xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision be709d48)
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_meta = 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 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
134 					block_t blkaddr, int type)
135 {
136 	switch (type) {
137 	case META_NAT:
138 		break;
139 	case META_SIT:
140 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
141 			return false;
142 		break;
143 	case META_SSA:
144 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
145 			blkaddr < SM_I(sbi)->ssa_blkaddr))
146 			return false;
147 		break;
148 	case META_CP:
149 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
150 			blkaddr < __start_cp_addr(sbi)))
151 			return false;
152 		break;
153 	case META_POR:
154 	case DATA_GENERIC:
155 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
156 			blkaddr < MAIN_BLKADDR(sbi))) {
157 			if (type == DATA_GENERIC) {
158 				f2fs_msg(sbi->sb, KERN_WARNING,
159 					"access invalid blkaddr:%u", blkaddr);
160 				WARN_ON(1);
161 			}
162 			return false;
163 		}
164 		break;
165 	case META_GENERIC:
166 		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
167 			blkaddr >= MAIN_BLKADDR(sbi)))
168 			return false;
169 		break;
170 	default:
171 		BUG();
172 	}
173 
174 	return true;
175 }
176 
177 /*
178  * Readahead CP/NAT/SIT/SSA pages
179  */
180 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
181 							int type, bool sync)
182 {
183 	struct page *page;
184 	block_t blkno = start;
185 	struct f2fs_io_info fio = {
186 		.sbi = sbi,
187 		.type = META,
188 		.op = REQ_OP_READ,
189 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
190 		.encrypted_page = NULL,
191 		.in_list = false,
192 		.is_meta = (type != META_POR),
193 	};
194 	struct blk_plug plug;
195 
196 	if (unlikely(type == META_POR))
197 		fio.op_flags &= ~REQ_META;
198 
199 	blk_start_plug(&plug);
200 	for (; nrpages-- > 0; blkno++) {
201 
202 		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
203 			goto out;
204 
205 		switch (type) {
206 		case META_NAT:
207 			if (unlikely(blkno >=
208 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
209 				blkno = 0;
210 			/* get nat block addr */
211 			fio.new_blkaddr = current_nat_addr(sbi,
212 					blkno * NAT_ENTRY_PER_BLOCK);
213 			break;
214 		case META_SIT:
215 			/* get sit block addr */
216 			fio.new_blkaddr = current_sit_addr(sbi,
217 					blkno * SIT_ENTRY_PER_BLOCK);
218 			break;
219 		case META_SSA:
220 		case META_CP:
221 		case META_POR:
222 			fio.new_blkaddr = blkno;
223 			break;
224 		default:
225 			BUG();
226 		}
227 
228 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
229 						fio.new_blkaddr, false);
230 		if (!page)
231 			continue;
232 		if (PageUptodate(page)) {
233 			f2fs_put_page(page, 1);
234 			continue;
235 		}
236 
237 		fio.page = page;
238 		f2fs_submit_page_bio(&fio);
239 		f2fs_put_page(page, 0);
240 	}
241 out:
242 	blk_finish_plug(&plug);
243 	return blkno - start;
244 }
245 
246 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
247 {
248 	struct page *page;
249 	bool readahead = false;
250 
251 	page = find_get_page(META_MAPPING(sbi), index);
252 	if (!page || !PageUptodate(page))
253 		readahead = true;
254 	f2fs_put_page(page, 0);
255 
256 	if (readahead)
257 		f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
258 }
259 
260 static int __f2fs_write_meta_page(struct page *page,
261 				struct writeback_control *wbc,
262 				enum iostat_type io_type)
263 {
264 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
265 
266 	trace_f2fs_writepage(page, META);
267 
268 	if (unlikely(f2fs_cp_error(sbi)))
269 		goto redirty_out;
270 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
271 		goto redirty_out;
272 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
273 		goto redirty_out;
274 
275 	f2fs_do_write_meta_page(sbi, page, io_type);
276 	dec_page_count(sbi, F2FS_DIRTY_META);
277 
278 	if (wbc->for_reclaim)
279 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
280 
281 	unlock_page(page);
282 
283 	if (unlikely(f2fs_cp_error(sbi)))
284 		f2fs_submit_merged_write(sbi, META);
285 
286 	return 0;
287 
288 redirty_out:
289 	redirty_page_for_writepage(wbc, page);
290 	return AOP_WRITEPAGE_ACTIVATE;
291 }
292 
293 static int f2fs_write_meta_page(struct page *page,
294 				struct writeback_control *wbc)
295 {
296 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
297 }
298 
299 static int f2fs_write_meta_pages(struct address_space *mapping,
300 				struct writeback_control *wbc)
301 {
302 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
303 	long diff, written;
304 
305 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
306 		goto skip_write;
307 
308 	/* collect a number of dirty meta pages and write together */
309 	if (wbc->sync_mode != WB_SYNC_ALL &&
310 			get_pages(sbi, F2FS_DIRTY_META) <
311 					nr_pages_to_skip(sbi, META))
312 		goto skip_write;
313 
314 	/* if locked failed, cp will flush dirty pages instead */
315 	if (!mutex_trylock(&sbi->cp_mutex))
316 		goto skip_write;
317 
318 	trace_f2fs_writepages(mapping->host, wbc, META);
319 	diff = nr_pages_to_write(sbi, META, wbc);
320 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
321 	mutex_unlock(&sbi->cp_mutex);
322 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
323 	return 0;
324 
325 skip_write:
326 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
327 	trace_f2fs_writepages(mapping->host, wbc, META);
328 	return 0;
329 }
330 
331 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
332 				long nr_to_write, enum iostat_type io_type)
333 {
334 	struct address_space *mapping = META_MAPPING(sbi);
335 	pgoff_t index = 0, prev = ULONG_MAX;
336 	struct pagevec pvec;
337 	long nwritten = 0;
338 	int nr_pages;
339 	struct writeback_control wbc = {
340 		.for_reclaim = 0,
341 	};
342 	struct blk_plug plug;
343 
344 	pagevec_init(&pvec);
345 
346 	blk_start_plug(&plug);
347 
348 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
349 				PAGECACHE_TAG_DIRTY))) {
350 		int i;
351 
352 		for (i = 0; i < nr_pages; i++) {
353 			struct page *page = pvec.pages[i];
354 
355 			if (prev == ULONG_MAX)
356 				prev = page->index - 1;
357 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
358 				pagevec_release(&pvec);
359 				goto stop;
360 			}
361 
362 			lock_page(page);
363 
364 			if (unlikely(page->mapping != mapping)) {
365 continue_unlock:
366 				unlock_page(page);
367 				continue;
368 			}
369 			if (!PageDirty(page)) {
370 				/* someone wrote it for us */
371 				goto continue_unlock;
372 			}
373 
374 			f2fs_wait_on_page_writeback(page, META, true, true);
375 
376 			if (!clear_page_dirty_for_io(page))
377 				goto continue_unlock;
378 
379 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
380 				unlock_page(page);
381 				break;
382 			}
383 			nwritten++;
384 			prev = page->index;
385 			if (unlikely(nwritten >= nr_to_write))
386 				break;
387 		}
388 		pagevec_release(&pvec);
389 		cond_resched();
390 	}
391 stop:
392 	if (nwritten)
393 		f2fs_submit_merged_write(sbi, type);
394 
395 	blk_finish_plug(&plug);
396 
397 	return nwritten;
398 }
399 
400 static int f2fs_set_meta_page_dirty(struct page *page)
401 {
402 	trace_f2fs_set_page_dirty(page, META);
403 
404 	if (!PageUptodate(page))
405 		SetPageUptodate(page);
406 	if (!PageDirty(page)) {
407 		__set_page_dirty_nobuffers(page);
408 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
409 		f2fs_set_page_private(page, 0);
410 		f2fs_trace_pid(page);
411 		return 1;
412 	}
413 	return 0;
414 }
415 
416 const struct address_space_operations f2fs_meta_aops = {
417 	.writepage	= f2fs_write_meta_page,
418 	.writepages	= f2fs_write_meta_pages,
419 	.set_page_dirty	= f2fs_set_meta_page_dirty,
420 	.invalidatepage = f2fs_invalidate_page,
421 	.releasepage	= f2fs_release_page,
422 #ifdef CONFIG_MIGRATION
423 	.migratepage    = f2fs_migrate_page,
424 #endif
425 };
426 
427 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
428 						unsigned int devidx, int type)
429 {
430 	struct inode_management *im = &sbi->im[type];
431 	struct ino_entry *e, *tmp;
432 
433 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
434 
435 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
436 
437 	spin_lock(&im->ino_lock);
438 	e = radix_tree_lookup(&im->ino_root, ino);
439 	if (!e) {
440 		e = tmp;
441 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
442 			f2fs_bug_on(sbi, 1);
443 
444 		memset(e, 0, sizeof(struct ino_entry));
445 		e->ino = ino;
446 
447 		list_add_tail(&e->list, &im->ino_list);
448 		if (type != ORPHAN_INO)
449 			im->ino_num++;
450 	}
451 
452 	if (type == FLUSH_INO)
453 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
454 
455 	spin_unlock(&im->ino_lock);
456 	radix_tree_preload_end();
457 
458 	if (e != tmp)
459 		kmem_cache_free(ino_entry_slab, tmp);
460 }
461 
462 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
463 {
464 	struct inode_management *im = &sbi->im[type];
465 	struct ino_entry *e;
466 
467 	spin_lock(&im->ino_lock);
468 	e = radix_tree_lookup(&im->ino_root, ino);
469 	if (e) {
470 		list_del(&e->list);
471 		radix_tree_delete(&im->ino_root, ino);
472 		im->ino_num--;
473 		spin_unlock(&im->ino_lock);
474 		kmem_cache_free(ino_entry_slab, e);
475 		return;
476 	}
477 	spin_unlock(&im->ino_lock);
478 }
479 
480 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
481 {
482 	/* add new dirty ino entry into list */
483 	__add_ino_entry(sbi, ino, 0, type);
484 }
485 
486 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
487 {
488 	/* remove dirty ino entry from list */
489 	__remove_ino_entry(sbi, ino, type);
490 }
491 
492 /* mode should be APPEND_INO or UPDATE_INO */
493 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
494 {
495 	struct inode_management *im = &sbi->im[mode];
496 	struct ino_entry *e;
497 
498 	spin_lock(&im->ino_lock);
499 	e = radix_tree_lookup(&im->ino_root, ino);
500 	spin_unlock(&im->ino_lock);
501 	return e ? true : false;
502 }
503 
504 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
505 {
506 	struct ino_entry *e, *tmp;
507 	int i;
508 
509 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
510 		struct inode_management *im = &sbi->im[i];
511 
512 		spin_lock(&im->ino_lock);
513 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
514 			list_del(&e->list);
515 			radix_tree_delete(&im->ino_root, e->ino);
516 			kmem_cache_free(ino_entry_slab, e);
517 			im->ino_num--;
518 		}
519 		spin_unlock(&im->ino_lock);
520 	}
521 }
522 
523 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
524 					unsigned int devidx, int type)
525 {
526 	__add_ino_entry(sbi, ino, devidx, type);
527 }
528 
529 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
530 					unsigned int devidx, int type)
531 {
532 	struct inode_management *im = &sbi->im[type];
533 	struct ino_entry *e;
534 	bool is_dirty = false;
535 
536 	spin_lock(&im->ino_lock);
537 	e = radix_tree_lookup(&im->ino_root, ino);
538 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
539 		is_dirty = true;
540 	spin_unlock(&im->ino_lock);
541 	return is_dirty;
542 }
543 
544 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
545 {
546 	struct inode_management *im = &sbi->im[ORPHAN_INO];
547 	int err = 0;
548 
549 	spin_lock(&im->ino_lock);
550 
551 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
552 		spin_unlock(&im->ino_lock);
553 		f2fs_show_injection_info(FAULT_ORPHAN);
554 		return -ENOSPC;
555 	}
556 
557 	if (unlikely(im->ino_num >= sbi->max_orphans))
558 		err = -ENOSPC;
559 	else
560 		im->ino_num++;
561 	spin_unlock(&im->ino_lock);
562 
563 	return err;
564 }
565 
566 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
567 {
568 	struct inode_management *im = &sbi->im[ORPHAN_INO];
569 
570 	spin_lock(&im->ino_lock);
571 	f2fs_bug_on(sbi, im->ino_num == 0);
572 	im->ino_num--;
573 	spin_unlock(&im->ino_lock);
574 }
575 
576 void f2fs_add_orphan_inode(struct inode *inode)
577 {
578 	/* add new orphan ino entry into list */
579 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
580 	f2fs_update_inode_page(inode);
581 }
582 
583 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
584 {
585 	/* remove orphan entry from orphan list */
586 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
587 }
588 
589 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
590 {
591 	struct inode *inode;
592 	struct node_info ni;
593 	int err;
594 
595 	inode = f2fs_iget_retry(sbi->sb, ino);
596 	if (IS_ERR(inode)) {
597 		/*
598 		 * there should be a bug that we can't find the entry
599 		 * to orphan inode.
600 		 */
601 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
602 		return PTR_ERR(inode);
603 	}
604 
605 	err = dquot_initialize(inode);
606 	if (err) {
607 		iput(inode);
608 		goto err_out;
609 	}
610 
611 	clear_nlink(inode);
612 
613 	/* truncate all the data during iput */
614 	iput(inode);
615 
616 	err = f2fs_get_node_info(sbi, ino, &ni);
617 	if (err)
618 		goto err_out;
619 
620 	/* ENOMEM was fully retried in f2fs_evict_inode. */
621 	if (ni.blk_addr != NULL_ADDR) {
622 		err = -EIO;
623 		goto err_out;
624 	}
625 	return 0;
626 
627 err_out:
628 	set_sbi_flag(sbi, SBI_NEED_FSCK);
629 	f2fs_msg(sbi->sb, KERN_WARNING,
630 			"%s: orphan failed (ino=%x), run fsck to fix.",
631 			__func__, ino);
632 	return err;
633 }
634 
635 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
636 {
637 	block_t start_blk, orphan_blocks, i, j;
638 	unsigned int s_flags = sbi->sb->s_flags;
639 	int err = 0;
640 #ifdef CONFIG_QUOTA
641 	int quota_enabled;
642 #endif
643 
644 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
645 		return 0;
646 
647 	if (s_flags & SB_RDONLY) {
648 		f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
649 		sbi->sb->s_flags &= ~SB_RDONLY;
650 	}
651 
652 #ifdef CONFIG_QUOTA
653 	/* Needed for iput() to work correctly and not trash data */
654 	sbi->sb->s_flags |= SB_ACTIVE;
655 
656 	/*
657 	 * Turn on quotas which were not enabled for read-only mounts if
658 	 * filesystem has quota feature, so that they are updated correctly.
659 	 */
660 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
661 #endif
662 
663 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
664 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
665 
666 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
667 
668 	for (i = 0; i < orphan_blocks; i++) {
669 		struct page *page;
670 		struct f2fs_orphan_block *orphan_blk;
671 
672 		page = f2fs_get_meta_page(sbi, start_blk + i);
673 		if (IS_ERR(page)) {
674 			err = PTR_ERR(page);
675 			goto out;
676 		}
677 
678 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
679 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
680 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
681 			err = recover_orphan_inode(sbi, ino);
682 			if (err) {
683 				f2fs_put_page(page, 1);
684 				goto out;
685 			}
686 		}
687 		f2fs_put_page(page, 1);
688 	}
689 	/* clear Orphan Flag */
690 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
691 out:
692 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
693 
694 #ifdef CONFIG_QUOTA
695 	/* Turn quotas off */
696 	if (quota_enabled)
697 		f2fs_quota_off_umount(sbi->sb);
698 #endif
699 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
700 
701 	return err;
702 }
703 
704 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
705 {
706 	struct list_head *head;
707 	struct f2fs_orphan_block *orphan_blk = NULL;
708 	unsigned int nentries = 0;
709 	unsigned short index = 1;
710 	unsigned short orphan_blocks;
711 	struct page *page = NULL;
712 	struct ino_entry *orphan = NULL;
713 	struct inode_management *im = &sbi->im[ORPHAN_INO];
714 
715 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
716 
717 	/*
718 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
719 	 * orphan inode operations are covered under f2fs_lock_op().
720 	 * And, spin_lock should be avoided due to page operations below.
721 	 */
722 	head = &im->ino_list;
723 
724 	/* loop for each orphan inode entry and write them in Jornal block */
725 	list_for_each_entry(orphan, head, list) {
726 		if (!page) {
727 			page = f2fs_grab_meta_page(sbi, start_blk++);
728 			orphan_blk =
729 				(struct f2fs_orphan_block *)page_address(page);
730 			memset(orphan_blk, 0, sizeof(*orphan_blk));
731 		}
732 
733 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
734 
735 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
736 			/*
737 			 * an orphan block is full of 1020 entries,
738 			 * then we need to flush current orphan blocks
739 			 * and bring another one in memory
740 			 */
741 			orphan_blk->blk_addr = cpu_to_le16(index);
742 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
743 			orphan_blk->entry_count = cpu_to_le32(nentries);
744 			set_page_dirty(page);
745 			f2fs_put_page(page, 1);
746 			index++;
747 			nentries = 0;
748 			page = NULL;
749 		}
750 	}
751 
752 	if (page) {
753 		orphan_blk->blk_addr = cpu_to_le16(index);
754 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
755 		orphan_blk->entry_count = cpu_to_le32(nentries);
756 		set_page_dirty(page);
757 		f2fs_put_page(page, 1);
758 	}
759 }
760 
761 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
762 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
763 		unsigned long long *version)
764 {
765 	unsigned long blk_size = sbi->blocksize;
766 	size_t crc_offset = 0;
767 	__u32 crc = 0;
768 
769 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
770 	if (IS_ERR(*cp_page))
771 		return PTR_ERR(*cp_page);
772 
773 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
774 
775 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
776 	if (crc_offset > (blk_size - sizeof(__le32))) {
777 		f2fs_put_page(*cp_page, 1);
778 		f2fs_msg(sbi->sb, KERN_WARNING,
779 			"invalid crc_offset: %zu", crc_offset);
780 		return -EINVAL;
781 	}
782 
783 	crc = cur_cp_crc(*cp_block);
784 	if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
785 		f2fs_put_page(*cp_page, 1);
786 		f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
787 		return -EINVAL;
788 	}
789 
790 	*version = cur_cp_version(*cp_block);
791 	return 0;
792 }
793 
794 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
795 				block_t cp_addr, unsigned long long *version)
796 {
797 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
798 	struct f2fs_checkpoint *cp_block = NULL;
799 	unsigned long long cur_version = 0, pre_version = 0;
800 	int err;
801 
802 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
803 					&cp_page_1, version);
804 	if (err)
805 		return NULL;
806 
807 	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
808 					sbi->blocks_per_seg) {
809 		f2fs_msg(sbi->sb, KERN_WARNING,
810 			"invalid cp_pack_total_block_count:%u",
811 			le32_to_cpu(cp_block->cp_pack_total_block_count));
812 		goto invalid_cp;
813 	}
814 	pre_version = *version;
815 
816 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
817 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
818 					&cp_page_2, version);
819 	if (err)
820 		goto invalid_cp;
821 	cur_version = *version;
822 
823 	if (cur_version == pre_version) {
824 		*version = cur_version;
825 		f2fs_put_page(cp_page_2, 1);
826 		return cp_page_1;
827 	}
828 	f2fs_put_page(cp_page_2, 1);
829 invalid_cp:
830 	f2fs_put_page(cp_page_1, 1);
831 	return NULL;
832 }
833 
834 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
835 {
836 	struct f2fs_checkpoint *cp_block;
837 	struct f2fs_super_block *fsb = sbi->raw_super;
838 	struct page *cp1, *cp2, *cur_page;
839 	unsigned long blk_size = sbi->blocksize;
840 	unsigned long long cp1_version = 0, cp2_version = 0;
841 	unsigned long long cp_start_blk_no;
842 	unsigned int cp_blks = 1 + __cp_payload(sbi);
843 	block_t cp_blk_no;
844 	int i;
845 
846 	sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
847 				 GFP_KERNEL);
848 	if (!sbi->ckpt)
849 		return -ENOMEM;
850 	/*
851 	 * Finding out valid cp block involves read both
852 	 * sets( cp pack1 and cp pack 2)
853 	 */
854 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
855 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
856 
857 	/* The second checkpoint pack should start at the next segment */
858 	cp_start_blk_no += ((unsigned long long)1) <<
859 				le32_to_cpu(fsb->log_blocks_per_seg);
860 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
861 
862 	if (cp1 && cp2) {
863 		if (ver_after(cp2_version, cp1_version))
864 			cur_page = cp2;
865 		else
866 			cur_page = cp1;
867 	} else if (cp1) {
868 		cur_page = cp1;
869 	} else if (cp2) {
870 		cur_page = cp2;
871 	} else {
872 		goto fail_no_cp;
873 	}
874 
875 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
876 	memcpy(sbi->ckpt, cp_block, blk_size);
877 
878 	if (cur_page == cp1)
879 		sbi->cur_cp_pack = 1;
880 	else
881 		sbi->cur_cp_pack = 2;
882 
883 	/* Sanity checking of checkpoint */
884 	if (f2fs_sanity_check_ckpt(sbi))
885 		goto free_fail_no_cp;
886 
887 	if (cp_blks <= 1)
888 		goto done;
889 
890 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
891 	if (cur_page == cp2)
892 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
893 
894 	for (i = 1; i < cp_blks; i++) {
895 		void *sit_bitmap_ptr;
896 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
897 
898 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
899 		if (IS_ERR(cur_page))
900 			goto free_fail_no_cp;
901 		sit_bitmap_ptr = page_address(cur_page);
902 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
903 		f2fs_put_page(cur_page, 1);
904 	}
905 done:
906 	f2fs_put_page(cp1, 1);
907 	f2fs_put_page(cp2, 1);
908 	return 0;
909 
910 free_fail_no_cp:
911 	f2fs_put_page(cp1, 1);
912 	f2fs_put_page(cp2, 1);
913 fail_no_cp:
914 	kvfree(sbi->ckpt);
915 	return -EINVAL;
916 }
917 
918 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
919 {
920 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
921 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
922 
923 	if (is_inode_flag_set(inode, flag))
924 		return;
925 
926 	set_inode_flag(inode, flag);
927 	if (!f2fs_is_volatile_file(inode))
928 		list_add_tail(&F2FS_I(inode)->dirty_list,
929 						&sbi->inode_list[type]);
930 	stat_inc_dirty_inode(sbi, type);
931 }
932 
933 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
934 {
935 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
936 
937 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
938 		return;
939 
940 	list_del_init(&F2FS_I(inode)->dirty_list);
941 	clear_inode_flag(inode, flag);
942 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
943 }
944 
945 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
946 {
947 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
948 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
949 
950 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
951 			!S_ISLNK(inode->i_mode))
952 		return;
953 
954 	spin_lock(&sbi->inode_lock[type]);
955 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
956 		__add_dirty_inode(inode, type);
957 	inode_inc_dirty_pages(inode);
958 	spin_unlock(&sbi->inode_lock[type]);
959 
960 	f2fs_set_page_private(page, 0);
961 	f2fs_trace_pid(page);
962 }
963 
964 void f2fs_remove_dirty_inode(struct inode *inode)
965 {
966 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
967 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
968 
969 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
970 			!S_ISLNK(inode->i_mode))
971 		return;
972 
973 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
974 		return;
975 
976 	spin_lock(&sbi->inode_lock[type]);
977 	__remove_dirty_inode(inode, type);
978 	spin_unlock(&sbi->inode_lock[type]);
979 }
980 
981 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
982 {
983 	struct list_head *head;
984 	struct inode *inode;
985 	struct f2fs_inode_info *fi;
986 	bool is_dir = (type == DIR_INODE);
987 	unsigned long ino = 0;
988 
989 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
990 				get_pages(sbi, is_dir ?
991 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
992 retry:
993 	if (unlikely(f2fs_cp_error(sbi)))
994 		return -EIO;
995 
996 	spin_lock(&sbi->inode_lock[type]);
997 
998 	head = &sbi->inode_list[type];
999 	if (list_empty(head)) {
1000 		spin_unlock(&sbi->inode_lock[type]);
1001 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1002 				get_pages(sbi, is_dir ?
1003 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1004 		return 0;
1005 	}
1006 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1007 	inode = igrab(&fi->vfs_inode);
1008 	spin_unlock(&sbi->inode_lock[type]);
1009 	if (inode) {
1010 		unsigned long cur_ino = inode->i_ino;
1011 
1012 		if (is_dir)
1013 			F2FS_I(inode)->cp_task = current;
1014 
1015 		filemap_fdatawrite(inode->i_mapping);
1016 
1017 		if (is_dir)
1018 			F2FS_I(inode)->cp_task = NULL;
1019 
1020 		iput(inode);
1021 		/* We need to give cpu to another writers. */
1022 		if (ino == cur_ino)
1023 			cond_resched();
1024 		else
1025 			ino = cur_ino;
1026 	} else {
1027 		/*
1028 		 * We should submit bio, since it exists several
1029 		 * wribacking dentry pages in the freeing inode.
1030 		 */
1031 		f2fs_submit_merged_write(sbi, DATA);
1032 		cond_resched();
1033 	}
1034 	goto retry;
1035 }
1036 
1037 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1038 {
1039 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1040 	struct inode *inode;
1041 	struct f2fs_inode_info *fi;
1042 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1043 
1044 	while (total--) {
1045 		if (unlikely(f2fs_cp_error(sbi)))
1046 			return -EIO;
1047 
1048 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1049 		if (list_empty(head)) {
1050 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1051 			return 0;
1052 		}
1053 		fi = list_first_entry(head, struct f2fs_inode_info,
1054 							gdirty_list);
1055 		inode = igrab(&fi->vfs_inode);
1056 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1057 		if (inode) {
1058 			sync_inode_metadata(inode, 0);
1059 
1060 			/* it's on eviction */
1061 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1062 				f2fs_update_inode_page(inode);
1063 			iput(inode);
1064 		}
1065 	}
1066 	return 0;
1067 }
1068 
1069 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1070 {
1071 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1072 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1073 	nid_t last_nid = nm_i->next_scan_nid;
1074 
1075 	next_free_nid(sbi, &last_nid);
1076 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1077 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1078 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1079 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1080 }
1081 
1082 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1083 {
1084 	if (!is_journalled_quota(sbi))
1085 		return false;
1086 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1087 		return false;
1088 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1089 		return false;
1090 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH))
1091 		return true;
1092 	if (get_pages(sbi, F2FS_DIRTY_QDATA))
1093 		return true;
1094 	return false;
1095 }
1096 
1097 /*
1098  * Freeze all the FS-operations for checkpoint.
1099  */
1100 static int block_operations(struct f2fs_sb_info *sbi)
1101 {
1102 	struct writeback_control wbc = {
1103 		.sync_mode = WB_SYNC_ALL,
1104 		.nr_to_write = LONG_MAX,
1105 		.for_reclaim = 0,
1106 	};
1107 	struct blk_plug plug;
1108 	int err = 0, cnt = 0;
1109 
1110 	blk_start_plug(&plug);
1111 
1112 retry_flush_quotas:
1113 	if (__need_flush_quota(sbi)) {
1114 		int locked;
1115 
1116 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1117 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1118 			f2fs_lock_all(sbi);
1119 			goto retry_flush_dents;
1120 		}
1121 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1122 
1123 		/* only failed during mount/umount/freeze/quotactl */
1124 		locked = down_read_trylock(&sbi->sb->s_umount);
1125 		f2fs_quota_sync(sbi->sb, -1);
1126 		if (locked)
1127 			up_read(&sbi->sb->s_umount);
1128 	}
1129 
1130 	f2fs_lock_all(sbi);
1131 	if (__need_flush_quota(sbi)) {
1132 		f2fs_unlock_all(sbi);
1133 		cond_resched();
1134 		goto retry_flush_quotas;
1135 	}
1136 
1137 retry_flush_dents:
1138 	/* write all the dirty dentry pages */
1139 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1140 		f2fs_unlock_all(sbi);
1141 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1142 		if (err)
1143 			goto out;
1144 		cond_resched();
1145 		goto retry_flush_quotas;
1146 	}
1147 
1148 	/*
1149 	 * POR: we should ensure that there are no dirty node pages
1150 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1151 	 */
1152 	down_write(&sbi->node_change);
1153 
1154 	if (__need_flush_quota(sbi)) {
1155 		up_write(&sbi->node_change);
1156 		f2fs_unlock_all(sbi);
1157 		goto retry_flush_quotas;
1158 	}
1159 
1160 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1161 		up_write(&sbi->node_change);
1162 		f2fs_unlock_all(sbi);
1163 		err = f2fs_sync_inode_meta(sbi);
1164 		if (err)
1165 			goto out;
1166 		cond_resched();
1167 		goto retry_flush_quotas;
1168 	}
1169 
1170 retry_flush_nodes:
1171 	down_write(&sbi->node_write);
1172 
1173 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1174 		up_write(&sbi->node_write);
1175 		atomic_inc(&sbi->wb_sync_req[NODE]);
1176 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1177 		atomic_dec(&sbi->wb_sync_req[NODE]);
1178 		if (err) {
1179 			up_write(&sbi->node_change);
1180 			f2fs_unlock_all(sbi);
1181 			goto out;
1182 		}
1183 		cond_resched();
1184 		goto retry_flush_nodes;
1185 	}
1186 
1187 	/*
1188 	 * sbi->node_change is used only for AIO write_begin path which produces
1189 	 * dirty node blocks and some checkpoint values by block allocation.
1190 	 */
1191 	__prepare_cp_block(sbi);
1192 	up_write(&sbi->node_change);
1193 out:
1194 	blk_finish_plug(&plug);
1195 	return err;
1196 }
1197 
1198 static void unblock_operations(struct f2fs_sb_info *sbi)
1199 {
1200 	up_write(&sbi->node_write);
1201 	f2fs_unlock_all(sbi);
1202 }
1203 
1204 void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1205 {
1206 	DEFINE_WAIT(wait);
1207 
1208 	for (;;) {
1209 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1210 
1211 		if (!get_pages(sbi, F2FS_WB_CP_DATA))
1212 			break;
1213 
1214 		if (unlikely(f2fs_cp_error(sbi)))
1215 			break;
1216 
1217 		io_schedule_timeout(5*HZ);
1218 	}
1219 	finish_wait(&sbi->cp_wait, &wait);
1220 }
1221 
1222 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1223 {
1224 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1225 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1226 	unsigned long flags;
1227 
1228 	spin_lock_irqsave(&sbi->cp_lock, flags);
1229 
1230 	if ((cpc->reason & CP_UMOUNT) &&
1231 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1232 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1233 		disable_nat_bits(sbi, false);
1234 
1235 	if (cpc->reason & CP_TRIMMED)
1236 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1237 	else
1238 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1239 
1240 	if (cpc->reason & CP_UMOUNT)
1241 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1242 	else
1243 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1244 
1245 	if (cpc->reason & CP_FASTBOOT)
1246 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1247 	else
1248 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1249 
1250 	if (orphan_num)
1251 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1252 	else
1253 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1254 
1255 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1256 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1257 
1258 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1259 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1260 	else
1261 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1262 
1263 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1264 		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1265 	else
1266 		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1267 
1268 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1269 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1270 	/*
1271 	 * TODO: we count on fsck.f2fs to clear this flag until we figure out
1272 	 * missing cases which clear it incorrectly.
1273 	 */
1274 
1275 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1276 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1277 
1278 	/* set this flag to activate crc|cp_ver for recovery */
1279 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1280 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1281 
1282 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1283 }
1284 
1285 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1286 	void *src, block_t blk_addr)
1287 {
1288 	struct writeback_control wbc = {
1289 		.for_reclaim = 0,
1290 	};
1291 
1292 	/*
1293 	 * pagevec_lookup_tag and lock_page again will take
1294 	 * some extra time. Therefore, f2fs_update_meta_pages and
1295 	 * f2fs_sync_meta_pages are combined in this function.
1296 	 */
1297 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1298 	int err;
1299 
1300 	f2fs_wait_on_page_writeback(page, META, true, true);
1301 
1302 	memcpy(page_address(page), src, PAGE_SIZE);
1303 
1304 	set_page_dirty(page);
1305 	if (unlikely(!clear_page_dirty_for_io(page)))
1306 		f2fs_bug_on(sbi, 1);
1307 
1308 	/* writeout cp pack 2 page */
1309 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1310 	if (unlikely(err && f2fs_cp_error(sbi))) {
1311 		f2fs_put_page(page, 1);
1312 		return;
1313 	}
1314 
1315 	f2fs_bug_on(sbi, err);
1316 	f2fs_put_page(page, 0);
1317 
1318 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1319 	f2fs_submit_merged_write(sbi, META_FLUSH);
1320 }
1321 
1322 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1323 {
1324 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1325 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1326 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1327 	block_t start_blk;
1328 	unsigned int data_sum_blocks, orphan_blocks;
1329 	__u32 crc32 = 0;
1330 	int i;
1331 	int cp_payload_blks = __cp_payload(sbi);
1332 	struct super_block *sb = sbi->sb;
1333 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1334 	u64 kbytes_written;
1335 	int err;
1336 
1337 	/* Flush all the NAT/SIT pages */
1338 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1339 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
1340 					!f2fs_cp_error(sbi));
1341 
1342 	/*
1343 	 * modify checkpoint
1344 	 * version number is already updated
1345 	 */
1346 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1347 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1348 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1349 		ckpt->cur_node_segno[i] =
1350 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1351 		ckpt->cur_node_blkoff[i] =
1352 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1353 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1354 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1355 	}
1356 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1357 		ckpt->cur_data_segno[i] =
1358 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1359 		ckpt->cur_data_blkoff[i] =
1360 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1361 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1362 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1363 	}
1364 
1365 	/* 2 cp  + n data seg summary + orphan inode blocks */
1366 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1367 	spin_lock_irqsave(&sbi->cp_lock, flags);
1368 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1369 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1370 	else
1371 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1372 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1373 
1374 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1375 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1376 			orphan_blocks);
1377 
1378 	if (__remain_node_summaries(cpc->reason))
1379 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1380 				cp_payload_blks + data_sum_blocks +
1381 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1382 	else
1383 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1384 				cp_payload_blks + data_sum_blocks +
1385 				orphan_blocks);
1386 
1387 	/* update ckpt flag for checkpoint */
1388 	update_ckpt_flags(sbi, cpc);
1389 
1390 	/* update SIT/NAT bitmap */
1391 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1392 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1393 
1394 	crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1395 	*((__le32 *)((unsigned char *)ckpt +
1396 				le32_to_cpu(ckpt->checksum_offset)))
1397 				= cpu_to_le32(crc32);
1398 
1399 	start_blk = __start_cp_next_addr(sbi);
1400 
1401 	/* write nat bits */
1402 	if (enabled_nat_bits(sbi, cpc)) {
1403 		__u64 cp_ver = cur_cp_version(ckpt);
1404 		block_t blk;
1405 
1406 		cp_ver |= ((__u64)crc32 << 32);
1407 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1408 
1409 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1410 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1411 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1412 					(i << F2FS_BLKSIZE_BITS), blk + i);
1413 	}
1414 
1415 	/* write out checkpoint buffer at block 0 */
1416 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1417 
1418 	for (i = 1; i < 1 + cp_payload_blks; i++)
1419 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1420 							start_blk++);
1421 
1422 	if (orphan_num) {
1423 		write_orphan_inodes(sbi, start_blk);
1424 		start_blk += orphan_blocks;
1425 	}
1426 
1427 	f2fs_write_data_summaries(sbi, start_blk);
1428 	start_blk += data_sum_blocks;
1429 
1430 	/* Record write statistics in the hot node summary */
1431 	kbytes_written = sbi->kbytes_written;
1432 	if (sb->s_bdev->bd_part)
1433 		kbytes_written += BD_PART_WRITTEN(sbi);
1434 
1435 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1436 
1437 	if (__remain_node_summaries(cpc->reason)) {
1438 		f2fs_write_node_summaries(sbi, start_blk);
1439 		start_blk += NR_CURSEG_NODE_TYPE;
1440 	}
1441 
1442 	/* update user_block_counts */
1443 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1444 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1445 
1446 	/* Here, we have one bio having CP pack except cp pack 2 page */
1447 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1448 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
1449 					!f2fs_cp_error(sbi));
1450 
1451 	/* wait for previous submitted meta pages writeback */
1452 	f2fs_wait_on_all_pages_writeback(sbi);
1453 
1454 	/* flush all device cache */
1455 	err = f2fs_flush_device_cache(sbi);
1456 	if (err)
1457 		return err;
1458 
1459 	/* barrier and flush checkpoint cp pack 2 page if it can */
1460 	commit_checkpoint(sbi, ckpt, start_blk);
1461 	f2fs_wait_on_all_pages_writeback(sbi);
1462 
1463 	/*
1464 	 * invalidate intermediate page cache borrowed from meta inode
1465 	 * which are used for migration of encrypted inode's blocks.
1466 	 */
1467 	if (f2fs_sb_has_encrypt(sbi))
1468 		invalidate_mapping_pages(META_MAPPING(sbi),
1469 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1470 
1471 	f2fs_release_ino_entry(sbi, false);
1472 
1473 	f2fs_reset_fsync_node_info(sbi);
1474 
1475 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1476 	clear_sbi_flag(sbi, SBI_NEED_CP);
1477 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1478 	sbi->unusable_block_count = 0;
1479 	__set_cp_next_pack(sbi);
1480 
1481 	/*
1482 	 * redirty superblock if metadata like node page or inode cache is
1483 	 * updated during writing checkpoint.
1484 	 */
1485 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1486 			get_pages(sbi, F2FS_DIRTY_IMETA))
1487 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1488 
1489 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1490 
1491 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1492 }
1493 
1494 /*
1495  * We guarantee that this checkpoint procedure will not fail.
1496  */
1497 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1498 {
1499 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1500 	unsigned long long ckpt_ver;
1501 	int err = 0;
1502 
1503 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1504 		if (cpc->reason != CP_PAUSE)
1505 			return 0;
1506 		f2fs_msg(sbi->sb, KERN_WARNING,
1507 				"Start checkpoint disabled!");
1508 	}
1509 	mutex_lock(&sbi->cp_mutex);
1510 
1511 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1512 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1513 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1514 		goto out;
1515 	if (unlikely(f2fs_cp_error(sbi))) {
1516 		err = -EIO;
1517 		goto out;
1518 	}
1519 	if (f2fs_readonly(sbi->sb)) {
1520 		err = -EROFS;
1521 		goto out;
1522 	}
1523 
1524 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1525 
1526 	err = block_operations(sbi);
1527 	if (err)
1528 		goto out;
1529 
1530 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1531 
1532 	f2fs_flush_merged_writes(sbi);
1533 
1534 	/* this is the case of multiple fstrims without any changes */
1535 	if (cpc->reason & CP_DISCARD) {
1536 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1537 			unblock_operations(sbi);
1538 			goto out;
1539 		}
1540 
1541 		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1542 				SIT_I(sbi)->dirty_sentries == 0 &&
1543 				prefree_segments(sbi) == 0) {
1544 			f2fs_flush_sit_entries(sbi, cpc);
1545 			f2fs_clear_prefree_segments(sbi, cpc);
1546 			unblock_operations(sbi);
1547 			goto out;
1548 		}
1549 	}
1550 
1551 	/*
1552 	 * update checkpoint pack index
1553 	 * Increase the version number so that
1554 	 * SIT entries and seg summaries are written at correct place
1555 	 */
1556 	ckpt_ver = cur_cp_version(ckpt);
1557 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1558 
1559 	/* write cached NAT/SIT entries to NAT/SIT area */
1560 	err = f2fs_flush_nat_entries(sbi, cpc);
1561 	if (err)
1562 		goto stop;
1563 
1564 	f2fs_flush_sit_entries(sbi, cpc);
1565 
1566 	/* unlock all the fs_lock[] in do_checkpoint() */
1567 	err = do_checkpoint(sbi, cpc);
1568 	if (err)
1569 		f2fs_release_discard_addrs(sbi);
1570 	else
1571 		f2fs_clear_prefree_segments(sbi, cpc);
1572 stop:
1573 	unblock_operations(sbi);
1574 	stat_inc_cp_count(sbi->stat_info);
1575 
1576 	if (cpc->reason & CP_RECOVERY)
1577 		f2fs_msg(sbi->sb, KERN_NOTICE,
1578 			"checkpoint: version = %llx", ckpt_ver);
1579 
1580 	/* do checkpoint periodically */
1581 	f2fs_update_time(sbi, CP_TIME);
1582 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1583 out:
1584 	mutex_unlock(&sbi->cp_mutex);
1585 	return err;
1586 }
1587 
1588 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1589 {
1590 	int i;
1591 
1592 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1593 		struct inode_management *im = &sbi->im[i];
1594 
1595 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1596 		spin_lock_init(&im->ino_lock);
1597 		INIT_LIST_HEAD(&im->ino_list);
1598 		im->ino_num = 0;
1599 	}
1600 
1601 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1602 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1603 				F2FS_ORPHANS_PER_BLOCK;
1604 }
1605 
1606 int __init f2fs_create_checkpoint_caches(void)
1607 {
1608 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1609 			sizeof(struct ino_entry));
1610 	if (!ino_entry_slab)
1611 		return -ENOMEM;
1612 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1613 			sizeof(struct inode_entry));
1614 	if (!f2fs_inode_entry_slab) {
1615 		kmem_cache_destroy(ino_entry_slab);
1616 		return -ENOMEM;
1617 	}
1618 	return 0;
1619 }
1620 
1621 void f2fs_destroy_checkpoint_caches(void)
1622 {
1623 	kmem_cache_destroy(ino_entry_slab);
1624 	kmem_cache_destroy(f2fs_inode_entry_slab);
1625 }
1626