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.
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>
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)
36 #if BITS_PER_LONG == 64
37 if ((word & 0xffffffff) == 0) {
42 if ((word & 0xffff) == 0) {
46 if ((word & 0xff) == 0) {
50 if ((word & 0xf0) == 0)
54 if ((word & 0xc) == 0)
58 if ((word & 0x2) == 0)
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.
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);
77 unsigned long mask, submask;
78 unsigned long quot, rest;
84 offset %= BITS_PER_LONG;
89 quot = (offset >> 3) << 3;
92 submask = (unsigned char)(0xff << rest) >> rest;
96 if (size < BITS_PER_LONG)
101 size -= BITS_PER_LONG;
102 result += BITS_PER_LONG;
104 while (size & ~(BITS_PER_LONG-1)) {
108 result += BITS_PER_LONG;
109 size -= BITS_PER_LONG;
115 tmp &= (~0UL >> (BITS_PER_LONG - size));
116 if (tmp == 0UL) /* Are any bits set? */
117 return result + size; /* Nope. */
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);
128 unsigned long mask, submask;
129 unsigned long quot, rest;
135 offset %= BITS_PER_LONG;
140 quot = (offset >> 3) << 3;
142 mask = ~(~0UL << quot);
143 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
147 if (size < BITS_PER_LONG)
152 size -= BITS_PER_LONG;
153 result += BITS_PER_LONG;
155 while (size & ~(BITS_PER_LONG - 1)) {
159 result += BITS_PER_LONG;
160 size -= BITS_PER_LONG;
168 if (tmp == ~0UL) /* Are any bits zero? */
169 return result + size; /* Nope. */
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);
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))
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))
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);
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;
292 if (!test_opt(sbi, DISCARD))
295 /* zero block will be discarded through the prefree list */
296 if (!se->valid_blocks || se->valid_blocks == max_blocks)
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)
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);
331 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
333 if (segno >= total_segs)
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);
354 start = find_next_bit(prefree_map, total_segs, end + 1);
355 if (start >= total_segs)
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))
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);
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 */
419 if (f2fs_set_bit(offset, se->cur_valid_map))
422 if (!f2fs_clear_bit(offset, se->cur_valid_map))
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)
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;
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;
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)
496 else if ((valid_sum_count - sum_in_page) <=
497 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
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);
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;
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))
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,
563 f2fs_bug_on(secno >= TOTAL_SECS(sbi));
566 left_start = hint - 1;
572 while (test_bit(left_start, free_i->free_secmap)) {
573 if (left_start > 0) {
577 left_start = find_next_zero_bit(free_i->free_secmap,
579 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
585 segno = secno * sbi->segs_per_sec;
586 zoneno = secno / sbi->secs_per_zone;
588 /* give up on finding another zone */
591 if (sbi->secs_per_zone == 1)
593 if (zoneno == old_zoneno)
595 if (dir == ALLOC_LEFT) {
596 if (!go_left && zoneno + 1 >= total_zones)
598 if (go_left && zoneno == 0)
601 for (i = 0; i < NR_CURSEG_TYPE; i++)
602 if (CURSEG_I(sbi, i)->zone == zoneno)
605 if (i < NR_CURSEG_TYPE) {
606 /* zone is in user, try another */
608 hint = zoneno * sbi->secs_per_zone - 1;
609 else if (zoneno + 1 >= total_zones)
612 hint = (zoneno + 1) * sbi->secs_per_zone;
614 goto find_other_zone;
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);
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)
658 if (test_opt(sbi, NOHEAP))
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;
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);
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);
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,
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);
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);
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;
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 void f2fs_end_io_write(struct bio *bio, int err)
792 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
793 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
794 struct bio_private *p = bio->bi_private;
797 struct page *page = bvec->bv_page;
799 if (--bvec >= bio->bi_io_vec)
800 prefetchw(&bvec->bv_page->flags);
804 set_bit(AS_EIO, &page->mapping->flags);
805 set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
806 p->sbi->sb->s_flags |= MS_RDONLY;
808 end_page_writeback(page);
809 dec_page_count(p->sbi, F2FS_WRITEBACK);
810 } while (bvec >= bio->bi_io_vec);
815 if (!get_pages(p->sbi, F2FS_WRITEBACK) &&
816 !list_empty(&p->sbi->cp_wait.task_list))
817 wake_up(&p->sbi->cp_wait);
823 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
827 /* No failure on bio allocation */
828 bio = bio_alloc(GFP_NOIO, npages);
830 bio->bi_private = NULL;
835 static void do_submit_bio(struct f2fs_sb_info *sbi,
836 enum page_type type, bool sync)
838 int rw = sync ? WRITE_SYNC : WRITE;
839 enum page_type btype = PAGE_TYPE_OF_BIO(type);
840 struct f2fs_bio_info *io = &sbi->write_io[btype];
841 struct bio_private *p;
846 if (type >= META_FLUSH)
847 rw = WRITE_FLUSH_FUA;
852 trace_f2fs_submit_write_bio(sbi->sb, rw, btype, io->bio);
854 p = io->bio->bi_private;
856 io->bio->bi_end_io = f2fs_end_io_write;
858 if (type == META_FLUSH) {
859 DECLARE_COMPLETION_ONSTACK(wait);
862 submit_bio(rw, io->bio);
863 wait_for_completion(&wait);
866 submit_bio(rw, io->bio);
871 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
873 struct f2fs_bio_info *io = &sbi->write_io[PAGE_TYPE_OF_BIO(type)];
878 mutex_lock(&io->io_mutex);
879 do_submit_bio(sbi, type, sync);
880 mutex_unlock(&io->io_mutex);
883 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
884 block_t blk_addr, enum page_type type)
886 struct block_device *bdev = sbi->sb->s_bdev;
887 struct f2fs_bio_info *io = &sbi->write_io[type];
890 verify_block_addr(sbi, blk_addr);
892 mutex_lock(&io->io_mutex);
894 inc_page_count(sbi, F2FS_WRITEBACK);
896 if (io->bio && io->last_block_in_bio != blk_addr - 1)
897 do_submit_bio(sbi, type, false);
899 if (io->bio == NULL) {
900 struct bio_private *priv;
902 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
908 bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
909 io->bio = f2fs_bio_alloc(bdev, bio_blocks);
910 io->bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
911 io->bio->bi_private = priv;
913 * The end_io will be assigned at the sumbission phase.
914 * Until then, let bio_add_page() merge consecutive IOs as much
919 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
921 do_submit_bio(sbi, type, false);
925 io->last_block_in_bio = blk_addr;
927 mutex_unlock(&io->io_mutex);
928 trace_f2fs_submit_write_page(page, WRITE, type, blk_addr);
931 void f2fs_wait_on_page_writeback(struct page *page,
932 enum page_type type, bool sync)
934 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
935 if (PageWriteback(page)) {
936 f2fs_submit_bio(sbi, type, sync);
937 wait_on_page_writeback(page);
941 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
943 struct curseg_info *curseg = CURSEG_I(sbi, type);
944 if (curseg->next_blkoff < sbi->blocks_per_seg)
949 static int __get_segment_type_2(struct page *page, enum page_type p_type)
952 return CURSEG_HOT_DATA;
954 return CURSEG_HOT_NODE;
957 static int __get_segment_type_4(struct page *page, enum page_type p_type)
959 if (p_type == DATA) {
960 struct inode *inode = page->mapping->host;
962 if (S_ISDIR(inode->i_mode))
963 return CURSEG_HOT_DATA;
965 return CURSEG_COLD_DATA;
967 if (IS_DNODE(page) && !is_cold_node(page))
968 return CURSEG_HOT_NODE;
970 return CURSEG_COLD_NODE;
974 static int __get_segment_type_6(struct page *page, enum page_type p_type)
976 if (p_type == DATA) {
977 struct inode *inode = page->mapping->host;
979 if (S_ISDIR(inode->i_mode))
980 return CURSEG_HOT_DATA;
981 else if (is_cold_data(page) || file_is_cold(inode))
982 return CURSEG_COLD_DATA;
984 return CURSEG_WARM_DATA;
987 return is_cold_node(page) ? CURSEG_WARM_NODE :
990 return CURSEG_COLD_NODE;
994 static int __get_segment_type(struct page *page, enum page_type p_type)
996 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
997 switch (sbi->active_logs) {
999 return __get_segment_type_2(page, p_type);
1001 return __get_segment_type_4(page, p_type);
1003 /* NR_CURSEG_TYPE(6) logs by default */
1004 f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
1005 return __get_segment_type_6(page, p_type);
1008 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1009 block_t old_blkaddr, block_t *new_blkaddr,
1010 struct f2fs_summary *sum, enum page_type p_type)
1012 struct sit_info *sit_i = SIT_I(sbi);
1013 struct curseg_info *curseg;
1014 unsigned int old_cursegno;
1017 type = __get_segment_type(page, p_type);
1018 curseg = CURSEG_I(sbi, type);
1020 mutex_lock(&curseg->curseg_mutex);
1022 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1023 old_cursegno = curseg->segno;
1026 * __add_sum_entry should be resided under the curseg_mutex
1027 * because, this function updates a summary entry in the
1028 * current summary block.
1030 __add_sum_entry(sbi, type, sum);
1032 mutex_lock(&sit_i->sentry_lock);
1033 __refresh_next_blkoff(sbi, curseg);
1035 stat_inc_block_count(sbi, curseg);
1038 * SIT information should be updated before segment allocation,
1039 * since SSR needs latest valid block information.
1041 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1043 if (!__has_curseg_space(sbi, type))
1044 sit_i->s_ops->allocate_segment(sbi, type, false);
1046 locate_dirty_segment(sbi, old_cursegno);
1047 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1048 mutex_unlock(&sit_i->sentry_lock);
1051 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1053 /* writeout dirty page into bdev */
1054 submit_write_page(sbi, page, *new_blkaddr, p_type);
1056 mutex_unlock(&curseg->curseg_mutex);
1059 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1061 set_page_writeback(page);
1062 submit_write_page(sbi, page, page->index, META);
1065 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1066 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1068 struct f2fs_summary sum;
1069 set_summary(&sum, nid, 0, 0);
1070 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
1073 void write_data_page(struct inode *inode, struct page *page,
1074 struct dnode_of_data *dn, block_t old_blkaddr,
1075 block_t *new_blkaddr)
1077 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1078 struct f2fs_summary sum;
1079 struct node_info ni;
1081 f2fs_bug_on(old_blkaddr == NULL_ADDR);
1082 get_node_info(sbi, dn->nid, &ni);
1083 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1085 do_write_page(sbi, page, old_blkaddr,
1086 new_blkaddr, &sum, DATA);
1089 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
1090 block_t old_blk_addr)
1092 submit_write_page(sbi, page, old_blk_addr, DATA);
1095 void recover_data_page(struct f2fs_sb_info *sbi,
1096 struct page *page, struct f2fs_summary *sum,
1097 block_t old_blkaddr, block_t new_blkaddr)
1099 struct sit_info *sit_i = SIT_I(sbi);
1100 struct curseg_info *curseg;
1101 unsigned int segno, old_cursegno;
1102 struct seg_entry *se;
1105 segno = GET_SEGNO(sbi, new_blkaddr);
1106 se = get_seg_entry(sbi, segno);
1109 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1110 if (old_blkaddr == NULL_ADDR)
1111 type = CURSEG_COLD_DATA;
1113 type = CURSEG_WARM_DATA;
1115 curseg = CURSEG_I(sbi, type);
1117 mutex_lock(&curseg->curseg_mutex);
1118 mutex_lock(&sit_i->sentry_lock);
1120 old_cursegno = curseg->segno;
1122 /* change the current segment */
1123 if (segno != curseg->segno) {
1124 curseg->next_segno = segno;
1125 change_curseg(sbi, type, true);
1128 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
1129 (sbi->blocks_per_seg - 1);
1130 __add_sum_entry(sbi, type, sum);
1132 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1134 locate_dirty_segment(sbi, old_cursegno);
1135 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1137 mutex_unlock(&sit_i->sentry_lock);
1138 mutex_unlock(&curseg->curseg_mutex);
1141 void rewrite_node_page(struct f2fs_sb_info *sbi,
1142 struct page *page, struct f2fs_summary *sum,
1143 block_t old_blkaddr, block_t new_blkaddr)
1145 struct sit_info *sit_i = SIT_I(sbi);
1146 int type = CURSEG_WARM_NODE;
1147 struct curseg_info *curseg;
1148 unsigned int segno, old_cursegno;
1149 block_t next_blkaddr = next_blkaddr_of_node(page);
1150 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
1152 curseg = CURSEG_I(sbi, type);
1154 mutex_lock(&curseg->curseg_mutex);
1155 mutex_lock(&sit_i->sentry_lock);
1157 segno = GET_SEGNO(sbi, new_blkaddr);
1158 old_cursegno = curseg->segno;
1160 /* change the current segment */
1161 if (segno != curseg->segno) {
1162 curseg->next_segno = segno;
1163 change_curseg(sbi, type, true);
1165 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
1166 (sbi->blocks_per_seg - 1);
1167 __add_sum_entry(sbi, type, sum);
1169 /* change the current log to the next block addr in advance */
1170 if (next_segno != segno) {
1171 curseg->next_segno = next_segno;
1172 change_curseg(sbi, type, true);
1174 curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
1175 (sbi->blocks_per_seg - 1);
1177 /* rewrite node page */
1178 set_page_writeback(page);
1179 submit_write_page(sbi, page, new_blkaddr, NODE);
1180 f2fs_submit_bio(sbi, NODE, true);
1181 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1183 locate_dirty_segment(sbi, old_cursegno);
1184 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1186 mutex_unlock(&sit_i->sentry_lock);
1187 mutex_unlock(&curseg->curseg_mutex);
1190 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1192 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1193 struct curseg_info *seg_i;
1194 unsigned char *kaddr;
1199 start = start_sum_block(sbi);
1201 page = get_meta_page(sbi, start++);
1202 kaddr = (unsigned char *)page_address(page);
1204 /* Step 1: restore nat cache */
1205 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1206 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1208 /* Step 2: restore sit cache */
1209 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1210 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1212 offset = 2 * SUM_JOURNAL_SIZE;
1214 /* Step 3: restore summary entries */
1215 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1216 unsigned short blk_off;
1219 seg_i = CURSEG_I(sbi, i);
1220 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1221 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1222 seg_i->next_segno = segno;
1223 reset_curseg(sbi, i, 0);
1224 seg_i->alloc_type = ckpt->alloc_type[i];
1225 seg_i->next_blkoff = blk_off;
1227 if (seg_i->alloc_type == SSR)
1228 blk_off = sbi->blocks_per_seg;
1230 for (j = 0; j < blk_off; j++) {
1231 struct f2fs_summary *s;
1232 s = (struct f2fs_summary *)(kaddr + offset);
1233 seg_i->sum_blk->entries[j] = *s;
1234 offset += SUMMARY_SIZE;
1235 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1239 f2fs_put_page(page, 1);
1242 page = get_meta_page(sbi, start++);
1243 kaddr = (unsigned char *)page_address(page);
1247 f2fs_put_page(page, 1);
1251 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1253 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1254 struct f2fs_summary_block *sum;
1255 struct curseg_info *curseg;
1257 unsigned short blk_off;
1258 unsigned int segno = 0;
1259 block_t blk_addr = 0;
1261 /* get segment number and block addr */
1262 if (IS_DATASEG(type)) {
1263 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1264 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1266 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1267 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1269 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1271 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1273 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1275 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1276 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1277 type - CURSEG_HOT_NODE);
1279 blk_addr = GET_SUM_BLOCK(sbi, segno);
1282 new = get_meta_page(sbi, blk_addr);
1283 sum = (struct f2fs_summary_block *)page_address(new);
1285 if (IS_NODESEG(type)) {
1286 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1287 struct f2fs_summary *ns = &sum->entries[0];
1289 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1291 ns->ofs_in_node = 0;
1294 if (restore_node_summary(sbi, segno, sum)) {
1295 f2fs_put_page(new, 1);
1301 /* set uncompleted segment to curseg */
1302 curseg = CURSEG_I(sbi, type);
1303 mutex_lock(&curseg->curseg_mutex);
1304 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1305 curseg->next_segno = segno;
1306 reset_curseg(sbi, type, 0);
1307 curseg->alloc_type = ckpt->alloc_type[type];
1308 curseg->next_blkoff = blk_off;
1309 mutex_unlock(&curseg->curseg_mutex);
1310 f2fs_put_page(new, 1);
1314 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1316 int type = CURSEG_HOT_DATA;
1318 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1319 /* restore for compacted data summary */
1320 if (read_compacted_summaries(sbi))
1322 type = CURSEG_HOT_NODE;
1325 for (; type <= CURSEG_COLD_NODE; type++)
1326 if (read_normal_summaries(sbi, type))
1331 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1334 unsigned char *kaddr;
1335 struct f2fs_summary *summary;
1336 struct curseg_info *seg_i;
1337 int written_size = 0;
1340 page = grab_meta_page(sbi, blkaddr++);
1341 kaddr = (unsigned char *)page_address(page);
1343 /* Step 1: write nat cache */
1344 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1345 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1346 written_size += SUM_JOURNAL_SIZE;
1348 /* Step 2: write sit cache */
1349 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1350 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1352 written_size += SUM_JOURNAL_SIZE;
1354 /* Step 3: write summary entries */
1355 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1356 unsigned short blkoff;
1357 seg_i = CURSEG_I(sbi, i);
1358 if (sbi->ckpt->alloc_type[i] == SSR)
1359 blkoff = sbi->blocks_per_seg;
1361 blkoff = curseg_blkoff(sbi, i);
1363 for (j = 0; j < blkoff; j++) {
1365 page = grab_meta_page(sbi, blkaddr++);
1366 kaddr = (unsigned char *)page_address(page);
1369 summary = (struct f2fs_summary *)(kaddr + written_size);
1370 *summary = seg_i->sum_blk->entries[j];
1371 written_size += SUMMARY_SIZE;
1373 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1377 set_page_dirty(page);
1378 f2fs_put_page(page, 1);
1383 set_page_dirty(page);
1384 f2fs_put_page(page, 1);
1388 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1389 block_t blkaddr, int type)
1392 if (IS_DATASEG(type))
1393 end = type + NR_CURSEG_DATA_TYPE;
1395 end = type + NR_CURSEG_NODE_TYPE;
1397 for (i = type; i < end; i++) {
1398 struct curseg_info *sum = CURSEG_I(sbi, i);
1399 mutex_lock(&sum->curseg_mutex);
1400 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1401 mutex_unlock(&sum->curseg_mutex);
1405 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1407 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1408 write_compacted_summaries(sbi, start_blk);
1410 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1413 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1415 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1416 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1419 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1420 unsigned int val, int alloc)
1424 if (type == NAT_JOURNAL) {
1425 for (i = 0; i < nats_in_cursum(sum); i++) {
1426 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1429 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1430 return update_nats_in_cursum(sum, 1);
1431 } else if (type == SIT_JOURNAL) {
1432 for (i = 0; i < sits_in_cursum(sum); i++)
1433 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1435 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1436 return update_sits_in_cursum(sum, 1);
1441 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1444 struct sit_info *sit_i = SIT_I(sbi);
1445 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1446 block_t blk_addr = sit_i->sit_base_addr + offset;
1448 check_seg_range(sbi, segno);
1450 /* calculate sit block address */
1451 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1452 blk_addr += sit_i->sit_blocks;
1454 return get_meta_page(sbi, blk_addr);
1457 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1460 struct sit_info *sit_i = SIT_I(sbi);
1461 struct page *src_page, *dst_page;
1462 pgoff_t src_off, dst_off;
1463 void *src_addr, *dst_addr;
1465 src_off = current_sit_addr(sbi, start);
1466 dst_off = next_sit_addr(sbi, src_off);
1468 /* get current sit block page without lock */
1469 src_page = get_meta_page(sbi, src_off);
1470 dst_page = grab_meta_page(sbi, dst_off);
1471 f2fs_bug_on(PageDirty(src_page));
1473 src_addr = page_address(src_page);
1474 dst_addr = page_address(dst_page);
1475 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1477 set_page_dirty(dst_page);
1478 f2fs_put_page(src_page, 1);
1480 set_to_next_sit(sit_i, start);
1485 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1487 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1488 struct f2fs_summary_block *sum = curseg->sum_blk;
1492 * If the journal area in the current summary is full of sit entries,
1493 * all the sit entries will be flushed. Otherwise the sit entries
1494 * are not able to replace with newly hot sit entries.
1496 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1497 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1499 segno = le32_to_cpu(segno_in_journal(sum, i));
1500 __mark_sit_entry_dirty(sbi, segno);
1502 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1509 * CP calls this function, which flushes SIT entries including sit_journal,
1510 * and moves prefree segs to free segs.
1512 void flush_sit_entries(struct f2fs_sb_info *sbi)
1514 struct sit_info *sit_i = SIT_I(sbi);
1515 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1516 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1517 struct f2fs_summary_block *sum = curseg->sum_blk;
1518 unsigned long nsegs = TOTAL_SEGS(sbi);
1519 struct page *page = NULL;
1520 struct f2fs_sit_block *raw_sit = NULL;
1521 unsigned int start = 0, end = 0;
1522 unsigned int segno = -1;
1525 mutex_lock(&curseg->curseg_mutex);
1526 mutex_lock(&sit_i->sentry_lock);
1529 * "flushed" indicates whether sit entries in journal are flushed
1530 * to the SIT area or not.
1532 flushed = flush_sits_in_journal(sbi);
1534 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1535 struct seg_entry *se = get_seg_entry(sbi, segno);
1536 int sit_offset, offset;
1538 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1540 /* add discard candidates */
1541 if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1542 add_discard_addrs(sbi, segno, se);
1547 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1549 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1550 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1554 if (!page || (start > segno) || (segno > end)) {
1556 f2fs_put_page(page, 1);
1560 start = START_SEGNO(sit_i, segno);
1561 end = start + SIT_ENTRY_PER_BLOCK - 1;
1563 /* read sit block that will be updated */
1564 page = get_next_sit_page(sbi, start);
1565 raw_sit = page_address(page);
1568 /* udpate entry in SIT block */
1569 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1571 __clear_bit(segno, bitmap);
1572 sit_i->dirty_sentries--;
1574 mutex_unlock(&sit_i->sentry_lock);
1575 mutex_unlock(&curseg->curseg_mutex);
1577 /* writeout last modified SIT block */
1578 f2fs_put_page(page, 1);
1580 set_prefree_as_free_segments(sbi);
1583 static int build_sit_info(struct f2fs_sb_info *sbi)
1585 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1586 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1587 struct sit_info *sit_i;
1588 unsigned int sit_segs, start;
1589 char *src_bitmap, *dst_bitmap;
1590 unsigned int bitmap_size;
1592 /* allocate memory for SIT information */
1593 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1597 SM_I(sbi)->sit_info = sit_i;
1599 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1600 if (!sit_i->sentries)
1603 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1604 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1605 if (!sit_i->dirty_sentries_bitmap)
1608 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1609 sit_i->sentries[start].cur_valid_map
1610 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1611 sit_i->sentries[start].ckpt_valid_map
1612 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1613 if (!sit_i->sentries[start].cur_valid_map
1614 || !sit_i->sentries[start].ckpt_valid_map)
1618 if (sbi->segs_per_sec > 1) {
1619 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1620 sizeof(struct sec_entry));
1621 if (!sit_i->sec_entries)
1625 /* get information related with SIT */
1626 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1628 /* setup SIT bitmap from ckeckpoint pack */
1629 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1630 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1632 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1636 /* init SIT information */
1637 sit_i->s_ops = &default_salloc_ops;
1639 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1640 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1641 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1642 sit_i->sit_bitmap = dst_bitmap;
1643 sit_i->bitmap_size = bitmap_size;
1644 sit_i->dirty_sentries = 0;
1645 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1646 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1647 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1648 mutex_init(&sit_i->sentry_lock);
1652 static int build_free_segmap(struct f2fs_sb_info *sbi)
1654 struct f2fs_sm_info *sm_info = SM_I(sbi);
1655 struct free_segmap_info *free_i;
1656 unsigned int bitmap_size, sec_bitmap_size;
1658 /* allocate memory for free segmap information */
1659 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1663 SM_I(sbi)->free_info = free_i;
1665 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1666 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1667 if (!free_i->free_segmap)
1670 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1671 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1672 if (!free_i->free_secmap)
1675 /* set all segments as dirty temporarily */
1676 memset(free_i->free_segmap, 0xff, bitmap_size);
1677 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1679 /* init free segmap information */
1680 free_i->start_segno =
1681 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1682 free_i->free_segments = 0;
1683 free_i->free_sections = 0;
1684 rwlock_init(&free_i->segmap_lock);
1688 static int build_curseg(struct f2fs_sb_info *sbi)
1690 struct curseg_info *array;
1693 array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1697 SM_I(sbi)->curseg_array = array;
1699 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1700 mutex_init(&array[i].curseg_mutex);
1701 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1702 if (!array[i].sum_blk)
1704 array[i].segno = NULL_SEGNO;
1705 array[i].next_blkoff = 0;
1707 return restore_curseg_summaries(sbi);
1710 static int ra_sit_pages(struct f2fs_sb_info *sbi, int start, int nrpages)
1712 struct address_space *mapping = sbi->meta_inode->i_mapping;
1714 block_t blk_addr, prev_blk_addr = 0;
1715 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1718 for (; blkno < start + nrpages && blkno < sit_blk_cnt; blkno++) {
1720 blk_addr = current_sit_addr(sbi, blkno * SIT_ENTRY_PER_BLOCK);
1722 if (blkno != start && prev_blk_addr + 1 != blk_addr)
1724 prev_blk_addr = blk_addr;
1726 page = grab_cache_page(mapping, blk_addr);
1731 if (PageUptodate(page)) {
1732 mark_page_accessed(page);
1733 f2fs_put_page(page, 1);
1737 submit_read_page(sbi, page, blk_addr, READ_SYNC | REQ_META);
1739 mark_page_accessed(page);
1740 f2fs_put_page(page, 0);
1743 f2fs_submit_read_bio(sbi, READ_SYNC | REQ_META);
1744 return blkno - start;
1747 static void build_sit_entries(struct f2fs_sb_info *sbi)
1749 struct sit_info *sit_i = SIT_I(sbi);
1750 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1751 struct f2fs_summary_block *sum = curseg->sum_blk;
1752 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1753 unsigned int i, start, end;
1754 unsigned int readed, start_blk = 0;
1755 int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1758 readed = ra_sit_pages(sbi, start_blk, nrpages);
1760 start = start_blk * sit_i->sents_per_block;
1761 end = (start_blk + readed) * sit_i->sents_per_block;
1763 for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1764 struct seg_entry *se = &sit_i->sentries[start];
1765 struct f2fs_sit_block *sit_blk;
1766 struct f2fs_sit_entry sit;
1769 mutex_lock(&curseg->curseg_mutex);
1770 for (i = 0; i < sits_in_cursum(sum); i++) {
1771 if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1772 sit = sit_in_journal(sum, i);
1773 mutex_unlock(&curseg->curseg_mutex);
1777 mutex_unlock(&curseg->curseg_mutex);
1779 page = get_current_sit_page(sbi, start);
1780 sit_blk = (struct f2fs_sit_block *)page_address(page);
1781 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1782 f2fs_put_page(page, 1);
1784 check_block_count(sbi, start, &sit);
1785 seg_info_from_raw_sit(se, &sit);
1786 if (sbi->segs_per_sec > 1) {
1787 struct sec_entry *e = get_sec_entry(sbi, start);
1788 e->valid_blocks += se->valid_blocks;
1791 start_blk += readed;
1792 } while (start_blk < sit_blk_cnt);
1795 static void init_free_segmap(struct f2fs_sb_info *sbi)
1800 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1801 struct seg_entry *sentry = get_seg_entry(sbi, start);
1802 if (!sentry->valid_blocks)
1803 __set_free(sbi, start);
1806 /* set use the current segments */
1807 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1808 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1809 __set_test_and_inuse(sbi, curseg_t->segno);
1813 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1815 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1816 struct free_segmap_info *free_i = FREE_I(sbi);
1817 unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1818 unsigned short valid_blocks;
1821 /* find dirty segment based on free segmap */
1822 segno = find_next_inuse(free_i, total_segs, offset);
1823 if (segno >= total_segs)
1826 valid_blocks = get_valid_blocks(sbi, segno, 0);
1827 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1829 mutex_lock(&dirty_i->seglist_lock);
1830 __locate_dirty_segment(sbi, segno, DIRTY);
1831 mutex_unlock(&dirty_i->seglist_lock);
1835 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1837 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1838 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1840 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1841 if (!dirty_i->victim_secmap)
1846 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1848 struct dirty_seglist_info *dirty_i;
1849 unsigned int bitmap_size, i;
1851 /* allocate memory for dirty segments list information */
1852 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1856 SM_I(sbi)->dirty_info = dirty_i;
1857 mutex_init(&dirty_i->seglist_lock);
1859 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1861 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1862 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1863 if (!dirty_i->dirty_segmap[i])
1867 init_dirty_segmap(sbi);
1868 return init_victim_secmap(sbi);
1872 * Update min, max modified time for cost-benefit GC algorithm
1874 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1876 struct sit_info *sit_i = SIT_I(sbi);
1879 mutex_lock(&sit_i->sentry_lock);
1881 sit_i->min_mtime = LLONG_MAX;
1883 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1885 unsigned long long mtime = 0;
1887 for (i = 0; i < sbi->segs_per_sec; i++)
1888 mtime += get_seg_entry(sbi, segno + i)->mtime;
1890 mtime = div_u64(mtime, sbi->segs_per_sec);
1892 if (sit_i->min_mtime > mtime)
1893 sit_i->min_mtime = mtime;
1895 sit_i->max_mtime = get_mtime(sbi);
1896 mutex_unlock(&sit_i->sentry_lock);
1899 int build_segment_manager(struct f2fs_sb_info *sbi)
1901 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1902 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1903 struct f2fs_sm_info *sm_info;
1906 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1911 sbi->sm_info = sm_info;
1912 INIT_LIST_HEAD(&sm_info->wblist_head);
1913 spin_lock_init(&sm_info->wblist_lock);
1914 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1915 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1916 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1917 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1918 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1919 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1920 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1921 sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
1923 INIT_LIST_HEAD(&sm_info->discard_list);
1924 sm_info->nr_discards = 0;
1925 sm_info->max_discards = 0;
1927 err = build_sit_info(sbi);
1930 err = build_free_segmap(sbi);
1933 err = build_curseg(sbi);
1937 /* reinit free segmap based on SIT */
1938 build_sit_entries(sbi);
1940 init_free_segmap(sbi);
1941 err = build_dirty_segmap(sbi);
1945 init_min_max_mtime(sbi);
1949 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1950 enum dirty_type dirty_type)
1952 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1954 mutex_lock(&dirty_i->seglist_lock);
1955 kfree(dirty_i->dirty_segmap[dirty_type]);
1956 dirty_i->nr_dirty[dirty_type] = 0;
1957 mutex_unlock(&dirty_i->seglist_lock);
1960 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1962 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1963 kfree(dirty_i->victim_secmap);
1966 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1968 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1974 /* discard pre-free/dirty segments list */
1975 for (i = 0; i < NR_DIRTY_TYPE; i++)
1976 discard_dirty_segmap(sbi, i);
1978 destroy_victim_secmap(sbi);
1979 SM_I(sbi)->dirty_info = NULL;
1983 static void destroy_curseg(struct f2fs_sb_info *sbi)
1985 struct curseg_info *array = SM_I(sbi)->curseg_array;
1990 SM_I(sbi)->curseg_array = NULL;
1991 for (i = 0; i < NR_CURSEG_TYPE; i++)
1992 kfree(array[i].sum_blk);
1996 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1998 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2001 SM_I(sbi)->free_info = NULL;
2002 kfree(free_i->free_segmap);
2003 kfree(free_i->free_secmap);
2007 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2009 struct sit_info *sit_i = SIT_I(sbi);
2015 if (sit_i->sentries) {
2016 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
2017 kfree(sit_i->sentries[start].cur_valid_map);
2018 kfree(sit_i->sentries[start].ckpt_valid_map);
2021 vfree(sit_i->sentries);
2022 vfree(sit_i->sec_entries);
2023 kfree(sit_i->dirty_sentries_bitmap);
2025 SM_I(sbi)->sit_info = NULL;
2026 kfree(sit_i->sit_bitmap);
2030 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2032 struct f2fs_sm_info *sm_info = SM_I(sbi);
2035 destroy_dirty_segmap(sbi);
2036 destroy_curseg(sbi);
2037 destroy_free_segmap(sbi);
2038 destroy_sit_info(sbi);
2039 sbi->sm_info = NULL;
2043 int __init create_segment_manager_caches(void)
2045 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2046 sizeof(struct discard_entry), NULL);
2047 if (!discard_entry_slab)
2052 void destroy_segment_manager_caches(void)
2054 kmem_cache_destroy(discard_entry_slab);