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