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/swap.h>
18 #include <linux/timer.h>
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
32 static unsigned long __reverse_ulong(unsigned char *str)
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
37 #if BITS_PER_LONG == 64
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 static inline unsigned long __reverse_ffs(unsigned long word)
55 #if BITS_PER_LONG == 64
56 if ((word & 0xffffffff00000000UL) == 0)
61 if ((word & 0xffff0000) == 0)
66 if ((word & 0xff00) == 0)
71 if ((word & 0xf0) == 0)
76 if ((word & 0xc) == 0)
81 if ((word & 0x2) == 0)
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * @size must be integral times of unsigned long.
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
95 static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
98 const unsigned long *p = addr + BIT_WORD(offset);
99 unsigned long result = size;
105 size -= (offset & ~(BITS_PER_LONG - 1));
106 offset %= BITS_PER_LONG;
112 tmp = __reverse_ulong((unsigned char *)p);
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
120 if (size <= BITS_PER_LONG)
122 size -= BITS_PER_LONG;
128 return result - size + __reverse_ffs(tmp);
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
134 const unsigned long *p = addr + BIT_WORD(offset);
135 unsigned long result = size;
141 size -= (offset & ~(BITS_PER_LONG - 1));
142 offset %= BITS_PER_LONG;
148 tmp = __reverse_ulong((unsigned char *)p);
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
157 if (size <= BITS_PER_LONG)
159 size -= BITS_PER_LONG;
165 return result - size + __reverse_ffz(tmp);
168 void register_inmem_page(struct inode *inode, struct page *page)
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
173 f2fs_trace_pid(page);
175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180 /* add atomic page indices to the list */
182 INIT_LIST_HEAD(&new->list);
184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
187 list_add_tail(&new->list, &fi->inmem_pages);
188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
189 mutex_unlock(&fi->inmem_lock);
191 trace_f2fs_register_inmem_page(page, INMEM);
194 static int __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head, bool drop, bool recover)
197 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
198 struct inmem_pages *cur, *tmp;
201 list_for_each_entry_safe(cur, tmp, head, list) {
202 struct page *page = cur->page;
205 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
210 struct dnode_of_data dn;
213 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215 set_new_dnode(&dn, inode, NULL, NULL, 0);
216 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
220 get_node_info(sbi, dn.nid, &ni);
221 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
222 cur->old_addr, ni.version, true, true);
226 /* we don't need to invalidate this in the sccessful status */
228 ClearPageUptodate(page);
229 set_page_private(page, 0);
230 ClearPagePrivate(page);
231 f2fs_put_page(page, 1);
233 list_del(&cur->list);
234 kmem_cache_free(inmem_entry_slab, cur);
235 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
240 void drop_inmem_pages(struct inode *inode)
242 struct f2fs_inode_info *fi = F2FS_I(inode);
244 clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
246 mutex_lock(&fi->inmem_lock);
247 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
248 mutex_unlock(&fi->inmem_lock);
251 static int __commit_inmem_pages(struct inode *inode,
252 struct list_head *revoke_list)
254 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct inmem_pages *cur, *tmp;
257 struct f2fs_io_info fio = {
260 .rw = WRITE_SYNC | REQ_PRIO,
261 .encrypted_page = NULL,
263 bool submit_bio = false;
266 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
267 struct page *page = cur->page;
270 if (page->mapping == inode->i_mapping) {
271 trace_f2fs_commit_inmem_page(page, INMEM);
273 set_page_dirty(page);
274 f2fs_wait_on_page_writeback(page, DATA, true);
275 if (clear_page_dirty_for_io(page))
276 inode_dec_dirty_pages(inode);
279 err = do_write_data_page(&fio);
285 /* record old blkaddr for revoking */
286 cur->old_addr = fio.old_blkaddr;
288 clear_cold_data(page);
292 list_move_tail(&cur->list, revoke_list);
296 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
299 __revoke_inmem_pages(inode, revoke_list, false, false);
304 int commit_inmem_pages(struct inode *inode)
306 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
307 struct f2fs_inode_info *fi = F2FS_I(inode);
308 struct list_head revoke_list;
311 INIT_LIST_HEAD(&revoke_list);
312 f2fs_balance_fs(sbi, true);
315 mutex_lock(&fi->inmem_lock);
316 err = __commit_inmem_pages(inode, &revoke_list);
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
327 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
334 mutex_unlock(&fi->inmem_lock);
341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
344 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
349 * We should do GC or end up with checkpoint, if there are so many dirty
350 * dir/node pages without enough free segments.
352 if (has_not_enough_free_secs(sbi, 0)) {
353 mutex_lock(&sbi->gc_mutex);
358 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
360 /* try to shrink extent cache when there is no enough memory */
361 if (!available_free_memory(sbi, EXTENT_CACHE))
362 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
364 /* check the # of cached NAT entries */
365 if (!available_free_memory(sbi, NAT_ENTRIES))
366 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
368 if (!available_free_memory(sbi, FREE_NIDS))
369 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
371 /* checkpoint is the only way to shrink partial cached entries */
372 if (!available_free_memory(sbi, NAT_ENTRIES) ||
373 !available_free_memory(sbi, INO_ENTRIES) ||
374 excess_prefree_segs(sbi) ||
375 excess_dirty_nats(sbi) ||
376 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
377 if (test_opt(sbi, DATA_FLUSH)) {
378 struct blk_plug plug;
380 blk_start_plug(&plug);
381 sync_dirty_inodes(sbi, FILE_INODE);
382 blk_finish_plug(&plug);
384 f2fs_sync_fs(sbi->sb, true);
385 stat_inc_bg_cp_count(sbi->stat_info);
389 static int issue_flush_thread(void *data)
391 struct f2fs_sb_info *sbi = data;
392 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
393 wait_queue_head_t *q = &fcc->flush_wait_queue;
395 if (kthread_should_stop())
398 if (!llist_empty(&fcc->issue_list)) {
400 struct flush_cmd *cmd, *next;
403 bio = f2fs_bio_alloc(0);
405 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
406 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
408 bio->bi_bdev = sbi->sb->s_bdev;
409 bio->bi_rw = WRITE_FLUSH;
410 ret = submit_bio_wait(bio);
412 llist_for_each_entry_safe(cmd, next,
413 fcc->dispatch_list, llnode) {
415 complete(&cmd->wait);
418 fcc->dispatch_list = NULL;
421 wait_event_interruptible(*q,
422 kthread_should_stop() || !llist_empty(&fcc->issue_list));
426 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
428 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
429 struct flush_cmd cmd;
431 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
432 test_opt(sbi, FLUSH_MERGE));
434 if (test_opt(sbi, NOBARRIER))
437 if (!test_opt(sbi, FLUSH_MERGE)) {
438 struct bio *bio = f2fs_bio_alloc(0);
441 bio->bi_bdev = sbi->sb->s_bdev;
442 bio->bi_rw = WRITE_FLUSH;
443 ret = submit_bio_wait(bio);
448 init_completion(&cmd.wait);
450 llist_add(&cmd.llnode, &fcc->issue_list);
452 if (!fcc->dispatch_list)
453 wake_up(&fcc->flush_wait_queue);
455 wait_for_completion(&cmd.wait);
460 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
462 dev_t dev = sbi->sb->s_bdev->bd_dev;
463 struct flush_cmd_control *fcc;
466 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
469 init_waitqueue_head(&fcc->flush_wait_queue);
470 init_llist_head(&fcc->issue_list);
471 SM_I(sbi)->cmd_control_info = fcc;
472 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
473 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
474 if (IS_ERR(fcc->f2fs_issue_flush)) {
475 err = PTR_ERR(fcc->f2fs_issue_flush);
477 SM_I(sbi)->cmd_control_info = NULL;
484 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
486 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
488 if (fcc && fcc->f2fs_issue_flush)
489 kthread_stop(fcc->f2fs_issue_flush);
491 SM_I(sbi)->cmd_control_info = NULL;
494 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
495 enum dirty_type dirty_type)
497 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
499 /* need not be added */
500 if (IS_CURSEG(sbi, segno))
503 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
504 dirty_i->nr_dirty[dirty_type]++;
506 if (dirty_type == DIRTY) {
507 struct seg_entry *sentry = get_seg_entry(sbi, segno);
508 enum dirty_type t = sentry->type;
510 if (unlikely(t >= DIRTY)) {
514 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
515 dirty_i->nr_dirty[t]++;
519 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
520 enum dirty_type dirty_type)
522 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
524 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
525 dirty_i->nr_dirty[dirty_type]--;
527 if (dirty_type == DIRTY) {
528 struct seg_entry *sentry = get_seg_entry(sbi, segno);
529 enum dirty_type t = sentry->type;
531 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
532 dirty_i->nr_dirty[t]--;
534 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
535 clear_bit(GET_SECNO(sbi, segno),
536 dirty_i->victim_secmap);
541 * Should not occur error such as -ENOMEM.
542 * Adding dirty entry into seglist is not critical operation.
543 * If a given segment is one of current working segments, it won't be added.
545 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
547 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
548 unsigned short valid_blocks;
550 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
553 mutex_lock(&dirty_i->seglist_lock);
555 valid_blocks = get_valid_blocks(sbi, segno, 0);
557 if (valid_blocks == 0) {
558 __locate_dirty_segment(sbi, segno, PRE);
559 __remove_dirty_segment(sbi, segno, DIRTY);
560 } else if (valid_blocks < sbi->blocks_per_seg) {
561 __locate_dirty_segment(sbi, segno, DIRTY);
563 /* Recovery routine with SSR needs this */
564 __remove_dirty_segment(sbi, segno, DIRTY);
567 mutex_unlock(&dirty_i->seglist_lock);
570 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
571 block_t blkstart, block_t blklen)
573 sector_t start = SECTOR_FROM_BLOCK(blkstart);
574 sector_t len = SECTOR_FROM_BLOCK(blklen);
575 struct seg_entry *se;
579 for (i = blkstart; i < blkstart + blklen; i++) {
580 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
581 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
583 if (!f2fs_test_and_set_bit(offset, se->discard_map))
586 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
587 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
590 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
592 int err = -EOPNOTSUPP;
594 if (test_opt(sbi, DISCARD)) {
595 struct seg_entry *se = get_seg_entry(sbi,
596 GET_SEGNO(sbi, blkaddr));
597 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
599 if (f2fs_test_bit(offset, se->discard_map))
602 err = f2fs_issue_discard(sbi, blkaddr, 1);
606 update_meta_page(sbi, NULL, blkaddr);
612 static void __add_discard_entry(struct f2fs_sb_info *sbi,
613 struct cp_control *cpc, struct seg_entry *se,
614 unsigned int start, unsigned int end)
616 struct list_head *head = &SM_I(sbi)->discard_list;
617 struct discard_entry *new, *last;
619 if (!list_empty(head)) {
620 last = list_last_entry(head, struct discard_entry, list);
621 if (START_BLOCK(sbi, cpc->trim_start) + start ==
622 last->blkaddr + last->len) {
623 last->len += end - start;
628 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
629 INIT_LIST_HEAD(&new->list);
630 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
631 new->len = end - start;
632 list_add_tail(&new->list, head);
634 SM_I(sbi)->nr_discards += end - start;
637 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
639 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
640 int max_blocks = sbi->blocks_per_seg;
641 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
642 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
643 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
644 unsigned long *discard_map = (unsigned long *)se->discard_map;
645 unsigned long *dmap = SIT_I(sbi)->tmp_map;
646 unsigned int start = 0, end = -1;
647 bool force = (cpc->reason == CP_DISCARD);
650 if (se->valid_blocks == max_blocks)
654 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
655 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
659 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
660 for (i = 0; i < entries; i++)
661 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
662 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
664 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
665 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
666 if (start >= max_blocks)
669 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
670 __add_discard_entry(sbi, cpc, se, start, end);
674 void release_discard_addrs(struct f2fs_sb_info *sbi)
676 struct list_head *head = &(SM_I(sbi)->discard_list);
677 struct discard_entry *entry, *this;
680 list_for_each_entry_safe(entry, this, head, list) {
681 list_del(&entry->list);
682 kmem_cache_free(discard_entry_slab, entry);
687 * Should call clear_prefree_segments after checkpoint is done.
689 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
691 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
694 mutex_lock(&dirty_i->seglist_lock);
695 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
696 __set_test_and_free(sbi, segno);
697 mutex_unlock(&dirty_i->seglist_lock);
700 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
702 struct list_head *head = &(SM_I(sbi)->discard_list);
703 struct discard_entry *entry, *this;
704 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
705 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
706 unsigned int start = 0, end = -1;
708 mutex_lock(&dirty_i->seglist_lock);
712 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
713 if (start >= MAIN_SEGS(sbi))
715 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
718 for (i = start; i < end; i++)
719 clear_bit(i, prefree_map);
721 dirty_i->nr_dirty[PRE] -= end - start;
723 if (!test_opt(sbi, DISCARD))
726 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
727 (end - start) << sbi->log_blocks_per_seg);
729 mutex_unlock(&dirty_i->seglist_lock);
731 /* send small discards */
732 list_for_each_entry_safe(entry, this, head, list) {
733 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
735 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
736 cpc->trimmed += entry->len;
738 list_del(&entry->list);
739 SM_I(sbi)->nr_discards -= entry->len;
740 kmem_cache_free(discard_entry_slab, entry);
744 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
746 struct sit_info *sit_i = SIT_I(sbi);
748 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
749 sit_i->dirty_sentries++;
756 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
757 unsigned int segno, int modified)
759 struct seg_entry *se = get_seg_entry(sbi, segno);
762 __mark_sit_entry_dirty(sbi, segno);
765 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
767 struct seg_entry *se;
768 unsigned int segno, offset;
769 long int new_vblocks;
771 segno = GET_SEGNO(sbi, blkaddr);
773 se = get_seg_entry(sbi, segno);
774 new_vblocks = se->valid_blocks + del;
775 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
777 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
778 (new_vblocks > sbi->blocks_per_seg)));
780 se->valid_blocks = new_vblocks;
781 se->mtime = get_mtime(sbi);
782 SIT_I(sbi)->max_mtime = se->mtime;
784 /* Update valid block bitmap */
786 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
788 if (!f2fs_test_and_set_bit(offset, se->discard_map))
791 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
793 if (f2fs_test_and_clear_bit(offset, se->discard_map))
796 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
797 se->ckpt_valid_blocks += del;
799 __mark_sit_entry_dirty(sbi, segno);
801 /* update total number of valid blocks to be written in ckpt area */
802 SIT_I(sbi)->written_valid_blocks += del;
804 if (sbi->segs_per_sec > 1)
805 get_sec_entry(sbi, segno)->valid_blocks += del;
808 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
810 update_sit_entry(sbi, new, 1);
811 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
812 update_sit_entry(sbi, old, -1);
814 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
815 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
818 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
820 unsigned int segno = GET_SEGNO(sbi, addr);
821 struct sit_info *sit_i = SIT_I(sbi);
823 f2fs_bug_on(sbi, addr == NULL_ADDR);
824 if (addr == NEW_ADDR)
827 /* add it into sit main buffer */
828 mutex_lock(&sit_i->sentry_lock);
830 update_sit_entry(sbi, addr, -1);
832 /* add it into dirty seglist */
833 locate_dirty_segment(sbi, segno);
835 mutex_unlock(&sit_i->sentry_lock);
838 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
840 struct sit_info *sit_i = SIT_I(sbi);
841 unsigned int segno, offset;
842 struct seg_entry *se;
845 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
848 mutex_lock(&sit_i->sentry_lock);
850 segno = GET_SEGNO(sbi, blkaddr);
851 se = get_seg_entry(sbi, segno);
852 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
854 if (f2fs_test_bit(offset, se->ckpt_valid_map))
857 mutex_unlock(&sit_i->sentry_lock);
863 * This function should be resided under the curseg_mutex lock
865 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
866 struct f2fs_summary *sum)
868 struct curseg_info *curseg = CURSEG_I(sbi, type);
869 void *addr = curseg->sum_blk;
870 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
871 memcpy(addr, sum, sizeof(struct f2fs_summary));
875 * Calculate the number of current summary pages for writing
877 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
879 int valid_sum_count = 0;
882 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
883 if (sbi->ckpt->alloc_type[i] == SSR)
884 valid_sum_count += sbi->blocks_per_seg;
887 valid_sum_count += le16_to_cpu(
888 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
890 valid_sum_count += curseg_blkoff(sbi, i);
894 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
895 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
896 if (valid_sum_count <= sum_in_page)
898 else if ((valid_sum_count - sum_in_page) <=
899 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
905 * Caller should put this summary page
907 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
909 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
912 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
914 struct page *page = grab_meta_page(sbi, blk_addr);
915 void *dst = page_address(page);
918 memcpy(dst, src, PAGE_SIZE);
920 memset(dst, 0, PAGE_SIZE);
921 set_page_dirty(page);
922 f2fs_put_page(page, 1);
925 static void write_sum_page(struct f2fs_sb_info *sbi,
926 struct f2fs_summary_block *sum_blk, block_t blk_addr)
928 update_meta_page(sbi, (void *)sum_blk, blk_addr);
931 static void write_current_sum_page(struct f2fs_sb_info *sbi,
932 int type, block_t blk_addr)
934 struct curseg_info *curseg = CURSEG_I(sbi, type);
935 struct page *page = grab_meta_page(sbi, blk_addr);
936 struct f2fs_summary_block *src = curseg->sum_blk;
937 struct f2fs_summary_block *dst;
939 dst = (struct f2fs_summary_block *)page_address(page);
941 mutex_lock(&curseg->curseg_mutex);
943 down_read(&curseg->journal_rwsem);
944 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
945 up_read(&curseg->journal_rwsem);
947 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
948 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
950 mutex_unlock(&curseg->curseg_mutex);
952 set_page_dirty(page);
953 f2fs_put_page(page, 1);
956 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
958 struct curseg_info *curseg = CURSEG_I(sbi, type);
959 unsigned int segno = curseg->segno + 1;
960 struct free_segmap_info *free_i = FREE_I(sbi);
962 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
963 return !test_bit(segno, free_i->free_segmap);
968 * Find a new segment from the free segments bitmap to right order
969 * This function should be returned with success, otherwise BUG
971 static void get_new_segment(struct f2fs_sb_info *sbi,
972 unsigned int *newseg, bool new_sec, int dir)
974 struct free_segmap_info *free_i = FREE_I(sbi);
975 unsigned int segno, secno, zoneno;
976 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
977 unsigned int hint = *newseg / sbi->segs_per_sec;
978 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
979 unsigned int left_start = hint;
984 spin_lock(&free_i->segmap_lock);
986 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
987 segno = find_next_zero_bit(free_i->free_segmap,
988 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
989 if (segno < (hint + 1) * sbi->segs_per_sec)
993 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
994 if (secno >= MAIN_SECS(sbi)) {
995 if (dir == ALLOC_RIGHT) {
996 secno = find_next_zero_bit(free_i->free_secmap,
998 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1001 left_start = hint - 1;
1007 while (test_bit(left_start, free_i->free_secmap)) {
1008 if (left_start > 0) {
1012 left_start = find_next_zero_bit(free_i->free_secmap,
1014 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1020 segno = secno * sbi->segs_per_sec;
1021 zoneno = secno / sbi->secs_per_zone;
1023 /* give up on finding another zone */
1026 if (sbi->secs_per_zone == 1)
1028 if (zoneno == old_zoneno)
1030 if (dir == ALLOC_LEFT) {
1031 if (!go_left && zoneno + 1 >= total_zones)
1033 if (go_left && zoneno == 0)
1036 for (i = 0; i < NR_CURSEG_TYPE; i++)
1037 if (CURSEG_I(sbi, i)->zone == zoneno)
1040 if (i < NR_CURSEG_TYPE) {
1041 /* zone is in user, try another */
1043 hint = zoneno * sbi->secs_per_zone - 1;
1044 else if (zoneno + 1 >= total_zones)
1047 hint = (zoneno + 1) * sbi->secs_per_zone;
1049 goto find_other_zone;
1052 /* set it as dirty segment in free segmap */
1053 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1054 __set_inuse(sbi, segno);
1056 spin_unlock(&free_i->segmap_lock);
1059 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1061 struct curseg_info *curseg = CURSEG_I(sbi, type);
1062 struct summary_footer *sum_footer;
1064 curseg->segno = curseg->next_segno;
1065 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1066 curseg->next_blkoff = 0;
1067 curseg->next_segno = NULL_SEGNO;
1069 sum_footer = &(curseg->sum_blk->footer);
1070 memset(sum_footer, 0, sizeof(struct summary_footer));
1071 if (IS_DATASEG(type))
1072 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1073 if (IS_NODESEG(type))
1074 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1075 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1079 * Allocate a current working segment.
1080 * This function always allocates a free segment in LFS manner.
1082 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1084 struct curseg_info *curseg = CURSEG_I(sbi, type);
1085 unsigned int segno = curseg->segno;
1086 int dir = ALLOC_LEFT;
1088 write_sum_page(sbi, curseg->sum_blk,
1089 GET_SUM_BLOCK(sbi, segno));
1090 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1093 if (test_opt(sbi, NOHEAP))
1096 get_new_segment(sbi, &segno, new_sec, dir);
1097 curseg->next_segno = segno;
1098 reset_curseg(sbi, type, 1);
1099 curseg->alloc_type = LFS;
1102 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1103 struct curseg_info *seg, block_t start)
1105 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1106 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1107 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1108 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1109 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1112 for (i = 0; i < entries; i++)
1113 target_map[i] = ckpt_map[i] | cur_map[i];
1115 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1117 seg->next_blkoff = pos;
1121 * If a segment is written by LFS manner, next block offset is just obtained
1122 * by increasing the current block offset. However, if a segment is written by
1123 * SSR manner, next block offset obtained by calling __next_free_blkoff
1125 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1126 struct curseg_info *seg)
1128 if (seg->alloc_type == SSR)
1129 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1135 * This function always allocates a used segment(from dirty seglist) by SSR
1136 * manner, so it should recover the existing segment information of valid blocks
1138 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1140 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1141 struct curseg_info *curseg = CURSEG_I(sbi, type);
1142 unsigned int new_segno = curseg->next_segno;
1143 struct f2fs_summary_block *sum_node;
1144 struct page *sum_page;
1146 write_sum_page(sbi, curseg->sum_blk,
1147 GET_SUM_BLOCK(sbi, curseg->segno));
1148 __set_test_and_inuse(sbi, new_segno);
1150 mutex_lock(&dirty_i->seglist_lock);
1151 __remove_dirty_segment(sbi, new_segno, PRE);
1152 __remove_dirty_segment(sbi, new_segno, DIRTY);
1153 mutex_unlock(&dirty_i->seglist_lock);
1155 reset_curseg(sbi, type, 1);
1156 curseg->alloc_type = SSR;
1157 __next_free_blkoff(sbi, curseg, 0);
1160 sum_page = get_sum_page(sbi, new_segno);
1161 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1162 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1163 f2fs_put_page(sum_page, 1);
1167 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1169 struct curseg_info *curseg = CURSEG_I(sbi, type);
1170 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1172 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1173 return v_ops->get_victim(sbi,
1174 &(curseg)->next_segno, BG_GC, type, SSR);
1176 /* For data segments, let's do SSR more intensively */
1177 for (; type >= CURSEG_HOT_DATA; type--)
1178 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1185 * flush out current segment and replace it with new segment
1186 * This function should be returned with success, otherwise BUG
1188 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1189 int type, bool force)
1191 struct curseg_info *curseg = CURSEG_I(sbi, type);
1194 new_curseg(sbi, type, true);
1195 else if (type == CURSEG_WARM_NODE)
1196 new_curseg(sbi, type, false);
1197 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1198 new_curseg(sbi, type, false);
1199 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1200 change_curseg(sbi, type, true);
1202 new_curseg(sbi, type, false);
1204 stat_inc_seg_type(sbi, curseg);
1207 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1209 struct curseg_info *curseg = CURSEG_I(sbi, type);
1210 unsigned int old_segno;
1212 old_segno = curseg->segno;
1213 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1214 locate_dirty_segment(sbi, old_segno);
1217 void allocate_new_segments(struct f2fs_sb_info *sbi)
1221 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1222 __allocate_new_segments(sbi, i);
1225 static const struct segment_allocation default_salloc_ops = {
1226 .allocate_segment = allocate_segment_by_default,
1229 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1231 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1232 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1233 unsigned int start_segno, end_segno;
1234 struct cp_control cpc;
1237 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1241 if (end <= MAIN_BLKADDR(sbi))
1244 /* start/end segment number in main_area */
1245 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1246 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1247 GET_SEGNO(sbi, end);
1248 cpc.reason = CP_DISCARD;
1249 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1251 /* do checkpoint to issue discard commands safely */
1252 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1253 cpc.trim_start = start_segno;
1255 if (sbi->discard_blks == 0)
1257 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1258 cpc.trim_end = end_segno;
1260 cpc.trim_end = min_t(unsigned int,
1261 rounddown(start_segno +
1262 BATCHED_TRIM_SEGMENTS(sbi),
1263 sbi->segs_per_sec) - 1, end_segno);
1265 mutex_lock(&sbi->gc_mutex);
1266 err = write_checkpoint(sbi, &cpc);
1267 mutex_unlock(&sbi->gc_mutex);
1270 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1274 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1276 struct curseg_info *curseg = CURSEG_I(sbi, type);
1277 if (curseg->next_blkoff < sbi->blocks_per_seg)
1282 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1285 return CURSEG_HOT_DATA;
1287 return CURSEG_HOT_NODE;
1290 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1292 if (p_type == DATA) {
1293 struct inode *inode = page->mapping->host;
1295 if (S_ISDIR(inode->i_mode))
1296 return CURSEG_HOT_DATA;
1298 return CURSEG_COLD_DATA;
1300 if (IS_DNODE(page) && is_cold_node(page))
1301 return CURSEG_WARM_NODE;
1303 return CURSEG_COLD_NODE;
1307 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1309 if (p_type == DATA) {
1310 struct inode *inode = page->mapping->host;
1312 if (S_ISDIR(inode->i_mode))
1313 return CURSEG_HOT_DATA;
1314 else if (is_cold_data(page) || file_is_cold(inode))
1315 return CURSEG_COLD_DATA;
1317 return CURSEG_WARM_DATA;
1320 return is_cold_node(page) ? CURSEG_WARM_NODE :
1323 return CURSEG_COLD_NODE;
1327 static int __get_segment_type(struct page *page, enum page_type p_type)
1329 switch (F2FS_P_SB(page)->active_logs) {
1331 return __get_segment_type_2(page, p_type);
1333 return __get_segment_type_4(page, p_type);
1335 /* NR_CURSEG_TYPE(6) logs by default */
1336 f2fs_bug_on(F2FS_P_SB(page),
1337 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1338 return __get_segment_type_6(page, p_type);
1341 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1342 block_t old_blkaddr, block_t *new_blkaddr,
1343 struct f2fs_summary *sum, int type)
1345 struct sit_info *sit_i = SIT_I(sbi);
1346 struct curseg_info *curseg;
1347 bool direct_io = (type == CURSEG_DIRECT_IO);
1349 type = direct_io ? CURSEG_WARM_DATA : type;
1351 curseg = CURSEG_I(sbi, type);
1353 mutex_lock(&curseg->curseg_mutex);
1354 mutex_lock(&sit_i->sentry_lock);
1356 /* direct_io'ed data is aligned to the segment for better performance */
1357 if (direct_io && curseg->next_blkoff &&
1358 !has_not_enough_free_secs(sbi, 0))
1359 __allocate_new_segments(sbi, type);
1361 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1364 * __add_sum_entry should be resided under the curseg_mutex
1365 * because, this function updates a summary entry in the
1366 * current summary block.
1368 __add_sum_entry(sbi, type, sum);
1370 __refresh_next_blkoff(sbi, curseg);
1372 stat_inc_block_count(sbi, curseg);
1374 if (!__has_curseg_space(sbi, type))
1375 sit_i->s_ops->allocate_segment(sbi, type, false);
1377 * SIT information should be updated before segment allocation,
1378 * since SSR needs latest valid block information.
1380 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1382 mutex_unlock(&sit_i->sentry_lock);
1384 if (page && IS_NODESEG(type))
1385 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1387 mutex_unlock(&curseg->curseg_mutex);
1390 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1392 int type = __get_segment_type(fio->page, fio->type);
1394 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1395 &fio->new_blkaddr, sum, type);
1397 /* writeout dirty page into bdev */
1398 f2fs_submit_page_mbio(fio);
1401 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1403 struct f2fs_io_info fio = {
1406 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1407 .old_blkaddr = page->index,
1408 .new_blkaddr = page->index,
1410 .encrypted_page = NULL,
1413 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1414 fio.rw &= ~REQ_META;
1416 set_page_writeback(page);
1417 f2fs_submit_page_mbio(&fio);
1420 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1422 struct f2fs_summary sum;
1424 set_summary(&sum, nid, 0, 0);
1425 do_write_page(&sum, fio);
1428 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1430 struct f2fs_sb_info *sbi = fio->sbi;
1431 struct f2fs_summary sum;
1432 struct node_info ni;
1434 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1435 get_node_info(sbi, dn->nid, &ni);
1436 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1437 do_write_page(&sum, fio);
1438 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1441 void rewrite_data_page(struct f2fs_io_info *fio)
1443 fio->new_blkaddr = fio->old_blkaddr;
1444 stat_inc_inplace_blocks(fio->sbi);
1445 f2fs_submit_page_mbio(fio);
1448 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1449 block_t old_blkaddr, block_t new_blkaddr,
1450 bool recover_curseg, bool recover_newaddr)
1452 struct sit_info *sit_i = SIT_I(sbi);
1453 struct curseg_info *curseg;
1454 unsigned int segno, old_cursegno;
1455 struct seg_entry *se;
1457 unsigned short old_blkoff;
1459 segno = GET_SEGNO(sbi, new_blkaddr);
1460 se = get_seg_entry(sbi, segno);
1463 if (!recover_curseg) {
1464 /* for recovery flow */
1465 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1466 if (old_blkaddr == NULL_ADDR)
1467 type = CURSEG_COLD_DATA;
1469 type = CURSEG_WARM_DATA;
1472 if (!IS_CURSEG(sbi, segno))
1473 type = CURSEG_WARM_DATA;
1476 curseg = CURSEG_I(sbi, type);
1478 mutex_lock(&curseg->curseg_mutex);
1479 mutex_lock(&sit_i->sentry_lock);
1481 old_cursegno = curseg->segno;
1482 old_blkoff = curseg->next_blkoff;
1484 /* change the current segment */
1485 if (segno != curseg->segno) {
1486 curseg->next_segno = segno;
1487 change_curseg(sbi, type, true);
1490 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1491 __add_sum_entry(sbi, type, sum);
1493 if (!recover_curseg || recover_newaddr)
1494 update_sit_entry(sbi, new_blkaddr, 1);
1495 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1496 update_sit_entry(sbi, old_blkaddr, -1);
1498 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1499 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1501 locate_dirty_segment(sbi, old_cursegno);
1503 if (recover_curseg) {
1504 if (old_cursegno != curseg->segno) {
1505 curseg->next_segno = old_cursegno;
1506 change_curseg(sbi, type, true);
1508 curseg->next_blkoff = old_blkoff;
1511 mutex_unlock(&sit_i->sentry_lock);
1512 mutex_unlock(&curseg->curseg_mutex);
1515 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1516 block_t old_addr, block_t new_addr,
1517 unsigned char version, bool recover_curseg,
1518 bool recover_newaddr)
1520 struct f2fs_summary sum;
1522 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1524 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1525 recover_curseg, recover_newaddr);
1527 f2fs_update_data_blkaddr(dn, new_addr);
1530 void f2fs_wait_on_page_writeback(struct page *page,
1531 enum page_type type, bool ordered)
1533 if (PageWriteback(page)) {
1534 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1536 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1538 wait_on_page_writeback(page);
1540 wait_for_stable_page(page);
1544 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1549 if (blkaddr == NEW_ADDR)
1552 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1554 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1556 f2fs_wait_on_page_writeback(cpage, DATA, true);
1557 f2fs_put_page(cpage, 1);
1561 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1563 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1564 struct curseg_info *seg_i;
1565 unsigned char *kaddr;
1570 start = start_sum_block(sbi);
1572 page = get_meta_page(sbi, start++);
1573 kaddr = (unsigned char *)page_address(page);
1575 /* Step 1: restore nat cache */
1576 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1577 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1579 /* Step 2: restore sit cache */
1580 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1581 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1582 offset = 2 * SUM_JOURNAL_SIZE;
1584 /* Step 3: restore summary entries */
1585 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1586 unsigned short blk_off;
1589 seg_i = CURSEG_I(sbi, i);
1590 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1591 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1592 seg_i->next_segno = segno;
1593 reset_curseg(sbi, i, 0);
1594 seg_i->alloc_type = ckpt->alloc_type[i];
1595 seg_i->next_blkoff = blk_off;
1597 if (seg_i->alloc_type == SSR)
1598 blk_off = sbi->blocks_per_seg;
1600 for (j = 0; j < blk_off; j++) {
1601 struct f2fs_summary *s;
1602 s = (struct f2fs_summary *)(kaddr + offset);
1603 seg_i->sum_blk->entries[j] = *s;
1604 offset += SUMMARY_SIZE;
1605 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1609 f2fs_put_page(page, 1);
1612 page = get_meta_page(sbi, start++);
1613 kaddr = (unsigned char *)page_address(page);
1617 f2fs_put_page(page, 1);
1621 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1623 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1624 struct f2fs_summary_block *sum;
1625 struct curseg_info *curseg;
1627 unsigned short blk_off;
1628 unsigned int segno = 0;
1629 block_t blk_addr = 0;
1631 /* get segment number and block addr */
1632 if (IS_DATASEG(type)) {
1633 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1634 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1636 if (__exist_node_summaries(sbi))
1637 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1639 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1641 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1643 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1645 if (__exist_node_summaries(sbi))
1646 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1647 type - CURSEG_HOT_NODE);
1649 blk_addr = GET_SUM_BLOCK(sbi, segno);
1652 new = get_meta_page(sbi, blk_addr);
1653 sum = (struct f2fs_summary_block *)page_address(new);
1655 if (IS_NODESEG(type)) {
1656 if (__exist_node_summaries(sbi)) {
1657 struct f2fs_summary *ns = &sum->entries[0];
1659 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1661 ns->ofs_in_node = 0;
1666 err = restore_node_summary(sbi, segno, sum);
1668 f2fs_put_page(new, 1);
1674 /* set uncompleted segment to curseg */
1675 curseg = CURSEG_I(sbi, type);
1676 mutex_lock(&curseg->curseg_mutex);
1678 /* update journal info */
1679 down_write(&curseg->journal_rwsem);
1680 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1681 up_write(&curseg->journal_rwsem);
1683 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1684 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1685 curseg->next_segno = segno;
1686 reset_curseg(sbi, type, 0);
1687 curseg->alloc_type = ckpt->alloc_type[type];
1688 curseg->next_blkoff = blk_off;
1689 mutex_unlock(&curseg->curseg_mutex);
1690 f2fs_put_page(new, 1);
1694 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1696 int type = CURSEG_HOT_DATA;
1699 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1700 int npages = npages_for_summary_flush(sbi, true);
1703 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1706 /* restore for compacted data summary */
1707 if (read_compacted_summaries(sbi))
1709 type = CURSEG_HOT_NODE;
1712 if (__exist_node_summaries(sbi))
1713 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1714 NR_CURSEG_TYPE - type, META_CP, true);
1716 for (; type <= CURSEG_COLD_NODE; type++) {
1717 err = read_normal_summaries(sbi, type);
1725 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1728 unsigned char *kaddr;
1729 struct f2fs_summary *summary;
1730 struct curseg_info *seg_i;
1731 int written_size = 0;
1734 page = grab_meta_page(sbi, blkaddr++);
1735 kaddr = (unsigned char *)page_address(page);
1737 /* Step 1: write nat cache */
1738 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1739 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1740 written_size += SUM_JOURNAL_SIZE;
1742 /* Step 2: write sit cache */
1743 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1744 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1745 written_size += SUM_JOURNAL_SIZE;
1747 /* Step 3: write summary entries */
1748 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1749 unsigned short blkoff;
1750 seg_i = CURSEG_I(sbi, i);
1751 if (sbi->ckpt->alloc_type[i] == SSR)
1752 blkoff = sbi->blocks_per_seg;
1754 blkoff = curseg_blkoff(sbi, i);
1756 for (j = 0; j < blkoff; j++) {
1758 page = grab_meta_page(sbi, blkaddr++);
1759 kaddr = (unsigned char *)page_address(page);
1762 summary = (struct f2fs_summary *)(kaddr + written_size);
1763 *summary = seg_i->sum_blk->entries[j];
1764 written_size += SUMMARY_SIZE;
1766 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1770 set_page_dirty(page);
1771 f2fs_put_page(page, 1);
1776 set_page_dirty(page);
1777 f2fs_put_page(page, 1);
1781 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1782 block_t blkaddr, int type)
1785 if (IS_DATASEG(type))
1786 end = type + NR_CURSEG_DATA_TYPE;
1788 end = type + NR_CURSEG_NODE_TYPE;
1790 for (i = type; i < end; i++)
1791 write_current_sum_page(sbi, i, blkaddr + (i - type));
1794 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1796 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1797 write_compacted_summaries(sbi, start_blk);
1799 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1802 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1804 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1807 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1808 unsigned int val, int alloc)
1812 if (type == NAT_JOURNAL) {
1813 for (i = 0; i < nats_in_cursum(journal); i++) {
1814 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1817 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1818 return update_nats_in_cursum(journal, 1);
1819 } else if (type == SIT_JOURNAL) {
1820 for (i = 0; i < sits_in_cursum(journal); i++)
1821 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1823 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1824 return update_sits_in_cursum(journal, 1);
1829 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1832 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1835 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1838 struct sit_info *sit_i = SIT_I(sbi);
1839 struct page *src_page, *dst_page;
1840 pgoff_t src_off, dst_off;
1841 void *src_addr, *dst_addr;
1843 src_off = current_sit_addr(sbi, start);
1844 dst_off = next_sit_addr(sbi, src_off);
1846 /* get current sit block page without lock */
1847 src_page = get_meta_page(sbi, src_off);
1848 dst_page = grab_meta_page(sbi, dst_off);
1849 f2fs_bug_on(sbi, PageDirty(src_page));
1851 src_addr = page_address(src_page);
1852 dst_addr = page_address(dst_page);
1853 memcpy(dst_addr, src_addr, PAGE_SIZE);
1855 set_page_dirty(dst_page);
1856 f2fs_put_page(src_page, 1);
1858 set_to_next_sit(sit_i, start);
1863 static struct sit_entry_set *grab_sit_entry_set(void)
1865 struct sit_entry_set *ses =
1866 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1869 INIT_LIST_HEAD(&ses->set_list);
1873 static void release_sit_entry_set(struct sit_entry_set *ses)
1875 list_del(&ses->set_list);
1876 kmem_cache_free(sit_entry_set_slab, ses);
1879 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1880 struct list_head *head)
1882 struct sit_entry_set *next = ses;
1884 if (list_is_last(&ses->set_list, head))
1887 list_for_each_entry_continue(next, head, set_list)
1888 if (ses->entry_cnt <= next->entry_cnt)
1891 list_move_tail(&ses->set_list, &next->set_list);
1894 static void add_sit_entry(unsigned int segno, struct list_head *head)
1896 struct sit_entry_set *ses;
1897 unsigned int start_segno = START_SEGNO(segno);
1899 list_for_each_entry(ses, head, set_list) {
1900 if (ses->start_segno == start_segno) {
1902 adjust_sit_entry_set(ses, head);
1907 ses = grab_sit_entry_set();
1909 ses->start_segno = start_segno;
1911 list_add(&ses->set_list, head);
1914 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1916 struct f2fs_sm_info *sm_info = SM_I(sbi);
1917 struct list_head *set_list = &sm_info->sit_entry_set;
1918 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1921 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1922 add_sit_entry(segno, set_list);
1925 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1927 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1928 struct f2fs_journal *journal = curseg->journal;
1931 down_write(&curseg->journal_rwsem);
1932 for (i = 0; i < sits_in_cursum(journal); i++) {
1936 segno = le32_to_cpu(segno_in_journal(journal, i));
1937 dirtied = __mark_sit_entry_dirty(sbi, segno);
1940 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1942 update_sits_in_cursum(journal, -i);
1943 up_write(&curseg->journal_rwsem);
1947 * CP calls this function, which flushes SIT entries including sit_journal,
1948 * and moves prefree segs to free segs.
1950 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1952 struct sit_info *sit_i = SIT_I(sbi);
1953 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1954 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1955 struct f2fs_journal *journal = curseg->journal;
1956 struct sit_entry_set *ses, *tmp;
1957 struct list_head *head = &SM_I(sbi)->sit_entry_set;
1958 bool to_journal = true;
1959 struct seg_entry *se;
1961 mutex_lock(&sit_i->sentry_lock);
1963 if (!sit_i->dirty_sentries)
1967 * add and account sit entries of dirty bitmap in sit entry
1970 add_sits_in_set(sbi);
1973 * if there are no enough space in journal to store dirty sit
1974 * entries, remove all entries from journal and add and account
1975 * them in sit entry set.
1977 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
1978 remove_sits_in_journal(sbi);
1981 * there are two steps to flush sit entries:
1982 * #1, flush sit entries to journal in current cold data summary block.
1983 * #2, flush sit entries to sit page.
1985 list_for_each_entry_safe(ses, tmp, head, set_list) {
1986 struct page *page = NULL;
1987 struct f2fs_sit_block *raw_sit = NULL;
1988 unsigned int start_segno = ses->start_segno;
1989 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1990 (unsigned long)MAIN_SEGS(sbi));
1991 unsigned int segno = start_segno;
1994 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
1998 down_write(&curseg->journal_rwsem);
2000 page = get_next_sit_page(sbi, start_segno);
2001 raw_sit = page_address(page);
2004 /* flush dirty sit entries in region of current sit set */
2005 for_each_set_bit_from(segno, bitmap, end) {
2006 int offset, sit_offset;
2008 se = get_seg_entry(sbi, segno);
2010 /* add discard candidates */
2011 if (cpc->reason != CP_DISCARD) {
2012 cpc->trim_start = segno;
2013 add_discard_addrs(sbi, cpc);
2017 offset = lookup_journal_in_cursum(journal,
2018 SIT_JOURNAL, segno, 1);
2019 f2fs_bug_on(sbi, offset < 0);
2020 segno_in_journal(journal, offset) =
2022 seg_info_to_raw_sit(se,
2023 &sit_in_journal(journal, offset));
2025 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2026 seg_info_to_raw_sit(se,
2027 &raw_sit->entries[sit_offset]);
2030 __clear_bit(segno, bitmap);
2031 sit_i->dirty_sentries--;
2036 up_write(&curseg->journal_rwsem);
2038 f2fs_put_page(page, 1);
2040 f2fs_bug_on(sbi, ses->entry_cnt);
2041 release_sit_entry_set(ses);
2044 f2fs_bug_on(sbi, !list_empty(head));
2045 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2047 if (cpc->reason == CP_DISCARD) {
2048 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2049 add_discard_addrs(sbi, cpc);
2051 mutex_unlock(&sit_i->sentry_lock);
2053 set_prefree_as_free_segments(sbi);
2056 static int build_sit_info(struct f2fs_sb_info *sbi)
2058 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2059 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2060 struct sit_info *sit_i;
2061 unsigned int sit_segs, start;
2062 char *src_bitmap, *dst_bitmap;
2063 unsigned int bitmap_size;
2065 /* allocate memory for SIT information */
2066 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2070 SM_I(sbi)->sit_info = sit_i;
2072 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2073 sizeof(struct seg_entry), GFP_KERNEL);
2074 if (!sit_i->sentries)
2077 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2078 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2079 if (!sit_i->dirty_sentries_bitmap)
2082 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2083 sit_i->sentries[start].cur_valid_map
2084 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2085 sit_i->sentries[start].ckpt_valid_map
2086 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2087 sit_i->sentries[start].discard_map
2088 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2089 if (!sit_i->sentries[start].cur_valid_map ||
2090 !sit_i->sentries[start].ckpt_valid_map ||
2091 !sit_i->sentries[start].discard_map)
2095 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2096 if (!sit_i->tmp_map)
2099 if (sbi->segs_per_sec > 1) {
2100 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2101 sizeof(struct sec_entry), GFP_KERNEL);
2102 if (!sit_i->sec_entries)
2106 /* get information related with SIT */
2107 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2109 /* setup SIT bitmap from ckeckpoint pack */
2110 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2111 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2113 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2117 /* init SIT information */
2118 sit_i->s_ops = &default_salloc_ops;
2120 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2121 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2122 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2123 sit_i->sit_bitmap = dst_bitmap;
2124 sit_i->bitmap_size = bitmap_size;
2125 sit_i->dirty_sentries = 0;
2126 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2127 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2128 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2129 mutex_init(&sit_i->sentry_lock);
2133 static int build_free_segmap(struct f2fs_sb_info *sbi)
2135 struct free_segmap_info *free_i;
2136 unsigned int bitmap_size, sec_bitmap_size;
2138 /* allocate memory for free segmap information */
2139 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2143 SM_I(sbi)->free_info = free_i;
2145 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2146 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2147 if (!free_i->free_segmap)
2150 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2151 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2152 if (!free_i->free_secmap)
2155 /* set all segments as dirty temporarily */
2156 memset(free_i->free_segmap, 0xff, bitmap_size);
2157 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2159 /* init free segmap information */
2160 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2161 free_i->free_segments = 0;
2162 free_i->free_sections = 0;
2163 spin_lock_init(&free_i->segmap_lock);
2167 static int build_curseg(struct f2fs_sb_info *sbi)
2169 struct curseg_info *array;
2172 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2176 SM_I(sbi)->curseg_array = array;
2178 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2179 mutex_init(&array[i].curseg_mutex);
2180 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2181 if (!array[i].sum_blk)
2183 init_rwsem(&array[i].journal_rwsem);
2184 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2186 if (!array[i].journal)
2188 array[i].segno = NULL_SEGNO;
2189 array[i].next_blkoff = 0;
2191 return restore_curseg_summaries(sbi);
2194 static void build_sit_entries(struct f2fs_sb_info *sbi)
2196 struct sit_info *sit_i = SIT_I(sbi);
2197 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2198 struct f2fs_journal *journal = curseg->journal;
2199 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2200 unsigned int i, start, end;
2201 unsigned int readed, start_blk = 0;
2202 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2205 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2207 start = start_blk * sit_i->sents_per_block;
2208 end = (start_blk + readed) * sit_i->sents_per_block;
2210 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2211 struct seg_entry *se = &sit_i->sentries[start];
2212 struct f2fs_sit_block *sit_blk;
2213 struct f2fs_sit_entry sit;
2216 down_read(&curseg->journal_rwsem);
2217 for (i = 0; i < sits_in_cursum(journal); i++) {
2218 if (le32_to_cpu(segno_in_journal(journal, i))
2220 sit = sit_in_journal(journal, i);
2221 up_read(&curseg->journal_rwsem);
2225 up_read(&curseg->journal_rwsem);
2227 page = get_current_sit_page(sbi, start);
2228 sit_blk = (struct f2fs_sit_block *)page_address(page);
2229 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2230 f2fs_put_page(page, 1);
2232 check_block_count(sbi, start, &sit);
2233 seg_info_from_raw_sit(se, &sit);
2235 /* build discard map only one time */
2236 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2237 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2239 if (sbi->segs_per_sec > 1) {
2240 struct sec_entry *e = get_sec_entry(sbi, start);
2241 e->valid_blocks += se->valid_blocks;
2244 start_blk += readed;
2245 } while (start_blk < sit_blk_cnt);
2248 static void init_free_segmap(struct f2fs_sb_info *sbi)
2253 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2254 struct seg_entry *sentry = get_seg_entry(sbi, start);
2255 if (!sentry->valid_blocks)
2256 __set_free(sbi, start);
2259 /* set use the current segments */
2260 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2261 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2262 __set_test_and_inuse(sbi, curseg_t->segno);
2266 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2268 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2269 struct free_segmap_info *free_i = FREE_I(sbi);
2270 unsigned int segno = 0, offset = 0;
2271 unsigned short valid_blocks;
2274 /* find dirty segment based on free segmap */
2275 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2276 if (segno >= MAIN_SEGS(sbi))
2279 valid_blocks = get_valid_blocks(sbi, segno, 0);
2280 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2282 if (valid_blocks > sbi->blocks_per_seg) {
2283 f2fs_bug_on(sbi, 1);
2286 mutex_lock(&dirty_i->seglist_lock);
2287 __locate_dirty_segment(sbi, segno, DIRTY);
2288 mutex_unlock(&dirty_i->seglist_lock);
2292 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2294 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2295 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2297 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2298 if (!dirty_i->victim_secmap)
2303 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2305 struct dirty_seglist_info *dirty_i;
2306 unsigned int bitmap_size, i;
2308 /* allocate memory for dirty segments list information */
2309 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2313 SM_I(sbi)->dirty_info = dirty_i;
2314 mutex_init(&dirty_i->seglist_lock);
2316 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2318 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2319 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2320 if (!dirty_i->dirty_segmap[i])
2324 init_dirty_segmap(sbi);
2325 return init_victim_secmap(sbi);
2329 * Update min, max modified time for cost-benefit GC algorithm
2331 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2333 struct sit_info *sit_i = SIT_I(sbi);
2336 mutex_lock(&sit_i->sentry_lock);
2338 sit_i->min_mtime = LLONG_MAX;
2340 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2342 unsigned long long mtime = 0;
2344 for (i = 0; i < sbi->segs_per_sec; i++)
2345 mtime += get_seg_entry(sbi, segno + i)->mtime;
2347 mtime = div_u64(mtime, sbi->segs_per_sec);
2349 if (sit_i->min_mtime > mtime)
2350 sit_i->min_mtime = mtime;
2352 sit_i->max_mtime = get_mtime(sbi);
2353 mutex_unlock(&sit_i->sentry_lock);
2356 int build_segment_manager(struct f2fs_sb_info *sbi)
2358 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2359 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2360 struct f2fs_sm_info *sm_info;
2363 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2368 sbi->sm_info = sm_info;
2369 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2370 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2371 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2372 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2373 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2374 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2375 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2376 sm_info->rec_prefree_segments = sm_info->main_segments *
2377 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2378 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2379 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2380 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2382 INIT_LIST_HEAD(&sm_info->discard_list);
2383 sm_info->nr_discards = 0;
2384 sm_info->max_discards = 0;
2386 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2388 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2390 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2391 err = create_flush_cmd_control(sbi);
2396 err = build_sit_info(sbi);
2399 err = build_free_segmap(sbi);
2402 err = build_curseg(sbi);
2406 /* reinit free segmap based on SIT */
2407 build_sit_entries(sbi);
2409 init_free_segmap(sbi);
2410 err = build_dirty_segmap(sbi);
2414 init_min_max_mtime(sbi);
2418 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2419 enum dirty_type dirty_type)
2421 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2423 mutex_lock(&dirty_i->seglist_lock);
2424 kvfree(dirty_i->dirty_segmap[dirty_type]);
2425 dirty_i->nr_dirty[dirty_type] = 0;
2426 mutex_unlock(&dirty_i->seglist_lock);
2429 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2431 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2432 kvfree(dirty_i->victim_secmap);
2435 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2437 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2443 /* discard pre-free/dirty segments list */
2444 for (i = 0; i < NR_DIRTY_TYPE; i++)
2445 discard_dirty_segmap(sbi, i);
2447 destroy_victim_secmap(sbi);
2448 SM_I(sbi)->dirty_info = NULL;
2452 static void destroy_curseg(struct f2fs_sb_info *sbi)
2454 struct curseg_info *array = SM_I(sbi)->curseg_array;
2459 SM_I(sbi)->curseg_array = NULL;
2460 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2461 kfree(array[i].sum_blk);
2462 kfree(array[i].journal);
2467 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2469 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2472 SM_I(sbi)->free_info = NULL;
2473 kvfree(free_i->free_segmap);
2474 kvfree(free_i->free_secmap);
2478 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2480 struct sit_info *sit_i = SIT_I(sbi);
2486 if (sit_i->sentries) {
2487 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2488 kfree(sit_i->sentries[start].cur_valid_map);
2489 kfree(sit_i->sentries[start].ckpt_valid_map);
2490 kfree(sit_i->sentries[start].discard_map);
2493 kfree(sit_i->tmp_map);
2495 kvfree(sit_i->sentries);
2496 kvfree(sit_i->sec_entries);
2497 kvfree(sit_i->dirty_sentries_bitmap);
2499 SM_I(sbi)->sit_info = NULL;
2500 kfree(sit_i->sit_bitmap);
2504 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2506 struct f2fs_sm_info *sm_info = SM_I(sbi);
2510 destroy_flush_cmd_control(sbi);
2511 destroy_dirty_segmap(sbi);
2512 destroy_curseg(sbi);
2513 destroy_free_segmap(sbi);
2514 destroy_sit_info(sbi);
2515 sbi->sm_info = NULL;
2519 int __init create_segment_manager_caches(void)
2521 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2522 sizeof(struct discard_entry));
2523 if (!discard_entry_slab)
2526 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2527 sizeof(struct sit_entry_set));
2528 if (!sit_entry_set_slab)
2529 goto destory_discard_entry;
2531 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2532 sizeof(struct inmem_pages));
2533 if (!inmem_entry_slab)
2534 goto destroy_sit_entry_set;
2537 destroy_sit_entry_set:
2538 kmem_cache_destroy(sit_entry_set_slab);
2539 destory_discard_entry:
2540 kmem_cache_destroy(discard_entry_slab);
2545 void destroy_segment_manager_caches(void)
2547 kmem_cache_destroy(sit_entry_set_slab);
2548 kmem_cache_destroy(discard_entry_slab);
2549 kmem_cache_destroy(inmem_entry_slab);