xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision f5005f78)
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 /*
30  * We guarantee no failure on the returned page.
31  */
32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
33 {
34 	struct address_space *mapping = META_MAPPING(sbi);
35 	struct page *page = NULL;
36 repeat:
37 	page = grab_cache_page(mapping, index);
38 	if (!page) {
39 		cond_resched();
40 		goto repeat;
41 	}
42 	f2fs_wait_on_page_writeback(page, META);
43 	SetPageUptodate(page);
44 	return page;
45 }
46 
47 /*
48  * We guarantee no failure on the returned page.
49  */
50 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
51 {
52 	struct address_space *mapping = META_MAPPING(sbi);
53 	struct page *page;
54 	struct f2fs_io_info fio = {
55 		.sbi = sbi,
56 		.type = META,
57 		.rw = READ_SYNC | REQ_META | REQ_PRIO,
58 		.blk_addr = index,
59 		.encrypted_page = NULL,
60 	};
61 repeat:
62 	page = grab_cache_page(mapping, index);
63 	if (!page) {
64 		cond_resched();
65 		goto repeat;
66 	}
67 	if (PageUptodate(page))
68 		goto out;
69 
70 	fio.page = page;
71 
72 	if (f2fs_submit_page_bio(&fio))
73 		goto repeat;
74 
75 	lock_page(page);
76 	if (unlikely(page->mapping != mapping)) {
77 		f2fs_put_page(page, 1);
78 		goto repeat;
79 	}
80 out:
81 	return page;
82 }
83 
84 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
85 {
86 	switch (type) {
87 	case META_NAT:
88 		break;
89 	case META_SIT:
90 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
91 			return false;
92 		break;
93 	case META_SSA:
94 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
95 			blkaddr < SM_I(sbi)->ssa_blkaddr))
96 			return false;
97 		break;
98 	case META_CP:
99 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
100 			blkaddr < __start_cp_addr(sbi)))
101 			return false;
102 		break;
103 	case META_POR:
104 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
105 			blkaddr < MAIN_BLKADDR(sbi)))
106 			return false;
107 		break;
108 	default:
109 		BUG();
110 	}
111 
112 	return true;
113 }
114 
115 /*
116  * Readahead CP/NAT/SIT/SSA pages
117  */
118 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
119 {
120 	block_t prev_blk_addr = 0;
121 	struct page *page;
122 	block_t blkno = start;
123 	struct f2fs_io_info fio = {
124 		.sbi = sbi,
125 		.type = META,
126 		.rw = READ_SYNC | REQ_META | REQ_PRIO,
127 		.encrypted_page = NULL,
128 	};
129 
130 	for (; nrpages-- > 0; blkno++) {
131 
132 		if (!is_valid_blkaddr(sbi, blkno, type))
133 			goto out;
134 
135 		switch (type) {
136 		case META_NAT:
137 			if (unlikely(blkno >=
138 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
139 				blkno = 0;
140 			/* get nat block addr */
141 			fio.blk_addr = current_nat_addr(sbi,
142 					blkno * NAT_ENTRY_PER_BLOCK);
143 			break;
144 		case META_SIT:
145 			/* get sit block addr */
146 			fio.blk_addr = current_sit_addr(sbi,
147 					blkno * SIT_ENTRY_PER_BLOCK);
148 			if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
149 				goto out;
150 			prev_blk_addr = fio.blk_addr;
151 			break;
152 		case META_SSA:
153 		case META_CP:
154 		case META_POR:
155 			fio.blk_addr = blkno;
156 			break;
157 		default:
158 			BUG();
159 		}
160 
161 		page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
162 		if (!page)
163 			continue;
164 		if (PageUptodate(page)) {
165 			f2fs_put_page(page, 1);
166 			continue;
167 		}
168 
169 		fio.page = page;
170 		f2fs_submit_page_mbio(&fio);
171 		f2fs_put_page(page, 0);
172 	}
173 out:
174 	f2fs_submit_merged_bio(sbi, META, READ);
175 	return blkno - start;
176 }
177 
178 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
179 {
180 	struct page *page;
181 	bool readahead = false;
182 
183 	page = find_get_page(META_MAPPING(sbi), index);
184 	if (!page || (page && !PageUptodate(page)))
185 		readahead = true;
186 	f2fs_put_page(page, 0);
187 
188 	if (readahead)
189 		ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
190 }
191 
192 static int f2fs_write_meta_page(struct page *page,
193 				struct writeback_control *wbc)
194 {
195 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
196 
197 	trace_f2fs_writepage(page, META);
198 
199 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
200 		goto redirty_out;
201 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
202 		goto redirty_out;
203 	if (unlikely(f2fs_cp_error(sbi)))
204 		goto redirty_out;
205 
206 	f2fs_wait_on_page_writeback(page, META);
207 	write_meta_page(sbi, page);
208 	dec_page_count(sbi, F2FS_DIRTY_META);
209 	unlock_page(page);
210 
211 	if (wbc->for_reclaim)
212 		f2fs_submit_merged_bio(sbi, META, WRITE);
213 	return 0;
214 
215 redirty_out:
216 	redirty_page_for_writepage(wbc, page);
217 	return AOP_WRITEPAGE_ACTIVATE;
218 }
219 
220 static int f2fs_write_meta_pages(struct address_space *mapping,
221 				struct writeback_control *wbc)
222 {
223 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
224 	long diff, written;
225 
226 	trace_f2fs_writepages(mapping->host, wbc, META);
227 
228 	/* collect a number of dirty meta pages and write together */
229 	if (wbc->for_kupdate ||
230 		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
231 		goto skip_write;
232 
233 	/* if mounting is failed, skip writing node pages */
234 	mutex_lock(&sbi->cp_mutex);
235 	diff = nr_pages_to_write(sbi, META, wbc);
236 	written = sync_meta_pages(sbi, META, wbc->nr_to_write);
237 	mutex_unlock(&sbi->cp_mutex);
238 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
239 	return 0;
240 
241 skip_write:
242 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
243 	return 0;
244 }
245 
246 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
247 						long nr_to_write)
248 {
249 	struct address_space *mapping = META_MAPPING(sbi);
250 	pgoff_t index = 0, end = LONG_MAX;
251 	struct pagevec pvec;
252 	long nwritten = 0;
253 	struct writeback_control wbc = {
254 		.for_reclaim = 0,
255 	};
256 
257 	pagevec_init(&pvec, 0);
258 
259 	while (index <= end) {
260 		int i, nr_pages;
261 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
262 				PAGECACHE_TAG_DIRTY,
263 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
264 		if (unlikely(nr_pages == 0))
265 			break;
266 
267 		for (i = 0; i < nr_pages; i++) {
268 			struct page *page = pvec.pages[i];
269 
270 			lock_page(page);
271 
272 			if (unlikely(page->mapping != mapping)) {
273 continue_unlock:
274 				unlock_page(page);
275 				continue;
276 			}
277 			if (!PageDirty(page)) {
278 				/* someone wrote it for us */
279 				goto continue_unlock;
280 			}
281 
282 			if (!clear_page_dirty_for_io(page))
283 				goto continue_unlock;
284 
285 			if (mapping->a_ops->writepage(page, &wbc)) {
286 				unlock_page(page);
287 				break;
288 			}
289 			nwritten++;
290 			if (unlikely(nwritten >= nr_to_write))
291 				break;
292 		}
293 		pagevec_release(&pvec);
294 		cond_resched();
295 	}
296 
297 	if (nwritten)
298 		f2fs_submit_merged_bio(sbi, type, WRITE);
299 
300 	return nwritten;
301 }
302 
303 static int f2fs_set_meta_page_dirty(struct page *page)
304 {
305 	trace_f2fs_set_page_dirty(page, META);
306 
307 	SetPageUptodate(page);
308 	if (!PageDirty(page)) {
309 		__set_page_dirty_nobuffers(page);
310 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
311 		SetPagePrivate(page);
312 		f2fs_trace_pid(page);
313 		return 1;
314 	}
315 	return 0;
316 }
317 
318 const struct address_space_operations f2fs_meta_aops = {
319 	.writepage	= f2fs_write_meta_page,
320 	.writepages	= f2fs_write_meta_pages,
321 	.set_page_dirty	= f2fs_set_meta_page_dirty,
322 	.invalidatepage = f2fs_invalidate_page,
323 	.releasepage	= f2fs_release_page,
324 };
325 
326 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
327 {
328 	struct inode_management *im = &sbi->im[type];
329 	struct ino_entry *e;
330 retry:
331 	if (radix_tree_preload(GFP_NOFS)) {
332 		cond_resched();
333 		goto retry;
334 	}
335 
336 	spin_lock(&im->ino_lock);
337 
338 	e = radix_tree_lookup(&im->ino_root, ino);
339 	if (!e) {
340 		e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
341 		if (!e) {
342 			spin_unlock(&im->ino_lock);
343 			radix_tree_preload_end();
344 			goto retry;
345 		}
346 		if (radix_tree_insert(&im->ino_root, ino, e)) {
347 			spin_unlock(&im->ino_lock);
348 			kmem_cache_free(ino_entry_slab, e);
349 			radix_tree_preload_end();
350 			goto retry;
351 		}
352 		memset(e, 0, sizeof(struct ino_entry));
353 		e->ino = ino;
354 
355 		list_add_tail(&e->list, &im->ino_list);
356 		if (type != ORPHAN_INO)
357 			im->ino_num++;
358 	}
359 	spin_unlock(&im->ino_lock);
360 	radix_tree_preload_end();
361 }
362 
363 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
364 {
365 	struct inode_management *im = &sbi->im[type];
366 	struct ino_entry *e;
367 
368 	spin_lock(&im->ino_lock);
369 	e = radix_tree_lookup(&im->ino_root, ino);
370 	if (e) {
371 		list_del(&e->list);
372 		radix_tree_delete(&im->ino_root, ino);
373 		im->ino_num--;
374 		spin_unlock(&im->ino_lock);
375 		kmem_cache_free(ino_entry_slab, e);
376 		return;
377 	}
378 	spin_unlock(&im->ino_lock);
379 }
380 
381 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
382 {
383 	/* add new dirty ino entry into list */
384 	__add_ino_entry(sbi, ino, type);
385 }
386 
387 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
388 {
389 	/* remove dirty ino entry from list */
390 	__remove_ino_entry(sbi, ino, type);
391 }
392 
393 /* mode should be APPEND_INO or UPDATE_INO */
394 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
395 {
396 	struct inode_management *im = &sbi->im[mode];
397 	struct ino_entry *e;
398 
399 	spin_lock(&im->ino_lock);
400 	e = radix_tree_lookup(&im->ino_root, ino);
401 	spin_unlock(&im->ino_lock);
402 	return e ? true : false;
403 }
404 
405 void release_dirty_inode(struct f2fs_sb_info *sbi)
406 {
407 	struct ino_entry *e, *tmp;
408 	int i;
409 
410 	for (i = APPEND_INO; i <= UPDATE_INO; i++) {
411 		struct inode_management *im = &sbi->im[i];
412 
413 		spin_lock(&im->ino_lock);
414 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
415 			list_del(&e->list);
416 			radix_tree_delete(&im->ino_root, e->ino);
417 			kmem_cache_free(ino_entry_slab, e);
418 			im->ino_num--;
419 		}
420 		spin_unlock(&im->ino_lock);
421 	}
422 }
423 
424 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
425 {
426 	struct inode_management *im = &sbi->im[ORPHAN_INO];
427 	int err = 0;
428 
429 	spin_lock(&im->ino_lock);
430 	if (unlikely(im->ino_num >= sbi->max_orphans))
431 		err = -ENOSPC;
432 	else
433 		im->ino_num++;
434 	spin_unlock(&im->ino_lock);
435 
436 	return err;
437 }
438 
439 void release_orphan_inode(struct f2fs_sb_info *sbi)
440 {
441 	struct inode_management *im = &sbi->im[ORPHAN_INO];
442 
443 	spin_lock(&im->ino_lock);
444 	f2fs_bug_on(sbi, im->ino_num == 0);
445 	im->ino_num--;
446 	spin_unlock(&im->ino_lock);
447 }
448 
449 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
450 {
451 	/* add new orphan ino entry into list */
452 	__add_ino_entry(sbi, ino, ORPHAN_INO);
453 }
454 
455 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
456 {
457 	/* remove orphan entry from orphan list */
458 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
459 }
460 
461 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
462 {
463 	struct inode *inode = f2fs_iget(sbi->sb, ino);
464 	f2fs_bug_on(sbi, IS_ERR(inode));
465 	clear_nlink(inode);
466 
467 	/* truncate all the data during iput */
468 	iput(inode);
469 }
470 
471 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
472 {
473 	block_t start_blk, orphan_blocks, i, j;
474 
475 	if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
476 		return;
477 
478 	set_sbi_flag(sbi, SBI_POR_DOING);
479 
480 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
481 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
482 
483 	ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP);
484 
485 	for (i = 0; i < orphan_blocks; i++) {
486 		struct page *page = get_meta_page(sbi, start_blk + i);
487 		struct f2fs_orphan_block *orphan_blk;
488 
489 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
490 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
491 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
492 			recover_orphan_inode(sbi, ino);
493 		}
494 		f2fs_put_page(page, 1);
495 	}
496 	/* clear Orphan Flag */
497 	clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
498 	clear_sbi_flag(sbi, SBI_POR_DOING);
499 	return;
500 }
501 
502 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
503 {
504 	struct list_head *head;
505 	struct f2fs_orphan_block *orphan_blk = NULL;
506 	unsigned int nentries = 0;
507 	unsigned short index;
508 	unsigned short orphan_blocks;
509 	struct page *page = NULL;
510 	struct ino_entry *orphan = NULL;
511 	struct inode_management *im = &sbi->im[ORPHAN_INO];
512 
513 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
514 
515 	for (index = 0; index < orphan_blocks; index++)
516 		grab_meta_page(sbi, start_blk + index);
517 
518 	index = 1;
519 
520 	/*
521 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
522 	 * orphan inode operations are covered under f2fs_lock_op().
523 	 * And, spin_lock should be avoided due to page operations below.
524 	 */
525 	head = &im->ino_list;
526 
527 	/* loop for each orphan inode entry and write them in Jornal block */
528 	list_for_each_entry(orphan, head, list) {
529 		if (!page) {
530 			page = find_get_page(META_MAPPING(sbi), start_blk++);
531 			f2fs_bug_on(sbi, !page);
532 			orphan_blk =
533 				(struct f2fs_orphan_block *)page_address(page);
534 			memset(orphan_blk, 0, sizeof(*orphan_blk));
535 			f2fs_put_page(page, 0);
536 		}
537 
538 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
539 
540 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
541 			/*
542 			 * an orphan block is full of 1020 entries,
543 			 * then we need to flush current orphan blocks
544 			 * and bring another one in memory
545 			 */
546 			orphan_blk->blk_addr = cpu_to_le16(index);
547 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
548 			orphan_blk->entry_count = cpu_to_le32(nentries);
549 			set_page_dirty(page);
550 			f2fs_put_page(page, 1);
551 			index++;
552 			nentries = 0;
553 			page = NULL;
554 		}
555 	}
556 
557 	if (page) {
558 		orphan_blk->blk_addr = cpu_to_le16(index);
559 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
560 		orphan_blk->entry_count = cpu_to_le32(nentries);
561 		set_page_dirty(page);
562 		f2fs_put_page(page, 1);
563 	}
564 }
565 
566 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
567 				block_t cp_addr, unsigned long long *version)
568 {
569 	struct page *cp_page_1, *cp_page_2 = NULL;
570 	unsigned long blk_size = sbi->blocksize;
571 	struct f2fs_checkpoint *cp_block;
572 	unsigned long long cur_version = 0, pre_version = 0;
573 	size_t crc_offset;
574 	__u32 crc = 0;
575 
576 	/* Read the 1st cp block in this CP pack */
577 	cp_page_1 = get_meta_page(sbi, cp_addr);
578 
579 	/* get the version number */
580 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
581 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
582 	if (crc_offset >= blk_size)
583 		goto invalid_cp1;
584 
585 	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
586 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
587 		goto invalid_cp1;
588 
589 	pre_version = cur_cp_version(cp_block);
590 
591 	/* Read the 2nd cp block in this CP pack */
592 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
593 	cp_page_2 = get_meta_page(sbi, cp_addr);
594 
595 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
596 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
597 	if (crc_offset >= blk_size)
598 		goto invalid_cp2;
599 
600 	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
601 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
602 		goto invalid_cp2;
603 
604 	cur_version = cur_cp_version(cp_block);
605 
606 	if (cur_version == pre_version) {
607 		*version = cur_version;
608 		f2fs_put_page(cp_page_2, 1);
609 		return cp_page_1;
610 	}
611 invalid_cp2:
612 	f2fs_put_page(cp_page_2, 1);
613 invalid_cp1:
614 	f2fs_put_page(cp_page_1, 1);
615 	return NULL;
616 }
617 
618 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
619 {
620 	struct f2fs_checkpoint *cp_block;
621 	struct f2fs_super_block *fsb = sbi->raw_super;
622 	struct page *cp1, *cp2, *cur_page;
623 	unsigned long blk_size = sbi->blocksize;
624 	unsigned long long cp1_version = 0, cp2_version = 0;
625 	unsigned long long cp_start_blk_no;
626 	unsigned int cp_blks = 1 + __cp_payload(sbi);
627 	block_t cp_blk_no;
628 	int i;
629 
630 	sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
631 	if (!sbi->ckpt)
632 		return -ENOMEM;
633 	/*
634 	 * Finding out valid cp block involves read both
635 	 * sets( cp pack1 and cp pack 2)
636 	 */
637 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
638 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
639 
640 	/* The second checkpoint pack should start at the next segment */
641 	cp_start_blk_no += ((unsigned long long)1) <<
642 				le32_to_cpu(fsb->log_blocks_per_seg);
643 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
644 
645 	if (cp1 && cp2) {
646 		if (ver_after(cp2_version, cp1_version))
647 			cur_page = cp2;
648 		else
649 			cur_page = cp1;
650 	} else if (cp1) {
651 		cur_page = cp1;
652 	} else if (cp2) {
653 		cur_page = cp2;
654 	} else {
655 		goto fail_no_cp;
656 	}
657 
658 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
659 	memcpy(sbi->ckpt, cp_block, blk_size);
660 
661 	if (cp_blks <= 1)
662 		goto done;
663 
664 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
665 	if (cur_page == cp2)
666 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
667 
668 	for (i = 1; i < cp_blks; i++) {
669 		void *sit_bitmap_ptr;
670 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
671 
672 		cur_page = get_meta_page(sbi, cp_blk_no + i);
673 		sit_bitmap_ptr = page_address(cur_page);
674 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
675 		f2fs_put_page(cur_page, 1);
676 	}
677 done:
678 	f2fs_put_page(cp1, 1);
679 	f2fs_put_page(cp2, 1);
680 	return 0;
681 
682 fail_no_cp:
683 	kfree(sbi->ckpt);
684 	return -EINVAL;
685 }
686 
687 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
688 {
689 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
690 
691 	if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
692 		return -EEXIST;
693 
694 	set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
695 	F2FS_I(inode)->dirty_dir = new;
696 	list_add_tail(&new->list, &sbi->dir_inode_list);
697 	stat_inc_dirty_dir(sbi);
698 	return 0;
699 }
700 
701 void update_dirty_page(struct inode *inode, struct page *page)
702 {
703 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
704 	struct inode_entry *new;
705 	int ret = 0;
706 
707 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
708 		return;
709 
710 	if (!S_ISDIR(inode->i_mode)) {
711 		inode_inc_dirty_pages(inode);
712 		goto out;
713 	}
714 
715 	new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
716 	new->inode = inode;
717 	INIT_LIST_HEAD(&new->list);
718 
719 	spin_lock(&sbi->dir_inode_lock);
720 	ret = __add_dirty_inode(inode, new);
721 	inode_inc_dirty_pages(inode);
722 	spin_unlock(&sbi->dir_inode_lock);
723 
724 	if (ret)
725 		kmem_cache_free(inode_entry_slab, new);
726 out:
727 	SetPagePrivate(page);
728 	f2fs_trace_pid(page);
729 }
730 
731 void add_dirty_dir_inode(struct inode *inode)
732 {
733 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
734 	struct inode_entry *new =
735 			f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
736 	int ret = 0;
737 
738 	new->inode = inode;
739 	INIT_LIST_HEAD(&new->list);
740 
741 	spin_lock(&sbi->dir_inode_lock);
742 	ret = __add_dirty_inode(inode, new);
743 	spin_unlock(&sbi->dir_inode_lock);
744 
745 	if (ret)
746 		kmem_cache_free(inode_entry_slab, new);
747 }
748 
749 void remove_dirty_dir_inode(struct inode *inode)
750 {
751 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
752 	struct inode_entry *entry;
753 
754 	if (!S_ISDIR(inode->i_mode))
755 		return;
756 
757 	spin_lock(&sbi->dir_inode_lock);
758 	if (get_dirty_pages(inode) ||
759 			!is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
760 		spin_unlock(&sbi->dir_inode_lock);
761 		return;
762 	}
763 
764 	entry = F2FS_I(inode)->dirty_dir;
765 	list_del(&entry->list);
766 	F2FS_I(inode)->dirty_dir = NULL;
767 	clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
768 	stat_dec_dirty_dir(sbi);
769 	spin_unlock(&sbi->dir_inode_lock);
770 	kmem_cache_free(inode_entry_slab, entry);
771 
772 	/* Only from the recovery routine */
773 	if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
774 		clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
775 		iput(inode);
776 	}
777 }
778 
779 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
780 {
781 	struct list_head *head;
782 	struct inode_entry *entry;
783 	struct inode *inode;
784 retry:
785 	if (unlikely(f2fs_cp_error(sbi)))
786 		return;
787 
788 	spin_lock(&sbi->dir_inode_lock);
789 
790 	head = &sbi->dir_inode_list;
791 	if (list_empty(head)) {
792 		spin_unlock(&sbi->dir_inode_lock);
793 		return;
794 	}
795 	entry = list_entry(head->next, struct inode_entry, list);
796 	inode = igrab(entry->inode);
797 	spin_unlock(&sbi->dir_inode_lock);
798 	if (inode) {
799 		filemap_fdatawrite(inode->i_mapping);
800 		iput(inode);
801 	} else {
802 		/*
803 		 * We should submit bio, since it exists several
804 		 * wribacking dentry pages in the freeing inode.
805 		 */
806 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
807 		cond_resched();
808 	}
809 	goto retry;
810 }
811 
812 /*
813  * Freeze all the FS-operations for checkpoint.
814  */
815 static int block_operations(struct f2fs_sb_info *sbi)
816 {
817 	struct writeback_control wbc = {
818 		.sync_mode = WB_SYNC_ALL,
819 		.nr_to_write = LONG_MAX,
820 		.for_reclaim = 0,
821 	};
822 	struct blk_plug plug;
823 	int err = 0;
824 
825 	blk_start_plug(&plug);
826 
827 retry_flush_dents:
828 	f2fs_lock_all(sbi);
829 	/* write all the dirty dentry pages */
830 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
831 		f2fs_unlock_all(sbi);
832 		sync_dirty_dir_inodes(sbi);
833 		if (unlikely(f2fs_cp_error(sbi))) {
834 			err = -EIO;
835 			goto out;
836 		}
837 		goto retry_flush_dents;
838 	}
839 
840 	/*
841 	 * POR: we should ensure that there are no dirty node pages
842 	 * until finishing nat/sit flush.
843 	 */
844 retry_flush_nodes:
845 	down_write(&sbi->node_write);
846 
847 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
848 		up_write(&sbi->node_write);
849 		sync_node_pages(sbi, 0, &wbc);
850 		if (unlikely(f2fs_cp_error(sbi))) {
851 			f2fs_unlock_all(sbi);
852 			err = -EIO;
853 			goto out;
854 		}
855 		goto retry_flush_nodes;
856 	}
857 out:
858 	blk_finish_plug(&plug);
859 	return err;
860 }
861 
862 static void unblock_operations(struct f2fs_sb_info *sbi)
863 {
864 	up_write(&sbi->node_write);
865 	f2fs_unlock_all(sbi);
866 }
867 
868 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
869 {
870 	DEFINE_WAIT(wait);
871 
872 	for (;;) {
873 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
874 
875 		if (!get_pages(sbi, F2FS_WRITEBACK))
876 			break;
877 
878 		io_schedule();
879 	}
880 	finish_wait(&sbi->cp_wait, &wait);
881 }
882 
883 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
884 {
885 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
886 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
887 	struct f2fs_nm_info *nm_i = NM_I(sbi);
888 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
889 	nid_t last_nid = nm_i->next_scan_nid;
890 	block_t start_blk;
891 	unsigned int data_sum_blocks, orphan_blocks;
892 	__u32 crc32 = 0;
893 	int i;
894 	int cp_payload_blks = __cp_payload(sbi);
895 
896 	/*
897 	 * This avoids to conduct wrong roll-forward operations and uses
898 	 * metapages, so should be called prior to sync_meta_pages below.
899 	 */
900 	discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
901 
902 	/* Flush all the NAT/SIT pages */
903 	while (get_pages(sbi, F2FS_DIRTY_META)) {
904 		sync_meta_pages(sbi, META, LONG_MAX);
905 		if (unlikely(f2fs_cp_error(sbi)))
906 			return;
907 	}
908 
909 	next_free_nid(sbi, &last_nid);
910 
911 	/*
912 	 * modify checkpoint
913 	 * version number is already updated
914 	 */
915 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
916 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
917 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
918 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
919 		ckpt->cur_node_segno[i] =
920 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
921 		ckpt->cur_node_blkoff[i] =
922 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
923 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
924 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
925 	}
926 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
927 		ckpt->cur_data_segno[i] =
928 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
929 		ckpt->cur_data_blkoff[i] =
930 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
931 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
932 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
933 	}
934 
935 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
936 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
937 	ckpt->next_free_nid = cpu_to_le32(last_nid);
938 
939 	/* 2 cp  + n data seg summary + orphan inode blocks */
940 	data_sum_blocks = npages_for_summary_flush(sbi, false);
941 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
942 		set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
943 	else
944 		clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
945 
946 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
947 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
948 			orphan_blocks);
949 
950 	if (__remain_node_summaries(cpc->reason))
951 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
952 				cp_payload_blks + data_sum_blocks +
953 				orphan_blocks + NR_CURSEG_NODE_TYPE);
954 	else
955 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
956 				cp_payload_blks + data_sum_blocks +
957 				orphan_blocks);
958 
959 	if (cpc->reason == CP_UMOUNT)
960 		set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
961 	else
962 		clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
963 
964 	if (cpc->reason == CP_FASTBOOT)
965 		set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
966 	else
967 		clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
968 
969 	if (orphan_num)
970 		set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
971 	else
972 		clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
973 
974 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
975 		set_ckpt_flags(ckpt, CP_FSCK_FLAG);
976 
977 	/* update SIT/NAT bitmap */
978 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
979 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
980 
981 	crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
982 	*((__le32 *)((unsigned char *)ckpt +
983 				le32_to_cpu(ckpt->checksum_offset)))
984 				= cpu_to_le32(crc32);
985 
986 	start_blk = __start_cp_addr(sbi);
987 
988 	/* write out checkpoint buffer at block 0 */
989 	update_meta_page(sbi, ckpt, start_blk++);
990 
991 	for (i = 1; i < 1 + cp_payload_blks; i++)
992 		update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
993 							start_blk++);
994 
995 	if (orphan_num) {
996 		write_orphan_inodes(sbi, start_blk);
997 		start_blk += orphan_blocks;
998 	}
999 
1000 	write_data_summaries(sbi, start_blk);
1001 	start_blk += data_sum_blocks;
1002 	if (__remain_node_summaries(cpc->reason)) {
1003 		write_node_summaries(sbi, start_blk);
1004 		start_blk += NR_CURSEG_NODE_TYPE;
1005 	}
1006 
1007 	/* writeout checkpoint block */
1008 	update_meta_page(sbi, ckpt, start_blk);
1009 
1010 	/* wait for previous submitted node/meta pages writeback */
1011 	wait_on_all_pages_writeback(sbi);
1012 
1013 	if (unlikely(f2fs_cp_error(sbi)))
1014 		return;
1015 
1016 	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1017 	filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1018 
1019 	/* update user_block_counts */
1020 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1021 	sbi->alloc_valid_block_count = 0;
1022 
1023 	/* Here, we only have one bio having CP pack */
1024 	sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1025 
1026 	/* wait for previous submitted meta pages writeback */
1027 	wait_on_all_pages_writeback(sbi);
1028 
1029 	release_dirty_inode(sbi);
1030 
1031 	if (unlikely(f2fs_cp_error(sbi)))
1032 		return;
1033 
1034 	clear_prefree_segments(sbi, cpc);
1035 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1036 }
1037 
1038 /*
1039  * We guarantee that this checkpoint procedure will not fail.
1040  */
1041 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1042 {
1043 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1044 	unsigned long long ckpt_ver;
1045 
1046 	mutex_lock(&sbi->cp_mutex);
1047 
1048 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1049 		(cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1050 		(cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1051 		goto out;
1052 	if (unlikely(f2fs_cp_error(sbi)))
1053 		goto out;
1054 	if (f2fs_readonly(sbi->sb))
1055 		goto out;
1056 
1057 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1058 
1059 	if (block_operations(sbi))
1060 		goto out;
1061 
1062 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1063 
1064 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
1065 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
1066 	f2fs_submit_merged_bio(sbi, META, WRITE);
1067 
1068 	/*
1069 	 * update checkpoint pack index
1070 	 * Increase the version number so that
1071 	 * SIT entries and seg summaries are written at correct place
1072 	 */
1073 	ckpt_ver = cur_cp_version(ckpt);
1074 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1075 
1076 	/* write cached NAT/SIT entries to NAT/SIT area */
1077 	flush_nat_entries(sbi);
1078 	flush_sit_entries(sbi, cpc);
1079 
1080 	/* unlock all the fs_lock[] in do_checkpoint() */
1081 	do_checkpoint(sbi, cpc);
1082 
1083 	unblock_operations(sbi);
1084 	stat_inc_cp_count(sbi->stat_info);
1085 
1086 	if (cpc->reason == CP_RECOVERY)
1087 		f2fs_msg(sbi->sb, KERN_NOTICE,
1088 			"checkpoint: version = %llx", ckpt_ver);
1089 out:
1090 	mutex_unlock(&sbi->cp_mutex);
1091 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1092 }
1093 
1094 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1095 {
1096 	int i;
1097 
1098 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1099 		struct inode_management *im = &sbi->im[i];
1100 
1101 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1102 		spin_lock_init(&im->ino_lock);
1103 		INIT_LIST_HEAD(&im->ino_list);
1104 		im->ino_num = 0;
1105 	}
1106 
1107 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1108 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1109 				F2FS_ORPHANS_PER_BLOCK;
1110 }
1111 
1112 int __init create_checkpoint_caches(void)
1113 {
1114 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1115 			sizeof(struct ino_entry));
1116 	if (!ino_entry_slab)
1117 		return -ENOMEM;
1118 	inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1119 			sizeof(struct inode_entry));
1120 	if (!inode_entry_slab) {
1121 		kmem_cache_destroy(ino_entry_slab);
1122 		return -ENOMEM;
1123 	}
1124 	return 0;
1125 }
1126 
1127 void destroy_checkpoint_caches(void)
1128 {
1129 	kmem_cache_destroy(ino_entry_slab);
1130 	kmem_cache_destroy(inode_entry_slab);
1131 }
1132