]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/f2fs/checkpoint.c
Merge branch 'akpm-current/current'
[karo-tx-linux.git] / fs / f2fs / checkpoint.c
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 *orphan_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 inode *inode = page->mapping->host;
155         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
156
157         trace_f2fs_writepage(page, META);
158
159         if (unlikely(sbi->por_doing))
160                 goto redirty_out;
161         if (wbc->for_reclaim)
162                 goto redirty_out;
163
164         /* Should not write any meta pages, if any IO error was occurred */
165         if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
166                 goto no_write;
167
168         f2fs_wait_on_page_writeback(page, META);
169         write_meta_page(sbi, page);
170 no_write:
171         dec_page_count(sbi, F2FS_DIRTY_META);
172         unlock_page(page);
173         return 0;
174
175 redirty_out:
176         redirty_page_for_writepage(wbc, page);
177         return AOP_WRITEPAGE_ACTIVATE;
178 }
179
180 static int f2fs_write_meta_pages(struct address_space *mapping,
181                                 struct writeback_control *wbc)
182 {
183         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
184         long diff, written;
185
186         trace_f2fs_writepages(mapping->host, wbc, META);
187
188         /* collect a number of dirty meta pages and write together */
189         if (wbc->for_kupdate ||
190                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
191                 goto skip_write;
192
193         /* if mounting is failed, skip writing node pages */
194         mutex_lock(&sbi->cp_mutex);
195         diff = nr_pages_to_write(sbi, META, wbc);
196         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
197         mutex_unlock(&sbi->cp_mutex);
198         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
199         return 0;
200
201 skip_write:
202         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
203         return 0;
204 }
205
206 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
207                                                 long nr_to_write)
208 {
209         struct address_space *mapping = META_MAPPING(sbi);
210         pgoff_t index = 0, end = LONG_MAX;
211         struct pagevec pvec;
212         long nwritten = 0;
213         struct writeback_control wbc = {
214                 .for_reclaim = 0,
215         };
216
217         pagevec_init(&pvec, 0);
218
219         while (index <= end) {
220                 int i, nr_pages;
221                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
222                                 PAGECACHE_TAG_DIRTY,
223                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
224                 if (unlikely(nr_pages == 0))
225                         break;
226
227                 for (i = 0; i < nr_pages; i++) {
228                         struct page *page = pvec.pages[i];
229
230                         lock_page(page);
231
232                         if (unlikely(page->mapping != mapping)) {
233 continue_unlock:
234                                 unlock_page(page);
235                                 continue;
236                         }
237                         if (!PageDirty(page)) {
238                                 /* someone wrote it for us */
239                                 goto continue_unlock;
240                         }
241
242                         if (!clear_page_dirty_for_io(page))
243                                 goto continue_unlock;
244
245                         if (f2fs_write_meta_page(page, &wbc)) {
246                                 unlock_page(page);
247                                 break;
248                         }
249                         nwritten++;
250                         if (unlikely(nwritten >= nr_to_write))
251                                 break;
252                 }
253                 pagevec_release(&pvec);
254                 cond_resched();
255         }
256
257         if (nwritten)
258                 f2fs_submit_merged_bio(sbi, type, WRITE);
259
260         return nwritten;
261 }
262
263 static int f2fs_set_meta_page_dirty(struct page *page)
264 {
265         struct address_space *mapping = page->mapping;
266         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
267
268         trace_f2fs_set_page_dirty(page, META);
269
270         SetPageUptodate(page);
271         if (!PageDirty(page)) {
272                 __set_page_dirty_nobuffers(page);
273                 inc_page_count(sbi, F2FS_DIRTY_META);
274                 return 1;
275         }
276         return 0;
277 }
278
279 const struct address_space_operations f2fs_meta_aops = {
280         .writepage      = f2fs_write_meta_page,
281         .writepages     = f2fs_write_meta_pages,
282         .set_page_dirty = f2fs_set_meta_page_dirty,
283 };
284
285 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
286 {
287         int err = 0;
288
289         spin_lock(&sbi->orphan_inode_lock);
290         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
291                 err = -ENOSPC;
292         else
293                 sbi->n_orphans++;
294         spin_unlock(&sbi->orphan_inode_lock);
295
296         return err;
297 }
298
299 void release_orphan_inode(struct f2fs_sb_info *sbi)
300 {
301         spin_lock(&sbi->orphan_inode_lock);
302         f2fs_bug_on(sbi->n_orphans == 0);
303         sbi->n_orphans--;
304         spin_unlock(&sbi->orphan_inode_lock);
305 }
306
307 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
308 {
309         struct list_head *head;
310         struct orphan_inode_entry *new, *orphan;
311
312         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
313         new->ino = ino;
314
315         spin_lock(&sbi->orphan_inode_lock);
316         head = &sbi->orphan_inode_list;
317         list_for_each_entry(orphan, head, list) {
318                 if (orphan->ino == ino) {
319                         spin_unlock(&sbi->orphan_inode_lock);
320                         kmem_cache_free(orphan_entry_slab, new);
321                         return;
322                 }
323
324                 if (orphan->ino > ino)
325                         break;
326         }
327
328         /* add new orphan entry into list which is sorted by inode number */
329         list_add_tail(&new->list, &orphan->list);
330         spin_unlock(&sbi->orphan_inode_lock);
331 }
332
333 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
334 {
335         struct list_head *head;
336         struct orphan_inode_entry *orphan;
337
338         spin_lock(&sbi->orphan_inode_lock);
339         head = &sbi->orphan_inode_list;
340         list_for_each_entry(orphan, head, list) {
341                 if (orphan->ino == ino) {
342                         list_del(&orphan->list);
343                         f2fs_bug_on(sbi->n_orphans == 0);
344                         sbi->n_orphans--;
345                         spin_unlock(&sbi->orphan_inode_lock);
346                         kmem_cache_free(orphan_entry_slab, orphan);
347                         return;
348                 }
349         }
350         spin_unlock(&sbi->orphan_inode_lock);
351 }
352
353 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
354 {
355         struct inode *inode = f2fs_iget(sbi->sb, ino);
356         f2fs_bug_on(IS_ERR(inode));
357         clear_nlink(inode);
358
359         /* truncate all the data during iput */
360         iput(inode);
361 }
362
363 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
364 {
365         block_t start_blk, orphan_blkaddr, i, j;
366
367         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
368                 return;
369
370         sbi->por_doing = true;
371         start_blk = __start_cp_addr(sbi) + 1;
372         orphan_blkaddr = __start_sum_addr(sbi) - 1;
373
374         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
375
376         for (i = 0; i < orphan_blkaddr; i++) {
377                 struct page *page = get_meta_page(sbi, start_blk + i);
378                 struct f2fs_orphan_block *orphan_blk;
379
380                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
381                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
382                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
383                         recover_orphan_inode(sbi, ino);
384                 }
385                 f2fs_put_page(page, 1);
386         }
387         /* clear Orphan Flag */
388         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
389         sbi->por_doing = false;
390         return;
391 }
392
393 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
394 {
395         struct list_head *head;
396         struct f2fs_orphan_block *orphan_blk = NULL;
397         unsigned int nentries = 0;
398         unsigned short index;
399         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
400                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
401         struct page *page = NULL;
402         struct orphan_inode_entry *orphan = NULL;
403
404         for (index = 0; index < orphan_blocks; index++)
405                 grab_meta_page(sbi, start_blk + index);
406
407         index = 1;
408         spin_lock(&sbi->orphan_inode_lock);
409         head = &sbi->orphan_inode_list;
410
411         /* loop for each orphan inode entry and write them in Jornal block */
412         list_for_each_entry(orphan, head, list) {
413                 if (!page) {
414                         page = find_get_page(META_MAPPING(sbi), start_blk++);
415                         f2fs_bug_on(!page);
416                         orphan_blk =
417                                 (struct f2fs_orphan_block *)page_address(page);
418                         memset(orphan_blk, 0, sizeof(*orphan_blk));
419                         f2fs_put_page(page, 0);
420                 }
421
422                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
423
424                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
425                         /*
426                          * an orphan block is full of 1020 entries,
427                          * then we need to flush current orphan blocks
428                          * and bring another one in memory
429                          */
430                         orphan_blk->blk_addr = cpu_to_le16(index);
431                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
432                         orphan_blk->entry_count = cpu_to_le32(nentries);
433                         set_page_dirty(page);
434                         f2fs_put_page(page, 1);
435                         index++;
436                         nentries = 0;
437                         page = NULL;
438                 }
439         }
440
441         if (page) {
442                 orphan_blk->blk_addr = cpu_to_le16(index);
443                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
444                 orphan_blk->entry_count = cpu_to_le32(nentries);
445                 set_page_dirty(page);
446                 f2fs_put_page(page, 1);
447         }
448
449         spin_unlock(&sbi->orphan_inode_lock);
450 }
451
452 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
453                                 block_t cp_addr, unsigned long long *version)
454 {
455         struct page *cp_page_1, *cp_page_2 = NULL;
456         unsigned long blk_size = sbi->blocksize;
457         struct f2fs_checkpoint *cp_block;
458         unsigned long long cur_version = 0, pre_version = 0;
459         size_t crc_offset;
460         __u32 crc = 0;
461
462         /* Read the 1st cp block in this CP pack */
463         cp_page_1 = get_meta_page(sbi, cp_addr);
464
465         /* get the version number */
466         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
467         crc_offset = le32_to_cpu(cp_block->checksum_offset);
468         if (crc_offset >= blk_size)
469                 goto invalid_cp1;
470
471         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
472         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
473                 goto invalid_cp1;
474
475         pre_version = cur_cp_version(cp_block);
476
477         /* Read the 2nd cp block in this CP pack */
478         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
479         cp_page_2 = get_meta_page(sbi, cp_addr);
480
481         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
482         crc_offset = le32_to_cpu(cp_block->checksum_offset);
483         if (crc_offset >= blk_size)
484                 goto invalid_cp2;
485
486         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
487         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
488                 goto invalid_cp2;
489
490         cur_version = cur_cp_version(cp_block);
491
492         if (cur_version == pre_version) {
493                 *version = cur_version;
494                 f2fs_put_page(cp_page_2, 1);
495                 return cp_page_1;
496         }
497 invalid_cp2:
498         f2fs_put_page(cp_page_2, 1);
499 invalid_cp1:
500         f2fs_put_page(cp_page_1, 1);
501         return NULL;
502 }
503
504 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
505 {
506         struct f2fs_checkpoint *cp_block;
507         struct f2fs_super_block *fsb = sbi->raw_super;
508         struct page *cp1, *cp2, *cur_page;
509         unsigned long blk_size = sbi->blocksize;
510         unsigned long long cp1_version = 0, cp2_version = 0;
511         unsigned long long cp_start_blk_no;
512
513         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
514         if (!sbi->ckpt)
515                 return -ENOMEM;
516         /*
517          * Finding out valid cp block involves read both
518          * sets( cp pack1 and cp pack 2)
519          */
520         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
521         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
522
523         /* The second checkpoint pack should start at the next segment */
524         cp_start_blk_no += ((unsigned long long)1) <<
525                                 le32_to_cpu(fsb->log_blocks_per_seg);
526         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
527
528         if (cp1 && cp2) {
529                 if (ver_after(cp2_version, cp1_version))
530                         cur_page = cp2;
531                 else
532                         cur_page = cp1;
533         } else if (cp1) {
534                 cur_page = cp1;
535         } else if (cp2) {
536                 cur_page = cp2;
537         } else {
538                 goto fail_no_cp;
539         }
540
541         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
542         memcpy(sbi->ckpt, cp_block, blk_size);
543
544         f2fs_put_page(cp1, 1);
545         f2fs_put_page(cp2, 1);
546         return 0;
547
548 fail_no_cp:
549         kfree(sbi->ckpt);
550         return -EINVAL;
551 }
552
553 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
554 {
555         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
556
557         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
558                 return -EEXIST;
559
560         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
561         F2FS_I(inode)->dirty_dir = new;
562         list_add_tail(&new->list, &sbi->dir_inode_list);
563         stat_inc_dirty_dir(sbi);
564         return 0;
565 }
566
567 void set_dirty_dir_page(struct inode *inode, struct page *page)
568 {
569         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
570         struct dir_inode_entry *new;
571         int ret = 0;
572
573         if (!S_ISDIR(inode->i_mode))
574                 return;
575
576         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
577         new->inode = inode;
578         INIT_LIST_HEAD(&new->list);
579
580         spin_lock(&sbi->dir_inode_lock);
581         ret = __add_dirty_inode(inode, new);
582         inode_inc_dirty_dents(inode);
583         SetPagePrivate(page);
584         spin_unlock(&sbi->dir_inode_lock);
585
586         if (ret)
587                 kmem_cache_free(inode_entry_slab, new);
588 }
589
590 void add_dirty_dir_inode(struct inode *inode)
591 {
592         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
593         struct dir_inode_entry *new =
594                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
595         int ret = 0;
596
597         new->inode = inode;
598         INIT_LIST_HEAD(&new->list);
599
600         spin_lock(&sbi->dir_inode_lock);
601         ret = __add_dirty_inode(inode, new);
602         spin_unlock(&sbi->dir_inode_lock);
603
604         if (ret)
605                 kmem_cache_free(inode_entry_slab, new);
606 }
607
608 void remove_dirty_dir_inode(struct inode *inode)
609 {
610         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
611         struct dir_inode_entry *entry;
612
613         if (!S_ISDIR(inode->i_mode))
614                 return;
615
616         spin_lock(&sbi->dir_inode_lock);
617         if (get_dirty_dents(inode) ||
618                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
619                 spin_unlock(&sbi->dir_inode_lock);
620                 return;
621         }
622
623         entry = F2FS_I(inode)->dirty_dir;
624         list_del(&entry->list);
625         F2FS_I(inode)->dirty_dir = NULL;
626         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
627         stat_dec_dirty_dir(sbi);
628         spin_unlock(&sbi->dir_inode_lock);
629         kmem_cache_free(inode_entry_slab, entry);
630
631         /* Only from the recovery routine */
632         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
633                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
634                 iput(inode);
635         }
636 }
637
638 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
639 {
640         struct list_head *head;
641         struct dir_inode_entry *entry;
642         struct inode *inode;
643 retry:
644         spin_lock(&sbi->dir_inode_lock);
645
646         head = &sbi->dir_inode_list;
647         if (list_empty(head)) {
648                 spin_unlock(&sbi->dir_inode_lock);
649                 return;
650         }
651         entry = list_entry(head->next, struct dir_inode_entry, list);
652         inode = igrab(entry->inode);
653         spin_unlock(&sbi->dir_inode_lock);
654         if (inode) {
655                 filemap_fdatawrite(inode->i_mapping);
656                 iput(inode);
657         } else {
658                 /*
659                  * We should submit bio, since it exists several
660                  * wribacking dentry pages in the freeing inode.
661                  */
662                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
663         }
664         goto retry;
665 }
666
667 /*
668  * Freeze all the FS-operations for checkpoint.
669  */
670 static void block_operations(struct f2fs_sb_info *sbi)
671 {
672         struct writeback_control wbc = {
673                 .sync_mode = WB_SYNC_ALL,
674                 .nr_to_write = LONG_MAX,
675                 .for_reclaim = 0,
676         };
677         struct blk_plug plug;
678
679         blk_start_plug(&plug);
680
681 retry_flush_dents:
682         f2fs_lock_all(sbi);
683         /* write all the dirty dentry pages */
684         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
685                 f2fs_unlock_all(sbi);
686                 sync_dirty_dir_inodes(sbi);
687                 goto retry_flush_dents;
688         }
689
690         /*
691          * POR: we should ensure that there is no dirty node pages
692          * until finishing nat/sit flush.
693          */
694 retry_flush_nodes:
695         mutex_lock(&sbi->node_write);
696
697         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
698                 mutex_unlock(&sbi->node_write);
699                 sync_node_pages(sbi, 0, &wbc);
700                 goto retry_flush_nodes;
701         }
702         blk_finish_plug(&plug);
703 }
704
705 static void unblock_operations(struct f2fs_sb_info *sbi)
706 {
707         mutex_unlock(&sbi->node_write);
708         f2fs_unlock_all(sbi);
709 }
710
711 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
712 {
713         DEFINE_WAIT(wait);
714
715         for (;;) {
716                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
717
718                 if (!get_pages(sbi, F2FS_WRITEBACK))
719                         break;
720
721                 io_schedule();
722         }
723         finish_wait(&sbi->cp_wait, &wait);
724 }
725
726 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
727 {
728         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
729         nid_t last_nid = 0;
730         block_t start_blk;
731         struct page *cp_page;
732         unsigned int data_sum_blocks, orphan_blocks;
733         __u32 crc32 = 0;
734         void *kaddr;
735         int i;
736
737         /*
738          * This avoids to conduct wrong roll-forward operations and uses
739          * metapages, so should be called prior to sync_meta_pages below.
740          */
741         discard_next_dnode(sbi);
742
743         /* Flush all the NAT/SIT pages */
744         while (get_pages(sbi, F2FS_DIRTY_META))
745                 sync_meta_pages(sbi, META, LONG_MAX);
746
747         next_free_nid(sbi, &last_nid);
748
749         /*
750          * modify checkpoint
751          * version number is already updated
752          */
753         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
754         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
755         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
756         for (i = 0; i < 3; i++) {
757                 ckpt->cur_node_segno[i] =
758                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
759                 ckpt->cur_node_blkoff[i] =
760                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
761                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
762                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
763         }
764         for (i = 0; i < 3; i++) {
765                 ckpt->cur_data_segno[i] =
766                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
767                 ckpt->cur_data_blkoff[i] =
768                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
769                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
770                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
771         }
772
773         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
774         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
775         ckpt->next_free_nid = cpu_to_le32(last_nid);
776
777         /* 2 cp  + n data seg summary + orphan inode blocks */
778         data_sum_blocks = npages_for_summary_flush(sbi);
779         if (data_sum_blocks < 3)
780                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
781         else
782                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
783
784         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
785                                         / F2FS_ORPHANS_PER_BLOCK;
786         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
787
788         if (is_umount) {
789                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
790                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
791                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
792         } else {
793                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
794                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
795                         data_sum_blocks + orphan_blocks);
796         }
797
798         if (sbi->n_orphans)
799                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
800         else
801                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
802
803         /* update SIT/NAT bitmap */
804         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
805         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
806
807         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
808         *((__le32 *)((unsigned char *)ckpt +
809                                 le32_to_cpu(ckpt->checksum_offset)))
810                                 = cpu_to_le32(crc32);
811
812         start_blk = __start_cp_addr(sbi);
813
814         /* write out checkpoint buffer at block 0 */
815         cp_page = grab_meta_page(sbi, start_blk++);
816         kaddr = page_address(cp_page);
817         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
818         set_page_dirty(cp_page);
819         f2fs_put_page(cp_page, 1);
820
821         if (sbi->n_orphans) {
822                 write_orphan_inodes(sbi, start_blk);
823                 start_blk += orphan_blocks;
824         }
825
826         write_data_summaries(sbi, start_blk);
827         start_blk += data_sum_blocks;
828         if (is_umount) {
829                 write_node_summaries(sbi, start_blk);
830                 start_blk += NR_CURSEG_NODE_TYPE;
831         }
832
833         /* writeout checkpoint block */
834         cp_page = grab_meta_page(sbi, start_blk);
835         kaddr = page_address(cp_page);
836         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
837         set_page_dirty(cp_page);
838         f2fs_put_page(cp_page, 1);
839
840         /* wait for previous submitted node/meta pages writeback */
841         wait_on_all_pages_writeback(sbi);
842
843         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
844         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
845
846         /* update user_block_counts */
847         sbi->last_valid_block_count = sbi->total_valid_block_count;
848         sbi->alloc_valid_block_count = 0;
849
850         /* Here, we only have one bio having CP pack */
851         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
852
853         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
854                 clear_prefree_segments(sbi);
855                 F2FS_RESET_SB_DIRT(sbi);
856         }
857 }
858
859 /*
860  * We guarantee that this checkpoint procedure should not fail.
861  */
862 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
863 {
864         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
865         unsigned long long ckpt_ver;
866
867         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
868
869         mutex_lock(&sbi->cp_mutex);
870         block_operations(sbi);
871
872         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
873
874         f2fs_submit_merged_bio(sbi, DATA, WRITE);
875         f2fs_submit_merged_bio(sbi, NODE, WRITE);
876         f2fs_submit_merged_bio(sbi, META, WRITE);
877
878         /*
879          * update checkpoint pack index
880          * Increase the version number so that
881          * SIT entries and seg summaries are written at correct place
882          */
883         ckpt_ver = cur_cp_version(ckpt);
884         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
885
886         /* write cached NAT/SIT entries to NAT/SIT area */
887         flush_nat_entries(sbi);
888         flush_sit_entries(sbi);
889
890         /* unlock all the fs_lock[] in do_checkpoint() */
891         do_checkpoint(sbi, is_umount);
892
893         unblock_operations(sbi);
894         mutex_unlock(&sbi->cp_mutex);
895
896         stat_inc_cp_count(sbi->stat_info);
897         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
898 }
899
900 void init_orphan_info(struct f2fs_sb_info *sbi)
901 {
902         spin_lock_init(&sbi->orphan_inode_lock);
903         INIT_LIST_HEAD(&sbi->orphan_inode_list);
904         sbi->n_orphans = 0;
905         /*
906          * considering 512 blocks in a segment 8 blocks are needed for cp
907          * and log segment summaries. Remaining blocks are used to keep
908          * orphan entries with the limitation one reserved segment
909          * for cp pack we can have max 1020*504 orphan entries
910          */
911         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
912                                 * F2FS_ORPHANS_PER_BLOCK;
913 }
914
915 int __init create_checkpoint_caches(void)
916 {
917         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
918                         sizeof(struct orphan_inode_entry));
919         if (!orphan_entry_slab)
920                 return -ENOMEM;
921         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
922                         sizeof(struct dir_inode_entry));
923         if (!inode_entry_slab) {
924                 kmem_cache_destroy(orphan_entry_slab);
925                 return -ENOMEM;
926         }
927         return 0;
928 }
929
930 void destroy_checkpoint_caches(void)
931 {
932         kmem_cache_destroy(orphan_entry_slab);
933         kmem_cache_destroy(inode_entry_slab);
934 }