04fe1df052b2b9e6a0c3f662797c3f46727ccdbe
[sfrench/cifs-2.6.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.h"
24 #include <trace/events/f2fs.h>
25
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *inode_entry_slab;
28
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
30 {
31         set_ckpt_flags(sbi, CP_ERROR_FLAG);
32         sbi->sb->s_flags |= MS_RDONLY;
33         if (!end_io)
34                 f2fs_flush_merged_writes(sbi);
35 }
36
37 /*
38  * We guarantee no failure on the returned page.
39  */
40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
41 {
42         struct address_space *mapping = META_MAPPING(sbi);
43         struct page *page = NULL;
44 repeat:
45         page = f2fs_grab_cache_page(mapping, index, false);
46         if (!page) {
47                 cond_resched();
48                 goto repeat;
49         }
50         f2fs_wait_on_page_writeback(page, META, true);
51         if (!PageUptodate(page))
52                 SetPageUptodate(page);
53         return page;
54 }
55
56 /*
57  * We guarantee no failure on the returned page.
58  */
59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
60                                                         bool is_meta)
61 {
62         struct address_space *mapping = META_MAPPING(sbi);
63         struct page *page;
64         struct f2fs_io_info fio = {
65                 .sbi = sbi,
66                 .type = META,
67                 .op = REQ_OP_READ,
68                 .op_flags = REQ_META | REQ_PRIO,
69                 .old_blkaddr = index,
70                 .new_blkaddr = index,
71                 .encrypted_page = NULL,
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                 f2fs_stop_checkpoint(sbi, false);
105 out:
106         return page;
107 }
108
109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
110 {
111         return __get_meta_page(sbi, index, true);
112 }
113
114 /* for POR only */
115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
116 {
117         return __get_meta_page(sbi, index, false);
118 }
119
120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
121 {
122         switch (type) {
123         case META_NAT:
124                 break;
125         case META_SIT:
126                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
127                         return false;
128                 break;
129         case META_SSA:
130                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
131                         blkaddr < SM_I(sbi)->ssa_blkaddr))
132                         return false;
133                 break;
134         case META_CP:
135                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
136                         blkaddr < __start_cp_addr(sbi)))
137                         return false;
138                 break;
139         case META_POR:
140                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
141                         blkaddr < MAIN_BLKADDR(sbi)))
142                         return false;
143                 break;
144         default:
145                 BUG();
146         }
147
148         return true;
149 }
150
151 /*
152  * Readahead CP/NAT/SIT/SSA pages
153  */
154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
155                                                         int type, bool sync)
156 {
157         struct page *page;
158         block_t blkno = start;
159         struct f2fs_io_info fio = {
160                 .sbi = sbi,
161                 .type = META,
162                 .op = REQ_OP_READ,
163                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
164                 .encrypted_page = NULL,
165                 .in_list = false,
166         };
167         struct blk_plug plug;
168
169         if (unlikely(type == META_POR))
170                 fio.op_flags &= ~REQ_META;
171
172         blk_start_plug(&plug);
173         for (; nrpages-- > 0; blkno++) {
174
175                 if (!is_valid_blkaddr(sbi, blkno, type))
176                         goto out;
177
178                 switch (type) {
179                 case META_NAT:
180                         if (unlikely(blkno >=
181                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
182                                 blkno = 0;
183                         /* get nat block addr */
184                         fio.new_blkaddr = current_nat_addr(sbi,
185                                         blkno * NAT_ENTRY_PER_BLOCK);
186                         break;
187                 case META_SIT:
188                         /* get sit block addr */
189                         fio.new_blkaddr = current_sit_addr(sbi,
190                                         blkno * SIT_ENTRY_PER_BLOCK);
191                         break;
192                 case META_SSA:
193                 case META_CP:
194                 case META_POR:
195                         fio.new_blkaddr = blkno;
196                         break;
197                 default:
198                         BUG();
199                 }
200
201                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
202                                                 fio.new_blkaddr, false);
203                 if (!page)
204                         continue;
205                 if (PageUptodate(page)) {
206                         f2fs_put_page(page, 1);
207                         continue;
208                 }
209
210                 fio.page = page;
211                 f2fs_submit_page_bio(&fio);
212                 f2fs_put_page(page, 0);
213         }
214 out:
215         blk_finish_plug(&plug);
216         return blkno - start;
217 }
218
219 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
220 {
221         struct page *page;
222         bool readahead = false;
223
224         page = find_get_page(META_MAPPING(sbi), index);
225         if (!page || !PageUptodate(page))
226                 readahead = true;
227         f2fs_put_page(page, 0);
228
229         if (readahead)
230                 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
231 }
232
233 static int __f2fs_write_meta_page(struct page *page,
234                                 struct writeback_control *wbc,
235                                 enum iostat_type io_type)
236 {
237         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
238
239         trace_f2fs_writepage(page, META);
240
241         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
242                 goto redirty_out;
243         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
244                 goto redirty_out;
245         if (unlikely(f2fs_cp_error(sbi)))
246                 goto redirty_out;
247
248         write_meta_page(sbi, page, io_type);
249         dec_page_count(sbi, F2FS_DIRTY_META);
250
251         if (wbc->for_reclaim)
252                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
253                                                 0, page->index, META);
254
255         unlock_page(page);
256
257         if (unlikely(f2fs_cp_error(sbi)))
258                 f2fs_submit_merged_write(sbi, META);
259
260         return 0;
261
262 redirty_out:
263         redirty_page_for_writepage(wbc, page);
264         return AOP_WRITEPAGE_ACTIVATE;
265 }
266
267 static int f2fs_write_meta_page(struct page *page,
268                                 struct writeback_control *wbc)
269 {
270         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
271 }
272
273 static int f2fs_write_meta_pages(struct address_space *mapping,
274                                 struct writeback_control *wbc)
275 {
276         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
277         long diff, written;
278
279         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
280                 goto skip_write;
281
282         /* collect a number of dirty meta pages and write together */
283         if (wbc->for_kupdate ||
284                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
285                 goto skip_write;
286
287         /* if locked failed, cp will flush dirty pages instead */
288         if (!mutex_trylock(&sbi->cp_mutex))
289                 goto skip_write;
290
291         trace_f2fs_writepages(mapping->host, wbc, META);
292         diff = nr_pages_to_write(sbi, META, wbc);
293         written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
294         mutex_unlock(&sbi->cp_mutex);
295         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
296         return 0;
297
298 skip_write:
299         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
300         trace_f2fs_writepages(mapping->host, wbc, META);
301         return 0;
302 }
303
304 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
305                                 long nr_to_write, enum iostat_type io_type)
306 {
307         struct address_space *mapping = META_MAPPING(sbi);
308         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
309         struct pagevec pvec;
310         long nwritten = 0;
311         struct writeback_control wbc = {
312                 .for_reclaim = 0,
313         };
314         struct blk_plug plug;
315
316         pagevec_init(&pvec, 0);
317
318         blk_start_plug(&plug);
319
320         while (index <= end) {
321                 int i, nr_pages;
322                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
323                                 PAGECACHE_TAG_DIRTY,
324                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
325                 if (unlikely(nr_pages == 0))
326                         break;
327
328                 for (i = 0; i < nr_pages; i++) {
329                         struct page *page = pvec.pages[i];
330
331                         if (prev == ULONG_MAX)
332                                 prev = page->index - 1;
333                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
334                                 pagevec_release(&pvec);
335                                 goto stop;
336                         }
337
338                         lock_page(page);
339
340                         if (unlikely(page->mapping != mapping)) {
341 continue_unlock:
342                                 unlock_page(page);
343                                 continue;
344                         }
345                         if (!PageDirty(page)) {
346                                 /* someone wrote it for us */
347                                 goto continue_unlock;
348                         }
349
350                         f2fs_wait_on_page_writeback(page, META, true);
351
352                         BUG_ON(PageWriteback(page));
353                         if (!clear_page_dirty_for_io(page))
354                                 goto continue_unlock;
355
356                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
357                                 unlock_page(page);
358                                 break;
359                         }
360                         nwritten++;
361                         prev = page->index;
362                         if (unlikely(nwritten >= nr_to_write))
363                                 break;
364                 }
365                 pagevec_release(&pvec);
366                 cond_resched();
367         }
368 stop:
369         if (nwritten)
370                 f2fs_submit_merged_write(sbi, type);
371
372         blk_finish_plug(&plug);
373
374         return nwritten;
375 }
376
377 static int f2fs_set_meta_page_dirty(struct page *page)
378 {
379         trace_f2fs_set_page_dirty(page, META);
380
381         if (!PageUptodate(page))
382                 SetPageUptodate(page);
383         if (!PageDirty(page)) {
384                 f2fs_set_page_dirty_nobuffers(page);
385                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
386                 SetPagePrivate(page);
387                 f2fs_trace_pid(page);
388                 return 1;
389         }
390         return 0;
391 }
392
393 const struct address_space_operations f2fs_meta_aops = {
394         .writepage      = f2fs_write_meta_page,
395         .writepages     = f2fs_write_meta_pages,
396         .set_page_dirty = f2fs_set_meta_page_dirty,
397         .invalidatepage = f2fs_invalidate_page,
398         .releasepage    = f2fs_release_page,
399 #ifdef CONFIG_MIGRATION
400         .migratepage    = f2fs_migrate_page,
401 #endif
402 };
403
404 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
405 {
406         struct inode_management *im = &sbi->im[type];
407         struct ino_entry *e, *tmp;
408
409         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
410 retry:
411         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
412
413         spin_lock(&im->ino_lock);
414         e = radix_tree_lookup(&im->ino_root, ino);
415         if (!e) {
416                 e = tmp;
417                 if (radix_tree_insert(&im->ino_root, ino, e)) {
418                         spin_unlock(&im->ino_lock);
419                         radix_tree_preload_end();
420                         goto retry;
421                 }
422                 memset(e, 0, sizeof(struct ino_entry));
423                 e->ino = ino;
424
425                 list_add_tail(&e->list, &im->ino_list);
426                 if (type != ORPHAN_INO)
427                         im->ino_num++;
428         }
429         spin_unlock(&im->ino_lock);
430         radix_tree_preload_end();
431
432         if (e != tmp)
433                 kmem_cache_free(ino_entry_slab, tmp);
434 }
435
436 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
437 {
438         struct inode_management *im = &sbi->im[type];
439         struct ino_entry *e;
440
441         spin_lock(&im->ino_lock);
442         e = radix_tree_lookup(&im->ino_root, ino);
443         if (e) {
444                 list_del(&e->list);
445                 radix_tree_delete(&im->ino_root, ino);
446                 im->ino_num--;
447                 spin_unlock(&im->ino_lock);
448                 kmem_cache_free(ino_entry_slab, e);
449                 return;
450         }
451         spin_unlock(&im->ino_lock);
452 }
453
454 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
455 {
456         /* add new dirty ino entry into list */
457         __add_ino_entry(sbi, ino, type);
458 }
459
460 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
461 {
462         /* remove dirty ino entry from list */
463         __remove_ino_entry(sbi, ino, type);
464 }
465
466 /* mode should be APPEND_INO or UPDATE_INO */
467 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
468 {
469         struct inode_management *im = &sbi->im[mode];
470         struct ino_entry *e;
471
472         spin_lock(&im->ino_lock);
473         e = radix_tree_lookup(&im->ino_root, ino);
474         spin_unlock(&im->ino_lock);
475         return e ? true : false;
476 }
477
478 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
479 {
480         struct ino_entry *e, *tmp;
481         int i;
482
483         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
484                 struct inode_management *im = &sbi->im[i];
485
486                 spin_lock(&im->ino_lock);
487                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
488                         list_del(&e->list);
489                         radix_tree_delete(&im->ino_root, e->ino);
490                         kmem_cache_free(ino_entry_slab, e);
491                         im->ino_num--;
492                 }
493                 spin_unlock(&im->ino_lock);
494         }
495 }
496
497 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
498 {
499         struct inode_management *im = &sbi->im[ORPHAN_INO];
500         int err = 0;
501
502         spin_lock(&im->ino_lock);
503
504 #ifdef CONFIG_F2FS_FAULT_INJECTION
505         if (time_to_inject(sbi, FAULT_ORPHAN)) {
506                 spin_unlock(&im->ino_lock);
507                 f2fs_show_injection_info(FAULT_ORPHAN);
508                 return -ENOSPC;
509         }
510 #endif
511         if (unlikely(im->ino_num >= sbi->max_orphans))
512                 err = -ENOSPC;
513         else
514                 im->ino_num++;
515         spin_unlock(&im->ino_lock);
516
517         return err;
518 }
519
520 void release_orphan_inode(struct f2fs_sb_info *sbi)
521 {
522         struct inode_management *im = &sbi->im[ORPHAN_INO];
523
524         spin_lock(&im->ino_lock);
525         f2fs_bug_on(sbi, im->ino_num == 0);
526         im->ino_num--;
527         spin_unlock(&im->ino_lock);
528 }
529
530 void add_orphan_inode(struct inode *inode)
531 {
532         /* add new orphan ino entry into list */
533         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
534         update_inode_page(inode);
535 }
536
537 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
538 {
539         /* remove orphan entry from orphan list */
540         __remove_ino_entry(sbi, ino, ORPHAN_INO);
541 }
542
543 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
544 {
545         struct inode *inode;
546         struct node_info ni;
547         int err = acquire_orphan_inode(sbi);
548
549         if (err) {
550                 set_sbi_flag(sbi, SBI_NEED_FSCK);
551                 f2fs_msg(sbi->sb, KERN_WARNING,
552                                 "%s: orphan failed (ino=%x), run fsck to fix.",
553                                 __func__, ino);
554                 return err;
555         }
556
557         __add_ino_entry(sbi, ino, ORPHAN_INO);
558
559         inode = f2fs_iget_retry(sbi->sb, ino);
560         if (IS_ERR(inode)) {
561                 /*
562                  * there should be a bug that we can't find the entry
563                  * to orphan inode.
564                  */
565                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
566                 return PTR_ERR(inode);
567         }
568
569         clear_nlink(inode);
570
571         /* truncate all the data during iput */
572         iput(inode);
573
574         get_node_info(sbi, ino, &ni);
575
576         /* ENOMEM was fully retried in f2fs_evict_inode. */
577         if (ni.blk_addr != NULL_ADDR) {
578                 set_sbi_flag(sbi, SBI_NEED_FSCK);
579                 f2fs_msg(sbi->sb, KERN_WARNING,
580                         "%s: orphan failed (ino=%x) by kernel, retry mount.",
581                                 __func__, ino);
582                 return -EIO;
583         }
584         __remove_ino_entry(sbi, ino, ORPHAN_INO);
585         return 0;
586 }
587
588 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
589 {
590         block_t start_blk, orphan_blocks, i, j;
591         unsigned int s_flags = sbi->sb->s_flags;
592         int err = 0;
593
594         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
595                 return 0;
596
597         if (s_flags & MS_RDONLY) {
598                 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
599                 sbi->sb->s_flags &= ~MS_RDONLY;
600         }
601
602 #ifdef CONFIG_QUOTA
603         /* Needed for iput() to work correctly and not trash data */
604         sbi->sb->s_flags |= MS_ACTIVE;
605         /* Turn on quotas so that they are updated correctly */
606         f2fs_enable_quota_files(sbi);
607 #endif
608
609         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
610         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
611
612         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
613
614         for (i = 0; i < orphan_blocks; i++) {
615                 struct page *page = get_meta_page(sbi, start_blk + i);
616                 struct f2fs_orphan_block *orphan_blk;
617
618                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
619                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
620                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
621                         err = recover_orphan_inode(sbi, ino);
622                         if (err) {
623                                 f2fs_put_page(page, 1);
624                                 goto out;
625                         }
626                 }
627                 f2fs_put_page(page, 1);
628         }
629         /* clear Orphan Flag */
630         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
631 out:
632 #ifdef CONFIG_QUOTA
633         /* Turn quotas off */
634         f2fs_quota_off_umount(sbi->sb);
635 #endif
636         sbi->sb->s_flags = s_flags; /* Restore MS_RDONLY status */
637
638         return err;
639 }
640
641 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
642 {
643         struct list_head *head;
644         struct f2fs_orphan_block *orphan_blk = NULL;
645         unsigned int nentries = 0;
646         unsigned short index = 1;
647         unsigned short orphan_blocks;
648         struct page *page = NULL;
649         struct ino_entry *orphan = NULL;
650         struct inode_management *im = &sbi->im[ORPHAN_INO];
651
652         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
653
654         /*
655          * we don't need to do spin_lock(&im->ino_lock) here, since all the
656          * orphan inode operations are covered under f2fs_lock_op().
657          * And, spin_lock should be avoided due to page operations below.
658          */
659         head = &im->ino_list;
660
661         /* loop for each orphan inode entry and write them in Jornal block */
662         list_for_each_entry(orphan, head, list) {
663                 if (!page) {
664                         page = grab_meta_page(sbi, start_blk++);
665                         orphan_blk =
666                                 (struct f2fs_orphan_block *)page_address(page);
667                         memset(orphan_blk, 0, sizeof(*orphan_blk));
668                 }
669
670                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
671
672                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
673                         /*
674                          * an orphan block is full of 1020 entries,
675                          * then we need to flush current orphan blocks
676                          * and bring another one in memory
677                          */
678                         orphan_blk->blk_addr = cpu_to_le16(index);
679                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
680                         orphan_blk->entry_count = cpu_to_le32(nentries);
681                         set_page_dirty(page);
682                         f2fs_put_page(page, 1);
683                         index++;
684                         nentries = 0;
685                         page = NULL;
686                 }
687         }
688
689         if (page) {
690                 orphan_blk->blk_addr = cpu_to_le16(index);
691                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
692                 orphan_blk->entry_count = cpu_to_le32(nentries);
693                 set_page_dirty(page);
694                 f2fs_put_page(page, 1);
695         }
696 }
697
698 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
699                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
700                 unsigned long long *version)
701 {
702         unsigned long blk_size = sbi->blocksize;
703         size_t crc_offset = 0;
704         __u32 crc = 0;
705
706         *cp_page = get_meta_page(sbi, cp_addr);
707         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
708
709         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
710         if (crc_offset > (blk_size - sizeof(__le32))) {
711                 f2fs_msg(sbi->sb, KERN_WARNING,
712                         "invalid crc_offset: %zu", crc_offset);
713                 return -EINVAL;
714         }
715
716         crc = cur_cp_crc(*cp_block);
717         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
718                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
719                 return -EINVAL;
720         }
721
722         *version = cur_cp_version(*cp_block);
723         return 0;
724 }
725
726 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
727                                 block_t cp_addr, unsigned long long *version)
728 {
729         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
730         struct f2fs_checkpoint *cp_block = NULL;
731         unsigned long long cur_version = 0, pre_version = 0;
732         int err;
733
734         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
735                                         &cp_page_1, version);
736         if (err)
737                 goto invalid_cp1;
738         pre_version = *version;
739
740         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
741         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
742                                         &cp_page_2, version);
743         if (err)
744                 goto invalid_cp2;
745         cur_version = *version;
746
747         if (cur_version == pre_version) {
748                 *version = cur_version;
749                 f2fs_put_page(cp_page_2, 1);
750                 return cp_page_1;
751         }
752 invalid_cp2:
753         f2fs_put_page(cp_page_2, 1);
754 invalid_cp1:
755         f2fs_put_page(cp_page_1, 1);
756         return NULL;
757 }
758
759 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
760 {
761         struct f2fs_checkpoint *cp_block;
762         struct f2fs_super_block *fsb = sbi->raw_super;
763         struct page *cp1, *cp2, *cur_page;
764         unsigned long blk_size = sbi->blocksize;
765         unsigned long long cp1_version = 0, cp2_version = 0;
766         unsigned long long cp_start_blk_no;
767         unsigned int cp_blks = 1 + __cp_payload(sbi);
768         block_t cp_blk_no;
769         int i;
770
771         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
772         if (!sbi->ckpt)
773                 return -ENOMEM;
774         /*
775          * Finding out valid cp block involves read both
776          * sets( cp pack1 and cp pack 2)
777          */
778         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
779         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
780
781         /* The second checkpoint pack should start at the next segment */
782         cp_start_blk_no += ((unsigned long long)1) <<
783                                 le32_to_cpu(fsb->log_blocks_per_seg);
784         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
785
786         if (cp1 && cp2) {
787                 if (ver_after(cp2_version, cp1_version))
788                         cur_page = cp2;
789                 else
790                         cur_page = cp1;
791         } else if (cp1) {
792                 cur_page = cp1;
793         } else if (cp2) {
794                 cur_page = cp2;
795         } else {
796                 goto fail_no_cp;
797         }
798
799         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
800         memcpy(sbi->ckpt, cp_block, blk_size);
801
802         /* Sanity checking of checkpoint */
803         if (sanity_check_ckpt(sbi))
804                 goto free_fail_no_cp;
805
806         if (cur_page == cp1)
807                 sbi->cur_cp_pack = 1;
808         else
809                 sbi->cur_cp_pack = 2;
810
811         if (cp_blks <= 1)
812                 goto done;
813
814         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
815         if (cur_page == cp2)
816                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
817
818         for (i = 1; i < cp_blks; i++) {
819                 void *sit_bitmap_ptr;
820                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
821
822                 cur_page = get_meta_page(sbi, cp_blk_no + i);
823                 sit_bitmap_ptr = page_address(cur_page);
824                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
825                 f2fs_put_page(cur_page, 1);
826         }
827 done:
828         f2fs_put_page(cp1, 1);
829         f2fs_put_page(cp2, 1);
830         return 0;
831
832 free_fail_no_cp:
833         f2fs_put_page(cp1, 1);
834         f2fs_put_page(cp2, 1);
835 fail_no_cp:
836         kfree(sbi->ckpt);
837         return -EINVAL;
838 }
839
840 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
841 {
842         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
843         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
844
845         if (is_inode_flag_set(inode, flag))
846                 return;
847
848         set_inode_flag(inode, flag);
849         if (!f2fs_is_volatile_file(inode))
850                 list_add_tail(&F2FS_I(inode)->dirty_list,
851                                                 &sbi->inode_list[type]);
852         stat_inc_dirty_inode(sbi, type);
853 }
854
855 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
856 {
857         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
858
859         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
860                 return;
861
862         list_del_init(&F2FS_I(inode)->dirty_list);
863         clear_inode_flag(inode, flag);
864         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
865 }
866
867 void update_dirty_page(struct inode *inode, struct page *page)
868 {
869         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
870         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
871
872         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
873                         !S_ISLNK(inode->i_mode))
874                 return;
875
876         spin_lock(&sbi->inode_lock[type]);
877         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
878                 __add_dirty_inode(inode, type);
879         inode_inc_dirty_pages(inode);
880         spin_unlock(&sbi->inode_lock[type]);
881
882         SetPagePrivate(page);
883         f2fs_trace_pid(page);
884 }
885
886 void remove_dirty_inode(struct inode *inode)
887 {
888         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
889         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
890
891         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
892                         !S_ISLNK(inode->i_mode))
893                 return;
894
895         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
896                 return;
897
898         spin_lock(&sbi->inode_lock[type]);
899         __remove_dirty_inode(inode, type);
900         spin_unlock(&sbi->inode_lock[type]);
901 }
902
903 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
904 {
905         struct list_head *head;
906         struct inode *inode;
907         struct f2fs_inode_info *fi;
908         bool is_dir = (type == DIR_INODE);
909         unsigned long ino = 0;
910
911         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
912                                 get_pages(sbi, is_dir ?
913                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
914 retry:
915         if (unlikely(f2fs_cp_error(sbi)))
916                 return -EIO;
917
918         spin_lock(&sbi->inode_lock[type]);
919
920         head = &sbi->inode_list[type];
921         if (list_empty(head)) {
922                 spin_unlock(&sbi->inode_lock[type]);
923                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
924                                 get_pages(sbi, is_dir ?
925                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
926                 return 0;
927         }
928         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
929         inode = igrab(&fi->vfs_inode);
930         spin_unlock(&sbi->inode_lock[type]);
931         if (inode) {
932                 unsigned long cur_ino = inode->i_ino;
933
934                 if (is_dir)
935                         F2FS_I(inode)->cp_task = current;
936
937                 filemap_fdatawrite(inode->i_mapping);
938
939                 if (is_dir)
940                         F2FS_I(inode)->cp_task = NULL;
941
942                 iput(inode);
943                 /* We need to give cpu to another writers. */
944                 if (ino == cur_ino) {
945                         congestion_wait(BLK_RW_ASYNC, HZ/50);
946                         cond_resched();
947                 } else {
948                         ino = cur_ino;
949                 }
950         } else {
951                 /*
952                  * We should submit bio, since it exists several
953                  * wribacking dentry pages in the freeing inode.
954                  */
955                 f2fs_submit_merged_write(sbi, DATA);
956                 cond_resched();
957         }
958         goto retry;
959 }
960
961 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
962 {
963         struct list_head *head = &sbi->inode_list[DIRTY_META];
964         struct inode *inode;
965         struct f2fs_inode_info *fi;
966         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
967
968         while (total--) {
969                 if (unlikely(f2fs_cp_error(sbi)))
970                         return -EIO;
971
972                 spin_lock(&sbi->inode_lock[DIRTY_META]);
973                 if (list_empty(head)) {
974                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
975                         return 0;
976                 }
977                 fi = list_first_entry(head, struct f2fs_inode_info,
978                                                         gdirty_list);
979                 inode = igrab(&fi->vfs_inode);
980                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
981                 if (inode) {
982                         sync_inode_metadata(inode, 0);
983
984                         /* it's on eviction */
985                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
986                                 update_inode_page(inode);
987                         iput(inode);
988                 }
989         };
990         return 0;
991 }
992
993 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
994 {
995         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
996         struct f2fs_nm_info *nm_i = NM_I(sbi);
997         nid_t last_nid = nm_i->next_scan_nid;
998
999         next_free_nid(sbi, &last_nid);
1000         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1001         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1002         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1003         ckpt->next_free_nid = cpu_to_le32(last_nid);
1004 }
1005
1006 /*
1007  * Freeze all the FS-operations for checkpoint.
1008  */
1009 static int block_operations(struct f2fs_sb_info *sbi)
1010 {
1011         struct writeback_control wbc = {
1012                 .sync_mode = WB_SYNC_ALL,
1013                 .nr_to_write = LONG_MAX,
1014                 .for_reclaim = 0,
1015         };
1016         struct blk_plug plug;
1017         int err = 0;
1018
1019         blk_start_plug(&plug);
1020
1021 retry_flush_dents:
1022         f2fs_lock_all(sbi);
1023         /* write all the dirty dentry pages */
1024         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1025                 f2fs_unlock_all(sbi);
1026                 err = sync_dirty_inodes(sbi, DIR_INODE);
1027                 if (err)
1028                         goto out;
1029                 cond_resched();
1030                 goto retry_flush_dents;
1031         }
1032
1033         /*
1034          * POR: we should ensure that there are no dirty node pages
1035          * until finishing nat/sit flush. inode->i_blocks can be updated.
1036          */
1037         down_write(&sbi->node_change);
1038
1039         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1040                 up_write(&sbi->node_change);
1041                 f2fs_unlock_all(sbi);
1042                 err = f2fs_sync_inode_meta(sbi);
1043                 if (err)
1044                         goto out;
1045                 cond_resched();
1046                 goto retry_flush_dents;
1047         }
1048
1049 retry_flush_nodes:
1050         down_write(&sbi->node_write);
1051
1052         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1053                 up_write(&sbi->node_write);
1054                 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1055                 if (err) {
1056                         up_write(&sbi->node_change);
1057                         f2fs_unlock_all(sbi);
1058                         goto out;
1059                 }
1060                 cond_resched();
1061                 goto retry_flush_nodes;
1062         }
1063
1064         /*
1065          * sbi->node_change is used only for AIO write_begin path which produces
1066          * dirty node blocks and some checkpoint values by block allocation.
1067          */
1068         __prepare_cp_block(sbi);
1069         up_write(&sbi->node_change);
1070 out:
1071         blk_finish_plug(&plug);
1072         return err;
1073 }
1074
1075 static void unblock_operations(struct f2fs_sb_info *sbi)
1076 {
1077         up_write(&sbi->node_write);
1078         f2fs_unlock_all(sbi);
1079 }
1080
1081 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1082 {
1083         DEFINE_WAIT(wait);
1084
1085         for (;;) {
1086                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1087
1088                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1089                         break;
1090
1091                 io_schedule_timeout(5*HZ);
1092         }
1093         finish_wait(&sbi->cp_wait, &wait);
1094 }
1095
1096 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1097 {
1098         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1099         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1100         unsigned long flags;
1101
1102         spin_lock_irqsave(&sbi->cp_lock, flags);
1103
1104         if ((cpc->reason & CP_UMOUNT) &&
1105                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1106                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1107                 disable_nat_bits(sbi, false);
1108
1109         if (cpc->reason & CP_TRIMMED)
1110                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1111
1112         if (cpc->reason & CP_UMOUNT)
1113                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1114         else
1115                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1116
1117         if (cpc->reason & CP_FASTBOOT)
1118                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1119         else
1120                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1121
1122         if (orphan_num)
1123                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1124         else
1125                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1126
1127         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1128                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1129
1130         /* set this flag to activate crc|cp_ver for recovery */
1131         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1132
1133         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1134 }
1135
1136 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1137 {
1138         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1139         struct f2fs_nm_info *nm_i = NM_I(sbi);
1140         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1141         block_t start_blk;
1142         unsigned int data_sum_blocks, orphan_blocks;
1143         __u32 crc32 = 0;
1144         int i;
1145         int cp_payload_blks = __cp_payload(sbi);
1146         struct super_block *sb = sbi->sb;
1147         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1148         u64 kbytes_written;
1149
1150         /* Flush all the NAT/SIT pages */
1151         while (get_pages(sbi, F2FS_DIRTY_META)) {
1152                 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1153                 if (unlikely(f2fs_cp_error(sbi)))
1154                         return -EIO;
1155         }
1156
1157         /*
1158          * modify checkpoint
1159          * version number is already updated
1160          */
1161         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1162         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1163         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1164                 ckpt->cur_node_segno[i] =
1165                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1166                 ckpt->cur_node_blkoff[i] =
1167                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1168                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1169                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1170         }
1171         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1172                 ckpt->cur_data_segno[i] =
1173                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1174                 ckpt->cur_data_blkoff[i] =
1175                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1176                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1177                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1178         }
1179
1180         /* 2 cp  + n data seg summary + orphan inode blocks */
1181         data_sum_blocks = npages_for_summary_flush(sbi, false);
1182         spin_lock_irqsave(&sbi->cp_lock, flags);
1183         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1184                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1185         else
1186                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1187         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1188
1189         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1190         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1191                         orphan_blocks);
1192
1193         if (__remain_node_summaries(cpc->reason))
1194                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1195                                 cp_payload_blks + data_sum_blocks +
1196                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1197         else
1198                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1199                                 cp_payload_blks + data_sum_blocks +
1200                                 orphan_blocks);
1201
1202         /* update ckpt flag for checkpoint */
1203         update_ckpt_flags(sbi, cpc);
1204
1205         /* update SIT/NAT bitmap */
1206         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1207         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1208
1209         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1210         *((__le32 *)((unsigned char *)ckpt +
1211                                 le32_to_cpu(ckpt->checksum_offset)))
1212                                 = cpu_to_le32(crc32);
1213
1214         start_blk = __start_cp_next_addr(sbi);
1215
1216         /* write nat bits */
1217         if (enabled_nat_bits(sbi, cpc)) {
1218                 __u64 cp_ver = cur_cp_version(ckpt);
1219                 block_t blk;
1220
1221                 cp_ver |= ((__u64)crc32 << 32);
1222                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1223
1224                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1225                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1226                         update_meta_page(sbi, nm_i->nat_bits +
1227                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1228
1229                 /* Flush all the NAT BITS pages */
1230                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1231                         sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1232                         if (unlikely(f2fs_cp_error(sbi)))
1233                                 return -EIO;
1234                 }
1235         }
1236
1237         /* need to wait for end_io results */
1238         wait_on_all_pages_writeback(sbi);
1239         if (unlikely(f2fs_cp_error(sbi)))
1240                 return -EIO;
1241
1242         /* write out checkpoint buffer at block 0 */
1243         update_meta_page(sbi, ckpt, start_blk++);
1244
1245         for (i = 1; i < 1 + cp_payload_blks; i++)
1246                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1247                                                         start_blk++);
1248
1249         if (orphan_num) {
1250                 write_orphan_inodes(sbi, start_blk);
1251                 start_blk += orphan_blocks;
1252         }
1253
1254         write_data_summaries(sbi, start_blk);
1255         start_blk += data_sum_blocks;
1256
1257         /* Record write statistics in the hot node summary */
1258         kbytes_written = sbi->kbytes_written;
1259         if (sb->s_bdev->bd_part)
1260                 kbytes_written += BD_PART_WRITTEN(sbi);
1261
1262         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1263
1264         if (__remain_node_summaries(cpc->reason)) {
1265                 write_node_summaries(sbi, start_blk);
1266                 start_blk += NR_CURSEG_NODE_TYPE;
1267         }
1268
1269         /* writeout checkpoint block */
1270         update_meta_page(sbi, ckpt, start_blk);
1271
1272         /* wait for previous submitted node/meta pages writeback */
1273         wait_on_all_pages_writeback(sbi);
1274
1275         if (unlikely(f2fs_cp_error(sbi)))
1276                 return -EIO;
1277
1278         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1279         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1280
1281         /* update user_block_counts */
1282         sbi->last_valid_block_count = sbi->total_valid_block_count;
1283         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1284
1285         /* Here, we only have one bio having CP pack */
1286         sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1287
1288         /* wait for previous submitted meta pages writeback */
1289         wait_on_all_pages_writeback(sbi);
1290
1291         release_ino_entry(sbi, false);
1292
1293         if (unlikely(f2fs_cp_error(sbi)))
1294                 return -EIO;
1295
1296         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1297         clear_sbi_flag(sbi, SBI_NEED_CP);
1298         __set_cp_next_pack(sbi);
1299
1300         /*
1301          * redirty superblock if metadata like node page or inode cache is
1302          * updated during writing checkpoint.
1303          */
1304         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1305                         get_pages(sbi, F2FS_DIRTY_IMETA))
1306                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1307
1308         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1309
1310         return 0;
1311 }
1312
1313 /*
1314  * We guarantee that this checkpoint procedure will not fail.
1315  */
1316 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1317 {
1318         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1319         unsigned long long ckpt_ver;
1320         int err = 0;
1321
1322         mutex_lock(&sbi->cp_mutex);
1323
1324         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1325                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1326                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1327                 goto out;
1328         if (unlikely(f2fs_cp_error(sbi))) {
1329                 err = -EIO;
1330                 goto out;
1331         }
1332         if (f2fs_readonly(sbi->sb)) {
1333                 err = -EROFS;
1334                 goto out;
1335         }
1336
1337         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1338
1339         err = block_operations(sbi);
1340         if (err)
1341                 goto out;
1342
1343         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1344
1345         f2fs_flush_merged_writes(sbi);
1346
1347         /* this is the case of multiple fstrims without any changes */
1348         if (cpc->reason & CP_DISCARD) {
1349                 if (!exist_trim_candidates(sbi, cpc)) {
1350                         unblock_operations(sbi);
1351                         goto out;
1352                 }
1353
1354                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1355                                 SIT_I(sbi)->dirty_sentries == 0 &&
1356                                 prefree_segments(sbi) == 0) {
1357                         flush_sit_entries(sbi, cpc);
1358                         clear_prefree_segments(sbi, cpc);
1359                         unblock_operations(sbi);
1360                         goto out;
1361                 }
1362         }
1363
1364         /*
1365          * update checkpoint pack index
1366          * Increase the version number so that
1367          * SIT entries and seg summaries are written at correct place
1368          */
1369         ckpt_ver = cur_cp_version(ckpt);
1370         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1371
1372         /* write cached NAT/SIT entries to NAT/SIT area */
1373         flush_nat_entries(sbi, cpc);
1374         flush_sit_entries(sbi, cpc);
1375
1376         /* unlock all the fs_lock[] in do_checkpoint() */
1377         err = do_checkpoint(sbi, cpc);
1378         if (err)
1379                 release_discard_addrs(sbi);
1380         else
1381                 clear_prefree_segments(sbi, cpc);
1382
1383         unblock_operations(sbi);
1384         stat_inc_cp_count(sbi->stat_info);
1385
1386         if (cpc->reason & CP_RECOVERY)
1387                 f2fs_msg(sbi->sb, KERN_NOTICE,
1388                         "checkpoint: version = %llx", ckpt_ver);
1389
1390         /* do checkpoint periodically */
1391         f2fs_update_time(sbi, CP_TIME);
1392         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1393 out:
1394         mutex_unlock(&sbi->cp_mutex);
1395         return err;
1396 }
1397
1398 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1399 {
1400         int i;
1401
1402         for (i = 0; i < MAX_INO_ENTRY; i++) {
1403                 struct inode_management *im = &sbi->im[i];
1404
1405                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1406                 spin_lock_init(&im->ino_lock);
1407                 INIT_LIST_HEAD(&im->ino_list);
1408                 im->ino_num = 0;
1409         }
1410
1411         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1412                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1413                                 F2FS_ORPHANS_PER_BLOCK;
1414 }
1415
1416 int __init create_checkpoint_caches(void)
1417 {
1418         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1419                         sizeof(struct ino_entry));
1420         if (!ino_entry_slab)
1421                 return -ENOMEM;
1422         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1423                         sizeof(struct inode_entry));
1424         if (!inode_entry_slab) {
1425                 kmem_cache_destroy(ino_entry_slab);
1426                 return -ENOMEM;
1427         }
1428         return 0;
1429 }
1430
1431 void destroy_checkpoint_caches(void)
1432 {
1433         kmem_cache_destroy(ino_entry_slab);
1434         kmem_cache_destroy(inode_entry_slab);
1435 }