xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision 5927145e)
1 /*
2  * fs/f2fs/checkpoint.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/bio.h>
13 #include <linux/mpage.h>
14 #include <linux/writeback.h>
15 #include <linux/blkdev.h>
16 #include <linux/f2fs_fs.h>
17 #include <linux/pagevec.h>
18 #include <linux/swap.h>
19 
20 #include "f2fs.h"
21 #include "node.h"
22 #include "segment.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25 
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *inode_entry_slab;
28 
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
30 {
31 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
32 	if (!end_io)
33 		f2fs_flush_merged_writes(sbi);
34 }
35 
36 /*
37  * We guarantee no failure on the returned page.
38  */
39 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
40 {
41 	struct address_space *mapping = META_MAPPING(sbi);
42 	struct page *page = NULL;
43 repeat:
44 	page = f2fs_grab_cache_page(mapping, index, false);
45 	if (!page) {
46 		cond_resched();
47 		goto repeat;
48 	}
49 	f2fs_wait_on_page_writeback(page, META, true);
50 	if (!PageUptodate(page))
51 		SetPageUptodate(page);
52 	return page;
53 }
54 
55 /*
56  * We guarantee no failure on the returned page.
57  */
58 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
59 							bool is_meta)
60 {
61 	struct address_space *mapping = META_MAPPING(sbi);
62 	struct page *page;
63 	struct f2fs_io_info fio = {
64 		.sbi = sbi,
65 		.type = META,
66 		.op = REQ_OP_READ,
67 		.op_flags = REQ_META | REQ_PRIO,
68 		.old_blkaddr = index,
69 		.new_blkaddr = index,
70 		.encrypted_page = NULL,
71 	};
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 	if (f2fs_submit_page_bio(&fio)) {
87 		f2fs_put_page(page, 1);
88 		goto repeat;
89 	}
90 
91 	lock_page(page);
92 	if (unlikely(page->mapping != mapping)) {
93 		f2fs_put_page(page, 1);
94 		goto repeat;
95 	}
96 
97 	/*
98 	 * if there is any IO error when accessing device, make our filesystem
99 	 * readonly and make sure do not write checkpoint with non-uptodate
100 	 * meta page.
101 	 */
102 	if (unlikely(!PageUptodate(page)))
103 		f2fs_stop_checkpoint(sbi, false);
104 out:
105 	return page;
106 }
107 
108 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
109 {
110 	return __get_meta_page(sbi, index, true);
111 }
112 
113 /* for POR only */
114 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
115 {
116 	return __get_meta_page(sbi, index, false);
117 }
118 
119 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
120 {
121 	switch (type) {
122 	case META_NAT:
123 		break;
124 	case META_SIT:
125 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
126 			return false;
127 		break;
128 	case META_SSA:
129 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
130 			blkaddr < SM_I(sbi)->ssa_blkaddr))
131 			return false;
132 		break;
133 	case META_CP:
134 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
135 			blkaddr < __start_cp_addr(sbi)))
136 			return false;
137 		break;
138 	case META_POR:
139 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
140 			blkaddr < MAIN_BLKADDR(sbi)))
141 			return false;
142 		break;
143 	default:
144 		BUG();
145 	}
146 
147 	return true;
148 }
149 
150 /*
151  * Readahead CP/NAT/SIT/SSA pages
152  */
153 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
154 							int type, bool sync)
155 {
156 	struct page *page;
157 	block_t blkno = start;
158 	struct f2fs_io_info fio = {
159 		.sbi = sbi,
160 		.type = META,
161 		.op = REQ_OP_READ,
162 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
163 		.encrypted_page = NULL,
164 		.in_list = false,
165 	};
166 	struct blk_plug plug;
167 
168 	if (unlikely(type == META_POR))
169 		fio.op_flags &= ~REQ_META;
170 
171 	blk_start_plug(&plug);
172 	for (; nrpages-- > 0; blkno++) {
173 
174 		if (!is_valid_blkaddr(sbi, blkno, type))
175 			goto out;
176 
177 		switch (type) {
178 		case META_NAT:
179 			if (unlikely(blkno >=
180 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
181 				blkno = 0;
182 			/* get nat block addr */
183 			fio.new_blkaddr = current_nat_addr(sbi,
184 					blkno * NAT_ENTRY_PER_BLOCK);
185 			break;
186 		case META_SIT:
187 			/* get sit block addr */
188 			fio.new_blkaddr = current_sit_addr(sbi,
189 					blkno * SIT_ENTRY_PER_BLOCK);
190 			break;
191 		case META_SSA:
192 		case META_CP:
193 		case META_POR:
194 			fio.new_blkaddr = blkno;
195 			break;
196 		default:
197 			BUG();
198 		}
199 
200 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
201 						fio.new_blkaddr, false);
202 		if (!page)
203 			continue;
204 		if (PageUptodate(page)) {
205 			f2fs_put_page(page, 1);
206 			continue;
207 		}
208 
209 		fio.page = page;
210 		f2fs_submit_page_bio(&fio);
211 		f2fs_put_page(page, 0);
212 	}
213 out:
214 	blk_finish_plug(&plug);
215 	return blkno - start;
216 }
217 
218 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
219 {
220 	struct page *page;
221 	bool readahead = false;
222 
223 	page = find_get_page(META_MAPPING(sbi), index);
224 	if (!page || !PageUptodate(page))
225 		readahead = true;
226 	f2fs_put_page(page, 0);
227 
228 	if (readahead)
229 		ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
230 }
231 
232 static int __f2fs_write_meta_page(struct page *page,
233 				struct writeback_control *wbc,
234 				enum iostat_type io_type)
235 {
236 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
237 
238 	trace_f2fs_writepage(page, META);
239 
240 	if (unlikely(f2fs_cp_error(sbi))) {
241 		dec_page_count(sbi, F2FS_DIRTY_META);
242 		unlock_page(page);
243 		return 0;
244 	}
245 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
246 		goto redirty_out;
247 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
248 		goto redirty_out;
249 
250 	write_meta_page(sbi, page, io_type);
251 	dec_page_count(sbi, F2FS_DIRTY_META);
252 
253 	if (wbc->for_reclaim)
254 		f2fs_submit_merged_write_cond(sbi, page->mapping->host,
255 						0, page->index, META);
256 
257 	unlock_page(page);
258 
259 	if (unlikely(f2fs_cp_error(sbi)))
260 		f2fs_submit_merged_write(sbi, META);
261 
262 	return 0;
263 
264 redirty_out:
265 	redirty_page_for_writepage(wbc, page);
266 	return AOP_WRITEPAGE_ACTIVATE;
267 }
268 
269 static int f2fs_write_meta_page(struct page *page,
270 				struct writeback_control *wbc)
271 {
272 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
273 }
274 
275 static int f2fs_write_meta_pages(struct address_space *mapping,
276 				struct writeback_control *wbc)
277 {
278 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
279 	long diff, written;
280 
281 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
282 		goto skip_write;
283 
284 	/* collect a number of dirty meta pages and write together */
285 	if (wbc->for_kupdate ||
286 		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
287 		goto skip_write;
288 
289 	/* if locked failed, cp will flush dirty pages instead */
290 	if (!mutex_trylock(&sbi->cp_mutex))
291 		goto skip_write;
292 
293 	trace_f2fs_writepages(mapping->host, wbc, META);
294 	diff = nr_pages_to_write(sbi, META, wbc);
295 	written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
296 	mutex_unlock(&sbi->cp_mutex);
297 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
298 	return 0;
299 
300 skip_write:
301 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
302 	trace_f2fs_writepages(mapping->host, wbc, META);
303 	return 0;
304 }
305 
306 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
307 				long nr_to_write, enum iostat_type io_type)
308 {
309 	struct address_space *mapping = META_MAPPING(sbi);
310 	pgoff_t index = 0, prev = ULONG_MAX;
311 	struct pagevec pvec;
312 	long nwritten = 0;
313 	int nr_pages;
314 	struct writeback_control wbc = {
315 		.for_reclaim = 0,
316 	};
317 	struct blk_plug plug;
318 
319 	pagevec_init(&pvec);
320 
321 	blk_start_plug(&plug);
322 
323 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
324 				PAGECACHE_TAG_DIRTY))) {
325 		int i;
326 
327 		for (i = 0; i < nr_pages; i++) {
328 			struct page *page = pvec.pages[i];
329 
330 			if (prev == ULONG_MAX)
331 				prev = page->index - 1;
332 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
333 				pagevec_release(&pvec);
334 				goto stop;
335 			}
336 
337 			lock_page(page);
338 
339 			if (unlikely(page->mapping != mapping)) {
340 continue_unlock:
341 				unlock_page(page);
342 				continue;
343 			}
344 			if (!PageDirty(page)) {
345 				/* someone wrote it for us */
346 				goto continue_unlock;
347 			}
348 
349 			f2fs_wait_on_page_writeback(page, META, true);
350 
351 			BUG_ON(PageWriteback(page));
352 			if (!clear_page_dirty_for_io(page))
353 				goto continue_unlock;
354 
355 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
356 				unlock_page(page);
357 				break;
358 			}
359 			nwritten++;
360 			prev = page->index;
361 			if (unlikely(nwritten >= nr_to_write))
362 				break;
363 		}
364 		pagevec_release(&pvec);
365 		cond_resched();
366 	}
367 stop:
368 	if (nwritten)
369 		f2fs_submit_merged_write(sbi, type);
370 
371 	blk_finish_plug(&plug);
372 
373 	return nwritten;
374 }
375 
376 static int f2fs_set_meta_page_dirty(struct page *page)
377 {
378 	trace_f2fs_set_page_dirty(page, META);
379 
380 	if (!PageUptodate(page))
381 		SetPageUptodate(page);
382 	if (!PageDirty(page)) {
383 		f2fs_set_page_dirty_nobuffers(page);
384 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
385 		SetPagePrivate(page);
386 		f2fs_trace_pid(page);
387 		return 1;
388 	}
389 	return 0;
390 }
391 
392 const struct address_space_operations f2fs_meta_aops = {
393 	.writepage	= f2fs_write_meta_page,
394 	.writepages	= f2fs_write_meta_pages,
395 	.set_page_dirty	= f2fs_set_meta_page_dirty,
396 	.invalidatepage = f2fs_invalidate_page,
397 	.releasepage	= f2fs_release_page,
398 #ifdef CONFIG_MIGRATION
399 	.migratepage    = f2fs_migrate_page,
400 #endif
401 };
402 
403 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
404 						unsigned int devidx, int type)
405 {
406 	struct inode_management *im = &sbi->im[type];
407 	struct ino_entry *e, *tmp;
408 
409 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
410 
411 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
412 
413 	spin_lock(&im->ino_lock);
414 	e = radix_tree_lookup(&im->ino_root, ino);
415 	if (!e) {
416 		e = tmp;
417 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
418 			f2fs_bug_on(sbi, 1);
419 
420 		memset(e, 0, sizeof(struct ino_entry));
421 		e->ino = ino;
422 
423 		list_add_tail(&e->list, &im->ino_list);
424 		if (type != ORPHAN_INO)
425 			im->ino_num++;
426 	}
427 
428 	if (type == FLUSH_INO)
429 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
430 
431 	spin_unlock(&im->ino_lock);
432 	radix_tree_preload_end();
433 
434 	if (e != tmp)
435 		kmem_cache_free(ino_entry_slab, tmp);
436 }
437 
438 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
439 {
440 	struct inode_management *im = &sbi->im[type];
441 	struct ino_entry *e;
442 
443 	spin_lock(&im->ino_lock);
444 	e = radix_tree_lookup(&im->ino_root, ino);
445 	if (e) {
446 		list_del(&e->list);
447 		radix_tree_delete(&im->ino_root, ino);
448 		im->ino_num--;
449 		spin_unlock(&im->ino_lock);
450 		kmem_cache_free(ino_entry_slab, e);
451 		return;
452 	}
453 	spin_unlock(&im->ino_lock);
454 }
455 
456 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
457 {
458 	/* add new dirty ino entry into list */
459 	__add_ino_entry(sbi, ino, 0, type);
460 }
461 
462 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
463 {
464 	/* remove dirty ino entry from list */
465 	__remove_ino_entry(sbi, ino, type);
466 }
467 
468 /* mode should be APPEND_INO or UPDATE_INO */
469 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
470 {
471 	struct inode_management *im = &sbi->im[mode];
472 	struct ino_entry *e;
473 
474 	spin_lock(&im->ino_lock);
475 	e = radix_tree_lookup(&im->ino_root, ino);
476 	spin_unlock(&im->ino_lock);
477 	return e ? true : false;
478 }
479 
480 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
481 {
482 	struct ino_entry *e, *tmp;
483 	int i;
484 
485 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
486 		struct inode_management *im = &sbi->im[i];
487 
488 		spin_lock(&im->ino_lock);
489 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
490 			list_del(&e->list);
491 			radix_tree_delete(&im->ino_root, e->ino);
492 			kmem_cache_free(ino_entry_slab, e);
493 			im->ino_num--;
494 		}
495 		spin_unlock(&im->ino_lock);
496 	}
497 }
498 
499 void set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
500 					unsigned int devidx, int type)
501 {
502 	__add_ino_entry(sbi, ino, devidx, type);
503 }
504 
505 bool is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
506 					unsigned int devidx, int type)
507 {
508 	struct inode_management *im = &sbi->im[type];
509 	struct ino_entry *e;
510 	bool is_dirty = false;
511 
512 	spin_lock(&im->ino_lock);
513 	e = radix_tree_lookup(&im->ino_root, ino);
514 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
515 		is_dirty = true;
516 	spin_unlock(&im->ino_lock);
517 	return is_dirty;
518 }
519 
520 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
521 {
522 	struct inode_management *im = &sbi->im[ORPHAN_INO];
523 	int err = 0;
524 
525 	spin_lock(&im->ino_lock);
526 
527 #ifdef CONFIG_F2FS_FAULT_INJECTION
528 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
529 		spin_unlock(&im->ino_lock);
530 		f2fs_show_injection_info(FAULT_ORPHAN);
531 		return -ENOSPC;
532 	}
533 #endif
534 	if (unlikely(im->ino_num >= sbi->max_orphans))
535 		err = -ENOSPC;
536 	else
537 		im->ino_num++;
538 	spin_unlock(&im->ino_lock);
539 
540 	return err;
541 }
542 
543 void release_orphan_inode(struct f2fs_sb_info *sbi)
544 {
545 	struct inode_management *im = &sbi->im[ORPHAN_INO];
546 
547 	spin_lock(&im->ino_lock);
548 	f2fs_bug_on(sbi, im->ino_num == 0);
549 	im->ino_num--;
550 	spin_unlock(&im->ino_lock);
551 }
552 
553 void add_orphan_inode(struct inode *inode)
554 {
555 	/* add new orphan ino entry into list */
556 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
557 	update_inode_page(inode);
558 }
559 
560 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
561 {
562 	/* remove orphan entry from orphan list */
563 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
564 }
565 
566 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
567 {
568 	struct inode *inode;
569 	struct node_info ni;
570 	int err = acquire_orphan_inode(sbi);
571 
572 	if (err) {
573 		set_sbi_flag(sbi, SBI_NEED_FSCK);
574 		f2fs_msg(sbi->sb, KERN_WARNING,
575 				"%s: orphan failed (ino=%x), run fsck to fix.",
576 				__func__, ino);
577 		return err;
578 	}
579 
580 	__add_ino_entry(sbi, ino, 0, ORPHAN_INO);
581 
582 	inode = f2fs_iget_retry(sbi->sb, ino);
583 	if (IS_ERR(inode)) {
584 		/*
585 		 * there should be a bug that we can't find the entry
586 		 * to orphan inode.
587 		 */
588 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
589 		return PTR_ERR(inode);
590 	}
591 
592 	clear_nlink(inode);
593 
594 	/* truncate all the data during iput */
595 	iput(inode);
596 
597 	get_node_info(sbi, ino, &ni);
598 
599 	/* ENOMEM was fully retried in f2fs_evict_inode. */
600 	if (ni.blk_addr != NULL_ADDR) {
601 		set_sbi_flag(sbi, SBI_NEED_FSCK);
602 		f2fs_msg(sbi->sb, KERN_WARNING,
603 			"%s: orphan failed (ino=%x) by kernel, retry mount.",
604 				__func__, ino);
605 		return -EIO;
606 	}
607 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
608 	return 0;
609 }
610 
611 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
612 {
613 	block_t start_blk, orphan_blocks, i, j;
614 	unsigned int s_flags = sbi->sb->s_flags;
615 	int err = 0;
616 #ifdef CONFIG_QUOTA
617 	int quota_enabled;
618 #endif
619 
620 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
621 		return 0;
622 
623 	if (s_flags & SB_RDONLY) {
624 		f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
625 		sbi->sb->s_flags &= ~SB_RDONLY;
626 	}
627 
628 #ifdef CONFIG_QUOTA
629 	/* Needed for iput() to work correctly and not trash data */
630 	sbi->sb->s_flags |= SB_ACTIVE;
631 
632 	/* Turn on quotas so that they are updated correctly */
633 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
634 #endif
635 
636 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
637 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
638 
639 	ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
640 
641 	for (i = 0; i < orphan_blocks; i++) {
642 		struct page *page = get_meta_page(sbi, start_blk + i);
643 		struct f2fs_orphan_block *orphan_blk;
644 
645 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
646 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
647 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
648 			err = recover_orphan_inode(sbi, ino);
649 			if (err) {
650 				f2fs_put_page(page, 1);
651 				goto out;
652 			}
653 		}
654 		f2fs_put_page(page, 1);
655 	}
656 	/* clear Orphan Flag */
657 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
658 out:
659 #ifdef CONFIG_QUOTA
660 	/* Turn quotas off */
661 	if (quota_enabled)
662 		f2fs_quota_off_umount(sbi->sb);
663 #endif
664 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
665 
666 	return err;
667 }
668 
669 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
670 {
671 	struct list_head *head;
672 	struct f2fs_orphan_block *orphan_blk = NULL;
673 	unsigned int nentries = 0;
674 	unsigned short index = 1;
675 	unsigned short orphan_blocks;
676 	struct page *page = NULL;
677 	struct ino_entry *orphan = NULL;
678 	struct inode_management *im = &sbi->im[ORPHAN_INO];
679 
680 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
681 
682 	/*
683 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
684 	 * orphan inode operations are covered under f2fs_lock_op().
685 	 * And, spin_lock should be avoided due to page operations below.
686 	 */
687 	head = &im->ino_list;
688 
689 	/* loop for each orphan inode entry and write them in Jornal block */
690 	list_for_each_entry(orphan, head, list) {
691 		if (!page) {
692 			page = grab_meta_page(sbi, start_blk++);
693 			orphan_blk =
694 				(struct f2fs_orphan_block *)page_address(page);
695 			memset(orphan_blk, 0, sizeof(*orphan_blk));
696 		}
697 
698 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
699 
700 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
701 			/*
702 			 * an orphan block is full of 1020 entries,
703 			 * then we need to flush current orphan blocks
704 			 * and bring another one in memory
705 			 */
706 			orphan_blk->blk_addr = cpu_to_le16(index);
707 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
708 			orphan_blk->entry_count = cpu_to_le32(nentries);
709 			set_page_dirty(page);
710 			f2fs_put_page(page, 1);
711 			index++;
712 			nentries = 0;
713 			page = NULL;
714 		}
715 	}
716 
717 	if (page) {
718 		orphan_blk->blk_addr = cpu_to_le16(index);
719 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
720 		orphan_blk->entry_count = cpu_to_le32(nentries);
721 		set_page_dirty(page);
722 		f2fs_put_page(page, 1);
723 	}
724 }
725 
726 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
727 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
728 		unsigned long long *version)
729 {
730 	unsigned long blk_size = sbi->blocksize;
731 	size_t crc_offset = 0;
732 	__u32 crc = 0;
733 
734 	*cp_page = get_meta_page(sbi, cp_addr);
735 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
736 
737 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
738 	if (crc_offset > (blk_size - sizeof(__le32))) {
739 		f2fs_msg(sbi->sb, KERN_WARNING,
740 			"invalid crc_offset: %zu", crc_offset);
741 		return -EINVAL;
742 	}
743 
744 	crc = cur_cp_crc(*cp_block);
745 	if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
746 		f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
747 		return -EINVAL;
748 	}
749 
750 	*version = cur_cp_version(*cp_block);
751 	return 0;
752 }
753 
754 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
755 				block_t cp_addr, unsigned long long *version)
756 {
757 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
758 	struct f2fs_checkpoint *cp_block = NULL;
759 	unsigned long long cur_version = 0, pre_version = 0;
760 	int err;
761 
762 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
763 					&cp_page_1, version);
764 	if (err)
765 		goto invalid_cp1;
766 	pre_version = *version;
767 
768 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
769 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
770 					&cp_page_2, version);
771 	if (err)
772 		goto invalid_cp2;
773 	cur_version = *version;
774 
775 	if (cur_version == pre_version) {
776 		*version = cur_version;
777 		f2fs_put_page(cp_page_2, 1);
778 		return cp_page_1;
779 	}
780 invalid_cp2:
781 	f2fs_put_page(cp_page_2, 1);
782 invalid_cp1:
783 	f2fs_put_page(cp_page_1, 1);
784 	return NULL;
785 }
786 
787 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
788 {
789 	struct f2fs_checkpoint *cp_block;
790 	struct f2fs_super_block *fsb = sbi->raw_super;
791 	struct page *cp1, *cp2, *cur_page;
792 	unsigned long blk_size = sbi->blocksize;
793 	unsigned long long cp1_version = 0, cp2_version = 0;
794 	unsigned long long cp_start_blk_no;
795 	unsigned int cp_blks = 1 + __cp_payload(sbi);
796 	block_t cp_blk_no;
797 	int i;
798 
799 	sbi->ckpt = f2fs_kzalloc(sbi, cp_blks * blk_size, GFP_KERNEL);
800 	if (!sbi->ckpt)
801 		return -ENOMEM;
802 	/*
803 	 * Finding out valid cp block involves read both
804 	 * sets( cp pack1 and cp pack 2)
805 	 */
806 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
807 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
808 
809 	/* The second checkpoint pack should start at the next segment */
810 	cp_start_blk_no += ((unsigned long long)1) <<
811 				le32_to_cpu(fsb->log_blocks_per_seg);
812 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
813 
814 	if (cp1 && cp2) {
815 		if (ver_after(cp2_version, cp1_version))
816 			cur_page = cp2;
817 		else
818 			cur_page = cp1;
819 	} else if (cp1) {
820 		cur_page = cp1;
821 	} else if (cp2) {
822 		cur_page = cp2;
823 	} else {
824 		goto fail_no_cp;
825 	}
826 
827 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
828 	memcpy(sbi->ckpt, cp_block, blk_size);
829 
830 	/* Sanity checking of checkpoint */
831 	if (sanity_check_ckpt(sbi))
832 		goto free_fail_no_cp;
833 
834 	if (cur_page == cp1)
835 		sbi->cur_cp_pack = 1;
836 	else
837 		sbi->cur_cp_pack = 2;
838 
839 	if (cp_blks <= 1)
840 		goto done;
841 
842 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
843 	if (cur_page == cp2)
844 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
845 
846 	for (i = 1; i < cp_blks; i++) {
847 		void *sit_bitmap_ptr;
848 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
849 
850 		cur_page = get_meta_page(sbi, cp_blk_no + i);
851 		sit_bitmap_ptr = page_address(cur_page);
852 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
853 		f2fs_put_page(cur_page, 1);
854 	}
855 done:
856 	f2fs_put_page(cp1, 1);
857 	f2fs_put_page(cp2, 1);
858 	return 0;
859 
860 free_fail_no_cp:
861 	f2fs_put_page(cp1, 1);
862 	f2fs_put_page(cp2, 1);
863 fail_no_cp:
864 	kfree(sbi->ckpt);
865 	return -EINVAL;
866 }
867 
868 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
869 {
870 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
871 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
872 
873 	if (is_inode_flag_set(inode, flag))
874 		return;
875 
876 	set_inode_flag(inode, flag);
877 	if (!f2fs_is_volatile_file(inode))
878 		list_add_tail(&F2FS_I(inode)->dirty_list,
879 						&sbi->inode_list[type]);
880 	stat_inc_dirty_inode(sbi, type);
881 }
882 
883 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
884 {
885 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
886 
887 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
888 		return;
889 
890 	list_del_init(&F2FS_I(inode)->dirty_list);
891 	clear_inode_flag(inode, flag);
892 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
893 }
894 
895 void update_dirty_page(struct inode *inode, struct page *page)
896 {
897 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
898 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
899 
900 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
901 			!S_ISLNK(inode->i_mode))
902 		return;
903 
904 	spin_lock(&sbi->inode_lock[type]);
905 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
906 		__add_dirty_inode(inode, type);
907 	inode_inc_dirty_pages(inode);
908 	spin_unlock(&sbi->inode_lock[type]);
909 
910 	SetPagePrivate(page);
911 	f2fs_trace_pid(page);
912 }
913 
914 void remove_dirty_inode(struct inode *inode)
915 {
916 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
917 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
918 
919 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
920 			!S_ISLNK(inode->i_mode))
921 		return;
922 
923 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
924 		return;
925 
926 	spin_lock(&sbi->inode_lock[type]);
927 	__remove_dirty_inode(inode, type);
928 	spin_unlock(&sbi->inode_lock[type]);
929 }
930 
931 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
932 {
933 	struct list_head *head;
934 	struct inode *inode;
935 	struct f2fs_inode_info *fi;
936 	bool is_dir = (type == DIR_INODE);
937 	unsigned long ino = 0;
938 
939 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
940 				get_pages(sbi, is_dir ?
941 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
942 retry:
943 	if (unlikely(f2fs_cp_error(sbi)))
944 		return -EIO;
945 
946 	spin_lock(&sbi->inode_lock[type]);
947 
948 	head = &sbi->inode_list[type];
949 	if (list_empty(head)) {
950 		spin_unlock(&sbi->inode_lock[type]);
951 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
952 				get_pages(sbi, is_dir ?
953 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
954 		return 0;
955 	}
956 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
957 	inode = igrab(&fi->vfs_inode);
958 	spin_unlock(&sbi->inode_lock[type]);
959 	if (inode) {
960 		unsigned long cur_ino = inode->i_ino;
961 
962 		if (is_dir)
963 			F2FS_I(inode)->cp_task = current;
964 
965 		filemap_fdatawrite(inode->i_mapping);
966 
967 		if (is_dir)
968 			F2FS_I(inode)->cp_task = NULL;
969 
970 		iput(inode);
971 		/* We need to give cpu to another writers. */
972 		if (ino == cur_ino) {
973 			congestion_wait(BLK_RW_ASYNC, HZ/50);
974 			cond_resched();
975 		} else {
976 			ino = cur_ino;
977 		}
978 	} else {
979 		/*
980 		 * We should submit bio, since it exists several
981 		 * wribacking dentry pages in the freeing inode.
982 		 */
983 		f2fs_submit_merged_write(sbi, DATA);
984 		cond_resched();
985 	}
986 	goto retry;
987 }
988 
989 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
990 {
991 	struct list_head *head = &sbi->inode_list[DIRTY_META];
992 	struct inode *inode;
993 	struct f2fs_inode_info *fi;
994 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
995 
996 	while (total--) {
997 		if (unlikely(f2fs_cp_error(sbi)))
998 			return -EIO;
999 
1000 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1001 		if (list_empty(head)) {
1002 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1003 			return 0;
1004 		}
1005 		fi = list_first_entry(head, struct f2fs_inode_info,
1006 							gdirty_list);
1007 		inode = igrab(&fi->vfs_inode);
1008 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1009 		if (inode) {
1010 			sync_inode_metadata(inode, 0);
1011 
1012 			/* it's on eviction */
1013 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1014 				update_inode_page(inode);
1015 			iput(inode);
1016 		}
1017 	}
1018 	return 0;
1019 }
1020 
1021 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1022 {
1023 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1024 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1025 	nid_t last_nid = nm_i->next_scan_nid;
1026 
1027 	next_free_nid(sbi, &last_nid);
1028 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1029 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1030 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1031 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1032 }
1033 
1034 /*
1035  * Freeze all the FS-operations for checkpoint.
1036  */
1037 static int block_operations(struct f2fs_sb_info *sbi)
1038 {
1039 	struct writeback_control wbc = {
1040 		.sync_mode = WB_SYNC_ALL,
1041 		.nr_to_write = LONG_MAX,
1042 		.for_reclaim = 0,
1043 	};
1044 	struct blk_plug plug;
1045 	int err = 0;
1046 
1047 	blk_start_plug(&plug);
1048 
1049 retry_flush_dents:
1050 	f2fs_lock_all(sbi);
1051 	/* write all the dirty dentry pages */
1052 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1053 		f2fs_unlock_all(sbi);
1054 		err = sync_dirty_inodes(sbi, DIR_INODE);
1055 		if (err)
1056 			goto out;
1057 		cond_resched();
1058 		goto retry_flush_dents;
1059 	}
1060 
1061 	/*
1062 	 * POR: we should ensure that there are no dirty node pages
1063 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1064 	 */
1065 	down_write(&sbi->node_change);
1066 
1067 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1068 		up_write(&sbi->node_change);
1069 		f2fs_unlock_all(sbi);
1070 		err = f2fs_sync_inode_meta(sbi);
1071 		if (err)
1072 			goto out;
1073 		cond_resched();
1074 		goto retry_flush_dents;
1075 	}
1076 
1077 retry_flush_nodes:
1078 	down_write(&sbi->node_write);
1079 
1080 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1081 		up_write(&sbi->node_write);
1082 		err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1083 		if (err) {
1084 			up_write(&sbi->node_change);
1085 			f2fs_unlock_all(sbi);
1086 			goto out;
1087 		}
1088 		cond_resched();
1089 		goto retry_flush_nodes;
1090 	}
1091 
1092 	/*
1093 	 * sbi->node_change is used only for AIO write_begin path which produces
1094 	 * dirty node blocks and some checkpoint values by block allocation.
1095 	 */
1096 	__prepare_cp_block(sbi);
1097 	up_write(&sbi->node_change);
1098 out:
1099 	blk_finish_plug(&plug);
1100 	return err;
1101 }
1102 
1103 static void unblock_operations(struct f2fs_sb_info *sbi)
1104 {
1105 	up_write(&sbi->node_write);
1106 	f2fs_unlock_all(sbi);
1107 }
1108 
1109 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1110 {
1111 	DEFINE_WAIT(wait);
1112 
1113 	for (;;) {
1114 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1115 
1116 		if (!get_pages(sbi, F2FS_WB_CP_DATA))
1117 			break;
1118 
1119 		io_schedule_timeout(5*HZ);
1120 	}
1121 	finish_wait(&sbi->cp_wait, &wait);
1122 }
1123 
1124 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1125 {
1126 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1127 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1128 	unsigned long flags;
1129 
1130 	spin_lock_irqsave(&sbi->cp_lock, flags);
1131 
1132 	if ((cpc->reason & CP_UMOUNT) &&
1133 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1134 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1135 		disable_nat_bits(sbi, false);
1136 
1137 	if (cpc->reason & CP_TRIMMED)
1138 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1139 
1140 	if (cpc->reason & CP_UMOUNT)
1141 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1142 	else
1143 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1144 
1145 	if (cpc->reason & CP_FASTBOOT)
1146 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1147 	else
1148 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1149 
1150 	if (orphan_num)
1151 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1152 	else
1153 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1154 
1155 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1156 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1157 
1158 	/* set this flag to activate crc|cp_ver for recovery */
1159 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1160 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1161 
1162 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1163 }
1164 
1165 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1166 {
1167 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1168 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1169 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1170 	block_t start_blk;
1171 	unsigned int data_sum_blocks, orphan_blocks;
1172 	__u32 crc32 = 0;
1173 	int i;
1174 	int cp_payload_blks = __cp_payload(sbi);
1175 	struct super_block *sb = sbi->sb;
1176 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1177 	u64 kbytes_written;
1178 	int err;
1179 
1180 	/* Flush all the NAT/SIT pages */
1181 	while (get_pages(sbi, F2FS_DIRTY_META)) {
1182 		sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1183 		if (unlikely(f2fs_cp_error(sbi)))
1184 			return -EIO;
1185 	}
1186 
1187 	/*
1188 	 * modify checkpoint
1189 	 * version number is already updated
1190 	 */
1191 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1192 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1193 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1194 		ckpt->cur_node_segno[i] =
1195 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1196 		ckpt->cur_node_blkoff[i] =
1197 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1198 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1199 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1200 	}
1201 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1202 		ckpt->cur_data_segno[i] =
1203 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1204 		ckpt->cur_data_blkoff[i] =
1205 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1206 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1207 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1208 	}
1209 
1210 	/* 2 cp  + n data seg summary + orphan inode blocks */
1211 	data_sum_blocks = npages_for_summary_flush(sbi, false);
1212 	spin_lock_irqsave(&sbi->cp_lock, flags);
1213 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1214 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1215 	else
1216 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1217 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1218 
1219 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1220 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1221 			orphan_blocks);
1222 
1223 	if (__remain_node_summaries(cpc->reason))
1224 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1225 				cp_payload_blks + data_sum_blocks +
1226 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1227 	else
1228 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1229 				cp_payload_blks + data_sum_blocks +
1230 				orphan_blocks);
1231 
1232 	/* update ckpt flag for checkpoint */
1233 	update_ckpt_flags(sbi, cpc);
1234 
1235 	/* update SIT/NAT bitmap */
1236 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1237 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1238 
1239 	crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1240 	*((__le32 *)((unsigned char *)ckpt +
1241 				le32_to_cpu(ckpt->checksum_offset)))
1242 				= cpu_to_le32(crc32);
1243 
1244 	start_blk = __start_cp_next_addr(sbi);
1245 
1246 	/* write nat bits */
1247 	if (enabled_nat_bits(sbi, cpc)) {
1248 		__u64 cp_ver = cur_cp_version(ckpt);
1249 		block_t blk;
1250 
1251 		cp_ver |= ((__u64)crc32 << 32);
1252 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1253 
1254 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1255 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1256 			update_meta_page(sbi, nm_i->nat_bits +
1257 					(i << F2FS_BLKSIZE_BITS), blk + i);
1258 
1259 		/* Flush all the NAT BITS pages */
1260 		while (get_pages(sbi, F2FS_DIRTY_META)) {
1261 			sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1262 			if (unlikely(f2fs_cp_error(sbi)))
1263 				return -EIO;
1264 		}
1265 	}
1266 
1267 	/* need to wait for end_io results */
1268 	wait_on_all_pages_writeback(sbi);
1269 	if (unlikely(f2fs_cp_error(sbi)))
1270 		return -EIO;
1271 
1272 	/* flush all device cache */
1273 	err = f2fs_flush_device_cache(sbi);
1274 	if (err)
1275 		return err;
1276 
1277 	/* write out checkpoint buffer at block 0 */
1278 	update_meta_page(sbi, ckpt, start_blk++);
1279 
1280 	for (i = 1; i < 1 + cp_payload_blks; i++)
1281 		update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1282 							start_blk++);
1283 
1284 	if (orphan_num) {
1285 		write_orphan_inodes(sbi, start_blk);
1286 		start_blk += orphan_blocks;
1287 	}
1288 
1289 	write_data_summaries(sbi, start_blk);
1290 	start_blk += data_sum_blocks;
1291 
1292 	/* Record write statistics in the hot node summary */
1293 	kbytes_written = sbi->kbytes_written;
1294 	if (sb->s_bdev->bd_part)
1295 		kbytes_written += BD_PART_WRITTEN(sbi);
1296 
1297 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1298 
1299 	if (__remain_node_summaries(cpc->reason)) {
1300 		write_node_summaries(sbi, start_blk);
1301 		start_blk += NR_CURSEG_NODE_TYPE;
1302 	}
1303 
1304 	/* writeout checkpoint block */
1305 	update_meta_page(sbi, ckpt, start_blk);
1306 
1307 	/* wait for previous submitted node/meta pages writeback */
1308 	wait_on_all_pages_writeback(sbi);
1309 
1310 	if (unlikely(f2fs_cp_error(sbi)))
1311 		return -EIO;
1312 
1313 	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1314 	filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1315 
1316 	/* update user_block_counts */
1317 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1318 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1319 
1320 	/* Here, we only have one bio having CP pack */
1321 	sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1322 
1323 	/* wait for previous submitted meta pages writeback */
1324 	wait_on_all_pages_writeback(sbi);
1325 
1326 	release_ino_entry(sbi, false);
1327 
1328 	if (unlikely(f2fs_cp_error(sbi)))
1329 		return -EIO;
1330 
1331 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1332 	clear_sbi_flag(sbi, SBI_NEED_CP);
1333 	__set_cp_next_pack(sbi);
1334 
1335 	/*
1336 	 * redirty superblock if metadata like node page or inode cache is
1337 	 * updated during writing checkpoint.
1338 	 */
1339 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1340 			get_pages(sbi, F2FS_DIRTY_IMETA))
1341 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1342 
1343 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1344 
1345 	return 0;
1346 }
1347 
1348 /*
1349  * We guarantee that this checkpoint procedure will not fail.
1350  */
1351 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1352 {
1353 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1354 	unsigned long long ckpt_ver;
1355 	int err = 0;
1356 
1357 	mutex_lock(&sbi->cp_mutex);
1358 
1359 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1360 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1361 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1362 		goto out;
1363 	if (unlikely(f2fs_cp_error(sbi))) {
1364 		err = -EIO;
1365 		goto out;
1366 	}
1367 	if (f2fs_readonly(sbi->sb)) {
1368 		err = -EROFS;
1369 		goto out;
1370 	}
1371 
1372 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1373 
1374 	err = block_operations(sbi);
1375 	if (err)
1376 		goto out;
1377 
1378 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1379 
1380 	f2fs_flush_merged_writes(sbi);
1381 
1382 	/* this is the case of multiple fstrims without any changes */
1383 	if (cpc->reason & CP_DISCARD) {
1384 		if (!exist_trim_candidates(sbi, cpc)) {
1385 			unblock_operations(sbi);
1386 			goto out;
1387 		}
1388 
1389 		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1390 				SIT_I(sbi)->dirty_sentries == 0 &&
1391 				prefree_segments(sbi) == 0) {
1392 			flush_sit_entries(sbi, cpc);
1393 			clear_prefree_segments(sbi, cpc);
1394 			unblock_operations(sbi);
1395 			goto out;
1396 		}
1397 	}
1398 
1399 	/*
1400 	 * update checkpoint pack index
1401 	 * Increase the version number so that
1402 	 * SIT entries and seg summaries are written at correct place
1403 	 */
1404 	ckpt_ver = cur_cp_version(ckpt);
1405 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1406 
1407 	/* write cached NAT/SIT entries to NAT/SIT area */
1408 	flush_nat_entries(sbi, cpc);
1409 	flush_sit_entries(sbi, cpc);
1410 
1411 	/* unlock all the fs_lock[] in do_checkpoint() */
1412 	err = do_checkpoint(sbi, cpc);
1413 	if (err)
1414 		release_discard_addrs(sbi);
1415 	else
1416 		clear_prefree_segments(sbi, cpc);
1417 
1418 	unblock_operations(sbi);
1419 	stat_inc_cp_count(sbi->stat_info);
1420 
1421 	if (cpc->reason & CP_RECOVERY)
1422 		f2fs_msg(sbi->sb, KERN_NOTICE,
1423 			"checkpoint: version = %llx", ckpt_ver);
1424 
1425 	/* do checkpoint periodically */
1426 	f2fs_update_time(sbi, CP_TIME);
1427 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1428 out:
1429 	mutex_unlock(&sbi->cp_mutex);
1430 	return err;
1431 }
1432 
1433 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1434 {
1435 	int i;
1436 
1437 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1438 		struct inode_management *im = &sbi->im[i];
1439 
1440 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1441 		spin_lock_init(&im->ino_lock);
1442 		INIT_LIST_HEAD(&im->ino_list);
1443 		im->ino_num = 0;
1444 	}
1445 
1446 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1447 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1448 				F2FS_ORPHANS_PER_BLOCK;
1449 }
1450 
1451 int __init create_checkpoint_caches(void)
1452 {
1453 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1454 			sizeof(struct ino_entry));
1455 	if (!ino_entry_slab)
1456 		return -ENOMEM;
1457 	inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1458 			sizeof(struct inode_entry));
1459 	if (!inode_entry_slab) {
1460 		kmem_cache_destroy(ino_entry_slab);
1461 		return -ENOMEM;
1462 	}
1463 	return 0;
1464 }
1465 
1466 void destroy_checkpoint_caches(void)
1467 {
1468 	kmem_cache_destroy(ino_entry_slab);
1469 	kmem_cache_destroy(inode_entry_slab);
1470 }
1471