xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision a2cab953)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/checkpoint.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/bio.h>
10 #include <linux/mpage.h>
11 #include <linux/writeback.h>
12 #include <linux/blkdev.h>
13 #include <linux/f2fs_fs.h>
14 #include <linux/pagevec.h>
15 #include <linux/swap.h>
16 #include <linux/kthread.h>
17 
18 #include "f2fs.h"
19 #include "node.h"
20 #include "segment.h"
21 #include "iostat.h"
22 #include <trace/events/f2fs.h>
23 
24 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
25 
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *f2fs_inode_entry_slab;
28 
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io,
30 						unsigned char reason)
31 {
32 	f2fs_build_fault_attr(sbi, 0, 0);
33 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
34 	if (!end_io) {
35 		f2fs_flush_merged_writes(sbi);
36 
37 		f2fs_handle_stop(sbi, reason);
38 	}
39 }
40 
41 /*
42  * We guarantee no failure on the returned page.
43  */
44 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
45 {
46 	struct address_space *mapping = META_MAPPING(sbi);
47 	struct page *page;
48 repeat:
49 	page = f2fs_grab_cache_page(mapping, index, false);
50 	if (!page) {
51 		cond_resched();
52 		goto repeat;
53 	}
54 	f2fs_wait_on_page_writeback(page, META, true, true);
55 	if (!PageUptodate(page))
56 		SetPageUptodate(page);
57 	return page;
58 }
59 
60 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
61 							bool is_meta)
62 {
63 	struct address_space *mapping = META_MAPPING(sbi);
64 	struct page *page;
65 	struct f2fs_io_info fio = {
66 		.sbi = sbi,
67 		.type = META,
68 		.op = REQ_OP_READ,
69 		.op_flags = REQ_META | REQ_PRIO,
70 		.old_blkaddr = index,
71 		.new_blkaddr = index,
72 		.encrypted_page = NULL,
73 		.is_por = !is_meta,
74 	};
75 	int err;
76 
77 	if (unlikely(!is_meta))
78 		fio.op_flags &= ~REQ_META;
79 repeat:
80 	page = f2fs_grab_cache_page(mapping, index, false);
81 	if (!page) {
82 		cond_resched();
83 		goto repeat;
84 	}
85 	if (PageUptodate(page))
86 		goto out;
87 
88 	fio.page = page;
89 
90 	err = f2fs_submit_page_bio(&fio);
91 	if (err) {
92 		f2fs_put_page(page, 1);
93 		return ERR_PTR(err);
94 	}
95 
96 	f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
97 
98 	lock_page(page);
99 	if (unlikely(page->mapping != mapping)) {
100 		f2fs_put_page(page, 1);
101 		goto repeat;
102 	}
103 
104 	if (unlikely(!PageUptodate(page))) {
105 		f2fs_handle_page_eio(sbi, page->index, META);
106 		f2fs_put_page(page, 1);
107 		return ERR_PTR(-EIO);
108 	}
109 out:
110 	return page;
111 }
112 
113 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
114 {
115 	return __get_meta_page(sbi, index, true);
116 }
117 
118 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
119 {
120 	struct page *page;
121 	int count = 0;
122 
123 retry:
124 	page = __get_meta_page(sbi, index, true);
125 	if (IS_ERR(page)) {
126 		if (PTR_ERR(page) == -EIO &&
127 				++count <= DEFAULT_RETRY_IO_COUNT)
128 			goto retry;
129 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
130 	}
131 	return page;
132 }
133 
134 /* for POR only */
135 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
136 {
137 	return __get_meta_page(sbi, index, false);
138 }
139 
140 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
141 							int type)
142 {
143 	struct seg_entry *se;
144 	unsigned int segno, offset;
145 	bool exist;
146 
147 	if (type == DATA_GENERIC)
148 		return true;
149 
150 	segno = GET_SEGNO(sbi, blkaddr);
151 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
152 	se = get_seg_entry(sbi, segno);
153 
154 	exist = f2fs_test_bit(offset, se->cur_valid_map);
155 	if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
156 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
157 			 blkaddr, exist);
158 		set_sbi_flag(sbi, SBI_NEED_FSCK);
159 		return exist;
160 	}
161 
162 	if (!exist && type == DATA_GENERIC_ENHANCE) {
163 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
164 			 blkaddr, exist);
165 		set_sbi_flag(sbi, SBI_NEED_FSCK);
166 		dump_stack();
167 	}
168 	return exist;
169 }
170 
171 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
172 					block_t blkaddr, int type)
173 {
174 	switch (type) {
175 	case META_NAT:
176 		break;
177 	case META_SIT:
178 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
179 			return false;
180 		break;
181 	case META_SSA:
182 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
183 			blkaddr < SM_I(sbi)->ssa_blkaddr))
184 			return false;
185 		break;
186 	case META_CP:
187 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
188 			blkaddr < __start_cp_addr(sbi)))
189 			return false;
190 		break;
191 	case META_POR:
192 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
193 			blkaddr < MAIN_BLKADDR(sbi)))
194 			return false;
195 		break;
196 	case DATA_GENERIC:
197 	case DATA_GENERIC_ENHANCE:
198 	case DATA_GENERIC_ENHANCE_READ:
199 	case DATA_GENERIC_ENHANCE_UPDATE:
200 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
201 				blkaddr < MAIN_BLKADDR(sbi))) {
202 			f2fs_warn(sbi, "access invalid blkaddr:%u",
203 				  blkaddr);
204 			set_sbi_flag(sbi, SBI_NEED_FSCK);
205 			dump_stack();
206 			return false;
207 		} else {
208 			return __is_bitmap_valid(sbi, blkaddr, type);
209 		}
210 		break;
211 	case META_GENERIC:
212 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
213 			blkaddr >= MAIN_BLKADDR(sbi)))
214 			return false;
215 		break;
216 	default:
217 		BUG();
218 	}
219 
220 	return true;
221 }
222 
223 /*
224  * Readahead CP/NAT/SIT/SSA/POR pages
225  */
226 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
227 							int type, bool sync)
228 {
229 	struct page *page;
230 	block_t blkno = start;
231 	struct f2fs_io_info fio = {
232 		.sbi = sbi,
233 		.type = META,
234 		.op = REQ_OP_READ,
235 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
236 		.encrypted_page = NULL,
237 		.in_list = false,
238 		.is_por = (type == META_POR),
239 	};
240 	struct blk_plug plug;
241 	int err;
242 
243 	if (unlikely(type == META_POR))
244 		fio.op_flags &= ~REQ_META;
245 
246 	blk_start_plug(&plug);
247 	for (; nrpages-- > 0; blkno++) {
248 
249 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
250 			goto out;
251 
252 		switch (type) {
253 		case META_NAT:
254 			if (unlikely(blkno >=
255 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
256 				blkno = 0;
257 			/* get nat block addr */
258 			fio.new_blkaddr = current_nat_addr(sbi,
259 					blkno * NAT_ENTRY_PER_BLOCK);
260 			break;
261 		case META_SIT:
262 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
263 				goto out;
264 			/* get sit block addr */
265 			fio.new_blkaddr = current_sit_addr(sbi,
266 					blkno * SIT_ENTRY_PER_BLOCK);
267 			break;
268 		case META_SSA:
269 		case META_CP:
270 		case META_POR:
271 			fio.new_blkaddr = blkno;
272 			break;
273 		default:
274 			BUG();
275 		}
276 
277 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
278 						fio.new_blkaddr, false);
279 		if (!page)
280 			continue;
281 		if (PageUptodate(page)) {
282 			f2fs_put_page(page, 1);
283 			continue;
284 		}
285 
286 		fio.page = page;
287 		err = f2fs_submit_page_bio(&fio);
288 		f2fs_put_page(page, err ? 1 : 0);
289 
290 		if (!err)
291 			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
292 							F2FS_BLKSIZE);
293 	}
294 out:
295 	blk_finish_plug(&plug);
296 	return blkno - start;
297 }
298 
299 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
300 							unsigned int ra_blocks)
301 {
302 	struct page *page;
303 	bool readahead = false;
304 
305 	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
306 		return;
307 
308 	page = find_get_page(META_MAPPING(sbi), index);
309 	if (!page || !PageUptodate(page))
310 		readahead = true;
311 	f2fs_put_page(page, 0);
312 
313 	if (readahead)
314 		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
315 }
316 
317 static int __f2fs_write_meta_page(struct page *page,
318 				struct writeback_control *wbc,
319 				enum iostat_type io_type)
320 {
321 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
322 
323 	trace_f2fs_writepage(page, META);
324 
325 	if (unlikely(f2fs_cp_error(sbi)))
326 		goto redirty_out;
327 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
328 		goto redirty_out;
329 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
330 		goto redirty_out;
331 
332 	f2fs_do_write_meta_page(sbi, page, io_type);
333 	dec_page_count(sbi, F2FS_DIRTY_META);
334 
335 	if (wbc->for_reclaim)
336 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
337 
338 	unlock_page(page);
339 
340 	if (unlikely(f2fs_cp_error(sbi)))
341 		f2fs_submit_merged_write(sbi, META);
342 
343 	return 0;
344 
345 redirty_out:
346 	redirty_page_for_writepage(wbc, page);
347 	return AOP_WRITEPAGE_ACTIVATE;
348 }
349 
350 static int f2fs_write_meta_page(struct page *page,
351 				struct writeback_control *wbc)
352 {
353 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
354 }
355 
356 static int f2fs_write_meta_pages(struct address_space *mapping,
357 				struct writeback_control *wbc)
358 {
359 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
360 	long diff, written;
361 
362 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
363 		goto skip_write;
364 
365 	/* collect a number of dirty meta pages and write together */
366 	if (wbc->sync_mode != WB_SYNC_ALL &&
367 			get_pages(sbi, F2FS_DIRTY_META) <
368 					nr_pages_to_skip(sbi, META))
369 		goto skip_write;
370 
371 	/* if locked failed, cp will flush dirty pages instead */
372 	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
373 		goto skip_write;
374 
375 	trace_f2fs_writepages(mapping->host, wbc, META);
376 	diff = nr_pages_to_write(sbi, META, wbc);
377 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
378 	f2fs_up_write(&sbi->cp_global_sem);
379 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
380 	return 0;
381 
382 skip_write:
383 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
384 	trace_f2fs_writepages(mapping->host, wbc, META);
385 	return 0;
386 }
387 
388 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
389 				long nr_to_write, enum iostat_type io_type)
390 {
391 	struct address_space *mapping = META_MAPPING(sbi);
392 	pgoff_t index = 0, prev = ULONG_MAX;
393 	struct pagevec pvec;
394 	long nwritten = 0;
395 	int nr_pages;
396 	struct writeback_control wbc = {
397 		.for_reclaim = 0,
398 	};
399 	struct blk_plug plug;
400 
401 	pagevec_init(&pvec);
402 
403 	blk_start_plug(&plug);
404 
405 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
406 				PAGECACHE_TAG_DIRTY))) {
407 		int i;
408 
409 		for (i = 0; i < nr_pages; i++) {
410 			struct page *page = pvec.pages[i];
411 
412 			if (prev == ULONG_MAX)
413 				prev = page->index - 1;
414 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
415 				pagevec_release(&pvec);
416 				goto stop;
417 			}
418 
419 			lock_page(page);
420 
421 			if (unlikely(page->mapping != mapping)) {
422 continue_unlock:
423 				unlock_page(page);
424 				continue;
425 			}
426 			if (!PageDirty(page)) {
427 				/* someone wrote it for us */
428 				goto continue_unlock;
429 			}
430 
431 			f2fs_wait_on_page_writeback(page, META, true, true);
432 
433 			if (!clear_page_dirty_for_io(page))
434 				goto continue_unlock;
435 
436 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
437 				unlock_page(page);
438 				break;
439 			}
440 			nwritten++;
441 			prev = page->index;
442 			if (unlikely(nwritten >= nr_to_write))
443 				break;
444 		}
445 		pagevec_release(&pvec);
446 		cond_resched();
447 	}
448 stop:
449 	if (nwritten)
450 		f2fs_submit_merged_write(sbi, type);
451 
452 	blk_finish_plug(&plug);
453 
454 	return nwritten;
455 }
456 
457 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
458 		struct folio *folio)
459 {
460 	trace_f2fs_set_page_dirty(&folio->page, META);
461 
462 	if (!folio_test_uptodate(folio))
463 		folio_mark_uptodate(folio);
464 	if (filemap_dirty_folio(mapping, folio)) {
465 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
466 		set_page_private_reference(&folio->page);
467 		return true;
468 	}
469 	return false;
470 }
471 
472 const struct address_space_operations f2fs_meta_aops = {
473 	.writepage	= f2fs_write_meta_page,
474 	.writepages	= f2fs_write_meta_pages,
475 	.dirty_folio	= f2fs_dirty_meta_folio,
476 	.invalidate_folio = f2fs_invalidate_folio,
477 	.release_folio	= f2fs_release_folio,
478 	.migrate_folio	= filemap_migrate_folio,
479 };
480 
481 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
482 						unsigned int devidx, int type)
483 {
484 	struct inode_management *im = &sbi->im[type];
485 	struct ino_entry *e = NULL, *new = NULL;
486 
487 	if (type == FLUSH_INO) {
488 		rcu_read_lock();
489 		e = radix_tree_lookup(&im->ino_root, ino);
490 		rcu_read_unlock();
491 	}
492 
493 retry:
494 	if (!e)
495 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
496 						GFP_NOFS, true, NULL);
497 
498 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
499 
500 	spin_lock(&im->ino_lock);
501 	e = radix_tree_lookup(&im->ino_root, ino);
502 	if (!e) {
503 		if (!new) {
504 			spin_unlock(&im->ino_lock);
505 			goto retry;
506 		}
507 		e = new;
508 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
509 			f2fs_bug_on(sbi, 1);
510 
511 		memset(e, 0, sizeof(struct ino_entry));
512 		e->ino = ino;
513 
514 		list_add_tail(&e->list, &im->ino_list);
515 		if (type != ORPHAN_INO)
516 			im->ino_num++;
517 	}
518 
519 	if (type == FLUSH_INO)
520 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
521 
522 	spin_unlock(&im->ino_lock);
523 	radix_tree_preload_end();
524 
525 	if (new && e != new)
526 		kmem_cache_free(ino_entry_slab, new);
527 }
528 
529 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
530 {
531 	struct inode_management *im = &sbi->im[type];
532 	struct ino_entry *e;
533 
534 	spin_lock(&im->ino_lock);
535 	e = radix_tree_lookup(&im->ino_root, ino);
536 	if (e) {
537 		list_del(&e->list);
538 		radix_tree_delete(&im->ino_root, ino);
539 		im->ino_num--;
540 		spin_unlock(&im->ino_lock);
541 		kmem_cache_free(ino_entry_slab, e);
542 		return;
543 	}
544 	spin_unlock(&im->ino_lock);
545 }
546 
547 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
548 {
549 	/* add new dirty ino entry into list */
550 	__add_ino_entry(sbi, ino, 0, type);
551 }
552 
553 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
554 {
555 	/* remove dirty ino entry from list */
556 	__remove_ino_entry(sbi, ino, type);
557 }
558 
559 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
560 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
561 {
562 	struct inode_management *im = &sbi->im[mode];
563 	struct ino_entry *e;
564 
565 	spin_lock(&im->ino_lock);
566 	e = radix_tree_lookup(&im->ino_root, ino);
567 	spin_unlock(&im->ino_lock);
568 	return e ? true : false;
569 }
570 
571 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
572 {
573 	struct ino_entry *e, *tmp;
574 	int i;
575 
576 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
577 		struct inode_management *im = &sbi->im[i];
578 
579 		spin_lock(&im->ino_lock);
580 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
581 			list_del(&e->list);
582 			radix_tree_delete(&im->ino_root, e->ino);
583 			kmem_cache_free(ino_entry_slab, e);
584 			im->ino_num--;
585 		}
586 		spin_unlock(&im->ino_lock);
587 	}
588 }
589 
590 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
591 					unsigned int devidx, int type)
592 {
593 	__add_ino_entry(sbi, ino, devidx, type);
594 }
595 
596 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
597 					unsigned int devidx, int type)
598 {
599 	struct inode_management *im = &sbi->im[type];
600 	struct ino_entry *e;
601 	bool is_dirty = false;
602 
603 	spin_lock(&im->ino_lock);
604 	e = radix_tree_lookup(&im->ino_root, ino);
605 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
606 		is_dirty = true;
607 	spin_unlock(&im->ino_lock);
608 	return is_dirty;
609 }
610 
611 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
612 {
613 	struct inode_management *im = &sbi->im[ORPHAN_INO];
614 	int err = 0;
615 
616 	spin_lock(&im->ino_lock);
617 
618 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
619 		spin_unlock(&im->ino_lock);
620 		f2fs_show_injection_info(sbi, FAULT_ORPHAN);
621 		return -ENOSPC;
622 	}
623 
624 	if (unlikely(im->ino_num >= sbi->max_orphans))
625 		err = -ENOSPC;
626 	else
627 		im->ino_num++;
628 	spin_unlock(&im->ino_lock);
629 
630 	return err;
631 }
632 
633 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
634 {
635 	struct inode_management *im = &sbi->im[ORPHAN_INO];
636 
637 	spin_lock(&im->ino_lock);
638 	f2fs_bug_on(sbi, im->ino_num == 0);
639 	im->ino_num--;
640 	spin_unlock(&im->ino_lock);
641 }
642 
643 void f2fs_add_orphan_inode(struct inode *inode)
644 {
645 	/* add new orphan ino entry into list */
646 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
647 	f2fs_update_inode_page(inode);
648 }
649 
650 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
651 {
652 	/* remove orphan entry from orphan list */
653 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
654 }
655 
656 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
657 {
658 	struct inode *inode;
659 	struct node_info ni;
660 	int err;
661 
662 	inode = f2fs_iget_retry(sbi->sb, ino);
663 	if (IS_ERR(inode)) {
664 		/*
665 		 * there should be a bug that we can't find the entry
666 		 * to orphan inode.
667 		 */
668 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
669 		return PTR_ERR(inode);
670 	}
671 
672 	err = f2fs_dquot_initialize(inode);
673 	if (err) {
674 		iput(inode);
675 		goto err_out;
676 	}
677 
678 	clear_nlink(inode);
679 
680 	/* truncate all the data during iput */
681 	iput(inode);
682 
683 	err = f2fs_get_node_info(sbi, ino, &ni, false);
684 	if (err)
685 		goto err_out;
686 
687 	/* ENOMEM was fully retried in f2fs_evict_inode. */
688 	if (ni.blk_addr != NULL_ADDR) {
689 		err = -EIO;
690 		goto err_out;
691 	}
692 	return 0;
693 
694 err_out:
695 	set_sbi_flag(sbi, SBI_NEED_FSCK);
696 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
697 		  __func__, ino);
698 	return err;
699 }
700 
701 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
702 {
703 	block_t start_blk, orphan_blocks, i, j;
704 	unsigned int s_flags = sbi->sb->s_flags;
705 	int err = 0;
706 #ifdef CONFIG_QUOTA
707 	int quota_enabled;
708 #endif
709 
710 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
711 		return 0;
712 
713 	if (bdev_read_only(sbi->sb->s_bdev)) {
714 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
715 		return 0;
716 	}
717 
718 	if (s_flags & SB_RDONLY) {
719 		f2fs_info(sbi, "orphan cleanup on readonly fs");
720 		sbi->sb->s_flags &= ~SB_RDONLY;
721 	}
722 
723 #ifdef CONFIG_QUOTA
724 	/*
725 	 * Turn on quotas which were not enabled for read-only mounts if
726 	 * filesystem has quota feature, so that they are updated correctly.
727 	 */
728 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
729 #endif
730 
731 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
732 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
733 
734 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
735 
736 	for (i = 0; i < orphan_blocks; i++) {
737 		struct page *page;
738 		struct f2fs_orphan_block *orphan_blk;
739 
740 		page = f2fs_get_meta_page(sbi, start_blk + i);
741 		if (IS_ERR(page)) {
742 			err = PTR_ERR(page);
743 			goto out;
744 		}
745 
746 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
747 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
748 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
749 
750 			err = recover_orphan_inode(sbi, ino);
751 			if (err) {
752 				f2fs_put_page(page, 1);
753 				goto out;
754 			}
755 		}
756 		f2fs_put_page(page, 1);
757 	}
758 	/* clear Orphan Flag */
759 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
760 out:
761 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
762 
763 #ifdef CONFIG_QUOTA
764 	/* Turn quotas off */
765 	if (quota_enabled)
766 		f2fs_quota_off_umount(sbi->sb);
767 #endif
768 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
769 
770 	return err;
771 }
772 
773 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
774 {
775 	struct list_head *head;
776 	struct f2fs_orphan_block *orphan_blk = NULL;
777 	unsigned int nentries = 0;
778 	unsigned short index = 1;
779 	unsigned short orphan_blocks;
780 	struct page *page = NULL;
781 	struct ino_entry *orphan = NULL;
782 	struct inode_management *im = &sbi->im[ORPHAN_INO];
783 
784 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
785 
786 	/*
787 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
788 	 * orphan inode operations are covered under f2fs_lock_op().
789 	 * And, spin_lock should be avoided due to page operations below.
790 	 */
791 	head = &im->ino_list;
792 
793 	/* loop for each orphan inode entry and write them in Jornal block */
794 	list_for_each_entry(orphan, head, list) {
795 		if (!page) {
796 			page = f2fs_grab_meta_page(sbi, start_blk++);
797 			orphan_blk =
798 				(struct f2fs_orphan_block *)page_address(page);
799 			memset(orphan_blk, 0, sizeof(*orphan_blk));
800 		}
801 
802 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
803 
804 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
805 			/*
806 			 * an orphan block is full of 1020 entries,
807 			 * then we need to flush current orphan blocks
808 			 * and bring another one in memory
809 			 */
810 			orphan_blk->blk_addr = cpu_to_le16(index);
811 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
812 			orphan_blk->entry_count = cpu_to_le32(nentries);
813 			set_page_dirty(page);
814 			f2fs_put_page(page, 1);
815 			index++;
816 			nentries = 0;
817 			page = NULL;
818 		}
819 	}
820 
821 	if (page) {
822 		orphan_blk->blk_addr = cpu_to_le16(index);
823 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
824 		orphan_blk->entry_count = cpu_to_le32(nentries);
825 		set_page_dirty(page);
826 		f2fs_put_page(page, 1);
827 	}
828 }
829 
830 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
831 						struct f2fs_checkpoint *ckpt)
832 {
833 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
834 	__u32 chksum;
835 
836 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
837 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
838 		chksum_ofs += sizeof(chksum);
839 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
840 						F2FS_BLKSIZE - chksum_ofs);
841 	}
842 	return chksum;
843 }
844 
845 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
846 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
847 		unsigned long long *version)
848 {
849 	size_t crc_offset = 0;
850 	__u32 crc;
851 
852 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
853 	if (IS_ERR(*cp_page))
854 		return PTR_ERR(*cp_page);
855 
856 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
857 
858 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
859 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
860 			crc_offset > CP_CHKSUM_OFFSET) {
861 		f2fs_put_page(*cp_page, 1);
862 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
863 		return -EINVAL;
864 	}
865 
866 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
867 	if (crc != cur_cp_crc(*cp_block)) {
868 		f2fs_put_page(*cp_page, 1);
869 		f2fs_warn(sbi, "invalid crc value");
870 		return -EINVAL;
871 	}
872 
873 	*version = cur_cp_version(*cp_block);
874 	return 0;
875 }
876 
877 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
878 				block_t cp_addr, unsigned long long *version)
879 {
880 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
881 	struct f2fs_checkpoint *cp_block = NULL;
882 	unsigned long long cur_version = 0, pre_version = 0;
883 	unsigned int cp_blocks;
884 	int err;
885 
886 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
887 					&cp_page_1, version);
888 	if (err)
889 		return NULL;
890 
891 	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
892 
893 	if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
894 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
895 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
896 		goto invalid_cp;
897 	}
898 	pre_version = *version;
899 
900 	cp_addr += cp_blocks - 1;
901 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
902 					&cp_page_2, version);
903 	if (err)
904 		goto invalid_cp;
905 	cur_version = *version;
906 
907 	if (cur_version == pre_version) {
908 		*version = cur_version;
909 		f2fs_put_page(cp_page_2, 1);
910 		return cp_page_1;
911 	}
912 	f2fs_put_page(cp_page_2, 1);
913 invalid_cp:
914 	f2fs_put_page(cp_page_1, 1);
915 	return NULL;
916 }
917 
918 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
919 {
920 	struct f2fs_checkpoint *cp_block;
921 	struct f2fs_super_block *fsb = sbi->raw_super;
922 	struct page *cp1, *cp2, *cur_page;
923 	unsigned long blk_size = sbi->blocksize;
924 	unsigned long long cp1_version = 0, cp2_version = 0;
925 	unsigned long long cp_start_blk_no;
926 	unsigned int cp_blks = 1 + __cp_payload(sbi);
927 	block_t cp_blk_no;
928 	int i;
929 	int err;
930 
931 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
932 				  GFP_KERNEL);
933 	if (!sbi->ckpt)
934 		return -ENOMEM;
935 	/*
936 	 * Finding out valid cp block involves read both
937 	 * sets( cp pack 1 and cp pack 2)
938 	 */
939 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
940 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
941 
942 	/* The second checkpoint pack should start at the next segment */
943 	cp_start_blk_no += ((unsigned long long)1) <<
944 				le32_to_cpu(fsb->log_blocks_per_seg);
945 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
946 
947 	if (cp1 && cp2) {
948 		if (ver_after(cp2_version, cp1_version))
949 			cur_page = cp2;
950 		else
951 			cur_page = cp1;
952 	} else if (cp1) {
953 		cur_page = cp1;
954 	} else if (cp2) {
955 		cur_page = cp2;
956 	} else {
957 		err = -EFSCORRUPTED;
958 		goto fail_no_cp;
959 	}
960 
961 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
962 	memcpy(sbi->ckpt, cp_block, blk_size);
963 
964 	if (cur_page == cp1)
965 		sbi->cur_cp_pack = 1;
966 	else
967 		sbi->cur_cp_pack = 2;
968 
969 	/* Sanity checking of checkpoint */
970 	if (f2fs_sanity_check_ckpt(sbi)) {
971 		err = -EFSCORRUPTED;
972 		goto free_fail_no_cp;
973 	}
974 
975 	if (cp_blks <= 1)
976 		goto done;
977 
978 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
979 	if (cur_page == cp2)
980 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
981 
982 	for (i = 1; i < cp_blks; i++) {
983 		void *sit_bitmap_ptr;
984 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
985 
986 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
987 		if (IS_ERR(cur_page)) {
988 			err = PTR_ERR(cur_page);
989 			goto free_fail_no_cp;
990 		}
991 		sit_bitmap_ptr = page_address(cur_page);
992 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
993 		f2fs_put_page(cur_page, 1);
994 	}
995 done:
996 	f2fs_put_page(cp1, 1);
997 	f2fs_put_page(cp2, 1);
998 	return 0;
999 
1000 free_fail_no_cp:
1001 	f2fs_put_page(cp1, 1);
1002 	f2fs_put_page(cp2, 1);
1003 fail_no_cp:
1004 	kvfree(sbi->ckpt);
1005 	return err;
1006 }
1007 
1008 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1009 {
1010 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1011 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1012 
1013 	if (is_inode_flag_set(inode, flag))
1014 		return;
1015 
1016 	set_inode_flag(inode, flag);
1017 	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1018 	stat_inc_dirty_inode(sbi, type);
1019 }
1020 
1021 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1022 {
1023 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1024 
1025 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1026 		return;
1027 
1028 	list_del_init(&F2FS_I(inode)->dirty_list);
1029 	clear_inode_flag(inode, flag);
1030 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1031 }
1032 
1033 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1034 {
1035 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1036 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1037 
1038 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1039 			!S_ISLNK(inode->i_mode))
1040 		return;
1041 
1042 	spin_lock(&sbi->inode_lock[type]);
1043 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1044 		__add_dirty_inode(inode, type);
1045 	inode_inc_dirty_pages(inode);
1046 	spin_unlock(&sbi->inode_lock[type]);
1047 
1048 	set_page_private_reference(&folio->page);
1049 }
1050 
1051 void f2fs_remove_dirty_inode(struct inode *inode)
1052 {
1053 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1054 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1055 
1056 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1057 			!S_ISLNK(inode->i_mode))
1058 		return;
1059 
1060 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1061 		return;
1062 
1063 	spin_lock(&sbi->inode_lock[type]);
1064 	__remove_dirty_inode(inode, type);
1065 	spin_unlock(&sbi->inode_lock[type]);
1066 }
1067 
1068 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1069 						bool from_cp)
1070 {
1071 	struct list_head *head;
1072 	struct inode *inode;
1073 	struct f2fs_inode_info *fi;
1074 	bool is_dir = (type == DIR_INODE);
1075 	unsigned long ino = 0;
1076 
1077 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1078 				get_pages(sbi, is_dir ?
1079 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1080 retry:
1081 	if (unlikely(f2fs_cp_error(sbi))) {
1082 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1083 				get_pages(sbi, is_dir ?
1084 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085 		return -EIO;
1086 	}
1087 
1088 	spin_lock(&sbi->inode_lock[type]);
1089 
1090 	head = &sbi->inode_list[type];
1091 	if (list_empty(head)) {
1092 		spin_unlock(&sbi->inode_lock[type]);
1093 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1094 				get_pages(sbi, is_dir ?
1095 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1096 		return 0;
1097 	}
1098 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1099 	inode = igrab(&fi->vfs_inode);
1100 	spin_unlock(&sbi->inode_lock[type]);
1101 	if (inode) {
1102 		unsigned long cur_ino = inode->i_ino;
1103 
1104 		if (from_cp)
1105 			F2FS_I(inode)->cp_task = current;
1106 		F2FS_I(inode)->wb_task = current;
1107 
1108 		filemap_fdatawrite(inode->i_mapping);
1109 
1110 		F2FS_I(inode)->wb_task = NULL;
1111 		if (from_cp)
1112 			F2FS_I(inode)->cp_task = NULL;
1113 
1114 		iput(inode);
1115 		/* We need to give cpu to another writers. */
1116 		if (ino == cur_ino)
1117 			cond_resched();
1118 		else
1119 			ino = cur_ino;
1120 	} else {
1121 		/*
1122 		 * We should submit bio, since it exists several
1123 		 * wribacking dentry pages in the freeing inode.
1124 		 */
1125 		f2fs_submit_merged_write(sbi, DATA);
1126 		cond_resched();
1127 	}
1128 	goto retry;
1129 }
1130 
1131 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1132 {
1133 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1134 	struct inode *inode;
1135 	struct f2fs_inode_info *fi;
1136 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1137 
1138 	while (total--) {
1139 		if (unlikely(f2fs_cp_error(sbi)))
1140 			return -EIO;
1141 
1142 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1143 		if (list_empty(head)) {
1144 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1145 			return 0;
1146 		}
1147 		fi = list_first_entry(head, struct f2fs_inode_info,
1148 							gdirty_list);
1149 		inode = igrab(&fi->vfs_inode);
1150 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1151 		if (inode) {
1152 			sync_inode_metadata(inode, 0);
1153 
1154 			/* it's on eviction */
1155 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1156 				f2fs_update_inode_page(inode);
1157 			iput(inode);
1158 		}
1159 	}
1160 	return 0;
1161 }
1162 
1163 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1164 {
1165 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1166 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1167 	nid_t last_nid = nm_i->next_scan_nid;
1168 
1169 	next_free_nid(sbi, &last_nid);
1170 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1171 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1172 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1173 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1174 }
1175 
1176 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1177 {
1178 	bool ret = false;
1179 
1180 	if (!is_journalled_quota(sbi))
1181 		return false;
1182 
1183 	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1184 		return true;
1185 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1186 		ret = false;
1187 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1188 		ret = false;
1189 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1190 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1191 		ret = true;
1192 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1193 		ret = true;
1194 	}
1195 	f2fs_up_write(&sbi->quota_sem);
1196 	return ret;
1197 }
1198 
1199 /*
1200  * Freeze all the FS-operations for checkpoint.
1201  */
1202 static int block_operations(struct f2fs_sb_info *sbi)
1203 {
1204 	struct writeback_control wbc = {
1205 		.sync_mode = WB_SYNC_ALL,
1206 		.nr_to_write = LONG_MAX,
1207 		.for_reclaim = 0,
1208 	};
1209 	int err = 0, cnt = 0;
1210 
1211 	/*
1212 	 * Let's flush inline_data in dirty node pages.
1213 	 */
1214 	f2fs_flush_inline_data(sbi);
1215 
1216 retry_flush_quotas:
1217 	f2fs_lock_all(sbi);
1218 	if (__need_flush_quota(sbi)) {
1219 		int locked;
1220 
1221 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1222 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1223 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1224 			goto retry_flush_dents;
1225 		}
1226 		f2fs_unlock_all(sbi);
1227 
1228 		/* only failed during mount/umount/freeze/quotactl */
1229 		locked = down_read_trylock(&sbi->sb->s_umount);
1230 		f2fs_quota_sync(sbi->sb, -1);
1231 		if (locked)
1232 			up_read(&sbi->sb->s_umount);
1233 		cond_resched();
1234 		goto retry_flush_quotas;
1235 	}
1236 
1237 retry_flush_dents:
1238 	/* write all the dirty dentry pages */
1239 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1240 		f2fs_unlock_all(sbi);
1241 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1242 		if (err)
1243 			return err;
1244 		cond_resched();
1245 		goto retry_flush_quotas;
1246 	}
1247 
1248 	/*
1249 	 * POR: we should ensure that there are no dirty node pages
1250 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1251 	 */
1252 	f2fs_down_write(&sbi->node_change);
1253 
1254 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1255 		f2fs_up_write(&sbi->node_change);
1256 		f2fs_unlock_all(sbi);
1257 		err = f2fs_sync_inode_meta(sbi);
1258 		if (err)
1259 			return err;
1260 		cond_resched();
1261 		goto retry_flush_quotas;
1262 	}
1263 
1264 retry_flush_nodes:
1265 	f2fs_down_write(&sbi->node_write);
1266 
1267 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1268 		f2fs_up_write(&sbi->node_write);
1269 		atomic_inc(&sbi->wb_sync_req[NODE]);
1270 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1271 		atomic_dec(&sbi->wb_sync_req[NODE]);
1272 		if (err) {
1273 			f2fs_up_write(&sbi->node_change);
1274 			f2fs_unlock_all(sbi);
1275 			return err;
1276 		}
1277 		cond_resched();
1278 		goto retry_flush_nodes;
1279 	}
1280 
1281 	/*
1282 	 * sbi->node_change is used only for AIO write_begin path which produces
1283 	 * dirty node blocks and some checkpoint values by block allocation.
1284 	 */
1285 	__prepare_cp_block(sbi);
1286 	f2fs_up_write(&sbi->node_change);
1287 	return err;
1288 }
1289 
1290 static void unblock_operations(struct f2fs_sb_info *sbi)
1291 {
1292 	f2fs_up_write(&sbi->node_write);
1293 	f2fs_unlock_all(sbi);
1294 }
1295 
1296 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1297 {
1298 	DEFINE_WAIT(wait);
1299 
1300 	for (;;) {
1301 		if (!get_pages(sbi, type))
1302 			break;
1303 
1304 		if (unlikely(f2fs_cp_error(sbi)))
1305 			break;
1306 
1307 		if (type == F2FS_DIRTY_META)
1308 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1309 							FS_CP_META_IO);
1310 		else if (type == F2FS_WB_CP_DATA)
1311 			f2fs_submit_merged_write(sbi, DATA);
1312 
1313 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1314 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1315 	}
1316 	finish_wait(&sbi->cp_wait, &wait);
1317 }
1318 
1319 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1320 {
1321 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1322 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1323 	unsigned long flags;
1324 
1325 	if (cpc->reason & CP_UMOUNT) {
1326 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1327 			NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1328 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1329 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1330 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1331 						f2fs_nat_bitmap_enabled(sbi)) {
1332 			f2fs_enable_nat_bits(sbi);
1333 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1334 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1335 		}
1336 	}
1337 
1338 	spin_lock_irqsave(&sbi->cp_lock, flags);
1339 
1340 	if (cpc->reason & CP_TRIMMED)
1341 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1342 	else
1343 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1344 
1345 	if (cpc->reason & CP_UMOUNT)
1346 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1347 	else
1348 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1349 
1350 	if (cpc->reason & CP_FASTBOOT)
1351 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1352 	else
1353 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1354 
1355 	if (orphan_num)
1356 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1357 	else
1358 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1359 
1360 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1361 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1362 
1363 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1364 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1365 	else
1366 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1367 
1368 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1369 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1370 	else
1371 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1372 
1373 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1374 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1375 	else
1376 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1377 
1378 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1379 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1380 	else
1381 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1382 
1383 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1384 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1385 
1386 	/* set this flag to activate crc|cp_ver for recovery */
1387 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1388 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1389 
1390 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1391 }
1392 
1393 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1394 	void *src, block_t blk_addr)
1395 {
1396 	struct writeback_control wbc = {
1397 		.for_reclaim = 0,
1398 	};
1399 
1400 	/*
1401 	 * pagevec_lookup_tag and lock_page again will take
1402 	 * some extra time. Therefore, f2fs_update_meta_pages and
1403 	 * f2fs_sync_meta_pages are combined in this function.
1404 	 */
1405 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1406 	int err;
1407 
1408 	f2fs_wait_on_page_writeback(page, META, true, true);
1409 
1410 	memcpy(page_address(page), src, PAGE_SIZE);
1411 
1412 	set_page_dirty(page);
1413 	if (unlikely(!clear_page_dirty_for_io(page)))
1414 		f2fs_bug_on(sbi, 1);
1415 
1416 	/* writeout cp pack 2 page */
1417 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1418 	if (unlikely(err && f2fs_cp_error(sbi))) {
1419 		f2fs_put_page(page, 1);
1420 		return;
1421 	}
1422 
1423 	f2fs_bug_on(sbi, err);
1424 	f2fs_put_page(page, 0);
1425 
1426 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1427 	f2fs_submit_merged_write(sbi, META_FLUSH);
1428 }
1429 
1430 static inline u64 get_sectors_written(struct block_device *bdev)
1431 {
1432 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1433 }
1434 
1435 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1436 {
1437 	if (f2fs_is_multi_device(sbi)) {
1438 		u64 sectors = 0;
1439 		int i;
1440 
1441 		for (i = 0; i < sbi->s_ndevs; i++)
1442 			sectors += get_sectors_written(FDEV(i).bdev);
1443 
1444 		return sectors;
1445 	}
1446 
1447 	return get_sectors_written(sbi->sb->s_bdev);
1448 }
1449 
1450 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1451 {
1452 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1453 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1454 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1455 	block_t start_blk;
1456 	unsigned int data_sum_blocks, orphan_blocks;
1457 	__u32 crc32 = 0;
1458 	int i;
1459 	int cp_payload_blks = __cp_payload(sbi);
1460 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1461 	u64 kbytes_written;
1462 	int err;
1463 
1464 	/* Flush all the NAT/SIT pages */
1465 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1466 
1467 	/* start to update checkpoint, cp ver is already updated previously */
1468 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1469 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1470 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1471 		ckpt->cur_node_segno[i] =
1472 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1473 		ckpt->cur_node_blkoff[i] =
1474 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1475 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1476 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1477 	}
1478 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1479 		ckpt->cur_data_segno[i] =
1480 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1481 		ckpt->cur_data_blkoff[i] =
1482 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1483 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1484 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1485 	}
1486 
1487 	/* 2 cp + n data seg summary + orphan inode blocks */
1488 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1489 	spin_lock_irqsave(&sbi->cp_lock, flags);
1490 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1491 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1492 	else
1493 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1494 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1495 
1496 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1497 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1498 			orphan_blocks);
1499 
1500 	if (__remain_node_summaries(cpc->reason))
1501 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1502 				cp_payload_blks + data_sum_blocks +
1503 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1504 	else
1505 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1506 				cp_payload_blks + data_sum_blocks +
1507 				orphan_blocks);
1508 
1509 	/* update ckpt flag for checkpoint */
1510 	update_ckpt_flags(sbi, cpc);
1511 
1512 	/* update SIT/NAT bitmap */
1513 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1514 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1515 
1516 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1517 	*((__le32 *)((unsigned char *)ckpt +
1518 				le32_to_cpu(ckpt->checksum_offset)))
1519 				= cpu_to_le32(crc32);
1520 
1521 	start_blk = __start_cp_next_addr(sbi);
1522 
1523 	/* write nat bits */
1524 	if ((cpc->reason & CP_UMOUNT) &&
1525 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1526 		__u64 cp_ver = cur_cp_version(ckpt);
1527 		block_t blk;
1528 
1529 		cp_ver |= ((__u64)crc32 << 32);
1530 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1531 
1532 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1533 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1534 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1535 					(i << F2FS_BLKSIZE_BITS), blk + i);
1536 	}
1537 
1538 	/* write out checkpoint buffer at block 0 */
1539 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1540 
1541 	for (i = 1; i < 1 + cp_payload_blks; i++)
1542 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1543 							start_blk++);
1544 
1545 	if (orphan_num) {
1546 		write_orphan_inodes(sbi, start_blk);
1547 		start_blk += orphan_blocks;
1548 	}
1549 
1550 	f2fs_write_data_summaries(sbi, start_blk);
1551 	start_blk += data_sum_blocks;
1552 
1553 	/* Record write statistics in the hot node summary */
1554 	kbytes_written = sbi->kbytes_written;
1555 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1556 				sbi->sectors_written_start) >> 1;
1557 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1558 
1559 	if (__remain_node_summaries(cpc->reason)) {
1560 		f2fs_write_node_summaries(sbi, start_blk);
1561 		start_blk += NR_CURSEG_NODE_TYPE;
1562 	}
1563 
1564 	/* update user_block_counts */
1565 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1566 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1567 	percpu_counter_set(&sbi->rf_node_block_count, 0);
1568 
1569 	/* Here, we have one bio having CP pack except cp pack 2 page */
1570 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1571 	/* Wait for all dirty meta pages to be submitted for IO */
1572 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1573 
1574 	/* wait for previous submitted meta pages writeback */
1575 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1576 
1577 	/* flush all device cache */
1578 	err = f2fs_flush_device_cache(sbi);
1579 	if (err)
1580 		return err;
1581 
1582 	/* barrier and flush checkpoint cp pack 2 page if it can */
1583 	commit_checkpoint(sbi, ckpt, start_blk);
1584 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1585 
1586 	/*
1587 	 * invalidate intermediate page cache borrowed from meta inode which are
1588 	 * used for migration of encrypted, verity or compressed inode's blocks.
1589 	 */
1590 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1591 		f2fs_sb_has_compression(sbi))
1592 		invalidate_mapping_pages(META_MAPPING(sbi),
1593 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1594 
1595 	f2fs_release_ino_entry(sbi, false);
1596 
1597 	f2fs_reset_fsync_node_info(sbi);
1598 
1599 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1600 	clear_sbi_flag(sbi, SBI_NEED_CP);
1601 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1602 
1603 	spin_lock(&sbi->stat_lock);
1604 	sbi->unusable_block_count = 0;
1605 	spin_unlock(&sbi->stat_lock);
1606 
1607 	__set_cp_next_pack(sbi);
1608 
1609 	/*
1610 	 * redirty superblock if metadata like node page or inode cache is
1611 	 * updated during writing checkpoint.
1612 	 */
1613 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1614 			get_pages(sbi, F2FS_DIRTY_IMETA))
1615 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1616 
1617 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1618 
1619 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1620 }
1621 
1622 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1623 {
1624 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1625 	unsigned long long ckpt_ver;
1626 	int err = 0;
1627 
1628 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1629 		return -EROFS;
1630 
1631 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1632 		if (cpc->reason != CP_PAUSE)
1633 			return 0;
1634 		f2fs_warn(sbi, "Start checkpoint disabled!");
1635 	}
1636 	if (cpc->reason != CP_RESIZE)
1637 		f2fs_down_write(&sbi->cp_global_sem);
1638 
1639 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1640 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1641 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1642 		goto out;
1643 	if (unlikely(f2fs_cp_error(sbi))) {
1644 		err = -EIO;
1645 		goto out;
1646 	}
1647 
1648 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1649 
1650 	err = block_operations(sbi);
1651 	if (err)
1652 		goto out;
1653 
1654 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1655 
1656 	f2fs_flush_merged_writes(sbi);
1657 
1658 	/* this is the case of multiple fstrims without any changes */
1659 	if (cpc->reason & CP_DISCARD) {
1660 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1661 			unblock_operations(sbi);
1662 			goto out;
1663 		}
1664 
1665 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1666 				SIT_I(sbi)->dirty_sentries == 0 &&
1667 				prefree_segments(sbi) == 0) {
1668 			f2fs_flush_sit_entries(sbi, cpc);
1669 			f2fs_clear_prefree_segments(sbi, cpc);
1670 			unblock_operations(sbi);
1671 			goto out;
1672 		}
1673 	}
1674 
1675 	/*
1676 	 * update checkpoint pack index
1677 	 * Increase the version number so that
1678 	 * SIT entries and seg summaries are written at correct place
1679 	 */
1680 	ckpt_ver = cur_cp_version(ckpt);
1681 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1682 
1683 	/* write cached NAT/SIT entries to NAT/SIT area */
1684 	err = f2fs_flush_nat_entries(sbi, cpc);
1685 	if (err) {
1686 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1687 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1688 		goto stop;
1689 	}
1690 
1691 	f2fs_flush_sit_entries(sbi, cpc);
1692 
1693 	/* save inmem log status */
1694 	f2fs_save_inmem_curseg(sbi);
1695 
1696 	err = do_checkpoint(sbi, cpc);
1697 	if (err) {
1698 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1699 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1700 		f2fs_release_discard_addrs(sbi);
1701 	} else {
1702 		f2fs_clear_prefree_segments(sbi, cpc);
1703 	}
1704 
1705 	f2fs_restore_inmem_curseg(sbi);
1706 stop:
1707 	unblock_operations(sbi);
1708 	stat_inc_cp_count(sbi->stat_info);
1709 
1710 	if (cpc->reason & CP_RECOVERY)
1711 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1712 
1713 	/* update CP_TIME to trigger checkpoint periodically */
1714 	f2fs_update_time(sbi, CP_TIME);
1715 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1716 out:
1717 	if (cpc->reason != CP_RESIZE)
1718 		f2fs_up_write(&sbi->cp_global_sem);
1719 	return err;
1720 }
1721 
1722 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1723 {
1724 	int i;
1725 
1726 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1727 		struct inode_management *im = &sbi->im[i];
1728 
1729 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1730 		spin_lock_init(&im->ino_lock);
1731 		INIT_LIST_HEAD(&im->ino_list);
1732 		im->ino_num = 0;
1733 	}
1734 
1735 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1736 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1737 				F2FS_ORPHANS_PER_BLOCK;
1738 }
1739 
1740 int __init f2fs_create_checkpoint_caches(void)
1741 {
1742 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1743 			sizeof(struct ino_entry));
1744 	if (!ino_entry_slab)
1745 		return -ENOMEM;
1746 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1747 			sizeof(struct inode_entry));
1748 	if (!f2fs_inode_entry_slab) {
1749 		kmem_cache_destroy(ino_entry_slab);
1750 		return -ENOMEM;
1751 	}
1752 	return 0;
1753 }
1754 
1755 void f2fs_destroy_checkpoint_caches(void)
1756 {
1757 	kmem_cache_destroy(ino_entry_slab);
1758 	kmem_cache_destroy(f2fs_inode_entry_slab);
1759 }
1760 
1761 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1762 {
1763 	struct cp_control cpc = { .reason = CP_SYNC, };
1764 	int err;
1765 
1766 	f2fs_down_write(&sbi->gc_lock);
1767 	err = f2fs_write_checkpoint(sbi, &cpc);
1768 	f2fs_up_write(&sbi->gc_lock);
1769 
1770 	return err;
1771 }
1772 
1773 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1774 {
1775 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1776 	struct ckpt_req *req, *next;
1777 	struct llist_node *dispatch_list;
1778 	u64 sum_diff = 0, diff, count = 0;
1779 	int ret;
1780 
1781 	dispatch_list = llist_del_all(&cprc->issue_list);
1782 	if (!dispatch_list)
1783 		return;
1784 	dispatch_list = llist_reverse_order(dispatch_list);
1785 
1786 	ret = __write_checkpoint_sync(sbi);
1787 	atomic_inc(&cprc->issued_ckpt);
1788 
1789 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1790 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1791 		req->ret = ret;
1792 		complete(&req->wait);
1793 
1794 		sum_diff += diff;
1795 		count++;
1796 	}
1797 	atomic_sub(count, &cprc->queued_ckpt);
1798 	atomic_add(count, &cprc->total_ckpt);
1799 
1800 	spin_lock(&cprc->stat_lock);
1801 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1802 	if (cprc->peak_time < cprc->cur_time)
1803 		cprc->peak_time = cprc->cur_time;
1804 	spin_unlock(&cprc->stat_lock);
1805 }
1806 
1807 static int issue_checkpoint_thread(void *data)
1808 {
1809 	struct f2fs_sb_info *sbi = data;
1810 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1811 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1812 repeat:
1813 	if (kthread_should_stop())
1814 		return 0;
1815 
1816 	if (!llist_empty(&cprc->issue_list))
1817 		__checkpoint_and_complete_reqs(sbi);
1818 
1819 	wait_event_interruptible(*q,
1820 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1821 	goto repeat;
1822 }
1823 
1824 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1825 		struct ckpt_req *wait_req)
1826 {
1827 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1828 
1829 	if (!llist_empty(&cprc->issue_list)) {
1830 		__checkpoint_and_complete_reqs(sbi);
1831 	} else {
1832 		/* already dispatched by issue_checkpoint_thread */
1833 		if (wait_req)
1834 			wait_for_completion(&wait_req->wait);
1835 	}
1836 }
1837 
1838 static void init_ckpt_req(struct ckpt_req *req)
1839 {
1840 	memset(req, 0, sizeof(struct ckpt_req));
1841 
1842 	init_completion(&req->wait);
1843 	req->queue_time = ktime_get();
1844 }
1845 
1846 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1847 {
1848 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1849 	struct ckpt_req req;
1850 	struct cp_control cpc;
1851 
1852 	cpc.reason = __get_cp_reason(sbi);
1853 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1854 		int ret;
1855 
1856 		f2fs_down_write(&sbi->gc_lock);
1857 		ret = f2fs_write_checkpoint(sbi, &cpc);
1858 		f2fs_up_write(&sbi->gc_lock);
1859 
1860 		return ret;
1861 	}
1862 
1863 	if (!cprc->f2fs_issue_ckpt)
1864 		return __write_checkpoint_sync(sbi);
1865 
1866 	init_ckpt_req(&req);
1867 
1868 	llist_add(&req.llnode, &cprc->issue_list);
1869 	atomic_inc(&cprc->queued_ckpt);
1870 
1871 	/*
1872 	 * update issue_list before we wake up issue_checkpoint thread,
1873 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1874 	 * see more details in comments of waitqueue_active().
1875 	 */
1876 	smp_mb();
1877 
1878 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1879 		wake_up(&cprc->ckpt_wait_queue);
1880 
1881 	if (cprc->f2fs_issue_ckpt)
1882 		wait_for_completion(&req.wait);
1883 	else
1884 		flush_remained_ckpt_reqs(sbi, &req);
1885 
1886 	return req.ret;
1887 }
1888 
1889 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1890 {
1891 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1892 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1893 
1894 	if (cprc->f2fs_issue_ckpt)
1895 		return 0;
1896 
1897 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1898 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1899 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1900 		cprc->f2fs_issue_ckpt = NULL;
1901 		return -ENOMEM;
1902 	}
1903 
1904 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1905 
1906 	return 0;
1907 }
1908 
1909 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1910 {
1911 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1912 	struct task_struct *ckpt_task;
1913 
1914 	if (!cprc->f2fs_issue_ckpt)
1915 		return;
1916 
1917 	ckpt_task = cprc->f2fs_issue_ckpt;
1918 	cprc->f2fs_issue_ckpt = NULL;
1919 	kthread_stop(ckpt_task);
1920 
1921 	f2fs_flush_ckpt_thread(sbi);
1922 }
1923 
1924 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1925 {
1926 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1927 
1928 	flush_remained_ckpt_reqs(sbi, NULL);
1929 
1930 	/* Let's wait for the previous dispatched checkpoint. */
1931 	while (atomic_read(&cprc->queued_ckpt))
1932 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1933 }
1934 
1935 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1936 {
1937 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1938 
1939 	atomic_set(&cprc->issued_ckpt, 0);
1940 	atomic_set(&cprc->total_ckpt, 0);
1941 	atomic_set(&cprc->queued_ckpt, 0);
1942 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1943 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1944 	init_llist_head(&cprc->issue_list);
1945 	spin_lock_init(&cprc->stat_lock);
1946 }
1947