xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision d0e22329)
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->for_kupdate ||
310 		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
311 		goto skip_write;
312 
313 	/* if locked failed, cp will flush dirty pages instead */
314 	if (!mutex_trylock(&sbi->cp_mutex))
315 		goto skip_write;
316 
317 	trace_f2fs_writepages(mapping->host, wbc, META);
318 	diff = nr_pages_to_write(sbi, META, wbc);
319 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
320 	mutex_unlock(&sbi->cp_mutex);
321 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
322 	return 0;
323 
324 skip_write:
325 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
326 	trace_f2fs_writepages(mapping->host, wbc, META);
327 	return 0;
328 }
329 
330 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
331 				long nr_to_write, enum iostat_type io_type)
332 {
333 	struct address_space *mapping = META_MAPPING(sbi);
334 	pgoff_t index = 0, prev = ULONG_MAX;
335 	struct pagevec pvec;
336 	long nwritten = 0;
337 	int nr_pages;
338 	struct writeback_control wbc = {
339 		.for_reclaim = 0,
340 	};
341 	struct blk_plug plug;
342 
343 	pagevec_init(&pvec);
344 
345 	blk_start_plug(&plug);
346 
347 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
348 				PAGECACHE_TAG_DIRTY))) {
349 		int i;
350 
351 		for (i = 0; i < nr_pages; i++) {
352 			struct page *page = pvec.pages[i];
353 
354 			if (prev == ULONG_MAX)
355 				prev = page->index - 1;
356 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
357 				pagevec_release(&pvec);
358 				goto stop;
359 			}
360 
361 			lock_page(page);
362 
363 			if (unlikely(page->mapping != mapping)) {
364 continue_unlock:
365 				unlock_page(page);
366 				continue;
367 			}
368 			if (!PageDirty(page)) {
369 				/* someone wrote it for us */
370 				goto continue_unlock;
371 			}
372 
373 			f2fs_wait_on_page_writeback(page, META, true, true);
374 
375 			if (!clear_page_dirty_for_io(page))
376 				goto continue_unlock;
377 
378 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
379 				unlock_page(page);
380 				break;
381 			}
382 			nwritten++;
383 			prev = page->index;
384 			if (unlikely(nwritten >= nr_to_write))
385 				break;
386 		}
387 		pagevec_release(&pvec);
388 		cond_resched();
389 	}
390 stop:
391 	if (nwritten)
392 		f2fs_submit_merged_write(sbi, type);
393 
394 	blk_finish_plug(&plug);
395 
396 	return nwritten;
397 }
398 
399 static int f2fs_set_meta_page_dirty(struct page *page)
400 {
401 	trace_f2fs_set_page_dirty(page, META);
402 
403 	if (!PageUptodate(page))
404 		SetPageUptodate(page);
405 	if (!PageDirty(page)) {
406 		__set_page_dirty_nobuffers(page);
407 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
408 		SetPagePrivate(page);
409 		f2fs_trace_pid(page);
410 		return 1;
411 	}
412 	return 0;
413 }
414 
415 const struct address_space_operations f2fs_meta_aops = {
416 	.writepage	= f2fs_write_meta_page,
417 	.writepages	= f2fs_write_meta_pages,
418 	.set_page_dirty	= f2fs_set_meta_page_dirty,
419 	.invalidatepage = f2fs_invalidate_page,
420 	.releasepage	= f2fs_release_page,
421 #ifdef CONFIG_MIGRATION
422 	.migratepage    = f2fs_migrate_page,
423 #endif
424 };
425 
426 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
427 						unsigned int devidx, int type)
428 {
429 	struct inode_management *im = &sbi->im[type];
430 	struct ino_entry *e, *tmp;
431 
432 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
433 
434 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
435 
436 	spin_lock(&im->ino_lock);
437 	e = radix_tree_lookup(&im->ino_root, ino);
438 	if (!e) {
439 		e = tmp;
440 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
441 			f2fs_bug_on(sbi, 1);
442 
443 		memset(e, 0, sizeof(struct ino_entry));
444 		e->ino = ino;
445 
446 		list_add_tail(&e->list, &im->ino_list);
447 		if (type != ORPHAN_INO)
448 			im->ino_num++;
449 	}
450 
451 	if (type == FLUSH_INO)
452 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
453 
454 	spin_unlock(&im->ino_lock);
455 	radix_tree_preload_end();
456 
457 	if (e != tmp)
458 		kmem_cache_free(ino_entry_slab, tmp);
459 }
460 
461 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
462 {
463 	struct inode_management *im = &sbi->im[type];
464 	struct ino_entry *e;
465 
466 	spin_lock(&im->ino_lock);
467 	e = radix_tree_lookup(&im->ino_root, ino);
468 	if (e) {
469 		list_del(&e->list);
470 		radix_tree_delete(&im->ino_root, ino);
471 		im->ino_num--;
472 		spin_unlock(&im->ino_lock);
473 		kmem_cache_free(ino_entry_slab, e);
474 		return;
475 	}
476 	spin_unlock(&im->ino_lock);
477 }
478 
479 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
480 {
481 	/* add new dirty ino entry into list */
482 	__add_ino_entry(sbi, ino, 0, type);
483 }
484 
485 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
486 {
487 	/* remove dirty ino entry from list */
488 	__remove_ino_entry(sbi, ino, type);
489 }
490 
491 /* mode should be APPEND_INO or UPDATE_INO */
492 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
493 {
494 	struct inode_management *im = &sbi->im[mode];
495 	struct ino_entry *e;
496 
497 	spin_lock(&im->ino_lock);
498 	e = radix_tree_lookup(&im->ino_root, ino);
499 	spin_unlock(&im->ino_lock);
500 	return e ? true : false;
501 }
502 
503 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
504 {
505 	struct ino_entry *e, *tmp;
506 	int i;
507 
508 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
509 		struct inode_management *im = &sbi->im[i];
510 
511 		spin_lock(&im->ino_lock);
512 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
513 			list_del(&e->list);
514 			radix_tree_delete(&im->ino_root, e->ino);
515 			kmem_cache_free(ino_entry_slab, e);
516 			im->ino_num--;
517 		}
518 		spin_unlock(&im->ino_lock);
519 	}
520 }
521 
522 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
523 					unsigned int devidx, int type)
524 {
525 	__add_ino_entry(sbi, ino, devidx, type);
526 }
527 
528 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
529 					unsigned int devidx, int type)
530 {
531 	struct inode_management *im = &sbi->im[type];
532 	struct ino_entry *e;
533 	bool is_dirty = false;
534 
535 	spin_lock(&im->ino_lock);
536 	e = radix_tree_lookup(&im->ino_root, ino);
537 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
538 		is_dirty = true;
539 	spin_unlock(&im->ino_lock);
540 	return is_dirty;
541 }
542 
543 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
544 {
545 	struct inode_management *im = &sbi->im[ORPHAN_INO];
546 	int err = 0;
547 
548 	spin_lock(&im->ino_lock);
549 
550 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
551 		spin_unlock(&im->ino_lock);
552 		f2fs_show_injection_info(FAULT_ORPHAN);
553 		return -ENOSPC;
554 	}
555 
556 	if (unlikely(im->ino_num >= sbi->max_orphans))
557 		err = -ENOSPC;
558 	else
559 		im->ino_num++;
560 	spin_unlock(&im->ino_lock);
561 
562 	return err;
563 }
564 
565 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
566 {
567 	struct inode_management *im = &sbi->im[ORPHAN_INO];
568 
569 	spin_lock(&im->ino_lock);
570 	f2fs_bug_on(sbi, im->ino_num == 0);
571 	im->ino_num--;
572 	spin_unlock(&im->ino_lock);
573 }
574 
575 void f2fs_add_orphan_inode(struct inode *inode)
576 {
577 	/* add new orphan ino entry into list */
578 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
579 	f2fs_update_inode_page(inode);
580 }
581 
582 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
583 {
584 	/* remove orphan entry from orphan list */
585 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
586 }
587 
588 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
589 {
590 	struct inode *inode;
591 	struct node_info ni;
592 	int err;
593 
594 	inode = f2fs_iget_retry(sbi->sb, ino);
595 	if (IS_ERR(inode)) {
596 		/*
597 		 * there should be a bug that we can't find the entry
598 		 * to orphan inode.
599 		 */
600 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
601 		return PTR_ERR(inode);
602 	}
603 
604 	err = dquot_initialize(inode);
605 	if (err) {
606 		iput(inode);
607 		goto err_out;
608 	}
609 
610 	clear_nlink(inode);
611 
612 	/* truncate all the data during iput */
613 	iput(inode);
614 
615 	err = f2fs_get_node_info(sbi, ino, &ni);
616 	if (err)
617 		goto err_out;
618 
619 	/* ENOMEM was fully retried in f2fs_evict_inode. */
620 	if (ni.blk_addr != NULL_ADDR) {
621 		err = -EIO;
622 		goto err_out;
623 	}
624 	return 0;
625 
626 err_out:
627 	set_sbi_flag(sbi, SBI_NEED_FSCK);
628 	f2fs_msg(sbi->sb, KERN_WARNING,
629 			"%s: orphan failed (ino=%x), run fsck to fix.",
630 			__func__, ino);
631 	return err;
632 }
633 
634 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
635 {
636 	block_t start_blk, orphan_blocks, i, j;
637 	unsigned int s_flags = sbi->sb->s_flags;
638 	int err = 0;
639 #ifdef CONFIG_QUOTA
640 	int quota_enabled;
641 #endif
642 
643 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
644 		return 0;
645 
646 	if (s_flags & SB_RDONLY) {
647 		f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
648 		sbi->sb->s_flags &= ~SB_RDONLY;
649 	}
650 
651 #ifdef CONFIG_QUOTA
652 	/* Needed for iput() to work correctly and not trash data */
653 	sbi->sb->s_flags |= SB_ACTIVE;
654 
655 	/*
656 	 * Turn on quotas which were not enabled for read-only mounts if
657 	 * filesystem has quota feature, so that they are updated correctly.
658 	 */
659 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
660 #endif
661 
662 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
663 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
664 
665 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
666 
667 	for (i = 0; i < orphan_blocks; i++) {
668 		struct page *page;
669 		struct f2fs_orphan_block *orphan_blk;
670 
671 		page = f2fs_get_meta_page(sbi, start_blk + i);
672 		if (IS_ERR(page)) {
673 			err = PTR_ERR(page);
674 			goto out;
675 		}
676 
677 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
678 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
679 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
680 			err = recover_orphan_inode(sbi, ino);
681 			if (err) {
682 				f2fs_put_page(page, 1);
683 				goto out;
684 			}
685 		}
686 		f2fs_put_page(page, 1);
687 	}
688 	/* clear Orphan Flag */
689 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
690 out:
691 	set_sbi_flag(sbi, SBI_IS_RECOVERED);
692 
693 #ifdef CONFIG_QUOTA
694 	/* Turn quotas off */
695 	if (quota_enabled)
696 		f2fs_quota_off_umount(sbi->sb);
697 #endif
698 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
699 
700 	return err;
701 }
702 
703 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
704 {
705 	struct list_head *head;
706 	struct f2fs_orphan_block *orphan_blk = NULL;
707 	unsigned int nentries = 0;
708 	unsigned short index = 1;
709 	unsigned short orphan_blocks;
710 	struct page *page = NULL;
711 	struct ino_entry *orphan = NULL;
712 	struct inode_management *im = &sbi->im[ORPHAN_INO];
713 
714 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
715 
716 	/*
717 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
718 	 * orphan inode operations are covered under f2fs_lock_op().
719 	 * And, spin_lock should be avoided due to page operations below.
720 	 */
721 	head = &im->ino_list;
722 
723 	/* loop for each orphan inode entry and write them in Jornal block */
724 	list_for_each_entry(orphan, head, list) {
725 		if (!page) {
726 			page = f2fs_grab_meta_page(sbi, start_blk++);
727 			orphan_blk =
728 				(struct f2fs_orphan_block *)page_address(page);
729 			memset(orphan_blk, 0, sizeof(*orphan_blk));
730 		}
731 
732 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
733 
734 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
735 			/*
736 			 * an orphan block is full of 1020 entries,
737 			 * then we need to flush current orphan blocks
738 			 * and bring another one in memory
739 			 */
740 			orphan_blk->blk_addr = cpu_to_le16(index);
741 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
742 			orphan_blk->entry_count = cpu_to_le32(nentries);
743 			set_page_dirty(page);
744 			f2fs_put_page(page, 1);
745 			index++;
746 			nentries = 0;
747 			page = NULL;
748 		}
749 	}
750 
751 	if (page) {
752 		orphan_blk->blk_addr = cpu_to_le16(index);
753 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
754 		orphan_blk->entry_count = cpu_to_le32(nentries);
755 		set_page_dirty(page);
756 		f2fs_put_page(page, 1);
757 	}
758 }
759 
760 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
761 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
762 		unsigned long long *version)
763 {
764 	unsigned long blk_size = sbi->blocksize;
765 	size_t crc_offset = 0;
766 	__u32 crc = 0;
767 
768 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
769 	if (IS_ERR(*cp_page))
770 		return PTR_ERR(*cp_page);
771 
772 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
773 
774 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
775 	if (crc_offset > (blk_size - sizeof(__le32))) {
776 		f2fs_put_page(*cp_page, 1);
777 		f2fs_msg(sbi->sb, KERN_WARNING,
778 			"invalid crc_offset: %zu", crc_offset);
779 		return -EINVAL;
780 	}
781 
782 	crc = cur_cp_crc(*cp_block);
783 	if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
784 		f2fs_put_page(*cp_page, 1);
785 		f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
786 		return -EINVAL;
787 	}
788 
789 	*version = cur_cp_version(*cp_block);
790 	return 0;
791 }
792 
793 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
794 				block_t cp_addr, unsigned long long *version)
795 {
796 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
797 	struct f2fs_checkpoint *cp_block = NULL;
798 	unsigned long long cur_version = 0, pre_version = 0;
799 	int err;
800 
801 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
802 					&cp_page_1, version);
803 	if (err)
804 		return NULL;
805 
806 	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
807 					sbi->blocks_per_seg) {
808 		f2fs_msg(sbi->sb, KERN_WARNING,
809 			"invalid cp_pack_total_block_count:%u",
810 			le32_to_cpu(cp_block->cp_pack_total_block_count));
811 		goto invalid_cp;
812 	}
813 	pre_version = *version;
814 
815 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
816 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
817 					&cp_page_2, version);
818 	if (err)
819 		goto invalid_cp;
820 	cur_version = *version;
821 
822 	if (cur_version == pre_version) {
823 		*version = cur_version;
824 		f2fs_put_page(cp_page_2, 1);
825 		return cp_page_1;
826 	}
827 	f2fs_put_page(cp_page_2, 1);
828 invalid_cp:
829 	f2fs_put_page(cp_page_1, 1);
830 	return NULL;
831 }
832 
833 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
834 {
835 	struct f2fs_checkpoint *cp_block;
836 	struct f2fs_super_block *fsb = sbi->raw_super;
837 	struct page *cp1, *cp2, *cur_page;
838 	unsigned long blk_size = sbi->blocksize;
839 	unsigned long long cp1_version = 0, cp2_version = 0;
840 	unsigned long long cp_start_blk_no;
841 	unsigned int cp_blks = 1 + __cp_payload(sbi);
842 	block_t cp_blk_no;
843 	int i;
844 
845 	sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
846 				 GFP_KERNEL);
847 	if (!sbi->ckpt)
848 		return -ENOMEM;
849 	/*
850 	 * Finding out valid cp block involves read both
851 	 * sets( cp pack1 and cp pack 2)
852 	 */
853 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
854 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
855 
856 	/* The second checkpoint pack should start at the next segment */
857 	cp_start_blk_no += ((unsigned long long)1) <<
858 				le32_to_cpu(fsb->log_blocks_per_seg);
859 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
860 
861 	if (cp1 && cp2) {
862 		if (ver_after(cp2_version, cp1_version))
863 			cur_page = cp2;
864 		else
865 			cur_page = cp1;
866 	} else if (cp1) {
867 		cur_page = cp1;
868 	} else if (cp2) {
869 		cur_page = cp2;
870 	} else {
871 		goto fail_no_cp;
872 	}
873 
874 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
875 	memcpy(sbi->ckpt, cp_block, blk_size);
876 
877 	if (cur_page == cp1)
878 		sbi->cur_cp_pack = 1;
879 	else
880 		sbi->cur_cp_pack = 2;
881 
882 	/* Sanity checking of checkpoint */
883 	if (f2fs_sanity_check_ckpt(sbi))
884 		goto free_fail_no_cp;
885 
886 	if (cp_blks <= 1)
887 		goto done;
888 
889 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
890 	if (cur_page == cp2)
891 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
892 
893 	for (i = 1; i < cp_blks; i++) {
894 		void *sit_bitmap_ptr;
895 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
896 
897 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
898 		if (IS_ERR(cur_page))
899 			goto free_fail_no_cp;
900 		sit_bitmap_ptr = page_address(cur_page);
901 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
902 		f2fs_put_page(cur_page, 1);
903 	}
904 done:
905 	f2fs_put_page(cp1, 1);
906 	f2fs_put_page(cp2, 1);
907 	return 0;
908 
909 free_fail_no_cp:
910 	f2fs_put_page(cp1, 1);
911 	f2fs_put_page(cp2, 1);
912 fail_no_cp:
913 	kvfree(sbi->ckpt);
914 	return -EINVAL;
915 }
916 
917 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
918 {
919 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
920 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
921 
922 	if (is_inode_flag_set(inode, flag))
923 		return;
924 
925 	set_inode_flag(inode, flag);
926 	if (!f2fs_is_volatile_file(inode))
927 		list_add_tail(&F2FS_I(inode)->dirty_list,
928 						&sbi->inode_list[type]);
929 	stat_inc_dirty_inode(sbi, type);
930 }
931 
932 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
933 {
934 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
935 
936 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
937 		return;
938 
939 	list_del_init(&F2FS_I(inode)->dirty_list);
940 	clear_inode_flag(inode, flag);
941 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
942 }
943 
944 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
945 {
946 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
947 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
948 
949 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
950 			!S_ISLNK(inode->i_mode))
951 		return;
952 
953 	spin_lock(&sbi->inode_lock[type]);
954 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
955 		__add_dirty_inode(inode, type);
956 	inode_inc_dirty_pages(inode);
957 	spin_unlock(&sbi->inode_lock[type]);
958 
959 	SetPagePrivate(page);
960 	f2fs_trace_pid(page);
961 }
962 
963 void f2fs_remove_dirty_inode(struct inode *inode)
964 {
965 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
966 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
967 
968 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
969 			!S_ISLNK(inode->i_mode))
970 		return;
971 
972 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
973 		return;
974 
975 	spin_lock(&sbi->inode_lock[type]);
976 	__remove_dirty_inode(inode, type);
977 	spin_unlock(&sbi->inode_lock[type]);
978 }
979 
980 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
981 {
982 	struct list_head *head;
983 	struct inode *inode;
984 	struct f2fs_inode_info *fi;
985 	bool is_dir = (type == DIR_INODE);
986 	unsigned long ino = 0;
987 
988 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
989 				get_pages(sbi, is_dir ?
990 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
991 retry:
992 	if (unlikely(f2fs_cp_error(sbi)))
993 		return -EIO;
994 
995 	spin_lock(&sbi->inode_lock[type]);
996 
997 	head = &sbi->inode_list[type];
998 	if (list_empty(head)) {
999 		spin_unlock(&sbi->inode_lock[type]);
1000 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1001 				get_pages(sbi, is_dir ?
1002 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1003 		return 0;
1004 	}
1005 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1006 	inode = igrab(&fi->vfs_inode);
1007 	spin_unlock(&sbi->inode_lock[type]);
1008 	if (inode) {
1009 		unsigned long cur_ino = inode->i_ino;
1010 
1011 		if (is_dir)
1012 			F2FS_I(inode)->cp_task = current;
1013 
1014 		filemap_fdatawrite(inode->i_mapping);
1015 
1016 		if (is_dir)
1017 			F2FS_I(inode)->cp_task = NULL;
1018 
1019 		iput(inode);
1020 		/* We need to give cpu to another writers. */
1021 		if (ino == cur_ino)
1022 			cond_resched();
1023 		else
1024 			ino = cur_ino;
1025 	} else {
1026 		/*
1027 		 * We should submit bio, since it exists several
1028 		 * wribacking dentry pages in the freeing inode.
1029 		 */
1030 		f2fs_submit_merged_write(sbi, DATA);
1031 		cond_resched();
1032 	}
1033 	goto retry;
1034 }
1035 
1036 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1037 {
1038 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1039 	struct inode *inode;
1040 	struct f2fs_inode_info *fi;
1041 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1042 
1043 	while (total--) {
1044 		if (unlikely(f2fs_cp_error(sbi)))
1045 			return -EIO;
1046 
1047 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1048 		if (list_empty(head)) {
1049 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1050 			return 0;
1051 		}
1052 		fi = list_first_entry(head, struct f2fs_inode_info,
1053 							gdirty_list);
1054 		inode = igrab(&fi->vfs_inode);
1055 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1056 		if (inode) {
1057 			sync_inode_metadata(inode, 0);
1058 
1059 			/* it's on eviction */
1060 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1061 				f2fs_update_inode_page(inode);
1062 			iput(inode);
1063 		}
1064 	}
1065 	return 0;
1066 }
1067 
1068 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1069 {
1070 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1071 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1072 	nid_t last_nid = nm_i->next_scan_nid;
1073 
1074 	next_free_nid(sbi, &last_nid);
1075 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1076 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1077 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1078 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1079 }
1080 
1081 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1082 {
1083 	if (!is_journalled_quota(sbi))
1084 		return false;
1085 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1086 		return false;
1087 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1088 		return false;
1089 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH))
1090 		return true;
1091 	if (get_pages(sbi, F2FS_DIRTY_QDATA))
1092 		return true;
1093 	return false;
1094 }
1095 
1096 /*
1097  * Freeze all the FS-operations for checkpoint.
1098  */
1099 static int block_operations(struct f2fs_sb_info *sbi)
1100 {
1101 	struct writeback_control wbc = {
1102 		.sync_mode = WB_SYNC_ALL,
1103 		.nr_to_write = LONG_MAX,
1104 		.for_reclaim = 0,
1105 	};
1106 	struct blk_plug plug;
1107 	int err = 0, cnt = 0;
1108 
1109 	blk_start_plug(&plug);
1110 
1111 retry_flush_quotas:
1112 	if (__need_flush_quota(sbi)) {
1113 		int locked;
1114 
1115 		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1116 			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1117 			f2fs_lock_all(sbi);
1118 			goto retry_flush_dents;
1119 		}
1120 		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1121 
1122 		/* only failed during mount/umount/freeze/quotactl */
1123 		locked = down_read_trylock(&sbi->sb->s_umount);
1124 		f2fs_quota_sync(sbi->sb, -1);
1125 		if (locked)
1126 			up_read(&sbi->sb->s_umount);
1127 	}
1128 
1129 	f2fs_lock_all(sbi);
1130 	if (__need_flush_quota(sbi)) {
1131 		f2fs_unlock_all(sbi);
1132 		cond_resched();
1133 		goto retry_flush_quotas;
1134 	}
1135 
1136 retry_flush_dents:
1137 	/* write all the dirty dentry pages */
1138 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1139 		f2fs_unlock_all(sbi);
1140 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1141 		if (err)
1142 			goto out;
1143 		cond_resched();
1144 		goto retry_flush_quotas;
1145 	}
1146 
1147 	/*
1148 	 * POR: we should ensure that there are no dirty node pages
1149 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1150 	 */
1151 	down_write(&sbi->node_change);
1152 
1153 	if (__need_flush_quota(sbi)) {
1154 		up_write(&sbi->node_change);
1155 		f2fs_unlock_all(sbi);
1156 		goto retry_flush_quotas;
1157 	}
1158 
1159 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1160 		up_write(&sbi->node_change);
1161 		f2fs_unlock_all(sbi);
1162 		err = f2fs_sync_inode_meta(sbi);
1163 		if (err)
1164 			goto out;
1165 		cond_resched();
1166 		goto retry_flush_quotas;
1167 	}
1168 
1169 retry_flush_nodes:
1170 	down_write(&sbi->node_write);
1171 
1172 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1173 		up_write(&sbi->node_write);
1174 		atomic_inc(&sbi->wb_sync_req[NODE]);
1175 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1176 		atomic_dec(&sbi->wb_sync_req[NODE]);
1177 		if (err) {
1178 			up_write(&sbi->node_change);
1179 			f2fs_unlock_all(sbi);
1180 			goto out;
1181 		}
1182 		cond_resched();
1183 		goto retry_flush_nodes;
1184 	}
1185 
1186 	/*
1187 	 * sbi->node_change is used only for AIO write_begin path which produces
1188 	 * dirty node blocks and some checkpoint values by block allocation.
1189 	 */
1190 	__prepare_cp_block(sbi);
1191 	up_write(&sbi->node_change);
1192 out:
1193 	blk_finish_plug(&plug);
1194 	return err;
1195 }
1196 
1197 static void unblock_operations(struct f2fs_sb_info *sbi)
1198 {
1199 	up_write(&sbi->node_write);
1200 	f2fs_unlock_all(sbi);
1201 }
1202 
1203 void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1204 {
1205 	DEFINE_WAIT(wait);
1206 
1207 	for (;;) {
1208 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1209 
1210 		if (!get_pages(sbi, F2FS_WB_CP_DATA))
1211 			break;
1212 
1213 		if (unlikely(f2fs_cp_error(sbi)))
1214 			break;
1215 
1216 		io_schedule_timeout(5*HZ);
1217 	}
1218 	finish_wait(&sbi->cp_wait, &wait);
1219 }
1220 
1221 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1222 {
1223 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1224 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1225 	unsigned long flags;
1226 
1227 	spin_lock_irqsave(&sbi->cp_lock, flags);
1228 
1229 	if ((cpc->reason & CP_UMOUNT) &&
1230 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1231 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1232 		disable_nat_bits(sbi, false);
1233 
1234 	if (cpc->reason & CP_TRIMMED)
1235 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1236 	else
1237 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1238 
1239 	if (cpc->reason & CP_UMOUNT)
1240 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1241 	else
1242 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1243 
1244 	if (cpc->reason & CP_FASTBOOT)
1245 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1246 	else
1247 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1248 
1249 	if (orphan_num)
1250 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1251 	else
1252 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1253 
1254 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1255 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1256 
1257 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1258 		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1259 	else
1260 		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1261 
1262 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1263 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1264 	else
1265 		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1266 
1267 	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1268 		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1269 
1270 	/* set this flag to activate crc|cp_ver for recovery */
1271 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1272 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1273 
1274 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1275 }
1276 
1277 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1278 	void *src, block_t blk_addr)
1279 {
1280 	struct writeback_control wbc = {
1281 		.for_reclaim = 0,
1282 	};
1283 
1284 	/*
1285 	 * pagevec_lookup_tag and lock_page again will take
1286 	 * some extra time. Therefore, f2fs_update_meta_pages and
1287 	 * f2fs_sync_meta_pages are combined in this function.
1288 	 */
1289 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1290 	int err;
1291 
1292 	f2fs_wait_on_page_writeback(page, META, true, true);
1293 
1294 	memcpy(page_address(page), src, PAGE_SIZE);
1295 
1296 	set_page_dirty(page);
1297 	if (unlikely(!clear_page_dirty_for_io(page)))
1298 		f2fs_bug_on(sbi, 1);
1299 
1300 	/* writeout cp pack 2 page */
1301 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1302 	if (unlikely(err && f2fs_cp_error(sbi))) {
1303 		f2fs_put_page(page, 1);
1304 		return;
1305 	}
1306 
1307 	f2fs_bug_on(sbi, err);
1308 	f2fs_put_page(page, 0);
1309 
1310 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1311 	f2fs_submit_merged_write(sbi, META_FLUSH);
1312 }
1313 
1314 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1315 {
1316 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1317 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1318 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1319 	block_t start_blk;
1320 	unsigned int data_sum_blocks, orphan_blocks;
1321 	__u32 crc32 = 0;
1322 	int i;
1323 	int cp_payload_blks = __cp_payload(sbi);
1324 	struct super_block *sb = sbi->sb;
1325 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1326 	u64 kbytes_written;
1327 	int err;
1328 
1329 	/* Flush all the NAT/SIT pages */
1330 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1331 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
1332 					!f2fs_cp_error(sbi));
1333 
1334 	/*
1335 	 * modify checkpoint
1336 	 * version number is already updated
1337 	 */
1338 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1339 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1340 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1341 		ckpt->cur_node_segno[i] =
1342 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1343 		ckpt->cur_node_blkoff[i] =
1344 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1345 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1346 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1347 	}
1348 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1349 		ckpt->cur_data_segno[i] =
1350 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1351 		ckpt->cur_data_blkoff[i] =
1352 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1353 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1354 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1355 	}
1356 
1357 	/* 2 cp  + n data seg summary + orphan inode blocks */
1358 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1359 	spin_lock_irqsave(&sbi->cp_lock, flags);
1360 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1361 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1362 	else
1363 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1364 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1365 
1366 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1367 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1368 			orphan_blocks);
1369 
1370 	if (__remain_node_summaries(cpc->reason))
1371 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1372 				cp_payload_blks + data_sum_blocks +
1373 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1374 	else
1375 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1376 				cp_payload_blks + data_sum_blocks +
1377 				orphan_blocks);
1378 
1379 	/* update ckpt flag for checkpoint */
1380 	update_ckpt_flags(sbi, cpc);
1381 
1382 	/* update SIT/NAT bitmap */
1383 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1384 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1385 
1386 	crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1387 	*((__le32 *)((unsigned char *)ckpt +
1388 				le32_to_cpu(ckpt->checksum_offset)))
1389 				= cpu_to_le32(crc32);
1390 
1391 	start_blk = __start_cp_next_addr(sbi);
1392 
1393 	/* write nat bits */
1394 	if (enabled_nat_bits(sbi, cpc)) {
1395 		__u64 cp_ver = cur_cp_version(ckpt);
1396 		block_t blk;
1397 
1398 		cp_ver |= ((__u64)crc32 << 32);
1399 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1400 
1401 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1402 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1403 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1404 					(i << F2FS_BLKSIZE_BITS), blk + i);
1405 	}
1406 
1407 	/* write out checkpoint buffer at block 0 */
1408 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1409 
1410 	for (i = 1; i < 1 + cp_payload_blks; i++)
1411 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1412 							start_blk++);
1413 
1414 	if (orphan_num) {
1415 		write_orphan_inodes(sbi, start_blk);
1416 		start_blk += orphan_blocks;
1417 	}
1418 
1419 	f2fs_write_data_summaries(sbi, start_blk);
1420 	start_blk += data_sum_blocks;
1421 
1422 	/* Record write statistics in the hot node summary */
1423 	kbytes_written = sbi->kbytes_written;
1424 	if (sb->s_bdev->bd_part)
1425 		kbytes_written += BD_PART_WRITTEN(sbi);
1426 
1427 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1428 
1429 	if (__remain_node_summaries(cpc->reason)) {
1430 		f2fs_write_node_summaries(sbi, start_blk);
1431 		start_blk += NR_CURSEG_NODE_TYPE;
1432 	}
1433 
1434 	/* update user_block_counts */
1435 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1436 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1437 
1438 	/* Here, we have one bio having CP pack except cp pack 2 page */
1439 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1440 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
1441 					!f2fs_cp_error(sbi));
1442 
1443 	/* wait for previous submitted meta pages writeback */
1444 	f2fs_wait_on_all_pages_writeback(sbi);
1445 
1446 	/* flush all device cache */
1447 	err = f2fs_flush_device_cache(sbi);
1448 	if (err)
1449 		return err;
1450 
1451 	/* barrier and flush checkpoint cp pack 2 page if it can */
1452 	commit_checkpoint(sbi, ckpt, start_blk);
1453 	f2fs_wait_on_all_pages_writeback(sbi);
1454 
1455 	/*
1456 	 * invalidate intermediate page cache borrowed from meta inode
1457 	 * which are used for migration of encrypted inode's blocks.
1458 	 */
1459 	if (f2fs_sb_has_encrypt(sbi))
1460 		invalidate_mapping_pages(META_MAPPING(sbi),
1461 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1462 
1463 	f2fs_release_ino_entry(sbi, false);
1464 
1465 	f2fs_reset_fsync_node_info(sbi);
1466 
1467 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1468 	clear_sbi_flag(sbi, SBI_NEED_CP);
1469 	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1470 	sbi->unusable_block_count = 0;
1471 	__set_cp_next_pack(sbi);
1472 
1473 	/*
1474 	 * redirty superblock if metadata like node page or inode cache is
1475 	 * updated during writing checkpoint.
1476 	 */
1477 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1478 			get_pages(sbi, F2FS_DIRTY_IMETA))
1479 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1480 
1481 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1482 
1483 	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1484 }
1485 
1486 /*
1487  * We guarantee that this checkpoint procedure will not fail.
1488  */
1489 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1490 {
1491 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1492 	unsigned long long ckpt_ver;
1493 	int err = 0;
1494 
1495 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1496 		if (cpc->reason != CP_PAUSE)
1497 			return 0;
1498 		f2fs_msg(sbi->sb, KERN_WARNING,
1499 				"Start checkpoint disabled!");
1500 	}
1501 	mutex_lock(&sbi->cp_mutex);
1502 
1503 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1504 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1505 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1506 		goto out;
1507 	if (unlikely(f2fs_cp_error(sbi))) {
1508 		err = -EIO;
1509 		goto out;
1510 	}
1511 	if (f2fs_readonly(sbi->sb)) {
1512 		err = -EROFS;
1513 		goto out;
1514 	}
1515 
1516 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1517 
1518 	err = block_operations(sbi);
1519 	if (err)
1520 		goto out;
1521 
1522 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1523 
1524 	f2fs_flush_merged_writes(sbi);
1525 
1526 	/* this is the case of multiple fstrims without any changes */
1527 	if (cpc->reason & CP_DISCARD) {
1528 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1529 			unblock_operations(sbi);
1530 			goto out;
1531 		}
1532 
1533 		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1534 				SIT_I(sbi)->dirty_sentries == 0 &&
1535 				prefree_segments(sbi) == 0) {
1536 			f2fs_flush_sit_entries(sbi, cpc);
1537 			f2fs_clear_prefree_segments(sbi, cpc);
1538 			unblock_operations(sbi);
1539 			goto out;
1540 		}
1541 	}
1542 
1543 	/*
1544 	 * update checkpoint pack index
1545 	 * Increase the version number so that
1546 	 * SIT entries and seg summaries are written at correct place
1547 	 */
1548 	ckpt_ver = cur_cp_version(ckpt);
1549 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1550 
1551 	/* write cached NAT/SIT entries to NAT/SIT area */
1552 	err = f2fs_flush_nat_entries(sbi, cpc);
1553 	if (err)
1554 		goto stop;
1555 
1556 	f2fs_flush_sit_entries(sbi, cpc);
1557 
1558 	/* unlock all the fs_lock[] in do_checkpoint() */
1559 	err = do_checkpoint(sbi, cpc);
1560 	if (err)
1561 		f2fs_release_discard_addrs(sbi);
1562 	else
1563 		f2fs_clear_prefree_segments(sbi, cpc);
1564 stop:
1565 	unblock_operations(sbi);
1566 	stat_inc_cp_count(sbi->stat_info);
1567 
1568 	if (cpc->reason & CP_RECOVERY)
1569 		f2fs_msg(sbi->sb, KERN_NOTICE,
1570 			"checkpoint: version = %llx", ckpt_ver);
1571 
1572 	/* do checkpoint periodically */
1573 	f2fs_update_time(sbi, CP_TIME);
1574 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1575 out:
1576 	mutex_unlock(&sbi->cp_mutex);
1577 	return err;
1578 }
1579 
1580 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1581 {
1582 	int i;
1583 
1584 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1585 		struct inode_management *im = &sbi->im[i];
1586 
1587 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1588 		spin_lock_init(&im->ino_lock);
1589 		INIT_LIST_HEAD(&im->ino_list);
1590 		im->ino_num = 0;
1591 	}
1592 
1593 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1594 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1595 				F2FS_ORPHANS_PER_BLOCK;
1596 }
1597 
1598 int __init f2fs_create_checkpoint_caches(void)
1599 {
1600 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1601 			sizeof(struct ino_entry));
1602 	if (!ino_entry_slab)
1603 		return -ENOMEM;
1604 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1605 			sizeof(struct inode_entry));
1606 	if (!f2fs_inode_entry_slab) {
1607 		kmem_cache_destroy(ino_entry_slab);
1608 		return -ENOMEM;
1609 	}
1610 	return 0;
1611 }
1612 
1613 void f2fs_destroy_checkpoint_caches(void)
1614 {
1615 	kmem_cache_destroy(ino_entry_slab);
1616 	kmem_cache_destroy(f2fs_inode_entry_slab);
1617 }
1618