xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision a7ffdbe22cecaed59b5d76a5f003d68907d64240)
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/events/f2fs.h>
24 
25 static struct kmem_cache *ino_entry_slab;
26 static struct kmem_cache *inode_entry_slab;
27 
28 /*
29  * We guarantee no failure on the returned page.
30  */
31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
32 {
33 	struct address_space *mapping = META_MAPPING(sbi);
34 	struct page *page = NULL;
35 repeat:
36 	page = grab_cache_page(mapping, index);
37 	if (!page) {
38 		cond_resched();
39 		goto repeat;
40 	}
41 	f2fs_wait_on_page_writeback(page, META);
42 	SetPageUptodate(page);
43 	return page;
44 }
45 
46 /*
47  * We guarantee no failure on the returned page.
48  */
49 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
50 {
51 	struct address_space *mapping = META_MAPPING(sbi);
52 	struct page *page;
53 repeat:
54 	page = grab_cache_page(mapping, index);
55 	if (!page) {
56 		cond_resched();
57 		goto repeat;
58 	}
59 	if (PageUptodate(page))
60 		goto out;
61 
62 	if (f2fs_submit_page_bio(sbi, page, index,
63 				READ_SYNC | REQ_META | REQ_PRIO))
64 		goto repeat;
65 
66 	lock_page(page);
67 	if (unlikely(page->mapping != mapping)) {
68 		f2fs_put_page(page, 1);
69 		goto repeat;
70 	}
71 out:
72 	return page;
73 }
74 
75 static inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
76 {
77 	switch (type) {
78 	case META_NAT:
79 		return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
80 	case META_SIT:
81 		return SIT_BLK_CNT(sbi);
82 	case META_SSA:
83 	case META_CP:
84 		return 0;
85 	default:
86 		BUG();
87 	}
88 }
89 
90 /*
91  * Readahead CP/NAT/SIT/SSA pages
92  */
93 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
94 {
95 	block_t prev_blk_addr = 0;
96 	struct page *page;
97 	int blkno = start;
98 	int max_blks = get_max_meta_blks(sbi, type);
99 
100 	struct f2fs_io_info fio = {
101 		.type = META,
102 		.rw = READ_SYNC | REQ_META | REQ_PRIO
103 	};
104 
105 	for (; nrpages-- > 0; blkno++) {
106 		block_t blk_addr;
107 
108 		switch (type) {
109 		case META_NAT:
110 			/* get nat block addr */
111 			if (unlikely(blkno >= max_blks))
112 				blkno = 0;
113 			blk_addr = current_nat_addr(sbi,
114 					blkno * NAT_ENTRY_PER_BLOCK);
115 			break;
116 		case META_SIT:
117 			/* get sit block addr */
118 			if (unlikely(blkno >= max_blks))
119 				goto out;
120 			blk_addr = current_sit_addr(sbi,
121 					blkno * SIT_ENTRY_PER_BLOCK);
122 			if (blkno != start && prev_blk_addr + 1 != blk_addr)
123 				goto out;
124 			prev_blk_addr = blk_addr;
125 			break;
126 		case META_SSA:
127 		case META_CP:
128 			/* get ssa/cp block addr */
129 			blk_addr = blkno;
130 			break;
131 		default:
132 			BUG();
133 		}
134 
135 		page = grab_cache_page(META_MAPPING(sbi), blk_addr);
136 		if (!page)
137 			continue;
138 		if (PageUptodate(page)) {
139 			f2fs_put_page(page, 1);
140 			continue;
141 		}
142 
143 		f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
144 		f2fs_put_page(page, 0);
145 	}
146 out:
147 	f2fs_submit_merged_bio(sbi, META, READ);
148 	return blkno - start;
149 }
150 
151 static int f2fs_write_meta_page(struct page *page,
152 				struct writeback_control *wbc)
153 {
154 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
155 
156 	trace_f2fs_writepage(page, META);
157 
158 	if (unlikely(sbi->por_doing))
159 		goto redirty_out;
160 	if (wbc->for_reclaim)
161 		goto redirty_out;
162 	if (unlikely(f2fs_cp_error(sbi)))
163 		goto redirty_out;
164 
165 	f2fs_wait_on_page_writeback(page, META);
166 	write_meta_page(sbi, page);
167 	dec_page_count(sbi, F2FS_DIRTY_META);
168 	unlock_page(page);
169 	return 0;
170 
171 redirty_out:
172 	redirty_page_for_writepage(wbc, page);
173 	return AOP_WRITEPAGE_ACTIVATE;
174 }
175 
176 static int f2fs_write_meta_pages(struct address_space *mapping,
177 				struct writeback_control *wbc)
178 {
179 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
180 	long diff, written;
181 
182 	trace_f2fs_writepages(mapping->host, wbc, META);
183 
184 	/* collect a number of dirty meta pages and write together */
185 	if (wbc->for_kupdate ||
186 		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
187 		goto skip_write;
188 
189 	/* if mounting is failed, skip writing node pages */
190 	mutex_lock(&sbi->cp_mutex);
191 	diff = nr_pages_to_write(sbi, META, wbc);
192 	written = sync_meta_pages(sbi, META, wbc->nr_to_write);
193 	mutex_unlock(&sbi->cp_mutex);
194 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
195 	return 0;
196 
197 skip_write:
198 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
199 	return 0;
200 }
201 
202 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
203 						long nr_to_write)
204 {
205 	struct address_space *mapping = META_MAPPING(sbi);
206 	pgoff_t index = 0, end = LONG_MAX;
207 	struct pagevec pvec;
208 	long nwritten = 0;
209 	struct writeback_control wbc = {
210 		.for_reclaim = 0,
211 	};
212 
213 	pagevec_init(&pvec, 0);
214 
215 	while (index <= end) {
216 		int i, nr_pages;
217 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
218 				PAGECACHE_TAG_DIRTY,
219 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
220 		if (unlikely(nr_pages == 0))
221 			break;
222 
223 		for (i = 0; i < nr_pages; i++) {
224 			struct page *page = pvec.pages[i];
225 
226 			lock_page(page);
227 
228 			if (unlikely(page->mapping != mapping)) {
229 continue_unlock:
230 				unlock_page(page);
231 				continue;
232 			}
233 			if (!PageDirty(page)) {
234 				/* someone wrote it for us */
235 				goto continue_unlock;
236 			}
237 
238 			if (!clear_page_dirty_for_io(page))
239 				goto continue_unlock;
240 
241 			if (f2fs_write_meta_page(page, &wbc)) {
242 				unlock_page(page);
243 				break;
244 			}
245 			nwritten++;
246 			if (unlikely(nwritten >= nr_to_write))
247 				break;
248 		}
249 		pagevec_release(&pvec);
250 		cond_resched();
251 	}
252 
253 	if (nwritten)
254 		f2fs_submit_merged_bio(sbi, type, WRITE);
255 
256 	return nwritten;
257 }
258 
259 static int f2fs_set_meta_page_dirty(struct page *page)
260 {
261 	trace_f2fs_set_page_dirty(page, META);
262 
263 	SetPageUptodate(page);
264 	if (!PageDirty(page)) {
265 		__set_page_dirty_nobuffers(page);
266 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
267 		return 1;
268 	}
269 	return 0;
270 }
271 
272 const struct address_space_operations f2fs_meta_aops = {
273 	.writepage	= f2fs_write_meta_page,
274 	.writepages	= f2fs_write_meta_pages,
275 	.set_page_dirty	= f2fs_set_meta_page_dirty,
276 };
277 
278 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
279 {
280 	struct ino_entry *e;
281 retry:
282 	spin_lock(&sbi->ino_lock[type]);
283 
284 	e = radix_tree_lookup(&sbi->ino_root[type], ino);
285 	if (!e) {
286 		e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
287 		if (!e) {
288 			spin_unlock(&sbi->ino_lock[type]);
289 			goto retry;
290 		}
291 		if (radix_tree_insert(&sbi->ino_root[type], ino, e)) {
292 			spin_unlock(&sbi->ino_lock[type]);
293 			kmem_cache_free(ino_entry_slab, e);
294 			goto retry;
295 		}
296 		memset(e, 0, sizeof(struct ino_entry));
297 		e->ino = ino;
298 
299 		list_add_tail(&e->list, &sbi->ino_list[type]);
300 	}
301 	spin_unlock(&sbi->ino_lock[type]);
302 }
303 
304 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
305 {
306 	struct ino_entry *e;
307 
308 	spin_lock(&sbi->ino_lock[type]);
309 	e = radix_tree_lookup(&sbi->ino_root[type], ino);
310 	if (e) {
311 		list_del(&e->list);
312 		radix_tree_delete(&sbi->ino_root[type], ino);
313 		if (type == ORPHAN_INO)
314 			sbi->n_orphans--;
315 		spin_unlock(&sbi->ino_lock[type]);
316 		kmem_cache_free(ino_entry_slab, e);
317 		return;
318 	}
319 	spin_unlock(&sbi->ino_lock[type]);
320 }
321 
322 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
323 {
324 	/* add new dirty ino entry into list */
325 	__add_ino_entry(sbi, ino, type);
326 }
327 
328 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
329 {
330 	/* remove dirty ino entry from list */
331 	__remove_ino_entry(sbi, ino, type);
332 }
333 
334 /* mode should be APPEND_INO or UPDATE_INO */
335 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
336 {
337 	struct ino_entry *e;
338 	spin_lock(&sbi->ino_lock[mode]);
339 	e = radix_tree_lookup(&sbi->ino_root[mode], ino);
340 	spin_unlock(&sbi->ino_lock[mode]);
341 	return e ? true : false;
342 }
343 
344 void release_dirty_inode(struct f2fs_sb_info *sbi)
345 {
346 	struct ino_entry *e, *tmp;
347 	int i;
348 
349 	for (i = APPEND_INO; i <= UPDATE_INO; i++) {
350 		spin_lock(&sbi->ino_lock[i]);
351 		list_for_each_entry_safe(e, tmp, &sbi->ino_list[i], list) {
352 			list_del(&e->list);
353 			radix_tree_delete(&sbi->ino_root[i], e->ino);
354 			kmem_cache_free(ino_entry_slab, e);
355 		}
356 		spin_unlock(&sbi->ino_lock[i]);
357 	}
358 }
359 
360 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
361 {
362 	int err = 0;
363 
364 	spin_lock(&sbi->ino_lock[ORPHAN_INO]);
365 	if (unlikely(sbi->n_orphans >= sbi->max_orphans))
366 		err = -ENOSPC;
367 	else
368 		sbi->n_orphans++;
369 	spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
370 
371 	return err;
372 }
373 
374 void release_orphan_inode(struct f2fs_sb_info *sbi)
375 {
376 	spin_lock(&sbi->ino_lock[ORPHAN_INO]);
377 	f2fs_bug_on(sbi, sbi->n_orphans == 0);
378 	sbi->n_orphans--;
379 	spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
380 }
381 
382 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
383 {
384 	/* add new orphan ino entry into list */
385 	__add_ino_entry(sbi, ino, ORPHAN_INO);
386 }
387 
388 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
389 {
390 	/* remove orphan entry from orphan list */
391 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
392 }
393 
394 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
395 {
396 	struct inode *inode = f2fs_iget(sbi->sb, ino);
397 	f2fs_bug_on(sbi, IS_ERR(inode));
398 	clear_nlink(inode);
399 
400 	/* truncate all the data during iput */
401 	iput(inode);
402 }
403 
404 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
405 {
406 	block_t start_blk, orphan_blkaddr, i, j;
407 
408 	if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
409 		return;
410 
411 	sbi->por_doing = true;
412 
413 	start_blk = __start_cp_addr(sbi) + 1 +
414 		le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
415 	orphan_blkaddr = __start_sum_addr(sbi) - 1;
416 
417 	ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
418 
419 	for (i = 0; i < orphan_blkaddr; i++) {
420 		struct page *page = get_meta_page(sbi, start_blk + i);
421 		struct f2fs_orphan_block *orphan_blk;
422 
423 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
424 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
425 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
426 			recover_orphan_inode(sbi, ino);
427 		}
428 		f2fs_put_page(page, 1);
429 	}
430 	/* clear Orphan Flag */
431 	clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
432 	sbi->por_doing = false;
433 	return;
434 }
435 
436 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
437 {
438 	struct list_head *head;
439 	struct f2fs_orphan_block *orphan_blk = NULL;
440 	unsigned int nentries = 0;
441 	unsigned short index;
442 	unsigned short orphan_blocks =
443 			(unsigned short)GET_ORPHAN_BLOCKS(sbi->n_orphans);
444 	struct page *page = NULL;
445 	struct ino_entry *orphan = NULL;
446 
447 	for (index = 0; index < orphan_blocks; index++)
448 		grab_meta_page(sbi, start_blk + index);
449 
450 	index = 1;
451 	spin_lock(&sbi->ino_lock[ORPHAN_INO]);
452 	head = &sbi->ino_list[ORPHAN_INO];
453 
454 	/* loop for each orphan inode entry and write them in Jornal block */
455 	list_for_each_entry(orphan, head, list) {
456 		if (!page) {
457 			page = find_get_page(META_MAPPING(sbi), start_blk++);
458 			f2fs_bug_on(sbi, !page);
459 			orphan_blk =
460 				(struct f2fs_orphan_block *)page_address(page);
461 			memset(orphan_blk, 0, sizeof(*orphan_blk));
462 			f2fs_put_page(page, 0);
463 		}
464 
465 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
466 
467 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
468 			/*
469 			 * an orphan block is full of 1020 entries,
470 			 * then we need to flush current orphan blocks
471 			 * and bring another one in memory
472 			 */
473 			orphan_blk->blk_addr = cpu_to_le16(index);
474 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
475 			orphan_blk->entry_count = cpu_to_le32(nentries);
476 			set_page_dirty(page);
477 			f2fs_put_page(page, 1);
478 			index++;
479 			nentries = 0;
480 			page = NULL;
481 		}
482 	}
483 
484 	if (page) {
485 		orphan_blk->blk_addr = cpu_to_le16(index);
486 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
487 		orphan_blk->entry_count = cpu_to_le32(nentries);
488 		set_page_dirty(page);
489 		f2fs_put_page(page, 1);
490 	}
491 
492 	spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
493 }
494 
495 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
496 				block_t cp_addr, unsigned long long *version)
497 {
498 	struct page *cp_page_1, *cp_page_2 = NULL;
499 	unsigned long blk_size = sbi->blocksize;
500 	struct f2fs_checkpoint *cp_block;
501 	unsigned long long cur_version = 0, pre_version = 0;
502 	size_t crc_offset;
503 	__u32 crc = 0;
504 
505 	/* Read the 1st cp block in this CP pack */
506 	cp_page_1 = get_meta_page(sbi, cp_addr);
507 
508 	/* get the version number */
509 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
510 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
511 	if (crc_offset >= blk_size)
512 		goto invalid_cp1;
513 
514 	crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
515 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
516 		goto invalid_cp1;
517 
518 	pre_version = cur_cp_version(cp_block);
519 
520 	/* Read the 2nd cp block in this CP pack */
521 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
522 	cp_page_2 = get_meta_page(sbi, cp_addr);
523 
524 	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
525 	crc_offset = le32_to_cpu(cp_block->checksum_offset);
526 	if (crc_offset >= blk_size)
527 		goto invalid_cp2;
528 
529 	crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
530 	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
531 		goto invalid_cp2;
532 
533 	cur_version = cur_cp_version(cp_block);
534 
535 	if (cur_version == pre_version) {
536 		*version = cur_version;
537 		f2fs_put_page(cp_page_2, 1);
538 		return cp_page_1;
539 	}
540 invalid_cp2:
541 	f2fs_put_page(cp_page_2, 1);
542 invalid_cp1:
543 	f2fs_put_page(cp_page_1, 1);
544 	return NULL;
545 }
546 
547 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
548 {
549 	struct f2fs_checkpoint *cp_block;
550 	struct f2fs_super_block *fsb = sbi->raw_super;
551 	struct page *cp1, *cp2, *cur_page;
552 	unsigned long blk_size = sbi->blocksize;
553 	unsigned long long cp1_version = 0, cp2_version = 0;
554 	unsigned long long cp_start_blk_no;
555 	unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
556 	block_t cp_blk_no;
557 	int i;
558 
559 	sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
560 	if (!sbi->ckpt)
561 		return -ENOMEM;
562 	/*
563 	 * Finding out valid cp block involves read both
564 	 * sets( cp pack1 and cp pack 2)
565 	 */
566 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
567 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
568 
569 	/* The second checkpoint pack should start at the next segment */
570 	cp_start_blk_no += ((unsigned long long)1) <<
571 				le32_to_cpu(fsb->log_blocks_per_seg);
572 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
573 
574 	if (cp1 && cp2) {
575 		if (ver_after(cp2_version, cp1_version))
576 			cur_page = cp2;
577 		else
578 			cur_page = cp1;
579 	} else if (cp1) {
580 		cur_page = cp1;
581 	} else if (cp2) {
582 		cur_page = cp2;
583 	} else {
584 		goto fail_no_cp;
585 	}
586 
587 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
588 	memcpy(sbi->ckpt, cp_block, blk_size);
589 
590 	if (cp_blks <= 1)
591 		goto done;
592 
593 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
594 	if (cur_page == cp2)
595 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
596 
597 	for (i = 1; i < cp_blks; i++) {
598 		void *sit_bitmap_ptr;
599 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
600 
601 		cur_page = get_meta_page(sbi, cp_blk_no + i);
602 		sit_bitmap_ptr = page_address(cur_page);
603 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
604 		f2fs_put_page(cur_page, 1);
605 	}
606 done:
607 	f2fs_put_page(cp1, 1);
608 	f2fs_put_page(cp2, 1);
609 	return 0;
610 
611 fail_no_cp:
612 	kfree(sbi->ckpt);
613 	return -EINVAL;
614 }
615 
616 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
617 {
618 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
619 
620 	if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
621 		return -EEXIST;
622 
623 	set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
624 	F2FS_I(inode)->dirty_dir = new;
625 	list_add_tail(&new->list, &sbi->dir_inode_list);
626 	stat_inc_dirty_dir(sbi);
627 	return 0;
628 }
629 
630 void update_dirty_page(struct inode *inode, struct page *page)
631 {
632 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
633 	struct dir_inode_entry *new;
634 	int ret = 0;
635 
636 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
637 		return;
638 
639 	if (!S_ISDIR(inode->i_mode)) {
640 		inode_inc_dirty_pages(inode);
641 		goto out;
642 	}
643 
644 	new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
645 	new->inode = inode;
646 	INIT_LIST_HEAD(&new->list);
647 
648 	spin_lock(&sbi->dir_inode_lock);
649 	ret = __add_dirty_inode(inode, new);
650 	inode_inc_dirty_pages(inode);
651 	spin_unlock(&sbi->dir_inode_lock);
652 
653 	if (ret)
654 		kmem_cache_free(inode_entry_slab, new);
655 out:
656 	SetPagePrivate(page);
657 }
658 
659 void add_dirty_dir_inode(struct inode *inode)
660 {
661 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
662 	struct dir_inode_entry *new =
663 			f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
664 	int ret = 0;
665 
666 	new->inode = inode;
667 	INIT_LIST_HEAD(&new->list);
668 
669 	spin_lock(&sbi->dir_inode_lock);
670 	ret = __add_dirty_inode(inode, new);
671 	spin_unlock(&sbi->dir_inode_lock);
672 
673 	if (ret)
674 		kmem_cache_free(inode_entry_slab, new);
675 }
676 
677 void remove_dirty_dir_inode(struct inode *inode)
678 {
679 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
680 	struct dir_inode_entry *entry;
681 
682 	if (!S_ISDIR(inode->i_mode))
683 		return;
684 
685 	spin_lock(&sbi->dir_inode_lock);
686 	if (get_dirty_pages(inode) ||
687 			!is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
688 		spin_unlock(&sbi->dir_inode_lock);
689 		return;
690 	}
691 
692 	entry = F2FS_I(inode)->dirty_dir;
693 	list_del(&entry->list);
694 	F2FS_I(inode)->dirty_dir = NULL;
695 	clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
696 	stat_dec_dirty_dir(sbi);
697 	spin_unlock(&sbi->dir_inode_lock);
698 	kmem_cache_free(inode_entry_slab, entry);
699 
700 	/* Only from the recovery routine */
701 	if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
702 		clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
703 		iput(inode);
704 	}
705 }
706 
707 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
708 {
709 	struct list_head *head;
710 	struct dir_inode_entry *entry;
711 	struct inode *inode;
712 retry:
713 	spin_lock(&sbi->dir_inode_lock);
714 
715 	head = &sbi->dir_inode_list;
716 	if (list_empty(head)) {
717 		spin_unlock(&sbi->dir_inode_lock);
718 		return;
719 	}
720 	entry = list_entry(head->next, struct dir_inode_entry, list);
721 	inode = igrab(entry->inode);
722 	spin_unlock(&sbi->dir_inode_lock);
723 	if (inode) {
724 		filemap_fdatawrite(inode->i_mapping);
725 		iput(inode);
726 	} else {
727 		/*
728 		 * We should submit bio, since it exists several
729 		 * wribacking dentry pages in the freeing inode.
730 		 */
731 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
732 	}
733 	goto retry;
734 }
735 
736 /*
737  * Freeze all the FS-operations for checkpoint.
738  */
739 static int block_operations(struct f2fs_sb_info *sbi)
740 {
741 	struct writeback_control wbc = {
742 		.sync_mode = WB_SYNC_ALL,
743 		.nr_to_write = LONG_MAX,
744 		.for_reclaim = 0,
745 	};
746 	struct blk_plug plug;
747 	int err = 0;
748 
749 	blk_start_plug(&plug);
750 
751 retry_flush_dents:
752 	f2fs_lock_all(sbi);
753 	/* write all the dirty dentry pages */
754 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
755 		f2fs_unlock_all(sbi);
756 		sync_dirty_dir_inodes(sbi);
757 		if (unlikely(f2fs_cp_error(sbi))) {
758 			err = -EIO;
759 			goto out;
760 		}
761 		goto retry_flush_dents;
762 	}
763 
764 	/*
765 	 * POR: we should ensure that there are no dirty node pages
766 	 * until finishing nat/sit flush.
767 	 */
768 retry_flush_nodes:
769 	down_write(&sbi->node_write);
770 
771 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
772 		up_write(&sbi->node_write);
773 		sync_node_pages(sbi, 0, &wbc);
774 		if (unlikely(f2fs_cp_error(sbi))) {
775 			f2fs_unlock_all(sbi);
776 			err = -EIO;
777 			goto out;
778 		}
779 		goto retry_flush_nodes;
780 	}
781 out:
782 	blk_finish_plug(&plug);
783 	return err;
784 }
785 
786 static void unblock_operations(struct f2fs_sb_info *sbi)
787 {
788 	up_write(&sbi->node_write);
789 	f2fs_unlock_all(sbi);
790 }
791 
792 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
793 {
794 	DEFINE_WAIT(wait);
795 
796 	for (;;) {
797 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
798 
799 		if (!get_pages(sbi, F2FS_WRITEBACK))
800 			break;
801 
802 		io_schedule();
803 	}
804 	finish_wait(&sbi->cp_wait, &wait);
805 }
806 
807 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
808 {
809 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
810 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
811 	nid_t last_nid = 0;
812 	block_t start_blk;
813 	struct page *cp_page;
814 	unsigned int data_sum_blocks, orphan_blocks;
815 	__u32 crc32 = 0;
816 	void *kaddr;
817 	int i;
818 	int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
819 
820 	/*
821 	 * This avoids to conduct wrong roll-forward operations and uses
822 	 * metapages, so should be called prior to sync_meta_pages below.
823 	 */
824 	discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
825 
826 	/* Flush all the NAT/SIT pages */
827 	while (get_pages(sbi, F2FS_DIRTY_META)) {
828 		sync_meta_pages(sbi, META, LONG_MAX);
829 		if (unlikely(f2fs_cp_error(sbi)))
830 			return;
831 	}
832 
833 	next_free_nid(sbi, &last_nid);
834 
835 	/*
836 	 * modify checkpoint
837 	 * version number is already updated
838 	 */
839 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
840 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
841 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
842 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
843 		ckpt->cur_node_segno[i] =
844 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
845 		ckpt->cur_node_blkoff[i] =
846 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
847 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
848 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
849 	}
850 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
851 		ckpt->cur_data_segno[i] =
852 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
853 		ckpt->cur_data_blkoff[i] =
854 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
855 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
856 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
857 	}
858 
859 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
860 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
861 	ckpt->next_free_nid = cpu_to_le32(last_nid);
862 
863 	/* 2 cp  + n data seg summary + orphan inode blocks */
864 	data_sum_blocks = npages_for_summary_flush(sbi);
865 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
866 		set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
867 	else
868 		clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
869 
870 	orphan_blocks = GET_ORPHAN_BLOCKS(sbi->n_orphans);
871 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
872 			orphan_blocks);
873 
874 	if (is_umount) {
875 		set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
876 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
877 				cp_payload_blks + data_sum_blocks +
878 				orphan_blocks + NR_CURSEG_NODE_TYPE);
879 	} else {
880 		clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
881 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
882 				cp_payload_blks + data_sum_blocks +
883 				orphan_blocks);
884 	}
885 
886 	if (sbi->n_orphans)
887 		set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
888 	else
889 		clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
890 
891 	if (sbi->need_fsck)
892 		set_ckpt_flags(ckpt, CP_FSCK_FLAG);
893 
894 	/* update SIT/NAT bitmap */
895 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
896 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
897 
898 	crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
899 	*((__le32 *)((unsigned char *)ckpt +
900 				le32_to_cpu(ckpt->checksum_offset)))
901 				= cpu_to_le32(crc32);
902 
903 	start_blk = __start_cp_addr(sbi);
904 
905 	/* write out checkpoint buffer at block 0 */
906 	cp_page = grab_meta_page(sbi, start_blk++);
907 	kaddr = page_address(cp_page);
908 	memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
909 	set_page_dirty(cp_page);
910 	f2fs_put_page(cp_page, 1);
911 
912 	for (i = 1; i < 1 + cp_payload_blks; i++) {
913 		cp_page = grab_meta_page(sbi, start_blk++);
914 		kaddr = page_address(cp_page);
915 		memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
916 				(1 << sbi->log_blocksize));
917 		set_page_dirty(cp_page);
918 		f2fs_put_page(cp_page, 1);
919 	}
920 
921 	if (sbi->n_orphans) {
922 		write_orphan_inodes(sbi, start_blk);
923 		start_blk += orphan_blocks;
924 	}
925 
926 	write_data_summaries(sbi, start_blk);
927 	start_blk += data_sum_blocks;
928 	if (is_umount) {
929 		write_node_summaries(sbi, start_blk);
930 		start_blk += NR_CURSEG_NODE_TYPE;
931 	}
932 
933 	/* writeout checkpoint block */
934 	cp_page = grab_meta_page(sbi, start_blk);
935 	kaddr = page_address(cp_page);
936 	memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
937 	set_page_dirty(cp_page);
938 	f2fs_put_page(cp_page, 1);
939 
940 	/* wait for previous submitted node/meta pages writeback */
941 	wait_on_all_pages_writeback(sbi);
942 
943 	if (unlikely(f2fs_cp_error(sbi)))
944 		return;
945 
946 	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
947 	filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
948 
949 	/* update user_block_counts */
950 	sbi->last_valid_block_count = sbi->total_valid_block_count;
951 	sbi->alloc_valid_block_count = 0;
952 
953 	/* Here, we only have one bio having CP pack */
954 	sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
955 
956 	release_dirty_inode(sbi);
957 
958 	if (unlikely(f2fs_cp_error(sbi)))
959 		return;
960 
961 	clear_prefree_segments(sbi);
962 	F2FS_RESET_SB_DIRT(sbi);
963 }
964 
965 /*
966  * We guarantee that this checkpoint procedure will not fail.
967  */
968 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
969 {
970 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
971 	unsigned long long ckpt_ver;
972 
973 	trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
974 
975 	mutex_lock(&sbi->cp_mutex);
976 
977 	if (!sbi->s_dirty)
978 		goto out;
979 	if (unlikely(f2fs_cp_error(sbi)))
980 		goto out;
981 	if (block_operations(sbi))
982 		goto out;
983 
984 	trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
985 
986 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
987 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
988 	f2fs_submit_merged_bio(sbi, META, WRITE);
989 
990 	/*
991 	 * update checkpoint pack index
992 	 * Increase the version number so that
993 	 * SIT entries and seg summaries are written at correct place
994 	 */
995 	ckpt_ver = cur_cp_version(ckpt);
996 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
997 
998 	/* write cached NAT/SIT entries to NAT/SIT area */
999 	flush_nat_entries(sbi);
1000 	flush_sit_entries(sbi);
1001 
1002 	/* unlock all the fs_lock[] in do_checkpoint() */
1003 	do_checkpoint(sbi, is_umount);
1004 
1005 	unblock_operations(sbi);
1006 	stat_inc_cp_count(sbi->stat_info);
1007 out:
1008 	mutex_unlock(&sbi->cp_mutex);
1009 	trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
1010 }
1011 
1012 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1013 {
1014 	int i;
1015 
1016 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1017 		INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
1018 		spin_lock_init(&sbi->ino_lock[i]);
1019 		INIT_LIST_HEAD(&sbi->ino_list[i]);
1020 	}
1021 
1022 	/*
1023 	 * considering 512 blocks in a segment 8 blocks are needed for cp
1024 	 * and log segment summaries. Remaining blocks are used to keep
1025 	 * orphan entries with the limitation one reserved segment
1026 	 * for cp pack we can have max 1020*504 orphan entries
1027 	 */
1028 	sbi->n_orphans = 0;
1029 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1030 			NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1031 }
1032 
1033 int __init create_checkpoint_caches(void)
1034 {
1035 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1036 			sizeof(struct ino_entry));
1037 	if (!ino_entry_slab)
1038 		return -ENOMEM;
1039 	inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1040 			sizeof(struct dir_inode_entry));
1041 	if (!inode_entry_slab) {
1042 		kmem_cache_destroy(ino_entry_slab);
1043 		return -ENOMEM;
1044 	}
1045 	return 0;
1046 }
1047 
1048 void destroy_checkpoint_caches(void)
1049 {
1050 	kmem_cache_destroy(ino_entry_slab);
1051 	kmem_cache_destroy(inode_entry_slab);
1052 }
1053