xref: /openbmc/linux/fs/f2fs/checkpoint.c (revision 8bd1369b)
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 *f2fs_inode_entry_slab;
28 
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
30 {
31 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
32 	if (!end_io)
33 		f2fs_flush_merged_writes(sbi);
34 }
35 
36 /*
37  * We guarantee no failure on the returned page.
38  */
39 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
40 {
41 	struct address_space *mapping = META_MAPPING(sbi);
42 	struct page *page = NULL;
43 repeat:
44 	page = f2fs_grab_cache_page(mapping, index, false);
45 	if (!page) {
46 		cond_resched();
47 		goto repeat;
48 	}
49 	f2fs_wait_on_page_writeback(page, META, true);
50 	if (!PageUptodate(page))
51 		SetPageUptodate(page);
52 	return page;
53 }
54 
55 /*
56  * We guarantee no failure on the returned page.
57  */
58 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
59 							bool is_meta)
60 {
61 	struct address_space *mapping = META_MAPPING(sbi);
62 	struct page *page;
63 	struct f2fs_io_info fio = {
64 		.sbi = sbi,
65 		.type = META,
66 		.op = REQ_OP_READ,
67 		.op_flags = REQ_META | REQ_PRIO,
68 		.old_blkaddr = index,
69 		.new_blkaddr = index,
70 		.encrypted_page = NULL,
71 		.is_meta = is_meta,
72 	};
73 
74 	if (unlikely(!is_meta))
75 		fio.op_flags &= ~REQ_META;
76 repeat:
77 	page = f2fs_grab_cache_page(mapping, index, false);
78 	if (!page) {
79 		cond_resched();
80 		goto repeat;
81 	}
82 	if (PageUptodate(page))
83 		goto out;
84 
85 	fio.page = page;
86 
87 	if (f2fs_submit_page_bio(&fio)) {
88 		f2fs_put_page(page, 1);
89 		goto repeat;
90 	}
91 
92 	lock_page(page);
93 	if (unlikely(page->mapping != mapping)) {
94 		f2fs_put_page(page, 1);
95 		goto repeat;
96 	}
97 
98 	/*
99 	 * if there is any IO error when accessing device, make our filesystem
100 	 * readonly and make sure do not write checkpoint with non-uptodate
101 	 * meta page.
102 	 */
103 	if (unlikely(!PageUptodate(page))) {
104 		memset(page_address(page), 0, PAGE_SIZE);
105 		f2fs_stop_checkpoint(sbi, false);
106 	}
107 out:
108 	return page;
109 }
110 
111 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
112 {
113 	return __get_meta_page(sbi, index, true);
114 }
115 
116 /* for POR only */
117 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
118 {
119 	return __get_meta_page(sbi, index, false);
120 }
121 
122 bool f2fs_is_valid_meta_blkaddr(struct f2fs_sb_info *sbi,
123 					block_t blkaddr, int type)
124 {
125 	switch (type) {
126 	case META_NAT:
127 		break;
128 	case META_SIT:
129 		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
130 			return false;
131 		break;
132 	case META_SSA:
133 		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
134 			blkaddr < SM_I(sbi)->ssa_blkaddr))
135 			return false;
136 		break;
137 	case META_CP:
138 		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
139 			blkaddr < __start_cp_addr(sbi)))
140 			return false;
141 		break;
142 	case META_POR:
143 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
144 			blkaddr < MAIN_BLKADDR(sbi)))
145 			return false;
146 		break;
147 	default:
148 		BUG();
149 	}
150 
151 	return true;
152 }
153 
154 /*
155  * Readahead CP/NAT/SIT/SSA pages
156  */
157 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
158 							int type, bool sync)
159 {
160 	struct page *page;
161 	block_t blkno = start;
162 	struct f2fs_io_info fio = {
163 		.sbi = sbi,
164 		.type = META,
165 		.op = REQ_OP_READ,
166 		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
167 		.encrypted_page = NULL,
168 		.in_list = false,
169 		.is_meta = (type != META_POR),
170 	};
171 	struct blk_plug plug;
172 
173 	if (unlikely(type == META_POR))
174 		fio.op_flags &= ~REQ_META;
175 
176 	blk_start_plug(&plug);
177 	for (; nrpages-- > 0; blkno++) {
178 
179 		if (!f2fs_is_valid_meta_blkaddr(sbi, blkno, type))
180 			goto out;
181 
182 		switch (type) {
183 		case META_NAT:
184 			if (unlikely(blkno >=
185 					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
186 				blkno = 0;
187 			/* get nat block addr */
188 			fio.new_blkaddr = current_nat_addr(sbi,
189 					blkno * NAT_ENTRY_PER_BLOCK);
190 			break;
191 		case META_SIT:
192 			/* get sit block addr */
193 			fio.new_blkaddr = current_sit_addr(sbi,
194 					blkno * SIT_ENTRY_PER_BLOCK);
195 			break;
196 		case META_SSA:
197 		case META_CP:
198 		case META_POR:
199 			fio.new_blkaddr = blkno;
200 			break;
201 		default:
202 			BUG();
203 		}
204 
205 		page = f2fs_grab_cache_page(META_MAPPING(sbi),
206 						fio.new_blkaddr, false);
207 		if (!page)
208 			continue;
209 		if (PageUptodate(page)) {
210 			f2fs_put_page(page, 1);
211 			continue;
212 		}
213 
214 		fio.page = page;
215 		f2fs_submit_page_bio(&fio);
216 		f2fs_put_page(page, 0);
217 	}
218 out:
219 	blk_finish_plug(&plug);
220 	return blkno - start;
221 }
222 
223 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
224 {
225 	struct page *page;
226 	bool readahead = false;
227 
228 	page = find_get_page(META_MAPPING(sbi), index);
229 	if (!page || !PageUptodate(page))
230 		readahead = true;
231 	f2fs_put_page(page, 0);
232 
233 	if (readahead)
234 		f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
235 }
236 
237 static int __f2fs_write_meta_page(struct page *page,
238 				struct writeback_control *wbc,
239 				enum iostat_type io_type)
240 {
241 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
242 
243 	trace_f2fs_writepage(page, META);
244 
245 	if (unlikely(f2fs_cp_error(sbi))) {
246 		dec_page_count(sbi, F2FS_DIRTY_META);
247 		unlock_page(page);
248 		return 0;
249 	}
250 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
251 		goto redirty_out;
252 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
253 		goto redirty_out;
254 
255 	f2fs_do_write_meta_page(sbi, page, io_type);
256 	dec_page_count(sbi, F2FS_DIRTY_META);
257 
258 	if (wbc->for_reclaim)
259 		f2fs_submit_merged_write_cond(sbi, page->mapping->host,
260 						0, page->index, META);
261 
262 	unlock_page(page);
263 
264 	if (unlikely(f2fs_cp_error(sbi)))
265 		f2fs_submit_merged_write(sbi, META);
266 
267 	return 0;
268 
269 redirty_out:
270 	redirty_page_for_writepage(wbc, page);
271 	return AOP_WRITEPAGE_ACTIVATE;
272 }
273 
274 static int f2fs_write_meta_page(struct page *page,
275 				struct writeback_control *wbc)
276 {
277 	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
278 }
279 
280 static int f2fs_write_meta_pages(struct address_space *mapping,
281 				struct writeback_control *wbc)
282 {
283 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
284 	long diff, written;
285 
286 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
287 		goto skip_write;
288 
289 	/* collect a number of dirty meta pages and write together */
290 	if (wbc->for_kupdate ||
291 		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
292 		goto skip_write;
293 
294 	/* if locked failed, cp will flush dirty pages instead */
295 	if (!mutex_trylock(&sbi->cp_mutex))
296 		goto skip_write;
297 
298 	trace_f2fs_writepages(mapping->host, wbc, META);
299 	diff = nr_pages_to_write(sbi, META, wbc);
300 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
301 	mutex_unlock(&sbi->cp_mutex);
302 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
303 	return 0;
304 
305 skip_write:
306 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
307 	trace_f2fs_writepages(mapping->host, wbc, META);
308 	return 0;
309 }
310 
311 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
312 				long nr_to_write, enum iostat_type io_type)
313 {
314 	struct address_space *mapping = META_MAPPING(sbi);
315 	pgoff_t index = 0, prev = ULONG_MAX;
316 	struct pagevec pvec;
317 	long nwritten = 0;
318 	int nr_pages;
319 	struct writeback_control wbc = {
320 		.for_reclaim = 0,
321 	};
322 	struct blk_plug plug;
323 
324 	pagevec_init(&pvec);
325 
326 	blk_start_plug(&plug);
327 
328 	while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
329 				PAGECACHE_TAG_DIRTY))) {
330 		int i;
331 
332 		for (i = 0; i < nr_pages; i++) {
333 			struct page *page = pvec.pages[i];
334 
335 			if (prev == ULONG_MAX)
336 				prev = page->index - 1;
337 			if (nr_to_write != LONG_MAX && page->index != prev + 1) {
338 				pagevec_release(&pvec);
339 				goto stop;
340 			}
341 
342 			lock_page(page);
343 
344 			if (unlikely(page->mapping != mapping)) {
345 continue_unlock:
346 				unlock_page(page);
347 				continue;
348 			}
349 			if (!PageDirty(page)) {
350 				/* someone wrote it for us */
351 				goto continue_unlock;
352 			}
353 
354 			f2fs_wait_on_page_writeback(page, META, true);
355 
356 			BUG_ON(PageWriteback(page));
357 			if (!clear_page_dirty_for_io(page))
358 				goto continue_unlock;
359 
360 			if (__f2fs_write_meta_page(page, &wbc, io_type)) {
361 				unlock_page(page);
362 				break;
363 			}
364 			nwritten++;
365 			prev = page->index;
366 			if (unlikely(nwritten >= nr_to_write))
367 				break;
368 		}
369 		pagevec_release(&pvec);
370 		cond_resched();
371 	}
372 stop:
373 	if (nwritten)
374 		f2fs_submit_merged_write(sbi, type);
375 
376 	blk_finish_plug(&plug);
377 
378 	return nwritten;
379 }
380 
381 static int f2fs_set_meta_page_dirty(struct page *page)
382 {
383 	trace_f2fs_set_page_dirty(page, META);
384 
385 	if (!PageUptodate(page))
386 		SetPageUptodate(page);
387 	if (!PageDirty(page)) {
388 		__set_page_dirty_nobuffers(page);
389 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
390 		SetPagePrivate(page);
391 		f2fs_trace_pid(page);
392 		return 1;
393 	}
394 	return 0;
395 }
396 
397 const struct address_space_operations f2fs_meta_aops = {
398 	.writepage	= f2fs_write_meta_page,
399 	.writepages	= f2fs_write_meta_pages,
400 	.set_page_dirty	= f2fs_set_meta_page_dirty,
401 	.invalidatepage = f2fs_invalidate_page,
402 	.releasepage	= f2fs_release_page,
403 #ifdef CONFIG_MIGRATION
404 	.migratepage    = f2fs_migrate_page,
405 #endif
406 };
407 
408 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
409 						unsigned int devidx, int type)
410 {
411 	struct inode_management *im = &sbi->im[type];
412 	struct ino_entry *e, *tmp;
413 
414 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
415 
416 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
417 
418 	spin_lock(&im->ino_lock);
419 	e = radix_tree_lookup(&im->ino_root, ino);
420 	if (!e) {
421 		e = tmp;
422 		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
423 			f2fs_bug_on(sbi, 1);
424 
425 		memset(e, 0, sizeof(struct ino_entry));
426 		e->ino = ino;
427 
428 		list_add_tail(&e->list, &im->ino_list);
429 		if (type != ORPHAN_INO)
430 			im->ino_num++;
431 	}
432 
433 	if (type == FLUSH_INO)
434 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
435 
436 	spin_unlock(&im->ino_lock);
437 	radix_tree_preload_end();
438 
439 	if (e != tmp)
440 		kmem_cache_free(ino_entry_slab, tmp);
441 }
442 
443 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
444 {
445 	struct inode_management *im = &sbi->im[type];
446 	struct ino_entry *e;
447 
448 	spin_lock(&im->ino_lock);
449 	e = radix_tree_lookup(&im->ino_root, ino);
450 	if (e) {
451 		list_del(&e->list);
452 		radix_tree_delete(&im->ino_root, ino);
453 		im->ino_num--;
454 		spin_unlock(&im->ino_lock);
455 		kmem_cache_free(ino_entry_slab, e);
456 		return;
457 	}
458 	spin_unlock(&im->ino_lock);
459 }
460 
461 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
462 {
463 	/* add new dirty ino entry into list */
464 	__add_ino_entry(sbi, ino, 0, type);
465 }
466 
467 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
468 {
469 	/* remove dirty ino entry from list */
470 	__remove_ino_entry(sbi, ino, type);
471 }
472 
473 /* mode should be APPEND_INO or UPDATE_INO */
474 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
475 {
476 	struct inode_management *im = &sbi->im[mode];
477 	struct ino_entry *e;
478 
479 	spin_lock(&im->ino_lock);
480 	e = radix_tree_lookup(&im->ino_root, ino);
481 	spin_unlock(&im->ino_lock);
482 	return e ? true : false;
483 }
484 
485 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
486 {
487 	struct ino_entry *e, *tmp;
488 	int i;
489 
490 	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
491 		struct inode_management *im = &sbi->im[i];
492 
493 		spin_lock(&im->ino_lock);
494 		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
495 			list_del(&e->list);
496 			radix_tree_delete(&im->ino_root, e->ino);
497 			kmem_cache_free(ino_entry_slab, e);
498 			im->ino_num--;
499 		}
500 		spin_unlock(&im->ino_lock);
501 	}
502 }
503 
504 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
505 					unsigned int devidx, int type)
506 {
507 	__add_ino_entry(sbi, ino, devidx, type);
508 }
509 
510 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
511 					unsigned int devidx, int type)
512 {
513 	struct inode_management *im = &sbi->im[type];
514 	struct ino_entry *e;
515 	bool is_dirty = false;
516 
517 	spin_lock(&im->ino_lock);
518 	e = radix_tree_lookup(&im->ino_root, ino);
519 	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
520 		is_dirty = true;
521 	spin_unlock(&im->ino_lock);
522 	return is_dirty;
523 }
524 
525 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
526 {
527 	struct inode_management *im = &sbi->im[ORPHAN_INO];
528 	int err = 0;
529 
530 	spin_lock(&im->ino_lock);
531 
532 #ifdef CONFIG_F2FS_FAULT_INJECTION
533 	if (time_to_inject(sbi, FAULT_ORPHAN)) {
534 		spin_unlock(&im->ino_lock);
535 		f2fs_show_injection_info(FAULT_ORPHAN);
536 		return -ENOSPC;
537 	}
538 #endif
539 	if (unlikely(im->ino_num >= sbi->max_orphans))
540 		err = -ENOSPC;
541 	else
542 		im->ino_num++;
543 	spin_unlock(&im->ino_lock);
544 
545 	return err;
546 }
547 
548 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
549 {
550 	struct inode_management *im = &sbi->im[ORPHAN_INO];
551 
552 	spin_lock(&im->ino_lock);
553 	f2fs_bug_on(sbi, im->ino_num == 0);
554 	im->ino_num--;
555 	spin_unlock(&im->ino_lock);
556 }
557 
558 void f2fs_add_orphan_inode(struct inode *inode)
559 {
560 	/* add new orphan ino entry into list */
561 	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
562 	f2fs_update_inode_page(inode);
563 }
564 
565 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
566 {
567 	/* remove orphan entry from orphan list */
568 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
569 }
570 
571 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
572 {
573 	struct inode *inode;
574 	struct node_info ni;
575 	int err = f2fs_acquire_orphan_inode(sbi);
576 
577 	if (err)
578 		goto err_out;
579 
580 	__add_ino_entry(sbi, ino, 0, ORPHAN_INO);
581 
582 	inode = f2fs_iget_retry(sbi->sb, ino);
583 	if (IS_ERR(inode)) {
584 		/*
585 		 * there should be a bug that we can't find the entry
586 		 * to orphan inode.
587 		 */
588 		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
589 		return PTR_ERR(inode);
590 	}
591 
592 	err = dquot_initialize(inode);
593 	if (err) {
594 		iput(inode);
595 		goto err_out;
596 	}
597 
598 	clear_nlink(inode);
599 
600 	/* truncate all the data during iput */
601 	iput(inode);
602 
603 	f2fs_get_node_info(sbi, ino, &ni);
604 
605 	/* ENOMEM was fully retried in f2fs_evict_inode. */
606 	if (ni.blk_addr != NULL_ADDR) {
607 		err = -EIO;
608 		goto err_out;
609 	}
610 	__remove_ino_entry(sbi, ino, ORPHAN_INO);
611 	return 0;
612 
613 err_out:
614 	set_sbi_flag(sbi, SBI_NEED_FSCK);
615 	f2fs_msg(sbi->sb, KERN_WARNING,
616 			"%s: orphan failed (ino=%x), run fsck to fix.",
617 			__func__, ino);
618 	return err;
619 }
620 
621 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
622 {
623 	block_t start_blk, orphan_blocks, i, j;
624 	unsigned int s_flags = sbi->sb->s_flags;
625 	int err = 0;
626 #ifdef CONFIG_QUOTA
627 	int quota_enabled;
628 #endif
629 
630 	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
631 		return 0;
632 
633 	if (s_flags & SB_RDONLY) {
634 		f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
635 		sbi->sb->s_flags &= ~SB_RDONLY;
636 	}
637 
638 #ifdef CONFIG_QUOTA
639 	/* Needed for iput() to work correctly and not trash data */
640 	sbi->sb->s_flags |= SB_ACTIVE;
641 
642 	/* Turn on quotas so that they are updated correctly */
643 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
644 #endif
645 
646 	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
647 	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
648 
649 	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
650 
651 	for (i = 0; i < orphan_blocks; i++) {
652 		struct page *page = f2fs_get_meta_page(sbi, start_blk + i);
653 		struct f2fs_orphan_block *orphan_blk;
654 
655 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
656 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
657 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
658 			err = recover_orphan_inode(sbi, ino);
659 			if (err) {
660 				f2fs_put_page(page, 1);
661 				goto out;
662 			}
663 		}
664 		f2fs_put_page(page, 1);
665 	}
666 	/* clear Orphan Flag */
667 	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
668 out:
669 #ifdef CONFIG_QUOTA
670 	/* Turn quotas off */
671 	if (quota_enabled)
672 		f2fs_quota_off_umount(sbi->sb);
673 #endif
674 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
675 
676 	return err;
677 }
678 
679 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
680 {
681 	struct list_head *head;
682 	struct f2fs_orphan_block *orphan_blk = NULL;
683 	unsigned int nentries = 0;
684 	unsigned short index = 1;
685 	unsigned short orphan_blocks;
686 	struct page *page = NULL;
687 	struct ino_entry *orphan = NULL;
688 	struct inode_management *im = &sbi->im[ORPHAN_INO];
689 
690 	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
691 
692 	/*
693 	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
694 	 * orphan inode operations are covered under f2fs_lock_op().
695 	 * And, spin_lock should be avoided due to page operations below.
696 	 */
697 	head = &im->ino_list;
698 
699 	/* loop for each orphan inode entry and write them in Jornal block */
700 	list_for_each_entry(orphan, head, list) {
701 		if (!page) {
702 			page = f2fs_grab_meta_page(sbi, start_blk++);
703 			orphan_blk =
704 				(struct f2fs_orphan_block *)page_address(page);
705 			memset(orphan_blk, 0, sizeof(*orphan_blk));
706 		}
707 
708 		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
709 
710 		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
711 			/*
712 			 * an orphan block is full of 1020 entries,
713 			 * then we need to flush current orphan blocks
714 			 * and bring another one in memory
715 			 */
716 			orphan_blk->blk_addr = cpu_to_le16(index);
717 			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
718 			orphan_blk->entry_count = cpu_to_le32(nentries);
719 			set_page_dirty(page);
720 			f2fs_put_page(page, 1);
721 			index++;
722 			nentries = 0;
723 			page = NULL;
724 		}
725 	}
726 
727 	if (page) {
728 		orphan_blk->blk_addr = cpu_to_le16(index);
729 		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
730 		orphan_blk->entry_count = cpu_to_le32(nentries);
731 		set_page_dirty(page);
732 		f2fs_put_page(page, 1);
733 	}
734 }
735 
736 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
737 		struct f2fs_checkpoint **cp_block, struct page **cp_page,
738 		unsigned long long *version)
739 {
740 	unsigned long blk_size = sbi->blocksize;
741 	size_t crc_offset = 0;
742 	__u32 crc = 0;
743 
744 	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
745 	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
746 
747 	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
748 	if (crc_offset > (blk_size - sizeof(__le32))) {
749 		f2fs_msg(sbi->sb, KERN_WARNING,
750 			"invalid crc_offset: %zu", crc_offset);
751 		return -EINVAL;
752 	}
753 
754 	crc = cur_cp_crc(*cp_block);
755 	if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
756 		f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
757 		return -EINVAL;
758 	}
759 
760 	*version = cur_cp_version(*cp_block);
761 	return 0;
762 }
763 
764 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
765 				block_t cp_addr, unsigned long long *version)
766 {
767 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
768 	struct f2fs_checkpoint *cp_block = NULL;
769 	unsigned long long cur_version = 0, pre_version = 0;
770 	int err;
771 
772 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
773 					&cp_page_1, version);
774 	if (err)
775 		goto invalid_cp1;
776 	pre_version = *version;
777 
778 	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
779 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
780 					&cp_page_2, version);
781 	if (err)
782 		goto invalid_cp2;
783 	cur_version = *version;
784 
785 	if (cur_version == pre_version) {
786 		*version = cur_version;
787 		f2fs_put_page(cp_page_2, 1);
788 		return cp_page_1;
789 	}
790 invalid_cp2:
791 	f2fs_put_page(cp_page_2, 1);
792 invalid_cp1:
793 	f2fs_put_page(cp_page_1, 1);
794 	return NULL;
795 }
796 
797 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
798 {
799 	struct f2fs_checkpoint *cp_block;
800 	struct f2fs_super_block *fsb = sbi->raw_super;
801 	struct page *cp1, *cp2, *cur_page;
802 	unsigned long blk_size = sbi->blocksize;
803 	unsigned long long cp1_version = 0, cp2_version = 0;
804 	unsigned long long cp_start_blk_no;
805 	unsigned int cp_blks = 1 + __cp_payload(sbi);
806 	block_t cp_blk_no;
807 	int i;
808 
809 	sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
810 				 GFP_KERNEL);
811 	if (!sbi->ckpt)
812 		return -ENOMEM;
813 	/*
814 	 * Finding out valid cp block involves read both
815 	 * sets( cp pack1 and cp pack 2)
816 	 */
817 	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
818 	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
819 
820 	/* The second checkpoint pack should start at the next segment */
821 	cp_start_blk_no += ((unsigned long long)1) <<
822 				le32_to_cpu(fsb->log_blocks_per_seg);
823 	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
824 
825 	if (cp1 && cp2) {
826 		if (ver_after(cp2_version, cp1_version))
827 			cur_page = cp2;
828 		else
829 			cur_page = cp1;
830 	} else if (cp1) {
831 		cur_page = cp1;
832 	} else if (cp2) {
833 		cur_page = cp2;
834 	} else {
835 		goto fail_no_cp;
836 	}
837 
838 	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
839 	memcpy(sbi->ckpt, cp_block, blk_size);
840 
841 	/* Sanity checking of checkpoint */
842 	if (f2fs_sanity_check_ckpt(sbi))
843 		goto free_fail_no_cp;
844 
845 	if (cur_page == cp1)
846 		sbi->cur_cp_pack = 1;
847 	else
848 		sbi->cur_cp_pack = 2;
849 
850 	if (cp_blks <= 1)
851 		goto done;
852 
853 	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
854 	if (cur_page == cp2)
855 		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
856 
857 	for (i = 1; i < cp_blks; i++) {
858 		void *sit_bitmap_ptr;
859 		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
860 
861 		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
862 		sit_bitmap_ptr = page_address(cur_page);
863 		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
864 		f2fs_put_page(cur_page, 1);
865 	}
866 done:
867 	f2fs_put_page(cp1, 1);
868 	f2fs_put_page(cp2, 1);
869 	return 0;
870 
871 free_fail_no_cp:
872 	f2fs_put_page(cp1, 1);
873 	f2fs_put_page(cp2, 1);
874 fail_no_cp:
875 	kfree(sbi->ckpt);
876 	return -EINVAL;
877 }
878 
879 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
880 {
881 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
882 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
883 
884 	if (is_inode_flag_set(inode, flag))
885 		return;
886 
887 	set_inode_flag(inode, flag);
888 	if (!f2fs_is_volatile_file(inode))
889 		list_add_tail(&F2FS_I(inode)->dirty_list,
890 						&sbi->inode_list[type]);
891 	stat_inc_dirty_inode(sbi, type);
892 }
893 
894 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
895 {
896 	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
897 
898 	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
899 		return;
900 
901 	list_del_init(&F2FS_I(inode)->dirty_list);
902 	clear_inode_flag(inode, flag);
903 	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
904 }
905 
906 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
907 {
908 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
909 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
910 
911 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
912 			!S_ISLNK(inode->i_mode))
913 		return;
914 
915 	spin_lock(&sbi->inode_lock[type]);
916 	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
917 		__add_dirty_inode(inode, type);
918 	inode_inc_dirty_pages(inode);
919 	spin_unlock(&sbi->inode_lock[type]);
920 
921 	SetPagePrivate(page);
922 	f2fs_trace_pid(page);
923 }
924 
925 void f2fs_remove_dirty_inode(struct inode *inode)
926 {
927 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
928 	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
929 
930 	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
931 			!S_ISLNK(inode->i_mode))
932 		return;
933 
934 	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
935 		return;
936 
937 	spin_lock(&sbi->inode_lock[type]);
938 	__remove_dirty_inode(inode, type);
939 	spin_unlock(&sbi->inode_lock[type]);
940 }
941 
942 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
943 {
944 	struct list_head *head;
945 	struct inode *inode;
946 	struct f2fs_inode_info *fi;
947 	bool is_dir = (type == DIR_INODE);
948 	unsigned long ino = 0;
949 
950 	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
951 				get_pages(sbi, is_dir ?
952 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
953 retry:
954 	if (unlikely(f2fs_cp_error(sbi)))
955 		return -EIO;
956 
957 	spin_lock(&sbi->inode_lock[type]);
958 
959 	head = &sbi->inode_list[type];
960 	if (list_empty(head)) {
961 		spin_unlock(&sbi->inode_lock[type]);
962 		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
963 				get_pages(sbi, is_dir ?
964 				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
965 		return 0;
966 	}
967 	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
968 	inode = igrab(&fi->vfs_inode);
969 	spin_unlock(&sbi->inode_lock[type]);
970 	if (inode) {
971 		unsigned long cur_ino = inode->i_ino;
972 
973 		if (is_dir)
974 			F2FS_I(inode)->cp_task = current;
975 
976 		filemap_fdatawrite(inode->i_mapping);
977 
978 		if (is_dir)
979 			F2FS_I(inode)->cp_task = NULL;
980 
981 		iput(inode);
982 		/* We need to give cpu to another writers. */
983 		if (ino == cur_ino) {
984 			congestion_wait(BLK_RW_ASYNC, HZ/50);
985 			cond_resched();
986 		} else {
987 			ino = cur_ino;
988 		}
989 	} else {
990 		/*
991 		 * We should submit bio, since it exists several
992 		 * wribacking dentry pages in the freeing inode.
993 		 */
994 		f2fs_submit_merged_write(sbi, DATA);
995 		cond_resched();
996 	}
997 	goto retry;
998 }
999 
1000 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1001 {
1002 	struct list_head *head = &sbi->inode_list[DIRTY_META];
1003 	struct inode *inode;
1004 	struct f2fs_inode_info *fi;
1005 	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1006 
1007 	while (total--) {
1008 		if (unlikely(f2fs_cp_error(sbi)))
1009 			return -EIO;
1010 
1011 		spin_lock(&sbi->inode_lock[DIRTY_META]);
1012 		if (list_empty(head)) {
1013 			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1014 			return 0;
1015 		}
1016 		fi = list_first_entry(head, struct f2fs_inode_info,
1017 							gdirty_list);
1018 		inode = igrab(&fi->vfs_inode);
1019 		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1020 		if (inode) {
1021 			sync_inode_metadata(inode, 0);
1022 
1023 			/* it's on eviction */
1024 			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1025 				f2fs_update_inode_page(inode);
1026 			iput(inode);
1027 		}
1028 	}
1029 	return 0;
1030 }
1031 
1032 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1033 {
1034 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1035 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1036 	nid_t last_nid = nm_i->next_scan_nid;
1037 
1038 	next_free_nid(sbi, &last_nid);
1039 	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1040 	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1041 	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1042 	ckpt->next_free_nid = cpu_to_le32(last_nid);
1043 }
1044 
1045 /*
1046  * Freeze all the FS-operations for checkpoint.
1047  */
1048 static int block_operations(struct f2fs_sb_info *sbi)
1049 {
1050 	struct writeback_control wbc = {
1051 		.sync_mode = WB_SYNC_ALL,
1052 		.nr_to_write = LONG_MAX,
1053 		.for_reclaim = 0,
1054 	};
1055 	struct blk_plug plug;
1056 	int err = 0;
1057 
1058 	blk_start_plug(&plug);
1059 
1060 retry_flush_dents:
1061 	f2fs_lock_all(sbi);
1062 	/* write all the dirty dentry pages */
1063 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1064 		f2fs_unlock_all(sbi);
1065 		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1066 		if (err)
1067 			goto out;
1068 		cond_resched();
1069 		goto retry_flush_dents;
1070 	}
1071 
1072 	/*
1073 	 * POR: we should ensure that there are no dirty node pages
1074 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1075 	 */
1076 	down_write(&sbi->node_change);
1077 
1078 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1079 		up_write(&sbi->node_change);
1080 		f2fs_unlock_all(sbi);
1081 		err = f2fs_sync_inode_meta(sbi);
1082 		if (err)
1083 			goto out;
1084 		cond_resched();
1085 		goto retry_flush_dents;
1086 	}
1087 
1088 retry_flush_nodes:
1089 	down_write(&sbi->node_write);
1090 
1091 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1092 		up_write(&sbi->node_write);
1093 		atomic_inc(&sbi->wb_sync_req[NODE]);
1094 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1095 		atomic_dec(&sbi->wb_sync_req[NODE]);
1096 		if (err) {
1097 			up_write(&sbi->node_change);
1098 			f2fs_unlock_all(sbi);
1099 			goto out;
1100 		}
1101 		cond_resched();
1102 		goto retry_flush_nodes;
1103 	}
1104 
1105 	/*
1106 	 * sbi->node_change is used only for AIO write_begin path which produces
1107 	 * dirty node blocks and some checkpoint values by block allocation.
1108 	 */
1109 	__prepare_cp_block(sbi);
1110 	up_write(&sbi->node_change);
1111 out:
1112 	blk_finish_plug(&plug);
1113 	return err;
1114 }
1115 
1116 static void unblock_operations(struct f2fs_sb_info *sbi)
1117 {
1118 	up_write(&sbi->node_write);
1119 	f2fs_unlock_all(sbi);
1120 }
1121 
1122 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1123 {
1124 	DEFINE_WAIT(wait);
1125 
1126 	for (;;) {
1127 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1128 
1129 		if (!get_pages(sbi, F2FS_WB_CP_DATA))
1130 			break;
1131 
1132 		io_schedule_timeout(5*HZ);
1133 	}
1134 	finish_wait(&sbi->cp_wait, &wait);
1135 }
1136 
1137 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1138 {
1139 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1140 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1141 	unsigned long flags;
1142 
1143 	spin_lock_irqsave(&sbi->cp_lock, flags);
1144 
1145 	if ((cpc->reason & CP_UMOUNT) &&
1146 			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1147 			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1148 		disable_nat_bits(sbi, false);
1149 
1150 	if (cpc->reason & CP_TRIMMED)
1151 		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1152 	else
1153 		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1154 
1155 	if (cpc->reason & CP_UMOUNT)
1156 		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1157 	else
1158 		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1159 
1160 	if (cpc->reason & CP_FASTBOOT)
1161 		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1162 	else
1163 		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1164 
1165 	if (orphan_num)
1166 		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1167 	else
1168 		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1169 
1170 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1171 		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1172 
1173 	/* set this flag to activate crc|cp_ver for recovery */
1174 	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1175 	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1176 
1177 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1178 }
1179 
1180 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1181 	void *src, block_t blk_addr)
1182 {
1183 	struct writeback_control wbc = {
1184 		.for_reclaim = 0,
1185 	};
1186 
1187 	/*
1188 	 * pagevec_lookup_tag and lock_page again will take
1189 	 * some extra time. Therefore, f2fs_update_meta_pages and
1190 	 * f2fs_sync_meta_pages are combined in this function.
1191 	 */
1192 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1193 	int err;
1194 
1195 	memcpy(page_address(page), src, PAGE_SIZE);
1196 	set_page_dirty(page);
1197 
1198 	f2fs_wait_on_page_writeback(page, META, true);
1199 	f2fs_bug_on(sbi, PageWriteback(page));
1200 	if (unlikely(!clear_page_dirty_for_io(page)))
1201 		f2fs_bug_on(sbi, 1);
1202 
1203 	/* writeout cp pack 2 page */
1204 	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1205 	f2fs_bug_on(sbi, err);
1206 
1207 	f2fs_put_page(page, 0);
1208 
1209 	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1210 	f2fs_submit_merged_write(sbi, META_FLUSH);
1211 }
1212 
1213 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1214 {
1215 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1216 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1217 	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1218 	block_t start_blk;
1219 	unsigned int data_sum_blocks, orphan_blocks;
1220 	__u32 crc32 = 0;
1221 	int i;
1222 	int cp_payload_blks = __cp_payload(sbi);
1223 	struct super_block *sb = sbi->sb;
1224 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1225 	u64 kbytes_written;
1226 	int err;
1227 
1228 	/* Flush all the NAT/SIT pages */
1229 	while (get_pages(sbi, F2FS_DIRTY_META)) {
1230 		f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1231 		if (unlikely(f2fs_cp_error(sbi)))
1232 			return -EIO;
1233 	}
1234 
1235 	/*
1236 	 * modify checkpoint
1237 	 * version number is already updated
1238 	 */
1239 	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1240 	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1241 	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1242 		ckpt->cur_node_segno[i] =
1243 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1244 		ckpt->cur_node_blkoff[i] =
1245 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1246 		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1247 				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1248 	}
1249 	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1250 		ckpt->cur_data_segno[i] =
1251 			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1252 		ckpt->cur_data_blkoff[i] =
1253 			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1254 		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1255 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1256 	}
1257 
1258 	/* 2 cp  + n data seg summary + orphan inode blocks */
1259 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1260 	spin_lock_irqsave(&sbi->cp_lock, flags);
1261 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1262 		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1263 	else
1264 		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1265 	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1266 
1267 	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1268 	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1269 			orphan_blocks);
1270 
1271 	if (__remain_node_summaries(cpc->reason))
1272 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1273 				cp_payload_blks + data_sum_blocks +
1274 				orphan_blocks + NR_CURSEG_NODE_TYPE);
1275 	else
1276 		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1277 				cp_payload_blks + data_sum_blocks +
1278 				orphan_blocks);
1279 
1280 	/* update ckpt flag for checkpoint */
1281 	update_ckpt_flags(sbi, cpc);
1282 
1283 	/* update SIT/NAT bitmap */
1284 	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1285 	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1286 
1287 	crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1288 	*((__le32 *)((unsigned char *)ckpt +
1289 				le32_to_cpu(ckpt->checksum_offset)))
1290 				= cpu_to_le32(crc32);
1291 
1292 	start_blk = __start_cp_next_addr(sbi);
1293 
1294 	/* write nat bits */
1295 	if (enabled_nat_bits(sbi, cpc)) {
1296 		__u64 cp_ver = cur_cp_version(ckpt);
1297 		block_t blk;
1298 
1299 		cp_ver |= ((__u64)crc32 << 32);
1300 		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1301 
1302 		blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1303 		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1304 			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1305 					(i << F2FS_BLKSIZE_BITS), blk + i);
1306 
1307 		/* Flush all the NAT BITS pages */
1308 		while (get_pages(sbi, F2FS_DIRTY_META)) {
1309 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1310 							FS_CP_META_IO);
1311 			if (unlikely(f2fs_cp_error(sbi)))
1312 				return -EIO;
1313 		}
1314 	}
1315 
1316 	/* write out checkpoint buffer at block 0 */
1317 	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1318 
1319 	for (i = 1; i < 1 + cp_payload_blks; i++)
1320 		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1321 							start_blk++);
1322 
1323 	if (orphan_num) {
1324 		write_orphan_inodes(sbi, start_blk);
1325 		start_blk += orphan_blocks;
1326 	}
1327 
1328 	f2fs_write_data_summaries(sbi, start_blk);
1329 	start_blk += data_sum_blocks;
1330 
1331 	/* Record write statistics in the hot node summary */
1332 	kbytes_written = sbi->kbytes_written;
1333 	if (sb->s_bdev->bd_part)
1334 		kbytes_written += BD_PART_WRITTEN(sbi);
1335 
1336 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1337 
1338 	if (__remain_node_summaries(cpc->reason)) {
1339 		f2fs_write_node_summaries(sbi, start_blk);
1340 		start_blk += NR_CURSEG_NODE_TYPE;
1341 	}
1342 
1343 	/* update user_block_counts */
1344 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1345 	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1346 
1347 	/* Here, we have one bio having CP pack except cp pack 2 page */
1348 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1349 
1350 	/* wait for previous submitted meta pages writeback */
1351 	wait_on_all_pages_writeback(sbi);
1352 
1353 	if (unlikely(f2fs_cp_error(sbi)))
1354 		return -EIO;
1355 
1356 	/* flush all device cache */
1357 	err = f2fs_flush_device_cache(sbi);
1358 	if (err)
1359 		return err;
1360 
1361 	/* barrier and flush checkpoint cp pack 2 page if it can */
1362 	commit_checkpoint(sbi, ckpt, start_blk);
1363 	wait_on_all_pages_writeback(sbi);
1364 
1365 	f2fs_release_ino_entry(sbi, false);
1366 
1367 	if (unlikely(f2fs_cp_error(sbi)))
1368 		return -EIO;
1369 
1370 	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1371 	clear_sbi_flag(sbi, SBI_NEED_CP);
1372 	__set_cp_next_pack(sbi);
1373 
1374 	/*
1375 	 * redirty superblock if metadata like node page or inode cache is
1376 	 * updated during writing checkpoint.
1377 	 */
1378 	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1379 			get_pages(sbi, F2FS_DIRTY_IMETA))
1380 		set_sbi_flag(sbi, SBI_IS_DIRTY);
1381 
1382 	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1383 
1384 	return 0;
1385 }
1386 
1387 /*
1388  * We guarantee that this checkpoint procedure will not fail.
1389  */
1390 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1391 {
1392 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1393 	unsigned long long ckpt_ver;
1394 	int err = 0;
1395 
1396 	mutex_lock(&sbi->cp_mutex);
1397 
1398 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1399 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1400 		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1401 		goto out;
1402 	if (unlikely(f2fs_cp_error(sbi))) {
1403 		err = -EIO;
1404 		goto out;
1405 	}
1406 	if (f2fs_readonly(sbi->sb)) {
1407 		err = -EROFS;
1408 		goto out;
1409 	}
1410 
1411 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1412 
1413 	err = block_operations(sbi);
1414 	if (err)
1415 		goto out;
1416 
1417 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1418 
1419 	f2fs_flush_merged_writes(sbi);
1420 
1421 	/* this is the case of multiple fstrims without any changes */
1422 	if (cpc->reason & CP_DISCARD) {
1423 		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1424 			unblock_operations(sbi);
1425 			goto out;
1426 		}
1427 
1428 		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1429 				SIT_I(sbi)->dirty_sentries == 0 &&
1430 				prefree_segments(sbi) == 0) {
1431 			f2fs_flush_sit_entries(sbi, cpc);
1432 			f2fs_clear_prefree_segments(sbi, cpc);
1433 			unblock_operations(sbi);
1434 			goto out;
1435 		}
1436 	}
1437 
1438 	/*
1439 	 * update checkpoint pack index
1440 	 * Increase the version number so that
1441 	 * SIT entries and seg summaries are written at correct place
1442 	 */
1443 	ckpt_ver = cur_cp_version(ckpt);
1444 	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1445 
1446 	/* write cached NAT/SIT entries to NAT/SIT area */
1447 	f2fs_flush_nat_entries(sbi, cpc);
1448 	f2fs_flush_sit_entries(sbi, cpc);
1449 
1450 	/* unlock all the fs_lock[] in do_checkpoint() */
1451 	err = do_checkpoint(sbi, cpc);
1452 	if (err)
1453 		f2fs_release_discard_addrs(sbi);
1454 	else
1455 		f2fs_clear_prefree_segments(sbi, cpc);
1456 
1457 	unblock_operations(sbi);
1458 	stat_inc_cp_count(sbi->stat_info);
1459 
1460 	if (cpc->reason & CP_RECOVERY)
1461 		f2fs_msg(sbi->sb, KERN_NOTICE,
1462 			"checkpoint: version = %llx", ckpt_ver);
1463 
1464 	/* do checkpoint periodically */
1465 	f2fs_update_time(sbi, CP_TIME);
1466 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1467 out:
1468 	mutex_unlock(&sbi->cp_mutex);
1469 	return err;
1470 }
1471 
1472 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1473 {
1474 	int i;
1475 
1476 	for (i = 0; i < MAX_INO_ENTRY; i++) {
1477 		struct inode_management *im = &sbi->im[i];
1478 
1479 		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1480 		spin_lock_init(&im->ino_lock);
1481 		INIT_LIST_HEAD(&im->ino_list);
1482 		im->ino_num = 0;
1483 	}
1484 
1485 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1486 			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1487 				F2FS_ORPHANS_PER_BLOCK;
1488 }
1489 
1490 int __init f2fs_create_checkpoint_caches(void)
1491 {
1492 	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1493 			sizeof(struct ino_entry));
1494 	if (!ino_entry_slab)
1495 		return -ENOMEM;
1496 	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1497 			sizeof(struct inode_entry));
1498 	if (!f2fs_inode_entry_slab) {
1499 		kmem_cache_destroy(ino_entry_slab);
1500 		return -ENOMEM;
1501 	}
1502 	return 0;
1503 }
1504 
1505 void f2fs_destroy_checkpoint_caches(void)
1506 {
1507 	kmem_cache_destroy(ino_entry_slab);
1508 	kmem_cache_destroy(f2fs_inode_entry_slab);
1509 }
1510