f2fs: push some variables to debug part
[platform/adaptation/renesas_rcar/renesas_kernel.git] / fs / f2fs / segment.c
1 /*
2  * fs/f2fs/segment.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/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/vmalloc.h>
17
18 #include "f2fs.h"
19 #include "segment.h"
20 #include "node.h"
21 #include <trace/events/f2fs.h>
22
23 /*
24  * This function balances dirty node and dentry pages.
25  * In addition, it controls garbage collection.
26  */
27 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
28 {
29         /*
30          * We should do GC or end up with checkpoint, if there are so many dirty
31          * dir/node pages without enough free segments.
32          */
33         if (has_not_enough_free_secs(sbi, 0)) {
34                 mutex_lock(&sbi->gc_mutex);
35                 f2fs_gc(sbi);
36         }
37 }
38
39 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
40                 enum dirty_type dirty_type)
41 {
42         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
43
44         /* need not be added */
45         if (IS_CURSEG(sbi, segno))
46                 return;
47
48         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
49                 dirty_i->nr_dirty[dirty_type]++;
50
51         if (dirty_type == DIRTY) {
52                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
53                 enum dirty_type t = DIRTY_HOT_DATA;
54
55                 dirty_type = sentry->type;
56
57                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
58                         dirty_i->nr_dirty[dirty_type]++;
59
60                 /* Only one bitmap should be set */
61                 for (; t <= DIRTY_COLD_NODE; t++) {
62                         if (t == dirty_type)
63                                 continue;
64                         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
65                                 dirty_i->nr_dirty[t]--;
66                 }
67         }
68 }
69
70 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
71                 enum dirty_type dirty_type)
72 {
73         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
74
75         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
76                 dirty_i->nr_dirty[dirty_type]--;
77
78         if (dirty_type == DIRTY) {
79                 enum dirty_type t = DIRTY_HOT_DATA;
80
81                 /* clear all the bitmaps */
82                 for (; t <= DIRTY_COLD_NODE; t++)
83                         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
84                                 dirty_i->nr_dirty[t]--;
85
86                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
87                         clear_bit(GET_SECNO(sbi, segno),
88                                                 dirty_i->victim_secmap);
89         }
90 }
91
92 /*
93  * Should not occur error such as -ENOMEM.
94  * Adding dirty entry into seglist is not critical operation.
95  * If a given segment is one of current working segments, it won't be added.
96  */
97 void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
98 {
99         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
100         unsigned short valid_blocks;
101
102         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
103                 return;
104
105         mutex_lock(&dirty_i->seglist_lock);
106
107         valid_blocks = get_valid_blocks(sbi, segno, 0);
108
109         if (valid_blocks == 0) {
110                 __locate_dirty_segment(sbi, segno, PRE);
111                 __remove_dirty_segment(sbi, segno, DIRTY);
112         } else if (valid_blocks < sbi->blocks_per_seg) {
113                 __locate_dirty_segment(sbi, segno, DIRTY);
114         } else {
115                 /* Recovery routine with SSR needs this */
116                 __remove_dirty_segment(sbi, segno, DIRTY);
117         }
118
119         mutex_unlock(&dirty_i->seglist_lock);
120         return;
121 }
122
123 /*
124  * Should call clear_prefree_segments after checkpoint is done.
125  */
126 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
127 {
128         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
129         unsigned int segno = -1;
130         unsigned int total_segs = TOTAL_SEGS(sbi);
131
132         mutex_lock(&dirty_i->seglist_lock);
133         while (1) {
134                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
135                                 segno + 1);
136                 if (segno >= total_segs)
137                         break;
138                 __set_test_and_free(sbi, segno);
139         }
140         mutex_unlock(&dirty_i->seglist_lock);
141 }
142
143 void clear_prefree_segments(struct f2fs_sb_info *sbi)
144 {
145         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
146         unsigned int segno = -1;
147         unsigned int total_segs = TOTAL_SEGS(sbi);
148
149         mutex_lock(&dirty_i->seglist_lock);
150         while (1) {
151                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
152                                 segno + 1);
153                 if (segno >= total_segs)
154                         break;
155
156                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
157                         dirty_i->nr_dirty[PRE]--;
158
159                 /* Let's use trim */
160                 if (test_opt(sbi, DISCARD))
161                         blkdev_issue_discard(sbi->sb->s_bdev,
162                                         START_BLOCK(sbi, segno) <<
163                                         sbi->log_sectors_per_block,
164                                         1 << (sbi->log_sectors_per_block +
165                                                 sbi->log_blocks_per_seg),
166                                         GFP_NOFS, 0);
167         }
168         mutex_unlock(&dirty_i->seglist_lock);
169 }
170
171 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
172 {
173         struct sit_info *sit_i = SIT_I(sbi);
174         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
175                 sit_i->dirty_sentries++;
176 }
177
178 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
179                                         unsigned int segno, int modified)
180 {
181         struct seg_entry *se = get_seg_entry(sbi, segno);
182         se->type = type;
183         if (modified)
184                 __mark_sit_entry_dirty(sbi, segno);
185 }
186
187 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
188 {
189         struct seg_entry *se;
190         unsigned int segno, offset;
191         long int new_vblocks;
192
193         segno = GET_SEGNO(sbi, blkaddr);
194
195         se = get_seg_entry(sbi, segno);
196         new_vblocks = se->valid_blocks + del;
197         offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
198
199         BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
200                                 (new_vblocks > sbi->blocks_per_seg)));
201
202         se->valid_blocks = new_vblocks;
203         se->mtime = get_mtime(sbi);
204         SIT_I(sbi)->max_mtime = se->mtime;
205
206         /* Update valid block bitmap */
207         if (del > 0) {
208                 if (f2fs_set_bit(offset, se->cur_valid_map))
209                         BUG();
210         } else {
211                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
212                         BUG();
213         }
214         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
215                 se->ckpt_valid_blocks += del;
216
217         __mark_sit_entry_dirty(sbi, segno);
218
219         /* update total number of valid blocks to be written in ckpt area */
220         SIT_I(sbi)->written_valid_blocks += del;
221
222         if (sbi->segs_per_sec > 1)
223                 get_sec_entry(sbi, segno)->valid_blocks += del;
224 }
225
226 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
227                         block_t old_blkaddr, block_t new_blkaddr)
228 {
229         update_sit_entry(sbi, new_blkaddr, 1);
230         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
231                 update_sit_entry(sbi, old_blkaddr, -1);
232 }
233
234 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
235 {
236         unsigned int segno = GET_SEGNO(sbi, addr);
237         struct sit_info *sit_i = SIT_I(sbi);
238
239         BUG_ON(addr == NULL_ADDR);
240         if (addr == NEW_ADDR)
241                 return;
242
243         /* add it into sit main buffer */
244         mutex_lock(&sit_i->sentry_lock);
245
246         update_sit_entry(sbi, addr, -1);
247
248         /* add it into dirty seglist */
249         locate_dirty_segment(sbi, segno);
250
251         mutex_unlock(&sit_i->sentry_lock);
252 }
253
254 /*
255  * This function should be resided under the curseg_mutex lock
256  */
257 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
258                 struct f2fs_summary *sum, unsigned short offset)
259 {
260         struct curseg_info *curseg = CURSEG_I(sbi, type);
261         void *addr = curseg->sum_blk;
262         addr += offset * sizeof(struct f2fs_summary);
263         memcpy(addr, sum, sizeof(struct f2fs_summary));
264         return;
265 }
266
267 /*
268  * Calculate the number of current summary pages for writing
269  */
270 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
271 {
272         int total_size_bytes = 0;
273         int valid_sum_count = 0;
274         int i, sum_space;
275
276         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
277                 if (sbi->ckpt->alloc_type[i] == SSR)
278                         valid_sum_count += sbi->blocks_per_seg;
279                 else
280                         valid_sum_count += curseg_blkoff(sbi, i);
281         }
282
283         total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
284                         + sizeof(struct nat_journal) + 2
285                         + sizeof(struct sit_journal) + 2;
286         sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
287         if (total_size_bytes < sum_space)
288                 return 1;
289         else if (total_size_bytes < 2 * sum_space)
290                 return 2;
291         return 3;
292 }
293
294 /*
295  * Caller should put this summary page
296  */
297 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
298 {
299         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
300 }
301
302 static void write_sum_page(struct f2fs_sb_info *sbi,
303                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
304 {
305         struct page *page = grab_meta_page(sbi, blk_addr);
306         void *kaddr = page_address(page);
307         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
308         set_page_dirty(page);
309         f2fs_put_page(page, 1);
310 }
311
312 static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi, int type)
313 {
314         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
315         unsigned long *prefree_segmap = dirty_i->dirty_segmap[PRE];
316         unsigned int segno;
317         unsigned int ofs = 0;
318
319         /*
320          * If there is not enough reserved sections,
321          * we should not reuse prefree segments.
322          */
323         if (has_not_enough_free_secs(sbi, 0))
324                 return NULL_SEGNO;
325
326         /*
327          * NODE page should not reuse prefree segment,
328          * since those information is used for SPOR.
329          */
330         if (IS_NODESEG(type))
331                 return NULL_SEGNO;
332 next:
333         segno = find_next_bit(prefree_segmap, TOTAL_SEGS(sbi), ofs);
334         ofs += sbi->segs_per_sec;
335
336         if (segno < TOTAL_SEGS(sbi)) {
337                 int i;
338
339                 /* skip intermediate segments in a section */
340                 if (segno % sbi->segs_per_sec)
341                         goto next;
342
343                 /* skip if the section is currently used */
344                 if (sec_usage_check(sbi, GET_SECNO(sbi, segno)))
345                         goto next;
346
347                 /* skip if whole section is not prefree */
348                 for (i = 1; i < sbi->segs_per_sec; i++)
349                         if (!test_bit(segno + i, prefree_segmap))
350                                 goto next;
351
352                 /* skip if whole section was not free at the last checkpoint */
353                 for (i = 0; i < sbi->segs_per_sec; i++)
354                         if (get_seg_entry(sbi, segno + i)->ckpt_valid_blocks)
355                                 goto next;
356
357                 return segno;
358         }
359         return NULL_SEGNO;
360 }
361
362 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
363 {
364         struct curseg_info *curseg = CURSEG_I(sbi, type);
365         unsigned int segno = curseg->segno + 1;
366         struct free_segmap_info *free_i = FREE_I(sbi);
367
368         if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
369                 return !test_bit(segno, free_i->free_segmap);
370         return 0;
371 }
372
373 /*
374  * Find a new segment from the free segments bitmap to right order
375  * This function should be returned with success, otherwise BUG
376  */
377 static void get_new_segment(struct f2fs_sb_info *sbi,
378                         unsigned int *newseg, bool new_sec, int dir)
379 {
380         struct free_segmap_info *free_i = FREE_I(sbi);
381         unsigned int segno, secno, zoneno;
382         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
383         unsigned int hint = *newseg / sbi->segs_per_sec;
384         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
385         unsigned int left_start = hint;
386         bool init = true;
387         int go_left = 0;
388         int i;
389
390         write_lock(&free_i->segmap_lock);
391
392         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
393                 segno = find_next_zero_bit(free_i->free_segmap,
394                                         TOTAL_SEGS(sbi), *newseg + 1);
395                 if (segno - *newseg < sbi->segs_per_sec -
396                                         (*newseg % sbi->segs_per_sec))
397                         goto got_it;
398         }
399 find_other_zone:
400         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
401         if (secno >= TOTAL_SECS(sbi)) {
402                 if (dir == ALLOC_RIGHT) {
403                         secno = find_next_zero_bit(free_i->free_secmap,
404                                                         TOTAL_SECS(sbi), 0);
405                         BUG_ON(secno >= TOTAL_SECS(sbi));
406                 } else {
407                         go_left = 1;
408                         left_start = hint - 1;
409                 }
410         }
411         if (go_left == 0)
412                 goto skip_left;
413
414         while (test_bit(left_start, free_i->free_secmap)) {
415                 if (left_start > 0) {
416                         left_start--;
417                         continue;
418                 }
419                 left_start = find_next_zero_bit(free_i->free_secmap,
420                                                         TOTAL_SECS(sbi), 0);
421                 BUG_ON(left_start >= TOTAL_SECS(sbi));
422                 break;
423         }
424         secno = left_start;
425 skip_left:
426         hint = secno;
427         segno = secno * sbi->segs_per_sec;
428         zoneno = secno / sbi->secs_per_zone;
429
430         /* give up on finding another zone */
431         if (!init)
432                 goto got_it;
433         if (sbi->secs_per_zone == 1)
434                 goto got_it;
435         if (zoneno == old_zoneno)
436                 goto got_it;
437         if (dir == ALLOC_LEFT) {
438                 if (!go_left && zoneno + 1 >= total_zones)
439                         goto got_it;
440                 if (go_left && zoneno == 0)
441                         goto got_it;
442         }
443         for (i = 0; i < NR_CURSEG_TYPE; i++)
444                 if (CURSEG_I(sbi, i)->zone == zoneno)
445                         break;
446
447         if (i < NR_CURSEG_TYPE) {
448                 /* zone is in user, try another */
449                 if (go_left)
450                         hint = zoneno * sbi->secs_per_zone - 1;
451                 else if (zoneno + 1 >= total_zones)
452                         hint = 0;
453                 else
454                         hint = (zoneno + 1) * sbi->secs_per_zone;
455                 init = false;
456                 goto find_other_zone;
457         }
458 got_it:
459         /* set it as dirty segment in free segmap */
460         BUG_ON(test_bit(segno, free_i->free_segmap));
461         __set_inuse(sbi, segno);
462         *newseg = segno;
463         write_unlock(&free_i->segmap_lock);
464 }
465
466 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
467 {
468         struct curseg_info *curseg = CURSEG_I(sbi, type);
469         struct summary_footer *sum_footer;
470
471         curseg->segno = curseg->next_segno;
472         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
473         curseg->next_blkoff = 0;
474         curseg->next_segno = NULL_SEGNO;
475
476         sum_footer = &(curseg->sum_blk->footer);
477         memset(sum_footer, 0, sizeof(struct summary_footer));
478         if (IS_DATASEG(type))
479                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
480         if (IS_NODESEG(type))
481                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
482         __set_sit_entry_type(sbi, type, curseg->segno, modified);
483 }
484
485 /*
486  * Allocate a current working segment.
487  * This function always allocates a free segment in LFS manner.
488  */
489 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
490 {
491         struct curseg_info *curseg = CURSEG_I(sbi, type);
492         unsigned int segno = curseg->segno;
493         int dir = ALLOC_LEFT;
494
495         write_sum_page(sbi, curseg->sum_blk,
496                                 GET_SUM_BLOCK(sbi, segno));
497         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
498                 dir = ALLOC_RIGHT;
499
500         if (test_opt(sbi, NOHEAP))
501                 dir = ALLOC_RIGHT;
502
503         get_new_segment(sbi, &segno, new_sec, dir);
504         curseg->next_segno = segno;
505         reset_curseg(sbi, type, 1);
506         curseg->alloc_type = LFS;
507 }
508
509 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
510                         struct curseg_info *seg, block_t start)
511 {
512         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
513         block_t ofs;
514         for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
515                 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
516                         && !f2fs_test_bit(ofs, se->cur_valid_map))
517                         break;
518         }
519         seg->next_blkoff = ofs;
520 }
521
522 /*
523  * If a segment is written by LFS manner, next block offset is just obtained
524  * by increasing the current block offset. However, if a segment is written by
525  * SSR manner, next block offset obtained by calling __next_free_blkoff
526  */
527 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
528                                 struct curseg_info *seg)
529 {
530         if (seg->alloc_type == SSR)
531                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
532         else
533                 seg->next_blkoff++;
534 }
535
536 /*
537  * This function always allocates a used segment (from dirty seglist) by SSR
538  * manner, so it should recover the existing segment information of valid blocks
539  */
540 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
541 {
542         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
543         struct curseg_info *curseg = CURSEG_I(sbi, type);
544         unsigned int new_segno = curseg->next_segno;
545         struct f2fs_summary_block *sum_node;
546         struct page *sum_page;
547
548         write_sum_page(sbi, curseg->sum_blk,
549                                 GET_SUM_BLOCK(sbi, curseg->segno));
550         __set_test_and_inuse(sbi, new_segno);
551
552         mutex_lock(&dirty_i->seglist_lock);
553         __remove_dirty_segment(sbi, new_segno, PRE);
554         __remove_dirty_segment(sbi, new_segno, DIRTY);
555         mutex_unlock(&dirty_i->seglist_lock);
556
557         reset_curseg(sbi, type, 1);
558         curseg->alloc_type = SSR;
559         __next_free_blkoff(sbi, curseg, 0);
560
561         if (reuse) {
562                 sum_page = get_sum_page(sbi, new_segno);
563                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
564                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
565                 f2fs_put_page(sum_page, 1);
566         }
567 }
568
569 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
570 {
571         struct curseg_info *curseg = CURSEG_I(sbi, type);
572         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
573
574         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
575                 return v_ops->get_victim(sbi,
576                                 &(curseg)->next_segno, BG_GC, type, SSR);
577
578         /* For data segments, let's do SSR more intensively */
579         for (; type >= CURSEG_HOT_DATA; type--)
580                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
581                                                 BG_GC, type, SSR))
582                         return 1;
583         return 0;
584 }
585
586 /*
587  * flush out current segment and replace it with new segment
588  * This function should be returned with success, otherwise BUG
589  */
590 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
591                                                 int type, bool force)
592 {
593         struct curseg_info *curseg = CURSEG_I(sbi, type);
594
595         if (force) {
596                 new_curseg(sbi, type, true);
597                 goto out;
598         }
599
600         curseg->next_segno = check_prefree_segments(sbi, type);
601
602         if (curseg->next_segno != NULL_SEGNO)
603                 change_curseg(sbi, type, false);
604         else if (type == CURSEG_WARM_NODE)
605                 new_curseg(sbi, type, false);
606         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
607                 new_curseg(sbi, type, false);
608         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
609                 change_curseg(sbi, type, true);
610         else
611                 new_curseg(sbi, type, false);
612 out:
613 #ifdef CONFIG_F2FS_STAT_FS
614         sbi->segment_count[curseg->alloc_type]++;
615 #endif
616         return;
617 }
618
619 void allocate_new_segments(struct f2fs_sb_info *sbi)
620 {
621         struct curseg_info *curseg;
622         unsigned int old_curseg;
623         int i;
624
625         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
626                 curseg = CURSEG_I(sbi, i);
627                 old_curseg = curseg->segno;
628                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
629                 locate_dirty_segment(sbi, old_curseg);
630         }
631 }
632
633 static const struct segment_allocation default_salloc_ops = {
634         .allocate_segment = allocate_segment_by_default,
635 };
636
637 static void f2fs_end_io_write(struct bio *bio, int err)
638 {
639         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
640         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
641         struct bio_private *p = bio->bi_private;
642
643         do {
644                 struct page *page = bvec->bv_page;
645
646                 if (--bvec >= bio->bi_io_vec)
647                         prefetchw(&bvec->bv_page->flags);
648                 if (!uptodate) {
649                         SetPageError(page);
650                         if (page->mapping)
651                                 set_bit(AS_EIO, &page->mapping->flags);
652                         set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
653                         p->sbi->sb->s_flags |= MS_RDONLY;
654                 }
655                 end_page_writeback(page);
656                 dec_page_count(p->sbi, F2FS_WRITEBACK);
657         } while (bvec >= bio->bi_io_vec);
658
659         if (p->is_sync)
660                 complete(p->wait);
661         kfree(p);
662         bio_put(bio);
663 }
664
665 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
666 {
667         struct bio *bio;
668         struct bio_private *priv;
669 retry:
670         priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
671         if (!priv) {
672                 cond_resched();
673                 goto retry;
674         }
675
676         /* No failure on bio allocation */
677         bio = bio_alloc(GFP_NOIO, npages);
678         bio->bi_bdev = bdev;
679         bio->bi_private = priv;
680         return bio;
681 }
682
683 static void do_submit_bio(struct f2fs_sb_info *sbi,
684                                 enum page_type type, bool sync)
685 {
686         int rw = sync ? WRITE_SYNC : WRITE;
687         enum page_type btype = type > META ? META : type;
688
689         if (type >= META_FLUSH)
690                 rw = WRITE_FLUSH_FUA;
691
692         if (btype == META)
693                 rw |= REQ_META;
694
695         if (sbi->bio[btype]) {
696                 struct bio_private *p = sbi->bio[btype]->bi_private;
697                 p->sbi = sbi;
698                 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
699
700                 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
701
702                 if (type == META_FLUSH) {
703                         DECLARE_COMPLETION_ONSTACK(wait);
704                         p->is_sync = true;
705                         p->wait = &wait;
706                         submit_bio(rw, sbi->bio[btype]);
707                         wait_for_completion(&wait);
708                 } else {
709                         p->is_sync = false;
710                         submit_bio(rw, sbi->bio[btype]);
711                 }
712                 sbi->bio[btype] = NULL;
713         }
714 }
715
716 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
717 {
718         down_write(&sbi->bio_sem);
719         do_submit_bio(sbi, type, sync);
720         up_write(&sbi->bio_sem);
721 }
722
723 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
724                                 block_t blk_addr, enum page_type type)
725 {
726         struct block_device *bdev = sbi->sb->s_bdev;
727
728         verify_block_addr(sbi, blk_addr);
729
730         down_write(&sbi->bio_sem);
731
732         inc_page_count(sbi, F2FS_WRITEBACK);
733
734         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
735                 do_submit_bio(sbi, type, false);
736 alloc_new:
737         if (sbi->bio[type] == NULL) {
738                 sbi->bio[type] = f2fs_bio_alloc(bdev, max_hw_blocks(sbi));
739                 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
740                 /*
741                  * The end_io will be assigned at the sumbission phase.
742                  * Until then, let bio_add_page() merge consecutive IOs as much
743                  * as possible.
744                  */
745         }
746
747         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
748                                                         PAGE_CACHE_SIZE) {
749                 do_submit_bio(sbi, type, false);
750                 goto alloc_new;
751         }
752
753         sbi->last_block_in_bio[type] = blk_addr;
754
755         up_write(&sbi->bio_sem);
756         trace_f2fs_submit_write_page(page, blk_addr, type);
757 }
758
759 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
760 {
761         struct curseg_info *curseg = CURSEG_I(sbi, type);
762         if (curseg->next_blkoff < sbi->blocks_per_seg)
763                 return true;
764         return false;
765 }
766
767 static int __get_segment_type_2(struct page *page, enum page_type p_type)
768 {
769         if (p_type == DATA)
770                 return CURSEG_HOT_DATA;
771         else
772                 return CURSEG_HOT_NODE;
773 }
774
775 static int __get_segment_type_4(struct page *page, enum page_type p_type)
776 {
777         if (p_type == DATA) {
778                 struct inode *inode = page->mapping->host;
779
780                 if (S_ISDIR(inode->i_mode))
781                         return CURSEG_HOT_DATA;
782                 else
783                         return CURSEG_COLD_DATA;
784         } else {
785                 if (IS_DNODE(page) && !is_cold_node(page))
786                         return CURSEG_HOT_NODE;
787                 else
788                         return CURSEG_COLD_NODE;
789         }
790 }
791
792 static int __get_segment_type_6(struct page *page, enum page_type p_type)
793 {
794         if (p_type == DATA) {
795                 struct inode *inode = page->mapping->host;
796
797                 if (S_ISDIR(inode->i_mode))
798                         return CURSEG_HOT_DATA;
799                 else if (is_cold_data(page) || is_cold_file(inode))
800                         return CURSEG_COLD_DATA;
801                 else
802                         return CURSEG_WARM_DATA;
803         } else {
804                 if (IS_DNODE(page))
805                         return is_cold_node(page) ? CURSEG_WARM_NODE :
806                                                 CURSEG_HOT_NODE;
807                 else
808                         return CURSEG_COLD_NODE;
809         }
810 }
811
812 static int __get_segment_type(struct page *page, enum page_type p_type)
813 {
814         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
815         switch (sbi->active_logs) {
816         case 2:
817                 return __get_segment_type_2(page, p_type);
818         case 4:
819                 return __get_segment_type_4(page, p_type);
820         }
821         /* NR_CURSEG_TYPE(6) logs by default */
822         BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
823         return __get_segment_type_6(page, p_type);
824 }
825
826 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
827                         block_t old_blkaddr, block_t *new_blkaddr,
828                         struct f2fs_summary *sum, enum page_type p_type)
829 {
830         struct sit_info *sit_i = SIT_I(sbi);
831         struct curseg_info *curseg;
832         unsigned int old_cursegno;
833         int type;
834
835         type = __get_segment_type(page, p_type);
836         curseg = CURSEG_I(sbi, type);
837
838         mutex_lock(&curseg->curseg_mutex);
839
840         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
841         old_cursegno = curseg->segno;
842
843         /*
844          * __add_sum_entry should be resided under the curseg_mutex
845          * because, this function updates a summary entry in the
846          * current summary block.
847          */
848         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
849
850         mutex_lock(&sit_i->sentry_lock);
851         __refresh_next_blkoff(sbi, curseg);
852 #ifdef CONFIG_F2FS_STAT_FS
853         sbi->block_count[curseg->alloc_type]++;
854 #endif
855
856         /*
857          * SIT information should be updated before segment allocation,
858          * since SSR needs latest valid block information.
859          */
860         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
861
862         if (!__has_curseg_space(sbi, type))
863                 sit_i->s_ops->allocate_segment(sbi, type, false);
864
865         locate_dirty_segment(sbi, old_cursegno);
866         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
867         mutex_unlock(&sit_i->sentry_lock);
868
869         if (p_type == NODE)
870                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
871
872         /* writeout dirty page into bdev */
873         submit_write_page(sbi, page, *new_blkaddr, p_type);
874
875         mutex_unlock(&curseg->curseg_mutex);
876 }
877
878 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
879 {
880         set_page_writeback(page);
881         submit_write_page(sbi, page, page->index, META);
882 }
883
884 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
885                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
886 {
887         struct f2fs_summary sum;
888         set_summary(&sum, nid, 0, 0);
889         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
890 }
891
892 void write_data_page(struct inode *inode, struct page *page,
893                 struct dnode_of_data *dn, block_t old_blkaddr,
894                 block_t *new_blkaddr)
895 {
896         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
897         struct f2fs_summary sum;
898         struct node_info ni;
899
900         BUG_ON(old_blkaddr == NULL_ADDR);
901         get_node_info(sbi, dn->nid, &ni);
902         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
903
904         do_write_page(sbi, page, old_blkaddr,
905                         new_blkaddr, &sum, DATA);
906 }
907
908 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
909                                         block_t old_blk_addr)
910 {
911         submit_write_page(sbi, page, old_blk_addr, DATA);
912 }
913
914 void recover_data_page(struct f2fs_sb_info *sbi,
915                         struct page *page, struct f2fs_summary *sum,
916                         block_t old_blkaddr, block_t new_blkaddr)
917 {
918         struct sit_info *sit_i = SIT_I(sbi);
919         struct curseg_info *curseg;
920         unsigned int segno, old_cursegno;
921         struct seg_entry *se;
922         int type;
923
924         segno = GET_SEGNO(sbi, new_blkaddr);
925         se = get_seg_entry(sbi, segno);
926         type = se->type;
927
928         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
929                 if (old_blkaddr == NULL_ADDR)
930                         type = CURSEG_COLD_DATA;
931                 else
932                         type = CURSEG_WARM_DATA;
933         }
934         curseg = CURSEG_I(sbi, type);
935
936         mutex_lock(&curseg->curseg_mutex);
937         mutex_lock(&sit_i->sentry_lock);
938
939         old_cursegno = curseg->segno;
940
941         /* change the current segment */
942         if (segno != curseg->segno) {
943                 curseg->next_segno = segno;
944                 change_curseg(sbi, type, true);
945         }
946
947         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
948                                         (sbi->blocks_per_seg - 1);
949         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
950
951         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
952
953         locate_dirty_segment(sbi, old_cursegno);
954         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
955
956         mutex_unlock(&sit_i->sentry_lock);
957         mutex_unlock(&curseg->curseg_mutex);
958 }
959
960 void rewrite_node_page(struct f2fs_sb_info *sbi,
961                         struct page *page, struct f2fs_summary *sum,
962                         block_t old_blkaddr, block_t new_blkaddr)
963 {
964         struct sit_info *sit_i = SIT_I(sbi);
965         int type = CURSEG_WARM_NODE;
966         struct curseg_info *curseg;
967         unsigned int segno, old_cursegno;
968         block_t next_blkaddr = next_blkaddr_of_node(page);
969         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
970
971         curseg = CURSEG_I(sbi, type);
972
973         mutex_lock(&curseg->curseg_mutex);
974         mutex_lock(&sit_i->sentry_lock);
975
976         segno = GET_SEGNO(sbi, new_blkaddr);
977         old_cursegno = curseg->segno;
978
979         /* change the current segment */
980         if (segno != curseg->segno) {
981                 curseg->next_segno = segno;
982                 change_curseg(sbi, type, true);
983         }
984         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
985                                         (sbi->blocks_per_seg - 1);
986         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
987
988         /* change the current log to the next block addr in advance */
989         if (next_segno != segno) {
990                 curseg->next_segno = next_segno;
991                 change_curseg(sbi, type, true);
992         }
993         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
994                                         (sbi->blocks_per_seg - 1);
995
996         /* rewrite node page */
997         set_page_writeback(page);
998         submit_write_page(sbi, page, new_blkaddr, NODE);
999         f2fs_submit_bio(sbi, NODE, true);
1000         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1001
1002         locate_dirty_segment(sbi, old_cursegno);
1003         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1004
1005         mutex_unlock(&sit_i->sentry_lock);
1006         mutex_unlock(&curseg->curseg_mutex);
1007 }
1008
1009 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1010 {
1011         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1012         struct curseg_info *seg_i;
1013         unsigned char *kaddr;
1014         struct page *page;
1015         block_t start;
1016         int i, j, offset;
1017
1018         start = start_sum_block(sbi);
1019
1020         page = get_meta_page(sbi, start++);
1021         kaddr = (unsigned char *)page_address(page);
1022
1023         /* Step 1: restore nat cache */
1024         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1025         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1026
1027         /* Step 2: restore sit cache */
1028         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1029         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1030                                                 SUM_JOURNAL_SIZE);
1031         offset = 2 * SUM_JOURNAL_SIZE;
1032
1033         /* Step 3: restore summary entries */
1034         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1035                 unsigned short blk_off;
1036                 unsigned int segno;
1037
1038                 seg_i = CURSEG_I(sbi, i);
1039                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1040                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1041                 seg_i->next_segno = segno;
1042                 reset_curseg(sbi, i, 0);
1043                 seg_i->alloc_type = ckpt->alloc_type[i];
1044                 seg_i->next_blkoff = blk_off;
1045
1046                 if (seg_i->alloc_type == SSR)
1047                         blk_off = sbi->blocks_per_seg;
1048
1049                 for (j = 0; j < blk_off; j++) {
1050                         struct f2fs_summary *s;
1051                         s = (struct f2fs_summary *)(kaddr + offset);
1052                         seg_i->sum_blk->entries[j] = *s;
1053                         offset += SUMMARY_SIZE;
1054                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1055                                                 SUM_FOOTER_SIZE)
1056                                 continue;
1057
1058                         f2fs_put_page(page, 1);
1059                         page = NULL;
1060
1061                         page = get_meta_page(sbi, start++);
1062                         kaddr = (unsigned char *)page_address(page);
1063                         offset = 0;
1064                 }
1065         }
1066         f2fs_put_page(page, 1);
1067         return 0;
1068 }
1069
1070 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1071 {
1072         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1073         struct f2fs_summary_block *sum;
1074         struct curseg_info *curseg;
1075         struct page *new;
1076         unsigned short blk_off;
1077         unsigned int segno = 0;
1078         block_t blk_addr = 0;
1079
1080         /* get segment number and block addr */
1081         if (IS_DATASEG(type)) {
1082                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1083                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1084                                                         CURSEG_HOT_DATA]);
1085                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1086                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1087                 else
1088                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1089         } else {
1090                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1091                                                         CURSEG_HOT_NODE]);
1092                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1093                                                         CURSEG_HOT_NODE]);
1094                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1095                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1096                                                         type - CURSEG_HOT_NODE);
1097                 else
1098                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1099         }
1100
1101         new = get_meta_page(sbi, blk_addr);
1102         sum = (struct f2fs_summary_block *)page_address(new);
1103
1104         if (IS_NODESEG(type)) {
1105                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1106                         struct f2fs_summary *ns = &sum->entries[0];
1107                         int i;
1108                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1109                                 ns->version = 0;
1110                                 ns->ofs_in_node = 0;
1111                         }
1112                 } else {
1113                         if (restore_node_summary(sbi, segno, sum)) {
1114                                 f2fs_put_page(new, 1);
1115                                 return -EINVAL;
1116                         }
1117                 }
1118         }
1119
1120         /* set uncompleted segment to curseg */
1121         curseg = CURSEG_I(sbi, type);
1122         mutex_lock(&curseg->curseg_mutex);
1123         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1124         curseg->next_segno = segno;
1125         reset_curseg(sbi, type, 0);
1126         curseg->alloc_type = ckpt->alloc_type[type];
1127         curseg->next_blkoff = blk_off;
1128         mutex_unlock(&curseg->curseg_mutex);
1129         f2fs_put_page(new, 1);
1130         return 0;
1131 }
1132
1133 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1134 {
1135         int type = CURSEG_HOT_DATA;
1136
1137         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1138                 /* restore for compacted data summary */
1139                 if (read_compacted_summaries(sbi))
1140                         return -EINVAL;
1141                 type = CURSEG_HOT_NODE;
1142         }
1143
1144         for (; type <= CURSEG_COLD_NODE; type++)
1145                 if (read_normal_summaries(sbi, type))
1146                         return -EINVAL;
1147         return 0;
1148 }
1149
1150 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1151 {
1152         struct page *page;
1153         unsigned char *kaddr;
1154         struct f2fs_summary *summary;
1155         struct curseg_info *seg_i;
1156         int written_size = 0;
1157         int i, j;
1158
1159         page = grab_meta_page(sbi, blkaddr++);
1160         kaddr = (unsigned char *)page_address(page);
1161
1162         /* Step 1: write nat cache */
1163         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1164         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1165         written_size += SUM_JOURNAL_SIZE;
1166
1167         /* Step 2: write sit cache */
1168         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1169         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1170                                                 SUM_JOURNAL_SIZE);
1171         written_size += SUM_JOURNAL_SIZE;
1172
1173         set_page_dirty(page);
1174
1175         /* Step 3: write summary entries */
1176         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1177                 unsigned short blkoff;
1178                 seg_i = CURSEG_I(sbi, i);
1179                 if (sbi->ckpt->alloc_type[i] == SSR)
1180                         blkoff = sbi->blocks_per_seg;
1181                 else
1182                         blkoff = curseg_blkoff(sbi, i);
1183
1184                 for (j = 0; j < blkoff; j++) {
1185                         if (!page) {
1186                                 page = grab_meta_page(sbi, blkaddr++);
1187                                 kaddr = (unsigned char *)page_address(page);
1188                                 written_size = 0;
1189                         }
1190                         summary = (struct f2fs_summary *)(kaddr + written_size);
1191                         *summary = seg_i->sum_blk->entries[j];
1192                         written_size += SUMMARY_SIZE;
1193                         set_page_dirty(page);
1194
1195                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1196                                                         SUM_FOOTER_SIZE)
1197                                 continue;
1198
1199                         f2fs_put_page(page, 1);
1200                         page = NULL;
1201                 }
1202         }
1203         if (page)
1204                 f2fs_put_page(page, 1);
1205 }
1206
1207 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1208                                         block_t blkaddr, int type)
1209 {
1210         int i, end;
1211         if (IS_DATASEG(type))
1212                 end = type + NR_CURSEG_DATA_TYPE;
1213         else
1214                 end = type + NR_CURSEG_NODE_TYPE;
1215
1216         for (i = type; i < end; i++) {
1217                 struct curseg_info *sum = CURSEG_I(sbi, i);
1218                 mutex_lock(&sum->curseg_mutex);
1219                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1220                 mutex_unlock(&sum->curseg_mutex);
1221         }
1222 }
1223
1224 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1225 {
1226         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1227                 write_compacted_summaries(sbi, start_blk);
1228         else
1229                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1230 }
1231
1232 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1233 {
1234         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1235                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1236         return;
1237 }
1238
1239 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1240                                         unsigned int val, int alloc)
1241 {
1242         int i;
1243
1244         if (type == NAT_JOURNAL) {
1245                 for (i = 0; i < nats_in_cursum(sum); i++) {
1246                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1247                                 return i;
1248                 }
1249                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1250                         return update_nats_in_cursum(sum, 1);
1251         } else if (type == SIT_JOURNAL) {
1252                 for (i = 0; i < sits_in_cursum(sum); i++)
1253                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1254                                 return i;
1255                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1256                         return update_sits_in_cursum(sum, 1);
1257         }
1258         return -1;
1259 }
1260
1261 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1262                                         unsigned int segno)
1263 {
1264         struct sit_info *sit_i = SIT_I(sbi);
1265         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1266         block_t blk_addr = sit_i->sit_base_addr + offset;
1267
1268         check_seg_range(sbi, segno);
1269
1270         /* calculate sit block address */
1271         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1272                 blk_addr += sit_i->sit_blocks;
1273
1274         return get_meta_page(sbi, blk_addr);
1275 }
1276
1277 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1278                                         unsigned int start)
1279 {
1280         struct sit_info *sit_i = SIT_I(sbi);
1281         struct page *src_page, *dst_page;
1282         pgoff_t src_off, dst_off;
1283         void *src_addr, *dst_addr;
1284
1285         src_off = current_sit_addr(sbi, start);
1286         dst_off = next_sit_addr(sbi, src_off);
1287
1288         /* get current sit block page without lock */
1289         src_page = get_meta_page(sbi, src_off);
1290         dst_page = grab_meta_page(sbi, dst_off);
1291         BUG_ON(PageDirty(src_page));
1292
1293         src_addr = page_address(src_page);
1294         dst_addr = page_address(dst_page);
1295         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1296
1297         set_page_dirty(dst_page);
1298         f2fs_put_page(src_page, 1);
1299
1300         set_to_next_sit(sit_i, start);
1301
1302         return dst_page;
1303 }
1304
1305 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1306 {
1307         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1308         struct f2fs_summary_block *sum = curseg->sum_blk;
1309         int i;
1310
1311         /*
1312          * If the journal area in the current summary is full of sit entries,
1313          * all the sit entries will be flushed. Otherwise the sit entries
1314          * are not able to replace with newly hot sit entries.
1315          */
1316         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1317                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1318                         unsigned int segno;
1319                         segno = le32_to_cpu(segno_in_journal(sum, i));
1320                         __mark_sit_entry_dirty(sbi, segno);
1321                 }
1322                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1323                 return 1;
1324         }
1325         return 0;
1326 }
1327
1328 /*
1329  * CP calls this function, which flushes SIT entries including sit_journal,
1330  * and moves prefree segs to free segs.
1331  */
1332 void flush_sit_entries(struct f2fs_sb_info *sbi)
1333 {
1334         struct sit_info *sit_i = SIT_I(sbi);
1335         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1336         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1337         struct f2fs_summary_block *sum = curseg->sum_blk;
1338         unsigned long nsegs = TOTAL_SEGS(sbi);
1339         struct page *page = NULL;
1340         struct f2fs_sit_block *raw_sit = NULL;
1341         unsigned int start = 0, end = 0;
1342         unsigned int segno = -1;
1343         bool flushed;
1344
1345         mutex_lock(&curseg->curseg_mutex);
1346         mutex_lock(&sit_i->sentry_lock);
1347
1348         /*
1349          * "flushed" indicates whether sit entries in journal are flushed
1350          * to the SIT area or not.
1351          */
1352         flushed = flush_sits_in_journal(sbi);
1353
1354         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1355                 struct seg_entry *se = get_seg_entry(sbi, segno);
1356                 int sit_offset, offset;
1357
1358                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1359
1360                 if (flushed)
1361                         goto to_sit_page;
1362
1363                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1364                 if (offset >= 0) {
1365                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1366                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1367                         goto flush_done;
1368                 }
1369 to_sit_page:
1370                 if (!page || (start > segno) || (segno > end)) {
1371                         if (page) {
1372                                 f2fs_put_page(page, 1);
1373                                 page = NULL;
1374                         }
1375
1376                         start = START_SEGNO(sit_i, segno);
1377                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1378
1379                         /* read sit block that will be updated */
1380                         page = get_next_sit_page(sbi, start);
1381                         raw_sit = page_address(page);
1382                 }
1383
1384                 /* udpate entry in SIT block */
1385                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1386 flush_done:
1387                 __clear_bit(segno, bitmap);
1388                 sit_i->dirty_sentries--;
1389         }
1390         mutex_unlock(&sit_i->sentry_lock);
1391         mutex_unlock(&curseg->curseg_mutex);
1392
1393         /* writeout last modified SIT block */
1394         f2fs_put_page(page, 1);
1395
1396         set_prefree_as_free_segments(sbi);
1397 }
1398
1399 static int build_sit_info(struct f2fs_sb_info *sbi)
1400 {
1401         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1402         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1403         struct sit_info *sit_i;
1404         unsigned int sit_segs, start;
1405         char *src_bitmap, *dst_bitmap;
1406         unsigned int bitmap_size;
1407
1408         /* allocate memory for SIT information */
1409         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1410         if (!sit_i)
1411                 return -ENOMEM;
1412
1413         SM_I(sbi)->sit_info = sit_i;
1414
1415         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1416         if (!sit_i->sentries)
1417                 return -ENOMEM;
1418
1419         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1420         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1421         if (!sit_i->dirty_sentries_bitmap)
1422                 return -ENOMEM;
1423
1424         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1425                 sit_i->sentries[start].cur_valid_map
1426                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1427                 sit_i->sentries[start].ckpt_valid_map
1428                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1429                 if (!sit_i->sentries[start].cur_valid_map
1430                                 || !sit_i->sentries[start].ckpt_valid_map)
1431                         return -ENOMEM;
1432         }
1433
1434         if (sbi->segs_per_sec > 1) {
1435                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1436                                         sizeof(struct sec_entry));
1437                 if (!sit_i->sec_entries)
1438                         return -ENOMEM;
1439         }
1440
1441         /* get information related with SIT */
1442         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1443
1444         /* setup SIT bitmap from ckeckpoint pack */
1445         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1446         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1447
1448         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1449         if (!dst_bitmap)
1450                 return -ENOMEM;
1451
1452         /* init SIT information */
1453         sit_i->s_ops = &default_salloc_ops;
1454
1455         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1456         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1457         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1458         sit_i->sit_bitmap = dst_bitmap;
1459         sit_i->bitmap_size = bitmap_size;
1460         sit_i->dirty_sentries = 0;
1461         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1462         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1463         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1464         mutex_init(&sit_i->sentry_lock);
1465         return 0;
1466 }
1467
1468 static int build_free_segmap(struct f2fs_sb_info *sbi)
1469 {
1470         struct f2fs_sm_info *sm_info = SM_I(sbi);
1471         struct free_segmap_info *free_i;
1472         unsigned int bitmap_size, sec_bitmap_size;
1473
1474         /* allocate memory for free segmap information */
1475         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1476         if (!free_i)
1477                 return -ENOMEM;
1478
1479         SM_I(sbi)->free_info = free_i;
1480
1481         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1482         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1483         if (!free_i->free_segmap)
1484                 return -ENOMEM;
1485
1486         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1487         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1488         if (!free_i->free_secmap)
1489                 return -ENOMEM;
1490
1491         /* set all segments as dirty temporarily */
1492         memset(free_i->free_segmap, 0xff, bitmap_size);
1493         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1494
1495         /* init free segmap information */
1496         free_i->start_segno =
1497                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1498         free_i->free_segments = 0;
1499         free_i->free_sections = 0;
1500         rwlock_init(&free_i->segmap_lock);
1501         return 0;
1502 }
1503
1504 static int build_curseg(struct f2fs_sb_info *sbi)
1505 {
1506         struct curseg_info *array;
1507         int i;
1508
1509         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1510         if (!array)
1511                 return -ENOMEM;
1512
1513         SM_I(sbi)->curseg_array = array;
1514
1515         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1516                 mutex_init(&array[i].curseg_mutex);
1517                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1518                 if (!array[i].sum_blk)
1519                         return -ENOMEM;
1520                 array[i].segno = NULL_SEGNO;
1521                 array[i].next_blkoff = 0;
1522         }
1523         return restore_curseg_summaries(sbi);
1524 }
1525
1526 static void build_sit_entries(struct f2fs_sb_info *sbi)
1527 {
1528         struct sit_info *sit_i = SIT_I(sbi);
1529         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1530         struct f2fs_summary_block *sum = curseg->sum_blk;
1531         unsigned int start;
1532
1533         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1534                 struct seg_entry *se = &sit_i->sentries[start];
1535                 struct f2fs_sit_block *sit_blk;
1536                 struct f2fs_sit_entry sit;
1537                 struct page *page;
1538                 int i;
1539
1540                 mutex_lock(&curseg->curseg_mutex);
1541                 for (i = 0; i < sits_in_cursum(sum); i++) {
1542                         if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1543                                 sit = sit_in_journal(sum, i);
1544                                 mutex_unlock(&curseg->curseg_mutex);
1545                                 goto got_it;
1546                         }
1547                 }
1548                 mutex_unlock(&curseg->curseg_mutex);
1549                 page = get_current_sit_page(sbi, start);
1550                 sit_blk = (struct f2fs_sit_block *)page_address(page);
1551                 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1552                 f2fs_put_page(page, 1);
1553 got_it:
1554                 check_block_count(sbi, start, &sit);
1555                 seg_info_from_raw_sit(se, &sit);
1556                 if (sbi->segs_per_sec > 1) {
1557                         struct sec_entry *e = get_sec_entry(sbi, start);
1558                         e->valid_blocks += se->valid_blocks;
1559                 }
1560         }
1561 }
1562
1563 static void init_free_segmap(struct f2fs_sb_info *sbi)
1564 {
1565         unsigned int start;
1566         int type;
1567
1568         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1569                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1570                 if (!sentry->valid_blocks)
1571                         __set_free(sbi, start);
1572         }
1573
1574         /* set use the current segments */
1575         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1576                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1577                 __set_test_and_inuse(sbi, curseg_t->segno);
1578         }
1579 }
1580
1581 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1582 {
1583         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1584         struct free_segmap_info *free_i = FREE_I(sbi);
1585         unsigned int segno = 0, offset = 0;
1586         unsigned short valid_blocks;
1587
1588         while (segno < TOTAL_SEGS(sbi)) {
1589                 /* find dirty segment based on free segmap */
1590                 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset);
1591                 if (segno >= TOTAL_SEGS(sbi))
1592                         break;
1593                 offset = segno + 1;
1594                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1595                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1596                         continue;
1597                 mutex_lock(&dirty_i->seglist_lock);
1598                 __locate_dirty_segment(sbi, segno, DIRTY);
1599                 mutex_unlock(&dirty_i->seglist_lock);
1600         }
1601 }
1602
1603 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1604 {
1605         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1606         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1607
1608         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1609         if (!dirty_i->victim_secmap)
1610                 return -ENOMEM;
1611         return 0;
1612 }
1613
1614 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1615 {
1616         struct dirty_seglist_info *dirty_i;
1617         unsigned int bitmap_size, i;
1618
1619         /* allocate memory for dirty segments list information */
1620         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1621         if (!dirty_i)
1622                 return -ENOMEM;
1623
1624         SM_I(sbi)->dirty_info = dirty_i;
1625         mutex_init(&dirty_i->seglist_lock);
1626
1627         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1628
1629         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1630                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1631                 if (!dirty_i->dirty_segmap[i])
1632                         return -ENOMEM;
1633         }
1634
1635         init_dirty_segmap(sbi);
1636         return init_victim_secmap(sbi);
1637 }
1638
1639 /*
1640  * Update min, max modified time for cost-benefit GC algorithm
1641  */
1642 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1643 {
1644         struct sit_info *sit_i = SIT_I(sbi);
1645         unsigned int segno;
1646
1647         mutex_lock(&sit_i->sentry_lock);
1648
1649         sit_i->min_mtime = LLONG_MAX;
1650
1651         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1652                 unsigned int i;
1653                 unsigned long long mtime = 0;
1654
1655                 for (i = 0; i < sbi->segs_per_sec; i++)
1656                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1657
1658                 mtime = div_u64(mtime, sbi->segs_per_sec);
1659
1660                 if (sit_i->min_mtime > mtime)
1661                         sit_i->min_mtime = mtime;
1662         }
1663         sit_i->max_mtime = get_mtime(sbi);
1664         mutex_unlock(&sit_i->sentry_lock);
1665 }
1666
1667 int build_segment_manager(struct f2fs_sb_info *sbi)
1668 {
1669         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1670         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1671         struct f2fs_sm_info *sm_info;
1672         int err;
1673
1674         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1675         if (!sm_info)
1676                 return -ENOMEM;
1677
1678         /* init sm info */
1679         sbi->sm_info = sm_info;
1680         INIT_LIST_HEAD(&sm_info->wblist_head);
1681         spin_lock_init(&sm_info->wblist_lock);
1682         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1683         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1684         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1685         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1686         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1687         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1688         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1689
1690         err = build_sit_info(sbi);
1691         if (err)
1692                 return err;
1693         err = build_free_segmap(sbi);
1694         if (err)
1695                 return err;
1696         err = build_curseg(sbi);
1697         if (err)
1698                 return err;
1699
1700         /* reinit free segmap based on SIT */
1701         build_sit_entries(sbi);
1702
1703         init_free_segmap(sbi);
1704         err = build_dirty_segmap(sbi);
1705         if (err)
1706                 return err;
1707
1708         init_min_max_mtime(sbi);
1709         return 0;
1710 }
1711
1712 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1713                 enum dirty_type dirty_type)
1714 {
1715         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1716
1717         mutex_lock(&dirty_i->seglist_lock);
1718         kfree(dirty_i->dirty_segmap[dirty_type]);
1719         dirty_i->nr_dirty[dirty_type] = 0;
1720         mutex_unlock(&dirty_i->seglist_lock);
1721 }
1722
1723 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1724 {
1725         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1726         kfree(dirty_i->victim_secmap);
1727 }
1728
1729 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1730 {
1731         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1732         int i;
1733
1734         if (!dirty_i)
1735                 return;
1736
1737         /* discard pre-free/dirty segments list */
1738         for (i = 0; i < NR_DIRTY_TYPE; i++)
1739                 discard_dirty_segmap(sbi, i);
1740
1741         destroy_victim_secmap(sbi);
1742         SM_I(sbi)->dirty_info = NULL;
1743         kfree(dirty_i);
1744 }
1745
1746 static void destroy_curseg(struct f2fs_sb_info *sbi)
1747 {
1748         struct curseg_info *array = SM_I(sbi)->curseg_array;
1749         int i;
1750
1751         if (!array)
1752                 return;
1753         SM_I(sbi)->curseg_array = NULL;
1754         for (i = 0; i < NR_CURSEG_TYPE; i++)
1755                 kfree(array[i].sum_blk);
1756         kfree(array);
1757 }
1758
1759 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1760 {
1761         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1762         if (!free_i)
1763                 return;
1764         SM_I(sbi)->free_info = NULL;
1765         kfree(free_i->free_segmap);
1766         kfree(free_i->free_secmap);
1767         kfree(free_i);
1768 }
1769
1770 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1771 {
1772         struct sit_info *sit_i = SIT_I(sbi);
1773         unsigned int start;
1774
1775         if (!sit_i)
1776                 return;
1777
1778         if (sit_i->sentries) {
1779                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1780                         kfree(sit_i->sentries[start].cur_valid_map);
1781                         kfree(sit_i->sentries[start].ckpt_valid_map);
1782                 }
1783         }
1784         vfree(sit_i->sentries);
1785         vfree(sit_i->sec_entries);
1786         kfree(sit_i->dirty_sentries_bitmap);
1787
1788         SM_I(sbi)->sit_info = NULL;
1789         kfree(sit_i->sit_bitmap);
1790         kfree(sit_i);
1791 }
1792
1793 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1794 {
1795         struct f2fs_sm_info *sm_info = SM_I(sbi);
1796         destroy_dirty_segmap(sbi);
1797         destroy_curseg(sbi);
1798         destroy_free_segmap(sbi);
1799         destroy_sit_info(sbi);
1800         sbi->sm_info = NULL;
1801         kfree(sm_info);
1802 }