xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision f3f338caad3428fbc4bb563828efc6ecce4d956b)
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;
340 retry:
341 	if (radix_tree_preload(GFP_NOFS)) {
342 		cond_resched();
343 		goto retry;
344 	}
345 
346 	spin_lock(&im->ino_lock);
347 
348 	e = radix_tree_lookup(&im->ino_root, ino);
349 	if (!e) {
350 		e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
351 		if (!e) {
352 			spin_unlock(&im->ino_lock);
353 			radix_tree_preload_end();
354 			goto retry;
355 		}
356 		if (radix_tree_insert(&im->ino_root, ino, e)) {
357 			spin_unlock(&im->ino_lock);
358 			kmem_cache_free(ino_entry_slab, e);
359 			radix_tree_preload_end();
360 			goto retry;
361 		}
362 		memset(e, 0, sizeof(struct ino_entry));
363 		e->ino = ino;
364 
365 		list_add_tail(&e->list, &im->ino_list);
366 		if (type != ORPHAN_INO)
367 			im->ino_num++;
368 	}
369 	spin_unlock(&im->ino_lock);
370 	radix_tree_preload_end();
371 }
372 
373 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
374 {
375 	struct inode_management *im = &sbi->im[type];
376 	struct ino_entry *e;
377 
378 	spin_lock(&im->ino_lock);
379 	e = radix_tree_lookup(&im->ino_root, ino);
380 	if (e) {
381 		list_del(&e->list);
382 		radix_tree_delete(&im->ino_root, ino);
383 		im->ino_num--;
384 		spin_unlock(&im->ino_lock);
385 		kmem_cache_free(ino_entry_slab, e);
386 		return;
387 	}
388 	spin_unlock(&im->ino_lock);
389 }
390 
391 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
392 {
393 	/* add new dirty ino entry into list */
394 	__add_ino_entry(sbi, ino, type);
395 }
396 
397 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
398 {
399 	/* remove dirty ino entry from list */
400 	__remove_ino_entry(sbi, ino, type);
401 }
402 
403 /* mode should be APPEND_INO or UPDATE_INO */
404 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
405 {
406 	struct inode_management *im = &sbi->im[mode];
407 	struct ino_entry *e;
408 
409 	spin_lock(&im->ino_lock);
410 	e = radix_tree_lookup(&im->ino_root, ino);
411 	spin_unlock(&im->ino_lock);
412 	return e ? true : false;
413 }
414 
415 void release_dirty_inode(struct f2fs_sb_info *sbi)
416 {
417 	struct ino_entry *e, *tmp;
418 	int i;
419 
420 	for (i = APPEND_INO; i <= UPDATE_INO; i++) {
421 		struct inode_management *im = &sbi->im[i];
422 
423 		spin_lock(&im->ino_lock);
424 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
425 			list_del(&e->list);
426 			radix_tree_delete(&im->ino_root, e->ino);
427 			kmem_cache_free(ino_entry_slab, e);
428 			im->ino_num--;
429 		}
430 		spin_unlock(&im->ino_lock);
431 	}
432 }
433 
434 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
435 {
436 	struct inode_management *im = &sbi->im[ORPHAN_INO];
437 	int err = 0;
438 
439 	spin_lock(&im->ino_lock);
440 	if (unlikely(im->ino_num >= sbi->max_orphans))
441 		err = -ENOSPC;
442 	else
443 		im->ino_num++;
444 	spin_unlock(&im->ino_lock);
445 
446 	return err;
447 }
448 
449 void release_orphan_inode(struct f2fs_sb_info *sbi)
450 {
451 	struct inode_management *im = &sbi->im[ORPHAN_INO];
452 
453 	spin_lock(&im->ino_lock);
454 	f2fs_bug_on(sbi, im->ino_num == 0);
455 	im->ino_num--;
456 	spin_unlock(&im->ino_lock);
457 }
458 
459 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
460 {
461 	/* add new orphan ino entry into list */
462 	__add_ino_entry(sbi, ino, ORPHAN_INO);
463 }
464 
465 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
466 {
467 	/* remove orphan entry from orphan list */
468 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
469 }
470 
471 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
472 {
473 	struct inode *inode = f2fs_iget(sbi->sb, ino);
474 	f2fs_bug_on(sbi, IS_ERR(inode));
475 	clear_nlink(inode);
476 
477 	/* truncate all the data during iput */
478 	iput(inode);
479 }
480 
481 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
482 {
483 	block_t start_blk, orphan_blocks, i, j;
484 
485 	if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
486 		return;
487 
488 	set_sbi_flag(sbi, SBI_POR_DOING);
489 
490 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
491 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
492 
493 	ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP);
494 
495 	for (i = 0; i < orphan_blocks; i++) {
496 		struct page *page = get_meta_page(sbi, start_blk + i);
497 		struct f2fs_orphan_block *orphan_blk;
498 
499 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
500 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
501 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
502 			recover_orphan_inode(sbi, ino);
503 		}
504 		f2fs_put_page(page, 1);
505 	}
506 	/* clear Orphan Flag */
507 	clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
508 	clear_sbi_flag(sbi, SBI_POR_DOING);
509 	return;
510 }
511 
512 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
513 {
514 	struct list_head *head;
515 	struct f2fs_orphan_block *orphan_blk = NULL;
516 	unsigned int nentries = 0;
517 	unsigned short index = 1;
518 	unsigned short orphan_blocks;
519 	struct page *page = NULL;
520 	struct ino_entry *orphan = NULL;
521 	struct inode_management *im = &sbi->im[ORPHAN_INO];
522 
523 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
524 
525 	/*
526 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
527 	 * orphan inode operations are covered under f2fs_lock_op().
528 	 * And, spin_lock should be avoided due to page operations below.
529 	 */
530 	head = &im->ino_list;
531 
532 	/* loop for each orphan inode entry and write them in Jornal block */
533 	list_for_each_entry(orphan, head, list) {
534 		if (!page) {
535 			page = grab_meta_page(sbi, start_blk++);
536 			orphan_blk =
537 				(struct f2fs_orphan_block *)page_address(page);
538 			memset(orphan_blk, 0, sizeof(*orphan_blk));
539 		}
540 
541 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
542 
543 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
544 			/*
545 			 * an orphan block is full of 1020 entries,
546 			 * then we need to flush current orphan blocks
547 			 * and bring another one in memory
548 			 */
549 			orphan_blk->blk_addr = cpu_to_le16(index);
550 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
551 			orphan_blk->entry_count = cpu_to_le32(nentries);
552 			set_page_dirty(page);
553 			f2fs_put_page(page, 1);
554 			index++;
555 			nentries = 0;
556 			page = NULL;
557 		}
558 	}
559 
560 	if (page) {
561 		orphan_blk->blk_addr = cpu_to_le16(index);
562 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
563 		orphan_blk->entry_count = cpu_to_le32(nentries);
564 		set_page_dirty(page);
565 		f2fs_put_page(page, 1);
566 	}
567 }
568 
569 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
570 				block_t cp_addr, unsigned long long *version)
571 {
572 	struct page *cp_page_1, *cp_page_2 = NULL;
573 	unsigned long blk_size = sbi->blocksize;
574 	struct f2fs_checkpoint *cp_block;
575 	unsigned long long cur_version = 0, pre_version = 0;
576 	size_t crc_offset;
577 	__u32 crc = 0;
578 
579 	/* Read the 1st cp block in this CP pack */
580 	cp_page_1 = get_meta_page(sbi, cp_addr);
581 
582 	/* get the version number */
583 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
584 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
585 	if (crc_offset >= blk_size)
586 		goto invalid_cp1;
587 
588 	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
589 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
590 		goto invalid_cp1;
591 
592 	pre_version = cur_cp_version(cp_block);
593 
594 	/* Read the 2nd cp block in this CP pack */
595 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
596 	cp_page_2 = get_meta_page(sbi, cp_addr);
597 
598 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
599 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
600 	if (crc_offset >= blk_size)
601 		goto invalid_cp2;
602 
603 	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
604 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
605 		goto invalid_cp2;
606 
607 	cur_version = cur_cp_version(cp_block);
608 
609 	if (cur_version == pre_version) {
610 		*version = cur_version;
611 		f2fs_put_page(cp_page_2, 1);
612 		return cp_page_1;
613 	}
614 invalid_cp2:
615 	f2fs_put_page(cp_page_2, 1);
616 invalid_cp1:
617 	f2fs_put_page(cp_page_1, 1);
618 	return NULL;
619 }
620 
621 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
622 {
623 	struct f2fs_checkpoint *cp_block;
624 	struct f2fs_super_block *fsb = sbi->raw_super;
625 	struct page *cp1, *cp2, *cur_page;
626 	unsigned long blk_size = sbi->blocksize;
627 	unsigned long long cp1_version = 0, cp2_version = 0;
628 	unsigned long long cp_start_blk_no;
629 	unsigned int cp_blks = 1 + __cp_payload(sbi);
630 	block_t cp_blk_no;
631 	int i;
632 
633 	sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
634 	if (!sbi->ckpt)
635 		return -ENOMEM;
636 	/*
637 	 * Finding out valid cp block involves read both
638 	 * sets( cp pack1 and cp pack 2)
639 	 */
640 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
641 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
642 
643 	/* The second checkpoint pack should start at the next segment */
644 	cp_start_blk_no += ((unsigned long long)1) <<
645 				le32_to_cpu(fsb->log_blocks_per_seg);
646 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
647 
648 	if (cp1 && cp2) {
649 		if (ver_after(cp2_version, cp1_version))
650 			cur_page = cp2;
651 		else
652 			cur_page = cp1;
653 	} else if (cp1) {
654 		cur_page = cp1;
655 	} else if (cp2) {
656 		cur_page = cp2;
657 	} else {
658 		goto fail_no_cp;
659 	}
660 
661 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
662 	memcpy(sbi->ckpt, cp_block, blk_size);
663 
664 	if (cp_blks <= 1)
665 		goto done;
666 
667 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
668 	if (cur_page == cp2)
669 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
670 
671 	for (i = 1; i < cp_blks; i++) {
672 		void *sit_bitmap_ptr;
673 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
674 
675 		cur_page = get_meta_page(sbi, cp_blk_no + i);
676 		sit_bitmap_ptr = page_address(cur_page);
677 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
678 		f2fs_put_page(cur_page, 1);
679 	}
680 done:
681 	f2fs_put_page(cp1, 1);
682 	f2fs_put_page(cp2, 1);
683 	return 0;
684 
685 fail_no_cp:
686 	kfree(sbi->ckpt);
687 	return -EINVAL;
688 }
689 
690 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
691 {
692 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
693 
694 	if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
695 		return -EEXIST;
696 
697 	set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
698 	F2FS_I(inode)->dirty_dir = new;
699 	list_add_tail(&new->list, &sbi->dir_inode_list);
700 	stat_inc_dirty_dir(sbi);
701 	return 0;
702 }
703 
704 void update_dirty_page(struct inode *inode, struct page *page)
705 {
706 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
707 	struct inode_entry *new;
708 	int ret = 0;
709 
710 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
711 			!S_ISLNK(inode->i_mode))
712 		return;
713 
714 	if (!S_ISDIR(inode->i_mode)) {
715 		inode_inc_dirty_pages(inode);
716 		goto out;
717 	}
718 
719 	new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
720 	new->inode = inode;
721 	INIT_LIST_HEAD(&new->list);
722 
723 	spin_lock(&sbi->dir_inode_lock);
724 	ret = __add_dirty_inode(inode, new);
725 	inode_inc_dirty_pages(inode);
726 	spin_unlock(&sbi->dir_inode_lock);
727 
728 	if (ret)
729 		kmem_cache_free(inode_entry_slab, new);
730 out:
731 	SetPagePrivate(page);
732 	f2fs_trace_pid(page);
733 }
734 
735 void add_dirty_dir_inode(struct inode *inode)
736 {
737 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
738 	struct inode_entry *new =
739 			f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
740 	int ret = 0;
741 
742 	new->inode = inode;
743 	INIT_LIST_HEAD(&new->list);
744 
745 	spin_lock(&sbi->dir_inode_lock);
746 	ret = __add_dirty_inode(inode, new);
747 	spin_unlock(&sbi->dir_inode_lock);
748 
749 	if (ret)
750 		kmem_cache_free(inode_entry_slab, new);
751 }
752 
753 void remove_dirty_dir_inode(struct inode *inode)
754 {
755 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
756 	struct inode_entry *entry;
757 
758 	if (!S_ISDIR(inode->i_mode))
759 		return;
760 
761 	spin_lock(&sbi->dir_inode_lock);
762 	if (get_dirty_pages(inode) ||
763 			!is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
764 		spin_unlock(&sbi->dir_inode_lock);
765 		return;
766 	}
767 
768 	entry = F2FS_I(inode)->dirty_dir;
769 	list_del(&entry->list);
770 	F2FS_I(inode)->dirty_dir = NULL;
771 	clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
772 	stat_dec_dirty_dir(sbi);
773 	spin_unlock(&sbi->dir_inode_lock);
774 	kmem_cache_free(inode_entry_slab, entry);
775 
776 	/* Only from the recovery routine */
777 	if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
778 		clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
779 		iput(inode);
780 	}
781 }
782 
783 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
784 {
785 	struct list_head *head;
786 	struct inode_entry *entry;
787 	struct inode *inode;
788 retry:
789 	if (unlikely(f2fs_cp_error(sbi)))
790 		return;
791 
792 	spin_lock(&sbi->dir_inode_lock);
793 
794 	head = &sbi->dir_inode_list;
795 	if (list_empty(head)) {
796 		spin_unlock(&sbi->dir_inode_lock);
797 		return;
798 	}
799 	entry = list_entry(head->next, struct inode_entry, list);
800 	inode = igrab(entry->inode);
801 	spin_unlock(&sbi->dir_inode_lock);
802 	if (inode) {
803 		filemap_fdatawrite(inode->i_mapping);
804 		iput(inode);
805 	} else {
806 		/*
807 		 * We should submit bio, since it exists several
808 		 * wribacking dentry pages in the freeing inode.
809 		 */
810 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
811 		cond_resched();
812 	}
813 	goto retry;
814 }
815 
816 /*
817  * Freeze all the FS-operations for checkpoint.
818  */
819 static int block_operations(struct f2fs_sb_info *sbi)
820 {
821 	struct writeback_control wbc = {
822 		.sync_mode = WB_SYNC_ALL,
823 		.nr_to_write = LONG_MAX,
824 		.for_reclaim = 0,
825 	};
826 	struct blk_plug plug;
827 	int err = 0;
828 
829 	blk_start_plug(&plug);
830 
831 retry_flush_dents:
832 	f2fs_lock_all(sbi);
833 	/* write all the dirty dentry pages */
834 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
835 		f2fs_unlock_all(sbi);
836 		sync_dirty_dir_inodes(sbi);
837 		if (unlikely(f2fs_cp_error(sbi))) {
838 			err = -EIO;
839 			goto out;
840 		}
841 		goto retry_flush_dents;
842 	}
843 
844 	/*
845 	 * POR: we should ensure that there are no dirty node pages
846 	 * until finishing nat/sit flush.
847 	 */
848 retry_flush_nodes:
849 	down_write(&sbi->node_write);
850 
851 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
852 		up_write(&sbi->node_write);
853 		sync_node_pages(sbi, 0, &wbc);
854 		if (unlikely(f2fs_cp_error(sbi))) {
855 			f2fs_unlock_all(sbi);
856 			err = -EIO;
857 			goto out;
858 		}
859 		goto retry_flush_nodes;
860 	}
861 out:
862 	blk_finish_plug(&plug);
863 	return err;
864 }
865 
866 static void unblock_operations(struct f2fs_sb_info *sbi)
867 {
868 	up_write(&sbi->node_write);
869 	f2fs_unlock_all(sbi);
870 }
871 
872 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
873 {
874 	DEFINE_WAIT(wait);
875 
876 	for (;;) {
877 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
878 
879 		if (!get_pages(sbi, F2FS_WRITEBACK))
880 			break;
881 
882 		io_schedule();
883 	}
884 	finish_wait(&sbi->cp_wait, &wait);
885 }
886 
887 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
888 {
889 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
890 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
891 	struct f2fs_nm_info *nm_i = NM_I(sbi);
892 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
893 	nid_t last_nid = nm_i->next_scan_nid;
894 	block_t start_blk;
895 	unsigned int data_sum_blocks, orphan_blocks;
896 	__u32 crc32 = 0;
897 	int i;
898 	int cp_payload_blks = __cp_payload(sbi);
899 
900 	/*
901 	 * This avoids to conduct wrong roll-forward operations and uses
902 	 * metapages, so should be called prior to sync_meta_pages below.
903 	 */
904 	discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
905 
906 	/* Flush all the NAT/SIT pages */
907 	while (get_pages(sbi, F2FS_DIRTY_META)) {
908 		sync_meta_pages(sbi, META, LONG_MAX);
909 		if (unlikely(f2fs_cp_error(sbi)))
910 			return;
911 	}
912 
913 	next_free_nid(sbi, &last_nid);
914 
915 	/*
916 	 * modify checkpoint
917 	 * version number is already updated
918 	 */
919 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
920 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
921 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
922 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
923 		ckpt->cur_node_segno[i] =
924 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
925 		ckpt->cur_node_blkoff[i] =
926 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
927 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
928 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
929 	}
930 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
931 		ckpt->cur_data_segno[i] =
932 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
933 		ckpt->cur_data_blkoff[i] =
934 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
935 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
936 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
937 	}
938 
939 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
940 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
941 	ckpt->next_free_nid = cpu_to_le32(last_nid);
942 
943 	/* 2 cp  + n data seg summary + orphan inode blocks */
944 	data_sum_blocks = npages_for_summary_flush(sbi, false);
945 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
946 		set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
947 	else
948 		clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
949 
950 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
951 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
952 			orphan_blocks);
953 
954 	if (__remain_node_summaries(cpc->reason))
955 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
956 				cp_payload_blks + data_sum_blocks +
957 				orphan_blocks + NR_CURSEG_NODE_TYPE);
958 	else
959 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
960 				cp_payload_blks + data_sum_blocks +
961 				orphan_blocks);
962 
963 	if (cpc->reason == CP_UMOUNT)
964 		set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
965 	else
966 		clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
967 
968 	if (cpc->reason == CP_FASTBOOT)
969 		set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
970 	else
971 		clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
972 
973 	if (orphan_num)
974 		set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
975 	else
976 		clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
977 
978 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
979 		set_ckpt_flags(ckpt, CP_FSCK_FLAG);
980 
981 	/* update SIT/NAT bitmap */
982 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
983 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
984 
985 	crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
986 	*((__le32 *)((unsigned char *)ckpt +
987 				le32_to_cpu(ckpt->checksum_offset)))
988 				= cpu_to_le32(crc32);
989 
990 	start_blk = __start_cp_addr(sbi);
991 
992 	/* write out checkpoint buffer at block 0 */
993 	update_meta_page(sbi, ckpt, start_blk++);
994 
995 	for (i = 1; i < 1 + cp_payload_blks; i++)
996 		update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
997 							start_blk++);
998 
999 	if (orphan_num) {
1000 		write_orphan_inodes(sbi, start_blk);
1001 		start_blk += orphan_blocks;
1002 	}
1003 
1004 	write_data_summaries(sbi, start_blk);
1005 	start_blk += data_sum_blocks;
1006 	if (__remain_node_summaries(cpc->reason)) {
1007 		write_node_summaries(sbi, start_blk);
1008 		start_blk += NR_CURSEG_NODE_TYPE;
1009 	}
1010 
1011 	/* writeout checkpoint block */
1012 	update_meta_page(sbi, ckpt, start_blk);
1013 
1014 	/* wait for previous submitted node/meta pages writeback */
1015 	wait_on_all_pages_writeback(sbi);
1016 
1017 	if (unlikely(f2fs_cp_error(sbi)))
1018 		return;
1019 
1020 	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1021 	filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1022 
1023 	/* update user_block_counts */
1024 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1025 	sbi->alloc_valid_block_count = 0;
1026 
1027 	/* Here, we only have one bio having CP pack */
1028 	sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1029 
1030 	/* wait for previous submitted meta pages writeback */
1031 	wait_on_all_pages_writeback(sbi);
1032 
1033 	release_dirty_inode(sbi);
1034 
1035 	if (unlikely(f2fs_cp_error(sbi)))
1036 		return;
1037 
1038 	clear_prefree_segments(sbi, cpc);
1039 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1040 }
1041 
1042 /*
1043  * We guarantee that this checkpoint procedure will not fail.
1044  */
1045 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1046 {
1047 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1048 	unsigned long long ckpt_ver;
1049 
1050 	mutex_lock(&sbi->cp_mutex);
1051 
1052 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1053 		(cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1054 		(cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1055 		goto out;
1056 	if (unlikely(f2fs_cp_error(sbi)))
1057 		goto out;
1058 	if (f2fs_readonly(sbi->sb))
1059 		goto out;
1060 
1061 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1062 
1063 	if (block_operations(sbi))
1064 		goto out;
1065 
1066 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1067 
1068 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
1069 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
1070 	f2fs_submit_merged_bio(sbi, META, WRITE);
1071 
1072 	/*
1073 	 * update checkpoint pack index
1074 	 * Increase the version number so that
1075 	 * SIT entries and seg summaries are written at correct place
1076 	 */
1077 	ckpt_ver = cur_cp_version(ckpt);
1078 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1079 
1080 	/* write cached NAT/SIT entries to NAT/SIT area */
1081 	flush_nat_entries(sbi);
1082 	flush_sit_entries(sbi, cpc);
1083 
1084 	/* unlock all the fs_lock[] in do_checkpoint() */
1085 	do_checkpoint(sbi, cpc);
1086 
1087 	unblock_operations(sbi);
1088 	stat_inc_cp_count(sbi->stat_info);
1089 
1090 	if (cpc->reason == CP_RECOVERY)
1091 		f2fs_msg(sbi->sb, KERN_NOTICE,
1092 			"checkpoint: version = %llx", ckpt_ver);
1093 out:
1094 	mutex_unlock(&sbi->cp_mutex);
1095 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1096 }
1097 
1098 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1099 {
1100 	int i;
1101 
1102 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1103 		struct inode_management *im = &sbi->im[i];
1104 
1105 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1106 		spin_lock_init(&im->ino_lock);
1107 		INIT_LIST_HEAD(&im->ino_list);
1108 		im->ino_num = 0;
1109 	}
1110 
1111 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1112 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1113 				F2FS_ORPHANS_PER_BLOCK;
1114 }
1115 
1116 int __init create_checkpoint_caches(void)
1117 {
1118 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1119 			sizeof(struct ino_entry));
1120 	if (!ino_entry_slab)
1121 		return -ENOMEM;
1122 	inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1123 			sizeof(struct inode_entry));
1124 	if (!inode_entry_slab) {
1125 		kmem_cache_destroy(ino_entry_slab);
1126 		return -ENOMEM;
1127 	}
1128 	return 0;
1129 }
1130 
1131 void destroy_checkpoint_caches(void)
1132 {
1133 	kmem_cache_destroy(ino_entry_slab);
1134 	kmem_cache_destroy(inode_entry_slab);
1135 }
1136