xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision 66127f0d)
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);
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 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
159 			 blkaddr, exist);
160 		set_sbi_flag(sbi, SBI_NEED_FSCK);
161 		return exist;
162 	}
163 
164 	if (!exist && type == DATA_GENERIC_ENHANCE) {
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 	}
170 	return exist;
171 }
172 
173 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
174 					block_t blkaddr, int type)
175 {
176 	if (time_to_inject(sbi, FAULT_BLKADDR))
177 		return false;
178 
179 	switch (type) {
180 	case META_NAT:
181 		break;
182 	case META_SIT:
183 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
184 			return false;
185 		break;
186 	case META_SSA:
187 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
188 			blkaddr < SM_I(sbi)->ssa_blkaddr))
189 			return false;
190 		break;
191 	case META_CP:
192 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
193 			blkaddr < __start_cp_addr(sbi)))
194 			return false;
195 		break;
196 	case META_POR:
197 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
198 			blkaddr < MAIN_BLKADDR(sbi)))
199 			return false;
200 		break;
201 	case DATA_GENERIC:
202 	case DATA_GENERIC_ENHANCE:
203 	case DATA_GENERIC_ENHANCE_READ:
204 	case DATA_GENERIC_ENHANCE_UPDATE:
205 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
206 				blkaddr < MAIN_BLKADDR(sbi))) {
207 
208 			/* Skip to emit an error message. */
209 			if (unlikely(f2fs_cp_error(sbi)))
210 				return false;
211 
212 			f2fs_warn(sbi, "access invalid blkaddr:%u",
213 				  blkaddr);
214 			set_sbi_flag(sbi, SBI_NEED_FSCK);
215 			dump_stack();
216 			return false;
217 		} else {
218 			return __is_bitmap_valid(sbi, blkaddr, type);
219 		}
220 		break;
221 	case META_GENERIC:
222 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
223 			blkaddr >= MAIN_BLKADDR(sbi)))
224 			return false;
225 		break;
226 	default:
227 		BUG();
228 	}
229 
230 	return true;
231 }
232 
233 /*
234  * Readahead CP/NAT/SIT/SSA/POR pages
235  */
236 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
237 							int type, bool sync)
238 {
239 	struct page *page;
240 	block_t blkno = start;
241 	struct f2fs_io_info fio = {
242 		.sbi = sbi,
243 		.type = META,
244 		.op = REQ_OP_READ,
245 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
246 		.encrypted_page = NULL,
247 		.in_list = 0,
248 		.is_por = (type == META_POR) ? 1 : 0,
249 	};
250 	struct blk_plug plug;
251 	int err;
252 
253 	if (unlikely(type == META_POR))
254 		fio.op_flags &= ~REQ_META;
255 
256 	blk_start_plug(&plug);
257 	for (; nrpages-- > 0; blkno++) {
258 
259 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
260 			goto out;
261 
262 		switch (type) {
263 		case META_NAT:
264 			if (unlikely(blkno >=
265 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
266 				blkno = 0;
267 			/* get nat block addr */
268 			fio.new_blkaddr = current_nat_addr(sbi,
269 					blkno * NAT_ENTRY_PER_BLOCK);
270 			break;
271 		case META_SIT:
272 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
273 				goto out;
274 			/* get sit block addr */
275 			fio.new_blkaddr = current_sit_addr(sbi,
276 					blkno * SIT_ENTRY_PER_BLOCK);
277 			break;
278 		case META_SSA:
279 		case META_CP:
280 		case META_POR:
281 			fio.new_blkaddr = blkno;
282 			break;
283 		default:
284 			BUG();
285 		}
286 
287 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
288 						fio.new_blkaddr, false);
289 		if (!page)
290 			continue;
291 		if (PageUptodate(page)) {
292 			f2fs_put_page(page, 1);
293 			continue;
294 		}
295 
296 		fio.page = page;
297 		err = f2fs_submit_page_bio(&fio);
298 		f2fs_put_page(page, err ? 1 : 0);
299 
300 		if (!err)
301 			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
302 							F2FS_BLKSIZE);
303 	}
304 out:
305 	blk_finish_plug(&plug);
306 	return blkno - start;
307 }
308 
309 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
310 							unsigned int ra_blocks)
311 {
312 	struct page *page;
313 	bool readahead = false;
314 
315 	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
316 		return;
317 
318 	page = find_get_page(META_MAPPING(sbi), index);
319 	if (!page || !PageUptodate(page))
320 		readahead = true;
321 	f2fs_put_page(page, 0);
322 
323 	if (readahead)
324 		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
325 }
326 
327 static int __f2fs_write_meta_page(struct page *page,
328 				struct writeback_control *wbc,
329 				enum iostat_type io_type)
330 {
331 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
332 
333 	trace_f2fs_writepage(page, META);
334 
335 	if (unlikely(f2fs_cp_error(sbi))) {
336 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
337 			ClearPageUptodate(page);
338 			dec_page_count(sbi, F2FS_DIRTY_META);
339 			unlock_page(page);
340 			return 0;
341 		}
342 		goto redirty_out;
343 	}
344 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
345 		goto redirty_out;
346 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
347 		goto redirty_out;
348 
349 	f2fs_do_write_meta_page(sbi, page, io_type);
350 	dec_page_count(sbi, F2FS_DIRTY_META);
351 
352 	if (wbc->for_reclaim)
353 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
354 
355 	unlock_page(page);
356 
357 	if (unlikely(f2fs_cp_error(sbi)))
358 		f2fs_submit_merged_write(sbi, META);
359 
360 	return 0;
361 
362 redirty_out:
363 	redirty_page_for_writepage(wbc, page);
364 	return AOP_WRITEPAGE_ACTIVATE;
365 }
366 
367 static int f2fs_write_meta_page(struct page *page,
368 				struct writeback_control *wbc)
369 {
370 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
371 }
372 
373 static int f2fs_write_meta_pages(struct address_space *mapping,
374 				struct writeback_control *wbc)
375 {
376 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
377 	long diff, written;
378 
379 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
380 		goto skip_write;
381 
382 	/* collect a number of dirty meta pages and write together */
383 	if (wbc->sync_mode != WB_SYNC_ALL &&
384 			get_pages(sbi, F2FS_DIRTY_META) <
385 					nr_pages_to_skip(sbi, META))
386 		goto skip_write;
387 
388 	/* if locked failed, cp will flush dirty pages instead */
389 	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
390 		goto skip_write;
391 
392 	trace_f2fs_writepages(mapping->host, wbc, META);
393 	diff = nr_pages_to_write(sbi, META, wbc);
394 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
395 	f2fs_up_write(&sbi->cp_global_sem);
396 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
397 	return 0;
398 
399 skip_write:
400 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
401 	trace_f2fs_writepages(mapping->host, wbc, META);
402 	return 0;
403 }
404 
405 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
406 				long nr_to_write, enum iostat_type io_type)
407 {
408 	struct address_space *mapping = META_MAPPING(sbi);
409 	pgoff_t index = 0, prev = ULONG_MAX;
410 	struct folio_batch fbatch;
411 	long nwritten = 0;
412 	int nr_folios;
413 	struct writeback_control wbc = {
414 		.for_reclaim = 0,
415 	};
416 	struct blk_plug plug;
417 
418 	folio_batch_init(&fbatch);
419 
420 	blk_start_plug(&plug);
421 
422 	while ((nr_folios = filemap_get_folios_tag(mapping, &index,
423 					(pgoff_t)-1,
424 					PAGECACHE_TAG_DIRTY, &fbatch))) {
425 		int i;
426 
427 		for (i = 0; i < nr_folios; i++) {
428 			struct folio *folio = fbatch.folios[i];
429 
430 			if (nr_to_write != LONG_MAX && i != 0 &&
431 					folio->index != prev +
432 					folio_nr_pages(fbatch.folios[i-1])) {
433 				folio_batch_release(&fbatch);
434 				goto stop;
435 			}
436 
437 			folio_lock(folio);
438 
439 			if (unlikely(folio->mapping != mapping)) {
440 continue_unlock:
441 				folio_unlock(folio);
442 				continue;
443 			}
444 			if (!folio_test_dirty(folio)) {
445 				/* someone wrote it for us */
446 				goto continue_unlock;
447 			}
448 
449 			f2fs_wait_on_page_writeback(&folio->page, META,
450 					true, true);
451 
452 			if (!folio_clear_dirty_for_io(folio))
453 				goto continue_unlock;
454 
455 			if (__f2fs_write_meta_page(&folio->page, &wbc,
456 						io_type)) {
457 				folio_unlock(folio);
458 				break;
459 			}
460 			nwritten += folio_nr_pages(folio);
461 			prev = folio->index;
462 			if (unlikely(nwritten >= nr_to_write))
463 				break;
464 		}
465 		folio_batch_release(&fbatch);
466 		cond_resched();
467 	}
468 stop:
469 	if (nwritten)
470 		f2fs_submit_merged_write(sbi, type);
471 
472 	blk_finish_plug(&plug);
473 
474 	return nwritten;
475 }
476 
477 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
478 		struct folio *folio)
479 {
480 	trace_f2fs_set_page_dirty(&folio->page, META);
481 
482 	if (!folio_test_uptodate(folio))
483 		folio_mark_uptodate(folio);
484 	if (filemap_dirty_folio(mapping, folio)) {
485 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
486 		set_page_private_reference(&folio->page);
487 		return true;
488 	}
489 	return false;
490 }
491 
492 const struct address_space_operations f2fs_meta_aops = {
493 	.writepage	= f2fs_write_meta_page,
494 	.writepages	= f2fs_write_meta_pages,
495 	.dirty_folio	= f2fs_dirty_meta_folio,
496 	.invalidate_folio = f2fs_invalidate_folio,
497 	.release_folio	= f2fs_release_folio,
498 	.migrate_folio	= filemap_migrate_folio,
499 };
500 
501 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
502 						unsigned int devidx, int type)
503 {
504 	struct inode_management *im = &sbi->im[type];
505 	struct ino_entry *e = NULL, *new = NULL;
506 
507 	if (type == FLUSH_INO) {
508 		rcu_read_lock();
509 		e = radix_tree_lookup(&im->ino_root, ino);
510 		rcu_read_unlock();
511 	}
512 
513 retry:
514 	if (!e)
515 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
516 						GFP_NOFS, true, NULL);
517 
518 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
519 
520 	spin_lock(&im->ino_lock);
521 	e = radix_tree_lookup(&im->ino_root, ino);
522 	if (!e) {
523 		if (!new) {
524 			spin_unlock(&im->ino_lock);
525 			radix_tree_preload_end();
526 			goto retry;
527 		}
528 		e = new;
529 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
530 			f2fs_bug_on(sbi, 1);
531 
532 		memset(e, 0, sizeof(struct ino_entry));
533 		e->ino = ino;
534 
535 		list_add_tail(&e->list, &im->ino_list);
536 		if (type != ORPHAN_INO)
537 			im->ino_num++;
538 	}
539 
540 	if (type == FLUSH_INO)
541 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
542 
543 	spin_unlock(&im->ino_lock);
544 	radix_tree_preload_end();
545 
546 	if (new && e != new)
547 		kmem_cache_free(ino_entry_slab, new);
548 }
549 
550 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
551 {
552 	struct inode_management *im = &sbi->im[type];
553 	struct ino_entry *e;
554 
555 	spin_lock(&im->ino_lock);
556 	e = radix_tree_lookup(&im->ino_root, ino);
557 	if (e) {
558 		list_del(&e->list);
559 		radix_tree_delete(&im->ino_root, ino);
560 		im->ino_num--;
561 		spin_unlock(&im->ino_lock);
562 		kmem_cache_free(ino_entry_slab, e);
563 		return;
564 	}
565 	spin_unlock(&im->ino_lock);
566 }
567 
568 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
569 {
570 	/* add new dirty ino entry into list */
571 	__add_ino_entry(sbi, ino, 0, type);
572 }
573 
574 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
575 {
576 	/* remove dirty ino entry from list */
577 	__remove_ino_entry(sbi, ino, type);
578 }
579 
580 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
581 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
582 {
583 	struct inode_management *im = &sbi->im[mode];
584 	struct ino_entry *e;
585 
586 	spin_lock(&im->ino_lock);
587 	e = radix_tree_lookup(&im->ino_root, ino);
588 	spin_unlock(&im->ino_lock);
589 	return e ? true : false;
590 }
591 
592 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
593 {
594 	struct ino_entry *e, *tmp;
595 	int i;
596 
597 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
598 		struct inode_management *im = &sbi->im[i];
599 
600 		spin_lock(&im->ino_lock);
601 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
602 			list_del(&e->list);
603 			radix_tree_delete(&im->ino_root, e->ino);
604 			kmem_cache_free(ino_entry_slab, e);
605 			im->ino_num--;
606 		}
607 		spin_unlock(&im->ino_lock);
608 	}
609 }
610 
611 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
612 					unsigned int devidx, int type)
613 {
614 	__add_ino_entry(sbi, ino, devidx, type);
615 }
616 
617 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
618 					unsigned int devidx, int type)
619 {
620 	struct inode_management *im = &sbi->im[type];
621 	struct ino_entry *e;
622 	bool is_dirty = false;
623 
624 	spin_lock(&im->ino_lock);
625 	e = radix_tree_lookup(&im->ino_root, ino);
626 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
627 		is_dirty = true;
628 	spin_unlock(&im->ino_lock);
629 	return is_dirty;
630 }
631 
632 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
633 {
634 	struct inode_management *im = &sbi->im[ORPHAN_INO];
635 	int err = 0;
636 
637 	spin_lock(&im->ino_lock);
638 
639 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
640 		spin_unlock(&im->ino_lock);
641 		return -ENOSPC;
642 	}
643 
644 	if (unlikely(im->ino_num >= sbi->max_orphans))
645 		err = -ENOSPC;
646 	else
647 		im->ino_num++;
648 	spin_unlock(&im->ino_lock);
649 
650 	return err;
651 }
652 
653 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
654 {
655 	struct inode_management *im = &sbi->im[ORPHAN_INO];
656 
657 	spin_lock(&im->ino_lock);
658 	f2fs_bug_on(sbi, im->ino_num == 0);
659 	im->ino_num--;
660 	spin_unlock(&im->ino_lock);
661 }
662 
663 void f2fs_add_orphan_inode(struct inode *inode)
664 {
665 	/* add new orphan ino entry into list */
666 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
667 	f2fs_update_inode_page(inode);
668 }
669 
670 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
671 {
672 	/* remove orphan entry from orphan list */
673 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
674 }
675 
676 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
677 {
678 	struct inode *inode;
679 	struct node_info ni;
680 	int err;
681 
682 	inode = f2fs_iget_retry(sbi->sb, ino);
683 	if (IS_ERR(inode)) {
684 		/*
685 		 * there should be a bug that we can't find the entry
686 		 * to orphan inode.
687 		 */
688 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
689 		return PTR_ERR(inode);
690 	}
691 
692 	err = f2fs_dquot_initialize(inode);
693 	if (err) {
694 		iput(inode);
695 		goto err_out;
696 	}
697 
698 	clear_nlink(inode);
699 
700 	/* truncate all the data during iput */
701 	iput(inode);
702 
703 	err = f2fs_get_node_info(sbi, ino, &ni, false);
704 	if (err)
705 		goto err_out;
706 
707 	/* ENOMEM was fully retried in f2fs_evict_inode. */
708 	if (ni.blk_addr != NULL_ADDR) {
709 		err = -EIO;
710 		goto err_out;
711 	}
712 	return 0;
713 
714 err_out:
715 	set_sbi_flag(sbi, SBI_NEED_FSCK);
716 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
717 		  __func__, ino);
718 	return err;
719 }
720 
721 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
722 {
723 	block_t start_blk, orphan_blocks, i, j;
724 	int err = 0;
725 
726 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
727 		return 0;
728 
729 	if (f2fs_hw_is_readonly(sbi)) {
730 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
731 		return 0;
732 	}
733 
734 	if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
735 		f2fs_info(sbi, "orphan cleanup on readonly fs");
736 
737 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
738 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
739 
740 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
741 
742 	for (i = 0; i < orphan_blocks; i++) {
743 		struct page *page;
744 		struct f2fs_orphan_block *orphan_blk;
745 
746 		page = f2fs_get_meta_page(sbi, start_blk + i);
747 		if (IS_ERR(page)) {
748 			err = PTR_ERR(page);
749 			goto out;
750 		}
751 
752 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
753 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
754 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
755 
756 			err = recover_orphan_inode(sbi, ino);
757 			if (err) {
758 				f2fs_put_page(page, 1);
759 				goto out;
760 			}
761 		}
762 		f2fs_put_page(page, 1);
763 	}
764 	/* clear Orphan Flag */
765 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
766 out:
767 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
768 
769 	return err;
770 }
771 
772 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
773 {
774 	struct list_head *head;
775 	struct f2fs_orphan_block *orphan_blk = NULL;
776 	unsigned int nentries = 0;
777 	unsigned short index = 1;
778 	unsigned short orphan_blocks;
779 	struct page *page = NULL;
780 	struct ino_entry *orphan = NULL;
781 	struct inode_management *im = &sbi->im[ORPHAN_INO];
782 
783 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
784 
785 	/*
786 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
787 	 * orphan inode operations are covered under f2fs_lock_op().
788 	 * And, spin_lock should be avoided due to page operations below.
789 	 */
790 	head = &im->ino_list;
791 
792 	/* loop for each orphan inode entry and write them in journal block */
793 	list_for_each_entry(orphan, head, list) {
794 		if (!page) {
795 			page = f2fs_grab_meta_page(sbi, start_blk++);
796 			orphan_blk =
797 				(struct f2fs_orphan_block *)page_address(page);
798 			memset(orphan_blk, 0, sizeof(*orphan_blk));
799 		}
800 
801 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
802 
803 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
804 			/*
805 			 * an orphan block is full of 1020 entries,
806 			 * then we need to flush current orphan blocks
807 			 * and bring another one in memory
808 			 */
809 			orphan_blk->blk_addr = cpu_to_le16(index);
810 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
811 			orphan_blk->entry_count = cpu_to_le32(nentries);
812 			set_page_dirty(page);
813 			f2fs_put_page(page, 1);
814 			index++;
815 			nentries = 0;
816 			page = NULL;
817 		}
818 	}
819 
820 	if (page) {
821 		orphan_blk->blk_addr = cpu_to_le16(index);
822 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
823 		orphan_blk->entry_count = cpu_to_le32(nentries);
824 		set_page_dirty(page);
825 		f2fs_put_page(page, 1);
826 	}
827 }
828 
829 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
830 						struct f2fs_checkpoint *ckpt)
831 {
832 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
833 	__u32 chksum;
834 
835 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
836 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
837 		chksum_ofs += sizeof(chksum);
838 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
839 						F2FS_BLKSIZE - chksum_ofs);
840 	}
841 	return chksum;
842 }
843 
844 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
845 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
846 		unsigned long long *version)
847 {
848 	size_t crc_offset = 0;
849 	__u32 crc;
850 
851 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
852 	if (IS_ERR(*cp_page))
853 		return PTR_ERR(*cp_page);
854 
855 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
856 
857 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
858 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
859 			crc_offset > CP_CHKSUM_OFFSET) {
860 		f2fs_put_page(*cp_page, 1);
861 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
862 		return -EINVAL;
863 	}
864 
865 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
866 	if (crc != cur_cp_crc(*cp_block)) {
867 		f2fs_put_page(*cp_page, 1);
868 		f2fs_warn(sbi, "invalid crc value");
869 		return -EINVAL;
870 	}
871 
872 	*version = cur_cp_version(*cp_block);
873 	return 0;
874 }
875 
876 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
877 				block_t cp_addr, unsigned long long *version)
878 {
879 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
880 	struct f2fs_checkpoint *cp_block = NULL;
881 	unsigned long long cur_version = 0, pre_version = 0;
882 	unsigned int cp_blocks;
883 	int err;
884 
885 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
886 					&cp_page_1, version);
887 	if (err)
888 		return NULL;
889 
890 	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
891 
892 	if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
893 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
894 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
895 		goto invalid_cp;
896 	}
897 	pre_version = *version;
898 
899 	cp_addr += cp_blocks - 1;
900 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
901 					&cp_page_2, version);
902 	if (err)
903 		goto invalid_cp;
904 	cur_version = *version;
905 
906 	if (cur_version == pre_version) {
907 		*version = cur_version;
908 		f2fs_put_page(cp_page_2, 1);
909 		return cp_page_1;
910 	}
911 	f2fs_put_page(cp_page_2, 1);
912 invalid_cp:
913 	f2fs_put_page(cp_page_1, 1);
914 	return NULL;
915 }
916 
917 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
918 {
919 	struct f2fs_checkpoint *cp_block;
920 	struct f2fs_super_block *fsb = sbi->raw_super;
921 	struct page *cp1, *cp2, *cur_page;
922 	unsigned long blk_size = sbi->blocksize;
923 	unsigned long long cp1_version = 0, cp2_version = 0;
924 	unsigned long long cp_start_blk_no;
925 	unsigned int cp_blks = 1 + __cp_payload(sbi);
926 	block_t cp_blk_no;
927 	int i;
928 	int err;
929 
930 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
931 				  GFP_KERNEL);
932 	if (!sbi->ckpt)
933 		return -ENOMEM;
934 	/*
935 	 * Finding out valid cp block involves read both
936 	 * sets( cp pack 1 and cp pack 2)
937 	 */
938 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
939 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
940 
941 	/* The second checkpoint pack should start at the next segment */
942 	cp_start_blk_no += ((unsigned long long)1) <<
943 				le32_to_cpu(fsb->log_blocks_per_seg);
944 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
945 
946 	if (cp1 && cp2) {
947 		if (ver_after(cp2_version, cp1_version))
948 			cur_page = cp2;
949 		else
950 			cur_page = cp1;
951 	} else if (cp1) {
952 		cur_page = cp1;
953 	} else if (cp2) {
954 		cur_page = cp2;
955 	} else {
956 		err = -EFSCORRUPTED;
957 		goto fail_no_cp;
958 	}
959 
960 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
961 	memcpy(sbi->ckpt, cp_block, blk_size);
962 
963 	if (cur_page == cp1)
964 		sbi->cur_cp_pack = 1;
965 	else
966 		sbi->cur_cp_pack = 2;
967 
968 	/* Sanity checking of checkpoint */
969 	if (f2fs_sanity_check_ckpt(sbi)) {
970 		err = -EFSCORRUPTED;
971 		goto free_fail_no_cp;
972 	}
973 
974 	if (cp_blks <= 1)
975 		goto done;
976 
977 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
978 	if (cur_page == cp2)
979 		cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
980 
981 	for (i = 1; i < cp_blks; i++) {
982 		void *sit_bitmap_ptr;
983 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
984 
985 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
986 		if (IS_ERR(cur_page)) {
987 			err = PTR_ERR(cur_page);
988 			goto free_fail_no_cp;
989 		}
990 		sit_bitmap_ptr = page_address(cur_page);
991 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
992 		f2fs_put_page(cur_page, 1);
993 	}
994 done:
995 	f2fs_put_page(cp1, 1);
996 	f2fs_put_page(cp2, 1);
997 	return 0;
998 
999 free_fail_no_cp:
1000 	f2fs_put_page(cp1, 1);
1001 	f2fs_put_page(cp2, 1);
1002 fail_no_cp:
1003 	kvfree(sbi->ckpt);
1004 	return err;
1005 }
1006 
1007 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1008 {
1009 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1010 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1011 
1012 	if (is_inode_flag_set(inode, flag))
1013 		return;
1014 
1015 	set_inode_flag(inode, flag);
1016 	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1017 	stat_inc_dirty_inode(sbi, type);
1018 }
1019 
1020 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1021 {
1022 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1023 
1024 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1025 		return;
1026 
1027 	list_del_init(&F2FS_I(inode)->dirty_list);
1028 	clear_inode_flag(inode, flag);
1029 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1030 }
1031 
1032 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1033 {
1034 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1035 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1036 
1037 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1038 			!S_ISLNK(inode->i_mode))
1039 		return;
1040 
1041 	spin_lock(&sbi->inode_lock[type]);
1042 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1043 		__add_dirty_inode(inode, type);
1044 	inode_inc_dirty_pages(inode);
1045 	spin_unlock(&sbi->inode_lock[type]);
1046 
1047 	set_page_private_reference(&folio->page);
1048 }
1049 
1050 void f2fs_remove_dirty_inode(struct inode *inode)
1051 {
1052 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1053 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1054 
1055 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1056 			!S_ISLNK(inode->i_mode))
1057 		return;
1058 
1059 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1060 		return;
1061 
1062 	spin_lock(&sbi->inode_lock[type]);
1063 	__remove_dirty_inode(inode, type);
1064 	spin_unlock(&sbi->inode_lock[type]);
1065 }
1066 
1067 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1068 						bool from_cp)
1069 {
1070 	struct list_head *head;
1071 	struct inode *inode;
1072 	struct f2fs_inode_info *fi;
1073 	bool is_dir = (type == DIR_INODE);
1074 	unsigned long ino = 0;
1075 
1076 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1077 				get_pages(sbi, is_dir ?
1078 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1079 retry:
1080 	if (unlikely(f2fs_cp_error(sbi))) {
1081 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1082 				get_pages(sbi, is_dir ?
1083 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1084 		return -EIO;
1085 	}
1086 
1087 	spin_lock(&sbi->inode_lock[type]);
1088 
1089 	head = &sbi->inode_list[type];
1090 	if (list_empty(head)) {
1091 		spin_unlock(&sbi->inode_lock[type]);
1092 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1093 				get_pages(sbi, is_dir ?
1094 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1095 		return 0;
1096 	}
1097 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1098 	inode = igrab(&fi->vfs_inode);
1099 	spin_unlock(&sbi->inode_lock[type]);
1100 	if (inode) {
1101 		unsigned long cur_ino = inode->i_ino;
1102 
1103 		if (from_cp)
1104 			F2FS_I(inode)->cp_task = current;
1105 		F2FS_I(inode)->wb_task = current;
1106 
1107 		filemap_fdatawrite(inode->i_mapping);
1108 
1109 		F2FS_I(inode)->wb_task = NULL;
1110 		if (from_cp)
1111 			F2FS_I(inode)->cp_task = NULL;
1112 
1113 		iput(inode);
1114 		/* We need to give cpu to another writers. */
1115 		if (ino == cur_ino)
1116 			cond_resched();
1117 		else
1118 			ino = cur_ino;
1119 	} else {
1120 		/*
1121 		 * We should submit bio, since it exists several
1122 		 * writebacking dentry pages in the freeing inode.
1123 		 */
1124 		f2fs_submit_merged_write(sbi, DATA);
1125 		cond_resched();
1126 	}
1127 	goto retry;
1128 }
1129 
1130 static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1131 {
1132 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1133 	struct inode *inode;
1134 	struct f2fs_inode_info *fi;
1135 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1136 
1137 	while (total--) {
1138 		if (unlikely(f2fs_cp_error(sbi)))
1139 			return -EIO;
1140 
1141 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1142 		if (list_empty(head)) {
1143 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1144 			return 0;
1145 		}
1146 		fi = list_first_entry(head, struct f2fs_inode_info,
1147 							gdirty_list);
1148 		inode = igrab(&fi->vfs_inode);
1149 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1150 		if (inode) {
1151 			sync_inode_metadata(inode, 0);
1152 
1153 			/* it's on eviction */
1154 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1155 				f2fs_update_inode_page(inode);
1156 			iput(inode);
1157 		}
1158 	}
1159 	return 0;
1160 }
1161 
1162 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1163 {
1164 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1165 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1166 	nid_t last_nid = nm_i->next_scan_nid;
1167 
1168 	next_free_nid(sbi, &last_nid);
1169 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1170 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1171 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1172 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1173 
1174 	/* update user_block_counts */
1175 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1176 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1177 	percpu_counter_set(&sbi->rf_node_block_count, 0);
1178 }
1179 
1180 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1181 {
1182 	bool ret = false;
1183 
1184 	if (!is_journalled_quota(sbi))
1185 		return false;
1186 
1187 	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1188 		return true;
1189 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1190 		ret = false;
1191 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1192 		ret = false;
1193 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1194 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1195 		ret = true;
1196 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1197 		ret = true;
1198 	}
1199 	f2fs_up_write(&sbi->quota_sem);
1200 	return ret;
1201 }
1202 
1203 /*
1204  * Freeze all the FS-operations for checkpoint.
1205  */
1206 static int block_operations(struct f2fs_sb_info *sbi)
1207 {
1208 	struct writeback_control wbc = {
1209 		.sync_mode = WB_SYNC_ALL,
1210 		.nr_to_write = LONG_MAX,
1211 		.for_reclaim = 0,
1212 	};
1213 	int err = 0, cnt = 0;
1214 
1215 	/*
1216 	 * Let's flush inline_data in dirty node pages.
1217 	 */
1218 	f2fs_flush_inline_data(sbi);
1219 
1220 retry_flush_quotas:
1221 	f2fs_lock_all(sbi);
1222 	if (__need_flush_quota(sbi)) {
1223 		int locked;
1224 
1225 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1226 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1227 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1228 			goto retry_flush_dents;
1229 		}
1230 		f2fs_unlock_all(sbi);
1231 
1232 		/* only failed during mount/umount/freeze/quotactl */
1233 		locked = down_read_trylock(&sbi->sb->s_umount);
1234 		f2fs_quota_sync(sbi->sb, -1);
1235 		if (locked)
1236 			up_read(&sbi->sb->s_umount);
1237 		cond_resched();
1238 		goto retry_flush_quotas;
1239 	}
1240 
1241 retry_flush_dents:
1242 	/* write all the dirty dentry pages */
1243 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1244 		f2fs_unlock_all(sbi);
1245 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1246 		if (err)
1247 			return err;
1248 		cond_resched();
1249 		goto retry_flush_quotas;
1250 	}
1251 
1252 	/*
1253 	 * POR: we should ensure that there are no dirty node pages
1254 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1255 	 */
1256 	f2fs_down_write(&sbi->node_change);
1257 
1258 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1259 		f2fs_up_write(&sbi->node_change);
1260 		f2fs_unlock_all(sbi);
1261 		err = f2fs_sync_inode_meta(sbi);
1262 		if (err)
1263 			return err;
1264 		cond_resched();
1265 		goto retry_flush_quotas;
1266 	}
1267 
1268 retry_flush_nodes:
1269 	f2fs_down_write(&sbi->node_write);
1270 
1271 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1272 		f2fs_up_write(&sbi->node_write);
1273 		atomic_inc(&sbi->wb_sync_req[NODE]);
1274 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1275 		atomic_dec(&sbi->wb_sync_req[NODE]);
1276 		if (err) {
1277 			f2fs_up_write(&sbi->node_change);
1278 			f2fs_unlock_all(sbi);
1279 			return err;
1280 		}
1281 		cond_resched();
1282 		goto retry_flush_nodes;
1283 	}
1284 
1285 	/*
1286 	 * sbi->node_change is used only for AIO write_begin path which produces
1287 	 * dirty node blocks and some checkpoint values by block allocation.
1288 	 */
1289 	__prepare_cp_block(sbi);
1290 	f2fs_up_write(&sbi->node_change);
1291 	return err;
1292 }
1293 
1294 static void unblock_operations(struct f2fs_sb_info *sbi)
1295 {
1296 	f2fs_up_write(&sbi->node_write);
1297 	f2fs_unlock_all(sbi);
1298 }
1299 
1300 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1301 {
1302 	DEFINE_WAIT(wait);
1303 
1304 	for (;;) {
1305 		if (!get_pages(sbi, type))
1306 			break;
1307 
1308 		if (unlikely(f2fs_cp_error(sbi) &&
1309 			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1310 			break;
1311 
1312 		if (type == F2FS_DIRTY_META)
1313 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1314 							FS_CP_META_IO);
1315 		else if (type == F2FS_WB_CP_DATA)
1316 			f2fs_submit_merged_write(sbi, DATA);
1317 
1318 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1319 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1320 	}
1321 	finish_wait(&sbi->cp_wait, &wait);
1322 }
1323 
1324 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1325 {
1326 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1327 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1328 	unsigned long flags;
1329 
1330 	if (cpc->reason & CP_UMOUNT) {
1331 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1332 			NM_I(sbi)->nat_bits_blocks > BLKS_PER_SEG(sbi)) {
1333 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1334 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1335 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1336 						f2fs_nat_bitmap_enabled(sbi)) {
1337 			f2fs_enable_nat_bits(sbi);
1338 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1339 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1340 		}
1341 	}
1342 
1343 	spin_lock_irqsave(&sbi->cp_lock, flags);
1344 
1345 	if (cpc->reason & CP_TRIMMED)
1346 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1347 	else
1348 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1349 
1350 	if (cpc->reason & CP_UMOUNT)
1351 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1352 	else
1353 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1354 
1355 	if (cpc->reason & CP_FASTBOOT)
1356 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1357 	else
1358 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1359 
1360 	if (orphan_num)
1361 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1362 	else
1363 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1364 
1365 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1366 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1367 
1368 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1369 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1370 	else
1371 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1372 
1373 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1374 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1375 	else
1376 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1377 
1378 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1379 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1380 	else
1381 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1382 
1383 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1384 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1385 	else
1386 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1387 
1388 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1389 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1390 
1391 	/* set this flag to activate crc|cp_ver for recovery */
1392 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1393 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1394 
1395 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1396 }
1397 
1398 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1399 	void *src, block_t blk_addr)
1400 {
1401 	struct writeback_control wbc = {
1402 		.for_reclaim = 0,
1403 	};
1404 
1405 	/*
1406 	 * filemap_get_folios_tag and lock_page again will take
1407 	 * some extra time. Therefore, f2fs_update_meta_pages and
1408 	 * f2fs_sync_meta_pages are combined in this function.
1409 	 */
1410 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1411 	int err;
1412 
1413 	f2fs_wait_on_page_writeback(page, META, true, true);
1414 
1415 	memcpy(page_address(page), src, PAGE_SIZE);
1416 
1417 	set_page_dirty(page);
1418 	if (unlikely(!clear_page_dirty_for_io(page)))
1419 		f2fs_bug_on(sbi, 1);
1420 
1421 	/* writeout cp pack 2 page */
1422 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1423 	if (unlikely(err && f2fs_cp_error(sbi))) {
1424 		f2fs_put_page(page, 1);
1425 		return;
1426 	}
1427 
1428 	f2fs_bug_on(sbi, err);
1429 	f2fs_put_page(page, 0);
1430 
1431 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1432 	f2fs_submit_merged_write(sbi, META_FLUSH);
1433 }
1434 
1435 static inline u64 get_sectors_written(struct block_device *bdev)
1436 {
1437 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1438 }
1439 
1440 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1441 {
1442 	if (f2fs_is_multi_device(sbi)) {
1443 		u64 sectors = 0;
1444 		int i;
1445 
1446 		for (i = 0; i < sbi->s_ndevs; i++)
1447 			sectors += get_sectors_written(FDEV(i).bdev);
1448 
1449 		return sectors;
1450 	}
1451 
1452 	return get_sectors_written(sbi->sb->s_bdev);
1453 }
1454 
1455 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1456 {
1457 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1458 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1459 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1460 	block_t start_blk;
1461 	unsigned int data_sum_blocks, orphan_blocks;
1462 	__u32 crc32 = 0;
1463 	int i;
1464 	int cp_payload_blks = __cp_payload(sbi);
1465 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1466 	u64 kbytes_written;
1467 	int err;
1468 
1469 	/* Flush all the NAT/SIT pages */
1470 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1471 
1472 	/* start to update checkpoint, cp ver is already updated previously */
1473 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1474 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1475 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1476 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1477 
1478 		ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1479 		ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1480 		ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1481 	}
1482 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1483 		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1484 
1485 		ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1486 		ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1487 		ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1488 	}
1489 
1490 	/* 2 cp + n data seg summary + orphan inode blocks */
1491 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1492 	spin_lock_irqsave(&sbi->cp_lock, flags);
1493 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1494 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1495 	else
1496 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1497 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1498 
1499 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1500 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1501 			orphan_blocks);
1502 
1503 	if (__remain_node_summaries(cpc->reason))
1504 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1505 				cp_payload_blks + data_sum_blocks +
1506 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1507 	else
1508 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1509 				cp_payload_blks + data_sum_blocks +
1510 				orphan_blocks);
1511 
1512 	/* update ckpt flag for checkpoint */
1513 	update_ckpt_flags(sbi, cpc);
1514 
1515 	/* update SIT/NAT bitmap */
1516 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1517 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1518 
1519 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1520 	*((__le32 *)((unsigned char *)ckpt +
1521 				le32_to_cpu(ckpt->checksum_offset)))
1522 				= cpu_to_le32(crc32);
1523 
1524 	start_blk = __start_cp_next_addr(sbi);
1525 
1526 	/* write nat bits */
1527 	if ((cpc->reason & CP_UMOUNT) &&
1528 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1529 		__u64 cp_ver = cur_cp_version(ckpt);
1530 		block_t blk;
1531 
1532 		cp_ver |= ((__u64)crc32 << 32);
1533 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1534 
1535 		blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1536 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1537 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1538 					(i << F2FS_BLKSIZE_BITS), blk + i);
1539 	}
1540 
1541 	/* write out checkpoint buffer at block 0 */
1542 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1543 
1544 	for (i = 1; i < 1 + cp_payload_blks; i++)
1545 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1546 							start_blk++);
1547 
1548 	if (orphan_num) {
1549 		write_orphan_inodes(sbi, start_blk);
1550 		start_blk += orphan_blocks;
1551 	}
1552 
1553 	f2fs_write_data_summaries(sbi, start_blk);
1554 	start_blk += data_sum_blocks;
1555 
1556 	/* Record write statistics in the hot node summary */
1557 	kbytes_written = sbi->kbytes_written;
1558 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1559 				sbi->sectors_written_start) >> 1;
1560 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1561 
1562 	if (__remain_node_summaries(cpc->reason)) {
1563 		f2fs_write_node_summaries(sbi, start_blk);
1564 		start_blk += NR_CURSEG_NODE_TYPE;
1565 	}
1566 
1567 	/* Here, we have one bio having CP pack except cp pack 2 page */
1568 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1569 	/* Wait for all dirty meta pages to be submitted for IO */
1570 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1571 
1572 	/* wait for previous submitted meta pages writeback */
1573 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1574 
1575 	/* flush all device cache */
1576 	err = f2fs_flush_device_cache(sbi);
1577 	if (err)
1578 		return err;
1579 
1580 	/* barrier and flush checkpoint cp pack 2 page if it can */
1581 	commit_checkpoint(sbi, ckpt, start_blk);
1582 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1583 
1584 	/*
1585 	 * invalidate intermediate page cache borrowed from meta inode which are
1586 	 * used for migration of encrypted, verity or compressed inode's blocks.
1587 	 */
1588 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1589 		f2fs_sb_has_compression(sbi))
1590 		f2fs_bug_on(sbi,
1591 			invalidate_inode_pages2_range(META_MAPPING(sbi),
1592 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1593 
1594 	f2fs_release_ino_entry(sbi, false);
1595 
1596 	f2fs_reset_fsync_node_info(sbi);
1597 
1598 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1599 	clear_sbi_flag(sbi, SBI_NEED_CP);
1600 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1601 
1602 	spin_lock(&sbi->stat_lock);
1603 	sbi->unusable_block_count = 0;
1604 	spin_unlock(&sbi->stat_lock);
1605 
1606 	__set_cp_next_pack(sbi);
1607 
1608 	/*
1609 	 * redirty superblock if metadata like node page or inode cache is
1610 	 * updated during writing checkpoint.
1611 	 */
1612 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1613 			get_pages(sbi, F2FS_DIRTY_IMETA))
1614 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1615 
1616 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1617 
1618 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1619 }
1620 
1621 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1622 {
1623 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1624 	unsigned long long ckpt_ver;
1625 	int err = 0;
1626 
1627 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1628 		return -EROFS;
1629 
1630 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1631 		if (cpc->reason != CP_PAUSE)
1632 			return 0;
1633 		f2fs_warn(sbi, "Start checkpoint disabled!");
1634 	}
1635 	if (cpc->reason != CP_RESIZE)
1636 		f2fs_down_write(&sbi->cp_global_sem);
1637 
1638 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1639 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1640 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1641 		goto out;
1642 	if (unlikely(f2fs_cp_error(sbi))) {
1643 		err = -EIO;
1644 		goto out;
1645 	}
1646 
1647 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1648 
1649 	err = block_operations(sbi);
1650 	if (err)
1651 		goto out;
1652 
1653 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1654 
1655 	f2fs_flush_merged_writes(sbi);
1656 
1657 	/* this is the case of multiple fstrims without any changes */
1658 	if (cpc->reason & CP_DISCARD) {
1659 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1660 			unblock_operations(sbi);
1661 			goto out;
1662 		}
1663 
1664 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1665 				SIT_I(sbi)->dirty_sentries == 0 &&
1666 				prefree_segments(sbi) == 0) {
1667 			f2fs_flush_sit_entries(sbi, cpc);
1668 			f2fs_clear_prefree_segments(sbi, cpc);
1669 			unblock_operations(sbi);
1670 			goto out;
1671 		}
1672 	}
1673 
1674 	/*
1675 	 * update checkpoint pack index
1676 	 * Increase the version number so that
1677 	 * SIT entries and seg summaries are written at correct place
1678 	 */
1679 	ckpt_ver = cur_cp_version(ckpt);
1680 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1681 
1682 	/* write cached NAT/SIT entries to NAT/SIT area */
1683 	err = f2fs_flush_nat_entries(sbi, cpc);
1684 	if (err) {
1685 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1686 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1687 		goto stop;
1688 	}
1689 
1690 	f2fs_flush_sit_entries(sbi, cpc);
1691 
1692 	/* save inmem log status */
1693 	f2fs_save_inmem_curseg(sbi);
1694 
1695 	err = do_checkpoint(sbi, cpc);
1696 	if (err) {
1697 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1698 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1699 		f2fs_release_discard_addrs(sbi);
1700 	} else {
1701 		f2fs_clear_prefree_segments(sbi, cpc);
1702 	}
1703 
1704 	f2fs_restore_inmem_curseg(sbi);
1705 	stat_inc_cp_count(sbi);
1706 stop:
1707 	unblock_operations(sbi);
1708 
1709 	if (cpc->reason & CP_RECOVERY)
1710 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1711 
1712 	/* update CP_TIME to trigger checkpoint periodically */
1713 	f2fs_update_time(sbi, CP_TIME);
1714 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1715 out:
1716 	if (cpc->reason != CP_RESIZE)
1717 		f2fs_up_write(&sbi->cp_global_sem);
1718 	return err;
1719 }
1720 
1721 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1722 {
1723 	int i;
1724 
1725 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1726 		struct inode_management *im = &sbi->im[i];
1727 
1728 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1729 		spin_lock_init(&im->ino_lock);
1730 		INIT_LIST_HEAD(&im->ino_list);
1731 		im->ino_num = 0;
1732 	}
1733 
1734 	sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1735 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1736 			F2FS_ORPHANS_PER_BLOCK;
1737 }
1738 
1739 int __init f2fs_create_checkpoint_caches(void)
1740 {
1741 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1742 			sizeof(struct ino_entry));
1743 	if (!ino_entry_slab)
1744 		return -ENOMEM;
1745 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1746 			sizeof(struct inode_entry));
1747 	if (!f2fs_inode_entry_slab) {
1748 		kmem_cache_destroy(ino_entry_slab);
1749 		return -ENOMEM;
1750 	}
1751 	return 0;
1752 }
1753 
1754 void f2fs_destroy_checkpoint_caches(void)
1755 {
1756 	kmem_cache_destroy(ino_entry_slab);
1757 	kmem_cache_destroy(f2fs_inode_entry_slab);
1758 }
1759 
1760 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1761 {
1762 	struct cp_control cpc = { .reason = CP_SYNC, };
1763 	int err;
1764 
1765 	f2fs_down_write(&sbi->gc_lock);
1766 	err = f2fs_write_checkpoint(sbi, &cpc);
1767 	f2fs_up_write(&sbi->gc_lock);
1768 
1769 	return err;
1770 }
1771 
1772 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1773 {
1774 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1775 	struct ckpt_req *req, *next;
1776 	struct llist_node *dispatch_list;
1777 	u64 sum_diff = 0, diff, count = 0;
1778 	int ret;
1779 
1780 	dispatch_list = llist_del_all(&cprc->issue_list);
1781 	if (!dispatch_list)
1782 		return;
1783 	dispatch_list = llist_reverse_order(dispatch_list);
1784 
1785 	ret = __write_checkpoint_sync(sbi);
1786 	atomic_inc(&cprc->issued_ckpt);
1787 
1788 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1789 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1790 		req->ret = ret;
1791 		complete(&req->wait);
1792 
1793 		sum_diff += diff;
1794 		count++;
1795 	}
1796 	atomic_sub(count, &cprc->queued_ckpt);
1797 	atomic_add(count, &cprc->total_ckpt);
1798 
1799 	spin_lock(&cprc->stat_lock);
1800 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1801 	if (cprc->peak_time < cprc->cur_time)
1802 		cprc->peak_time = cprc->cur_time;
1803 	spin_unlock(&cprc->stat_lock);
1804 }
1805 
1806 static int issue_checkpoint_thread(void *data)
1807 {
1808 	struct f2fs_sb_info *sbi = data;
1809 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1810 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1811 repeat:
1812 	if (kthread_should_stop())
1813 		return 0;
1814 
1815 	if (!llist_empty(&cprc->issue_list))
1816 		__checkpoint_and_complete_reqs(sbi);
1817 
1818 	wait_event_interruptible(*q,
1819 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1820 	goto repeat;
1821 }
1822 
1823 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1824 		struct ckpt_req *wait_req)
1825 {
1826 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1827 
1828 	if (!llist_empty(&cprc->issue_list)) {
1829 		__checkpoint_and_complete_reqs(sbi);
1830 	} else {
1831 		/* already dispatched by issue_checkpoint_thread */
1832 		if (wait_req)
1833 			wait_for_completion(&wait_req->wait);
1834 	}
1835 }
1836 
1837 static void init_ckpt_req(struct ckpt_req *req)
1838 {
1839 	memset(req, 0, sizeof(struct ckpt_req));
1840 
1841 	init_completion(&req->wait);
1842 	req->queue_time = ktime_get();
1843 }
1844 
1845 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1846 {
1847 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1848 	struct ckpt_req req;
1849 	struct cp_control cpc;
1850 
1851 	cpc.reason = __get_cp_reason(sbi);
1852 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1853 		int ret;
1854 
1855 		f2fs_down_write(&sbi->gc_lock);
1856 		ret = f2fs_write_checkpoint(sbi, &cpc);
1857 		f2fs_up_write(&sbi->gc_lock);
1858 
1859 		return ret;
1860 	}
1861 
1862 	if (!cprc->f2fs_issue_ckpt)
1863 		return __write_checkpoint_sync(sbi);
1864 
1865 	init_ckpt_req(&req);
1866 
1867 	llist_add(&req.llnode, &cprc->issue_list);
1868 	atomic_inc(&cprc->queued_ckpt);
1869 
1870 	/*
1871 	 * update issue_list before we wake up issue_checkpoint thread,
1872 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1873 	 * see more details in comments of waitqueue_active().
1874 	 */
1875 	smp_mb();
1876 
1877 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1878 		wake_up(&cprc->ckpt_wait_queue);
1879 
1880 	if (cprc->f2fs_issue_ckpt)
1881 		wait_for_completion(&req.wait);
1882 	else
1883 		flush_remained_ckpt_reqs(sbi, &req);
1884 
1885 	return req.ret;
1886 }
1887 
1888 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1889 {
1890 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1891 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1892 
1893 	if (cprc->f2fs_issue_ckpt)
1894 		return 0;
1895 
1896 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1897 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1898 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1899 		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1900 
1901 		cprc->f2fs_issue_ckpt = NULL;
1902 		return err;
1903 	}
1904 
1905 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1906 
1907 	return 0;
1908 }
1909 
1910 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1911 {
1912 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1913 	struct task_struct *ckpt_task;
1914 
1915 	if (!cprc->f2fs_issue_ckpt)
1916 		return;
1917 
1918 	ckpt_task = cprc->f2fs_issue_ckpt;
1919 	cprc->f2fs_issue_ckpt = NULL;
1920 	kthread_stop(ckpt_task);
1921 
1922 	f2fs_flush_ckpt_thread(sbi);
1923 }
1924 
1925 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1926 {
1927 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1928 
1929 	flush_remained_ckpt_reqs(sbi, NULL);
1930 
1931 	/* Let's wait for the previous dispatched checkpoint. */
1932 	while (atomic_read(&cprc->queued_ckpt))
1933 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1934 }
1935 
1936 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1937 {
1938 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1939 
1940 	atomic_set(&cprc->issued_ckpt, 0);
1941 	atomic_set(&cprc->total_ckpt, 0);
1942 	atomic_set(&cprc->queued_ckpt, 0);
1943 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1944 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1945 	init_llist_head(&cprc->issue_list);
1946 	spin_lock_init(&cprc->stat_lock);
1947 }
1948