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))
196 truncate_inode_pages_final(inode->i_mapping);
198 release_atomic_write_cnt(inode);
199 clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
200 clear_inode_flag(inode, FI_ATOMIC_REPLACE);
201 clear_inode_flag(inode, FI_ATOMIC_FILE);
202 stat_dec_atomic_inode(inode);
204 F2FS_I(inode)->atomic_write_task = NULL;
207 f2fs_i_size_write(inode, fi->original_i_size);
208 fi->original_i_size = 0;
210 /* avoid stale dirty inode during eviction */
211 sync_inode_metadata(inode, 0);
214 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
215 block_t new_addr, block_t *old_addr, bool recover)
217 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
218 struct dnode_of_data dn;
223 set_new_dnode(&dn, inode, NULL, NULL, 0);
224 err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
226 if (err == -ENOMEM) {
227 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
233 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
240 /* dn.data_blkaddr is always valid */
241 if (!__is_valid_data_blkaddr(new_addr)) {
242 if (new_addr == NULL_ADDR)
243 dec_valid_block_count(sbi, inode, 1);
244 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
245 f2fs_update_data_blkaddr(&dn, new_addr);
247 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
248 new_addr, ni.version, true, true);
253 err = inc_valid_block_count(sbi, inode, &count, true);
259 *old_addr = dn.data_blkaddr;
260 f2fs_truncate_data_blocks_range(&dn, 1);
261 dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
263 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
264 ni.version, true, false);
269 trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
270 index, old_addr ? *old_addr : 0, new_addr, recover);
274 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
277 struct revoke_entry *cur, *tmp;
278 pgoff_t start_index = 0;
279 bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
281 list_for_each_entry_safe(cur, tmp, head, list) {
283 __replace_atomic_write_block(inode, cur->index,
284 cur->old_addr, NULL, true);
285 } else if (truncate) {
286 f2fs_truncate_hole(inode, start_index, cur->index);
287 start_index = cur->index + 1;
290 list_del(&cur->list);
291 kmem_cache_free(revoke_entry_slab, cur);
294 if (!revoke && truncate)
295 f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
298 static int __f2fs_commit_atomic_write(struct inode *inode)
300 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
301 struct f2fs_inode_info *fi = F2FS_I(inode);
302 struct inode *cow_inode = fi->cow_inode;
303 struct revoke_entry *new;
304 struct list_head revoke_list;
306 struct dnode_of_data dn;
307 pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
308 pgoff_t off = 0, blen, index;
311 INIT_LIST_HEAD(&revoke_list);
314 blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
316 set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
317 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
318 if (ret && ret != -ENOENT) {
320 } else if (ret == -ENOENT) {
322 if (dn.max_level == 0)
327 blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
330 for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
331 blkaddr = f2fs_data_blkaddr(&dn);
333 if (!__is_valid_data_blkaddr(blkaddr)) {
335 } else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
336 DATA_GENERIC_ENHANCE)) {
339 f2fs_handle_error(sbi,
340 ERROR_INVALID_BLKADDR);
344 new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
347 ret = __replace_atomic_write_block(inode, index, blkaddr,
348 &new->old_addr, false);
351 kmem_cache_free(revoke_entry_slab, new);
355 f2fs_update_data_blkaddr(&dn, NULL_ADDR);
357 list_add_tail(&new->list, &revoke_list);
367 sbi->revoked_atomic_block += fi->atomic_write_cnt;
369 sbi->committed_atomic_block += fi->atomic_write_cnt;
370 set_inode_flag(inode, FI_ATOMIC_COMMITTED);
373 __complete_revoke_list(inode, &revoke_list, ret ? true : false);
378 int f2fs_commit_atomic_write(struct inode *inode)
380 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
381 struct f2fs_inode_info *fi = F2FS_I(inode);
384 err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
388 f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
391 err = __f2fs_commit_atomic_write(inode);
394 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
400 * This function balances dirty node and dentry pages.
401 * In addition, it controls garbage collection.
403 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
405 if (time_to_inject(sbi, FAULT_CHECKPOINT))
406 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
408 /* balance_fs_bg is able to be pending */
409 if (need && excess_cached_nats(sbi))
410 f2fs_balance_fs_bg(sbi, false);
412 if (!f2fs_is_checkpoint_ready(sbi))
416 * We should do GC or end up with checkpoint, if there are so many dirty
417 * dir/node pages without enough free segments.
419 if (has_enough_free_secs(sbi, 0, 0))
422 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
423 sbi->gc_thread->f2fs_gc_task) {
426 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
427 TASK_UNINTERRUPTIBLE);
428 wake_up(&sbi->gc_thread->gc_wait_queue_head);
430 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
432 struct f2fs_gc_control gc_control = {
433 .victim_segno = NULL_SEGNO,
434 .init_gc_type = BG_GC,
436 .should_migrate_blocks = false,
437 .err_gc_skipped = false,
439 f2fs_down_write(&sbi->gc_lock);
440 stat_inc_gc_call_count(sbi, FOREGROUND);
441 f2fs_gc(sbi, &gc_control);
445 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
447 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
448 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
449 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
450 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
451 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
452 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
453 unsigned int threshold = sbi->blocks_per_seg * factor *
454 DEFAULT_DIRTY_THRESHOLD;
455 unsigned int global_threshold = threshold * 3 / 2;
457 if (dents >= threshold || qdata >= threshold ||
458 nodes >= threshold || meta >= threshold ||
461 return dents + qdata + nodes + meta + imeta > global_threshold;
464 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
466 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
469 /* try to shrink extent cache when there is no enough memory */
470 if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
471 f2fs_shrink_read_extent_tree(sbi,
472 READ_EXTENT_CACHE_SHRINK_NUMBER);
474 /* try to shrink age extent cache when there is no enough memory */
475 if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
476 f2fs_shrink_age_extent_tree(sbi,
477 AGE_EXTENT_CACHE_SHRINK_NUMBER);
479 /* check the # of cached NAT entries */
480 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
481 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
483 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
484 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
486 f2fs_build_free_nids(sbi, false, false);
488 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
489 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
492 /* there is background inflight IO or foreground operation recently */
493 if (is_inflight_io(sbi, REQ_TIME) ||
494 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
497 /* exceed periodical checkpoint timeout threshold */
498 if (f2fs_time_over(sbi, CP_TIME))
501 /* checkpoint is the only way to shrink partial cached entries */
502 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
503 f2fs_available_free_memory(sbi, INO_ENTRIES))
507 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
508 struct blk_plug plug;
510 mutex_lock(&sbi->flush_lock);
512 blk_start_plug(&plug);
513 f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
514 blk_finish_plug(&plug);
516 mutex_unlock(&sbi->flush_lock);
518 stat_inc_cp_call_count(sbi, BACKGROUND);
519 f2fs_sync_fs(sbi->sb, 1);
522 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
523 struct block_device *bdev)
525 int ret = blkdev_issue_flush(bdev);
527 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
528 test_opt(sbi, FLUSH_MERGE), ret);
530 f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
534 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
539 if (!f2fs_is_multi_device(sbi))
540 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
542 for (i = 0; i < sbi->s_ndevs; i++) {
543 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
545 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
552 static int issue_flush_thread(void *data)
554 struct f2fs_sb_info *sbi = data;
555 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
556 wait_queue_head_t *q = &fcc->flush_wait_queue;
558 if (kthread_should_stop())
561 if (!llist_empty(&fcc->issue_list)) {
562 struct flush_cmd *cmd, *next;
565 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
566 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
568 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
570 ret = submit_flush_wait(sbi, cmd->ino);
571 atomic_inc(&fcc->issued_flush);
573 llist_for_each_entry_safe(cmd, next,
574 fcc->dispatch_list, llnode) {
576 complete(&cmd->wait);
578 fcc->dispatch_list = NULL;
581 wait_event_interruptible(*q,
582 kthread_should_stop() || !llist_empty(&fcc->issue_list));
586 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
588 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
589 struct flush_cmd cmd;
592 if (test_opt(sbi, NOBARRIER))
595 if (!test_opt(sbi, FLUSH_MERGE)) {
596 atomic_inc(&fcc->queued_flush);
597 ret = submit_flush_wait(sbi, ino);
598 atomic_dec(&fcc->queued_flush);
599 atomic_inc(&fcc->issued_flush);
603 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
604 f2fs_is_multi_device(sbi)) {
605 ret = submit_flush_wait(sbi, ino);
606 atomic_dec(&fcc->queued_flush);
608 atomic_inc(&fcc->issued_flush);
613 init_completion(&cmd.wait);
615 llist_add(&cmd.llnode, &fcc->issue_list);
618 * update issue_list before we wake up issue_flush thread, this
619 * smp_mb() pairs with another barrier in ___wait_event(), see
620 * more details in comments of waitqueue_active().
624 if (waitqueue_active(&fcc->flush_wait_queue))
625 wake_up(&fcc->flush_wait_queue);
627 if (fcc->f2fs_issue_flush) {
628 wait_for_completion(&cmd.wait);
629 atomic_dec(&fcc->queued_flush);
631 struct llist_node *list;
633 list = llist_del_all(&fcc->issue_list);
635 wait_for_completion(&cmd.wait);
636 atomic_dec(&fcc->queued_flush);
638 struct flush_cmd *tmp, *next;
640 ret = submit_flush_wait(sbi, ino);
642 llist_for_each_entry_safe(tmp, next, list, llnode) {
645 atomic_dec(&fcc->queued_flush);
649 complete(&tmp->wait);
657 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
659 dev_t dev = sbi->sb->s_bdev->bd_dev;
660 struct flush_cmd_control *fcc;
662 if (SM_I(sbi)->fcc_info) {
663 fcc = SM_I(sbi)->fcc_info;
664 if (fcc->f2fs_issue_flush)
669 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
672 atomic_set(&fcc->issued_flush, 0);
673 atomic_set(&fcc->queued_flush, 0);
674 init_waitqueue_head(&fcc->flush_wait_queue);
675 init_llist_head(&fcc->issue_list);
676 SM_I(sbi)->fcc_info = fcc;
677 if (!test_opt(sbi, FLUSH_MERGE))
681 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
682 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
683 if (IS_ERR(fcc->f2fs_issue_flush)) {
684 int err = PTR_ERR(fcc->f2fs_issue_flush);
686 fcc->f2fs_issue_flush = NULL;
693 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
695 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
697 if (fcc && fcc->f2fs_issue_flush) {
698 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
700 fcc->f2fs_issue_flush = NULL;
701 kthread_stop(flush_thread);
705 SM_I(sbi)->fcc_info = NULL;
709 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
713 if (!f2fs_is_multi_device(sbi))
716 if (test_opt(sbi, NOBARRIER))
719 for (i = 1; i < sbi->s_ndevs; i++) {
720 int count = DEFAULT_RETRY_IO_COUNT;
722 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
726 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
728 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
729 } while (ret && --count);
732 f2fs_stop_checkpoint(sbi, false,
733 STOP_CP_REASON_FLUSH_FAIL);
737 spin_lock(&sbi->dev_lock);
738 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
739 spin_unlock(&sbi->dev_lock);
745 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
746 enum dirty_type dirty_type)
748 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
750 /* need not be added */
751 if (IS_CURSEG(sbi, segno))
754 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
755 dirty_i->nr_dirty[dirty_type]++;
757 if (dirty_type == DIRTY) {
758 struct seg_entry *sentry = get_seg_entry(sbi, segno);
759 enum dirty_type t = sentry->type;
761 if (unlikely(t >= DIRTY)) {
765 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
766 dirty_i->nr_dirty[t]++;
768 if (__is_large_section(sbi)) {
769 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
770 block_t valid_blocks =
771 get_valid_blocks(sbi, segno, true);
773 f2fs_bug_on(sbi, unlikely(!valid_blocks ||
774 valid_blocks == CAP_BLKS_PER_SEC(sbi)));
776 if (!IS_CURSEC(sbi, secno))
777 set_bit(secno, dirty_i->dirty_secmap);
782 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
783 enum dirty_type dirty_type)
785 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
786 block_t valid_blocks;
788 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
789 dirty_i->nr_dirty[dirty_type]--;
791 if (dirty_type == DIRTY) {
792 struct seg_entry *sentry = get_seg_entry(sbi, segno);
793 enum dirty_type t = sentry->type;
795 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
796 dirty_i->nr_dirty[t]--;
798 valid_blocks = get_valid_blocks(sbi, segno, true);
799 if (valid_blocks == 0) {
800 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
801 dirty_i->victim_secmap);
802 #ifdef CONFIG_F2FS_CHECK_FS
803 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
806 if (__is_large_section(sbi)) {
807 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
810 valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
811 clear_bit(secno, dirty_i->dirty_secmap);
815 if (!IS_CURSEC(sbi, secno))
816 set_bit(secno, dirty_i->dirty_secmap);
822 * Should not occur error such as -ENOMEM.
823 * Adding dirty entry into seglist is not critical operation.
824 * If a given segment is one of current working segments, it won't be added.
826 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
828 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
829 unsigned short valid_blocks, ckpt_valid_blocks;
830 unsigned int usable_blocks;
832 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
835 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
836 mutex_lock(&dirty_i->seglist_lock);
838 valid_blocks = get_valid_blocks(sbi, segno, false);
839 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
841 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
842 ckpt_valid_blocks == usable_blocks)) {
843 __locate_dirty_segment(sbi, segno, PRE);
844 __remove_dirty_segment(sbi, segno, DIRTY);
845 } else if (valid_blocks < usable_blocks) {
846 __locate_dirty_segment(sbi, segno, DIRTY);
848 /* Recovery routine with SSR needs this */
849 __remove_dirty_segment(sbi, segno, DIRTY);
852 mutex_unlock(&dirty_i->seglist_lock);
855 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
856 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
858 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
861 mutex_lock(&dirty_i->seglist_lock);
862 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
863 if (get_valid_blocks(sbi, segno, false))
865 if (IS_CURSEG(sbi, segno))
867 __locate_dirty_segment(sbi, segno, PRE);
868 __remove_dirty_segment(sbi, segno, DIRTY);
870 mutex_unlock(&dirty_i->seglist_lock);
873 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
876 (overprovision_segments(sbi) - reserved_segments(sbi));
877 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
878 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
879 block_t holes[2] = {0, 0}; /* DATA and NODE */
881 struct seg_entry *se;
884 mutex_lock(&dirty_i->seglist_lock);
885 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
886 se = get_seg_entry(sbi, segno);
887 if (IS_NODESEG(se->type))
888 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
891 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
894 mutex_unlock(&dirty_i->seglist_lock);
896 unusable = max(holes[DATA], holes[NODE]);
897 if (unusable > ovp_holes)
898 return unusable - ovp_holes;
902 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
905 (overprovision_segments(sbi) - reserved_segments(sbi));
906 if (unusable > F2FS_OPTION(sbi).unusable_cap)
908 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
909 dirty_segments(sbi) > ovp_hole_segs)
914 /* This is only used by SBI_CP_DISABLED */
915 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
917 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
918 unsigned int segno = 0;
920 mutex_lock(&dirty_i->seglist_lock);
921 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
922 if (get_valid_blocks(sbi, segno, false))
924 if (get_ckpt_valid_blocks(sbi, segno, false))
926 mutex_unlock(&dirty_i->seglist_lock);
929 mutex_unlock(&dirty_i->seglist_lock);
933 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
934 struct block_device *bdev, block_t lstart,
935 block_t start, block_t len)
937 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
938 struct list_head *pend_list;
939 struct discard_cmd *dc;
941 f2fs_bug_on(sbi, !len);
943 pend_list = &dcc->pend_list[plist_idx(len)];
945 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
946 INIT_LIST_HEAD(&dc->list);
948 dc->di.lstart = lstart;
949 dc->di.start = start;
955 init_completion(&dc->wait);
956 list_add_tail(&dc->list, pend_list);
957 spin_lock_init(&dc->lock);
959 atomic_inc(&dcc->discard_cmd_cnt);
960 dcc->undiscard_blks += len;
965 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
967 #ifdef CONFIG_F2FS_CHECK_FS
968 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
969 struct rb_node *cur = rb_first_cached(&dcc->root), *next;
970 struct discard_cmd *cur_dc, *next_dc;
977 cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
978 next_dc = rb_entry(next, struct discard_cmd, rb_node);
980 if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
981 f2fs_info(sbi, "broken discard_rbtree, "
982 "cur(%u, %u) next(%u, %u)",
983 cur_dc->di.lstart, cur_dc->di.len,
984 next_dc->di.lstart, next_dc->di.len);
993 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
996 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
997 struct rb_node *node = dcc->root.rb_root.rb_node;
998 struct discard_cmd *dc;
1001 dc = rb_entry(node, struct discard_cmd, rb_node);
1003 if (blkaddr < dc->di.lstart)
1004 node = node->rb_left;
1005 else if (blkaddr >= dc->di.lstart + dc->di.len)
1006 node = node->rb_right;
1013 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1015 struct discard_cmd **prev_entry,
1016 struct discard_cmd **next_entry,
1017 struct rb_node ***insert_p,
1018 struct rb_node **insert_parent)
1020 struct rb_node **pnode = &root->rb_root.rb_node;
1021 struct rb_node *parent = NULL, *tmp_node;
1022 struct discard_cmd *dc;
1025 *insert_parent = NULL;
1029 if (RB_EMPTY_ROOT(&root->rb_root))
1034 dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1036 if (blkaddr < dc->di.lstart)
1037 pnode = &(*pnode)->rb_left;
1038 else if (blkaddr >= dc->di.lstart + dc->di.len)
1039 pnode = &(*pnode)->rb_right;
1041 goto lookup_neighbors;
1045 *insert_parent = parent;
1047 dc = rb_entry(parent, struct discard_cmd, rb_node);
1049 if (parent && blkaddr > dc->di.lstart)
1050 tmp_node = rb_next(parent);
1051 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1054 if (parent && blkaddr < dc->di.lstart)
1055 tmp_node = rb_prev(parent);
1056 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1060 /* lookup prev node for merging backward later */
1061 tmp_node = rb_prev(&dc->rb_node);
1062 *prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1064 /* lookup next node for merging frontward later */
1065 tmp_node = rb_next(&dc->rb_node);
1066 *next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1070 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1071 struct discard_cmd *dc)
1073 if (dc->state == D_DONE)
1074 atomic_sub(dc->queued, &dcc->queued_discard);
1076 list_del(&dc->list);
1077 rb_erase_cached(&dc->rb_node, &dcc->root);
1078 dcc->undiscard_blks -= dc->di.len;
1080 kmem_cache_free(discard_cmd_slab, dc);
1082 atomic_dec(&dcc->discard_cmd_cnt);
1085 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1086 struct discard_cmd *dc)
1088 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1089 unsigned long flags;
1091 trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1093 spin_lock_irqsave(&dc->lock, flags);
1095 spin_unlock_irqrestore(&dc->lock, flags);
1098 spin_unlock_irqrestore(&dc->lock, flags);
1100 f2fs_bug_on(sbi, dc->ref);
1102 if (dc->error == -EOPNOTSUPP)
1107 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1108 KERN_INFO, sbi->sb->s_id,
1109 dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1110 __detach_discard_cmd(dcc, dc);
1113 static void f2fs_submit_discard_endio(struct bio *bio)
1115 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1116 unsigned long flags;
1118 spin_lock_irqsave(&dc->lock, flags);
1120 dc->error = blk_status_to_errno(bio->bi_status);
1122 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1124 complete_all(&dc->wait);
1126 spin_unlock_irqrestore(&dc->lock, flags);
1130 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1131 block_t start, block_t end)
1133 #ifdef CONFIG_F2FS_CHECK_FS
1134 struct seg_entry *sentry;
1136 block_t blk = start;
1137 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1141 segno = GET_SEGNO(sbi, blk);
1142 sentry = get_seg_entry(sbi, segno);
1143 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1145 if (end < START_BLOCK(sbi, segno + 1))
1146 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1149 map = (unsigned long *)(sentry->cur_valid_map);
1150 offset = __find_rev_next_bit(map, size, offset);
1151 f2fs_bug_on(sbi, offset != size);
1152 blk = START_BLOCK(sbi, segno + 1);
1157 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1158 struct discard_policy *dpolicy,
1159 int discard_type, unsigned int granularity)
1161 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1164 dpolicy->type = discard_type;
1165 dpolicy->sync = true;
1166 dpolicy->ordered = false;
1167 dpolicy->granularity = granularity;
1169 dpolicy->max_requests = dcc->max_discard_request;
1170 dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1171 dpolicy->timeout = false;
1173 if (discard_type == DPOLICY_BG) {
1174 dpolicy->min_interval = dcc->min_discard_issue_time;
1175 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1176 dpolicy->max_interval = dcc->max_discard_issue_time;
1177 if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE)
1178 dpolicy->io_aware = true;
1179 else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE)
1180 dpolicy->io_aware = false;
1181 dpolicy->sync = false;
1182 dpolicy->ordered = true;
1183 if (utilization(sbi) > dcc->discard_urgent_util) {
1184 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1185 if (atomic_read(&dcc->discard_cmd_cnt))
1186 dpolicy->max_interval =
1187 dcc->min_discard_issue_time;
1189 } else if (discard_type == DPOLICY_FORCE) {
1190 dpolicy->min_interval = dcc->min_discard_issue_time;
1191 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1192 dpolicy->max_interval = dcc->max_discard_issue_time;
1193 dpolicy->io_aware = false;
1194 } else if (discard_type == DPOLICY_FSTRIM) {
1195 dpolicy->io_aware = false;
1196 } else if (discard_type == DPOLICY_UMOUNT) {
1197 dpolicy->io_aware = false;
1198 /* we need to issue all to keep CP_TRIMMED_FLAG */
1199 dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1200 dpolicy->timeout = true;
1204 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1205 struct block_device *bdev, block_t lstart,
1206 block_t start, block_t len);
1208 #ifdef CONFIG_BLK_DEV_ZONED
1209 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1210 struct discard_cmd *dc, blk_opf_t flag,
1211 struct list_head *wait_list,
1212 unsigned int *issued)
1214 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1215 struct block_device *bdev = dc->bdev;
1216 struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1217 unsigned long flags;
1219 trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1221 spin_lock_irqsave(&dc->lock, flags);
1222 dc->state = D_SUBMIT;
1224 spin_unlock_irqrestore(&dc->lock, flags);
1229 atomic_inc(&dcc->queued_discard);
1231 list_move_tail(&dc->list, wait_list);
1233 /* sanity check on discard range */
1234 __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1236 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1237 bio->bi_private = dc;
1238 bio->bi_end_io = f2fs_submit_discard_endio;
1241 atomic_inc(&dcc->issued_discard);
1242 f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1246 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1247 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1248 struct discard_policy *dpolicy,
1249 struct discard_cmd *dc, int *issued)
1251 struct block_device *bdev = dc->bdev;
1252 unsigned int max_discard_blocks =
1253 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1254 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1255 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1256 &(dcc->fstrim_list) : &(dcc->wait_list);
1257 blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1258 block_t lstart, start, len, total_len;
1261 if (dc->state != D_PREP)
1264 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1267 #ifdef CONFIG_BLK_DEV_ZONED
1268 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1269 int devi = f2fs_bdev_index(sbi, bdev);
1274 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1275 __submit_zone_reset_cmd(sbi, dc, flag,
1282 trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1284 lstart = dc->di.lstart;
1285 start = dc->di.start;
1291 while (total_len && *issued < dpolicy->max_requests && !err) {
1292 struct bio *bio = NULL;
1293 unsigned long flags;
1296 if (len > max_discard_blocks) {
1297 len = max_discard_blocks;
1302 if (*issued == dpolicy->max_requests)
1307 if (time_to_inject(sbi, FAULT_DISCARD)) {
1310 err = __blkdev_issue_discard(bdev,
1311 SECTOR_FROM_BLOCK(start),
1312 SECTOR_FROM_BLOCK(len),
1316 spin_lock_irqsave(&dc->lock, flags);
1317 if (dc->state == D_PARTIAL)
1318 dc->state = D_SUBMIT;
1319 spin_unlock_irqrestore(&dc->lock, flags);
1324 f2fs_bug_on(sbi, !bio);
1327 * should keep before submission to avoid D_DONE
1330 spin_lock_irqsave(&dc->lock, flags);
1332 dc->state = D_SUBMIT;
1334 dc->state = D_PARTIAL;
1336 spin_unlock_irqrestore(&dc->lock, flags);
1338 atomic_inc(&dcc->queued_discard);
1340 list_move_tail(&dc->list, wait_list);
1342 /* sanity check on discard range */
1343 __check_sit_bitmap(sbi, lstart, lstart + len);
1345 bio->bi_private = dc;
1346 bio->bi_end_io = f2fs_submit_discard_endio;
1347 bio->bi_opf |= flag;
1350 atomic_inc(&dcc->issued_discard);
1352 f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1361 dcc->undiscard_blks -= len;
1362 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1367 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1368 struct block_device *bdev, block_t lstart,
1369 block_t start, block_t len)
1371 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1372 struct rb_node **p = &dcc->root.rb_root.rb_node;
1373 struct rb_node *parent = NULL;
1374 struct discard_cmd *dc;
1375 bool leftmost = true;
1377 /* look up rb tree to find parent node */
1380 dc = rb_entry(parent, struct discard_cmd, rb_node);
1382 if (lstart < dc->di.lstart) {
1384 } else if (lstart >= dc->di.lstart + dc->di.len) {
1385 p = &(*p)->rb_right;
1388 /* Let's skip to add, if exists */
1393 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1395 rb_link_node(&dc->rb_node, parent, p);
1396 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1399 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1400 struct discard_cmd *dc)
1402 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1405 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1406 struct discard_cmd *dc, block_t blkaddr)
1408 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1409 struct discard_info di = dc->di;
1410 bool modified = false;
1412 if (dc->state == D_DONE || dc->di.len == 1) {
1413 __remove_discard_cmd(sbi, dc);
1417 dcc->undiscard_blks -= di.len;
1419 if (blkaddr > di.lstart) {
1420 dc->di.len = blkaddr - dc->di.lstart;
1421 dcc->undiscard_blks += dc->di.len;
1422 __relocate_discard_cmd(dcc, dc);
1426 if (blkaddr < di.lstart + di.len - 1) {
1428 __insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1429 di.start + blkaddr + 1 - di.lstart,
1430 di.lstart + di.len - 1 - blkaddr);
1435 dcc->undiscard_blks += dc->di.len;
1436 __relocate_discard_cmd(dcc, dc);
1441 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1442 struct block_device *bdev, block_t lstart,
1443 block_t start, block_t len)
1445 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1446 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1447 struct discard_cmd *dc;
1448 struct discard_info di = {0};
1449 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1450 unsigned int max_discard_blocks =
1451 SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1452 block_t end = lstart + len;
1454 dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1455 &prev_dc, &next_dc, &insert_p, &insert_parent);
1461 di.len = next_dc ? next_dc->di.lstart - lstart : len;
1462 di.len = min(di.len, len);
1467 struct rb_node *node;
1468 bool merged = false;
1469 struct discard_cmd *tdc = NULL;
1472 di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1473 if (di.lstart < lstart)
1475 if (di.lstart >= end)
1478 if (!next_dc || next_dc->di.lstart > end)
1479 di.len = end - di.lstart;
1481 di.len = next_dc->di.lstart - di.lstart;
1482 di.start = start + di.lstart - lstart;
1488 if (prev_dc && prev_dc->state == D_PREP &&
1489 prev_dc->bdev == bdev &&
1490 __is_discard_back_mergeable(&di, &prev_dc->di,
1491 max_discard_blocks)) {
1492 prev_dc->di.len += di.len;
1493 dcc->undiscard_blks += di.len;
1494 __relocate_discard_cmd(dcc, prev_dc);
1500 if (next_dc && next_dc->state == D_PREP &&
1501 next_dc->bdev == bdev &&
1502 __is_discard_front_mergeable(&di, &next_dc->di,
1503 max_discard_blocks)) {
1504 next_dc->di.lstart = di.lstart;
1505 next_dc->di.len += di.len;
1506 next_dc->di.start = di.start;
1507 dcc->undiscard_blks += di.len;
1508 __relocate_discard_cmd(dcc, next_dc);
1510 __remove_discard_cmd(sbi, tdc);
1515 __insert_discard_cmd(sbi, bdev,
1516 di.lstart, di.start, di.len);
1522 node = rb_next(&prev_dc->rb_node);
1523 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1527 #ifdef CONFIG_BLK_DEV_ZONED
1528 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1529 struct block_device *bdev, block_t blkstart, block_t lblkstart,
1532 trace_f2fs_queue_reset_zone(bdev, blkstart);
1534 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1535 __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1536 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1540 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1541 struct block_device *bdev, block_t blkstart, block_t blklen)
1543 block_t lblkstart = blkstart;
1545 if (!f2fs_bdev_support_discard(bdev))
1548 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1550 if (f2fs_is_multi_device(sbi)) {
1551 int devi = f2fs_target_device_index(sbi, blkstart);
1553 blkstart -= FDEV(devi).start_blk;
1555 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1556 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1557 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1560 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1561 struct discard_policy *dpolicy, int *issued)
1563 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1564 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1565 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1566 struct discard_cmd *dc;
1567 struct blk_plug plug;
1568 bool io_interrupted = false;
1570 mutex_lock(&dcc->cmd_lock);
1571 dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1572 &prev_dc, &next_dc, &insert_p, &insert_parent);
1576 blk_start_plug(&plug);
1579 struct rb_node *node;
1582 if (dc->state != D_PREP)
1585 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1586 io_interrupted = true;
1590 dcc->next_pos = dc->di.lstart + dc->di.len;
1591 err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1593 if (*issued >= dpolicy->max_requests)
1596 node = rb_next(&dc->rb_node);
1598 __remove_discard_cmd(sbi, dc);
1599 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1602 blk_finish_plug(&plug);
1607 mutex_unlock(&dcc->cmd_lock);
1609 if (!(*issued) && io_interrupted)
1612 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1613 struct discard_policy *dpolicy);
1615 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1616 struct discard_policy *dpolicy)
1618 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1619 struct list_head *pend_list;
1620 struct discard_cmd *dc, *tmp;
1621 struct blk_plug plug;
1623 bool io_interrupted = false;
1625 if (dpolicy->timeout)
1626 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1630 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1631 if (dpolicy->timeout &&
1632 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1635 if (i + 1 < dpolicy->granularity)
1638 if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1639 __issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1643 pend_list = &dcc->pend_list[i];
1645 mutex_lock(&dcc->cmd_lock);
1646 if (list_empty(pend_list))
1648 if (unlikely(dcc->rbtree_check))
1649 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1650 blk_start_plug(&plug);
1651 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1652 f2fs_bug_on(sbi, dc->state != D_PREP);
1654 if (dpolicy->timeout &&
1655 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1658 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1659 !is_idle(sbi, DISCARD_TIME)) {
1660 io_interrupted = true;
1664 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1666 if (issued >= dpolicy->max_requests)
1669 blk_finish_plug(&plug);
1671 mutex_unlock(&dcc->cmd_lock);
1673 if (issued >= dpolicy->max_requests || io_interrupted)
1677 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1678 __wait_all_discard_cmd(sbi, dpolicy);
1682 if (!issued && io_interrupted)
1688 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1690 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1691 struct list_head *pend_list;
1692 struct discard_cmd *dc, *tmp;
1694 bool dropped = false;
1696 mutex_lock(&dcc->cmd_lock);
1697 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1698 pend_list = &dcc->pend_list[i];
1699 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1700 f2fs_bug_on(sbi, dc->state != D_PREP);
1701 __remove_discard_cmd(sbi, dc);
1705 mutex_unlock(&dcc->cmd_lock);
1710 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1712 __drop_discard_cmd(sbi);
1715 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1716 struct discard_cmd *dc)
1718 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1719 unsigned int len = 0;
1721 wait_for_completion_io(&dc->wait);
1722 mutex_lock(&dcc->cmd_lock);
1723 f2fs_bug_on(sbi, dc->state != D_DONE);
1728 __remove_discard_cmd(sbi, dc);
1730 mutex_unlock(&dcc->cmd_lock);
1735 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1736 struct discard_policy *dpolicy,
1737 block_t start, block_t end)
1739 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1740 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1741 &(dcc->fstrim_list) : &(dcc->wait_list);
1742 struct discard_cmd *dc = NULL, *iter, *tmp;
1743 unsigned int trimmed = 0;
1748 mutex_lock(&dcc->cmd_lock);
1749 list_for_each_entry_safe(iter, tmp, wait_list, list) {
1750 if (iter->di.lstart + iter->di.len <= start ||
1751 end <= iter->di.lstart)
1753 if (iter->di.len < dpolicy->granularity)
1755 if (iter->state == D_DONE && !iter->ref) {
1756 wait_for_completion_io(&iter->wait);
1758 trimmed += iter->di.len;
1759 __remove_discard_cmd(sbi, iter);
1766 mutex_unlock(&dcc->cmd_lock);
1769 trimmed += __wait_one_discard_bio(sbi, dc);
1776 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1777 struct discard_policy *dpolicy)
1779 struct discard_policy dp;
1780 unsigned int discard_blks;
1783 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1786 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1787 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1788 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1789 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1791 return discard_blks;
1794 /* This should be covered by global mutex, &sit_i->sentry_lock */
1795 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1797 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1798 struct discard_cmd *dc;
1799 bool need_wait = false;
1801 mutex_lock(&dcc->cmd_lock);
1802 dc = __lookup_discard_cmd(sbi, blkaddr);
1803 #ifdef CONFIG_BLK_DEV_ZONED
1804 if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1805 int devi = f2fs_bdev_index(sbi, dc->bdev);
1808 mutex_unlock(&dcc->cmd_lock);
1812 if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1813 /* force submit zone reset */
1814 if (dc->state == D_PREP)
1815 __submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1816 &dcc->wait_list, NULL);
1818 mutex_unlock(&dcc->cmd_lock);
1819 /* wait zone reset */
1820 __wait_one_discard_bio(sbi, dc);
1826 if (dc->state == D_PREP) {
1827 __punch_discard_cmd(sbi, dc, blkaddr);
1833 mutex_unlock(&dcc->cmd_lock);
1836 __wait_one_discard_bio(sbi, dc);
1839 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1841 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1843 if (dcc && dcc->f2fs_issue_discard) {
1844 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1846 dcc->f2fs_issue_discard = NULL;
1847 kthread_stop(discard_thread);
1852 * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1853 * @sbi: the f2fs_sb_info data for discard cmd to issue
1855 * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1857 * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1859 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1861 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1862 struct discard_policy dpolicy;
1865 if (!atomic_read(&dcc->discard_cmd_cnt))
1868 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1869 dcc->discard_granularity);
1870 __issue_discard_cmd(sbi, &dpolicy);
1871 dropped = __drop_discard_cmd(sbi);
1873 /* just to make sure there is no pending discard commands */
1874 __wait_all_discard_cmd(sbi, NULL);
1876 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1880 static int issue_discard_thread(void *data)
1882 struct f2fs_sb_info *sbi = data;
1883 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1884 wait_queue_head_t *q = &dcc->discard_wait_queue;
1885 struct discard_policy dpolicy;
1886 unsigned int wait_ms = dcc->min_discard_issue_time;
1892 wait_event_freezable_timeout(*q,
1893 kthread_should_stop() || dcc->discard_wake,
1894 msecs_to_jiffies(wait_ms));
1896 if (sbi->gc_mode == GC_URGENT_HIGH ||
1897 !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1898 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1899 MIN_DISCARD_GRANULARITY);
1901 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1902 dcc->discard_granularity);
1904 if (dcc->discard_wake)
1905 dcc->discard_wake = false;
1907 /* clean up pending candidates before going to sleep */
1908 if (atomic_read(&dcc->queued_discard))
1909 __wait_all_discard_cmd(sbi, NULL);
1911 if (f2fs_readonly(sbi->sb))
1913 if (kthread_should_stop())
1915 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1916 !atomic_read(&dcc->discard_cmd_cnt)) {
1917 wait_ms = dpolicy.max_interval;
1921 sb_start_intwrite(sbi->sb);
1923 issued = __issue_discard_cmd(sbi, &dpolicy);
1925 __wait_all_discard_cmd(sbi, &dpolicy);
1926 wait_ms = dpolicy.min_interval;
1927 } else if (issued == -1) {
1928 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1930 wait_ms = dpolicy.mid_interval;
1932 wait_ms = dpolicy.max_interval;
1934 if (!atomic_read(&dcc->discard_cmd_cnt))
1935 wait_ms = dpolicy.max_interval;
1937 sb_end_intwrite(sbi->sb);
1939 } while (!kthread_should_stop());
1943 #ifdef CONFIG_BLK_DEV_ZONED
1944 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1945 struct block_device *bdev, block_t blkstart, block_t blklen)
1947 sector_t sector, nr_sects;
1948 block_t lblkstart = blkstart;
1952 if (f2fs_is_multi_device(sbi)) {
1953 devi = f2fs_target_device_index(sbi, blkstart);
1954 if (blkstart < FDEV(devi).start_blk ||
1955 blkstart > FDEV(devi).end_blk) {
1956 f2fs_err(sbi, "Invalid block %x", blkstart);
1959 blkstart -= FDEV(devi).start_blk;
1962 /* For sequential zones, reset the zone write pointer */
1963 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1964 sector = SECTOR_FROM_BLOCK(blkstart);
1965 nr_sects = SECTOR_FROM_BLOCK(blklen);
1966 div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1968 if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1969 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1970 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1975 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1976 trace_f2fs_issue_reset_zone(bdev, blkstart);
1977 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1978 sector, nr_sects, GFP_NOFS);
1981 __queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1985 /* For conventional zones, use regular discard if supported */
1986 __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1991 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1992 struct block_device *bdev, block_t blkstart, block_t blklen)
1994 #ifdef CONFIG_BLK_DEV_ZONED
1995 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1996 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1998 __queue_discard_cmd(sbi, bdev, blkstart, blklen);
2002 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2003 block_t blkstart, block_t blklen)
2005 sector_t start = blkstart, len = 0;
2006 struct block_device *bdev;
2007 struct seg_entry *se;
2008 unsigned int offset;
2012 bdev = f2fs_target_device(sbi, blkstart, NULL);
2014 for (i = blkstart; i < blkstart + blklen; i++, len++) {
2016 struct block_device *bdev2 =
2017 f2fs_target_device(sbi, i, NULL);
2019 if (bdev2 != bdev) {
2020 err = __issue_discard_async(sbi, bdev,
2030 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2031 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2033 if (f2fs_block_unit_discard(sbi) &&
2034 !f2fs_test_and_set_bit(offset, se->discard_map))
2035 sbi->discard_blks--;
2039 err = __issue_discard_async(sbi, bdev, start, len);
2043 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2046 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2047 int max_blocks = sbi->blocks_per_seg;
2048 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2049 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2050 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2051 unsigned long *discard_map = (unsigned long *)se->discard_map;
2052 unsigned long *dmap = SIT_I(sbi)->tmp_map;
2053 unsigned int start = 0, end = -1;
2054 bool force = (cpc->reason & CP_DISCARD);
2055 struct discard_entry *de = NULL;
2056 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2059 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
2060 !f2fs_block_unit_discard(sbi))
2064 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2065 SM_I(sbi)->dcc_info->nr_discards >=
2066 SM_I(sbi)->dcc_info->max_discards)
2070 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2071 for (i = 0; i < entries; i++)
2072 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2073 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2075 while (force || SM_I(sbi)->dcc_info->nr_discards <=
2076 SM_I(sbi)->dcc_info->max_discards) {
2077 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
2078 if (start >= max_blocks)
2081 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
2082 if (force && start && end != max_blocks
2083 && (end - start) < cpc->trim_minlen)
2090 de = f2fs_kmem_cache_alloc(discard_entry_slab,
2091 GFP_F2FS_ZERO, true, NULL);
2092 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2093 list_add_tail(&de->list, head);
2096 for (i = start; i < end; i++)
2097 __set_bit_le(i, (void *)de->discard_map);
2099 SM_I(sbi)->dcc_info->nr_discards += end - start;
2104 static void release_discard_addr(struct discard_entry *entry)
2106 list_del(&entry->list);
2107 kmem_cache_free(discard_entry_slab, entry);
2110 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2112 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2113 struct discard_entry *entry, *this;
2116 list_for_each_entry_safe(entry, this, head, list)
2117 release_discard_addr(entry);
2121 * Should call f2fs_clear_prefree_segments after checkpoint is done.
2123 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2125 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2128 mutex_lock(&dirty_i->seglist_lock);
2129 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2130 __set_test_and_free(sbi, segno, false);
2131 mutex_unlock(&dirty_i->seglist_lock);
2134 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2135 struct cp_control *cpc)
2137 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2138 struct list_head *head = &dcc->entry_list;
2139 struct discard_entry *entry, *this;
2140 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2141 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2142 unsigned int start = 0, end = -1;
2143 unsigned int secno, start_segno;
2144 bool force = (cpc->reason & CP_DISCARD);
2145 bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2146 DISCARD_UNIT_SECTION;
2148 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2149 section_alignment = true;
2151 mutex_lock(&dirty_i->seglist_lock);
2156 if (section_alignment && end != -1)
2158 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2159 if (start >= MAIN_SEGS(sbi))
2161 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2164 if (section_alignment) {
2165 start = rounddown(start, sbi->segs_per_sec);
2166 end = roundup(end, sbi->segs_per_sec);
2169 for (i = start; i < end; i++) {
2170 if (test_and_clear_bit(i, prefree_map))
2171 dirty_i->nr_dirty[PRE]--;
2174 if (!f2fs_realtime_discard_enable(sbi))
2177 if (force && start >= cpc->trim_start &&
2178 (end - 1) <= cpc->trim_end)
2181 /* Should cover 2MB zoned device for zone-based reset */
2182 if (!f2fs_sb_has_blkzoned(sbi) &&
2183 (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2184 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2185 (end - start) << sbi->log_blocks_per_seg);
2189 secno = GET_SEC_FROM_SEG(sbi, start);
2190 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2191 if (!IS_CURSEC(sbi, secno) &&
2192 !get_valid_blocks(sbi, start, true))
2193 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2194 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2196 start = start_segno + sbi->segs_per_sec;
2202 mutex_unlock(&dirty_i->seglist_lock);
2204 if (!f2fs_block_unit_discard(sbi))
2207 /* send small discards */
2208 list_for_each_entry_safe(entry, this, head, list) {
2209 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2210 bool is_valid = test_bit_le(0, entry->discard_map);
2214 next_pos = find_next_zero_bit_le(entry->discard_map,
2215 sbi->blocks_per_seg, cur_pos);
2216 len = next_pos - cur_pos;
2218 if (f2fs_sb_has_blkzoned(sbi) ||
2219 (force && len < cpc->trim_minlen))
2222 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2226 next_pos = find_next_bit_le(entry->discard_map,
2227 sbi->blocks_per_seg, cur_pos);
2231 is_valid = !is_valid;
2233 if (cur_pos < sbi->blocks_per_seg)
2236 release_discard_addr(entry);
2237 dcc->nr_discards -= total_len;
2241 wake_up_discard_thread(sbi, false);
2244 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2246 dev_t dev = sbi->sb->s_bdev->bd_dev;
2247 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2250 if (!f2fs_realtime_discard_enable(sbi))
2253 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2254 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2255 if (IS_ERR(dcc->f2fs_issue_discard)) {
2256 err = PTR_ERR(dcc->f2fs_issue_discard);
2257 dcc->f2fs_issue_discard = NULL;
2263 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2265 struct discard_cmd_control *dcc;
2268 if (SM_I(sbi)->dcc_info) {
2269 dcc = SM_I(sbi)->dcc_info;
2273 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2277 dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2278 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2279 dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2280 dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE;
2281 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2282 dcc->discard_granularity = sbi->blocks_per_seg;
2283 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2284 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2286 INIT_LIST_HEAD(&dcc->entry_list);
2287 for (i = 0; i < MAX_PLIST_NUM; i++)
2288 INIT_LIST_HEAD(&dcc->pend_list[i]);
2289 INIT_LIST_HEAD(&dcc->wait_list);
2290 INIT_LIST_HEAD(&dcc->fstrim_list);
2291 mutex_init(&dcc->cmd_lock);
2292 atomic_set(&dcc->issued_discard, 0);
2293 atomic_set(&dcc->queued_discard, 0);
2294 atomic_set(&dcc->discard_cmd_cnt, 0);
2295 dcc->nr_discards = 0;
2296 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2297 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2298 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2299 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2300 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2301 dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2302 dcc->undiscard_blks = 0;
2304 dcc->root = RB_ROOT_CACHED;
2305 dcc->rbtree_check = false;
2307 init_waitqueue_head(&dcc->discard_wait_queue);
2308 SM_I(sbi)->dcc_info = dcc;
2310 err = f2fs_start_discard_thread(sbi);
2313 SM_I(sbi)->dcc_info = NULL;
2319 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2321 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2326 f2fs_stop_discard_thread(sbi);
2329 * Recovery can cache discard commands, so in error path of
2330 * fill_super(), it needs to give a chance to handle them.
2332 f2fs_issue_discard_timeout(sbi);
2335 SM_I(sbi)->dcc_info = NULL;
2338 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2340 struct sit_info *sit_i = SIT_I(sbi);
2342 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2343 sit_i->dirty_sentries++;
2350 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2351 unsigned int segno, int modified)
2353 struct seg_entry *se = get_seg_entry(sbi, segno);
2357 __mark_sit_entry_dirty(sbi, segno);
2360 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2363 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2365 if (segno == NULL_SEGNO)
2367 return get_seg_entry(sbi, segno)->mtime;
2370 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2371 unsigned long long old_mtime)
2373 struct seg_entry *se;
2374 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2375 unsigned long long ctime = get_mtime(sbi, false);
2376 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2378 if (segno == NULL_SEGNO)
2381 se = get_seg_entry(sbi, segno);
2386 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2387 se->valid_blocks + 1);
2389 if (ctime > SIT_I(sbi)->max_mtime)
2390 SIT_I(sbi)->max_mtime = ctime;
2393 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2395 struct seg_entry *se;
2396 unsigned int segno, offset;
2397 long int new_vblocks;
2399 #ifdef CONFIG_F2FS_CHECK_FS
2403 segno = GET_SEGNO(sbi, blkaddr);
2405 se = get_seg_entry(sbi, segno);
2406 new_vblocks = se->valid_blocks + del;
2407 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2409 f2fs_bug_on(sbi, (new_vblocks < 0 ||
2410 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2412 se->valid_blocks = new_vblocks;
2414 /* Update valid block bitmap */
2416 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2417 #ifdef CONFIG_F2FS_CHECK_FS
2418 mir_exist = f2fs_test_and_set_bit(offset,
2419 se->cur_valid_map_mir);
2420 if (unlikely(exist != mir_exist)) {
2421 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2423 f2fs_bug_on(sbi, 1);
2426 if (unlikely(exist)) {
2427 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2429 f2fs_bug_on(sbi, 1);
2434 if (f2fs_block_unit_discard(sbi) &&
2435 !f2fs_test_and_set_bit(offset, se->discard_map))
2436 sbi->discard_blks--;
2439 * SSR should never reuse block which is checkpointed
2440 * or newly invalidated.
2442 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2443 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2444 se->ckpt_valid_blocks++;
2447 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2448 #ifdef CONFIG_F2FS_CHECK_FS
2449 mir_exist = f2fs_test_and_clear_bit(offset,
2450 se->cur_valid_map_mir);
2451 if (unlikely(exist != mir_exist)) {
2452 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2454 f2fs_bug_on(sbi, 1);
2457 if (unlikely(!exist)) {
2458 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2460 f2fs_bug_on(sbi, 1);
2463 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2465 * If checkpoints are off, we must not reuse data that
2466 * was used in the previous checkpoint. If it was used
2467 * before, we must track that to know how much space we
2470 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2471 spin_lock(&sbi->stat_lock);
2472 sbi->unusable_block_count++;
2473 spin_unlock(&sbi->stat_lock);
2477 if (f2fs_block_unit_discard(sbi) &&
2478 f2fs_test_and_clear_bit(offset, se->discard_map))
2479 sbi->discard_blks++;
2481 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2482 se->ckpt_valid_blocks += del;
2484 __mark_sit_entry_dirty(sbi, segno);
2486 /* update total number of valid blocks to be written in ckpt area */
2487 SIT_I(sbi)->written_valid_blocks += del;
2489 if (__is_large_section(sbi))
2490 get_sec_entry(sbi, segno)->valid_blocks += del;
2493 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2495 unsigned int segno = GET_SEGNO(sbi, addr);
2496 struct sit_info *sit_i = SIT_I(sbi);
2498 f2fs_bug_on(sbi, addr == NULL_ADDR);
2499 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2502 f2fs_invalidate_internal_cache(sbi, addr);
2504 /* add it into sit main buffer */
2505 down_write(&sit_i->sentry_lock);
2507 update_segment_mtime(sbi, addr, 0);
2508 update_sit_entry(sbi, addr, -1);
2510 /* add it into dirty seglist */
2511 locate_dirty_segment(sbi, segno);
2513 up_write(&sit_i->sentry_lock);
2516 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2518 struct sit_info *sit_i = SIT_I(sbi);
2519 unsigned int segno, offset;
2520 struct seg_entry *se;
2523 if (!__is_valid_data_blkaddr(blkaddr))
2526 down_read(&sit_i->sentry_lock);
2528 segno = GET_SEGNO(sbi, blkaddr);
2529 se = get_seg_entry(sbi, segno);
2530 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2532 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2535 up_read(&sit_i->sentry_lock);
2540 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2542 struct curseg_info *curseg = CURSEG_I(sbi, type);
2544 if (sbi->ckpt->alloc_type[type] == SSR)
2545 return sbi->blocks_per_seg;
2546 return curseg->next_blkoff;
2550 * Calculate the number of current summary pages for writing
2552 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2554 int valid_sum_count = 0;
2557 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2558 if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2560 le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2562 valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2565 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2566 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2567 if (valid_sum_count <= sum_in_page)
2569 else if ((valid_sum_count - sum_in_page) <=
2570 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2576 * Caller should put this summary page
2578 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2580 if (unlikely(f2fs_cp_error(sbi)))
2581 return ERR_PTR(-EIO);
2582 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2585 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2586 void *src, block_t blk_addr)
2588 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2590 memcpy(page_address(page), src, PAGE_SIZE);
2591 set_page_dirty(page);
2592 f2fs_put_page(page, 1);
2595 static void write_sum_page(struct f2fs_sb_info *sbi,
2596 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2598 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2601 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2602 int type, block_t blk_addr)
2604 struct curseg_info *curseg = CURSEG_I(sbi, type);
2605 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2606 struct f2fs_summary_block *src = curseg->sum_blk;
2607 struct f2fs_summary_block *dst;
2609 dst = (struct f2fs_summary_block *)page_address(page);
2610 memset(dst, 0, PAGE_SIZE);
2612 mutex_lock(&curseg->curseg_mutex);
2614 down_read(&curseg->journal_rwsem);
2615 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2616 up_read(&curseg->journal_rwsem);
2618 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2619 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2621 mutex_unlock(&curseg->curseg_mutex);
2623 set_page_dirty(page);
2624 f2fs_put_page(page, 1);
2627 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2628 struct curseg_info *curseg, int type)
2630 unsigned int segno = curseg->segno + 1;
2631 struct free_segmap_info *free_i = FREE_I(sbi);
2633 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2634 return !test_bit(segno, free_i->free_segmap);
2639 * Find a new segment from the free segments bitmap to right order
2640 * This function should be returned with success, otherwise BUG
2642 static void get_new_segment(struct f2fs_sb_info *sbi,
2643 unsigned int *newseg, bool new_sec, int dir)
2645 struct free_segmap_info *free_i = FREE_I(sbi);
2646 unsigned int segno, secno, zoneno;
2647 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2648 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2649 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2650 unsigned int left_start = hint;
2655 spin_lock(&free_i->segmap_lock);
2657 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2658 segno = find_next_zero_bit(free_i->free_segmap,
2659 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2660 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2664 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2665 if (secno >= MAIN_SECS(sbi)) {
2666 if (dir == ALLOC_RIGHT) {
2667 secno = find_first_zero_bit(free_i->free_secmap,
2669 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2672 left_start = hint - 1;
2678 while (test_bit(left_start, free_i->free_secmap)) {
2679 if (left_start > 0) {
2683 left_start = find_first_zero_bit(free_i->free_secmap,
2685 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2690 segno = GET_SEG_FROM_SEC(sbi, secno);
2691 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2693 /* give up on finding another zone */
2696 if (sbi->secs_per_zone == 1)
2698 if (zoneno == old_zoneno)
2700 if (dir == ALLOC_LEFT) {
2701 if (!go_left && zoneno + 1 >= total_zones)
2703 if (go_left && zoneno == 0)
2706 for (i = 0; i < NR_CURSEG_TYPE; i++)
2707 if (CURSEG_I(sbi, i)->zone == zoneno)
2710 if (i < NR_CURSEG_TYPE) {
2711 /* zone is in user, try another */
2713 hint = zoneno * sbi->secs_per_zone - 1;
2714 else if (zoneno + 1 >= total_zones)
2717 hint = (zoneno + 1) * sbi->secs_per_zone;
2719 goto find_other_zone;
2722 /* set it as dirty segment in free segmap */
2723 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2724 __set_inuse(sbi, segno);
2726 spin_unlock(&free_i->segmap_lock);
2729 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2731 struct curseg_info *curseg = CURSEG_I(sbi, type);
2732 struct summary_footer *sum_footer;
2733 unsigned short seg_type = curseg->seg_type;
2735 curseg->inited = true;
2736 curseg->segno = curseg->next_segno;
2737 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2738 curseg->next_blkoff = 0;
2739 curseg->next_segno = NULL_SEGNO;
2741 sum_footer = &(curseg->sum_blk->footer);
2742 memset(sum_footer, 0, sizeof(struct summary_footer));
2744 sanity_check_seg_type(sbi, seg_type);
2746 if (IS_DATASEG(seg_type))
2747 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2748 if (IS_NODESEG(seg_type))
2749 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2750 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2753 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2755 struct curseg_info *curseg = CURSEG_I(sbi, type);
2756 unsigned short seg_type = curseg->seg_type;
2758 sanity_check_seg_type(sbi, seg_type);
2759 if (f2fs_need_rand_seg(sbi))
2760 return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
2762 /* if segs_per_sec is large than 1, we need to keep original policy. */
2763 if (__is_large_section(sbi))
2764 return curseg->segno;
2766 /* inmem log may not locate on any segment after mount */
2767 if (!curseg->inited)
2770 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2773 if (test_opt(sbi, NOHEAP) &&
2774 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2777 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2778 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2780 /* find segments from 0 to reuse freed segments */
2781 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2784 return curseg->segno;
2788 * Allocate a current working segment.
2789 * This function always allocates a free segment in LFS manner.
2791 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2793 struct curseg_info *curseg = CURSEG_I(sbi, type);
2794 unsigned short seg_type = curseg->seg_type;
2795 unsigned int segno = curseg->segno;
2796 int dir = ALLOC_LEFT;
2799 write_sum_page(sbi, curseg->sum_blk,
2800 GET_SUM_BLOCK(sbi, segno));
2801 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2804 if (test_opt(sbi, NOHEAP))
2807 segno = __get_next_segno(sbi, type);
2808 get_new_segment(sbi, &segno, new_sec, dir);
2809 curseg->next_segno = segno;
2810 reset_curseg(sbi, type, 1);
2811 curseg->alloc_type = LFS;
2812 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2813 curseg->fragment_remained_chunk =
2814 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2817 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2818 int segno, block_t start)
2820 struct seg_entry *se = get_seg_entry(sbi, segno);
2821 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2822 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2823 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2824 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2827 for (i = 0; i < entries; i++)
2828 target_map[i] = ckpt_map[i] | cur_map[i];
2830 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2833 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2834 struct curseg_info *seg)
2836 return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2839 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2841 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2845 * This function always allocates a used segment(from dirty seglist) by SSR
2846 * manner, so it should recover the existing segment information of valid blocks
2848 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2850 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2851 struct curseg_info *curseg = CURSEG_I(sbi, type);
2852 unsigned int new_segno = curseg->next_segno;
2853 struct f2fs_summary_block *sum_node;
2854 struct page *sum_page;
2856 write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2858 __set_test_and_inuse(sbi, new_segno);
2860 mutex_lock(&dirty_i->seglist_lock);
2861 __remove_dirty_segment(sbi, new_segno, PRE);
2862 __remove_dirty_segment(sbi, new_segno, DIRTY);
2863 mutex_unlock(&dirty_i->seglist_lock);
2865 reset_curseg(sbi, type, 1);
2866 curseg->alloc_type = SSR;
2867 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2869 sum_page = f2fs_get_sum_page(sbi, new_segno);
2870 if (IS_ERR(sum_page)) {
2871 /* GC won't be able to use stale summary pages by cp_error */
2872 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2875 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2876 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2877 f2fs_put_page(sum_page, 1);
2880 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2881 int alloc_mode, unsigned long long age);
2883 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2884 int target_type, int alloc_mode,
2885 unsigned long long age)
2887 struct curseg_info *curseg = CURSEG_I(sbi, type);
2889 curseg->seg_type = target_type;
2891 if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2892 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2894 curseg->seg_type = se->type;
2895 change_curseg(sbi, type);
2897 /* allocate cold segment by default */
2898 curseg->seg_type = CURSEG_COLD_DATA;
2899 new_curseg(sbi, type, true);
2901 stat_inc_seg_type(sbi, curseg);
2904 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2906 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2908 if (!sbi->am.atgc_enabled)
2911 f2fs_down_read(&SM_I(sbi)->curseg_lock);
2913 mutex_lock(&curseg->curseg_mutex);
2914 down_write(&SIT_I(sbi)->sentry_lock);
2916 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2918 up_write(&SIT_I(sbi)->sentry_lock);
2919 mutex_unlock(&curseg->curseg_mutex);
2921 f2fs_up_read(&SM_I(sbi)->curseg_lock);
2924 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2926 __f2fs_init_atgc_curseg(sbi);
2929 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2931 struct curseg_info *curseg = CURSEG_I(sbi, type);
2933 mutex_lock(&curseg->curseg_mutex);
2934 if (!curseg->inited)
2937 if (get_valid_blocks(sbi, curseg->segno, false)) {
2938 write_sum_page(sbi, curseg->sum_blk,
2939 GET_SUM_BLOCK(sbi, curseg->segno));
2941 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2942 __set_test_and_free(sbi, curseg->segno, true);
2943 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2946 mutex_unlock(&curseg->curseg_mutex);
2949 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2951 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2953 if (sbi->am.atgc_enabled)
2954 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2957 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2959 struct curseg_info *curseg = CURSEG_I(sbi, type);
2961 mutex_lock(&curseg->curseg_mutex);
2962 if (!curseg->inited)
2964 if (get_valid_blocks(sbi, curseg->segno, false))
2967 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2968 __set_test_and_inuse(sbi, curseg->segno);
2969 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2971 mutex_unlock(&curseg->curseg_mutex);
2974 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2976 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2978 if (sbi->am.atgc_enabled)
2979 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2982 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2983 int alloc_mode, unsigned long long age)
2985 struct curseg_info *curseg = CURSEG_I(sbi, type);
2986 unsigned segno = NULL_SEGNO;
2987 unsigned short seg_type = curseg->seg_type;
2989 bool reversed = false;
2991 sanity_check_seg_type(sbi, seg_type);
2993 /* f2fs_need_SSR() already forces to do this */
2994 if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2995 curseg->next_segno = segno;
2999 /* For node segments, let's do SSR more intensively */
3000 if (IS_NODESEG(seg_type)) {
3001 if (seg_type >= CURSEG_WARM_NODE) {
3003 i = CURSEG_COLD_NODE;
3005 i = CURSEG_HOT_NODE;
3007 cnt = NR_CURSEG_NODE_TYPE;
3009 if (seg_type >= CURSEG_WARM_DATA) {
3011 i = CURSEG_COLD_DATA;
3013 i = CURSEG_HOT_DATA;
3015 cnt = NR_CURSEG_DATA_TYPE;
3018 for (; cnt-- > 0; reversed ? i-- : i++) {
3021 if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
3022 curseg->next_segno = segno;
3027 /* find valid_blocks=0 in dirty list */
3028 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3029 segno = get_free_segment(sbi);
3030 if (segno != NULL_SEGNO) {
3031 curseg->next_segno = segno;
3038 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3040 struct curseg_info *curseg = CURSEG_I(sbi, type);
3042 if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3043 curseg->seg_type == CURSEG_WARM_NODE)
3045 if (curseg->alloc_type == LFS &&
3046 is_next_segment_free(sbi, curseg, type) &&
3047 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3049 if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3054 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3055 unsigned int start, unsigned int end)
3057 struct curseg_info *curseg = CURSEG_I(sbi, type);
3060 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3061 mutex_lock(&curseg->curseg_mutex);
3062 down_write(&SIT_I(sbi)->sentry_lock);
3064 segno = CURSEG_I(sbi, type)->segno;
3065 if (segno < start || segno > end)
3068 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3069 change_curseg(sbi, type);
3071 new_curseg(sbi, type, true);
3073 stat_inc_seg_type(sbi, curseg);
3075 locate_dirty_segment(sbi, segno);
3077 up_write(&SIT_I(sbi)->sentry_lock);
3079 if (segno != curseg->segno)
3080 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3081 type, segno, curseg->segno);
3083 mutex_unlock(&curseg->curseg_mutex);
3084 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3087 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3088 bool new_sec, bool force)
3090 struct curseg_info *curseg = CURSEG_I(sbi, type);
3091 unsigned int old_segno;
3093 if (!force && curseg->inited &&
3094 !curseg->next_blkoff &&
3095 !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3096 !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3099 old_segno = curseg->segno;
3100 new_curseg(sbi, type, true);
3101 stat_inc_seg_type(sbi, curseg);
3102 locate_dirty_segment(sbi, old_segno);
3105 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3107 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3108 down_write(&SIT_I(sbi)->sentry_lock);
3109 __allocate_new_segment(sbi, type, true, force);
3110 up_write(&SIT_I(sbi)->sentry_lock);
3111 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3114 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3118 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3119 down_write(&SIT_I(sbi)->sentry_lock);
3120 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3121 __allocate_new_segment(sbi, i, false, false);
3122 up_write(&SIT_I(sbi)->sentry_lock);
3123 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3126 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3127 struct cp_control *cpc)
3129 __u64 trim_start = cpc->trim_start;
3130 bool has_candidate = false;
3132 down_write(&SIT_I(sbi)->sentry_lock);
3133 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3134 if (add_discard_addrs(sbi, cpc, true)) {
3135 has_candidate = true;
3139 up_write(&SIT_I(sbi)->sentry_lock);
3141 cpc->trim_start = trim_start;
3142 return has_candidate;
3145 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3146 struct discard_policy *dpolicy,
3147 unsigned int start, unsigned int end)
3149 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3150 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3151 struct rb_node **insert_p = NULL, *insert_parent = NULL;
3152 struct discard_cmd *dc;
3153 struct blk_plug plug;
3155 unsigned int trimmed = 0;
3160 mutex_lock(&dcc->cmd_lock);
3161 if (unlikely(dcc->rbtree_check))
3162 f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3164 dc = __lookup_discard_cmd_ret(&dcc->root, start,
3165 &prev_dc, &next_dc, &insert_p, &insert_parent);
3169 blk_start_plug(&plug);
3171 while (dc && dc->di.lstart <= end) {
3172 struct rb_node *node;
3175 if (dc->di.len < dpolicy->granularity)
3178 if (dc->state != D_PREP) {
3179 list_move_tail(&dc->list, &dcc->fstrim_list);
3183 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3185 if (issued >= dpolicy->max_requests) {
3186 start = dc->di.lstart + dc->di.len;
3189 __remove_discard_cmd(sbi, dc);
3191 blk_finish_plug(&plug);
3192 mutex_unlock(&dcc->cmd_lock);
3193 trimmed += __wait_all_discard_cmd(sbi, NULL);
3194 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3198 node = rb_next(&dc->rb_node);
3200 __remove_discard_cmd(sbi, dc);
3201 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3203 if (fatal_signal_pending(current))
3207 blk_finish_plug(&plug);
3208 mutex_unlock(&dcc->cmd_lock);
3213 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3215 __u64 start = F2FS_BYTES_TO_BLK(range->start);
3216 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3217 unsigned int start_segno, end_segno;
3218 block_t start_block, end_block;
3219 struct cp_control cpc;
3220 struct discard_policy dpolicy;
3221 unsigned long long trimmed = 0;
3223 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3225 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3228 if (end < MAIN_BLKADDR(sbi))
3231 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3232 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3233 return -EFSCORRUPTED;
3236 /* start/end segment number in main_area */
3237 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3238 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3239 GET_SEGNO(sbi, end);
3241 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3242 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3245 cpc.reason = CP_DISCARD;
3246 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3247 cpc.trim_start = start_segno;
3248 cpc.trim_end = end_segno;
3250 if (sbi->discard_blks == 0)
3253 f2fs_down_write(&sbi->gc_lock);
3254 stat_inc_cp_call_count(sbi, TOTAL_CALL);
3255 err = f2fs_write_checkpoint(sbi, &cpc);
3256 f2fs_up_write(&sbi->gc_lock);
3261 * We filed discard candidates, but actually we don't need to wait for
3262 * all of them, since they'll be issued in idle time along with runtime
3263 * discard option. User configuration looks like using runtime discard
3264 * or periodic fstrim instead of it.
3266 if (f2fs_realtime_discard_enable(sbi))
3269 start_block = START_BLOCK(sbi, start_segno);
3270 end_block = START_BLOCK(sbi, end_segno + 1);
3272 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3273 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3274 start_block, end_block);
3276 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3277 start_block, end_block);
3280 range->len = F2FS_BLK_TO_BYTES(trimmed);
3284 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3287 case WRITE_LIFE_SHORT:
3288 return CURSEG_HOT_DATA;
3289 case WRITE_LIFE_EXTREME:
3290 return CURSEG_COLD_DATA;
3292 return CURSEG_WARM_DATA;
3296 static int __get_segment_type_2(struct f2fs_io_info *fio)
3298 if (fio->type == DATA)
3299 return CURSEG_HOT_DATA;
3301 return CURSEG_HOT_NODE;
3304 static int __get_segment_type_4(struct f2fs_io_info *fio)
3306 if (fio->type == DATA) {
3307 struct inode *inode = fio->page->mapping->host;
3309 if (S_ISDIR(inode->i_mode))
3310 return CURSEG_HOT_DATA;
3312 return CURSEG_COLD_DATA;
3314 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3315 return CURSEG_WARM_NODE;
3317 return CURSEG_COLD_NODE;
3321 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3323 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3324 struct extent_info ei = {};
3326 if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3328 return NO_CHECK_TYPE;
3329 if (ei.age <= sbi->hot_data_age_threshold)
3330 return CURSEG_HOT_DATA;
3331 if (ei.age <= sbi->warm_data_age_threshold)
3332 return CURSEG_WARM_DATA;
3333 return CURSEG_COLD_DATA;
3335 return NO_CHECK_TYPE;
3338 static int __get_segment_type_6(struct f2fs_io_info *fio)
3340 if (fio->type == DATA) {
3341 struct inode *inode = fio->page->mapping->host;
3344 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3345 return CURSEG_COLD_DATA_PINNED;
3347 if (page_private_gcing(fio->page)) {
3348 if (fio->sbi->am.atgc_enabled &&
3349 (fio->io_type == FS_DATA_IO) &&
3350 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3351 return CURSEG_ALL_DATA_ATGC;
3353 return CURSEG_COLD_DATA;
3355 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3356 return CURSEG_COLD_DATA;
3358 type = __get_age_segment_type(inode, fio->page->index);
3359 if (type != NO_CHECK_TYPE)
3362 if (file_is_hot(inode) ||
3363 is_inode_flag_set(inode, FI_HOT_DATA) ||
3364 f2fs_is_cow_file(inode))
3365 return CURSEG_HOT_DATA;
3366 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3368 if (IS_DNODE(fio->page))
3369 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3371 return CURSEG_COLD_NODE;
3375 static int __get_segment_type(struct f2fs_io_info *fio)
3379 switch (F2FS_OPTION(fio->sbi).active_logs) {
3381 type = __get_segment_type_2(fio);
3384 type = __get_segment_type_4(fio);
3387 type = __get_segment_type_6(fio);
3390 f2fs_bug_on(fio->sbi, true);
3395 else if (IS_WARM(type))
3402 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3403 struct curseg_info *seg)
3405 /* To allocate block chunks in different sizes, use random number */
3406 if (--seg->fragment_remained_chunk > 0)
3409 seg->fragment_remained_chunk =
3410 get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3412 get_random_u32_inclusive(1, sbi->max_fragment_hole);
3415 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3416 block_t old_blkaddr, block_t *new_blkaddr,
3417 struct f2fs_summary *sum, int type,
3418 struct f2fs_io_info *fio)
3420 struct sit_info *sit_i = SIT_I(sbi);
3421 struct curseg_info *curseg = CURSEG_I(sbi, type);
3422 unsigned long long old_mtime;
3423 bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3424 struct seg_entry *se = NULL;
3425 bool segment_full = false;
3427 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3429 mutex_lock(&curseg->curseg_mutex);
3430 down_write(&sit_i->sentry_lock);
3433 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3434 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3435 sanity_check_seg_type(sbi, se->type);
3436 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3438 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3440 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3442 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3444 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3445 if (curseg->alloc_type == SSR) {
3446 curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3448 curseg->next_blkoff++;
3449 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3450 f2fs_randomize_chunk(sbi, curseg);
3452 if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3453 segment_full = true;
3454 stat_inc_block_count(sbi, curseg);
3457 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3459 update_segment_mtime(sbi, old_blkaddr, 0);
3462 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3465 * SIT information should be updated before segment allocation,
3466 * since SSR needs latest valid block information.
3468 update_sit_entry(sbi, *new_blkaddr, 1);
3469 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3470 update_sit_entry(sbi, old_blkaddr, -1);
3473 * If the current segment is full, flush it out and replace it with a
3478 get_atssr_segment(sbi, type, se->type,
3481 if (need_new_seg(sbi, type))
3482 new_curseg(sbi, type, false);
3484 change_curseg(sbi, type);
3485 stat_inc_seg_type(sbi, curseg);
3489 * segment dirty status should be updated after segment allocation,
3490 * so we just need to update status only one time after previous
3491 * segment being closed.
3493 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3494 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3496 if (IS_DATASEG(curseg->seg_type))
3497 atomic64_inc(&sbi->allocated_data_blocks);
3499 up_write(&sit_i->sentry_lock);
3501 if (page && IS_NODESEG(curseg->seg_type)) {
3502 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3504 f2fs_inode_chksum_set(sbi, page);
3508 struct f2fs_bio_info *io;
3510 if (F2FS_IO_ALIGNED(sbi))
3513 INIT_LIST_HEAD(&fio->list);
3515 io = sbi->write_io[fio->type] + fio->temp;
3516 spin_lock(&io->io_lock);
3517 list_add_tail(&fio->list, &io->io_list);
3518 spin_unlock(&io->io_lock);
3521 mutex_unlock(&curseg->curseg_mutex);
3523 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3526 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3527 block_t blkaddr, unsigned int blkcnt)
3529 if (!f2fs_is_multi_device(sbi))
3533 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3534 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3536 /* update device state for fsync */
3537 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3539 /* update device state for checkpoint */
3540 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3541 spin_lock(&sbi->dev_lock);
3542 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3543 spin_unlock(&sbi->dev_lock);
3553 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3555 int type = __get_segment_type(fio);
3556 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3559 f2fs_down_read(&fio->sbi->io_order_lock);
3561 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3562 &fio->new_blkaddr, sum, type, fio);
3563 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3564 f2fs_invalidate_internal_cache(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 f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3660 stat_inc_inplace_blocks(fio->sbi);
3662 if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3663 err = f2fs_merge_page_bio(fio);
3665 err = f2fs_submit_page_bio(fio);
3667 f2fs_update_device_state(fio->sbi, fio->ino,
3668 fio->new_blkaddr, 1);
3669 f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3670 fio->io_type, F2FS_BLKSIZE);
3675 if (fio->bio && *(fio->bio)) {
3676 struct bio *bio = *(fio->bio);
3678 bio->bi_status = BLK_STS_IOERR;
3685 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3690 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3691 if (CURSEG_I(sbi, i)->segno == segno)
3697 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3698 block_t old_blkaddr, block_t new_blkaddr,
3699 bool recover_curseg, bool recover_newaddr,
3702 struct sit_info *sit_i = SIT_I(sbi);
3703 struct curseg_info *curseg;
3704 unsigned int segno, old_cursegno;
3705 struct seg_entry *se;
3707 unsigned short old_blkoff;
3708 unsigned char old_alloc_type;
3710 segno = GET_SEGNO(sbi, new_blkaddr);
3711 se = get_seg_entry(sbi, segno);
3714 f2fs_down_write(&SM_I(sbi)->curseg_lock);
3716 if (!recover_curseg) {
3717 /* for recovery flow */
3718 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3719 if (old_blkaddr == NULL_ADDR)
3720 type = CURSEG_COLD_DATA;
3722 type = CURSEG_WARM_DATA;
3725 if (IS_CURSEG(sbi, segno)) {
3726 /* se->type is volatile as SSR allocation */
3727 type = __f2fs_get_curseg(sbi, segno);
3728 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3730 type = CURSEG_WARM_DATA;
3734 f2fs_bug_on(sbi, !IS_DATASEG(type));
3735 curseg = CURSEG_I(sbi, type);
3737 mutex_lock(&curseg->curseg_mutex);
3738 down_write(&sit_i->sentry_lock);
3740 old_cursegno = curseg->segno;
3741 old_blkoff = curseg->next_blkoff;
3742 old_alloc_type = curseg->alloc_type;
3744 /* change the current segment */
3745 if (segno != curseg->segno) {
3746 curseg->next_segno = segno;
3747 change_curseg(sbi, type);
3750 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3751 curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3753 if (!recover_curseg || recover_newaddr) {
3755 update_segment_mtime(sbi, new_blkaddr, 0);
3756 update_sit_entry(sbi, new_blkaddr, 1);
3758 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3759 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
3761 update_segment_mtime(sbi, old_blkaddr, 0);
3762 update_sit_entry(sbi, old_blkaddr, -1);
3765 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3766 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3768 locate_dirty_segment(sbi, old_cursegno);
3770 if (recover_curseg) {
3771 if (old_cursegno != curseg->segno) {
3772 curseg->next_segno = old_cursegno;
3773 change_curseg(sbi, type);
3775 curseg->next_blkoff = old_blkoff;
3776 curseg->alloc_type = old_alloc_type;
3779 up_write(&sit_i->sentry_lock);
3780 mutex_unlock(&curseg->curseg_mutex);
3781 f2fs_up_write(&SM_I(sbi)->curseg_lock);
3784 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3785 block_t old_addr, block_t new_addr,
3786 unsigned char version, bool recover_curseg,
3787 bool recover_newaddr)
3789 struct f2fs_summary sum;
3791 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3793 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3794 recover_curseg, recover_newaddr, false);
3796 f2fs_update_data_blkaddr(dn, new_addr);
3799 void f2fs_wait_on_page_writeback(struct page *page,
3800 enum page_type type, bool ordered, bool locked)
3802 if (PageWriteback(page)) {
3803 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3805 /* submit cached LFS IO */
3806 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3807 /* submit cached IPU IO */
3808 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3810 wait_on_page_writeback(page);
3811 f2fs_bug_on(sbi, locked && PageWriteback(page));
3813 wait_for_stable_page(page);
3818 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3820 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3823 if (!f2fs_post_read_required(inode))
3826 if (!__is_valid_data_blkaddr(blkaddr))
3829 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3831 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3832 f2fs_put_page(cpage, 1);
3836 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3839 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3842 if (!f2fs_post_read_required(inode))
3845 for (i = 0; i < len; i++)
3846 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3848 f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
3851 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3853 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3854 struct curseg_info *seg_i;
3855 unsigned char *kaddr;
3860 start = start_sum_block(sbi);
3862 page = f2fs_get_meta_page(sbi, start++);
3864 return PTR_ERR(page);
3865 kaddr = (unsigned char *)page_address(page);
3867 /* Step 1: restore nat cache */
3868 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3869 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3871 /* Step 2: restore sit cache */
3872 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3873 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3874 offset = 2 * SUM_JOURNAL_SIZE;
3876 /* Step 3: restore summary entries */
3877 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3878 unsigned short blk_off;
3881 seg_i = CURSEG_I(sbi, i);
3882 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3883 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3884 seg_i->next_segno = segno;
3885 reset_curseg(sbi, i, 0);
3886 seg_i->alloc_type = ckpt->alloc_type[i];
3887 seg_i->next_blkoff = blk_off;
3889 if (seg_i->alloc_type == SSR)
3890 blk_off = sbi->blocks_per_seg;
3892 for (j = 0; j < blk_off; j++) {
3893 struct f2fs_summary *s;
3895 s = (struct f2fs_summary *)(kaddr + offset);
3896 seg_i->sum_blk->entries[j] = *s;
3897 offset += SUMMARY_SIZE;
3898 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3902 f2fs_put_page(page, 1);
3905 page = f2fs_get_meta_page(sbi, start++);
3907 return PTR_ERR(page);
3908 kaddr = (unsigned char *)page_address(page);
3912 f2fs_put_page(page, 1);
3916 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3918 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3919 struct f2fs_summary_block *sum;
3920 struct curseg_info *curseg;
3922 unsigned short blk_off;
3923 unsigned int segno = 0;
3924 block_t blk_addr = 0;
3927 /* get segment number and block addr */
3928 if (IS_DATASEG(type)) {
3929 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3930 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3932 if (__exist_node_summaries(sbi))
3933 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3935 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3937 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3939 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3941 if (__exist_node_summaries(sbi))
3942 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3943 type - CURSEG_HOT_NODE);
3945 blk_addr = GET_SUM_BLOCK(sbi, segno);
3948 new = f2fs_get_meta_page(sbi, blk_addr);
3950 return PTR_ERR(new);
3951 sum = (struct f2fs_summary_block *)page_address(new);
3953 if (IS_NODESEG(type)) {
3954 if (__exist_node_summaries(sbi)) {
3955 struct f2fs_summary *ns = &sum->entries[0];
3958 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3960 ns->ofs_in_node = 0;
3963 err = f2fs_restore_node_summary(sbi, segno, sum);
3969 /* set uncompleted segment to curseg */
3970 curseg = CURSEG_I(sbi, type);
3971 mutex_lock(&curseg->curseg_mutex);
3973 /* update journal info */
3974 down_write(&curseg->journal_rwsem);
3975 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3976 up_write(&curseg->journal_rwsem);
3978 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3979 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3980 curseg->next_segno = segno;
3981 reset_curseg(sbi, type, 0);
3982 curseg->alloc_type = ckpt->alloc_type[type];
3983 curseg->next_blkoff = blk_off;
3984 mutex_unlock(&curseg->curseg_mutex);
3986 f2fs_put_page(new, 1);
3990 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3992 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3993 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3994 int type = CURSEG_HOT_DATA;
3997 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3998 int npages = f2fs_npages_for_summary_flush(sbi, true);
4001 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4004 /* restore for compacted data summary */
4005 err = read_compacted_summaries(sbi);
4008 type = CURSEG_HOT_NODE;
4011 if (__exist_node_summaries(sbi))
4012 f2fs_ra_meta_pages(sbi,
4013 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4014 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4016 for (; type <= CURSEG_COLD_NODE; type++) {
4017 err = read_normal_summaries(sbi, type);
4022 /* sanity check for summary blocks */
4023 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4024 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4025 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4026 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4033 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4036 unsigned char *kaddr;
4037 struct f2fs_summary *summary;
4038 struct curseg_info *seg_i;
4039 int written_size = 0;
4042 page = f2fs_grab_meta_page(sbi, blkaddr++);
4043 kaddr = (unsigned char *)page_address(page);
4044 memset(kaddr, 0, PAGE_SIZE);
4046 /* Step 1: write nat cache */
4047 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4048 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4049 written_size += SUM_JOURNAL_SIZE;
4051 /* Step 2: write sit cache */
4052 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4053 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4054 written_size += SUM_JOURNAL_SIZE;
4056 /* Step 3: write summary entries */
4057 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4058 seg_i = CURSEG_I(sbi, i);
4059 for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4061 page = f2fs_grab_meta_page(sbi, blkaddr++);
4062 kaddr = (unsigned char *)page_address(page);
4063 memset(kaddr, 0, PAGE_SIZE);
4066 summary = (struct f2fs_summary *)(kaddr + written_size);
4067 *summary = seg_i->sum_blk->entries[j];
4068 written_size += SUMMARY_SIZE;
4070 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4074 set_page_dirty(page);
4075 f2fs_put_page(page, 1);
4080 set_page_dirty(page);
4081 f2fs_put_page(page, 1);
4085 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4086 block_t blkaddr, int type)
4090 if (IS_DATASEG(type))
4091 end = type + NR_CURSEG_DATA_TYPE;
4093 end = type + NR_CURSEG_NODE_TYPE;
4095 for (i = type; i < end; i++)
4096 write_current_sum_page(sbi, i, blkaddr + (i - type));
4099 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4101 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4102 write_compacted_summaries(sbi, start_blk);
4104 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4107 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4109 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4112 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4113 unsigned int val, int alloc)
4117 if (type == NAT_JOURNAL) {
4118 for (i = 0; i < nats_in_cursum(journal); i++) {
4119 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4122 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4123 return update_nats_in_cursum(journal, 1);
4124 } else if (type == SIT_JOURNAL) {
4125 for (i = 0; i < sits_in_cursum(journal); i++)
4126 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4128 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4129 return update_sits_in_cursum(journal, 1);
4134 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4137 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4140 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4143 struct sit_info *sit_i = SIT_I(sbi);
4145 pgoff_t src_off, dst_off;
4147 src_off = current_sit_addr(sbi, start);
4148 dst_off = next_sit_addr(sbi, src_off);
4150 page = f2fs_grab_meta_page(sbi, dst_off);
4151 seg_info_to_sit_page(sbi, page, start);
4153 set_page_dirty(page);
4154 set_to_next_sit(sit_i, start);
4159 static struct sit_entry_set *grab_sit_entry_set(void)
4161 struct sit_entry_set *ses =
4162 f2fs_kmem_cache_alloc(sit_entry_set_slab,
4163 GFP_NOFS, true, NULL);
4166 INIT_LIST_HEAD(&ses->set_list);
4170 static void release_sit_entry_set(struct sit_entry_set *ses)
4172 list_del(&ses->set_list);
4173 kmem_cache_free(sit_entry_set_slab, ses);
4176 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4177 struct list_head *head)
4179 struct sit_entry_set *next = ses;
4181 if (list_is_last(&ses->set_list, head))
4184 list_for_each_entry_continue(next, head, set_list)
4185 if (ses->entry_cnt <= next->entry_cnt) {
4186 list_move_tail(&ses->set_list, &next->set_list);
4190 list_move_tail(&ses->set_list, head);
4193 static void add_sit_entry(unsigned int segno, struct list_head *head)
4195 struct sit_entry_set *ses;
4196 unsigned int start_segno = START_SEGNO(segno);
4198 list_for_each_entry(ses, head, set_list) {
4199 if (ses->start_segno == start_segno) {
4201 adjust_sit_entry_set(ses, head);
4206 ses = grab_sit_entry_set();
4208 ses->start_segno = start_segno;
4210 list_add(&ses->set_list, head);
4213 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4215 struct f2fs_sm_info *sm_info = SM_I(sbi);
4216 struct list_head *set_list = &sm_info->sit_entry_set;
4217 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4220 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4221 add_sit_entry(segno, set_list);
4224 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4226 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4227 struct f2fs_journal *journal = curseg->journal;
4230 down_write(&curseg->journal_rwsem);
4231 for (i = 0; i < sits_in_cursum(journal); i++) {
4235 segno = le32_to_cpu(segno_in_journal(journal, i));
4236 dirtied = __mark_sit_entry_dirty(sbi, segno);
4239 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4241 update_sits_in_cursum(journal, -i);
4242 up_write(&curseg->journal_rwsem);
4246 * CP calls this function, which flushes SIT entries including sit_journal,
4247 * and moves prefree segs to free segs.
4249 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4251 struct sit_info *sit_i = SIT_I(sbi);
4252 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4253 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4254 struct f2fs_journal *journal = curseg->journal;
4255 struct sit_entry_set *ses, *tmp;
4256 struct list_head *head = &SM_I(sbi)->sit_entry_set;
4257 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4258 struct seg_entry *se;
4260 down_write(&sit_i->sentry_lock);
4262 if (!sit_i->dirty_sentries)
4266 * add and account sit entries of dirty bitmap in sit entry
4269 add_sits_in_set(sbi);
4272 * if there are no enough space in journal to store dirty sit
4273 * entries, remove all entries from journal and add and account
4274 * them in sit entry set.
4276 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4278 remove_sits_in_journal(sbi);
4281 * there are two steps to flush sit entries:
4282 * #1, flush sit entries to journal in current cold data summary block.
4283 * #2, flush sit entries to sit page.
4285 list_for_each_entry_safe(ses, tmp, head, set_list) {
4286 struct page *page = NULL;
4287 struct f2fs_sit_block *raw_sit = NULL;
4288 unsigned int start_segno = ses->start_segno;
4289 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4290 (unsigned long)MAIN_SEGS(sbi));
4291 unsigned int segno = start_segno;
4294 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4298 down_write(&curseg->journal_rwsem);
4300 page = get_next_sit_page(sbi, start_segno);
4301 raw_sit = page_address(page);
4304 /* flush dirty sit entries in region of current sit set */
4305 for_each_set_bit_from(segno, bitmap, end) {
4306 int offset, sit_offset;
4308 se = get_seg_entry(sbi, segno);
4309 #ifdef CONFIG_F2FS_CHECK_FS
4310 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4311 SIT_VBLOCK_MAP_SIZE))
4312 f2fs_bug_on(sbi, 1);
4315 /* add discard candidates */
4316 if (!(cpc->reason & CP_DISCARD)) {
4317 cpc->trim_start = segno;
4318 add_discard_addrs(sbi, cpc, false);
4322 offset = f2fs_lookup_journal_in_cursum(journal,
4323 SIT_JOURNAL, segno, 1);
4324 f2fs_bug_on(sbi, offset < 0);
4325 segno_in_journal(journal, offset) =
4327 seg_info_to_raw_sit(se,
4328 &sit_in_journal(journal, offset));
4329 check_block_count(sbi, segno,
4330 &sit_in_journal(journal, offset));
4332 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4333 seg_info_to_raw_sit(se,
4334 &raw_sit->entries[sit_offset]);
4335 check_block_count(sbi, segno,
4336 &raw_sit->entries[sit_offset]);
4339 __clear_bit(segno, bitmap);
4340 sit_i->dirty_sentries--;
4345 up_write(&curseg->journal_rwsem);
4347 f2fs_put_page(page, 1);
4349 f2fs_bug_on(sbi, ses->entry_cnt);
4350 release_sit_entry_set(ses);
4353 f2fs_bug_on(sbi, !list_empty(head));
4354 f2fs_bug_on(sbi, sit_i->dirty_sentries);
4356 if (cpc->reason & CP_DISCARD) {
4357 __u64 trim_start = cpc->trim_start;
4359 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4360 add_discard_addrs(sbi, cpc, false);
4362 cpc->trim_start = trim_start;
4364 up_write(&sit_i->sentry_lock);
4366 set_prefree_as_free_segments(sbi);
4369 static int build_sit_info(struct f2fs_sb_info *sbi)
4371 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4372 struct sit_info *sit_i;
4373 unsigned int sit_segs, start;
4374 char *src_bitmap, *bitmap;
4375 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4376 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4378 /* allocate memory for SIT information */
4379 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4383 SM_I(sbi)->sit_info = sit_i;
4386 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4389 if (!sit_i->sentries)
4392 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4393 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4395 if (!sit_i->dirty_sentries_bitmap)
4398 #ifdef CONFIG_F2FS_CHECK_FS
4399 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4401 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4403 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4407 bitmap = sit_i->bitmap;
4409 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4410 sit_i->sentries[start].cur_valid_map = bitmap;
4411 bitmap += SIT_VBLOCK_MAP_SIZE;
4413 sit_i->sentries[start].ckpt_valid_map = bitmap;
4414 bitmap += SIT_VBLOCK_MAP_SIZE;
4416 #ifdef CONFIG_F2FS_CHECK_FS
4417 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4418 bitmap += SIT_VBLOCK_MAP_SIZE;
4422 sit_i->sentries[start].discard_map = bitmap;
4423 bitmap += SIT_VBLOCK_MAP_SIZE;
4427 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4428 if (!sit_i->tmp_map)
4431 if (__is_large_section(sbi)) {
4432 sit_i->sec_entries =
4433 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4436 if (!sit_i->sec_entries)
4440 /* get information related with SIT */
4441 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4443 /* setup SIT bitmap from ckeckpoint pack */
4444 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4445 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4447 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4448 if (!sit_i->sit_bitmap)
4451 #ifdef CONFIG_F2FS_CHECK_FS
4452 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4453 sit_bitmap_size, GFP_KERNEL);
4454 if (!sit_i->sit_bitmap_mir)
4457 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4458 main_bitmap_size, GFP_KERNEL);
4459 if (!sit_i->invalid_segmap)
4463 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4464 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4465 sit_i->written_valid_blocks = 0;
4466 sit_i->bitmap_size = sit_bitmap_size;
4467 sit_i->dirty_sentries = 0;
4468 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4469 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4470 sit_i->mounted_time = ktime_get_boottime_seconds();
4471 init_rwsem(&sit_i->sentry_lock);
4475 static int build_free_segmap(struct f2fs_sb_info *sbi)
4477 struct free_segmap_info *free_i;
4478 unsigned int bitmap_size, sec_bitmap_size;
4480 /* allocate memory for free segmap information */
4481 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4485 SM_I(sbi)->free_info = free_i;
4487 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4488 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4489 if (!free_i->free_segmap)
4492 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4493 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4494 if (!free_i->free_secmap)
4497 /* set all segments as dirty temporarily */
4498 memset(free_i->free_segmap, 0xff, bitmap_size);
4499 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4501 /* init free segmap information */
4502 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4503 free_i->free_segments = 0;
4504 free_i->free_sections = 0;
4505 spin_lock_init(&free_i->segmap_lock);
4509 static int build_curseg(struct f2fs_sb_info *sbi)
4511 struct curseg_info *array;
4514 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4515 sizeof(*array)), GFP_KERNEL);
4519 SM_I(sbi)->curseg_array = array;
4521 for (i = 0; i < NO_CHECK_TYPE; i++) {
4522 mutex_init(&array[i].curseg_mutex);
4523 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4524 if (!array[i].sum_blk)
4526 init_rwsem(&array[i].journal_rwsem);
4527 array[i].journal = f2fs_kzalloc(sbi,
4528 sizeof(struct f2fs_journal), GFP_KERNEL);
4529 if (!array[i].journal)
4531 if (i < NR_PERSISTENT_LOG)
4532 array[i].seg_type = CURSEG_HOT_DATA + i;
4533 else if (i == CURSEG_COLD_DATA_PINNED)
4534 array[i].seg_type = CURSEG_COLD_DATA;
4535 else if (i == CURSEG_ALL_DATA_ATGC)
4536 array[i].seg_type = CURSEG_COLD_DATA;
4537 array[i].segno = NULL_SEGNO;
4538 array[i].next_blkoff = 0;
4539 array[i].inited = false;
4541 return restore_curseg_summaries(sbi);
4544 static int build_sit_entries(struct f2fs_sb_info *sbi)
4546 struct sit_info *sit_i = SIT_I(sbi);
4547 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4548 struct f2fs_journal *journal = curseg->journal;
4549 struct seg_entry *se;
4550 struct f2fs_sit_entry sit;
4551 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4552 unsigned int i, start, end;
4553 unsigned int readed, start_blk = 0;
4555 block_t sit_valid_blocks[2] = {0, 0};
4558 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4561 start = start_blk * sit_i->sents_per_block;
4562 end = (start_blk + readed) * sit_i->sents_per_block;
4564 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4565 struct f2fs_sit_block *sit_blk;
4568 se = &sit_i->sentries[start];
4569 page = get_current_sit_page(sbi, start);
4571 return PTR_ERR(page);
4572 sit_blk = (struct f2fs_sit_block *)page_address(page);
4573 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4574 f2fs_put_page(page, 1);
4576 err = check_block_count(sbi, start, &sit);
4579 seg_info_from_raw_sit(se, &sit);
4581 if (se->type >= NR_PERSISTENT_LOG) {
4582 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4584 f2fs_handle_error(sbi,
4585 ERROR_INCONSISTENT_SUM_TYPE);
4586 return -EFSCORRUPTED;
4589 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4591 if (f2fs_block_unit_discard(sbi)) {
4592 /* build discard map only one time */
4593 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4594 memset(se->discard_map, 0xff,
4595 SIT_VBLOCK_MAP_SIZE);
4597 memcpy(se->discard_map,
4599 SIT_VBLOCK_MAP_SIZE);
4600 sbi->discard_blks +=
4601 sbi->blocks_per_seg -
4606 if (__is_large_section(sbi))
4607 get_sec_entry(sbi, start)->valid_blocks +=
4610 start_blk += readed;
4611 } while (start_blk < sit_blk_cnt);
4613 down_read(&curseg->journal_rwsem);
4614 for (i = 0; i < sits_in_cursum(journal); i++) {
4615 unsigned int old_valid_blocks;
4617 start = le32_to_cpu(segno_in_journal(journal, i));
4618 if (start >= MAIN_SEGS(sbi)) {
4619 f2fs_err(sbi, "Wrong journal entry on segno %u",
4621 err = -EFSCORRUPTED;
4622 f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4626 se = &sit_i->sentries[start];
4627 sit = sit_in_journal(journal, i);
4629 old_valid_blocks = se->valid_blocks;
4631 sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4633 err = check_block_count(sbi, start, &sit);
4636 seg_info_from_raw_sit(se, &sit);
4638 if (se->type >= NR_PERSISTENT_LOG) {
4639 f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4641 err = -EFSCORRUPTED;
4642 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4646 sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4648 if (f2fs_block_unit_discard(sbi)) {
4649 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4650 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4652 memcpy(se->discard_map, se->cur_valid_map,
4653 SIT_VBLOCK_MAP_SIZE);
4654 sbi->discard_blks += old_valid_blocks;
4655 sbi->discard_blks -= se->valid_blocks;
4659 if (__is_large_section(sbi)) {
4660 get_sec_entry(sbi, start)->valid_blocks +=
4662 get_sec_entry(sbi, start)->valid_blocks -=
4666 up_read(&curseg->journal_rwsem);
4671 if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4672 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4673 sit_valid_blocks[NODE], valid_node_count(sbi));
4674 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4675 return -EFSCORRUPTED;
4678 if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4679 valid_user_blocks(sbi)) {
4680 f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4681 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4682 valid_user_blocks(sbi));
4683 f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4684 return -EFSCORRUPTED;
4690 static void init_free_segmap(struct f2fs_sb_info *sbi)
4694 struct seg_entry *sentry;
4696 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4697 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4699 sentry = get_seg_entry(sbi, start);
4700 if (!sentry->valid_blocks)
4701 __set_free(sbi, start);
4703 SIT_I(sbi)->written_valid_blocks +=
4704 sentry->valid_blocks;
4707 /* set use the current segments */
4708 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4709 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4711 __set_test_and_inuse(sbi, curseg_t->segno);
4715 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4717 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4718 struct free_segmap_info *free_i = FREE_I(sbi);
4719 unsigned int segno = 0, offset = 0, secno;
4720 block_t valid_blocks, usable_blks_in_seg;
4723 /* find dirty segment based on free segmap */
4724 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4725 if (segno >= MAIN_SEGS(sbi))
4728 valid_blocks = get_valid_blocks(sbi, segno, false);
4729 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4730 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4732 if (valid_blocks > usable_blks_in_seg) {
4733 f2fs_bug_on(sbi, 1);
4736 mutex_lock(&dirty_i->seglist_lock);
4737 __locate_dirty_segment(sbi, segno, DIRTY);
4738 mutex_unlock(&dirty_i->seglist_lock);
4741 if (!__is_large_section(sbi))
4744 mutex_lock(&dirty_i->seglist_lock);
4745 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4746 valid_blocks = get_valid_blocks(sbi, segno, true);
4747 secno = GET_SEC_FROM_SEG(sbi, segno);
4749 if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4751 if (IS_CURSEC(sbi, secno))
4753 set_bit(secno, dirty_i->dirty_secmap);
4755 mutex_unlock(&dirty_i->seglist_lock);
4758 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4760 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4761 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4763 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4764 if (!dirty_i->victim_secmap)
4767 dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4768 if (!dirty_i->pinned_secmap)
4771 dirty_i->pinned_secmap_cnt = 0;
4772 dirty_i->enable_pin_section = true;
4776 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4778 struct dirty_seglist_info *dirty_i;
4779 unsigned int bitmap_size, i;
4781 /* allocate memory for dirty segments list information */
4782 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4787 SM_I(sbi)->dirty_info = dirty_i;
4788 mutex_init(&dirty_i->seglist_lock);
4790 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4792 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4793 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4795 if (!dirty_i->dirty_segmap[i])
4799 if (__is_large_section(sbi)) {
4800 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4801 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4802 bitmap_size, GFP_KERNEL);
4803 if (!dirty_i->dirty_secmap)
4807 init_dirty_segmap(sbi);
4808 return init_victim_secmap(sbi);
4811 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4816 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4817 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4819 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4820 struct curseg_info *curseg = CURSEG_I(sbi, i);
4821 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4822 unsigned int blkofs = curseg->next_blkoff;
4824 if (f2fs_sb_has_readonly(sbi) &&
4825 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4828 sanity_check_seg_type(sbi, curseg->seg_type);
4830 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4832 "Current segment has invalid alloc_type:%d",
4833 curseg->alloc_type);
4834 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4835 return -EFSCORRUPTED;
4838 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4841 if (curseg->alloc_type == SSR)
4844 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4845 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4849 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4850 i, curseg->segno, curseg->alloc_type,
4851 curseg->next_blkoff, blkofs);
4852 f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4853 return -EFSCORRUPTED;
4859 #ifdef CONFIG_BLK_DEV_ZONED
4861 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4862 struct f2fs_dev_info *fdev,
4863 struct blk_zone *zone)
4865 unsigned int zone_segno;
4866 block_t zone_block, valid_block_cnt;
4867 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4870 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4873 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4874 zone_segno = GET_SEGNO(sbi, zone_block);
4877 * Skip check of zones cursegs point to, since
4878 * fix_curseg_write_pointer() checks them.
4880 if (zone_segno >= MAIN_SEGS(sbi) ||
4881 IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno)))
4885 * Get # of valid block of the zone.
4887 valid_block_cnt = get_valid_blocks(sbi, zone_segno, true);
4889 if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) ||
4890 (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL))
4893 if (!valid_block_cnt) {
4894 f2fs_notice(sbi, "Zone without valid block has non-zero write "
4895 "pointer. Reset the write pointer: cond[0x%x]",
4897 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4898 zone->len >> log_sectors_per_block);
4900 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4906 * If there are valid blocks and the write pointer doesn't match
4907 * with them, we need to report the inconsistency and fill
4908 * the zone till the end to close the zone. This inconsistency
4909 * does not cause write error because the zone will not be
4910 * selected for write operation until it get discarded.
4912 f2fs_notice(sbi, "Valid blocks are not aligned with write "
4913 "pointer: valid block[0x%x,0x%x] cond[0x%x]",
4914 zone_segno, valid_block_cnt, zone->cond);
4916 ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
4917 zone->start, zone->len, GFP_NOFS);
4918 if (ret == -EOPNOTSUPP) {
4919 ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
4920 zone->len - (zone->wp - zone->start),
4923 f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
4926 f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
4933 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4934 block_t zone_blkaddr)
4938 for (i = 0; i < sbi->s_ndevs; i++) {
4939 if (!bdev_is_zoned(FDEV(i).bdev))
4941 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4942 zone_blkaddr <= FDEV(i).end_blk))
4949 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4952 memcpy(data, zone, sizeof(struct blk_zone));
4956 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4958 struct curseg_info *cs = CURSEG_I(sbi, type);
4959 struct f2fs_dev_info *zbd;
4960 struct blk_zone zone;
4961 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4962 block_t cs_zone_block, wp_block;
4963 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4964 sector_t zone_sector;
4967 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4968 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4970 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4974 /* report zone for the sector the curseg points to */
4975 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4976 << log_sectors_per_block;
4977 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4978 report_one_zone_cb, &zone);
4980 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4985 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4989 * When safely unmounted in the previous mount, we could use current
4990 * segments. Otherwise, allocate new sections.
4992 if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) {
4993 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4994 wp_segno = GET_SEGNO(sbi, wp_block);
4995 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4996 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4998 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5002 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5003 "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno,
5004 cs->next_blkoff, wp_segno, wp_blkoff);
5007 /* Allocate a new section if it's not new. */
5008 if (cs->next_blkoff) {
5009 unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff;
5011 f2fs_allocate_new_section(sbi, type, true);
5012 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5013 "[0x%x,0x%x] -> [0x%x,0x%x]",
5014 type, old_segno, old_blkoff,
5015 cs->segno, cs->next_blkoff);
5018 /* check consistency of the zone curseg pointed to */
5019 if (check_zone_write_pointer(sbi, zbd, &zone))
5022 /* check newly assigned zone */
5023 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5024 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5026 zbd = get_target_zoned_dev(sbi, cs_zone_block);
5030 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5031 << log_sectors_per_block;
5032 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5033 report_one_zone_cb, &zone);
5035 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5040 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5043 if (zone.wp != zone.start) {
5045 "New zone for curseg[%d] is not yet discarded. "
5046 "Reset the zone: curseg[0x%x,0x%x]",
5047 type, cs->segno, cs->next_blkoff);
5048 err = __f2fs_issue_discard_zone(sbi, zbd->bdev, cs_zone_block,
5049 zone.len >> log_sectors_per_block);
5051 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5060 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5064 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5065 ret = fix_curseg_write_pointer(sbi, i);
5073 struct check_zone_write_pointer_args {
5074 struct f2fs_sb_info *sbi;
5075 struct f2fs_dev_info *fdev;
5078 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5081 struct check_zone_write_pointer_args *args;
5083 args = (struct check_zone_write_pointer_args *)data;
5085 return check_zone_write_pointer(args->sbi, args->fdev, zone);
5088 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5091 struct check_zone_write_pointer_args args;
5093 for (i = 0; i < sbi->s_ndevs; i++) {
5094 if (!bdev_is_zoned(FDEV(i).bdev))
5098 args.fdev = &FDEV(i);
5099 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5100 check_zone_write_pointer_cb, &args);
5109 * Return the number of usable blocks in a segment. The number of blocks
5110 * returned is always equal to the number of blocks in a segment for
5111 * segments fully contained within a sequential zone capacity or a
5112 * conventional zone. For segments partially contained in a sequential
5113 * zone capacity, the number of usable blocks up to the zone capacity
5114 * is returned. 0 is returned in all other cases.
5116 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5117 struct f2fs_sb_info *sbi, unsigned int segno)
5119 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5122 if (!sbi->unusable_blocks_per_sec)
5123 return sbi->blocks_per_seg;
5125 secno = GET_SEC_FROM_SEG(sbi, segno);
5126 seg_start = START_BLOCK(sbi, segno);
5127 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5128 sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5131 * If segment starts before zone capacity and spans beyond
5132 * zone capacity, then usable blocks are from seg start to
5133 * zone capacity. If the segment starts after the zone capacity,
5134 * then there are no usable blocks.
5136 if (seg_start >= sec_cap_blkaddr)
5138 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5139 return sec_cap_blkaddr - seg_start;
5141 return sbi->blocks_per_seg;
5144 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5149 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5154 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5161 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5164 if (f2fs_sb_has_blkzoned(sbi))
5165 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5167 return sbi->blocks_per_seg;
5170 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5173 if (f2fs_sb_has_blkzoned(sbi))
5174 return CAP_SEGS_PER_SEC(sbi);
5176 return sbi->segs_per_sec;
5180 * Update min, max modified time for cost-benefit GC algorithm
5182 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5184 struct sit_info *sit_i = SIT_I(sbi);
5187 down_write(&sit_i->sentry_lock);
5189 sit_i->min_mtime = ULLONG_MAX;
5191 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5193 unsigned long long mtime = 0;
5195 for (i = 0; i < sbi->segs_per_sec; i++)
5196 mtime += get_seg_entry(sbi, segno + i)->mtime;
5198 mtime = div_u64(mtime, sbi->segs_per_sec);
5200 if (sit_i->min_mtime > mtime)
5201 sit_i->min_mtime = mtime;
5203 sit_i->max_mtime = get_mtime(sbi, false);
5204 sit_i->dirty_max_mtime = 0;
5205 up_write(&sit_i->sentry_lock);
5208 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5210 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5211 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5212 struct f2fs_sm_info *sm_info;
5215 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5220 sbi->sm_info = sm_info;
5221 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5222 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5223 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5224 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5225 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5226 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5227 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5228 sm_info->rec_prefree_segments = sm_info->main_segments *
5229 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5230 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5231 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5233 if (!f2fs_lfs_mode(sbi))
5234 sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5235 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5236 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5237 sm_info->min_seq_blocks = sbi->blocks_per_seg;
5238 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5239 sm_info->min_ssr_sections = reserved_sections(sbi);
5241 INIT_LIST_HEAD(&sm_info->sit_entry_set);
5243 init_f2fs_rwsem(&sm_info->curseg_lock);
5245 err = f2fs_create_flush_cmd_control(sbi);
5249 err = create_discard_cmd_control(sbi);
5253 err = build_sit_info(sbi);
5256 err = build_free_segmap(sbi);
5259 err = build_curseg(sbi);
5263 /* reinit free segmap based on SIT */
5264 err = build_sit_entries(sbi);
5268 init_free_segmap(sbi);
5269 err = build_dirty_segmap(sbi);
5273 err = sanity_check_curseg(sbi);
5277 init_min_max_mtime(sbi);
5281 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5282 enum dirty_type dirty_type)
5284 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5286 mutex_lock(&dirty_i->seglist_lock);
5287 kvfree(dirty_i->dirty_segmap[dirty_type]);
5288 dirty_i->nr_dirty[dirty_type] = 0;
5289 mutex_unlock(&dirty_i->seglist_lock);
5292 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5294 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5296 kvfree(dirty_i->pinned_secmap);
5297 kvfree(dirty_i->victim_secmap);
5300 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5302 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5308 /* discard pre-free/dirty segments list */
5309 for (i = 0; i < NR_DIRTY_TYPE; i++)
5310 discard_dirty_segmap(sbi, i);
5312 if (__is_large_section(sbi)) {
5313 mutex_lock(&dirty_i->seglist_lock);
5314 kvfree(dirty_i->dirty_secmap);
5315 mutex_unlock(&dirty_i->seglist_lock);
5318 destroy_victim_secmap(sbi);
5319 SM_I(sbi)->dirty_info = NULL;
5323 static void destroy_curseg(struct f2fs_sb_info *sbi)
5325 struct curseg_info *array = SM_I(sbi)->curseg_array;
5330 SM_I(sbi)->curseg_array = NULL;
5331 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5332 kfree(array[i].sum_blk);
5333 kfree(array[i].journal);
5338 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5340 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5344 SM_I(sbi)->free_info = NULL;
5345 kvfree(free_i->free_segmap);
5346 kvfree(free_i->free_secmap);
5350 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5352 struct sit_info *sit_i = SIT_I(sbi);
5357 if (sit_i->sentries)
5358 kvfree(sit_i->bitmap);
5359 kfree(sit_i->tmp_map);
5361 kvfree(sit_i->sentries);
5362 kvfree(sit_i->sec_entries);
5363 kvfree(sit_i->dirty_sentries_bitmap);
5365 SM_I(sbi)->sit_info = NULL;
5366 kvfree(sit_i->sit_bitmap);
5367 #ifdef CONFIG_F2FS_CHECK_FS
5368 kvfree(sit_i->sit_bitmap_mir);
5369 kvfree(sit_i->invalid_segmap);
5374 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5376 struct f2fs_sm_info *sm_info = SM_I(sbi);
5380 f2fs_destroy_flush_cmd_control(sbi, true);
5381 destroy_discard_cmd_control(sbi);
5382 destroy_dirty_segmap(sbi);
5383 destroy_curseg(sbi);
5384 destroy_free_segmap(sbi);
5385 destroy_sit_info(sbi);
5386 sbi->sm_info = NULL;
5390 int __init f2fs_create_segment_manager_caches(void)
5392 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5393 sizeof(struct discard_entry));
5394 if (!discard_entry_slab)
5397 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5398 sizeof(struct discard_cmd));
5399 if (!discard_cmd_slab)
5400 goto destroy_discard_entry;
5402 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5403 sizeof(struct sit_entry_set));
5404 if (!sit_entry_set_slab)
5405 goto destroy_discard_cmd;
5407 revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5408 sizeof(struct revoke_entry));
5409 if (!revoke_entry_slab)
5410 goto destroy_sit_entry_set;
5413 destroy_sit_entry_set:
5414 kmem_cache_destroy(sit_entry_set_slab);
5415 destroy_discard_cmd:
5416 kmem_cache_destroy(discard_cmd_slab);
5417 destroy_discard_entry:
5418 kmem_cache_destroy(discard_entry_slab);
5423 void f2fs_destroy_segment_manager_caches(void)
5425 kmem_cache_destroy(sit_entry_set_slab);
5426 kmem_cache_destroy(discard_cmd_slab);
5427 kmem_cache_destroy(discard_entry_slab);
5428 kmem_cache_destroy(revoke_entry_slab);