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 *bio_entry_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
33 static unsigned long __reverse_ulong(unsigned char *str)
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
38 #if BITS_PER_LONG == 64
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
52 static inline unsigned long __reverse_ffs(unsigned long word)
56 #if BITS_PER_LONG == 64
57 if ((word & 0xffffffff00000000UL) == 0)
62 if ((word & 0xffff0000) == 0)
67 if ((word & 0xff00) == 0)
72 if ((word & 0xf0) == 0)
77 if ((word & 0xc) == 0)
82 if ((word & 0x2) == 0)
88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
90 * @size must be integral times of unsigned long.
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
99 const unsigned long *p = addr + BIT_WORD(offset);
100 unsigned long result = size;
106 size -= (offset & ~(BITS_PER_LONG - 1));
107 offset %= BITS_PER_LONG;
113 tmp = __reverse_ulong((unsigned char *)p);
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
121 if (size <= BITS_PER_LONG)
123 size -= BITS_PER_LONG;
129 return result - size + __reverse_ffs(tmp);
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
135 const unsigned long *p = addr + BIT_WORD(offset);
136 unsigned long result = size;
142 size -= (offset & ~(BITS_PER_LONG - 1));
143 offset %= BITS_PER_LONG;
149 tmp = __reverse_ulong((unsigned char *)p);
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
158 if (size <= BITS_PER_LONG)
160 size -= BITS_PER_LONG;
166 return result - size + __reverse_ffz(tmp);
169 void register_inmem_page(struct inode *inode, struct page *page)
171 struct f2fs_inode_info *fi = F2FS_I(inode);
172 struct inmem_pages *new;
174 f2fs_trace_pid(page);
176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 SetPagePrivate(page);
179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
181 /* add atomic page indices to the list */
183 INIT_LIST_HEAD(&new->list);
185 /* increase reference count with clean state */
186 mutex_lock(&fi->inmem_lock);
188 list_add_tail(&new->list, &fi->inmem_pages);
189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
190 mutex_unlock(&fi->inmem_lock);
192 trace_f2fs_register_inmem_page(page, INMEM);
195 static int __revoke_inmem_pages(struct inode *inode,
196 struct list_head *head, bool drop, bool recover)
198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
199 struct inmem_pages *cur, *tmp;
202 list_for_each_entry_safe(cur, tmp, head, list) {
203 struct page *page = cur->page;
206 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
211 struct dnode_of_data dn;
214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
216 set_new_dnode(&dn, inode, NULL, NULL, 0);
217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
221 get_node_info(sbi, dn.nid, &ni);
222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 cur->old_addr, ni.version, true, true);
227 /* we don't need to invalidate this in the sccessful status */
229 ClearPageUptodate(page);
230 set_page_private(page, 0);
231 ClearPagePrivate(page);
232 f2fs_put_page(page, 1);
234 list_del(&cur->list);
235 kmem_cache_free(inmem_entry_slab, cur);
236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
241 void drop_inmem_pages(struct inode *inode)
243 struct f2fs_inode_info *fi = F2FS_I(inode);
245 clear_inode_flag(inode, FI_ATOMIC_FILE);
247 mutex_lock(&fi->inmem_lock);
248 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
249 mutex_unlock(&fi->inmem_lock);
252 static int __commit_inmem_pages(struct inode *inode,
253 struct list_head *revoke_list)
255 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
256 struct f2fs_inode_info *fi = F2FS_I(inode);
257 struct inmem_pages *cur, *tmp;
258 struct f2fs_io_info fio = {
262 .op_flags = WRITE_SYNC | REQ_PRIO,
263 .encrypted_page = NULL,
265 bool submit_bio = false;
268 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
269 struct page *page = cur->page;
272 if (page->mapping == inode->i_mapping) {
273 trace_f2fs_commit_inmem_page(page, INMEM);
275 set_page_dirty(page);
276 f2fs_wait_on_page_writeback(page, DATA, true);
277 if (clear_page_dirty_for_io(page))
278 inode_dec_dirty_pages(inode);
281 err = do_write_data_page(&fio);
287 /* record old blkaddr for revoking */
288 cur->old_addr = fio.old_blkaddr;
290 clear_cold_data(page);
294 list_move_tail(&cur->list, revoke_list);
298 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
301 __revoke_inmem_pages(inode, revoke_list, false, false);
306 int commit_inmem_pages(struct inode *inode)
308 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
309 struct f2fs_inode_info *fi = F2FS_I(inode);
310 struct list_head revoke_list;
313 INIT_LIST_HEAD(&revoke_list);
314 f2fs_balance_fs(sbi, true);
317 mutex_lock(&fi->inmem_lock);
318 err = __commit_inmem_pages(inode, &revoke_list);
322 * try to revoke all committed pages, but still we could fail
323 * due to no memory or other reason, if that happened, EAGAIN
324 * will be returned, which means in such case, transaction is
325 * already not integrity, caller should use journal to do the
326 * recovery or rewrite & commit last transaction. For other
327 * error number, revoking was done by filesystem itself.
329 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
333 /* drop all uncommitted pages */
334 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
336 mutex_unlock(&fi->inmem_lock);
343 * This function balances dirty node and dentry pages.
344 * In addition, it controls garbage collection.
346 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351 /* balance_fs_bg is able to be pending */
352 if (excess_cached_nats(sbi))
353 f2fs_balance_fs_bg(sbi);
356 * We should do GC or end up with checkpoint, if there are so many dirty
357 * dir/node pages without enough free segments.
359 if (has_not_enough_free_secs(sbi, 0, 0)) {
360 mutex_lock(&sbi->gc_mutex);
365 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
367 /* try to shrink extent cache when there is no enough memory */
368 if (!available_free_memory(sbi, EXTENT_CACHE))
369 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
371 /* check the # of cached NAT entries */
372 if (!available_free_memory(sbi, NAT_ENTRIES))
373 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
375 if (!available_free_memory(sbi, FREE_NIDS))
376 try_to_free_nids(sbi, MAX_FREE_NIDS);
378 build_free_nids(sbi);
380 /* checkpoint is the only way to shrink partial cached entries */
381 if (!available_free_memory(sbi, NAT_ENTRIES) ||
382 !available_free_memory(sbi, INO_ENTRIES) ||
383 excess_prefree_segs(sbi) ||
384 excess_dirty_nats(sbi) ||
385 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
386 if (test_opt(sbi, DATA_FLUSH)) {
387 struct blk_plug plug;
389 blk_start_plug(&plug);
390 sync_dirty_inodes(sbi, FILE_INODE);
391 blk_finish_plug(&plug);
393 f2fs_sync_fs(sbi->sb, true);
394 stat_inc_bg_cp_count(sbi->stat_info);
398 static int issue_flush_thread(void *data)
400 struct f2fs_sb_info *sbi = data;
401 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
402 wait_queue_head_t *q = &fcc->flush_wait_queue;
404 if (kthread_should_stop())
407 if (!llist_empty(&fcc->issue_list)) {
409 struct flush_cmd *cmd, *next;
412 bio = f2fs_bio_alloc(0);
414 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
415 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
417 bio->bi_bdev = sbi->sb->s_bdev;
418 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
419 ret = submit_bio_wait(bio);
421 llist_for_each_entry_safe(cmd, next,
422 fcc->dispatch_list, llnode) {
424 complete(&cmd->wait);
427 fcc->dispatch_list = NULL;
430 wait_event_interruptible(*q,
431 kthread_should_stop() || !llist_empty(&fcc->issue_list));
435 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
437 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
438 struct flush_cmd cmd;
440 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
441 test_opt(sbi, FLUSH_MERGE));
443 if (test_opt(sbi, NOBARRIER))
446 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
447 struct bio *bio = f2fs_bio_alloc(0);
450 atomic_inc(&fcc->submit_flush);
451 bio->bi_bdev = sbi->sb->s_bdev;
452 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
453 ret = submit_bio_wait(bio);
454 atomic_dec(&fcc->submit_flush);
459 init_completion(&cmd.wait);
461 atomic_inc(&fcc->submit_flush);
462 llist_add(&cmd.llnode, &fcc->issue_list);
464 if (!fcc->dispatch_list)
465 wake_up(&fcc->flush_wait_queue);
467 wait_for_completion(&cmd.wait);
468 atomic_dec(&fcc->submit_flush);
473 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
475 dev_t dev = sbi->sb->s_bdev->bd_dev;
476 struct flush_cmd_control *fcc;
479 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
482 atomic_set(&fcc->submit_flush, 0);
483 init_waitqueue_head(&fcc->flush_wait_queue);
484 init_llist_head(&fcc->issue_list);
485 SM_I(sbi)->cmd_control_info = fcc;
486 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
487 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
488 if (IS_ERR(fcc->f2fs_issue_flush)) {
489 err = PTR_ERR(fcc->f2fs_issue_flush);
491 SM_I(sbi)->cmd_control_info = NULL;
498 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
500 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
502 if (fcc && fcc->f2fs_issue_flush)
503 kthread_stop(fcc->f2fs_issue_flush);
505 SM_I(sbi)->cmd_control_info = NULL;
508 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
509 enum dirty_type dirty_type)
511 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
513 /* need not be added */
514 if (IS_CURSEG(sbi, segno))
517 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
518 dirty_i->nr_dirty[dirty_type]++;
520 if (dirty_type == DIRTY) {
521 struct seg_entry *sentry = get_seg_entry(sbi, segno);
522 enum dirty_type t = sentry->type;
524 if (unlikely(t >= DIRTY)) {
528 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
529 dirty_i->nr_dirty[t]++;
533 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
534 enum dirty_type dirty_type)
536 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
538 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
539 dirty_i->nr_dirty[dirty_type]--;
541 if (dirty_type == DIRTY) {
542 struct seg_entry *sentry = get_seg_entry(sbi, segno);
543 enum dirty_type t = sentry->type;
545 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
546 dirty_i->nr_dirty[t]--;
548 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
549 clear_bit(GET_SECNO(sbi, segno),
550 dirty_i->victim_secmap);
555 * Should not occur error such as -ENOMEM.
556 * Adding dirty entry into seglist is not critical operation.
557 * If a given segment is one of current working segments, it won't be added.
559 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
561 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
562 unsigned short valid_blocks;
564 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
567 mutex_lock(&dirty_i->seglist_lock);
569 valid_blocks = get_valid_blocks(sbi, segno, 0);
571 if (valid_blocks == 0) {
572 __locate_dirty_segment(sbi, segno, PRE);
573 __remove_dirty_segment(sbi, segno, DIRTY);
574 } else if (valid_blocks < sbi->blocks_per_seg) {
575 __locate_dirty_segment(sbi, segno, DIRTY);
577 /* Recovery routine with SSR needs this */
578 __remove_dirty_segment(sbi, segno, DIRTY);
581 mutex_unlock(&dirty_i->seglist_lock);
584 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
587 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
588 struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
590 INIT_LIST_HEAD(&be->list);
592 init_completion(&be->event);
593 list_add_tail(&be->list, wait_list);
598 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
600 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
601 struct bio_entry *be, *tmp;
603 list_for_each_entry_safe(be, tmp, wait_list, list) {
604 struct bio *bio = be->bio;
607 wait_for_completion_io(&be->event);
609 if (err == -EOPNOTSUPP)
613 f2fs_msg(sbi->sb, KERN_INFO,
614 "Issue discard failed, ret: %d", err);
618 kmem_cache_free(bio_entry_slab, be);
622 static void f2fs_submit_bio_wait_endio(struct bio *bio)
624 struct bio_entry *be = (struct bio_entry *)bio->bi_private;
626 be->error = bio->bi_error;
627 complete(&be->event);
630 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
631 int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi, sector_t sector,
632 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
634 struct block_device *bdev = sbi->sb->s_bdev;
635 struct bio *bio = NULL;
638 err = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
641 struct bio_entry *be = __add_bio_entry(sbi, bio);
643 bio->bi_private = be;
644 bio->bi_end_io = f2fs_submit_bio_wait_endio;
645 bio->bi_opf |= REQ_SYNC;
652 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
653 block_t blkstart, block_t blklen)
655 sector_t start = SECTOR_FROM_BLOCK(blkstart);
656 sector_t len = SECTOR_FROM_BLOCK(blklen);
657 struct seg_entry *se;
661 for (i = blkstart; i < blkstart + blklen; i++) {
662 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
663 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
665 if (!f2fs_test_and_set_bit(offset, se->discard_map))
668 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
669 return __f2fs_issue_discard_async(sbi, start, len, GFP_NOFS, 0);
672 static void __add_discard_entry(struct f2fs_sb_info *sbi,
673 struct cp_control *cpc, struct seg_entry *se,
674 unsigned int start, unsigned int end)
676 struct list_head *head = &SM_I(sbi)->discard_list;
677 struct discard_entry *new, *last;
679 if (!list_empty(head)) {
680 last = list_last_entry(head, struct discard_entry, list);
681 if (START_BLOCK(sbi, cpc->trim_start) + start ==
682 last->blkaddr + last->len) {
683 last->len += end - start;
688 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
689 INIT_LIST_HEAD(&new->list);
690 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
691 new->len = end - start;
692 list_add_tail(&new->list, head);
694 SM_I(sbi)->nr_discards += end - start;
697 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
699 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
700 int max_blocks = sbi->blocks_per_seg;
701 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
702 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
703 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
704 unsigned long *discard_map = (unsigned long *)se->discard_map;
705 unsigned long *dmap = SIT_I(sbi)->tmp_map;
706 unsigned int start = 0, end = -1;
707 bool force = (cpc->reason == CP_DISCARD);
710 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
714 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
715 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
719 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
720 for (i = 0; i < entries; i++)
721 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
722 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
724 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
725 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
726 if (start >= max_blocks)
729 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
730 if (force && start && end != max_blocks
731 && (end - start) < cpc->trim_minlen)
734 __add_discard_entry(sbi, cpc, se, start, end);
738 void release_discard_addrs(struct f2fs_sb_info *sbi)
740 struct list_head *head = &(SM_I(sbi)->discard_list);
741 struct discard_entry *entry, *this;
744 list_for_each_entry_safe(entry, this, head, list) {
745 list_del(&entry->list);
746 kmem_cache_free(discard_entry_slab, entry);
751 * Should call clear_prefree_segments after checkpoint is done.
753 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
755 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
758 mutex_lock(&dirty_i->seglist_lock);
759 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
760 __set_test_and_free(sbi, segno);
761 mutex_unlock(&dirty_i->seglist_lock);
764 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
766 struct list_head *head = &(SM_I(sbi)->discard_list);
767 struct discard_entry *entry, *this;
768 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
769 struct blk_plug plug;
770 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
771 unsigned int start = 0, end = -1;
772 unsigned int secno, start_segno;
773 bool force = (cpc->reason == CP_DISCARD);
775 blk_start_plug(&plug);
777 mutex_lock(&dirty_i->seglist_lock);
781 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
782 if (start >= MAIN_SEGS(sbi))
784 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
787 for (i = start; i < end; i++)
788 clear_bit(i, prefree_map);
790 dirty_i->nr_dirty[PRE] -= end - start;
792 if (force || !test_opt(sbi, DISCARD))
795 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
796 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
797 (end - start) << sbi->log_blocks_per_seg);
801 secno = GET_SECNO(sbi, start);
802 start_segno = secno * sbi->segs_per_sec;
803 if (!IS_CURSEC(sbi, secno) &&
804 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
805 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
806 sbi->segs_per_sec << sbi->log_blocks_per_seg);
808 start = start_segno + sbi->segs_per_sec;
812 mutex_unlock(&dirty_i->seglist_lock);
814 /* send small discards */
815 list_for_each_entry_safe(entry, this, head, list) {
816 if (force && entry->len < cpc->trim_minlen)
818 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
819 cpc->trimmed += entry->len;
821 list_del(&entry->list);
822 SM_I(sbi)->nr_discards -= entry->len;
823 kmem_cache_free(discard_entry_slab, entry);
826 blk_finish_plug(&plug);
829 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
831 struct sit_info *sit_i = SIT_I(sbi);
833 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
834 sit_i->dirty_sentries++;
841 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
842 unsigned int segno, int modified)
844 struct seg_entry *se = get_seg_entry(sbi, segno);
847 __mark_sit_entry_dirty(sbi, segno);
850 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
852 struct seg_entry *se;
853 unsigned int segno, offset;
854 long int new_vblocks;
856 segno = GET_SEGNO(sbi, blkaddr);
858 se = get_seg_entry(sbi, segno);
859 new_vblocks = se->valid_blocks + del;
860 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
862 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
863 (new_vblocks > sbi->blocks_per_seg)));
865 se->valid_blocks = new_vblocks;
866 se->mtime = get_mtime(sbi);
867 SIT_I(sbi)->max_mtime = se->mtime;
869 /* Update valid block bitmap */
871 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
873 if (f2fs_discard_en(sbi) &&
874 !f2fs_test_and_set_bit(offset, se->discard_map))
877 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
879 if (f2fs_discard_en(sbi) &&
880 f2fs_test_and_clear_bit(offset, se->discard_map))
883 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
884 se->ckpt_valid_blocks += del;
886 __mark_sit_entry_dirty(sbi, segno);
888 /* update total number of valid blocks to be written in ckpt area */
889 SIT_I(sbi)->written_valid_blocks += del;
891 if (sbi->segs_per_sec > 1)
892 get_sec_entry(sbi, segno)->valid_blocks += del;
895 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
897 update_sit_entry(sbi, new, 1);
898 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
899 update_sit_entry(sbi, old, -1);
901 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
902 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
905 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
907 unsigned int segno = GET_SEGNO(sbi, addr);
908 struct sit_info *sit_i = SIT_I(sbi);
910 f2fs_bug_on(sbi, addr == NULL_ADDR);
911 if (addr == NEW_ADDR)
914 /* add it into sit main buffer */
915 mutex_lock(&sit_i->sentry_lock);
917 update_sit_entry(sbi, addr, -1);
919 /* add it into dirty seglist */
920 locate_dirty_segment(sbi, segno);
922 mutex_unlock(&sit_i->sentry_lock);
925 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
927 struct sit_info *sit_i = SIT_I(sbi);
928 unsigned int segno, offset;
929 struct seg_entry *se;
932 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
935 mutex_lock(&sit_i->sentry_lock);
937 segno = GET_SEGNO(sbi, blkaddr);
938 se = get_seg_entry(sbi, segno);
939 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
941 if (f2fs_test_bit(offset, se->ckpt_valid_map))
944 mutex_unlock(&sit_i->sentry_lock);
950 * This function should be resided under the curseg_mutex lock
952 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
953 struct f2fs_summary *sum)
955 struct curseg_info *curseg = CURSEG_I(sbi, type);
956 void *addr = curseg->sum_blk;
957 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
958 memcpy(addr, sum, sizeof(struct f2fs_summary));
962 * Calculate the number of current summary pages for writing
964 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
966 int valid_sum_count = 0;
969 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
970 if (sbi->ckpt->alloc_type[i] == SSR)
971 valid_sum_count += sbi->blocks_per_seg;
974 valid_sum_count += le16_to_cpu(
975 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
977 valid_sum_count += curseg_blkoff(sbi, i);
981 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
982 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
983 if (valid_sum_count <= sum_in_page)
985 else if ((valid_sum_count - sum_in_page) <=
986 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
992 * Caller should put this summary page
994 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
996 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
999 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1001 struct page *page = grab_meta_page(sbi, blk_addr);
1002 void *dst = page_address(page);
1005 memcpy(dst, src, PAGE_SIZE);
1007 memset(dst, 0, PAGE_SIZE);
1008 set_page_dirty(page);
1009 f2fs_put_page(page, 1);
1012 static void write_sum_page(struct f2fs_sb_info *sbi,
1013 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1015 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1018 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1019 int type, block_t blk_addr)
1021 struct curseg_info *curseg = CURSEG_I(sbi, type);
1022 struct page *page = grab_meta_page(sbi, blk_addr);
1023 struct f2fs_summary_block *src = curseg->sum_blk;
1024 struct f2fs_summary_block *dst;
1026 dst = (struct f2fs_summary_block *)page_address(page);
1028 mutex_lock(&curseg->curseg_mutex);
1030 down_read(&curseg->journal_rwsem);
1031 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1032 up_read(&curseg->journal_rwsem);
1034 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1035 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1037 mutex_unlock(&curseg->curseg_mutex);
1039 set_page_dirty(page);
1040 f2fs_put_page(page, 1);
1043 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1045 struct curseg_info *curseg = CURSEG_I(sbi, type);
1046 unsigned int segno = curseg->segno + 1;
1047 struct free_segmap_info *free_i = FREE_I(sbi);
1049 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1050 return !test_bit(segno, free_i->free_segmap);
1055 * Find a new segment from the free segments bitmap to right order
1056 * This function should be returned with success, otherwise BUG
1058 static void get_new_segment(struct f2fs_sb_info *sbi,
1059 unsigned int *newseg, bool new_sec, int dir)
1061 struct free_segmap_info *free_i = FREE_I(sbi);
1062 unsigned int segno, secno, zoneno;
1063 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1064 unsigned int hint = *newseg / sbi->segs_per_sec;
1065 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1066 unsigned int left_start = hint;
1071 spin_lock(&free_i->segmap_lock);
1073 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1074 segno = find_next_zero_bit(free_i->free_segmap,
1075 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1076 if (segno < (hint + 1) * sbi->segs_per_sec)
1080 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1081 if (secno >= MAIN_SECS(sbi)) {
1082 if (dir == ALLOC_RIGHT) {
1083 secno = find_next_zero_bit(free_i->free_secmap,
1085 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1088 left_start = hint - 1;
1094 while (test_bit(left_start, free_i->free_secmap)) {
1095 if (left_start > 0) {
1099 left_start = find_next_zero_bit(free_i->free_secmap,
1101 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1107 segno = secno * sbi->segs_per_sec;
1108 zoneno = secno / sbi->secs_per_zone;
1110 /* give up on finding another zone */
1113 if (sbi->secs_per_zone == 1)
1115 if (zoneno == old_zoneno)
1117 if (dir == ALLOC_LEFT) {
1118 if (!go_left && zoneno + 1 >= total_zones)
1120 if (go_left && zoneno == 0)
1123 for (i = 0; i < NR_CURSEG_TYPE; i++)
1124 if (CURSEG_I(sbi, i)->zone == zoneno)
1127 if (i < NR_CURSEG_TYPE) {
1128 /* zone is in user, try another */
1130 hint = zoneno * sbi->secs_per_zone - 1;
1131 else if (zoneno + 1 >= total_zones)
1134 hint = (zoneno + 1) * sbi->secs_per_zone;
1136 goto find_other_zone;
1139 /* set it as dirty segment in free segmap */
1140 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1141 __set_inuse(sbi, segno);
1143 spin_unlock(&free_i->segmap_lock);
1146 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1148 struct curseg_info *curseg = CURSEG_I(sbi, type);
1149 struct summary_footer *sum_footer;
1151 curseg->segno = curseg->next_segno;
1152 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1153 curseg->next_blkoff = 0;
1154 curseg->next_segno = NULL_SEGNO;
1156 sum_footer = &(curseg->sum_blk->footer);
1157 memset(sum_footer, 0, sizeof(struct summary_footer));
1158 if (IS_DATASEG(type))
1159 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1160 if (IS_NODESEG(type))
1161 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1162 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1166 * Allocate a current working segment.
1167 * This function always allocates a free segment in LFS manner.
1169 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1171 struct curseg_info *curseg = CURSEG_I(sbi, type);
1172 unsigned int segno = curseg->segno;
1173 int dir = ALLOC_LEFT;
1175 write_sum_page(sbi, curseg->sum_blk,
1176 GET_SUM_BLOCK(sbi, segno));
1177 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1180 if (test_opt(sbi, NOHEAP))
1183 get_new_segment(sbi, &segno, new_sec, dir);
1184 curseg->next_segno = segno;
1185 reset_curseg(sbi, type, 1);
1186 curseg->alloc_type = LFS;
1189 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1190 struct curseg_info *seg, block_t start)
1192 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1193 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1194 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1195 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1196 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1199 for (i = 0; i < entries; i++)
1200 target_map[i] = ckpt_map[i] | cur_map[i];
1202 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1204 seg->next_blkoff = pos;
1208 * If a segment is written by LFS manner, next block offset is just obtained
1209 * by increasing the current block offset. However, if a segment is written by
1210 * SSR manner, next block offset obtained by calling __next_free_blkoff
1212 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1213 struct curseg_info *seg)
1215 if (seg->alloc_type == SSR)
1216 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1222 * This function always allocates a used segment(from dirty seglist) by SSR
1223 * manner, so it should recover the existing segment information of valid blocks
1225 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1227 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1228 struct curseg_info *curseg = CURSEG_I(sbi, type);
1229 unsigned int new_segno = curseg->next_segno;
1230 struct f2fs_summary_block *sum_node;
1231 struct page *sum_page;
1233 write_sum_page(sbi, curseg->sum_blk,
1234 GET_SUM_BLOCK(sbi, curseg->segno));
1235 __set_test_and_inuse(sbi, new_segno);
1237 mutex_lock(&dirty_i->seglist_lock);
1238 __remove_dirty_segment(sbi, new_segno, PRE);
1239 __remove_dirty_segment(sbi, new_segno, DIRTY);
1240 mutex_unlock(&dirty_i->seglist_lock);
1242 reset_curseg(sbi, type, 1);
1243 curseg->alloc_type = SSR;
1244 __next_free_blkoff(sbi, curseg, 0);
1247 sum_page = get_sum_page(sbi, new_segno);
1248 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1249 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1250 f2fs_put_page(sum_page, 1);
1254 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1256 struct curseg_info *curseg = CURSEG_I(sbi, type);
1257 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1259 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1260 return v_ops->get_victim(sbi,
1261 &(curseg)->next_segno, BG_GC, type, SSR);
1263 /* For data segments, let's do SSR more intensively */
1264 for (; type >= CURSEG_HOT_DATA; type--)
1265 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1272 * flush out current segment and replace it with new segment
1273 * This function should be returned with success, otherwise BUG
1275 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1276 int type, bool force)
1278 struct curseg_info *curseg = CURSEG_I(sbi, type);
1281 new_curseg(sbi, type, true);
1282 else if (type == CURSEG_WARM_NODE)
1283 new_curseg(sbi, type, false);
1284 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1285 new_curseg(sbi, type, false);
1286 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1287 change_curseg(sbi, type, true);
1289 new_curseg(sbi, type, false);
1291 stat_inc_seg_type(sbi, curseg);
1294 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1296 struct curseg_info *curseg = CURSEG_I(sbi, type);
1297 unsigned int old_segno;
1299 old_segno = curseg->segno;
1300 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1301 locate_dirty_segment(sbi, old_segno);
1304 void allocate_new_segments(struct f2fs_sb_info *sbi)
1308 if (test_opt(sbi, LFS))
1311 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1312 __allocate_new_segments(sbi, i);
1315 static const struct segment_allocation default_salloc_ops = {
1316 .allocate_segment = allocate_segment_by_default,
1319 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1321 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1322 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1323 unsigned int start_segno, end_segno;
1324 struct cp_control cpc;
1327 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1331 if (end <= MAIN_BLKADDR(sbi))
1334 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1335 f2fs_msg(sbi->sb, KERN_WARNING,
1336 "Found FS corruption, run fsck to fix.");
1340 /* start/end segment number in main_area */
1341 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1342 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1343 GET_SEGNO(sbi, end);
1344 cpc.reason = CP_DISCARD;
1345 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1347 /* do checkpoint to issue discard commands safely */
1348 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1349 cpc.trim_start = start_segno;
1351 if (sbi->discard_blks == 0)
1353 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1354 cpc.trim_end = end_segno;
1356 cpc.trim_end = min_t(unsigned int,
1357 rounddown(start_segno +
1358 BATCHED_TRIM_SEGMENTS(sbi),
1359 sbi->segs_per_sec) - 1, end_segno);
1361 mutex_lock(&sbi->gc_mutex);
1362 err = write_checkpoint(sbi, &cpc);
1363 mutex_unlock(&sbi->gc_mutex);
1370 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1374 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1376 struct curseg_info *curseg = CURSEG_I(sbi, type);
1377 if (curseg->next_blkoff < sbi->blocks_per_seg)
1382 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1385 return CURSEG_HOT_DATA;
1387 return CURSEG_HOT_NODE;
1390 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1392 if (p_type == DATA) {
1393 struct inode *inode = page->mapping->host;
1395 if (S_ISDIR(inode->i_mode))
1396 return CURSEG_HOT_DATA;
1398 return CURSEG_COLD_DATA;
1400 if (IS_DNODE(page) && is_cold_node(page))
1401 return CURSEG_WARM_NODE;
1403 return CURSEG_COLD_NODE;
1407 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1409 if (p_type == DATA) {
1410 struct inode *inode = page->mapping->host;
1412 if (S_ISDIR(inode->i_mode))
1413 return CURSEG_HOT_DATA;
1414 else if (is_cold_data(page) || file_is_cold(inode))
1415 return CURSEG_COLD_DATA;
1417 return CURSEG_WARM_DATA;
1420 return is_cold_node(page) ? CURSEG_WARM_NODE :
1423 return CURSEG_COLD_NODE;
1427 static int __get_segment_type(struct page *page, enum page_type p_type)
1429 switch (F2FS_P_SB(page)->active_logs) {
1431 return __get_segment_type_2(page, p_type);
1433 return __get_segment_type_4(page, p_type);
1435 /* NR_CURSEG_TYPE(6) logs by default */
1436 f2fs_bug_on(F2FS_P_SB(page),
1437 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1438 return __get_segment_type_6(page, p_type);
1441 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1442 block_t old_blkaddr, block_t *new_blkaddr,
1443 struct f2fs_summary *sum, int type)
1445 struct sit_info *sit_i = SIT_I(sbi);
1446 struct curseg_info *curseg;
1447 bool direct_io = (type == CURSEG_DIRECT_IO);
1449 type = direct_io ? CURSEG_WARM_DATA : type;
1451 curseg = CURSEG_I(sbi, type);
1453 mutex_lock(&curseg->curseg_mutex);
1454 mutex_lock(&sit_i->sentry_lock);
1456 /* direct_io'ed data is aligned to the segment for better performance */
1457 if (direct_io && curseg->next_blkoff &&
1458 !has_not_enough_free_secs(sbi, 0, 0))
1459 __allocate_new_segments(sbi, type);
1461 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1464 * __add_sum_entry should be resided under the curseg_mutex
1465 * because, this function updates a summary entry in the
1466 * current summary block.
1468 __add_sum_entry(sbi, type, sum);
1470 __refresh_next_blkoff(sbi, curseg);
1472 stat_inc_block_count(sbi, curseg);
1474 if (!__has_curseg_space(sbi, type))
1475 sit_i->s_ops->allocate_segment(sbi, type, false);
1477 * SIT information should be updated before segment allocation,
1478 * since SSR needs latest valid block information.
1480 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1482 mutex_unlock(&sit_i->sentry_lock);
1484 if (page && IS_NODESEG(type))
1485 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1487 mutex_unlock(&curseg->curseg_mutex);
1490 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1492 int type = __get_segment_type(fio->page, fio->type);
1494 if (fio->type == NODE || fio->type == DATA)
1495 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1497 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1498 &fio->new_blkaddr, sum, type);
1500 /* writeout dirty page into bdev */
1501 f2fs_submit_page_mbio(fio);
1503 if (fio->type == NODE || fio->type == DATA)
1504 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1507 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1509 struct f2fs_io_info fio = {
1513 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
1514 .old_blkaddr = page->index,
1515 .new_blkaddr = page->index,
1517 .encrypted_page = NULL,
1520 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1521 fio.op_flags &= ~REQ_META;
1523 set_page_writeback(page);
1524 f2fs_submit_page_mbio(&fio);
1527 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1529 struct f2fs_summary sum;
1531 set_summary(&sum, nid, 0, 0);
1532 do_write_page(&sum, fio);
1535 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1537 struct f2fs_sb_info *sbi = fio->sbi;
1538 struct f2fs_summary sum;
1539 struct node_info ni;
1541 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1542 get_node_info(sbi, dn->nid, &ni);
1543 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1544 do_write_page(&sum, fio);
1545 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1548 void rewrite_data_page(struct f2fs_io_info *fio)
1550 fio->new_blkaddr = fio->old_blkaddr;
1551 stat_inc_inplace_blocks(fio->sbi);
1552 f2fs_submit_page_mbio(fio);
1555 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1556 block_t old_blkaddr, block_t new_blkaddr,
1557 bool recover_curseg, bool recover_newaddr)
1559 struct sit_info *sit_i = SIT_I(sbi);
1560 struct curseg_info *curseg;
1561 unsigned int segno, old_cursegno;
1562 struct seg_entry *se;
1564 unsigned short old_blkoff;
1566 segno = GET_SEGNO(sbi, new_blkaddr);
1567 se = get_seg_entry(sbi, segno);
1570 if (!recover_curseg) {
1571 /* for recovery flow */
1572 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1573 if (old_blkaddr == NULL_ADDR)
1574 type = CURSEG_COLD_DATA;
1576 type = CURSEG_WARM_DATA;
1579 if (!IS_CURSEG(sbi, segno))
1580 type = CURSEG_WARM_DATA;
1583 curseg = CURSEG_I(sbi, type);
1585 mutex_lock(&curseg->curseg_mutex);
1586 mutex_lock(&sit_i->sentry_lock);
1588 old_cursegno = curseg->segno;
1589 old_blkoff = curseg->next_blkoff;
1591 /* change the current segment */
1592 if (segno != curseg->segno) {
1593 curseg->next_segno = segno;
1594 change_curseg(sbi, type, true);
1597 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1598 __add_sum_entry(sbi, type, sum);
1600 if (!recover_curseg || recover_newaddr)
1601 update_sit_entry(sbi, new_blkaddr, 1);
1602 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1603 update_sit_entry(sbi, old_blkaddr, -1);
1605 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1606 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1608 locate_dirty_segment(sbi, old_cursegno);
1610 if (recover_curseg) {
1611 if (old_cursegno != curseg->segno) {
1612 curseg->next_segno = old_cursegno;
1613 change_curseg(sbi, type, true);
1615 curseg->next_blkoff = old_blkoff;
1618 mutex_unlock(&sit_i->sentry_lock);
1619 mutex_unlock(&curseg->curseg_mutex);
1622 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1623 block_t old_addr, block_t new_addr,
1624 unsigned char version, bool recover_curseg,
1625 bool recover_newaddr)
1627 struct f2fs_summary sum;
1629 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1631 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1632 recover_curseg, recover_newaddr);
1634 f2fs_update_data_blkaddr(dn, new_addr);
1637 void f2fs_wait_on_page_writeback(struct page *page,
1638 enum page_type type, bool ordered)
1640 if (PageWriteback(page)) {
1641 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1643 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1645 wait_on_page_writeback(page);
1647 wait_for_stable_page(page);
1651 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1656 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1659 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1661 f2fs_wait_on_page_writeback(cpage, DATA, true);
1662 f2fs_put_page(cpage, 1);
1666 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1668 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1669 struct curseg_info *seg_i;
1670 unsigned char *kaddr;
1675 start = start_sum_block(sbi);
1677 page = get_meta_page(sbi, start++);
1678 kaddr = (unsigned char *)page_address(page);
1680 /* Step 1: restore nat cache */
1681 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1682 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1684 /* Step 2: restore sit cache */
1685 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1686 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1687 offset = 2 * SUM_JOURNAL_SIZE;
1689 /* Step 3: restore summary entries */
1690 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1691 unsigned short blk_off;
1694 seg_i = CURSEG_I(sbi, i);
1695 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1696 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1697 seg_i->next_segno = segno;
1698 reset_curseg(sbi, i, 0);
1699 seg_i->alloc_type = ckpt->alloc_type[i];
1700 seg_i->next_blkoff = blk_off;
1702 if (seg_i->alloc_type == SSR)
1703 blk_off = sbi->blocks_per_seg;
1705 for (j = 0; j < blk_off; j++) {
1706 struct f2fs_summary *s;
1707 s = (struct f2fs_summary *)(kaddr + offset);
1708 seg_i->sum_blk->entries[j] = *s;
1709 offset += SUMMARY_SIZE;
1710 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1714 f2fs_put_page(page, 1);
1717 page = get_meta_page(sbi, start++);
1718 kaddr = (unsigned char *)page_address(page);
1722 f2fs_put_page(page, 1);
1726 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1728 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1729 struct f2fs_summary_block *sum;
1730 struct curseg_info *curseg;
1732 unsigned short blk_off;
1733 unsigned int segno = 0;
1734 block_t blk_addr = 0;
1736 /* get segment number and block addr */
1737 if (IS_DATASEG(type)) {
1738 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1739 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1741 if (__exist_node_summaries(sbi))
1742 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1744 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1746 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1748 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1750 if (__exist_node_summaries(sbi))
1751 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1752 type - CURSEG_HOT_NODE);
1754 blk_addr = GET_SUM_BLOCK(sbi, segno);
1757 new = get_meta_page(sbi, blk_addr);
1758 sum = (struct f2fs_summary_block *)page_address(new);
1760 if (IS_NODESEG(type)) {
1761 if (__exist_node_summaries(sbi)) {
1762 struct f2fs_summary *ns = &sum->entries[0];
1764 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1766 ns->ofs_in_node = 0;
1771 err = restore_node_summary(sbi, segno, sum);
1773 f2fs_put_page(new, 1);
1779 /* set uncompleted segment to curseg */
1780 curseg = CURSEG_I(sbi, type);
1781 mutex_lock(&curseg->curseg_mutex);
1783 /* update journal info */
1784 down_write(&curseg->journal_rwsem);
1785 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1786 up_write(&curseg->journal_rwsem);
1788 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1789 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1790 curseg->next_segno = segno;
1791 reset_curseg(sbi, type, 0);
1792 curseg->alloc_type = ckpt->alloc_type[type];
1793 curseg->next_blkoff = blk_off;
1794 mutex_unlock(&curseg->curseg_mutex);
1795 f2fs_put_page(new, 1);
1799 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1801 int type = CURSEG_HOT_DATA;
1804 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1805 int npages = npages_for_summary_flush(sbi, true);
1808 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1811 /* restore for compacted data summary */
1812 if (read_compacted_summaries(sbi))
1814 type = CURSEG_HOT_NODE;
1817 if (__exist_node_summaries(sbi))
1818 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1819 NR_CURSEG_TYPE - type, META_CP, true);
1821 for (; type <= CURSEG_COLD_NODE; type++) {
1822 err = read_normal_summaries(sbi, type);
1830 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1833 unsigned char *kaddr;
1834 struct f2fs_summary *summary;
1835 struct curseg_info *seg_i;
1836 int written_size = 0;
1839 page = grab_meta_page(sbi, blkaddr++);
1840 kaddr = (unsigned char *)page_address(page);
1842 /* Step 1: write nat cache */
1843 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1844 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1845 written_size += SUM_JOURNAL_SIZE;
1847 /* Step 2: write sit cache */
1848 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1849 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1850 written_size += SUM_JOURNAL_SIZE;
1852 /* Step 3: write summary entries */
1853 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1854 unsigned short blkoff;
1855 seg_i = CURSEG_I(sbi, i);
1856 if (sbi->ckpt->alloc_type[i] == SSR)
1857 blkoff = sbi->blocks_per_seg;
1859 blkoff = curseg_blkoff(sbi, i);
1861 for (j = 0; j < blkoff; j++) {
1863 page = grab_meta_page(sbi, blkaddr++);
1864 kaddr = (unsigned char *)page_address(page);
1867 summary = (struct f2fs_summary *)(kaddr + written_size);
1868 *summary = seg_i->sum_blk->entries[j];
1869 written_size += SUMMARY_SIZE;
1871 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1875 set_page_dirty(page);
1876 f2fs_put_page(page, 1);
1881 set_page_dirty(page);
1882 f2fs_put_page(page, 1);
1886 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1887 block_t blkaddr, int type)
1890 if (IS_DATASEG(type))
1891 end = type + NR_CURSEG_DATA_TYPE;
1893 end = type + NR_CURSEG_NODE_TYPE;
1895 for (i = type; i < end; i++)
1896 write_current_sum_page(sbi, i, blkaddr + (i - type));
1899 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1901 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
1902 write_compacted_summaries(sbi, start_blk);
1904 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1907 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1909 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1912 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1913 unsigned int val, int alloc)
1917 if (type == NAT_JOURNAL) {
1918 for (i = 0; i < nats_in_cursum(journal); i++) {
1919 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1922 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1923 return update_nats_in_cursum(journal, 1);
1924 } else if (type == SIT_JOURNAL) {
1925 for (i = 0; i < sits_in_cursum(journal); i++)
1926 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1928 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1929 return update_sits_in_cursum(journal, 1);
1934 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1937 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1940 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1943 struct sit_info *sit_i = SIT_I(sbi);
1944 struct page *src_page, *dst_page;
1945 pgoff_t src_off, dst_off;
1946 void *src_addr, *dst_addr;
1948 src_off = current_sit_addr(sbi, start);
1949 dst_off = next_sit_addr(sbi, src_off);
1951 /* get current sit block page without lock */
1952 src_page = get_meta_page(sbi, src_off);
1953 dst_page = grab_meta_page(sbi, dst_off);
1954 f2fs_bug_on(sbi, PageDirty(src_page));
1956 src_addr = page_address(src_page);
1957 dst_addr = page_address(dst_page);
1958 memcpy(dst_addr, src_addr, PAGE_SIZE);
1960 set_page_dirty(dst_page);
1961 f2fs_put_page(src_page, 1);
1963 set_to_next_sit(sit_i, start);
1968 static struct sit_entry_set *grab_sit_entry_set(void)
1970 struct sit_entry_set *ses =
1971 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1974 INIT_LIST_HEAD(&ses->set_list);
1978 static void release_sit_entry_set(struct sit_entry_set *ses)
1980 list_del(&ses->set_list);
1981 kmem_cache_free(sit_entry_set_slab, ses);
1984 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1985 struct list_head *head)
1987 struct sit_entry_set *next = ses;
1989 if (list_is_last(&ses->set_list, head))
1992 list_for_each_entry_continue(next, head, set_list)
1993 if (ses->entry_cnt <= next->entry_cnt)
1996 list_move_tail(&ses->set_list, &next->set_list);
1999 static void add_sit_entry(unsigned int segno, struct list_head *head)
2001 struct sit_entry_set *ses;
2002 unsigned int start_segno = START_SEGNO(segno);
2004 list_for_each_entry(ses, head, set_list) {
2005 if (ses->start_segno == start_segno) {
2007 adjust_sit_entry_set(ses, head);
2012 ses = grab_sit_entry_set();
2014 ses->start_segno = start_segno;
2016 list_add(&ses->set_list, head);
2019 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2021 struct f2fs_sm_info *sm_info = SM_I(sbi);
2022 struct list_head *set_list = &sm_info->sit_entry_set;
2023 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2026 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2027 add_sit_entry(segno, set_list);
2030 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2032 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2033 struct f2fs_journal *journal = curseg->journal;
2036 down_write(&curseg->journal_rwsem);
2037 for (i = 0; i < sits_in_cursum(journal); i++) {
2041 segno = le32_to_cpu(segno_in_journal(journal, i));
2042 dirtied = __mark_sit_entry_dirty(sbi, segno);
2045 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2047 update_sits_in_cursum(journal, -i);
2048 up_write(&curseg->journal_rwsem);
2052 * CP calls this function, which flushes SIT entries including sit_journal,
2053 * and moves prefree segs to free segs.
2055 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2057 struct sit_info *sit_i = SIT_I(sbi);
2058 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2059 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2060 struct f2fs_journal *journal = curseg->journal;
2061 struct sit_entry_set *ses, *tmp;
2062 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2063 bool to_journal = true;
2064 struct seg_entry *se;
2066 mutex_lock(&sit_i->sentry_lock);
2068 if (!sit_i->dirty_sentries)
2072 * add and account sit entries of dirty bitmap in sit entry
2075 add_sits_in_set(sbi);
2078 * if there are no enough space in journal to store dirty sit
2079 * entries, remove all entries from journal and add and account
2080 * them in sit entry set.
2082 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2083 remove_sits_in_journal(sbi);
2086 * there are two steps to flush sit entries:
2087 * #1, flush sit entries to journal in current cold data summary block.
2088 * #2, flush sit entries to sit page.
2090 list_for_each_entry_safe(ses, tmp, head, set_list) {
2091 struct page *page = NULL;
2092 struct f2fs_sit_block *raw_sit = NULL;
2093 unsigned int start_segno = ses->start_segno;
2094 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2095 (unsigned long)MAIN_SEGS(sbi));
2096 unsigned int segno = start_segno;
2099 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2103 down_write(&curseg->journal_rwsem);
2105 page = get_next_sit_page(sbi, start_segno);
2106 raw_sit = page_address(page);
2109 /* flush dirty sit entries in region of current sit set */
2110 for_each_set_bit_from(segno, bitmap, end) {
2111 int offset, sit_offset;
2113 se = get_seg_entry(sbi, segno);
2115 /* add discard candidates */
2116 if (cpc->reason != CP_DISCARD) {
2117 cpc->trim_start = segno;
2118 add_discard_addrs(sbi, cpc);
2122 offset = lookup_journal_in_cursum(journal,
2123 SIT_JOURNAL, segno, 1);
2124 f2fs_bug_on(sbi, offset < 0);
2125 segno_in_journal(journal, offset) =
2127 seg_info_to_raw_sit(se,
2128 &sit_in_journal(journal, offset));
2130 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2131 seg_info_to_raw_sit(se,
2132 &raw_sit->entries[sit_offset]);
2135 __clear_bit(segno, bitmap);
2136 sit_i->dirty_sentries--;
2141 up_write(&curseg->journal_rwsem);
2143 f2fs_put_page(page, 1);
2145 f2fs_bug_on(sbi, ses->entry_cnt);
2146 release_sit_entry_set(ses);
2149 f2fs_bug_on(sbi, !list_empty(head));
2150 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2152 if (cpc->reason == CP_DISCARD) {
2153 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2154 add_discard_addrs(sbi, cpc);
2156 mutex_unlock(&sit_i->sentry_lock);
2158 set_prefree_as_free_segments(sbi);
2161 static int build_sit_info(struct f2fs_sb_info *sbi)
2163 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2164 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2165 struct sit_info *sit_i;
2166 unsigned int sit_segs, start;
2167 char *src_bitmap, *dst_bitmap;
2168 unsigned int bitmap_size;
2170 /* allocate memory for SIT information */
2171 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2175 SM_I(sbi)->sit_info = sit_i;
2177 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2178 sizeof(struct seg_entry), GFP_KERNEL);
2179 if (!sit_i->sentries)
2182 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2183 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2184 if (!sit_i->dirty_sentries_bitmap)
2187 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2188 sit_i->sentries[start].cur_valid_map
2189 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2190 sit_i->sentries[start].ckpt_valid_map
2191 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2192 if (!sit_i->sentries[start].cur_valid_map ||
2193 !sit_i->sentries[start].ckpt_valid_map)
2196 if (f2fs_discard_en(sbi)) {
2197 sit_i->sentries[start].discard_map
2198 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2199 if (!sit_i->sentries[start].discard_map)
2204 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2205 if (!sit_i->tmp_map)
2208 if (sbi->segs_per_sec > 1) {
2209 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2210 sizeof(struct sec_entry), GFP_KERNEL);
2211 if (!sit_i->sec_entries)
2215 /* get information related with SIT */
2216 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2218 /* setup SIT bitmap from ckeckpoint pack */
2219 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2220 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2222 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2226 /* init SIT information */
2227 sit_i->s_ops = &default_salloc_ops;
2229 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2230 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2231 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2232 sit_i->sit_bitmap = dst_bitmap;
2233 sit_i->bitmap_size = bitmap_size;
2234 sit_i->dirty_sentries = 0;
2235 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2236 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2237 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2238 mutex_init(&sit_i->sentry_lock);
2242 static int build_free_segmap(struct f2fs_sb_info *sbi)
2244 struct free_segmap_info *free_i;
2245 unsigned int bitmap_size, sec_bitmap_size;
2247 /* allocate memory for free segmap information */
2248 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2252 SM_I(sbi)->free_info = free_i;
2254 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2255 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2256 if (!free_i->free_segmap)
2259 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2260 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2261 if (!free_i->free_secmap)
2264 /* set all segments as dirty temporarily */
2265 memset(free_i->free_segmap, 0xff, bitmap_size);
2266 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2268 /* init free segmap information */
2269 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2270 free_i->free_segments = 0;
2271 free_i->free_sections = 0;
2272 spin_lock_init(&free_i->segmap_lock);
2276 static int build_curseg(struct f2fs_sb_info *sbi)
2278 struct curseg_info *array;
2281 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2285 SM_I(sbi)->curseg_array = array;
2287 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2288 mutex_init(&array[i].curseg_mutex);
2289 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2290 if (!array[i].sum_blk)
2292 init_rwsem(&array[i].journal_rwsem);
2293 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2295 if (!array[i].journal)
2297 array[i].segno = NULL_SEGNO;
2298 array[i].next_blkoff = 0;
2300 return restore_curseg_summaries(sbi);
2303 static void build_sit_entries(struct f2fs_sb_info *sbi)
2305 struct sit_info *sit_i = SIT_I(sbi);
2306 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2307 struct f2fs_journal *journal = curseg->journal;
2308 struct seg_entry *se;
2309 struct f2fs_sit_entry sit;
2310 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2311 unsigned int i, start, end;
2312 unsigned int readed, start_blk = 0;
2313 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2316 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2318 start = start_blk * sit_i->sents_per_block;
2319 end = (start_blk + readed) * sit_i->sents_per_block;
2321 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2322 struct f2fs_sit_block *sit_blk;
2325 se = &sit_i->sentries[start];
2326 page = get_current_sit_page(sbi, start);
2327 sit_blk = (struct f2fs_sit_block *)page_address(page);
2328 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2329 f2fs_put_page(page, 1);
2331 check_block_count(sbi, start, &sit);
2332 seg_info_from_raw_sit(se, &sit);
2334 /* build discard map only one time */
2335 if (f2fs_discard_en(sbi)) {
2336 memcpy(se->discard_map, se->cur_valid_map,
2337 SIT_VBLOCK_MAP_SIZE);
2338 sbi->discard_blks += sbi->blocks_per_seg -
2342 if (sbi->segs_per_sec > 1)
2343 get_sec_entry(sbi, start)->valid_blocks +=
2346 start_blk += readed;
2347 } while (start_blk < sit_blk_cnt);
2349 down_read(&curseg->journal_rwsem);
2350 for (i = 0; i < sits_in_cursum(journal); i++) {
2351 unsigned int old_valid_blocks;
2353 start = le32_to_cpu(segno_in_journal(journal, i));
2354 se = &sit_i->sentries[start];
2355 sit = sit_in_journal(journal, i);
2357 old_valid_blocks = se->valid_blocks;
2359 check_block_count(sbi, start, &sit);
2360 seg_info_from_raw_sit(se, &sit);
2362 if (f2fs_discard_en(sbi)) {
2363 memcpy(se->discard_map, se->cur_valid_map,
2364 SIT_VBLOCK_MAP_SIZE);
2365 sbi->discard_blks += old_valid_blocks -
2369 if (sbi->segs_per_sec > 1)
2370 get_sec_entry(sbi, start)->valid_blocks +=
2371 se->valid_blocks - old_valid_blocks;
2373 up_read(&curseg->journal_rwsem);
2376 static void init_free_segmap(struct f2fs_sb_info *sbi)
2381 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2382 struct seg_entry *sentry = get_seg_entry(sbi, start);
2383 if (!sentry->valid_blocks)
2384 __set_free(sbi, start);
2387 /* set use the current segments */
2388 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2389 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2390 __set_test_and_inuse(sbi, curseg_t->segno);
2394 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2396 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2397 struct free_segmap_info *free_i = FREE_I(sbi);
2398 unsigned int segno = 0, offset = 0;
2399 unsigned short valid_blocks;
2402 /* find dirty segment based on free segmap */
2403 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2404 if (segno >= MAIN_SEGS(sbi))
2407 valid_blocks = get_valid_blocks(sbi, segno, 0);
2408 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2410 if (valid_blocks > sbi->blocks_per_seg) {
2411 f2fs_bug_on(sbi, 1);
2414 mutex_lock(&dirty_i->seglist_lock);
2415 __locate_dirty_segment(sbi, segno, DIRTY);
2416 mutex_unlock(&dirty_i->seglist_lock);
2420 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2422 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2423 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2425 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2426 if (!dirty_i->victim_secmap)
2431 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2433 struct dirty_seglist_info *dirty_i;
2434 unsigned int bitmap_size, i;
2436 /* allocate memory for dirty segments list information */
2437 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2441 SM_I(sbi)->dirty_info = dirty_i;
2442 mutex_init(&dirty_i->seglist_lock);
2444 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2446 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2447 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2448 if (!dirty_i->dirty_segmap[i])
2452 init_dirty_segmap(sbi);
2453 return init_victim_secmap(sbi);
2457 * Update min, max modified time for cost-benefit GC algorithm
2459 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2461 struct sit_info *sit_i = SIT_I(sbi);
2464 mutex_lock(&sit_i->sentry_lock);
2466 sit_i->min_mtime = LLONG_MAX;
2468 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2470 unsigned long long mtime = 0;
2472 for (i = 0; i < sbi->segs_per_sec; i++)
2473 mtime += get_seg_entry(sbi, segno + i)->mtime;
2475 mtime = div_u64(mtime, sbi->segs_per_sec);
2477 if (sit_i->min_mtime > mtime)
2478 sit_i->min_mtime = mtime;
2480 sit_i->max_mtime = get_mtime(sbi);
2481 mutex_unlock(&sit_i->sentry_lock);
2484 int build_segment_manager(struct f2fs_sb_info *sbi)
2486 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2487 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2488 struct f2fs_sm_info *sm_info;
2491 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2496 sbi->sm_info = sm_info;
2497 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2498 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2499 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2500 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2501 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2502 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2503 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2504 sm_info->rec_prefree_segments = sm_info->main_segments *
2505 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2506 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2507 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2509 if (!test_opt(sbi, LFS))
2510 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2511 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2512 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2514 INIT_LIST_HEAD(&sm_info->discard_list);
2515 INIT_LIST_HEAD(&sm_info->wait_list);
2516 sm_info->nr_discards = 0;
2517 sm_info->max_discards = 0;
2519 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2521 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2523 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2524 err = create_flush_cmd_control(sbi);
2529 err = build_sit_info(sbi);
2532 err = build_free_segmap(sbi);
2535 err = build_curseg(sbi);
2539 /* reinit free segmap based on SIT */
2540 build_sit_entries(sbi);
2542 init_free_segmap(sbi);
2543 err = build_dirty_segmap(sbi);
2547 init_min_max_mtime(sbi);
2551 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2552 enum dirty_type dirty_type)
2554 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2556 mutex_lock(&dirty_i->seglist_lock);
2557 kvfree(dirty_i->dirty_segmap[dirty_type]);
2558 dirty_i->nr_dirty[dirty_type] = 0;
2559 mutex_unlock(&dirty_i->seglist_lock);
2562 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2564 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2565 kvfree(dirty_i->victim_secmap);
2568 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2570 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2576 /* discard pre-free/dirty segments list */
2577 for (i = 0; i < NR_DIRTY_TYPE; i++)
2578 discard_dirty_segmap(sbi, i);
2580 destroy_victim_secmap(sbi);
2581 SM_I(sbi)->dirty_info = NULL;
2585 static void destroy_curseg(struct f2fs_sb_info *sbi)
2587 struct curseg_info *array = SM_I(sbi)->curseg_array;
2592 SM_I(sbi)->curseg_array = NULL;
2593 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2594 kfree(array[i].sum_blk);
2595 kfree(array[i].journal);
2600 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2602 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2605 SM_I(sbi)->free_info = NULL;
2606 kvfree(free_i->free_segmap);
2607 kvfree(free_i->free_secmap);
2611 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2613 struct sit_info *sit_i = SIT_I(sbi);
2619 if (sit_i->sentries) {
2620 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2621 kfree(sit_i->sentries[start].cur_valid_map);
2622 kfree(sit_i->sentries[start].ckpt_valid_map);
2623 kfree(sit_i->sentries[start].discard_map);
2626 kfree(sit_i->tmp_map);
2628 kvfree(sit_i->sentries);
2629 kvfree(sit_i->sec_entries);
2630 kvfree(sit_i->dirty_sentries_bitmap);
2632 SM_I(sbi)->sit_info = NULL;
2633 kfree(sit_i->sit_bitmap);
2637 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2639 struct f2fs_sm_info *sm_info = SM_I(sbi);
2643 destroy_flush_cmd_control(sbi);
2644 destroy_dirty_segmap(sbi);
2645 destroy_curseg(sbi);
2646 destroy_free_segmap(sbi);
2647 destroy_sit_info(sbi);
2648 sbi->sm_info = NULL;
2652 int __init create_segment_manager_caches(void)
2654 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2655 sizeof(struct discard_entry));
2656 if (!discard_entry_slab)
2659 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2660 sizeof(struct bio_entry));
2661 if (!bio_entry_slab)
2662 goto destroy_discard_entry;
2664 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2665 sizeof(struct sit_entry_set));
2666 if (!sit_entry_set_slab)
2667 goto destroy_bio_entry;
2669 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2670 sizeof(struct inmem_pages));
2671 if (!inmem_entry_slab)
2672 goto destroy_sit_entry_set;
2675 destroy_sit_entry_set:
2676 kmem_cache_destroy(sit_entry_set_slab);
2678 kmem_cache_destroy(bio_entry_slab);
2679 destroy_discard_entry:
2680 kmem_cache_destroy(discard_entry_slab);
2685 void destroy_segment_manager_caches(void)
2687 kmem_cache_destroy(sit_entry_set_slab);
2688 kmem_cache_destroy(bio_entry_slab);
2689 kmem_cache_destroy(discard_entry_slab);
2690 kmem_cache_destroy(inmem_entry_slab);