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/kthread.h>
17 #include <linux/vmalloc.h>
18 #include <linux/swap.h>
23 #include <trace/events/f2fs.h>
25 #define __reverse_ffz(x) __reverse_ffs(~(x))
27 static struct kmem_cache *discard_entry_slab;
30 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
31 * MSB and LSB are reversed in a byte by f2fs_set_bit.
33 static inline unsigned long __reverse_ffs(unsigned long word)
37 #if BITS_PER_LONG == 64
38 if ((word & 0xffffffff) == 0) {
43 if ((word & 0xffff) == 0) {
47 if ((word & 0xff) == 0) {
51 if ((word & 0xf0) == 0)
55 if ((word & 0xc) == 0)
59 if ((word & 0x2) == 0)
65 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c becasue
66 * f2fs_set_bit makes MSB and LSB reversed in a byte.
69 * f2fs_set_bit(0, bitmap) => 0000 0001
70 * f2fs_set_bit(7, bitmap) => 1000 0000
72 static unsigned long __find_rev_next_bit(const unsigned long *addr,
73 unsigned long size, unsigned long offset)
75 const unsigned long *p = addr + BIT_WORD(offset);
76 unsigned long result = offset & ~(BITS_PER_LONG - 1);
78 unsigned long mask, submask;
79 unsigned long quot, rest;
85 offset %= BITS_PER_LONG;
90 quot = (offset >> 3) << 3;
93 submask = (unsigned char)(0xff << rest) >> rest;
97 if (size < BITS_PER_LONG)
102 size -= BITS_PER_LONG;
103 result += BITS_PER_LONG;
105 while (size & ~(BITS_PER_LONG-1)) {
109 result += BITS_PER_LONG;
110 size -= BITS_PER_LONG;
116 tmp &= (~0UL >> (BITS_PER_LONG - size));
117 if (tmp == 0UL) /* Are any bits set? */
118 return result + size; /* Nope. */
120 return result + __reverse_ffs(tmp);
123 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
124 unsigned long size, unsigned long offset)
126 const unsigned long *p = addr + BIT_WORD(offset);
127 unsigned long result = offset & ~(BITS_PER_LONG - 1);
129 unsigned long mask, submask;
130 unsigned long quot, rest;
136 offset %= BITS_PER_LONG;
141 quot = (offset >> 3) << 3;
143 mask = ~(~0UL << quot);
144 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
148 if (size < BITS_PER_LONG)
153 size -= BITS_PER_LONG;
154 result += BITS_PER_LONG;
156 while (size & ~(BITS_PER_LONG - 1)) {
160 result += BITS_PER_LONG;
161 size -= BITS_PER_LONG;
169 if (tmp == ~0UL) /* Are any bits zero? */
170 return result + size; /* Nope. */
172 return result + __reverse_ffz(tmp);
176 * This function balances dirty node and dentry pages.
177 * In addition, it controls garbage collection.
179 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
182 * We should do GC or end up with checkpoint, if there are so many dirty
183 * dir/node pages without enough free segments.
185 if (has_not_enough_free_secs(sbi, 0)) {
186 mutex_lock(&sbi->gc_mutex);
191 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
193 /* check the # of cached NAT entries and prefree segments */
194 if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
195 excess_prefree_segs(sbi))
196 f2fs_sync_fs(sbi->sb, true);
199 static int issue_flush_thread(void *data)
201 struct f2fs_sb_info *sbi = data;
202 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
203 wait_queue_head_t *q = &fcc->flush_wait_queue;
205 if (kthread_should_stop())
208 spin_lock(&fcc->issue_lock);
209 if (fcc->issue_list) {
210 fcc->dispatch_list = fcc->issue_list;
211 fcc->issue_list = fcc->issue_tail = NULL;
213 spin_unlock(&fcc->issue_lock);
215 if (fcc->dispatch_list) {
216 struct bio *bio = bio_alloc(GFP_NOIO, 0);
217 struct flush_cmd *cmd, *next;
220 bio->bi_bdev = sbi->sb->s_bdev;
221 ret = submit_bio_wait(WRITE_FLUSH, bio);
223 for (cmd = fcc->dispatch_list; cmd; cmd = next) {
226 complete(&cmd->wait);
229 fcc->dispatch_list = NULL;
232 wait_event_interruptible(*q,
233 kthread_should_stop() || fcc->issue_list);
237 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
239 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
240 struct flush_cmd cmd;
242 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
243 test_opt(sbi, FLUSH_MERGE));
245 if (test_opt(sbi, NOBARRIER))
248 if (!test_opt(sbi, FLUSH_MERGE))
249 return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
251 init_completion(&cmd.wait);
254 spin_lock(&fcc->issue_lock);
256 fcc->issue_tail->next = &cmd;
258 fcc->issue_list = &cmd;
259 fcc->issue_tail = &cmd;
260 spin_unlock(&fcc->issue_lock);
262 if (!fcc->dispatch_list)
263 wake_up(&fcc->flush_wait_queue);
265 wait_for_completion(&cmd.wait);
270 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
272 dev_t dev = sbi->sb->s_bdev->bd_dev;
273 struct flush_cmd_control *fcc;
276 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
279 spin_lock_init(&fcc->issue_lock);
280 init_waitqueue_head(&fcc->flush_wait_queue);
281 SM_I(sbi)->cmd_control_info = fcc;
282 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
283 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
284 if (IS_ERR(fcc->f2fs_issue_flush)) {
285 err = PTR_ERR(fcc->f2fs_issue_flush);
287 SM_I(sbi)->cmd_control_info = NULL;
294 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
296 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
298 if (fcc && fcc->f2fs_issue_flush)
299 kthread_stop(fcc->f2fs_issue_flush);
301 SM_I(sbi)->cmd_control_info = NULL;
304 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
305 enum dirty_type dirty_type)
307 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
309 /* need not be added */
310 if (IS_CURSEG(sbi, segno))
313 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
314 dirty_i->nr_dirty[dirty_type]++;
316 if (dirty_type == DIRTY) {
317 struct seg_entry *sentry = get_seg_entry(sbi, segno);
318 enum dirty_type t = sentry->type;
320 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
321 dirty_i->nr_dirty[t]++;
325 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
326 enum dirty_type dirty_type)
328 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
330 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
331 dirty_i->nr_dirty[dirty_type]--;
333 if (dirty_type == DIRTY) {
334 struct seg_entry *sentry = get_seg_entry(sbi, segno);
335 enum dirty_type t = sentry->type;
337 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
338 dirty_i->nr_dirty[t]--;
340 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
341 clear_bit(GET_SECNO(sbi, segno),
342 dirty_i->victim_secmap);
347 * Should not occur error such as -ENOMEM.
348 * Adding dirty entry into seglist is not critical operation.
349 * If a given segment is one of current working segments, it won't be added.
351 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
353 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
354 unsigned short valid_blocks;
356 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
359 mutex_lock(&dirty_i->seglist_lock);
361 valid_blocks = get_valid_blocks(sbi, segno, 0);
363 if (valid_blocks == 0) {
364 __locate_dirty_segment(sbi, segno, PRE);
365 __remove_dirty_segment(sbi, segno, DIRTY);
366 } else if (valid_blocks < sbi->blocks_per_seg) {
367 __locate_dirty_segment(sbi, segno, DIRTY);
369 /* Recovery routine with SSR needs this */
370 __remove_dirty_segment(sbi, segno, DIRTY);
373 mutex_unlock(&dirty_i->seglist_lock);
376 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
377 block_t blkstart, block_t blklen)
379 sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
380 sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
381 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
382 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
385 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
387 if (f2fs_issue_discard(sbi, blkaddr, 1)) {
388 struct page *page = grab_meta_page(sbi, blkaddr);
389 /* zero-filled page */
390 set_page_dirty(page);
391 f2fs_put_page(page, 1);
395 static void add_discard_addrs(struct f2fs_sb_info *sbi,
396 unsigned int segno, struct seg_entry *se)
398 struct list_head *head = &SM_I(sbi)->discard_list;
399 struct discard_entry *new;
400 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
401 int max_blocks = sbi->blocks_per_seg;
402 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
403 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
404 unsigned long dmap[entries];
405 unsigned int start = 0, end = -1;
408 if (!test_opt(sbi, DISCARD))
411 /* zero block will be discarded through the prefree list */
412 if (!se->valid_blocks || se->valid_blocks == max_blocks)
415 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
416 for (i = 0; i < entries; i++)
417 dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
419 while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
420 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
421 if (start >= max_blocks)
424 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
426 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
427 INIT_LIST_HEAD(&new->list);
428 new->blkaddr = START_BLOCK(sbi, segno) + start;
429 new->len = end - start;
431 list_add_tail(&new->list, head);
432 SM_I(sbi)->nr_discards += end - start;
437 * Should call clear_prefree_segments after checkpoint is done.
439 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
441 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
442 unsigned int segno = -1;
443 unsigned int total_segs = TOTAL_SEGS(sbi);
445 mutex_lock(&dirty_i->seglist_lock);
447 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
449 if (segno >= total_segs)
451 __set_test_and_free(sbi, segno);
453 mutex_unlock(&dirty_i->seglist_lock);
456 void clear_prefree_segments(struct f2fs_sb_info *sbi)
458 struct list_head *head = &(SM_I(sbi)->discard_list);
459 struct discard_entry *entry, *this;
460 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
461 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
462 unsigned int total_segs = TOTAL_SEGS(sbi);
463 unsigned int start = 0, end = -1;
465 mutex_lock(&dirty_i->seglist_lock);
469 start = find_next_bit(prefree_map, total_segs, end + 1);
470 if (start >= total_segs)
472 end = find_next_zero_bit(prefree_map, total_segs, start + 1);
474 for (i = start; i < end; i++)
475 clear_bit(i, prefree_map);
477 dirty_i->nr_dirty[PRE] -= end - start;
479 if (!test_opt(sbi, DISCARD))
482 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
483 (end - start) << sbi->log_blocks_per_seg);
485 mutex_unlock(&dirty_i->seglist_lock);
487 /* send small discards */
488 list_for_each_entry_safe(entry, this, head, list) {
489 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
490 list_del(&entry->list);
491 SM_I(sbi)->nr_discards -= entry->len;
492 kmem_cache_free(discard_entry_slab, entry);
496 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
498 struct sit_info *sit_i = SIT_I(sbi);
499 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
500 sit_i->dirty_sentries++;
503 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
504 unsigned int segno, int modified)
506 struct seg_entry *se = get_seg_entry(sbi, segno);
509 __mark_sit_entry_dirty(sbi, segno);
512 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
514 struct seg_entry *se;
515 unsigned int segno, offset;
516 long int new_vblocks;
518 segno = GET_SEGNO(sbi, blkaddr);
520 se = get_seg_entry(sbi, segno);
521 new_vblocks = se->valid_blocks + del;
522 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
524 f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
525 (new_vblocks > sbi->blocks_per_seg)));
527 se->valid_blocks = new_vblocks;
528 se->mtime = get_mtime(sbi);
529 SIT_I(sbi)->max_mtime = se->mtime;
531 /* Update valid block bitmap */
533 if (f2fs_set_bit(offset, se->cur_valid_map))
536 if (!f2fs_clear_bit(offset, se->cur_valid_map))
539 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
540 se->ckpt_valid_blocks += del;
542 __mark_sit_entry_dirty(sbi, segno);
544 /* update total number of valid blocks to be written in ckpt area */
545 SIT_I(sbi)->written_valid_blocks += del;
547 if (sbi->segs_per_sec > 1)
548 get_sec_entry(sbi, segno)->valid_blocks += del;
551 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
553 update_sit_entry(sbi, new, 1);
554 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
555 update_sit_entry(sbi, old, -1);
557 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
558 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
561 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
563 unsigned int segno = GET_SEGNO(sbi, addr);
564 struct sit_info *sit_i = SIT_I(sbi);
566 f2fs_bug_on(addr == NULL_ADDR);
567 if (addr == NEW_ADDR)
570 /* add it into sit main buffer */
571 mutex_lock(&sit_i->sentry_lock);
573 update_sit_entry(sbi, addr, -1);
575 /* add it into dirty seglist */
576 locate_dirty_segment(sbi, segno);
578 mutex_unlock(&sit_i->sentry_lock);
582 * This function should be resided under the curseg_mutex lock
584 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
585 struct f2fs_summary *sum)
587 struct curseg_info *curseg = CURSEG_I(sbi, type);
588 void *addr = curseg->sum_blk;
589 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
590 memcpy(addr, sum, sizeof(struct f2fs_summary));
594 * Calculate the number of current summary pages for writing
596 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
598 int valid_sum_count = 0;
601 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
602 if (sbi->ckpt->alloc_type[i] == SSR)
603 valid_sum_count += sbi->blocks_per_seg;
605 valid_sum_count += curseg_blkoff(sbi, i);
608 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
609 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
610 if (valid_sum_count <= sum_in_page)
612 else if ((valid_sum_count - sum_in_page) <=
613 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
619 * Caller should put this summary page
621 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
623 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
626 static void write_sum_page(struct f2fs_sb_info *sbi,
627 struct f2fs_summary_block *sum_blk, block_t blk_addr)
629 struct page *page = grab_meta_page(sbi, blk_addr);
630 void *kaddr = page_address(page);
631 memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
632 set_page_dirty(page);
633 f2fs_put_page(page, 1);
636 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
638 struct curseg_info *curseg = CURSEG_I(sbi, type);
639 unsigned int segno = curseg->segno + 1;
640 struct free_segmap_info *free_i = FREE_I(sbi);
642 if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
643 return !test_bit(segno, free_i->free_segmap);
648 * Find a new segment from the free segments bitmap to right order
649 * This function should be returned with success, otherwise BUG
651 static void get_new_segment(struct f2fs_sb_info *sbi,
652 unsigned int *newseg, bool new_sec, int dir)
654 struct free_segmap_info *free_i = FREE_I(sbi);
655 unsigned int segno, secno, zoneno;
656 unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
657 unsigned int hint = *newseg / sbi->segs_per_sec;
658 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
659 unsigned int left_start = hint;
664 write_lock(&free_i->segmap_lock);
666 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
667 segno = find_next_zero_bit(free_i->free_segmap,
668 TOTAL_SEGS(sbi), *newseg + 1);
669 if (segno - *newseg < sbi->segs_per_sec -
670 (*newseg % sbi->segs_per_sec))
674 secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
675 if (secno >= TOTAL_SECS(sbi)) {
676 if (dir == ALLOC_RIGHT) {
677 secno = find_next_zero_bit(free_i->free_secmap,
679 f2fs_bug_on(secno >= TOTAL_SECS(sbi));
682 left_start = hint - 1;
688 while (test_bit(left_start, free_i->free_secmap)) {
689 if (left_start > 0) {
693 left_start = find_next_zero_bit(free_i->free_secmap,
695 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
701 segno = secno * sbi->segs_per_sec;
702 zoneno = secno / sbi->secs_per_zone;
704 /* give up on finding another zone */
707 if (sbi->secs_per_zone == 1)
709 if (zoneno == old_zoneno)
711 if (dir == ALLOC_LEFT) {
712 if (!go_left && zoneno + 1 >= total_zones)
714 if (go_left && zoneno == 0)
717 for (i = 0; i < NR_CURSEG_TYPE; i++)
718 if (CURSEG_I(sbi, i)->zone == zoneno)
721 if (i < NR_CURSEG_TYPE) {
722 /* zone is in user, try another */
724 hint = zoneno * sbi->secs_per_zone - 1;
725 else if (zoneno + 1 >= total_zones)
728 hint = (zoneno + 1) * sbi->secs_per_zone;
730 goto find_other_zone;
733 /* set it as dirty segment in free segmap */
734 f2fs_bug_on(test_bit(segno, free_i->free_segmap));
735 __set_inuse(sbi, segno);
737 write_unlock(&free_i->segmap_lock);
740 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
742 struct curseg_info *curseg = CURSEG_I(sbi, type);
743 struct summary_footer *sum_footer;
745 curseg->segno = curseg->next_segno;
746 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
747 curseg->next_blkoff = 0;
748 curseg->next_segno = NULL_SEGNO;
750 sum_footer = &(curseg->sum_blk->footer);
751 memset(sum_footer, 0, sizeof(struct summary_footer));
752 if (IS_DATASEG(type))
753 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
754 if (IS_NODESEG(type))
755 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
756 __set_sit_entry_type(sbi, type, curseg->segno, modified);
760 * Allocate a current working segment.
761 * This function always allocates a free segment in LFS manner.
763 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
765 struct curseg_info *curseg = CURSEG_I(sbi, type);
766 unsigned int segno = curseg->segno;
767 int dir = ALLOC_LEFT;
769 write_sum_page(sbi, curseg->sum_blk,
770 GET_SUM_BLOCK(sbi, segno));
771 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
774 if (test_opt(sbi, NOHEAP))
777 get_new_segment(sbi, &segno, new_sec, dir);
778 curseg->next_segno = segno;
779 reset_curseg(sbi, type, 1);
780 curseg->alloc_type = LFS;
783 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
784 struct curseg_info *seg, block_t start)
786 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
787 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
788 unsigned long target_map[entries];
789 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
790 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
793 for (i = 0; i < entries; i++)
794 target_map[i] = ckpt_map[i] | cur_map[i];
796 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
798 seg->next_blkoff = pos;
802 * If a segment is written by LFS manner, next block offset is just obtained
803 * by increasing the current block offset. However, if a segment is written by
804 * SSR manner, next block offset obtained by calling __next_free_blkoff
806 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
807 struct curseg_info *seg)
809 if (seg->alloc_type == SSR)
810 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
816 * This function always allocates a used segment (from dirty seglist) by SSR
817 * manner, so it should recover the existing segment information of valid blocks
819 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
821 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
822 struct curseg_info *curseg = CURSEG_I(sbi, type);
823 unsigned int new_segno = curseg->next_segno;
824 struct f2fs_summary_block *sum_node;
825 struct page *sum_page;
827 write_sum_page(sbi, curseg->sum_blk,
828 GET_SUM_BLOCK(sbi, curseg->segno));
829 __set_test_and_inuse(sbi, new_segno);
831 mutex_lock(&dirty_i->seglist_lock);
832 __remove_dirty_segment(sbi, new_segno, PRE);
833 __remove_dirty_segment(sbi, new_segno, DIRTY);
834 mutex_unlock(&dirty_i->seglist_lock);
836 reset_curseg(sbi, type, 1);
837 curseg->alloc_type = SSR;
838 __next_free_blkoff(sbi, curseg, 0);
841 sum_page = get_sum_page(sbi, new_segno);
842 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
843 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
844 f2fs_put_page(sum_page, 1);
848 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
850 struct curseg_info *curseg = CURSEG_I(sbi, type);
851 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
853 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
854 return v_ops->get_victim(sbi,
855 &(curseg)->next_segno, BG_GC, type, SSR);
857 /* For data segments, let's do SSR more intensively */
858 for (; type >= CURSEG_HOT_DATA; type--)
859 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
866 * flush out current segment and replace it with new segment
867 * This function should be returned with success, otherwise BUG
869 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
870 int type, bool force)
872 struct curseg_info *curseg = CURSEG_I(sbi, type);
875 new_curseg(sbi, type, true);
876 else if (type == CURSEG_WARM_NODE)
877 new_curseg(sbi, type, false);
878 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
879 new_curseg(sbi, type, false);
880 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
881 change_curseg(sbi, type, true);
883 new_curseg(sbi, type, false);
885 stat_inc_seg_type(sbi, curseg);
888 void allocate_new_segments(struct f2fs_sb_info *sbi)
890 struct curseg_info *curseg;
891 unsigned int old_curseg;
894 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
895 curseg = CURSEG_I(sbi, i);
896 old_curseg = curseg->segno;
897 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
898 locate_dirty_segment(sbi, old_curseg);
902 static const struct segment_allocation default_salloc_ops = {
903 .allocate_segment = allocate_segment_by_default,
906 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
908 struct curseg_info *curseg = CURSEG_I(sbi, type);
909 if (curseg->next_blkoff < sbi->blocks_per_seg)
914 static int __get_segment_type_2(struct page *page, enum page_type p_type)
917 return CURSEG_HOT_DATA;
919 return CURSEG_HOT_NODE;
922 static int __get_segment_type_4(struct page *page, enum page_type p_type)
924 if (p_type == DATA) {
925 struct inode *inode = page->mapping->host;
927 if (S_ISDIR(inode->i_mode))
928 return CURSEG_HOT_DATA;
930 return CURSEG_COLD_DATA;
932 if (IS_DNODE(page) && !is_cold_node(page))
933 return CURSEG_HOT_NODE;
935 return CURSEG_COLD_NODE;
939 static int __get_segment_type_6(struct page *page, enum page_type p_type)
941 if (p_type == DATA) {
942 struct inode *inode = page->mapping->host;
944 if (S_ISDIR(inode->i_mode))
945 return CURSEG_HOT_DATA;
946 else if (is_cold_data(page) || file_is_cold(inode))
947 return CURSEG_COLD_DATA;
949 return CURSEG_WARM_DATA;
952 return is_cold_node(page) ? CURSEG_WARM_NODE :
955 return CURSEG_COLD_NODE;
959 static int __get_segment_type(struct page *page, enum page_type p_type)
961 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
962 switch (sbi->active_logs) {
964 return __get_segment_type_2(page, p_type);
966 return __get_segment_type_4(page, p_type);
968 /* NR_CURSEG_TYPE(6) logs by default */
969 f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
970 return __get_segment_type_6(page, p_type);
973 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
974 block_t old_blkaddr, block_t *new_blkaddr,
975 struct f2fs_summary *sum, int type)
977 struct sit_info *sit_i = SIT_I(sbi);
978 struct curseg_info *curseg;
979 unsigned int old_cursegno;
981 curseg = CURSEG_I(sbi, type);
983 mutex_lock(&curseg->curseg_mutex);
985 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
986 old_cursegno = curseg->segno;
989 * __add_sum_entry should be resided under the curseg_mutex
990 * because, this function updates a summary entry in the
991 * current summary block.
993 __add_sum_entry(sbi, type, sum);
995 mutex_lock(&sit_i->sentry_lock);
996 __refresh_next_blkoff(sbi, curseg);
998 stat_inc_block_count(sbi, curseg);
1000 if (!__has_curseg_space(sbi, type))
1001 sit_i->s_ops->allocate_segment(sbi, type, false);
1003 * SIT information should be updated before segment allocation,
1004 * since SSR needs latest valid block information.
1006 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1007 locate_dirty_segment(sbi, old_cursegno);
1009 mutex_unlock(&sit_i->sentry_lock);
1011 if (page && IS_NODESEG(type))
1012 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1014 mutex_unlock(&curseg->curseg_mutex);
1017 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1018 block_t old_blkaddr, block_t *new_blkaddr,
1019 struct f2fs_summary *sum, struct f2fs_io_info *fio)
1021 int type = __get_segment_type(page, fio->type);
1023 allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1025 /* writeout dirty page into bdev */
1026 f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1029 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1031 struct f2fs_io_info fio = {
1033 .rw = WRITE_SYNC | REQ_META | REQ_PRIO
1036 set_page_writeback(page);
1037 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1040 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1041 struct f2fs_io_info *fio,
1042 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1044 struct f2fs_summary sum;
1045 set_summary(&sum, nid, 0, 0);
1046 do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1049 void write_data_page(struct page *page, struct dnode_of_data *dn,
1050 block_t *new_blkaddr, struct f2fs_io_info *fio)
1052 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
1053 struct f2fs_summary sum;
1054 struct node_info ni;
1056 f2fs_bug_on(dn->data_blkaddr == NULL_ADDR);
1057 get_node_info(sbi, dn->nid, &ni);
1058 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1060 do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1063 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1064 struct f2fs_io_info *fio)
1066 struct inode *inode = page->mapping->host;
1067 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1068 f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio);
1071 void recover_data_page(struct f2fs_sb_info *sbi,
1072 struct page *page, struct f2fs_summary *sum,
1073 block_t old_blkaddr, block_t new_blkaddr)
1075 struct sit_info *sit_i = SIT_I(sbi);
1076 struct curseg_info *curseg;
1077 unsigned int segno, old_cursegno;
1078 struct seg_entry *se;
1081 segno = GET_SEGNO(sbi, new_blkaddr);
1082 se = get_seg_entry(sbi, segno);
1085 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1086 if (old_blkaddr == NULL_ADDR)
1087 type = CURSEG_COLD_DATA;
1089 type = CURSEG_WARM_DATA;
1091 curseg = CURSEG_I(sbi, type);
1093 mutex_lock(&curseg->curseg_mutex);
1094 mutex_lock(&sit_i->sentry_lock);
1096 old_cursegno = curseg->segno;
1098 /* change the current segment */
1099 if (segno != curseg->segno) {
1100 curseg->next_segno = segno;
1101 change_curseg(sbi, type, true);
1104 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1105 __add_sum_entry(sbi, type, sum);
1107 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1108 locate_dirty_segment(sbi, old_cursegno);
1110 mutex_unlock(&sit_i->sentry_lock);
1111 mutex_unlock(&curseg->curseg_mutex);
1114 void rewrite_node_page(struct f2fs_sb_info *sbi,
1115 struct page *page, struct f2fs_summary *sum,
1116 block_t old_blkaddr, block_t new_blkaddr)
1118 struct sit_info *sit_i = SIT_I(sbi);
1119 int type = CURSEG_WARM_NODE;
1120 struct curseg_info *curseg;
1121 unsigned int segno, old_cursegno;
1122 block_t next_blkaddr = next_blkaddr_of_node(page);
1123 unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
1124 struct f2fs_io_info fio = {
1129 curseg = CURSEG_I(sbi, type);
1131 mutex_lock(&curseg->curseg_mutex);
1132 mutex_lock(&sit_i->sentry_lock);
1134 segno = GET_SEGNO(sbi, new_blkaddr);
1135 old_cursegno = curseg->segno;
1137 /* change the current segment */
1138 if (segno != curseg->segno) {
1139 curseg->next_segno = segno;
1140 change_curseg(sbi, type, true);
1142 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1143 __add_sum_entry(sbi, type, sum);
1145 /* change the current log to the next block addr in advance */
1146 if (next_segno != segno) {
1147 curseg->next_segno = next_segno;
1148 change_curseg(sbi, type, true);
1150 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, next_blkaddr);
1152 /* rewrite node page */
1153 set_page_writeback(page);
1154 f2fs_submit_page_mbio(sbi, page, new_blkaddr, &fio);
1155 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1156 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1157 locate_dirty_segment(sbi, old_cursegno);
1159 mutex_unlock(&sit_i->sentry_lock);
1160 mutex_unlock(&curseg->curseg_mutex);
1163 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1164 struct page *page, enum page_type type)
1166 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1167 struct f2fs_bio_info *io = &sbi->write_io[btype];
1168 struct bio_vec *bvec;
1171 down_read(&io->io_rwsem);
1175 bio_for_each_segment_all(bvec, io->bio, i) {
1176 if (page == bvec->bv_page) {
1177 up_read(&io->io_rwsem);
1183 up_read(&io->io_rwsem);
1187 void f2fs_wait_on_page_writeback(struct page *page,
1188 enum page_type type)
1190 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1191 if (PageWriteback(page)) {
1192 if (is_merged_page(sbi, page, type))
1193 f2fs_submit_merged_bio(sbi, type, WRITE);
1194 wait_on_page_writeback(page);
1198 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1200 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1201 struct curseg_info *seg_i;
1202 unsigned char *kaddr;
1207 start = start_sum_block(sbi);
1209 page = get_meta_page(sbi, start++);
1210 kaddr = (unsigned char *)page_address(page);
1212 /* Step 1: restore nat cache */
1213 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1214 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1216 /* Step 2: restore sit cache */
1217 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1218 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1220 offset = 2 * SUM_JOURNAL_SIZE;
1222 /* Step 3: restore summary entries */
1223 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1224 unsigned short blk_off;
1227 seg_i = CURSEG_I(sbi, i);
1228 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1229 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1230 seg_i->next_segno = segno;
1231 reset_curseg(sbi, i, 0);
1232 seg_i->alloc_type = ckpt->alloc_type[i];
1233 seg_i->next_blkoff = blk_off;
1235 if (seg_i->alloc_type == SSR)
1236 blk_off = sbi->blocks_per_seg;
1238 for (j = 0; j < blk_off; j++) {
1239 struct f2fs_summary *s;
1240 s = (struct f2fs_summary *)(kaddr + offset);
1241 seg_i->sum_blk->entries[j] = *s;
1242 offset += SUMMARY_SIZE;
1243 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1247 f2fs_put_page(page, 1);
1250 page = get_meta_page(sbi, start++);
1251 kaddr = (unsigned char *)page_address(page);
1255 f2fs_put_page(page, 1);
1259 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1261 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1262 struct f2fs_summary_block *sum;
1263 struct curseg_info *curseg;
1265 unsigned short blk_off;
1266 unsigned int segno = 0;
1267 block_t blk_addr = 0;
1269 /* get segment number and block addr */
1270 if (IS_DATASEG(type)) {
1271 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1272 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1274 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1275 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1277 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1279 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1281 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1283 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1284 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1285 type - CURSEG_HOT_NODE);
1287 blk_addr = GET_SUM_BLOCK(sbi, segno);
1290 new = get_meta_page(sbi, blk_addr);
1291 sum = (struct f2fs_summary_block *)page_address(new);
1293 if (IS_NODESEG(type)) {
1294 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1295 struct f2fs_summary *ns = &sum->entries[0];
1297 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1299 ns->ofs_in_node = 0;
1304 err = restore_node_summary(sbi, segno, sum);
1306 f2fs_put_page(new, 1);
1312 /* set uncompleted segment to curseg */
1313 curseg = CURSEG_I(sbi, type);
1314 mutex_lock(&curseg->curseg_mutex);
1315 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1316 curseg->next_segno = segno;
1317 reset_curseg(sbi, type, 0);
1318 curseg->alloc_type = ckpt->alloc_type[type];
1319 curseg->next_blkoff = blk_off;
1320 mutex_unlock(&curseg->curseg_mutex);
1321 f2fs_put_page(new, 1);
1325 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1327 int type = CURSEG_HOT_DATA;
1330 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1331 /* restore for compacted data summary */
1332 if (read_compacted_summaries(sbi))
1334 type = CURSEG_HOT_NODE;
1337 for (; type <= CURSEG_COLD_NODE; type++) {
1338 err = read_normal_summaries(sbi, type);
1346 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1349 unsigned char *kaddr;
1350 struct f2fs_summary *summary;
1351 struct curseg_info *seg_i;
1352 int written_size = 0;
1355 page = grab_meta_page(sbi, blkaddr++);
1356 kaddr = (unsigned char *)page_address(page);
1358 /* Step 1: write nat cache */
1359 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1360 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1361 written_size += SUM_JOURNAL_SIZE;
1363 /* Step 2: write sit cache */
1364 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1365 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1367 written_size += SUM_JOURNAL_SIZE;
1369 /* Step 3: write summary entries */
1370 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1371 unsigned short blkoff;
1372 seg_i = CURSEG_I(sbi, i);
1373 if (sbi->ckpt->alloc_type[i] == SSR)
1374 blkoff = sbi->blocks_per_seg;
1376 blkoff = curseg_blkoff(sbi, i);
1378 for (j = 0; j < blkoff; j++) {
1380 page = grab_meta_page(sbi, blkaddr++);
1381 kaddr = (unsigned char *)page_address(page);
1384 summary = (struct f2fs_summary *)(kaddr + written_size);
1385 *summary = seg_i->sum_blk->entries[j];
1386 written_size += SUMMARY_SIZE;
1388 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1392 set_page_dirty(page);
1393 f2fs_put_page(page, 1);
1398 set_page_dirty(page);
1399 f2fs_put_page(page, 1);
1403 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1404 block_t blkaddr, int type)
1407 if (IS_DATASEG(type))
1408 end = type + NR_CURSEG_DATA_TYPE;
1410 end = type + NR_CURSEG_NODE_TYPE;
1412 for (i = type; i < end; i++) {
1413 struct curseg_info *sum = CURSEG_I(sbi, i);
1414 mutex_lock(&sum->curseg_mutex);
1415 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1416 mutex_unlock(&sum->curseg_mutex);
1420 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1422 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1423 write_compacted_summaries(sbi, start_blk);
1425 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1428 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1430 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1431 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1434 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1435 unsigned int val, int alloc)
1439 if (type == NAT_JOURNAL) {
1440 for (i = 0; i < nats_in_cursum(sum); i++) {
1441 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1444 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1445 return update_nats_in_cursum(sum, 1);
1446 } else if (type == SIT_JOURNAL) {
1447 for (i = 0; i < sits_in_cursum(sum); i++)
1448 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1450 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1451 return update_sits_in_cursum(sum, 1);
1456 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1459 struct sit_info *sit_i = SIT_I(sbi);
1460 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1461 block_t blk_addr = sit_i->sit_base_addr + offset;
1463 check_seg_range(sbi, segno);
1465 /* calculate sit block address */
1466 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1467 blk_addr += sit_i->sit_blocks;
1469 return get_meta_page(sbi, blk_addr);
1472 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1475 struct sit_info *sit_i = SIT_I(sbi);
1476 struct page *src_page, *dst_page;
1477 pgoff_t src_off, dst_off;
1478 void *src_addr, *dst_addr;
1480 src_off = current_sit_addr(sbi, start);
1481 dst_off = next_sit_addr(sbi, src_off);
1483 /* get current sit block page without lock */
1484 src_page = get_meta_page(sbi, src_off);
1485 dst_page = grab_meta_page(sbi, dst_off);
1486 f2fs_bug_on(PageDirty(src_page));
1488 src_addr = page_address(src_page);
1489 dst_addr = page_address(dst_page);
1490 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1492 set_page_dirty(dst_page);
1493 f2fs_put_page(src_page, 1);
1495 set_to_next_sit(sit_i, start);
1500 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1502 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1503 struct f2fs_summary_block *sum = curseg->sum_blk;
1507 * If the journal area in the current summary is full of sit entries,
1508 * all the sit entries will be flushed. Otherwise the sit entries
1509 * are not able to replace with newly hot sit entries.
1511 if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1512 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1514 segno = le32_to_cpu(segno_in_journal(sum, i));
1515 __mark_sit_entry_dirty(sbi, segno);
1517 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1524 * CP calls this function, which flushes SIT entries including sit_journal,
1525 * and moves prefree segs to free segs.
1527 void flush_sit_entries(struct f2fs_sb_info *sbi)
1529 struct sit_info *sit_i = SIT_I(sbi);
1530 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1531 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1532 struct f2fs_summary_block *sum = curseg->sum_blk;
1533 unsigned long nsegs = TOTAL_SEGS(sbi);
1534 struct page *page = NULL;
1535 struct f2fs_sit_block *raw_sit = NULL;
1536 unsigned int start = 0, end = 0;
1537 unsigned int segno = -1;
1540 mutex_lock(&curseg->curseg_mutex);
1541 mutex_lock(&sit_i->sentry_lock);
1544 * "flushed" indicates whether sit entries in journal are flushed
1545 * to the SIT area or not.
1547 flushed = flush_sits_in_journal(sbi);
1549 while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1550 struct seg_entry *se = get_seg_entry(sbi, segno);
1551 int sit_offset, offset;
1553 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1555 /* add discard candidates */
1556 if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1557 add_discard_addrs(sbi, segno, se);
1562 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1564 segno_in_journal(sum, offset) = cpu_to_le32(segno);
1565 seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1569 if (!page || (start > segno) || (segno > end)) {
1571 f2fs_put_page(page, 1);
1575 start = START_SEGNO(sit_i, segno);
1576 end = start + SIT_ENTRY_PER_BLOCK - 1;
1578 /* read sit block that will be updated */
1579 page = get_next_sit_page(sbi, start);
1580 raw_sit = page_address(page);
1583 /* udpate entry in SIT block */
1584 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1586 __clear_bit(segno, bitmap);
1587 sit_i->dirty_sentries--;
1589 mutex_unlock(&sit_i->sentry_lock);
1590 mutex_unlock(&curseg->curseg_mutex);
1592 /* writeout last modified SIT block */
1593 f2fs_put_page(page, 1);
1595 set_prefree_as_free_segments(sbi);
1598 static int build_sit_info(struct f2fs_sb_info *sbi)
1600 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1601 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1602 struct sit_info *sit_i;
1603 unsigned int sit_segs, start;
1604 char *src_bitmap, *dst_bitmap;
1605 unsigned int bitmap_size;
1607 /* allocate memory for SIT information */
1608 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1612 SM_I(sbi)->sit_info = sit_i;
1614 sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1615 if (!sit_i->sentries)
1618 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1619 sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1620 if (!sit_i->dirty_sentries_bitmap)
1623 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1624 sit_i->sentries[start].cur_valid_map
1625 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1626 sit_i->sentries[start].ckpt_valid_map
1627 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1628 if (!sit_i->sentries[start].cur_valid_map
1629 || !sit_i->sentries[start].ckpt_valid_map)
1633 if (sbi->segs_per_sec > 1) {
1634 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1635 sizeof(struct sec_entry));
1636 if (!sit_i->sec_entries)
1640 /* get information related with SIT */
1641 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1643 /* setup SIT bitmap from ckeckpoint pack */
1644 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1645 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1647 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1651 /* init SIT information */
1652 sit_i->s_ops = &default_salloc_ops;
1654 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1655 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1656 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1657 sit_i->sit_bitmap = dst_bitmap;
1658 sit_i->bitmap_size = bitmap_size;
1659 sit_i->dirty_sentries = 0;
1660 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1661 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1662 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1663 mutex_init(&sit_i->sentry_lock);
1667 static int build_free_segmap(struct f2fs_sb_info *sbi)
1669 struct f2fs_sm_info *sm_info = SM_I(sbi);
1670 struct free_segmap_info *free_i;
1671 unsigned int bitmap_size, sec_bitmap_size;
1673 /* allocate memory for free segmap information */
1674 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1678 SM_I(sbi)->free_info = free_i;
1680 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1681 free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1682 if (!free_i->free_segmap)
1685 sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1686 free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1687 if (!free_i->free_secmap)
1690 /* set all segments as dirty temporarily */
1691 memset(free_i->free_segmap, 0xff, bitmap_size);
1692 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1694 /* init free segmap information */
1695 free_i->start_segno =
1696 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1697 free_i->free_segments = 0;
1698 free_i->free_sections = 0;
1699 rwlock_init(&free_i->segmap_lock);
1703 static int build_curseg(struct f2fs_sb_info *sbi)
1705 struct curseg_info *array;
1708 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1712 SM_I(sbi)->curseg_array = array;
1714 for (i = 0; i < NR_CURSEG_TYPE; i++) {
1715 mutex_init(&array[i].curseg_mutex);
1716 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1717 if (!array[i].sum_blk)
1719 array[i].segno = NULL_SEGNO;
1720 array[i].next_blkoff = 0;
1722 return restore_curseg_summaries(sbi);
1725 static void build_sit_entries(struct f2fs_sb_info *sbi)
1727 struct sit_info *sit_i = SIT_I(sbi);
1728 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1729 struct f2fs_summary_block *sum = curseg->sum_blk;
1730 int sit_blk_cnt = SIT_BLK_CNT(sbi);
1731 unsigned int i, start, end;
1732 unsigned int readed, start_blk = 0;
1733 int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1736 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1738 start = start_blk * sit_i->sents_per_block;
1739 end = (start_blk + readed) * sit_i->sents_per_block;
1741 for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1742 struct seg_entry *se = &sit_i->sentries[start];
1743 struct f2fs_sit_block *sit_blk;
1744 struct f2fs_sit_entry sit;
1747 mutex_lock(&curseg->curseg_mutex);
1748 for (i = 0; i < sits_in_cursum(sum); i++) {
1749 if (le32_to_cpu(segno_in_journal(sum, i))
1751 sit = sit_in_journal(sum, i);
1752 mutex_unlock(&curseg->curseg_mutex);
1756 mutex_unlock(&curseg->curseg_mutex);
1758 page = get_current_sit_page(sbi, start);
1759 sit_blk = (struct f2fs_sit_block *)page_address(page);
1760 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1761 f2fs_put_page(page, 1);
1763 check_block_count(sbi, start, &sit);
1764 seg_info_from_raw_sit(se, &sit);
1765 if (sbi->segs_per_sec > 1) {
1766 struct sec_entry *e = get_sec_entry(sbi, start);
1767 e->valid_blocks += se->valid_blocks;
1770 start_blk += readed;
1771 } while (start_blk < sit_blk_cnt);
1774 static void init_free_segmap(struct f2fs_sb_info *sbi)
1779 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1780 struct seg_entry *sentry = get_seg_entry(sbi, start);
1781 if (!sentry->valid_blocks)
1782 __set_free(sbi, start);
1785 /* set use the current segments */
1786 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1787 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1788 __set_test_and_inuse(sbi, curseg_t->segno);
1792 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1794 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1795 struct free_segmap_info *free_i = FREE_I(sbi);
1796 unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1797 unsigned short valid_blocks;
1800 /* find dirty segment based on free segmap */
1801 segno = find_next_inuse(free_i, total_segs, offset);
1802 if (segno >= total_segs)
1805 valid_blocks = get_valid_blocks(sbi, segno, 0);
1806 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1808 mutex_lock(&dirty_i->seglist_lock);
1809 __locate_dirty_segment(sbi, segno, DIRTY);
1810 mutex_unlock(&dirty_i->seglist_lock);
1814 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1816 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1817 unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1819 dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1820 if (!dirty_i->victim_secmap)
1825 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1827 struct dirty_seglist_info *dirty_i;
1828 unsigned int bitmap_size, i;
1830 /* allocate memory for dirty segments list information */
1831 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1835 SM_I(sbi)->dirty_info = dirty_i;
1836 mutex_init(&dirty_i->seglist_lock);
1838 bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1840 for (i = 0; i < NR_DIRTY_TYPE; i++) {
1841 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1842 if (!dirty_i->dirty_segmap[i])
1846 init_dirty_segmap(sbi);
1847 return init_victim_secmap(sbi);
1851 * Update min, max modified time for cost-benefit GC algorithm
1853 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1855 struct sit_info *sit_i = SIT_I(sbi);
1858 mutex_lock(&sit_i->sentry_lock);
1860 sit_i->min_mtime = LLONG_MAX;
1862 for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1864 unsigned long long mtime = 0;
1866 for (i = 0; i < sbi->segs_per_sec; i++)
1867 mtime += get_seg_entry(sbi, segno + i)->mtime;
1869 mtime = div_u64(mtime, sbi->segs_per_sec);
1871 if (sit_i->min_mtime > mtime)
1872 sit_i->min_mtime = mtime;
1874 sit_i->max_mtime = get_mtime(sbi);
1875 mutex_unlock(&sit_i->sentry_lock);
1878 int build_segment_manager(struct f2fs_sb_info *sbi)
1880 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1881 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1882 struct f2fs_sm_info *sm_info;
1885 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1890 sbi->sm_info = sm_info;
1891 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1892 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1893 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1894 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1895 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1896 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1897 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1898 sm_info->rec_prefree_segments = sm_info->main_segments *
1899 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
1900 sm_info->ipu_policy = F2FS_IPU_DISABLE;
1901 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
1903 INIT_LIST_HEAD(&sm_info->discard_list);
1904 sm_info->nr_discards = 0;
1905 sm_info->max_discards = 0;
1907 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
1908 err = create_flush_cmd_control(sbi);
1913 err = build_sit_info(sbi);
1916 err = build_free_segmap(sbi);
1919 err = build_curseg(sbi);
1923 /* reinit free segmap based on SIT */
1924 build_sit_entries(sbi);
1926 init_free_segmap(sbi);
1927 err = build_dirty_segmap(sbi);
1931 init_min_max_mtime(sbi);
1935 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1936 enum dirty_type dirty_type)
1938 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1940 mutex_lock(&dirty_i->seglist_lock);
1941 kfree(dirty_i->dirty_segmap[dirty_type]);
1942 dirty_i->nr_dirty[dirty_type] = 0;
1943 mutex_unlock(&dirty_i->seglist_lock);
1946 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1948 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1949 kfree(dirty_i->victim_secmap);
1952 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1954 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1960 /* discard pre-free/dirty segments list */
1961 for (i = 0; i < NR_DIRTY_TYPE; i++)
1962 discard_dirty_segmap(sbi, i);
1964 destroy_victim_secmap(sbi);
1965 SM_I(sbi)->dirty_info = NULL;
1969 static void destroy_curseg(struct f2fs_sb_info *sbi)
1971 struct curseg_info *array = SM_I(sbi)->curseg_array;
1976 SM_I(sbi)->curseg_array = NULL;
1977 for (i = 0; i < NR_CURSEG_TYPE; i++)
1978 kfree(array[i].sum_blk);
1982 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1984 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1987 SM_I(sbi)->free_info = NULL;
1988 kfree(free_i->free_segmap);
1989 kfree(free_i->free_secmap);
1993 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1995 struct sit_info *sit_i = SIT_I(sbi);
2001 if (sit_i->sentries) {
2002 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
2003 kfree(sit_i->sentries[start].cur_valid_map);
2004 kfree(sit_i->sentries[start].ckpt_valid_map);
2007 vfree(sit_i->sentries);
2008 vfree(sit_i->sec_entries);
2009 kfree(sit_i->dirty_sentries_bitmap);
2011 SM_I(sbi)->sit_info = NULL;
2012 kfree(sit_i->sit_bitmap);
2016 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2018 struct f2fs_sm_info *sm_info = SM_I(sbi);
2022 destroy_flush_cmd_control(sbi);
2023 destroy_dirty_segmap(sbi);
2024 destroy_curseg(sbi);
2025 destroy_free_segmap(sbi);
2026 destroy_sit_info(sbi);
2027 sbi->sm_info = NULL;
2031 int __init create_segment_manager_caches(void)
2033 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2034 sizeof(struct discard_entry));
2035 if (!discard_entry_slab)
2040 void destroy_segment_manager_caches(void)
2042 kmem_cache_destroy(discard_entry_slab);