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