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)
348 #ifdef CONFIG_F2FS_FAULT_INJECTION
349 if (time_to_inject(sbi, FAULT_CHECKPOINT))
350 f2fs_stop_checkpoint(sbi, false);
356 /* balance_fs_bg is able to be pending */
357 if (excess_cached_nats(sbi))
358 f2fs_balance_fs_bg(sbi);
361 * We should do GC or end up with checkpoint, if there are so many dirty
362 * dir/node pages without enough free segments.
364 if (has_not_enough_free_secs(sbi, 0, 0)) {
365 mutex_lock(&sbi->gc_mutex);
370 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
372 /* try to shrink extent cache when there is no enough memory */
373 if (!available_free_memory(sbi, EXTENT_CACHE))
374 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
376 /* check the # of cached NAT entries */
377 if (!available_free_memory(sbi, NAT_ENTRIES))
378 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
380 if (!available_free_memory(sbi, FREE_NIDS))
381 try_to_free_nids(sbi, MAX_FREE_NIDS);
383 build_free_nids(sbi);
385 /* checkpoint is the only way to shrink partial cached entries */
386 if (!available_free_memory(sbi, NAT_ENTRIES) ||
387 !available_free_memory(sbi, INO_ENTRIES) ||
388 excess_prefree_segs(sbi) ||
389 excess_dirty_nats(sbi) ||
390 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
391 if (test_opt(sbi, DATA_FLUSH)) {
392 struct blk_plug plug;
394 blk_start_plug(&plug);
395 sync_dirty_inodes(sbi, FILE_INODE);
396 blk_finish_plug(&plug);
398 f2fs_sync_fs(sbi->sb, true);
399 stat_inc_bg_cp_count(sbi->stat_info);
403 static int issue_flush_thread(void *data)
405 struct f2fs_sb_info *sbi = data;
406 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
407 wait_queue_head_t *q = &fcc->flush_wait_queue;
409 if (kthread_should_stop())
412 if (!llist_empty(&fcc->issue_list)) {
414 struct flush_cmd *cmd, *next;
417 bio = f2fs_bio_alloc(0);
419 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
420 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
422 bio->bi_bdev = sbi->sb->s_bdev;
423 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
424 ret = submit_bio_wait(bio);
426 llist_for_each_entry_safe(cmd, next,
427 fcc->dispatch_list, llnode) {
429 complete(&cmd->wait);
432 fcc->dispatch_list = NULL;
435 wait_event_interruptible(*q,
436 kthread_should_stop() || !llist_empty(&fcc->issue_list));
440 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
442 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
443 struct flush_cmd cmd;
445 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
446 test_opt(sbi, FLUSH_MERGE));
448 if (test_opt(sbi, NOBARRIER))
451 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
452 struct bio *bio = f2fs_bio_alloc(0);
455 atomic_inc(&fcc->submit_flush);
456 bio->bi_bdev = sbi->sb->s_bdev;
457 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
458 ret = submit_bio_wait(bio);
459 atomic_dec(&fcc->submit_flush);
464 init_completion(&cmd.wait);
466 atomic_inc(&fcc->submit_flush);
467 llist_add(&cmd.llnode, &fcc->issue_list);
469 if (!fcc->dispatch_list)
470 wake_up(&fcc->flush_wait_queue);
472 wait_for_completion(&cmd.wait);
473 atomic_dec(&fcc->submit_flush);
478 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
480 dev_t dev = sbi->sb->s_bdev->bd_dev;
481 struct flush_cmd_control *fcc;
484 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
487 atomic_set(&fcc->submit_flush, 0);
488 init_waitqueue_head(&fcc->flush_wait_queue);
489 init_llist_head(&fcc->issue_list);
490 SM_I(sbi)->cmd_control_info = fcc;
491 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
492 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
493 if (IS_ERR(fcc->f2fs_issue_flush)) {
494 err = PTR_ERR(fcc->f2fs_issue_flush);
496 SM_I(sbi)->cmd_control_info = NULL;
503 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
505 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
507 if (fcc && fcc->f2fs_issue_flush)
508 kthread_stop(fcc->f2fs_issue_flush);
510 SM_I(sbi)->cmd_control_info = NULL;
513 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
514 enum dirty_type dirty_type)
516 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
518 /* need not be added */
519 if (IS_CURSEG(sbi, segno))
522 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
523 dirty_i->nr_dirty[dirty_type]++;
525 if (dirty_type == DIRTY) {
526 struct seg_entry *sentry = get_seg_entry(sbi, segno);
527 enum dirty_type t = sentry->type;
529 if (unlikely(t >= DIRTY)) {
533 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
534 dirty_i->nr_dirty[t]++;
538 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
539 enum dirty_type dirty_type)
541 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
543 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
544 dirty_i->nr_dirty[dirty_type]--;
546 if (dirty_type == DIRTY) {
547 struct seg_entry *sentry = get_seg_entry(sbi, segno);
548 enum dirty_type t = sentry->type;
550 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
551 dirty_i->nr_dirty[t]--;
553 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
554 clear_bit(GET_SECNO(sbi, segno),
555 dirty_i->victim_secmap);
560 * Should not occur error such as -ENOMEM.
561 * Adding dirty entry into seglist is not critical operation.
562 * If a given segment is one of current working segments, it won't be added.
564 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
566 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
567 unsigned short valid_blocks;
569 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
572 mutex_lock(&dirty_i->seglist_lock);
574 valid_blocks = get_valid_blocks(sbi, segno, 0);
576 if (valid_blocks == 0) {
577 __locate_dirty_segment(sbi, segno, PRE);
578 __remove_dirty_segment(sbi, segno, DIRTY);
579 } else if (valid_blocks < sbi->blocks_per_seg) {
580 __locate_dirty_segment(sbi, segno, DIRTY);
582 /* Recovery routine with SSR needs this */
583 __remove_dirty_segment(sbi, segno, DIRTY);
586 mutex_unlock(&dirty_i->seglist_lock);
589 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
592 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
593 struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
595 INIT_LIST_HEAD(&be->list);
597 init_completion(&be->event);
598 list_add_tail(&be->list, wait_list);
603 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
605 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
606 struct bio_entry *be, *tmp;
608 list_for_each_entry_safe(be, tmp, wait_list, list) {
609 struct bio *bio = be->bio;
612 wait_for_completion_io(&be->event);
614 if (err == -EOPNOTSUPP)
618 f2fs_msg(sbi->sb, KERN_INFO,
619 "Issue discard failed, ret: %d", err);
623 kmem_cache_free(bio_entry_slab, be);
627 static void f2fs_submit_bio_wait_endio(struct bio *bio)
629 struct bio_entry *be = (struct bio_entry *)bio->bi_private;
631 be->error = bio->bi_error;
632 complete(&be->event);
635 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
636 int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi, sector_t sector,
637 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
639 struct block_device *bdev = sbi->sb->s_bdev;
640 struct bio *bio = NULL;
643 err = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
646 struct bio_entry *be = __add_bio_entry(sbi, bio);
648 bio->bi_private = be;
649 bio->bi_end_io = f2fs_submit_bio_wait_endio;
650 bio->bi_opf |= REQ_SYNC;
657 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
658 block_t blkstart, block_t blklen)
660 sector_t start = SECTOR_FROM_BLOCK(blkstart);
661 sector_t len = SECTOR_FROM_BLOCK(blklen);
662 struct seg_entry *se;
666 for (i = blkstart; i < blkstart + blklen; i++) {
667 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
668 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
670 if (!f2fs_test_and_set_bit(offset, se->discard_map))
673 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
674 return __f2fs_issue_discard_async(sbi, start, len, GFP_NOFS, 0);
677 static void __add_discard_entry(struct f2fs_sb_info *sbi,
678 struct cp_control *cpc, struct seg_entry *se,
679 unsigned int start, unsigned int end)
681 struct list_head *head = &SM_I(sbi)->discard_list;
682 struct discard_entry *new, *last;
684 if (!list_empty(head)) {
685 last = list_last_entry(head, struct discard_entry, list);
686 if (START_BLOCK(sbi, cpc->trim_start) + start ==
687 last->blkaddr + last->len) {
688 last->len += end - start;
693 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
694 INIT_LIST_HEAD(&new->list);
695 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
696 new->len = end - start;
697 list_add_tail(&new->list, head);
699 SM_I(sbi)->nr_discards += end - start;
702 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
704 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
705 int max_blocks = sbi->blocks_per_seg;
706 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
707 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
708 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
709 unsigned long *discard_map = (unsigned long *)se->discard_map;
710 unsigned long *dmap = SIT_I(sbi)->tmp_map;
711 unsigned int start = 0, end = -1;
712 bool force = (cpc->reason == CP_DISCARD);
715 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
719 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
720 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
724 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
725 for (i = 0; i < entries; i++)
726 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
727 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
729 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
730 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
731 if (start >= max_blocks)
734 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
735 if (force && start && end != max_blocks
736 && (end - start) < cpc->trim_minlen)
739 __add_discard_entry(sbi, cpc, se, start, end);
743 void release_discard_addrs(struct f2fs_sb_info *sbi)
745 struct list_head *head = &(SM_I(sbi)->discard_list);
746 struct discard_entry *entry, *this;
749 list_for_each_entry_safe(entry, this, head, list) {
750 list_del(&entry->list);
751 kmem_cache_free(discard_entry_slab, entry);
756 * Should call clear_prefree_segments after checkpoint is done.
758 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
760 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
763 mutex_lock(&dirty_i->seglist_lock);
764 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
765 __set_test_and_free(sbi, segno);
766 mutex_unlock(&dirty_i->seglist_lock);
769 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
771 struct list_head *head = &(SM_I(sbi)->discard_list);
772 struct discard_entry *entry, *this;
773 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
774 struct blk_plug plug;
775 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
776 unsigned int start = 0, end = -1;
777 unsigned int secno, start_segno;
778 bool force = (cpc->reason == CP_DISCARD);
780 blk_start_plug(&plug);
782 mutex_lock(&dirty_i->seglist_lock);
786 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
787 if (start >= MAIN_SEGS(sbi))
789 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
792 for (i = start; i < end; i++)
793 clear_bit(i, prefree_map);
795 dirty_i->nr_dirty[PRE] -= end - start;
797 if (force || !test_opt(sbi, DISCARD))
800 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
801 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
802 (end - start) << sbi->log_blocks_per_seg);
806 secno = GET_SECNO(sbi, start);
807 start_segno = secno * sbi->segs_per_sec;
808 if (!IS_CURSEC(sbi, secno) &&
809 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
810 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
811 sbi->segs_per_sec << sbi->log_blocks_per_seg);
813 start = start_segno + sbi->segs_per_sec;
819 mutex_unlock(&dirty_i->seglist_lock);
821 /* send small discards */
822 list_for_each_entry_safe(entry, this, head, list) {
823 if (force && entry->len < cpc->trim_minlen)
825 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
826 cpc->trimmed += entry->len;
828 list_del(&entry->list);
829 SM_I(sbi)->nr_discards -= entry->len;
830 kmem_cache_free(discard_entry_slab, entry);
833 blk_finish_plug(&plug);
836 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
838 struct sit_info *sit_i = SIT_I(sbi);
840 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
841 sit_i->dirty_sentries++;
848 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
849 unsigned int segno, int modified)
851 struct seg_entry *se = get_seg_entry(sbi, segno);
854 __mark_sit_entry_dirty(sbi, segno);
857 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
859 struct seg_entry *se;
860 unsigned int segno, offset;
861 long int new_vblocks;
863 segno = GET_SEGNO(sbi, blkaddr);
865 se = get_seg_entry(sbi, segno);
866 new_vblocks = se->valid_blocks + del;
867 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
869 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
870 (new_vblocks > sbi->blocks_per_seg)));
872 se->valid_blocks = new_vblocks;
873 se->mtime = get_mtime(sbi);
874 SIT_I(sbi)->max_mtime = se->mtime;
876 /* Update valid block bitmap */
878 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
880 if (f2fs_discard_en(sbi) &&
881 !f2fs_test_and_set_bit(offset, se->discard_map))
884 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
886 if (f2fs_discard_en(sbi) &&
887 f2fs_test_and_clear_bit(offset, se->discard_map))
890 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
891 se->ckpt_valid_blocks += del;
893 __mark_sit_entry_dirty(sbi, segno);
895 /* update total number of valid blocks to be written in ckpt area */
896 SIT_I(sbi)->written_valid_blocks += del;
898 if (sbi->segs_per_sec > 1)
899 get_sec_entry(sbi, segno)->valid_blocks += del;
902 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
904 update_sit_entry(sbi, new, 1);
905 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
906 update_sit_entry(sbi, old, -1);
908 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
909 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
912 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
914 unsigned int segno = GET_SEGNO(sbi, addr);
915 struct sit_info *sit_i = SIT_I(sbi);
917 f2fs_bug_on(sbi, addr == NULL_ADDR);
918 if (addr == NEW_ADDR)
921 /* add it into sit main buffer */
922 mutex_lock(&sit_i->sentry_lock);
924 update_sit_entry(sbi, addr, -1);
926 /* add it into dirty seglist */
927 locate_dirty_segment(sbi, segno);
929 mutex_unlock(&sit_i->sentry_lock);
932 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
934 struct sit_info *sit_i = SIT_I(sbi);
935 unsigned int segno, offset;
936 struct seg_entry *se;
939 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
942 mutex_lock(&sit_i->sentry_lock);
944 segno = GET_SEGNO(sbi, blkaddr);
945 se = get_seg_entry(sbi, segno);
946 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
948 if (f2fs_test_bit(offset, se->ckpt_valid_map))
951 mutex_unlock(&sit_i->sentry_lock);
957 * This function should be resided under the curseg_mutex lock
959 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
960 struct f2fs_summary *sum)
962 struct curseg_info *curseg = CURSEG_I(sbi, type);
963 void *addr = curseg->sum_blk;
964 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
965 memcpy(addr, sum, sizeof(struct f2fs_summary));
969 * Calculate the number of current summary pages for writing
971 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
973 int valid_sum_count = 0;
976 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
977 if (sbi->ckpt->alloc_type[i] == SSR)
978 valid_sum_count += sbi->blocks_per_seg;
981 valid_sum_count += le16_to_cpu(
982 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
984 valid_sum_count += curseg_blkoff(sbi, i);
988 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
989 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
990 if (valid_sum_count <= sum_in_page)
992 else if ((valid_sum_count - sum_in_page) <=
993 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
999 * Caller should put this summary page
1001 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1003 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1006 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1008 struct page *page = grab_meta_page(sbi, blk_addr);
1009 void *dst = page_address(page);
1012 memcpy(dst, src, PAGE_SIZE);
1014 memset(dst, 0, PAGE_SIZE);
1015 set_page_dirty(page);
1016 f2fs_put_page(page, 1);
1019 static void write_sum_page(struct f2fs_sb_info *sbi,
1020 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1022 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1025 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1026 int type, block_t blk_addr)
1028 struct curseg_info *curseg = CURSEG_I(sbi, type);
1029 struct page *page = grab_meta_page(sbi, blk_addr);
1030 struct f2fs_summary_block *src = curseg->sum_blk;
1031 struct f2fs_summary_block *dst;
1033 dst = (struct f2fs_summary_block *)page_address(page);
1035 mutex_lock(&curseg->curseg_mutex);
1037 down_read(&curseg->journal_rwsem);
1038 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1039 up_read(&curseg->journal_rwsem);
1041 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1042 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1044 mutex_unlock(&curseg->curseg_mutex);
1046 set_page_dirty(page);
1047 f2fs_put_page(page, 1);
1050 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1052 struct curseg_info *curseg = CURSEG_I(sbi, type);
1053 unsigned int segno = curseg->segno + 1;
1054 struct free_segmap_info *free_i = FREE_I(sbi);
1056 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1057 return !test_bit(segno, free_i->free_segmap);
1062 * Find a new segment from the free segments bitmap to right order
1063 * This function should be returned with success, otherwise BUG
1065 static void get_new_segment(struct f2fs_sb_info *sbi,
1066 unsigned int *newseg, bool new_sec, int dir)
1068 struct free_segmap_info *free_i = FREE_I(sbi);
1069 unsigned int segno, secno, zoneno;
1070 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1071 unsigned int hint = *newseg / sbi->segs_per_sec;
1072 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1073 unsigned int left_start = hint;
1078 spin_lock(&free_i->segmap_lock);
1080 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1081 segno = find_next_zero_bit(free_i->free_segmap,
1082 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1083 if (segno < (hint + 1) * sbi->segs_per_sec)
1087 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1088 if (secno >= MAIN_SECS(sbi)) {
1089 if (dir == ALLOC_RIGHT) {
1090 secno = find_next_zero_bit(free_i->free_secmap,
1092 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1095 left_start = hint - 1;
1101 while (test_bit(left_start, free_i->free_secmap)) {
1102 if (left_start > 0) {
1106 left_start = find_next_zero_bit(free_i->free_secmap,
1108 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1114 segno = secno * sbi->segs_per_sec;
1115 zoneno = secno / sbi->secs_per_zone;
1117 /* give up on finding another zone */
1120 if (sbi->secs_per_zone == 1)
1122 if (zoneno == old_zoneno)
1124 if (dir == ALLOC_LEFT) {
1125 if (!go_left && zoneno + 1 >= total_zones)
1127 if (go_left && zoneno == 0)
1130 for (i = 0; i < NR_CURSEG_TYPE; i++)
1131 if (CURSEG_I(sbi, i)->zone == zoneno)
1134 if (i < NR_CURSEG_TYPE) {
1135 /* zone is in user, try another */
1137 hint = zoneno * sbi->secs_per_zone - 1;
1138 else if (zoneno + 1 >= total_zones)
1141 hint = (zoneno + 1) * sbi->secs_per_zone;
1143 goto find_other_zone;
1146 /* set it as dirty segment in free segmap */
1147 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1148 __set_inuse(sbi, segno);
1150 spin_unlock(&free_i->segmap_lock);
1153 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1155 struct curseg_info *curseg = CURSEG_I(sbi, type);
1156 struct summary_footer *sum_footer;
1158 curseg->segno = curseg->next_segno;
1159 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1160 curseg->next_blkoff = 0;
1161 curseg->next_segno = NULL_SEGNO;
1163 sum_footer = &(curseg->sum_blk->footer);
1164 memset(sum_footer, 0, sizeof(struct summary_footer));
1165 if (IS_DATASEG(type))
1166 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1167 if (IS_NODESEG(type))
1168 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1169 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1173 * Allocate a current working segment.
1174 * This function always allocates a free segment in LFS manner.
1176 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1178 struct curseg_info *curseg = CURSEG_I(sbi, type);
1179 unsigned int segno = curseg->segno;
1180 int dir = ALLOC_LEFT;
1182 write_sum_page(sbi, curseg->sum_blk,
1183 GET_SUM_BLOCK(sbi, segno));
1184 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1187 if (test_opt(sbi, NOHEAP))
1190 get_new_segment(sbi, &segno, new_sec, dir);
1191 curseg->next_segno = segno;
1192 reset_curseg(sbi, type, 1);
1193 curseg->alloc_type = LFS;
1196 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1197 struct curseg_info *seg, block_t start)
1199 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1200 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1201 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1202 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1203 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1206 for (i = 0; i < entries; i++)
1207 target_map[i] = ckpt_map[i] | cur_map[i];
1209 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1211 seg->next_blkoff = pos;
1215 * If a segment is written by LFS manner, next block offset is just obtained
1216 * by increasing the current block offset. However, if a segment is written by
1217 * SSR manner, next block offset obtained by calling __next_free_blkoff
1219 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1220 struct curseg_info *seg)
1222 if (seg->alloc_type == SSR)
1223 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1229 * This function always allocates a used segment(from dirty seglist) by SSR
1230 * manner, so it should recover the existing segment information of valid blocks
1232 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1234 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1235 struct curseg_info *curseg = CURSEG_I(sbi, type);
1236 unsigned int new_segno = curseg->next_segno;
1237 struct f2fs_summary_block *sum_node;
1238 struct page *sum_page;
1240 write_sum_page(sbi, curseg->sum_blk,
1241 GET_SUM_BLOCK(sbi, curseg->segno));
1242 __set_test_and_inuse(sbi, new_segno);
1244 mutex_lock(&dirty_i->seglist_lock);
1245 __remove_dirty_segment(sbi, new_segno, PRE);
1246 __remove_dirty_segment(sbi, new_segno, DIRTY);
1247 mutex_unlock(&dirty_i->seglist_lock);
1249 reset_curseg(sbi, type, 1);
1250 curseg->alloc_type = SSR;
1251 __next_free_blkoff(sbi, curseg, 0);
1254 sum_page = get_sum_page(sbi, new_segno);
1255 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1256 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1257 f2fs_put_page(sum_page, 1);
1261 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1263 struct curseg_info *curseg = CURSEG_I(sbi, type);
1264 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1266 if (IS_NODESEG(type))
1267 return v_ops->get_victim(sbi,
1268 &(curseg)->next_segno, BG_GC, type, SSR);
1270 /* For data segments, let's do SSR more intensively */
1271 for (; type >= CURSEG_HOT_DATA; type--)
1272 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1279 * flush out current segment and replace it with new segment
1280 * This function should be returned with success, otherwise BUG
1282 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1283 int type, bool force)
1285 struct curseg_info *curseg = CURSEG_I(sbi, type);
1288 new_curseg(sbi, type, true);
1289 else if (type == CURSEG_WARM_NODE)
1290 new_curseg(sbi, type, false);
1291 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1292 new_curseg(sbi, type, false);
1293 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1294 change_curseg(sbi, type, true);
1296 new_curseg(sbi, type, false);
1298 stat_inc_seg_type(sbi, curseg);
1301 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1303 struct curseg_info *curseg = CURSEG_I(sbi, type);
1304 unsigned int old_segno;
1306 old_segno = curseg->segno;
1307 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1308 locate_dirty_segment(sbi, old_segno);
1311 void allocate_new_segments(struct f2fs_sb_info *sbi)
1315 if (test_opt(sbi, LFS))
1318 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1319 __allocate_new_segments(sbi, i);
1322 static const struct segment_allocation default_salloc_ops = {
1323 .allocate_segment = allocate_segment_by_default,
1326 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1328 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1329 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1330 unsigned int start_segno, end_segno;
1331 struct cp_control cpc;
1334 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1338 if (end <= MAIN_BLKADDR(sbi))
1341 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1342 f2fs_msg(sbi->sb, KERN_WARNING,
1343 "Found FS corruption, run fsck to fix.");
1347 /* start/end segment number in main_area */
1348 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1349 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1350 GET_SEGNO(sbi, end);
1351 cpc.reason = CP_DISCARD;
1352 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1354 /* do checkpoint to issue discard commands safely */
1355 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1356 cpc.trim_start = start_segno;
1358 if (sbi->discard_blks == 0)
1360 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1361 cpc.trim_end = end_segno;
1363 cpc.trim_end = min_t(unsigned int,
1364 rounddown(start_segno +
1365 BATCHED_TRIM_SEGMENTS(sbi),
1366 sbi->segs_per_sec) - 1, end_segno);
1368 mutex_lock(&sbi->gc_mutex);
1369 err = write_checkpoint(sbi, &cpc);
1370 mutex_unlock(&sbi->gc_mutex);
1377 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1381 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1383 struct curseg_info *curseg = CURSEG_I(sbi, type);
1384 if (curseg->next_blkoff < sbi->blocks_per_seg)
1389 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1392 return CURSEG_HOT_DATA;
1394 return CURSEG_HOT_NODE;
1397 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1399 if (p_type == DATA) {
1400 struct inode *inode = page->mapping->host;
1402 if (S_ISDIR(inode->i_mode))
1403 return CURSEG_HOT_DATA;
1405 return CURSEG_COLD_DATA;
1407 if (IS_DNODE(page) && is_cold_node(page))
1408 return CURSEG_WARM_NODE;
1410 return CURSEG_COLD_NODE;
1414 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1416 if (p_type == DATA) {
1417 struct inode *inode = page->mapping->host;
1419 if (S_ISDIR(inode->i_mode))
1420 return CURSEG_HOT_DATA;
1421 else if (is_cold_data(page) || file_is_cold(inode))
1422 return CURSEG_COLD_DATA;
1424 return CURSEG_WARM_DATA;
1427 return is_cold_node(page) ? CURSEG_WARM_NODE :
1430 return CURSEG_COLD_NODE;
1434 static int __get_segment_type(struct page *page, enum page_type p_type)
1436 switch (F2FS_P_SB(page)->active_logs) {
1438 return __get_segment_type_2(page, p_type);
1440 return __get_segment_type_4(page, p_type);
1442 /* NR_CURSEG_TYPE(6) logs by default */
1443 f2fs_bug_on(F2FS_P_SB(page),
1444 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1445 return __get_segment_type_6(page, p_type);
1448 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1449 block_t old_blkaddr, block_t *new_blkaddr,
1450 struct f2fs_summary *sum, int type)
1452 struct sit_info *sit_i = SIT_I(sbi);
1453 struct curseg_info *curseg;
1454 bool direct_io = (type == CURSEG_DIRECT_IO);
1456 type = direct_io ? CURSEG_WARM_DATA : type;
1458 curseg = CURSEG_I(sbi, type);
1460 mutex_lock(&curseg->curseg_mutex);
1461 mutex_lock(&sit_i->sentry_lock);
1463 /* direct_io'ed data is aligned to the segment for better performance */
1464 if (direct_io && curseg->next_blkoff &&
1465 !has_not_enough_free_secs(sbi, 0, 0))
1466 __allocate_new_segments(sbi, type);
1468 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1471 * __add_sum_entry should be resided under the curseg_mutex
1472 * because, this function updates a summary entry in the
1473 * current summary block.
1475 __add_sum_entry(sbi, type, sum);
1477 __refresh_next_blkoff(sbi, curseg);
1479 stat_inc_block_count(sbi, curseg);
1481 if (!__has_curseg_space(sbi, type))
1482 sit_i->s_ops->allocate_segment(sbi, type, false);
1484 * SIT information should be updated before segment allocation,
1485 * since SSR needs latest valid block information.
1487 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1489 mutex_unlock(&sit_i->sentry_lock);
1491 if (page && IS_NODESEG(type))
1492 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1494 mutex_unlock(&curseg->curseg_mutex);
1497 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1499 int type = __get_segment_type(fio->page, fio->type);
1501 if (fio->type == NODE || fio->type == DATA)
1502 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1504 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1505 &fio->new_blkaddr, sum, type);
1507 /* writeout dirty page into bdev */
1508 f2fs_submit_page_mbio(fio);
1510 if (fio->type == NODE || fio->type == DATA)
1511 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1514 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1516 struct f2fs_io_info fio = {
1520 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
1521 .old_blkaddr = page->index,
1522 .new_blkaddr = page->index,
1524 .encrypted_page = NULL,
1527 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1528 fio.op_flags &= ~REQ_META;
1530 set_page_writeback(page);
1531 f2fs_submit_page_mbio(&fio);
1534 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1536 struct f2fs_summary sum;
1538 set_summary(&sum, nid, 0, 0);
1539 do_write_page(&sum, fio);
1542 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1544 struct f2fs_sb_info *sbi = fio->sbi;
1545 struct f2fs_summary sum;
1546 struct node_info ni;
1548 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1549 get_node_info(sbi, dn->nid, &ni);
1550 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1551 do_write_page(&sum, fio);
1552 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1555 void rewrite_data_page(struct f2fs_io_info *fio)
1557 fio->new_blkaddr = fio->old_blkaddr;
1558 stat_inc_inplace_blocks(fio->sbi);
1559 f2fs_submit_page_mbio(fio);
1562 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1563 block_t old_blkaddr, block_t new_blkaddr,
1564 bool recover_curseg, bool recover_newaddr)
1566 struct sit_info *sit_i = SIT_I(sbi);
1567 struct curseg_info *curseg;
1568 unsigned int segno, old_cursegno;
1569 struct seg_entry *se;
1571 unsigned short old_blkoff;
1573 segno = GET_SEGNO(sbi, new_blkaddr);
1574 se = get_seg_entry(sbi, segno);
1577 if (!recover_curseg) {
1578 /* for recovery flow */
1579 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1580 if (old_blkaddr == NULL_ADDR)
1581 type = CURSEG_COLD_DATA;
1583 type = CURSEG_WARM_DATA;
1586 if (!IS_CURSEG(sbi, segno))
1587 type = CURSEG_WARM_DATA;
1590 curseg = CURSEG_I(sbi, type);
1592 mutex_lock(&curseg->curseg_mutex);
1593 mutex_lock(&sit_i->sentry_lock);
1595 old_cursegno = curseg->segno;
1596 old_blkoff = curseg->next_blkoff;
1598 /* change the current segment */
1599 if (segno != curseg->segno) {
1600 curseg->next_segno = segno;
1601 change_curseg(sbi, type, true);
1604 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1605 __add_sum_entry(sbi, type, sum);
1607 if (!recover_curseg || recover_newaddr)
1608 update_sit_entry(sbi, new_blkaddr, 1);
1609 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1610 update_sit_entry(sbi, old_blkaddr, -1);
1612 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1613 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1615 locate_dirty_segment(sbi, old_cursegno);
1617 if (recover_curseg) {
1618 if (old_cursegno != curseg->segno) {
1619 curseg->next_segno = old_cursegno;
1620 change_curseg(sbi, type, true);
1622 curseg->next_blkoff = old_blkoff;
1625 mutex_unlock(&sit_i->sentry_lock);
1626 mutex_unlock(&curseg->curseg_mutex);
1629 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1630 block_t old_addr, block_t new_addr,
1631 unsigned char version, bool recover_curseg,
1632 bool recover_newaddr)
1634 struct f2fs_summary sum;
1636 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1638 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1639 recover_curseg, recover_newaddr);
1641 f2fs_update_data_blkaddr(dn, new_addr);
1644 void f2fs_wait_on_page_writeback(struct page *page,
1645 enum page_type type, bool ordered)
1647 if (PageWriteback(page)) {
1648 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1650 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1652 wait_on_page_writeback(page);
1654 wait_for_stable_page(page);
1658 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1663 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1666 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1668 f2fs_wait_on_page_writeback(cpage, DATA, true);
1669 f2fs_put_page(cpage, 1);
1673 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1675 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1676 struct curseg_info *seg_i;
1677 unsigned char *kaddr;
1682 start = start_sum_block(sbi);
1684 page = get_meta_page(sbi, start++);
1685 kaddr = (unsigned char *)page_address(page);
1687 /* Step 1: restore nat cache */
1688 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1689 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1691 /* Step 2: restore sit cache */
1692 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1693 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1694 offset = 2 * SUM_JOURNAL_SIZE;
1696 /* Step 3: restore summary entries */
1697 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1698 unsigned short blk_off;
1701 seg_i = CURSEG_I(sbi, i);
1702 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1703 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1704 seg_i->next_segno = segno;
1705 reset_curseg(sbi, i, 0);
1706 seg_i->alloc_type = ckpt->alloc_type[i];
1707 seg_i->next_blkoff = blk_off;
1709 if (seg_i->alloc_type == SSR)
1710 blk_off = sbi->blocks_per_seg;
1712 for (j = 0; j < blk_off; j++) {
1713 struct f2fs_summary *s;
1714 s = (struct f2fs_summary *)(kaddr + offset);
1715 seg_i->sum_blk->entries[j] = *s;
1716 offset += SUMMARY_SIZE;
1717 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1721 f2fs_put_page(page, 1);
1724 page = get_meta_page(sbi, start++);
1725 kaddr = (unsigned char *)page_address(page);
1729 f2fs_put_page(page, 1);
1733 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1735 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1736 struct f2fs_summary_block *sum;
1737 struct curseg_info *curseg;
1739 unsigned short blk_off;
1740 unsigned int segno = 0;
1741 block_t blk_addr = 0;
1743 /* get segment number and block addr */
1744 if (IS_DATASEG(type)) {
1745 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1746 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1748 if (__exist_node_summaries(sbi))
1749 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1751 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1753 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1755 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1757 if (__exist_node_summaries(sbi))
1758 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1759 type - CURSEG_HOT_NODE);
1761 blk_addr = GET_SUM_BLOCK(sbi, segno);
1764 new = get_meta_page(sbi, blk_addr);
1765 sum = (struct f2fs_summary_block *)page_address(new);
1767 if (IS_NODESEG(type)) {
1768 if (__exist_node_summaries(sbi)) {
1769 struct f2fs_summary *ns = &sum->entries[0];
1771 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1773 ns->ofs_in_node = 0;
1778 err = restore_node_summary(sbi, segno, sum);
1780 f2fs_put_page(new, 1);
1786 /* set uncompleted segment to curseg */
1787 curseg = CURSEG_I(sbi, type);
1788 mutex_lock(&curseg->curseg_mutex);
1790 /* update journal info */
1791 down_write(&curseg->journal_rwsem);
1792 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1793 up_write(&curseg->journal_rwsem);
1795 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1796 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1797 curseg->next_segno = segno;
1798 reset_curseg(sbi, type, 0);
1799 curseg->alloc_type = ckpt->alloc_type[type];
1800 curseg->next_blkoff = blk_off;
1801 mutex_unlock(&curseg->curseg_mutex);
1802 f2fs_put_page(new, 1);
1806 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1808 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
1809 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
1810 int type = CURSEG_HOT_DATA;
1813 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1814 int npages = npages_for_summary_flush(sbi, true);
1817 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1820 /* restore for compacted data summary */
1821 if (read_compacted_summaries(sbi))
1823 type = CURSEG_HOT_NODE;
1826 if (__exist_node_summaries(sbi))
1827 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1828 NR_CURSEG_TYPE - type, META_CP, true);
1830 for (; type <= CURSEG_COLD_NODE; type++) {
1831 err = read_normal_summaries(sbi, type);
1836 /* sanity check for summary blocks */
1837 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
1838 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
1844 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1847 unsigned char *kaddr;
1848 struct f2fs_summary *summary;
1849 struct curseg_info *seg_i;
1850 int written_size = 0;
1853 page = grab_meta_page(sbi, blkaddr++);
1854 kaddr = (unsigned char *)page_address(page);
1856 /* Step 1: write nat cache */
1857 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1858 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1859 written_size += SUM_JOURNAL_SIZE;
1861 /* Step 2: write sit cache */
1862 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1863 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1864 written_size += SUM_JOURNAL_SIZE;
1866 /* Step 3: write summary entries */
1867 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1868 unsigned short blkoff;
1869 seg_i = CURSEG_I(sbi, i);
1870 if (sbi->ckpt->alloc_type[i] == SSR)
1871 blkoff = sbi->blocks_per_seg;
1873 blkoff = curseg_blkoff(sbi, i);
1875 for (j = 0; j < blkoff; j++) {
1877 page = grab_meta_page(sbi, blkaddr++);
1878 kaddr = (unsigned char *)page_address(page);
1881 summary = (struct f2fs_summary *)(kaddr + written_size);
1882 *summary = seg_i->sum_blk->entries[j];
1883 written_size += SUMMARY_SIZE;
1885 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1889 set_page_dirty(page);
1890 f2fs_put_page(page, 1);
1895 set_page_dirty(page);
1896 f2fs_put_page(page, 1);
1900 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1901 block_t blkaddr, int type)
1904 if (IS_DATASEG(type))
1905 end = type + NR_CURSEG_DATA_TYPE;
1907 end = type + NR_CURSEG_NODE_TYPE;
1909 for (i = type; i < end; i++)
1910 write_current_sum_page(sbi, i, blkaddr + (i - type));
1913 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1915 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
1916 write_compacted_summaries(sbi, start_blk);
1918 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1921 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1923 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1926 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1927 unsigned int val, int alloc)
1931 if (type == NAT_JOURNAL) {
1932 for (i = 0; i < nats_in_cursum(journal); i++) {
1933 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1936 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1937 return update_nats_in_cursum(journal, 1);
1938 } else if (type == SIT_JOURNAL) {
1939 for (i = 0; i < sits_in_cursum(journal); i++)
1940 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1942 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1943 return update_sits_in_cursum(journal, 1);
1948 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1951 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1954 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1957 struct sit_info *sit_i = SIT_I(sbi);
1958 struct page *src_page, *dst_page;
1959 pgoff_t src_off, dst_off;
1960 void *src_addr, *dst_addr;
1962 src_off = current_sit_addr(sbi, start);
1963 dst_off = next_sit_addr(sbi, src_off);
1965 /* get current sit block page without lock */
1966 src_page = get_meta_page(sbi, src_off);
1967 dst_page = grab_meta_page(sbi, dst_off);
1968 f2fs_bug_on(sbi, PageDirty(src_page));
1970 src_addr = page_address(src_page);
1971 dst_addr = page_address(dst_page);
1972 memcpy(dst_addr, src_addr, PAGE_SIZE);
1974 set_page_dirty(dst_page);
1975 f2fs_put_page(src_page, 1);
1977 set_to_next_sit(sit_i, start);
1982 static struct sit_entry_set *grab_sit_entry_set(void)
1984 struct sit_entry_set *ses =
1985 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1988 INIT_LIST_HEAD(&ses->set_list);
1992 static void release_sit_entry_set(struct sit_entry_set *ses)
1994 list_del(&ses->set_list);
1995 kmem_cache_free(sit_entry_set_slab, ses);
1998 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1999 struct list_head *head)
2001 struct sit_entry_set *next = ses;
2003 if (list_is_last(&ses->set_list, head))
2006 list_for_each_entry_continue(next, head, set_list)
2007 if (ses->entry_cnt <= next->entry_cnt)
2010 list_move_tail(&ses->set_list, &next->set_list);
2013 static void add_sit_entry(unsigned int segno, struct list_head *head)
2015 struct sit_entry_set *ses;
2016 unsigned int start_segno = START_SEGNO(segno);
2018 list_for_each_entry(ses, head, set_list) {
2019 if (ses->start_segno == start_segno) {
2021 adjust_sit_entry_set(ses, head);
2026 ses = grab_sit_entry_set();
2028 ses->start_segno = start_segno;
2030 list_add(&ses->set_list, head);
2033 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2035 struct f2fs_sm_info *sm_info = SM_I(sbi);
2036 struct list_head *set_list = &sm_info->sit_entry_set;
2037 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2040 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2041 add_sit_entry(segno, set_list);
2044 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2046 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2047 struct f2fs_journal *journal = curseg->journal;
2050 down_write(&curseg->journal_rwsem);
2051 for (i = 0; i < sits_in_cursum(journal); i++) {
2055 segno = le32_to_cpu(segno_in_journal(journal, i));
2056 dirtied = __mark_sit_entry_dirty(sbi, segno);
2059 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2061 update_sits_in_cursum(journal, -i);
2062 up_write(&curseg->journal_rwsem);
2066 * CP calls this function, which flushes SIT entries including sit_journal,
2067 * and moves prefree segs to free segs.
2069 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2071 struct sit_info *sit_i = SIT_I(sbi);
2072 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2073 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2074 struct f2fs_journal *journal = curseg->journal;
2075 struct sit_entry_set *ses, *tmp;
2076 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2077 bool to_journal = true;
2078 struct seg_entry *se;
2080 mutex_lock(&sit_i->sentry_lock);
2082 if (!sit_i->dirty_sentries)
2086 * add and account sit entries of dirty bitmap in sit entry
2089 add_sits_in_set(sbi);
2092 * if there are no enough space in journal to store dirty sit
2093 * entries, remove all entries from journal and add and account
2094 * them in sit entry set.
2096 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2097 remove_sits_in_journal(sbi);
2100 * there are two steps to flush sit entries:
2101 * #1, flush sit entries to journal in current cold data summary block.
2102 * #2, flush sit entries to sit page.
2104 list_for_each_entry_safe(ses, tmp, head, set_list) {
2105 struct page *page = NULL;
2106 struct f2fs_sit_block *raw_sit = NULL;
2107 unsigned int start_segno = ses->start_segno;
2108 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2109 (unsigned long)MAIN_SEGS(sbi));
2110 unsigned int segno = start_segno;
2113 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2117 down_write(&curseg->journal_rwsem);
2119 page = get_next_sit_page(sbi, start_segno);
2120 raw_sit = page_address(page);
2123 /* flush dirty sit entries in region of current sit set */
2124 for_each_set_bit_from(segno, bitmap, end) {
2125 int offset, sit_offset;
2127 se = get_seg_entry(sbi, segno);
2129 /* add discard candidates */
2130 if (cpc->reason != CP_DISCARD) {
2131 cpc->trim_start = segno;
2132 add_discard_addrs(sbi, cpc);
2136 offset = lookup_journal_in_cursum(journal,
2137 SIT_JOURNAL, segno, 1);
2138 f2fs_bug_on(sbi, offset < 0);
2139 segno_in_journal(journal, offset) =
2141 seg_info_to_raw_sit(se,
2142 &sit_in_journal(journal, offset));
2144 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2145 seg_info_to_raw_sit(se,
2146 &raw_sit->entries[sit_offset]);
2149 __clear_bit(segno, bitmap);
2150 sit_i->dirty_sentries--;
2155 up_write(&curseg->journal_rwsem);
2157 f2fs_put_page(page, 1);
2159 f2fs_bug_on(sbi, ses->entry_cnt);
2160 release_sit_entry_set(ses);
2163 f2fs_bug_on(sbi, !list_empty(head));
2164 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2166 if (cpc->reason == CP_DISCARD) {
2167 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2168 add_discard_addrs(sbi, cpc);
2170 mutex_unlock(&sit_i->sentry_lock);
2172 set_prefree_as_free_segments(sbi);
2175 static int build_sit_info(struct f2fs_sb_info *sbi)
2177 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2178 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2179 struct sit_info *sit_i;
2180 unsigned int sit_segs, start;
2181 char *src_bitmap, *dst_bitmap;
2182 unsigned int bitmap_size;
2184 /* allocate memory for SIT information */
2185 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2189 SM_I(sbi)->sit_info = sit_i;
2191 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2192 sizeof(struct seg_entry), GFP_KERNEL);
2193 if (!sit_i->sentries)
2196 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2197 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2198 if (!sit_i->dirty_sentries_bitmap)
2201 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2202 sit_i->sentries[start].cur_valid_map
2203 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2204 sit_i->sentries[start].ckpt_valid_map
2205 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2206 if (!sit_i->sentries[start].cur_valid_map ||
2207 !sit_i->sentries[start].ckpt_valid_map)
2210 if (f2fs_discard_en(sbi)) {
2211 sit_i->sentries[start].discard_map
2212 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2213 if (!sit_i->sentries[start].discard_map)
2218 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2219 if (!sit_i->tmp_map)
2222 if (sbi->segs_per_sec > 1) {
2223 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2224 sizeof(struct sec_entry), GFP_KERNEL);
2225 if (!sit_i->sec_entries)
2229 /* get information related with SIT */
2230 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2232 /* setup SIT bitmap from ckeckpoint pack */
2233 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2234 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2236 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2240 /* init SIT information */
2241 sit_i->s_ops = &default_salloc_ops;
2243 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2244 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2245 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2246 sit_i->sit_bitmap = dst_bitmap;
2247 sit_i->bitmap_size = bitmap_size;
2248 sit_i->dirty_sentries = 0;
2249 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2250 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2251 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2252 mutex_init(&sit_i->sentry_lock);
2256 static int build_free_segmap(struct f2fs_sb_info *sbi)
2258 struct free_segmap_info *free_i;
2259 unsigned int bitmap_size, sec_bitmap_size;
2261 /* allocate memory for free segmap information */
2262 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2266 SM_I(sbi)->free_info = free_i;
2268 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2269 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2270 if (!free_i->free_segmap)
2273 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2274 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2275 if (!free_i->free_secmap)
2278 /* set all segments as dirty temporarily */
2279 memset(free_i->free_segmap, 0xff, bitmap_size);
2280 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2282 /* init free segmap information */
2283 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2284 free_i->free_segments = 0;
2285 free_i->free_sections = 0;
2286 spin_lock_init(&free_i->segmap_lock);
2290 static int build_curseg(struct f2fs_sb_info *sbi)
2292 struct curseg_info *array;
2295 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2299 SM_I(sbi)->curseg_array = array;
2301 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2302 mutex_init(&array[i].curseg_mutex);
2303 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2304 if (!array[i].sum_blk)
2306 init_rwsem(&array[i].journal_rwsem);
2307 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2309 if (!array[i].journal)
2311 array[i].segno = NULL_SEGNO;
2312 array[i].next_blkoff = 0;
2314 return restore_curseg_summaries(sbi);
2317 static void build_sit_entries(struct f2fs_sb_info *sbi)
2319 struct sit_info *sit_i = SIT_I(sbi);
2320 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2321 struct f2fs_journal *journal = curseg->journal;
2322 struct seg_entry *se;
2323 struct f2fs_sit_entry sit;
2324 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2325 unsigned int i, start, end;
2326 unsigned int readed, start_blk = 0;
2327 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2330 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2332 start = start_blk * sit_i->sents_per_block;
2333 end = (start_blk + readed) * sit_i->sents_per_block;
2335 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2336 struct f2fs_sit_block *sit_blk;
2339 se = &sit_i->sentries[start];
2340 page = get_current_sit_page(sbi, start);
2341 sit_blk = (struct f2fs_sit_block *)page_address(page);
2342 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2343 f2fs_put_page(page, 1);
2345 check_block_count(sbi, start, &sit);
2346 seg_info_from_raw_sit(se, &sit);
2348 /* build discard map only one time */
2349 if (f2fs_discard_en(sbi)) {
2350 memcpy(se->discard_map, se->cur_valid_map,
2351 SIT_VBLOCK_MAP_SIZE);
2352 sbi->discard_blks += sbi->blocks_per_seg -
2356 if (sbi->segs_per_sec > 1)
2357 get_sec_entry(sbi, start)->valid_blocks +=
2360 start_blk += readed;
2361 } while (start_blk < sit_blk_cnt);
2363 down_read(&curseg->journal_rwsem);
2364 for (i = 0; i < sits_in_cursum(journal); i++) {
2365 unsigned int old_valid_blocks;
2367 start = le32_to_cpu(segno_in_journal(journal, i));
2368 se = &sit_i->sentries[start];
2369 sit = sit_in_journal(journal, i);
2371 old_valid_blocks = se->valid_blocks;
2373 check_block_count(sbi, start, &sit);
2374 seg_info_from_raw_sit(se, &sit);
2376 if (f2fs_discard_en(sbi)) {
2377 memcpy(se->discard_map, se->cur_valid_map,
2378 SIT_VBLOCK_MAP_SIZE);
2379 sbi->discard_blks += old_valid_blocks -
2383 if (sbi->segs_per_sec > 1)
2384 get_sec_entry(sbi, start)->valid_blocks +=
2385 se->valid_blocks - old_valid_blocks;
2387 up_read(&curseg->journal_rwsem);
2390 static void init_free_segmap(struct f2fs_sb_info *sbi)
2395 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2396 struct seg_entry *sentry = get_seg_entry(sbi, start);
2397 if (!sentry->valid_blocks)
2398 __set_free(sbi, start);
2401 /* set use the current segments */
2402 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2403 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2404 __set_test_and_inuse(sbi, curseg_t->segno);
2408 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2410 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2411 struct free_segmap_info *free_i = FREE_I(sbi);
2412 unsigned int segno = 0, offset = 0;
2413 unsigned short valid_blocks;
2416 /* find dirty segment based on free segmap */
2417 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2418 if (segno >= MAIN_SEGS(sbi))
2421 valid_blocks = get_valid_blocks(sbi, segno, 0);
2422 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2424 if (valid_blocks > sbi->blocks_per_seg) {
2425 f2fs_bug_on(sbi, 1);
2428 mutex_lock(&dirty_i->seglist_lock);
2429 __locate_dirty_segment(sbi, segno, DIRTY);
2430 mutex_unlock(&dirty_i->seglist_lock);
2434 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2436 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2437 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2439 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2440 if (!dirty_i->victim_secmap)
2445 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2447 struct dirty_seglist_info *dirty_i;
2448 unsigned int bitmap_size, i;
2450 /* allocate memory for dirty segments list information */
2451 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2455 SM_I(sbi)->dirty_info = dirty_i;
2456 mutex_init(&dirty_i->seglist_lock);
2458 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2460 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2461 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2462 if (!dirty_i->dirty_segmap[i])
2466 init_dirty_segmap(sbi);
2467 return init_victim_secmap(sbi);
2471 * Update min, max modified time for cost-benefit GC algorithm
2473 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2475 struct sit_info *sit_i = SIT_I(sbi);
2478 mutex_lock(&sit_i->sentry_lock);
2480 sit_i->min_mtime = LLONG_MAX;
2482 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2484 unsigned long long mtime = 0;
2486 for (i = 0; i < sbi->segs_per_sec; i++)
2487 mtime += get_seg_entry(sbi, segno + i)->mtime;
2489 mtime = div_u64(mtime, sbi->segs_per_sec);
2491 if (sit_i->min_mtime > mtime)
2492 sit_i->min_mtime = mtime;
2494 sit_i->max_mtime = get_mtime(sbi);
2495 mutex_unlock(&sit_i->sentry_lock);
2498 int build_segment_manager(struct f2fs_sb_info *sbi)
2500 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2501 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2502 struct f2fs_sm_info *sm_info;
2505 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2510 sbi->sm_info = sm_info;
2511 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2512 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2513 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2514 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2515 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2516 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2517 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2518 sm_info->rec_prefree_segments = sm_info->main_segments *
2519 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2520 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2521 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2523 if (!test_opt(sbi, LFS))
2524 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2525 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2526 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2528 INIT_LIST_HEAD(&sm_info->discard_list);
2529 INIT_LIST_HEAD(&sm_info->wait_list);
2530 sm_info->nr_discards = 0;
2531 sm_info->max_discards = 0;
2533 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2535 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2537 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2538 err = create_flush_cmd_control(sbi);
2543 err = build_sit_info(sbi);
2546 err = build_free_segmap(sbi);
2549 err = build_curseg(sbi);
2553 /* reinit free segmap based on SIT */
2554 build_sit_entries(sbi);
2556 init_free_segmap(sbi);
2557 err = build_dirty_segmap(sbi);
2561 init_min_max_mtime(sbi);
2565 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2566 enum dirty_type dirty_type)
2568 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2570 mutex_lock(&dirty_i->seglist_lock);
2571 kvfree(dirty_i->dirty_segmap[dirty_type]);
2572 dirty_i->nr_dirty[dirty_type] = 0;
2573 mutex_unlock(&dirty_i->seglist_lock);
2576 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2578 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2579 kvfree(dirty_i->victim_secmap);
2582 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2584 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2590 /* discard pre-free/dirty segments list */
2591 for (i = 0; i < NR_DIRTY_TYPE; i++)
2592 discard_dirty_segmap(sbi, i);
2594 destroy_victim_secmap(sbi);
2595 SM_I(sbi)->dirty_info = NULL;
2599 static void destroy_curseg(struct f2fs_sb_info *sbi)
2601 struct curseg_info *array = SM_I(sbi)->curseg_array;
2606 SM_I(sbi)->curseg_array = NULL;
2607 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2608 kfree(array[i].sum_blk);
2609 kfree(array[i].journal);
2614 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2616 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2619 SM_I(sbi)->free_info = NULL;
2620 kvfree(free_i->free_segmap);
2621 kvfree(free_i->free_secmap);
2625 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2627 struct sit_info *sit_i = SIT_I(sbi);
2633 if (sit_i->sentries) {
2634 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2635 kfree(sit_i->sentries[start].cur_valid_map);
2636 kfree(sit_i->sentries[start].ckpt_valid_map);
2637 kfree(sit_i->sentries[start].discard_map);
2640 kfree(sit_i->tmp_map);
2642 kvfree(sit_i->sentries);
2643 kvfree(sit_i->sec_entries);
2644 kvfree(sit_i->dirty_sentries_bitmap);
2646 SM_I(sbi)->sit_info = NULL;
2647 kfree(sit_i->sit_bitmap);
2651 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2653 struct f2fs_sm_info *sm_info = SM_I(sbi);
2657 destroy_flush_cmd_control(sbi);
2658 destroy_dirty_segmap(sbi);
2659 destroy_curseg(sbi);
2660 destroy_free_segmap(sbi);
2661 destroy_sit_info(sbi);
2662 sbi->sm_info = NULL;
2666 int __init create_segment_manager_caches(void)
2668 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2669 sizeof(struct discard_entry));
2670 if (!discard_entry_slab)
2673 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2674 sizeof(struct bio_entry));
2675 if (!bio_entry_slab)
2676 goto destroy_discard_entry;
2678 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2679 sizeof(struct sit_entry_set));
2680 if (!sit_entry_set_slab)
2681 goto destroy_bio_entry;
2683 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2684 sizeof(struct inmem_pages));
2685 if (!inmem_entry_slab)
2686 goto destroy_sit_entry_set;
2689 destroy_sit_entry_set:
2690 kmem_cache_destroy(sit_entry_set_slab);
2692 kmem_cache_destroy(bio_entry_slab);
2693 destroy_discard_entry:
2694 kmem_cache_destroy(discard_entry_slab);
2699 void destroy_segment_manager_caches(void)
2701 kmem_cache_destroy(sit_entry_set_slab);
2702 kmem_cache_destroy(bio_entry_slab);
2703 kmem_cache_destroy(discard_entry_slab);
2704 kmem_cache_destroy(inmem_entry_slab);