xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision 37744fee)
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 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 	lock_page(page);
90 	if (unlikely(page->mapping != mapping)) {
91 		f2fs_put_page(page, 1);
92 		goto repeat;
93 	}
94 
95 	if (unlikely(!PageUptodate(page))) {
96 		f2fs_put_page(page, 1);
97 		return ERR_PTR(-EIO);
98 	}
99 out:
100 	return page;
101 }
102 
103 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
104 {
105 	return __get_meta_page(sbi, index, true);
106 }
107 
108 struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
109 {
110 	struct page *page;
111 	int count = 0;
112 
113 retry:
114 	page = __get_meta_page(sbi, index, true);
115 	if (IS_ERR(page)) {
116 		if (PTR_ERR(page) == -EIO &&
117 				++count <= DEFAULT_RETRY_IO_COUNT)
118 			goto retry;
119 		f2fs_stop_checkpoint(sbi, false);
120 	}
121 	return page;
122 }
123 
124 /* for POR only */
125 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
126 {
127 	return __get_meta_page(sbi, index, false);
128 }
129 
130 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
131 							int type)
132 {
133 	struct seg_entry *se;
134 	unsigned int segno, offset;
135 	bool exist;
136 
137 	if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
138 		return true;
139 
140 	segno = GET_SEGNO(sbi, blkaddr);
141 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
142 	se = get_seg_entry(sbi, segno);
143 
144 	exist = f2fs_test_bit(offset, se->cur_valid_map);
145 	if (!exist && type == DATA_GENERIC_ENHANCE) {
146 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
147 			 blkaddr, exist);
148 		set_sbi_flag(sbi, SBI_NEED_FSCK);
149 		WARN_ON(1);
150 	}
151 	return exist;
152 }
153 
154 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
155 					block_t blkaddr, int type)
156 {
157 	switch (type) {
158 	case META_NAT:
159 		break;
160 	case META_SIT:
161 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
162 			return false;
163 		break;
164 	case META_SSA:
165 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
166 			blkaddr < SM_I(sbi)->ssa_blkaddr))
167 			return false;
168 		break;
169 	case META_CP:
170 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
171 			blkaddr < __start_cp_addr(sbi)))
172 			return false;
173 		break;
174 	case META_POR:
175 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
176 			blkaddr < MAIN_BLKADDR(sbi)))
177 			return false;
178 		break;
179 	case DATA_GENERIC:
180 	case DATA_GENERIC_ENHANCE:
181 	case DATA_GENERIC_ENHANCE_READ:
182 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
183 				blkaddr < MAIN_BLKADDR(sbi))) {
184 			f2fs_warn(sbi, "access invalid blkaddr:%u",
185 				  blkaddr);
186 			set_sbi_flag(sbi, SBI_NEED_FSCK);
187 			WARN_ON(1);
188 			return false;
189 		} else {
190 			return __is_bitmap_valid(sbi, blkaddr, type);
191 		}
192 		break;
193 	case META_GENERIC:
194 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
195 			blkaddr >= MAIN_BLKADDR(sbi)))
196 			return false;
197 		break;
198 	default:
199 		BUG();
200 	}
201 
202 	return true;
203 }
204 
205 /*
206  * Readahead CP/NAT/SIT/SSA/POR pages
207  */
208 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
209 							int type, bool sync)
210 {
211 	struct page *page;
212 	block_t blkno = start;
213 	struct f2fs_io_info fio = {
214 		.sbi = sbi,
215 		.type = META,
216 		.op = REQ_OP_READ,
217 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
218 		.encrypted_page = NULL,
219 		.in_list = false,
220 		.is_por = (type == META_POR),
221 	};
222 	struct blk_plug plug;
223 
224 	if (unlikely(type == META_POR))
225 		fio.op_flags &= ~REQ_META;
226 
227 	blk_start_plug(&plug);
228 	for (; nrpages-- > 0; blkno++) {
229 
230 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
231 			goto out;
232 
233 		switch (type) {
234 		case META_NAT:
235 			if (unlikely(blkno >=
236 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
237 				blkno = 0;
238 			/* get nat block addr */
239 			fio.new_blkaddr = current_nat_addr(sbi,
240 					blkno * NAT_ENTRY_PER_BLOCK);
241 			break;
242 		case META_SIT:
243 			/* get sit block addr */
244 			fio.new_blkaddr = current_sit_addr(sbi,
245 					blkno * SIT_ENTRY_PER_BLOCK);
246 			break;
247 		case META_SSA:
248 		case META_CP:
249 		case META_POR:
250 			fio.new_blkaddr = blkno;
251 			break;
252 		default:
253 			BUG();
254 		}
255 
256 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
257 						fio.new_blkaddr, false);
258 		if (!page)
259 			continue;
260 		if (PageUptodate(page)) {
261 			f2fs_put_page(page, 1);
262 			continue;
263 		}
264 
265 		fio.page = page;
266 		f2fs_submit_page_bio(&fio);
267 		f2fs_put_page(page, 0);
268 	}
269 out:
270 	blk_finish_plug(&plug);
271 	return blkno - start;
272 }
273 
274 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
275 {
276 	struct page *page;
277 	bool readahead = false;
278 
279 	page = find_get_page(META_MAPPING(sbi), index);
280 	if (!page || !PageUptodate(page))
281 		readahead = true;
282 	f2fs_put_page(page, 0);
283 
284 	if (readahead)
285 		f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
286 }
287 
288 static int __f2fs_write_meta_page(struct page *page,
289 				struct writeback_control *wbc,
290 				enum iostat_type io_type)
291 {
292 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
293 
294 	trace_f2fs_writepage(page, META);
295 
296 	if (unlikely(f2fs_cp_error(sbi)))
297 		goto redirty_out;
298 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
299 		goto redirty_out;
300 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
301 		goto redirty_out;
302 
303 	f2fs_do_write_meta_page(sbi, page, io_type);
304 	dec_page_count(sbi, F2FS_DIRTY_META);
305 
306 	if (wbc->for_reclaim)
307 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
308 
309 	unlock_page(page);
310 
311 	if (unlikely(f2fs_cp_error(sbi)))
312 		f2fs_submit_merged_write(sbi, META);
313 
314 	return 0;
315 
316 redirty_out:
317 	redirty_page_for_writepage(wbc, page);
318 	return AOP_WRITEPAGE_ACTIVATE;
319 }
320 
321 static int f2fs_write_meta_page(struct page *page,
322 				struct writeback_control *wbc)
323 {
324 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
325 }
326 
327 static int f2fs_write_meta_pages(struct address_space *mapping,
328 				struct writeback_control *wbc)
329 {
330 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
331 	long diff, written;
332 
333 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
334 		goto skip_write;
335 
336 	/* collect a number of dirty meta pages and write together */
337 	if (wbc->sync_mode != WB_SYNC_ALL &&
338 			get_pages(sbi, F2FS_DIRTY_META) <
339 					nr_pages_to_skip(sbi, META))
340 		goto skip_write;
341 
342 	/* if locked failed, cp will flush dirty pages instead */
343 	if (!mutex_trylock(&sbi->cp_mutex))
344 		goto skip_write;
345 
346 	trace_f2fs_writepages(mapping->host, wbc, META);
347 	diff = nr_pages_to_write(sbi, META, wbc);
348 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
349 	mutex_unlock(&sbi->cp_mutex);
350 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
351 	return 0;
352 
353 skip_write:
354 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
355 	trace_f2fs_writepages(mapping->host, wbc, META);
356 	return 0;
357 }
358 
359 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
360 				long nr_to_write, enum iostat_type io_type)
361 {
362 	struct address_space *mapping = META_MAPPING(sbi);
363 	pgoff_t index = 0, prev = ULONG_MAX;
364 	struct pagevec pvec;
365 	long nwritten = 0;
366 	int nr_pages;
367 	struct writeback_control wbc = {
368 		.for_reclaim = 0,
369 	};
370 	struct blk_plug plug;
371 
372 	pagevec_init(&pvec);
373 
374 	blk_start_plug(&plug);
375 
376 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
377 				PAGECACHE_TAG_DIRTY))) {
378 		int i;
379 
380 		for (i = 0; i < nr_pages; i++) {
381 			struct page *page = pvec.pages[i];
382 
383 			if (prev == ULONG_MAX)
384 				prev = page->index - 1;
385 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
386 				pagevec_release(&pvec);
387 				goto stop;
388 			}
389 
390 			lock_page(page);
391 
392 			if (unlikely(page->mapping != mapping)) {
393 continue_unlock:
394 				unlock_page(page);
395 				continue;
396 			}
397 			if (!PageDirty(page)) {
398 				/* someone wrote it for us */
399 				goto continue_unlock;
400 			}
401 
402 			f2fs_wait_on_page_writeback(page, META, true, true);
403 
404 			if (!clear_page_dirty_for_io(page))
405 				goto continue_unlock;
406 
407 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
408 				unlock_page(page);
409 				break;
410 			}
411 			nwritten++;
412 			prev = page->index;
413 			if (unlikely(nwritten >= nr_to_write))
414 				break;
415 		}
416 		pagevec_release(&pvec);
417 		cond_resched();
418 	}
419 stop:
420 	if (nwritten)
421 		f2fs_submit_merged_write(sbi, type);
422 
423 	blk_finish_plug(&plug);
424 
425 	return nwritten;
426 }
427 
428 static int f2fs_set_meta_page_dirty(struct page *page)
429 {
430 	trace_f2fs_set_page_dirty(page, META);
431 
432 	if (!PageUptodate(page))
433 		SetPageUptodate(page);
434 	if (!PageDirty(page)) {
435 		__set_page_dirty_nobuffers(page);
436 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
437 		f2fs_set_page_private(page, 0);
438 		f2fs_trace_pid(page);
439 		return 1;
440 	}
441 	return 0;
442 }
443 
444 const struct address_space_operations f2fs_meta_aops = {
445 	.writepage	= f2fs_write_meta_page,
446 	.writepages	= f2fs_write_meta_pages,
447 	.set_page_dirty	= f2fs_set_meta_page_dirty,
448 	.invalidatepage = f2fs_invalidate_page,
449 	.releasepage	= f2fs_release_page,
450 #ifdef CONFIG_MIGRATION
451 	.migratepage    = f2fs_migrate_page,
452 #endif
453 };
454 
455 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
456 						unsigned int devidx, int type)
457 {
458 	struct inode_management *im = &sbi->im[type];
459 	struct ino_entry *e, *tmp;
460 
461 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
462 
463 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
464 
465 	spin_lock(&im->ino_lock);
466 	e = radix_tree_lookup(&im->ino_root, ino);
467 	if (!e) {
468 		e = tmp;
469 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
470 			f2fs_bug_on(sbi, 1);
471 
472 		memset(e, 0, sizeof(struct ino_entry));
473 		e->ino = ino;
474 
475 		list_add_tail(&e->list, &im->ino_list);
476 		if (type != ORPHAN_INO)
477 			im->ino_num++;
478 	}
479 
480 	if (type == FLUSH_INO)
481 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
482 
483 	spin_unlock(&im->ino_lock);
484 	radix_tree_preload_end();
485 
486 	if (e != tmp)
487 		kmem_cache_free(ino_entry_slab, tmp);
488 }
489 
490 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
491 {
492 	struct inode_management *im = &sbi->im[type];
493 	struct ino_entry *e;
494 
495 	spin_lock(&im->ino_lock);
496 	e = radix_tree_lookup(&im->ino_root, ino);
497 	if (e) {
498 		list_del(&e->list);
499 		radix_tree_delete(&im->ino_root, ino);
500 		im->ino_num--;
501 		spin_unlock(&im->ino_lock);
502 		kmem_cache_free(ino_entry_slab, e);
503 		return;
504 	}
505 	spin_unlock(&im->ino_lock);
506 }
507 
508 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
509 {
510 	/* add new dirty ino entry into list */
511 	__add_ino_entry(sbi, ino, 0, type);
512 }
513 
514 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
515 {
516 	/* remove dirty ino entry from list */
517 	__remove_ino_entry(sbi, ino, type);
518 }
519 
520 /* mode should be APPEND_INO or UPDATE_INO */
521 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
522 {
523 	struct inode_management *im = &sbi->im[mode];
524 	struct ino_entry *e;
525 
526 	spin_lock(&im->ino_lock);
527 	e = radix_tree_lookup(&im->ino_root, ino);
528 	spin_unlock(&im->ino_lock);
529 	return e ? true : false;
530 }
531 
532 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
533 {
534 	struct ino_entry *e, *tmp;
535 	int i;
536 
537 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
538 		struct inode_management *im = &sbi->im[i];
539 
540 		spin_lock(&im->ino_lock);
541 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
542 			list_del(&e->list);
543 			radix_tree_delete(&im->ino_root, e->ino);
544 			kmem_cache_free(ino_entry_slab, e);
545 			im->ino_num--;
546 		}
547 		spin_unlock(&im->ino_lock);
548 	}
549 }
550 
551 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
552 					unsigned int devidx, int type)
553 {
554 	__add_ino_entry(sbi, ino, devidx, type);
555 }
556 
557 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
558 					unsigned int devidx, int type)
559 {
560 	struct inode_management *im = &sbi->im[type];
561 	struct ino_entry *e;
562 	bool is_dirty = false;
563 
564 	spin_lock(&im->ino_lock);
565 	e = radix_tree_lookup(&im->ino_root, ino);
566 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
567 		is_dirty = true;
568 	spin_unlock(&im->ino_lock);
569 	return is_dirty;
570 }
571 
572 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
573 {
574 	struct inode_management *im = &sbi->im[ORPHAN_INO];
575 	int err = 0;
576 
577 	spin_lock(&im->ino_lock);
578 
579 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
580 		spin_unlock(&im->ino_lock);
581 		f2fs_show_injection_info(sbi, FAULT_ORPHAN);
582 		return -ENOSPC;
583 	}
584 
585 	if (unlikely(im->ino_num >= sbi->max_orphans))
586 		err = -ENOSPC;
587 	else
588 		im->ino_num++;
589 	spin_unlock(&im->ino_lock);
590 
591 	return err;
592 }
593 
594 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
595 {
596 	struct inode_management *im = &sbi->im[ORPHAN_INO];
597 
598 	spin_lock(&im->ino_lock);
599 	f2fs_bug_on(sbi, im->ino_num == 0);
600 	im->ino_num--;
601 	spin_unlock(&im->ino_lock);
602 }
603 
604 void f2fs_add_orphan_inode(struct inode *inode)
605 {
606 	/* add new orphan ino entry into list */
607 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
608 	f2fs_update_inode_page(inode);
609 }
610 
611 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
612 {
613 	/* remove orphan entry from orphan list */
614 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
615 }
616 
617 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
618 {
619 	struct inode *inode;
620 	struct node_info ni;
621 	int err;
622 
623 	inode = f2fs_iget_retry(sbi->sb, ino);
624 	if (IS_ERR(inode)) {
625 		/*
626 		 * there should be a bug that we can't find the entry
627 		 * to orphan inode.
628 		 */
629 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
630 		return PTR_ERR(inode);
631 	}
632 
633 	err = dquot_initialize(inode);
634 	if (err) {
635 		iput(inode);
636 		goto err_out;
637 	}
638 
639 	clear_nlink(inode);
640 
641 	/* truncate all the data during iput */
642 	iput(inode);
643 
644 	err = f2fs_get_node_info(sbi, ino, &ni);
645 	if (err)
646 		goto err_out;
647 
648 	/* ENOMEM was fully retried in f2fs_evict_inode. */
649 	if (ni.blk_addr != NULL_ADDR) {
650 		err = -EIO;
651 		goto err_out;
652 	}
653 	return 0;
654 
655 err_out:
656 	set_sbi_flag(sbi, SBI_NEED_FSCK);
657 	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
658 		  __func__, ino);
659 	return err;
660 }
661 
662 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
663 {
664 	block_t start_blk, orphan_blocks, i, j;
665 	unsigned int s_flags = sbi->sb->s_flags;
666 	int err = 0;
667 #ifdef CONFIG_QUOTA
668 	int quota_enabled;
669 #endif
670 
671 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
672 		return 0;
673 
674 	if (bdev_read_only(sbi->sb->s_bdev)) {
675 		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
676 		return 0;
677 	}
678 
679 	if (s_flags & SB_RDONLY) {
680 		f2fs_info(sbi, "orphan cleanup on readonly fs");
681 		sbi->sb->s_flags &= ~SB_RDONLY;
682 	}
683 
684 #ifdef CONFIG_QUOTA
685 	/* Needed for iput() to work correctly and not trash data */
686 	sbi->sb->s_flags |= SB_ACTIVE;
687 
688 	/*
689 	 * Turn on quotas which were not enabled for read-only mounts if
690 	 * filesystem has quota feature, so that they are updated correctly.
691 	 */
692 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
693 #endif
694 
695 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
696 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
697 
698 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
699 
700 	for (i = 0; i < orphan_blocks; i++) {
701 		struct page *page;
702 		struct f2fs_orphan_block *orphan_blk;
703 
704 		page = f2fs_get_meta_page(sbi, start_blk + i);
705 		if (IS_ERR(page)) {
706 			err = PTR_ERR(page);
707 			goto out;
708 		}
709 
710 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
711 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
712 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
713 			err = recover_orphan_inode(sbi, ino);
714 			if (err) {
715 				f2fs_put_page(page, 1);
716 				goto out;
717 			}
718 		}
719 		f2fs_put_page(page, 1);
720 	}
721 	/* clear Orphan Flag */
722 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
723 out:
724 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
725 
726 #ifdef CONFIG_QUOTA
727 	/* Turn quotas off */
728 	if (quota_enabled)
729 		f2fs_quota_off_umount(sbi->sb);
730 #endif
731 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
732 
733 	return err;
734 }
735 
736 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
737 {
738 	struct list_head *head;
739 	struct f2fs_orphan_block *orphan_blk = NULL;
740 	unsigned int nentries = 0;
741 	unsigned short index = 1;
742 	unsigned short orphan_blocks;
743 	struct page *page = NULL;
744 	struct ino_entry *orphan = NULL;
745 	struct inode_management *im = &sbi->im[ORPHAN_INO];
746 
747 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
748 
749 	/*
750 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
751 	 * orphan inode operations are covered under f2fs_lock_op().
752 	 * And, spin_lock should be avoided due to page operations below.
753 	 */
754 	head = &im->ino_list;
755 
756 	/* loop for each orphan inode entry and write them in Jornal block */
757 	list_for_each_entry(orphan, head, list) {
758 		if (!page) {
759 			page = f2fs_grab_meta_page(sbi, start_blk++);
760 			orphan_blk =
761 				(struct f2fs_orphan_block *)page_address(page);
762 			memset(orphan_blk, 0, sizeof(*orphan_blk));
763 		}
764 
765 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
766 
767 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
768 			/*
769 			 * an orphan block is full of 1020 entries,
770 			 * then we need to flush current orphan blocks
771 			 * and bring another one in memory
772 			 */
773 			orphan_blk->blk_addr = cpu_to_le16(index);
774 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
775 			orphan_blk->entry_count = cpu_to_le32(nentries);
776 			set_page_dirty(page);
777 			f2fs_put_page(page, 1);
778 			index++;
779 			nentries = 0;
780 			page = NULL;
781 		}
782 	}
783 
784 	if (page) {
785 		orphan_blk->blk_addr = cpu_to_le16(index);
786 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
787 		orphan_blk->entry_count = cpu_to_le32(nentries);
788 		set_page_dirty(page);
789 		f2fs_put_page(page, 1);
790 	}
791 }
792 
793 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
794 						struct f2fs_checkpoint *ckpt)
795 {
796 	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
797 	__u32 chksum;
798 
799 	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
800 	if (chksum_ofs < CP_CHKSUM_OFFSET) {
801 		chksum_ofs += sizeof(chksum);
802 		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
803 						F2FS_BLKSIZE - chksum_ofs);
804 	}
805 	return chksum;
806 }
807 
808 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
809 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
810 		unsigned long long *version)
811 {
812 	size_t crc_offset = 0;
813 	__u32 crc;
814 
815 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
816 	if (IS_ERR(*cp_page))
817 		return PTR_ERR(*cp_page);
818 
819 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
820 
821 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
822 	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
823 			crc_offset > CP_CHKSUM_OFFSET) {
824 		f2fs_put_page(*cp_page, 1);
825 		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
826 		return -EINVAL;
827 	}
828 
829 	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
830 	if (crc != cur_cp_crc(*cp_block)) {
831 		f2fs_put_page(*cp_page, 1);
832 		f2fs_warn(sbi, "invalid crc value");
833 		return -EINVAL;
834 	}
835 
836 	*version = cur_cp_version(*cp_block);
837 	return 0;
838 }
839 
840 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
841 				block_t cp_addr, unsigned long long *version)
842 {
843 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
844 	struct f2fs_checkpoint *cp_block = NULL;
845 	unsigned long long cur_version = 0, pre_version = 0;
846 	int err;
847 
848 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
849 					&cp_page_1, version);
850 	if (err)
851 		return NULL;
852 
853 	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
854 					sbi->blocks_per_seg) {
855 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
856 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
857 		goto invalid_cp;
858 	}
859 	pre_version = *version;
860 
861 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
862 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
863 					&cp_page_2, version);
864 	if (err)
865 		goto invalid_cp;
866 	cur_version = *version;
867 
868 	if (cur_version == pre_version) {
869 		*version = cur_version;
870 		f2fs_put_page(cp_page_2, 1);
871 		return cp_page_1;
872 	}
873 	f2fs_put_page(cp_page_2, 1);
874 invalid_cp:
875 	f2fs_put_page(cp_page_1, 1);
876 	return NULL;
877 }
878 
879 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
880 {
881 	struct f2fs_checkpoint *cp_block;
882 	struct f2fs_super_block *fsb = sbi->raw_super;
883 	struct page *cp1, *cp2, *cur_page;
884 	unsigned long blk_size = sbi->blocksize;
885 	unsigned long long cp1_version = 0, cp2_version = 0;
886 	unsigned long long cp_start_blk_no;
887 	unsigned int cp_blks = 1 + __cp_payload(sbi);
888 	block_t cp_blk_no;
889 	int i;
890 	int err;
891 
892 	sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
893 				 GFP_KERNEL);
894 	if (!sbi->ckpt)
895 		return -ENOMEM;
896 	/*
897 	 * Finding out valid cp block involves read both
898 	 * sets( cp pack 1 and cp pack 2)
899 	 */
900 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
901 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
902 
903 	/* The second checkpoint pack should start at the next segment */
904 	cp_start_blk_no += ((unsigned long long)1) <<
905 				le32_to_cpu(fsb->log_blocks_per_seg);
906 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
907 
908 	if (cp1 && cp2) {
909 		if (ver_after(cp2_version, cp1_version))
910 			cur_page = cp2;
911 		else
912 			cur_page = cp1;
913 	} else if (cp1) {
914 		cur_page = cp1;
915 	} else if (cp2) {
916 		cur_page = cp2;
917 	} else {
918 		err = -EFSCORRUPTED;
919 		goto fail_no_cp;
920 	}
921 
922 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
923 	memcpy(sbi->ckpt, cp_block, blk_size);
924 
925 	if (cur_page == cp1)
926 		sbi->cur_cp_pack = 1;
927 	else
928 		sbi->cur_cp_pack = 2;
929 
930 	/* Sanity checking of checkpoint */
931 	if (f2fs_sanity_check_ckpt(sbi)) {
932 		err = -EFSCORRUPTED;
933 		goto free_fail_no_cp;
934 	}
935 
936 	if (cp_blks <= 1)
937 		goto done;
938 
939 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
940 	if (cur_page == cp2)
941 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
942 
943 	for (i = 1; i < cp_blks; i++) {
944 		void *sit_bitmap_ptr;
945 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
946 
947 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
948 		if (IS_ERR(cur_page)) {
949 			err = PTR_ERR(cur_page);
950 			goto free_fail_no_cp;
951 		}
952 		sit_bitmap_ptr = page_address(cur_page);
953 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
954 		f2fs_put_page(cur_page, 1);
955 	}
956 done:
957 	f2fs_put_page(cp1, 1);
958 	f2fs_put_page(cp2, 1);
959 	return 0;
960 
961 free_fail_no_cp:
962 	f2fs_put_page(cp1, 1);
963 	f2fs_put_page(cp2, 1);
964 fail_no_cp:
965 	kvfree(sbi->ckpt);
966 	return err;
967 }
968 
969 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
970 {
971 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
972 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
973 
974 	if (is_inode_flag_set(inode, flag))
975 		return;
976 
977 	set_inode_flag(inode, flag);
978 	if (!f2fs_is_volatile_file(inode))
979 		list_add_tail(&F2FS_I(inode)->dirty_list,
980 						&sbi->inode_list[type]);
981 	stat_inc_dirty_inode(sbi, type);
982 }
983 
984 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
985 {
986 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
987 
988 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
989 		return;
990 
991 	list_del_init(&F2FS_I(inode)->dirty_list);
992 	clear_inode_flag(inode, flag);
993 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
994 }
995 
996 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
997 {
998 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
999 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1000 
1001 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1002 			!S_ISLNK(inode->i_mode))
1003 		return;
1004 
1005 	spin_lock(&sbi->inode_lock[type]);
1006 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1007 		__add_dirty_inode(inode, type);
1008 	inode_inc_dirty_pages(inode);
1009 	spin_unlock(&sbi->inode_lock[type]);
1010 
1011 	f2fs_set_page_private(page, 0);
1012 	f2fs_trace_pid(page);
1013 }
1014 
1015 void f2fs_remove_dirty_inode(struct inode *inode)
1016 {
1017 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1018 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1019 
1020 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1021 			!S_ISLNK(inode->i_mode))
1022 		return;
1023 
1024 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1025 		return;
1026 
1027 	spin_lock(&sbi->inode_lock[type]);
1028 	__remove_dirty_inode(inode, type);
1029 	spin_unlock(&sbi->inode_lock[type]);
1030 }
1031 
1032 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1033 {
1034 	struct list_head *head;
1035 	struct inode *inode;
1036 	struct f2fs_inode_info *fi;
1037 	bool is_dir = (type == DIR_INODE);
1038 	unsigned long ino = 0;
1039 
1040 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1041 				get_pages(sbi, is_dir ?
1042 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1043 retry:
1044 	if (unlikely(f2fs_cp_error(sbi)))
1045 		return -EIO;
1046 
1047 	spin_lock(&sbi->inode_lock[type]);
1048 
1049 	head = &sbi->inode_list[type];
1050 	if (list_empty(head)) {
1051 		spin_unlock(&sbi->inode_lock[type]);
1052 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1053 				get_pages(sbi, is_dir ?
1054 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1055 		return 0;
1056 	}
1057 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1058 	inode = igrab(&fi->vfs_inode);
1059 	spin_unlock(&sbi->inode_lock[type]);
1060 	if (inode) {
1061 		unsigned long cur_ino = inode->i_ino;
1062 
1063 		F2FS_I(inode)->cp_task = current;
1064 
1065 		filemap_fdatawrite(inode->i_mapping);
1066 
1067 		F2FS_I(inode)->cp_task = NULL;
1068 
1069 		iput(inode);
1070 		/* We need to give cpu to another writers. */
1071 		if (ino == cur_ino)
1072 			cond_resched();
1073 		else
1074 			ino = cur_ino;
1075 	} else {
1076 		/*
1077 		 * We should submit bio, since it exists several
1078 		 * wribacking dentry pages in the freeing inode.
1079 		 */
1080 		f2fs_submit_merged_write(sbi, DATA);
1081 		cond_resched();
1082 	}
1083 	goto retry;
1084 }
1085 
1086 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1087 {
1088 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1089 	struct inode *inode;
1090 	struct f2fs_inode_info *fi;
1091 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1092 
1093 	while (total--) {
1094 		if (unlikely(f2fs_cp_error(sbi)))
1095 			return -EIO;
1096 
1097 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1098 		if (list_empty(head)) {
1099 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1100 			return 0;
1101 		}
1102 		fi = list_first_entry(head, struct f2fs_inode_info,
1103 							gdirty_list);
1104 		inode = igrab(&fi->vfs_inode);
1105 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1106 		if (inode) {
1107 			sync_inode_metadata(inode, 0);
1108 
1109 			/* it's on eviction */
1110 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1111 				f2fs_update_inode_page(inode);
1112 			iput(inode);
1113 		}
1114 	}
1115 	return 0;
1116 }
1117 
1118 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1119 {
1120 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1121 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1122 	nid_t last_nid = nm_i->next_scan_nid;
1123 
1124 	next_free_nid(sbi, &last_nid);
1125 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1126 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1127 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1128 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1129 }
1130 
1131 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1132 {
1133 	bool ret = false;
1134 
1135 	if (!is_journalled_quota(sbi))
1136 		return false;
1137 
1138 	down_write(&sbi->quota_sem);
1139 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1140 		ret = false;
1141 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1142 		ret = false;
1143 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1144 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1145 		ret = true;
1146 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1147 		ret = true;
1148 	}
1149 	up_write(&sbi->quota_sem);
1150 	return ret;
1151 }
1152 
1153 /*
1154  * Freeze all the FS-operations for checkpoint.
1155  */
1156 static int block_operations(struct f2fs_sb_info *sbi)
1157 {
1158 	struct writeback_control wbc = {
1159 		.sync_mode = WB_SYNC_ALL,
1160 		.nr_to_write = LONG_MAX,
1161 		.for_reclaim = 0,
1162 	};
1163 	struct blk_plug plug;
1164 	int err = 0, cnt = 0;
1165 
1166 	blk_start_plug(&plug);
1167 
1168 retry_flush_quotas:
1169 	f2fs_lock_all(sbi);
1170 	if (__need_flush_quota(sbi)) {
1171 		int locked;
1172 
1173 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1174 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1175 			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1176 			goto retry_flush_dents;
1177 		}
1178 		f2fs_unlock_all(sbi);
1179 
1180 		/* only failed during mount/umount/freeze/quotactl */
1181 		locked = down_read_trylock(&sbi->sb->s_umount);
1182 		f2fs_quota_sync(sbi->sb, -1);
1183 		if (locked)
1184 			up_read(&sbi->sb->s_umount);
1185 		cond_resched();
1186 		goto retry_flush_quotas;
1187 	}
1188 
1189 retry_flush_dents:
1190 	/* write all the dirty dentry pages */
1191 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1192 		f2fs_unlock_all(sbi);
1193 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1194 		if (err)
1195 			goto out;
1196 		cond_resched();
1197 		goto retry_flush_quotas;
1198 	}
1199 
1200 	/*
1201 	 * POR: we should ensure that there are no dirty node pages
1202 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1203 	 */
1204 	down_write(&sbi->node_change);
1205 
1206 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1207 		up_write(&sbi->node_change);
1208 		f2fs_unlock_all(sbi);
1209 		err = f2fs_sync_inode_meta(sbi);
1210 		if (err)
1211 			goto out;
1212 		cond_resched();
1213 		goto retry_flush_quotas;
1214 	}
1215 
1216 retry_flush_nodes:
1217 	down_write(&sbi->node_write);
1218 
1219 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1220 		up_write(&sbi->node_write);
1221 		atomic_inc(&sbi->wb_sync_req[NODE]);
1222 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1223 		atomic_dec(&sbi->wb_sync_req[NODE]);
1224 		if (err) {
1225 			up_write(&sbi->node_change);
1226 			f2fs_unlock_all(sbi);
1227 			goto out;
1228 		}
1229 		cond_resched();
1230 		goto retry_flush_nodes;
1231 	}
1232 
1233 	/*
1234 	 * sbi->node_change is used only for AIO write_begin path which produces
1235 	 * dirty node blocks and some checkpoint values by block allocation.
1236 	 */
1237 	__prepare_cp_block(sbi);
1238 	up_write(&sbi->node_change);
1239 out:
1240 	blk_finish_plug(&plug);
1241 	return err;
1242 }
1243 
1244 static void unblock_operations(struct f2fs_sb_info *sbi)
1245 {
1246 	up_write(&sbi->node_write);
1247 	f2fs_unlock_all(sbi);
1248 }
1249 
1250 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1251 {
1252 	DEFINE_WAIT(wait);
1253 
1254 	for (;;) {
1255 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1256 
1257 		if (!get_pages(sbi, type))
1258 			break;
1259 
1260 		if (unlikely(f2fs_cp_error(sbi)))
1261 			break;
1262 
1263 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1264 	}
1265 	finish_wait(&sbi->cp_wait, &wait);
1266 }
1267 
1268 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1269 {
1270 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1271 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1272 	unsigned long flags;
1273 
1274 	spin_lock_irqsave(&sbi->cp_lock, flags);
1275 
1276 	if ((cpc->reason & CP_UMOUNT) &&
1277 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1278 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1279 		disable_nat_bits(sbi, false);
1280 
1281 	if (cpc->reason & CP_TRIMMED)
1282 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1283 	else
1284 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1285 
1286 	if (cpc->reason & CP_UMOUNT)
1287 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1288 	else
1289 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1290 
1291 	if (cpc->reason & CP_FASTBOOT)
1292 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1293 	else
1294 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1295 
1296 	if (orphan_num)
1297 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1298 	else
1299 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1300 
1301 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1302 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1303 
1304 	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1305 		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1306 	else
1307 		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1308 
1309 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1310 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1311 	else
1312 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1313 
1314 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1315 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1316 	else
1317 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1318 
1319 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1320 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1321 	else
1322 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1323 
1324 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1325 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1326 
1327 	/* set this flag to activate crc|cp_ver for recovery */
1328 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1329 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1330 
1331 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1332 }
1333 
1334 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1335 	void *src, block_t blk_addr)
1336 {
1337 	struct writeback_control wbc = {
1338 		.for_reclaim = 0,
1339 	};
1340 
1341 	/*
1342 	 * pagevec_lookup_tag and lock_page again will take
1343 	 * some extra time. Therefore, f2fs_update_meta_pages and
1344 	 * f2fs_sync_meta_pages are combined in this function.
1345 	 */
1346 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1347 	int err;
1348 
1349 	f2fs_wait_on_page_writeback(page, META, true, true);
1350 
1351 	memcpy(page_address(page), src, PAGE_SIZE);
1352 
1353 	set_page_dirty(page);
1354 	if (unlikely(!clear_page_dirty_for_io(page)))
1355 		f2fs_bug_on(sbi, 1);
1356 
1357 	/* writeout cp pack 2 page */
1358 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1359 	if (unlikely(err && f2fs_cp_error(sbi))) {
1360 		f2fs_put_page(page, 1);
1361 		return;
1362 	}
1363 
1364 	f2fs_bug_on(sbi, err);
1365 	f2fs_put_page(page, 0);
1366 
1367 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1368 	f2fs_submit_merged_write(sbi, META_FLUSH);
1369 }
1370 
1371 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1372 {
1373 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1374 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1375 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1376 	block_t start_blk;
1377 	unsigned int data_sum_blocks, orphan_blocks;
1378 	__u32 crc32 = 0;
1379 	int i;
1380 	int cp_payload_blks = __cp_payload(sbi);
1381 	struct super_block *sb = sbi->sb;
1382 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1383 	u64 kbytes_written;
1384 	int err;
1385 
1386 	/* Flush all the NAT/SIT pages */
1387 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1388 
1389 	/* start to update checkpoint, cp ver is already updated previously */
1390 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1391 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1392 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1393 		ckpt->cur_node_segno[i] =
1394 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1395 		ckpt->cur_node_blkoff[i] =
1396 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1397 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1398 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1399 	}
1400 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1401 		ckpt->cur_data_segno[i] =
1402 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1403 		ckpt->cur_data_blkoff[i] =
1404 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1405 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1406 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1407 	}
1408 
1409 	/* 2 cp  + n data seg summary + orphan inode blocks */
1410 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1411 	spin_lock_irqsave(&sbi->cp_lock, flags);
1412 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1413 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1414 	else
1415 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1416 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1417 
1418 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1419 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1420 			orphan_blocks);
1421 
1422 	if (__remain_node_summaries(cpc->reason))
1423 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1424 				cp_payload_blks + data_sum_blocks +
1425 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1426 	else
1427 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1428 				cp_payload_blks + data_sum_blocks +
1429 				orphan_blocks);
1430 
1431 	/* update ckpt flag for checkpoint */
1432 	update_ckpt_flags(sbi, cpc);
1433 
1434 	/* update SIT/NAT bitmap */
1435 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1436 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1437 
1438 	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1439 	*((__le32 *)((unsigned char *)ckpt +
1440 				le32_to_cpu(ckpt->checksum_offset)))
1441 				= cpu_to_le32(crc32);
1442 
1443 	start_blk = __start_cp_next_addr(sbi);
1444 
1445 	/* write nat bits */
1446 	if (enabled_nat_bits(sbi, cpc)) {
1447 		__u64 cp_ver = cur_cp_version(ckpt);
1448 		block_t blk;
1449 
1450 		cp_ver |= ((__u64)crc32 << 32);
1451 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1452 
1453 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1454 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1455 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1456 					(i << F2FS_BLKSIZE_BITS), blk + i);
1457 	}
1458 
1459 	/* write out checkpoint buffer at block 0 */
1460 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1461 
1462 	for (i = 1; i < 1 + cp_payload_blks; i++)
1463 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1464 							start_blk++);
1465 
1466 	if (orphan_num) {
1467 		write_orphan_inodes(sbi, start_blk);
1468 		start_blk += orphan_blocks;
1469 	}
1470 
1471 	f2fs_write_data_summaries(sbi, start_blk);
1472 	start_blk += data_sum_blocks;
1473 
1474 	/* Record write statistics in the hot node summary */
1475 	kbytes_written = sbi->kbytes_written;
1476 	if (sb->s_bdev->bd_part)
1477 		kbytes_written += BD_PART_WRITTEN(sbi);
1478 
1479 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1480 
1481 	if (__remain_node_summaries(cpc->reason)) {
1482 		f2fs_write_node_summaries(sbi, start_blk);
1483 		start_blk += NR_CURSEG_NODE_TYPE;
1484 	}
1485 
1486 	/* update user_block_counts */
1487 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1488 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1489 
1490 	/* Here, we have one bio having CP pack except cp pack 2 page */
1491 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1492 	/* Wait for all dirty meta pages to be submitted for IO */
1493 	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1494 
1495 	/* wait for previous submitted meta pages writeback */
1496 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1497 
1498 	/* flush all device cache */
1499 	err = f2fs_flush_device_cache(sbi);
1500 	if (err)
1501 		return err;
1502 
1503 	/* barrier and flush checkpoint cp pack 2 page if it can */
1504 	commit_checkpoint(sbi, ckpt, start_blk);
1505 	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1506 
1507 	/*
1508 	 * invalidate intermediate page cache borrowed from meta inode which are
1509 	 * used for migration of encrypted or verity inode's blocks.
1510 	 */
1511 	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi))
1512 		invalidate_mapping_pages(META_MAPPING(sbi),
1513 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1514 
1515 	f2fs_release_ino_entry(sbi, false);
1516 
1517 	f2fs_reset_fsync_node_info(sbi);
1518 
1519 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1520 	clear_sbi_flag(sbi, SBI_NEED_CP);
1521 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1522 
1523 	spin_lock(&sbi->stat_lock);
1524 	sbi->unusable_block_count = 0;
1525 	spin_unlock(&sbi->stat_lock);
1526 
1527 	__set_cp_next_pack(sbi);
1528 
1529 	/*
1530 	 * redirty superblock if metadata like node page or inode cache is
1531 	 * updated during writing checkpoint.
1532 	 */
1533 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1534 			get_pages(sbi, F2FS_DIRTY_IMETA))
1535 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1536 
1537 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1538 
1539 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1540 }
1541 
1542 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1543 {
1544 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1545 	unsigned long long ckpt_ver;
1546 	int err = 0;
1547 
1548 	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1549 		return -EROFS;
1550 
1551 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1552 		if (cpc->reason != CP_PAUSE)
1553 			return 0;
1554 		f2fs_warn(sbi, "Start checkpoint disabled!");
1555 	}
1556 	mutex_lock(&sbi->cp_mutex);
1557 
1558 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1559 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1560 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1561 		goto out;
1562 	if (unlikely(f2fs_cp_error(sbi))) {
1563 		err = -EIO;
1564 		goto out;
1565 	}
1566 
1567 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1568 
1569 	err = block_operations(sbi);
1570 	if (err)
1571 		goto out;
1572 
1573 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1574 
1575 	f2fs_flush_merged_writes(sbi);
1576 
1577 	/* this is the case of multiple fstrims without any changes */
1578 	if (cpc->reason & CP_DISCARD) {
1579 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1580 			unblock_operations(sbi);
1581 			goto out;
1582 		}
1583 
1584 		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1585 				SIT_I(sbi)->dirty_sentries == 0 &&
1586 				prefree_segments(sbi) == 0) {
1587 			f2fs_flush_sit_entries(sbi, cpc);
1588 			f2fs_clear_prefree_segments(sbi, cpc);
1589 			unblock_operations(sbi);
1590 			goto out;
1591 		}
1592 	}
1593 
1594 	/*
1595 	 * update checkpoint pack index
1596 	 * Increase the version number so that
1597 	 * SIT entries and seg summaries are written at correct place
1598 	 */
1599 	ckpt_ver = cur_cp_version(ckpt);
1600 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1601 
1602 	/* write cached NAT/SIT entries to NAT/SIT area */
1603 	err = f2fs_flush_nat_entries(sbi, cpc);
1604 	if (err)
1605 		goto stop;
1606 
1607 	f2fs_flush_sit_entries(sbi, cpc);
1608 
1609 	err = do_checkpoint(sbi, cpc);
1610 	if (err)
1611 		f2fs_release_discard_addrs(sbi);
1612 	else
1613 		f2fs_clear_prefree_segments(sbi, cpc);
1614 stop:
1615 	unblock_operations(sbi);
1616 	stat_inc_cp_count(sbi->stat_info);
1617 
1618 	if (cpc->reason & CP_RECOVERY)
1619 		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1620 
1621 	/* update CP_TIME to trigger checkpoint periodically */
1622 	f2fs_update_time(sbi, CP_TIME);
1623 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1624 out:
1625 	mutex_unlock(&sbi->cp_mutex);
1626 	return err;
1627 }
1628 
1629 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1630 {
1631 	int i;
1632 
1633 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1634 		struct inode_management *im = &sbi->im[i];
1635 
1636 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1637 		spin_lock_init(&im->ino_lock);
1638 		INIT_LIST_HEAD(&im->ino_list);
1639 		im->ino_num = 0;
1640 	}
1641 
1642 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1643 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1644 				F2FS_ORPHANS_PER_BLOCK;
1645 }
1646 
1647 int __init f2fs_create_checkpoint_caches(void)
1648 {
1649 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1650 			sizeof(struct ino_entry));
1651 	if (!ino_entry_slab)
1652 		return -ENOMEM;
1653 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1654 			sizeof(struct inode_entry));
1655 	if (!f2fs_inode_entry_slab) {
1656 		kmem_cache_destroy(ino_entry_slab);
1657 		return -ENOMEM;
1658 	}
1659 	return 0;
1660 }
1661 
1662 void f2fs_destroy_checkpoint_caches(void)
1663 {
1664 	kmem_cache_destroy(ino_entry_slab);
1665 	kmem_cache_destroy(f2fs_inode_entry_slab);
1666 }
1667