xref: /linux/fs/f2fs/checkpoint.c (revision 64b14a184e83eb62ea0615e31a409956049d40e7)
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 {
31 	f2fs_build_fault_attr(sbi, 0, 0);
32 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
33 	if (!end_io)
34 		f2fs_flush_merged_writes(sbi);
35 }
36 
37 /*
38  * We guarantee no failure on the returned page.
39  */
40 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
41 {
42 	struct address_space *mapping = META_MAPPING(sbi);
43 	struct page *page;
44 repeat:
45 	page = f2fs_grab_cache_page(mapping, index, false);
46 	if (!page) {
47 		cond_resched();
48 		goto repeat;
49 	}
50 	f2fs_wait_on_page_writeback(page, META, true, true);
51 	if (!PageUptodate(page))
52 		SetPageUptodate(page);
53 	return page;
54 }
55 
56 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
57 							bool is_meta)
58 {
59 	struct address_space *mapping = META_MAPPING(sbi);
60 	struct page *page;
61 	struct f2fs_io_info fio = {
62 		.sbi = sbi,
63 		.type = META,
64 		.op = REQ_OP_READ,
65 		.op_flags = REQ_META | REQ_PRIO,
66 		.old_blkaddr = index,
67 		.new_blkaddr = index,
68 		.encrypted_page = NULL,
69 		.is_por = !is_meta,
70 	};
71 	int err;
72 
73 	if (unlikely(!is_meta))
74 		fio.op_flags &= ~REQ_META;
75 repeat:
76 	page = f2fs_grab_cache_page(mapping, index, false);
77 	if (!page) {
78 		cond_resched();
79 		goto repeat;
80 	}
81 	if (PageUptodate(page))
82 		goto out;
83 
84 	fio.page = page;
85 
86 	err = f2fs_submit_page_bio(&fio);
87 	if (err) {
88 		f2fs_put_page(page, 1);
89 		return ERR_PTR(err);
90 	}
91 
92 	f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
93 
94 	lock_page(page);
95 	if (unlikely(page->mapping != mapping)) {
96 		f2fs_put_page(page, 1);
97 		goto repeat;
98 	}
99 
100 	if (unlikely(!PageUptodate(page))) {
101 		f2fs_put_page(page, 1);
102 		return ERR_PTR(-EIO);
103 	}
104 out:
105 	return page;
106 }
107 
108 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
109 {
110 	return __get_meta_page(sbi, index, true);
111 }
112 
113 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
114 {
115 	struct page *page;
116 	int count = 0;
117 
118 retry:
119 	page = __get_meta_page(sbi, index, true);
120 	if (IS_ERR(page)) {
121 		if (PTR_ERR(page) == -EIO &&
122 				++count <= DEFAULT_RETRY_IO_COUNT)
123 			goto retry;
124 		f2fs_stop_checkpoint(sbi, false);
125 	}
126 	return page;
127 }
128 
129 /* for POR only */
130 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
131 {
132 	return __get_meta_page(sbi, index, false);
133 }
134 
135 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
136 							int type)
137 {
138 	struct seg_entry *se;
139 	unsigned int segno, offset;
140 	bool exist;
141 
142 	if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
143 		return true;
144 
145 	segno = GET_SEGNO(sbi, blkaddr);
146 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
147 	se = get_seg_entry(sbi, segno);
148 
149 	exist = f2fs_test_bit(offset, se->cur_valid_map);
150 	if (!exist && type == DATA_GENERIC_ENHANCE) {
151 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
152 			 blkaddr, exist);
153 		set_sbi_flag(sbi, SBI_NEED_FSCK);
154 		WARN_ON(1);
155 	}
156 	return exist;
157 }
158 
159 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
160 					block_t blkaddr, int type)
161 {
162 	switch (type) {
163 	case META_NAT:
164 		break;
165 	case META_SIT:
166 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
167 			return false;
168 		break;
169 	case META_SSA:
170 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
171 			blkaddr < SM_I(sbi)->ssa_blkaddr))
172 			return false;
173 		break;
174 	case META_CP:
175 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
176 			blkaddr < __start_cp_addr(sbi)))
177 			return false;
178 		break;
179 	case META_POR:
180 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
181 			blkaddr < MAIN_BLKADDR(sbi)))
182 			return false;
183 		break;
184 	case DATA_GENERIC:
185 	case DATA_GENERIC_ENHANCE:
186 	case DATA_GENERIC_ENHANCE_READ:
187 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
188 				blkaddr < MAIN_BLKADDR(sbi))) {
189 			f2fs_warn(sbi, "access invalid blkaddr:%u",
190 				  blkaddr);
191 			set_sbi_flag(sbi, SBI_NEED_FSCK);
192 			WARN_ON(1);
193 			return false;
194 		} else {
195 			return __is_bitmap_valid(sbi, blkaddr, type);
196 		}
197 		break;
198 	case META_GENERIC:
199 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
200 			blkaddr >= MAIN_BLKADDR(sbi)))
201 			return false;
202 		break;
203 	default:
204 		BUG();
205 	}
206 
207 	return true;
208 }
209 
210 /*
211  * Readahead CP/NAT/SIT/SSA/POR pages
212  */
213 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
214 							int type, bool sync)
215 {
216 	struct page *page;
217 	block_t blkno = start;
218 	struct f2fs_io_info fio = {
219 		.sbi = sbi,
220 		.type = META,
221 		.op = REQ_OP_READ,
222 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
223 		.encrypted_page = NULL,
224 		.in_list = false,
225 		.is_por = (type == META_POR),
226 	};
227 	struct blk_plug plug;
228 	int err;
229 
230 	if (unlikely(type == META_POR))
231 		fio.op_flags &= ~REQ_META;
232 
233 	blk_start_plug(&plug);
234 	for (; nrpages-- > 0; blkno++) {
235 
236 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
237 			goto out;
238 
239 		switch (type) {
240 		case META_NAT:
241 			if (unlikely(blkno >=
242 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
243 				blkno = 0;
244 			/* get nat block addr */
245 			fio.new_blkaddr = current_nat_addr(sbi,
246 					blkno * NAT_ENTRY_PER_BLOCK);
247 			break;
248 		case META_SIT:
249 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
250 				goto out;
251 			/* get sit block addr */
252 			fio.new_blkaddr = current_sit_addr(sbi,
253 					blkno * SIT_ENTRY_PER_BLOCK);
254 			break;
255 		case META_SSA:
256 		case META_CP:
257 		case META_POR:
258 			fio.new_blkaddr = blkno;
259 			break;
260 		default:
261 			BUG();
262 		}
263 
264 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
265 						fio.new_blkaddr, false);
266 		if (!page)
267 			continue;
268 		if (PageUptodate(page)) {
269 			f2fs_put_page(page, 1);
270 			continue;
271 		}
272 
273 		fio.page = page;
274 		err = f2fs_submit_page_bio(&fio);
275 		f2fs_put_page(page, err ? 1 : 0);
276 
277 		if (!err)
278 			f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
279 	}
280 out:
281 	blk_finish_plug(&plug);
282 	return blkno - start;
283 }
284 
285 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
286 {
287 	struct page *page;
288 	bool readahead = false;
289 
290 	page = find_get_page(META_MAPPING(sbi), index);
291 	if (!page || !PageUptodate(page))
292 		readahead = true;
293 	f2fs_put_page(page, 0);
294 
295 	if (readahead)
296 		f2fs_ra_meta_pages(sbi, index, BIO_MAX_VECS, META_POR, true);
297 }
298 
299 static int __f2fs_write_meta_page(struct page *page,
300 				struct writeback_control *wbc,
301 				enum iostat_type io_type)
302 {
303 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
304 
305 	trace_f2fs_writepage(page, META);
306 
307 	if (unlikely(f2fs_cp_error(sbi)))
308 		goto redirty_out;
309 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
310 		goto redirty_out;
311 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
312 		goto redirty_out;
313 
314 	f2fs_do_write_meta_page(sbi, page, io_type);
315 	dec_page_count(sbi, F2FS_DIRTY_META);
316 
317 	if (wbc->for_reclaim)
318 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
319 
320 	unlock_page(page);
321 
322 	if (unlikely(f2fs_cp_error(sbi)))
323 		f2fs_submit_merged_write(sbi, META);
324 
325 	return 0;
326 
327 redirty_out:
328 	redirty_page_for_writepage(wbc, page);
329 	return AOP_WRITEPAGE_ACTIVATE;
330 }
331 
332 static int f2fs_write_meta_page(struct page *page,
333 				struct writeback_control *wbc)
334 {
335 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
336 }
337 
338 static int f2fs_write_meta_pages(struct address_space *mapping,
339 				struct writeback_control *wbc)
340 {
341 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
342 	long diff, written;
343 
344 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
345 		goto skip_write;
346 
347 	/* collect a number of dirty meta pages and write together */
348 	if (wbc->sync_mode != WB_SYNC_ALL &&
349 			get_pages(sbi, F2FS_DIRTY_META) <
350 					nr_pages_to_skip(sbi, META))
351 		goto skip_write;
352 
353 	/* if locked failed, cp will flush dirty pages instead */
354 	if (!down_write_trylock(&sbi->cp_global_sem))
355 		goto skip_write;
356 
357 	trace_f2fs_writepages(mapping->host, wbc, META);
358 	diff = nr_pages_to_write(sbi, META, wbc);
359 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
360 	up_write(&sbi->cp_global_sem);
361 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
362 	return 0;
363 
364 skip_write:
365 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
366 	trace_f2fs_writepages(mapping->host, wbc, META);
367 	return 0;
368 }
369 
370 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
371 				long nr_to_write, enum iostat_type io_type)
372 {
373 	struct address_space *mapping = META_MAPPING(sbi);
374 	pgoff_t index = 0, prev = ULONG_MAX;
375 	struct pagevec pvec;
376 	long nwritten = 0;
377 	int nr_pages;
378 	struct writeback_control wbc = {
379 		.for_reclaim = 0,
380 	};
381 	struct blk_plug plug;
382 
383 	pagevec_init(&pvec);
384 
385 	blk_start_plug(&plug);
386 
387 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
388 				PAGECACHE_TAG_DIRTY))) {
389 		int i;
390 
391 		for (i = 0; i < nr_pages; i++) {
392 			struct page *page = pvec.pages[i];
393 
394 			if (prev == ULONG_MAX)
395 				prev = page->index - 1;
396 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
397 				pagevec_release(&pvec);
398 				goto stop;
399 			}
400 
401 			lock_page(page);
402 
403 			if (unlikely(page->mapping != mapping)) {
404 continue_unlock:
405 				unlock_page(page);
406 				continue;
407 			}
408 			if (!PageDirty(page)) {
409 				/* someone wrote it for us */
410 				goto continue_unlock;
411 			}
412 
413 			f2fs_wait_on_page_writeback(page, META, true, true);
414 
415 			if (!clear_page_dirty_for_io(page))
416 				goto continue_unlock;
417 
418 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
419 				unlock_page(page);
420 				break;
421 			}
422 			nwritten++;
423 			prev = page->index;
424 			if (unlikely(nwritten >= nr_to_write))
425 				break;
426 		}
427 		pagevec_release(&pvec);
428 		cond_resched();
429 	}
430 stop:
431 	if (nwritten)
432 		f2fs_submit_merged_write(sbi, type);
433 
434 	blk_finish_plug(&plug);
435 
436 	return nwritten;
437 }
438 
439 static int f2fs_set_meta_page_dirty(struct page *page)
440 {
441 	trace_f2fs_set_page_dirty(page, META);
442 
443 	if (!PageUptodate(page))
444 		SetPageUptodate(page);
445 	if (!PageDirty(page)) {
446 		__set_page_dirty_nobuffers(page);
447 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
448 		set_page_private_reference(page);
449 		return 1;
450 	}
451 	return 0;
452 }
453 
454 const struct address_space_operations f2fs_meta_aops = {
455 	.writepage	= f2fs_write_meta_page,
456 	.writepages	= f2fs_write_meta_pages,
457 	.set_page_dirty	= f2fs_set_meta_page_dirty,
458 	.invalidatepage = f2fs_invalidate_page,
459 	.releasepage	= f2fs_release_page,
460 #ifdef CONFIG_MIGRATION
461 	.migratepage    = f2fs_migrate_page,
462 #endif
463 };
464 
465 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
466 						unsigned int devidx, int type)
467 {
468 	struct inode_management *im = &sbi->im[type];
469 	struct ino_entry *e = NULL, *new = NULL;
470 
471 	if (type == FLUSH_INO) {
472 		rcu_read_lock();
473 		e = radix_tree_lookup(&im->ino_root, ino);
474 		rcu_read_unlock();
475 	}
476 
477 retry:
478 	if (!e)
479 		new = f2fs_kmem_cache_alloc(ino_entry_slab,
480 						GFP_NOFS, true, NULL);
481 
482 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
483 
484 	spin_lock(&im->ino_lock);
485 	e = radix_tree_lookup(&im->ino_root, ino);
486 	if (!e) {
487 		if (!new) {
488 			spin_unlock(&im->ino_lock);
489 			goto retry;
490 		}
491 		e = new;
492 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
493 			f2fs_bug_on(sbi, 1);
494 
495 		memset(e, 0, sizeof(struct ino_entry));
496 		e->ino = ino;
497 
498 		list_add_tail(&e->list, &im->ino_list);
499 		if (type != ORPHAN_INO)
500 			im->ino_num++;
501 	}
502 
503 	if (type == FLUSH_INO)
504 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
505 
506 	spin_unlock(&im->ino_lock);
507 	radix_tree_preload_end();
508 
509 	if (new && e != new)
510 		kmem_cache_free(ino_entry_slab, new);
511 }
512 
513 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
514 {
515 	struct inode_management *im = &sbi->im[type];
516 	struct ino_entry *e;
517 
518 	spin_lock(&im->ino_lock);
519 	e = radix_tree_lookup(&im->ino_root, ino);
520 	if (e) {
521 		list_del(&e->list);
522 		radix_tree_delete(&im->ino_root, ino);
523 		im->ino_num--;
524 		spin_unlock(&im->ino_lock);
525 		kmem_cache_free(ino_entry_slab, e);
526 		return;
527 	}
528 	spin_unlock(&im->ino_lock);
529 }
530 
531 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
532 {
533 	/* add new dirty ino entry into list */
534 	__add_ino_entry(sbi, ino, 0, type);
535 }
536 
537 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
538 {
539 	/* remove dirty ino entry from list */
540 	__remove_ino_entry(sbi, ino, type);
541 }
542 
543 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
544 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
545 {
546 	struct inode_management *im = &sbi->im[mode];
547 	struct ino_entry *e;
548 
549 	spin_lock(&im->ino_lock);
550 	e = radix_tree_lookup(&im->ino_root, ino);
551 	spin_unlock(&im->ino_lock);
552 	return e ? true : false;
553 }
554 
555 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
556 {
557 	struct ino_entry *e, *tmp;
558 	int i;
559 
560 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
561 		struct inode_management *im = &sbi->im[i];
562 
563 		spin_lock(&im->ino_lock);
564 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
565 			list_del(&e->list);
566 			radix_tree_delete(&im->ino_root, e->ino);
567 			kmem_cache_free(ino_entry_slab, e);
568 			im->ino_num--;
569 		}
570 		spin_unlock(&im->ino_lock);
571 	}
572 }
573 
574 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
575 					unsigned int devidx, int type)
576 {
577 	__add_ino_entry(sbi, ino, devidx, type);
578 }
579 
580 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
581 					unsigned int devidx, int type)
582 {
583 	struct inode_management *im = &sbi->im[type];
584 	struct ino_entry *e;
585 	bool is_dirty = false;
586 
587 	spin_lock(&im->ino_lock);
588 	e = radix_tree_lookup(&im->ino_root, ino);
589 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
590 		is_dirty = true;
591 	spin_unlock(&im->ino_lock);
592 	return is_dirty;
593 }
594 
595 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
596 {
597 	struct inode_management *im = &sbi->im[ORPHAN_INO];
598 	int err = 0;
599 
600 	spin_lock(&im->ino_lock);
601 
602 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
603 		spin_unlock(&im->ino_lock);
604 		f2fs_show_injection_info(sbi, FAULT_ORPHAN);
605 		return -ENOSPC;
606 	}
607 
608 	if (unlikely(im->ino_num >= sbi->max_orphans))
609 		err = -ENOSPC;
610 	else
611 		im->ino_num++;
612 	spin_unlock(&im->ino_lock);
613 
614 	return err;
615 }
616 
617 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
618 {
619 	struct inode_management *im = &sbi->im[ORPHAN_INO];
620 
621 	spin_lock(&im->ino_lock);
622 	f2fs_bug_on(sbi, im->ino_num == 0);
623 	im->ino_num--;
624 	spin_unlock(&im->ino_lock);
625 }
626 
627 void f2fs_add_orphan_inode(struct inode *inode)
628 {
629 	/* add new orphan ino entry into list */
630 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
631 	f2fs_update_inode_page(inode);
632 }
633 
634 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
635 {
636 	/* remove orphan entry from orphan list */
637 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
638 }
639 
640 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
641 {
642 	struct inode *inode;
643 	struct node_info ni;
644 	int err;
645 
646 	inode = f2fs_iget_retry(sbi->sb, ino);
647 	if (IS_ERR(inode)) {
648 		/*
649 		 * there should be a bug that we can't find the entry
650 		 * to orphan inode.
651 		 */
652 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
653 		return PTR_ERR(inode);
654 	}
655 
656 	err = f2fs_dquot_initialize(inode);
657 	if (err) {
658 		iput(inode);
659 		goto err_out;
660 	}
661 
662 	clear_nlink(inode);
663 
664 	/* truncate all the data during iput */
665 	iput(inode);
666 
667 	err = f2fs_get_node_info(sbi, ino, &ni, false);
668 	if (err)
669 		goto err_out;
670 
671 	/* ENOMEM was fully retried in f2fs_evict_inode. */
672 	if (ni.blk_addr != NULL_ADDR) {
673 		err = -EIO;
674 		goto err_out;
675 	}
676 	return 0;
677 
678 err_out:
679 	set_sbi_flag(sbi, SBI_NEED_FSCK);
680 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
681 		  __func__, ino);
682 	return err;
683 }
684 
685 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
686 {
687 	block_t start_blk, orphan_blocks, i, j;
688 	unsigned int s_flags = sbi->sb->s_flags;
689 	int err = 0;
690 #ifdef CONFIG_QUOTA
691 	int quota_enabled;
692 #endif
693 
694 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
695 		return 0;
696 
697 	if (bdev_read_only(sbi->sb->s_bdev)) {
698 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
699 		return 0;
700 	}
701 
702 	if (s_flags & SB_RDONLY) {
703 		f2fs_info(sbi, "orphan cleanup on readonly fs");
704 		sbi->sb->s_flags &= ~SB_RDONLY;
705 	}
706 
707 #ifdef CONFIG_QUOTA
708 	/*
709 	 * Turn on quotas which were not enabled for read-only mounts if
710 	 * filesystem has quota feature, so that they are updated correctly.
711 	 */
712 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
713 #endif
714 
715 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
716 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
717 
718 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
719 
720 	for (i = 0; i < orphan_blocks; i++) {
721 		struct page *page;
722 		struct f2fs_orphan_block *orphan_blk;
723 
724 		page = f2fs_get_meta_page(sbi, start_blk + i);
725 		if (IS_ERR(page)) {
726 			err = PTR_ERR(page);
727 			goto out;
728 		}
729 
730 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
731 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
732 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
733 
734 			err = recover_orphan_inode(sbi, ino);
735 			if (err) {
736 				f2fs_put_page(page, 1);
737 				goto out;
738 			}
739 		}
740 		f2fs_put_page(page, 1);
741 	}
742 	/* clear Orphan Flag */
743 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
744 out:
745 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
746 
747 #ifdef CONFIG_QUOTA
748 	/* Turn quotas off */
749 	if (quota_enabled)
750 		f2fs_quota_off_umount(sbi->sb);
751 #endif
752 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
753 
754 	return err;
755 }
756 
757 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
758 {
759 	struct list_head *head;
760 	struct f2fs_orphan_block *orphan_blk = NULL;
761 	unsigned int nentries = 0;
762 	unsigned short index = 1;
763 	unsigned short orphan_blocks;
764 	struct page *page = NULL;
765 	struct ino_entry *orphan = NULL;
766 	struct inode_management *im = &sbi->im[ORPHAN_INO];
767 
768 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
769 
770 	/*
771 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
772 	 * orphan inode operations are covered under f2fs_lock_op().
773 	 * And, spin_lock should be avoided due to page operations below.
774 	 */
775 	head = &im->ino_list;
776 
777 	/* loop for each orphan inode entry and write them in Jornal block */
778 	list_for_each_entry(orphan, head, list) {
779 		if (!page) {
780 			page = f2fs_grab_meta_page(sbi, start_blk++);
781 			orphan_blk =
782 				(struct f2fs_orphan_block *)page_address(page);
783 			memset(orphan_blk, 0, sizeof(*orphan_blk));
784 		}
785 
786 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
787 
788 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
789 			/*
790 			 * an orphan block is full of 1020 entries,
791 			 * then we need to flush current orphan blocks
792 			 * and bring another one in memory
793 			 */
794 			orphan_blk->blk_addr = cpu_to_le16(index);
795 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
796 			orphan_blk->entry_count = cpu_to_le32(nentries);
797 			set_page_dirty(page);
798 			f2fs_put_page(page, 1);
799 			index++;
800 			nentries = 0;
801 			page = NULL;
802 		}
803 	}
804 
805 	if (page) {
806 		orphan_blk->blk_addr = cpu_to_le16(index);
807 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
808 		orphan_blk->entry_count = cpu_to_le32(nentries);
809 		set_page_dirty(page);
810 		f2fs_put_page(page, 1);
811 	}
812 }
813 
814 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
815 						struct f2fs_checkpoint *ckpt)
816 {
817 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
818 	__u32 chksum;
819 
820 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
821 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
822 		chksum_ofs += sizeof(chksum);
823 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
824 						F2FS_BLKSIZE - chksum_ofs);
825 	}
826 	return chksum;
827 }
828 
829 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
830 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
831 		unsigned long long *version)
832 {
833 	size_t crc_offset = 0;
834 	__u32 crc;
835 
836 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
837 	if (IS_ERR(*cp_page))
838 		return PTR_ERR(*cp_page);
839 
840 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
841 
842 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
843 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
844 			crc_offset > CP_CHKSUM_OFFSET) {
845 		f2fs_put_page(*cp_page, 1);
846 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
847 		return -EINVAL;
848 	}
849 
850 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
851 	if (crc != cur_cp_crc(*cp_block)) {
852 		f2fs_put_page(*cp_page, 1);
853 		f2fs_warn(sbi, "invalid crc value");
854 		return -EINVAL;
855 	}
856 
857 	*version = cur_cp_version(*cp_block);
858 	return 0;
859 }
860 
861 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
862 				block_t cp_addr, unsigned long long *version)
863 {
864 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
865 	struct f2fs_checkpoint *cp_block = NULL;
866 	unsigned long long cur_version = 0, pre_version = 0;
867 	int err;
868 
869 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
870 					&cp_page_1, version);
871 	if (err)
872 		return NULL;
873 
874 	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
875 					sbi->blocks_per_seg) {
876 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
877 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
878 		goto invalid_cp;
879 	}
880 	pre_version = *version;
881 
882 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
883 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
884 					&cp_page_2, version);
885 	if (err)
886 		goto invalid_cp;
887 	cur_version = *version;
888 
889 	if (cur_version == pre_version) {
890 		*version = cur_version;
891 		f2fs_put_page(cp_page_2, 1);
892 		return cp_page_1;
893 	}
894 	f2fs_put_page(cp_page_2, 1);
895 invalid_cp:
896 	f2fs_put_page(cp_page_1, 1);
897 	return NULL;
898 }
899 
900 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
901 {
902 	struct f2fs_checkpoint *cp_block;
903 	struct f2fs_super_block *fsb = sbi->raw_super;
904 	struct page *cp1, *cp2, *cur_page;
905 	unsigned long blk_size = sbi->blocksize;
906 	unsigned long long cp1_version = 0, cp2_version = 0;
907 	unsigned long long cp_start_blk_no;
908 	unsigned int cp_blks = 1 + __cp_payload(sbi);
909 	block_t cp_blk_no;
910 	int i;
911 	int err;
912 
913 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
914 				  GFP_KERNEL);
915 	if (!sbi->ckpt)
916 		return -ENOMEM;
917 	/*
918 	 * Finding out valid cp block involves read both
919 	 * sets( cp pack 1 and cp pack 2)
920 	 */
921 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
922 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
923 
924 	/* The second checkpoint pack should start at the next segment */
925 	cp_start_blk_no += ((unsigned long long)1) <<
926 				le32_to_cpu(fsb->log_blocks_per_seg);
927 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
928 
929 	if (cp1 && cp2) {
930 		if (ver_after(cp2_version, cp1_version))
931 			cur_page = cp2;
932 		else
933 			cur_page = cp1;
934 	} else if (cp1) {
935 		cur_page = cp1;
936 	} else if (cp2) {
937 		cur_page = cp2;
938 	} else {
939 		err = -EFSCORRUPTED;
940 		goto fail_no_cp;
941 	}
942 
943 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
944 	memcpy(sbi->ckpt, cp_block, blk_size);
945 
946 	if (cur_page == cp1)
947 		sbi->cur_cp_pack = 1;
948 	else
949 		sbi->cur_cp_pack = 2;
950 
951 	/* Sanity checking of checkpoint */
952 	if (f2fs_sanity_check_ckpt(sbi)) {
953 		err = -EFSCORRUPTED;
954 		goto free_fail_no_cp;
955 	}
956 
957 	if (cp_blks <= 1)
958 		goto done;
959 
960 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
961 	if (cur_page == cp2)
962 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
963 
964 	for (i = 1; i < cp_blks; i++) {
965 		void *sit_bitmap_ptr;
966 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
967 
968 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
969 		if (IS_ERR(cur_page)) {
970 			err = PTR_ERR(cur_page);
971 			goto free_fail_no_cp;
972 		}
973 		sit_bitmap_ptr = page_address(cur_page);
974 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
975 		f2fs_put_page(cur_page, 1);
976 	}
977 done:
978 	f2fs_put_page(cp1, 1);
979 	f2fs_put_page(cp2, 1);
980 	return 0;
981 
982 free_fail_no_cp:
983 	f2fs_put_page(cp1, 1);
984 	f2fs_put_page(cp2, 1);
985 fail_no_cp:
986 	kvfree(sbi->ckpt);
987 	return err;
988 }
989 
990 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
991 {
992 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
993 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
994 
995 	if (is_inode_flag_set(inode, flag))
996 		return;
997 
998 	set_inode_flag(inode, flag);
999 	if (!f2fs_is_volatile_file(inode))
1000 		list_add_tail(&F2FS_I(inode)->dirty_list,
1001 						&sbi->inode_list[type]);
1002 	stat_inc_dirty_inode(sbi, type);
1003 }
1004 
1005 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1006 {
1007 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1008 
1009 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1010 		return;
1011 
1012 	list_del_init(&F2FS_I(inode)->dirty_list);
1013 	clear_inode_flag(inode, flag);
1014 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1015 }
1016 
1017 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1018 {
1019 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1020 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1021 
1022 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1023 			!S_ISLNK(inode->i_mode))
1024 		return;
1025 
1026 	spin_lock(&sbi->inode_lock[type]);
1027 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1028 		__add_dirty_inode(inode, type);
1029 	inode_inc_dirty_pages(inode);
1030 	spin_unlock(&sbi->inode_lock[type]);
1031 
1032 	set_page_private_reference(page);
1033 }
1034 
1035 void f2fs_remove_dirty_inode(struct inode *inode)
1036 {
1037 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1038 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1039 
1040 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1041 			!S_ISLNK(inode->i_mode))
1042 		return;
1043 
1044 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1045 		return;
1046 
1047 	spin_lock(&sbi->inode_lock[type]);
1048 	__remove_dirty_inode(inode, type);
1049 	spin_unlock(&sbi->inode_lock[type]);
1050 }
1051 
1052 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1053 {
1054 	struct list_head *head;
1055 	struct inode *inode;
1056 	struct f2fs_inode_info *fi;
1057 	bool is_dir = (type == DIR_INODE);
1058 	unsigned long ino = 0;
1059 
1060 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1061 				get_pages(sbi, is_dir ?
1062 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1063 retry:
1064 	if (unlikely(f2fs_cp_error(sbi))) {
1065 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1066 				get_pages(sbi, is_dir ?
1067 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1068 		return -EIO;
1069 	}
1070 
1071 	spin_lock(&sbi->inode_lock[type]);
1072 
1073 	head = &sbi->inode_list[type];
1074 	if (list_empty(head)) {
1075 		spin_unlock(&sbi->inode_lock[type]);
1076 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1077 				get_pages(sbi, is_dir ?
1078 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1079 		return 0;
1080 	}
1081 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1082 	inode = igrab(&fi->vfs_inode);
1083 	spin_unlock(&sbi->inode_lock[type]);
1084 	if (inode) {
1085 		unsigned long cur_ino = inode->i_ino;
1086 
1087 		F2FS_I(inode)->cp_task = current;
1088 
1089 		filemap_fdatawrite(inode->i_mapping);
1090 
1091 		F2FS_I(inode)->cp_task = NULL;
1092 
1093 		iput(inode);
1094 		/* We need to give cpu to another writers. */
1095 		if (ino == cur_ino)
1096 			cond_resched();
1097 		else
1098 			ino = cur_ino;
1099 	} else {
1100 		/*
1101 		 * We should submit bio, since it exists several
1102 		 * wribacking dentry pages in the freeing inode.
1103 		 */
1104 		f2fs_submit_merged_write(sbi, DATA);
1105 		cond_resched();
1106 	}
1107 	goto retry;
1108 }
1109 
1110 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1111 {
1112 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1113 	struct inode *inode;
1114 	struct f2fs_inode_info *fi;
1115 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1116 
1117 	while (total--) {
1118 		if (unlikely(f2fs_cp_error(sbi)))
1119 			return -EIO;
1120 
1121 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1122 		if (list_empty(head)) {
1123 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1124 			return 0;
1125 		}
1126 		fi = list_first_entry(head, struct f2fs_inode_info,
1127 							gdirty_list);
1128 		inode = igrab(&fi->vfs_inode);
1129 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1130 		if (inode) {
1131 			sync_inode_metadata(inode, 0);
1132 
1133 			/* it's on eviction */
1134 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1135 				f2fs_update_inode_page(inode);
1136 			iput(inode);
1137 		}
1138 	}
1139 	return 0;
1140 }
1141 
1142 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1143 {
1144 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1146 	nid_t last_nid = nm_i->next_scan_nid;
1147 
1148 	next_free_nid(sbi, &last_nid);
1149 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1150 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1151 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1152 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1153 }
1154 
1155 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1156 {
1157 	bool ret = false;
1158 
1159 	if (!is_journalled_quota(sbi))
1160 		return false;
1161 
1162 	if (!down_write_trylock(&sbi->quota_sem))
1163 		return true;
1164 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1165 		ret = false;
1166 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1167 		ret = false;
1168 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1169 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1170 		ret = true;
1171 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1172 		ret = true;
1173 	}
1174 	up_write(&sbi->quota_sem);
1175 	return ret;
1176 }
1177 
1178 /*
1179  * Freeze all the FS-operations for checkpoint.
1180  */
1181 static int block_operations(struct f2fs_sb_info *sbi)
1182 {
1183 	struct writeback_control wbc = {
1184 		.sync_mode = WB_SYNC_ALL,
1185 		.nr_to_write = LONG_MAX,
1186 		.for_reclaim = 0,
1187 	};
1188 	int err = 0, cnt = 0;
1189 
1190 	/*
1191 	 * Let's flush inline_data in dirty node pages.
1192 	 */
1193 	f2fs_flush_inline_data(sbi);
1194 
1195 retry_flush_quotas:
1196 	f2fs_lock_all(sbi);
1197 	if (__need_flush_quota(sbi)) {
1198 		int locked;
1199 
1200 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1201 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1202 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1203 			goto retry_flush_dents;
1204 		}
1205 		f2fs_unlock_all(sbi);
1206 
1207 		/* only failed during mount/umount/freeze/quotactl */
1208 		locked = down_read_trylock(&sbi->sb->s_umount);
1209 		f2fs_quota_sync(sbi->sb, -1);
1210 		if (locked)
1211 			up_read(&sbi->sb->s_umount);
1212 		cond_resched();
1213 		goto retry_flush_quotas;
1214 	}
1215 
1216 retry_flush_dents:
1217 	/* write all the dirty dentry pages */
1218 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1219 		f2fs_unlock_all(sbi);
1220 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1221 		if (err)
1222 			return err;
1223 		cond_resched();
1224 		goto retry_flush_quotas;
1225 	}
1226 
1227 	/*
1228 	 * POR: we should ensure that there are no dirty node pages
1229 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1230 	 */
1231 	down_write(&sbi->node_change);
1232 
1233 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1234 		up_write(&sbi->node_change);
1235 		f2fs_unlock_all(sbi);
1236 		err = f2fs_sync_inode_meta(sbi);
1237 		if (err)
1238 			return err;
1239 		cond_resched();
1240 		goto retry_flush_quotas;
1241 	}
1242 
1243 retry_flush_nodes:
1244 	down_write(&sbi->node_write);
1245 
1246 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1247 		up_write(&sbi->node_write);
1248 		atomic_inc(&sbi->wb_sync_req[NODE]);
1249 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1250 		atomic_dec(&sbi->wb_sync_req[NODE]);
1251 		if (err) {
1252 			up_write(&sbi->node_change);
1253 			f2fs_unlock_all(sbi);
1254 			return err;
1255 		}
1256 		cond_resched();
1257 		goto retry_flush_nodes;
1258 	}
1259 
1260 	/*
1261 	 * sbi->node_change is used only for AIO write_begin path which produces
1262 	 * dirty node blocks and some checkpoint values by block allocation.
1263 	 */
1264 	__prepare_cp_block(sbi);
1265 	up_write(&sbi->node_change);
1266 	return err;
1267 }
1268 
1269 static void unblock_operations(struct f2fs_sb_info *sbi)
1270 {
1271 	up_write(&sbi->node_write);
1272 	f2fs_unlock_all(sbi);
1273 }
1274 
1275 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1276 {
1277 	DEFINE_WAIT(wait);
1278 
1279 	for (;;) {
1280 		if (!get_pages(sbi, type))
1281 			break;
1282 
1283 		if (unlikely(f2fs_cp_error(sbi)))
1284 			break;
1285 
1286 		if (type == F2FS_DIRTY_META)
1287 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1288 							FS_CP_META_IO);
1289 		else if (type == F2FS_WB_CP_DATA)
1290 			f2fs_submit_merged_write(sbi, DATA);
1291 
1292 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1293 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1294 	}
1295 	finish_wait(&sbi->cp_wait, &wait);
1296 }
1297 
1298 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1299 {
1300 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1301 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1302 	unsigned long flags;
1303 
1304 	if (cpc->reason & CP_UMOUNT) {
1305 		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1306 			NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1307 			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1308 			f2fs_notice(sbi, "Disable nat_bits due to no space");
1309 		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1310 						f2fs_nat_bitmap_enabled(sbi)) {
1311 			f2fs_enable_nat_bits(sbi);
1312 			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1313 			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1314 		}
1315 	}
1316 
1317 	spin_lock_irqsave(&sbi->cp_lock, flags);
1318 
1319 	if (cpc->reason & CP_TRIMMED)
1320 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1321 	else
1322 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1323 
1324 	if (cpc->reason & CP_UMOUNT)
1325 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1326 	else
1327 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1328 
1329 	if (cpc->reason & CP_FASTBOOT)
1330 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1331 	else
1332 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1333 
1334 	if (orphan_num)
1335 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1336 	else
1337 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1338 
1339 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1340 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1341 
1342 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1343 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1344 	else
1345 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1346 
1347 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1348 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1349 	else
1350 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1351 
1352 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1353 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1354 	else
1355 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1356 
1357 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1358 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1359 	else
1360 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1361 
1362 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1363 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1364 
1365 	/* set this flag to activate crc|cp_ver for recovery */
1366 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1367 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1368 
1369 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1370 }
1371 
1372 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1373 	void *src, block_t blk_addr)
1374 {
1375 	struct writeback_control wbc = {
1376 		.for_reclaim = 0,
1377 	};
1378 
1379 	/*
1380 	 * pagevec_lookup_tag and lock_page again will take
1381 	 * some extra time. Therefore, f2fs_update_meta_pages and
1382 	 * f2fs_sync_meta_pages are combined in this function.
1383 	 */
1384 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1385 	int err;
1386 
1387 	f2fs_wait_on_page_writeback(page, META, true, true);
1388 
1389 	memcpy(page_address(page), src, PAGE_SIZE);
1390 
1391 	set_page_dirty(page);
1392 	if (unlikely(!clear_page_dirty_for_io(page)))
1393 		f2fs_bug_on(sbi, 1);
1394 
1395 	/* writeout cp pack 2 page */
1396 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1397 	if (unlikely(err && f2fs_cp_error(sbi))) {
1398 		f2fs_put_page(page, 1);
1399 		return;
1400 	}
1401 
1402 	f2fs_bug_on(sbi, err);
1403 	f2fs_put_page(page, 0);
1404 
1405 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1406 	f2fs_submit_merged_write(sbi, META_FLUSH);
1407 }
1408 
1409 static inline u64 get_sectors_written(struct block_device *bdev)
1410 {
1411 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1412 }
1413 
1414 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1415 {
1416 	if (f2fs_is_multi_device(sbi)) {
1417 		u64 sectors = 0;
1418 		int i;
1419 
1420 		for (i = 0; i < sbi->s_ndevs; i++)
1421 			sectors += get_sectors_written(FDEV(i).bdev);
1422 
1423 		return sectors;
1424 	}
1425 
1426 	return get_sectors_written(sbi->sb->s_bdev);
1427 }
1428 
1429 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1430 {
1431 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1432 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1433 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1434 	block_t start_blk;
1435 	unsigned int data_sum_blocks, orphan_blocks;
1436 	__u32 crc32 = 0;
1437 	int i;
1438 	int cp_payload_blks = __cp_payload(sbi);
1439 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1440 	u64 kbytes_written;
1441 	int err;
1442 
1443 	/* Flush all the NAT/SIT pages */
1444 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1445 
1446 	/* start to update checkpoint, cp ver is already updated previously */
1447 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1448 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1449 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1450 		ckpt->cur_node_segno[i] =
1451 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1452 		ckpt->cur_node_blkoff[i] =
1453 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1454 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1455 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1456 	}
1457 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1458 		ckpt->cur_data_segno[i] =
1459 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1460 		ckpt->cur_data_blkoff[i] =
1461 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1462 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1463 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1464 	}
1465 
1466 	/* 2 cp + n data seg summary + orphan inode blocks */
1467 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1468 	spin_lock_irqsave(&sbi->cp_lock, flags);
1469 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1470 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1471 	else
1472 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1473 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1474 
1475 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1476 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1477 			orphan_blocks);
1478 
1479 	if (__remain_node_summaries(cpc->reason))
1480 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1481 				cp_payload_blks + data_sum_blocks +
1482 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1483 	else
1484 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1485 				cp_payload_blks + data_sum_blocks +
1486 				orphan_blocks);
1487 
1488 	/* update ckpt flag for checkpoint */
1489 	update_ckpt_flags(sbi, cpc);
1490 
1491 	/* update SIT/NAT bitmap */
1492 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1493 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1494 
1495 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1496 	*((__le32 *)((unsigned char *)ckpt +
1497 				le32_to_cpu(ckpt->checksum_offset)))
1498 				= cpu_to_le32(crc32);
1499 
1500 	start_blk = __start_cp_next_addr(sbi);
1501 
1502 	/* write nat bits */
1503 	if ((cpc->reason & CP_UMOUNT) &&
1504 			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1505 		__u64 cp_ver = cur_cp_version(ckpt);
1506 		block_t blk;
1507 
1508 		cp_ver |= ((__u64)crc32 << 32);
1509 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1510 
1511 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1512 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1513 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1514 					(i << F2FS_BLKSIZE_BITS), blk + i);
1515 	}
1516 
1517 	/* write out checkpoint buffer at block 0 */
1518 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1519 
1520 	for (i = 1; i < 1 + cp_payload_blks; i++)
1521 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1522 							start_blk++);
1523 
1524 	if (orphan_num) {
1525 		write_orphan_inodes(sbi, start_blk);
1526 		start_blk += orphan_blocks;
1527 	}
1528 
1529 	f2fs_write_data_summaries(sbi, start_blk);
1530 	start_blk += data_sum_blocks;
1531 
1532 	/* Record write statistics in the hot node summary */
1533 	kbytes_written = sbi->kbytes_written;
1534 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1535 				sbi->sectors_written_start) >> 1;
1536 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1537 
1538 	if (__remain_node_summaries(cpc->reason)) {
1539 		f2fs_write_node_summaries(sbi, start_blk);
1540 		start_blk += NR_CURSEG_NODE_TYPE;
1541 	}
1542 
1543 	/* update user_block_counts */
1544 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1545 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1546 
1547 	/* Here, we have one bio having CP pack except cp pack 2 page */
1548 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1549 	/* Wait for all dirty meta pages to be submitted for IO */
1550 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1551 
1552 	/* wait for previous submitted meta pages writeback */
1553 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1554 
1555 	/* flush all device cache */
1556 	err = f2fs_flush_device_cache(sbi);
1557 	if (err)
1558 		return err;
1559 
1560 	/* barrier and flush checkpoint cp pack 2 page if it can */
1561 	commit_checkpoint(sbi, ckpt, start_blk);
1562 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1563 
1564 	/*
1565 	 * invalidate intermediate page cache borrowed from meta inode which are
1566 	 * used for migration of encrypted, verity or compressed inode's blocks.
1567 	 */
1568 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1569 		f2fs_sb_has_compression(sbi))
1570 		invalidate_mapping_pages(META_MAPPING(sbi),
1571 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1572 
1573 	f2fs_release_ino_entry(sbi, false);
1574 
1575 	f2fs_reset_fsync_node_info(sbi);
1576 
1577 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1578 	clear_sbi_flag(sbi, SBI_NEED_CP);
1579 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1580 
1581 	spin_lock(&sbi->stat_lock);
1582 	sbi->unusable_block_count = 0;
1583 	spin_unlock(&sbi->stat_lock);
1584 
1585 	__set_cp_next_pack(sbi);
1586 
1587 	/*
1588 	 * redirty superblock if metadata like node page or inode cache is
1589 	 * updated during writing checkpoint.
1590 	 */
1591 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1592 			get_pages(sbi, F2FS_DIRTY_IMETA))
1593 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1594 
1595 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1596 
1597 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1598 }
1599 
1600 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1601 {
1602 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1603 	unsigned long long ckpt_ver;
1604 	int err = 0;
1605 
1606 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1607 		return -EROFS;
1608 
1609 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1610 		if (cpc->reason != CP_PAUSE)
1611 			return 0;
1612 		f2fs_warn(sbi, "Start checkpoint disabled!");
1613 	}
1614 	if (cpc->reason != CP_RESIZE)
1615 		down_write(&sbi->cp_global_sem);
1616 
1617 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1618 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1619 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1620 		goto out;
1621 	if (unlikely(f2fs_cp_error(sbi))) {
1622 		err = -EIO;
1623 		goto out;
1624 	}
1625 
1626 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1627 
1628 	err = block_operations(sbi);
1629 	if (err)
1630 		goto out;
1631 
1632 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1633 
1634 	f2fs_flush_merged_writes(sbi);
1635 
1636 	/* this is the case of multiple fstrims without any changes */
1637 	if (cpc->reason & CP_DISCARD) {
1638 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1639 			unblock_operations(sbi);
1640 			goto out;
1641 		}
1642 
1643 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1644 				SIT_I(sbi)->dirty_sentries == 0 &&
1645 				prefree_segments(sbi) == 0) {
1646 			f2fs_flush_sit_entries(sbi, cpc);
1647 			f2fs_clear_prefree_segments(sbi, cpc);
1648 			unblock_operations(sbi);
1649 			goto out;
1650 		}
1651 	}
1652 
1653 	/*
1654 	 * update checkpoint pack index
1655 	 * Increase the version number so that
1656 	 * SIT entries and seg summaries are written at correct place
1657 	 */
1658 	ckpt_ver = cur_cp_version(ckpt);
1659 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1660 
1661 	/* write cached NAT/SIT entries to NAT/SIT area */
1662 	err = f2fs_flush_nat_entries(sbi, cpc);
1663 	if (err) {
1664 		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1665 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1666 		goto stop;
1667 	}
1668 
1669 	f2fs_flush_sit_entries(sbi, cpc);
1670 
1671 	/* save inmem log status */
1672 	f2fs_save_inmem_curseg(sbi);
1673 
1674 	err = do_checkpoint(sbi, cpc);
1675 	if (err) {
1676 		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1677 		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1678 		f2fs_release_discard_addrs(sbi);
1679 	} else {
1680 		f2fs_clear_prefree_segments(sbi, cpc);
1681 	}
1682 
1683 	f2fs_restore_inmem_curseg(sbi);
1684 stop:
1685 	unblock_operations(sbi);
1686 	stat_inc_cp_count(sbi->stat_info);
1687 
1688 	if (cpc->reason & CP_RECOVERY)
1689 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1690 
1691 	/* update CP_TIME to trigger checkpoint periodically */
1692 	f2fs_update_time(sbi, CP_TIME);
1693 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1694 out:
1695 	if (cpc->reason != CP_RESIZE)
1696 		up_write(&sbi->cp_global_sem);
1697 	return err;
1698 }
1699 
1700 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1701 {
1702 	int i;
1703 
1704 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1705 		struct inode_management *im = &sbi->im[i];
1706 
1707 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1708 		spin_lock_init(&im->ino_lock);
1709 		INIT_LIST_HEAD(&im->ino_list);
1710 		im->ino_num = 0;
1711 	}
1712 
1713 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1714 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1715 				F2FS_ORPHANS_PER_BLOCK;
1716 }
1717 
1718 int __init f2fs_create_checkpoint_caches(void)
1719 {
1720 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1721 			sizeof(struct ino_entry));
1722 	if (!ino_entry_slab)
1723 		return -ENOMEM;
1724 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1725 			sizeof(struct inode_entry));
1726 	if (!f2fs_inode_entry_slab) {
1727 		kmem_cache_destroy(ino_entry_slab);
1728 		return -ENOMEM;
1729 	}
1730 	return 0;
1731 }
1732 
1733 void f2fs_destroy_checkpoint_caches(void)
1734 {
1735 	kmem_cache_destroy(ino_entry_slab);
1736 	kmem_cache_destroy(f2fs_inode_entry_slab);
1737 }
1738 
1739 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1740 {
1741 	struct cp_control cpc = { .reason = CP_SYNC, };
1742 	int err;
1743 
1744 	down_write(&sbi->gc_lock);
1745 	err = f2fs_write_checkpoint(sbi, &cpc);
1746 	up_write(&sbi->gc_lock);
1747 
1748 	return err;
1749 }
1750 
1751 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1752 {
1753 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1754 	struct ckpt_req *req, *next;
1755 	struct llist_node *dispatch_list;
1756 	u64 sum_diff = 0, diff, count = 0;
1757 	int ret;
1758 
1759 	dispatch_list = llist_del_all(&cprc->issue_list);
1760 	if (!dispatch_list)
1761 		return;
1762 	dispatch_list = llist_reverse_order(dispatch_list);
1763 
1764 	ret = __write_checkpoint_sync(sbi);
1765 	atomic_inc(&cprc->issued_ckpt);
1766 
1767 	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1768 		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1769 		req->ret = ret;
1770 		complete(&req->wait);
1771 
1772 		sum_diff += diff;
1773 		count++;
1774 	}
1775 	atomic_sub(count, &cprc->queued_ckpt);
1776 	atomic_add(count, &cprc->total_ckpt);
1777 
1778 	spin_lock(&cprc->stat_lock);
1779 	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1780 	if (cprc->peak_time < cprc->cur_time)
1781 		cprc->peak_time = cprc->cur_time;
1782 	spin_unlock(&cprc->stat_lock);
1783 }
1784 
1785 static int issue_checkpoint_thread(void *data)
1786 {
1787 	struct f2fs_sb_info *sbi = data;
1788 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1789 	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1790 repeat:
1791 	if (kthread_should_stop())
1792 		return 0;
1793 
1794 	if (!llist_empty(&cprc->issue_list))
1795 		__checkpoint_and_complete_reqs(sbi);
1796 
1797 	wait_event_interruptible(*q,
1798 		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1799 	goto repeat;
1800 }
1801 
1802 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1803 		struct ckpt_req *wait_req)
1804 {
1805 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1806 
1807 	if (!llist_empty(&cprc->issue_list)) {
1808 		__checkpoint_and_complete_reqs(sbi);
1809 	} else {
1810 		/* already dispatched by issue_checkpoint_thread */
1811 		if (wait_req)
1812 			wait_for_completion(&wait_req->wait);
1813 	}
1814 }
1815 
1816 static void init_ckpt_req(struct ckpt_req *req)
1817 {
1818 	memset(req, 0, sizeof(struct ckpt_req));
1819 
1820 	init_completion(&req->wait);
1821 	req->queue_time = ktime_get();
1822 }
1823 
1824 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1825 {
1826 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1827 	struct ckpt_req req;
1828 	struct cp_control cpc;
1829 
1830 	cpc.reason = __get_cp_reason(sbi);
1831 	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1832 		int ret;
1833 
1834 		down_write(&sbi->gc_lock);
1835 		ret = f2fs_write_checkpoint(sbi, &cpc);
1836 		up_write(&sbi->gc_lock);
1837 
1838 		return ret;
1839 	}
1840 
1841 	if (!cprc->f2fs_issue_ckpt)
1842 		return __write_checkpoint_sync(sbi);
1843 
1844 	init_ckpt_req(&req);
1845 
1846 	llist_add(&req.llnode, &cprc->issue_list);
1847 	atomic_inc(&cprc->queued_ckpt);
1848 
1849 	/*
1850 	 * update issue_list before we wake up issue_checkpoint thread,
1851 	 * this smp_mb() pairs with another barrier in ___wait_event(),
1852 	 * see more details in comments of waitqueue_active().
1853 	 */
1854 	smp_mb();
1855 
1856 	if (waitqueue_active(&cprc->ckpt_wait_queue))
1857 		wake_up(&cprc->ckpt_wait_queue);
1858 
1859 	if (cprc->f2fs_issue_ckpt)
1860 		wait_for_completion(&req.wait);
1861 	else
1862 		flush_remained_ckpt_reqs(sbi, &req);
1863 
1864 	return req.ret;
1865 }
1866 
1867 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1868 {
1869 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1870 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1871 
1872 	if (cprc->f2fs_issue_ckpt)
1873 		return 0;
1874 
1875 	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1876 			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1877 	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1878 		cprc->f2fs_issue_ckpt = NULL;
1879 		return -ENOMEM;
1880 	}
1881 
1882 	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1883 
1884 	return 0;
1885 }
1886 
1887 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1888 {
1889 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1890 
1891 	if (cprc->f2fs_issue_ckpt) {
1892 		struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt;
1893 
1894 		cprc->f2fs_issue_ckpt = NULL;
1895 		kthread_stop(ckpt_task);
1896 
1897 		flush_remained_ckpt_reqs(sbi, NULL);
1898 	}
1899 }
1900 
1901 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1902 {
1903 	struct ckpt_req_control *cprc = &sbi->cprc_info;
1904 
1905 	atomic_set(&cprc->issued_ckpt, 0);
1906 	atomic_set(&cprc->total_ckpt, 0);
1907 	atomic_set(&cprc->queued_ckpt, 0);
1908 	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1909 	init_waitqueue_head(&cprc->ckpt_wait_queue);
1910 	init_llist_head(&cprc->issue_list);
1911 	spin_lock_init(&cprc->stat_lock);
1912 }
1913