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/sched/mm.h>
13 #include <linux/prefetch.h>
14 #include <linux/kthread.h>
15 #include <linux/swap.h>
16 #include <linux/timer.h>
17 #include <linux/freezer.h>
18 #include <linux/sched/signal.h>
19 #include <linux/random.h>
26 #include <trace/events/f2fs.h>
28 #define __reverse_ffz(x) __reverse_ffs(~(x))
30 static struct kmem_cache *discard_entry_slab;
31 static struct kmem_cache *discard_cmd_slab;
32 static struct kmem_cache *sit_entry_set_slab;
33 static struct kmem_cache *revoke_entry_slab;
35 static unsigned long __reverse_ulong(unsigned char *str)
37 unsigned long tmp = 0;
38 int shift = 24, idx = 0;
40 #if BITS_PER_LONG == 64
44 tmp |= (unsigned long)str[idx++] << shift;
45 shift -= BITS_PER_BYTE;
51 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
52 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 static inline unsigned long __reverse_ffs(unsigned long word)
58 #if BITS_PER_LONG == 64
59 if ((word & 0xffffffff00000000UL) == 0)
64 if ((word & 0xffff0000) == 0)
69 if ((word & 0xff00) == 0)
74 if ((word & 0xf0) == 0)
79 if ((word & 0xc) == 0)
84 if ((word & 0x2) == 0)
90 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
91 * f2fs_set_bit makes MSB and LSB reversed in a byte.
92 * @size must be integral times of unsigned long.
95 * f2fs_set_bit(0, bitmap) => 1000 0000
96 * f2fs_set_bit(7, bitmap) => 0000 0001
98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
99 unsigned long size, unsigned long offset)
101 const unsigned long *p = addr + BIT_WORD(offset);
102 unsigned long result = size;
108 size -= (offset & ~(BITS_PER_LONG - 1));
109 offset %= BITS_PER_LONG;
115 tmp = __reverse_ulong((unsigned char *)p);
117 tmp &= ~0UL >> offset;
118 if (size < BITS_PER_LONG)
119 tmp &= (~0UL << (BITS_PER_LONG - size));
123 if (size <= BITS_PER_LONG)
125 size -= BITS_PER_LONG;
131 return result - size + __reverse_ffs(tmp);
134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 unsigned long size, unsigned long offset)
137 const unsigned long *p = addr + BIT_WORD(offset);
138 unsigned long result = size;
144 size -= (offset & ~(BITS_PER_LONG - 1));
145 offset %= BITS_PER_LONG;
151 tmp = __reverse_ulong((unsigned char *)p);
154 tmp |= ~0UL << (BITS_PER_LONG - offset);
155 if (size < BITS_PER_LONG)
160 if (size <= BITS_PER_LONG)
162 size -= BITS_PER_LONG;
168 return result - size + __reverse_ffz(tmp);
171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
173 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
174 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
175 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177 if (f2fs_lfs_mode(sbi))
179 if (sbi->gc_mode == GC_URGENT_HIGH)
181 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
184 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
185 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
188 void f2fs_abort_atomic_write(struct inode *inode, bool clean)
190 struct f2fs_inode_info *fi = F2FS_I(inode);
192 if (!f2fs_is_atomic_file(inode))
195 release_atomic_write_cnt(inode);
196 clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
197 clear_inode_flag(inode, FI_ATOMIC_REPLACE);
198 clear_inode_flag(inode, FI_ATOMIC_FILE);
199 stat_dec_atomic_inode(inode);
201 F2FS_I(inode)->atomic_write_task = NULL;
204 truncate_inode_pages_final(inode->i_mapping);
205 f2fs_i_size_write(inode, fi->original_i_size);
206 fi->original_i_size = 0;
208 /* avoid stale dirty inode during eviction */
209 sync_inode_metadata(inode, 0);
212 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
213 block_t new_addr, block_t *old_addr, bool recover)
215 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
216 struct dnode_of_data dn;
221 set_new_dnode(&dn, inode, NULL, NULL, 0);
222 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
224 if (err == -ENOMEM) {
225 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
231 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
238 /* dn.data_blkaddr is always valid */
239 if (!__is_valid_data_blkaddr(new_addr)) {
240 if (new_addr == NULL_ADDR)
241 dec_valid_block_count(sbi, inode, 1);
242 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
243 f2fs_update_data_blkaddr(&dn, new_addr);
245 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
246 new_addr, ni.version, true, true);
251 err = inc_valid_block_count(sbi, inode, &count);
257 *old_addr = dn.data_blkaddr;
258 f2fs_truncate_data_blocks_range(&dn, 1);
259 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
261 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
262 ni.version, true, false);
267 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
268 index, old_addr ? *old_addr : 0, new_addr, recover);
272 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
275 struct revoke_entry *cur, *tmp;
276 pgoff_t start_index = 0;
277 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
279 list_for_each_entry_safe(cur, tmp, head, list) {
281 __replace_atomic_write_block(inode, cur->index,
282 cur->old_addr, NULL, true);
283 } else if (truncate) {
284 f2fs_truncate_hole(inode, start_index, cur->index);
285 start_index = cur->index + 1;
288 list_del(&cur->list);
289 kmem_cache_free(revoke_entry_slab, cur);
292 if (!revoke && truncate)
293 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
296 static int __f2fs_commit_atomic_write(struct inode *inode)
298 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
299 struct f2fs_inode_info *fi = F2FS_I(inode);
300 struct inode *cow_inode = fi->cow_inode;
301 struct revoke_entry *new;
302 struct list_head revoke_list;
304 struct dnode_of_data dn;
305 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
306 pgoff_t off = 0, blen, index;
309 INIT_LIST_HEAD(&revoke_list);
312 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
314 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
315 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
316 if (ret && ret != -ENOENT) {
318 } else if (ret == -ENOENT) {
320 if (dn.max_level == 0)
325 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
328 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
329 blkaddr = f2fs_data_blkaddr(&dn);
331 if (!__is_valid_data_blkaddr(blkaddr)) {
333 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
334 DATA_GENERIC_ENHANCE)) {
337 f2fs_handle_error(sbi,
338 ERROR_INVALID_BLKADDR);
342 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
345 ret = __replace_atomic_write_block(inode, index, blkaddr,
346 &new->old_addr, false);
349 kmem_cache_free(revoke_entry_slab, new);
353 f2fs_update_data_blkaddr(&dn, NULL_ADDR);
355 list_add_tail(&new->list, &revoke_list);
365 sbi->revoked_atomic_block += fi->atomic_write_cnt;
367 sbi->committed_atomic_block += fi->atomic_write_cnt;
368 set_inode_flag(inode, FI_ATOMIC_COMMITTED);
371 __complete_revoke_list(inode, &revoke_list, ret ? true : false);
376 int f2fs_commit_atomic_write(struct inode *inode)
378 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
379 struct f2fs_inode_info *fi = F2FS_I(inode);
382 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
386 f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
389 err = __f2fs_commit_atomic_write(inode);
392 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
398 * This function balances dirty node and dentry pages.
399 * In addition, it controls garbage collection.
401 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
403 if (time_to_inject(sbi, FAULT_CHECKPOINT))
404 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
406 /* balance_fs_bg is able to be pending */
407 if (need && excess_cached_nats(sbi))
408 f2fs_balance_fs_bg(sbi, false);
410 if (!f2fs_is_checkpoint_ready(sbi))
414 * We should do GC or end up with checkpoint, if there are so many dirty
415 * dir/node pages without enough free segments.
417 if (has_enough_free_secs(sbi, 0, 0))
420 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
421 sbi->gc_thread->f2fs_gc_task) {
424 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
425 TASK_UNINTERRUPTIBLE);
426 wake_up(&sbi->gc_thread->gc_wait_queue_head);
428 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
430 struct f2fs_gc_control gc_control = {
431 .victim_segno = NULL_SEGNO,
432 .init_gc_type = BG_GC,
434 .should_migrate_blocks = false,
435 .err_gc_skipped = false,
437 f2fs_down_write(&sbi->gc_lock);
438 stat_inc_gc_call_count(sbi, FOREGROUND);
439 f2fs_gc(sbi, &gc_control);
443 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
445 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
446 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
447 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
448 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
449 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
450 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
451 unsigned int threshold = sbi->blocks_per_seg * factor *
452 DEFAULT_DIRTY_THRESHOLD;
453 unsigned int global_threshold = threshold * 3 / 2;
455 if (dents >= threshold || qdata >= threshold ||
456 nodes >= threshold || meta >= threshold ||
459 return dents + qdata + nodes + meta + imeta > global_threshold;
462 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
464 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
467 /* try to shrink extent cache when there is no enough memory */
468 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
469 f2fs_shrink_read_extent_tree(sbi,
470 READ_EXTENT_CACHE_SHRINK_NUMBER);
472 /* try to shrink age extent cache when there is no enough memory */
473 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
474 f2fs_shrink_age_extent_tree(sbi,
475 AGE_EXTENT_CACHE_SHRINK_NUMBER);
477 /* check the # of cached NAT entries */
478 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
479 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
481 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
482 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
484 f2fs_build_free_nids(sbi, false, false);
486 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
487 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
490 /* there is background inflight IO or foreground operation recently */
491 if (is_inflight_io(sbi, REQ_TIME) ||
492 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
495 /* exceed periodical checkpoint timeout threshold */
496 if (f2fs_time_over(sbi, CP_TIME))
499 /* checkpoint is the only way to shrink partial cached entries */
500 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
501 f2fs_available_free_memory(sbi, INO_ENTRIES))
505 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
506 struct blk_plug plug;
508 mutex_lock(&sbi->flush_lock);
510 blk_start_plug(&plug);
511 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
512 blk_finish_plug(&plug);
514 mutex_unlock(&sbi->flush_lock);
516 stat_inc_cp_call_count(sbi, BACKGROUND);
517 f2fs_sync_fs(sbi->sb, 1);
520 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
521 struct block_device *bdev)
523 int ret = blkdev_issue_flush(bdev);
525 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
526 test_opt(sbi, FLUSH_MERGE), ret);
528 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
532 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
537 if (!f2fs_is_multi_device(sbi))
538 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
540 for (i = 0; i < sbi->s_ndevs; i++) {
541 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
543 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
550 static int issue_flush_thread(void *data)
552 struct f2fs_sb_info *sbi = data;
553 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
554 wait_queue_head_t *q = &fcc->flush_wait_queue;
556 if (kthread_should_stop())
559 if (!llist_empty(&fcc->issue_list)) {
560 struct flush_cmd *cmd, *next;
563 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
564 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
566 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
568 ret = submit_flush_wait(sbi, cmd->ino);
569 atomic_inc(&fcc->issued_flush);
571 llist_for_each_entry_safe(cmd, next,
572 fcc->dispatch_list, llnode) {
574 complete(&cmd->wait);
576 fcc->dispatch_list = NULL;
579 wait_event_interruptible(*q,
580 kthread_should_stop() || !llist_empty(&fcc->issue_list));
584 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
586 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
587 struct flush_cmd cmd;
590 if (test_opt(sbi, NOBARRIER))
593 if (!test_opt(sbi, FLUSH_MERGE)) {
594 atomic_inc(&fcc->queued_flush);
595 ret = submit_flush_wait(sbi, ino);
596 atomic_dec(&fcc->queued_flush);
597 atomic_inc(&fcc->issued_flush);
601 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
602 f2fs_is_multi_device(sbi)) {
603 ret = submit_flush_wait(sbi, ino);
604 atomic_dec(&fcc->queued_flush);
606 atomic_inc(&fcc->issued_flush);
611 init_completion(&cmd.wait);
613 llist_add(&cmd.llnode, &fcc->issue_list);
616 * update issue_list before we wake up issue_flush thread, this
617 * smp_mb() pairs with another barrier in ___wait_event(), see
618 * more details in comments of waitqueue_active().
622 if (waitqueue_active(&fcc->flush_wait_queue))
623 wake_up(&fcc->flush_wait_queue);
625 if (fcc->f2fs_issue_flush) {
626 wait_for_completion(&cmd.wait);
627 atomic_dec(&fcc->queued_flush);
629 struct llist_node *list;
631 list = llist_del_all(&fcc->issue_list);
633 wait_for_completion(&cmd.wait);
634 atomic_dec(&fcc->queued_flush);
636 struct flush_cmd *tmp, *next;
638 ret = submit_flush_wait(sbi, ino);
640 llist_for_each_entry_safe(tmp, next, list, llnode) {
643 atomic_dec(&fcc->queued_flush);
647 complete(&tmp->wait);
655 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
657 dev_t dev = sbi->sb->s_bdev->bd_dev;
658 struct flush_cmd_control *fcc;
660 if (SM_I(sbi)->fcc_info) {
661 fcc = SM_I(sbi)->fcc_info;
662 if (fcc->f2fs_issue_flush)
667 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
670 atomic_set(&fcc->issued_flush, 0);
671 atomic_set(&fcc->queued_flush, 0);
672 init_waitqueue_head(&fcc->flush_wait_queue);
673 init_llist_head(&fcc->issue_list);
674 SM_I(sbi)->fcc_info = fcc;
675 if (!test_opt(sbi, FLUSH_MERGE))
679 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
680 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
681 if (IS_ERR(fcc->f2fs_issue_flush)) {
682 int err = PTR_ERR(fcc->f2fs_issue_flush);
684 fcc->f2fs_issue_flush = NULL;
691 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
693 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
695 if (fcc && fcc->f2fs_issue_flush) {
696 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
698 fcc->f2fs_issue_flush = NULL;
699 kthread_stop(flush_thread);
703 SM_I(sbi)->fcc_info = NULL;
707 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
711 if (!f2fs_is_multi_device(sbi))
714 if (test_opt(sbi, NOBARRIER))
717 for (i = 1; i < sbi->s_ndevs; i++) {
718 int count = DEFAULT_RETRY_IO_COUNT;
720 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
724 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
726 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
727 } while (ret && --count);
730 f2fs_stop_checkpoint(sbi, false,
731 STOP_CP_REASON_FLUSH_FAIL);
735 spin_lock(&sbi->dev_lock);
736 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
737 spin_unlock(&sbi->dev_lock);
743 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
744 enum dirty_type dirty_type)
746 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
748 /* need not be added */
749 if (IS_CURSEG(sbi, segno))
752 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
753 dirty_i->nr_dirty[dirty_type]++;
755 if (dirty_type == DIRTY) {
756 struct seg_entry *sentry = get_seg_entry(sbi, segno);
757 enum dirty_type t = sentry->type;
759 if (unlikely(t >= DIRTY)) {
763 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
764 dirty_i->nr_dirty[t]++;
766 if (__is_large_section(sbi)) {
767 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
768 block_t valid_blocks =
769 get_valid_blocks(sbi, segno, true);
771 f2fs_bug_on(sbi, unlikely(!valid_blocks ||
772 valid_blocks == CAP_BLKS_PER_SEC(sbi)));
774 if (!IS_CURSEC(sbi, secno))
775 set_bit(secno, dirty_i->dirty_secmap);
780 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
781 enum dirty_type dirty_type)
783 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
784 block_t valid_blocks;
786 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
787 dirty_i->nr_dirty[dirty_type]--;
789 if (dirty_type == DIRTY) {
790 struct seg_entry *sentry = get_seg_entry(sbi, segno);
791 enum dirty_type t = sentry->type;
793 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
794 dirty_i->nr_dirty[t]--;
796 valid_blocks = get_valid_blocks(sbi, segno, true);
797 if (valid_blocks == 0) {
798 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
799 dirty_i->victim_secmap);
800 #ifdef CONFIG_F2FS_CHECK_FS
801 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
804 if (__is_large_section(sbi)) {
805 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
808 valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
809 clear_bit(secno, dirty_i->dirty_secmap);
813 if (!IS_CURSEC(sbi, secno))
814 set_bit(secno, dirty_i->dirty_secmap);
820 * Should not occur error such as -ENOMEM.
821 * Adding dirty entry into seglist is not critical operation.
822 * If a given segment is one of current working segments, it won't be added.
824 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
826 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
827 unsigned short valid_blocks, ckpt_valid_blocks;
828 unsigned int usable_blocks;
830 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
833 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
834 mutex_lock(&dirty_i->seglist_lock);
836 valid_blocks = get_valid_blocks(sbi, segno, false);
837 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
839 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
840 ckpt_valid_blocks == usable_blocks)) {
841 __locate_dirty_segment(sbi, segno, PRE);
842 __remove_dirty_segment(sbi, segno, DIRTY);
843 } else if (valid_blocks < usable_blocks) {
844 __locate_dirty_segment(sbi, segno, DIRTY);
846 /* Recovery routine with SSR needs this */
847 __remove_dirty_segment(sbi, segno, DIRTY);
850 mutex_unlock(&dirty_i->seglist_lock);
853 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
854 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
856 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
859 mutex_lock(&dirty_i->seglist_lock);
860 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
861 if (get_valid_blocks(sbi, segno, false))
863 if (IS_CURSEG(sbi, segno))
865 __locate_dirty_segment(sbi, segno, PRE);
866 __remove_dirty_segment(sbi, segno, DIRTY);
868 mutex_unlock(&dirty_i->seglist_lock);
871 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
874 (overprovision_segments(sbi) - reserved_segments(sbi));
875 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
876 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
877 block_t holes[2] = {0, 0}; /* DATA and NODE */
879 struct seg_entry *se;
882 mutex_lock(&dirty_i->seglist_lock);
883 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
884 se = get_seg_entry(sbi, segno);
885 if (IS_NODESEG(se->type))
886 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
889 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
892 mutex_unlock(&dirty_i->seglist_lock);
894 unusable = max(holes[DATA], holes[NODE]);
895 if (unusable > ovp_holes)
896 return unusable - ovp_holes;
900 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
903 (overprovision_segments(sbi) - reserved_segments(sbi));
904 if (unusable > F2FS_OPTION(sbi).unusable_cap)
906 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
907 dirty_segments(sbi) > ovp_hole_segs)
912 /* This is only used by SBI_CP_DISABLED */
913 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
915 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
916 unsigned int segno = 0;
918 mutex_lock(&dirty_i->seglist_lock);
919 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
920 if (get_valid_blocks(sbi, segno, false))
922 if (get_ckpt_valid_blocks(sbi, segno, false))
924 mutex_unlock(&dirty_i->seglist_lock);
927 mutex_unlock(&dirty_i->seglist_lock);
931 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
932 struct block_device *bdev, block_t lstart,
933 block_t start, block_t len)
935 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
936 struct list_head *pend_list;
937 struct discard_cmd *dc;
939 f2fs_bug_on(sbi, !len);
941 pend_list = &dcc->pend_list[plist_idx(len)];
943 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
944 INIT_LIST_HEAD(&dc->list);
946 dc->di.lstart = lstart;
947 dc->di.start = start;
953 init_completion(&dc->wait);
954 list_add_tail(&dc->list, pend_list);
955 spin_lock_init(&dc->lock);
957 atomic_inc(&dcc->discard_cmd_cnt);
958 dcc->undiscard_blks += len;
963 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
965 #ifdef CONFIG_F2FS_CHECK_FS
966 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
967 struct rb_node *cur = rb_first_cached(&dcc->root), *next;
968 struct discard_cmd *cur_dc, *next_dc;
975 cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
976 next_dc = rb_entry(next, struct discard_cmd, rb_node);
978 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
979 f2fs_info(sbi, "broken discard_rbtree, "
980 "cur(%u, %u) next(%u, %u)",
981 cur_dc->di.lstart, cur_dc->di.len,
982 next_dc->di.lstart, next_dc->di.len);
991 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
994 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
995 struct rb_node *node = dcc->root.rb_root.rb_node;
996 struct discard_cmd *dc;
999 dc = rb_entry(node, struct discard_cmd, rb_node);
1001 if (blkaddr < dc->di.lstart)
1002 node = node->rb_left;
1003 else if (blkaddr >= dc->di.lstart + dc->di.len)
1004 node = node->rb_right;
1011 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1013 struct discard_cmd **prev_entry,
1014 struct discard_cmd **next_entry,
1015 struct rb_node ***insert_p,
1016 struct rb_node **insert_parent)
1018 struct rb_node **pnode = &root->rb_root.rb_node;
1019 struct rb_node *parent = NULL, *tmp_node;
1020 struct discard_cmd *dc;
1023 *insert_parent = NULL;
1027 if (RB_EMPTY_ROOT(&root->rb_root))
1032 dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1034 if (blkaddr < dc->di.lstart)
1035 pnode = &(*pnode)->rb_left;
1036 else if (blkaddr >= dc->di.lstart + dc->di.len)
1037 pnode = &(*pnode)->rb_right;
1039 goto lookup_neighbors;
1043 *insert_parent = parent;
1045 dc = rb_entry(parent, struct discard_cmd, rb_node);
1047 if (parent && blkaddr > dc->di.lstart)
1048 tmp_node = rb_next(parent);
1049 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1052 if (parent && blkaddr < dc->di.lstart)
1053 tmp_node = rb_prev(parent);
1054 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1058 /* lookup prev node for merging backward later */
1059 tmp_node = rb_prev(&dc->rb_node);
1060 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1062 /* lookup next node for merging frontward later */
1063 tmp_node = rb_next(&dc->rb_node);
1064 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1068 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1069 struct discard_cmd *dc)
1071 if (dc->state == D_DONE)
1072 atomic_sub(dc->queued, &dcc->queued_discard);
1074 list_del(&dc->list);
1075 rb_erase_cached(&dc->rb_node, &dcc->root);
1076 dcc->undiscard_blks -= dc->di.len;
1078 kmem_cache_free(discard_cmd_slab, dc);
1080 atomic_dec(&dcc->discard_cmd_cnt);
1083 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1084 struct discard_cmd *dc)
1086 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1087 unsigned long flags;
1089 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1091 spin_lock_irqsave(&dc->lock, flags);
1093 spin_unlock_irqrestore(&dc->lock, flags);
1096 spin_unlock_irqrestore(&dc->lock, flags);
1098 f2fs_bug_on(sbi, dc->ref);
1100 if (dc->error == -EOPNOTSUPP)
1105 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1106 KERN_INFO, sbi->sb->s_id,
1107 dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1108 __detach_discard_cmd(dcc, dc);
1111 static void f2fs_submit_discard_endio(struct bio *bio)
1113 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1114 unsigned long flags;
1116 spin_lock_irqsave(&dc->lock, flags);
1118 dc->error = blk_status_to_errno(bio->bi_status);
1120 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1122 complete_all(&dc->wait);
1124 spin_unlock_irqrestore(&dc->lock, flags);
1128 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1129 block_t start, block_t end)
1131 #ifdef CONFIG_F2FS_CHECK_FS
1132 struct seg_entry *sentry;
1134 block_t blk = start;
1135 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1139 segno = GET_SEGNO(sbi, blk);
1140 sentry = get_seg_entry(sbi, segno);
1141 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1143 if (end < START_BLOCK(sbi, segno + 1))
1144 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1147 map = (unsigned long *)(sentry->cur_valid_map);
1148 offset = __find_rev_next_bit(map, size, offset);
1149 f2fs_bug_on(sbi, offset != size);
1150 blk = START_BLOCK(sbi, segno + 1);
1155 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1156 struct discard_policy *dpolicy,
1157 int discard_type, unsigned int granularity)
1159 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1162 dpolicy->type = discard_type;
1163 dpolicy->sync = true;
1164 dpolicy->ordered = false;
1165 dpolicy->granularity = granularity;
1167 dpolicy->max_requests = dcc->max_discard_request;
1168 dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1169 dpolicy->timeout = false;
1171 if (discard_type == DPOLICY_BG) {
1172 dpolicy->min_interval = dcc->min_discard_issue_time;
1173 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1174 dpolicy->max_interval = dcc->max_discard_issue_time;
1175 dpolicy->io_aware = true;
1176 dpolicy->sync = false;
1177 dpolicy->ordered = true;
1178 if (utilization(sbi) > dcc->discard_urgent_util) {
1179 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1180 if (atomic_read(&dcc->discard_cmd_cnt))
1181 dpolicy->max_interval =
1182 dcc->min_discard_issue_time;
1184 } else if (discard_type == DPOLICY_FORCE) {
1185 dpolicy->min_interval = dcc->min_discard_issue_time;
1186 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1187 dpolicy->max_interval = dcc->max_discard_issue_time;
1188 dpolicy->io_aware = false;
1189 } else if (discard_type == DPOLICY_FSTRIM) {
1190 dpolicy->io_aware = false;
1191 } else if (discard_type == DPOLICY_UMOUNT) {
1192 dpolicy->io_aware = false;
1193 /* we need to issue all to keep CP_TRIMMED_FLAG */
1194 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1195 dpolicy->timeout = true;
1199 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1200 struct block_device *bdev, block_t lstart,
1201 block_t start, block_t len);
1203 #ifdef CONFIG_BLK_DEV_ZONED
1204 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1205 struct discard_cmd *dc, blk_opf_t flag,
1206 struct list_head *wait_list,
1207 unsigned int *issued)
1209 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1210 struct block_device *bdev = dc->bdev;
1211 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1212 unsigned long flags;
1214 trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1216 spin_lock_irqsave(&dc->lock, flags);
1217 dc->state = D_SUBMIT;
1219 spin_unlock_irqrestore(&dc->lock, flags);
1224 atomic_inc(&dcc->queued_discard);
1226 list_move_tail(&dc->list, wait_list);
1228 /* sanity check on discard range */
1229 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1231 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1232 bio->bi_private = dc;
1233 bio->bi_end_io = f2fs_submit_discard_endio;
1236 atomic_inc(&dcc->issued_discard);
1237 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1241 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1242 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1243 struct discard_policy *dpolicy,
1244 struct discard_cmd *dc, int *issued)
1246 struct block_device *bdev = dc->bdev;
1247 unsigned int max_discard_blocks =
1248 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1249 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1250 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1251 &(dcc->fstrim_list) : &(dcc->wait_list);
1252 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1253 block_t lstart, start, len, total_len;
1256 if (dc->state != D_PREP)
1259 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1262 #ifdef CONFIG_BLK_DEV_ZONED
1263 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1264 int devi = f2fs_bdev_index(sbi, bdev);
1269 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1270 __submit_zone_reset_cmd(sbi, dc, flag,
1277 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1279 lstart = dc->di.lstart;
1280 start = dc->di.start;
1286 while (total_len && *issued < dpolicy->max_requests && !err) {
1287 struct bio *bio = NULL;
1288 unsigned long flags;
1291 if (len > max_discard_blocks) {
1292 len = max_discard_blocks;
1297 if (*issued == dpolicy->max_requests)
1302 if (time_to_inject(sbi, FAULT_DISCARD)) {
1305 err = __blkdev_issue_discard(bdev,
1306 SECTOR_FROM_BLOCK(start),
1307 SECTOR_FROM_BLOCK(len),
1311 spin_lock_irqsave(&dc->lock, flags);
1312 if (dc->state == D_PARTIAL)
1313 dc->state = D_SUBMIT;
1314 spin_unlock_irqrestore(&dc->lock, flags);
1319 f2fs_bug_on(sbi, !bio);
1322 * should keep before submission to avoid D_DONE
1325 spin_lock_irqsave(&dc->lock, flags);
1327 dc->state = D_SUBMIT;
1329 dc->state = D_PARTIAL;
1331 spin_unlock_irqrestore(&dc->lock, flags);
1333 atomic_inc(&dcc->queued_discard);
1335 list_move_tail(&dc->list, wait_list);
1337 /* sanity check on discard range */
1338 __check_sit_bitmap(sbi, lstart, lstart + len);
1340 bio->bi_private = dc;
1341 bio->bi_end_io = f2fs_submit_discard_endio;
1342 bio->bi_opf |= flag;
1345 atomic_inc(&dcc->issued_discard);
1347 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1356 dcc->undiscard_blks -= len;
1357 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1362 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1363 struct block_device *bdev, block_t lstart,
1364 block_t start, block_t len)
1366 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1367 struct rb_node **p = &dcc->root.rb_root.rb_node;
1368 struct rb_node *parent = NULL;
1369 struct discard_cmd *dc;
1370 bool leftmost = true;
1372 /* look up rb tree to find parent node */
1375 dc = rb_entry(parent, struct discard_cmd, rb_node);
1377 if (lstart < dc->di.lstart) {
1379 } else if (lstart >= dc->di.lstart + dc->di.len) {
1380 p = &(*p)->rb_right;
1383 f2fs_bug_on(sbi, 1);
1387 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1389 rb_link_node(&dc->rb_node, parent, p);
1390 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1393 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1394 struct discard_cmd *dc)
1396 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1399 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1400 struct discard_cmd *dc, block_t blkaddr)
1402 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1403 struct discard_info di = dc->di;
1404 bool modified = false;
1406 if (dc->state == D_DONE || dc->di.len == 1) {
1407 __remove_discard_cmd(sbi, dc);
1411 dcc->undiscard_blks -= di.len;
1413 if (blkaddr > di.lstart) {
1414 dc->di.len = blkaddr - dc->di.lstart;
1415 dcc->undiscard_blks += dc->di.len;
1416 __relocate_discard_cmd(dcc, dc);
1420 if (blkaddr < di.lstart + di.len - 1) {
1422 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1423 di.start + blkaddr + 1 - di.lstart,
1424 di.lstart + di.len - 1 - blkaddr);
1429 dcc->undiscard_blks += dc->di.len;
1430 __relocate_discard_cmd(dcc, dc);
1435 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1436 struct block_device *bdev, block_t lstart,
1437 block_t start, block_t len)
1439 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1440 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1441 struct discard_cmd *dc;
1442 struct discard_info di = {0};
1443 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1444 unsigned int max_discard_blocks =
1445 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1446 block_t end = lstart + len;
1448 dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1449 &prev_dc, &next_dc, &insert_p, &insert_parent);
1455 di.len = next_dc ? next_dc->di.lstart - lstart : len;
1456 di.len = min(di.len, len);
1461 struct rb_node *node;
1462 bool merged = false;
1463 struct discard_cmd *tdc = NULL;
1466 di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1467 if (di.lstart < lstart)
1469 if (di.lstart >= end)
1472 if (!next_dc || next_dc->di.lstart > end)
1473 di.len = end - di.lstart;
1475 di.len = next_dc->di.lstart - di.lstart;
1476 di.start = start + di.lstart - lstart;
1482 if (prev_dc && prev_dc->state == D_PREP &&
1483 prev_dc->bdev == bdev &&
1484 __is_discard_back_mergeable(&di, &prev_dc->di,
1485 max_discard_blocks)) {
1486 prev_dc->di.len += di.len;
1487 dcc->undiscard_blks += di.len;
1488 __relocate_discard_cmd(dcc, prev_dc);
1494 if (next_dc && next_dc->state == D_PREP &&
1495 next_dc->bdev == bdev &&
1496 __is_discard_front_mergeable(&di, &next_dc->di,
1497 max_discard_blocks)) {
1498 next_dc->di.lstart = di.lstart;
1499 next_dc->di.len += di.len;
1500 next_dc->di.start = di.start;
1501 dcc->undiscard_blks += di.len;
1502 __relocate_discard_cmd(dcc, next_dc);
1504 __remove_discard_cmd(sbi, tdc);
1509 __insert_discard_cmd(sbi, bdev,
1510 di.lstart, di.start, di.len);
1516 node = rb_next(&prev_dc->rb_node);
1517 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1521 #ifdef CONFIG_BLK_DEV_ZONED
1522 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1523 struct block_device *bdev, block_t blkstart, block_t lblkstart,
1526 trace_f2fs_queue_reset_zone(bdev, blkstart);
1528 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1529 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1530 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1534 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1535 struct block_device *bdev, block_t blkstart, block_t blklen)
1537 block_t lblkstart = blkstart;
1539 if (!f2fs_bdev_support_discard(bdev))
1542 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1544 if (f2fs_is_multi_device(sbi)) {
1545 int devi = f2fs_target_device_index(sbi, blkstart);
1547 blkstart -= FDEV(devi).start_blk;
1549 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1550 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1551 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1554 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1555 struct discard_policy *dpolicy, int *issued)
1557 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1558 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1559 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1560 struct discard_cmd *dc;
1561 struct blk_plug plug;
1562 bool io_interrupted = false;
1564 mutex_lock(&dcc->cmd_lock);
1565 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1566 &prev_dc, &next_dc, &insert_p, &insert_parent);
1570 blk_start_plug(&plug);
1573 struct rb_node *node;
1576 if (dc->state != D_PREP)
1579 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1580 io_interrupted = true;
1584 dcc->next_pos = dc->di.lstart + dc->di.len;
1585 err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1587 if (*issued >= dpolicy->max_requests)
1590 node = rb_next(&dc->rb_node);
1592 __remove_discard_cmd(sbi, dc);
1593 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1596 blk_finish_plug(&plug);
1601 mutex_unlock(&dcc->cmd_lock);
1603 if (!(*issued) && io_interrupted)
1606 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1607 struct discard_policy *dpolicy);
1609 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1610 struct discard_policy *dpolicy)
1612 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1613 struct list_head *pend_list;
1614 struct discard_cmd *dc, *tmp;
1615 struct blk_plug plug;
1617 bool io_interrupted = false;
1619 if (dpolicy->timeout)
1620 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1624 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1625 if (dpolicy->timeout &&
1626 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1629 if (i + 1 < dpolicy->granularity)
1632 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1633 __issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1637 pend_list = &dcc->pend_list[i];
1639 mutex_lock(&dcc->cmd_lock);
1640 if (list_empty(pend_list))
1642 if (unlikely(dcc->rbtree_check))
1643 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1644 blk_start_plug(&plug);
1645 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1646 f2fs_bug_on(sbi, dc->state != D_PREP);
1648 if (dpolicy->timeout &&
1649 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1652 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1653 !is_idle(sbi, DISCARD_TIME)) {
1654 io_interrupted = true;
1658 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1660 if (issued >= dpolicy->max_requests)
1663 blk_finish_plug(&plug);
1665 mutex_unlock(&dcc->cmd_lock);
1667 if (issued >= dpolicy->max_requests || io_interrupted)
1671 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1672 __wait_all_discard_cmd(sbi, dpolicy);
1676 if (!issued && io_interrupted)
1682 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1684 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1685 struct list_head *pend_list;
1686 struct discard_cmd *dc, *tmp;
1688 bool dropped = false;
1690 mutex_lock(&dcc->cmd_lock);
1691 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1692 pend_list = &dcc->pend_list[i];
1693 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1694 f2fs_bug_on(sbi, dc->state != D_PREP);
1695 __remove_discard_cmd(sbi, dc);
1699 mutex_unlock(&dcc->cmd_lock);
1704 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1706 __drop_discard_cmd(sbi);
1709 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1710 struct discard_cmd *dc)
1712 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1713 unsigned int len = 0;
1715 wait_for_completion_io(&dc->wait);
1716 mutex_lock(&dcc->cmd_lock);
1717 f2fs_bug_on(sbi, dc->state != D_DONE);
1722 __remove_discard_cmd(sbi, dc);
1724 mutex_unlock(&dcc->cmd_lock);
1729 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1730 struct discard_policy *dpolicy,
1731 block_t start, block_t end)
1733 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1734 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1735 &(dcc->fstrim_list) : &(dcc->wait_list);
1736 struct discard_cmd *dc = NULL, *iter, *tmp;
1737 unsigned int trimmed = 0;
1742 mutex_lock(&dcc->cmd_lock);
1743 list_for_each_entry_safe(iter, tmp, wait_list, list) {
1744 if (iter->di.lstart + iter->di.len <= start ||
1745 end <= iter->di.lstart)
1747 if (iter->di.len < dpolicy->granularity)
1749 if (iter->state == D_DONE && !iter->ref) {
1750 wait_for_completion_io(&iter->wait);
1752 trimmed += iter->di.len;
1753 __remove_discard_cmd(sbi, iter);
1760 mutex_unlock(&dcc->cmd_lock);
1763 trimmed += __wait_one_discard_bio(sbi, dc);
1770 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1771 struct discard_policy *dpolicy)
1773 struct discard_policy dp;
1774 unsigned int discard_blks;
1777 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1780 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1781 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1782 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1783 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1785 return discard_blks;
1788 /* This should be covered by global mutex, &sit_i->sentry_lock */
1789 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1791 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1792 struct discard_cmd *dc;
1793 bool need_wait = false;
1795 mutex_lock(&dcc->cmd_lock);
1796 dc = __lookup_discard_cmd(sbi, blkaddr);
1797 #ifdef CONFIG_BLK_DEV_ZONED
1798 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1799 int devi = f2fs_bdev_index(sbi, dc->bdev);
1802 mutex_unlock(&dcc->cmd_lock);
1806 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1807 /* force submit zone reset */
1808 if (dc->state == D_PREP)
1809 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1810 &dcc->wait_list, NULL);
1812 mutex_unlock(&dcc->cmd_lock);
1813 /* wait zone reset */
1814 __wait_one_discard_bio(sbi, dc);
1820 if (dc->state == D_PREP) {
1821 __punch_discard_cmd(sbi, dc, blkaddr);
1827 mutex_unlock(&dcc->cmd_lock);
1830 __wait_one_discard_bio(sbi, dc);
1833 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1835 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1837 if (dcc && dcc->f2fs_issue_discard) {
1838 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1840 dcc->f2fs_issue_discard = NULL;
1841 kthread_stop(discard_thread);
1846 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1847 * @sbi: the f2fs_sb_info data for discard cmd to issue
1849 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1851 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1853 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1855 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1856 struct discard_policy dpolicy;
1859 if (!atomic_read(&dcc->discard_cmd_cnt))
1862 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1863 dcc->discard_granularity);
1864 __issue_discard_cmd(sbi, &dpolicy);
1865 dropped = __drop_discard_cmd(sbi);
1867 /* just to make sure there is no pending discard commands */
1868 __wait_all_discard_cmd(sbi, NULL);
1870 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1874 static int issue_discard_thread(void *data)
1876 struct f2fs_sb_info *sbi = data;
1877 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1878 wait_queue_head_t *q = &dcc->discard_wait_queue;
1879 struct discard_policy dpolicy;
1880 unsigned int wait_ms = dcc->min_discard_issue_time;
1886 wait_event_interruptible_timeout(*q,
1887 kthread_should_stop() || freezing(current) ||
1889 msecs_to_jiffies(wait_ms));
1891 if (sbi->gc_mode == GC_URGENT_HIGH ||
1892 !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1893 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1894 MIN_DISCARD_GRANULARITY);
1896 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1897 dcc->discard_granularity);
1899 if (dcc->discard_wake)
1900 dcc->discard_wake = false;
1902 /* clean up pending candidates before going to sleep */
1903 if (atomic_read(&dcc->queued_discard))
1904 __wait_all_discard_cmd(sbi, NULL);
1906 if (try_to_freeze())
1908 if (f2fs_readonly(sbi->sb))
1910 if (kthread_should_stop())
1912 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1913 !atomic_read(&dcc->discard_cmd_cnt)) {
1914 wait_ms = dpolicy.max_interval;
1918 sb_start_intwrite(sbi->sb);
1920 issued = __issue_discard_cmd(sbi, &dpolicy);
1922 __wait_all_discard_cmd(sbi, &dpolicy);
1923 wait_ms = dpolicy.min_interval;
1924 } else if (issued == -1) {
1925 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1927 wait_ms = dpolicy.mid_interval;
1929 wait_ms = dpolicy.max_interval;
1931 if (!atomic_read(&dcc->discard_cmd_cnt))
1932 wait_ms = dpolicy.max_interval;
1934 sb_end_intwrite(sbi->sb);
1936 } while (!kthread_should_stop());
1940 #ifdef CONFIG_BLK_DEV_ZONED
1941 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1942 struct block_device *bdev, block_t blkstart, block_t blklen)
1944 sector_t sector, nr_sects;
1945 block_t lblkstart = blkstart;
1949 if (f2fs_is_multi_device(sbi)) {
1950 devi = f2fs_target_device_index(sbi, blkstart);
1951 if (blkstart < FDEV(devi).start_blk ||
1952 blkstart > FDEV(devi).end_blk) {
1953 f2fs_err(sbi, "Invalid block %x", blkstart);
1956 blkstart -= FDEV(devi).start_blk;
1959 /* For sequential zones, reset the zone write pointer */
1960 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1961 sector = SECTOR_FROM_BLOCK(blkstart);
1962 nr_sects = SECTOR_FROM_BLOCK(blklen);
1963 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1965 if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1966 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1967 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1972 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1973 trace_f2fs_issue_reset_zone(bdev, blkstart);
1974 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1975 sector, nr_sects, GFP_NOFS);
1978 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1982 /* For conventional zones, use regular discard if supported */
1983 __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1988 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1989 struct block_device *bdev, block_t blkstart, block_t blklen)
1991 #ifdef CONFIG_BLK_DEV_ZONED
1992 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1993 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1995 __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1999 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2000 block_t blkstart, block_t blklen)
2002 sector_t start = blkstart, len = 0;
2003 struct block_device *bdev;
2004 struct seg_entry *se;
2005 unsigned int offset;
2009 bdev = f2fs_target_device(sbi, blkstart, NULL);
2011 for (i = blkstart; i < blkstart + blklen; i++, len++) {
2013 struct block_device *bdev2 =
2014 f2fs_target_device(sbi, i, NULL);
2016 if (bdev2 != bdev) {
2017 err = __issue_discard_async(sbi, bdev,
2027 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2028 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2030 if (f2fs_block_unit_discard(sbi) &&
2031 !f2fs_test_and_set_bit(offset, se->discard_map))
2032 sbi->discard_blks--;
2036 err = __issue_discard_async(sbi, bdev, start, len);
2040 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2043 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2044 int max_blocks = sbi->blocks_per_seg;
2045 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2046 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2047 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2048 unsigned long *discard_map = (unsigned long *)se->discard_map;
2049 unsigned long *dmap = SIT_I(sbi)->tmp_map;
2050 unsigned int start = 0, end = -1;
2051 bool force = (cpc->reason & CP_DISCARD);
2052 struct discard_entry *de = NULL;
2053 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2056 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
2057 !f2fs_block_unit_discard(sbi))
2061 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2062 SM_I(sbi)->dcc_info->nr_discards >=
2063 SM_I(sbi)->dcc_info->max_discards)
2067 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2068 for (i = 0; i < entries; i++)
2069 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2070 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2072 while (force || SM_I(sbi)->dcc_info->nr_discards <=
2073 SM_I(sbi)->dcc_info->max_discards) {
2074 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
2075 if (start >= max_blocks)
2078 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
2079 if (force && start && end != max_blocks
2080 && (end - start) < cpc->trim_minlen)
2087 de = f2fs_kmem_cache_alloc(discard_entry_slab,
2088 GFP_F2FS_ZERO, true, NULL);
2089 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2090 list_add_tail(&de->list, head);
2093 for (i = start; i < end; i++)
2094 __set_bit_le(i, (void *)de->discard_map);
2096 SM_I(sbi)->dcc_info->nr_discards += end - start;
2101 static void release_discard_addr(struct discard_entry *entry)
2103 list_del(&entry->list);
2104 kmem_cache_free(discard_entry_slab, entry);
2107 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2109 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2110 struct discard_entry *entry, *this;
2113 list_for_each_entry_safe(entry, this, head, list)
2114 release_discard_addr(entry);
2118 * Should call f2fs_clear_prefree_segments after checkpoint is done.
2120 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2122 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2125 mutex_lock(&dirty_i->seglist_lock);
2126 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2127 __set_test_and_free(sbi, segno, false);
2128 mutex_unlock(&dirty_i->seglist_lock);
2131 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2132 struct cp_control *cpc)
2134 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2135 struct list_head *head = &dcc->entry_list;
2136 struct discard_entry *entry, *this;
2137 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2138 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2139 unsigned int start = 0, end = -1;
2140 unsigned int secno, start_segno;
2141 bool force = (cpc->reason & CP_DISCARD);
2142 bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2143 DISCARD_UNIT_SECTION;
2145 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2146 section_alignment = true;
2148 mutex_lock(&dirty_i->seglist_lock);
2153 if (section_alignment && end != -1)
2155 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2156 if (start >= MAIN_SEGS(sbi))
2158 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2161 if (section_alignment) {
2162 start = rounddown(start, sbi->segs_per_sec);
2163 end = roundup(end, sbi->segs_per_sec);
2166 for (i = start; i < end; i++) {
2167 if (test_and_clear_bit(i, prefree_map))
2168 dirty_i->nr_dirty[PRE]--;
2171 if (!f2fs_realtime_discard_enable(sbi))
2174 if (force && start >= cpc->trim_start &&
2175 (end - 1) <= cpc->trim_end)
2178 /* Should cover 2MB zoned device for zone-based reset */
2179 if (!f2fs_sb_has_blkzoned(sbi) &&
2180 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2181 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2182 (end - start) << sbi->log_blocks_per_seg);
2186 secno = GET_SEC_FROM_SEG(sbi, start);
2187 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2188 if (!IS_CURSEC(sbi, secno) &&
2189 !get_valid_blocks(sbi, start, true))
2190 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2191 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2193 start = start_segno + sbi->segs_per_sec;
2199 mutex_unlock(&dirty_i->seglist_lock);
2201 if (!f2fs_block_unit_discard(sbi))
2204 /* send small discards */
2205 list_for_each_entry_safe(entry, this, head, list) {
2206 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2207 bool is_valid = test_bit_le(0, entry->discard_map);
2211 next_pos = find_next_zero_bit_le(entry->discard_map,
2212 sbi->blocks_per_seg, cur_pos);
2213 len = next_pos - cur_pos;
2215 if (f2fs_sb_has_blkzoned(sbi) ||
2216 (force && len < cpc->trim_minlen))
2219 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2223 next_pos = find_next_bit_le(entry->discard_map,
2224 sbi->blocks_per_seg, cur_pos);
2228 is_valid = !is_valid;
2230 if (cur_pos < sbi->blocks_per_seg)
2233 release_discard_addr(entry);
2234 dcc->nr_discards -= total_len;
2238 wake_up_discard_thread(sbi, false);
2241 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2243 dev_t dev = sbi->sb->s_bdev->bd_dev;
2244 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2247 if (!f2fs_realtime_discard_enable(sbi))
2250 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2251 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2252 if (IS_ERR(dcc->f2fs_issue_discard)) {
2253 err = PTR_ERR(dcc->f2fs_issue_discard);
2254 dcc->f2fs_issue_discard = NULL;
2260 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2262 struct discard_cmd_control *dcc;
2265 if (SM_I(sbi)->dcc_info) {
2266 dcc = SM_I(sbi)->dcc_info;
2270 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2274 dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2275 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2276 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2277 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2278 dcc->discard_granularity = sbi->blocks_per_seg;
2279 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2280 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2282 INIT_LIST_HEAD(&dcc->entry_list);
2283 for (i = 0; i < MAX_PLIST_NUM; i++)
2284 INIT_LIST_HEAD(&dcc->pend_list[i]);
2285 INIT_LIST_HEAD(&dcc->wait_list);
2286 INIT_LIST_HEAD(&dcc->fstrim_list);
2287 mutex_init(&dcc->cmd_lock);
2288 atomic_set(&dcc->issued_discard, 0);
2289 atomic_set(&dcc->queued_discard, 0);
2290 atomic_set(&dcc->discard_cmd_cnt, 0);
2291 dcc->nr_discards = 0;
2292 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2293 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2294 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2295 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2296 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2297 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2298 dcc->undiscard_blks = 0;
2300 dcc->root = RB_ROOT_CACHED;
2301 dcc->rbtree_check = false;
2303 init_waitqueue_head(&dcc->discard_wait_queue);
2304 SM_I(sbi)->dcc_info = dcc;
2306 err = f2fs_start_discard_thread(sbi);
2309 SM_I(sbi)->dcc_info = NULL;
2315 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2317 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2322 f2fs_stop_discard_thread(sbi);
2325 * Recovery can cache discard commands, so in error path of
2326 * fill_super(), it needs to give a chance to handle them.
2328 f2fs_issue_discard_timeout(sbi);
2331 SM_I(sbi)->dcc_info = NULL;
2334 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2336 struct sit_info *sit_i = SIT_I(sbi);
2338 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2339 sit_i->dirty_sentries++;
2346 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2347 unsigned int segno, int modified)
2349 struct seg_entry *se = get_seg_entry(sbi, segno);
2353 __mark_sit_entry_dirty(sbi, segno);
2356 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2359 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2361 if (segno == NULL_SEGNO)
2363 return get_seg_entry(sbi, segno)->mtime;
2366 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2367 unsigned long long old_mtime)
2369 struct seg_entry *se;
2370 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2371 unsigned long long ctime = get_mtime(sbi, false);
2372 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2374 if (segno == NULL_SEGNO)
2377 se = get_seg_entry(sbi, segno);
2382 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2383 se->valid_blocks + 1);
2385 if (ctime > SIT_I(sbi)->max_mtime)
2386 SIT_I(sbi)->max_mtime = ctime;
2389 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2391 struct seg_entry *se;
2392 unsigned int segno, offset;
2393 long int new_vblocks;
2395 #ifdef CONFIG_F2FS_CHECK_FS
2399 segno = GET_SEGNO(sbi, blkaddr);
2401 se = get_seg_entry(sbi, segno);
2402 new_vblocks = se->valid_blocks + del;
2403 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2405 f2fs_bug_on(sbi, (new_vblocks < 0 ||
2406 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2408 se->valid_blocks = new_vblocks;
2410 /* Update valid block bitmap */
2412 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2413 #ifdef CONFIG_F2FS_CHECK_FS
2414 mir_exist = f2fs_test_and_set_bit(offset,
2415 se->cur_valid_map_mir);
2416 if (unlikely(exist != mir_exist)) {
2417 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2419 f2fs_bug_on(sbi, 1);
2422 if (unlikely(exist)) {
2423 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2425 f2fs_bug_on(sbi, 1);
2430 if (f2fs_block_unit_discard(sbi) &&
2431 !f2fs_test_and_set_bit(offset, se->discard_map))
2432 sbi->discard_blks--;
2435 * SSR should never reuse block which is checkpointed
2436 * or newly invalidated.
2438 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2439 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2440 se->ckpt_valid_blocks++;
2443 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2444 #ifdef CONFIG_F2FS_CHECK_FS
2445 mir_exist = f2fs_test_and_clear_bit(offset,
2446 se->cur_valid_map_mir);
2447 if (unlikely(exist != mir_exist)) {
2448 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2450 f2fs_bug_on(sbi, 1);
2453 if (unlikely(!exist)) {
2454 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2456 f2fs_bug_on(sbi, 1);
2459 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2461 * If checkpoints are off, we must not reuse data that
2462 * was used in the previous checkpoint. If it was used
2463 * before, we must track that to know how much space we
2466 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2467 spin_lock(&sbi->stat_lock);
2468 sbi->unusable_block_count++;
2469 spin_unlock(&sbi->stat_lock);
2473 if (f2fs_block_unit_discard(sbi) &&
2474 f2fs_test_and_clear_bit(offset, se->discard_map))
2475 sbi->discard_blks++;
2477 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2478 se->ckpt_valid_blocks += del;
2480 __mark_sit_entry_dirty(sbi, segno);
2482 /* update total number of valid blocks to be written in ckpt area */
2483 SIT_I(sbi)->written_valid_blocks += del;
2485 if (__is_large_section(sbi))
2486 get_sec_entry(sbi, segno)->valid_blocks += del;
2489 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2491 unsigned int segno = GET_SEGNO(sbi, addr);
2492 struct sit_info *sit_i = SIT_I(sbi);
2494 f2fs_bug_on(sbi, addr == NULL_ADDR);
2495 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2498 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2499 f2fs_invalidate_compress_page(sbi, addr);
2501 /* add it into sit main buffer */
2502 down_write(&sit_i->sentry_lock);
2504 update_segment_mtime(sbi, addr, 0);
2505 update_sit_entry(sbi, addr, -1);
2507 /* add it into dirty seglist */
2508 locate_dirty_segment(sbi, segno);
2510 up_write(&sit_i->sentry_lock);
2513 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2515 struct sit_info *sit_i = SIT_I(sbi);
2516 unsigned int segno, offset;
2517 struct seg_entry *se;
2520 if (!__is_valid_data_blkaddr(blkaddr))
2523 down_read(&sit_i->sentry_lock);
2525 segno = GET_SEGNO(sbi, blkaddr);
2526 se = get_seg_entry(sbi, segno);
2527 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2529 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2532 up_read(&sit_i->sentry_lock);
2537 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2539 struct curseg_info *curseg = CURSEG_I(sbi, type);
2541 if (sbi->ckpt->alloc_type[type] == SSR)
2542 return sbi->blocks_per_seg;
2543 return curseg->next_blkoff;
2547 * Calculate the number of current summary pages for writing
2549 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2551 int valid_sum_count = 0;
2554 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2555 if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2557 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2559 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2562 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2563 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2564 if (valid_sum_count <= sum_in_page)
2566 else if ((valid_sum_count - sum_in_page) <=
2567 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2573 * Caller should put this summary page
2575 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2577 if (unlikely(f2fs_cp_error(sbi)))
2578 return ERR_PTR(-EIO);
2579 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2582 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2583 void *src, block_t blk_addr)
2585 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2587 memcpy(page_address(page), src, PAGE_SIZE);
2588 set_page_dirty(page);
2589 f2fs_put_page(page, 1);
2592 static void write_sum_page(struct f2fs_sb_info *sbi,
2593 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2595 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2598 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2599 int type, block_t blk_addr)
2601 struct curseg_info *curseg = CURSEG_I(sbi, type);
2602 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2603 struct f2fs_summary_block *src = curseg->sum_blk;
2604 struct f2fs_summary_block *dst;
2606 dst = (struct f2fs_summary_block *)page_address(page);
2607 memset(dst, 0, PAGE_SIZE);
2609 mutex_lock(&curseg->curseg_mutex);
2611 down_read(&curseg->journal_rwsem);
2612 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2613 up_read(&curseg->journal_rwsem);
2615 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2616 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2618 mutex_unlock(&curseg->curseg_mutex);
2620 set_page_dirty(page);
2621 f2fs_put_page(page, 1);
2624 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2625 struct curseg_info *curseg, int type)
2627 unsigned int segno = curseg->segno + 1;
2628 struct free_segmap_info *free_i = FREE_I(sbi);
2630 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2631 return !test_bit(segno, free_i->free_segmap);
2636 * Find a new segment from the free segments bitmap to right order
2637 * This function should be returned with success, otherwise BUG
2639 static void get_new_segment(struct f2fs_sb_info *sbi,
2640 unsigned int *newseg, bool new_sec, int dir)
2642 struct free_segmap_info *free_i = FREE_I(sbi);
2643 unsigned int segno, secno, zoneno;
2644 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2645 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2646 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2647 unsigned int left_start = hint;
2652 spin_lock(&free_i->segmap_lock);
2654 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2655 segno = find_next_zero_bit(free_i->free_segmap,
2656 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2657 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2661 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2662 if (secno >= MAIN_SECS(sbi)) {
2663 if (dir == ALLOC_RIGHT) {
2664 secno = find_first_zero_bit(free_i->free_secmap,
2666 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2669 left_start = hint - 1;
2675 while (test_bit(left_start, free_i->free_secmap)) {
2676 if (left_start > 0) {
2680 left_start = find_first_zero_bit(free_i->free_secmap,
2682 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2687 segno = GET_SEG_FROM_SEC(sbi, secno);
2688 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2690 /* give up on finding another zone */
2693 if (sbi->secs_per_zone == 1)
2695 if (zoneno == old_zoneno)
2697 if (dir == ALLOC_LEFT) {
2698 if (!go_left && zoneno + 1 >= total_zones)
2700 if (go_left && zoneno == 0)
2703 for (i = 0; i < NR_CURSEG_TYPE; i++)
2704 if (CURSEG_I(sbi, i)->zone == zoneno)
2707 if (i < NR_CURSEG_TYPE) {
2708 /* zone is in user, try another */
2710 hint = zoneno * sbi->secs_per_zone - 1;
2711 else if (zoneno + 1 >= total_zones)
2714 hint = (zoneno + 1) * sbi->secs_per_zone;
2716 goto find_other_zone;
2719 /* set it as dirty segment in free segmap */
2720 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2721 __set_inuse(sbi, segno);
2723 spin_unlock(&free_i->segmap_lock);
2726 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2728 struct curseg_info *curseg = CURSEG_I(sbi, type);
2729 struct summary_footer *sum_footer;
2730 unsigned short seg_type = curseg->seg_type;
2732 curseg->inited = true;
2733 curseg->segno = curseg->next_segno;
2734 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2735 curseg->next_blkoff = 0;
2736 curseg->next_segno = NULL_SEGNO;
2738 sum_footer = &(curseg->sum_blk->footer);
2739 memset(sum_footer, 0, sizeof(struct summary_footer));
2741 sanity_check_seg_type(sbi, seg_type);
2743 if (IS_DATASEG(seg_type))
2744 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2745 if (IS_NODESEG(seg_type))
2746 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2747 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2750 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2752 struct curseg_info *curseg = CURSEG_I(sbi, type);
2753 unsigned short seg_type = curseg->seg_type;
2755 sanity_check_seg_type(sbi, seg_type);
2756 if (f2fs_need_rand_seg(sbi))
2757 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
2759 /* if segs_per_sec is large than 1, we need to keep original policy. */
2760 if (__is_large_section(sbi))
2761 return curseg->segno;
2763 /* inmem log may not locate on any segment after mount */
2764 if (!curseg->inited)
2767 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2770 if (test_opt(sbi, NOHEAP) &&
2771 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2774 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2775 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2777 /* find segments from 0 to reuse freed segments */
2778 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2781 return curseg->segno;
2785 * Allocate a current working segment.
2786 * This function always allocates a free segment in LFS manner.
2788 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2790 struct curseg_info *curseg = CURSEG_I(sbi, type);
2791 unsigned short seg_type = curseg->seg_type;
2792 unsigned int segno = curseg->segno;
2793 int dir = ALLOC_LEFT;
2796 write_sum_page(sbi, curseg->sum_blk,
2797 GET_SUM_BLOCK(sbi, segno));
2798 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2801 if (test_opt(sbi, NOHEAP))
2804 segno = __get_next_segno(sbi, type);
2805 get_new_segment(sbi, &segno, new_sec, dir);
2806 curseg->next_segno = segno;
2807 reset_curseg(sbi, type, 1);
2808 curseg->alloc_type = LFS;
2809 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2810 curseg->fragment_remained_chunk =
2811 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2814 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2815 int segno, block_t start)
2817 struct seg_entry *se = get_seg_entry(sbi, segno);
2818 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2819 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2820 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2821 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2824 for (i = 0; i < entries; i++)
2825 target_map[i] = ckpt_map[i] | cur_map[i];
2827 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2830 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2831 struct curseg_info *seg)
2833 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2836 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2838 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2842 * This function always allocates a used segment(from dirty seglist) by SSR
2843 * manner, so it should recover the existing segment information of valid blocks
2845 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2847 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2848 struct curseg_info *curseg = CURSEG_I(sbi, type);
2849 unsigned int new_segno = curseg->next_segno;
2850 struct f2fs_summary_block *sum_node;
2851 struct page *sum_page;
2853 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2855 __set_test_and_inuse(sbi, new_segno);
2857 mutex_lock(&dirty_i->seglist_lock);
2858 __remove_dirty_segment(sbi, new_segno, PRE);
2859 __remove_dirty_segment(sbi, new_segno, DIRTY);
2860 mutex_unlock(&dirty_i->seglist_lock);
2862 reset_curseg(sbi, type, 1);
2863 curseg->alloc_type = SSR;
2864 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2866 sum_page = f2fs_get_sum_page(sbi, new_segno);
2867 if (IS_ERR(sum_page)) {
2868 /* GC won't be able to use stale summary pages by cp_error */
2869 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2872 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2873 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2874 f2fs_put_page(sum_page, 1);
2877 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2878 int alloc_mode, unsigned long long age);
2880 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2881 int target_type, int alloc_mode,
2882 unsigned long long age)
2884 struct curseg_info *curseg = CURSEG_I(sbi, type);
2886 curseg->seg_type = target_type;
2888 if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2889 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2891 curseg->seg_type = se->type;
2892 change_curseg(sbi, type);
2894 /* allocate cold segment by default */
2895 curseg->seg_type = CURSEG_COLD_DATA;
2896 new_curseg(sbi, type, true);
2898 stat_inc_seg_type(sbi, curseg);
2901 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2903 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2905 if (!sbi->am.atgc_enabled)
2908 f2fs_down_read(&SM_I(sbi)->curseg_lock);
2910 mutex_lock(&curseg->curseg_mutex);
2911 down_write(&SIT_I(sbi)->sentry_lock);
2913 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2915 up_write(&SIT_I(sbi)->sentry_lock);
2916 mutex_unlock(&curseg->curseg_mutex);
2918 f2fs_up_read(&SM_I(sbi)->curseg_lock);
2921 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2923 __f2fs_init_atgc_curseg(sbi);
2926 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2928 struct curseg_info *curseg = CURSEG_I(sbi, type);
2930 mutex_lock(&curseg->curseg_mutex);
2931 if (!curseg->inited)
2934 if (get_valid_blocks(sbi, curseg->segno, false)) {
2935 write_sum_page(sbi, curseg->sum_blk,
2936 GET_SUM_BLOCK(sbi, curseg->segno));
2938 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2939 __set_test_and_free(sbi, curseg->segno, true);
2940 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2943 mutex_unlock(&curseg->curseg_mutex);
2946 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2948 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2950 if (sbi->am.atgc_enabled)
2951 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2954 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2956 struct curseg_info *curseg = CURSEG_I(sbi, type);
2958 mutex_lock(&curseg->curseg_mutex);
2959 if (!curseg->inited)
2961 if (get_valid_blocks(sbi, curseg->segno, false))
2964 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2965 __set_test_and_inuse(sbi, curseg->segno);
2966 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2968 mutex_unlock(&curseg->curseg_mutex);
2971 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2973 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2975 if (sbi->am.atgc_enabled)
2976 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2979 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2980 int alloc_mode, unsigned long long age)
2982 struct curseg_info *curseg = CURSEG_I(sbi, type);
2983 unsigned segno = NULL_SEGNO;
2984 unsigned short seg_type = curseg->seg_type;
2986 bool reversed = false;
2988 sanity_check_seg_type(sbi, seg_type);
2990 /* f2fs_need_SSR() already forces to do this */
2991 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2992 curseg->next_segno = segno;
2996 /* For node segments, let's do SSR more intensively */
2997 if (IS_NODESEG(seg_type)) {
2998 if (seg_type >= CURSEG_WARM_NODE) {
3000 i = CURSEG_COLD_NODE;
3002 i = CURSEG_HOT_NODE;
3004 cnt = NR_CURSEG_NODE_TYPE;
3006 if (seg_type >= CURSEG_WARM_DATA) {
3008 i = CURSEG_COLD_DATA;
3010 i = CURSEG_HOT_DATA;
3012 cnt = NR_CURSEG_DATA_TYPE;
3015 for (; cnt-- > 0; reversed ? i-- : i++) {
3018 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
3019 curseg->next_segno = segno;
3024 /* find valid_blocks=0 in dirty list */
3025 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3026 segno = get_free_segment(sbi);
3027 if (segno != NULL_SEGNO) {
3028 curseg->next_segno = segno;
3035 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3037 struct curseg_info *curseg = CURSEG_I(sbi, type);
3039 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3040 curseg->seg_type == CURSEG_WARM_NODE)
3042 if (curseg->alloc_type == LFS &&
3043 is_next_segment_free(sbi, curseg, type) &&
3044 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3046 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3051 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3052 unsigned int start, unsigned int end)
3054 struct curseg_info *curseg = CURSEG_I(sbi, type);
3057 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3058 mutex_lock(&curseg->curseg_mutex);
3059 down_write(&SIT_I(sbi)->sentry_lock);
3061 segno = CURSEG_I(sbi, type)->segno;
3062 if (segno < start || segno > end)
3065 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3066 change_curseg(sbi, type);
3068 new_curseg(sbi, type, true);
3070 stat_inc_seg_type(sbi, curseg);
3072 locate_dirty_segment(sbi, segno);
3074 up_write(&SIT_I(sbi)->sentry_lock);
3076 if (segno != curseg->segno)
3077 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3078 type, segno, curseg->segno);
3080 mutex_unlock(&curseg->curseg_mutex);
3081 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3084 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3085 bool new_sec, bool force)
3087 struct curseg_info *curseg = CURSEG_I(sbi, type);
3088 unsigned int old_segno;
3090 if (!force && curseg->inited &&
3091 !curseg->next_blkoff &&
3092 !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3093 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3096 old_segno = curseg->segno;
3097 new_curseg(sbi, type, true);
3098 stat_inc_seg_type(sbi, curseg);
3099 locate_dirty_segment(sbi, old_segno);
3102 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3104 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3105 down_write(&SIT_I(sbi)->sentry_lock);
3106 __allocate_new_segment(sbi, type, true, force);
3107 up_write(&SIT_I(sbi)->sentry_lock);
3108 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3111 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3115 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3116 down_write(&SIT_I(sbi)->sentry_lock);
3117 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3118 __allocate_new_segment(sbi, i, false, false);
3119 up_write(&SIT_I(sbi)->sentry_lock);
3120 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3123 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3124 struct cp_control *cpc)
3126 __u64 trim_start = cpc->trim_start;
3127 bool has_candidate = false;
3129 down_write(&SIT_I(sbi)->sentry_lock);
3130 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3131 if (add_discard_addrs(sbi, cpc, true)) {
3132 has_candidate = true;
3136 up_write(&SIT_I(sbi)->sentry_lock);
3138 cpc->trim_start = trim_start;
3139 return has_candidate;
3142 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3143 struct discard_policy *dpolicy,
3144 unsigned int start, unsigned int end)
3146 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3147 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3148 struct rb_node **insert_p = NULL, *insert_parent = NULL;
3149 struct discard_cmd *dc;
3150 struct blk_plug plug;
3152 unsigned int trimmed = 0;
3157 mutex_lock(&dcc->cmd_lock);
3158 if (unlikely(dcc->rbtree_check))
3159 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3161 dc = __lookup_discard_cmd_ret(&dcc->root, start,
3162 &prev_dc, &next_dc, &insert_p, &insert_parent);
3166 blk_start_plug(&plug);
3168 while (dc && dc->di.lstart <= end) {
3169 struct rb_node *node;
3172 if (dc->di.len < dpolicy->granularity)
3175 if (dc->state != D_PREP) {
3176 list_move_tail(&dc->list, &dcc->fstrim_list);
3180 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3182 if (issued >= dpolicy->max_requests) {
3183 start = dc->di.lstart + dc->di.len;
3186 __remove_discard_cmd(sbi, dc);
3188 blk_finish_plug(&plug);
3189 mutex_unlock(&dcc->cmd_lock);
3190 trimmed += __wait_all_discard_cmd(sbi, NULL);
3191 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3195 node = rb_next(&dc->rb_node);
3197 __remove_discard_cmd(sbi, dc);
3198 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3200 if (fatal_signal_pending(current))
3204 blk_finish_plug(&plug);
3205 mutex_unlock(&dcc->cmd_lock);
3210 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3212 __u64 start = F2FS_BYTES_TO_BLK(range->start);
3213 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3214 unsigned int start_segno, end_segno;
3215 block_t start_block, end_block;
3216 struct cp_control cpc;
3217 struct discard_policy dpolicy;
3218 unsigned long long trimmed = 0;
3220 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3222 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3225 if (end < MAIN_BLKADDR(sbi))
3228 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3229 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3230 return -EFSCORRUPTED;
3233 /* start/end segment number in main_area */
3234 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3235 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3236 GET_SEGNO(sbi, end);
3238 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3239 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3242 cpc.reason = CP_DISCARD;
3243 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3244 cpc.trim_start = start_segno;
3245 cpc.trim_end = end_segno;
3247 if (sbi->discard_blks == 0)
3250 f2fs_down_write(&sbi->gc_lock);
3251 stat_inc_cp_call_count(sbi, TOTAL_CALL);
3252 err = f2fs_write_checkpoint(sbi, &cpc);
3253 f2fs_up_write(&sbi->gc_lock);
3258 * We filed discard candidates, but actually we don't need to wait for
3259 * all of them, since they'll be issued in idle time along with runtime
3260 * discard option. User configuration looks like using runtime discard
3261 * or periodic fstrim instead of it.
3263 if (f2fs_realtime_discard_enable(sbi))
3266 start_block = START_BLOCK(sbi, start_segno);
3267 end_block = START_BLOCK(sbi, end_segno + 1);
3269 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3270 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3271 start_block, end_block);
3273 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3274 start_block, end_block);
3277 range->len = F2FS_BLK_TO_BYTES(trimmed);
3281 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3284 case WRITE_LIFE_SHORT:
3285 return CURSEG_HOT_DATA;
3286 case WRITE_LIFE_EXTREME:
3287 return CURSEG_COLD_DATA;
3289 return CURSEG_WARM_DATA;
3293 static int __get_segment_type_2(struct f2fs_io_info *fio)
3295 if (fio->type == DATA)
3296 return CURSEG_HOT_DATA;
3298 return CURSEG_HOT_NODE;
3301 static int __get_segment_type_4(struct f2fs_io_info *fio)
3303 if (fio->type == DATA) {
3304 struct inode *inode = fio->page->mapping->host;
3306 if (S_ISDIR(inode->i_mode))
3307 return CURSEG_HOT_DATA;
3309 return CURSEG_COLD_DATA;
3311 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3312 return CURSEG_WARM_NODE;
3314 return CURSEG_COLD_NODE;
3318 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3320 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3321 struct extent_info ei = {};
3323 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3325 return NO_CHECK_TYPE;
3326 if (ei.age <= sbi->hot_data_age_threshold)
3327 return CURSEG_HOT_DATA;
3328 if (ei.age <= sbi->warm_data_age_threshold)
3329 return CURSEG_WARM_DATA;
3330 return CURSEG_COLD_DATA;
3332 return NO_CHECK_TYPE;
3335 static int __get_segment_type_6(struct f2fs_io_info *fio)
3337 if (fio->type == DATA) {
3338 struct inode *inode = fio->page->mapping->host;
3341 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3342 return CURSEG_COLD_DATA_PINNED;
3344 if (page_private_gcing(fio->page)) {
3345 if (fio->sbi->am.atgc_enabled &&
3346 (fio->io_type == FS_DATA_IO) &&
3347 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3348 return CURSEG_ALL_DATA_ATGC;
3350 return CURSEG_COLD_DATA;
3352 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3353 return CURSEG_COLD_DATA;
3355 type = __get_age_segment_type(inode, fio->page->index);
3356 if (type != NO_CHECK_TYPE)
3359 if (file_is_hot(inode) ||
3360 is_inode_flag_set(inode, FI_HOT_DATA) ||
3361 f2fs_is_cow_file(inode))
3362 return CURSEG_HOT_DATA;
3363 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3365 if (IS_DNODE(fio->page))
3366 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3368 return CURSEG_COLD_NODE;
3372 static int __get_segment_type(struct f2fs_io_info *fio)
3376 switch (F2FS_OPTION(fio->sbi).active_logs) {
3378 type = __get_segment_type_2(fio);
3381 type = __get_segment_type_4(fio);
3384 type = __get_segment_type_6(fio);
3387 f2fs_bug_on(fio->sbi, true);
3392 else if (IS_WARM(type))
3399 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3400 struct curseg_info *seg)
3402 /* To allocate block chunks in different sizes, use random number */
3403 if (--seg->fragment_remained_chunk > 0)
3406 seg->fragment_remained_chunk =
3407 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3409 get_random_u32_inclusive(1, sbi->max_fragment_hole);
3412 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3413 block_t old_blkaddr, block_t *new_blkaddr,
3414 struct f2fs_summary *sum, int type,
3415 struct f2fs_io_info *fio)
3417 struct sit_info *sit_i = SIT_I(sbi);
3418 struct curseg_info *curseg = CURSEG_I(sbi, type);
3419 unsigned long long old_mtime;
3420 bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3421 struct seg_entry *se = NULL;
3422 bool segment_full = false;
3424 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3426 mutex_lock(&curseg->curseg_mutex);
3427 down_write(&sit_i->sentry_lock);
3430 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3431 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3432 sanity_check_seg_type(sbi, se->type);
3433 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3435 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3437 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3439 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3441 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3442 if (curseg->alloc_type == SSR) {
3443 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3445 curseg->next_blkoff++;
3446 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3447 f2fs_randomize_chunk(sbi, curseg);
3449 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3450 segment_full = true;
3451 stat_inc_block_count(sbi, curseg);
3454 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3456 update_segment_mtime(sbi, old_blkaddr, 0);
3459 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3462 * SIT information should be updated before segment allocation,
3463 * since SSR needs latest valid block information.
3465 update_sit_entry(sbi, *new_blkaddr, 1);
3466 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3467 update_sit_entry(sbi, old_blkaddr, -1);
3470 * If the current segment is full, flush it out and replace it with a
3475 get_atssr_segment(sbi, type, se->type,
3478 if (need_new_seg(sbi, type))
3479 new_curseg(sbi, type, false);
3481 change_curseg(sbi, type);
3482 stat_inc_seg_type(sbi, curseg);
3486 * segment dirty status should be updated after segment allocation,
3487 * so we just need to update status only one time after previous
3488 * segment being closed.
3490 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3491 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3493 if (IS_DATASEG(type))
3494 atomic64_inc(&sbi->allocated_data_blocks);
3496 up_write(&sit_i->sentry_lock);
3498 if (page && IS_NODESEG(type)) {
3499 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3501 f2fs_inode_chksum_set(sbi, page);
3505 struct f2fs_bio_info *io;
3507 if (F2FS_IO_ALIGNED(sbi))
3510 INIT_LIST_HEAD(&fio->list);
3512 io = sbi->write_io[fio->type] + fio->temp;
3513 spin_lock(&io->io_lock);
3514 list_add_tail(&fio->list, &io->io_list);
3515 spin_unlock(&io->io_lock);
3518 mutex_unlock(&curseg->curseg_mutex);
3520 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3523 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3524 block_t blkaddr, unsigned int blkcnt)
3526 if (!f2fs_is_multi_device(sbi))
3530 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3531 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3533 /* update device state for fsync */
3534 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3536 /* update device state for checkpoint */
3537 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3538 spin_lock(&sbi->dev_lock);
3539 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3540 spin_unlock(&sbi->dev_lock);
3550 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3552 int type = __get_segment_type(fio);
3553 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3556 f2fs_down_read(&fio->sbi->io_order_lock);
3558 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3559 &fio->new_blkaddr, sum, type, fio);
3560 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3561 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3562 fio->old_blkaddr, fio->old_blkaddr);
3563 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3566 /* writeout dirty page into bdev */
3567 f2fs_submit_page_write(fio);
3569 fio->old_blkaddr = fio->new_blkaddr;
3573 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3576 f2fs_up_read(&fio->sbi->io_order_lock);
3579 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3580 enum iostat_type io_type)
3582 struct f2fs_io_info fio = {
3587 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3588 .old_blkaddr = page->index,
3589 .new_blkaddr = page->index,
3591 .encrypted_page = NULL,
3595 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3596 fio.op_flags &= ~REQ_META;
3598 set_page_writeback(page);
3599 f2fs_submit_page_write(&fio);
3601 stat_inc_meta_count(sbi, page->index);
3602 f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3605 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3607 struct f2fs_summary sum;
3609 set_summary(&sum, nid, 0, 0);
3610 do_write_page(&sum, fio);
3612 f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3615 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3616 struct f2fs_io_info *fio)
3618 struct f2fs_sb_info *sbi = fio->sbi;
3619 struct f2fs_summary sum;
3621 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3622 if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3623 f2fs_update_age_extent_cache(dn);
3624 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3625 do_write_page(&sum, fio);
3626 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3628 f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3631 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3634 struct f2fs_sb_info *sbi = fio->sbi;
3637 fio->new_blkaddr = fio->old_blkaddr;
3638 /* i/o temperature is needed for passing down write hints */
3639 __get_segment_type(fio);
3641 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3643 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3644 set_sbi_flag(sbi, SBI_NEED_FSCK);
3645 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3647 err = -EFSCORRUPTED;
3648 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3652 if (f2fs_cp_error(sbi)) {
3658 invalidate_mapping_pages(META_MAPPING(sbi),
3659 fio->new_blkaddr, fio->new_blkaddr);
3661 stat_inc_inplace_blocks(fio->sbi);
3663 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3664 err = f2fs_merge_page_bio(fio);
3666 err = f2fs_submit_page_bio(fio);
3668 f2fs_update_device_state(fio->sbi, fio->ino,
3669 fio->new_blkaddr, 1);
3670 f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3671 fio->io_type, F2FS_BLKSIZE);
3676 if (fio->bio && *(fio->bio)) {
3677 struct bio *bio = *(fio->bio);
3679 bio->bi_status = BLK_STS_IOERR;
3686 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3691 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3692 if (CURSEG_I(sbi, i)->segno == segno)
3698 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3699 block_t old_blkaddr, block_t new_blkaddr,
3700 bool recover_curseg, bool recover_newaddr,
3703 struct sit_info *sit_i = SIT_I(sbi);
3704 struct curseg_info *curseg;
3705 unsigned int segno, old_cursegno;
3706 struct seg_entry *se;
3708 unsigned short old_blkoff;
3709 unsigned char old_alloc_type;
3711 segno = GET_SEGNO(sbi, new_blkaddr);
3712 se = get_seg_entry(sbi, segno);
3715 f2fs_down_write(&SM_I(sbi)->curseg_lock);
3717 if (!recover_curseg) {
3718 /* for recovery flow */
3719 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3720 if (old_blkaddr == NULL_ADDR)
3721 type = CURSEG_COLD_DATA;
3723 type = CURSEG_WARM_DATA;
3726 if (IS_CURSEG(sbi, segno)) {
3727 /* se->type is volatile as SSR allocation */
3728 type = __f2fs_get_curseg(sbi, segno);
3729 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3731 type = CURSEG_WARM_DATA;
3735 f2fs_bug_on(sbi, !IS_DATASEG(type));
3736 curseg = CURSEG_I(sbi, type);
3738 mutex_lock(&curseg->curseg_mutex);
3739 down_write(&sit_i->sentry_lock);
3741 old_cursegno = curseg->segno;
3742 old_blkoff = curseg->next_blkoff;
3743 old_alloc_type = curseg->alloc_type;
3745 /* change the current segment */
3746 if (segno != curseg->segno) {
3747 curseg->next_segno = segno;
3748 change_curseg(sbi, type);
3751 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3752 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3754 if (!recover_curseg || recover_newaddr) {
3756 update_segment_mtime(sbi, new_blkaddr, 0);
3757 update_sit_entry(sbi, new_blkaddr, 1);
3759 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3760 invalidate_mapping_pages(META_MAPPING(sbi),
3761 old_blkaddr, old_blkaddr);
3762 f2fs_invalidate_compress_page(sbi, old_blkaddr);
3764 update_segment_mtime(sbi, old_blkaddr, 0);
3765 update_sit_entry(sbi, old_blkaddr, -1);
3768 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3769 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3771 locate_dirty_segment(sbi, old_cursegno);
3773 if (recover_curseg) {
3774 if (old_cursegno != curseg->segno) {
3775 curseg->next_segno = old_cursegno;
3776 change_curseg(sbi, type);
3778 curseg->next_blkoff = old_blkoff;
3779 curseg->alloc_type = old_alloc_type;
3782 up_write(&sit_i->sentry_lock);
3783 mutex_unlock(&curseg->curseg_mutex);
3784 f2fs_up_write(&SM_I(sbi)->curseg_lock);
3787 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3788 block_t old_addr, block_t new_addr,
3789 unsigned char version, bool recover_curseg,
3790 bool recover_newaddr)
3792 struct f2fs_summary sum;
3794 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3796 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3797 recover_curseg, recover_newaddr, false);
3799 f2fs_update_data_blkaddr(dn, new_addr);
3802 void f2fs_wait_on_page_writeback(struct page *page,
3803 enum page_type type, bool ordered, bool locked)
3805 if (PageWriteback(page)) {
3806 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3808 /* submit cached LFS IO */
3809 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3810 /* submit cached IPU IO */
3811 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3813 wait_on_page_writeback(page);
3814 f2fs_bug_on(sbi, locked && PageWriteback(page));
3816 wait_for_stable_page(page);
3821 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3823 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3826 if (!f2fs_post_read_required(inode))
3829 if (!__is_valid_data_blkaddr(blkaddr))
3832 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3834 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3835 f2fs_put_page(cpage, 1);
3839 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3842 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3845 if (!f2fs_post_read_required(inode))
3848 for (i = 0; i < len; i++)
3849 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3851 invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1);
3854 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3856 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3857 struct curseg_info *seg_i;
3858 unsigned char *kaddr;
3863 start = start_sum_block(sbi);
3865 page = f2fs_get_meta_page(sbi, start++);
3867 return PTR_ERR(page);
3868 kaddr = (unsigned char *)page_address(page);
3870 /* Step 1: restore nat cache */
3871 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3872 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3874 /* Step 2: restore sit cache */
3875 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3876 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3877 offset = 2 * SUM_JOURNAL_SIZE;
3879 /* Step 3: restore summary entries */
3880 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3881 unsigned short blk_off;
3884 seg_i = CURSEG_I(sbi, i);
3885 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3886 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3887 seg_i->next_segno = segno;
3888 reset_curseg(sbi, i, 0);
3889 seg_i->alloc_type = ckpt->alloc_type[i];
3890 seg_i->next_blkoff = blk_off;
3892 if (seg_i->alloc_type == SSR)
3893 blk_off = sbi->blocks_per_seg;
3895 for (j = 0; j < blk_off; j++) {
3896 struct f2fs_summary *s;
3898 s = (struct f2fs_summary *)(kaddr + offset);
3899 seg_i->sum_blk->entries[j] = *s;
3900 offset += SUMMARY_SIZE;
3901 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3905 f2fs_put_page(page, 1);
3908 page = f2fs_get_meta_page(sbi, start++);
3910 return PTR_ERR(page);
3911 kaddr = (unsigned char *)page_address(page);
3915 f2fs_put_page(page, 1);
3919 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3921 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3922 struct f2fs_summary_block *sum;
3923 struct curseg_info *curseg;
3925 unsigned short blk_off;
3926 unsigned int segno = 0;
3927 block_t blk_addr = 0;
3930 /* get segment number and block addr */
3931 if (IS_DATASEG(type)) {
3932 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3933 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3935 if (__exist_node_summaries(sbi))
3936 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3938 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3940 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3942 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3944 if (__exist_node_summaries(sbi))
3945 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3946 type - CURSEG_HOT_NODE);
3948 blk_addr = GET_SUM_BLOCK(sbi, segno);
3951 new = f2fs_get_meta_page(sbi, blk_addr);
3953 return PTR_ERR(new);
3954 sum = (struct f2fs_summary_block *)page_address(new);
3956 if (IS_NODESEG(type)) {
3957 if (__exist_node_summaries(sbi)) {
3958 struct f2fs_summary *ns = &sum->entries[0];
3961 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3963 ns->ofs_in_node = 0;
3966 err = f2fs_restore_node_summary(sbi, segno, sum);
3972 /* set uncompleted segment to curseg */
3973 curseg = CURSEG_I(sbi, type);
3974 mutex_lock(&curseg->curseg_mutex);
3976 /* update journal info */
3977 down_write(&curseg->journal_rwsem);
3978 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3979 up_write(&curseg->journal_rwsem);
3981 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3982 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3983 curseg->next_segno = segno;
3984 reset_curseg(sbi, type, 0);
3985 curseg->alloc_type = ckpt->alloc_type[type];
3986 curseg->next_blkoff = blk_off;
3987 mutex_unlock(&curseg->curseg_mutex);
3989 f2fs_put_page(new, 1);
3993 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3995 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3996 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3997 int type = CURSEG_HOT_DATA;
4000 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4001 int npages = f2fs_npages_for_summary_flush(sbi, true);
4004 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4007 /* restore for compacted data summary */
4008 err = read_compacted_summaries(sbi);
4011 type = CURSEG_HOT_NODE;
4014 if (__exist_node_summaries(sbi))
4015 f2fs_ra_meta_pages(sbi,
4016 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4017 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4019 for (; type <= CURSEG_COLD_NODE; type++) {
4020 err = read_normal_summaries(sbi, type);
4025 /* sanity check for summary blocks */
4026 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4027 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4028 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4029 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4036 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4039 unsigned char *kaddr;
4040 struct f2fs_summary *summary;
4041 struct curseg_info *seg_i;
4042 int written_size = 0;
4045 page = f2fs_grab_meta_page(sbi, blkaddr++);
4046 kaddr = (unsigned char *)page_address(page);
4047 memset(kaddr, 0, PAGE_SIZE);
4049 /* Step 1: write nat cache */
4050 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4051 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4052 written_size += SUM_JOURNAL_SIZE;
4054 /* Step 2: write sit cache */
4055 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4056 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4057 written_size += SUM_JOURNAL_SIZE;
4059 /* Step 3: write summary entries */
4060 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4061 seg_i = CURSEG_I(sbi, i);
4062 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4064 page = f2fs_grab_meta_page(sbi, blkaddr++);
4065 kaddr = (unsigned char *)page_address(page);
4066 memset(kaddr, 0, PAGE_SIZE);
4069 summary = (struct f2fs_summary *)(kaddr + written_size);
4070 *summary = seg_i->sum_blk->entries[j];
4071 written_size += SUMMARY_SIZE;
4073 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4077 set_page_dirty(page);
4078 f2fs_put_page(page, 1);
4083 set_page_dirty(page);
4084 f2fs_put_page(page, 1);
4088 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4089 block_t blkaddr, int type)
4093 if (IS_DATASEG(type))
4094 end = type + NR_CURSEG_DATA_TYPE;
4096 end = type + NR_CURSEG_NODE_TYPE;
4098 for (i = type; i < end; i++)
4099 write_current_sum_page(sbi, i, blkaddr + (i - type));
4102 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4104 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4105 write_compacted_summaries(sbi, start_blk);
4107 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4110 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4112 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4115 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4116 unsigned int val, int alloc)
4120 if (type == NAT_JOURNAL) {
4121 for (i = 0; i < nats_in_cursum(journal); i++) {
4122 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4125 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4126 return update_nats_in_cursum(journal, 1);
4127 } else if (type == SIT_JOURNAL) {
4128 for (i = 0; i < sits_in_cursum(journal); i++)
4129 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4131 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4132 return update_sits_in_cursum(journal, 1);
4137 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4140 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4143 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4146 struct sit_info *sit_i = SIT_I(sbi);
4148 pgoff_t src_off, dst_off;
4150 src_off = current_sit_addr(sbi, start);
4151 dst_off = next_sit_addr(sbi, src_off);
4153 page = f2fs_grab_meta_page(sbi, dst_off);
4154 seg_info_to_sit_page(sbi, page, start);
4156 set_page_dirty(page);
4157 set_to_next_sit(sit_i, start);
4162 static struct sit_entry_set *grab_sit_entry_set(void)
4164 struct sit_entry_set *ses =
4165 f2fs_kmem_cache_alloc(sit_entry_set_slab,
4166 GFP_NOFS, true, NULL);
4169 INIT_LIST_HEAD(&ses->set_list);
4173 static void release_sit_entry_set(struct sit_entry_set *ses)
4175 list_del(&ses->set_list);
4176 kmem_cache_free(sit_entry_set_slab, ses);
4179 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4180 struct list_head *head)
4182 struct sit_entry_set *next = ses;
4184 if (list_is_last(&ses->set_list, head))
4187 list_for_each_entry_continue(next, head, set_list)
4188 if (ses->entry_cnt <= next->entry_cnt) {
4189 list_move_tail(&ses->set_list, &next->set_list);
4193 list_move_tail(&ses->set_list, head);
4196 static void add_sit_entry(unsigned int segno, struct list_head *head)
4198 struct sit_entry_set *ses;
4199 unsigned int start_segno = START_SEGNO(segno);
4201 list_for_each_entry(ses, head, set_list) {
4202 if (ses->start_segno == start_segno) {
4204 adjust_sit_entry_set(ses, head);
4209 ses = grab_sit_entry_set();
4211 ses->start_segno = start_segno;
4213 list_add(&ses->set_list, head);
4216 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4218 struct f2fs_sm_info *sm_info = SM_I(sbi);
4219 struct list_head *set_list = &sm_info->sit_entry_set;
4220 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4223 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4224 add_sit_entry(segno, set_list);
4227 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4229 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4230 struct f2fs_journal *journal = curseg->journal;
4233 down_write(&curseg->journal_rwsem);
4234 for (i = 0; i < sits_in_cursum(journal); i++) {
4238 segno = le32_to_cpu(segno_in_journal(journal, i));
4239 dirtied = __mark_sit_entry_dirty(sbi, segno);
4242 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4244 update_sits_in_cursum(journal, -i);
4245 up_write(&curseg->journal_rwsem);
4249 * CP calls this function, which flushes SIT entries including sit_journal,
4250 * and moves prefree segs to free segs.
4252 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4254 struct sit_info *sit_i = SIT_I(sbi);
4255 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4256 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4257 struct f2fs_journal *journal = curseg->journal;
4258 struct sit_entry_set *ses, *tmp;
4259 struct list_head *head = &SM_I(sbi)->sit_entry_set;
4260 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4261 struct seg_entry *se;
4263 down_write(&sit_i->sentry_lock);
4265 if (!sit_i->dirty_sentries)
4269 * add and account sit entries of dirty bitmap in sit entry
4272 add_sits_in_set(sbi);
4275 * if there are no enough space in journal to store dirty sit
4276 * entries, remove all entries from journal and add and account
4277 * them in sit entry set.
4279 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4281 remove_sits_in_journal(sbi);
4284 * there are two steps to flush sit entries:
4285 * #1, flush sit entries to journal in current cold data summary block.
4286 * #2, flush sit entries to sit page.
4288 list_for_each_entry_safe(ses, tmp, head, set_list) {
4289 struct page *page = NULL;
4290 struct f2fs_sit_block *raw_sit = NULL;
4291 unsigned int start_segno = ses->start_segno;
4292 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4293 (unsigned long)MAIN_SEGS(sbi));
4294 unsigned int segno = start_segno;
4297 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4301 down_write(&curseg->journal_rwsem);
4303 page = get_next_sit_page(sbi, start_segno);
4304 raw_sit = page_address(page);
4307 /* flush dirty sit entries in region of current sit set */
4308 for_each_set_bit_from(segno, bitmap, end) {
4309 int offset, sit_offset;
4311 se = get_seg_entry(sbi, segno);
4312 #ifdef CONFIG_F2FS_CHECK_FS
4313 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4314 SIT_VBLOCK_MAP_SIZE))
4315 f2fs_bug_on(sbi, 1);
4318 /* add discard candidates */
4319 if (!(cpc->reason & CP_DISCARD)) {
4320 cpc->trim_start = segno;
4321 add_discard_addrs(sbi, cpc, false);
4325 offset = f2fs_lookup_journal_in_cursum(journal,
4326 SIT_JOURNAL, segno, 1);
4327 f2fs_bug_on(sbi, offset < 0);
4328 segno_in_journal(journal, offset) =
4330 seg_info_to_raw_sit(se,
4331 &sit_in_journal(journal, offset));
4332 check_block_count(sbi, segno,
4333 &sit_in_journal(journal, offset));
4335 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4336 seg_info_to_raw_sit(se,
4337 &raw_sit->entries[sit_offset]);
4338 check_block_count(sbi, segno,
4339 &raw_sit->entries[sit_offset]);
4342 __clear_bit(segno, bitmap);
4343 sit_i->dirty_sentries--;
4348 up_write(&curseg->journal_rwsem);
4350 f2fs_put_page(page, 1);
4352 f2fs_bug_on(sbi, ses->entry_cnt);
4353 release_sit_entry_set(ses);
4356 f2fs_bug_on(sbi, !list_empty(head));
4357 f2fs_bug_on(sbi, sit_i->dirty_sentries);
4359 if (cpc->reason & CP_DISCARD) {
4360 __u64 trim_start = cpc->trim_start;
4362 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4363 add_discard_addrs(sbi, cpc, false);
4365 cpc->trim_start = trim_start;
4367 up_write(&sit_i->sentry_lock);
4369 set_prefree_as_free_segments(sbi);
4372 static int build_sit_info(struct f2fs_sb_info *sbi)
4374 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4375 struct sit_info *sit_i;
4376 unsigned int sit_segs, start;
4377 char *src_bitmap, *bitmap;
4378 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4379 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4381 /* allocate memory for SIT information */
4382 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4386 SM_I(sbi)->sit_info = sit_i;
4389 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4392 if (!sit_i->sentries)
4395 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4396 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4398 if (!sit_i->dirty_sentries_bitmap)
4401 #ifdef CONFIG_F2FS_CHECK_FS
4402 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4404 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4406 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4410 bitmap = sit_i->bitmap;
4412 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4413 sit_i->sentries[start].cur_valid_map = bitmap;
4414 bitmap += SIT_VBLOCK_MAP_SIZE;
4416 sit_i->sentries[start].ckpt_valid_map = bitmap;
4417 bitmap += SIT_VBLOCK_MAP_SIZE;
4419 #ifdef CONFIG_F2FS_CHECK_FS
4420 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4421 bitmap += SIT_VBLOCK_MAP_SIZE;
4425 sit_i->sentries[start].discard_map = bitmap;
4426 bitmap += SIT_VBLOCK_MAP_SIZE;
4430 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4431 if (!sit_i->tmp_map)
4434 if (__is_large_section(sbi)) {
4435 sit_i->sec_entries =
4436 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4439 if (!sit_i->sec_entries)
4443 /* get information related with SIT */
4444 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4446 /* setup SIT bitmap from ckeckpoint pack */
4447 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4448 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4450 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4451 if (!sit_i->sit_bitmap)
4454 #ifdef CONFIG_F2FS_CHECK_FS
4455 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4456 sit_bitmap_size, GFP_KERNEL);
4457 if (!sit_i->sit_bitmap_mir)
4460 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4461 main_bitmap_size, GFP_KERNEL);
4462 if (!sit_i->invalid_segmap)
4466 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4467 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4468 sit_i->written_valid_blocks = 0;
4469 sit_i->bitmap_size = sit_bitmap_size;
4470 sit_i->dirty_sentries = 0;
4471 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4472 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4473 sit_i->mounted_time = ktime_get_boottime_seconds();
4474 init_rwsem(&sit_i->sentry_lock);
4478 static int build_free_segmap(struct f2fs_sb_info *sbi)
4480 struct free_segmap_info *free_i;
4481 unsigned int bitmap_size, sec_bitmap_size;
4483 /* allocate memory for free segmap information */
4484 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4488 SM_I(sbi)->free_info = free_i;
4490 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4491 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4492 if (!free_i->free_segmap)
4495 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4496 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4497 if (!free_i->free_secmap)
4500 /* set all segments as dirty temporarily */
4501 memset(free_i->free_segmap, 0xff, bitmap_size);
4502 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4504 /* init free segmap information */
4505 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4506 free_i->free_segments = 0;
4507 free_i->free_sections = 0;
4508 spin_lock_init(&free_i->segmap_lock);
4512 static int build_curseg(struct f2fs_sb_info *sbi)
4514 struct curseg_info *array;
4517 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4518 sizeof(*array)), GFP_KERNEL);
4522 SM_I(sbi)->curseg_array = array;
4524 for (i = 0; i < NO_CHECK_TYPE; i++) {
4525 mutex_init(&array[i].curseg_mutex);
4526 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4527 if (!array[i].sum_blk)
4529 init_rwsem(&array[i].journal_rwsem);
4530 array[i].journal = f2fs_kzalloc(sbi,
4531 sizeof(struct f2fs_journal), GFP_KERNEL);
4532 if (!array[i].journal)
4534 if (i < NR_PERSISTENT_LOG)
4535 array[i].seg_type = CURSEG_HOT_DATA + i;
4536 else if (i == CURSEG_COLD_DATA_PINNED)
4537 array[i].seg_type = CURSEG_COLD_DATA;
4538 else if (i == CURSEG_ALL_DATA_ATGC)
4539 array[i].seg_type = CURSEG_COLD_DATA;
4540 array[i].segno = NULL_SEGNO;
4541 array[i].next_blkoff = 0;
4542 array[i].inited = false;
4544 return restore_curseg_summaries(sbi);
4547 static int build_sit_entries(struct f2fs_sb_info *sbi)
4549 struct sit_info *sit_i = SIT_I(sbi);
4550 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4551 struct f2fs_journal *journal = curseg->journal;
4552 struct seg_entry *se;
4553 struct f2fs_sit_entry sit;
4554 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4555 unsigned int i, start, end;
4556 unsigned int readed, start_blk = 0;
4558 block_t sit_valid_blocks[2] = {0, 0};
4561 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4564 start = start_blk * sit_i->sents_per_block;
4565 end = (start_blk + readed) * sit_i->sents_per_block;
4567 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4568 struct f2fs_sit_block *sit_blk;
4571 se = &sit_i->sentries[start];
4572 page = get_current_sit_page(sbi, start);
4574 return PTR_ERR(page);
4575 sit_blk = (struct f2fs_sit_block *)page_address(page);
4576 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4577 f2fs_put_page(page, 1);
4579 err = check_block_count(sbi, start, &sit);
4582 seg_info_from_raw_sit(se, &sit);
4584 if (se->type >= NR_PERSISTENT_LOG) {
4585 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4587 f2fs_handle_error(sbi,
4588 ERROR_INCONSISTENT_SUM_TYPE);
4589 return -EFSCORRUPTED;
4592 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4594 if (f2fs_block_unit_discard(sbi)) {
4595 /* build discard map only one time */
4596 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4597 memset(se->discard_map, 0xff,
4598 SIT_VBLOCK_MAP_SIZE);
4600 memcpy(se->discard_map,
4602 SIT_VBLOCK_MAP_SIZE);
4603 sbi->discard_blks +=
4604 sbi->blocks_per_seg -
4609 if (__is_large_section(sbi))
4610 get_sec_entry(sbi, start)->valid_blocks +=
4613 start_blk += readed;
4614 } while (start_blk < sit_blk_cnt);
4616 down_read(&curseg->journal_rwsem);
4617 for (i = 0; i < sits_in_cursum(journal); i++) {
4618 unsigned int old_valid_blocks;
4620 start = le32_to_cpu(segno_in_journal(journal, i));
4621 if (start >= MAIN_SEGS(sbi)) {
4622 f2fs_err(sbi, "Wrong journal entry on segno %u",
4624 err = -EFSCORRUPTED;
4625 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4629 se = &sit_i->sentries[start];
4630 sit = sit_in_journal(journal, i);
4632 old_valid_blocks = se->valid_blocks;
4634 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4636 err = check_block_count(sbi, start, &sit);
4639 seg_info_from_raw_sit(se, &sit);
4641 if (se->type >= NR_PERSISTENT_LOG) {
4642 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4644 err = -EFSCORRUPTED;
4645 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4649 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4651 if (f2fs_block_unit_discard(sbi)) {
4652 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4653 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4655 memcpy(se->discard_map, se->cur_valid_map,
4656 SIT_VBLOCK_MAP_SIZE);
4657 sbi->discard_blks += old_valid_blocks;
4658 sbi->discard_blks -= se->valid_blocks;
4662 if (__is_large_section(sbi)) {
4663 get_sec_entry(sbi, start)->valid_blocks +=
4665 get_sec_entry(sbi, start)->valid_blocks -=
4669 up_read(&curseg->journal_rwsem);
4674 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4675 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4676 sit_valid_blocks[NODE], valid_node_count(sbi));
4677 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4678 return -EFSCORRUPTED;
4681 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4682 valid_user_blocks(sbi)) {
4683 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4684 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4685 valid_user_blocks(sbi));
4686 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4687 return -EFSCORRUPTED;
4693 static void init_free_segmap(struct f2fs_sb_info *sbi)
4697 struct seg_entry *sentry;
4699 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4700 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4702 sentry = get_seg_entry(sbi, start);
4703 if (!sentry->valid_blocks)
4704 __set_free(sbi, start);
4706 SIT_I(sbi)->written_valid_blocks +=
4707 sentry->valid_blocks;
4710 /* set use the current segments */
4711 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4712 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4714 __set_test_and_inuse(sbi, curseg_t->segno);
4718 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4720 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4721 struct free_segmap_info *free_i = FREE_I(sbi);
4722 unsigned int segno = 0, offset = 0, secno;
4723 block_t valid_blocks, usable_blks_in_seg;
4726 /* find dirty segment based on free segmap */
4727 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4728 if (segno >= MAIN_SEGS(sbi))
4731 valid_blocks = get_valid_blocks(sbi, segno, false);
4732 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4733 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4735 if (valid_blocks > usable_blks_in_seg) {
4736 f2fs_bug_on(sbi, 1);
4739 mutex_lock(&dirty_i->seglist_lock);
4740 __locate_dirty_segment(sbi, segno, DIRTY);
4741 mutex_unlock(&dirty_i->seglist_lock);
4744 if (!__is_large_section(sbi))
4747 mutex_lock(&dirty_i->seglist_lock);
4748 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4749 valid_blocks = get_valid_blocks(sbi, segno, true);
4750 secno = GET_SEC_FROM_SEG(sbi, segno);
4752 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4754 if (IS_CURSEC(sbi, secno))
4756 set_bit(secno, dirty_i->dirty_secmap);
4758 mutex_unlock(&dirty_i->seglist_lock);
4761 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4763 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4764 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4766 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4767 if (!dirty_i->victim_secmap)
4770 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4771 if (!dirty_i->pinned_secmap)
4774 dirty_i->pinned_secmap_cnt = 0;
4775 dirty_i->enable_pin_section = true;
4779 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4781 struct dirty_seglist_info *dirty_i;
4782 unsigned int bitmap_size, i;
4784 /* allocate memory for dirty segments list information */
4785 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4790 SM_I(sbi)->dirty_info = dirty_i;
4791 mutex_init(&dirty_i->seglist_lock);
4793 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4795 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4796 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4798 if (!dirty_i->dirty_segmap[i])
4802 if (__is_large_section(sbi)) {
4803 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4804 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4805 bitmap_size, GFP_KERNEL);
4806 if (!dirty_i->dirty_secmap)
4810 init_dirty_segmap(sbi);
4811 return init_victim_secmap(sbi);
4814 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4819 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4820 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4822 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4823 struct curseg_info *curseg = CURSEG_I(sbi, i);
4824 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4825 unsigned int blkofs = curseg->next_blkoff;
4827 if (f2fs_sb_has_readonly(sbi) &&
4828 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4831 sanity_check_seg_type(sbi, curseg->seg_type);
4833 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4835 "Current segment has invalid alloc_type:%d",
4836 curseg->alloc_type);
4837 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4838 return -EFSCORRUPTED;
4841 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4844 if (curseg->alloc_type == SSR)
4847 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4848 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4852 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4853 i, curseg->segno, curseg->alloc_type,
4854 curseg->next_blkoff, blkofs);
4855 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4856 return -EFSCORRUPTED;
4862 #ifdef CONFIG_BLK_DEV_ZONED
4864 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4865 struct f2fs_dev_info *fdev,
4866 struct blk_zone *zone)
4868 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4869 block_t zone_block, wp_block, last_valid_block;
4870 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4872 struct seg_entry *se;
4874 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4877 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4878 wp_segno = GET_SEGNO(sbi, wp_block);
4879 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4880 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4881 zone_segno = GET_SEGNO(sbi, zone_block);
4882 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4884 if (zone_segno >= MAIN_SEGS(sbi))
4888 * Skip check of zones cursegs point to, since
4889 * fix_curseg_write_pointer() checks them.
4891 for (i = 0; i < NO_CHECK_TYPE; i++)
4892 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4893 CURSEG_I(sbi, i)->segno))
4897 * Get last valid block of the zone.
4899 last_valid_block = zone_block - 1;
4900 for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4901 segno = zone_segno + s;
4902 se = get_seg_entry(sbi, segno);
4903 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4904 if (f2fs_test_bit(b, se->cur_valid_map)) {
4905 last_valid_block = START_BLOCK(sbi, segno) + b;
4908 if (last_valid_block >= zone_block)
4913 * The write pointer matches with the valid blocks or
4914 * already points to the end of the zone.
4916 if ((last_valid_block + 1 == wp_block) ||
4917 (zone->wp == zone->start + zone->len))
4920 if (last_valid_block + 1 == zone_block) {
4922 * If there is no valid block in the zone and if write pointer
4923 * is not at zone start, reset the write pointer.
4926 "Zone without valid block has non-zero write "
4927 "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4928 wp_segno, wp_blkoff);
4929 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4930 zone->len >> log_sectors_per_block);
4932 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4939 * If there are valid blocks and the write pointer doesn't
4940 * match with them, we need to report the inconsistency and
4941 * fill the zone till the end to close the zone. This inconsistency
4942 * does not cause write error because the zone will not be selected
4943 * for write operation until it get discarded.
4945 f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: "
4946 "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4947 GET_SEGNO(sbi, last_valid_block),
4948 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4949 wp_segno, wp_blkoff);
4951 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
4952 zone->start, zone->len, GFP_NOFS);
4953 if (ret == -EOPNOTSUPP) {
4954 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
4955 zone->len - (zone->wp - zone->start),
4958 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
4961 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
4968 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4969 block_t zone_blkaddr)
4973 for (i = 0; i < sbi->s_ndevs; i++) {
4974 if (!bdev_is_zoned(FDEV(i).bdev))
4976 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4977 zone_blkaddr <= FDEV(i).end_blk))
4984 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4987 memcpy(data, zone, sizeof(struct blk_zone));
4991 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4993 struct curseg_info *cs = CURSEG_I(sbi, type);
4994 struct f2fs_dev_info *zbd;
4995 struct blk_zone zone;
4996 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4997 block_t cs_zone_block, wp_block;
4998 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4999 sector_t zone_sector;
5002 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5003 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5005 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5009 /* report zone for the sector the curseg points to */
5010 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5011 << log_sectors_per_block;
5012 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5013 report_one_zone_cb, &zone);
5015 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5020 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5023 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5024 wp_segno = GET_SEGNO(sbi, wp_block);
5025 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5026 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5028 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5032 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5033 "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
5034 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
5036 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5037 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
5039 f2fs_allocate_new_section(sbi, type, true);
5041 /* check consistency of the zone curseg pointed to */
5042 if (check_zone_write_pointer(sbi, zbd, &zone))
5045 /* check newly assigned zone */
5046 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5047 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5049 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5053 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5054 << log_sectors_per_block;
5055 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5056 report_one_zone_cb, &zone);
5058 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5063 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5066 if (zone.wp != zone.start) {
5068 "New zone for curseg[%d] is not yet discarded. "
5069 "Reset the zone: curseg[0x%x,0x%x]",
5070 type, cs->segno, cs->next_blkoff);
5071 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block,
5072 zone.len >> log_sectors_per_block);
5074 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5083 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5087 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5088 ret = fix_curseg_write_pointer(sbi, i);
5096 struct check_zone_write_pointer_args {
5097 struct f2fs_sb_info *sbi;
5098 struct f2fs_dev_info *fdev;
5101 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5104 struct check_zone_write_pointer_args *args;
5106 args = (struct check_zone_write_pointer_args *)data;
5108 return check_zone_write_pointer(args->sbi, args->fdev, zone);
5111 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5114 struct check_zone_write_pointer_args args;
5116 for (i = 0; i < sbi->s_ndevs; i++) {
5117 if (!bdev_is_zoned(FDEV(i).bdev))
5121 args.fdev = &FDEV(i);
5122 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5123 check_zone_write_pointer_cb, &args);
5132 * Return the number of usable blocks in a segment. The number of blocks
5133 * returned is always equal to the number of blocks in a segment for
5134 * segments fully contained within a sequential zone capacity or a
5135 * conventional zone. For segments partially contained in a sequential
5136 * zone capacity, the number of usable blocks up to the zone capacity
5137 * is returned. 0 is returned in all other cases.
5139 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5140 struct f2fs_sb_info *sbi, unsigned int segno)
5142 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5145 if (!sbi->unusable_blocks_per_sec)
5146 return sbi->blocks_per_seg;
5148 secno = GET_SEC_FROM_SEG(sbi, segno);
5149 seg_start = START_BLOCK(sbi, segno);
5150 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5151 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5154 * If segment starts before zone capacity and spans beyond
5155 * zone capacity, then usable blocks are from seg start to
5156 * zone capacity. If the segment starts after the zone capacity,
5157 * then there are no usable blocks.
5159 if (seg_start >= sec_cap_blkaddr)
5161 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5162 return sec_cap_blkaddr - seg_start;
5164 return sbi->blocks_per_seg;
5167 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5172 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5177 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5184 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5187 if (f2fs_sb_has_blkzoned(sbi))
5188 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5190 return sbi->blocks_per_seg;
5193 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5196 if (f2fs_sb_has_blkzoned(sbi))
5197 return CAP_SEGS_PER_SEC(sbi);
5199 return sbi->segs_per_sec;
5203 * Update min, max modified time for cost-benefit GC algorithm
5205 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5207 struct sit_info *sit_i = SIT_I(sbi);
5210 down_write(&sit_i->sentry_lock);
5212 sit_i->min_mtime = ULLONG_MAX;
5214 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5216 unsigned long long mtime = 0;
5218 for (i = 0; i < sbi->segs_per_sec; i++)
5219 mtime += get_seg_entry(sbi, segno + i)->mtime;
5221 mtime = div_u64(mtime, sbi->segs_per_sec);
5223 if (sit_i->min_mtime > mtime)
5224 sit_i->min_mtime = mtime;
5226 sit_i->max_mtime = get_mtime(sbi, false);
5227 sit_i->dirty_max_mtime = 0;
5228 up_write(&sit_i->sentry_lock);
5231 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5233 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5234 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5235 struct f2fs_sm_info *sm_info;
5238 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5243 sbi->sm_info = sm_info;
5244 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5245 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5246 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5247 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5248 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5249 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5250 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5251 sm_info->rec_prefree_segments = sm_info->main_segments *
5252 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5253 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5254 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5256 if (!f2fs_lfs_mode(sbi))
5257 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5258 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5259 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5260 sm_info->min_seq_blocks = sbi->blocks_per_seg;
5261 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5262 sm_info->min_ssr_sections = reserved_sections(sbi);
5264 INIT_LIST_HEAD(&sm_info->sit_entry_set);
5266 init_f2fs_rwsem(&sm_info->curseg_lock);
5268 err = f2fs_create_flush_cmd_control(sbi);
5272 err = create_discard_cmd_control(sbi);
5276 err = build_sit_info(sbi);
5279 err = build_free_segmap(sbi);
5282 err = build_curseg(sbi);
5286 /* reinit free segmap based on SIT */
5287 err = build_sit_entries(sbi);
5291 init_free_segmap(sbi);
5292 err = build_dirty_segmap(sbi);
5296 err = sanity_check_curseg(sbi);
5300 init_min_max_mtime(sbi);
5304 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5305 enum dirty_type dirty_type)
5307 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5309 mutex_lock(&dirty_i->seglist_lock);
5310 kvfree(dirty_i->dirty_segmap[dirty_type]);
5311 dirty_i->nr_dirty[dirty_type] = 0;
5312 mutex_unlock(&dirty_i->seglist_lock);
5315 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5319 kvfree(dirty_i->pinned_secmap);
5320 kvfree(dirty_i->victim_secmap);
5323 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5325 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5331 /* discard pre-free/dirty segments list */
5332 for (i = 0; i < NR_DIRTY_TYPE; i++)
5333 discard_dirty_segmap(sbi, i);
5335 if (__is_large_section(sbi)) {
5336 mutex_lock(&dirty_i->seglist_lock);
5337 kvfree(dirty_i->dirty_secmap);
5338 mutex_unlock(&dirty_i->seglist_lock);
5341 destroy_victim_secmap(sbi);
5342 SM_I(sbi)->dirty_info = NULL;
5346 static void destroy_curseg(struct f2fs_sb_info *sbi)
5348 struct curseg_info *array = SM_I(sbi)->curseg_array;
5353 SM_I(sbi)->curseg_array = NULL;
5354 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5355 kfree(array[i].sum_blk);
5356 kfree(array[i].journal);
5361 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5363 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5367 SM_I(sbi)->free_info = NULL;
5368 kvfree(free_i->free_segmap);
5369 kvfree(free_i->free_secmap);
5373 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5375 struct sit_info *sit_i = SIT_I(sbi);
5380 if (sit_i->sentries)
5381 kvfree(sit_i->bitmap);
5382 kfree(sit_i->tmp_map);
5384 kvfree(sit_i->sentries);
5385 kvfree(sit_i->sec_entries);
5386 kvfree(sit_i->dirty_sentries_bitmap);
5388 SM_I(sbi)->sit_info = NULL;
5389 kvfree(sit_i->sit_bitmap);
5390 #ifdef CONFIG_F2FS_CHECK_FS
5391 kvfree(sit_i->sit_bitmap_mir);
5392 kvfree(sit_i->invalid_segmap);
5397 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5399 struct f2fs_sm_info *sm_info = SM_I(sbi);
5403 f2fs_destroy_flush_cmd_control(sbi, true);
5404 destroy_discard_cmd_control(sbi);
5405 destroy_dirty_segmap(sbi);
5406 destroy_curseg(sbi);
5407 destroy_free_segmap(sbi);
5408 destroy_sit_info(sbi);
5409 sbi->sm_info = NULL;
5413 int __init f2fs_create_segment_manager_caches(void)
5415 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5416 sizeof(struct discard_entry));
5417 if (!discard_entry_slab)
5420 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5421 sizeof(struct discard_cmd));
5422 if (!discard_cmd_slab)
5423 goto destroy_discard_entry;
5425 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5426 sizeof(struct sit_entry_set));
5427 if (!sit_entry_set_slab)
5428 goto destroy_discard_cmd;
5430 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5431 sizeof(struct revoke_entry));
5432 if (!revoke_entry_slab)
5433 goto destroy_sit_entry_set;
5436 destroy_sit_entry_set:
5437 kmem_cache_destroy(sit_entry_set_slab);
5438 destroy_discard_cmd:
5439 kmem_cache_destroy(discard_cmd_slab);
5440 destroy_discard_entry:
5441 kmem_cache_destroy(discard_entry_slab);
5446 void f2fs_destroy_segment_manager_caches(void)
5448 kmem_cache_destroy(sit_entry_set_slab);
5449 kmem_cache_destroy(discard_cmd_slab);
5450 kmem_cache_destroy(discard_entry_slab);
5451 kmem_cache_destroy(revoke_entry_slab);