]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/f2fs/checkpoint.c
Merge remote-tracking branch 'fscache/fscache'
[karo-tx-linux.git] / fs / f2fs / checkpoint.c
1 /*
2  * fs/f2fs/checkpoint.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/bio.h>
13 #include <linux/mpage.h>
14 #include <linux/writeback.h>
15 #include <linux/blkdev.h>
16 #include <linux/f2fs_fs.h>
17 #include <linux/pagevec.h>
18 #include <linux/swap.h>
19
20 #include "f2fs.h"
21 #include "node.h"
22 #include "segment.h"
23 #include <trace/events/f2fs.h>
24
25 static struct kmem_cache *orphan_entry_slab;
26 static struct kmem_cache *inode_entry_slab;
27
28 /*
29  * We guarantee no failure on the returned page.
30  */
31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
32 {
33         struct address_space *mapping = sbi->meta_inode->i_mapping;
34         struct page *page = NULL;
35 repeat:
36         page = grab_cache_page(mapping, index);
37         if (!page) {
38                 cond_resched();
39                 goto repeat;
40         }
41
42         /* We wait writeback only inside grab_meta_page() */
43         wait_on_page_writeback(page);
44         SetPageUptodate(page);
45         return page;
46 }
47
48 /*
49  * We guarantee no failure on the returned page.
50  */
51 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
52 {
53         struct address_space *mapping = sbi->meta_inode->i_mapping;
54         struct page *page;
55 repeat:
56         page = grab_cache_page(mapping, index);
57         if (!page) {
58                 cond_resched();
59                 goto repeat;
60         }
61         if (PageUptodate(page))
62                 goto out;
63
64         if (f2fs_readpage(sbi, page, index, READ_SYNC))
65                 goto repeat;
66
67         lock_page(page);
68         if (page->mapping != mapping) {
69                 f2fs_put_page(page, 1);
70                 goto repeat;
71         }
72 out:
73         mark_page_accessed(page);
74         return page;
75 }
76
77 static int f2fs_write_meta_page(struct page *page,
78                                 struct writeback_control *wbc)
79 {
80         struct inode *inode = page->mapping->host;
81         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
82
83         /* Should not write any meta pages, if any IO error was occurred */
84         if (wbc->for_reclaim || sbi->por_doing ||
85                         is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)) {
86                 dec_page_count(sbi, F2FS_DIRTY_META);
87                 wbc->pages_skipped++;
88                 set_page_dirty(page);
89                 return AOP_WRITEPAGE_ACTIVATE;
90         }
91
92         wait_on_page_writeback(page);
93
94         write_meta_page(sbi, page);
95         dec_page_count(sbi, F2FS_DIRTY_META);
96         unlock_page(page);
97         return 0;
98 }
99
100 static int f2fs_write_meta_pages(struct address_space *mapping,
101                                 struct writeback_control *wbc)
102 {
103         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
104         struct block_device *bdev = sbi->sb->s_bdev;
105         long written;
106
107         if (wbc->for_kupdate)
108                 return 0;
109
110         if (get_pages(sbi, F2FS_DIRTY_META) == 0)
111                 return 0;
112
113         /* if mounting is failed, skip writing node pages */
114         mutex_lock(&sbi->cp_mutex);
115         written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev));
116         mutex_unlock(&sbi->cp_mutex);
117         wbc->nr_to_write -= written;
118         return 0;
119 }
120
121 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
122                                                 long nr_to_write)
123 {
124         struct address_space *mapping = sbi->meta_inode->i_mapping;
125         pgoff_t index = 0, end = LONG_MAX;
126         struct pagevec pvec;
127         long nwritten = 0;
128         struct writeback_control wbc = {
129                 .for_reclaim = 0,
130         };
131
132         pagevec_init(&pvec, 0);
133
134         while (index <= end) {
135                 int i, nr_pages;
136                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
137                                 PAGECACHE_TAG_DIRTY,
138                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
139                 if (nr_pages == 0)
140                         break;
141
142                 for (i = 0; i < nr_pages; i++) {
143                         struct page *page = pvec.pages[i];
144                         lock_page(page);
145                         BUG_ON(page->mapping != mapping);
146                         BUG_ON(!PageDirty(page));
147                         clear_page_dirty_for_io(page);
148                         if (f2fs_write_meta_page(page, &wbc)) {
149                                 unlock_page(page);
150                                 break;
151                         }
152                         if (nwritten++ >= nr_to_write)
153                                 break;
154                 }
155                 pagevec_release(&pvec);
156                 cond_resched();
157         }
158
159         if (nwritten)
160                 f2fs_submit_bio(sbi, type, nr_to_write == LONG_MAX);
161
162         return nwritten;
163 }
164
165 static int f2fs_set_meta_page_dirty(struct page *page)
166 {
167         struct address_space *mapping = page->mapping;
168         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
169
170         SetPageUptodate(page);
171         if (!PageDirty(page)) {
172                 __set_page_dirty_nobuffers(page);
173                 inc_page_count(sbi, F2FS_DIRTY_META);
174                 return 1;
175         }
176         return 0;
177 }
178
179 const struct address_space_operations f2fs_meta_aops = {
180         .writepage      = f2fs_write_meta_page,
181         .writepages     = f2fs_write_meta_pages,
182         .set_page_dirty = f2fs_set_meta_page_dirty,
183 };
184
185 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
186 {
187         unsigned int max_orphans;
188         int err = 0;
189
190         /*
191          * considering 512 blocks in a segment 5 blocks are needed for cp
192          * and log segment summaries. Remaining blocks are used to keep
193          * orphan entries with the limitation one reserved segment
194          * for cp pack we can have max 1020*507 orphan entries
195          */
196         max_orphans = (sbi->blocks_per_seg - 5) * F2FS_ORPHANS_PER_BLOCK;
197         mutex_lock(&sbi->orphan_inode_mutex);
198         if (sbi->n_orphans >= max_orphans)
199                 err = -ENOSPC;
200         else
201                 sbi->n_orphans++;
202         mutex_unlock(&sbi->orphan_inode_mutex);
203         return err;
204 }
205
206 void release_orphan_inode(struct f2fs_sb_info *sbi)
207 {
208         mutex_lock(&sbi->orphan_inode_mutex);
209         BUG_ON(sbi->n_orphans == 0);
210         sbi->n_orphans--;
211         mutex_unlock(&sbi->orphan_inode_mutex);
212 }
213
214 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
215 {
216         struct list_head *head, *this;
217         struct orphan_inode_entry *new = NULL, *orphan = NULL;
218
219         mutex_lock(&sbi->orphan_inode_mutex);
220         head = &sbi->orphan_inode_list;
221         list_for_each(this, head) {
222                 orphan = list_entry(this, struct orphan_inode_entry, list);
223                 if (orphan->ino == ino)
224                         goto out;
225                 if (orphan->ino > ino)
226                         break;
227                 orphan = NULL;
228         }
229
230         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
231         new->ino = ino;
232
233         /* add new_oentry into list which is sorted by inode number */
234         if (orphan)
235                 list_add(&new->list, this->prev);
236         else
237                 list_add_tail(&new->list, head);
238 out:
239         mutex_unlock(&sbi->orphan_inode_mutex);
240 }
241
242 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
243 {
244         struct list_head *head;
245         struct orphan_inode_entry *orphan;
246
247         mutex_lock(&sbi->orphan_inode_mutex);
248         head = &sbi->orphan_inode_list;
249         list_for_each_entry(orphan, head, list) {
250                 if (orphan->ino == ino) {
251                         list_del(&orphan->list);
252                         kmem_cache_free(orphan_entry_slab, orphan);
253                         BUG_ON(sbi->n_orphans == 0);
254                         sbi->n_orphans--;
255                         break;
256                 }
257         }
258         mutex_unlock(&sbi->orphan_inode_mutex);
259 }
260
261 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
262 {
263         struct inode *inode = f2fs_iget(sbi->sb, ino);
264         BUG_ON(IS_ERR(inode));
265         clear_nlink(inode);
266
267         /* truncate all the data during iput */
268         iput(inode);
269 }
270
271 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
272 {
273         block_t start_blk, orphan_blkaddr, i, j;
274
275         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
276                 return 0;
277
278         sbi->por_doing = true;
279         start_blk = __start_cp_addr(sbi) + 1;
280         orphan_blkaddr = __start_sum_addr(sbi) - 1;
281
282         for (i = 0; i < orphan_blkaddr; i++) {
283                 struct page *page = get_meta_page(sbi, start_blk + i);
284                 struct f2fs_orphan_block *orphan_blk;
285
286                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
287                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
288                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
289                         recover_orphan_inode(sbi, ino);
290                 }
291                 f2fs_put_page(page, 1);
292         }
293         /* clear Orphan Flag */
294         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
295         sbi->por_doing = false;
296         return 0;
297 }
298
299 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
300 {
301         struct list_head *head, *this, *next;
302         struct f2fs_orphan_block *orphan_blk = NULL;
303         struct page *page = NULL;
304         unsigned int nentries = 0;
305         unsigned short index = 1;
306         unsigned short orphan_blocks;
307
308         orphan_blocks = (unsigned short)((sbi->n_orphans +
309                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
310
311         mutex_lock(&sbi->orphan_inode_mutex);
312         head = &sbi->orphan_inode_list;
313
314         /* loop for each orphan inode entry and write them in Jornal block */
315         list_for_each_safe(this, next, head) {
316                 struct orphan_inode_entry *orphan;
317
318                 orphan = list_entry(this, struct orphan_inode_entry, list);
319
320                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
321                         /*
322                          * an orphan block is full of 1020 entries,
323                          * then we need to flush current orphan blocks
324                          * and bring another one in memory
325                          */
326                         orphan_blk->blk_addr = cpu_to_le16(index);
327                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
328                         orphan_blk->entry_count = cpu_to_le32(nentries);
329                         set_page_dirty(page);
330                         f2fs_put_page(page, 1);
331                         index++;
332                         start_blk++;
333                         nentries = 0;
334                         page = NULL;
335                 }
336                 if (page)
337                         goto page_exist;
338
339                 page = grab_meta_page(sbi, start_blk);
340                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
341                 memset(orphan_blk, 0, sizeof(*orphan_blk));
342 page_exist:
343                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
344         }
345         if (!page)
346                 goto end;
347
348         orphan_blk->blk_addr = cpu_to_le16(index);
349         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
350         orphan_blk->entry_count = cpu_to_le32(nentries);
351         set_page_dirty(page);
352         f2fs_put_page(page, 1);
353 end:
354         mutex_unlock(&sbi->orphan_inode_mutex);
355 }
356
357 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
358                                 block_t cp_addr, unsigned long long *version)
359 {
360         struct page *cp_page_1, *cp_page_2 = NULL;
361         unsigned long blk_size = sbi->blocksize;
362         struct f2fs_checkpoint *cp_block;
363         unsigned long long cur_version = 0, pre_version = 0;
364         size_t crc_offset;
365         __u32 crc = 0;
366
367         /* Read the 1st cp block in this CP pack */
368         cp_page_1 = get_meta_page(sbi, cp_addr);
369
370         /* get the version number */
371         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
372         crc_offset = le32_to_cpu(cp_block->checksum_offset);
373         if (crc_offset >= blk_size)
374                 goto invalid_cp1;
375
376         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
377         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
378                 goto invalid_cp1;
379
380         pre_version = cur_cp_version(cp_block);
381
382         /* Read the 2nd cp block in this CP pack */
383         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
384         cp_page_2 = get_meta_page(sbi, cp_addr);
385
386         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
387         crc_offset = le32_to_cpu(cp_block->checksum_offset);
388         if (crc_offset >= blk_size)
389                 goto invalid_cp2;
390
391         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
392         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
393                 goto invalid_cp2;
394
395         cur_version = cur_cp_version(cp_block);
396
397         if (cur_version == pre_version) {
398                 *version = cur_version;
399                 f2fs_put_page(cp_page_2, 1);
400                 return cp_page_1;
401         }
402 invalid_cp2:
403         f2fs_put_page(cp_page_2, 1);
404 invalid_cp1:
405         f2fs_put_page(cp_page_1, 1);
406         return NULL;
407 }
408
409 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
410 {
411         struct f2fs_checkpoint *cp_block;
412         struct f2fs_super_block *fsb = sbi->raw_super;
413         struct page *cp1, *cp2, *cur_page;
414         unsigned long blk_size = sbi->blocksize;
415         unsigned long long cp1_version = 0, cp2_version = 0;
416         unsigned long long cp_start_blk_no;
417
418         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
419         if (!sbi->ckpt)
420                 return -ENOMEM;
421         /*
422          * Finding out valid cp block involves read both
423          * sets( cp pack1 and cp pack 2)
424          */
425         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
426         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
427
428         /* The second checkpoint pack should start at the next segment */
429         cp_start_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
430         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
431
432         if (cp1 && cp2) {
433                 if (ver_after(cp2_version, cp1_version))
434                         cur_page = cp2;
435                 else
436                         cur_page = cp1;
437         } else if (cp1) {
438                 cur_page = cp1;
439         } else if (cp2) {
440                 cur_page = cp2;
441         } else {
442                 goto fail_no_cp;
443         }
444
445         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
446         memcpy(sbi->ckpt, cp_block, blk_size);
447
448         f2fs_put_page(cp1, 1);
449         f2fs_put_page(cp2, 1);
450         return 0;
451
452 fail_no_cp:
453         kfree(sbi->ckpt);
454         return -EINVAL;
455 }
456
457 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
458 {
459         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
460         struct list_head *head = &sbi->dir_inode_list;
461         struct list_head *this;
462
463         list_for_each(this, head) {
464                 struct dir_inode_entry *entry;
465                 entry = list_entry(this, struct dir_inode_entry, list);
466                 if (entry->inode == inode)
467                         return -EEXIST;
468         }
469         list_add_tail(&new->list, head);
470         stat_inc_dirty_dir(sbi);
471         return 0;
472 }
473
474 void set_dirty_dir_page(struct inode *inode, struct page *page)
475 {
476         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
477         struct dir_inode_entry *new;
478
479         if (!S_ISDIR(inode->i_mode))
480                 return;
481
482         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
483         new->inode = inode;
484         INIT_LIST_HEAD(&new->list);
485
486         spin_lock(&sbi->dir_inode_lock);
487         if (__add_dirty_inode(inode, new))
488                 kmem_cache_free(inode_entry_slab, new);
489
490         inc_page_count(sbi, F2FS_DIRTY_DENTS);
491         inode_inc_dirty_dents(inode);
492         SetPagePrivate(page);
493         spin_unlock(&sbi->dir_inode_lock);
494 }
495
496 void add_dirty_dir_inode(struct inode *inode)
497 {
498         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
499         struct dir_inode_entry *new =
500                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
501
502         new->inode = inode;
503         INIT_LIST_HEAD(&new->list);
504
505         spin_lock(&sbi->dir_inode_lock);
506         if (__add_dirty_inode(inode, new))
507                 kmem_cache_free(inode_entry_slab, new);
508         spin_unlock(&sbi->dir_inode_lock);
509 }
510
511 void remove_dirty_dir_inode(struct inode *inode)
512 {
513         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
514         struct list_head *head = &sbi->dir_inode_list;
515         struct list_head *this;
516
517         if (!S_ISDIR(inode->i_mode))
518                 return;
519
520         spin_lock(&sbi->dir_inode_lock);
521         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
522                 spin_unlock(&sbi->dir_inode_lock);
523                 return;
524         }
525
526         list_for_each(this, head) {
527                 struct dir_inode_entry *entry;
528                 entry = list_entry(this, struct dir_inode_entry, list);
529                 if (entry->inode == inode) {
530                         list_del(&entry->list);
531                         kmem_cache_free(inode_entry_slab, entry);
532                         stat_dec_dirty_dir(sbi);
533                         break;
534                 }
535         }
536         spin_unlock(&sbi->dir_inode_lock);
537
538         /* Only from the recovery routine */
539         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
540                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
541                 iput(inode);
542         }
543 }
544
545 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
546 {
547         struct list_head *head = &sbi->dir_inode_list;
548         struct list_head *this;
549         struct inode *inode = NULL;
550
551         spin_lock(&sbi->dir_inode_lock);
552         list_for_each(this, head) {
553                 struct dir_inode_entry *entry;
554                 entry = list_entry(this, struct dir_inode_entry, list);
555                 if (entry->inode->i_ino == ino) {
556                         inode = entry->inode;
557                         break;
558                 }
559         }
560         spin_unlock(&sbi->dir_inode_lock);
561         return inode;
562 }
563
564 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
565 {
566         struct list_head *head = &sbi->dir_inode_list;
567         struct dir_inode_entry *entry;
568         struct inode *inode;
569 retry:
570         spin_lock(&sbi->dir_inode_lock);
571         if (list_empty(head)) {
572                 spin_unlock(&sbi->dir_inode_lock);
573                 return;
574         }
575         entry = list_entry(head->next, struct dir_inode_entry, list);
576         inode = igrab(entry->inode);
577         spin_unlock(&sbi->dir_inode_lock);
578         if (inode) {
579                 filemap_flush(inode->i_mapping);
580                 iput(inode);
581         } else {
582                 /*
583                  * We should submit bio, since it exists several
584                  * wribacking dentry pages in the freeing inode.
585                  */
586                 f2fs_submit_bio(sbi, DATA, true);
587         }
588         goto retry;
589 }
590
591 /*
592  * Freeze all the FS-operations for checkpoint.
593  */
594 static void block_operations(struct f2fs_sb_info *sbi)
595 {
596         struct writeback_control wbc = {
597                 .sync_mode = WB_SYNC_ALL,
598                 .nr_to_write = LONG_MAX,
599                 .for_reclaim = 0,
600         };
601         struct blk_plug plug;
602
603         blk_start_plug(&plug);
604
605 retry_flush_dents:
606         f2fs_lock_all(sbi);
607         /* write all the dirty dentry pages */
608         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
609                 f2fs_unlock_all(sbi);
610                 sync_dirty_dir_inodes(sbi);
611                 goto retry_flush_dents;
612         }
613
614         /*
615          * POR: we should ensure that there is no dirty node pages
616          * until finishing nat/sit flush.
617          */
618 retry_flush_nodes:
619         mutex_lock(&sbi->node_write);
620
621         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
622                 mutex_unlock(&sbi->node_write);
623                 sync_node_pages(sbi, 0, &wbc);
624                 goto retry_flush_nodes;
625         }
626         blk_finish_plug(&plug);
627 }
628
629 static void unblock_operations(struct f2fs_sb_info *sbi)
630 {
631         mutex_unlock(&sbi->node_write);
632         f2fs_unlock_all(sbi);
633 }
634
635 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
636 {
637         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
638         nid_t last_nid = 0;
639         block_t start_blk;
640         struct page *cp_page;
641         unsigned int data_sum_blocks, orphan_blocks;
642         __u32 crc32 = 0;
643         void *kaddr;
644         int i;
645
646         /* Flush all the NAT/SIT pages */
647         while (get_pages(sbi, F2FS_DIRTY_META))
648                 sync_meta_pages(sbi, META, LONG_MAX);
649
650         next_free_nid(sbi, &last_nid);
651
652         /*
653          * modify checkpoint
654          * version number is already updated
655          */
656         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
657         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
658         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
659         for (i = 0; i < 3; i++) {
660                 ckpt->cur_node_segno[i] =
661                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
662                 ckpt->cur_node_blkoff[i] =
663                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
664                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
665                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
666         }
667         for (i = 0; i < 3; i++) {
668                 ckpt->cur_data_segno[i] =
669                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
670                 ckpt->cur_data_blkoff[i] =
671                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
672                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
673                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
674         }
675
676         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
677         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
678         ckpt->next_free_nid = cpu_to_le32(last_nid);
679
680         /* 2 cp  + n data seg summary + orphan inode blocks */
681         data_sum_blocks = npages_for_summary_flush(sbi);
682         if (data_sum_blocks < 3)
683                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
684         else
685                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
686
687         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
688                                         / F2FS_ORPHANS_PER_BLOCK;
689         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
690
691         if (is_umount) {
692                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
693                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
694                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
695         } else {
696                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
697                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
698                         data_sum_blocks + orphan_blocks);
699         }
700
701         if (sbi->n_orphans)
702                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
703         else
704                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
705
706         /* update SIT/NAT bitmap */
707         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
708         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
709
710         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
711         *((__le32 *)((unsigned char *)ckpt +
712                                 le32_to_cpu(ckpt->checksum_offset)))
713                                 = cpu_to_le32(crc32);
714
715         start_blk = __start_cp_addr(sbi);
716
717         /* write out checkpoint buffer at block 0 */
718         cp_page = grab_meta_page(sbi, start_blk++);
719         kaddr = page_address(cp_page);
720         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
721         set_page_dirty(cp_page);
722         f2fs_put_page(cp_page, 1);
723
724         if (sbi->n_orphans) {
725                 write_orphan_inodes(sbi, start_blk);
726                 start_blk += orphan_blocks;
727         }
728
729         write_data_summaries(sbi, start_blk);
730         start_blk += data_sum_blocks;
731         if (is_umount) {
732                 write_node_summaries(sbi, start_blk);
733                 start_blk += NR_CURSEG_NODE_TYPE;
734         }
735
736         /* writeout checkpoint block */
737         cp_page = grab_meta_page(sbi, start_blk);
738         kaddr = page_address(cp_page);
739         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
740         set_page_dirty(cp_page);
741         f2fs_put_page(cp_page, 1);
742
743         /* wait for previous submitted node/meta pages writeback */
744         sbi->cp_task = current;
745         while (get_pages(sbi, F2FS_WRITEBACK)) {
746                 set_current_state(TASK_UNINTERRUPTIBLE);
747                 if (!get_pages(sbi, F2FS_WRITEBACK))
748                         break;
749                 io_schedule();
750         }
751         __set_current_state(TASK_RUNNING);
752         sbi->cp_task = NULL;
753
754         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
755         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
756
757         /* update user_block_counts */
758         sbi->last_valid_block_count = sbi->total_valid_block_count;
759         sbi->alloc_valid_block_count = 0;
760
761         /* Here, we only have one bio having CP pack */
762         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
763
764         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
765                 clear_prefree_segments(sbi);
766                 F2FS_RESET_SB_DIRT(sbi);
767         }
768 }
769
770 /*
771  * We guarantee that this checkpoint procedure should not fail.
772  */
773 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
774 {
775         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
776         unsigned long long ckpt_ver;
777
778         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
779
780         mutex_lock(&sbi->cp_mutex);
781         block_operations(sbi);
782
783         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
784
785         f2fs_submit_bio(sbi, DATA, true);
786         f2fs_submit_bio(sbi, NODE, true);
787         f2fs_submit_bio(sbi, META, true);
788
789         /*
790          * update checkpoint pack index
791          * Increase the version number so that
792          * SIT entries and seg summaries are written at correct place
793          */
794         ckpt_ver = cur_cp_version(ckpt);
795         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
796
797         /* write cached NAT/SIT entries to NAT/SIT area */
798         flush_nat_entries(sbi);
799         flush_sit_entries(sbi);
800
801         /* unlock all the fs_lock[] in do_checkpoint() */
802         do_checkpoint(sbi, is_umount);
803
804         unblock_operations(sbi);
805         mutex_unlock(&sbi->cp_mutex);
806
807         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
808 }
809
810 void init_orphan_info(struct f2fs_sb_info *sbi)
811 {
812         mutex_init(&sbi->orphan_inode_mutex);
813         INIT_LIST_HEAD(&sbi->orphan_inode_list);
814         sbi->n_orphans = 0;
815 }
816
817 int __init create_checkpoint_caches(void)
818 {
819         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
820                         sizeof(struct orphan_inode_entry), NULL);
821         if (unlikely(!orphan_entry_slab))
822                 return -ENOMEM;
823         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
824                         sizeof(struct dir_inode_entry), NULL);
825         if (unlikely(!inode_entry_slab)) {
826                 kmem_cache_destroy(orphan_entry_slab);
827                 return -ENOMEM;
828         }
829         return 0;
830 }
831
832 void destroy_checkpoint_caches(void)
833 {
834         kmem_cache_destroy(orphan_entry_slab);
835         kmem_cache_destroy(inode_entry_slab);
836 }