Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / fs / f2fs / segment.c
blob7caac5f2ca9eec01fe1c92a494c8f32ce238e72e
1 /*
2 * fs/f2fs/segment.c
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
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 #include <linux/swap.h>
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include <trace/events/f2fs.h>
24 #define __reverse_ffz(x) __reverse_ffs(~(x))
26 static struct kmem_cache *discard_entry_slab;
29 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
30 * MSB and LSB are reversed in a byte by f2fs_set_bit.
32 static inline unsigned long __reverse_ffs(unsigned long word)
34 int num = 0;
36 #if BITS_PER_LONG == 64
37 if ((word & 0xffffffff) == 0) {
38 num += 32;
39 word >>= 32;
41 #endif
42 if ((word & 0xffff) == 0) {
43 num += 16;
44 word >>= 16;
46 if ((word & 0xff) == 0) {
47 num += 8;
48 word >>= 8;
50 if ((word & 0xf0) == 0)
51 num += 4;
52 else
53 word >>= 4;
54 if ((word & 0xc) == 0)
55 num += 2;
56 else
57 word >>= 2;
58 if ((word & 0x2) == 0)
59 num += 1;
60 return num;
64 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c becasue
65 * f2fs_set_bit makes MSB and LSB reversed in a byte.
66 * Example:
67 * LSB <--> MSB
68 * f2fs_set_bit(0, bitmap) => 0000 0001
69 * f2fs_set_bit(7, bitmap) => 1000 0000
71 static unsigned long __find_rev_next_bit(const unsigned long *addr,
72 unsigned long size, unsigned long offset)
74 const unsigned long *p = addr + BIT_WORD(offset);
75 unsigned long result = offset & ~(BITS_PER_LONG - 1);
76 unsigned long tmp;
77 unsigned long mask, submask;
78 unsigned long quot, rest;
80 if (offset >= size)
81 return size;
83 size -= result;
84 offset %= BITS_PER_LONG;
85 if (!offset)
86 goto aligned;
88 tmp = *(p++);
89 quot = (offset >> 3) << 3;
90 rest = offset & 0x7;
91 mask = ~0UL << quot;
92 submask = (unsigned char)(0xff << rest) >> rest;
93 submask <<= quot;
94 mask &= submask;
95 tmp &= mask;
96 if (size < BITS_PER_LONG)
97 goto found_first;
98 if (tmp)
99 goto found_middle;
101 size -= BITS_PER_LONG;
102 result += BITS_PER_LONG;
103 aligned:
104 while (size & ~(BITS_PER_LONG-1)) {
105 tmp = *(p++);
106 if (tmp)
107 goto found_middle;
108 result += BITS_PER_LONG;
109 size -= BITS_PER_LONG;
111 if (!size)
112 return result;
113 tmp = *p;
114 found_first:
115 tmp &= (~0UL >> (BITS_PER_LONG - size));
116 if (tmp == 0UL) /* Are any bits set? */
117 return result + size; /* Nope. */
118 found_middle:
119 return result + __reverse_ffs(tmp);
122 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
123 unsigned long size, unsigned long offset)
125 const unsigned long *p = addr + BIT_WORD(offset);
126 unsigned long result = offset & ~(BITS_PER_LONG - 1);
127 unsigned long tmp;
128 unsigned long mask, submask;
129 unsigned long quot, rest;
131 if (offset >= size)
132 return size;
134 size -= result;
135 offset %= BITS_PER_LONG;
136 if (!offset)
137 goto aligned;
139 tmp = *(p++);
140 quot = (offset >> 3) << 3;
141 rest = offset & 0x7;
142 mask = ~(~0UL << quot);
143 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
144 submask <<= quot;
145 mask += submask;
146 tmp |= mask;
147 if (size < BITS_PER_LONG)
148 goto found_first;
149 if (~tmp)
150 goto found_middle;
152 size -= BITS_PER_LONG;
153 result += BITS_PER_LONG;
154 aligned:
155 while (size & ~(BITS_PER_LONG - 1)) {
156 tmp = *(p++);
157 if (~tmp)
158 goto found_middle;
159 result += BITS_PER_LONG;
160 size -= BITS_PER_LONG;
162 if (!size)
163 return result;
164 tmp = *p;
166 found_first:
167 tmp |= ~0UL << size;
168 if (tmp == ~0UL) /* Are any bits zero? */
169 return result + size; /* Nope. */
170 found_middle:
171 return result + __reverse_ffz(tmp);
175 * This function balances dirty node and dentry pages.
176 * In addition, it controls garbage collection.
178 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
181 * We should do GC or end up with checkpoint, if there are so many dirty
182 * dir/node pages without enough free segments.
184 if (has_not_enough_free_secs(sbi, 0)) {
185 mutex_lock(&sbi->gc_mutex);
186 f2fs_gc(sbi);
190 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
192 /* check the # of cached NAT entries and prefree segments */
193 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
194 excess_prefree_segs(sbi))
195 f2fs_sync_fs(sbi->sb, true);
198 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
199 enum dirty_type dirty_type)
201 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
203 /* need not be added */
204 if (IS_CURSEG(sbi, segno))
205 return;
207 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
208 dirty_i->nr_dirty[dirty_type]++;
210 if (dirty_type == DIRTY) {
211 struct seg_entry *sentry = get_seg_entry(sbi, segno);
212 enum dirty_type t = sentry->type;
214 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
215 dirty_i->nr_dirty[t]++;
219 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
220 enum dirty_type dirty_type)
222 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
224 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
225 dirty_i->nr_dirty[dirty_type]--;
227 if (dirty_type == DIRTY) {
228 struct seg_entry *sentry = get_seg_entry(sbi, segno);
229 enum dirty_type t = sentry->type;
231 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
232 dirty_i->nr_dirty[t]--;
234 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
235 clear_bit(GET_SECNO(sbi, segno),
236 dirty_i->victim_secmap);
241 * Should not occur error such as -ENOMEM.
242 * Adding dirty entry into seglist is not critical operation.
243 * If a given segment is one of current working segments, it won't be added.
245 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
247 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
248 unsigned short valid_blocks;
250 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
251 return;
253 mutex_lock(&dirty_i->seglist_lock);
255 valid_blocks = get_valid_blocks(sbi, segno, 0);
257 if (valid_blocks == 0) {
258 __locate_dirty_segment(sbi, segno, PRE);
259 __remove_dirty_segment(sbi, segno, DIRTY);
260 } else if (valid_blocks < sbi->blocks_per_seg) {
261 __locate_dirty_segment(sbi, segno, DIRTY);
262 } else {
263 /* Recovery routine with SSR needs this */
264 __remove_dirty_segment(sbi, segno, DIRTY);
267 mutex_unlock(&dirty_i->seglist_lock);
270 static void f2fs_issue_discard(struct f2fs_sb_info *sbi,
271 block_t blkstart, block_t blklen)
273 sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
274 sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
275 blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
276 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
279 static void add_discard_addrs(struct f2fs_sb_info *sbi,
280 unsigned int segno, struct seg_entry *se)
282 struct list_head *head = &SM_I(sbi)->discard_list;
283 struct discard_entry *new;
284 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
285 int max_blocks = sbi->blocks_per_seg;
286 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
287 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
288 unsigned long dmap[entries];
289 unsigned int start = 0, end = -1;
290 int i;
292 if (!test_opt(sbi, DISCARD))
293 return;
295 /* zero block will be discarded through the prefree list */
296 if (!se->valid_blocks || se->valid_blocks == max_blocks)
297 return;
299 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
300 for (i = 0; i < entries; i++)
301 dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
303 while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
304 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
305 if (start >= max_blocks)
306 break;
308 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
310 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
311 INIT_LIST_HEAD(&new->list);
312 new->blkaddr = START_BLOCK(sbi, segno) + start;
313 new->len = end - start;
315 list_add_tail(&new->list, head);
316 SM_I(sbi)->nr_discards += end - start;
321 * Should call clear_prefree_segments after checkpoint is done.
323 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
325 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
326 unsigned int segno = -1;
327 unsigned int total_segs = TOTAL_SEGS(sbi);
329 mutex_lock(&dirty_i->seglist_lock);
330 while (1) {
331 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
332 segno + 1);
333 if (segno >= total_segs)
334 break;
335 __set_test_and_free(sbi, segno);
337 mutex_unlock(&dirty_i->seglist_lock);
340 void clear_prefree_segments(struct f2fs_sb_info *sbi)
342 struct list_head *head = &(SM_I(sbi)->discard_list);
343 struct list_head *this, *next;
344 struct discard_entry *entry;
345 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
346 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
347 unsigned int total_segs = TOTAL_SEGS(sbi);
348 unsigned int start = 0, end = -1;
350 mutex_lock(&dirty_i->seglist_lock);
352 while (1) {
353 int i;
354 start = find_next_bit(prefree_map, total_segs, end + 1);
355 if (start >= total_segs)
356 break;
357 end = find_next_zero_bit(prefree_map, total_segs, start + 1);
359 for (i = start; i < end; i++)
360 clear_bit(i, prefree_map);
362 dirty_i->nr_dirty[PRE] -= end - start;
364 if (!test_opt(sbi, DISCARD))
365 continue;
367 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
368 (end - start) << sbi->log_blocks_per_seg);
370 mutex_unlock(&dirty_i->seglist_lock);
372 /* send small discards */
373 list_for_each_safe(this, next, head) {
374 entry = list_entry(this, struct discard_entry, list);
375 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
376 list_del(&entry->list);
377 SM_I(sbi)->nr_discards -= entry->len;
378 kmem_cache_free(discard_entry_slab, entry);
382 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
384 struct sit_info *sit_i = SIT_I(sbi);
385 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
386 sit_i->dirty_sentries++;
389 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
390 unsigned int segno, int modified)
392 struct seg_entry *se = get_seg_entry(sbi, segno);
393 se->type = type;
394 if (modified)
395 __mark_sit_entry_dirty(sbi, segno);
398 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
400 struct seg_entry *se;
401 unsigned int segno, offset;
402 long int new_vblocks;
404 segno = GET_SEGNO(sbi, blkaddr);
406 se = get_seg_entry(sbi, segno);
407 new_vblocks = se->valid_blocks + del;
408 offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
410 f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
411 (new_vblocks > sbi->blocks_per_seg)));
413 se->valid_blocks = new_vblocks;
414 se->mtime = get_mtime(sbi);
415 SIT_I(sbi)->max_mtime = se->mtime;
417 /* Update valid block bitmap */
418 if (del > 0) {
419 if (f2fs_set_bit(offset, se->cur_valid_map))
420 BUG();
421 } else {
422 if (!f2fs_clear_bit(offset, se->cur_valid_map))
423 BUG();
425 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
426 se->ckpt_valid_blocks += del;
428 __mark_sit_entry_dirty(sbi, segno);
430 /* update total number of valid blocks to be written in ckpt area */
431 SIT_I(sbi)->written_valid_blocks += del;
433 if (sbi->segs_per_sec > 1)
434 get_sec_entry(sbi, segno)->valid_blocks += del;
437 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
438 block_t old_blkaddr, block_t new_blkaddr)
440 update_sit_entry(sbi, new_blkaddr, 1);
441 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
442 update_sit_entry(sbi, old_blkaddr, -1);
445 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
447 unsigned int segno = GET_SEGNO(sbi, addr);
448 struct sit_info *sit_i = SIT_I(sbi);
450 f2fs_bug_on(addr == NULL_ADDR);
451 if (addr == NEW_ADDR)
452 return;
454 /* add it into sit main buffer */
455 mutex_lock(&sit_i->sentry_lock);
457 update_sit_entry(sbi, addr, -1);
459 /* add it into dirty seglist */
460 locate_dirty_segment(sbi, segno);
462 mutex_unlock(&sit_i->sentry_lock);
466 * This function should be resided under the curseg_mutex lock
468 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
469 struct f2fs_summary *sum)
471 struct curseg_info *curseg = CURSEG_I(sbi, type);
472 void *addr = curseg->sum_blk;
473 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
474 memcpy(addr, sum, sizeof(struct f2fs_summary));
478 * Calculate the number of current summary pages for writing
480 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
482 int valid_sum_count = 0;
483 int i, sum_in_page;
485 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
486 if (sbi->ckpt->alloc_type[i] == SSR)
487 valid_sum_count += sbi->blocks_per_seg;
488 else
489 valid_sum_count += curseg_blkoff(sbi, i);
492 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
493 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
494 if (valid_sum_count <= sum_in_page)
495 return 1;
496 else if ((valid_sum_count - sum_in_page) <=
497 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
498 return 2;
499 return 3;
503 * Caller should put this summary page
505 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
507 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
510 static void write_sum_page(struct f2fs_sb_info *sbi,
511 struct f2fs_summary_block *sum_blk, block_t blk_addr)
513 struct page *page = grab_meta_page(sbi, blk_addr);
514 void *kaddr = page_address(page);
515 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
516 set_page_dirty(page);
517 f2fs_put_page(page, 1);
520 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
522 struct curseg_info *curseg = CURSEG_I(sbi, type);
523 unsigned int segno = curseg->segno + 1;
524 struct free_segmap_info *free_i = FREE_I(sbi);
526 if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
527 return !test_bit(segno, free_i->free_segmap);
528 return 0;
532 * Find a new segment from the free segments bitmap to right order
533 * This function should be returned with success, otherwise BUG
535 static void get_new_segment(struct f2fs_sb_info *sbi,
536 unsigned int *newseg, bool new_sec, int dir)
538 struct free_segmap_info *free_i = FREE_I(sbi);
539 unsigned int segno, secno, zoneno;
540 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
541 unsigned int hint = *newseg / sbi->segs_per_sec;
542 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
543 unsigned int left_start = hint;
544 bool init = true;
545 int go_left = 0;
546 int i;
548 write_lock(&free_i->segmap_lock);
550 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
551 segno = find_next_zero_bit(free_i->free_segmap,
552 TOTAL_SEGS(sbi), *newseg + 1);
553 if (segno - *newseg < sbi->segs_per_sec -
554 (*newseg % sbi->segs_per_sec))
555 goto got_it;
557 find_other_zone:
558 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
559 if (secno >= TOTAL_SECS(sbi)) {
560 if (dir == ALLOC_RIGHT) {
561 secno = find_next_zero_bit(free_i->free_secmap,
562 TOTAL_SECS(sbi), 0);
563 f2fs_bug_on(secno >= TOTAL_SECS(sbi));
564 } else {
565 go_left = 1;
566 left_start = hint - 1;
569 if (go_left == 0)
570 goto skip_left;
572 while (test_bit(left_start, free_i->free_secmap)) {
573 if (left_start > 0) {
574 left_start--;
575 continue;
577 left_start = find_next_zero_bit(free_i->free_secmap,
578 TOTAL_SECS(sbi), 0);
579 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
580 break;
582 secno = left_start;
583 skip_left:
584 hint = secno;
585 segno = secno * sbi->segs_per_sec;
586 zoneno = secno / sbi->secs_per_zone;
588 /* give up on finding another zone */
589 if (!init)
590 goto got_it;
591 if (sbi->secs_per_zone == 1)
592 goto got_it;
593 if (zoneno == old_zoneno)
594 goto got_it;
595 if (dir == ALLOC_LEFT) {
596 if (!go_left && zoneno + 1 >= total_zones)
597 goto got_it;
598 if (go_left && zoneno == 0)
599 goto got_it;
601 for (i = 0; i < NR_CURSEG_TYPE; i++)
602 if (CURSEG_I(sbi, i)->zone == zoneno)
603 break;
605 if (i < NR_CURSEG_TYPE) {
606 /* zone is in user, try another */
607 if (go_left)
608 hint = zoneno * sbi->secs_per_zone - 1;
609 else if (zoneno + 1 >= total_zones)
610 hint = 0;
611 else
612 hint = (zoneno + 1) * sbi->secs_per_zone;
613 init = false;
614 goto find_other_zone;
616 got_it:
617 /* set it as dirty segment in free segmap */
618 f2fs_bug_on(test_bit(segno, free_i->free_segmap));
619 __set_inuse(sbi, segno);
620 *newseg = segno;
621 write_unlock(&free_i->segmap_lock);
624 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
626 struct curseg_info *curseg = CURSEG_I(sbi, type);
627 struct summary_footer *sum_footer;
629 curseg->segno = curseg->next_segno;
630 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
631 curseg->next_blkoff = 0;
632 curseg->next_segno = NULL_SEGNO;
634 sum_footer = &(curseg->sum_blk->footer);
635 memset(sum_footer, 0, sizeof(struct summary_footer));
636 if (IS_DATASEG(type))
637 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
638 if (IS_NODESEG(type))
639 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
640 __set_sit_entry_type(sbi, type, curseg->segno, modified);
644 * Allocate a current working segment.
645 * This function always allocates a free segment in LFS manner.
647 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
649 struct curseg_info *curseg = CURSEG_I(sbi, type);
650 unsigned int segno = curseg->segno;
651 int dir = ALLOC_LEFT;
653 write_sum_page(sbi, curseg->sum_blk,
654 GET_SUM_BLOCK(sbi, segno));
655 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
656 dir = ALLOC_RIGHT;
658 if (test_opt(sbi, NOHEAP))
659 dir = ALLOC_RIGHT;
661 get_new_segment(sbi, &segno, new_sec, dir);
662 curseg->next_segno = segno;
663 reset_curseg(sbi, type, 1);
664 curseg->alloc_type = LFS;
667 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
668 struct curseg_info *seg, block_t start)
670 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
671 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
672 unsigned long target_map[entries];
673 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
674 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
675 int i, pos;
677 for (i = 0; i < entries; i++)
678 target_map[i] = ckpt_map[i] | cur_map[i];
680 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
682 seg->next_blkoff = pos;
686 * If a segment is written by LFS manner, next block offset is just obtained
687 * by increasing the current block offset. However, if a segment is written by
688 * SSR manner, next block offset obtained by calling __next_free_blkoff
690 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
691 struct curseg_info *seg)
693 if (seg->alloc_type == SSR)
694 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
695 else
696 seg->next_blkoff++;
700 * This function always allocates a used segment (from dirty seglist) by SSR
701 * manner, so it should recover the existing segment information of valid blocks
703 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
705 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
706 struct curseg_info *curseg = CURSEG_I(sbi, type);
707 unsigned int new_segno = curseg->next_segno;
708 struct f2fs_summary_block *sum_node;
709 struct page *sum_page;
711 write_sum_page(sbi, curseg->sum_blk,
712 GET_SUM_BLOCK(sbi, curseg->segno));
713 __set_test_and_inuse(sbi, new_segno);
715 mutex_lock(&dirty_i->seglist_lock);
716 __remove_dirty_segment(sbi, new_segno, PRE);
717 __remove_dirty_segment(sbi, new_segno, DIRTY);
718 mutex_unlock(&dirty_i->seglist_lock);
720 reset_curseg(sbi, type, 1);
721 curseg->alloc_type = SSR;
722 __next_free_blkoff(sbi, curseg, 0);
724 if (reuse) {
725 sum_page = get_sum_page(sbi, new_segno);
726 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
727 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
728 f2fs_put_page(sum_page, 1);
732 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
734 struct curseg_info *curseg = CURSEG_I(sbi, type);
735 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
737 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
738 return v_ops->get_victim(sbi,
739 &(curseg)->next_segno, BG_GC, type, SSR);
741 /* For data segments, let's do SSR more intensively */
742 for (; type >= CURSEG_HOT_DATA; type--)
743 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
744 BG_GC, type, SSR))
745 return 1;
746 return 0;
750 * flush out current segment and replace it with new segment
751 * This function should be returned with success, otherwise BUG
753 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
754 int type, bool force)
756 struct curseg_info *curseg = CURSEG_I(sbi, type);
758 if (force)
759 new_curseg(sbi, type, true);
760 else if (type == CURSEG_WARM_NODE)
761 new_curseg(sbi, type, false);
762 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
763 new_curseg(sbi, type, false);
764 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
765 change_curseg(sbi, type, true);
766 else
767 new_curseg(sbi, type, false);
769 stat_inc_seg_type(sbi, curseg);
772 void allocate_new_segments(struct f2fs_sb_info *sbi)
774 struct curseg_info *curseg;
775 unsigned int old_curseg;
776 int i;
778 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
779 curseg = CURSEG_I(sbi, i);
780 old_curseg = curseg->segno;
781 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
782 locate_dirty_segment(sbi, old_curseg);
786 static const struct segment_allocation default_salloc_ops = {
787 .allocate_segment = allocate_segment_by_default,
790 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
792 struct curseg_info *curseg = CURSEG_I(sbi, type);
793 if (curseg->next_blkoff < sbi->blocks_per_seg)
794 return true;
795 return false;
798 static int __get_segment_type_2(struct page *page, enum page_type p_type)
800 if (p_type == DATA)
801 return CURSEG_HOT_DATA;
802 else
803 return CURSEG_HOT_NODE;
806 static int __get_segment_type_4(struct page *page, enum page_type p_type)
808 if (p_type == DATA) {
809 struct inode *inode = page->mapping->host;
811 if (S_ISDIR(inode->i_mode))
812 return CURSEG_HOT_DATA;
813 else
814 return CURSEG_COLD_DATA;
815 } else {
816 if (IS_DNODE(page) && !is_cold_node(page))
817 return CURSEG_HOT_NODE;
818 else
819 return CURSEG_COLD_NODE;
823 static int __get_segment_type_6(struct page *page, enum page_type p_type)
825 if (p_type == DATA) {
826 struct inode *inode = page->mapping->host;
828 if (S_ISDIR(inode->i_mode))
829 return CURSEG_HOT_DATA;
830 else if (is_cold_data(page) || file_is_cold(inode))
831 return CURSEG_COLD_DATA;
832 else
833 return CURSEG_WARM_DATA;
834 } else {
835 if (IS_DNODE(page))
836 return is_cold_node(page) ? CURSEG_WARM_NODE :
837 CURSEG_HOT_NODE;
838 else
839 return CURSEG_COLD_NODE;
843 static int __get_segment_type(struct page *page, enum page_type p_type)
845 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
846 switch (sbi->active_logs) {
847 case 2:
848 return __get_segment_type_2(page, p_type);
849 case 4:
850 return __get_segment_type_4(page, p_type);
852 /* NR_CURSEG_TYPE(6) logs by default */
853 f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
854 return __get_segment_type_6(page, p_type);
857 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
858 block_t old_blkaddr, block_t *new_blkaddr,
859 struct f2fs_summary *sum, int type)
861 struct sit_info *sit_i = SIT_I(sbi);
862 struct curseg_info *curseg;
863 unsigned int old_cursegno;
865 curseg = CURSEG_I(sbi, type);
867 mutex_lock(&curseg->curseg_mutex);
869 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
870 old_cursegno = curseg->segno;
873 * __add_sum_entry should be resided under the curseg_mutex
874 * because, this function updates a summary entry in the
875 * current summary block.
877 __add_sum_entry(sbi, type, sum);
879 mutex_lock(&sit_i->sentry_lock);
880 __refresh_next_blkoff(sbi, curseg);
882 stat_inc_block_count(sbi, curseg);
885 * SIT information should be updated before segment allocation,
886 * since SSR needs latest valid block information.
888 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
890 if (!__has_curseg_space(sbi, type))
891 sit_i->s_ops->allocate_segment(sbi, type, false);
893 locate_dirty_segment(sbi, old_cursegno);
894 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
895 mutex_unlock(&sit_i->sentry_lock);
897 if (page && IS_NODESEG(type))
898 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
900 mutex_unlock(&curseg->curseg_mutex);
903 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
904 block_t old_blkaddr, block_t *new_blkaddr,
905 struct f2fs_summary *sum, struct f2fs_io_info *fio)
907 int type = __get_segment_type(page, fio->type);
909 allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
911 /* writeout dirty page into bdev */
912 f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
915 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
917 struct f2fs_io_info fio = {
918 .type = META,
919 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
922 set_page_writeback(page);
923 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
926 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
927 struct f2fs_io_info *fio,
928 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
930 struct f2fs_summary sum;
931 set_summary(&sum, nid, 0, 0);
932 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
935 void write_data_page(struct page *page, struct dnode_of_data *dn,
936 block_t *new_blkaddr, struct f2fs_io_info *fio)
938 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
939 struct f2fs_summary sum;
940 struct node_info ni;
942 f2fs_bug_on(dn->data_blkaddr == NULL_ADDR);
943 get_node_info(sbi, dn->nid, &ni);
944 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
946 do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
949 void rewrite_data_page(struct page *page, block_t old_blkaddr,
950 struct f2fs_io_info *fio)
952 struct inode *inode = page->mapping->host;
953 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
954 f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio);
957 void recover_data_page(struct f2fs_sb_info *sbi,
958 struct page *page, struct f2fs_summary *sum,
959 block_t old_blkaddr, block_t new_blkaddr)
961 struct sit_info *sit_i = SIT_I(sbi);
962 struct curseg_info *curseg;
963 unsigned int segno, old_cursegno;
964 struct seg_entry *se;
965 int type;
967 segno = GET_SEGNO(sbi, new_blkaddr);
968 se = get_seg_entry(sbi, segno);
969 type = se->type;
971 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
972 if (old_blkaddr == NULL_ADDR)
973 type = CURSEG_COLD_DATA;
974 else
975 type = CURSEG_WARM_DATA;
977 curseg = CURSEG_I(sbi, type);
979 mutex_lock(&curseg->curseg_mutex);
980 mutex_lock(&sit_i->sentry_lock);
982 old_cursegno = curseg->segno;
984 /* change the current segment */
985 if (segno != curseg->segno) {
986 curseg->next_segno = segno;
987 change_curseg(sbi, type, true);
990 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
991 (sbi->blocks_per_seg - 1);
992 __add_sum_entry(sbi, type, sum);
994 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
996 locate_dirty_segment(sbi, old_cursegno);
997 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
999 mutex_unlock(&sit_i->sentry_lock);
1000 mutex_unlock(&curseg->curseg_mutex);
1003 void rewrite_node_page(struct f2fs_sb_info *sbi,
1004 struct page *page, struct f2fs_summary *sum,
1005 block_t old_blkaddr, block_t new_blkaddr)
1007 struct sit_info *sit_i = SIT_I(sbi);
1008 int type = CURSEG_WARM_NODE;
1009 struct curseg_info *curseg;
1010 unsigned int segno, old_cursegno;
1011 block_t next_blkaddr = next_blkaddr_of_node(page);
1012 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
1013 struct f2fs_io_info fio = {
1014 .type = NODE,
1015 .rw = WRITE_SYNC,
1018 curseg = CURSEG_I(sbi, type);
1020 mutex_lock(&curseg->curseg_mutex);
1021 mutex_lock(&sit_i->sentry_lock);
1023 segno = GET_SEGNO(sbi, new_blkaddr);
1024 old_cursegno = curseg->segno;
1026 /* change the current segment */
1027 if (segno != curseg->segno) {
1028 curseg->next_segno = segno;
1029 change_curseg(sbi, type, true);
1031 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
1032 (sbi->blocks_per_seg - 1);
1033 __add_sum_entry(sbi, type, sum);
1035 /* change the current log to the next block addr in advance */
1036 if (next_segno != segno) {
1037 curseg->next_segno = next_segno;
1038 change_curseg(sbi, type, true);
1040 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
1041 (sbi->blocks_per_seg - 1);
1043 /* rewrite node page */
1044 set_page_writeback(page);
1045 f2fs_submit_page_mbio(sbi, page, new_blkaddr, &fio);
1046 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1047 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1049 locate_dirty_segment(sbi, old_cursegno);
1050 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1052 mutex_unlock(&sit_i->sentry_lock);
1053 mutex_unlock(&curseg->curseg_mutex);
1056 void f2fs_wait_on_page_writeback(struct page *page,
1057 enum page_type type)
1059 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1060 if (PageWriteback(page)) {
1061 f2fs_submit_merged_bio(sbi, type, WRITE);
1062 wait_on_page_writeback(page);
1066 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1068 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1069 struct curseg_info *seg_i;
1070 unsigned char *kaddr;
1071 struct page *page;
1072 block_t start;
1073 int i, j, offset;
1075 start = start_sum_block(sbi);
1077 page = get_meta_page(sbi, start++);
1078 kaddr = (unsigned char *)page_address(page);
1080 /* Step 1: restore nat cache */
1081 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1082 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1084 /* Step 2: restore sit cache */
1085 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1086 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1087 SUM_JOURNAL_SIZE);
1088 offset = 2 * SUM_JOURNAL_SIZE;
1090 /* Step 3: restore summary entries */
1091 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1092 unsigned short blk_off;
1093 unsigned int segno;
1095 seg_i = CURSEG_I(sbi, i);
1096 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1097 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1098 seg_i->next_segno = segno;
1099 reset_curseg(sbi, i, 0);
1100 seg_i->alloc_type = ckpt->alloc_type[i];
1101 seg_i->next_blkoff = blk_off;
1103 if (seg_i->alloc_type == SSR)
1104 blk_off = sbi->blocks_per_seg;
1106 for (j = 0; j < blk_off; j++) {
1107 struct f2fs_summary *s;
1108 s = (struct f2fs_summary *)(kaddr + offset);
1109 seg_i->sum_blk->entries[j] = *s;
1110 offset += SUMMARY_SIZE;
1111 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1112 SUM_FOOTER_SIZE)
1113 continue;
1115 f2fs_put_page(page, 1);
1116 page = NULL;
1118 page = get_meta_page(sbi, start++);
1119 kaddr = (unsigned char *)page_address(page);
1120 offset = 0;
1123 f2fs_put_page(page, 1);
1124 return 0;
1127 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1129 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1130 struct f2fs_summary_block *sum;
1131 struct curseg_info *curseg;
1132 struct page *new;
1133 unsigned short blk_off;
1134 unsigned int segno = 0;
1135 block_t blk_addr = 0;
1137 /* get segment number and block addr */
1138 if (IS_DATASEG(type)) {
1139 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1140 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1141 CURSEG_HOT_DATA]);
1142 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1143 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1144 else
1145 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1146 } else {
1147 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1148 CURSEG_HOT_NODE]);
1149 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1150 CURSEG_HOT_NODE]);
1151 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1152 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1153 type - CURSEG_HOT_NODE);
1154 else
1155 blk_addr = GET_SUM_BLOCK(sbi, segno);
1158 new = get_meta_page(sbi, blk_addr);
1159 sum = (struct f2fs_summary_block *)page_address(new);
1161 if (IS_NODESEG(type)) {
1162 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1163 struct f2fs_summary *ns = &sum->entries[0];
1164 int i;
1165 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1166 ns->version = 0;
1167 ns->ofs_in_node = 0;
1169 } else {
1170 if (restore_node_summary(sbi, segno, sum)) {
1171 f2fs_put_page(new, 1);
1172 return -EINVAL;
1177 /* set uncompleted segment to curseg */
1178 curseg = CURSEG_I(sbi, type);
1179 mutex_lock(&curseg->curseg_mutex);
1180 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1181 curseg->next_segno = segno;
1182 reset_curseg(sbi, type, 0);
1183 curseg->alloc_type = ckpt->alloc_type[type];
1184 curseg->next_blkoff = blk_off;
1185 mutex_unlock(&curseg->curseg_mutex);
1186 f2fs_put_page(new, 1);
1187 return 0;
1190 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1192 int type = CURSEG_HOT_DATA;
1194 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1195 /* restore for compacted data summary */
1196 if (read_compacted_summaries(sbi))
1197 return -EINVAL;
1198 type = CURSEG_HOT_NODE;
1201 for (; type <= CURSEG_COLD_NODE; type++)
1202 if (read_normal_summaries(sbi, type))
1203 return -EINVAL;
1204 return 0;
1207 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1209 struct page *page;
1210 unsigned char *kaddr;
1211 struct f2fs_summary *summary;
1212 struct curseg_info *seg_i;
1213 int written_size = 0;
1214 int i, j;
1216 page = grab_meta_page(sbi, blkaddr++);
1217 kaddr = (unsigned char *)page_address(page);
1219 /* Step 1: write nat cache */
1220 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1221 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1222 written_size += SUM_JOURNAL_SIZE;
1224 /* Step 2: write sit cache */
1225 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1226 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1227 SUM_JOURNAL_SIZE);
1228 written_size += SUM_JOURNAL_SIZE;
1230 /* Step 3: write summary entries */
1231 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1232 unsigned short blkoff;
1233 seg_i = CURSEG_I(sbi, i);
1234 if (sbi->ckpt->alloc_type[i] == SSR)
1235 blkoff = sbi->blocks_per_seg;
1236 else
1237 blkoff = curseg_blkoff(sbi, i);
1239 for (j = 0; j < blkoff; j++) {
1240 if (!page) {
1241 page = grab_meta_page(sbi, blkaddr++);
1242 kaddr = (unsigned char *)page_address(page);
1243 written_size = 0;
1245 summary = (struct f2fs_summary *)(kaddr + written_size);
1246 *summary = seg_i->sum_blk->entries[j];
1247 written_size += SUMMARY_SIZE;
1249 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1250 SUM_FOOTER_SIZE)
1251 continue;
1253 set_page_dirty(page);
1254 f2fs_put_page(page, 1);
1255 page = NULL;
1258 if (page) {
1259 set_page_dirty(page);
1260 f2fs_put_page(page, 1);
1264 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1265 block_t blkaddr, int type)
1267 int i, end;
1268 if (IS_DATASEG(type))
1269 end = type + NR_CURSEG_DATA_TYPE;
1270 else
1271 end = type + NR_CURSEG_NODE_TYPE;
1273 for (i = type; i < end; i++) {
1274 struct curseg_info *sum = CURSEG_I(sbi, i);
1275 mutex_lock(&sum->curseg_mutex);
1276 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1277 mutex_unlock(&sum->curseg_mutex);
1281 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1283 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1284 write_compacted_summaries(sbi, start_blk);
1285 else
1286 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1289 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1291 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1292 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1295 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1296 unsigned int val, int alloc)
1298 int i;
1300 if (type == NAT_JOURNAL) {
1301 for (i = 0; i < nats_in_cursum(sum); i++) {
1302 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1303 return i;
1305 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1306 return update_nats_in_cursum(sum, 1);
1307 } else if (type == SIT_JOURNAL) {
1308 for (i = 0; i < sits_in_cursum(sum); i++)
1309 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1310 return i;
1311 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1312 return update_sits_in_cursum(sum, 1);
1314 return -1;
1317 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1318 unsigned int segno)
1320 struct sit_info *sit_i = SIT_I(sbi);
1321 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1322 block_t blk_addr = sit_i->sit_base_addr + offset;
1324 check_seg_range(sbi, segno);
1326 /* calculate sit block address */
1327 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1328 blk_addr += sit_i->sit_blocks;
1330 return get_meta_page(sbi, blk_addr);
1333 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1334 unsigned int start)
1336 struct sit_info *sit_i = SIT_I(sbi);
1337 struct page *src_page, *dst_page;
1338 pgoff_t src_off, dst_off;
1339 void *src_addr, *dst_addr;
1341 src_off = current_sit_addr(sbi, start);
1342 dst_off = next_sit_addr(sbi, src_off);
1344 /* get current sit block page without lock */
1345 src_page = get_meta_page(sbi, src_off);
1346 dst_page = grab_meta_page(sbi, dst_off);
1347 f2fs_bug_on(PageDirty(src_page));
1349 src_addr = page_address(src_page);
1350 dst_addr = page_address(dst_page);
1351 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1353 set_page_dirty(dst_page);
1354 f2fs_put_page(src_page, 1);
1356 set_to_next_sit(sit_i, start);
1358 return dst_page;
1361 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1363 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1364 struct f2fs_summary_block *sum = curseg->sum_blk;
1365 int i;
1368 * If the journal area in the current summary is full of sit entries,
1369 * all the sit entries will be flushed. Otherwise the sit entries
1370 * are not able to replace with newly hot sit entries.
1372 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1373 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1374 unsigned int segno;
1375 segno = le32_to_cpu(segno_in_journal(sum, i));
1376 __mark_sit_entry_dirty(sbi, segno);
1378 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1379 return true;
1381 return false;
1385 * CP calls this function, which flushes SIT entries including sit_journal,
1386 * and moves prefree segs to free segs.
1388 void flush_sit_entries(struct f2fs_sb_info *sbi)
1390 struct sit_info *sit_i = SIT_I(sbi);
1391 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1392 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1393 struct f2fs_summary_block *sum = curseg->sum_blk;
1394 unsigned long nsegs = TOTAL_SEGS(sbi);
1395 struct page *page = NULL;
1396 struct f2fs_sit_block *raw_sit = NULL;
1397 unsigned int start = 0, end = 0;
1398 unsigned int segno = -1;
1399 bool flushed;
1401 mutex_lock(&curseg->curseg_mutex);
1402 mutex_lock(&sit_i->sentry_lock);
1405 * "flushed" indicates whether sit entries in journal are flushed
1406 * to the SIT area or not.
1408 flushed = flush_sits_in_journal(sbi);
1410 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1411 struct seg_entry *se = get_seg_entry(sbi, segno);
1412 int sit_offset, offset;
1414 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1416 /* add discard candidates */
1417 if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1418 add_discard_addrs(sbi, segno, se);
1420 if (flushed)
1421 goto to_sit_page;
1423 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1424 if (offset >= 0) {
1425 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1426 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1427 goto flush_done;
1429 to_sit_page:
1430 if (!page || (start > segno) || (segno > end)) {
1431 if (page) {
1432 f2fs_put_page(page, 1);
1433 page = NULL;
1436 start = START_SEGNO(sit_i, segno);
1437 end = start + SIT_ENTRY_PER_BLOCK - 1;
1439 /* read sit block that will be updated */
1440 page = get_next_sit_page(sbi, start);
1441 raw_sit = page_address(page);
1444 /* udpate entry in SIT block */
1445 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1446 flush_done:
1447 __clear_bit(segno, bitmap);
1448 sit_i->dirty_sentries--;
1450 mutex_unlock(&sit_i->sentry_lock);
1451 mutex_unlock(&curseg->curseg_mutex);
1453 /* writeout last modified SIT block */
1454 f2fs_put_page(page, 1);
1456 set_prefree_as_free_segments(sbi);
1459 static int build_sit_info(struct f2fs_sb_info *sbi)
1461 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1462 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1463 struct sit_info *sit_i;
1464 unsigned int sit_segs, start;
1465 char *src_bitmap, *dst_bitmap;
1466 unsigned int bitmap_size;
1468 /* allocate memory for SIT information */
1469 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1470 if (!sit_i)
1471 return -ENOMEM;
1473 SM_I(sbi)->sit_info = sit_i;
1475 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1476 if (!sit_i->sentries)
1477 return -ENOMEM;
1479 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1480 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1481 if (!sit_i->dirty_sentries_bitmap)
1482 return -ENOMEM;
1484 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1485 sit_i->sentries[start].cur_valid_map
1486 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1487 sit_i->sentries[start].ckpt_valid_map
1488 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1489 if (!sit_i->sentries[start].cur_valid_map
1490 || !sit_i->sentries[start].ckpt_valid_map)
1491 return -ENOMEM;
1494 if (sbi->segs_per_sec > 1) {
1495 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1496 sizeof(struct sec_entry));
1497 if (!sit_i->sec_entries)
1498 return -ENOMEM;
1501 /* get information related with SIT */
1502 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1504 /* setup SIT bitmap from ckeckpoint pack */
1505 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1506 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1508 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1509 if (!dst_bitmap)
1510 return -ENOMEM;
1512 /* init SIT information */
1513 sit_i->s_ops = &default_salloc_ops;
1515 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1516 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1517 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1518 sit_i->sit_bitmap = dst_bitmap;
1519 sit_i->bitmap_size = bitmap_size;
1520 sit_i->dirty_sentries = 0;
1521 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1522 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1523 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1524 mutex_init(&sit_i->sentry_lock);
1525 return 0;
1528 static int build_free_segmap(struct f2fs_sb_info *sbi)
1530 struct f2fs_sm_info *sm_info = SM_I(sbi);
1531 struct free_segmap_info *free_i;
1532 unsigned int bitmap_size, sec_bitmap_size;
1534 /* allocate memory for free segmap information */
1535 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1536 if (!free_i)
1537 return -ENOMEM;
1539 SM_I(sbi)->free_info = free_i;
1541 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1542 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1543 if (!free_i->free_segmap)
1544 return -ENOMEM;
1546 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1547 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1548 if (!free_i->free_secmap)
1549 return -ENOMEM;
1551 /* set all segments as dirty temporarily */
1552 memset(free_i->free_segmap, 0xff, bitmap_size);
1553 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1555 /* init free segmap information */
1556 free_i->start_segno =
1557 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1558 free_i->free_segments = 0;
1559 free_i->free_sections = 0;
1560 rwlock_init(&free_i->segmap_lock);
1561 return 0;
1564 static int build_curseg(struct f2fs_sb_info *sbi)
1566 struct curseg_info *array;
1567 int i;
1569 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1570 if (!array)
1571 return -ENOMEM;
1573 SM_I(sbi)->curseg_array = array;
1575 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1576 mutex_init(&array[i].curseg_mutex);
1577 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1578 if (!array[i].sum_blk)
1579 return -ENOMEM;
1580 array[i].segno = NULL_SEGNO;
1581 array[i].next_blkoff = 0;
1583 return restore_curseg_summaries(sbi);
1586 static int ra_sit_pages(struct f2fs_sb_info *sbi, int start, int nrpages)
1588 struct address_space *mapping = META_MAPPING(sbi);
1589 struct page *page;
1590 block_t blk_addr, prev_blk_addr = 0;
1591 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1592 int blkno = start;
1593 struct f2fs_io_info fio = {
1594 .type = META,
1595 .rw = READ_SYNC | REQ_META | REQ_PRIO
1598 for (; blkno < start + nrpages && blkno < sit_blk_cnt; blkno++) {
1600 blk_addr = current_sit_addr(sbi, blkno * SIT_ENTRY_PER_BLOCK);
1602 if (blkno != start && prev_blk_addr + 1 != blk_addr)
1603 break;
1604 prev_blk_addr = blk_addr;
1605 repeat:
1606 page = grab_cache_page(mapping, blk_addr);
1607 if (!page) {
1608 cond_resched();
1609 goto repeat;
1611 if (PageUptodate(page)) {
1612 mark_page_accessed(page);
1613 f2fs_put_page(page, 1);
1614 continue;
1617 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
1619 mark_page_accessed(page);
1620 f2fs_put_page(page, 0);
1623 f2fs_submit_merged_bio(sbi, META, READ);
1624 return blkno - start;
1627 static void build_sit_entries(struct f2fs_sb_info *sbi)
1629 struct sit_info *sit_i = SIT_I(sbi);
1630 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1631 struct f2fs_summary_block *sum = curseg->sum_blk;
1632 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1633 unsigned int i, start, end;
1634 unsigned int readed, start_blk = 0;
1635 int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1637 do {
1638 readed = ra_sit_pages(sbi, start_blk, nrpages);
1640 start = start_blk * sit_i->sents_per_block;
1641 end = (start_blk + readed) * sit_i->sents_per_block;
1643 for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1644 struct seg_entry *se = &sit_i->sentries[start];
1645 struct f2fs_sit_block *sit_blk;
1646 struct f2fs_sit_entry sit;
1647 struct page *page;
1649 mutex_lock(&curseg->curseg_mutex);
1650 for (i = 0; i < sits_in_cursum(sum); i++) {
1651 if (le32_to_cpu(segno_in_journal(sum, i))
1652 == start) {
1653 sit = sit_in_journal(sum, i);
1654 mutex_unlock(&curseg->curseg_mutex);
1655 goto got_it;
1658 mutex_unlock(&curseg->curseg_mutex);
1660 page = get_current_sit_page(sbi, start);
1661 sit_blk = (struct f2fs_sit_block *)page_address(page);
1662 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1663 f2fs_put_page(page, 1);
1664 got_it:
1665 check_block_count(sbi, start, &sit);
1666 seg_info_from_raw_sit(se, &sit);
1667 if (sbi->segs_per_sec > 1) {
1668 struct sec_entry *e = get_sec_entry(sbi, start);
1669 e->valid_blocks += se->valid_blocks;
1672 start_blk += readed;
1673 } while (start_blk < sit_blk_cnt);
1676 static void init_free_segmap(struct f2fs_sb_info *sbi)
1678 unsigned int start;
1679 int type;
1681 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1682 struct seg_entry *sentry = get_seg_entry(sbi, start);
1683 if (!sentry->valid_blocks)
1684 __set_free(sbi, start);
1687 /* set use the current segments */
1688 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1689 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1690 __set_test_and_inuse(sbi, curseg_t->segno);
1694 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1696 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1697 struct free_segmap_info *free_i = FREE_I(sbi);
1698 unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1699 unsigned short valid_blocks;
1701 while (1) {
1702 /* find dirty segment based on free segmap */
1703 segno = find_next_inuse(free_i, total_segs, offset);
1704 if (segno >= total_segs)
1705 break;
1706 offset = segno + 1;
1707 valid_blocks = get_valid_blocks(sbi, segno, 0);
1708 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1709 continue;
1710 mutex_lock(&dirty_i->seglist_lock);
1711 __locate_dirty_segment(sbi, segno, DIRTY);
1712 mutex_unlock(&dirty_i->seglist_lock);
1716 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1718 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1719 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1721 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1722 if (!dirty_i->victim_secmap)
1723 return -ENOMEM;
1724 return 0;
1727 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1729 struct dirty_seglist_info *dirty_i;
1730 unsigned int bitmap_size, i;
1732 /* allocate memory for dirty segments list information */
1733 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1734 if (!dirty_i)
1735 return -ENOMEM;
1737 SM_I(sbi)->dirty_info = dirty_i;
1738 mutex_init(&dirty_i->seglist_lock);
1740 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1742 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1743 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1744 if (!dirty_i->dirty_segmap[i])
1745 return -ENOMEM;
1748 init_dirty_segmap(sbi);
1749 return init_victim_secmap(sbi);
1753 * Update min, max modified time for cost-benefit GC algorithm
1755 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1757 struct sit_info *sit_i = SIT_I(sbi);
1758 unsigned int segno;
1760 mutex_lock(&sit_i->sentry_lock);
1762 sit_i->min_mtime = LLONG_MAX;
1764 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1765 unsigned int i;
1766 unsigned long long mtime = 0;
1768 for (i = 0; i < sbi->segs_per_sec; i++)
1769 mtime += get_seg_entry(sbi, segno + i)->mtime;
1771 mtime = div_u64(mtime, sbi->segs_per_sec);
1773 if (sit_i->min_mtime > mtime)
1774 sit_i->min_mtime = mtime;
1776 sit_i->max_mtime = get_mtime(sbi);
1777 mutex_unlock(&sit_i->sentry_lock);
1780 int build_segment_manager(struct f2fs_sb_info *sbi)
1782 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1783 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1784 struct f2fs_sm_info *sm_info;
1785 int err;
1787 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1788 if (!sm_info)
1789 return -ENOMEM;
1791 /* init sm info */
1792 sbi->sm_info = sm_info;
1793 INIT_LIST_HEAD(&sm_info->wblist_head);
1794 spin_lock_init(&sm_info->wblist_lock);
1795 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1796 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1797 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1798 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1799 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1800 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1801 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1802 sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
1803 sm_info->ipu_policy = F2FS_IPU_DISABLE;
1804 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
1806 INIT_LIST_HEAD(&sm_info->discard_list);
1807 sm_info->nr_discards = 0;
1808 sm_info->max_discards = 0;
1810 err = build_sit_info(sbi);
1811 if (err)
1812 return err;
1813 err = build_free_segmap(sbi);
1814 if (err)
1815 return err;
1816 err = build_curseg(sbi);
1817 if (err)
1818 return err;
1820 /* reinit free segmap based on SIT */
1821 build_sit_entries(sbi);
1823 init_free_segmap(sbi);
1824 err = build_dirty_segmap(sbi);
1825 if (err)
1826 return err;
1828 init_min_max_mtime(sbi);
1829 return 0;
1832 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1833 enum dirty_type dirty_type)
1835 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1837 mutex_lock(&dirty_i->seglist_lock);
1838 kfree(dirty_i->dirty_segmap[dirty_type]);
1839 dirty_i->nr_dirty[dirty_type] = 0;
1840 mutex_unlock(&dirty_i->seglist_lock);
1843 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1845 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1846 kfree(dirty_i->victim_secmap);
1849 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1851 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1852 int i;
1854 if (!dirty_i)
1855 return;
1857 /* discard pre-free/dirty segments list */
1858 for (i = 0; i < NR_DIRTY_TYPE; i++)
1859 discard_dirty_segmap(sbi, i);
1861 destroy_victim_secmap(sbi);
1862 SM_I(sbi)->dirty_info = NULL;
1863 kfree(dirty_i);
1866 static void destroy_curseg(struct f2fs_sb_info *sbi)
1868 struct curseg_info *array = SM_I(sbi)->curseg_array;
1869 int i;
1871 if (!array)
1872 return;
1873 SM_I(sbi)->curseg_array = NULL;
1874 for (i = 0; i < NR_CURSEG_TYPE; i++)
1875 kfree(array[i].sum_blk);
1876 kfree(array);
1879 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1881 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1882 if (!free_i)
1883 return;
1884 SM_I(sbi)->free_info = NULL;
1885 kfree(free_i->free_segmap);
1886 kfree(free_i->free_secmap);
1887 kfree(free_i);
1890 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1892 struct sit_info *sit_i = SIT_I(sbi);
1893 unsigned int start;
1895 if (!sit_i)
1896 return;
1898 if (sit_i->sentries) {
1899 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1900 kfree(sit_i->sentries[start].cur_valid_map);
1901 kfree(sit_i->sentries[start].ckpt_valid_map);
1904 vfree(sit_i->sentries);
1905 vfree(sit_i->sec_entries);
1906 kfree(sit_i->dirty_sentries_bitmap);
1908 SM_I(sbi)->sit_info = NULL;
1909 kfree(sit_i->sit_bitmap);
1910 kfree(sit_i);
1913 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1915 struct f2fs_sm_info *sm_info = SM_I(sbi);
1916 if (!sm_info)
1917 return;
1918 destroy_dirty_segmap(sbi);
1919 destroy_curseg(sbi);
1920 destroy_free_segmap(sbi);
1921 destroy_sit_info(sbi);
1922 sbi->sm_info = NULL;
1923 kfree(sm_info);
1926 int __init create_segment_manager_caches(void)
1928 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
1929 sizeof(struct discard_entry), NULL);
1930 if (!discard_entry_slab)
1931 return -ENOMEM;
1932 return 0;
1935 void destroy_segment_manager_caches(void)
1937 kmem_cache_destroy(discard_entry_slab);