xref: /linux/fs/f2fs/checkpoint.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 		f2fs_put_page(page, 1);
74 		goto repeat;
75 	}
76 
77 	lock_page(page);
78 	if (unlikely(page->mapping != mapping)) {
79 		f2fs_put_page(page, 1);
80 		goto repeat;
81 	}
82 
83 	/*
84 	 * if there is any IO error when accessing device, make our filesystem
85 	 * readonly and make sure do not write checkpoint with non-uptodate
86 	 * meta page.
87 	 */
88 	if (unlikely(!PageUptodate(page)))
89 		f2fs_stop_checkpoint(sbi);
90 out:
91 	return page;
92 }
93 
94 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
95 {
96 	switch (type) {
97 	case META_NAT:
98 		break;
99 	case META_SIT:
100 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
101 			return false;
102 		break;
103 	case META_SSA:
104 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
105 			blkaddr < SM_I(sbi)->ssa_blkaddr))
106 			return false;
107 		break;
108 	case META_CP:
109 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
110 			blkaddr < __start_cp_addr(sbi)))
111 			return false;
112 		break;
113 	case META_POR:
114 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
115 			blkaddr < MAIN_BLKADDR(sbi)))
116 			return false;
117 		break;
118 	default:
119 		BUG();
120 	}
121 
122 	return true;
123 }
124 
125 /*
126  * Readahead CP/NAT/SIT/SSA pages
127  */
128 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
129 {
130 	block_t prev_blk_addr = 0;
131 	struct page *page;
132 	block_t blkno = start;
133 	struct f2fs_io_info fio = {
134 		.sbi = sbi,
135 		.type = META,
136 		.rw = READ_SYNC | REQ_META | REQ_PRIO,
137 		.encrypted_page = NULL,
138 	};
139 
140 	for (; nrpages-- > 0; blkno++) {
141 
142 		if (!is_valid_blkaddr(sbi, blkno, type))
143 			goto out;
144 
145 		switch (type) {
146 		case META_NAT:
147 			if (unlikely(blkno >=
148 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
149 				blkno = 0;
150 			/* get nat block addr */
151 			fio.blk_addr = current_nat_addr(sbi,
152 					blkno * NAT_ENTRY_PER_BLOCK);
153 			break;
154 		case META_SIT:
155 			/* get sit block addr */
156 			fio.blk_addr = current_sit_addr(sbi,
157 					blkno * SIT_ENTRY_PER_BLOCK);
158 			if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
159 				goto out;
160 			prev_blk_addr = fio.blk_addr;
161 			break;
162 		case META_SSA:
163 		case META_CP:
164 		case META_POR:
165 			fio.blk_addr = blkno;
166 			break;
167 		default:
168 			BUG();
169 		}
170 
171 		page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
172 		if (!page)
173 			continue;
174 		if (PageUptodate(page)) {
175 			f2fs_put_page(page, 1);
176 			continue;
177 		}
178 
179 		fio.page = page;
180 		f2fs_submit_page_mbio(&fio);
181 		f2fs_put_page(page, 0);
182 	}
183 out:
184 	f2fs_submit_merged_bio(sbi, META, READ);
185 	return blkno - start;
186 }
187 
188 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
189 {
190 	struct page *page;
191 	bool readahead = false;
192 
193 	page = find_get_page(META_MAPPING(sbi), index);
194 	if (!page || (page && !PageUptodate(page)))
195 		readahead = true;
196 	f2fs_put_page(page, 0);
197 
198 	if (readahead)
199 		ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
200 }
201 
202 static int f2fs_write_meta_page(struct page *page,
203 				struct writeback_control *wbc)
204 {
205 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
206 
207 	trace_f2fs_writepage(page, META);
208 
209 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
210 		goto redirty_out;
211 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
212 		goto redirty_out;
213 	if (unlikely(f2fs_cp_error(sbi)))
214 		goto redirty_out;
215 
216 	f2fs_wait_on_page_writeback(page, META);
217 	write_meta_page(sbi, page);
218 	dec_page_count(sbi, F2FS_DIRTY_META);
219 	unlock_page(page);
220 
221 	if (wbc->for_reclaim)
222 		f2fs_submit_merged_bio(sbi, META, WRITE);
223 	return 0;
224 
225 redirty_out:
226 	redirty_page_for_writepage(wbc, page);
227 	return AOP_WRITEPAGE_ACTIVATE;
228 }
229 
230 static int f2fs_write_meta_pages(struct address_space *mapping,
231 				struct writeback_control *wbc)
232 {
233 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
234 	long diff, written;
235 
236 	trace_f2fs_writepages(mapping->host, wbc, META);
237 
238 	/* collect a number of dirty meta pages and write together */
239 	if (wbc->for_kupdate ||
240 		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
241 		goto skip_write;
242 
243 	/* if mounting is failed, skip writing node pages */
244 	mutex_lock(&sbi->cp_mutex);
245 	diff = nr_pages_to_write(sbi, META, wbc);
246 	written = sync_meta_pages(sbi, META, wbc->nr_to_write);
247 	mutex_unlock(&sbi->cp_mutex);
248 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
249 	return 0;
250 
251 skip_write:
252 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
253 	return 0;
254 }
255 
256 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
257 						long nr_to_write)
258 {
259 	struct address_space *mapping = META_MAPPING(sbi);
260 	pgoff_t index = 0, end = LONG_MAX;
261 	struct pagevec pvec;
262 	long nwritten = 0;
263 	struct writeback_control wbc = {
264 		.for_reclaim = 0,
265 	};
266 
267 	pagevec_init(&pvec, 0);
268 
269 	while (index <= end) {
270 		int i, nr_pages;
271 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
272 				PAGECACHE_TAG_DIRTY,
273 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
274 		if (unlikely(nr_pages == 0))
275 			break;
276 
277 		for (i = 0; i < nr_pages; i++) {
278 			struct page *page = pvec.pages[i];
279 
280 			lock_page(page);
281 
282 			if (unlikely(page->mapping != mapping)) {
283 continue_unlock:
284 				unlock_page(page);
285 				continue;
286 			}
287 			if (!PageDirty(page)) {
288 				/* someone wrote it for us */
289 				goto continue_unlock;
290 			}
291 
292 			if (!clear_page_dirty_for_io(page))
293 				goto continue_unlock;
294 
295 			if (mapping->a_ops->writepage(page, &wbc)) {
296 				unlock_page(page);
297 				break;
298 			}
299 			nwritten++;
300 			if (unlikely(nwritten >= nr_to_write))
301 				break;
302 		}
303 		pagevec_release(&pvec);
304 		cond_resched();
305 	}
306 
307 	if (nwritten)
308 		f2fs_submit_merged_bio(sbi, type, WRITE);
309 
310 	return nwritten;
311 }
312 
313 static int f2fs_set_meta_page_dirty(struct page *page)
314 {
315 	trace_f2fs_set_page_dirty(page, META);
316 
317 	SetPageUptodate(page);
318 	if (!PageDirty(page)) {
319 		__set_page_dirty_nobuffers(page);
320 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
321 		SetPagePrivate(page);
322 		f2fs_trace_pid(page);
323 		return 1;
324 	}
325 	return 0;
326 }
327 
328 const struct address_space_operations f2fs_meta_aops = {
329 	.writepage	= f2fs_write_meta_page,
330 	.writepages	= f2fs_write_meta_pages,
331 	.set_page_dirty	= f2fs_set_meta_page_dirty,
332 	.invalidatepage = f2fs_invalidate_page,
333 	.releasepage	= f2fs_release_page,
334 };
335 
336 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
337 {
338 	struct inode_management *im = &sbi->im[type];
339 	struct ino_entry *e, *tmp;
340 
341 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
342 retry:
343 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
344 
345 	spin_lock(&im->ino_lock);
346 	e = radix_tree_lookup(&im->ino_root, ino);
347 	if (!e) {
348 		e = tmp;
349 		if (radix_tree_insert(&im->ino_root, ino, e)) {
350 			spin_unlock(&im->ino_lock);
351 			radix_tree_preload_end();
352 			goto retry;
353 		}
354 		memset(e, 0, sizeof(struct ino_entry));
355 		e->ino = ino;
356 
357 		list_add_tail(&e->list, &im->ino_list);
358 		if (type != ORPHAN_INO)
359 			im->ino_num++;
360 	}
361 	spin_unlock(&im->ino_lock);
362 	radix_tree_preload_end();
363 
364 	if (e != tmp)
365 		kmem_cache_free(ino_entry_slab, tmp);
366 }
367 
368 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
369 {
370 	struct inode_management *im = &sbi->im[type];
371 	struct ino_entry *e;
372 
373 	spin_lock(&im->ino_lock);
374 	e = radix_tree_lookup(&im->ino_root, ino);
375 	if (e) {
376 		list_del(&e->list);
377 		radix_tree_delete(&im->ino_root, ino);
378 		im->ino_num--;
379 		spin_unlock(&im->ino_lock);
380 		kmem_cache_free(ino_entry_slab, e);
381 		return;
382 	}
383 	spin_unlock(&im->ino_lock);
384 }
385 
386 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
387 {
388 	/* add new dirty ino entry into list */
389 	__add_ino_entry(sbi, ino, type);
390 }
391 
392 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
393 {
394 	/* remove dirty ino entry from list */
395 	__remove_ino_entry(sbi, ino, type);
396 }
397 
398 /* mode should be APPEND_INO or UPDATE_INO */
399 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
400 {
401 	struct inode_management *im = &sbi->im[mode];
402 	struct ino_entry *e;
403 
404 	spin_lock(&im->ino_lock);
405 	e = radix_tree_lookup(&im->ino_root, ino);
406 	spin_unlock(&im->ino_lock);
407 	return e ? true : false;
408 }
409 
410 void release_dirty_inode(struct f2fs_sb_info *sbi)
411 {
412 	struct ino_entry *e, *tmp;
413 	int i;
414 
415 	for (i = APPEND_INO; i <= UPDATE_INO; i++) {
416 		struct inode_management *im = &sbi->im[i];
417 
418 		spin_lock(&im->ino_lock);
419 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
420 			list_del(&e->list);
421 			radix_tree_delete(&im->ino_root, e->ino);
422 			kmem_cache_free(ino_entry_slab, e);
423 			im->ino_num--;
424 		}
425 		spin_unlock(&im->ino_lock);
426 	}
427 }
428 
429 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
430 {
431 	struct inode_management *im = &sbi->im[ORPHAN_INO];
432 	int err = 0;
433 
434 	spin_lock(&im->ino_lock);
435 	if (unlikely(im->ino_num >= sbi->max_orphans))
436 		err = -ENOSPC;
437 	else
438 		im->ino_num++;
439 	spin_unlock(&im->ino_lock);
440 
441 	return err;
442 }
443 
444 void release_orphan_inode(struct f2fs_sb_info *sbi)
445 {
446 	struct inode_management *im = &sbi->im[ORPHAN_INO];
447 
448 	spin_lock(&im->ino_lock);
449 	f2fs_bug_on(sbi, im->ino_num == 0);
450 	im->ino_num--;
451 	spin_unlock(&im->ino_lock);
452 }
453 
454 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
455 {
456 	/* add new orphan ino entry into list */
457 	__add_ino_entry(sbi, ino, ORPHAN_INO);
458 }
459 
460 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
461 {
462 	/* remove orphan entry from orphan list */
463 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
464 }
465 
466 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
467 {
468 	struct inode *inode;
469 
470 	inode = f2fs_iget(sbi->sb, ino);
471 	if (IS_ERR(inode)) {
472 		/*
473 		 * there should be a bug that we can't find the entry
474 		 * to orphan inode.
475 		 */
476 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
477 		return PTR_ERR(inode);
478 	}
479 
480 	clear_nlink(inode);
481 
482 	/* truncate all the data during iput */
483 	iput(inode);
484 	return 0;
485 }
486 
487 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
488 {
489 	block_t start_blk, orphan_blocks, i, j;
490 	int err;
491 
492 	if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
493 		return 0;
494 
495 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
496 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
497 
498 	ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP);
499 
500 	for (i = 0; i < orphan_blocks; i++) {
501 		struct page *page = get_meta_page(sbi, start_blk + i);
502 		struct f2fs_orphan_block *orphan_blk;
503 
504 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
505 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
506 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
507 			err = recover_orphan_inode(sbi, ino);
508 			if (err) {
509 				f2fs_put_page(page, 1);
510 				return err;
511 			}
512 		}
513 		f2fs_put_page(page, 1);
514 	}
515 	/* clear Orphan Flag */
516 	clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
517 	return 0;
518 }
519 
520 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
521 {
522 	struct list_head *head;
523 	struct f2fs_orphan_block *orphan_blk = NULL;
524 	unsigned int nentries = 0;
525 	unsigned short index = 1;
526 	unsigned short orphan_blocks;
527 	struct page *page = NULL;
528 	struct ino_entry *orphan = NULL;
529 	struct inode_management *im = &sbi->im[ORPHAN_INO];
530 
531 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
532 
533 	/*
534 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
535 	 * orphan inode operations are covered under f2fs_lock_op().
536 	 * And, spin_lock should be avoided due to page operations below.
537 	 */
538 	head = &im->ino_list;
539 
540 	/* loop for each orphan inode entry and write them in Jornal block */
541 	list_for_each_entry(orphan, head, list) {
542 		if (!page) {
543 			page = grab_meta_page(sbi, start_blk++);
544 			orphan_blk =
545 				(struct f2fs_orphan_block *)page_address(page);
546 			memset(orphan_blk, 0, sizeof(*orphan_blk));
547 		}
548 
549 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
550 
551 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
552 			/*
553 			 * an orphan block is full of 1020 entries,
554 			 * then we need to flush current orphan blocks
555 			 * and bring another one in memory
556 			 */
557 			orphan_blk->blk_addr = cpu_to_le16(index);
558 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
559 			orphan_blk->entry_count = cpu_to_le32(nentries);
560 			set_page_dirty(page);
561 			f2fs_put_page(page, 1);
562 			index++;
563 			nentries = 0;
564 			page = NULL;
565 		}
566 	}
567 
568 	if (page) {
569 		orphan_blk->blk_addr = cpu_to_le16(index);
570 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
571 		orphan_blk->entry_count = cpu_to_le32(nentries);
572 		set_page_dirty(page);
573 		f2fs_put_page(page, 1);
574 	}
575 }
576 
577 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
578 				block_t cp_addr, unsigned long long *version)
579 {
580 	struct page *cp_page_1, *cp_page_2 = NULL;
581 	unsigned long blk_size = sbi->blocksize;
582 	struct f2fs_checkpoint *cp_block;
583 	unsigned long long cur_version = 0, pre_version = 0;
584 	size_t crc_offset;
585 	__u32 crc = 0;
586 
587 	/* Read the 1st cp block in this CP pack */
588 	cp_page_1 = get_meta_page(sbi, cp_addr);
589 
590 	/* get the version number */
591 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
592 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
593 	if (crc_offset >= blk_size)
594 		goto invalid_cp1;
595 
596 	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
597 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
598 		goto invalid_cp1;
599 
600 	pre_version = cur_cp_version(cp_block);
601 
602 	/* Read the 2nd cp block in this CP pack */
603 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
604 	cp_page_2 = get_meta_page(sbi, cp_addr);
605 
606 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
607 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
608 	if (crc_offset >= blk_size)
609 		goto invalid_cp2;
610 
611 	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
612 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
613 		goto invalid_cp2;
614 
615 	cur_version = cur_cp_version(cp_block);
616 
617 	if (cur_version == pre_version) {
618 		*version = cur_version;
619 		f2fs_put_page(cp_page_2, 1);
620 		return cp_page_1;
621 	}
622 invalid_cp2:
623 	f2fs_put_page(cp_page_2, 1);
624 invalid_cp1:
625 	f2fs_put_page(cp_page_1, 1);
626 	return NULL;
627 }
628 
629 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
630 {
631 	struct f2fs_checkpoint *cp_block;
632 	struct f2fs_super_block *fsb = sbi->raw_super;
633 	struct page *cp1, *cp2, *cur_page;
634 	unsigned long blk_size = sbi->blocksize;
635 	unsigned long long cp1_version = 0, cp2_version = 0;
636 	unsigned long long cp_start_blk_no;
637 	unsigned int cp_blks = 1 + __cp_payload(sbi);
638 	block_t cp_blk_no;
639 	int i;
640 
641 	sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
642 	if (!sbi->ckpt)
643 		return -ENOMEM;
644 	/*
645 	 * Finding out valid cp block involves read both
646 	 * sets( cp pack1 and cp pack 2)
647 	 */
648 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
649 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
650 
651 	/* The second checkpoint pack should start at the next segment */
652 	cp_start_blk_no += ((unsigned long long)1) <<
653 				le32_to_cpu(fsb->log_blocks_per_seg);
654 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
655 
656 	if (cp1 && cp2) {
657 		if (ver_after(cp2_version, cp1_version))
658 			cur_page = cp2;
659 		else
660 			cur_page = cp1;
661 	} else if (cp1) {
662 		cur_page = cp1;
663 	} else if (cp2) {
664 		cur_page = cp2;
665 	} else {
666 		goto fail_no_cp;
667 	}
668 
669 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
670 	memcpy(sbi->ckpt, cp_block, blk_size);
671 
672 	if (cp_blks <= 1)
673 		goto done;
674 
675 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
676 	if (cur_page == cp2)
677 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
678 
679 	for (i = 1; i < cp_blks; i++) {
680 		void *sit_bitmap_ptr;
681 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
682 
683 		cur_page = get_meta_page(sbi, cp_blk_no + i);
684 		sit_bitmap_ptr = page_address(cur_page);
685 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
686 		f2fs_put_page(cur_page, 1);
687 	}
688 done:
689 	f2fs_put_page(cp1, 1);
690 	f2fs_put_page(cp2, 1);
691 	return 0;
692 
693 fail_no_cp:
694 	kfree(sbi->ckpt);
695 	return -EINVAL;
696 }
697 
698 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
699 {
700 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
701 
702 	if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
703 		return -EEXIST;
704 
705 	set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
706 	F2FS_I(inode)->dirty_dir = new;
707 	list_add_tail(&new->list, &sbi->dir_inode_list);
708 	stat_inc_dirty_dir(sbi);
709 	return 0;
710 }
711 
712 void update_dirty_page(struct inode *inode, struct page *page)
713 {
714 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
715 	struct inode_entry *new;
716 	int ret = 0;
717 
718 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
719 			!S_ISLNK(inode->i_mode))
720 		return;
721 
722 	if (!S_ISDIR(inode->i_mode)) {
723 		inode_inc_dirty_pages(inode);
724 		goto out;
725 	}
726 
727 	new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
728 	new->inode = inode;
729 	INIT_LIST_HEAD(&new->list);
730 
731 	spin_lock(&sbi->dir_inode_lock);
732 	ret = __add_dirty_inode(inode, new);
733 	inode_inc_dirty_pages(inode);
734 	spin_unlock(&sbi->dir_inode_lock);
735 
736 	if (ret)
737 		kmem_cache_free(inode_entry_slab, new);
738 out:
739 	SetPagePrivate(page);
740 	f2fs_trace_pid(page);
741 }
742 
743 void add_dirty_dir_inode(struct inode *inode)
744 {
745 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
746 	struct inode_entry *new =
747 			f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
748 	int ret = 0;
749 
750 	new->inode = inode;
751 	INIT_LIST_HEAD(&new->list);
752 
753 	spin_lock(&sbi->dir_inode_lock);
754 	ret = __add_dirty_inode(inode, new);
755 	spin_unlock(&sbi->dir_inode_lock);
756 
757 	if (ret)
758 		kmem_cache_free(inode_entry_slab, new);
759 }
760 
761 void remove_dirty_dir_inode(struct inode *inode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
764 	struct inode_entry *entry;
765 
766 	if (!S_ISDIR(inode->i_mode))
767 		return;
768 
769 	spin_lock(&sbi->dir_inode_lock);
770 	if (get_dirty_pages(inode) ||
771 			!is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
772 		spin_unlock(&sbi->dir_inode_lock);
773 		return;
774 	}
775 
776 	entry = F2FS_I(inode)->dirty_dir;
777 	list_del(&entry->list);
778 	F2FS_I(inode)->dirty_dir = NULL;
779 	clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
780 	stat_dec_dirty_dir(sbi);
781 	spin_unlock(&sbi->dir_inode_lock);
782 	kmem_cache_free(inode_entry_slab, entry);
783 
784 	/* Only from the recovery routine */
785 	if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
786 		clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
787 		iput(inode);
788 	}
789 }
790 
791 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
792 {
793 	struct list_head *head;
794 	struct inode_entry *entry;
795 	struct inode *inode;
796 retry:
797 	if (unlikely(f2fs_cp_error(sbi)))
798 		return;
799 
800 	spin_lock(&sbi->dir_inode_lock);
801 
802 	head = &sbi->dir_inode_list;
803 	if (list_empty(head)) {
804 		spin_unlock(&sbi->dir_inode_lock);
805 		return;
806 	}
807 	entry = list_entry(head->next, struct inode_entry, list);
808 	inode = igrab(entry->inode);
809 	spin_unlock(&sbi->dir_inode_lock);
810 	if (inode) {
811 		filemap_fdatawrite(inode->i_mapping);
812 		iput(inode);
813 	} else {
814 		/*
815 		 * We should submit bio, since it exists several
816 		 * wribacking dentry pages in the freeing inode.
817 		 */
818 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
819 		cond_resched();
820 	}
821 	goto retry;
822 }
823 
824 /*
825  * Freeze all the FS-operations for checkpoint.
826  */
827 static int block_operations(struct f2fs_sb_info *sbi)
828 {
829 	struct writeback_control wbc = {
830 		.sync_mode = WB_SYNC_ALL,
831 		.nr_to_write = LONG_MAX,
832 		.for_reclaim = 0,
833 	};
834 	struct blk_plug plug;
835 	int err = 0;
836 
837 	blk_start_plug(&plug);
838 
839 retry_flush_dents:
840 	f2fs_lock_all(sbi);
841 	/* write all the dirty dentry pages */
842 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
843 		f2fs_unlock_all(sbi);
844 		sync_dirty_dir_inodes(sbi);
845 		if (unlikely(f2fs_cp_error(sbi))) {
846 			err = -EIO;
847 			goto out;
848 		}
849 		goto retry_flush_dents;
850 	}
851 
852 	/*
853 	 * POR: we should ensure that there are no dirty node pages
854 	 * until finishing nat/sit flush.
855 	 */
856 retry_flush_nodes:
857 	down_write(&sbi->node_write);
858 
859 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
860 		up_write(&sbi->node_write);
861 		sync_node_pages(sbi, 0, &wbc);
862 		if (unlikely(f2fs_cp_error(sbi))) {
863 			f2fs_unlock_all(sbi);
864 			err = -EIO;
865 			goto out;
866 		}
867 		goto retry_flush_nodes;
868 	}
869 out:
870 	blk_finish_plug(&plug);
871 	return err;
872 }
873 
874 static void unblock_operations(struct f2fs_sb_info *sbi)
875 {
876 	up_write(&sbi->node_write);
877 	f2fs_unlock_all(sbi);
878 }
879 
880 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
881 {
882 	DEFINE_WAIT(wait);
883 
884 	for (;;) {
885 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
886 
887 		if (!get_pages(sbi, F2FS_WRITEBACK))
888 			break;
889 
890 		io_schedule();
891 	}
892 	finish_wait(&sbi->cp_wait, &wait);
893 }
894 
895 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
896 {
897 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
898 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
899 	struct f2fs_nm_info *nm_i = NM_I(sbi);
900 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
901 	nid_t last_nid = nm_i->next_scan_nid;
902 	block_t start_blk;
903 	unsigned int data_sum_blocks, orphan_blocks;
904 	__u32 crc32 = 0;
905 	int i;
906 	int cp_payload_blks = __cp_payload(sbi);
907 	block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
908 	bool invalidate = false;
909 
910 	/*
911 	 * This avoids to conduct wrong roll-forward operations and uses
912 	 * metapages, so should be called prior to sync_meta_pages below.
913 	 */
914 	if (discard_next_dnode(sbi, discard_blk))
915 		invalidate = true;
916 
917 	/* Flush all the NAT/SIT pages */
918 	while (get_pages(sbi, F2FS_DIRTY_META)) {
919 		sync_meta_pages(sbi, META, LONG_MAX);
920 		if (unlikely(f2fs_cp_error(sbi)))
921 			return;
922 	}
923 
924 	next_free_nid(sbi, &last_nid);
925 
926 	/*
927 	 * modify checkpoint
928 	 * version number is already updated
929 	 */
930 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
931 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
932 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
933 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
934 		ckpt->cur_node_segno[i] =
935 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
936 		ckpt->cur_node_blkoff[i] =
937 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
938 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
939 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
940 	}
941 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
942 		ckpt->cur_data_segno[i] =
943 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
944 		ckpt->cur_data_blkoff[i] =
945 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
946 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
947 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
948 	}
949 
950 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
951 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
952 	ckpt->next_free_nid = cpu_to_le32(last_nid);
953 
954 	/* 2 cp  + n data seg summary + orphan inode blocks */
955 	data_sum_blocks = npages_for_summary_flush(sbi, false);
956 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
957 		set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
958 	else
959 		clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
960 
961 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
962 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
963 			orphan_blocks);
964 
965 	if (__remain_node_summaries(cpc->reason))
966 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
967 				cp_payload_blks + data_sum_blocks +
968 				orphan_blocks + NR_CURSEG_NODE_TYPE);
969 	else
970 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
971 				cp_payload_blks + data_sum_blocks +
972 				orphan_blocks);
973 
974 	if (cpc->reason == CP_UMOUNT)
975 		set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
976 	else
977 		clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
978 
979 	if (cpc->reason == CP_FASTBOOT)
980 		set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
981 	else
982 		clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
983 
984 	if (orphan_num)
985 		set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
986 	else
987 		clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
988 
989 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
990 		set_ckpt_flags(ckpt, CP_FSCK_FLAG);
991 
992 	/* update SIT/NAT bitmap */
993 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
994 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
995 
996 	crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
997 	*((__le32 *)((unsigned char *)ckpt +
998 				le32_to_cpu(ckpt->checksum_offset)))
999 				= cpu_to_le32(crc32);
1000 
1001 	start_blk = __start_cp_addr(sbi);
1002 
1003 	/* write out checkpoint buffer at block 0 */
1004 	update_meta_page(sbi, ckpt, start_blk++);
1005 
1006 	for (i = 1; i < 1 + cp_payload_blks; i++)
1007 		update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1008 							start_blk++);
1009 
1010 	if (orphan_num) {
1011 		write_orphan_inodes(sbi, start_blk);
1012 		start_blk += orphan_blocks;
1013 	}
1014 
1015 	write_data_summaries(sbi, start_blk);
1016 	start_blk += data_sum_blocks;
1017 	if (__remain_node_summaries(cpc->reason)) {
1018 		write_node_summaries(sbi, start_blk);
1019 		start_blk += NR_CURSEG_NODE_TYPE;
1020 	}
1021 
1022 	/* writeout checkpoint block */
1023 	update_meta_page(sbi, ckpt, start_blk);
1024 
1025 	/* wait for previous submitted node/meta pages writeback */
1026 	wait_on_all_pages_writeback(sbi);
1027 
1028 	if (unlikely(f2fs_cp_error(sbi)))
1029 		return;
1030 
1031 	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1032 	filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1033 
1034 	/* update user_block_counts */
1035 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1036 	sbi->alloc_valid_block_count = 0;
1037 
1038 	/* Here, we only have one bio having CP pack */
1039 	sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1040 
1041 	/* wait for previous submitted meta pages writeback */
1042 	wait_on_all_pages_writeback(sbi);
1043 
1044 	/*
1045 	 * invalidate meta page which is used temporarily for zeroing out
1046 	 * block at the end of warm node chain.
1047 	 */
1048 	if (invalidate)
1049 		invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1050 								discard_blk);
1051 
1052 	release_dirty_inode(sbi);
1053 
1054 	if (unlikely(f2fs_cp_error(sbi)))
1055 		return;
1056 
1057 	clear_prefree_segments(sbi, cpc);
1058 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1059 }
1060 
1061 /*
1062  * We guarantee that this checkpoint procedure will not fail.
1063  */
1064 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1065 {
1066 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1067 	unsigned long long ckpt_ver;
1068 
1069 	mutex_lock(&sbi->cp_mutex);
1070 
1071 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1072 		(cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1073 		(cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1074 		goto out;
1075 	if (unlikely(f2fs_cp_error(sbi)))
1076 		goto out;
1077 	if (f2fs_readonly(sbi->sb))
1078 		goto out;
1079 
1080 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1081 
1082 	if (block_operations(sbi))
1083 		goto out;
1084 
1085 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1086 
1087 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
1088 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
1089 	f2fs_submit_merged_bio(sbi, META, WRITE);
1090 
1091 	/*
1092 	 * update checkpoint pack index
1093 	 * Increase the version number so that
1094 	 * SIT entries and seg summaries are written at correct place
1095 	 */
1096 	ckpt_ver = cur_cp_version(ckpt);
1097 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1098 
1099 	/* write cached NAT/SIT entries to NAT/SIT area */
1100 	flush_nat_entries(sbi);
1101 	flush_sit_entries(sbi, cpc);
1102 
1103 	/* unlock all the fs_lock[] in do_checkpoint() */
1104 	do_checkpoint(sbi, cpc);
1105 
1106 	unblock_operations(sbi);
1107 	stat_inc_cp_count(sbi->stat_info);
1108 
1109 	if (cpc->reason == CP_RECOVERY)
1110 		f2fs_msg(sbi->sb, KERN_NOTICE,
1111 			"checkpoint: version = %llx", ckpt_ver);
1112 out:
1113 	mutex_unlock(&sbi->cp_mutex);
1114 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1115 }
1116 
1117 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1118 {
1119 	int i;
1120 
1121 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1122 		struct inode_management *im = &sbi->im[i];
1123 
1124 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1125 		spin_lock_init(&im->ino_lock);
1126 		INIT_LIST_HEAD(&im->ino_list);
1127 		im->ino_num = 0;
1128 	}
1129 
1130 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1131 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1132 				F2FS_ORPHANS_PER_BLOCK;
1133 }
1134 
1135 int __init create_checkpoint_caches(void)
1136 {
1137 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1138 			sizeof(struct ino_entry));
1139 	if (!ino_entry_slab)
1140 		return -ENOMEM;
1141 	inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1142 			sizeof(struct inode_entry));
1143 	if (!inode_entry_slab) {
1144 		kmem_cache_destroy(ino_entry_slab);
1145 		return -ENOMEM;
1146 	}
1147 	return 0;
1148 }
1149 
1150 void destroy_checkpoint_caches(void)
1151 {
1152 	kmem_cache_destroy(ino_entry_slab);
1153 	kmem_cache_destroy(inode_entry_slab);
1154 }
1155