page) {
811 			page = f2fs_grab_meta_page(sbi, start_blk++);
812 			orphan_blk =
813 				(struct f2fs_orphan_block *)page_address(page);
814 			memset(orphan_blk, 0, sizeof(*orphan_blk));
815 		}
816 
817 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
818 
819 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
820 			/*
821 			 * an orphan block is full of 1020 entries,
822 			 * then we need to flush current orphan blocks
823 			 * and bring another one in memory
824 			 */
825 			orphan_blk->blk_addr = cpu_to_le16(index);
826 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
827 			orphan_blk->entry_count = cpu_to_le32(nentries);
828 			set_page_dirty(page);
829 			f2fs_put_page(page, 1);
830 			index++;
831 			nentries = 0;
832 			page = NULL;
833 		}
834 	}
835 
836 	if (page) {
837 		orphan_blk->blk_addr = cpu_to_le16(index);
838 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
839 		orphan_blk->entry_count = cpu_to_le32(nentries);
840 		set_page_dirty(page);
841 		f2fs_put_page(page, 1);
842 	}
843 }
844 
845 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
846 						struct f2fs_checkpoint *ckpt)
847 {
848 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
849 	__u32 chksum;
850 
851 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
852 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
853 		chksum_ofs += sizeof(chksum);
854 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
855 						F2FS_BLKSIZE - chksum_ofs);
856 	}
857 	return chksum;
858 }
859 
860 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
861 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
862 		unsigned long long *version)
863 {
864 	size_t crc_offset = 0;
865 	__u32 crc;
866 
867 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
868 	if (IS_ERR(*cp_page))
869 		return PTR_ERR(*cp_page);
870 
871 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
872 
873 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
874 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
875 			crc_offset > CP_CHKSUM_OFFSET) {
876 		f2fs_put_page(*cp_page, 1);
877 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
878 		return -EINVAL;
879 	}
880 
881 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
882 	if (crc != cur_cp_crc(*cp_block)) {
883 		f2fs_put_page(*cp_page, 1);
884 		f2fs_warn(sbi, "invalid crc value");
885 		return -EINVAL;
886 	}
887 
888 	*version = cur_cp_version(*cp_block);
889 	return 0;
890 }
891 
892 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
893 				block_t cp_addr, unsigned long long *version)
894 {
895 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
896 	struct f2fs_checkpoint *cp_block = NULL;
897 	unsigned long long cur_version = 0, pre_version = 0;
898 	unsigned int cp_blocks;
899 	int err;
900 
901 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
902 					&cp_page_1, version);
903 	if (err)
904 		return NULL;
905 
906 	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
907 
908 	if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
909 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
910 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
911 		goto invalid_cp;
912 	}
913 	pre_version = *version;
914 
915 	cp_addr += cp_blocks - 1;
916 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
917 					&cp_page_2, version);
918 	if (err)
919 		goto invalid_cp;
920 	cur_version = *version;
921 
922 	if (cur_version == pre_version) {
923 		*version = cur_version;
924 		f2fs_put_page(cp_page_2, 1);
925 		return cp_page_1;
926 	}
927 	f2fs_put_page(cp_page_2, 1);
928 invalid_cp:
929 	f2fs_put_page(cp_page_1, 1);
930 	return NULL;
931 }
932 
933 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
934 {
935 	struct f2fs_checkpoint *cp_block;
936 	struct f2fs_super_block *fsb = sbi->raw_super;
937 	struct page *cp1, *cp2, *cur_page;
938 	unsigned long blk_size = sbi->blocksize;
939 	unsigned long long cp1_version = 0, cp2_version = 0;
940 	unsigned long long cp_start_blk_no;
941 	unsigned int cp_blks = 1 + __cp_payload(sbi);
942 	block_t cp_blk_no;
943 	int i;
944 	int err;
945 
946 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
947 				  GFP_KERNEL);
948 	if (!sbi->ckpt)
949 		return -ENOMEM;
950 	/*
951 	 * Finding out valid cp block involves read both
952 	 * sets( cp pack 1 and cp pack 2)
953 	 */
954 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
955 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
956 
957 	/* The second checkpoint pack should start at the next segment */
958 	cp_start_blk_no += ((unsigned long long)1) <<
959 				le32_to_cpu(fsb->log_blocks_per_seg);
960 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
961 
962 	if (cp1 && cp2) {
963 		if (ver_after(cp2_version, cp1_version))
964 			cur_page = cp2;
965 		else
966 			cur_page = cp1;
967 	} else if (cp1) {
968 		cur_page = cp1;
969 	} else if (cp2) {
970 		cur_page = cp2;
971 	} else {
972 		err = -EFSCORRUPTED;
973 		goto fail_no_cp;
974 	}
975 
976 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
977 	memcpy(sbi->ckpt, cp_block, blk_size);
978 
979 	if (cur_page == cp1)
980 		sbi->cur_cp_pack = 1;
981 	else
982 		sbi->cur_cp_pack = 2;
983 
984 	/* Sanity checking of checkpoint */
985 	if (f2fs_sanity_check_ckpt(sbi)) {
986 		err = -EFSCORRUPTED;
987 		goto free_fail_no_cp;
988 	}
989 
990 	if (cp_blks <= 1)
991 		goto done;
992 
993 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
994 	if (cur_page == cp2)
995 		cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
996 
997 	for (i = 1; i < cp_blks; i++) {
998 		void *sit_bitmap_ptr;
999 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
1000 
1001 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
1002 		if (IS_ERR(cur_page)) {
1003 			err = PTR_ERR(cur_page);
1004 			goto free_fail_no_cp;
1005 		}
1006 		sit_bitmap_ptr = page_address(cur_page);
1007 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1008 		f2fs_put_page(cur_page, 1);
1009 	}
1010 done:
1011 	f2fs_put_page(cp1, 1);
1012 	f2fs_put_page(cp2, 1);
1013 	return 0;
1014 
1015 free_fail_no_cp:
1016 	f2fs_put_page(cp1, 1);
1017 	f2fs_put_page(cp2, 1);
1018 fail_no_cp:
1019 	kvfree(sbi->ckpt);
1020 	return err;
1021 }
1022 
1023 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1024 {
1025 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1026 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1027 
1028 	if (is_inode_flag_set(inode, flag))
1029 		return;
1030 
1031 	set_inode_flag(inode, flag);
1032 	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1033 	stat_inc_dirty_inode(sbi, type);
1034 }
1035 
1036 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1037 {
1038 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1039 
1040 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1041 		return;
1042 
1043 	list_del_init(&F2FS_I(inode)->dirty_list);
1044 	clear_inode_flag(inode, flag);
1045 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1046 }
1047 
1048 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1049 {
1050 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1051 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1052 
1053 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1054 			!S_ISLNK(inode->i_mode))
1055 		return;
1056 
1057 	spin_lock(&sbi->inode_lock[type]);
1058 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1059 		__add_dirty_inode(inode, type);
1060 	inode_inc_dirty_pages(inode);
1061 	spin_unlock(&sbi->inode_lock[type]);
1062 
1063 	set_page_private_reference(&folio->page);
1064 }
1065 
1066 void f2fs_remove_dirty_inode(struct inode *inode)
1067 {
1068 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1069 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1070 
1071 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1072 			!S_ISLNK(inode->i_mode))
1073 		return;
1074 
1075 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1076 		return;
1077 
1078 	spin_lock(&sbi->inode_lock[type]);
1079 	__remove_dirty_inode(inode, type);
1080 	spin_unlock(&sbi->inode_lock[type]);
1081 }
1082 
1083 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1084 						bool from_cp)
1085 {
1086 	struct list_head *head;
1087 	struct inode *inode;
1088 	struct f2fs_inode_info *fi;
1089 	bool is_dir = (type == DIR_INODE);
1090 	unsigned long ino = 0;
1091 
1092 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1093 				get_pages(sbi, is_dir ?
1094 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1095 retry:
1096 	if (unlikely(f2fs_cp_error(sbi))) {
1097 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1098 				get_pages(sbi, is_dir ?
1099 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1100 		return -EIO;
1101 	}
1102 
1103 	spin_lock(&sbi->inode_lock[type]);
1104 
1105 	head = &sbi->inode_list[type];
1106 	if (list_empty(head)) {
1107 		spin_unlock(&sbi->inode_lock[type]);
1108 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1109 				get_pages(sbi, is_dir ?
1110 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1111 		return 0;
1112 	}
1113 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1114 	inode = igrab(&fi->vfs_inode);
1115 	spin_unlock(&sbi->inode_lock[type]);
1116 	if (inode) {
1117 		unsigned long cur_ino = inode->i_ino;
1118 
1119 		if (from_cp)
1120 			F2FS_I(inode)->cp_task = current;
1121 		F2FS_I(inode)->wb_task = current;
1122 
1123 		filemap_fdatawrite(inode->i_mapping);
1124 
1125 		F2FS_I(inode)->wb_task = NULL;
1126 		if (from_cp)
1127 			F2FS_I(inode)->cp_task = NULL;
1128 
1129 		iput(inode);
1130 		/* We need to give cpu to another writers. */
1131 		if (ino == cur_ino)
1132 			cond_resched();
1133 		else
1134 			ino = cur_ino;
1135 	} else {
1136 		/*
1137 		 * We should submit bio, since it exists several
1138 		 * writebacking dentry pages in the freeing inode.
1139 		 */
1140 		f2fs_submit_merged_write(sbi, DATA);
1141 		cond_resched();
1142 	}
1143 	goto retry;
1144 }
1145 
1146 static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1147 {
1148 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1149 	struct inode *inode;
1150 	struct f2fs_inode_info *fi;
1151 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1152 
1153 	while (total--) {
1154 		if (unlikely(f2fs_cp_error(sbi)))
1155 			return -EIO;
1156 
1157 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1158 		if (list_empty(head)) {
1159 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1160 			return 0;
1161 		}
1162 		fi = list_first_entry(head, struct f2fs_inode_info,
1163 							gdirty_list);
1164 		inode = igrab(&fi->vfs_inode);
1165 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1166 		if (inode) {
1167 			sync_inode_metadata(inode, 0);
1168 
1169 			/* it's on eviction */
1170 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1171 				f2fs_update_inode_page(inode);
1172 			iput(inode);
1173 		}
1174 	}
1175 	return 0;
1176 }
1177 
1178 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1179 {
1180 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1181 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1182 	nid_t last_nid = nm_i->next_scan_nid;
1183 
1184 	next_free_nid(sbi, &last_nid);
1185 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1186 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1187 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1188 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1189 
1190 	/* update user_block_counts */
1191 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1192 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1193 	percpu_counter_set(&sbi->rf_node_block_count, 0);
1194 }
1195 
1196 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1197 {
1198 	bool ret = false;
1199 
1200 	if (!is_journalled_quota(sbi))
1201 		return false;
1202 
1203 	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1204 		return true;
1205 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1206 		ret = false;
1207 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1208 		ret = false;
1209 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1210 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1211 		ret = true;
1212 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1213 		ret = true;
1214 	}
1215 	f2fs_up_write(&sbi->quota_sem);
1216 	return ret;
1217 }
1218 
1219 /*
1220  * Freeze all the FS-operations for checkpoint.
1221  */
1222 static int block_operations(struct f2fs_sb_info *sbi)
1223 {
1224 	struct writeback_control wbc = {
1225 		.sync_mode = WB_SYNC_ALL,
1226 		.nr_to_write = LONG_MAX,
1227 		.for_reclaim = 0,
1228 	};
1229 	int err = 0, cnt = 0;
1230 
1231 	/*
1232 	 * Let's flush inline_data in dirty node pages.
1233 	 */
1234 	f2fs_flush_inline_data(sbi);
1235 
1236 retry_flush_quotas:
1237 	f2fs_lock_all(sbi);
1238 	if (__need_flush_quota(sbi)) {
1239 		int locked;
1240 
1241 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1242 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1243 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1244 			goto retry_flush_dents;
1245 		}
1246 		f2fs_unlock_all(sbi);
1247 
1248 		/* only failed during mount/umount/freeze/quotactl */
1249 		locked = down_read_trylock(&sbi->sb->s_umount);
1250 		f2fs_quota_sync(sbi->sb, -1);
1251 		if (locked)
1252 			up_read(&sbi->sb->s_umount);
1253 		cond_resched();
1254 		goto retry_flush_quotas;
1255 	}
1256 
1257 retry_flush_dents:
1258 	/* write all the dirty dentry pages */
1259 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1260 		f2fs_unlock_all(sbi);
1261 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1262 		if (err)
1263 			return err;
1264 		cond_resched();
1265 		goto retry_flush_quotas;
1266 	}
1267 
1268 	/*
1269 	 * POR: we should ensure that there are no dirty node pages
1270 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1271 	 */
1272 	f2fs_down_write(&sbi->node_change);
1273 
1274 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1275 		f2fs_up_write(&sbi->node_change);
1276 		f2fs_unlock_all(sbi);
1277 		err = f2fs_sync_inode_meta(sbi);
1278 		if (err)
1279 			return err;
1280 		cond_resched();
1281 		goto retry_flush_quotas;
1282 	}
1283 
1284 retry_flush_nodes:
1285 	f2fs_down_write(&sbi->node_write);
1286 
1287 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1288 		f2fs_up_write(&sbi->node_write);
1289 		atomic_inc(&sbi->wb_sync_req[NODE]);
1290 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1291 		atomic_dec(&sbi->wb_sync_req[NODE]);
1292 		if (err) {
1293 			f2fs_up_write(&sbi->node_change);
1294 			f2fs_unlock_all(sbi);
1295 			return err;
1296 		}
1297 		cond_resched();
1298 		goto retry_flush_nodes;
1299 	}
1300 
1301 	/*
1302 	 * sbi->node_change is used only for AIO write_begin path which produces
1303 	 * dirty node blocks and some checkpoint values by block allocation.
1304 	 */
1305 	__prepare_cp_block(sbi);
1306 	f2fs_up_write(&sbi->node_change);
1307 	return err;
1308 }
1309 
1310 static void unblock_operations(struct f2fs_sb_info *sbi)
1311 {
1312 	f2fs_up_write(&sbi->node_write);
1313 	f2fs_unlock_all(sbi);
1314 }
1315 
1316 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1317 {
1318 	DEFINE_WAIT(wait);
1319 
1320 	for (;;) {
1321 		if (!get_pages(sbi, type))
1322 			break;
1323 
1324 		if (unlikely(f2fs_cp_error(sbi) &&
1325 			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1326 			break;
1327 
1328 		if (type == F2FS_DIRTY_META)
1329 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1330 							FS_CP_META_IO);
1331 		else if (type == F2FS_WB_CP_DATA)
1332 			f2fs_submit_merged_write(sbi, DATA);
1333 
1334 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1335 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1336 	}
1337 	finish_wait(&sbi->cp_wait, &wait);
1338 }
1339 
1340 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1341 {
1342 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1343 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1344 	unsigned long flags;
1345 
1346 	if (cpc->reason & CP_UMOUNT) {
1347 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1348 			NM_I(sbi)->nat_bits_blocks > BLKS_PER_SEG(sbi)) {
1349 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1350 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1351 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1352 						f2fs_nat_bitmap_enabled(sbi)) {
1353 			f2fs_enable_nat_bits(sbi);
1354 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1355 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1356 		}
1357 	}
1358 
1359 	spin_lock_irqsave(&sbi->cp_lock, flags);
1360 
1361 	if (cpc->reason & CP_TRIMMED)
1362 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1363 	else
1364 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1365 
1366 	if (cpc->reason & CP_UMOUNT)
1367 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1368 	else
1369 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1370 
1371 	if (cpc->reason & CP_FASTBOOT)
1372 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1373 	else
1374 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1375 
1376 	if (orphan_num)
1377 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1378 	else
1379 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1380 
1381 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1382 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1383 
1384 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1385 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1386 	else
1387 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1388 
1389 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1390 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1391 	else
1392 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1393 
1394 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1395 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1396 	else
1397 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1398 
1399 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1400 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1401 	else
1402 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1403 
1404 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1405 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1406 
1407 	/* set this flag to activate crc|cp_ver for recovery */
1408 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1409 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1410 
1411 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1412 }
1413 
1414 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1415 	void *src, block_t blk_addr)
1416 {
1417 	struct writeback_control wbc = {
1418 		.for_reclaim = 0,
1419 	};
1420 
1421 	/*
1422 	 * filemap_get_folios_tag and lock_page again will take
1423 	 * some extra time. Therefore, f2fs_update_meta_pages and
1424 	 * f2fs_sync_meta_pages are combined in this function.
1425 	 */
1426 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1427 	int err;
1428 
1429 	f2fs_wait_on_page_writeback(page, META, true, true);
1430 
1431 	memcpy(page_address(page), src, PAGE_SIZE);
1432 
1433 	set_page_dirty(page);
1434 	if (unlikely(!clear_page_dirty_for_io(page)))
1435 		f2fs_bug_on(sbi, 1);
1436 
1437 	/* writeout cp pack 2 page */
1438 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1439 	if (unlikely(err && f2fs_cp_error(sbi))) {
1440 		f2fs_put_page(page, 1);
1441 		return;
1442 	}
1443 
1444 	f2fs_bug_on(sbi, err);
1445 	f2fs_put_page(page, 0);
1446 
1447 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1448 	f2fs_submit_merged_write(sbi, META_FLUSH);
1449 }
1450 
1451 static inline u64 get_sectors_written(struct block_device *bdev)
1452 {
1453 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1454 }
1455 
1456 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1457 {
1458 	if (f2fs_is_multi_device(sbi)) {
1459 		u64 sectors = 0;
1460 		int i;
1461 
1462 		for (i = 0; i < sbi->s_ndevs; i++)
1463 			sectors += get_sectors_written(FDEV(i).bdev);
1464 
1465 		return sectors;
1466 	}
1467 
1468 	return get_sectors_written(sbi->sb->s_bdev);
1469 }
1470 
1471 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1472 {
1473 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1474 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1475 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1476 	block_t start_blk;
1477 	unsigned int data_sum_blocks, orphan_blocks;
1478 	__u32 crc32 = 0;
1479 	int i;
1480 	int cp_payload_blks = __cp_payload(sbi);
1481 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1482 	u64 kbytes_written;
1483 	int err;
1484 
1485 	/* Flush all the NAT/SIT pages */
1486 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1487 
1488 	/* start to update checkpoint, cp ver is already updated previously */
1489 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1490 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1491 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1492 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1493 
1494 		ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1495 		ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1496 		ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1497 	}
1498 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1499 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1500 
1501 		ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1502 		ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1503 		ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1504 	}
1505 
1506 	/* 2 cp + n data seg summary + orphan inode blocks */
1507 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1508 	spin_lock_irqsave(&sbi->cp_lock, flags);
1509 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1510 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1511 	else
1512 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1513 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1514 
1515 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1516 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1517 			orphan_blocks);
1518 
1519 	if (__remain_node_summaries(cpc->reason))
1520 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1521 				cp_payload_blks + data_sum_blocks +
1522 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1523 	else
1524 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1525 				cp_payload_blks + data_sum_blocks +
1526 				orphan_blocks);
1527 
1528 	/* update ckpt flag for checkpoint */
1529 	update_ckpt_flags(sbi, cpc);
1530 
1531 	/* update SIT/NAT bitmap */
1532 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1533 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1534 
1535 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1536 	*((__le32 *)((unsigned char *)ckpt +
1537 				le32_to_cpu(ckpt->checksum_offset)))
1538 				= cpu_to_le32(crc32);
1539 
1540 	start_blk = __start_cp_next_addr(sbi);
1541 
1542 	/* write nat bits */
1543 	if ((cpc->reason & CP_UMOUNT) &&
1544 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1545 		__u64 cp_ver = cur_cp_version(ckpt);
1546 		block_t blk;
1547 
1548 		cp_ver |= ((__u64)crc32 << 32);
1549 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1550 
1551 		blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1552 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1553 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1554 					(i << F2FS_BLKSIZE_BITS), blk + i);
1555 	}
1556 
1557 	/* write out checkpoint buffer at block 0 */
1558 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1559 
1560 	for (i = 1; i < 1 + cp_payload_blks; i++)
1561 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1562 							start_blk++);
1563 
1564 	if (orphan_num) {
1565 		write_orphan_inodes(sbi, start_blk);
1566 		start_blk += orphan_blocks;
1567 	}
1568 
1569 	f2fs_write_data_summaries(sbi, start_blk);
1570 	start_blk += data_sum_blocks;
1571 
1572 	/* Record write statistics in the hot node summary */
1573 	kbytes_written = sbi->kbytes_written;
1574 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1575 				sbi->sectors_written_start) >> 1;
1576 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1577 
1578 	if (__remain_node_summaries(cpc->reason)) {
1579 		f2fs_write_node_summaries(sbi, start_blk);
1580 		start_blk += NR_CURSEG_NODE_TYPE;
1581 	}
1582 
1583 	/* Here, we have one bio having CP pack except cp pack 2 page */
1584 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1585 	/* Wait for all dirty meta pages to be submitted for IO */
1586 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1587 
1588 	/* wait for previous submitted meta pages writeback */
1589 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1590 
1591 	/* flush all device cache */
1592 	err = f2fs_flush_device_cache(sbi);
1593 	if (err)
1594 		return err;
1595 
1596 	/* barrier and flush checkpoint cp pack 2 page if it can */
1597 	commit_checkpoint(sbi, ckpt, start_blk);
1598 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1599 
1600 	/*
1601 	 * invalidate intermediate page cache borrowed from meta inode which are
1602 	 * used for migration of encrypted, verity or compressed inode's blocks.
1603 	 */
1604 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1605 		f2fs_sb_has_compression(sbi))
1606 		f2fs_bug_on(sbi,
1607 			invalidate_inode_pages2_range(META_MAPPING(sbi),
1608 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1609 
1610 	f2fs_release_ino_entry(sbi, false);
1611 
1612 	f2fs_reset_fsync_node_info(sbi);
1613 
1614 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1615 	clear_sbi_flag(sbi, SBI_NEED_CP);
1616 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1617 
1618 	spin_lock(&sbi->stat_lock);
1619 	sbi->unusable_block_count = 0;
1620 	spin_unlock(&sbi->stat_lock);
1621 
1622 	__set_cp_next_pack(sbi);
1623 
1624 	/*
1625 	 * redirty superblock if metadata like node page or inode cache is
1626 	 * updated during writing checkpoint.
1627 	 */
1628 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1629 			get_pages(sbi, F2FS_DIRTY_IMETA))
1630 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1631 
1632 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1633 
1634 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1635 }
1636 
1637 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1638 {
1639 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1640 	unsigned long long ckpt_ver;
1641 	int err = 0;
1642 
1643 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1644 		return -EROFS;
1645 
1646 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1647 		if (cpc->reason != CP_PAUSE)
1648 			return 0;
1649 		f2fs_warn(sbi, "Start checkpoint disabled!");
1650 	}
1651 	if (cpc->reason != CP_RESIZE)
1652 		f2fs_down_write(&sbi->cp_global_sem);
1653 
1654 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1655 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1656 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1657 		goto out;
1658 	if (unlikely(f2fs_cp_error(sbi))) {
1659 		err = -EIO;
1660 		goto out;
1661 	}
1662 
1663 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1664 
1665 	err = block_operations(sbi);
1666 	if (err)
1667 		goto out;
1668 
1669 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1670 
1671 	f2fs_flush_merged_writes(sbi);
1672 
1673 	/* this is the case of multiple fstrims without any changes */
1674 	if (cpc->reason & CP_DISCARD) {
1675 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1676 			unblock_operations(sbi);
1677 			goto out;
1678 		}
1679 
1680 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1681 				SIT_I(sbi)->dirty_sentries == 0 &&
1682 				prefree_segments(sbi) == 0) {
1683 			f2fs_flush_sit_entries(sbi, cpc);
1684 			f2fs_clear_prefree_segments(sbi, cpc);
1685 			unblock_operations(sbi);
1686 			goto out;
1687 		}
1688 	}
1689 
1690 	/*
1691 	 * update checkpoint pack index
1692 	 * Increase the version number so that
1693 	 * SIT entries and seg summaries are written at correct place
1694 	 */
1695 	ckpt_ver = cur_cp_version(ckpt);
1696 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1697 
1698 	/* write cached NAT/SIT entries to NAT/SIT area */
1699 	err = f2fs_flush_nat_entries(sbi, cpc);
1700 	if (err) {
1701 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1702 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1703 		goto stop;
1704 	}
1705 
1706 	f2fs_flush_sit_entries(sbi, cpc);
1707 
1708 	/* save inmem log status */
1709 	f2fs_save_inmem_curseg(sbi);
1710 
1711 	err = do_checkpoint(sbi, cpc);
1712 	if (err) {
1713 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1714 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1715 		f2fs_release_discard_addrs(sbi);
1716 	} else {
1717 		f2fs_clear_prefree_segments(sbi, cpc);
1718 	}
1719 
1720 	f2fs_restore_inmem_curseg(sbi);
1721 	f2fs_reinit_atgc_curseg(sbi);
1722 	stat_inc_cp_count(sbi);
1723 stop:
1724 	unblock_operations(sbi);
1725 
1726 	if (cpc->reason & CP_RECOVERY)
1727 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1728 
1729 	/* update CP_TIME to trigger checkpoint periodically */
1730 	f2fs_update_time(sbi, CP_TIME);
1731 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1732 out:
1733 	if (cpc->reason != CP_RESIZE)
1734 		f2fs_up_write(&sbi->cp_global_sem);
1735 	return err;
1736 }
1737 
1738 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1739 {
1740 	int i;
1741 
1742 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1743 		struct inode_management *im = &sbi->im[i];
1744 
1745 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1746 		spin_lock_init(&im->ino_lock);
1747 		INIT_LIST_HEAD(&im->ino_list);
1748 		im->ino_num = 0;
1749 	}
1750 
1751 	sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1752 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1753 			F2FS_ORPHANS_PER_BLOCK;
1754 }
1755 
1756 int __init f2fs_create_checkpoint_caches(void)
1757 {
1758 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1759 			sizeof(struct ino_entry));
1760 	if (!ino_entry_slab)
1761 		return -ENOMEM;
1762 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1763 			sizeof(struct inode_entry));
1764 	if (!f2fs_inode_entry_slab) {
1765 		kmem_cache_destroy(ino_entry_slab);
1766 		return -ENOMEM;
1767 	}
1768 	return 0;
1769 }
1770 
1771 void f2fs_destroy_checkpoint_caches(void)
1772 {
1773 	kmem_cache_destroy(ino_entry_slab);
1774 	kmem_cache_destroy(f2fs_inode_entry_slab);
1775 }
1776 
1777 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1778 {
1779 	struct cp_control cpc = { .reason = CP_SYNC, };
1780 	int err;
1781 
1782 	f2fs_down_write(&sbi->gc_lock);
1783 	err = f2fs_write_checkpoint(sbi, &cpc);
1784 	f2fs_up_write(&sbi->gc_lock);
1785 
1786 	return err;
1787 }
1788 
1789 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1790 {
1791 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1792 	struct ckpt_req *req, *next;
1793 	struct llist_node *dispatch_list;
1794 	u64 sum_diff = 0, diff, count = 0;
1795 	int ret;
1796 
1797 	dispatch_list = llist_del_all(&cprc->issue_list);
1798 	if (!dispatch_list)
1799 		return;
1800 	dispatch_list = llist_reverse_order(dispatch_list);
1801 
1802 	ret = __write_checkpoint_sync(sbi);
1803 	atomic_inc(&cprc->issued_ckpt);
1804 
1805 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1806 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1807 		req->ret = ret;
1808 		complete(&req->wait);
1809 
1810 		sum_diff += diff;
1811 		count++;
1812 	}
1813 	atomic_sub(count, &cprc->queued_ckpt);
1814 	atomic_add(count, &cprc->total_ckpt);
1815 
1816 	spin_lock(&cprc->stat_lock);
1817 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1818 	if (cprc->peak_time < cprc->cur_time)
1819 		cprc->peak_time = cprc->cur_time;
1820 	spin_unlock(&cprc->stat_lock);
1821 }
1822 
1823 static int issue_checkpoint_thread(void *data)
1824 {
1825 	struct f2fs_sb_info *sbi = data;
1826 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1827 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1828 repeat:
1829 	if (kthread_should_stop())
1830 		return 0;
1831 
1832 	if (!llist_empty(&cprc->issue_list))
1833 		__checkpoint_and_complete_reqs(sbi);
1834 
1835 	wait_event_interruptible(*q,
1836 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1837 	goto repeat;
1838 }
1839 
1840 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1841 		struct ckpt_req *wait_req)
1842 {
1843 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1844 
1845 	if (!llist_empty(&cprc->issue_list)) {
1846 		__checkpoint_and_complete_reqs(sbi);
1847 	} else {
1848 		/* already dispatched by issue_checkpoint_thread */
1849 		if (wait_req)
1850 			wait_for_completion(&wait_req->wait);
1851 	}
1852 }
1853 
1854 static void init_ckpt_req(struct ckpt_req *req)
1855 {
1856 	memset(req, 0, sizeof(struct ckpt_req));
1857 
1858 	init_completion(&req->wait);
1859 	req->queue_time = ktime_get();
1860 }
1861 
1862 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1863 {
1864 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1865 	struct ckpt_req req;
1866 	struct cp_control cpc;
1867 
1868 	cpc.reason = __get_cp_reason(sbi);
1869 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1870 		int ret;
1871 
1872 		f2fs_down_write(&sbi->gc_lock);
1873 		ret = f2fs_write_checkpoint(sbi, &cpc);
1874 		f2fs_up_write(&sbi->gc_lock);
1875 
1876 		return ret;
1877 	}
1878 
1879 	if (!cprc->f2fs_issue_ckpt)
1880 		return __write_checkpoint_sync(sbi);
1881 
1882 	init_ckpt_req(&req);
1883 
1884 	llist_add(&req.llnode, &cprc->issue_list);
1885 	atomic_inc(&cprc->queued_ckpt);
1886 
1887 	/*
1888 	 * update issue_list before we wake up issue_checkpoint thread,
1889 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1890 	 * see more details in comments of waitqueue_active().
1891 	 */
1892 	smp_mb();
1893 
1894 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1895 		wake_up(&cprc->ckpt_wait_queue);
1896 
1897 	if (cprc->f2fs_issue_ckpt)
1898 		wait_for_completion(&req.wait);
1899 	else
1900 		flush_remained_ckpt_reqs(sbi, &req);
1901 
1902 	return req.ret;
1903 }
1904 
1905 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1906 {
1907 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1908 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1909 
1910 	if (cprc->f2fs_issue_ckpt)
1911 		return 0;
1912 
1913 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1914 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1915 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1916 		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1917 
1918 		cprc->f2fs_issue_ckpt = NULL;
1919 		return err;
1920 	}
1921 
1922 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1923 
1924 	return 0;
1925 }
1926 
1927 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1928 {
1929 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1930 	struct task_struct *ckpt_task;
1931 
1932 	if (!cprc->f2fs_issue_ckpt)
1933 		return;
1934 
1935 	ckpt_task = cprc->f2fs_issue_ckpt;
1936 	cprc->f2fs_issue_ckpt = NULL;
1937 	kthread_stop(ckpt_task);
1938 
1939 	f2fs_flush_ckpt_thread(sbi);
1940 }
1941 
1942 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1943 {
1944 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1945 
1946 	flush_remained_ckpt_reqs(sbi, NULL);
1947 
1948 	/* Let's wait for the previous dispatched checkpoint. */
1949 	while (atomic_read(&cprc->queued_ckpt))
1950 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1951 }
1952 
1953 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1954 {
1955 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1956 
1957 	atomic_set(&cprc->issued_ckpt, 0);
1958 	atomic_set(&cprc->total_ckpt, 0);
1959 	atomic_set(&cprc->queued_ckpt, 0);
1960 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1961 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1962 	init_llist_head(&cprc->issue_list);
1963 	spin_lock_init(&cprc->stat_lock);
1964 }
1965