xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision f519cd13)
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 = NULL;
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 /*
54  * We guarantee no failure on the returned page.
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 	lock_page(page);
93 	if (unlikely(page->mapping != mapping)) {
94 		f2fs_put_page(page, 1);
95 		goto repeat;
96 	}
97 
98 	if (unlikely(!PageUptodate(page))) {
99 		f2fs_put_page(page, 1);
100 		return ERR_PTR(-EIO);
101 	}
102 out:
103 	return page;
104 }
105 
106 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
107 {
108 	return __get_meta_page(sbi, index, true);
109 }
110 
111 struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
112 {
113 	struct page *page;
114 	int count = 0;
115 
116 retry:
117 	page = __get_meta_page(sbi, index, true);
118 	if (IS_ERR(page)) {
119 		if (PTR_ERR(page) == -EIO &&
120 				++count <= DEFAULT_RETRY_IO_COUNT)
121 			goto retry;
122 		f2fs_stop_checkpoint(sbi, false);
123 	}
124 	return page;
125 }
126 
127 /* for POR only */
128 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
129 {
130 	return __get_meta_page(sbi, index, false);
131 }
132 
133 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
134 							int type)
135 {
136 	struct seg_entry *se;
137 	unsigned int segno, offset;
138 	bool exist;
139 
140 	if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
141 		return true;
142 
143 	segno = GET_SEGNO(sbi, blkaddr);
144 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
145 	se = get_seg_entry(sbi, segno);
146 
147 	exist = f2fs_test_bit(offset, se->cur_valid_map);
148 	if (!exist && type == DATA_GENERIC_ENHANCE) {
149 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
150 			 blkaddr, exist);
151 		set_sbi_flag(sbi, SBI_NEED_FSCK);
152 		WARN_ON(1);
153 	}
154 	return exist;
155 }
156 
157 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
158 					block_t blkaddr, int type)
159 {
160 	switch (type) {
161 	case META_NAT:
162 		break;
163 	case META_SIT:
164 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
165 			return false;
166 		break;
167 	case META_SSA:
168 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
169 			blkaddr < SM_I(sbi)->ssa_blkaddr))
170 			return false;
171 		break;
172 	case META_CP:
173 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
174 			blkaddr < __start_cp_addr(sbi)))
175 			return false;
176 		break;
177 	case META_POR:
178 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
179 			blkaddr < MAIN_BLKADDR(sbi)))
180 			return false;
181 		break;
182 	case DATA_GENERIC:
183 	case DATA_GENERIC_ENHANCE:
184 	case DATA_GENERIC_ENHANCE_READ:
185 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
186 				blkaddr < MAIN_BLKADDR(sbi))) {
187 			f2fs_warn(sbi, "access invalid blkaddr:%u",
188 				  blkaddr);
189 			set_sbi_flag(sbi, SBI_NEED_FSCK);
190 			WARN_ON(1);
191 			return false;
192 		} else {
193 			return __is_bitmap_valid(sbi, blkaddr, type);
194 		}
195 		break;
196 	case META_GENERIC:
197 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
198 			blkaddr >= MAIN_BLKADDR(sbi)))
199 			return false;
200 		break;
201 	default:
202 		BUG();
203 	}
204 
205 	return true;
206 }
207 
208 /*
209  * Readahead CP/NAT/SIT/SSA pages
210  */
211 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
212 							int type, bool sync)
213 {
214 	struct page *page;
215 	block_t blkno = start;
216 	struct f2fs_io_info fio = {
217 		.sbi = sbi,
218 		.type = META,
219 		.op = REQ_OP_READ,
220 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
221 		.encrypted_page = NULL,
222 		.in_list = false,
223 		.is_por = (type == META_POR),
224 	};
225 	struct blk_plug plug;
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 			/* get sit block addr */
247 			fio.new_blkaddr = current_sit_addr(sbi,
248 					blkno * SIT_ENTRY_PER_BLOCK);
249 			break;
250 		case META_SSA:
251 		case META_CP:
252 		case META_POR:
253 			fio.new_blkaddr = blkno;
254 			break;
255 		default:
256 			BUG();
257 		}
258 
259 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
260 						fio.new_blkaddr, false);
261 		if (!page)
262 			continue;
263 		if (PageUptodate(page)) {
264 			f2fs_put_page(page, 1);
265 			continue;
266 		}
267 
268 		fio.page = page;
269 		f2fs_submit_page_bio(&fio);
270 		f2fs_put_page(page, 0);
271 	}
272 out:
273 	blk_finish_plug(&plug);
274 	return blkno - start;
275 }
276 
277 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
278 {
279 	struct page *page;
280 	bool readahead = false;
281 
282 	page = find_get_page(META_MAPPING(sbi), index);
283 	if (!page || !PageUptodate(page))
284 		readahead = true;
285 	f2fs_put_page(page, 0);
286 
287 	if (readahead)
288 		f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
289 }
290 
291 static int __f2fs_write_meta_page(struct page *page,
292 				struct writeback_control *wbc,
293 				enum iostat_type io_type)
294 {
295 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
296 
297 	trace_f2fs_writepage(page, META);
298 
299 	if (unlikely(f2fs_cp_error(sbi)))
300 		goto redirty_out;
301 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
302 		goto redirty_out;
303 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
304 		goto redirty_out;
305 
306 	f2fs_do_write_meta_page(sbi, page, io_type);
307 	dec_page_count(sbi, F2FS_DIRTY_META);
308 
309 	if (wbc->for_reclaim)
310 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
311 
312 	unlock_page(page);
313 
314 	if (unlikely(f2fs_cp_error(sbi)))
315 		f2fs_submit_merged_write(sbi, META);
316 
317 	return 0;
318 
319 redirty_out:
320 	redirty_page_for_writepage(wbc, page);
321 	return AOP_WRITEPAGE_ACTIVATE;
322 }
323 
324 static int f2fs_write_meta_page(struct page *page,
325 				struct writeback_control *wbc)
326 {
327 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
328 }
329 
330 static int f2fs_write_meta_pages(struct address_space *mapping,
331 				struct writeback_control *wbc)
332 {
333 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
334 	long diff, written;
335 
336 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
337 		goto skip_write;
338 
339 	/* collect a number of dirty meta pages and write together */
340 	if (wbc->sync_mode != WB_SYNC_ALL &&
341 			get_pages(sbi, F2FS_DIRTY_META) <
342 					nr_pages_to_skip(sbi, META))
343 		goto skip_write;
344 
345 	/* if locked failed, cp will flush dirty pages instead */
346 	if (!mutex_trylock(&sbi->cp_mutex))
347 		goto skip_write;
348 
349 	trace_f2fs_writepages(mapping->host, wbc, META);
350 	diff = nr_pages_to_write(sbi, META, wbc);
351 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
352 	mutex_unlock(&sbi->cp_mutex);
353 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
354 	return 0;
355 
356 skip_write:
357 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
358 	trace_f2fs_writepages(mapping->host, wbc, META);
359 	return 0;
360 }
361 
362 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
363 				long nr_to_write, enum iostat_type io_type)
364 {
365 	struct address_space *mapping = META_MAPPING(sbi);
366 	pgoff_t index = 0, prev = ULONG_MAX;
367 	struct pagevec pvec;
368 	long nwritten = 0;
369 	int nr_pages;
370 	struct writeback_control wbc = {
371 		.for_reclaim = 0,
372 	};
373 	struct blk_plug plug;
374 
375 	pagevec_init(&pvec);
376 
377 	blk_start_plug(&plug);
378 
379 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
380 				PAGECACHE_TAG_DIRTY))) {
381 		int i;
382 
383 		for (i = 0; i < nr_pages; i++) {
384 			struct page *page = pvec.pages[i];
385 
386 			if (prev == ULONG_MAX)
387 				prev = page->index - 1;
388 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
389 				pagevec_release(&pvec);
390 				goto stop;
391 			}
392 
393 			lock_page(page);
394 
395 			if (unlikely(page->mapping != mapping)) {
396 continue_unlock:
397 				unlock_page(page);
398 				continue;
399 			}
400 			if (!PageDirty(page)) {
401 				/* someone wrote it for us */
402 				goto continue_unlock;
403 			}
404 
405 			f2fs_wait_on_page_writeback(page, META, true, true);
406 
407 			if (!clear_page_dirty_for_io(page))
408 				goto continue_unlock;
409 
410 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
411 				unlock_page(page);
412 				break;
413 			}
414 			nwritten++;
415 			prev = page->index;
416 			if (unlikely(nwritten >= nr_to_write))
417 				break;
418 		}
419 		pagevec_release(&pvec);
420 		cond_resched();
421 	}
422 stop:
423 	if (nwritten)
424 		f2fs_submit_merged_write(sbi, type);
425 
426 	blk_finish_plug(&plug);
427 
428 	return nwritten;
429 }
430 
431 static int f2fs_set_meta_page_dirty(struct page *page)
432 {
433 	trace_f2fs_set_page_dirty(page, META);
434 
435 	if (!PageUptodate(page))
436 		SetPageUptodate(page);
437 	if (!PageDirty(page)) {
438 		__set_page_dirty_nobuffers(page);
439 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
440 		f2fs_set_page_private(page, 0);
441 		f2fs_trace_pid(page);
442 		return 1;
443 	}
444 	return 0;
445 }
446 
447 const struct address_space_operations f2fs_meta_aops = {
448 	.writepage	= f2fs_write_meta_page,
449 	.writepages	= f2fs_write_meta_pages,
450 	.set_page_dirty	= f2fs_set_meta_page_dirty,
451 	.invalidatepage = f2fs_invalidate_page,
452 	.releasepage	= f2fs_release_page,
453 #ifdef CONFIG_MIGRATION
454 	.migratepage    = f2fs_migrate_page,
455 #endif
456 };
457 
458 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
459 						unsigned int devidx, int type)
460 {
461 	struct inode_management *im = &sbi->im[type];
462 	struct ino_entry *e, *tmp;
463 
464 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
465 
466 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
467 
468 	spin_lock(&im->ino_lock);
469 	e = radix_tree_lookup(&im->ino_root, ino);
470 	if (!e) {
471 		e = tmp;
472 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
473 			f2fs_bug_on(sbi, 1);
474 
475 		memset(e, 0, sizeof(struct ino_entry));
476 		e->ino = ino;
477 
478 		list_add_tail(&e->list, &im->ino_list);
479 		if (type != ORPHAN_INO)
480 			im->ino_num++;
481 	}
482 
483 	if (type == FLUSH_INO)
484 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
485 
486 	spin_unlock(&im->ino_lock);
487 	radix_tree_preload_end();
488 
489 	if (e != tmp)
490 		kmem_cache_free(ino_entry_slab, tmp);
491 }
492 
493 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
494 {
495 	struct inode_management *im = &sbi->im[type];
496 	struct ino_entry *e;
497 
498 	spin_lock(&im->ino_lock);
499 	e = radix_tree_lookup(&im->ino_root, ino);
500 	if (e) {
501 		list_del(&e->list);
502 		radix_tree_delete(&im->ino_root, ino);
503 		im->ino_num--;
504 		spin_unlock(&im->ino_lock);
505 		kmem_cache_free(ino_entry_slab, e);
506 		return;
507 	}
508 	spin_unlock(&im->ino_lock);
509 }
510 
511 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
512 {
513 	/* add new dirty ino entry into list */
514 	__add_ino_entry(sbi, ino, 0, type);
515 }
516 
517 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
518 {
519 	/* remove dirty ino entry from list */
520 	__remove_ino_entry(sbi, ino, type);
521 }
522 
523 /* mode should be APPEND_INO or UPDATE_INO */
524 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
525 {
526 	struct inode_management *im = &sbi->im[mode];
527 	struct ino_entry *e;
528 
529 	spin_lock(&im->ino_lock);
530 	e = radix_tree_lookup(&im->ino_root, ino);
531 	spin_unlock(&im->ino_lock);
532 	return e ? true : false;
533 }
534 
535 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
536 {
537 	struct ino_entry *e, *tmp;
538 	int i;
539 
540 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
541 		struct inode_management *im = &sbi->im[i];
542 
543 		spin_lock(&im->ino_lock);
544 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
545 			list_del(&e->list);
546 			radix_tree_delete(&im->ino_root, e->ino);
547 			kmem_cache_free(ino_entry_slab, e);
548 			im->ino_num--;
549 		}
550 		spin_unlock(&im->ino_lock);
551 	}
552 }
553 
554 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
555 					unsigned int devidx, int type)
556 {
557 	__add_ino_entry(sbi, ino, devidx, type);
558 }
559 
560 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
561 					unsigned int devidx, int type)
562 {
563 	struct inode_management *im = &sbi->im[type];
564 	struct ino_entry *e;
565 	bool is_dirty = false;
566 
567 	spin_lock(&im->ino_lock);
568 	e = radix_tree_lookup(&im->ino_root, ino);
569 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
570 		is_dirty = true;
571 	spin_unlock(&im->ino_lock);
572 	return is_dirty;
573 }
574 
575 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
576 {
577 	struct inode_management *im = &sbi->im[ORPHAN_INO];
578 	int err = 0;
579 
580 	spin_lock(&im->ino_lock);
581 
582 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
583 		spin_unlock(&im->ino_lock);
584 		f2fs_show_injection_info(sbi, FAULT_ORPHAN);
585 		return -ENOSPC;
586 	}
587 
588 	if (unlikely(im->ino_num >= sbi->max_orphans))
589 		err = -ENOSPC;
590 	else
591 		im->ino_num++;
592 	spin_unlock(&im->ino_lock);
593 
594 	return err;
595 }
596 
597 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
598 {
599 	struct inode_management *im = &sbi->im[ORPHAN_INO];
600 
601 	spin_lock(&im->ino_lock);
602 	f2fs_bug_on(sbi, im->ino_num == 0);
603 	im->ino_num--;
604 	spin_unlock(&im->ino_lock);
605 }
606 
607 void f2fs_add_orphan_inode(struct inode *inode)
608 {
609 	/* add new orphan ino entry into list */
610 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
611 	f2fs_update_inode_page(inode);
612 }
613 
614 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
615 {
616 	/* remove orphan entry from orphan list */
617 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
618 }
619 
620 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
621 {
622 	struct inode *inode;
623 	struct node_info ni;
624 	int err;
625 
626 	inode = f2fs_iget_retry(sbi->sb, ino);
627 	if (IS_ERR(inode)) {
628 		/*
629 		 * there should be a bug that we can't find the entry
630 		 * to orphan inode.
631 		 */
632 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
633 		return PTR_ERR(inode);
634 	}
635 
636 	err = dquot_initialize(inode);
637 	if (err) {
638 		iput(inode);
639 		goto err_out;
640 	}
641 
642 	clear_nlink(inode);
643 
644 	/* truncate all the data during iput */
645 	iput(inode);
646 
647 	err = f2fs_get_node_info(sbi, ino, &ni);
648 	if (err)
649 		goto err_out;
650 
651 	/* ENOMEM was fully retried in f2fs_evict_inode. */
652 	if (ni.blk_addr != NULL_ADDR) {
653 		err = -EIO;
654 		goto err_out;
655 	}
656 	return 0;
657 
658 err_out:
659 	set_sbi_flag(sbi, SBI_NEED_FSCK);
660 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
661 		  __func__, ino);
662 	return err;
663 }
664 
665 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
666 {
667 	block_t start_blk, orphan_blocks, i, j;
668 	unsigned int s_flags = sbi->sb->s_flags;
669 	int err = 0;
670 #ifdef CONFIG_QUOTA
671 	int quota_enabled;
672 #endif
673 
674 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
675 		return 0;
676 
677 	if (bdev_read_only(sbi->sb->s_bdev)) {
678 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
679 		return 0;
680 	}
681 
682 	if (s_flags & SB_RDONLY) {
683 		f2fs_info(sbi, "orphan cleanup on readonly fs");
684 		sbi->sb->s_flags &= ~SB_RDONLY;
685 	}
686 
687 #ifdef CONFIG_QUOTA
688 	/* Needed for iput() to work correctly and not trash data */
689 	sbi->sb->s_flags |= SB_ACTIVE;
690 
691 	/*
692 	 * Turn on quotas which were not enabled for read-only mounts if
693 	 * filesystem has quota feature, so that they are updated correctly.
694 	 */
695 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
696 #endif
697 
698 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
699 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
700 
701 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
702 
703 	for (i = 0; i < orphan_blocks; i++) {
704 		struct page *page;
705 		struct f2fs_orphan_block *orphan_blk;
706 
707 		page = f2fs_get_meta_page(sbi, start_blk + i);
708 		if (IS_ERR(page)) {
709 			err = PTR_ERR(page);
710 			goto out;
711 		}
712 
713 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
714 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
715 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
716 			err = recover_orphan_inode(sbi, ino);
717 			if (err) {
718 				f2fs_put_page(page, 1);
719 				goto out;
720 			}
721 		}
722 		f2fs_put_page(page, 1);
723 	}
724 	/* clear Orphan Flag */
725 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
726 out:
727 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
728 
729 #ifdef CONFIG_QUOTA
730 	/* Turn quotas off */
731 	if (quota_enabled)
732 		f2fs_quota_off_umount(sbi->sb);
733 #endif
734 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
735 
736 	return err;
737 }
738 
739 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
740 {
741 	struct list_head *head;
742 	struct f2fs_orphan_block *orphan_blk = NULL;
743 	unsigned int nentries = 0;
744 	unsigned short index = 1;
745 	unsigned short orphan_blocks;
746 	struct page *page = NULL;
747 	struct ino_entry *orphan = NULL;
748 	struct inode_management *im = &sbi->im[ORPHAN_INO];
749 
750 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
751 
752 	/*
753 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
754 	 * orphan inode operations are covered under f2fs_lock_op().
755 	 * And, spin_lock should be avoided due to page operations below.
756 	 */
757 	head = &im->ino_list;
758 
759 	/* loop for each orphan inode entry and write them in Jornal block */
760 	list_for_each_entry(orphan, head, list) {
761 		if (!page) {
762 			page = f2fs_grab_meta_page(sbi, start_blk++);
763 			orphan_blk =
764 				(struct f2fs_orphan_block *)page_address(page);
765 			memset(orphan_blk, 0, sizeof(*orphan_blk));
766 		}
767 
768 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
769 
770 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
771 			/*
772 			 * an orphan block is full of 1020 entries,
773 			 * then we need to flush current orphan blocks
774 			 * and bring another one in memory
775 			 */
776 			orphan_blk->blk_addr = cpu_to_le16(index);
777 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
778 			orphan_blk->entry_count = cpu_to_le32(nentries);
779 			set_page_dirty(page);
780 			f2fs_put_page(page, 1);
781 			index++;
782 			nentries = 0;
783 			page = NULL;
784 		}
785 	}
786 
787 	if (page) {
788 		orphan_blk->blk_addr = cpu_to_le16(index);
789 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
790 		orphan_blk->entry_count = cpu_to_le32(nentries);
791 		set_page_dirty(page);
792 		f2fs_put_page(page, 1);
793 	}
794 }
795 
796 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
797 						struct f2fs_checkpoint *ckpt)
798 {
799 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
800 	__u32 chksum;
801 
802 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
803 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
804 		chksum_ofs += sizeof(chksum);
805 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
806 						F2FS_BLKSIZE - chksum_ofs);
807 	}
808 	return chksum;
809 }
810 
811 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
812 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
813 		unsigned long long *version)
814 {
815 	size_t crc_offset = 0;
816 	__u32 crc;
817 
818 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
819 	if (IS_ERR(*cp_page))
820 		return PTR_ERR(*cp_page);
821 
822 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
823 
824 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
825 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
826 			crc_offset > CP_CHKSUM_OFFSET) {
827 		f2fs_put_page(*cp_page, 1);
828 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
829 		return -EINVAL;
830 	}
831 
832 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
833 	if (crc != cur_cp_crc(*cp_block)) {
834 		f2fs_put_page(*cp_page, 1);
835 		f2fs_warn(sbi, "invalid crc value");
836 		return -EINVAL;
837 	}
838 
839 	*version = cur_cp_version(*cp_block);
840 	return 0;
841 }
842 
843 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
844 				block_t cp_addr, unsigned long long *version)
845 {
846 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
847 	struct f2fs_checkpoint *cp_block = NULL;
848 	unsigned long long cur_version = 0, pre_version = 0;
849 	int err;
850 
851 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
852 					&cp_page_1, version);
853 	if (err)
854 		return NULL;
855 
856 	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
857 					sbi->blocks_per_seg) {
858 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
859 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
860 		goto invalid_cp;
861 	}
862 	pre_version = *version;
863 
864 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
865 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
866 					&cp_page_2, version);
867 	if (err)
868 		goto invalid_cp;
869 	cur_version = *version;
870 
871 	if (cur_version == pre_version) {
872 		*version = cur_version;
873 		f2fs_put_page(cp_page_2, 1);
874 		return cp_page_1;
875 	}
876 	f2fs_put_page(cp_page_2, 1);
877 invalid_cp:
878 	f2fs_put_page(cp_page_1, 1);
879 	return NULL;
880 }
881 
882 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
883 {
884 	struct f2fs_checkpoint *cp_block;
885 	struct f2fs_super_block *fsb = sbi->raw_super;
886 	struct page *cp1, *cp2, *cur_page;
887 	unsigned long blk_size = sbi->blocksize;
888 	unsigned long long cp1_version = 0, cp2_version = 0;
889 	unsigned long long cp_start_blk_no;
890 	unsigned int cp_blks = 1 + __cp_payload(sbi);
891 	block_t cp_blk_no;
892 	int i;
893 	int err;
894 
895 	sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
896 				 GFP_KERNEL);
897 	if (!sbi->ckpt)
898 		return -ENOMEM;
899 	/*
900 	 * Finding out valid cp block involves read both
901 	 * sets( cp pack1 and cp pack 2)
902 	 */
903 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
904 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
905 
906 	/* The second checkpoint pack should start at the next segment */
907 	cp_start_blk_no += ((unsigned long long)1) <<
908 				le32_to_cpu(fsb->log_blocks_per_seg);
909 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
910 
911 	if (cp1 && cp2) {
912 		if (ver_after(cp2_version, cp1_version))
913 			cur_page = cp2;
914 		else
915 			cur_page = cp1;
916 	} else if (cp1) {
917 		cur_page = cp1;
918 	} else if (cp2) {
919 		cur_page = cp2;
920 	} else {
921 		err = -EFSCORRUPTED;
922 		goto fail_no_cp;
923 	}
924 
925 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
926 	memcpy(sbi->ckpt, cp_block, blk_size);
927 
928 	if (cur_page == cp1)
929 		sbi->cur_cp_pack = 1;
930 	else
931 		sbi->cur_cp_pack = 2;
932 
933 	/* Sanity checking of checkpoint */
934 	if (f2fs_sanity_check_ckpt(sbi)) {
935 		err = -EFSCORRUPTED;
936 		goto free_fail_no_cp;
937 	}
938 
939 	if (cp_blks <= 1)
940 		goto done;
941 
942 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
943 	if (cur_page == cp2)
944 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
945 
946 	for (i = 1; i < cp_blks; i++) {
947 		void *sit_bitmap_ptr;
948 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
949 
950 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
951 		if (IS_ERR(cur_page)) {
952 			err = PTR_ERR(cur_page);
953 			goto free_fail_no_cp;
954 		}
955 		sit_bitmap_ptr = page_address(cur_page);
956 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
957 		f2fs_put_page(cur_page, 1);
958 	}
959 done:
960 	f2fs_put_page(cp1, 1);
961 	f2fs_put_page(cp2, 1);
962 	return 0;
963 
964 free_fail_no_cp:
965 	f2fs_put_page(cp1, 1);
966 	f2fs_put_page(cp2, 1);
967 fail_no_cp:
968 	kvfree(sbi->ckpt);
969 	return err;
970 }
971 
972 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
973 {
974 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
975 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
976 
977 	if (is_inode_flag_set(inode, flag))
978 		return;
979 
980 	set_inode_flag(inode, flag);
981 	if (!f2fs_is_volatile_file(inode))
982 		list_add_tail(&F2FS_I(inode)->dirty_list,
983 						&sbi->inode_list[type]);
984 	stat_inc_dirty_inode(sbi, type);
985 }
986 
987 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
988 {
989 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
990 
991 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
992 		return;
993 
994 	list_del_init(&F2FS_I(inode)->dirty_list);
995 	clear_inode_flag(inode, flag);
996 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
997 }
998 
999 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1000 {
1001 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1002 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1003 
1004 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1005 			!S_ISLNK(inode->i_mode))
1006 		return;
1007 
1008 	spin_lock(&sbi->inode_lock[type]);
1009 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1010 		__add_dirty_inode(inode, type);
1011 	inode_inc_dirty_pages(inode);
1012 	spin_unlock(&sbi->inode_lock[type]);
1013 
1014 	f2fs_set_page_private(page, 0);
1015 	f2fs_trace_pid(page);
1016 }
1017 
1018 void f2fs_remove_dirty_inode(struct inode *inode)
1019 {
1020 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1021 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1022 
1023 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1024 			!S_ISLNK(inode->i_mode))
1025 		return;
1026 
1027 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1028 		return;
1029 
1030 	spin_lock(&sbi->inode_lock[type]);
1031 	__remove_dirty_inode(inode, type);
1032 	spin_unlock(&sbi->inode_lock[type]);
1033 }
1034 
1035 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1036 {
1037 	struct list_head *head;
1038 	struct inode *inode;
1039 	struct f2fs_inode_info *fi;
1040 	bool is_dir = (type == DIR_INODE);
1041 	unsigned long ino = 0;
1042 
1043 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1044 				get_pages(sbi, is_dir ?
1045 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1046 retry:
1047 	if (unlikely(f2fs_cp_error(sbi)))
1048 		return -EIO;
1049 
1050 	spin_lock(&sbi->inode_lock[type]);
1051 
1052 	head = &sbi->inode_list[type];
1053 	if (list_empty(head)) {
1054 		spin_unlock(&sbi->inode_lock[type]);
1055 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1056 				get_pages(sbi, is_dir ?
1057 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1058 		return 0;
1059 	}
1060 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1061 	inode = igrab(&fi->vfs_inode);
1062 	spin_unlock(&sbi->inode_lock[type]);
1063 	if (inode) {
1064 		unsigned long cur_ino = inode->i_ino;
1065 
1066 		F2FS_I(inode)->cp_task = current;
1067 
1068 		filemap_fdatawrite(inode->i_mapping);
1069 
1070 		F2FS_I(inode)->cp_task = NULL;
1071 
1072 		iput(inode);
1073 		/* We need to give cpu to another writers. */
1074 		if (ino == cur_ino)
1075 			cond_resched();
1076 		else
1077 			ino = cur_ino;
1078 	} else {
1079 		/*
1080 		 * We should submit bio, since it exists several
1081 		 * wribacking dentry pages in the freeing inode.
1082 		 */
1083 		f2fs_submit_merged_write(sbi, DATA);
1084 		cond_resched();
1085 	}
1086 	goto retry;
1087 }
1088 
1089 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1090 {
1091 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1092 	struct inode *inode;
1093 	struct f2fs_inode_info *fi;
1094 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1095 
1096 	while (total--) {
1097 		if (unlikely(f2fs_cp_error(sbi)))
1098 			return -EIO;
1099 
1100 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1101 		if (list_empty(head)) {
1102 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1103 			return 0;
1104 		}
1105 		fi = list_first_entry(head, struct f2fs_inode_info,
1106 							gdirty_list);
1107 		inode = igrab(&fi->vfs_inode);
1108 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1109 		if (inode) {
1110 			sync_inode_metadata(inode, 0);
1111 
1112 			/* it's on eviction */
1113 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1114 				f2fs_update_inode_page(inode);
1115 			iput(inode);
1116 		}
1117 	}
1118 	return 0;
1119 }
1120 
1121 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1122 {
1123 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1124 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1125 	nid_t last_nid = nm_i->next_scan_nid;
1126 
1127 	next_free_nid(sbi, &last_nid);
1128 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1129 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1130 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1131 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1132 }
1133 
1134 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1135 {
1136 	bool ret = false;
1137 
1138 	if (!is_journalled_quota(sbi))
1139 		return false;
1140 
1141 	down_write(&sbi->quota_sem);
1142 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1143 		ret = false;
1144 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1145 		ret = false;
1146 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1147 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1148 		ret = true;
1149 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1150 		ret = true;
1151 	}
1152 	up_write(&sbi->quota_sem);
1153 	return ret;
1154 }
1155 
1156 /*
1157  * Freeze all the FS-operations for checkpoint.
1158  */
1159 static int block_operations(struct f2fs_sb_info *sbi)
1160 {
1161 	struct writeback_control wbc = {
1162 		.sync_mode = WB_SYNC_ALL,
1163 		.nr_to_write = LONG_MAX,
1164 		.for_reclaim = 0,
1165 	};
1166 	struct blk_plug plug;
1167 	int err = 0, cnt = 0;
1168 
1169 	blk_start_plug(&plug);
1170 
1171 retry_flush_quotas:
1172 	f2fs_lock_all(sbi);
1173 	if (__need_flush_quota(sbi)) {
1174 		int locked;
1175 
1176 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1177 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1178 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1179 			goto retry_flush_dents;
1180 		}
1181 		f2fs_unlock_all(sbi);
1182 
1183 		/* only failed during mount/umount/freeze/quotactl */
1184 		locked = down_read_trylock(&sbi->sb->s_umount);
1185 		f2fs_quota_sync(sbi->sb, -1);
1186 		if (locked)
1187 			up_read(&sbi->sb->s_umount);
1188 		cond_resched();
1189 		goto retry_flush_quotas;
1190 	}
1191 
1192 retry_flush_dents:
1193 	/* write all the dirty dentry pages */
1194 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1195 		f2fs_unlock_all(sbi);
1196 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1197 		if (err)
1198 			goto out;
1199 		cond_resched();
1200 		goto retry_flush_quotas;
1201 	}
1202 
1203 	/*
1204 	 * POR: we should ensure that there are no dirty node pages
1205 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1206 	 */
1207 	down_write(&sbi->node_change);
1208 
1209 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1210 		up_write(&sbi->node_change);
1211 		f2fs_unlock_all(sbi);
1212 		err = f2fs_sync_inode_meta(sbi);
1213 		if (err)
1214 			goto out;
1215 		cond_resched();
1216 		goto retry_flush_quotas;
1217 	}
1218 
1219 retry_flush_nodes:
1220 	down_write(&sbi->node_write);
1221 
1222 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1223 		up_write(&sbi->node_write);
1224 		atomic_inc(&sbi->wb_sync_req[NODE]);
1225 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1226 		atomic_dec(&sbi->wb_sync_req[NODE]);
1227 		if (err) {
1228 			up_write(&sbi->node_change);
1229 			f2fs_unlock_all(sbi);
1230 			goto out;
1231 		}
1232 		cond_resched();
1233 		goto retry_flush_nodes;
1234 	}
1235 
1236 	/*
1237 	 * sbi->node_change is used only for AIO write_begin path which produces
1238 	 * dirty node blocks and some checkpoint values by block allocation.
1239 	 */
1240 	__prepare_cp_block(sbi);
1241 	up_write(&sbi->node_change);
1242 out:
1243 	blk_finish_plug(&plug);
1244 	return err;
1245 }
1246 
1247 static void unblock_operations(struct f2fs_sb_info *sbi)
1248 {
1249 	up_write(&sbi->node_write);
1250 	f2fs_unlock_all(sbi);
1251 }
1252 
1253 void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1254 {
1255 	DEFINE_WAIT(wait);
1256 
1257 	for (;;) {
1258 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1259 
1260 		if (!get_pages(sbi, F2FS_WB_CP_DATA))
1261 			break;
1262 
1263 		if (unlikely(f2fs_cp_error(sbi)))
1264 			break;
1265 
1266 		io_schedule_timeout(5*HZ);
1267 	}
1268 	finish_wait(&sbi->cp_wait, &wait);
1269 }
1270 
1271 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1272 {
1273 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1274 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1275 	unsigned long flags;
1276 
1277 	spin_lock_irqsave(&sbi->cp_lock, flags);
1278 
1279 	if ((cpc->reason & CP_UMOUNT) &&
1280 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1281 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1282 		disable_nat_bits(sbi, false);
1283 
1284 	if (cpc->reason & CP_TRIMMED)
1285 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1286 	else
1287 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1288 
1289 	if (cpc->reason & CP_UMOUNT)
1290 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1291 	else
1292 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1293 
1294 	if (cpc->reason & CP_FASTBOOT)
1295 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1296 	else
1297 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1298 
1299 	if (orphan_num)
1300 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1301 	else
1302 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1303 
1304 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1305 		is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1306 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1307 
1308 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1309 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1310 	else
1311 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1312 
1313 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1314 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1315 	else
1316 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1317 
1318 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1319 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1320 	else
1321 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1322 
1323 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1324 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1325 
1326 	/* set this flag to activate crc|cp_ver for recovery */
1327 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1328 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1329 
1330 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1331 }
1332 
1333 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1334 	void *src, block_t blk_addr)
1335 {
1336 	struct writeback_control wbc = {
1337 		.for_reclaim = 0,
1338 	};
1339 
1340 	/*
1341 	 * pagevec_lookup_tag and lock_page again will take
1342 	 * some extra time. Therefore, f2fs_update_meta_pages and
1343 	 * f2fs_sync_meta_pages are combined in this function.
1344 	 */
1345 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1346 	int err;
1347 
1348 	f2fs_wait_on_page_writeback(page, META, true, true);
1349 
1350 	memcpy(page_address(page), src, PAGE_SIZE);
1351 
1352 	set_page_dirty(page);
1353 	if (unlikely(!clear_page_dirty_for_io(page)))
1354 		f2fs_bug_on(sbi, 1);
1355 
1356 	/* writeout cp pack 2 page */
1357 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1358 	if (unlikely(err && f2fs_cp_error(sbi))) {
1359 		f2fs_put_page(page, 1);
1360 		return;
1361 	}
1362 
1363 	f2fs_bug_on(sbi, err);
1364 	f2fs_put_page(page, 0);
1365 
1366 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1367 	f2fs_submit_merged_write(sbi, META_FLUSH);
1368 }
1369 
1370 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1371 {
1372 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1373 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1374 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1375 	block_t start_blk;
1376 	unsigned int data_sum_blocks, orphan_blocks;
1377 	__u32 crc32 = 0;
1378 	int i;
1379 	int cp_payload_blks = __cp_payload(sbi);
1380 	struct super_block *sb = sbi->sb;
1381 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1382 	u64 kbytes_written;
1383 	int err;
1384 
1385 	/* Flush all the NAT/SIT pages */
1386 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1387 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
1388 					!f2fs_cp_error(sbi));
1389 
1390 	/*
1391 	 * modify checkpoint
1392 	 * version number is already updated
1393 	 */
1394 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1395 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1396 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1397 		ckpt->cur_node_segno[i] =
1398 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1399 		ckpt->cur_node_blkoff[i] =
1400 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1401 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1402 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1403 	}
1404 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1405 		ckpt->cur_data_segno[i] =
1406 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1407 		ckpt->cur_data_blkoff[i] =
1408 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1409 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1410 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1411 	}
1412 
1413 	/* 2 cp  + n data seg summary + orphan inode blocks */
1414 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1415 	spin_lock_irqsave(&sbi->cp_lock, flags);
1416 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1417 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1418 	else
1419 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1420 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1421 
1422 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1423 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1424 			orphan_blocks);
1425 
1426 	if (__remain_node_summaries(cpc->reason))
1427 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1428 				cp_payload_blks + data_sum_blocks +
1429 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1430 	else
1431 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1432 				cp_payload_blks + data_sum_blocks +
1433 				orphan_blocks);
1434 
1435 	/* update ckpt flag for checkpoint */
1436 	update_ckpt_flags(sbi, cpc);
1437 
1438 	/* update SIT/NAT bitmap */
1439 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1440 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1441 
1442 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1443 	*((__le32 *)((unsigned char *)ckpt +
1444 				le32_to_cpu(ckpt->checksum_offset)))
1445 				= cpu_to_le32(crc32);
1446 
1447 	start_blk = __start_cp_next_addr(sbi);
1448 
1449 	/* write nat bits */
1450 	if (enabled_nat_bits(sbi, cpc)) {
1451 		__u64 cp_ver = cur_cp_version(ckpt);
1452 		block_t blk;
1453 
1454 		cp_ver |= ((__u64)crc32 << 32);
1455 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1456 
1457 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1458 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1459 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1460 					(i << F2FS_BLKSIZE_BITS), blk + i);
1461 	}
1462 
1463 	/* write out checkpoint buffer at block 0 */
1464 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1465 
1466 	for (i = 1; i < 1 + cp_payload_blks; i++)
1467 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1468 							start_blk++);
1469 
1470 	if (orphan_num) {
1471 		write_orphan_inodes(sbi, start_blk);
1472 		start_blk += orphan_blocks;
1473 	}
1474 
1475 	f2fs_write_data_summaries(sbi, start_blk);
1476 	start_blk += data_sum_blocks;
1477 
1478 	/* Record write statistics in the hot node summary */
1479 	kbytes_written = sbi->kbytes_written;
1480 	if (sb->s_bdev->bd_part)
1481 		kbytes_written += BD_PART_WRITTEN(sbi);
1482 
1483 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1484 
1485 	if (__remain_node_summaries(cpc->reason)) {
1486 		f2fs_write_node_summaries(sbi, start_blk);
1487 		start_blk += NR_CURSEG_NODE_TYPE;
1488 	}
1489 
1490 	/* update user_block_counts */
1491 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1492 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1493 
1494 	/* Here, we have one bio having CP pack except cp pack 2 page */
1495 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1496 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
1497 					!f2fs_cp_error(sbi));
1498 
1499 	/* wait for previous submitted meta pages writeback */
1500 	f2fs_wait_on_all_pages_writeback(sbi);
1501 
1502 	/* flush all device cache */
1503 	err = f2fs_flush_device_cache(sbi);
1504 	if (err)
1505 		return err;
1506 
1507 	/* barrier and flush checkpoint cp pack 2 page if it can */
1508 	commit_checkpoint(sbi, ckpt, start_blk);
1509 	f2fs_wait_on_all_pages_writeback(sbi);
1510 
1511 	/*
1512 	 * invalidate intermediate page cache borrowed from meta inode
1513 	 * which are used for migration of encrypted inode's blocks.
1514 	 */
1515 	if (f2fs_sb_has_encrypt(sbi))
1516 		invalidate_mapping_pages(META_MAPPING(sbi),
1517 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1518 
1519 	f2fs_release_ino_entry(sbi, false);
1520 
1521 	f2fs_reset_fsync_node_info(sbi);
1522 
1523 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1524 	clear_sbi_flag(sbi, SBI_NEED_CP);
1525 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1526 
1527 	spin_lock(&sbi->stat_lock);
1528 	sbi->unusable_block_count = 0;
1529 	spin_unlock(&sbi->stat_lock);
1530 
1531 	__set_cp_next_pack(sbi);
1532 
1533 	/*
1534 	 * redirty superblock if metadata like node page or inode cache is
1535 	 * updated during writing checkpoint.
1536 	 */
1537 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1538 			get_pages(sbi, F2FS_DIRTY_IMETA))
1539 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1540 
1541 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1542 
1543 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1544 }
1545 
1546 /*
1547  * We guarantee that this checkpoint procedure will not fail.
1548  */
1549 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1550 {
1551 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1552 	unsigned long long ckpt_ver;
1553 	int err = 0;
1554 
1555 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1556 		return -EROFS;
1557 
1558 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1559 		if (cpc->reason != CP_PAUSE)
1560 			return 0;
1561 		f2fs_warn(sbi, "Start checkpoint disabled!");
1562 	}
1563 	mutex_lock(&sbi->cp_mutex);
1564 
1565 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1566 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1567 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1568 		goto out;
1569 	if (unlikely(f2fs_cp_error(sbi))) {
1570 		err = -EIO;
1571 		goto out;
1572 	}
1573 
1574 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1575 
1576 	err = block_operations(sbi);
1577 	if (err)
1578 		goto out;
1579 
1580 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1581 
1582 	f2fs_flush_merged_writes(sbi);
1583 
1584 	/* this is the case of multiple fstrims without any changes */
1585 	if (cpc->reason & CP_DISCARD) {
1586 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1587 			unblock_operations(sbi);
1588 			goto out;
1589 		}
1590 
1591 		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1592 				SIT_I(sbi)->dirty_sentries == 0 &&
1593 				prefree_segments(sbi) == 0) {
1594 			f2fs_flush_sit_entries(sbi, cpc);
1595 			f2fs_clear_prefree_segments(sbi, cpc);
1596 			unblock_operations(sbi);
1597 			goto out;
1598 		}
1599 	}
1600 
1601 	/*
1602 	 * update checkpoint pack index
1603 	 * Increase the version number so that
1604 	 * SIT entries and seg summaries are written at correct place
1605 	 */
1606 	ckpt_ver = cur_cp_version(ckpt);
1607 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1608 
1609 	/* write cached NAT/SIT entries to NAT/SIT area */
1610 	err = f2fs_flush_nat_entries(sbi, cpc);
1611 	if (err)
1612 		goto stop;
1613 
1614 	f2fs_flush_sit_entries(sbi, cpc);
1615 
1616 	/* unlock all the fs_lock[] in do_checkpoint() */
1617 	err = do_checkpoint(sbi, cpc);
1618 	if (err)
1619 		f2fs_release_discard_addrs(sbi);
1620 	else
1621 		f2fs_clear_prefree_segments(sbi, cpc);
1622 stop:
1623 	unblock_operations(sbi);
1624 	stat_inc_cp_count(sbi->stat_info);
1625 
1626 	if (cpc->reason & CP_RECOVERY)
1627 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1628 
1629 	/* do checkpoint periodically */
1630 	f2fs_update_time(sbi, CP_TIME);
1631 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1632 out:
1633 	mutex_unlock(&sbi->cp_mutex);
1634 	return err;
1635 }
1636 
1637 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1638 {
1639 	int i;
1640 
1641 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1642 		struct inode_management *im = &sbi->im[i];
1643 
1644 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1645 		spin_lock_init(&im->ino_lock);
1646 		INIT_LIST_HEAD(&im->ino_list);
1647 		im->ino_num = 0;
1648 	}
1649 
1650 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1651 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1652 				F2FS_ORPHANS_PER_BLOCK;
1653 }
1654 
1655 int __init f2fs_create_checkpoint_caches(void)
1656 {
1657 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1658 			sizeof(struct ino_entry));
1659 	if (!ino_entry_slab)
1660 		return -ENOMEM;
1661 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1662 			sizeof(struct inode_entry));
1663 	if (!f2fs_inode_entry_slab) {
1664 		kmem_cache_destroy(ino_entry_slab);
1665 		return -ENOMEM;
1666 	}
1667 	return 0;
1668 }
1669 
1670 void f2fs_destroy_checkpoint_caches(void)
1671 {
1672 	kmem_cache_destroy(ino_entry_slab);
1673 	kmem_cache_destroy(f2fs_inode_entry_slab);
1674 }
1675