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
26static struct kmem_cache *ino_entry_slab;
27struct kmem_cache *inode_entry_slab;
28
29/*
30 * We guarantee no failure on the returned page.
31 */
32struct 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;
36repeat:
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 */
50struct 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		.type = META,
56		.rw = READ_SYNC | REQ_META | REQ_PRIO,
57		.blk_addr = index,
58	};
59repeat:
60	page = grab_cache_page(mapping, index);
61	if (!page) {
62		cond_resched();
63		goto repeat;
64	}
65	if (PageUptodate(page))
66		goto out;
67
68	if (f2fs_submit_page_bio(sbi, page, &fio))
69		goto repeat;
70
71	lock_page(page);
72	if (unlikely(page->mapping != mapping)) {
73		f2fs_put_page(page, 1);
74		goto repeat;
75	}
76out:
77	return page;
78}
79
80static inline bool is_valid_blkaddr(struct f2fs_sb_info *sbi,
81						block_t blkaddr, int type)
82{
83	switch (type) {
84	case META_NAT:
85		break;
86	case META_SIT:
87		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
88			return false;
89		break;
90	case META_SSA:
91		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
92			blkaddr < SM_I(sbi)->ssa_blkaddr))
93			return false;
94		break;
95	case META_CP:
96		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
97			blkaddr < __start_cp_addr(sbi)))
98			return false;
99		break;
100	case META_POR:
101		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
102			blkaddr < MAIN_BLKADDR(sbi)))
103			return false;
104		break;
105	default:
106		BUG();
107	}
108
109	return true;
110}
111
112/*
113 * Readahead CP/NAT/SIT/SSA pages
114 */
115int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
116{
117	block_t prev_blk_addr = 0;
118	struct page *page;
119	block_t blkno = start;
120	struct f2fs_io_info fio = {
121		.type = META,
122		.rw = READ_SYNC | REQ_META | REQ_PRIO
123	};
124
125	for (; nrpages-- > 0; blkno++) {
126
127		if (!is_valid_blkaddr(sbi, blkno, type))
128			goto out;
129
130		switch (type) {
131		case META_NAT:
132			if (unlikely(blkno >=
133					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
134				blkno = 0;
135			/* get nat block addr */
136			fio.blk_addr = current_nat_addr(sbi,
137					blkno * NAT_ENTRY_PER_BLOCK);
138			break;
139		case META_SIT:
140			/* get sit block addr */
141			fio.blk_addr = current_sit_addr(sbi,
142					blkno * SIT_ENTRY_PER_BLOCK);
143			if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
144				goto out;
145			prev_blk_addr = fio.blk_addr;
146			break;
147		case META_SSA:
148		case META_CP:
149		case META_POR:
150			fio.blk_addr = blkno;
151			break;
152		default:
153			BUG();
154		}
155
156		page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
157		if (!page)
158			continue;
159		if (PageUptodate(page)) {
160			f2fs_put_page(page, 1);
161			continue;
162		}
163
164		f2fs_submit_page_mbio(sbi, page, &fio);
165		f2fs_put_page(page, 0);
166	}
167out:
168	f2fs_submit_merged_bio(sbi, META, READ);
169	return blkno - start;
170}
171
172void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
173{
174	struct page *page;
175	bool readahead = false;
176
177	page = find_get_page(META_MAPPING(sbi), index);
178	if (!page || (page && !PageUptodate(page)))
179		readahead = true;
180	f2fs_put_page(page, 0);
181
182	if (readahead)
183		ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
184}
185
186static int f2fs_write_meta_page(struct page *page,
187				struct writeback_control *wbc)
188{
189	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
190
191	trace_f2fs_writepage(page, META);
192
193	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
194		goto redirty_out;
195	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
196		goto redirty_out;
197	if (unlikely(f2fs_cp_error(sbi)))
198		goto redirty_out;
199
200	f2fs_wait_on_page_writeback(page, META);
201	write_meta_page(sbi, page);
202	dec_page_count(sbi, F2FS_DIRTY_META);
203	unlock_page(page);
204
205	if (wbc->for_reclaim)
206		f2fs_submit_merged_bio(sbi, META, WRITE);
207	return 0;
208
209redirty_out:
210	redirty_page_for_writepage(wbc, page);
211	return AOP_WRITEPAGE_ACTIVATE;
212}
213
214static int f2fs_write_meta_pages(struct address_space *mapping,
215				struct writeback_control *wbc)
216{
217	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
218	long diff, written;
219
220	trace_f2fs_writepages(mapping->host, wbc, META);
221
222	/* collect a number of dirty meta pages and write together */
223	if (wbc->for_kupdate ||
224		get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
225		goto skip_write;
226
227	/* if mounting is failed, skip writing node pages */
228	mutex_lock(&sbi->cp_mutex);
229	diff = nr_pages_to_write(sbi, META, wbc);
230	written = sync_meta_pages(sbi, META, wbc->nr_to_write);
231	mutex_unlock(&sbi->cp_mutex);
232	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
233	return 0;
234
235skip_write:
236	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
237	return 0;
238}
239
240long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
241						long nr_to_write)
242{
243	struct address_space *mapping = META_MAPPING(sbi);
244	pgoff_t index = 0, end = LONG_MAX;
245	struct pagevec pvec;
246	long nwritten = 0;
247	struct writeback_control wbc = {
248		.for_reclaim = 0,
249	};
250
251	pagevec_init(&pvec, 0);
252
253	while (index <= end) {
254		int i, nr_pages;
255		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
256				PAGECACHE_TAG_DIRTY,
257				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
258		if (unlikely(nr_pages == 0))
259			break;
260
261		for (i = 0; i < nr_pages; i++) {
262			struct page *page = pvec.pages[i];
263
264			lock_page(page);
265
266			if (unlikely(page->mapping != mapping)) {
267continue_unlock:
268				unlock_page(page);
269				continue;
270			}
271			if (!PageDirty(page)) {
272				/* someone wrote it for us */
273				goto continue_unlock;
274			}
275
276			if (!clear_page_dirty_for_io(page))
277				goto continue_unlock;
278
279			if (mapping->a_ops->writepage(page, &wbc)) {
280				unlock_page(page);
281				break;
282			}
283			nwritten++;
284			if (unlikely(nwritten >= nr_to_write))
285				break;
286		}
287		pagevec_release(&pvec);
288		cond_resched();
289	}
290
291	if (nwritten)
292		f2fs_submit_merged_bio(sbi, type, WRITE);
293
294	return nwritten;
295}
296
297static int f2fs_set_meta_page_dirty(struct page *page)
298{
299	trace_f2fs_set_page_dirty(page, META);
300
301	SetPageUptodate(page);
302	if (!PageDirty(page)) {
303		__set_page_dirty_nobuffers(page);
304		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
305		SetPagePrivate(page);
306		f2fs_trace_pid(page);
307		return 1;
308	}
309	return 0;
310}
311
312const struct address_space_operations f2fs_meta_aops = {
313	.writepage	= f2fs_write_meta_page,
314	.writepages	= f2fs_write_meta_pages,
315	.set_page_dirty	= f2fs_set_meta_page_dirty,
316	.invalidatepage = f2fs_invalidate_page,
317	.releasepage	= f2fs_release_page,
318};
319
320static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
321{
322	struct inode_management *im = &sbi->im[type];
323	struct ino_entry *e;
324retry:
325	if (radix_tree_preload(GFP_NOFS)) {
326		cond_resched();
327		goto retry;
328	}
329
330	spin_lock(&im->ino_lock);
331
332	e = radix_tree_lookup(&im->ino_root, ino);
333	if (!e) {
334		e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
335		if (!e) {
336			spin_unlock(&im->ino_lock);
337			radix_tree_preload_end();
338			goto retry;
339		}
340		if (radix_tree_insert(&im->ino_root, ino, e)) {
341			spin_unlock(&im->ino_lock);
342			kmem_cache_free(ino_entry_slab, e);
343			radix_tree_preload_end();
344			goto retry;
345		}
346		memset(e, 0, sizeof(struct ino_entry));
347		e->ino = ino;
348
349		list_add_tail(&e->list, &im->ino_list);
350		if (type != ORPHAN_INO)
351			im->ino_num++;
352	}
353	spin_unlock(&im->ino_lock);
354	radix_tree_preload_end();
355}
356
357static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
358{
359	struct inode_management *im = &sbi->im[type];
360	struct ino_entry *e;
361
362	spin_lock(&im->ino_lock);
363	e = radix_tree_lookup(&im->ino_root, ino);
364	if (e) {
365		list_del(&e->list);
366		radix_tree_delete(&im->ino_root, ino);
367		im->ino_num--;
368		spin_unlock(&im->ino_lock);
369		kmem_cache_free(ino_entry_slab, e);
370		return;
371	}
372	spin_unlock(&im->ino_lock);
373}
374
375void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
376{
377	/* add new dirty ino entry into list */
378	__add_ino_entry(sbi, ino, type);
379}
380
381void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
382{
383	/* remove dirty ino entry from list */
384	__remove_ino_entry(sbi, ino, type);
385}
386
387/* mode should be APPEND_INO or UPDATE_INO */
388bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
389{
390	struct inode_management *im = &sbi->im[mode];
391	struct ino_entry *e;
392
393	spin_lock(&im->ino_lock);
394	e = radix_tree_lookup(&im->ino_root, ino);
395	spin_unlock(&im->ino_lock);
396	return e ? true : false;
397}
398
399void release_dirty_inode(struct f2fs_sb_info *sbi)
400{
401	struct ino_entry *e, *tmp;
402	int i;
403
404	for (i = APPEND_INO; i <= UPDATE_INO; i++) {
405		struct inode_management *im = &sbi->im[i];
406
407		spin_lock(&im->ino_lock);
408		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
409			list_del(&e->list);
410			radix_tree_delete(&im->ino_root, e->ino);
411			kmem_cache_free(ino_entry_slab, e);
412			im->ino_num--;
413		}
414		spin_unlock(&im->ino_lock);
415	}
416}
417
418int acquire_orphan_inode(struct f2fs_sb_info *sbi)
419{
420	struct inode_management *im = &sbi->im[ORPHAN_INO];
421	int err = 0;
422
423	spin_lock(&im->ino_lock);
424	if (unlikely(im->ino_num >= sbi->max_orphans))
425		err = -ENOSPC;
426	else
427		im->ino_num++;
428	spin_unlock(&im->ino_lock);
429
430	return err;
431}
432
433void release_orphan_inode(struct f2fs_sb_info *sbi)
434{
435	struct inode_management *im = &sbi->im[ORPHAN_INO];
436
437	spin_lock(&im->ino_lock);
438	f2fs_bug_on(sbi, im->ino_num == 0);
439	im->ino_num--;
440	spin_unlock(&im->ino_lock);
441}
442
443void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
444{
445	/* add new orphan ino entry into list */
446	__add_ino_entry(sbi, ino, ORPHAN_INO);
447}
448
449void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
450{
451	/* remove orphan entry from orphan list */
452	__remove_ino_entry(sbi, ino, ORPHAN_INO);
453}
454
455static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
456{
457	struct inode *inode = f2fs_iget(sbi->sb, ino);
458	f2fs_bug_on(sbi, IS_ERR(inode));
459	clear_nlink(inode);
460
461	/* truncate all the data during iput */
462	iput(inode);
463}
464
465void recover_orphan_inodes(struct f2fs_sb_info *sbi)
466{
467	block_t start_blk, orphan_blocks, i, j;
468
469	if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
470		return;
471
472	set_sbi_flag(sbi, SBI_POR_DOING);
473
474	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
475	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
476
477	ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP);
478
479	for (i = 0; i < orphan_blocks; i++) {
480		struct page *page = get_meta_page(sbi, start_blk + i);
481		struct f2fs_orphan_block *orphan_blk;
482
483		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
484		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
485			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
486			recover_orphan_inode(sbi, ino);
487		}
488		f2fs_put_page(page, 1);
489	}
490	/* clear Orphan Flag */
491	clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
492	clear_sbi_flag(sbi, SBI_POR_DOING);
493	return;
494}
495
496static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
497{
498	struct list_head *head;
499	struct f2fs_orphan_block *orphan_blk = NULL;
500	unsigned int nentries = 0;
501	unsigned short index;
502	unsigned short orphan_blocks;
503	struct page *page = NULL;
504	struct ino_entry *orphan = NULL;
505	struct inode_management *im = &sbi->im[ORPHAN_INO];
506
507	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
508
509	for (index = 0; index < orphan_blocks; index++)
510		grab_meta_page(sbi, start_blk + index);
511
512	index = 1;
513	spin_lock(&im->ino_lock);
514	head = &im->ino_list;
515
516	/* loop for each orphan inode entry and write them in Jornal block */
517	list_for_each_entry(orphan, head, list) {
518		if (!page) {
519			page = find_get_page(META_MAPPING(sbi), start_blk++);
520			f2fs_bug_on(sbi, !page);
521			orphan_blk =
522				(struct f2fs_orphan_block *)page_address(page);
523			memset(orphan_blk, 0, sizeof(*orphan_blk));
524			f2fs_put_page(page, 0);
525		}
526
527		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
528
529		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
530			/*
531			 * an orphan block is full of 1020 entries,
532			 * then we need to flush current orphan blocks
533			 * and bring another one in memory
534			 */
535			orphan_blk->blk_addr = cpu_to_le16(index);
536			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
537			orphan_blk->entry_count = cpu_to_le32(nentries);
538			set_page_dirty(page);
539			f2fs_put_page(page, 1);
540			index++;
541			nentries = 0;
542			page = NULL;
543		}
544	}
545
546	if (page) {
547		orphan_blk->blk_addr = cpu_to_le16(index);
548		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
549		orphan_blk->entry_count = cpu_to_le32(nentries);
550		set_page_dirty(page);
551		f2fs_put_page(page, 1);
552	}
553
554	spin_unlock(&im->ino_lock);
555}
556
557static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
558				block_t cp_addr, unsigned long long *version)
559{
560	struct page *cp_page_1, *cp_page_2 = NULL;
561	unsigned long blk_size = sbi->blocksize;
562	struct f2fs_checkpoint *cp_block;
563	unsigned long long cur_version = 0, pre_version = 0;
564	size_t crc_offset;
565	__u32 crc = 0;
566
567	/* Read the 1st cp block in this CP pack */
568	cp_page_1 = get_meta_page(sbi, cp_addr);
569
570	/* get the version number */
571	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
572	crc_offset = le32_to_cpu(cp_block->checksum_offset);
573	if (crc_offset >= blk_size)
574		goto invalid_cp1;
575
576	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
577	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
578		goto invalid_cp1;
579
580	pre_version = cur_cp_version(cp_block);
581
582	/* Read the 2nd cp block in this CP pack */
583	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
584	cp_page_2 = get_meta_page(sbi, cp_addr);
585
586	cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
587	crc_offset = le32_to_cpu(cp_block->checksum_offset);
588	if (crc_offset >= blk_size)
589		goto invalid_cp2;
590
591	crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
592	if (!f2fs_crc_valid(crc, cp_block, crc_offset))
593		goto invalid_cp2;
594
595	cur_version = cur_cp_version(cp_block);
596
597	if (cur_version == pre_version) {
598		*version = cur_version;
599		f2fs_put_page(cp_page_2, 1);
600		return cp_page_1;
601	}
602invalid_cp2:
603	f2fs_put_page(cp_page_2, 1);
604invalid_cp1:
605	f2fs_put_page(cp_page_1, 1);
606	return NULL;
607}
608
609int get_valid_checkpoint(struct f2fs_sb_info *sbi)
610{
611	struct f2fs_checkpoint *cp_block;
612	struct f2fs_super_block *fsb = sbi->raw_super;
613	struct page *cp1, *cp2, *cur_page;
614	unsigned long blk_size = sbi->blocksize;
615	unsigned long long cp1_version = 0, cp2_version = 0;
616	unsigned long long cp_start_blk_no;
617	unsigned int cp_blks = 1 + __cp_payload(sbi);
618	block_t cp_blk_no;
619	int i;
620
621	sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
622	if (!sbi->ckpt)
623		return -ENOMEM;
624	/*
625	 * Finding out valid cp block involves read both
626	 * sets( cp pack1 and cp pack 2)
627	 */
628	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
629	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
630
631	/* The second checkpoint pack should start at the next segment */
632	cp_start_blk_no += ((unsigned long long)1) <<
633				le32_to_cpu(fsb->log_blocks_per_seg);
634	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
635
636	if (cp1 && cp2) {
637		if (ver_after(cp2_version, cp1_version))
638			cur_page = cp2;
639		else
640			cur_page = cp1;
641	} else if (cp1) {
642		cur_page = cp1;
643	} else if (cp2) {
644		cur_page = cp2;
645	} else {
646		goto fail_no_cp;
647	}
648
649	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
650	memcpy(sbi->ckpt, cp_block, blk_size);
651
652	if (cp_blks <= 1)
653		goto done;
654
655	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
656	if (cur_page == cp2)
657		cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
658
659	for (i = 1; i < cp_blks; i++) {
660		void *sit_bitmap_ptr;
661		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
662
663		cur_page = get_meta_page(sbi, cp_blk_no + i);
664		sit_bitmap_ptr = page_address(cur_page);
665		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
666		f2fs_put_page(cur_page, 1);
667	}
668done:
669	f2fs_put_page(cp1, 1);
670	f2fs_put_page(cp2, 1);
671	return 0;
672
673fail_no_cp:
674	kfree(sbi->ckpt);
675	return -EINVAL;
676}
677
678static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
679{
680	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
681
682	if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
683		return -EEXIST;
684
685	set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
686	F2FS_I(inode)->dirty_dir = new;
687	list_add_tail(&new->list, &sbi->dir_inode_list);
688	stat_inc_dirty_dir(sbi);
689	return 0;
690}
691
692void update_dirty_page(struct inode *inode, struct page *page)
693{
694	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
695	struct inode_entry *new;
696	int ret = 0;
697
698	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
699		return;
700
701	if (!S_ISDIR(inode->i_mode)) {
702		inode_inc_dirty_pages(inode);
703		goto out;
704	}
705
706	new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
707	new->inode = inode;
708	INIT_LIST_HEAD(&new->list);
709
710	spin_lock(&sbi->dir_inode_lock);
711	ret = __add_dirty_inode(inode, new);
712	inode_inc_dirty_pages(inode);
713	spin_unlock(&sbi->dir_inode_lock);
714
715	if (ret)
716		kmem_cache_free(inode_entry_slab, new);
717out:
718	SetPagePrivate(page);
719	f2fs_trace_pid(page);
720}
721
722void add_dirty_dir_inode(struct inode *inode)
723{
724	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
725	struct inode_entry *new =
726			f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
727	int ret = 0;
728
729	new->inode = inode;
730	INIT_LIST_HEAD(&new->list);
731
732	spin_lock(&sbi->dir_inode_lock);
733	ret = __add_dirty_inode(inode, new);
734	spin_unlock(&sbi->dir_inode_lock);
735
736	if (ret)
737		kmem_cache_free(inode_entry_slab, new);
738}
739
740void remove_dirty_dir_inode(struct inode *inode)
741{
742	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
743	struct inode_entry *entry;
744
745	if (!S_ISDIR(inode->i_mode))
746		return;
747
748	spin_lock(&sbi->dir_inode_lock);
749	if (get_dirty_pages(inode) ||
750			!is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
751		spin_unlock(&sbi->dir_inode_lock);
752		return;
753	}
754
755	entry = F2FS_I(inode)->dirty_dir;
756	list_del(&entry->list);
757	F2FS_I(inode)->dirty_dir = NULL;
758	clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
759	stat_dec_dirty_dir(sbi);
760	spin_unlock(&sbi->dir_inode_lock);
761	kmem_cache_free(inode_entry_slab, entry);
762
763	/* Only from the recovery routine */
764	if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
765		clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
766		iput(inode);
767	}
768}
769
770void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
771{
772	struct list_head *head;
773	struct inode_entry *entry;
774	struct inode *inode;
775retry:
776	if (unlikely(f2fs_cp_error(sbi)))
777		return;
778
779	spin_lock(&sbi->dir_inode_lock);
780
781	head = &sbi->dir_inode_list;
782	if (list_empty(head)) {
783		spin_unlock(&sbi->dir_inode_lock);
784		return;
785	}
786	entry = list_entry(head->next, struct inode_entry, list);
787	inode = igrab(entry->inode);
788	spin_unlock(&sbi->dir_inode_lock);
789	if (inode) {
790		filemap_fdatawrite(inode->i_mapping);
791		iput(inode);
792	} else {
793		/*
794		 * We should submit bio, since it exists several
795		 * wribacking dentry pages in the freeing inode.
796		 */
797		f2fs_submit_merged_bio(sbi, DATA, WRITE);
798		cond_resched();
799	}
800	goto retry;
801}
802
803/*
804 * Freeze all the FS-operations for checkpoint.
805 */
806static int block_operations(struct f2fs_sb_info *sbi)
807{
808	struct writeback_control wbc = {
809		.sync_mode = WB_SYNC_ALL,
810		.nr_to_write = LONG_MAX,
811		.for_reclaim = 0,
812	};
813	struct blk_plug plug;
814	int err = 0;
815
816	blk_start_plug(&plug);
817
818retry_flush_dents:
819	f2fs_lock_all(sbi);
820	/* write all the dirty dentry pages */
821	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
822		f2fs_unlock_all(sbi);
823		sync_dirty_dir_inodes(sbi);
824		if (unlikely(f2fs_cp_error(sbi))) {
825			err = -EIO;
826			goto out;
827		}
828		goto retry_flush_dents;
829	}
830
831	/*
832	 * POR: we should ensure that there are no dirty node pages
833	 * until finishing nat/sit flush.
834	 */
835retry_flush_nodes:
836	down_write(&sbi->node_write);
837
838	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
839		up_write(&sbi->node_write);
840		sync_node_pages(sbi, 0, &wbc);
841		if (unlikely(f2fs_cp_error(sbi))) {
842			f2fs_unlock_all(sbi);
843			err = -EIO;
844			goto out;
845		}
846		goto retry_flush_nodes;
847	}
848out:
849	blk_finish_plug(&plug);
850	return err;
851}
852
853static void unblock_operations(struct f2fs_sb_info *sbi)
854{
855	up_write(&sbi->node_write);
856	f2fs_unlock_all(sbi);
857}
858
859static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
860{
861	DEFINE_WAIT(wait);
862
863	for (;;) {
864		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
865
866		if (!get_pages(sbi, F2FS_WRITEBACK))
867			break;
868
869		io_schedule();
870	}
871	finish_wait(&sbi->cp_wait, &wait);
872}
873
874static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
875{
876	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
877	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
878	struct f2fs_nm_info *nm_i = NM_I(sbi);
879	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
880	nid_t last_nid = nm_i->next_scan_nid;
881	block_t start_blk;
882	struct page *cp_page;
883	unsigned int data_sum_blocks, orphan_blocks;
884	__u32 crc32 = 0;
885	void *kaddr;
886	int i;
887	int cp_payload_blks = __cp_payload(sbi);
888
889	/*
890	 * This avoids to conduct wrong roll-forward operations and uses
891	 * metapages, so should be called prior to sync_meta_pages below.
892	 */
893	discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
894
895	/* Flush all the NAT/SIT pages */
896	while (get_pages(sbi, F2FS_DIRTY_META)) {
897		sync_meta_pages(sbi, META, LONG_MAX);
898		if (unlikely(f2fs_cp_error(sbi)))
899			return;
900	}
901
902	next_free_nid(sbi, &last_nid);
903
904	/*
905	 * modify checkpoint
906	 * version number is already updated
907	 */
908	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
909	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
910	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
911	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
912		ckpt->cur_node_segno[i] =
913			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
914		ckpt->cur_node_blkoff[i] =
915			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
916		ckpt->alloc_type[i + CURSEG_HOT_NODE] =
917				curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
918	}
919	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
920		ckpt->cur_data_segno[i] =
921			cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
922		ckpt->cur_data_blkoff[i] =
923			cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
924		ckpt->alloc_type[i + CURSEG_HOT_DATA] =
925				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
926	}
927
928	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
929	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
930	ckpt->next_free_nid = cpu_to_le32(last_nid);
931
932	/* 2 cp  + n data seg summary + orphan inode blocks */
933	data_sum_blocks = npages_for_summary_flush(sbi, false);
934	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
935		set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
936	else
937		clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
938
939	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
940	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
941			orphan_blocks);
942
943	if (__remain_node_summaries(cpc->reason))
944		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
945				cp_payload_blks + data_sum_blocks +
946				orphan_blocks + NR_CURSEG_NODE_TYPE);
947	else
948		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
949				cp_payload_blks + data_sum_blocks +
950				orphan_blocks);
951
952	if (cpc->reason == CP_UMOUNT)
953		set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
954	else
955		clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
956
957	if (cpc->reason == CP_FASTBOOT)
958		set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
959	else
960		clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
961
962	if (orphan_num)
963		set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
964	else
965		clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
966
967	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
968		set_ckpt_flags(ckpt, CP_FSCK_FLAG);
969
970	/* update SIT/NAT bitmap */
971	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
972	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
973
974	crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
975	*((__le32 *)((unsigned char *)ckpt +
976				le32_to_cpu(ckpt->checksum_offset)))
977				= cpu_to_le32(crc32);
978
979	start_blk = __start_cp_addr(sbi);
980
981	/* write out checkpoint buffer at block 0 */
982	cp_page = grab_meta_page(sbi, start_blk++);
983	kaddr = page_address(cp_page);
984	memcpy(kaddr, ckpt, F2FS_BLKSIZE);
985	set_page_dirty(cp_page);
986	f2fs_put_page(cp_page, 1);
987
988	for (i = 1; i < 1 + cp_payload_blks; i++) {
989		cp_page = grab_meta_page(sbi, start_blk++);
990		kaddr = page_address(cp_page);
991		memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE, F2FS_BLKSIZE);
992		set_page_dirty(cp_page);
993		f2fs_put_page(cp_page, 1);
994	}
995
996	if (orphan_num) {
997		write_orphan_inodes(sbi, start_blk);
998		start_blk += orphan_blocks;
999	}
1000
1001	write_data_summaries(sbi, start_blk);
1002	start_blk += data_sum_blocks;
1003	if (__remain_node_summaries(cpc->reason)) {
1004		write_node_summaries(sbi, start_blk);
1005		start_blk += NR_CURSEG_NODE_TYPE;
1006	}
1007
1008	/* writeout checkpoint block */
1009	cp_page = grab_meta_page(sbi, start_blk);
1010	kaddr = page_address(cp_page);
1011	memcpy(kaddr, ckpt, F2FS_BLKSIZE);
1012	set_page_dirty(cp_page);
1013	f2fs_put_page(cp_page, 1);
1014
1015	/* wait for previous submitted node/meta pages writeback */
1016	wait_on_all_pages_writeback(sbi);
1017
1018	if (unlikely(f2fs_cp_error(sbi)))
1019		return;
1020
1021	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1022	filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1023
1024	/* update user_block_counts */
1025	sbi->last_valid_block_count = sbi->total_valid_block_count;
1026	sbi->alloc_valid_block_count = 0;
1027
1028	/* Here, we only have one bio having CP pack */
1029	sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1030
1031	/* wait for previous submitted meta pages writeback */
1032	wait_on_all_pages_writeback(sbi);
1033
1034	release_dirty_inode(sbi);
1035
1036	if (unlikely(f2fs_cp_error(sbi)))
1037		return;
1038
1039	clear_prefree_segments(sbi);
1040	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1041}
1042
1043/*
1044 * We guarantee that this checkpoint procedure will not fail.
1045 */
1046void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1047{
1048	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1049	unsigned long long ckpt_ver;
1050
1051	mutex_lock(&sbi->cp_mutex);
1052
1053	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1054		(cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC))
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);
1093out:
1094	mutex_unlock(&sbi->cp_mutex);
1095	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1096}
1097
1098void 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
1116int __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
1131void destroy_checkpoint_caches(void)
1132{
1133	kmem_cache_destroy(ino_entry_slab);
1134	kmem_cache_destroy(inode_entry_slab);
1135}
1136