f2fs: clean up to remove parameter
[linux-block.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/kthread.h>
17 #include <linux/vmalloc.h>
18 #include <linux/swap.h>
19
20 #include "f2fs.h"
21 #include "segment.h"
22 #include "node.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
31
32 /*
33  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
34  * MSB and LSB are reversed in a byte by f2fs_set_bit.
35  */
36 static inline unsigned long __reverse_ffs(unsigned long word)
37 {
38         int num = 0;
39
40 #if BITS_PER_LONG == 64
41         if ((word & 0xffffffff) == 0) {
42                 num += 32;
43                 word >>= 32;
44         }
45 #endif
46         if ((word & 0xffff) == 0) {
47                 num += 16;
48                 word >>= 16;
49         }
50         if ((word & 0xff) == 0) {
51                 num += 8;
52                 word >>= 8;
53         }
54         if ((word & 0xf0) == 0)
55                 num += 4;
56         else
57                 word >>= 4;
58         if ((word & 0xc) == 0)
59                 num += 2;
60         else
61                 word >>= 2;
62         if ((word & 0x2) == 0)
63                 num += 1;
64         return num;
65 }
66
67 /*
68  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
69  * f2fs_set_bit makes MSB and LSB reversed in a byte.
70  * Example:
71  *                             LSB <--> MSB
72  *   f2fs_set_bit(0, bitmap) => 0000 0001
73  *   f2fs_set_bit(7, bitmap) => 1000 0000
74  */
75 static unsigned long __find_rev_next_bit(const unsigned long *addr,
76                         unsigned long size, unsigned long offset)
77 {
78         const unsigned long *p = addr + BIT_WORD(offset);
79         unsigned long result = offset & ~(BITS_PER_LONG - 1);
80         unsigned long tmp;
81         unsigned long mask, submask;
82         unsigned long quot, rest;
83
84         if (offset >= size)
85                 return size;
86
87         size -= result;
88         offset %= BITS_PER_LONG;
89         if (!offset)
90                 goto aligned;
91
92         tmp = *(p++);
93         quot = (offset >> 3) << 3;
94         rest = offset & 0x7;
95         mask = ~0UL << quot;
96         submask = (unsigned char)(0xff << rest) >> rest;
97         submask <<= quot;
98         mask &= submask;
99         tmp &= mask;
100         if (size < BITS_PER_LONG)
101                 goto found_first;
102         if (tmp)
103                 goto found_middle;
104
105         size -= BITS_PER_LONG;
106         result += BITS_PER_LONG;
107 aligned:
108         while (size & ~(BITS_PER_LONG-1)) {
109                 tmp = *(p++);
110                 if (tmp)
111                         goto found_middle;
112                 result += BITS_PER_LONG;
113                 size -= BITS_PER_LONG;
114         }
115         if (!size)
116                 return result;
117         tmp = *p;
118 found_first:
119         tmp &= (~0UL >> (BITS_PER_LONG - size));
120         if (tmp == 0UL)         /* Are any bits set? */
121                 return result + size;   /* Nope. */
122 found_middle:
123         return result + __reverse_ffs(tmp);
124 }
125
126 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
127                         unsigned long size, unsigned long offset)
128 {
129         const unsigned long *p = addr + BIT_WORD(offset);
130         unsigned long result = offset & ~(BITS_PER_LONG - 1);
131         unsigned long tmp;
132         unsigned long mask, submask;
133         unsigned long quot, rest;
134
135         if (offset >= size)
136                 return size;
137
138         size -= result;
139         offset %= BITS_PER_LONG;
140         if (!offset)
141                 goto aligned;
142
143         tmp = *(p++);
144         quot = (offset >> 3) << 3;
145         rest = offset & 0x7;
146         mask = ~(~0UL << quot);
147         submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
148         submask <<= quot;
149         mask += submask;
150         tmp |= mask;
151         if (size < BITS_PER_LONG)
152                 goto found_first;
153         if (~tmp)
154                 goto found_middle;
155
156         size -= BITS_PER_LONG;
157         result += BITS_PER_LONG;
158 aligned:
159         while (size & ~(BITS_PER_LONG - 1)) {
160                 tmp = *(p++);
161                 if (~tmp)
162                         goto found_middle;
163                 result += BITS_PER_LONG;
164                 size -= BITS_PER_LONG;
165         }
166         if (!size)
167                 return result;
168         tmp = *p;
169
170 found_first:
171         tmp |= ~0UL << size;
172         if (tmp == ~0UL)        /* Are any bits zero? */
173                 return result + size;   /* Nope. */
174 found_middle:
175         return result + __reverse_ffz(tmp);
176 }
177
178 void register_inmem_page(struct inode *inode, struct page *page)
179 {
180         struct f2fs_inode_info *fi = F2FS_I(inode);
181         struct inmem_pages *new;
182         int err;
183
184         SetPagePrivate(page);
185         f2fs_trace_pid(page);
186
187         new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
188
189         /* add atomic page indices to the list */
190         new->page = page;
191         INIT_LIST_HEAD(&new->list);
192 retry:
193         /* increase reference count with clean state */
194         mutex_lock(&fi->inmem_lock);
195         err = radix_tree_insert(&fi->inmem_root, page->index, new);
196         if (err == -EEXIST) {
197                 mutex_unlock(&fi->inmem_lock);
198                 kmem_cache_free(inmem_entry_slab, new);
199                 return;
200         } else if (err) {
201                 mutex_unlock(&fi->inmem_lock);
202                 goto retry;
203         }
204         get_page(page);
205         list_add_tail(&new->list, &fi->inmem_pages);
206         inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
207         mutex_unlock(&fi->inmem_lock);
208 }
209
210 void commit_inmem_pages(struct inode *inode, bool abort)
211 {
212         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
213         struct f2fs_inode_info *fi = F2FS_I(inode);
214         struct inmem_pages *cur, *tmp;
215         bool submit_bio = false;
216         struct f2fs_io_info fio = {
217                 .type = DATA,
218                 .rw = WRITE_SYNC | REQ_PRIO,
219         };
220
221         /*
222          * The abort is true only when f2fs_evict_inode is called.
223          * Basically, the f2fs_evict_inode doesn't produce any data writes, so
224          * that we don't need to call f2fs_balance_fs.
225          * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
226          * inode becomes free by iget_locked in f2fs_iget.
227          */
228         if (!abort) {
229                 f2fs_balance_fs(sbi);
230                 f2fs_lock_op(sbi);
231         }
232
233         mutex_lock(&fi->inmem_lock);
234         list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
235                 if (!abort) {
236                         lock_page(cur->page);
237                         if (cur->page->mapping == inode->i_mapping) {
238                                 f2fs_wait_on_page_writeback(cur->page, DATA);
239                                 if (clear_page_dirty_for_io(cur->page))
240                                         inode_dec_dirty_pages(inode);
241                                 do_write_data_page(cur->page, &fio);
242                                 submit_bio = true;
243                         }
244                         f2fs_put_page(cur->page, 1);
245                 } else {
246                         put_page(cur->page);
247                 }
248                 radix_tree_delete(&fi->inmem_root, cur->page->index);
249                 list_del(&cur->list);
250                 kmem_cache_free(inmem_entry_slab, cur);
251                 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
252         }
253         mutex_unlock(&fi->inmem_lock);
254
255         if (!abort) {
256                 f2fs_unlock_op(sbi);
257                 if (submit_bio)
258                         f2fs_submit_merged_bio(sbi, DATA, WRITE);
259         }
260 }
261
262 /*
263  * This function balances dirty node and dentry pages.
264  * In addition, it controls garbage collection.
265  */
266 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
267 {
268         /*
269          * We should do GC or end up with checkpoint, if there are so many dirty
270          * dir/node pages without enough free segments.
271          */
272         if (has_not_enough_free_secs(sbi, 0)) {
273                 mutex_lock(&sbi->gc_mutex);
274                 f2fs_gc(sbi);
275         }
276 }
277
278 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
279 {
280         /* check the # of cached NAT entries and prefree segments */
281         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
282                         excess_prefree_segs(sbi) ||
283                         !available_free_memory(sbi, INO_ENTRIES))
284                 f2fs_sync_fs(sbi->sb, true);
285 }
286
287 static int issue_flush_thread(void *data)
288 {
289         struct f2fs_sb_info *sbi = data;
290         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
291         wait_queue_head_t *q = &fcc->flush_wait_queue;
292 repeat:
293         if (kthread_should_stop())
294                 return 0;
295
296         if (!llist_empty(&fcc->issue_list)) {
297                 struct bio *bio = bio_alloc(GFP_NOIO, 0);
298                 struct flush_cmd *cmd, *next;
299                 int ret;
300
301                 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
302                 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
303
304                 bio->bi_bdev = sbi->sb->s_bdev;
305                 ret = submit_bio_wait(WRITE_FLUSH, bio);
306
307                 llist_for_each_entry_safe(cmd, next,
308                                           fcc->dispatch_list, llnode) {
309                         cmd->ret = ret;
310                         complete(&cmd->wait);
311                 }
312                 bio_put(bio);
313                 fcc->dispatch_list = NULL;
314         }
315
316         wait_event_interruptible(*q,
317                 kthread_should_stop() || !llist_empty(&fcc->issue_list));
318         goto repeat;
319 }
320
321 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
322 {
323         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
324         struct flush_cmd cmd;
325
326         trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
327                                         test_opt(sbi, FLUSH_MERGE));
328
329         if (test_opt(sbi, NOBARRIER))
330                 return 0;
331
332         if (!test_opt(sbi, FLUSH_MERGE))
333                 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
334
335         init_completion(&cmd.wait);
336
337         llist_add(&cmd.llnode, &fcc->issue_list);
338
339         if (!fcc->dispatch_list)
340                 wake_up(&fcc->flush_wait_queue);
341
342         wait_for_completion(&cmd.wait);
343
344         return cmd.ret;
345 }
346
347 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
348 {
349         dev_t dev = sbi->sb->s_bdev->bd_dev;
350         struct flush_cmd_control *fcc;
351         int err = 0;
352
353         fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
354         if (!fcc)
355                 return -ENOMEM;
356         init_waitqueue_head(&fcc->flush_wait_queue);
357         init_llist_head(&fcc->issue_list);
358         SM_I(sbi)->cmd_control_info = fcc;
359         fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
360                                 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
361         if (IS_ERR(fcc->f2fs_issue_flush)) {
362                 err = PTR_ERR(fcc->f2fs_issue_flush);
363                 kfree(fcc);
364                 SM_I(sbi)->cmd_control_info = NULL;
365                 return err;
366         }
367
368         return err;
369 }
370
371 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
372 {
373         struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
374
375         if (fcc && fcc->f2fs_issue_flush)
376                 kthread_stop(fcc->f2fs_issue_flush);
377         kfree(fcc);
378         SM_I(sbi)->cmd_control_info = NULL;
379 }
380
381 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
382                 enum dirty_type dirty_type)
383 {
384         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
385
386         /* need not be added */
387         if (IS_CURSEG(sbi, segno))
388                 return;
389
390         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
391                 dirty_i->nr_dirty[dirty_type]++;
392
393         if (dirty_type == DIRTY) {
394                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
395                 enum dirty_type t = sentry->type;
396
397                 if (unlikely(t >= DIRTY)) {
398                         f2fs_bug_on(sbi, 1);
399                         return;
400                 }
401                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
402                         dirty_i->nr_dirty[t]++;
403         }
404 }
405
406 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
407                 enum dirty_type dirty_type)
408 {
409         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
410
411         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
412                 dirty_i->nr_dirty[dirty_type]--;
413
414         if (dirty_type == DIRTY) {
415                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
416                 enum dirty_type t = sentry->type;
417
418                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
419                         dirty_i->nr_dirty[t]--;
420
421                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
422                         clear_bit(GET_SECNO(sbi, segno),
423                                                 dirty_i->victim_secmap);
424         }
425 }
426
427 /*
428  * Should not occur error such as -ENOMEM.
429  * Adding dirty entry into seglist is not critical operation.
430  * If a given segment is one of current working segments, it won't be added.
431  */
432 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
433 {
434         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
435         unsigned short valid_blocks;
436
437         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
438                 return;
439
440         mutex_lock(&dirty_i->seglist_lock);
441
442         valid_blocks = get_valid_blocks(sbi, segno, 0);
443
444         if (valid_blocks == 0) {
445                 __locate_dirty_segment(sbi, segno, PRE);
446                 __remove_dirty_segment(sbi, segno, DIRTY);
447         } else if (valid_blocks < sbi->blocks_per_seg) {
448                 __locate_dirty_segment(sbi, segno, DIRTY);
449         } else {
450                 /* Recovery routine with SSR needs this */
451                 __remove_dirty_segment(sbi, segno, DIRTY);
452         }
453
454         mutex_unlock(&dirty_i->seglist_lock);
455 }
456
457 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
458                                 block_t blkstart, block_t blklen)
459 {
460         sector_t start = SECTOR_FROM_BLOCK(blkstart);
461         sector_t len = SECTOR_FROM_BLOCK(blklen);
462         trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
463         return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
464 }
465
466 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
467 {
468         if (f2fs_issue_discard(sbi, blkaddr, 1)) {
469                 struct page *page = grab_meta_page(sbi, blkaddr);
470                 /* zero-filled page */
471                 set_page_dirty(page);
472                 f2fs_put_page(page, 1);
473         }
474 }
475
476 static void __add_discard_entry(struct f2fs_sb_info *sbi,
477                 struct cp_control *cpc, unsigned int start, unsigned int end)
478 {
479         struct list_head *head = &SM_I(sbi)->discard_list;
480         struct discard_entry *new, *last;
481
482         if (!list_empty(head)) {
483                 last = list_last_entry(head, struct discard_entry, list);
484                 if (START_BLOCK(sbi, cpc->trim_start) + start ==
485                                                 last->blkaddr + last->len) {
486                         last->len += end - start;
487                         goto done;
488                 }
489         }
490
491         new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
492         INIT_LIST_HEAD(&new->list);
493         new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
494         new->len = end - start;
495         list_add_tail(&new->list, head);
496 done:
497         SM_I(sbi)->nr_discards += end - start;
498         cpc->trimmed += end - start;
499 }
500
501 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
502 {
503         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
504         int max_blocks = sbi->blocks_per_seg;
505         struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
506         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
507         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
508         unsigned long dmap[entries];
509         unsigned int start = 0, end = -1;
510         bool force = (cpc->reason == CP_DISCARD);
511         int i;
512
513         if (!force && (!test_opt(sbi, DISCARD) ||
514                         SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards))
515                 return;
516
517         if (force && !se->valid_blocks) {
518                 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
519                 /*
520                  * if this segment is registered in the prefree list, then
521                  * we should skip adding a discard candidate, and let the
522                  * checkpoint do that later.
523                  */
524                 mutex_lock(&dirty_i->seglist_lock);
525                 if (test_bit(cpc->trim_start, dirty_i->dirty_segmap[PRE])) {
526                         mutex_unlock(&dirty_i->seglist_lock);
527                         cpc->trimmed += sbi->blocks_per_seg;
528                         return;
529                 }
530                 mutex_unlock(&dirty_i->seglist_lock);
531
532                 __add_discard_entry(sbi, cpc, 0, sbi->blocks_per_seg);
533                 return;
534         }
535
536         /* zero block will be discarded through the prefree list */
537         if (!se->valid_blocks || se->valid_blocks == max_blocks)
538                 return;
539
540         /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
541         for (i = 0; i < entries; i++)
542                 dmap[i] = force ? ~ckpt_map[i] :
543                                 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
544
545         while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
546                 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
547                 if (start >= max_blocks)
548                         break;
549
550                 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
551
552                 if (end - start < cpc->trim_minlen)
553                         continue;
554
555                 __add_discard_entry(sbi, cpc, start, end);
556         }
557 }
558
559 void release_discard_addrs(struct f2fs_sb_info *sbi)
560 {
561         struct list_head *head = &(SM_I(sbi)->discard_list);
562         struct discard_entry *entry, *this;
563
564         /* drop caches */
565         list_for_each_entry_safe(entry, this, head, list) {
566                 list_del(&entry->list);
567                 kmem_cache_free(discard_entry_slab, entry);
568         }
569 }
570
571 /*
572  * Should call clear_prefree_segments after checkpoint is done.
573  */
574 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
575 {
576         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
577         unsigned int segno;
578
579         mutex_lock(&dirty_i->seglist_lock);
580         for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
581                 __set_test_and_free(sbi, segno);
582         mutex_unlock(&dirty_i->seglist_lock);
583 }
584
585 void clear_prefree_segments(struct f2fs_sb_info *sbi)
586 {
587         struct list_head *head = &(SM_I(sbi)->discard_list);
588         struct discard_entry *entry, *this;
589         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
590         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
591         unsigned int start = 0, end = -1;
592
593         mutex_lock(&dirty_i->seglist_lock);
594
595         while (1) {
596                 int i;
597                 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
598                 if (start >= MAIN_SEGS(sbi))
599                         break;
600                 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
601                                                                 start + 1);
602
603                 for (i = start; i < end; i++)
604                         clear_bit(i, prefree_map);
605
606                 dirty_i->nr_dirty[PRE] -= end - start;
607
608                 if (!test_opt(sbi, DISCARD))
609                         continue;
610
611                 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
612                                 (end - start) << sbi->log_blocks_per_seg);
613         }
614         mutex_unlock(&dirty_i->seglist_lock);
615
616         /* send small discards */
617         list_for_each_entry_safe(entry, this, head, list) {
618                 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
619                 list_del(&entry->list);
620                 SM_I(sbi)->nr_discards -= entry->len;
621                 kmem_cache_free(discard_entry_slab, entry);
622         }
623 }
624
625 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
626 {
627         struct sit_info *sit_i = SIT_I(sbi);
628
629         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
630                 sit_i->dirty_sentries++;
631                 return false;
632         }
633
634         return true;
635 }
636
637 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
638                                         unsigned int segno, int modified)
639 {
640         struct seg_entry *se = get_seg_entry(sbi, segno);
641         se->type = type;
642         if (modified)
643                 __mark_sit_entry_dirty(sbi, segno);
644 }
645
646 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
647 {
648         struct seg_entry *se;
649         unsigned int segno, offset;
650         long int new_vblocks;
651
652         segno = GET_SEGNO(sbi, blkaddr);
653
654         se = get_seg_entry(sbi, segno);
655         new_vblocks = se->valid_blocks + del;
656         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
657
658         f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
659                                 (new_vblocks > sbi->blocks_per_seg)));
660
661         se->valid_blocks = new_vblocks;
662         se->mtime = get_mtime(sbi);
663         SIT_I(sbi)->max_mtime = se->mtime;
664
665         /* Update valid block bitmap */
666         if (del > 0) {
667                 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
668                         f2fs_bug_on(sbi, 1);
669         } else {
670                 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
671                         f2fs_bug_on(sbi, 1);
672         }
673         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
674                 se->ckpt_valid_blocks += del;
675
676         __mark_sit_entry_dirty(sbi, segno);
677
678         /* update total number of valid blocks to be written in ckpt area */
679         SIT_I(sbi)->written_valid_blocks += del;
680
681         if (sbi->segs_per_sec > 1)
682                 get_sec_entry(sbi, segno)->valid_blocks += del;
683 }
684
685 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
686 {
687         update_sit_entry(sbi, new, 1);
688         if (GET_SEGNO(sbi, old) != NULL_SEGNO)
689                 update_sit_entry(sbi, old, -1);
690
691         locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
692         locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
693 }
694
695 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
696 {
697         unsigned int segno = GET_SEGNO(sbi, addr);
698         struct sit_info *sit_i = SIT_I(sbi);
699
700         f2fs_bug_on(sbi, addr == NULL_ADDR);
701         if (addr == NEW_ADDR)
702                 return;
703
704         /* add it into sit main buffer */
705         mutex_lock(&sit_i->sentry_lock);
706
707         update_sit_entry(sbi, addr, -1);
708
709         /* add it into dirty seglist */
710         locate_dirty_segment(sbi, segno);
711
712         mutex_unlock(&sit_i->sentry_lock);
713 }
714
715 /*
716  * This function should be resided under the curseg_mutex lock
717  */
718 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
719                                         struct f2fs_summary *sum)
720 {
721         struct curseg_info *curseg = CURSEG_I(sbi, type);
722         void *addr = curseg->sum_blk;
723         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
724         memcpy(addr, sum, sizeof(struct f2fs_summary));
725 }
726
727 /*
728  * Calculate the number of current summary pages for writing
729  */
730 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
731 {
732         int valid_sum_count = 0;
733         int i, sum_in_page;
734
735         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
736                 if (sbi->ckpt->alloc_type[i] == SSR)
737                         valid_sum_count += sbi->blocks_per_seg;
738                 else {
739                         if (for_ra)
740                                 valid_sum_count += le16_to_cpu(
741                                         F2FS_CKPT(sbi)->cur_data_blkoff[i]);
742                         else
743                                 valid_sum_count += curseg_blkoff(sbi, i);
744                 }
745         }
746
747         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
748                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
749         if (valid_sum_count <= sum_in_page)
750                 return 1;
751         else if ((valid_sum_count - sum_in_page) <=
752                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
753                 return 2;
754         return 3;
755 }
756
757 /*
758  * Caller should put this summary page
759  */
760 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
761 {
762         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
763 }
764
765 static void write_sum_page(struct f2fs_sb_info *sbi,
766                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
767 {
768         struct page *page = grab_meta_page(sbi, blk_addr);
769         void *kaddr = page_address(page);
770         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
771         set_page_dirty(page);
772         f2fs_put_page(page, 1);
773 }
774
775 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
776 {
777         struct curseg_info *curseg = CURSEG_I(sbi, type);
778         unsigned int segno = curseg->segno + 1;
779         struct free_segmap_info *free_i = FREE_I(sbi);
780
781         if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
782                 return !test_bit(segno, free_i->free_segmap);
783         return 0;
784 }
785
786 /*
787  * Find a new segment from the free segments bitmap to right order
788  * This function should be returned with success, otherwise BUG
789  */
790 static void get_new_segment(struct f2fs_sb_info *sbi,
791                         unsigned int *newseg, bool new_sec, int dir)
792 {
793         struct free_segmap_info *free_i = FREE_I(sbi);
794         unsigned int segno, secno, zoneno;
795         unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
796         unsigned int hint = *newseg / sbi->segs_per_sec;
797         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
798         unsigned int left_start = hint;
799         bool init = true;
800         int go_left = 0;
801         int i;
802
803         write_lock(&free_i->segmap_lock);
804
805         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
806                 segno = find_next_zero_bit(free_i->free_segmap,
807                                         MAIN_SEGS(sbi), *newseg + 1);
808                 if (segno - *newseg < sbi->segs_per_sec -
809                                         (*newseg % sbi->segs_per_sec))
810                         goto got_it;
811         }
812 find_other_zone:
813         secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
814         if (secno >= MAIN_SECS(sbi)) {
815                 if (dir == ALLOC_RIGHT) {
816                         secno = find_next_zero_bit(free_i->free_secmap,
817                                                         MAIN_SECS(sbi), 0);
818                         f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
819                 } else {
820                         go_left = 1;
821                         left_start = hint - 1;
822                 }
823         }
824         if (go_left == 0)
825                 goto skip_left;
826
827         while (test_bit(left_start, free_i->free_secmap)) {
828                 if (left_start > 0) {
829                         left_start--;
830                         continue;
831                 }
832                 left_start = find_next_zero_bit(free_i->free_secmap,
833                                                         MAIN_SECS(sbi), 0);
834                 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
835                 break;
836         }
837         secno = left_start;
838 skip_left:
839         hint = secno;
840         segno = secno * sbi->segs_per_sec;
841         zoneno = secno / sbi->secs_per_zone;
842
843         /* give up on finding another zone */
844         if (!init)
845                 goto got_it;
846         if (sbi->secs_per_zone == 1)
847                 goto got_it;
848         if (zoneno == old_zoneno)
849                 goto got_it;
850         if (dir == ALLOC_LEFT) {
851                 if (!go_left && zoneno + 1 >= total_zones)
852                         goto got_it;
853                 if (go_left && zoneno == 0)
854                         goto got_it;
855         }
856         for (i = 0; i < NR_CURSEG_TYPE; i++)
857                 if (CURSEG_I(sbi, i)->zone == zoneno)
858                         break;
859
860         if (i < NR_CURSEG_TYPE) {
861                 /* zone is in user, try another */
862                 if (go_left)
863                         hint = zoneno * sbi->secs_per_zone - 1;
864                 else if (zoneno + 1 >= total_zones)
865                         hint = 0;
866                 else
867                         hint = (zoneno + 1) * sbi->secs_per_zone;
868                 init = false;
869                 goto find_other_zone;
870         }
871 got_it:
872         /* set it as dirty segment in free segmap */
873         f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
874         __set_inuse(sbi, segno);
875         *newseg = segno;
876         write_unlock(&free_i->segmap_lock);
877 }
878
879 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
880 {
881         struct curseg_info *curseg = CURSEG_I(sbi, type);
882         struct summary_footer *sum_footer;
883
884         curseg->segno = curseg->next_segno;
885         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
886         curseg->next_blkoff = 0;
887         curseg->next_segno = NULL_SEGNO;
888
889         sum_footer = &(curseg->sum_blk->footer);
890         memset(sum_footer, 0, sizeof(struct summary_footer));
891         if (IS_DATASEG(type))
892                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
893         if (IS_NODESEG(type))
894                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
895         __set_sit_entry_type(sbi, type, curseg->segno, modified);
896 }
897
898 /*
899  * Allocate a current working segment.
900  * This function always allocates a free segment in LFS manner.
901  */
902 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
903 {
904         struct curseg_info *curseg = CURSEG_I(sbi, type);
905         unsigned int segno = curseg->segno;
906         int dir = ALLOC_LEFT;
907
908         write_sum_page(sbi, curseg->sum_blk,
909                                 GET_SUM_BLOCK(sbi, segno));
910         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
911                 dir = ALLOC_RIGHT;
912
913         if (test_opt(sbi, NOHEAP))
914                 dir = ALLOC_RIGHT;
915
916         get_new_segment(sbi, &segno, new_sec, dir);
917         curseg->next_segno = segno;
918         reset_curseg(sbi, type, 1);
919         curseg->alloc_type = LFS;
920 }
921
922 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
923                         struct curseg_info *seg, block_t start)
924 {
925         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
926         int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
927         unsigned long target_map[entries];
928         unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
929         unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
930         int i, pos;
931
932         for (i = 0; i < entries; i++)
933                 target_map[i] = ckpt_map[i] | cur_map[i];
934
935         pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
936
937         seg->next_blkoff = pos;
938 }
939
940 /*
941  * If a segment is written by LFS manner, next block offset is just obtained
942  * by increasing the current block offset. However, if a segment is written by
943  * SSR manner, next block offset obtained by calling __next_free_blkoff
944  */
945 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
946                                 struct curseg_info *seg)
947 {
948         if (seg->alloc_type == SSR)
949                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
950         else
951                 seg->next_blkoff++;
952 }
953
954 /*
955  * This function always allocates a used segment(from dirty seglist) by SSR
956  * manner, so it should recover the existing segment information of valid blocks
957  */
958 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
959 {
960         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
961         struct curseg_info *curseg = CURSEG_I(sbi, type);
962         unsigned int new_segno = curseg->next_segno;
963         struct f2fs_summary_block *sum_node;
964         struct page *sum_page;
965
966         write_sum_page(sbi, curseg->sum_blk,
967                                 GET_SUM_BLOCK(sbi, curseg->segno));
968         __set_test_and_inuse(sbi, new_segno);
969
970         mutex_lock(&dirty_i->seglist_lock);
971         __remove_dirty_segment(sbi, new_segno, PRE);
972         __remove_dirty_segment(sbi, new_segno, DIRTY);
973         mutex_unlock(&dirty_i->seglist_lock);
974
975         reset_curseg(sbi, type, 1);
976         curseg->alloc_type = SSR;
977         __next_free_blkoff(sbi, curseg, 0);
978
979         if (reuse) {
980                 sum_page = get_sum_page(sbi, new_segno);
981                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
982                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
983                 f2fs_put_page(sum_page, 1);
984         }
985 }
986
987 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
988 {
989         struct curseg_info *curseg = CURSEG_I(sbi, type);
990         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
991
992         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
993                 return v_ops->get_victim(sbi,
994                                 &(curseg)->next_segno, BG_GC, type, SSR);
995
996         /* For data segments, let's do SSR more intensively */
997         for (; type >= CURSEG_HOT_DATA; type--)
998                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
999                                                 BG_GC, type, SSR))
1000                         return 1;
1001         return 0;
1002 }
1003
1004 /*
1005  * flush out current segment and replace it with new segment
1006  * This function should be returned with success, otherwise BUG
1007  */
1008 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1009                                                 int type, bool force)
1010 {
1011         struct curseg_info *curseg = CURSEG_I(sbi, type);
1012
1013         if (force)
1014                 new_curseg(sbi, type, true);
1015         else if (type == CURSEG_WARM_NODE)
1016                 new_curseg(sbi, type, false);
1017         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1018                 new_curseg(sbi, type, false);
1019         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1020                 change_curseg(sbi, type, true);
1021         else
1022                 new_curseg(sbi, type, false);
1023
1024         stat_inc_seg_type(sbi, curseg);
1025 }
1026
1027 void allocate_new_segments(struct f2fs_sb_info *sbi)
1028 {
1029         struct curseg_info *curseg;
1030         unsigned int old_curseg;
1031         int i;
1032
1033         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1034                 curseg = CURSEG_I(sbi, i);
1035                 old_curseg = curseg->segno;
1036                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1037                 locate_dirty_segment(sbi, old_curseg);
1038         }
1039 }
1040
1041 static const struct segment_allocation default_salloc_ops = {
1042         .allocate_segment = allocate_segment_by_default,
1043 };
1044
1045 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1046 {
1047         __u64 start = range->start >> sbi->log_blocksize;
1048         __u64 end = start + (range->len >> sbi->log_blocksize) - 1;
1049         unsigned int start_segno, end_segno;
1050         struct cp_control cpc;
1051
1052         if (range->minlen > SEGMENT_SIZE(sbi) || start >= MAX_BLKADDR(sbi) ||
1053                                                 range->len < sbi->blocksize)
1054                 return -EINVAL;
1055
1056         cpc.trimmed = 0;
1057         if (end <= MAIN_BLKADDR(sbi))
1058                 goto out;
1059
1060         /* start/end segment number in main_area */
1061         start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1062         end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1063                                                 GET_SEGNO(sbi, end);
1064         cpc.reason = CP_DISCARD;
1065         cpc.trim_start = start_segno;
1066         cpc.trim_end = end_segno;
1067         cpc.trim_minlen = range->minlen >> sbi->log_blocksize;
1068
1069         /* do checkpoint to issue discard commands safely */
1070         mutex_lock(&sbi->gc_mutex);
1071         write_checkpoint(sbi, &cpc);
1072         mutex_unlock(&sbi->gc_mutex);
1073 out:
1074         range->len = cpc.trimmed << sbi->log_blocksize;
1075         return 0;
1076 }
1077
1078 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1079 {
1080         struct curseg_info *curseg = CURSEG_I(sbi, type);
1081         if (curseg->next_blkoff < sbi->blocks_per_seg)
1082                 return true;
1083         return false;
1084 }
1085
1086 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1087 {
1088         if (p_type == DATA)
1089                 return CURSEG_HOT_DATA;
1090         else
1091                 return CURSEG_HOT_NODE;
1092 }
1093
1094 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1095 {
1096         if (p_type == DATA) {
1097                 struct inode *inode = page->mapping->host;
1098
1099                 if (S_ISDIR(inode->i_mode))
1100                         return CURSEG_HOT_DATA;
1101                 else
1102                         return CURSEG_COLD_DATA;
1103         } else {
1104                 if (IS_DNODE(page) && is_cold_node(page))
1105                         return CURSEG_WARM_NODE;
1106                 else
1107                         return CURSEG_COLD_NODE;
1108         }
1109 }
1110
1111 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1112 {
1113         if (p_type == DATA) {
1114                 struct inode *inode = page->mapping->host;
1115
1116                 if (S_ISDIR(inode->i_mode))
1117                         return CURSEG_HOT_DATA;
1118                 else if (is_cold_data(page) || file_is_cold(inode))
1119                         return CURSEG_COLD_DATA;
1120                 else
1121                         return CURSEG_WARM_DATA;
1122         } else {
1123                 if (IS_DNODE(page))
1124                         return is_cold_node(page) ? CURSEG_WARM_NODE :
1125                                                 CURSEG_HOT_NODE;
1126                 else
1127                         return CURSEG_COLD_NODE;
1128         }
1129 }
1130
1131 static int __get_segment_type(struct page *page, enum page_type p_type)
1132 {
1133         switch (F2FS_P_SB(page)->active_logs) {
1134         case 2:
1135                 return __get_segment_type_2(page, p_type);
1136         case 4:
1137                 return __get_segment_type_4(page, p_type);
1138         }
1139         /* NR_CURSEG_TYPE(6) logs by default */
1140         f2fs_bug_on(F2FS_P_SB(page),
1141                 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1142         return __get_segment_type_6(page, p_type);
1143 }
1144
1145 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1146                 block_t old_blkaddr, block_t *new_blkaddr,
1147                 struct f2fs_summary *sum, int type)
1148 {
1149         struct sit_info *sit_i = SIT_I(sbi);
1150         struct curseg_info *curseg;
1151
1152         curseg = CURSEG_I(sbi, type);
1153
1154         mutex_lock(&curseg->curseg_mutex);
1155
1156         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1157
1158         /*
1159          * __add_sum_entry should be resided under the curseg_mutex
1160          * because, this function updates a summary entry in the
1161          * current summary block.
1162          */
1163         __add_sum_entry(sbi, type, sum);
1164
1165         mutex_lock(&sit_i->sentry_lock);
1166         __refresh_next_blkoff(sbi, curseg);
1167
1168         stat_inc_block_count(sbi, curseg);
1169
1170         if (!__has_curseg_space(sbi, type))
1171                 sit_i->s_ops->allocate_segment(sbi, type, false);
1172         /*
1173          * SIT information should be updated before segment allocation,
1174          * since SSR needs latest valid block information.
1175          */
1176         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1177
1178         mutex_unlock(&sit_i->sentry_lock);
1179
1180         if (page && IS_NODESEG(type))
1181                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1182
1183         mutex_unlock(&curseg->curseg_mutex);
1184 }
1185
1186 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1187                         struct f2fs_summary *sum,
1188                         struct f2fs_io_info *fio)
1189 {
1190         int type = __get_segment_type(page, fio->type);
1191
1192         allocate_data_block(sbi, page, fio->blk_addr, &fio->blk_addr, sum, type);
1193
1194         /* writeout dirty page into bdev */
1195         f2fs_submit_page_mbio(sbi, page, fio);
1196 }
1197
1198 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1199 {
1200         struct f2fs_io_info fio = {
1201                 .type = META,
1202                 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1203                 .blk_addr = page->index,
1204         };
1205
1206         set_page_writeback(page);
1207         f2fs_submit_page_mbio(sbi, page, &fio);
1208 }
1209
1210 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1211                         unsigned int nid, struct f2fs_io_info *fio)
1212 {
1213         struct f2fs_summary sum;
1214         set_summary(&sum, nid, 0, 0);
1215         do_write_page(sbi, page, &sum, fio);
1216 }
1217
1218 void write_data_page(struct page *page, struct dnode_of_data *dn,
1219                                 struct f2fs_io_info *fio)
1220 {
1221         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1222         struct f2fs_summary sum;
1223         struct node_info ni;
1224
1225         f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1226         get_node_info(sbi, dn->nid, &ni);
1227         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1228         do_write_page(sbi, page, &sum, fio);
1229         dn->data_blkaddr = fio->blk_addr;
1230 }
1231
1232 void rewrite_data_page(struct page *page, struct f2fs_io_info *fio)
1233 {
1234         stat_inc_inplace_blocks(F2FS_P_SB(page));
1235         f2fs_submit_page_mbio(F2FS_P_SB(page), page, fio);
1236 }
1237
1238 void recover_data_page(struct f2fs_sb_info *sbi,
1239                         struct page *page, struct f2fs_summary *sum,
1240                         block_t old_blkaddr, block_t new_blkaddr)
1241 {
1242         struct sit_info *sit_i = SIT_I(sbi);
1243         struct curseg_info *curseg;
1244         unsigned int segno, old_cursegno;
1245         struct seg_entry *se;
1246         int type;
1247
1248         segno = GET_SEGNO(sbi, new_blkaddr);
1249         se = get_seg_entry(sbi, segno);
1250         type = se->type;
1251
1252         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1253                 if (old_blkaddr == NULL_ADDR)
1254                         type = CURSEG_COLD_DATA;
1255                 else
1256                         type = CURSEG_WARM_DATA;
1257         }
1258         curseg = CURSEG_I(sbi, type);
1259
1260         mutex_lock(&curseg->curseg_mutex);
1261         mutex_lock(&sit_i->sentry_lock);
1262
1263         old_cursegno = curseg->segno;
1264
1265         /* change the current segment */
1266         if (segno != curseg->segno) {
1267                 curseg->next_segno = segno;
1268                 change_curseg(sbi, type, true);
1269         }
1270
1271         curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1272         __add_sum_entry(sbi, type, sum);
1273
1274         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1275         locate_dirty_segment(sbi, old_cursegno);
1276
1277         mutex_unlock(&sit_i->sentry_lock);
1278         mutex_unlock(&curseg->curseg_mutex);
1279 }
1280
1281 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1282                                         struct page *page, enum page_type type)
1283 {
1284         enum page_type btype = PAGE_TYPE_OF_BIO(type);
1285         struct f2fs_bio_info *io = &sbi->write_io[btype];
1286         struct bio_vec *bvec;
1287         int i;
1288
1289         down_read(&io->io_rwsem);
1290         if (!io->bio)
1291                 goto out;
1292
1293         bio_for_each_segment_all(bvec, io->bio, i) {
1294                 if (page == bvec->bv_page) {
1295                         up_read(&io->io_rwsem);
1296                         return true;
1297                 }
1298         }
1299
1300 out:
1301         up_read(&io->io_rwsem);
1302         return false;
1303 }
1304
1305 void f2fs_wait_on_page_writeback(struct page *page,
1306                                 enum page_type type)
1307 {
1308         if (PageWriteback(page)) {
1309                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1310
1311                 if (is_merged_page(sbi, page, type))
1312                         f2fs_submit_merged_bio(sbi, type, WRITE);
1313                 wait_on_page_writeback(page);
1314         }
1315 }
1316
1317 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1318 {
1319         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1320         struct curseg_info *seg_i;
1321         unsigned char *kaddr;
1322         struct page *page;
1323         block_t start;
1324         int i, j, offset;
1325
1326         start = start_sum_block(sbi);
1327
1328         page = get_meta_page(sbi, start++);
1329         kaddr = (unsigned char *)page_address(page);
1330
1331         /* Step 1: restore nat cache */
1332         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1333         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1334
1335         /* Step 2: restore sit cache */
1336         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1337         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1338                                                 SUM_JOURNAL_SIZE);
1339         offset = 2 * SUM_JOURNAL_SIZE;
1340
1341         /* Step 3: restore summary entries */
1342         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1343                 unsigned short blk_off;
1344                 unsigned int segno;
1345
1346                 seg_i = CURSEG_I(sbi, i);
1347                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1348                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1349                 seg_i->next_segno = segno;
1350                 reset_curseg(sbi, i, 0);
1351                 seg_i->alloc_type = ckpt->alloc_type[i];
1352                 seg_i->next_blkoff = blk_off;
1353
1354                 if (seg_i->alloc_type == SSR)
1355                         blk_off = sbi->blocks_per_seg;
1356
1357                 for (j = 0; j < blk_off; j++) {
1358                         struct f2fs_summary *s;
1359                         s = (struct f2fs_summary *)(kaddr + offset);
1360                         seg_i->sum_blk->entries[j] = *s;
1361                         offset += SUMMARY_SIZE;
1362                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1363                                                 SUM_FOOTER_SIZE)
1364                                 continue;
1365
1366                         f2fs_put_page(page, 1);
1367                         page = NULL;
1368
1369                         page = get_meta_page(sbi, start++);
1370                         kaddr = (unsigned char *)page_address(page);
1371                         offset = 0;
1372                 }
1373         }
1374         f2fs_put_page(page, 1);
1375         return 0;
1376 }
1377
1378 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1379 {
1380         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1381         struct f2fs_summary_block *sum;
1382         struct curseg_info *curseg;
1383         struct page *new;
1384         unsigned short blk_off;
1385         unsigned int segno = 0;
1386         block_t blk_addr = 0;
1387
1388         /* get segment number and block addr */
1389         if (IS_DATASEG(type)) {
1390                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1391                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1392                                                         CURSEG_HOT_DATA]);
1393                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1394                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1395                 else
1396                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1397         } else {
1398                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1399                                                         CURSEG_HOT_NODE]);
1400                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1401                                                         CURSEG_HOT_NODE]);
1402                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1403                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1404                                                         type - CURSEG_HOT_NODE);
1405                 else
1406                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1407         }
1408
1409         new = get_meta_page(sbi, blk_addr);
1410         sum = (struct f2fs_summary_block *)page_address(new);
1411
1412         if (IS_NODESEG(type)) {
1413                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1414                         struct f2fs_summary *ns = &sum->entries[0];
1415                         int i;
1416                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1417                                 ns->version = 0;
1418                                 ns->ofs_in_node = 0;
1419                         }
1420                 } else {
1421                         int err;
1422
1423                         err = restore_node_summary(sbi, segno, sum);
1424                         if (err) {
1425                                 f2fs_put_page(new, 1);
1426                                 return err;
1427                         }
1428                 }
1429         }
1430
1431         /* set uncompleted segment to curseg */
1432         curseg = CURSEG_I(sbi, type);
1433         mutex_lock(&curseg->curseg_mutex);
1434         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1435         curseg->next_segno = segno;
1436         reset_curseg(sbi, type, 0);
1437         curseg->alloc_type = ckpt->alloc_type[type];
1438         curseg->next_blkoff = blk_off;
1439         mutex_unlock(&curseg->curseg_mutex);
1440         f2fs_put_page(new, 1);
1441         return 0;
1442 }
1443
1444 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1445 {
1446         int type = CURSEG_HOT_DATA;
1447         int err;
1448
1449         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1450                 int npages = npages_for_summary_flush(sbi, true);
1451
1452                 if (npages >= 2)
1453                         ra_meta_pages(sbi, start_sum_block(sbi), npages,
1454                                                                 META_CP);
1455
1456                 /* restore for compacted data summary */
1457                 if (read_compacted_summaries(sbi))
1458                         return -EINVAL;
1459                 type = CURSEG_HOT_NODE;
1460         }
1461
1462         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1463                 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1464                                         NR_CURSEG_TYPE - type, META_CP);
1465
1466         for (; type <= CURSEG_COLD_NODE; type++) {
1467                 err = read_normal_summaries(sbi, type);
1468                 if (err)
1469                         return err;
1470         }
1471
1472         return 0;
1473 }
1474
1475 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1476 {
1477         struct page *page;
1478         unsigned char *kaddr;
1479         struct f2fs_summary *summary;
1480         struct curseg_info *seg_i;
1481         int written_size = 0;
1482         int i, j;
1483
1484         page = grab_meta_page(sbi, blkaddr++);
1485         kaddr = (unsigned char *)page_address(page);
1486
1487         /* Step 1: write nat cache */
1488         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1489         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1490         written_size += SUM_JOURNAL_SIZE;
1491
1492         /* Step 2: write sit cache */
1493         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1494         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1495                                                 SUM_JOURNAL_SIZE);
1496         written_size += SUM_JOURNAL_SIZE;
1497
1498         /* Step 3: write summary entries */
1499         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1500                 unsigned short blkoff;
1501                 seg_i = CURSEG_I(sbi, i);
1502                 if (sbi->ckpt->alloc_type[i] == SSR)
1503                         blkoff = sbi->blocks_per_seg;
1504                 else
1505                         blkoff = curseg_blkoff(sbi, i);
1506
1507                 for (j = 0; j < blkoff; j++) {
1508                         if (!page) {
1509                                 page = grab_meta_page(sbi, blkaddr++);
1510                                 kaddr = (unsigned char *)page_address(page);
1511                                 written_size = 0;
1512                         }
1513                         summary = (struct f2fs_summary *)(kaddr + written_size);
1514                         *summary = seg_i->sum_blk->entries[j];
1515                         written_size += SUMMARY_SIZE;
1516
1517                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1518                                                         SUM_FOOTER_SIZE)
1519                                 continue;
1520
1521                         set_page_dirty(page);
1522                         f2fs_put_page(page, 1);
1523                         page = NULL;
1524                 }
1525         }
1526         if (page) {
1527                 set_page_dirty(page);
1528                 f2fs_put_page(page, 1);
1529         }
1530 }
1531
1532 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1533                                         block_t blkaddr, int type)
1534 {
1535         int i, end;
1536         if (IS_DATASEG(type))
1537                 end = type + NR_CURSEG_DATA_TYPE;
1538         else
1539                 end = type + NR_CURSEG_NODE_TYPE;
1540
1541         for (i = type; i < end; i++) {
1542                 struct curseg_info *sum = CURSEG_I(sbi, i);
1543                 mutex_lock(&sum->curseg_mutex);
1544                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1545                 mutex_unlock(&sum->curseg_mutex);
1546         }
1547 }
1548
1549 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1550 {
1551         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1552                 write_compacted_summaries(sbi, start_blk);
1553         else
1554                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1555 }
1556
1557 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1558 {
1559         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1560                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1561 }
1562
1563 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1564                                         unsigned int val, int alloc)
1565 {
1566         int i;
1567
1568         if (type == NAT_JOURNAL) {
1569                 for (i = 0; i < nats_in_cursum(sum); i++) {
1570                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1571                                 return i;
1572                 }
1573                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1574                         return update_nats_in_cursum(sum, 1);
1575         } else if (type == SIT_JOURNAL) {
1576                 for (i = 0; i < sits_in_cursum(sum); i++)
1577                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1578                                 return i;
1579                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1580                         return update_sits_in_cursum(sum, 1);
1581         }
1582         return -1;
1583 }
1584
1585 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1586                                         unsigned int segno)
1587 {
1588         return get_meta_page(sbi, current_sit_addr(sbi, segno));
1589 }
1590
1591 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1592                                         unsigned int start)
1593 {
1594         struct sit_info *sit_i = SIT_I(sbi);
1595         struct page *src_page, *dst_page;
1596         pgoff_t src_off, dst_off;
1597         void *src_addr, *dst_addr;
1598
1599         src_off = current_sit_addr(sbi, start);
1600         dst_off = next_sit_addr(sbi, src_off);
1601
1602         /* get current sit block page without lock */
1603         src_page = get_meta_page(sbi, src_off);
1604         dst_page = grab_meta_page(sbi, dst_off);
1605         f2fs_bug_on(sbi, PageDirty(src_page));
1606
1607         src_addr = page_address(src_page);
1608         dst_addr = page_address(dst_page);
1609         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1610
1611         set_page_dirty(dst_page);
1612         f2fs_put_page(src_page, 1);
1613
1614         set_to_next_sit(sit_i, start);
1615
1616         return dst_page;
1617 }
1618
1619 static struct sit_entry_set *grab_sit_entry_set(void)
1620 {
1621         struct sit_entry_set *ses =
1622                         f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1623
1624         ses->entry_cnt = 0;
1625         INIT_LIST_HEAD(&ses->set_list);
1626         return ses;
1627 }
1628
1629 static void release_sit_entry_set(struct sit_entry_set *ses)
1630 {
1631         list_del(&ses->set_list);
1632         kmem_cache_free(sit_entry_set_slab, ses);
1633 }
1634
1635 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1636                                                 struct list_head *head)
1637 {
1638         struct sit_entry_set *next = ses;
1639
1640         if (list_is_last(&ses->set_list, head))
1641                 return;
1642
1643         list_for_each_entry_continue(next, head, set_list)
1644                 if (ses->entry_cnt <= next->entry_cnt)
1645                         break;
1646
1647         list_move_tail(&ses->set_list, &next->set_list);
1648 }
1649
1650 static void add_sit_entry(unsigned int segno, struct list_head *head)
1651 {
1652         struct sit_entry_set *ses;
1653         unsigned int start_segno = START_SEGNO(segno);
1654
1655         list_for_each_entry(ses, head, set_list) {
1656                 if (ses->start_segno == start_segno) {
1657                         ses->entry_cnt++;
1658                         adjust_sit_entry_set(ses, head);
1659                         return;
1660                 }
1661         }
1662
1663         ses = grab_sit_entry_set();
1664
1665         ses->start_segno = start_segno;
1666         ses->entry_cnt++;
1667         list_add(&ses->set_list, head);
1668 }
1669
1670 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1671 {
1672         struct f2fs_sm_info *sm_info = SM_I(sbi);
1673         struct list_head *set_list = &sm_info->sit_entry_set;
1674         unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1675         unsigned int segno;
1676
1677         for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1678                 add_sit_entry(segno, set_list);
1679 }
1680
1681 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1682 {
1683         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1684         struct f2fs_summary_block *sum = curseg->sum_blk;
1685         int i;
1686
1687         for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1688                 unsigned int segno;
1689                 bool dirtied;
1690
1691                 segno = le32_to_cpu(segno_in_journal(sum, i));
1692                 dirtied = __mark_sit_entry_dirty(sbi, segno);
1693
1694                 if (!dirtied)
1695                         add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1696         }
1697         update_sits_in_cursum(sum, -sits_in_cursum(sum));
1698 }
1699
1700 /*
1701  * CP calls this function, which flushes SIT entries including sit_journal,
1702  * and moves prefree segs to free segs.
1703  */
1704 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1705 {
1706         struct sit_info *sit_i = SIT_I(sbi);
1707         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1708         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1709         struct f2fs_summary_block *sum = curseg->sum_blk;
1710         struct sit_entry_set *ses, *tmp;
1711         struct list_head *head = &SM_I(sbi)->sit_entry_set;
1712         bool to_journal = true;
1713         struct seg_entry *se;
1714
1715         mutex_lock(&curseg->curseg_mutex);
1716         mutex_lock(&sit_i->sentry_lock);
1717
1718         /*
1719          * add and account sit entries of dirty bitmap in sit entry
1720          * set temporarily
1721          */
1722         add_sits_in_set(sbi);
1723
1724         /*
1725          * if there are no enough space in journal to store dirty sit
1726          * entries, remove all entries from journal and add and account
1727          * them in sit entry set.
1728          */
1729         if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1730                 remove_sits_in_journal(sbi);
1731
1732         if (!sit_i->dirty_sentries)
1733                 goto out;
1734
1735         /*
1736          * there are two steps to flush sit entries:
1737          * #1, flush sit entries to journal in current cold data summary block.
1738          * #2, flush sit entries to sit page.
1739          */
1740         list_for_each_entry_safe(ses, tmp, head, set_list) {
1741                 struct page *page = NULL;
1742                 struct f2fs_sit_block *raw_sit = NULL;
1743                 unsigned int start_segno = ses->start_segno;
1744                 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1745                                                 (unsigned long)MAIN_SEGS(sbi));
1746                 unsigned int segno = start_segno;
1747
1748                 if (to_journal &&
1749                         !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1750                         to_journal = false;
1751
1752                 if (!to_journal) {
1753                         page = get_next_sit_page(sbi, start_segno);
1754                         raw_sit = page_address(page);
1755                 }
1756
1757                 /* flush dirty sit entries in region of current sit set */
1758                 for_each_set_bit_from(segno, bitmap, end) {
1759                         int offset, sit_offset;
1760
1761                         se = get_seg_entry(sbi, segno);
1762
1763                         /* add discard candidates */
1764                         if (cpc->reason != CP_DISCARD) {
1765                                 cpc->trim_start = segno;
1766                                 add_discard_addrs(sbi, cpc);
1767                         }
1768
1769                         if (to_journal) {
1770                                 offset = lookup_journal_in_cursum(sum,
1771                                                         SIT_JOURNAL, segno, 1);
1772                                 f2fs_bug_on(sbi, offset < 0);
1773                                 segno_in_journal(sum, offset) =
1774                                                         cpu_to_le32(segno);
1775                                 seg_info_to_raw_sit(se,
1776                                                 &sit_in_journal(sum, offset));
1777                         } else {
1778                                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1779                                 seg_info_to_raw_sit(se,
1780                                                 &raw_sit->entries[sit_offset]);
1781                         }
1782
1783                         __clear_bit(segno, bitmap);
1784                         sit_i->dirty_sentries--;
1785                         ses->entry_cnt--;
1786                 }
1787
1788                 if (!to_journal)
1789                         f2fs_put_page(page, 1);
1790
1791                 f2fs_bug_on(sbi, ses->entry_cnt);
1792                 release_sit_entry_set(ses);
1793         }
1794
1795         f2fs_bug_on(sbi, !list_empty(head));
1796         f2fs_bug_on(sbi, sit_i->dirty_sentries);
1797 out:
1798         if (cpc->reason == CP_DISCARD) {
1799                 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1800                         add_discard_addrs(sbi, cpc);
1801         }
1802         mutex_unlock(&sit_i->sentry_lock);
1803         mutex_unlock(&curseg->curseg_mutex);
1804
1805         set_prefree_as_free_segments(sbi);
1806 }
1807
1808 static int build_sit_info(struct f2fs_sb_info *sbi)
1809 {
1810         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1811         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1812         struct sit_info *sit_i;
1813         unsigned int sit_segs, start;
1814         char *src_bitmap, *dst_bitmap;
1815         unsigned int bitmap_size;
1816
1817         /* allocate memory for SIT information */
1818         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1819         if (!sit_i)
1820                 return -ENOMEM;
1821
1822         SM_I(sbi)->sit_info = sit_i;
1823
1824         sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1825         if (!sit_i->sentries)
1826                 return -ENOMEM;
1827
1828         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1829         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1830         if (!sit_i->dirty_sentries_bitmap)
1831                 return -ENOMEM;
1832
1833         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1834                 sit_i->sentries[start].cur_valid_map
1835                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1836                 sit_i->sentries[start].ckpt_valid_map
1837                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1838                 if (!sit_i->sentries[start].cur_valid_map
1839                                 || !sit_i->sentries[start].ckpt_valid_map)
1840                         return -ENOMEM;
1841         }
1842
1843         if (sbi->segs_per_sec > 1) {
1844                 sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1845                                         sizeof(struct sec_entry));
1846                 if (!sit_i->sec_entries)
1847                         return -ENOMEM;
1848         }
1849
1850         /* get information related with SIT */
1851         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1852
1853         /* setup SIT bitmap from ckeckpoint pack */
1854         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1855         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1856
1857         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1858         if (!dst_bitmap)
1859                 return -ENOMEM;
1860
1861         /* init SIT information */
1862         sit_i->s_ops = &default_salloc_ops;
1863
1864         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1865         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1866         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1867         sit_i->sit_bitmap = dst_bitmap;
1868         sit_i->bitmap_size = bitmap_size;
1869         sit_i->dirty_sentries = 0;
1870         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1871         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1872         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1873         mutex_init(&sit_i->sentry_lock);
1874         return 0;
1875 }
1876
1877 static int build_free_segmap(struct f2fs_sb_info *sbi)
1878 {
1879         struct free_segmap_info *free_i;
1880         unsigned int bitmap_size, sec_bitmap_size;
1881
1882         /* allocate memory for free segmap information */
1883         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1884         if (!free_i)
1885                 return -ENOMEM;
1886
1887         SM_I(sbi)->free_info = free_i;
1888
1889         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1890         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1891         if (!free_i->free_segmap)
1892                 return -ENOMEM;
1893
1894         sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1895         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1896         if (!free_i->free_secmap)
1897                 return -ENOMEM;
1898
1899         /* set all segments as dirty temporarily */
1900         memset(free_i->free_segmap, 0xff, bitmap_size);
1901         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1902
1903         /* init free segmap information */
1904         free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1905         free_i->free_segments = 0;
1906         free_i->free_sections = 0;
1907         rwlock_init(&free_i->segmap_lock);
1908         return 0;
1909 }
1910
1911 static int build_curseg(struct f2fs_sb_info *sbi)
1912 {
1913         struct curseg_info *array;
1914         int i;
1915
1916         array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1917         if (!array)
1918                 return -ENOMEM;
1919
1920         SM_I(sbi)->curseg_array = array;
1921
1922         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1923                 mutex_init(&array[i].curseg_mutex);
1924                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1925                 if (!array[i].sum_blk)
1926                         return -ENOMEM;
1927                 array[i].segno = NULL_SEGNO;
1928                 array[i].next_blkoff = 0;
1929         }
1930         return restore_curseg_summaries(sbi);
1931 }
1932
1933 static void build_sit_entries(struct f2fs_sb_info *sbi)
1934 {
1935         struct sit_info *sit_i = SIT_I(sbi);
1936         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1937         struct f2fs_summary_block *sum = curseg->sum_blk;
1938         int sit_blk_cnt = SIT_BLK_CNT(sbi);
1939         unsigned int i, start, end;
1940         unsigned int readed, start_blk = 0;
1941         int nrpages = MAX_BIO_BLOCKS(sbi);
1942
1943         do {
1944                 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1945
1946                 start = start_blk * sit_i->sents_per_block;
1947                 end = (start_blk + readed) * sit_i->sents_per_block;
1948
1949                 for (; start < end && start < MAIN_SEGS(sbi); start++) {
1950                         struct seg_entry *se = &sit_i->sentries[start];
1951                         struct f2fs_sit_block *sit_blk;
1952                         struct f2fs_sit_entry sit;
1953                         struct page *page;
1954
1955                         mutex_lock(&curseg->curseg_mutex);
1956                         for (i = 0; i < sits_in_cursum(sum); i++) {
1957                                 if (le32_to_cpu(segno_in_journal(sum, i))
1958                                                                 == start) {
1959                                         sit = sit_in_journal(sum, i);
1960                                         mutex_unlock(&curseg->curseg_mutex);
1961                                         goto got_it;
1962                                 }
1963                         }
1964                         mutex_unlock(&curseg->curseg_mutex);
1965
1966                         page = get_current_sit_page(sbi, start);
1967                         sit_blk = (struct f2fs_sit_block *)page_address(page);
1968                         sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1969                         f2fs_put_page(page, 1);
1970 got_it:
1971                         check_block_count(sbi, start, &sit);
1972                         seg_info_from_raw_sit(se, &sit);
1973                         if (sbi->segs_per_sec > 1) {
1974                                 struct sec_entry *e = get_sec_entry(sbi, start);
1975                                 e->valid_blocks += se->valid_blocks;
1976                         }
1977                 }
1978                 start_blk += readed;
1979         } while (start_blk < sit_blk_cnt);
1980 }
1981
1982 static void init_free_segmap(struct f2fs_sb_info *sbi)
1983 {
1984         unsigned int start;
1985         int type;
1986
1987         for (start = 0; start < MAIN_SEGS(sbi); start++) {
1988                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1989                 if (!sentry->valid_blocks)
1990                         __set_free(sbi, start);
1991         }
1992
1993         /* set use the current segments */
1994         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1995                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1996                 __set_test_and_inuse(sbi, curseg_t->segno);
1997         }
1998 }
1999
2000 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2001 {
2002         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2003         struct free_segmap_info *free_i = FREE_I(sbi);
2004         unsigned int segno = 0, offset = 0;
2005         unsigned short valid_blocks;
2006
2007         while (1) {
2008                 /* find dirty segment based on free segmap */
2009                 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2010                 if (segno >= MAIN_SEGS(sbi))
2011                         break;
2012                 offset = segno + 1;
2013                 valid_blocks = get_valid_blocks(sbi, segno, 0);
2014                 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2015                         continue;
2016                 if (valid_blocks > sbi->blocks_per_seg) {
2017                         f2fs_bug_on(sbi, 1);
2018                         continue;
2019                 }
2020                 mutex_lock(&dirty_i->seglist_lock);
2021                 __locate_dirty_segment(sbi, segno, DIRTY);
2022                 mutex_unlock(&dirty_i->seglist_lock);
2023         }
2024 }
2025
2026 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2027 {
2028         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2029         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2030
2031         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2032         if (!dirty_i->victim_secmap)
2033                 return -ENOMEM;
2034         return 0;
2035 }
2036
2037 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2038 {
2039         struct dirty_seglist_info *dirty_i;
2040         unsigned int bitmap_size, i;
2041
2042         /* allocate memory for dirty segments list information */
2043         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2044         if (!dirty_i)
2045                 return -ENOMEM;
2046
2047         SM_I(sbi)->dirty_info = dirty_i;
2048         mutex_init(&dirty_i->seglist_lock);
2049
2050         bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2051
2052         for (i = 0; i < NR_DIRTY_TYPE; i++) {
2053                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2054                 if (!dirty_i->dirty_segmap[i])
2055                         return -ENOMEM;
2056         }
2057
2058         init_dirty_segmap(sbi);
2059         return init_victim_secmap(sbi);
2060 }
2061
2062 /*
2063  * Update min, max modified time for cost-benefit GC algorithm
2064  */
2065 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2066 {
2067         struct sit_info *sit_i = SIT_I(sbi);
2068         unsigned int segno;
2069
2070         mutex_lock(&sit_i->sentry_lock);
2071
2072         sit_i->min_mtime = LLONG_MAX;
2073
2074         for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2075                 unsigned int i;
2076                 unsigned long long mtime = 0;
2077
2078                 for (i = 0; i < sbi->segs_per_sec; i++)
2079                         mtime += get_seg_entry(sbi, segno + i)->mtime;
2080
2081                 mtime = div_u64(mtime, sbi->segs_per_sec);
2082
2083                 if (sit_i->min_mtime > mtime)
2084                         sit_i->min_mtime = mtime;
2085         }
2086         sit_i->max_mtime = get_mtime(sbi);
2087         mutex_unlock(&sit_i->sentry_lock);
2088 }
2089
2090 int build_segment_manager(struct f2fs_sb_info *sbi)
2091 {
2092         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2093         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2094         struct f2fs_sm_info *sm_info;
2095         int err;
2096
2097         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2098         if (!sm_info)
2099                 return -ENOMEM;
2100
2101         /* init sm info */
2102         sbi->sm_info = sm_info;
2103         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2104         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2105         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2106         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2107         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2108         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2109         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2110         sm_info->rec_prefree_segments = sm_info->main_segments *
2111                                         DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2112         sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2113         sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2114         sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2115
2116         INIT_LIST_HEAD(&sm_info->discard_list);
2117         sm_info->nr_discards = 0;
2118         sm_info->max_discards = 0;
2119
2120         INIT_LIST_HEAD(&sm_info->sit_entry_set);
2121
2122         if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2123                 err = create_flush_cmd_control(sbi);
2124                 if (err)
2125                         return err;
2126         }
2127
2128         err = build_sit_info(sbi);
2129         if (err)
2130                 return err;
2131         err = build_free_segmap(sbi);
2132         if (err)
2133                 return err;
2134         err = build_curseg(sbi);
2135         if (err)
2136                 return err;
2137
2138         /* reinit free segmap based on SIT */
2139         build_sit_entries(sbi);
2140
2141         init_free_segmap(sbi);
2142         err = build_dirty_segmap(sbi);
2143         if (err)
2144                 return err;
2145
2146         init_min_max_mtime(sbi);
2147         return 0;
2148 }
2149
2150 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2151                 enum dirty_type dirty_type)
2152 {
2153         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2154
2155         mutex_lock(&dirty_i->seglist_lock);
2156         kfree(dirty_i->dirty_segmap[dirty_type]);
2157         dirty_i->nr_dirty[dirty_type] = 0;
2158         mutex_unlock(&dirty_i->seglist_lock);
2159 }
2160
2161 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2162 {
2163         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2164         kfree(dirty_i->victim_secmap);
2165 }
2166
2167 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2168 {
2169         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2170         int i;
2171
2172         if (!dirty_i)
2173                 return;
2174
2175         /* discard pre-free/dirty segments list */
2176         for (i = 0; i < NR_DIRTY_TYPE; i++)
2177                 discard_dirty_segmap(sbi, i);
2178
2179         destroy_victim_secmap(sbi);
2180         SM_I(sbi)->dirty_info = NULL;
2181         kfree(dirty_i);
2182 }
2183
2184 static void destroy_curseg(struct f2fs_sb_info *sbi)
2185 {
2186         struct curseg_info *array = SM_I(sbi)->curseg_array;
2187         int i;
2188
2189         if (!array)
2190                 return;
2191         SM_I(sbi)->curseg_array = NULL;
2192         for (i = 0; i < NR_CURSEG_TYPE; i++)
2193                 kfree(array[i].sum_blk);
2194         kfree(array);
2195 }
2196
2197 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2198 {
2199         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2200         if (!free_i)
2201                 return;
2202         SM_I(sbi)->free_info = NULL;
2203         kfree(free_i->free_segmap);
2204         kfree(free_i->free_secmap);
2205         kfree(free_i);
2206 }
2207
2208 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2209 {
2210         struct sit_info *sit_i = SIT_I(sbi);
2211         unsigned int start;
2212
2213         if (!sit_i)
2214                 return;
2215
2216         if (sit_i->sentries) {
2217                 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2218                         kfree(sit_i->sentries[start].cur_valid_map);
2219                         kfree(sit_i->sentries[start].ckpt_valid_map);
2220                 }
2221         }
2222         vfree(sit_i->sentries);
2223         vfree(sit_i->sec_entries);
2224         kfree(sit_i->dirty_sentries_bitmap);
2225
2226         SM_I(sbi)->sit_info = NULL;
2227         kfree(sit_i->sit_bitmap);
2228         kfree(sit_i);
2229 }
2230
2231 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2232 {
2233         struct f2fs_sm_info *sm_info = SM_I(sbi);
2234
2235         if (!sm_info)
2236                 return;
2237         destroy_flush_cmd_control(sbi);
2238         destroy_dirty_segmap(sbi);
2239         destroy_curseg(sbi);
2240         destroy_free_segmap(sbi);
2241         destroy_sit_info(sbi);
2242         sbi->sm_info = NULL;
2243         kfree(sm_info);
2244 }
2245
2246 int __init create_segment_manager_caches(void)
2247 {
2248         discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2249                         sizeof(struct discard_entry));
2250         if (!discard_entry_slab)
2251                 goto fail;
2252
2253         sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2254                         sizeof(struct sit_entry_set));
2255         if (!sit_entry_set_slab)
2256                 goto destory_discard_entry;
2257
2258         inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2259                         sizeof(struct inmem_pages));
2260         if (!inmem_entry_slab)
2261                 goto destroy_sit_entry_set;
2262         return 0;
2263
2264 destroy_sit_entry_set:
2265         kmem_cache_destroy(sit_entry_set_slab);
2266 destory_discard_entry:
2267         kmem_cache_destroy(discard_entry_slab);
2268 fail:
2269         return -ENOMEM;
2270 }
2271
2272 void destroy_segment_manager_caches(void)
2273 {
2274         kmem_cache_destroy(sit_entry_set_slab);
2275         kmem_cache_destroy(discard_entry_slab);
2276         kmem_cache_destroy(inmem_entry_slab);
2277 }