1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched.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 *discard_cmd_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 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
175 if (test_opt(sbi, LFS))
177 if (sbi->gc_mode == GC_URGENT)
179 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
182 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
183 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
188 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
189 struct f2fs_inode_info *fi = F2FS_I(inode);
190 struct inmem_pages *new;
192 f2fs_trace_pid(page);
194 f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
198 /* add atomic page indices to the list */
200 INIT_LIST_HEAD(&new->list);
202 /* increase reference count with clean state */
203 mutex_lock(&fi->inmem_lock);
205 list_add_tail(&new->list, &fi->inmem_pages);
206 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
207 if (list_empty(&fi->inmem_ilist))
208 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
209 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
210 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
211 mutex_unlock(&fi->inmem_lock);
213 trace_f2fs_register_inmem_page(page, INMEM);
216 static int __revoke_inmem_pages(struct inode *inode,
217 struct list_head *head, bool drop, bool recover,
220 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
221 struct inmem_pages *cur, *tmp;
224 list_for_each_entry_safe(cur, tmp, head, list) {
225 struct page *page = cur->page;
228 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
232 * to avoid deadlock in between page lock and
235 if (!trylock_page(page))
241 f2fs_wait_on_page_writeback(page, DATA, true, true);
244 struct dnode_of_data dn;
247 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
249 set_new_dnode(&dn, inode, NULL, NULL, 0);
250 err = f2fs_get_dnode_of_data(&dn, page->index,
253 if (err == -ENOMEM) {
254 congestion_wait(BLK_RW_ASYNC, HZ/50);
262 err = f2fs_get_node_info(sbi, dn.nid, &ni);
268 if (cur->old_addr == NEW_ADDR) {
269 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
270 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
272 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
273 cur->old_addr, ni.version, true, true);
277 /* we don't need to invalidate this in the sccessful status */
278 if (drop || recover) {
279 ClearPageUptodate(page);
280 clear_cold_data(page);
282 f2fs_clear_page_private(page);
283 f2fs_put_page(page, 1);
285 list_del(&cur->list);
286 kmem_cache_free(inmem_entry_slab, cur);
287 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
292 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
294 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
296 struct f2fs_inode_info *fi;
298 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
299 if (list_empty(head)) {
300 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
303 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
304 inode = igrab(&fi->vfs_inode);
305 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
309 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
314 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
315 f2fs_drop_inmem_pages(inode);
319 congestion_wait(BLK_RW_ASYNC, HZ/50);
324 void f2fs_drop_inmem_pages(struct inode *inode)
326 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
327 struct f2fs_inode_info *fi = F2FS_I(inode);
329 while (!list_empty(&fi->inmem_pages)) {
330 mutex_lock(&fi->inmem_lock);
331 __revoke_inmem_pages(inode, &fi->inmem_pages,
334 if (list_empty(&fi->inmem_pages)) {
335 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
336 if (!list_empty(&fi->inmem_ilist))
337 list_del_init(&fi->inmem_ilist);
338 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
340 mutex_unlock(&fi->inmem_lock);
343 clear_inode_flag(inode, FI_ATOMIC_FILE);
344 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
345 stat_dec_atomic_write(inode);
348 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
350 struct f2fs_inode_info *fi = F2FS_I(inode);
351 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
352 struct list_head *head = &fi->inmem_pages;
353 struct inmem_pages *cur = NULL;
355 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
357 mutex_lock(&fi->inmem_lock);
358 list_for_each_entry(cur, head, list) {
359 if (cur->page == page)
363 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
364 list_del(&cur->list);
365 mutex_unlock(&fi->inmem_lock);
367 dec_page_count(sbi, F2FS_INMEM_PAGES);
368 kmem_cache_free(inmem_entry_slab, cur);
370 ClearPageUptodate(page);
371 f2fs_clear_page_private(page);
372 f2fs_put_page(page, 0);
374 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
377 static int __f2fs_commit_inmem_pages(struct inode *inode)
379 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
380 struct f2fs_inode_info *fi = F2FS_I(inode);
381 struct inmem_pages *cur, *tmp;
382 struct f2fs_io_info fio = {
387 .op_flags = REQ_SYNC | REQ_PRIO,
388 .io_type = FS_DATA_IO,
390 struct list_head revoke_list;
391 bool submit_bio = false;
394 INIT_LIST_HEAD(&revoke_list);
396 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
397 struct page *page = cur->page;
400 if (page->mapping == inode->i_mapping) {
401 trace_f2fs_commit_inmem_page(page, INMEM);
403 f2fs_wait_on_page_writeback(page, DATA, true, true);
405 set_page_dirty(page);
406 if (clear_page_dirty_for_io(page)) {
407 inode_dec_dirty_pages(inode);
408 f2fs_remove_dirty_inode(inode);
412 fio.old_blkaddr = NULL_ADDR;
413 fio.encrypted_page = NULL;
414 fio.need_lock = LOCK_DONE;
415 err = f2fs_do_write_data_page(&fio);
417 if (err == -ENOMEM) {
418 congestion_wait(BLK_RW_ASYNC, HZ/50);
425 /* record old blkaddr for revoking */
426 cur->old_addr = fio.old_blkaddr;
430 list_move_tail(&cur->list, &revoke_list);
434 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
438 * try to revoke all committed pages, but still we could fail
439 * due to no memory or other reason, if that happened, EAGAIN
440 * will be returned, which means in such case, transaction is
441 * already not integrity, caller should use journal to do the
442 * recovery or rewrite & commit last transaction. For other
443 * error number, revoking was done by filesystem itself.
445 err = __revoke_inmem_pages(inode, &revoke_list,
448 /* drop all uncommitted pages */
449 __revoke_inmem_pages(inode, &fi->inmem_pages,
452 __revoke_inmem_pages(inode, &revoke_list,
453 false, false, false);
459 int f2fs_commit_inmem_pages(struct inode *inode)
461 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
462 struct f2fs_inode_info *fi = F2FS_I(inode);
465 f2fs_balance_fs(sbi, true);
467 down_write(&fi->i_gc_rwsem[WRITE]);
470 set_inode_flag(inode, FI_ATOMIC_COMMIT);
472 mutex_lock(&fi->inmem_lock);
473 err = __f2fs_commit_inmem_pages(inode);
475 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
476 if (!list_empty(&fi->inmem_ilist))
477 list_del_init(&fi->inmem_ilist);
478 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
479 mutex_unlock(&fi->inmem_lock);
481 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
484 up_write(&fi->i_gc_rwsem[WRITE]);
490 * This function balances dirty node and dentry pages.
491 * In addition, it controls garbage collection.
493 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
495 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
496 f2fs_show_injection_info(FAULT_CHECKPOINT);
497 f2fs_stop_checkpoint(sbi, false);
500 /* balance_fs_bg is able to be pending */
501 if (need && excess_cached_nats(sbi))
502 f2fs_balance_fs_bg(sbi);
504 if (f2fs_is_checkpoint_ready(sbi))
508 * We should do GC or end up with checkpoint, if there are so many dirty
509 * dir/node pages without enough free segments.
511 if (has_not_enough_free_secs(sbi, 0, 0)) {
512 mutex_lock(&sbi->gc_mutex);
513 f2fs_gc(sbi, false, false, NULL_SEGNO);
517 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
519 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
522 /* try to shrink extent cache when there is no enough memory */
523 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
524 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
526 /* check the # of cached NAT entries */
527 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
528 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
530 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
531 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
533 f2fs_build_free_nids(sbi, false, false);
535 if (!is_idle(sbi, REQ_TIME) &&
536 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
539 /* checkpoint is the only way to shrink partial cached entries */
540 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
541 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
542 excess_prefree_segs(sbi) ||
543 excess_dirty_nats(sbi) ||
544 excess_dirty_nodes(sbi) ||
545 f2fs_time_over(sbi, CP_TIME)) {
546 if (test_opt(sbi, DATA_FLUSH)) {
547 struct blk_plug plug;
549 blk_start_plug(&plug);
550 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
551 blk_finish_plug(&plug);
553 f2fs_sync_fs(sbi->sb, true);
554 stat_inc_bg_cp_count(sbi->stat_info);
558 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
559 struct block_device *bdev)
564 bio = f2fs_bio_alloc(sbi, 0, false);
568 bio->bi_rw = REQ_OP_WRITE;
570 ret = submit_bio_wait(WRITE_FLUSH, bio);
573 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
574 test_opt(sbi, FLUSH_MERGE), ret);
578 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
583 if (!f2fs_is_multi_device(sbi))
584 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
586 for (i = 0; i < sbi->s_ndevs; i++) {
587 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
589 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
596 static int issue_flush_thread(void *data)
598 struct f2fs_sb_info *sbi = data;
599 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
600 wait_queue_head_t *q = &fcc->flush_wait_queue;
602 if (kthread_should_stop())
605 sb_start_intwrite(sbi->sb);
607 if (!llist_empty(&fcc->issue_list)) {
608 struct flush_cmd *cmd, *next;
611 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
612 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
614 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
616 ret = submit_flush_wait(sbi, cmd->ino);
617 atomic_inc(&fcc->issued_flush);
619 llist_for_each_entry_safe(cmd, next,
620 fcc->dispatch_list, llnode) {
622 complete(&cmd->wait);
624 fcc->dispatch_list = NULL;
627 sb_end_intwrite(sbi->sb);
629 wait_event_interruptible(*q,
630 kthread_should_stop() || !llist_empty(&fcc->issue_list));
634 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
636 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
637 struct flush_cmd cmd;
640 if (test_opt(sbi, NOBARRIER))
643 if (!test_opt(sbi, FLUSH_MERGE)) {
644 atomic_inc(&fcc->queued_flush);
645 ret = submit_flush_wait(sbi, ino);
646 atomic_dec(&fcc->queued_flush);
647 atomic_inc(&fcc->issued_flush);
651 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
652 f2fs_is_multi_device(sbi)) {
653 ret = submit_flush_wait(sbi, ino);
654 atomic_dec(&fcc->queued_flush);
656 atomic_inc(&fcc->issued_flush);
661 init_completion(&cmd.wait);
663 llist_add(&cmd.llnode, &fcc->issue_list);
665 /* update issue_list before we wake up issue_flush thread */
668 if (waitqueue_active(&fcc->flush_wait_queue))
669 wake_up(&fcc->flush_wait_queue);
671 if (fcc->f2fs_issue_flush) {
672 wait_for_completion(&cmd.wait);
673 atomic_dec(&fcc->queued_flush);
675 struct llist_node *list;
677 list = llist_del_all(&fcc->issue_list);
679 wait_for_completion(&cmd.wait);
680 atomic_dec(&fcc->queued_flush);
682 struct flush_cmd *tmp, *next;
684 ret = submit_flush_wait(sbi, ino);
686 llist_for_each_entry_safe(tmp, next, list, llnode) {
689 atomic_dec(&fcc->queued_flush);
693 complete(&tmp->wait);
701 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
703 dev_t dev = sbi->sb->s_bdev->bd_dev;
704 struct flush_cmd_control *fcc;
707 if (SM_I(sbi)->fcc_info) {
708 fcc = SM_I(sbi)->fcc_info;
709 if (fcc->f2fs_issue_flush)
714 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
717 atomic_set(&fcc->issued_flush, 0);
718 atomic_set(&fcc->queued_flush, 0);
719 init_waitqueue_head(&fcc->flush_wait_queue);
720 init_llist_head(&fcc->issue_list);
721 SM_I(sbi)->fcc_info = fcc;
722 if (!test_opt(sbi, FLUSH_MERGE))
726 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
727 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
728 if (IS_ERR(fcc->f2fs_issue_flush)) {
729 err = PTR_ERR(fcc->f2fs_issue_flush);
731 SM_I(sbi)->fcc_info = NULL;
738 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
740 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
742 if (fcc && fcc->f2fs_issue_flush) {
743 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
745 fcc->f2fs_issue_flush = NULL;
746 kthread_stop(flush_thread);
750 SM_I(sbi)->fcc_info = NULL;
754 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
758 if (!f2fs_is_multi_device(sbi))
761 for (i = 1; i < sbi->s_ndevs; i++) {
762 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
764 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
768 spin_lock(&sbi->dev_lock);
769 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
770 spin_unlock(&sbi->dev_lock);
776 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
777 enum dirty_type dirty_type)
779 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
781 /* need not be added */
782 if (IS_CURSEG(sbi, segno))
785 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
786 dirty_i->nr_dirty[dirty_type]++;
788 if (dirty_type == DIRTY) {
789 struct seg_entry *sentry = get_seg_entry(sbi, segno);
790 enum dirty_type t = sentry->type;
792 if (unlikely(t >= DIRTY)) {
796 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
797 dirty_i->nr_dirty[t]++;
801 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
802 enum dirty_type dirty_type)
804 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
806 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
807 dirty_i->nr_dirty[dirty_type]--;
809 if (dirty_type == DIRTY) {
810 struct seg_entry *sentry = get_seg_entry(sbi, segno);
811 enum dirty_type t = sentry->type;
813 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
814 dirty_i->nr_dirty[t]--;
816 if (get_valid_blocks(sbi, segno, true) == 0)
817 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
818 dirty_i->victim_secmap);
823 * Should not occur error such as -ENOMEM.
824 * Adding dirty entry into seglist is not critical operation.
825 * If a given segment is one of current working segments, it won't be added.
827 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
829 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
830 unsigned short valid_blocks, ckpt_valid_blocks;
832 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
835 mutex_lock(&dirty_i->seglist_lock);
837 valid_blocks = get_valid_blocks(sbi, segno, false);
838 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
840 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
841 ckpt_valid_blocks == sbi->blocks_per_seg)) {
842 __locate_dirty_segment(sbi, segno, PRE);
843 __remove_dirty_segment(sbi, segno, DIRTY);
844 } else if (valid_blocks < sbi->blocks_per_seg) {
845 __locate_dirty_segment(sbi, segno, DIRTY);
847 /* Recovery routine with SSR needs this */
848 __remove_dirty_segment(sbi, segno, DIRTY);
851 mutex_unlock(&dirty_i->seglist_lock);
854 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
855 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
857 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
860 mutex_lock(&dirty_i->seglist_lock);
861 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
862 if (get_valid_blocks(sbi, segno, false))
864 if (IS_CURSEG(sbi, segno))
866 __locate_dirty_segment(sbi, segno, PRE);
867 __remove_dirty_segment(sbi, segno, DIRTY);
869 mutex_unlock(&dirty_i->seglist_lock);
872 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi)
874 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
875 block_t ovp = overprovision_segments(sbi) << sbi->log_blocks_per_seg;
876 block_t holes[2] = {0, 0}; /* DATA and NODE */
877 struct seg_entry *se;
880 mutex_lock(&dirty_i->seglist_lock);
881 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
882 se = get_seg_entry(sbi, segno);
883 if (IS_NODESEG(se->type))
884 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
886 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
888 mutex_unlock(&dirty_i->seglist_lock);
890 if (holes[DATA] > ovp || holes[NODE] > ovp)
892 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
893 dirty_segments(sbi) > overprovision_segments(sbi))
898 /* This is only used by SBI_CP_DISABLED */
899 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
901 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
902 unsigned int segno = 0;
904 mutex_lock(&dirty_i->seglist_lock);
905 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
906 if (get_valid_blocks(sbi, segno, false))
908 if (get_ckpt_valid_blocks(sbi, segno))
910 mutex_unlock(&dirty_i->seglist_lock);
913 mutex_unlock(&dirty_i->seglist_lock);
917 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
918 struct block_device *bdev, block_t lstart,
919 block_t start, block_t len)
921 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
922 struct list_head *pend_list;
923 struct discard_cmd *dc;
925 f2fs_bug_on(sbi, !len);
927 pend_list = &dcc->pend_list[plist_idx(len)];
929 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
930 INIT_LIST_HEAD(&dc->list);
939 init_completion(&dc->wait);
940 list_add_tail(&dc->list, pend_list);
941 spin_lock_init(&dc->lock);
943 atomic_inc(&dcc->discard_cmd_cnt);
944 dcc->undiscard_blks += len;
949 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
950 struct block_device *bdev, block_t lstart,
951 block_t start, block_t len,
952 struct rb_node *parent, struct rb_node **p)
954 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
955 struct discard_cmd *dc;
957 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
959 rb_link_node(&dc->rb_node, parent, p);
960 rb_insert_color(&dc->rb_node, &dcc->root);
965 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
966 struct discard_cmd *dc)
968 if (dc->state == D_DONE)
969 atomic_sub(dc->queued, &dcc->queued_discard);
972 rb_erase(&dc->rb_node, &dcc->root);
973 dcc->undiscard_blks -= dc->len;
975 kmem_cache_free(discard_cmd_slab, dc);
977 atomic_dec(&dcc->discard_cmd_cnt);
980 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
981 struct discard_cmd *dc)
983 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
986 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
988 spin_lock_irqsave(&dc->lock, flags);
990 spin_unlock_irqrestore(&dc->lock, flags);
993 spin_unlock_irqrestore(&dc->lock, flags);
995 f2fs_bug_on(sbi, dc->ref);
997 if (dc->error == -EOPNOTSUPP)
1002 "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1003 KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1004 __detach_discard_cmd(dcc, dc);
1007 static void f2fs_submit_discard_endio(struct bio *bio)
1009 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1010 unsigned long flags;
1012 dc->error = bio->bi_error;
1014 spin_lock_irqsave(&dc->lock, flags);
1016 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1018 complete_all(&dc->wait);
1020 spin_unlock_irqrestore(&dc->lock, flags);
1024 /* copied from block/blk-lib.c in 4.10-rc1 */
1025 static int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
1026 sector_t nr_sects, gfp_t gfp_mask, int flags,
1029 struct request_queue *q = bdev_get_queue(bdev);
1030 struct bio *bio = *biop;
1031 unsigned int granularity;
1032 int op = REQ_WRITE | REQ_DISCARD;
1039 if (!blk_queue_discard(q))
1042 if (flags & BLKDEV_DISCARD_SECURE) {
1043 if (!blk_queue_secdiscard(q))
1048 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
1049 if ((sector | nr_sects) & bs_mask)
1052 /* Zero-sector (unknown) and one-sector granularities are the same. */
1053 granularity = max(q->limits.discard_granularity >> 9, 1U);
1054 alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
1057 unsigned int req_sects;
1058 sector_t end_sect, tmp;
1060 /* Make sure bi_size doesn't overflow */
1061 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
1064 * If splitting a request, and the next starting sector would be
1065 * misaligned, stop the discard at the previous aligned sector.
1067 end_sect = sector + req_sects;
1069 if (req_sects < nr_sects &&
1070 sector_div(tmp, granularity) != alignment) {
1071 end_sect = end_sect - alignment;
1072 sector_div(end_sect, granularity);
1073 end_sect = end_sect * granularity + alignment;
1074 req_sects = end_sect - sector;
1078 int ret = submit_bio_wait(op, bio);
1083 bio = bio_alloc(GFP_NOIO | __GFP_NOFAIL, 1);
1084 bio->bi_iter.bi_sector = sector;
1085 bio->bi_bdev = bdev;
1086 bio_set_op_attrs(bio, op, 0);
1088 bio->bi_iter.bi_size = req_sects << 9;
1089 nr_sects -= req_sects;
1093 * We can loop for a long time in here, if someone does
1094 * full device discards (like mkfs). Be nice and allow
1095 * us to schedule out to avoid softlocking if preempt
1105 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1106 block_t start, block_t end)
1108 #ifdef CONFIG_F2FS_CHECK_FS
1109 struct seg_entry *sentry;
1111 block_t blk = start;
1112 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1116 segno = GET_SEGNO(sbi, blk);
1117 sentry = get_seg_entry(sbi, segno);
1118 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1120 if (end < START_BLOCK(sbi, segno + 1))
1121 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1124 map = (unsigned long *)(sentry->cur_valid_map);
1125 offset = __find_rev_next_bit(map, size, offset);
1126 f2fs_bug_on(sbi, offset != size);
1127 blk = START_BLOCK(sbi, segno + 1);
1132 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1133 struct discard_policy *dpolicy,
1134 int discard_type, unsigned int granularity)
1137 dpolicy->type = discard_type;
1138 dpolicy->sync = true;
1139 dpolicy->ordered = false;
1140 dpolicy->granularity = granularity;
1142 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1143 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1144 dpolicy->timeout = 0;
1146 if (discard_type == DPOLICY_BG) {
1147 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1148 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1149 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1150 dpolicy->io_aware = true;
1151 dpolicy->sync = false;
1152 dpolicy->ordered = true;
1153 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1154 dpolicy->granularity = 1;
1155 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1157 } else if (discard_type == DPOLICY_FORCE) {
1158 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1159 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1160 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1161 dpolicy->io_aware = false;
1162 } else if (discard_type == DPOLICY_FSTRIM) {
1163 dpolicy->io_aware = false;
1164 } else if (discard_type == DPOLICY_UMOUNT) {
1165 dpolicy->max_requests = UINT_MAX;
1166 dpolicy->io_aware = false;
1167 /* we need to issue all to keep CP_TRIMMED_FLAG */
1168 dpolicy->granularity = 1;
1172 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1173 struct block_device *bdev, block_t lstart,
1174 block_t start, block_t len);
1175 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1176 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1177 struct discard_policy *dpolicy,
1178 struct discard_cmd *dc,
1179 unsigned int *issued)
1181 struct block_device *bdev = dc->bdev;
1182 struct request_queue *q = bdev_get_queue(bdev);
1183 unsigned int max_discard_blocks =
1184 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1185 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1186 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1187 &(dcc->fstrim_list) : &(dcc->wait_list);
1188 int flag = dpolicy->sync ? REQ_SYNC : 0;
1189 block_t lstart, start, len, total_len;
1192 if (dc->state != D_PREP)
1195 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1198 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1200 lstart = dc->lstart;
1207 while (total_len && *issued < dpolicy->max_requests && !err) {
1208 struct bio *bio = NULL;
1209 unsigned long flags;
1212 if (len > max_discard_blocks) {
1213 len = max_discard_blocks;
1218 if (*issued == dpolicy->max_requests)
1223 if (time_to_inject(sbi, FAULT_DISCARD)) {
1224 f2fs_show_injection_info(FAULT_DISCARD);
1228 err = __blkdev_issue_discard(bdev,
1229 SECTOR_FROM_BLOCK(start),
1230 SECTOR_FROM_BLOCK(len),
1234 spin_lock_irqsave(&dc->lock, flags);
1235 if (dc->state == D_PARTIAL)
1236 dc->state = D_SUBMIT;
1237 spin_unlock_irqrestore(&dc->lock, flags);
1242 f2fs_bug_on(sbi, !bio);
1245 * should keep before submission to avoid D_DONE
1248 spin_lock_irqsave(&dc->lock, flags);
1250 dc->state = D_SUBMIT;
1252 dc->state = D_PARTIAL;
1254 spin_unlock_irqrestore(&dc->lock, flags);
1256 atomic_inc(&dcc->queued_discard);
1258 list_move_tail(&dc->list, wait_list);
1260 /* sanity check on discard range */
1261 __check_sit_bitmap(sbi, lstart, lstart + len);
1263 bio->bi_private = dc;
1264 bio->bi_end_io = f2fs_submit_discard_endio;
1265 submit_bio(flag, bio);
1267 atomic_inc(&dcc->issued_discard);
1269 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1278 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1282 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1283 struct block_device *bdev, block_t lstart,
1284 block_t start, block_t len,
1285 struct rb_node **insert_p,
1286 struct rb_node *insert_parent)
1288 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1290 struct rb_node *parent = NULL;
1291 struct discard_cmd *dc = NULL;
1293 if (insert_p && insert_parent) {
1294 parent = insert_parent;
1299 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1301 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1308 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1309 struct discard_cmd *dc)
1311 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1314 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1315 struct discard_cmd *dc, block_t blkaddr)
1317 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1318 struct discard_info di = dc->di;
1319 bool modified = false;
1321 if (dc->state == D_DONE || dc->len == 1) {
1322 __remove_discard_cmd(sbi, dc);
1326 dcc->undiscard_blks -= di.len;
1328 if (blkaddr > di.lstart) {
1329 dc->len = blkaddr - dc->lstart;
1330 dcc->undiscard_blks += dc->len;
1331 __relocate_discard_cmd(dcc, dc);
1335 if (blkaddr < di.lstart + di.len - 1) {
1337 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1338 di.start + blkaddr + 1 - di.lstart,
1339 di.lstart + di.len - 1 - blkaddr,
1345 dcc->undiscard_blks += dc->len;
1346 __relocate_discard_cmd(dcc, dc);
1351 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1352 struct block_device *bdev, block_t lstart,
1353 block_t start, block_t len)
1355 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1356 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1357 struct discard_cmd *dc;
1358 struct discard_info di = {0};
1359 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1360 struct request_queue *q = bdev_get_queue(bdev);
1361 unsigned int max_discard_blocks =
1362 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1363 block_t end = lstart + len;
1365 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1367 (struct rb_entry **)&prev_dc,
1368 (struct rb_entry **)&next_dc,
1369 &insert_p, &insert_parent, true);
1375 di.len = next_dc ? next_dc->lstart - lstart : len;
1376 di.len = min(di.len, len);
1381 struct rb_node *node;
1382 bool merged = false;
1383 struct discard_cmd *tdc = NULL;
1386 di.lstart = prev_dc->lstart + prev_dc->len;
1387 if (di.lstart < lstart)
1389 if (di.lstart >= end)
1392 if (!next_dc || next_dc->lstart > end)
1393 di.len = end - di.lstart;
1395 di.len = next_dc->lstart - di.lstart;
1396 di.start = start + di.lstart - lstart;
1402 if (prev_dc && prev_dc->state == D_PREP &&
1403 prev_dc->bdev == bdev &&
1404 __is_discard_back_mergeable(&di, &prev_dc->di,
1405 max_discard_blocks)) {
1406 prev_dc->di.len += di.len;
1407 dcc->undiscard_blks += di.len;
1408 __relocate_discard_cmd(dcc, prev_dc);
1414 if (next_dc && next_dc->state == D_PREP &&
1415 next_dc->bdev == bdev &&
1416 __is_discard_front_mergeable(&di, &next_dc->di,
1417 max_discard_blocks)) {
1418 next_dc->di.lstart = di.lstart;
1419 next_dc->di.len += di.len;
1420 next_dc->di.start = di.start;
1421 dcc->undiscard_blks += di.len;
1422 __relocate_discard_cmd(dcc, next_dc);
1424 __remove_discard_cmd(sbi, tdc);
1429 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1430 di.len, NULL, NULL);
1437 node = rb_next(&prev_dc->rb_node);
1438 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1442 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1443 struct block_device *bdev, block_t blkstart, block_t blklen)
1445 block_t lblkstart = blkstart;
1447 if (!f2fs_bdev_support_discard(bdev))
1450 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1452 if (f2fs_is_multi_device(sbi)) {
1453 int devi = f2fs_target_device_index(sbi, blkstart);
1455 blkstart -= FDEV(devi).start_blk;
1457 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1458 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1459 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1463 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1464 struct discard_policy *dpolicy)
1466 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1467 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1468 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1469 struct discard_cmd *dc;
1470 struct blk_plug plug;
1471 unsigned int pos = dcc->next_pos;
1472 unsigned int issued = 0;
1473 bool io_interrupted = false;
1475 mutex_lock(&dcc->cmd_lock);
1476 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1478 (struct rb_entry **)&prev_dc,
1479 (struct rb_entry **)&next_dc,
1480 &insert_p, &insert_parent, true);
1484 blk_start_plug(&plug);
1487 struct rb_node *node;
1490 if (dc->state != D_PREP)
1493 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1494 io_interrupted = true;
1498 dcc->next_pos = dc->lstart + dc->len;
1499 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1501 if (issued >= dpolicy->max_requests)
1504 node = rb_next(&dc->rb_node);
1506 __remove_discard_cmd(sbi, dc);
1507 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1510 blk_finish_plug(&plug);
1515 mutex_unlock(&dcc->cmd_lock);
1517 if (!issued && io_interrupted)
1523 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1524 struct discard_policy *dpolicy)
1526 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1527 struct list_head *pend_list;
1528 struct discard_cmd *dc, *tmp;
1529 struct blk_plug plug;
1531 bool io_interrupted = false;
1533 if (dpolicy->timeout != 0)
1534 f2fs_update_time(sbi, dpolicy->timeout);
1536 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1537 if (dpolicy->timeout != 0 &&
1538 f2fs_time_over(sbi, dpolicy->timeout))
1541 if (i + 1 < dpolicy->granularity)
1544 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1545 return __issue_discard_cmd_orderly(sbi, dpolicy);
1547 pend_list = &dcc->pend_list[i];
1549 mutex_lock(&dcc->cmd_lock);
1550 if (list_empty(pend_list))
1552 if (unlikely(dcc->rbtree_check))
1553 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1555 blk_start_plug(&plug);
1556 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1557 f2fs_bug_on(sbi, dc->state != D_PREP);
1559 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1560 !is_idle(sbi, DISCARD_TIME)) {
1561 io_interrupted = true;
1565 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1567 if (issued >= dpolicy->max_requests)
1570 blk_finish_plug(&plug);
1572 mutex_unlock(&dcc->cmd_lock);
1574 if (issued >= dpolicy->max_requests || io_interrupted)
1578 if (!issued && io_interrupted)
1584 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1586 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1587 struct list_head *pend_list;
1588 struct discard_cmd *dc, *tmp;
1590 bool dropped = false;
1592 mutex_lock(&dcc->cmd_lock);
1593 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1594 pend_list = &dcc->pend_list[i];
1595 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1596 f2fs_bug_on(sbi, dc->state != D_PREP);
1597 __remove_discard_cmd(sbi, dc);
1601 mutex_unlock(&dcc->cmd_lock);
1606 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1608 __drop_discard_cmd(sbi);
1611 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1612 struct discard_cmd *dc)
1614 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1615 unsigned int len = 0;
1617 wait_for_completion_io(&dc->wait);
1618 mutex_lock(&dcc->cmd_lock);
1619 f2fs_bug_on(sbi, dc->state != D_DONE);
1624 __remove_discard_cmd(sbi, dc);
1626 mutex_unlock(&dcc->cmd_lock);
1631 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1632 struct discard_policy *dpolicy,
1633 block_t start, block_t end)
1635 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1636 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1637 &(dcc->fstrim_list) : &(dcc->wait_list);
1638 struct discard_cmd *dc, *tmp;
1640 unsigned int trimmed = 0;
1645 mutex_lock(&dcc->cmd_lock);
1646 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1647 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1649 if (dc->len < dpolicy->granularity)
1651 if (dc->state == D_DONE && !dc->ref) {
1652 wait_for_completion_io(&dc->wait);
1655 __remove_discard_cmd(sbi, dc);
1662 mutex_unlock(&dcc->cmd_lock);
1665 trimmed += __wait_one_discard_bio(sbi, dc);
1672 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1673 struct discard_policy *dpolicy)
1675 struct discard_policy dp;
1676 unsigned int discard_blks;
1679 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1682 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1683 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1684 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1685 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1687 return discard_blks;
1690 /* This should be covered by global mutex, &sit_i->sentry_lock */
1691 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1693 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1694 struct discard_cmd *dc;
1695 bool need_wait = false;
1697 mutex_lock(&dcc->cmd_lock);
1698 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1701 if (dc->state == D_PREP) {
1702 __punch_discard_cmd(sbi, dc, blkaddr);
1708 mutex_unlock(&dcc->cmd_lock);
1711 __wait_one_discard_bio(sbi, dc);
1714 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1716 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1718 if (dcc && dcc->f2fs_issue_discard) {
1719 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1721 dcc->f2fs_issue_discard = NULL;
1722 kthread_stop(discard_thread);
1726 /* This comes from f2fs_put_super */
1727 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1729 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1730 struct discard_policy dpolicy;
1733 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1734 dcc->discard_granularity);
1735 dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1736 __issue_discard_cmd(sbi, &dpolicy);
1737 dropped = __drop_discard_cmd(sbi);
1739 /* just to make sure there is no pending discard commands */
1740 __wait_all_discard_cmd(sbi, NULL);
1742 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1746 static int issue_discard_thread(void *data)
1748 struct f2fs_sb_info *sbi = data;
1749 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1750 wait_queue_head_t *q = &dcc->discard_wait_queue;
1751 struct discard_policy dpolicy;
1752 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1758 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1759 dcc->discard_granularity);
1761 wait_event_interruptible_timeout(*q,
1762 kthread_should_stop() || freezing(current) ||
1764 msecs_to_jiffies(wait_ms));
1766 if (dcc->discard_wake)
1767 dcc->discard_wake = 0;
1769 /* clean up pending candidates before going to sleep */
1770 if (atomic_read(&dcc->queued_discard))
1771 __wait_all_discard_cmd(sbi, NULL);
1773 if (try_to_freeze())
1775 if (f2fs_readonly(sbi->sb))
1777 if (kthread_should_stop())
1779 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1780 wait_ms = dpolicy.max_interval;
1784 if (sbi->gc_mode == GC_URGENT)
1785 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1787 sb_start_intwrite(sbi->sb);
1789 issued = __issue_discard_cmd(sbi, &dpolicy);
1791 __wait_all_discard_cmd(sbi, &dpolicy);
1792 wait_ms = dpolicy.min_interval;
1793 } else if (issued == -1){
1794 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1796 wait_ms = dpolicy.mid_interval;
1798 wait_ms = dpolicy.max_interval;
1801 sb_end_intwrite(sbi->sb);
1803 } while (!kthread_should_stop());
1807 #ifdef CONFIG_BLK_DEV_ZONED
1808 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1809 struct block_device *bdev, block_t blkstart, block_t blklen)
1811 sector_t sector, nr_sects;
1812 block_t lblkstart = blkstart;
1815 if (f2fs_is_multi_device(sbi)) {
1816 devi = f2fs_target_device_index(sbi, blkstart);
1817 if (blkstart < FDEV(devi).start_blk ||
1818 blkstart > FDEV(devi).end_blk) {
1819 f2fs_msg(sbi->sb, KERN_ERR, "Invalid block %x",
1823 blkstart -= FDEV(devi).start_blk;
1826 /* For sequential zones, reset the zone write pointer */
1827 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1828 sector = SECTOR_FROM_BLOCK(blkstart);
1829 nr_sects = SECTOR_FROM_BLOCK(blklen);
1831 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1832 nr_sects != bdev_zone_sectors(bdev)) {
1833 f2fs_msg(sbi->sb, KERN_ERR,
1834 "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1835 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1839 trace_f2fs_issue_reset_zone(bdev, blkstart);
1840 return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
1843 /* For conventional zones, use regular discard if supported */
1844 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1848 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1849 struct block_device *bdev, block_t blkstart, block_t blklen)
1851 #ifdef CONFIG_BLK_DEV_ZONED
1852 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1853 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1855 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1858 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1859 block_t blkstart, block_t blklen)
1861 sector_t start = blkstart, len = 0;
1862 struct block_device *bdev;
1863 struct seg_entry *se;
1864 unsigned int offset;
1868 bdev = f2fs_target_device(sbi, blkstart, NULL);
1870 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1872 struct block_device *bdev2 =
1873 f2fs_target_device(sbi, i, NULL);
1875 if (bdev2 != bdev) {
1876 err = __issue_discard_async(sbi, bdev,
1886 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1887 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1889 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1890 sbi->discard_blks--;
1894 err = __issue_discard_async(sbi, bdev, start, len);
1898 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1901 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1902 int max_blocks = sbi->blocks_per_seg;
1903 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1904 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1905 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1906 unsigned long *discard_map = (unsigned long *)se->discard_map;
1907 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1908 unsigned int start = 0, end = -1;
1909 bool force = (cpc->reason & CP_DISCARD);
1910 struct discard_entry *de = NULL;
1911 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1914 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1918 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1919 SM_I(sbi)->dcc_info->nr_discards >=
1920 SM_I(sbi)->dcc_info->max_discards)
1924 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1925 for (i = 0; i < entries; i++)
1926 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1927 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1929 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1930 SM_I(sbi)->dcc_info->max_discards) {
1931 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1932 if (start >= max_blocks)
1935 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1936 if (force && start && end != max_blocks
1937 && (end - start) < cpc->trim_minlen)
1944 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1946 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1947 list_add_tail(&de->list, head);
1950 for (i = start; i < end; i++)
1951 __set_bit_le(i, (void *)de->discard_map);
1953 SM_I(sbi)->dcc_info->nr_discards += end - start;
1958 static void release_discard_addr(struct discard_entry *entry)
1960 list_del(&entry->list);
1961 kmem_cache_free(discard_entry_slab, entry);
1964 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1966 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1967 struct discard_entry *entry, *this;
1970 list_for_each_entry_safe(entry, this, head, list)
1971 release_discard_addr(entry);
1975 * Should call f2fs_clear_prefree_segments after checkpoint is done.
1977 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1979 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1982 mutex_lock(&dirty_i->seglist_lock);
1983 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1984 __set_test_and_free(sbi, segno);
1985 mutex_unlock(&dirty_i->seglist_lock);
1988 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1989 struct cp_control *cpc)
1991 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1992 struct list_head *head = &dcc->entry_list;
1993 struct discard_entry *entry, *this;
1994 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1995 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1996 unsigned int start = 0, end = -1;
1997 unsigned int secno, start_segno;
1998 bool force = (cpc->reason & CP_DISCARD);
1999 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2001 mutex_lock(&dirty_i->seglist_lock);
2006 if (need_align && end != -1)
2008 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2009 if (start >= MAIN_SEGS(sbi))
2011 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2015 start = rounddown(start, sbi->segs_per_sec);
2016 end = roundup(end, sbi->segs_per_sec);
2019 for (i = start; i < end; i++) {
2020 if (test_and_clear_bit(i, prefree_map))
2021 dirty_i->nr_dirty[PRE]--;
2024 if (!f2fs_realtime_discard_enable(sbi))
2027 if (force && start >= cpc->trim_start &&
2028 (end - 1) <= cpc->trim_end)
2031 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
2032 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2033 (end - start) << sbi->log_blocks_per_seg);
2037 secno = GET_SEC_FROM_SEG(sbi, start);
2038 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2039 if (!IS_CURSEC(sbi, secno) &&
2040 !get_valid_blocks(sbi, start, true))
2041 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2042 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2044 start = start_segno + sbi->segs_per_sec;
2050 mutex_unlock(&dirty_i->seglist_lock);
2052 /* send small discards */
2053 list_for_each_entry_safe(entry, this, head, list) {
2054 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2055 bool is_valid = test_bit_le(0, entry->discard_map);
2059 next_pos = find_next_zero_bit_le(entry->discard_map,
2060 sbi->blocks_per_seg, cur_pos);
2061 len = next_pos - cur_pos;
2063 if (f2fs_sb_has_blkzoned(sbi) ||
2064 (force && len < cpc->trim_minlen))
2067 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2071 next_pos = find_next_bit_le(entry->discard_map,
2072 sbi->blocks_per_seg, cur_pos);
2076 is_valid = !is_valid;
2078 if (cur_pos < sbi->blocks_per_seg)
2081 release_discard_addr(entry);
2082 dcc->nr_discards -= total_len;
2085 wake_up_discard_thread(sbi, false);
2088 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2090 dev_t dev = sbi->sb->s_bdev->bd_dev;
2091 struct discard_cmd_control *dcc;
2094 if (SM_I(sbi)->dcc_info) {
2095 dcc = SM_I(sbi)->dcc_info;
2099 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2103 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2104 INIT_LIST_HEAD(&dcc->entry_list);
2105 for (i = 0; i < MAX_PLIST_NUM; i++)
2106 INIT_LIST_HEAD(&dcc->pend_list[i]);
2107 INIT_LIST_HEAD(&dcc->wait_list);
2108 INIT_LIST_HEAD(&dcc->fstrim_list);
2109 mutex_init(&dcc->cmd_lock);
2110 atomic_set(&dcc->issued_discard, 0);
2111 atomic_set(&dcc->queued_discard, 0);
2112 atomic_set(&dcc->discard_cmd_cnt, 0);
2113 dcc->nr_discards = 0;
2114 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2115 dcc->undiscard_blks = 0;
2117 dcc->root = RB_ROOT;
2118 dcc->rbtree_check = false;
2120 init_waitqueue_head(&dcc->discard_wait_queue);
2121 SM_I(sbi)->dcc_info = dcc;
2123 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2124 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2125 if (IS_ERR(dcc->f2fs_issue_discard)) {
2126 err = PTR_ERR(dcc->f2fs_issue_discard);
2128 SM_I(sbi)->dcc_info = NULL;
2135 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2137 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2142 f2fs_stop_discard_thread(sbi);
2145 SM_I(sbi)->dcc_info = NULL;
2148 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2150 struct sit_info *sit_i = SIT_I(sbi);
2152 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2153 sit_i->dirty_sentries++;
2160 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2161 unsigned int segno, int modified)
2163 struct seg_entry *se = get_seg_entry(sbi, segno);
2166 __mark_sit_entry_dirty(sbi, segno);
2169 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2171 struct seg_entry *se;
2172 unsigned int segno, offset;
2173 long int new_vblocks;
2175 #ifdef CONFIG_F2FS_CHECK_FS
2179 segno = GET_SEGNO(sbi, blkaddr);
2181 se = get_seg_entry(sbi, segno);
2182 new_vblocks = se->valid_blocks + del;
2183 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2185 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2186 (new_vblocks > sbi->blocks_per_seg)));
2188 se->valid_blocks = new_vblocks;
2189 se->mtime = get_mtime(sbi, false);
2190 if (se->mtime > SIT_I(sbi)->max_mtime)
2191 SIT_I(sbi)->max_mtime = se->mtime;
2193 /* Update valid block bitmap */
2195 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2196 #ifdef CONFIG_F2FS_CHECK_FS
2197 mir_exist = f2fs_test_and_set_bit(offset,
2198 se->cur_valid_map_mir);
2199 if (unlikely(exist != mir_exist)) {
2200 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2201 "when setting bitmap, blk:%u, old bit:%d",
2203 f2fs_bug_on(sbi, 1);
2206 if (unlikely(exist)) {
2207 f2fs_msg(sbi->sb, KERN_ERR,
2208 "Bitmap was wrongly set, blk:%u", blkaddr);
2209 f2fs_bug_on(sbi, 1);
2214 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2215 sbi->discard_blks--;
2217 /* don't overwrite by SSR to keep node chain */
2218 if (IS_NODESEG(se->type) &&
2219 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2220 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2221 se->ckpt_valid_blocks++;
2224 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2225 #ifdef CONFIG_F2FS_CHECK_FS
2226 mir_exist = f2fs_test_and_clear_bit(offset,
2227 se->cur_valid_map_mir);
2228 if (unlikely(exist != mir_exist)) {
2229 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2230 "when clearing bitmap, blk:%u, old bit:%d",
2232 f2fs_bug_on(sbi, 1);
2235 if (unlikely(!exist)) {
2236 f2fs_msg(sbi->sb, KERN_ERR,
2237 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2238 f2fs_bug_on(sbi, 1);
2241 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2243 * If checkpoints are off, we must not reuse data that
2244 * was used in the previous checkpoint. If it was used
2245 * before, we must track that to know how much space we
2248 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2249 sbi->unusable_block_count++;
2252 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2253 sbi->discard_blks++;
2255 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2256 se->ckpt_valid_blocks += del;
2258 __mark_sit_entry_dirty(sbi, segno);
2260 /* update total number of valid blocks to be written in ckpt area */
2261 SIT_I(sbi)->written_valid_blocks += del;
2263 if (__is_large_section(sbi))
2264 get_sec_entry(sbi, segno)->valid_blocks += del;
2267 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2269 unsigned int segno = GET_SEGNO(sbi, addr);
2270 struct sit_info *sit_i = SIT_I(sbi);
2272 f2fs_bug_on(sbi, addr == NULL_ADDR);
2273 if (addr == NEW_ADDR)
2276 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2278 /* add it into sit main buffer */
2279 down_write(&sit_i->sentry_lock);
2281 update_sit_entry(sbi, addr, -1);
2283 /* add it into dirty seglist */
2284 locate_dirty_segment(sbi, segno);
2286 up_write(&sit_i->sentry_lock);
2289 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2291 struct sit_info *sit_i = SIT_I(sbi);
2292 unsigned int segno, offset;
2293 struct seg_entry *se;
2296 if (!is_valid_data_blkaddr(sbi, blkaddr))
2299 down_read(&sit_i->sentry_lock);
2301 segno = GET_SEGNO(sbi, blkaddr);
2302 se = get_seg_entry(sbi, segno);
2303 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2305 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2308 up_read(&sit_i->sentry_lock);
2314 * This function should be resided under the curseg_mutex lock
2316 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2317 struct f2fs_summary *sum)
2319 struct curseg_info *curseg = CURSEG_I(sbi, type);
2320 void *addr = curseg->sum_blk;
2321 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2322 memcpy(addr, sum, sizeof(struct f2fs_summary));
2326 * Calculate the number of current summary pages for writing
2328 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2330 int valid_sum_count = 0;
2333 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2334 if (sbi->ckpt->alloc_type[i] == SSR)
2335 valid_sum_count += sbi->blocks_per_seg;
2338 valid_sum_count += le16_to_cpu(
2339 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2341 valid_sum_count += curseg_blkoff(sbi, i);
2345 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2346 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2347 if (valid_sum_count <= sum_in_page)
2349 else if ((valid_sum_count - sum_in_page) <=
2350 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2356 * Caller should put this summary page
2358 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2360 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2363 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2364 void *src, block_t blk_addr)
2366 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2368 memcpy(page_address(page), src, PAGE_SIZE);
2369 set_page_dirty(page);
2370 f2fs_put_page(page, 1);
2373 static void write_sum_page(struct f2fs_sb_info *sbi,
2374 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2376 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2379 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2380 int type, block_t blk_addr)
2382 struct curseg_info *curseg = CURSEG_I(sbi, type);
2383 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2384 struct f2fs_summary_block *src = curseg->sum_blk;
2385 struct f2fs_summary_block *dst;
2387 dst = (struct f2fs_summary_block *)page_address(page);
2388 memset(dst, 0, PAGE_SIZE);
2390 mutex_lock(&curseg->curseg_mutex);
2392 down_read(&curseg->journal_rwsem);
2393 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2394 up_read(&curseg->journal_rwsem);
2396 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2397 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2399 mutex_unlock(&curseg->curseg_mutex);
2401 set_page_dirty(page);
2402 f2fs_put_page(page, 1);
2405 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2407 struct curseg_info *curseg = CURSEG_I(sbi, type);
2408 unsigned int segno = curseg->segno + 1;
2409 struct free_segmap_info *free_i = FREE_I(sbi);
2411 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2412 return !test_bit(segno, free_i->free_segmap);
2417 * Find a new segment from the free segments bitmap to right order
2418 * This function should be returned with success, otherwise BUG
2420 static void get_new_segment(struct f2fs_sb_info *sbi,
2421 unsigned int *newseg, bool new_sec, int dir)
2423 struct free_segmap_info *free_i = FREE_I(sbi);
2424 unsigned int segno, secno, zoneno;
2425 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2426 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2427 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2428 unsigned int left_start = hint;
2433 spin_lock(&free_i->segmap_lock);
2435 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2436 segno = find_next_zero_bit(free_i->free_segmap,
2437 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2438 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2442 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2443 if (secno >= MAIN_SECS(sbi)) {
2444 if (dir == ALLOC_RIGHT) {
2445 secno = find_next_zero_bit(free_i->free_secmap,
2447 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2450 left_start = hint - 1;
2456 while (test_bit(left_start, free_i->free_secmap)) {
2457 if (left_start > 0) {
2461 left_start = find_next_zero_bit(free_i->free_secmap,
2463 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2468 segno = GET_SEG_FROM_SEC(sbi, secno);
2469 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2471 /* give up on finding another zone */
2474 if (sbi->secs_per_zone == 1)
2476 if (zoneno == old_zoneno)
2478 if (dir == ALLOC_LEFT) {
2479 if (!go_left && zoneno + 1 >= total_zones)
2481 if (go_left && zoneno == 0)
2484 for (i = 0; i < NR_CURSEG_TYPE; i++)
2485 if (CURSEG_I(sbi, i)->zone == zoneno)
2488 if (i < NR_CURSEG_TYPE) {
2489 /* zone is in user, try another */
2491 hint = zoneno * sbi->secs_per_zone - 1;
2492 else if (zoneno + 1 >= total_zones)
2495 hint = (zoneno + 1) * sbi->secs_per_zone;
2497 goto find_other_zone;
2500 /* set it as dirty segment in free segmap */
2501 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2502 __set_inuse(sbi, segno);
2504 spin_unlock(&free_i->segmap_lock);
2507 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2509 struct curseg_info *curseg = CURSEG_I(sbi, type);
2510 struct summary_footer *sum_footer;
2512 curseg->segno = curseg->next_segno;
2513 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2514 curseg->next_blkoff = 0;
2515 curseg->next_segno = NULL_SEGNO;
2517 sum_footer = &(curseg->sum_blk->footer);
2518 memset(sum_footer, 0, sizeof(struct summary_footer));
2519 if (IS_DATASEG(type))
2520 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2521 if (IS_NODESEG(type))
2522 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2523 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2526 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2528 /* if segs_per_sec is large than 1, we need to keep original policy. */
2529 if (__is_large_section(sbi))
2530 return CURSEG_I(sbi, type)->segno;
2532 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2535 if (test_opt(sbi, NOHEAP) &&
2536 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2539 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2540 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2542 /* find segments from 0 to reuse freed segments */
2543 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2546 return CURSEG_I(sbi, type)->segno;
2550 * Allocate a current working segment.
2551 * This function always allocates a free segment in LFS manner.
2553 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2555 struct curseg_info *curseg = CURSEG_I(sbi, type);
2556 unsigned int segno = curseg->segno;
2557 int dir = ALLOC_LEFT;
2559 write_sum_page(sbi, curseg->sum_blk,
2560 GET_SUM_BLOCK(sbi, segno));
2561 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2564 if (test_opt(sbi, NOHEAP))
2567 segno = __get_next_segno(sbi, type);
2568 get_new_segment(sbi, &segno, new_sec, dir);
2569 curseg->next_segno = segno;
2570 reset_curseg(sbi, type, 1);
2571 curseg->alloc_type = LFS;
2574 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2575 struct curseg_info *seg, block_t start)
2577 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2578 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2579 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2580 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2581 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2584 for (i = 0; i < entries; i++)
2585 target_map[i] = ckpt_map[i] | cur_map[i];
2587 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2589 seg->next_blkoff = pos;
2593 * If a segment is written by LFS manner, next block offset is just obtained
2594 * by increasing the current block offset. However, if a segment is written by
2595 * SSR manner, next block offset obtained by calling __next_free_blkoff
2597 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2598 struct curseg_info *seg)
2600 if (seg->alloc_type == SSR)
2601 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2607 * This function always allocates a used segment(from dirty seglist) by SSR
2608 * manner, so it should recover the existing segment information of valid blocks
2610 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2612 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2613 struct curseg_info *curseg = CURSEG_I(sbi, type);
2614 unsigned int new_segno = curseg->next_segno;
2615 struct f2fs_summary_block *sum_node;
2616 struct page *sum_page;
2618 write_sum_page(sbi, curseg->sum_blk,
2619 GET_SUM_BLOCK(sbi, curseg->segno));
2620 __set_test_and_inuse(sbi, new_segno);
2622 mutex_lock(&dirty_i->seglist_lock);
2623 __remove_dirty_segment(sbi, new_segno, PRE);
2624 __remove_dirty_segment(sbi, new_segno, DIRTY);
2625 mutex_unlock(&dirty_i->seglist_lock);
2627 reset_curseg(sbi, type, 1);
2628 curseg->alloc_type = SSR;
2629 __next_free_blkoff(sbi, curseg, 0);
2631 sum_page = f2fs_get_sum_page(sbi, new_segno);
2632 f2fs_bug_on(sbi, IS_ERR(sum_page));
2633 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2634 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2635 f2fs_put_page(sum_page, 1);
2638 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2640 struct curseg_info *curseg = CURSEG_I(sbi, type);
2641 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2642 unsigned segno = NULL_SEGNO;
2644 bool reversed = false;
2646 /* f2fs_need_SSR() already forces to do this */
2647 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2648 curseg->next_segno = segno;
2652 /* For node segments, let's do SSR more intensively */
2653 if (IS_NODESEG(type)) {
2654 if (type >= CURSEG_WARM_NODE) {
2656 i = CURSEG_COLD_NODE;
2658 i = CURSEG_HOT_NODE;
2660 cnt = NR_CURSEG_NODE_TYPE;
2662 if (type >= CURSEG_WARM_DATA) {
2664 i = CURSEG_COLD_DATA;
2666 i = CURSEG_HOT_DATA;
2668 cnt = NR_CURSEG_DATA_TYPE;
2671 for (; cnt-- > 0; reversed ? i-- : i++) {
2674 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2675 curseg->next_segno = segno;
2680 /* find valid_blocks=0 in dirty list */
2681 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2682 segno = get_free_segment(sbi);
2683 if (segno != NULL_SEGNO) {
2684 curseg->next_segno = segno;
2692 * flush out current segment and replace it with new segment
2693 * This function should be returned with success, otherwise BUG
2695 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2696 int type, bool force)
2698 struct curseg_info *curseg = CURSEG_I(sbi, type);
2701 new_curseg(sbi, type, true);
2702 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2703 type == CURSEG_WARM_NODE)
2704 new_curseg(sbi, type, false);
2705 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2706 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2707 new_curseg(sbi, type, false);
2708 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2709 change_curseg(sbi, type);
2711 new_curseg(sbi, type, false);
2713 stat_inc_seg_type(sbi, curseg);
2716 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2718 struct curseg_info *curseg;
2719 unsigned int old_segno;
2722 down_write(&SIT_I(sbi)->sentry_lock);
2724 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2725 curseg = CURSEG_I(sbi, i);
2726 old_segno = curseg->segno;
2727 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2728 locate_dirty_segment(sbi, old_segno);
2731 up_write(&SIT_I(sbi)->sentry_lock);
2734 static const struct segment_allocation default_salloc_ops = {
2735 .allocate_segment = allocate_segment_by_default,
2738 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2739 struct cp_control *cpc)
2741 __u64 trim_start = cpc->trim_start;
2742 bool has_candidate = false;
2744 down_write(&SIT_I(sbi)->sentry_lock);
2745 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2746 if (add_discard_addrs(sbi, cpc, true)) {
2747 has_candidate = true;
2751 up_write(&SIT_I(sbi)->sentry_lock);
2753 cpc->trim_start = trim_start;
2754 return has_candidate;
2757 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2758 struct discard_policy *dpolicy,
2759 unsigned int start, unsigned int end)
2761 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2762 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2763 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2764 struct discard_cmd *dc;
2765 struct blk_plug plug;
2767 unsigned int trimmed = 0;
2772 mutex_lock(&dcc->cmd_lock);
2773 if (unlikely(dcc->rbtree_check))
2774 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2777 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2779 (struct rb_entry **)&prev_dc,
2780 (struct rb_entry **)&next_dc,
2781 &insert_p, &insert_parent, true);
2785 blk_start_plug(&plug);
2787 while (dc && dc->lstart <= end) {
2788 struct rb_node *node;
2791 if (dc->len < dpolicy->granularity)
2794 if (dc->state != D_PREP) {
2795 list_move_tail(&dc->list, &dcc->fstrim_list);
2799 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2801 if (issued >= dpolicy->max_requests) {
2802 start = dc->lstart + dc->len;
2805 __remove_discard_cmd(sbi, dc);
2807 blk_finish_plug(&plug);
2808 mutex_unlock(&dcc->cmd_lock);
2809 trimmed += __wait_all_discard_cmd(sbi, NULL);
2810 congestion_wait(BLK_RW_ASYNC, HZ/50);
2814 node = rb_next(&dc->rb_node);
2816 __remove_discard_cmd(sbi, dc);
2817 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2819 if (fatal_signal_pending(current))
2823 blk_finish_plug(&plug);
2824 mutex_unlock(&dcc->cmd_lock);
2829 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2831 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2832 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2833 unsigned int start_segno, end_segno;
2834 block_t start_block, end_block;
2835 struct cp_control cpc;
2836 struct discard_policy dpolicy;
2837 unsigned long long trimmed = 0;
2839 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2841 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2844 if (end < MAIN_BLKADDR(sbi))
2847 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2848 f2fs_msg(sbi->sb, KERN_WARNING,
2849 "Found FS corruption, run fsck to fix.");
2853 /* start/end segment number in main_area */
2854 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2855 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2856 GET_SEGNO(sbi, end);
2858 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2859 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2862 cpc.reason = CP_DISCARD;
2863 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2864 cpc.trim_start = start_segno;
2865 cpc.trim_end = end_segno;
2867 if (sbi->discard_blks == 0)
2870 mutex_lock(&sbi->gc_mutex);
2871 err = f2fs_write_checkpoint(sbi, &cpc);
2872 mutex_unlock(&sbi->gc_mutex);
2877 * We filed discard candidates, but actually we don't need to wait for
2878 * all of them, since they'll be issued in idle time along with runtime
2879 * discard option. User configuration looks like using runtime discard
2880 * or periodic fstrim instead of it.
2882 if (f2fs_realtime_discard_enable(sbi))
2885 start_block = START_BLOCK(sbi, start_segno);
2886 end_block = START_BLOCK(sbi, end_segno + 1);
2888 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2889 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2890 start_block, end_block);
2892 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2893 start_block, end_block);
2896 range->len = F2FS_BLK_TO_BYTES(trimmed);
2900 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2902 struct curseg_info *curseg = CURSEG_I(sbi, type);
2903 if (curseg->next_blkoff < sbi->blocks_per_seg)
2908 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2911 case WRITE_LIFE_SHORT:
2912 return CURSEG_HOT_DATA;
2913 case WRITE_LIFE_EXTREME:
2914 return CURSEG_COLD_DATA;
2916 return CURSEG_WARM_DATA;
2920 /* This returns write hints for each segment type. This hints will be
2921 * passed down to block layer. There are mapping tables which depend on
2922 * the mount option 'whint_mode'.
2924 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2926 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2930 * META WRITE_LIFE_NOT_SET
2934 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2935 * extension list " "
2938 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2939 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2940 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2941 * WRITE_LIFE_NONE " "
2942 * WRITE_LIFE_MEDIUM " "
2943 * WRITE_LIFE_LONG " "
2946 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2947 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2948 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2949 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2950 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2951 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2953 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2957 * META WRITE_LIFE_MEDIUM;
2958 * HOT_NODE WRITE_LIFE_NOT_SET
2960 * COLD_NODE WRITE_LIFE_NONE
2961 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2962 * extension list " "
2965 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2966 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2967 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2968 * WRITE_LIFE_NONE " "
2969 * WRITE_LIFE_MEDIUM " "
2970 * WRITE_LIFE_LONG " "
2973 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2974 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2975 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2976 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2977 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2978 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2981 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2982 enum page_type type, enum temp_type temp)
2984 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2987 return WRITE_LIFE_NOT_SET;
2988 else if (temp == HOT)
2989 return WRITE_LIFE_SHORT;
2990 else if (temp == COLD)
2991 return WRITE_LIFE_EXTREME;
2993 return WRITE_LIFE_NOT_SET;
2995 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2998 return WRITE_LIFE_LONG;
2999 else if (temp == HOT)
3000 return WRITE_LIFE_SHORT;
3001 else if (temp == COLD)
3002 return WRITE_LIFE_EXTREME;
3003 } else if (type == NODE) {
3004 if (temp == WARM || temp == HOT)
3005 return WRITE_LIFE_NOT_SET;
3006 else if (temp == COLD)
3007 return WRITE_LIFE_NONE;
3008 } else if (type == META) {
3009 return WRITE_LIFE_MEDIUM;
3012 return WRITE_LIFE_NOT_SET;
3015 static int __get_segment_type_2(struct f2fs_io_info *fio)
3017 if (fio->type == DATA)
3018 return CURSEG_HOT_DATA;
3020 return CURSEG_HOT_NODE;
3023 static int __get_segment_type_4(struct f2fs_io_info *fio)
3025 if (fio->type == DATA) {
3026 struct inode *inode = fio->page->mapping->host;
3028 if (S_ISDIR(inode->i_mode))
3029 return CURSEG_HOT_DATA;
3031 return CURSEG_COLD_DATA;
3033 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3034 return CURSEG_WARM_NODE;
3036 return CURSEG_COLD_NODE;
3040 static int __get_segment_type_6(struct f2fs_io_info *fio)
3042 if (fio->type == DATA) {
3043 struct inode *inode = fio->page->mapping->host;
3045 if (is_cold_data(fio->page) || file_is_cold(inode))
3046 return CURSEG_COLD_DATA;
3047 if (file_is_hot(inode) ||
3048 is_inode_flag_set(inode, FI_HOT_DATA) ||
3049 f2fs_is_atomic_file(inode) ||
3050 f2fs_is_volatile_file(inode))
3051 return CURSEG_HOT_DATA;
3052 /* f2fs_rw_hint_to_seg_type(inode->i_write_hint); */
3053 return CURSEG_WARM_DATA;
3055 if (IS_DNODE(fio->page))
3056 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3058 return CURSEG_COLD_NODE;
3062 static int __get_segment_type(struct f2fs_io_info *fio)
3066 switch (F2FS_OPTION(fio->sbi).active_logs) {
3068 type = __get_segment_type_2(fio);
3071 type = __get_segment_type_4(fio);
3074 type = __get_segment_type_6(fio);
3077 f2fs_bug_on(fio->sbi, true);
3082 else if (IS_WARM(type))
3089 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3090 block_t old_blkaddr, block_t *new_blkaddr,
3091 struct f2fs_summary *sum, int type,
3092 struct f2fs_io_info *fio, bool add_list)
3094 struct sit_info *sit_i = SIT_I(sbi);
3095 struct curseg_info *curseg = CURSEG_I(sbi, type);
3097 down_read(&SM_I(sbi)->curseg_lock);
3099 mutex_lock(&curseg->curseg_mutex);
3100 down_write(&sit_i->sentry_lock);
3102 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3104 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3107 * __add_sum_entry should be resided under the curseg_mutex
3108 * because, this function updates a summary entry in the
3109 * current summary block.
3111 __add_sum_entry(sbi, type, sum);
3113 __refresh_next_blkoff(sbi, curseg);
3115 stat_inc_block_count(sbi, curseg);
3118 * SIT information should be updated before segment allocation,
3119 * since SSR needs latest valid block information.
3121 update_sit_entry(sbi, *new_blkaddr, 1);
3122 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3123 update_sit_entry(sbi, old_blkaddr, -1);
3125 if (!__has_curseg_space(sbi, type))
3126 sit_i->s_ops->allocate_segment(sbi, type, false);
3129 * segment dirty status should be updated after segment allocation,
3130 * so we just need to update status only one time after previous
3131 * segment being closed.
3133 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3134 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3136 up_write(&sit_i->sentry_lock);
3138 if (page && IS_NODESEG(type)) {
3139 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3141 f2fs_inode_chksum_set(sbi, page);
3145 struct f2fs_bio_info *io;
3147 INIT_LIST_HEAD(&fio->list);
3148 fio->in_list = true;
3150 io = sbi->write_io[fio->type] + fio->temp;
3151 spin_lock(&io->io_lock);
3152 list_add_tail(&fio->list, &io->io_list);
3153 spin_unlock(&io->io_lock);
3156 mutex_unlock(&curseg->curseg_mutex);
3158 up_read(&SM_I(sbi)->curseg_lock);
3161 static void update_device_state(struct f2fs_io_info *fio)
3163 struct f2fs_sb_info *sbi = fio->sbi;
3164 unsigned int devidx;
3166 if (!f2fs_is_multi_device(sbi))
3169 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3171 /* update device state for fsync */
3172 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3174 /* update device state for checkpoint */
3175 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3176 spin_lock(&sbi->dev_lock);
3177 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3178 spin_unlock(&sbi->dev_lock);
3182 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3184 int type = __get_segment_type(fio);
3185 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3188 down_read(&fio->sbi->io_order_lock);
3190 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3191 &fio->new_blkaddr, sum, type, fio, true);
3192 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3193 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3194 fio->old_blkaddr, fio->old_blkaddr);
3196 /* writeout dirty page into bdev */
3197 f2fs_submit_page_write(fio);
3199 fio->old_blkaddr = fio->new_blkaddr;
3203 update_device_state(fio);
3206 up_read(&fio->sbi->io_order_lock);
3209 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3210 enum iostat_type io_type)
3212 struct f2fs_io_info fio = {
3217 .op_flags = REQ_SYNC | REQ_NOIDLE | REQ_META | REQ_PRIO,
3218 .old_blkaddr = page->index,
3219 .new_blkaddr = page->index,
3221 .encrypted_page = NULL,
3225 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3226 fio.op_flags &= ~REQ_META;
3228 set_page_writeback(page);
3229 ClearPageError(page);
3230 f2fs_submit_page_write(&fio);
3232 stat_inc_meta_count(sbi, page->index);
3233 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3236 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3238 struct f2fs_summary sum;
3240 set_summary(&sum, nid, 0, 0);
3241 do_write_page(&sum, fio);
3243 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3246 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3247 struct f2fs_io_info *fio)
3249 struct f2fs_sb_info *sbi = fio->sbi;
3250 struct f2fs_summary sum;
3252 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3253 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3254 do_write_page(&sum, fio);
3255 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3257 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3260 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3263 struct f2fs_sb_info *sbi = fio->sbi;
3265 fio->new_blkaddr = fio->old_blkaddr;
3266 /* i/o temperature is needed for passing down write hints */
3267 __get_segment_type(fio);
3269 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3270 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3272 stat_inc_inplace_blocks(fio->sbi);
3274 err = f2fs_submit_page_bio(fio);
3276 update_device_state(fio);
3277 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3283 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3288 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3289 if (CURSEG_I(sbi, i)->segno == segno)
3295 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3296 block_t old_blkaddr, block_t new_blkaddr,
3297 bool recover_curseg, bool recover_newaddr)
3299 struct sit_info *sit_i = SIT_I(sbi);
3300 struct curseg_info *curseg;
3301 unsigned int segno, old_cursegno;
3302 struct seg_entry *se;
3304 unsigned short old_blkoff;
3306 segno = GET_SEGNO(sbi, new_blkaddr);
3307 se = get_seg_entry(sbi, segno);
3310 down_write(&SM_I(sbi)->curseg_lock);
3312 if (!recover_curseg) {
3313 /* for recovery flow */
3314 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3315 if (old_blkaddr == NULL_ADDR)
3316 type = CURSEG_COLD_DATA;
3318 type = CURSEG_WARM_DATA;
3321 if (IS_CURSEG(sbi, segno)) {
3322 /* se->type is volatile as SSR allocation */
3323 type = __f2fs_get_curseg(sbi, segno);
3324 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3326 type = CURSEG_WARM_DATA;
3330 f2fs_bug_on(sbi, !IS_DATASEG(type));
3331 curseg = CURSEG_I(sbi, type);
3333 mutex_lock(&curseg->curseg_mutex);
3334 down_write(&sit_i->sentry_lock);
3336 old_cursegno = curseg->segno;
3337 old_blkoff = curseg->next_blkoff;
3339 /* change the current segment */
3340 if (segno != curseg->segno) {
3341 curseg->next_segno = segno;
3342 change_curseg(sbi, type);
3345 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3346 __add_sum_entry(sbi, type, sum);
3348 if (!recover_curseg || recover_newaddr)
3349 update_sit_entry(sbi, new_blkaddr, 1);
3350 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3351 invalidate_mapping_pages(META_MAPPING(sbi),
3352 old_blkaddr, old_blkaddr);
3353 update_sit_entry(sbi, old_blkaddr, -1);
3356 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3357 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3359 locate_dirty_segment(sbi, old_cursegno);
3361 if (recover_curseg) {
3362 if (old_cursegno != curseg->segno) {
3363 curseg->next_segno = old_cursegno;
3364 change_curseg(sbi, type);
3366 curseg->next_blkoff = old_blkoff;
3369 up_write(&sit_i->sentry_lock);
3370 mutex_unlock(&curseg->curseg_mutex);
3371 up_write(&SM_I(sbi)->curseg_lock);
3374 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3375 block_t old_addr, block_t new_addr,
3376 unsigned char version, bool recover_curseg,
3377 bool recover_newaddr)
3379 struct f2fs_summary sum;
3381 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3383 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3384 recover_curseg, recover_newaddr);
3386 f2fs_update_data_blkaddr(dn, new_addr);
3389 void f2fs_wait_on_page_writeback(struct page *page,
3390 enum page_type type, bool ordered, bool locked)
3392 if (PageWriteback(page)) {
3393 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3395 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3397 wait_on_page_writeback(page);
3398 f2fs_bug_on(sbi, locked && PageWriteback(page));
3400 wait_for_stable_page(page);
3405 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3407 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3410 if (!f2fs_post_read_required(inode))
3413 if (!is_valid_data_blkaddr(sbi, blkaddr))
3416 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3418 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3419 f2fs_put_page(cpage, 1);
3423 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3428 for (i = 0; i < len; i++)
3429 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3432 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3434 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3435 struct curseg_info *seg_i;
3436 unsigned char *kaddr;
3441 start = start_sum_block(sbi);
3443 page = f2fs_get_meta_page(sbi, start++);
3445 return PTR_ERR(page);
3446 kaddr = (unsigned char *)page_address(page);
3448 /* Step 1: restore nat cache */
3449 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3450 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3452 /* Step 2: restore sit cache */
3453 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3454 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3455 offset = 2 * SUM_JOURNAL_SIZE;
3457 /* Step 3: restore summary entries */
3458 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3459 unsigned short blk_off;
3462 seg_i = CURSEG_I(sbi, i);
3463 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3464 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3465 seg_i->next_segno = segno;
3466 reset_curseg(sbi, i, 0);
3467 seg_i->alloc_type = ckpt->alloc_type[i];
3468 seg_i->next_blkoff = blk_off;
3470 if (seg_i->alloc_type == SSR)
3471 blk_off = sbi->blocks_per_seg;
3473 for (j = 0; j < blk_off; j++) {
3474 struct f2fs_summary *s;
3475 s = (struct f2fs_summary *)(kaddr + offset);
3476 seg_i->sum_blk->entries[j] = *s;
3477 offset += SUMMARY_SIZE;
3478 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3482 f2fs_put_page(page, 1);
3485 page = f2fs_get_meta_page(sbi, start++);
3487 return PTR_ERR(page);
3488 kaddr = (unsigned char *)page_address(page);
3492 f2fs_put_page(page, 1);
3496 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3498 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3499 struct f2fs_summary_block *sum;
3500 struct curseg_info *curseg;
3502 unsigned short blk_off;
3503 unsigned int segno = 0;
3504 block_t blk_addr = 0;
3507 /* get segment number and block addr */
3508 if (IS_DATASEG(type)) {
3509 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3510 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3512 if (__exist_node_summaries(sbi))
3513 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3515 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3517 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3519 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3521 if (__exist_node_summaries(sbi))
3522 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3523 type - CURSEG_HOT_NODE);
3525 blk_addr = GET_SUM_BLOCK(sbi, segno);
3528 new = f2fs_get_meta_page(sbi, blk_addr);
3530 return PTR_ERR(new);
3531 sum = (struct f2fs_summary_block *)page_address(new);
3533 if (IS_NODESEG(type)) {
3534 if (__exist_node_summaries(sbi)) {
3535 struct f2fs_summary *ns = &sum->entries[0];
3537 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3539 ns->ofs_in_node = 0;
3542 err = f2fs_restore_node_summary(sbi, segno, sum);
3548 /* set uncompleted segment to curseg */
3549 curseg = CURSEG_I(sbi, type);
3550 mutex_lock(&curseg->curseg_mutex);
3552 /* update journal info */
3553 down_write(&curseg->journal_rwsem);
3554 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3555 up_write(&curseg->journal_rwsem);
3557 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3558 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3559 curseg->next_segno = segno;
3560 reset_curseg(sbi, type, 0);
3561 curseg->alloc_type = ckpt->alloc_type[type];
3562 curseg->next_blkoff = blk_off;
3563 mutex_unlock(&curseg->curseg_mutex);
3565 f2fs_put_page(new, 1);
3569 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3571 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3572 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3573 int type = CURSEG_HOT_DATA;
3576 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3577 int npages = f2fs_npages_for_summary_flush(sbi, true);
3580 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3583 /* restore for compacted data summary */
3584 err = read_compacted_summaries(sbi);
3587 type = CURSEG_HOT_NODE;
3590 if (__exist_node_summaries(sbi))
3591 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3592 NR_CURSEG_TYPE - type, META_CP, true);
3594 for (; type <= CURSEG_COLD_NODE; type++) {
3595 err = read_normal_summaries(sbi, type);
3600 /* sanity check for summary blocks */
3601 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3602 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3608 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3611 unsigned char *kaddr;
3612 struct f2fs_summary *summary;
3613 struct curseg_info *seg_i;
3614 int written_size = 0;
3617 page = f2fs_grab_meta_page(sbi, blkaddr++);
3618 kaddr = (unsigned char *)page_address(page);
3619 memset(kaddr, 0, PAGE_SIZE);
3621 /* Step 1: write nat cache */
3622 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3623 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3624 written_size += SUM_JOURNAL_SIZE;
3626 /* Step 2: write sit cache */
3627 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3628 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3629 written_size += SUM_JOURNAL_SIZE;
3631 /* Step 3: write summary entries */
3632 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3633 unsigned short blkoff;
3634 seg_i = CURSEG_I(sbi, i);
3635 if (sbi->ckpt->alloc_type[i] == SSR)
3636 blkoff = sbi->blocks_per_seg;
3638 blkoff = curseg_blkoff(sbi, i);
3640 for (j = 0; j < blkoff; j++) {
3642 page = f2fs_grab_meta_page(sbi, blkaddr++);
3643 kaddr = (unsigned char *)page_address(page);
3644 memset(kaddr, 0, PAGE_SIZE);
3647 summary = (struct f2fs_summary *)(kaddr + written_size);
3648 *summary = seg_i->sum_blk->entries[j];
3649 written_size += SUMMARY_SIZE;
3651 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3655 set_page_dirty(page);
3656 f2fs_put_page(page, 1);
3661 set_page_dirty(page);
3662 f2fs_put_page(page, 1);
3666 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3667 block_t blkaddr, int type)
3670 if (IS_DATASEG(type))
3671 end = type + NR_CURSEG_DATA_TYPE;
3673 end = type + NR_CURSEG_NODE_TYPE;
3675 for (i = type; i < end; i++)
3676 write_current_sum_page(sbi, i, blkaddr + (i - type));
3679 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3681 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3682 write_compacted_summaries(sbi, start_blk);
3684 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3687 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3689 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3692 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3693 unsigned int val, int alloc)
3697 if (type == NAT_JOURNAL) {
3698 for (i = 0; i < nats_in_cursum(journal); i++) {
3699 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3702 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3703 return update_nats_in_cursum(journal, 1);
3704 } else if (type == SIT_JOURNAL) {
3705 for (i = 0; i < sits_in_cursum(journal); i++)
3706 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3708 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3709 return update_sits_in_cursum(journal, 1);
3714 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3717 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3720 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3723 struct sit_info *sit_i = SIT_I(sbi);
3725 pgoff_t src_off, dst_off;
3727 src_off = current_sit_addr(sbi, start);
3728 dst_off = next_sit_addr(sbi, src_off);
3730 page = f2fs_grab_meta_page(sbi, dst_off);
3731 seg_info_to_sit_page(sbi, page, start);
3733 set_page_dirty(page);
3734 set_to_next_sit(sit_i, start);
3739 static struct sit_entry_set *grab_sit_entry_set(void)
3741 struct sit_entry_set *ses =
3742 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3745 INIT_LIST_HEAD(&ses->set_list);
3749 static void release_sit_entry_set(struct sit_entry_set *ses)
3751 list_del(&ses->set_list);
3752 kmem_cache_free(sit_entry_set_slab, ses);
3755 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3756 struct list_head *head)
3758 struct sit_entry_set *next = ses;
3760 if (list_is_last(&ses->set_list, head))
3763 list_for_each_entry_continue(next, head, set_list)
3764 if (ses->entry_cnt <= next->entry_cnt)
3767 list_move_tail(&ses->set_list, &next->set_list);
3770 static void add_sit_entry(unsigned int segno, struct list_head *head)
3772 struct sit_entry_set *ses;
3773 unsigned int start_segno = START_SEGNO(segno);
3775 list_for_each_entry(ses, head, set_list) {
3776 if (ses->start_segno == start_segno) {
3778 adjust_sit_entry_set(ses, head);
3783 ses = grab_sit_entry_set();
3785 ses->start_segno = start_segno;
3787 list_add(&ses->set_list, head);
3790 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3792 struct f2fs_sm_info *sm_info = SM_I(sbi);
3793 struct list_head *set_list = &sm_info->sit_entry_set;
3794 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3797 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3798 add_sit_entry(segno, set_list);
3801 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3803 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3804 struct f2fs_journal *journal = curseg->journal;
3807 down_write(&curseg->journal_rwsem);
3808 for (i = 0; i < sits_in_cursum(journal); i++) {
3812 segno = le32_to_cpu(segno_in_journal(journal, i));
3813 dirtied = __mark_sit_entry_dirty(sbi, segno);
3816 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3818 update_sits_in_cursum(journal, -i);
3819 up_write(&curseg->journal_rwsem);
3823 * CP calls this function, which flushes SIT entries including sit_journal,
3824 * and moves prefree segs to free segs.
3826 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3828 struct sit_info *sit_i = SIT_I(sbi);
3829 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3830 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3831 struct f2fs_journal *journal = curseg->journal;
3832 struct sit_entry_set *ses, *tmp;
3833 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3834 bool to_journal = true;
3835 struct seg_entry *se;
3837 down_write(&sit_i->sentry_lock);
3839 if (!sit_i->dirty_sentries)
3843 * add and account sit entries of dirty bitmap in sit entry
3846 add_sits_in_set(sbi);
3849 * if there are no enough space in journal to store dirty sit
3850 * entries, remove all entries from journal and add and account
3851 * them in sit entry set.
3853 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3854 remove_sits_in_journal(sbi);
3857 * there are two steps to flush sit entries:
3858 * #1, flush sit entries to journal in current cold data summary block.
3859 * #2, flush sit entries to sit page.
3861 list_for_each_entry_safe(ses, tmp, head, set_list) {
3862 struct page *page = NULL;
3863 struct f2fs_sit_block *raw_sit = NULL;
3864 unsigned int start_segno = ses->start_segno;
3865 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3866 (unsigned long)MAIN_SEGS(sbi));
3867 unsigned int segno = start_segno;
3870 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3874 down_write(&curseg->journal_rwsem);
3876 page = get_next_sit_page(sbi, start_segno);
3877 raw_sit = page_address(page);
3880 /* flush dirty sit entries in region of current sit set */
3881 for_each_set_bit_from(segno, bitmap, end) {
3882 int offset, sit_offset;
3884 se = get_seg_entry(sbi, segno);
3885 #ifdef CONFIG_F2FS_CHECK_FS
3886 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3887 SIT_VBLOCK_MAP_SIZE))
3888 f2fs_bug_on(sbi, 1);
3891 /* add discard candidates */
3892 if (!(cpc->reason & CP_DISCARD)) {
3893 cpc->trim_start = segno;
3894 add_discard_addrs(sbi, cpc, false);
3898 offset = f2fs_lookup_journal_in_cursum(journal,
3899 SIT_JOURNAL, segno, 1);
3900 f2fs_bug_on(sbi, offset < 0);
3901 segno_in_journal(journal, offset) =
3903 seg_info_to_raw_sit(se,
3904 &sit_in_journal(journal, offset));
3905 check_block_count(sbi, segno,
3906 &sit_in_journal(journal, offset));
3908 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3909 seg_info_to_raw_sit(se,
3910 &raw_sit->entries[sit_offset]);
3911 check_block_count(sbi, segno,
3912 &raw_sit->entries[sit_offset]);
3915 __clear_bit(segno, bitmap);
3916 sit_i->dirty_sentries--;
3921 up_write(&curseg->journal_rwsem);
3923 f2fs_put_page(page, 1);
3925 f2fs_bug_on(sbi, ses->entry_cnt);
3926 release_sit_entry_set(ses);
3929 f2fs_bug_on(sbi, !list_empty(head));
3930 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3932 if (cpc->reason & CP_DISCARD) {
3933 __u64 trim_start = cpc->trim_start;
3935 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3936 add_discard_addrs(sbi, cpc, false);
3938 cpc->trim_start = trim_start;
3940 up_write(&sit_i->sentry_lock);
3942 set_prefree_as_free_segments(sbi);
3945 static int build_sit_info(struct f2fs_sb_info *sbi)
3947 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3948 struct sit_info *sit_i;
3949 unsigned int sit_segs, start;
3951 unsigned int bitmap_size;
3953 /* allocate memory for SIT information */
3954 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3958 SM_I(sbi)->sit_info = sit_i;
3961 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3964 if (!sit_i->sentries)
3967 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3968 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3970 if (!sit_i->dirty_sentries_bitmap)
3973 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3974 sit_i->sentries[start].cur_valid_map
3975 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3976 sit_i->sentries[start].ckpt_valid_map
3977 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3978 if (!sit_i->sentries[start].cur_valid_map ||
3979 !sit_i->sentries[start].ckpt_valid_map)
3982 #ifdef CONFIG_F2FS_CHECK_FS
3983 sit_i->sentries[start].cur_valid_map_mir
3984 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3985 if (!sit_i->sentries[start].cur_valid_map_mir)
3989 sit_i->sentries[start].discard_map
3990 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3992 if (!sit_i->sentries[start].discard_map)
3996 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3997 if (!sit_i->tmp_map)
4000 if (__is_large_section(sbi)) {
4001 sit_i->sec_entries =
4002 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4005 if (!sit_i->sec_entries)
4009 /* get information related with SIT */
4010 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4012 /* setup SIT bitmap from ckeckpoint pack */
4013 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4014 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4016 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4017 if (!sit_i->sit_bitmap)
4020 #ifdef CONFIG_F2FS_CHECK_FS
4021 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4022 if (!sit_i->sit_bitmap_mir)
4026 /* init SIT information */
4027 sit_i->s_ops = &default_salloc_ops;
4029 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4030 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4031 sit_i->written_valid_blocks = 0;
4032 sit_i->bitmap_size = bitmap_size;
4033 sit_i->dirty_sentries = 0;
4034 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4035 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4036 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
4037 init_rwsem(&sit_i->sentry_lock);
4041 static int build_free_segmap(struct f2fs_sb_info *sbi)
4043 struct free_segmap_info *free_i;
4044 unsigned int bitmap_size, sec_bitmap_size;
4046 /* allocate memory for free segmap information */
4047 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4051 SM_I(sbi)->free_info = free_i;
4053 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4054 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4055 if (!free_i->free_segmap)
4058 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4059 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4060 if (!free_i->free_secmap)
4063 /* set all segments as dirty temporarily */
4064 memset(free_i->free_segmap, 0xff, bitmap_size);
4065 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4067 /* init free segmap information */
4068 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4069 free_i->free_segments = 0;
4070 free_i->free_sections = 0;
4071 spin_lock_init(&free_i->segmap_lock);
4075 static int build_curseg(struct f2fs_sb_info *sbi)
4077 struct curseg_info *array;
4080 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4085 SM_I(sbi)->curseg_array = array;
4087 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4088 mutex_init(&array[i].curseg_mutex);
4089 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4090 if (!array[i].sum_blk)
4092 init_rwsem(&array[i].journal_rwsem);
4093 array[i].journal = f2fs_kzalloc(sbi,
4094 sizeof(struct f2fs_journal), GFP_KERNEL);
4095 if (!array[i].journal)
4097 array[i].segno = NULL_SEGNO;
4098 array[i].next_blkoff = 0;
4100 return restore_curseg_summaries(sbi);
4103 static int build_sit_entries(struct f2fs_sb_info *sbi)
4105 struct sit_info *sit_i = SIT_I(sbi);
4106 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4107 struct f2fs_journal *journal = curseg->journal;
4108 struct seg_entry *se;
4109 struct f2fs_sit_entry sit;
4110 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4111 unsigned int i, start, end;
4112 unsigned int readed, start_blk = 0;
4114 block_t total_node_blocks = 0;
4117 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4120 start = start_blk * sit_i->sents_per_block;
4121 end = (start_blk + readed) * sit_i->sents_per_block;
4123 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4124 struct f2fs_sit_block *sit_blk;
4127 se = &sit_i->sentries[start];
4128 page = get_current_sit_page(sbi, start);
4130 return PTR_ERR(page);
4131 sit_blk = (struct f2fs_sit_block *)page_address(page);
4132 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4133 f2fs_put_page(page, 1);
4135 err = check_block_count(sbi, start, &sit);
4138 seg_info_from_raw_sit(se, &sit);
4139 if (IS_NODESEG(se->type))
4140 total_node_blocks += se->valid_blocks;
4142 /* build discard map only one time */
4143 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4144 memset(se->discard_map, 0xff,
4145 SIT_VBLOCK_MAP_SIZE);
4147 memcpy(se->discard_map,
4149 SIT_VBLOCK_MAP_SIZE);
4150 sbi->discard_blks +=
4151 sbi->blocks_per_seg -
4155 if (__is_large_section(sbi))
4156 get_sec_entry(sbi, start)->valid_blocks +=
4159 start_blk += readed;
4160 } while (start_blk < sit_blk_cnt);
4162 down_read(&curseg->journal_rwsem);
4163 for (i = 0; i < sits_in_cursum(journal); i++) {
4164 unsigned int old_valid_blocks;
4166 start = le32_to_cpu(segno_in_journal(journal, i));
4167 if (start >= MAIN_SEGS(sbi)) {
4168 f2fs_msg(sbi->sb, KERN_ERR,
4169 "Wrong journal entry on segno %u",
4171 set_sbi_flag(sbi, SBI_NEED_FSCK);
4176 se = &sit_i->sentries[start];
4177 sit = sit_in_journal(journal, i);
4179 old_valid_blocks = se->valid_blocks;
4180 if (IS_NODESEG(se->type))
4181 total_node_blocks -= old_valid_blocks;
4183 err = check_block_count(sbi, start, &sit);
4186 seg_info_from_raw_sit(se, &sit);
4187 if (IS_NODESEG(se->type))
4188 total_node_blocks += se->valid_blocks;
4190 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4191 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4193 memcpy(se->discard_map, se->cur_valid_map,
4194 SIT_VBLOCK_MAP_SIZE);
4195 sbi->discard_blks += old_valid_blocks;
4196 sbi->discard_blks -= se->valid_blocks;
4199 if (__is_large_section(sbi)) {
4200 get_sec_entry(sbi, start)->valid_blocks +=
4202 get_sec_entry(sbi, start)->valid_blocks -=
4206 up_read(&curseg->journal_rwsem);
4208 if (!err && total_node_blocks != valid_node_count(sbi)) {
4209 f2fs_msg(sbi->sb, KERN_ERR,
4210 "SIT is corrupted node# %u vs %u",
4211 total_node_blocks, valid_node_count(sbi));
4212 set_sbi_flag(sbi, SBI_NEED_FSCK);
4219 static void init_free_segmap(struct f2fs_sb_info *sbi)
4224 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4225 struct seg_entry *sentry = get_seg_entry(sbi, start);
4226 if (!sentry->valid_blocks)
4227 __set_free(sbi, start);
4229 SIT_I(sbi)->written_valid_blocks +=
4230 sentry->valid_blocks;
4233 /* set use the current segments */
4234 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4235 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4236 __set_test_and_inuse(sbi, curseg_t->segno);
4240 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4242 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4243 struct free_segmap_info *free_i = FREE_I(sbi);
4244 unsigned int segno = 0, offset = 0;
4245 unsigned short valid_blocks;
4248 /* find dirty segment based on free segmap */
4249 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4250 if (segno >= MAIN_SEGS(sbi))
4253 valid_blocks = get_valid_blocks(sbi, segno, false);
4254 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4256 if (valid_blocks > sbi->blocks_per_seg) {
4257 f2fs_bug_on(sbi, 1);
4260 mutex_lock(&dirty_i->seglist_lock);
4261 __locate_dirty_segment(sbi, segno, DIRTY);
4262 mutex_unlock(&dirty_i->seglist_lock);
4266 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4268 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4269 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4271 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4272 if (!dirty_i->victim_secmap)
4277 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4279 struct dirty_seglist_info *dirty_i;
4280 unsigned int bitmap_size, i;
4282 /* allocate memory for dirty segments list information */
4283 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4288 SM_I(sbi)->dirty_info = dirty_i;
4289 mutex_init(&dirty_i->seglist_lock);
4291 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4293 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4294 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4296 if (!dirty_i->dirty_segmap[i])
4300 init_dirty_segmap(sbi);
4301 return init_victim_secmap(sbi);
4305 * Update min, max modified time for cost-benefit GC algorithm
4307 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4309 struct sit_info *sit_i = SIT_I(sbi);
4312 down_write(&sit_i->sentry_lock);
4314 sit_i->min_mtime = ULLONG_MAX;
4316 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4318 unsigned long long mtime = 0;
4320 for (i = 0; i < sbi->segs_per_sec; i++)
4321 mtime += get_seg_entry(sbi, segno + i)->mtime;
4323 mtime = div_u64(mtime, sbi->segs_per_sec);
4325 if (sit_i->min_mtime > mtime)
4326 sit_i->min_mtime = mtime;
4328 sit_i->max_mtime = get_mtime(sbi, false);
4329 up_write(&sit_i->sentry_lock);
4332 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4334 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4335 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4336 struct f2fs_sm_info *sm_info;
4339 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4344 sbi->sm_info = sm_info;
4345 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4346 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4347 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4348 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4349 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4350 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4351 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4352 sm_info->rec_prefree_segments = sm_info->main_segments *
4353 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4354 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4355 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4357 if (!test_opt(sbi, LFS))
4358 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4359 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4360 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4361 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4362 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4363 sm_info->min_ssr_sections = reserved_sections(sbi);
4365 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4367 init_rwsem(&sm_info->curseg_lock);
4369 if (!f2fs_readonly(sbi->sb)) {
4370 err = f2fs_create_flush_cmd_control(sbi);
4375 err = create_discard_cmd_control(sbi);
4379 err = build_sit_info(sbi);
4382 err = build_free_segmap(sbi);
4385 err = build_curseg(sbi);
4389 /* reinit free segmap based on SIT */
4390 err = build_sit_entries(sbi);
4394 init_free_segmap(sbi);
4395 err = build_dirty_segmap(sbi);
4399 init_min_max_mtime(sbi);
4403 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4404 enum dirty_type dirty_type)
4406 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4408 mutex_lock(&dirty_i->seglist_lock);
4409 kvfree(dirty_i->dirty_segmap[dirty_type]);
4410 dirty_i->nr_dirty[dirty_type] = 0;
4411 mutex_unlock(&dirty_i->seglist_lock);
4414 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4416 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4417 kvfree(dirty_i->victim_secmap);
4420 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4422 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4428 /* discard pre-free/dirty segments list */
4429 for (i = 0; i < NR_DIRTY_TYPE; i++)
4430 discard_dirty_segmap(sbi, i);
4432 destroy_victim_secmap(sbi);
4433 SM_I(sbi)->dirty_info = NULL;
4437 static void destroy_curseg(struct f2fs_sb_info *sbi)
4439 struct curseg_info *array = SM_I(sbi)->curseg_array;
4444 SM_I(sbi)->curseg_array = NULL;
4445 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4446 kvfree(array[i].sum_blk);
4447 kvfree(array[i].journal);
4452 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4454 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4457 SM_I(sbi)->free_info = NULL;
4458 kvfree(free_i->free_segmap);
4459 kvfree(free_i->free_secmap);
4463 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4465 struct sit_info *sit_i = SIT_I(sbi);
4471 if (sit_i->sentries) {
4472 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4473 kvfree(sit_i->sentries[start].cur_valid_map);
4474 #ifdef CONFIG_F2FS_CHECK_FS
4475 kvfree(sit_i->sentries[start].cur_valid_map_mir);
4477 kvfree(sit_i->sentries[start].ckpt_valid_map);
4478 kvfree(sit_i->sentries[start].discard_map);
4481 kvfree(sit_i->tmp_map);
4483 kvfree(sit_i->sentries);
4484 kvfree(sit_i->sec_entries);
4485 kvfree(sit_i->dirty_sentries_bitmap);
4487 SM_I(sbi)->sit_info = NULL;
4488 kvfree(sit_i->sit_bitmap);
4489 #ifdef CONFIG_F2FS_CHECK_FS
4490 kvfree(sit_i->sit_bitmap_mir);
4495 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4497 struct f2fs_sm_info *sm_info = SM_I(sbi);
4501 f2fs_destroy_flush_cmd_control(sbi, true);
4502 destroy_discard_cmd_control(sbi);
4503 destroy_dirty_segmap(sbi);
4504 destroy_curseg(sbi);
4505 destroy_free_segmap(sbi);
4506 destroy_sit_info(sbi);
4507 sbi->sm_info = NULL;
4511 int __init f2fs_create_segment_manager_caches(void)
4513 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4514 sizeof(struct discard_entry));
4515 if (!discard_entry_slab)
4518 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4519 sizeof(struct discard_cmd));
4520 if (!discard_cmd_slab)
4521 goto destroy_discard_entry;
4523 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4524 sizeof(struct sit_entry_set));
4525 if (!sit_entry_set_slab)
4526 goto destroy_discard_cmd;
4528 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4529 sizeof(struct inmem_pages));
4530 if (!inmem_entry_slab)
4531 goto destroy_sit_entry_set;
4534 destroy_sit_entry_set:
4535 kmem_cache_destroy(sit_entry_set_slab);
4536 destroy_discard_cmd:
4537 kmem_cache_destroy(discard_cmd_slab);
4538 destroy_discard_entry:
4539 kmem_cache_destroy(discard_entry_slab);
4544 void f2fs_destroy_segment_manager_caches(void)
4546 kmem_cache_destroy(sit_entry_set_slab);
4547 kmem_cache_destroy(discard_cmd_slab);
4548 kmem_cache_destroy(discard_entry_slab);
4549 kmem_cache_destroy(inmem_entry_slab);