Merge tag 'for-f2fs-4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk...
[cascardo/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.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->ckpt, CP_ERROR_FLAG);
32         sbi->sb->s_flags |= MS_RDONLY;
33         if (!end_io)
34                 f2fs_flush_merged_bios(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 = READ_SYNC | 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 ? (READ_SYNC | REQ_META | REQ_PRIO) : REQ_RAHEAD,
164                 .encrypted_page = NULL,
165         };
166         struct blk_plug plug;
167
168         if (unlikely(type == META_POR))
169                 fio.op_flags &= ~REQ_META;
170
171         blk_start_plug(&plug);
172         for (; nrpages-- > 0; blkno++) {
173
174                 if (!is_valid_blkaddr(sbi, blkno, type))
175                         goto out;
176
177                 switch (type) {
178                 case META_NAT:
179                         if (unlikely(blkno >=
180                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
181                                 blkno = 0;
182                         /* get nat block addr */
183                         fio.new_blkaddr = current_nat_addr(sbi,
184                                         blkno * NAT_ENTRY_PER_BLOCK);
185                         break;
186                 case META_SIT:
187                         /* get sit block addr */
188                         fio.new_blkaddr = current_sit_addr(sbi,
189                                         blkno * SIT_ENTRY_PER_BLOCK);
190                         break;
191                 case META_SSA:
192                 case META_CP:
193                 case META_POR:
194                         fio.new_blkaddr = blkno;
195                         break;
196                 default:
197                         BUG();
198                 }
199
200                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
201                                                 fio.new_blkaddr, false);
202                 if (!page)
203                         continue;
204                 if (PageUptodate(page)) {
205                         f2fs_put_page(page, 1);
206                         continue;
207                 }
208
209                 fio.page = page;
210                 fio.old_blkaddr = fio.new_blkaddr;
211                 f2fs_submit_page_mbio(&fio);
212                 f2fs_put_page(page, 0);
213         }
214 out:
215         f2fs_submit_merged_bio(sbi, META, READ);
216         blk_finish_plug(&plug);
217         return blkno - start;
218 }
219
220 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
221 {
222         struct page *page;
223         bool readahead = false;
224
225         page = find_get_page(META_MAPPING(sbi), index);
226         if (!page || !PageUptodate(page))
227                 readahead = true;
228         f2fs_put_page(page, 0);
229
230         if (readahead)
231                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
232 }
233
234 static int f2fs_write_meta_page(struct page *page,
235                                 struct writeback_control *wbc)
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);
249         dec_page_count(sbi, F2FS_DIRTY_META);
250
251         if (wbc->for_reclaim)
252                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
253
254         unlock_page(page);
255
256         if (unlikely(f2fs_cp_error(sbi)))
257                 f2fs_submit_merged_bio(sbi, META, WRITE);
258
259         return 0;
260
261 redirty_out:
262         redirty_page_for_writepage(wbc, page);
263         return AOP_WRITEPAGE_ACTIVATE;
264 }
265
266 static int f2fs_write_meta_pages(struct address_space *mapping,
267                                 struct writeback_control *wbc)
268 {
269         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
270         struct blk_plug plug;
271         long diff, written;
272
273         /* collect a number of dirty meta pages and write together */
274         if (wbc->for_kupdate ||
275                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
276                 goto skip_write;
277
278         trace_f2fs_writepages(mapping->host, wbc, META);
279
280         /* if mounting is failed, skip writing node pages */
281         mutex_lock(&sbi->cp_mutex);
282         diff = nr_pages_to_write(sbi, META, wbc);
283         blk_start_plug(&plug);
284         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
285         blk_finish_plug(&plug);
286         mutex_unlock(&sbi->cp_mutex);
287         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
288         return 0;
289
290 skip_write:
291         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
292         trace_f2fs_writepages(mapping->host, wbc, META);
293         return 0;
294 }
295
296 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
297                                                 long nr_to_write)
298 {
299         struct address_space *mapping = META_MAPPING(sbi);
300         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
301         struct pagevec pvec;
302         long nwritten = 0;
303         struct writeback_control wbc = {
304                 .for_reclaim = 0,
305         };
306         struct blk_plug plug;
307
308         pagevec_init(&pvec, 0);
309
310         blk_start_plug(&plug);
311
312         while (index <= end) {
313                 int i, nr_pages;
314                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
315                                 PAGECACHE_TAG_DIRTY,
316                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
317                 if (unlikely(nr_pages == 0))
318                         break;
319
320                 for (i = 0; i < nr_pages; i++) {
321                         struct page *page = pvec.pages[i];
322
323                         if (prev == ULONG_MAX)
324                                 prev = page->index - 1;
325                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
326                                 pagevec_release(&pvec);
327                                 goto stop;
328                         }
329
330                         lock_page(page);
331
332                         if (unlikely(page->mapping != mapping)) {
333 continue_unlock:
334                                 unlock_page(page);
335                                 continue;
336                         }
337                         if (!PageDirty(page)) {
338                                 /* someone wrote it for us */
339                                 goto continue_unlock;
340                         }
341
342                         f2fs_wait_on_page_writeback(page, META, true);
343
344                         BUG_ON(PageWriteback(page));
345                         if (!clear_page_dirty_for_io(page))
346                                 goto continue_unlock;
347
348                         if (mapping->a_ops->writepage(page, &wbc)) {
349                                 unlock_page(page);
350                                 break;
351                         }
352                         nwritten++;
353                         prev = page->index;
354                         if (unlikely(nwritten >= nr_to_write))
355                                 break;
356                 }
357                 pagevec_release(&pvec);
358                 cond_resched();
359         }
360 stop:
361         if (nwritten)
362                 f2fs_submit_merged_bio(sbi, type, WRITE);
363
364         blk_finish_plug(&plug);
365
366         return nwritten;
367 }
368
369 static int f2fs_set_meta_page_dirty(struct page *page)
370 {
371         trace_f2fs_set_page_dirty(page, META);
372
373         if (!PageUptodate(page))
374                 SetPageUptodate(page);
375         if (!PageDirty(page)) {
376                 f2fs_set_page_dirty_nobuffers(page);
377                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
378                 SetPagePrivate(page);
379                 f2fs_trace_pid(page);
380                 return 1;
381         }
382         return 0;
383 }
384
385 const struct address_space_operations f2fs_meta_aops = {
386         .writepage      = f2fs_write_meta_page,
387         .writepages     = f2fs_write_meta_pages,
388         .set_page_dirty = f2fs_set_meta_page_dirty,
389         .invalidatepage = f2fs_invalidate_page,
390         .releasepage    = f2fs_release_page,
391 };
392
393 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
394 {
395         struct inode_management *im = &sbi->im[type];
396         struct ino_entry *e, *tmp;
397
398         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
399 retry:
400         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
401
402         spin_lock(&im->ino_lock);
403         e = radix_tree_lookup(&im->ino_root, ino);
404         if (!e) {
405                 e = tmp;
406                 if (radix_tree_insert(&im->ino_root, ino, e)) {
407                         spin_unlock(&im->ino_lock);
408                         radix_tree_preload_end();
409                         goto retry;
410                 }
411                 memset(e, 0, sizeof(struct ino_entry));
412                 e->ino = ino;
413
414                 list_add_tail(&e->list, &im->ino_list);
415                 if (type != ORPHAN_INO)
416                         im->ino_num++;
417         }
418         spin_unlock(&im->ino_lock);
419         radix_tree_preload_end();
420
421         if (e != tmp)
422                 kmem_cache_free(ino_entry_slab, tmp);
423 }
424
425 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
426 {
427         struct inode_management *im = &sbi->im[type];
428         struct ino_entry *e;
429
430         spin_lock(&im->ino_lock);
431         e = radix_tree_lookup(&im->ino_root, ino);
432         if (e) {
433                 list_del(&e->list);
434                 radix_tree_delete(&im->ino_root, ino);
435                 im->ino_num--;
436                 spin_unlock(&im->ino_lock);
437                 kmem_cache_free(ino_entry_slab, e);
438                 return;
439         }
440         spin_unlock(&im->ino_lock);
441 }
442
443 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
444 {
445         /* add new dirty ino entry into list */
446         __add_ino_entry(sbi, ino, type);
447 }
448
449 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
450 {
451         /* remove dirty ino entry from list */
452         __remove_ino_entry(sbi, ino, type);
453 }
454
455 /* mode should be APPEND_INO or UPDATE_INO */
456 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
457 {
458         struct inode_management *im = &sbi->im[mode];
459         struct ino_entry *e;
460
461         spin_lock(&im->ino_lock);
462         e = radix_tree_lookup(&im->ino_root, ino);
463         spin_unlock(&im->ino_lock);
464         return e ? true : false;
465 }
466
467 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
468 {
469         struct ino_entry *e, *tmp;
470         int i;
471
472         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
473                 struct inode_management *im = &sbi->im[i];
474
475                 spin_lock(&im->ino_lock);
476                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
477                         list_del(&e->list);
478                         radix_tree_delete(&im->ino_root, e->ino);
479                         kmem_cache_free(ino_entry_slab, e);
480                         im->ino_num--;
481                 }
482                 spin_unlock(&im->ino_lock);
483         }
484 }
485
486 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
487 {
488         struct inode_management *im = &sbi->im[ORPHAN_INO];
489         int err = 0;
490
491         spin_lock(&im->ino_lock);
492
493 #ifdef CONFIG_F2FS_FAULT_INJECTION
494         if (time_to_inject(FAULT_ORPHAN)) {
495                 spin_unlock(&im->ino_lock);
496                 return -ENOSPC;
497         }
498 #endif
499         if (unlikely(im->ino_num >= sbi->max_orphans))
500                 err = -ENOSPC;
501         else
502                 im->ino_num++;
503         spin_unlock(&im->ino_lock);
504
505         return err;
506 }
507
508 void release_orphan_inode(struct f2fs_sb_info *sbi)
509 {
510         struct inode_management *im = &sbi->im[ORPHAN_INO];
511
512         spin_lock(&im->ino_lock);
513         f2fs_bug_on(sbi, im->ino_num == 0);
514         im->ino_num--;
515         spin_unlock(&im->ino_lock);
516 }
517
518 void add_orphan_inode(struct inode *inode)
519 {
520         /* add new orphan ino entry into list */
521         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
522         update_inode_page(inode);
523 }
524
525 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
526 {
527         /* remove orphan entry from orphan list */
528         __remove_ino_entry(sbi, ino, ORPHAN_INO);
529 }
530
531 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
532 {
533         struct inode *inode;
534
535         inode = f2fs_iget(sbi->sb, ino);
536         if (IS_ERR(inode)) {
537                 /*
538                  * there should be a bug that we can't find the entry
539                  * to orphan inode.
540                  */
541                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
542                 return PTR_ERR(inode);
543         }
544
545         clear_nlink(inode);
546
547         /* truncate all the data during iput */
548         iput(inode);
549         return 0;
550 }
551
552 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
553 {
554         block_t start_blk, orphan_blocks, i, j;
555         int err;
556
557         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
558                 return 0;
559
560         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
561         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
562
563         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
564
565         for (i = 0; i < orphan_blocks; i++) {
566                 struct page *page = get_meta_page(sbi, start_blk + i);
567                 struct f2fs_orphan_block *orphan_blk;
568
569                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
570                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
571                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
572                         err = recover_orphan_inode(sbi, ino);
573                         if (err) {
574                                 f2fs_put_page(page, 1);
575                                 return err;
576                         }
577                 }
578                 f2fs_put_page(page, 1);
579         }
580         /* clear Orphan Flag */
581         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
582         return 0;
583 }
584
585 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
586 {
587         struct list_head *head;
588         struct f2fs_orphan_block *orphan_blk = NULL;
589         unsigned int nentries = 0;
590         unsigned short index = 1;
591         unsigned short orphan_blocks;
592         struct page *page = NULL;
593         struct ino_entry *orphan = NULL;
594         struct inode_management *im = &sbi->im[ORPHAN_INO];
595
596         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
597
598         /*
599          * we don't need to do spin_lock(&im->ino_lock) here, since all the
600          * orphan inode operations are covered under f2fs_lock_op().
601          * And, spin_lock should be avoided due to page operations below.
602          */
603         head = &im->ino_list;
604
605         /* loop for each orphan inode entry and write them in Jornal block */
606         list_for_each_entry(orphan, head, list) {
607                 if (!page) {
608                         page = grab_meta_page(sbi, start_blk++);
609                         orphan_blk =
610                                 (struct f2fs_orphan_block *)page_address(page);
611                         memset(orphan_blk, 0, sizeof(*orphan_blk));
612                 }
613
614                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
615
616                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
617                         /*
618                          * an orphan block is full of 1020 entries,
619                          * then we need to flush current orphan blocks
620                          * and bring another one in memory
621                          */
622                         orphan_blk->blk_addr = cpu_to_le16(index);
623                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
624                         orphan_blk->entry_count = cpu_to_le32(nentries);
625                         set_page_dirty(page);
626                         f2fs_put_page(page, 1);
627                         index++;
628                         nentries = 0;
629                         page = NULL;
630                 }
631         }
632
633         if (page) {
634                 orphan_blk->blk_addr = cpu_to_le16(index);
635                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
636                 orphan_blk->entry_count = cpu_to_le32(nentries);
637                 set_page_dirty(page);
638                 f2fs_put_page(page, 1);
639         }
640 }
641
642 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
643                                 block_t cp_addr, unsigned long long *version)
644 {
645         struct page *cp_page_1, *cp_page_2 = NULL;
646         unsigned long blk_size = sbi->blocksize;
647         struct f2fs_checkpoint *cp_block;
648         unsigned long long cur_version = 0, pre_version = 0;
649         size_t crc_offset;
650         __u32 crc = 0;
651
652         /* Read the 1st cp block in this CP pack */
653         cp_page_1 = get_meta_page(sbi, cp_addr);
654
655         /* get the version number */
656         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
657         crc_offset = le32_to_cpu(cp_block->checksum_offset);
658         if (crc_offset >= blk_size)
659                 goto invalid_cp1;
660
661         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
662         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
663                 goto invalid_cp1;
664
665         pre_version = cur_cp_version(cp_block);
666
667         /* Read the 2nd cp block in this CP pack */
668         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
669         cp_page_2 = get_meta_page(sbi, cp_addr);
670
671         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
672         crc_offset = le32_to_cpu(cp_block->checksum_offset);
673         if (crc_offset >= blk_size)
674                 goto invalid_cp2;
675
676         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
677         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
678                 goto invalid_cp2;
679
680         cur_version = cur_cp_version(cp_block);
681
682         if (cur_version == pre_version) {
683                 *version = cur_version;
684                 f2fs_put_page(cp_page_2, 1);
685                 return cp_page_1;
686         }
687 invalid_cp2:
688         f2fs_put_page(cp_page_2, 1);
689 invalid_cp1:
690         f2fs_put_page(cp_page_1, 1);
691         return NULL;
692 }
693
694 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
695 {
696         struct f2fs_checkpoint *cp_block;
697         struct f2fs_super_block *fsb = sbi->raw_super;
698         struct page *cp1, *cp2, *cur_page;
699         unsigned long blk_size = sbi->blocksize;
700         unsigned long long cp1_version = 0, cp2_version = 0;
701         unsigned long long cp_start_blk_no;
702         unsigned int cp_blks = 1 + __cp_payload(sbi);
703         block_t cp_blk_no;
704         int i;
705
706         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
707         if (!sbi->ckpt)
708                 return -ENOMEM;
709         /*
710          * Finding out valid cp block involves read both
711          * sets( cp pack1 and cp pack 2)
712          */
713         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
714         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
715
716         /* The second checkpoint pack should start at the next segment */
717         cp_start_blk_no += ((unsigned long long)1) <<
718                                 le32_to_cpu(fsb->log_blocks_per_seg);
719         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
720
721         if (cp1 && cp2) {
722                 if (ver_after(cp2_version, cp1_version))
723                         cur_page = cp2;
724                 else
725                         cur_page = cp1;
726         } else if (cp1) {
727                 cur_page = cp1;
728         } else if (cp2) {
729                 cur_page = cp2;
730         } else {
731                 goto fail_no_cp;
732         }
733
734         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
735         memcpy(sbi->ckpt, cp_block, blk_size);
736
737         /* Sanity checking of checkpoint */
738         if (sanity_check_ckpt(sbi))
739                 goto fail_no_cp;
740
741         if (cp_blks <= 1)
742                 goto done;
743
744         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
745         if (cur_page == cp2)
746                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
747
748         for (i = 1; i < cp_blks; i++) {
749                 void *sit_bitmap_ptr;
750                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
751
752                 cur_page = get_meta_page(sbi, cp_blk_no + i);
753                 sit_bitmap_ptr = page_address(cur_page);
754                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
755                 f2fs_put_page(cur_page, 1);
756         }
757 done:
758         f2fs_put_page(cp1, 1);
759         f2fs_put_page(cp2, 1);
760         return 0;
761
762 fail_no_cp:
763         kfree(sbi->ckpt);
764         return -EINVAL;
765 }
766
767 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
768 {
769         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
770         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
771
772         if (is_inode_flag_set(inode, flag))
773                 return;
774
775         set_inode_flag(inode, flag);
776         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
777         stat_inc_dirty_inode(sbi, type);
778 }
779
780 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
781 {
782         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
783
784         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
785                 return;
786
787         list_del_init(&F2FS_I(inode)->dirty_list);
788         clear_inode_flag(inode, flag);
789         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
790 }
791
792 void update_dirty_page(struct inode *inode, struct page *page)
793 {
794         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
795         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
796
797         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
798                         !S_ISLNK(inode->i_mode))
799                 return;
800
801         spin_lock(&sbi->inode_lock[type]);
802         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
803                 __add_dirty_inode(inode, type);
804         inode_inc_dirty_pages(inode);
805         spin_unlock(&sbi->inode_lock[type]);
806
807         SetPagePrivate(page);
808         f2fs_trace_pid(page);
809 }
810
811 void remove_dirty_inode(struct inode *inode)
812 {
813         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
814         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
815
816         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
817                         !S_ISLNK(inode->i_mode))
818                 return;
819
820         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
821                 return;
822
823         spin_lock(&sbi->inode_lock[type]);
824         __remove_dirty_inode(inode, type);
825         spin_unlock(&sbi->inode_lock[type]);
826 }
827
828 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
829 {
830         struct list_head *head;
831         struct inode *inode;
832         struct f2fs_inode_info *fi;
833         bool is_dir = (type == DIR_INODE);
834
835         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
836                                 get_pages(sbi, is_dir ?
837                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
838 retry:
839         if (unlikely(f2fs_cp_error(sbi)))
840                 return -EIO;
841
842         spin_lock(&sbi->inode_lock[type]);
843
844         head = &sbi->inode_list[type];
845         if (list_empty(head)) {
846                 spin_unlock(&sbi->inode_lock[type]);
847                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
848                                 get_pages(sbi, is_dir ?
849                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
850                 return 0;
851         }
852         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
853         inode = igrab(&fi->vfs_inode);
854         spin_unlock(&sbi->inode_lock[type]);
855         if (inode) {
856                 filemap_fdatawrite(inode->i_mapping);
857                 iput(inode);
858         } else {
859                 /*
860                  * We should submit bio, since it exists several
861                  * wribacking dentry pages in the freeing inode.
862                  */
863                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
864                 cond_resched();
865         }
866         goto retry;
867 }
868
869 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
870 {
871         struct list_head *head = &sbi->inode_list[DIRTY_META];
872         struct inode *inode;
873         struct f2fs_inode_info *fi;
874         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
875
876         while (total--) {
877                 if (unlikely(f2fs_cp_error(sbi)))
878                         return -EIO;
879
880                 spin_lock(&sbi->inode_lock[DIRTY_META]);
881                 if (list_empty(head)) {
882                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
883                         return 0;
884                 }
885                 fi = list_entry(head->next, struct f2fs_inode_info,
886                                                         gdirty_list);
887                 inode = igrab(&fi->vfs_inode);
888                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
889                 if (inode) {
890                         update_inode_page(inode);
891                         iput(inode);
892                 }
893         };
894         return 0;
895 }
896
897 /*
898  * Freeze all the FS-operations for checkpoint.
899  */
900 static int block_operations(struct f2fs_sb_info *sbi)
901 {
902         struct writeback_control wbc = {
903                 .sync_mode = WB_SYNC_ALL,
904                 .nr_to_write = LONG_MAX,
905                 .for_reclaim = 0,
906         };
907         struct blk_plug plug;
908         int err = 0;
909
910         blk_start_plug(&plug);
911
912 retry_flush_dents:
913         f2fs_lock_all(sbi);
914         /* write all the dirty dentry pages */
915         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
916                 f2fs_unlock_all(sbi);
917                 err = sync_dirty_inodes(sbi, DIR_INODE);
918                 if (err)
919                         goto out;
920                 goto retry_flush_dents;
921         }
922
923         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
924                 f2fs_unlock_all(sbi);
925                 err = f2fs_sync_inode_meta(sbi);
926                 if (err)
927                         goto out;
928                 goto retry_flush_dents;
929         }
930
931         /*
932          * POR: we should ensure that there are no dirty node pages
933          * until finishing nat/sit flush.
934          */
935 retry_flush_nodes:
936         down_write(&sbi->node_write);
937
938         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
939                 up_write(&sbi->node_write);
940                 err = sync_node_pages(sbi, &wbc);
941                 if (err) {
942                         f2fs_unlock_all(sbi);
943                         goto out;
944                 }
945                 goto retry_flush_nodes;
946         }
947 out:
948         blk_finish_plug(&plug);
949         return err;
950 }
951
952 static void unblock_operations(struct f2fs_sb_info *sbi)
953 {
954         up_write(&sbi->node_write);
955
956         build_free_nids(sbi);
957         f2fs_unlock_all(sbi);
958 }
959
960 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
961 {
962         DEFINE_WAIT(wait);
963
964         for (;;) {
965                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
966
967                 if (!atomic_read(&sbi->nr_wb_bios))
968                         break;
969
970                 io_schedule_timeout(5*HZ);
971         }
972         finish_wait(&sbi->cp_wait, &wait);
973 }
974
975 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
976 {
977         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
978         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
979         struct f2fs_nm_info *nm_i = NM_I(sbi);
980         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
981         nid_t last_nid = nm_i->next_scan_nid;
982         block_t start_blk;
983         unsigned int data_sum_blocks, orphan_blocks;
984         __u32 crc32 = 0;
985         int i;
986         int cp_payload_blks = __cp_payload(sbi);
987         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
988         bool invalidate = false;
989         struct super_block *sb = sbi->sb;
990         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
991         u64 kbytes_written;
992
993         /*
994          * This avoids to conduct wrong roll-forward operations and uses
995          * metapages, so should be called prior to sync_meta_pages below.
996          */
997         if (!test_opt(sbi, LFS) && discard_next_dnode(sbi, discard_blk))
998                 invalidate = true;
999
1000         /* Flush all the NAT/SIT pages */
1001         while (get_pages(sbi, F2FS_DIRTY_META)) {
1002                 sync_meta_pages(sbi, META, LONG_MAX);
1003                 if (unlikely(f2fs_cp_error(sbi)))
1004                         return -EIO;
1005         }
1006
1007         next_free_nid(sbi, &last_nid);
1008
1009         /*
1010          * modify checkpoint
1011          * version number is already updated
1012          */
1013         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1014         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1015         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1016         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1017                 ckpt->cur_node_segno[i] =
1018                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1019                 ckpt->cur_node_blkoff[i] =
1020                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1021                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1022                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1023         }
1024         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1025                 ckpt->cur_data_segno[i] =
1026                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1027                 ckpt->cur_data_blkoff[i] =
1028                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1029                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1030                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1031         }
1032
1033         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1034         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1035         ckpt->next_free_nid = cpu_to_le32(last_nid);
1036
1037         /* 2 cp  + n data seg summary + orphan inode blocks */
1038         data_sum_blocks = npages_for_summary_flush(sbi, false);
1039         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1040                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1041         else
1042                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1043
1044         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1045         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1046                         orphan_blocks);
1047
1048         if (__remain_node_summaries(cpc->reason))
1049                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1050                                 cp_payload_blks + data_sum_blocks +
1051                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1052         else
1053                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1054                                 cp_payload_blks + data_sum_blocks +
1055                                 orphan_blocks);
1056
1057         if (cpc->reason == CP_UMOUNT)
1058                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1059         else
1060                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1061
1062         if (cpc->reason == CP_FASTBOOT)
1063                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1064         else
1065                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1066
1067         if (orphan_num)
1068                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1069         else
1070                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1071
1072         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1073                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1074
1075         /* update SIT/NAT bitmap */
1076         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1077         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1078
1079         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1080         *((__le32 *)((unsigned char *)ckpt +
1081                                 le32_to_cpu(ckpt->checksum_offset)))
1082                                 = cpu_to_le32(crc32);
1083
1084         start_blk = __start_cp_addr(sbi);
1085
1086         /* need to wait for end_io results */
1087         wait_on_all_pages_writeback(sbi);
1088         if (unlikely(f2fs_cp_error(sbi)))
1089                 return -EIO;
1090
1091         /* write out checkpoint buffer at block 0 */
1092         update_meta_page(sbi, ckpt, start_blk++);
1093
1094         for (i = 1; i < 1 + cp_payload_blks; i++)
1095                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1096                                                         start_blk++);
1097
1098         if (orphan_num) {
1099                 write_orphan_inodes(sbi, start_blk);
1100                 start_blk += orphan_blocks;
1101         }
1102
1103         write_data_summaries(sbi, start_blk);
1104         start_blk += data_sum_blocks;
1105
1106         /* Record write statistics in the hot node summary */
1107         kbytes_written = sbi->kbytes_written;
1108         if (sb->s_bdev->bd_part)
1109                 kbytes_written += BD_PART_WRITTEN(sbi);
1110
1111         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1112
1113         if (__remain_node_summaries(cpc->reason)) {
1114                 write_node_summaries(sbi, start_blk);
1115                 start_blk += NR_CURSEG_NODE_TYPE;
1116         }
1117
1118         /* writeout checkpoint block */
1119         update_meta_page(sbi, ckpt, start_blk);
1120
1121         /* wait for previous submitted node/meta pages writeback */
1122         wait_on_all_pages_writeback(sbi);
1123
1124         if (unlikely(f2fs_cp_error(sbi)))
1125                 return -EIO;
1126
1127         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1128         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1129
1130         /* update user_block_counts */
1131         sbi->last_valid_block_count = sbi->total_valid_block_count;
1132         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1133
1134         /* Here, we only have one bio having CP pack */
1135         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1136
1137         /* wait for previous submitted meta pages writeback */
1138         wait_on_all_pages_writeback(sbi);
1139
1140         /*
1141          * invalidate meta page which is used temporarily for zeroing out
1142          * block at the end of warm node chain.
1143          */
1144         if (invalidate)
1145                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1146                                                                 discard_blk);
1147
1148         release_ino_entry(sbi, false);
1149
1150         if (unlikely(f2fs_cp_error(sbi)))
1151                 return -EIO;
1152
1153         clear_prefree_segments(sbi, cpc);
1154         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1155
1156         return 0;
1157 }
1158
1159 /*
1160  * We guarantee that this checkpoint procedure will not fail.
1161  */
1162 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1163 {
1164         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1165         unsigned long long ckpt_ver;
1166         int err = 0;
1167
1168         mutex_lock(&sbi->cp_mutex);
1169
1170         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1171                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1172                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1173                 goto out;
1174         if (unlikely(f2fs_cp_error(sbi))) {
1175                 err = -EIO;
1176                 goto out;
1177         }
1178         if (f2fs_readonly(sbi->sb)) {
1179                 err = -EROFS;
1180                 goto out;
1181         }
1182
1183         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1184
1185         err = block_operations(sbi);
1186         if (err)
1187                 goto out;
1188
1189         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1190
1191         f2fs_flush_merged_bios(sbi);
1192
1193         /*
1194          * update checkpoint pack index
1195          * Increase the version number so that
1196          * SIT entries and seg summaries are written at correct place
1197          */
1198         ckpt_ver = cur_cp_version(ckpt);
1199         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1200
1201         /* write cached NAT/SIT entries to NAT/SIT area */
1202         flush_nat_entries(sbi);
1203         flush_sit_entries(sbi, cpc);
1204
1205         /* unlock all the fs_lock[] in do_checkpoint() */
1206         err = do_checkpoint(sbi, cpc);
1207
1208         unblock_operations(sbi);
1209         stat_inc_cp_count(sbi->stat_info);
1210
1211         if (cpc->reason == CP_RECOVERY)
1212                 f2fs_msg(sbi->sb, KERN_NOTICE,
1213                         "checkpoint: version = %llx", ckpt_ver);
1214
1215         /* do checkpoint periodically */
1216         f2fs_update_time(sbi, CP_TIME);
1217         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1218 out:
1219         mutex_unlock(&sbi->cp_mutex);
1220         return err;
1221 }
1222
1223 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1224 {
1225         int i;
1226
1227         for (i = 0; i < MAX_INO_ENTRY; i++) {
1228                 struct inode_management *im = &sbi->im[i];
1229
1230                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1231                 spin_lock_init(&im->ino_lock);
1232                 INIT_LIST_HEAD(&im->ino_list);
1233                 im->ino_num = 0;
1234         }
1235
1236         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1237                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1238                                 F2FS_ORPHANS_PER_BLOCK;
1239 }
1240
1241 int __init create_checkpoint_caches(void)
1242 {
1243         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1244                         sizeof(struct ino_entry));
1245         if (!ino_entry_slab)
1246                 return -ENOMEM;
1247         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1248                         sizeof(struct inode_entry));
1249         if (!inode_entry_slab) {
1250                 kmem_cache_destroy(ino_entry_slab);
1251                 return -ENOMEM;
1252         }
1253         return 0;
1254 }
1255
1256 void destroy_checkpoint_caches(void)
1257 {
1258         kmem_cache_destroy(ino_entry_slab);
1259         kmem_cache_destroy(inode_entry_slab);
1260 }