f2fs: rewrite f2fs_bio_alloc to make it simpler
[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/vmalloc.h>
16
17 #include "f2fs.h"
18 #include "segment.h"
19 #include "node.h"
20
21 static int need_to_flush(struct f2fs_sb_info *sbi)
22 {
23         unsigned int pages_per_sec = (1 << sbi->log_blocks_per_seg) *
24                         sbi->segs_per_sec;
25         int node_secs = ((get_pages(sbi, F2FS_DIRTY_NODES) + pages_per_sec - 1)
26                 >> sbi->log_blocks_per_seg) / sbi->segs_per_sec;
27         int dent_secs = ((get_pages(sbi, F2FS_DIRTY_DENTS) + pages_per_sec - 1)
28                 >> sbi->log_blocks_per_seg) / sbi->segs_per_sec;
29
30         if (sbi->por_doing)
31                 return 0;
32
33         if (free_sections(sbi) <= (node_secs + 2 * dent_secs +
34                                                 reserved_sections(sbi)))
35                 return 1;
36         return 0;
37 }
38
39 /*
40  * This function balances dirty node and dentry pages.
41  * In addition, it controls garbage collection.
42  */
43 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
44 {
45         struct writeback_control wbc = {
46                 .sync_mode = WB_SYNC_ALL,
47                 .nr_to_write = LONG_MAX,
48                 .for_reclaim = 0,
49         };
50
51         if (sbi->por_doing)
52                 return;
53
54         /*
55          * We should do checkpoint when there are so many dirty node pages
56          * with enough free segments. After then, we should do GC.
57          */
58         if (need_to_flush(sbi)) {
59                 sync_dirty_dir_inodes(sbi);
60                 sync_node_pages(sbi, 0, &wbc);
61         }
62
63         if (has_not_enough_free_secs(sbi)) {
64                 mutex_lock(&sbi->gc_mutex);
65                 f2fs_gc(sbi, 1);
66         }
67 }
68
69 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
70                 enum dirty_type dirty_type)
71 {
72         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
73
74         /* need not be added */
75         if (IS_CURSEG(sbi, segno))
76                 return;
77
78         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
79                 dirty_i->nr_dirty[dirty_type]++;
80
81         if (dirty_type == DIRTY) {
82                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
83                 dirty_type = sentry->type;
84                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
85                         dirty_i->nr_dirty[dirty_type]++;
86         }
87 }
88
89 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
90                 enum dirty_type dirty_type)
91 {
92         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
93
94         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
95                 dirty_i->nr_dirty[dirty_type]--;
96
97         if (dirty_type == DIRTY) {
98                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
99                 dirty_type = sentry->type;
100                 if (test_and_clear_bit(segno,
101                                         dirty_i->dirty_segmap[dirty_type]))
102                         dirty_i->nr_dirty[dirty_type]--;
103                 clear_bit(segno, dirty_i->victim_segmap[FG_GC]);
104                 clear_bit(segno, dirty_i->victim_segmap[BG_GC]);
105         }
106 }
107
108 /*
109  * Should not occur error such as -ENOMEM.
110  * Adding dirty entry into seglist is not critical operation.
111  * If a given segment is one of current working segments, it won't be added.
112  */
113 void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
114 {
115         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
116         unsigned short valid_blocks;
117
118         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
119                 return;
120
121         mutex_lock(&dirty_i->seglist_lock);
122
123         valid_blocks = get_valid_blocks(sbi, segno, 0);
124
125         if (valid_blocks == 0) {
126                 __locate_dirty_segment(sbi, segno, PRE);
127                 __remove_dirty_segment(sbi, segno, DIRTY);
128         } else if (valid_blocks < sbi->blocks_per_seg) {
129                 __locate_dirty_segment(sbi, segno, DIRTY);
130         } else {
131                 /* Recovery routine with SSR needs this */
132                 __remove_dirty_segment(sbi, segno, DIRTY);
133         }
134
135         mutex_unlock(&dirty_i->seglist_lock);
136         return;
137 }
138
139 /*
140  * Should call clear_prefree_segments after checkpoint is done.
141  */
142 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
143 {
144         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
145         unsigned int segno, offset = 0;
146         unsigned int total_segs = TOTAL_SEGS(sbi);
147
148         mutex_lock(&dirty_i->seglist_lock);
149         while (1) {
150                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
151                                 offset);
152                 if (segno >= total_segs)
153                         break;
154                 __set_test_and_free(sbi, segno);
155                 offset = segno + 1;
156         }
157         mutex_unlock(&dirty_i->seglist_lock);
158 }
159
160 void clear_prefree_segments(struct f2fs_sb_info *sbi)
161 {
162         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
163         unsigned int segno, offset = 0;
164         unsigned int total_segs = TOTAL_SEGS(sbi);
165
166         mutex_lock(&dirty_i->seglist_lock);
167         while (1) {
168                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
169                                 offset);
170                 if (segno >= total_segs)
171                         break;
172
173                 offset = segno + 1;
174                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
175                         dirty_i->nr_dirty[PRE]--;
176
177                 /* Let's use trim */
178                 if (test_opt(sbi, DISCARD))
179                         blkdev_issue_discard(sbi->sb->s_bdev,
180                                         START_BLOCK(sbi, segno) <<
181                                         sbi->log_sectors_per_block,
182                                         1 << (sbi->log_sectors_per_block +
183                                                 sbi->log_blocks_per_seg),
184                                         GFP_NOFS, 0);
185         }
186         mutex_unlock(&dirty_i->seglist_lock);
187 }
188
189 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
190 {
191         struct sit_info *sit_i = SIT_I(sbi);
192         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
193                 sit_i->dirty_sentries++;
194 }
195
196 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
197                                         unsigned int segno, int modified)
198 {
199         struct seg_entry *se = get_seg_entry(sbi, segno);
200         se->type = type;
201         if (modified)
202                 __mark_sit_entry_dirty(sbi, segno);
203 }
204
205 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
206 {
207         struct seg_entry *se;
208         unsigned int segno, offset;
209         long int new_vblocks;
210
211         segno = GET_SEGNO(sbi, blkaddr);
212
213         se = get_seg_entry(sbi, segno);
214         new_vblocks = se->valid_blocks + del;
215         offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
216
217         BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
218                                 (new_vblocks > sbi->blocks_per_seg)));
219
220         se->valid_blocks = new_vblocks;
221         se->mtime = get_mtime(sbi);
222         SIT_I(sbi)->max_mtime = se->mtime;
223
224         /* Update valid block bitmap */
225         if (del > 0) {
226                 if (f2fs_set_bit(offset, se->cur_valid_map))
227                         BUG();
228         } else {
229                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
230                         BUG();
231         }
232         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
233                 se->ckpt_valid_blocks += del;
234
235         __mark_sit_entry_dirty(sbi, segno);
236
237         /* update total number of valid blocks to be written in ckpt area */
238         SIT_I(sbi)->written_valid_blocks += del;
239
240         if (sbi->segs_per_sec > 1)
241                 get_sec_entry(sbi, segno)->valid_blocks += del;
242 }
243
244 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
245                         block_t old_blkaddr, block_t new_blkaddr)
246 {
247         update_sit_entry(sbi, new_blkaddr, 1);
248         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
249                 update_sit_entry(sbi, old_blkaddr, -1);
250 }
251
252 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
253 {
254         unsigned int segno = GET_SEGNO(sbi, addr);
255         struct sit_info *sit_i = SIT_I(sbi);
256
257         BUG_ON(addr == NULL_ADDR);
258         if (addr == NEW_ADDR)
259                 return;
260
261         /* add it into sit main buffer */
262         mutex_lock(&sit_i->sentry_lock);
263
264         update_sit_entry(sbi, addr, -1);
265
266         /* add it into dirty seglist */
267         locate_dirty_segment(sbi, segno);
268
269         mutex_unlock(&sit_i->sentry_lock);
270 }
271
272 /*
273  * This function should be resided under the curseg_mutex lock
274  */
275 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
276                 struct f2fs_summary *sum, unsigned short offset)
277 {
278         struct curseg_info *curseg = CURSEG_I(sbi, type);
279         void *addr = curseg->sum_blk;
280         addr += offset * sizeof(struct f2fs_summary);
281         memcpy(addr, sum, sizeof(struct f2fs_summary));
282         return;
283 }
284
285 /*
286  * Calculate the number of current summary pages for writing
287  */
288 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
289 {
290         int total_size_bytes = 0;
291         int valid_sum_count = 0;
292         int i, sum_space;
293
294         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
295                 if (sbi->ckpt->alloc_type[i] == SSR)
296                         valid_sum_count += sbi->blocks_per_seg;
297                 else
298                         valid_sum_count += curseg_blkoff(sbi, i);
299         }
300
301         total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
302                         + sizeof(struct nat_journal) + 2
303                         + sizeof(struct sit_journal) + 2;
304         sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
305         if (total_size_bytes < sum_space)
306                 return 1;
307         else if (total_size_bytes < 2 * sum_space)
308                 return 2;
309         return 3;
310 }
311
312 /*
313  * Caller should put this summary page
314  */
315 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
316 {
317         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
318 }
319
320 static void write_sum_page(struct f2fs_sb_info *sbi,
321                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
322 {
323         struct page *page = grab_meta_page(sbi, blk_addr);
324         void *kaddr = page_address(page);
325         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
326         set_page_dirty(page);
327         f2fs_put_page(page, 1);
328 }
329
330 static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi,
331                                         int ofs_unit, int type)
332 {
333         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
334         unsigned long *prefree_segmap = dirty_i->dirty_segmap[PRE];
335         unsigned int segno, next_segno, i;
336         int ofs = 0;
337
338         /*
339          * If there is not enough reserved sections,
340          * we should not reuse prefree segments.
341          */
342         if (has_not_enough_free_secs(sbi))
343                 return NULL_SEGNO;
344
345         /*
346          * NODE page should not reuse prefree segment,
347          * since those information is used for SPOR.
348          */
349         if (IS_NODESEG(type))
350                 return NULL_SEGNO;
351 next:
352         segno = find_next_bit(prefree_segmap, TOTAL_SEGS(sbi), ofs++);
353         ofs = ((segno / ofs_unit) * ofs_unit) + ofs_unit;
354         if (segno < TOTAL_SEGS(sbi)) {
355                 /* skip intermediate segments in a section */
356                 if (segno % ofs_unit)
357                         goto next;
358
359                 /* skip if whole section is not prefree */
360                 next_segno = find_next_zero_bit(prefree_segmap,
361                                                 TOTAL_SEGS(sbi), segno + 1);
362                 if (next_segno - segno < ofs_unit)
363                         goto next;
364
365                 /* skip if whole section was not free at the last checkpoint */
366                 for (i = 0; i < ofs_unit; i++)
367                         if (get_seg_entry(sbi, segno)->ckpt_valid_blocks)
368                                 goto next;
369                 return segno;
370         }
371         return NULL_SEGNO;
372 }
373
374 /*
375  * Find a new segment from the free segments bitmap to right order
376  * This function should be returned with success, otherwise BUG
377  */
378 static void get_new_segment(struct f2fs_sb_info *sbi,
379                         unsigned int *newseg, bool new_sec, int dir)
380 {
381         struct free_segmap_info *free_i = FREE_I(sbi);
382         unsigned int total_secs = sbi->total_sections;
383         unsigned int segno, secno, zoneno;
384         unsigned int total_zones = sbi->total_sections / 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 < TOTAL_SEGS(sbi))
398                         goto got_it;
399         }
400 find_other_zone:
401         secno = find_next_zero_bit(free_i->free_secmap, total_secs, hint);
402         if (secno >= total_secs) {
403                 if (dir == ALLOC_RIGHT) {
404                         secno = find_next_zero_bit(free_i->free_secmap,
405                                                 total_secs, 0);
406                         BUG_ON(secno >= total_secs);
407                 } else {
408                         go_left = 1;
409                         left_start = hint - 1;
410                 }
411         }
412         if (go_left == 0)
413                 goto skip_left;
414
415         while (test_bit(left_start, free_i->free_secmap)) {
416                 if (left_start > 0) {
417                         left_start--;
418                         continue;
419                 }
420                 left_start = find_next_zero_bit(free_i->free_secmap,
421                                                 total_secs, 0);
422                 BUG_ON(left_start >= total_secs);
423                 break;
424         }
425         secno = left_start;
426 skip_left:
427         hint = secno;
428         segno = secno * sbi->segs_per_sec;
429         zoneno = secno / sbi->secs_per_zone;
430
431         /* give up on finding another zone */
432         if (!init)
433                 goto got_it;
434         if (sbi->secs_per_zone == 1)
435                 goto got_it;
436         if (zoneno == old_zoneno)
437                 goto got_it;
438         if (dir == ALLOC_LEFT) {
439                 if (!go_left && zoneno + 1 >= total_zones)
440                         goto got_it;
441                 if (go_left && zoneno == 0)
442                         goto got_it;
443         }
444         for (i = 0; i < NR_CURSEG_TYPE; i++)
445                 if (CURSEG_I(sbi, i)->zone == zoneno)
446                         break;
447
448         if (i < NR_CURSEG_TYPE) {
449                 /* zone is in user, try another */
450                 if (go_left)
451                         hint = zoneno * sbi->secs_per_zone - 1;
452                 else if (zoneno + 1 >= total_zones)
453                         hint = 0;
454                 else
455                         hint = (zoneno + 1) * sbi->secs_per_zone;
456                 init = false;
457                 goto find_other_zone;
458         }
459 got_it:
460         /* set it as dirty segment in free segmap */
461         BUG_ON(test_bit(segno, free_i->free_segmap));
462         __set_inuse(sbi, segno);
463         *newseg = segno;
464         write_unlock(&free_i->segmap_lock);
465 }
466
467 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
468 {
469         struct curseg_info *curseg = CURSEG_I(sbi, type);
470         struct summary_footer *sum_footer;
471
472         curseg->segno = curseg->next_segno;
473         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
474         curseg->next_blkoff = 0;
475         curseg->next_segno = NULL_SEGNO;
476
477         sum_footer = &(curseg->sum_blk->footer);
478         memset(sum_footer, 0, sizeof(struct summary_footer));
479         if (IS_DATASEG(type))
480                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
481         if (IS_NODESEG(type))
482                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
483         __set_sit_entry_type(sbi, type, curseg->segno, modified);
484 }
485
486 /*
487  * Allocate a current working segment.
488  * This function always allocates a free segment in LFS manner.
489  */
490 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
491 {
492         struct curseg_info *curseg = CURSEG_I(sbi, type);
493         unsigned int segno = curseg->segno;
494         int dir = ALLOC_LEFT;
495
496         write_sum_page(sbi, curseg->sum_blk,
497                                 GET_SUM_BLOCK(sbi, curseg->segno));
498         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
499                 dir = ALLOC_RIGHT;
500
501         if (test_opt(sbi, NOHEAP))
502                 dir = ALLOC_RIGHT;
503
504         get_new_segment(sbi, &segno, new_sec, dir);
505         curseg->next_segno = segno;
506         reset_curseg(sbi, type, 1);
507         curseg->alloc_type = LFS;
508 }
509
510 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
511                         struct curseg_info *seg, block_t start)
512 {
513         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
514         block_t ofs;
515         for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
516                 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
517                         && !f2fs_test_bit(ofs, se->cur_valid_map))
518                         break;
519         }
520         seg->next_blkoff = ofs;
521 }
522
523 /*
524  * If a segment is written by LFS manner, next block offset is just obtained
525  * by increasing the current block offset. However, if a segment is written by
526  * SSR manner, next block offset obtained by calling __next_free_blkoff
527  */
528 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
529                                 struct curseg_info *seg)
530 {
531         if (seg->alloc_type == SSR)
532                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
533         else
534                 seg->next_blkoff++;
535 }
536
537 /*
538  * This function always allocates a used segment (from dirty seglist) by SSR
539  * manner, so it should recover the existing segment information of valid blocks
540  */
541 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
542 {
543         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
544         struct curseg_info *curseg = CURSEG_I(sbi, type);
545         unsigned int new_segno = curseg->next_segno;
546         struct f2fs_summary_block *sum_node;
547         struct page *sum_page;
548
549         write_sum_page(sbi, curseg->sum_blk,
550                                 GET_SUM_BLOCK(sbi, curseg->segno));
551         __set_test_and_inuse(sbi, new_segno);
552
553         mutex_lock(&dirty_i->seglist_lock);
554         __remove_dirty_segment(sbi, new_segno, PRE);
555         __remove_dirty_segment(sbi, new_segno, DIRTY);
556         mutex_unlock(&dirty_i->seglist_lock);
557
558         reset_curseg(sbi, type, 1);
559         curseg->alloc_type = SSR;
560         __next_free_blkoff(sbi, curseg, 0);
561
562         if (reuse) {
563                 sum_page = get_sum_page(sbi, new_segno);
564                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
565                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
566                 f2fs_put_page(sum_page, 1);
567         }
568 }
569
570 /*
571  * flush out current segment and replace it with new segment
572  * This function should be returned with success, otherwise BUG
573  */
574 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
575                                                 int type, bool force)
576 {
577         struct curseg_info *curseg = CURSEG_I(sbi, type);
578         unsigned int ofs_unit;
579
580         if (force) {
581                 new_curseg(sbi, type, true);
582                 goto out;
583         }
584
585         ofs_unit = need_SSR(sbi) ? 1 : sbi->segs_per_sec;
586         curseg->next_segno = check_prefree_segments(sbi, ofs_unit, type);
587
588         if (curseg->next_segno != NULL_SEGNO)
589                 change_curseg(sbi, type, false);
590         else if (type == CURSEG_WARM_NODE)
591                 new_curseg(sbi, type, false);
592         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
593                 change_curseg(sbi, type, true);
594         else
595                 new_curseg(sbi, type, false);
596 out:
597         sbi->segment_count[curseg->alloc_type]++;
598 }
599
600 void allocate_new_segments(struct f2fs_sb_info *sbi)
601 {
602         struct curseg_info *curseg;
603         unsigned int old_curseg;
604         int i;
605
606         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
607                 curseg = CURSEG_I(sbi, i);
608                 old_curseg = curseg->segno;
609                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
610                 locate_dirty_segment(sbi, old_curseg);
611         }
612 }
613
614 static const struct segment_allocation default_salloc_ops = {
615         .allocate_segment = allocate_segment_by_default,
616 };
617
618 static void f2fs_end_io_write(struct bio *bio, int err)
619 {
620         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
621         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
622         struct bio_private *p = bio->bi_private;
623
624         do {
625                 struct page *page = bvec->bv_page;
626
627                 if (--bvec >= bio->bi_io_vec)
628                         prefetchw(&bvec->bv_page->flags);
629                 if (!uptodate) {
630                         SetPageError(page);
631                         if (page->mapping)
632                                 set_bit(AS_EIO, &page->mapping->flags);
633                         set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
634                         set_page_dirty(page);
635                 }
636                 end_page_writeback(page);
637                 dec_page_count(p->sbi, F2FS_WRITEBACK);
638         } while (bvec >= bio->bi_io_vec);
639
640         if (p->is_sync)
641                 complete(p->wait);
642         kfree(p);
643         bio_put(bio);
644 }
645
646 struct bio *f2fs_bio_alloc(struct block_device *bdev, sector_t first_sector,
647                                         int nr_vecs, gfp_t gfp_flags)
648 {
649         struct bio *bio;
650
651         /* allocate new bio */
652         bio = bio_alloc(gfp_flags, nr_vecs);
653
654         bio->bi_bdev = bdev;
655         bio->bi_sector = first_sector;
656 retry:
657         bio->bi_private = kmalloc(sizeof(struct bio_private),
658                                         GFP_NOFS | __GFP_HIGH);
659         if (!bio->bi_private) {
660                 cond_resched();
661                 goto retry;
662         }
663         return bio;
664 }
665
666 static void do_submit_bio(struct f2fs_sb_info *sbi,
667                                 enum page_type type, bool sync)
668 {
669         int rw = sync ? WRITE_SYNC : WRITE;
670         enum page_type btype = type > META ? META : type;
671
672         if (type >= META_FLUSH)
673                 rw = WRITE_FLUSH_FUA;
674
675         if (sbi->bio[btype]) {
676                 struct bio_private *p = sbi->bio[btype]->bi_private;
677                 p->sbi = sbi;
678                 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
679                 if (type == META_FLUSH) {
680                         DECLARE_COMPLETION_ONSTACK(wait);
681                         p->is_sync = true;
682                         p->wait = &wait;
683                         submit_bio(rw, sbi->bio[btype]);
684                         wait_for_completion(&wait);
685                 } else {
686                         p->is_sync = false;
687                         submit_bio(rw, sbi->bio[btype]);
688                 }
689                 sbi->bio[btype] = NULL;
690         }
691 }
692
693 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
694 {
695         down_write(&sbi->bio_sem);
696         do_submit_bio(sbi, type, sync);
697         up_write(&sbi->bio_sem);
698 }
699
700 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
701                                 block_t blk_addr, enum page_type type)
702 {
703         struct block_device *bdev = sbi->sb->s_bdev;
704
705         verify_block_addr(sbi, blk_addr);
706
707         down_write(&sbi->bio_sem);
708
709         inc_page_count(sbi, F2FS_WRITEBACK);
710
711         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
712                 do_submit_bio(sbi, type, false);
713 alloc_new:
714         if (sbi->bio[type] == NULL)
715                 sbi->bio[type] = f2fs_bio_alloc(bdev,
716                                 blk_addr << (sbi->log_blocksize - 9),
717                                 bio_get_nr_vecs(bdev), GFP_NOFS | __GFP_HIGH);
718
719         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
720                                                         PAGE_CACHE_SIZE) {
721                 do_submit_bio(sbi, type, false);
722                 goto alloc_new;
723         }
724
725         sbi->last_block_in_bio[type] = blk_addr;
726
727         up_write(&sbi->bio_sem);
728 }
729
730 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
731 {
732         struct curseg_info *curseg = CURSEG_I(sbi, type);
733         if (curseg->next_blkoff < sbi->blocks_per_seg)
734                 return true;
735         return false;
736 }
737
738 static int __get_segment_type_2(struct page *page, enum page_type p_type)
739 {
740         if (p_type == DATA)
741                 return CURSEG_HOT_DATA;
742         else
743                 return CURSEG_HOT_NODE;
744 }
745
746 static int __get_segment_type_4(struct page *page, enum page_type p_type)
747 {
748         if (p_type == DATA) {
749                 struct inode *inode = page->mapping->host;
750
751                 if (S_ISDIR(inode->i_mode))
752                         return CURSEG_HOT_DATA;
753                 else
754                         return CURSEG_COLD_DATA;
755         } else {
756                 if (IS_DNODE(page) && !is_cold_node(page))
757                         return CURSEG_HOT_NODE;
758                 else
759                         return CURSEG_COLD_NODE;
760         }
761 }
762
763 static int __get_segment_type_6(struct page *page, enum page_type p_type)
764 {
765         if (p_type == DATA) {
766                 struct inode *inode = page->mapping->host;
767
768                 if (S_ISDIR(inode->i_mode))
769                         return CURSEG_HOT_DATA;
770                 else if (is_cold_data(page) || is_cold_file(inode))
771                         return CURSEG_COLD_DATA;
772                 else
773                         return CURSEG_WARM_DATA;
774         } else {
775                 if (IS_DNODE(page))
776                         return is_cold_node(page) ? CURSEG_WARM_NODE :
777                                                 CURSEG_HOT_NODE;
778                 else
779                         return CURSEG_COLD_NODE;
780         }
781 }
782
783 static int __get_segment_type(struct page *page, enum page_type p_type)
784 {
785         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
786         switch (sbi->active_logs) {
787         case 2:
788                 return __get_segment_type_2(page, p_type);
789         case 4:
790                 return __get_segment_type_4(page, p_type);
791         case 6:
792                 return __get_segment_type_6(page, p_type);
793         default:
794                 BUG();
795         }
796 }
797
798 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
799                         block_t old_blkaddr, block_t *new_blkaddr,
800                         struct f2fs_summary *sum, enum page_type p_type)
801 {
802         struct sit_info *sit_i = SIT_I(sbi);
803         struct curseg_info *curseg;
804         unsigned int old_cursegno;
805         int type;
806
807         type = __get_segment_type(page, p_type);
808         curseg = CURSEG_I(sbi, type);
809
810         mutex_lock(&curseg->curseg_mutex);
811
812         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
813         old_cursegno = curseg->segno;
814
815         /*
816          * __add_sum_entry should be resided under the curseg_mutex
817          * because, this function updates a summary entry in the
818          * current summary block.
819          */
820         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
821
822         mutex_lock(&sit_i->sentry_lock);
823         __refresh_next_blkoff(sbi, curseg);
824         sbi->block_count[curseg->alloc_type]++;
825
826         /*
827          * SIT information should be updated before segment allocation,
828          * since SSR needs latest valid block information.
829          */
830         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
831
832         if (!__has_curseg_space(sbi, type))
833                 sit_i->s_ops->allocate_segment(sbi, type, false);
834
835         locate_dirty_segment(sbi, old_cursegno);
836         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
837         mutex_unlock(&sit_i->sentry_lock);
838
839         if (p_type == NODE)
840                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
841
842         /* writeout dirty page into bdev */
843         submit_write_page(sbi, page, *new_blkaddr, p_type);
844
845         mutex_unlock(&curseg->curseg_mutex);
846 }
847
848 int write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
849                         struct writeback_control *wbc)
850 {
851         if (wbc->for_reclaim)
852                 return AOP_WRITEPAGE_ACTIVATE;
853
854         set_page_writeback(page);
855         submit_write_page(sbi, page, page->index, META);
856         return 0;
857 }
858
859 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
860                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
861 {
862         struct f2fs_summary sum;
863         set_summary(&sum, nid, 0, 0);
864         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
865 }
866
867 void write_data_page(struct inode *inode, struct page *page,
868                 struct dnode_of_data *dn, block_t old_blkaddr,
869                 block_t *new_blkaddr)
870 {
871         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
872         struct f2fs_summary sum;
873         struct node_info ni;
874
875         BUG_ON(old_blkaddr == NULL_ADDR);
876         get_node_info(sbi, dn->nid, &ni);
877         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
878
879         do_write_page(sbi, page, old_blkaddr,
880                         new_blkaddr, &sum, DATA);
881 }
882
883 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
884                                         block_t old_blk_addr)
885 {
886         submit_write_page(sbi, page, old_blk_addr, DATA);
887 }
888
889 void recover_data_page(struct f2fs_sb_info *sbi,
890                         struct page *page, struct f2fs_summary *sum,
891                         block_t old_blkaddr, block_t new_blkaddr)
892 {
893         struct sit_info *sit_i = SIT_I(sbi);
894         struct curseg_info *curseg;
895         unsigned int segno, old_cursegno;
896         struct seg_entry *se;
897         int type;
898
899         segno = GET_SEGNO(sbi, new_blkaddr);
900         se = get_seg_entry(sbi, segno);
901         type = se->type;
902
903         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
904                 if (old_blkaddr == NULL_ADDR)
905                         type = CURSEG_COLD_DATA;
906                 else
907                         type = CURSEG_WARM_DATA;
908         }
909         curseg = CURSEG_I(sbi, type);
910
911         mutex_lock(&curseg->curseg_mutex);
912         mutex_lock(&sit_i->sentry_lock);
913
914         old_cursegno = curseg->segno;
915
916         /* change the current segment */
917         if (segno != curseg->segno) {
918                 curseg->next_segno = segno;
919                 change_curseg(sbi, type, true);
920         }
921
922         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
923                                         (sbi->blocks_per_seg - 1);
924         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
925
926         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
927
928         locate_dirty_segment(sbi, old_cursegno);
929         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
930
931         mutex_unlock(&sit_i->sentry_lock);
932         mutex_unlock(&curseg->curseg_mutex);
933 }
934
935 void rewrite_node_page(struct f2fs_sb_info *sbi,
936                         struct page *page, struct f2fs_summary *sum,
937                         block_t old_blkaddr, block_t new_blkaddr)
938 {
939         struct sit_info *sit_i = SIT_I(sbi);
940         int type = CURSEG_WARM_NODE;
941         struct curseg_info *curseg;
942         unsigned int segno, old_cursegno;
943         block_t next_blkaddr = next_blkaddr_of_node(page);
944         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
945
946         curseg = CURSEG_I(sbi, type);
947
948         mutex_lock(&curseg->curseg_mutex);
949         mutex_lock(&sit_i->sentry_lock);
950
951         segno = GET_SEGNO(sbi, new_blkaddr);
952         old_cursegno = curseg->segno;
953
954         /* change the current segment */
955         if (segno != curseg->segno) {
956                 curseg->next_segno = segno;
957                 change_curseg(sbi, type, true);
958         }
959         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
960                                         (sbi->blocks_per_seg - 1);
961         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
962
963         /* change the current log to the next block addr in advance */
964         if (next_segno != segno) {
965                 curseg->next_segno = next_segno;
966                 change_curseg(sbi, type, true);
967         }
968         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
969                                         (sbi->blocks_per_seg - 1);
970
971         /* rewrite node page */
972         set_page_writeback(page);
973         submit_write_page(sbi, page, new_blkaddr, NODE);
974         f2fs_submit_bio(sbi, NODE, true);
975         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
976
977         locate_dirty_segment(sbi, old_cursegno);
978         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
979
980         mutex_unlock(&sit_i->sentry_lock);
981         mutex_unlock(&curseg->curseg_mutex);
982 }
983
984 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
985 {
986         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
987         struct curseg_info *seg_i;
988         unsigned char *kaddr;
989         struct page *page;
990         block_t start;
991         int i, j, offset;
992
993         start = start_sum_block(sbi);
994
995         page = get_meta_page(sbi, start++);
996         kaddr = (unsigned char *)page_address(page);
997
998         /* Step 1: restore nat cache */
999         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1000         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1001
1002         /* Step 2: restore sit cache */
1003         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1004         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1005                                                 SUM_JOURNAL_SIZE);
1006         offset = 2 * SUM_JOURNAL_SIZE;
1007
1008         /* Step 3: restore summary entries */
1009         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1010                 unsigned short blk_off;
1011                 unsigned int segno;
1012
1013                 seg_i = CURSEG_I(sbi, i);
1014                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1015                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1016                 seg_i->next_segno = segno;
1017                 reset_curseg(sbi, i, 0);
1018                 seg_i->alloc_type = ckpt->alloc_type[i];
1019                 seg_i->next_blkoff = blk_off;
1020
1021                 if (seg_i->alloc_type == SSR)
1022                         blk_off = sbi->blocks_per_seg;
1023
1024                 for (j = 0; j < blk_off; j++) {
1025                         struct f2fs_summary *s;
1026                         s = (struct f2fs_summary *)(kaddr + offset);
1027                         seg_i->sum_blk->entries[j] = *s;
1028                         offset += SUMMARY_SIZE;
1029                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1030                                                 SUM_FOOTER_SIZE)
1031                                 continue;
1032
1033                         f2fs_put_page(page, 1);
1034                         page = NULL;
1035
1036                         page = get_meta_page(sbi, start++);
1037                         kaddr = (unsigned char *)page_address(page);
1038                         offset = 0;
1039                 }
1040         }
1041         f2fs_put_page(page, 1);
1042         return 0;
1043 }
1044
1045 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1046 {
1047         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1048         struct f2fs_summary_block *sum;
1049         struct curseg_info *curseg;
1050         struct page *new;
1051         unsigned short blk_off;
1052         unsigned int segno = 0;
1053         block_t blk_addr = 0;
1054
1055         /* get segment number and block addr */
1056         if (IS_DATASEG(type)) {
1057                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1058                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1059                                                         CURSEG_HOT_DATA]);
1060                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1061                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1062                 else
1063                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1064         } else {
1065                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1066                                                         CURSEG_HOT_NODE]);
1067                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1068                                                         CURSEG_HOT_NODE]);
1069                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1070                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1071                                                         type - CURSEG_HOT_NODE);
1072                 else
1073                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1074         }
1075
1076         new = get_meta_page(sbi, blk_addr);
1077         sum = (struct f2fs_summary_block *)page_address(new);
1078
1079         if (IS_NODESEG(type)) {
1080                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1081                         struct f2fs_summary *ns = &sum->entries[0];
1082                         int i;
1083                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1084                                 ns->version = 0;
1085                                 ns->ofs_in_node = 0;
1086                         }
1087                 } else {
1088                         if (restore_node_summary(sbi, segno, sum)) {
1089                                 f2fs_put_page(new, 1);
1090                                 return -EINVAL;
1091                         }
1092                 }
1093         }
1094
1095         /* set uncompleted segment to curseg */
1096         curseg = CURSEG_I(sbi, type);
1097         mutex_lock(&curseg->curseg_mutex);
1098         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1099         curseg->next_segno = segno;
1100         reset_curseg(sbi, type, 0);
1101         curseg->alloc_type = ckpt->alloc_type[type];
1102         curseg->next_blkoff = blk_off;
1103         mutex_unlock(&curseg->curseg_mutex);
1104         f2fs_put_page(new, 1);
1105         return 0;
1106 }
1107
1108 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1109 {
1110         int type = CURSEG_HOT_DATA;
1111
1112         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1113                 /* restore for compacted data summary */
1114                 if (read_compacted_summaries(sbi))
1115                         return -EINVAL;
1116                 type = CURSEG_HOT_NODE;
1117         }
1118
1119         for (; type <= CURSEG_COLD_NODE; type++)
1120                 if (read_normal_summaries(sbi, type))
1121                         return -EINVAL;
1122         return 0;
1123 }
1124
1125 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1126 {
1127         struct page *page;
1128         unsigned char *kaddr;
1129         struct f2fs_summary *summary;
1130         struct curseg_info *seg_i;
1131         int written_size = 0;
1132         int i, j;
1133
1134         page = grab_meta_page(sbi, blkaddr++);
1135         kaddr = (unsigned char *)page_address(page);
1136
1137         /* Step 1: write nat cache */
1138         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1139         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1140         written_size += SUM_JOURNAL_SIZE;
1141
1142         /* Step 2: write sit cache */
1143         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1144         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1145                                                 SUM_JOURNAL_SIZE);
1146         written_size += SUM_JOURNAL_SIZE;
1147
1148         set_page_dirty(page);
1149
1150         /* Step 3: write summary entries */
1151         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1152                 unsigned short blkoff;
1153                 seg_i = CURSEG_I(sbi, i);
1154                 if (sbi->ckpt->alloc_type[i] == SSR)
1155                         blkoff = sbi->blocks_per_seg;
1156                 else
1157                         blkoff = curseg_blkoff(sbi, i);
1158
1159                 for (j = 0; j < blkoff; j++) {
1160                         if (!page) {
1161                                 page = grab_meta_page(sbi, blkaddr++);
1162                                 kaddr = (unsigned char *)page_address(page);
1163                                 written_size = 0;
1164                         }
1165                         summary = (struct f2fs_summary *)(kaddr + written_size);
1166                         *summary = seg_i->sum_blk->entries[j];
1167                         written_size += SUMMARY_SIZE;
1168                         set_page_dirty(page);
1169
1170                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1171                                                         SUM_FOOTER_SIZE)
1172                                 continue;
1173
1174                         f2fs_put_page(page, 1);
1175                         page = NULL;
1176                 }
1177         }
1178         if (page)
1179                 f2fs_put_page(page, 1);
1180 }
1181
1182 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1183                                         block_t blkaddr, int type)
1184 {
1185         int i, end;
1186         if (IS_DATASEG(type))
1187                 end = type + NR_CURSEG_DATA_TYPE;
1188         else
1189                 end = type + NR_CURSEG_NODE_TYPE;
1190
1191         for (i = type; i < end; i++) {
1192                 struct curseg_info *sum = CURSEG_I(sbi, i);
1193                 mutex_lock(&sum->curseg_mutex);
1194                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1195                 mutex_unlock(&sum->curseg_mutex);
1196         }
1197 }
1198
1199 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1200 {
1201         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1202                 write_compacted_summaries(sbi, start_blk);
1203         else
1204                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1205 }
1206
1207 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1208 {
1209         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1210                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1211         return;
1212 }
1213
1214 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1215                                         unsigned int val, int alloc)
1216 {
1217         int i;
1218
1219         if (type == NAT_JOURNAL) {
1220                 for (i = 0; i < nats_in_cursum(sum); i++) {
1221                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1222                                 return i;
1223                 }
1224                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1225                         return update_nats_in_cursum(sum, 1);
1226         } else if (type == SIT_JOURNAL) {
1227                 for (i = 0; i < sits_in_cursum(sum); i++)
1228                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1229                                 return i;
1230                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1231                         return update_sits_in_cursum(sum, 1);
1232         }
1233         return -1;
1234 }
1235
1236 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1237                                         unsigned int segno)
1238 {
1239         struct sit_info *sit_i = SIT_I(sbi);
1240         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1241         block_t blk_addr = sit_i->sit_base_addr + offset;
1242
1243         check_seg_range(sbi, segno);
1244
1245         /* calculate sit block address */
1246         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1247                 blk_addr += sit_i->sit_blocks;
1248
1249         return get_meta_page(sbi, blk_addr);
1250 }
1251
1252 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1253                                         unsigned int start)
1254 {
1255         struct sit_info *sit_i = SIT_I(sbi);
1256         struct page *src_page, *dst_page;
1257         pgoff_t src_off, dst_off;
1258         void *src_addr, *dst_addr;
1259
1260         src_off = current_sit_addr(sbi, start);
1261         dst_off = next_sit_addr(sbi, src_off);
1262
1263         /* get current sit block page without lock */
1264         src_page = get_meta_page(sbi, src_off);
1265         dst_page = grab_meta_page(sbi, dst_off);
1266         BUG_ON(PageDirty(src_page));
1267
1268         src_addr = page_address(src_page);
1269         dst_addr = page_address(dst_page);
1270         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1271
1272         set_page_dirty(dst_page);
1273         f2fs_put_page(src_page, 1);
1274
1275         set_to_next_sit(sit_i, start);
1276
1277         return dst_page;
1278 }
1279
1280 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1281 {
1282         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1283         struct f2fs_summary_block *sum = curseg->sum_blk;
1284         int i;
1285
1286         /*
1287          * If the journal area in the current summary is full of sit entries,
1288          * all the sit entries will be flushed. Otherwise the sit entries
1289          * are not able to replace with newly hot sit entries.
1290          */
1291         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1292                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1293                         unsigned int segno;
1294                         segno = le32_to_cpu(segno_in_journal(sum, i));
1295                         __mark_sit_entry_dirty(sbi, segno);
1296                 }
1297                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1298                 return 1;
1299         }
1300         return 0;
1301 }
1302
1303 /*
1304  * CP calls this function, which flushes SIT entries including sit_journal,
1305  * and moves prefree segs to free segs.
1306  */
1307 void flush_sit_entries(struct f2fs_sb_info *sbi)
1308 {
1309         struct sit_info *sit_i = SIT_I(sbi);
1310         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1311         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1312         struct f2fs_summary_block *sum = curseg->sum_blk;
1313         unsigned long nsegs = TOTAL_SEGS(sbi);
1314         struct page *page = NULL;
1315         struct f2fs_sit_block *raw_sit = NULL;
1316         unsigned int start = 0, end = 0;
1317         unsigned int segno = -1;
1318         bool flushed;
1319
1320         mutex_lock(&curseg->curseg_mutex);
1321         mutex_lock(&sit_i->sentry_lock);
1322
1323         /*
1324          * "flushed" indicates whether sit entries in journal are flushed
1325          * to the SIT area or not.
1326          */
1327         flushed = flush_sits_in_journal(sbi);
1328
1329         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1330                 struct seg_entry *se = get_seg_entry(sbi, segno);
1331                 int sit_offset, offset;
1332
1333                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1334
1335                 if (flushed)
1336                         goto to_sit_page;
1337
1338                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1339                 if (offset >= 0) {
1340                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1341                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1342                         goto flush_done;
1343                 }
1344 to_sit_page:
1345                 if (!page || (start > segno) || (segno > end)) {
1346                         if (page) {
1347                                 f2fs_put_page(page, 1);
1348                                 page = NULL;
1349                         }
1350
1351                         start = START_SEGNO(sit_i, segno);
1352                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1353
1354                         /* read sit block that will be updated */
1355                         page = get_next_sit_page(sbi, start);
1356                         raw_sit = page_address(page);
1357                 }
1358
1359                 /* udpate entry in SIT block */
1360                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1361 flush_done:
1362                 __clear_bit(segno, bitmap);
1363                 sit_i->dirty_sentries--;
1364         }
1365         mutex_unlock(&sit_i->sentry_lock);
1366         mutex_unlock(&curseg->curseg_mutex);
1367
1368         /* writeout last modified SIT block */
1369         f2fs_put_page(page, 1);
1370
1371         set_prefree_as_free_segments(sbi);
1372 }
1373
1374 static int build_sit_info(struct f2fs_sb_info *sbi)
1375 {
1376         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1377         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1378         struct sit_info *sit_i;
1379         unsigned int sit_segs, start;
1380         char *src_bitmap, *dst_bitmap;
1381         unsigned int bitmap_size;
1382
1383         /* allocate memory for SIT information */
1384         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1385         if (!sit_i)
1386                 return -ENOMEM;
1387
1388         SM_I(sbi)->sit_info = sit_i;
1389
1390         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1391         if (!sit_i->sentries)
1392                 return -ENOMEM;
1393
1394         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1395         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1396         if (!sit_i->dirty_sentries_bitmap)
1397                 return -ENOMEM;
1398
1399         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1400                 sit_i->sentries[start].cur_valid_map
1401                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1402                 sit_i->sentries[start].ckpt_valid_map
1403                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1404                 if (!sit_i->sentries[start].cur_valid_map
1405                                 || !sit_i->sentries[start].ckpt_valid_map)
1406                         return -ENOMEM;
1407         }
1408
1409         if (sbi->segs_per_sec > 1) {
1410                 sit_i->sec_entries = vzalloc(sbi->total_sections *
1411                                         sizeof(struct sec_entry));
1412                 if (!sit_i->sec_entries)
1413                         return -ENOMEM;
1414         }
1415
1416         /* get information related with SIT */
1417         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1418
1419         /* setup SIT bitmap from ckeckpoint pack */
1420         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1421         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1422
1423         dst_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1424         if (!dst_bitmap)
1425                 return -ENOMEM;
1426         memcpy(dst_bitmap, src_bitmap, bitmap_size);
1427
1428         /* init SIT information */
1429         sit_i->s_ops = &default_salloc_ops;
1430
1431         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1432         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1433         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1434         sit_i->sit_bitmap = dst_bitmap;
1435         sit_i->bitmap_size = bitmap_size;
1436         sit_i->dirty_sentries = 0;
1437         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1438         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1439         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1440         mutex_init(&sit_i->sentry_lock);
1441         return 0;
1442 }
1443
1444 static int build_free_segmap(struct f2fs_sb_info *sbi)
1445 {
1446         struct f2fs_sm_info *sm_info = SM_I(sbi);
1447         struct free_segmap_info *free_i;
1448         unsigned int bitmap_size, sec_bitmap_size;
1449
1450         /* allocate memory for free segmap information */
1451         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1452         if (!free_i)
1453                 return -ENOMEM;
1454
1455         SM_I(sbi)->free_info = free_i;
1456
1457         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1458         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1459         if (!free_i->free_segmap)
1460                 return -ENOMEM;
1461
1462         sec_bitmap_size = f2fs_bitmap_size(sbi->total_sections);
1463         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1464         if (!free_i->free_secmap)
1465                 return -ENOMEM;
1466
1467         /* set all segments as dirty temporarily */
1468         memset(free_i->free_segmap, 0xff, bitmap_size);
1469         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1470
1471         /* init free segmap information */
1472         free_i->start_segno =
1473                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1474         free_i->free_segments = 0;
1475         free_i->free_sections = 0;
1476         rwlock_init(&free_i->segmap_lock);
1477         return 0;
1478 }
1479
1480 static int build_curseg(struct f2fs_sb_info *sbi)
1481 {
1482         struct curseg_info *array;
1483         int i;
1484
1485         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1486         if (!array)
1487                 return -ENOMEM;
1488
1489         SM_I(sbi)->curseg_array = array;
1490
1491         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1492                 mutex_init(&array[i].curseg_mutex);
1493                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1494                 if (!array[i].sum_blk)
1495                         return -ENOMEM;
1496                 array[i].segno = NULL_SEGNO;
1497                 array[i].next_blkoff = 0;
1498         }
1499         return restore_curseg_summaries(sbi);
1500 }
1501
1502 static void build_sit_entries(struct f2fs_sb_info *sbi)
1503 {
1504         struct sit_info *sit_i = SIT_I(sbi);
1505         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1506         struct f2fs_summary_block *sum = curseg->sum_blk;
1507         unsigned int start;
1508
1509         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1510                 struct seg_entry *se = &sit_i->sentries[start];
1511                 struct f2fs_sit_block *sit_blk;
1512                 struct f2fs_sit_entry sit;
1513                 struct page *page;
1514                 int i;
1515
1516                 mutex_lock(&curseg->curseg_mutex);
1517                 for (i = 0; i < sits_in_cursum(sum); i++) {
1518                         if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1519                                 sit = sit_in_journal(sum, i);
1520                                 mutex_unlock(&curseg->curseg_mutex);
1521                                 goto got_it;
1522                         }
1523                 }
1524                 mutex_unlock(&curseg->curseg_mutex);
1525                 page = get_current_sit_page(sbi, start);
1526                 sit_blk = (struct f2fs_sit_block *)page_address(page);
1527                 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1528                 f2fs_put_page(page, 1);
1529 got_it:
1530                 check_block_count(sbi, start, &sit);
1531                 seg_info_from_raw_sit(se, &sit);
1532                 if (sbi->segs_per_sec > 1) {
1533                         struct sec_entry *e = get_sec_entry(sbi, start);
1534                         e->valid_blocks += se->valid_blocks;
1535                 }
1536         }
1537 }
1538
1539 static void init_free_segmap(struct f2fs_sb_info *sbi)
1540 {
1541         unsigned int start;
1542         int type;
1543
1544         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1545                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1546                 if (!sentry->valid_blocks)
1547                         __set_free(sbi, start);
1548         }
1549
1550         /* set use the current segments */
1551         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1552                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1553                 __set_test_and_inuse(sbi, curseg_t->segno);
1554         }
1555 }
1556
1557 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1558 {
1559         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1560         struct free_segmap_info *free_i = FREE_I(sbi);
1561         unsigned int segno = 0, offset = 0;
1562         unsigned short valid_blocks;
1563
1564         while (segno < TOTAL_SEGS(sbi)) {
1565                 /* find dirty segment based on free segmap */
1566                 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset);
1567                 if (segno >= TOTAL_SEGS(sbi))
1568                         break;
1569                 offset = segno + 1;
1570                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1571                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1572                         continue;
1573                 mutex_lock(&dirty_i->seglist_lock);
1574                 __locate_dirty_segment(sbi, segno, DIRTY);
1575                 mutex_unlock(&dirty_i->seglist_lock);
1576         }
1577 }
1578
1579 static int init_victim_segmap(struct f2fs_sb_info *sbi)
1580 {
1581         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1582         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1583
1584         dirty_i->victim_segmap[FG_GC] = kzalloc(bitmap_size, GFP_KERNEL);
1585         dirty_i->victim_segmap[BG_GC] = kzalloc(bitmap_size, GFP_KERNEL);
1586         if (!dirty_i->victim_segmap[FG_GC] || !dirty_i->victim_segmap[BG_GC])
1587                 return -ENOMEM;
1588         return 0;
1589 }
1590
1591 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1592 {
1593         struct dirty_seglist_info *dirty_i;
1594         unsigned int bitmap_size, i;
1595
1596         /* allocate memory for dirty segments list information */
1597         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1598         if (!dirty_i)
1599                 return -ENOMEM;
1600
1601         SM_I(sbi)->dirty_info = dirty_i;
1602         mutex_init(&dirty_i->seglist_lock);
1603
1604         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1605
1606         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1607                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1608                 dirty_i->nr_dirty[i] = 0;
1609                 if (!dirty_i->dirty_segmap[i])
1610                         return -ENOMEM;
1611         }
1612
1613         init_dirty_segmap(sbi);
1614         return init_victim_segmap(sbi);
1615 }
1616
1617 /*
1618  * Update min, max modified time for cost-benefit GC algorithm
1619  */
1620 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1621 {
1622         struct sit_info *sit_i = SIT_I(sbi);
1623         unsigned int segno;
1624
1625         mutex_lock(&sit_i->sentry_lock);
1626
1627         sit_i->min_mtime = LLONG_MAX;
1628
1629         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1630                 unsigned int i;
1631                 unsigned long long mtime = 0;
1632
1633                 for (i = 0; i < sbi->segs_per_sec; i++)
1634                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1635
1636                 mtime = div_u64(mtime, sbi->segs_per_sec);
1637
1638                 if (sit_i->min_mtime > mtime)
1639                         sit_i->min_mtime = mtime;
1640         }
1641         sit_i->max_mtime = get_mtime(sbi);
1642         mutex_unlock(&sit_i->sentry_lock);
1643 }
1644
1645 int build_segment_manager(struct f2fs_sb_info *sbi)
1646 {
1647         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1648         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1649         struct f2fs_sm_info *sm_info;
1650         int err;
1651
1652         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1653         if (!sm_info)
1654                 return -ENOMEM;
1655
1656         /* init sm info */
1657         sbi->sm_info = sm_info;
1658         INIT_LIST_HEAD(&sm_info->wblist_head);
1659         spin_lock_init(&sm_info->wblist_lock);
1660         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1661         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1662         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1663         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1664         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1665         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1666         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1667
1668         err = build_sit_info(sbi);
1669         if (err)
1670                 return err;
1671         err = build_free_segmap(sbi);
1672         if (err)
1673                 return err;
1674         err = build_curseg(sbi);
1675         if (err)
1676                 return err;
1677
1678         /* reinit free segmap based on SIT */
1679         build_sit_entries(sbi);
1680
1681         init_free_segmap(sbi);
1682         err = build_dirty_segmap(sbi);
1683         if (err)
1684                 return err;
1685
1686         init_min_max_mtime(sbi);
1687         return 0;
1688 }
1689
1690 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1691                 enum dirty_type dirty_type)
1692 {
1693         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1694
1695         mutex_lock(&dirty_i->seglist_lock);
1696         kfree(dirty_i->dirty_segmap[dirty_type]);
1697         dirty_i->nr_dirty[dirty_type] = 0;
1698         mutex_unlock(&dirty_i->seglist_lock);
1699 }
1700
1701 void reset_victim_segmap(struct f2fs_sb_info *sbi)
1702 {
1703         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1704         memset(DIRTY_I(sbi)->victim_segmap[FG_GC], 0, bitmap_size);
1705 }
1706
1707 static void destroy_victim_segmap(struct f2fs_sb_info *sbi)
1708 {
1709         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1710
1711         kfree(dirty_i->victim_segmap[FG_GC]);
1712         kfree(dirty_i->victim_segmap[BG_GC]);
1713 }
1714
1715 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1716 {
1717         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1718         int i;
1719
1720         if (!dirty_i)
1721                 return;
1722
1723         /* discard pre-free/dirty segments list */
1724         for (i = 0; i < NR_DIRTY_TYPE; i++)
1725                 discard_dirty_segmap(sbi, i);
1726
1727         destroy_victim_segmap(sbi);
1728         SM_I(sbi)->dirty_info = NULL;
1729         kfree(dirty_i);
1730 }
1731
1732 static void destroy_curseg(struct f2fs_sb_info *sbi)
1733 {
1734         struct curseg_info *array = SM_I(sbi)->curseg_array;
1735         int i;
1736
1737         if (!array)
1738                 return;
1739         SM_I(sbi)->curseg_array = NULL;
1740         for (i = 0; i < NR_CURSEG_TYPE; i++)
1741                 kfree(array[i].sum_blk);
1742         kfree(array);
1743 }
1744
1745 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1746 {
1747         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1748         if (!free_i)
1749                 return;
1750         SM_I(sbi)->free_info = NULL;
1751         kfree(free_i->free_segmap);
1752         kfree(free_i->free_secmap);
1753         kfree(free_i);
1754 }
1755
1756 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1757 {
1758         struct sit_info *sit_i = SIT_I(sbi);
1759         unsigned int start;
1760
1761         if (!sit_i)
1762                 return;
1763
1764         if (sit_i->sentries) {
1765                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1766                         kfree(sit_i->sentries[start].cur_valid_map);
1767                         kfree(sit_i->sentries[start].ckpt_valid_map);
1768                 }
1769         }
1770         vfree(sit_i->sentries);
1771         vfree(sit_i->sec_entries);
1772         kfree(sit_i->dirty_sentries_bitmap);
1773
1774         SM_I(sbi)->sit_info = NULL;
1775         kfree(sit_i->sit_bitmap);
1776         kfree(sit_i);
1777 }
1778
1779 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1780 {
1781         struct f2fs_sm_info *sm_info = SM_I(sbi);
1782         destroy_dirty_segmap(sbi);
1783         destroy_curseg(sbi);
1784         destroy_free_segmap(sbi);
1785         destroy_sit_info(sbi);
1786         sbi->sm_info = NULL;
1787         kfree(sm_info);
1788 }