xref: /linux/fs/f2fs/checkpoint.c (revision 0bc1bd092af3c7c0b025ece93c3a86916f89f3ca)
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 
17 #include "f2fs.h"
18 #include "node.h"
19 #include "segment.h"
20 #include "trace.h"
21 #include <trace/events/f2fs.h>
22 
23 static struct kmem_cache *ino_entry_slab;
24 struct kmem_cache *f2fs_inode_entry_slab;
25 
26 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
27 {
28 	f2fs_build_fault_attr(sbi, 0, 0);
29 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
30 	if (!end_io)
31 		f2fs_flush_merged_writes(sbi);
32 }
33 
34 /*
35  * We guarantee no failure on the returned page.
36  */
37 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
38 {
39 	struct address_space *mapping = META_MAPPING(sbi);
40 	struct page *page;
41 repeat:
42 	page = f2fs_grab_cache_page(mapping, index, false);
43 	if (!page) {
44 		cond_resched();
45 		goto repeat;
46 	}
47 	f2fs_wait_on_page_writeback(page, META, true, true);
48 	if (!PageUptodate(page))
49 		SetPageUptodate(page);
50 	return page;
51 }
52 
53 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
54 							bool is_meta)
55 {
56 	struct address_space *mapping = META_MAPPING(sbi);
57 	struct page *page;
58 	struct f2fs_io_info fio = {
59 		.sbi = sbi,
60 		.type = META,
61 		.op = REQ_OP_READ,
62 		.op_flags = REQ_META | REQ_PRIO,
63 		.old_blkaddr = index,
64 		.new_blkaddr = index,
65 		.encrypted_page = NULL,
66 		.is_por = !is_meta,
67 	};
68 	int err;
69 
70 	if (unlikely(!is_meta))
71 		fio.op_flags &= ~REQ_META;
72 repeat:
73 	page = f2fs_grab_cache_page(mapping, index, false);
74 	if (!page) {
75 		cond_resched();
76 		goto repeat;
77 	}
78 	if (PageUptodate(page))
79 		goto out;
80 
81 	fio.page = page;
82 
83 	err = f2fs_submit_page_bio(&fio);
84 	if (err) {
85 		f2fs_put_page(page, 1);
86 		return ERR_PTR(err);
87 	}
88 
89 	f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
90 
91 	lock_page(page);
92 	if (unlikely(page->mapping != mapping)) {
93 		f2fs_put_page(page, 1);
94 		goto repeat;
95 	}
96 
97 	if (unlikely(!PageUptodate(page))) {
98 		f2fs_put_page(page, 1);
99 		return ERR_PTR(-EIO);
100 	}
101 out:
102 	return page;
103 }
104 
105 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
106 {
107 	return __get_meta_page(sbi, index, true);
108 }
109 
110 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
111 {
112 	struct page *page;
113 	int count = 0;
114 
115 retry:
116 	page = __get_meta_page(sbi, index, true);
117 	if (IS_ERR(page)) {
118 		if (PTR_ERR(page) == -EIO &&
119 				++count <= DEFAULT_RETRY_IO_COUNT)
120 			goto retry;
121 		f2fs_stop_checkpoint(sbi, false);
122 	}
123 	return page;
124 }
125 
126 /* for POR only */
127 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
128 {
129 	return __get_meta_page(sbi, index, false);
130 }
131 
132 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
133 							int type)
134 {
135 	struct seg_entry *se;
136 	unsigned int segno, offset;
137 	bool exist;
138 
139 	if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
140 		return true;
141 
142 	segno = GET_SEGNO(sbi, blkaddr);
143 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
144 	se = get_seg_entry(sbi, segno);
145 
146 	exist = f2fs_test_bit(offset, se->cur_valid_map);
147 	if (!exist && type == DATA_GENERIC_ENHANCE) {
148 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
149 			 blkaddr, exist);
150 		set_sbi_flag(sbi, SBI_NEED_FSCK);
151 		WARN_ON(1);
152 	}
153 	return exist;
154 }
155 
156 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
157 					block_t blkaddr, int type)
158 {
159 	switch (type) {
160 	case META_NAT:
161 		break;
162 	case META_SIT:
163 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
164 			return false;
165 		break;
166 	case META_SSA:
167 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
168 			blkaddr < SM_I(sbi)->ssa_blkaddr))
169 			return false;
170 		break;
171 	case META_CP:
172 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
173 			blkaddr < __start_cp_addr(sbi)))
174 			return false;
175 		break;
176 	case META_POR:
177 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
178 			blkaddr < MAIN_BLKADDR(sbi)))
179 			return false;
180 		break;
181 	case DATA_GENERIC:
182 	case DATA_GENERIC_ENHANCE:
183 	case DATA_GENERIC_ENHANCE_READ:
184 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
185 				blkaddr < MAIN_BLKADDR(sbi))) {
186 			f2fs_warn(sbi, "access invalid blkaddr:%u",
187 				  blkaddr);
188 			set_sbi_flag(sbi, SBI_NEED_FSCK);
189 			WARN_ON(1);
190 			return false;
191 		} else {
192 			return __is_bitmap_valid(sbi, blkaddr, type);
193 		}
194 		break;
195 	case META_GENERIC:
196 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
197 			blkaddr >= MAIN_BLKADDR(sbi)))
198 			return false;
199 		break;
200 	default:
201 		BUG();
202 	}
203 
204 	return true;
205 }
206 
207 /*
208  * Readahead CP/NAT/SIT/SSA/POR pages
209  */
210 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
211 							int type, bool sync)
212 {
213 	struct page *page;
214 	block_t blkno = start;
215 	struct f2fs_io_info fio = {
216 		.sbi = sbi,
217 		.type = META,
218 		.op = REQ_OP_READ,
219 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
220 		.encrypted_page = NULL,
221 		.in_list = false,
222 		.is_por = (type == META_POR),
223 	};
224 	struct blk_plug plug;
225 	int err;
226 
227 	if (unlikely(type == META_POR))
228 		fio.op_flags &= ~REQ_META;
229 
230 	blk_start_plug(&plug);
231 	for (; nrpages-- > 0; blkno++) {
232 
233 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
234 			goto out;
235 
236 		switch (type) {
237 		case META_NAT:
238 			if (unlikely(blkno >=
239 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
240 				blkno = 0;
241 			/* get nat block addr */
242 			fio.new_blkaddr = current_nat_addr(sbi,
243 					blkno * NAT_ENTRY_PER_BLOCK);
244 			break;
245 		case META_SIT:
246 			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
247 				goto out;
248 			/* get sit block addr */
249 			fio.new_blkaddr = current_sit_addr(sbi,
250 					blkno * SIT_ENTRY_PER_BLOCK);
251 			break;
252 		case META_SSA:
253 		case META_CP:
254 		case META_POR:
255 			fio.new_blkaddr = blkno;
256 			break;
257 		default:
258 			BUG();
259 		}
260 
261 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
262 						fio.new_blkaddr, false);
263 		if (!page)
264 			continue;
265 		if (PageUptodate(page)) {
266 			f2fs_put_page(page, 1);
267 			continue;
268 		}
269 
270 		fio.page = page;
271 		err = f2fs_submit_page_bio(&fio);
272 		f2fs_put_page(page, err ? 1 : 0);
273 
274 		if (!err)
275 			f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
276 	}
277 out:
278 	blk_finish_plug(&plug);
279 	return blkno - start;
280 }
281 
282 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
283 {
284 	struct page *page;
285 	bool readahead = false;
286 
287 	page = find_get_page(META_MAPPING(sbi), index);
288 	if (!page || !PageUptodate(page))
289 		readahead = true;
290 	f2fs_put_page(page, 0);
291 
292 	if (readahead)
293 		f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
294 }
295 
296 static int __f2fs_write_meta_page(struct page *page,
297 				struct writeback_control *wbc,
298 				enum iostat_type io_type)
299 {
300 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
301 
302 	trace_f2fs_writepage(page, META);
303 
304 	if (unlikely(f2fs_cp_error(sbi)))
305 		goto redirty_out;
306 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
307 		goto redirty_out;
308 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
309 		goto redirty_out;
310 
311 	f2fs_do_write_meta_page(sbi, page, io_type);
312 	dec_page_count(sbi, F2FS_DIRTY_META);
313 
314 	if (wbc->for_reclaim)
315 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
316 
317 	unlock_page(page);
318 
319 	if (unlikely(f2fs_cp_error(sbi)))
320 		f2fs_submit_merged_write(sbi, META);
321 
322 	return 0;
323 
324 redirty_out:
325 	redirty_page_for_writepage(wbc, page);
326 	return AOP_WRITEPAGE_ACTIVATE;
327 }
328 
329 static int f2fs_write_meta_page(struct page *page,
330 				struct writeback_control *wbc)
331 {
332 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
333 }
334 
335 static int f2fs_write_meta_pages(struct address_space *mapping,
336 				struct writeback_control *wbc)
337 {
338 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
339 	long diff, written;
340 
341 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
342 		goto skip_write;
343 
344 	/* collect a number of dirty meta pages and write together */
345 	if (wbc->sync_mode != WB_SYNC_ALL &&
346 			get_pages(sbi, F2FS_DIRTY_META) <
347 					nr_pages_to_skip(sbi, META))
348 		goto skip_write;
349 
350 	/* if locked failed, cp will flush dirty pages instead */
351 	if (!down_write_trylock(&sbi->cp_global_sem))
352 		goto skip_write;
353 
354 	trace_f2fs_writepages(mapping->host, wbc, META);
355 	diff = nr_pages_to_write(sbi, META, wbc);
356 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
357 	up_write(&sbi->cp_global_sem);
358 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
359 	return 0;
360 
361 skip_write:
362 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
363 	trace_f2fs_writepages(mapping->host, wbc, META);
364 	return 0;
365 }
366 
367 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
368 				long nr_to_write, enum iostat_type io_type)
369 {
370 	struct address_space *mapping = META_MAPPING(sbi);
371 	pgoff_t index = 0, prev = ULONG_MAX;
372 	struct pagevec pvec;
373 	long nwritten = 0;
374 	int nr_pages;
375 	struct writeback_control wbc = {
376 		.for_reclaim = 0,
377 	};
378 	struct blk_plug plug;
379 
380 	pagevec_init(&pvec);
381 
382 	blk_start_plug(&plug);
383 
384 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
385 				PAGECACHE_TAG_DIRTY))) {
386 		int i;
387 
388 		for (i = 0; i < nr_pages; i++) {
389 			struct page *page = pvec.pages[i];
390 
391 			if (prev == ULONG_MAX)
392 				prev = page->index - 1;
393 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
394 				pagevec_release(&pvec);
395 				goto stop;
396 			}
397 
398 			lock_page(page);
399 
400 			if (unlikely(page->mapping != mapping)) {
401 continue_unlock:
402 				unlock_page(page);
403 				continue;
404 			}
405 			if (!PageDirty(page)) {
406 				/* someone wrote it for us */
407 				goto continue_unlock;
408 			}
409 
410 			f2fs_wait_on_page_writeback(page, META, true, true);
411 
412 			if (!clear_page_dirty_for_io(page))
413 				goto continue_unlock;
414 
415 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
416 				unlock_page(page);
417 				break;
418 			}
419 			nwritten++;
420 			prev = page->index;
421 			if (unlikely(nwritten >= nr_to_write))
422 				break;
423 		}
424 		pagevec_release(&pvec);
425 		cond_resched();
426 	}
427 stop:
428 	if (nwritten)
429 		f2fs_submit_merged_write(sbi, type);
430 
431 	blk_finish_plug(&plug);
432 
433 	return nwritten;
434 }
435 
436 static int f2fs_set_meta_page_dirty(struct page *page)
437 {
438 	trace_f2fs_set_page_dirty(page, META);
439 
440 	if (!PageUptodate(page))
441 		SetPageUptodate(page);
442 	if (!PageDirty(page)) {
443 		__set_page_dirty_nobuffers(page);
444 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
445 		f2fs_set_page_private(page, 0);
446 		f2fs_trace_pid(page);
447 		return 1;
448 	}
449 	return 0;
450 }
451 
452 const struct address_space_operations f2fs_meta_aops = {
453 	.writepage	= f2fs_write_meta_page,
454 	.writepages	= f2fs_write_meta_pages,
455 	.set_page_dirty	= f2fs_set_meta_page_dirty,
456 	.invalidatepage = f2fs_invalidate_page,
457 	.releasepage	= f2fs_release_page,
458 #ifdef CONFIG_MIGRATION
459 	.migratepage    = f2fs_migrate_page,
460 #endif
461 };
462 
463 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
464 						unsigned int devidx, int type)
465 {
466 	struct inode_management *im = &sbi->im[type];
467 	struct ino_entry *e, *tmp;
468 
469 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
470 
471 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
472 
473 	spin_lock(&im->ino_lock);
474 	e = radix_tree_lookup(&im->ino_root, ino);
475 	if (!e) {
476 		e = tmp;
477 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
478 			f2fs_bug_on(sbi, 1);
479 
480 		memset(e, 0, sizeof(struct ino_entry));
481 		e->ino = ino;
482 
483 		list_add_tail(&e->list, &im->ino_list);
484 		if (type != ORPHAN_INO)
485 			im->ino_num++;
486 	}
487 
488 	if (type == FLUSH_INO)
489 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
490 
491 	spin_unlock(&im->ino_lock);
492 	radix_tree_preload_end();
493 
494 	if (e != tmp)
495 		kmem_cache_free(ino_entry_slab, tmp);
496 }
497 
498 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
499 {
500 	struct inode_management *im = &sbi->im[type];
501 	struct ino_entry *e;
502 
503 	spin_lock(&im->ino_lock);
504 	e = radix_tree_lookup(&im->ino_root, ino);
505 	if (e) {
506 		list_del(&e->list);
507 		radix_tree_delete(&im->ino_root, ino);
508 		im->ino_num--;
509 		spin_unlock(&im->ino_lock);
510 		kmem_cache_free(ino_entry_slab, e);
511 		return;
512 	}
513 	spin_unlock(&im->ino_lock);
514 }
515 
516 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
517 {
518 	/* add new dirty ino entry into list */
519 	__add_ino_entry(sbi, ino, 0, type);
520 }
521 
522 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
523 {
524 	/* remove dirty ino entry from list */
525 	__remove_ino_entry(sbi, ino, type);
526 }
527 
528 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
529 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
530 {
531 	struct inode_management *im = &sbi->im[mode];
532 	struct ino_entry *e;
533 
534 	spin_lock(&im->ino_lock);
535 	e = radix_tree_lookup(&im->ino_root, ino);
536 	spin_unlock(&im->ino_lock);
537 	return e ? true : false;
538 }
539 
540 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
541 {
542 	struct ino_entry *e, *tmp;
543 	int i;
544 
545 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
546 		struct inode_management *im = &sbi->im[i];
547 
548 		spin_lock(&im->ino_lock);
549 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
550 			list_del(&e->list);
551 			radix_tree_delete(&im->ino_root, e->ino);
552 			kmem_cache_free(ino_entry_slab, e);
553 			im->ino_num--;
554 		}
555 		spin_unlock(&im->ino_lock);
556 	}
557 }
558 
559 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
560 					unsigned int devidx, int type)
561 {
562 	__add_ino_entry(sbi, ino, devidx, type);
563 }
564 
565 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
566 					unsigned int devidx, int type)
567 {
568 	struct inode_management *im = &sbi->im[type];
569 	struct ino_entry *e;
570 	bool is_dirty = false;
571 
572 	spin_lock(&im->ino_lock);
573 	e = radix_tree_lookup(&im->ino_root, ino);
574 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
575 		is_dirty = true;
576 	spin_unlock(&im->ino_lock);
577 	return is_dirty;
578 }
579 
580 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
581 {
582 	struct inode_management *im = &sbi->im[ORPHAN_INO];
583 	int err = 0;
584 
585 	spin_lock(&im->ino_lock);
586 
587 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
588 		spin_unlock(&im->ino_lock);
589 		f2fs_show_injection_info(sbi, FAULT_ORPHAN);
590 		return -ENOSPC;
591 	}
592 
593 	if (unlikely(im->ino_num >= sbi->max_orphans))
594 		err = -ENOSPC;
595 	else
596 		im->ino_num++;
597 	spin_unlock(&im->ino_lock);
598 
599 	return err;
600 }
601 
602 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
603 {
604 	struct inode_management *im = &sbi->im[ORPHAN_INO];
605 
606 	spin_lock(&im->ino_lock);
607 	f2fs_bug_on(sbi, im->ino_num == 0);
608 	im->ino_num--;
609 	spin_unlock(&im->ino_lock);
610 }
611 
612 void f2fs_add_orphan_inode(struct inode *inode)
613 {
614 	/* add new orphan ino entry into list */
615 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
616 	f2fs_update_inode_page(inode);
617 }
618 
619 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
620 {
621 	/* remove orphan entry from orphan list */
622 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
623 }
624 
625 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
626 {
627 	struct inode *inode;
628 	struct node_info ni;
629 	int err;
630 
631 	inode = f2fs_iget_retry(sbi->sb, ino);
632 	if (IS_ERR(inode)) {
633 		/*
634 		 * there should be a bug that we can't find the entry
635 		 * to orphan inode.
636 		 */
637 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
638 		return PTR_ERR(inode);
639 	}
640 
641 	err = dquot_initialize(inode);
642 	if (err) {
643 		iput(inode);
644 		goto err_out;
645 	}
646 
647 	clear_nlink(inode);
648 
649 	/* truncate all the data during iput */
650 	iput(inode);
651 
652 	err = f2fs_get_node_info(sbi, ino, &ni);
653 	if (err)
654 		goto err_out;
655 
656 	/* ENOMEM was fully retried in f2fs_evict_inode. */
657 	if (ni.blk_addr != NULL_ADDR) {
658 		err = -EIO;
659 		goto err_out;
660 	}
661 	return 0;
662 
663 err_out:
664 	set_sbi_flag(sbi, SBI_NEED_FSCK);
665 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
666 		  __func__, ino);
667 	return err;
668 }
669 
670 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
671 {
672 	block_t start_blk, orphan_blocks, i, j;
673 	unsigned int s_flags = sbi->sb->s_flags;
674 	int err = 0;
675 #ifdef CONFIG_QUOTA
676 	int quota_enabled;
677 #endif
678 
679 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
680 		return 0;
681 
682 	if (bdev_read_only(sbi->sb->s_bdev)) {
683 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
684 		return 0;
685 	}
686 
687 	if (s_flags & SB_RDONLY) {
688 		f2fs_info(sbi, "orphan cleanup on readonly fs");
689 		sbi->sb->s_flags &= ~SB_RDONLY;
690 	}
691 
692 #ifdef CONFIG_QUOTA
693 	/* Needed for iput() to work correctly and not trash data */
694 	sbi->sb->s_flags |= SB_ACTIVE;
695 
696 	/*
697 	 * Turn on quotas which were not enabled for read-only mounts if
698 	 * filesystem has quota feature, so that they are updated correctly.
699 	 */
700 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
701 #endif
702 
703 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
704 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
705 
706 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
707 
708 	for (i = 0; i < orphan_blocks; i++) {
709 		struct page *page;
710 		struct f2fs_orphan_block *orphan_blk;
711 
712 		page = f2fs_get_meta_page(sbi, start_blk + i);
713 		if (IS_ERR(page)) {
714 			err = PTR_ERR(page);
715 			goto out;
716 		}
717 
718 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
719 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
720 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
721 			err = recover_orphan_inode(sbi, ino);
722 			if (err) {
723 				f2fs_put_page(page, 1);
724 				goto out;
725 			}
726 		}
727 		f2fs_put_page(page, 1);
728 	}
729 	/* clear Orphan Flag */
730 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
731 out:
732 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
733 
734 #ifdef CONFIG_QUOTA
735 	/* Turn quotas off */
736 	if (quota_enabled)
737 		f2fs_quota_off_umount(sbi->sb);
738 #endif
739 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
740 
741 	return err;
742 }
743 
744 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
745 {
746 	struct list_head *head;
747 	struct f2fs_orphan_block *orphan_blk = NULL;
748 	unsigned int nentries = 0;
749 	unsigned short index = 1;
750 	unsigned short orphan_blocks;
751 	struct page *page = NULL;
752 	struct ino_entry *orphan = NULL;
753 	struct inode_management *im = &sbi->im[ORPHAN_INO];
754 
755 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
756 
757 	/*
758 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
759 	 * orphan inode operations are covered under f2fs_lock_op().
760 	 * And, spin_lock should be avoided due to page operations below.
761 	 */
762 	head = &im->ino_list;
763 
764 	/* loop for each orphan inode entry and write them in Jornal block */
765 	list_for_each_entry(orphan, head, list) {
766 		if (!page) {
767 			page = f2fs_grab_meta_page(sbi, start_blk++);
768 			orphan_blk =
769 				(struct f2fs_orphan_block *)page_address(page);
770 			memset(orphan_blk, 0, sizeof(*orphan_blk));
771 		}
772 
773 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
774 
775 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
776 			/*
777 			 * an orphan block is full of 1020 entries,
778 			 * then we need to flush current orphan blocks
779 			 * and bring another one in memory
780 			 */
781 			orphan_blk->blk_addr = cpu_to_le16(index);
782 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
783 			orphan_blk->entry_count = cpu_to_le32(nentries);
784 			set_page_dirty(page);
785 			f2fs_put_page(page, 1);
786 			index++;
787 			nentries = 0;
788 			page = NULL;
789 		}
790 	}
791 
792 	if (page) {
793 		orphan_blk->blk_addr = cpu_to_le16(index);
794 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
795 		orphan_blk->entry_count = cpu_to_le32(nentries);
796 		set_page_dirty(page);
797 		f2fs_put_page(page, 1);
798 	}
799 }
800 
801 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
802 						struct f2fs_checkpoint *ckpt)
803 {
804 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
805 	__u32 chksum;
806 
807 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
808 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
809 		chksum_ofs += sizeof(chksum);
810 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
811 						F2FS_BLKSIZE - chksum_ofs);
812 	}
813 	return chksum;
814 }
815 
816 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
817 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
818 		unsigned long long *version)
819 {
820 	size_t crc_offset = 0;
821 	__u32 crc;
822 
823 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
824 	if (IS_ERR(*cp_page))
825 		return PTR_ERR(*cp_page);
826 
827 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
828 
829 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
830 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
831 			crc_offset > CP_CHKSUM_OFFSET) {
832 		f2fs_put_page(*cp_page, 1);
833 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
834 		return -EINVAL;
835 	}
836 
837 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
838 	if (crc != cur_cp_crc(*cp_block)) {
839 		f2fs_put_page(*cp_page, 1);
840 		f2fs_warn(sbi, "invalid crc value");
841 		return -EINVAL;
842 	}
843 
844 	*version = cur_cp_version(*cp_block);
845 	return 0;
846 }
847 
848 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
849 				block_t cp_addr, unsigned long long *version)
850 {
851 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
852 	struct f2fs_checkpoint *cp_block = NULL;
853 	unsigned long long cur_version = 0, pre_version = 0;
854 	int err;
855 
856 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
857 					&cp_page_1, version);
858 	if (err)
859 		return NULL;
860 
861 	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
862 					sbi->blocks_per_seg) {
863 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
864 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
865 		goto invalid_cp;
866 	}
867 	pre_version = *version;
868 
869 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
870 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
871 					&cp_page_2, version);
872 	if (err)
873 		goto invalid_cp;
874 	cur_version = *version;
875 
876 	if (cur_version == pre_version) {
877 		*version = cur_version;
878 		f2fs_put_page(cp_page_2, 1);
879 		return cp_page_1;
880 	}
881 	f2fs_put_page(cp_page_2, 1);
882 invalid_cp:
883 	f2fs_put_page(cp_page_1, 1);
884 	return NULL;
885 }
886 
887 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
888 {
889 	struct f2fs_checkpoint *cp_block;
890 	struct f2fs_super_block *fsb = sbi->raw_super;
891 	struct page *cp1, *cp2, *cur_page;
892 	unsigned long blk_size = sbi->blocksize;
893 	unsigned long long cp1_version = 0, cp2_version = 0;
894 	unsigned long long cp_start_blk_no;
895 	unsigned int cp_blks = 1 + __cp_payload(sbi);
896 	block_t cp_blk_no;
897 	int i;
898 	int err;
899 
900 	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
901 				  GFP_KERNEL);
902 	if (!sbi->ckpt)
903 		return -ENOMEM;
904 	/*
905 	 * Finding out valid cp block involves read both
906 	 * sets( cp pack 1 and cp pack 2)
907 	 */
908 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
909 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
910 
911 	/* The second checkpoint pack should start at the next segment */
912 	cp_start_blk_no += ((unsigned long long)1) <<
913 				le32_to_cpu(fsb->log_blocks_per_seg);
914 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
915 
916 	if (cp1 && cp2) {
917 		if (ver_after(cp2_version, cp1_version))
918 			cur_page = cp2;
919 		else
920 			cur_page = cp1;
921 	} else if (cp1) {
922 		cur_page = cp1;
923 	} else if (cp2) {
924 		cur_page = cp2;
925 	} else {
926 		err = -EFSCORRUPTED;
927 		goto fail_no_cp;
928 	}
929 
930 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
931 	memcpy(sbi->ckpt, cp_block, blk_size);
932 
933 	if (cur_page == cp1)
934 		sbi->cur_cp_pack = 1;
935 	else
936 		sbi->cur_cp_pack = 2;
937 
938 	/* Sanity checking of checkpoint */
939 	if (f2fs_sanity_check_ckpt(sbi)) {
940 		err = -EFSCORRUPTED;
941 		goto free_fail_no_cp;
942 	}
943 
944 	if (cp_blks <= 1)
945 		goto done;
946 
947 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
948 	if (cur_page == cp2)
949 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
950 
951 	for (i = 1; i < cp_blks; i++) {
952 		void *sit_bitmap_ptr;
953 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
954 
955 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
956 		if (IS_ERR(cur_page)) {
957 			err = PTR_ERR(cur_page);
958 			goto free_fail_no_cp;
959 		}
960 		sit_bitmap_ptr = page_address(cur_page);
961 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
962 		f2fs_put_page(cur_page, 1);
963 	}
964 done:
965 	f2fs_put_page(cp1, 1);
966 	f2fs_put_page(cp2, 1);
967 	return 0;
968 
969 free_fail_no_cp:
970 	f2fs_put_page(cp1, 1);
971 	f2fs_put_page(cp2, 1);
972 fail_no_cp:
973 	kvfree(sbi->ckpt);
974 	return err;
975 }
976 
977 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
978 {
979 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
980 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
981 
982 	if (is_inode_flag_set(inode, flag))
983 		return;
984 
985 	set_inode_flag(inode, flag);
986 	if (!f2fs_is_volatile_file(inode))
987 		list_add_tail(&F2FS_I(inode)->dirty_list,
988 						&sbi->inode_list[type]);
989 	stat_inc_dirty_inode(sbi, type);
990 }
991 
992 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
993 {
994 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
995 
996 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
997 		return;
998 
999 	list_del_init(&F2FS_I(inode)->dirty_list);
1000 	clear_inode_flag(inode, flag);
1001 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1002 }
1003 
1004 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1005 {
1006 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1007 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1008 
1009 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1010 			!S_ISLNK(inode->i_mode))
1011 		return;
1012 
1013 	spin_lock(&sbi->inode_lock[type]);
1014 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1015 		__add_dirty_inode(inode, type);
1016 	inode_inc_dirty_pages(inode);
1017 	spin_unlock(&sbi->inode_lock[type]);
1018 
1019 	f2fs_set_page_private(page, 0);
1020 	f2fs_trace_pid(page);
1021 }
1022 
1023 void f2fs_remove_dirty_inode(struct inode *inode)
1024 {
1025 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1026 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1027 
1028 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1029 			!S_ISLNK(inode->i_mode))
1030 		return;
1031 
1032 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1033 		return;
1034 
1035 	spin_lock(&sbi->inode_lock[type]);
1036 	__remove_dirty_inode(inode, type);
1037 	spin_unlock(&sbi->inode_lock[type]);
1038 }
1039 
1040 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1041 {
1042 	struct list_head *head;
1043 	struct inode *inode;
1044 	struct f2fs_inode_info *fi;
1045 	bool is_dir = (type == DIR_INODE);
1046 	unsigned long ino = 0;
1047 
1048 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1049 				get_pages(sbi, is_dir ?
1050 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1051 retry:
1052 	if (unlikely(f2fs_cp_error(sbi))) {
1053 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1054 				get_pages(sbi, is_dir ?
1055 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1056 		return -EIO;
1057 	}
1058 
1059 	spin_lock(&sbi->inode_lock[type]);
1060 
1061 	head = &sbi->inode_list[type];
1062 	if (list_empty(head)) {
1063 		spin_unlock(&sbi->inode_lock[type]);
1064 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1065 				get_pages(sbi, is_dir ?
1066 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1067 		return 0;
1068 	}
1069 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1070 	inode = igrab(&fi->vfs_inode);
1071 	spin_unlock(&sbi->inode_lock[type]);
1072 	if (inode) {
1073 		unsigned long cur_ino = inode->i_ino;
1074 
1075 		F2FS_I(inode)->cp_task = current;
1076 
1077 		filemap_fdatawrite(inode->i_mapping);
1078 
1079 		F2FS_I(inode)->cp_task = NULL;
1080 
1081 		iput(inode);
1082 		/* We need to give cpu to another writers. */
1083 		if (ino == cur_ino)
1084 			cond_resched();
1085 		else
1086 			ino = cur_ino;
1087 	} else {
1088 		/*
1089 		 * We should submit bio, since it exists several
1090 		 * wribacking dentry pages in the freeing inode.
1091 		 */
1092 		f2fs_submit_merged_write(sbi, DATA);
1093 		cond_resched();
1094 	}
1095 	goto retry;
1096 }
1097 
1098 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1099 {
1100 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1101 	struct inode *inode;
1102 	struct f2fs_inode_info *fi;
1103 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1104 
1105 	while (total--) {
1106 		if (unlikely(f2fs_cp_error(sbi)))
1107 			return -EIO;
1108 
1109 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1110 		if (list_empty(head)) {
1111 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1112 			return 0;
1113 		}
1114 		fi = list_first_entry(head, struct f2fs_inode_info,
1115 							gdirty_list);
1116 		inode = igrab(&fi->vfs_inode);
1117 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1118 		if (inode) {
1119 			sync_inode_metadata(inode, 0);
1120 
1121 			/* it's on eviction */
1122 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1123 				f2fs_update_inode_page(inode);
1124 			iput(inode);
1125 		}
1126 	}
1127 	return 0;
1128 }
1129 
1130 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1131 {
1132 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1133 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1134 	nid_t last_nid = nm_i->next_scan_nid;
1135 
1136 	next_free_nid(sbi, &last_nid);
1137 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1138 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1139 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1140 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1141 }
1142 
1143 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1144 {
1145 	bool ret = false;
1146 
1147 	if (!is_journalled_quota(sbi))
1148 		return false;
1149 
1150 	down_write(&sbi->quota_sem);
1151 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1152 		ret = false;
1153 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1154 		ret = false;
1155 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1156 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1157 		ret = true;
1158 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1159 		ret = true;
1160 	}
1161 	up_write(&sbi->quota_sem);
1162 	return ret;
1163 }
1164 
1165 /*
1166  * Freeze all the FS-operations for checkpoint.
1167  */
1168 static int block_operations(struct f2fs_sb_info *sbi)
1169 {
1170 	struct writeback_control wbc = {
1171 		.sync_mode = WB_SYNC_ALL,
1172 		.nr_to_write = LONG_MAX,
1173 		.for_reclaim = 0,
1174 	};
1175 	int err = 0, cnt = 0;
1176 
1177 	/*
1178 	 * Let's flush inline_data in dirty node pages.
1179 	 */
1180 	f2fs_flush_inline_data(sbi);
1181 
1182 retry_flush_quotas:
1183 	f2fs_lock_all(sbi);
1184 	if (__need_flush_quota(sbi)) {
1185 		int locked;
1186 
1187 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1188 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1189 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1190 			goto retry_flush_dents;
1191 		}
1192 		f2fs_unlock_all(sbi);
1193 
1194 		/* only failed during mount/umount/freeze/quotactl */
1195 		locked = down_read_trylock(&sbi->sb->s_umount);
1196 		f2fs_quota_sync(sbi->sb, -1);
1197 		if (locked)
1198 			up_read(&sbi->sb->s_umount);
1199 		cond_resched();
1200 		goto retry_flush_quotas;
1201 	}
1202 
1203 retry_flush_dents:
1204 	/* write all the dirty dentry pages */
1205 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1206 		f2fs_unlock_all(sbi);
1207 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1208 		if (err)
1209 			return err;
1210 		cond_resched();
1211 		goto retry_flush_quotas;
1212 	}
1213 
1214 	/*
1215 	 * POR: we should ensure that there are no dirty node pages
1216 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1217 	 */
1218 	down_write(&sbi->node_change);
1219 
1220 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1221 		up_write(&sbi->node_change);
1222 		f2fs_unlock_all(sbi);
1223 		err = f2fs_sync_inode_meta(sbi);
1224 		if (err)
1225 			return err;
1226 		cond_resched();
1227 		goto retry_flush_quotas;
1228 	}
1229 
1230 retry_flush_nodes:
1231 	down_write(&sbi->node_write);
1232 
1233 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1234 		up_write(&sbi->node_write);
1235 		atomic_inc(&sbi->wb_sync_req[NODE]);
1236 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1237 		atomic_dec(&sbi->wb_sync_req[NODE]);
1238 		if (err) {
1239 			up_write(&sbi->node_change);
1240 			f2fs_unlock_all(sbi);
1241 			return err;
1242 		}
1243 		cond_resched();
1244 		goto retry_flush_nodes;
1245 	}
1246 
1247 	/*
1248 	 * sbi->node_change is used only for AIO write_begin path which produces
1249 	 * dirty node blocks and some checkpoint values by block allocation.
1250 	 */
1251 	__prepare_cp_block(sbi);
1252 	up_write(&sbi->node_change);
1253 	return err;
1254 }
1255 
1256 static void unblock_operations(struct f2fs_sb_info *sbi)
1257 {
1258 	up_write(&sbi->node_write);
1259 	f2fs_unlock_all(sbi);
1260 }
1261 
1262 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1263 {
1264 	DEFINE_WAIT(wait);
1265 
1266 	for (;;) {
1267 		if (!get_pages(sbi, type))
1268 			break;
1269 
1270 		if (unlikely(f2fs_cp_error(sbi)))
1271 			break;
1272 
1273 		if (type == F2FS_DIRTY_META)
1274 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1275 							FS_CP_META_IO);
1276 		else if (type == F2FS_WB_CP_DATA)
1277 			f2fs_submit_merged_write(sbi, DATA);
1278 
1279 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1280 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1281 	}
1282 	finish_wait(&sbi->cp_wait, &wait);
1283 }
1284 
1285 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1286 {
1287 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1288 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1289 	unsigned long flags;
1290 
1291 	spin_lock_irqsave(&sbi->cp_lock, flags);
1292 
1293 	if ((cpc->reason & CP_UMOUNT) &&
1294 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1295 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1296 		disable_nat_bits(sbi, false);
1297 
1298 	if (cpc->reason & CP_TRIMMED)
1299 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1300 	else
1301 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1302 
1303 	if (cpc->reason & CP_UMOUNT)
1304 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1305 	else
1306 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1307 
1308 	if (cpc->reason & CP_FASTBOOT)
1309 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1310 	else
1311 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1312 
1313 	if (orphan_num)
1314 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1315 	else
1316 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1317 
1318 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1319 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1320 
1321 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1322 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1323 	else
1324 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1325 
1326 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1327 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1328 	else
1329 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1330 
1331 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1332 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1333 	else
1334 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1335 
1336 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1337 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1338 	else
1339 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1340 
1341 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1342 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1343 
1344 	/* set this flag to activate crc|cp_ver for recovery */
1345 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1346 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1347 
1348 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1349 }
1350 
1351 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1352 	void *src, block_t blk_addr)
1353 {
1354 	struct writeback_control wbc = {
1355 		.for_reclaim = 0,
1356 	};
1357 
1358 	/*
1359 	 * pagevec_lookup_tag and lock_page again will take
1360 	 * some extra time. Therefore, f2fs_update_meta_pages and
1361 	 * f2fs_sync_meta_pages are combined in this function.
1362 	 */
1363 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1364 	int err;
1365 
1366 	f2fs_wait_on_page_writeback(page, META, true, true);
1367 
1368 	memcpy(page_address(page), src, PAGE_SIZE);
1369 
1370 	set_page_dirty(page);
1371 	if (unlikely(!clear_page_dirty_for_io(page)))
1372 		f2fs_bug_on(sbi, 1);
1373 
1374 	/* writeout cp pack 2 page */
1375 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1376 	if (unlikely(err && f2fs_cp_error(sbi))) {
1377 		f2fs_put_page(page, 1);
1378 		return;
1379 	}
1380 
1381 	f2fs_bug_on(sbi, err);
1382 	f2fs_put_page(page, 0);
1383 
1384 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1385 	f2fs_submit_merged_write(sbi, META_FLUSH);
1386 }
1387 
1388 static inline u64 get_sectors_written(struct block_device *bdev)
1389 {
1390 	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1391 }
1392 
1393 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1394 {
1395 	if (f2fs_is_multi_device(sbi)) {
1396 		u64 sectors = 0;
1397 		int i;
1398 
1399 		for (i = 0; i < sbi->s_ndevs; i++)
1400 			sectors += get_sectors_written(FDEV(i).bdev);
1401 
1402 		return sectors;
1403 	}
1404 
1405 	return get_sectors_written(sbi->sb->s_bdev);
1406 }
1407 
1408 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1409 {
1410 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1411 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1412 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1413 	block_t start_blk;
1414 	unsigned int data_sum_blocks, orphan_blocks;
1415 	__u32 crc32 = 0;
1416 	int i;
1417 	int cp_payload_blks = __cp_payload(sbi);
1418 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1419 	u64 kbytes_written;
1420 	int err;
1421 
1422 	/* Flush all the NAT/SIT pages */
1423 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1424 
1425 	/* start to update checkpoint, cp ver is already updated previously */
1426 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1427 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1428 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1429 		ckpt->cur_node_segno[i] =
1430 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1431 		ckpt->cur_node_blkoff[i] =
1432 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1433 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1434 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1435 	}
1436 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1437 		ckpt->cur_data_segno[i] =
1438 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1439 		ckpt->cur_data_blkoff[i] =
1440 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1441 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1442 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1443 	}
1444 
1445 	/* 2 cp + n data seg summary + orphan inode blocks */
1446 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1447 	spin_lock_irqsave(&sbi->cp_lock, flags);
1448 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1449 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1450 	else
1451 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1452 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1453 
1454 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1455 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1456 			orphan_blocks);
1457 
1458 	if (__remain_node_summaries(cpc->reason))
1459 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1460 				cp_payload_blks + data_sum_blocks +
1461 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1462 	else
1463 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1464 				cp_payload_blks + data_sum_blocks +
1465 				orphan_blocks);
1466 
1467 	/* update ckpt flag for checkpoint */
1468 	update_ckpt_flags(sbi, cpc);
1469 
1470 	/* update SIT/NAT bitmap */
1471 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1472 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1473 
1474 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1475 	*((__le32 *)((unsigned char *)ckpt +
1476 				le32_to_cpu(ckpt->checksum_offset)))
1477 				= cpu_to_le32(crc32);
1478 
1479 	start_blk = __start_cp_next_addr(sbi);
1480 
1481 	/* write nat bits */
1482 	if (enabled_nat_bits(sbi, cpc)) {
1483 		__u64 cp_ver = cur_cp_version(ckpt);
1484 		block_t blk;
1485 
1486 		cp_ver |= ((__u64)crc32 << 32);
1487 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1488 
1489 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1490 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1491 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1492 					(i << F2FS_BLKSIZE_BITS), blk + i);
1493 	}
1494 
1495 	/* write out checkpoint buffer at block 0 */
1496 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1497 
1498 	for (i = 1; i < 1 + cp_payload_blks; i++)
1499 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1500 							start_blk++);
1501 
1502 	if (orphan_num) {
1503 		write_orphan_inodes(sbi, start_blk);
1504 		start_blk += orphan_blocks;
1505 	}
1506 
1507 	f2fs_write_data_summaries(sbi, start_blk);
1508 	start_blk += data_sum_blocks;
1509 
1510 	/* Record write statistics in the hot node summary */
1511 	kbytes_written = sbi->kbytes_written;
1512 	kbytes_written += (f2fs_get_sectors_written(sbi) -
1513 				sbi->sectors_written_start) >> 1;
1514 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1515 
1516 	if (__remain_node_summaries(cpc->reason)) {
1517 		f2fs_write_node_summaries(sbi, start_blk);
1518 		start_blk += NR_CURSEG_NODE_TYPE;
1519 	}
1520 
1521 	/* update user_block_counts */
1522 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1523 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1524 
1525 	/* Here, we have one bio having CP pack except cp pack 2 page */
1526 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1527 	/* Wait for all dirty meta pages to be submitted for IO */
1528 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1529 
1530 	/* wait for previous submitted meta pages writeback */
1531 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1532 
1533 	/* flush all device cache */
1534 	err = f2fs_flush_device_cache(sbi);
1535 	if (err)
1536 		return err;
1537 
1538 	/* barrier and flush checkpoint cp pack 2 page if it can */
1539 	commit_checkpoint(sbi, ckpt, start_blk);
1540 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1541 
1542 	/*
1543 	 * invalidate intermediate page cache borrowed from meta inode which are
1544 	 * used for migration of encrypted, verity or compressed inode's blocks.
1545 	 */
1546 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1547 		f2fs_sb_has_compression(sbi))
1548 		invalidate_mapping_pages(META_MAPPING(sbi),
1549 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1550 
1551 	f2fs_release_ino_entry(sbi, false);
1552 
1553 	f2fs_reset_fsync_node_info(sbi);
1554 
1555 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1556 	clear_sbi_flag(sbi, SBI_NEED_CP);
1557 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1558 
1559 	spin_lock(&sbi->stat_lock);
1560 	sbi->unusable_block_count = 0;
1561 	spin_unlock(&sbi->stat_lock);
1562 
1563 	__set_cp_next_pack(sbi);
1564 
1565 	/*
1566 	 * redirty superblock if metadata like node page or inode cache is
1567 	 * updated during writing checkpoint.
1568 	 */
1569 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1570 			get_pages(sbi, F2FS_DIRTY_IMETA))
1571 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1572 
1573 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1574 
1575 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1576 }
1577 
1578 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1579 {
1580 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1581 	unsigned long long ckpt_ver;
1582 	int err = 0;
1583 
1584 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1585 		return -EROFS;
1586 
1587 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1588 		if (cpc->reason != CP_PAUSE)
1589 			return 0;
1590 		f2fs_warn(sbi, "Start checkpoint disabled!");
1591 	}
1592 	if (cpc->reason != CP_RESIZE)
1593 		down_write(&sbi->cp_global_sem);
1594 
1595 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1596 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1597 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1598 		goto out;
1599 	if (unlikely(f2fs_cp_error(sbi))) {
1600 		err = -EIO;
1601 		goto out;
1602 	}
1603 
1604 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1605 
1606 	err = block_operations(sbi);
1607 	if (err)
1608 		goto out;
1609 
1610 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1611 
1612 	f2fs_flush_merged_writes(sbi);
1613 
1614 	/* this is the case of multiple fstrims without any changes */
1615 	if (cpc->reason & CP_DISCARD) {
1616 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1617 			unblock_operations(sbi);
1618 			goto out;
1619 		}
1620 
1621 		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1622 				SIT_I(sbi)->dirty_sentries == 0 &&
1623 				prefree_segments(sbi) == 0) {
1624 			f2fs_flush_sit_entries(sbi, cpc);
1625 			f2fs_clear_prefree_segments(sbi, cpc);
1626 			unblock_operations(sbi);
1627 			goto out;
1628 		}
1629 	}
1630 
1631 	/*
1632 	 * update checkpoint pack index
1633 	 * Increase the version number so that
1634 	 * SIT entries and seg summaries are written at correct place
1635 	 */
1636 	ckpt_ver = cur_cp_version(ckpt);
1637 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1638 
1639 	/* write cached NAT/SIT entries to NAT/SIT area */
1640 	err = f2fs_flush_nat_entries(sbi, cpc);
1641 	if (err)
1642 		goto stop;
1643 
1644 	f2fs_flush_sit_entries(sbi, cpc);
1645 
1646 	/* save inmem log status */
1647 	f2fs_save_inmem_curseg(sbi);
1648 
1649 	err = do_checkpoint(sbi, cpc);
1650 	if (err)
1651 		f2fs_release_discard_addrs(sbi);
1652 	else
1653 		f2fs_clear_prefree_segments(sbi, cpc);
1654 
1655 	f2fs_restore_inmem_curseg(sbi);
1656 stop:
1657 	unblock_operations(sbi);
1658 	stat_inc_cp_count(sbi->stat_info);
1659 
1660 	if (cpc->reason & CP_RECOVERY)
1661 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1662 
1663 	/* update CP_TIME to trigger checkpoint periodically */
1664 	f2fs_update_time(sbi, CP_TIME);
1665 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1666 out:
1667 	if (cpc->reason != CP_RESIZE)
1668 		up_write(&sbi->cp_global_sem);
1669 	return err;
1670 }
1671 
1672 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1673 {
1674 	int i;
1675 
1676 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1677 		struct inode_management *im = &sbi->im[i];
1678 
1679 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1680 		spin_lock_init(&im->ino_lock);
1681 		INIT_LIST_HEAD(&im->ino_list);
1682 		im->ino_num = 0;
1683 	}
1684 
1685 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1686 			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1687 				F2FS_ORPHANS_PER_BLOCK;
1688 }
1689 
1690 int __init f2fs_create_checkpoint_caches(void)
1691 {
1692 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1693 			sizeof(struct ino_entry));
1694 	if (!ino_entry_slab)
1695 		return -ENOMEM;
1696 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1697 			sizeof(struct inode_entry));
1698 	if (!f2fs_inode_entry_slab) {
1699 		kmem_cache_destroy(ino_entry_slab);
1700 		return -ENOMEM;
1701 	}
1702 	return 0;
1703 }
1704 
1705 void f2fs_destroy_checkpoint_caches(void)
1706 {
1707 	kmem_cache_destroy(ino_entry_slab);
1708 	kmem_cache_destroy(f2fs_inode_entry_slab);
1709 }
1710