1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched/signal.h>
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *discard_cmd_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
33 static unsigned long __reverse_ulong(unsigned char *str)
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
38 #if BITS_PER_LONG == 64
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
52 static inline unsigned long __reverse_ffs(unsigned long word)
56 #if BITS_PER_LONG == 64
57 if ((word & 0xffffffff00000000UL) == 0)
62 if ((word & 0xffff0000) == 0)
67 if ((word & 0xff00) == 0)
72 if ((word & 0xf0) == 0)
77 if ((word & 0xc) == 0)
82 if ((word & 0x2) == 0)
88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
90 * @size must be integral times of unsigned long.
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
99 const unsigned long *p = addr + BIT_WORD(offset);
100 unsigned long result = size;
106 size -= (offset & ~(BITS_PER_LONG - 1));
107 offset %= BITS_PER_LONG;
113 tmp = __reverse_ulong((unsigned char *)p);
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
121 if (size <= BITS_PER_LONG)
123 size -= BITS_PER_LONG;
129 return result - size + __reverse_ffs(tmp);
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
135 const unsigned long *p = addr + BIT_WORD(offset);
136 unsigned long result = size;
142 size -= (offset & ~(BITS_PER_LONG - 1));
143 offset %= BITS_PER_LONG;
149 tmp = __reverse_ulong((unsigned char *)p);
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
158 if (size <= BITS_PER_LONG)
160 size -= BITS_PER_LONG;
166 return result - size + __reverse_ffz(tmp);
169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
175 if (f2fs_lfs_mode(sbi))
177 if (sbi->gc_mode == GC_URGENT_HIGH)
179 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
182 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
183 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
188 struct inmem_pages *new;
190 f2fs_trace_pid(page);
192 f2fs_set_page_private(page, ATOMIC_WRITTEN_PAGE);
194 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
196 /* add atomic page indices to the list */
198 INIT_LIST_HEAD(&new->list);
200 /* increase reference count with clean state */
202 mutex_lock(&F2FS_I(inode)->inmem_lock);
203 list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
204 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
205 mutex_unlock(&F2FS_I(inode)->inmem_lock);
207 trace_f2fs_register_inmem_page(page, INMEM);
210 static int __revoke_inmem_pages(struct inode *inode,
211 struct list_head *head, bool drop, bool recover,
214 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
215 struct inmem_pages *cur, *tmp;
218 list_for_each_entry_safe(cur, tmp, head, list) {
219 struct page *page = cur->page;
222 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
226 * to avoid deadlock in between page lock and
229 if (!trylock_page(page))
235 f2fs_wait_on_page_writeback(page, DATA, true, true);
238 struct dnode_of_data dn;
241 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
243 set_new_dnode(&dn, inode, NULL, NULL, 0);
244 err = f2fs_get_dnode_of_data(&dn, page->index,
247 if (err == -ENOMEM) {
248 congestion_wait(BLK_RW_ASYNC,
257 err = f2fs_get_node_info(sbi, dn.nid, &ni);
263 if (cur->old_addr == NEW_ADDR) {
264 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
265 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
267 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
268 cur->old_addr, ni.version, true, true);
272 /* we don't need to invalidate this in the sccessful status */
273 if (drop || recover) {
274 ClearPageUptodate(page);
275 clear_cold_data(page);
277 f2fs_clear_page_private(page);
278 f2fs_put_page(page, 1);
280 list_del(&cur->list);
281 kmem_cache_free(inmem_entry_slab, cur);
282 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
287 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
289 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
291 struct f2fs_inode_info *fi;
292 unsigned int count = sbi->atomic_files;
293 unsigned int looped = 0;
295 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
296 if (list_empty(head)) {
297 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
300 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
301 inode = igrab(&fi->vfs_inode);
303 list_move_tail(&fi->inmem_ilist, head);
304 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
308 if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
311 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
312 f2fs_drop_inmem_pages(inode);
316 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
319 if (++looped >= count)
325 void f2fs_drop_inmem_pages(struct inode *inode)
327 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
328 struct f2fs_inode_info *fi = F2FS_I(inode);
330 while (!list_empty(&fi->inmem_pages)) {
331 mutex_lock(&fi->inmem_lock);
332 __revoke_inmem_pages(inode, &fi->inmem_pages,
334 mutex_unlock(&fi->inmem_lock);
337 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
339 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
340 if (!list_empty(&fi->inmem_ilist))
341 list_del_init(&fi->inmem_ilist);
342 if (f2fs_is_atomic_file(inode)) {
343 clear_inode_flag(inode, FI_ATOMIC_FILE);
346 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
349 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
351 struct f2fs_inode_info *fi = F2FS_I(inode);
352 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
353 struct list_head *head = &fi->inmem_pages;
354 struct inmem_pages *cur = NULL;
356 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
358 mutex_lock(&fi->inmem_lock);
359 list_for_each_entry(cur, head, list) {
360 if (cur->page == page)
364 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
365 list_del(&cur->list);
366 mutex_unlock(&fi->inmem_lock);
368 dec_page_count(sbi, F2FS_INMEM_PAGES);
369 kmem_cache_free(inmem_entry_slab, cur);
371 ClearPageUptodate(page);
372 f2fs_clear_page_private(page);
373 f2fs_put_page(page, 0);
375 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
378 static int __f2fs_commit_inmem_pages(struct inode *inode)
380 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
381 struct f2fs_inode_info *fi = F2FS_I(inode);
382 struct inmem_pages *cur, *tmp;
383 struct f2fs_io_info fio = {
388 .op_flags = REQ_SYNC | REQ_PRIO,
389 .io_type = FS_DATA_IO,
391 struct list_head revoke_list;
392 bool submit_bio = false;
395 INIT_LIST_HEAD(&revoke_list);
397 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
398 struct page *page = cur->page;
401 if (page->mapping == inode->i_mapping) {
402 trace_f2fs_commit_inmem_page(page, INMEM);
404 f2fs_wait_on_page_writeback(page, DATA, true, true);
406 set_page_dirty(page);
407 if (clear_page_dirty_for_io(page)) {
408 inode_dec_dirty_pages(inode);
409 f2fs_remove_dirty_inode(inode);
413 fio.old_blkaddr = NULL_ADDR;
414 fio.encrypted_page = NULL;
415 fio.need_lock = LOCK_DONE;
416 err = f2fs_do_write_data_page(&fio);
418 if (err == -ENOMEM) {
419 congestion_wait(BLK_RW_ASYNC,
427 /* record old blkaddr for revoking */
428 cur->old_addr = fio.old_blkaddr;
432 list_move_tail(&cur->list, &revoke_list);
436 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
440 * try to revoke all committed pages, but still we could fail
441 * due to no memory or other reason, if that happened, EAGAIN
442 * will be returned, which means in such case, transaction is
443 * already not integrity, caller should use journal to do the
444 * recovery or rewrite & commit last transaction. For other
445 * error number, revoking was done by filesystem itself.
447 err = __revoke_inmem_pages(inode, &revoke_list,
450 /* drop all uncommitted pages */
451 __revoke_inmem_pages(inode, &fi->inmem_pages,
454 __revoke_inmem_pages(inode, &revoke_list,
455 false, false, false);
461 int f2fs_commit_inmem_pages(struct inode *inode)
463 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
464 struct f2fs_inode_info *fi = F2FS_I(inode);
467 f2fs_balance_fs(sbi, true);
469 down_write(&fi->i_gc_rwsem[WRITE]);
472 set_inode_flag(inode, FI_ATOMIC_COMMIT);
474 mutex_lock(&fi->inmem_lock);
475 err = __f2fs_commit_inmem_pages(inode);
476 mutex_unlock(&fi->inmem_lock);
478 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
481 up_write(&fi->i_gc_rwsem[WRITE]);
487 * This function balances dirty node and dentry pages.
488 * In addition, it controls garbage collection.
490 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
492 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
493 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
494 f2fs_stop_checkpoint(sbi, false);
497 /* balance_fs_bg is able to be pending */
498 if (need && excess_cached_nats(sbi))
499 f2fs_balance_fs_bg(sbi, false);
501 if (!f2fs_is_checkpoint_ready(sbi))
505 * We should do GC or end up with checkpoint, if there are so many dirty
506 * dir/node pages without enough free segments.
508 if (has_not_enough_free_secs(sbi, 0, 0)) {
509 down_write(&sbi->gc_lock);
510 f2fs_gc(sbi, false, false, NULL_SEGNO);
514 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
516 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
519 /* try to shrink extent cache when there is no enough memory */
520 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
521 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
523 /* check the # of cached NAT entries */
524 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
525 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
527 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
528 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
530 f2fs_build_free_nids(sbi, false, false);
532 if (!is_idle(sbi, REQ_TIME) &&
533 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
536 /* checkpoint is the only way to shrink partial cached entries */
537 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
538 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
539 excess_prefree_segs(sbi) ||
540 excess_dirty_nats(sbi) ||
541 excess_dirty_nodes(sbi) ||
542 f2fs_time_over(sbi, CP_TIME)) {
543 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
544 struct blk_plug plug;
546 mutex_lock(&sbi->flush_lock);
548 blk_start_plug(&plug);
549 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
550 blk_finish_plug(&plug);
552 mutex_unlock(&sbi->flush_lock);
554 f2fs_sync_fs(sbi->sb, true);
555 stat_inc_bg_cp_count(sbi->stat_info);
559 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
560 struct block_device *bdev)
565 bio = f2fs_bio_alloc(sbi, 0, false);
569 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
570 bio_set_dev(bio, bdev);
571 ret = submit_bio_wait(bio);
574 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
575 test_opt(sbi, FLUSH_MERGE), ret);
579 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
584 if (!f2fs_is_multi_device(sbi))
585 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
587 for (i = 0; i < sbi->s_ndevs; i++) {
588 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
590 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
597 static int issue_flush_thread(void *data)
599 struct f2fs_sb_info *sbi = data;
600 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
601 wait_queue_head_t *q = &fcc->flush_wait_queue;
603 if (kthread_should_stop())
606 sb_start_intwrite(sbi->sb);
608 if (!llist_empty(&fcc->issue_list)) {
609 struct flush_cmd *cmd, *next;
612 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
613 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
615 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
617 ret = submit_flush_wait(sbi, cmd->ino);
618 atomic_inc(&fcc->issued_flush);
620 llist_for_each_entry_safe(cmd, next,
621 fcc->dispatch_list, llnode) {
623 complete(&cmd->wait);
625 fcc->dispatch_list = NULL;
628 sb_end_intwrite(sbi->sb);
630 wait_event_interruptible(*q,
631 kthread_should_stop() || !llist_empty(&fcc->issue_list));
635 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
637 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
638 struct flush_cmd cmd;
641 if (test_opt(sbi, NOBARRIER))
644 if (!test_opt(sbi, FLUSH_MERGE)) {
645 atomic_inc(&fcc->queued_flush);
646 ret = submit_flush_wait(sbi, ino);
647 atomic_dec(&fcc->queued_flush);
648 atomic_inc(&fcc->issued_flush);
652 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
653 f2fs_is_multi_device(sbi)) {
654 ret = submit_flush_wait(sbi, ino);
655 atomic_dec(&fcc->queued_flush);
657 atomic_inc(&fcc->issued_flush);
662 init_completion(&cmd.wait);
664 llist_add(&cmd.llnode, &fcc->issue_list);
666 /* update issue_list before we wake up issue_flush thread */
669 if (waitqueue_active(&fcc->flush_wait_queue))
670 wake_up(&fcc->flush_wait_queue);
672 if (fcc->f2fs_issue_flush) {
673 wait_for_completion(&cmd.wait);
674 atomic_dec(&fcc->queued_flush);
676 struct llist_node *list;
678 list = llist_del_all(&fcc->issue_list);
680 wait_for_completion(&cmd.wait);
681 atomic_dec(&fcc->queued_flush);
683 struct flush_cmd *tmp, *next;
685 ret = submit_flush_wait(sbi, ino);
687 llist_for_each_entry_safe(tmp, next, list, llnode) {
690 atomic_dec(&fcc->queued_flush);
694 complete(&tmp->wait);
702 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
704 dev_t dev = sbi->sb->s_bdev->bd_dev;
705 struct flush_cmd_control *fcc;
708 if (SM_I(sbi)->fcc_info) {
709 fcc = SM_I(sbi)->fcc_info;
710 if (fcc->f2fs_issue_flush)
715 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
718 atomic_set(&fcc->issued_flush, 0);
719 atomic_set(&fcc->queued_flush, 0);
720 init_waitqueue_head(&fcc->flush_wait_queue);
721 init_llist_head(&fcc->issue_list);
722 SM_I(sbi)->fcc_info = fcc;
723 if (!test_opt(sbi, FLUSH_MERGE))
727 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
728 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
729 if (IS_ERR(fcc->f2fs_issue_flush)) {
730 err = PTR_ERR(fcc->f2fs_issue_flush);
732 SM_I(sbi)->fcc_info = NULL;
739 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
741 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
743 if (fcc && fcc->f2fs_issue_flush) {
744 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
746 fcc->f2fs_issue_flush = NULL;
747 kthread_stop(flush_thread);
751 SM_I(sbi)->fcc_info = NULL;
755 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
759 if (!f2fs_is_multi_device(sbi))
762 for (i = 1; i < sbi->s_ndevs; i++) {
763 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
765 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
769 spin_lock(&sbi->dev_lock);
770 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
771 spin_unlock(&sbi->dev_lock);
777 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
778 enum dirty_type dirty_type)
780 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
782 /* need not be added */
783 if (IS_CURSEG(sbi, segno))
786 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
787 dirty_i->nr_dirty[dirty_type]++;
789 if (dirty_type == DIRTY) {
790 struct seg_entry *sentry = get_seg_entry(sbi, segno);
791 enum dirty_type t = sentry->type;
793 if (unlikely(t >= DIRTY)) {
797 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
798 dirty_i->nr_dirty[t]++;
800 if (__is_large_section(sbi)) {
801 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
802 block_t valid_blocks =
803 get_valid_blocks(sbi, segno, true);
805 f2fs_bug_on(sbi, unlikely(!valid_blocks ||
806 valid_blocks == BLKS_PER_SEC(sbi)));
808 if (!IS_CURSEC(sbi, secno))
809 set_bit(secno, dirty_i->dirty_secmap);
814 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
815 enum dirty_type dirty_type)
817 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
818 block_t valid_blocks;
820 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
821 dirty_i->nr_dirty[dirty_type]--;
823 if (dirty_type == DIRTY) {
824 struct seg_entry *sentry = get_seg_entry(sbi, segno);
825 enum dirty_type t = sentry->type;
827 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
828 dirty_i->nr_dirty[t]--;
830 valid_blocks = get_valid_blocks(sbi, segno, true);
831 if (valid_blocks == 0) {
832 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
833 dirty_i->victim_secmap);
834 #ifdef CONFIG_F2FS_CHECK_FS
835 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
838 if (__is_large_section(sbi)) {
839 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
842 valid_blocks == BLKS_PER_SEC(sbi)) {
843 clear_bit(secno, dirty_i->dirty_secmap);
847 if (!IS_CURSEC(sbi, secno))
848 set_bit(secno, dirty_i->dirty_secmap);
854 * Should not occur error such as -ENOMEM.
855 * Adding dirty entry into seglist is not critical operation.
856 * If a given segment is one of current working segments, it won't be added.
858 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
860 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
861 unsigned short valid_blocks, ckpt_valid_blocks;
862 unsigned int usable_blocks;
864 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
867 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
868 mutex_lock(&dirty_i->seglist_lock);
870 valid_blocks = get_valid_blocks(sbi, segno, false);
871 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
873 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
874 ckpt_valid_blocks == usable_blocks)) {
875 __locate_dirty_segment(sbi, segno, PRE);
876 __remove_dirty_segment(sbi, segno, DIRTY);
877 } else if (valid_blocks < usable_blocks) {
878 __locate_dirty_segment(sbi, segno, DIRTY);
880 /* Recovery routine with SSR needs this */
881 __remove_dirty_segment(sbi, segno, DIRTY);
884 mutex_unlock(&dirty_i->seglist_lock);
887 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
888 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
890 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
893 mutex_lock(&dirty_i->seglist_lock);
894 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
895 if (get_valid_blocks(sbi, segno, false))
897 if (IS_CURSEG(sbi, segno))
899 __locate_dirty_segment(sbi, segno, PRE);
900 __remove_dirty_segment(sbi, segno, DIRTY);
902 mutex_unlock(&dirty_i->seglist_lock);
905 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
908 (overprovision_segments(sbi) - reserved_segments(sbi));
909 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
910 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
911 block_t holes[2] = {0, 0}; /* DATA and NODE */
913 struct seg_entry *se;
916 mutex_lock(&dirty_i->seglist_lock);
917 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
918 se = get_seg_entry(sbi, segno);
919 if (IS_NODESEG(se->type))
920 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
923 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
926 mutex_unlock(&dirty_i->seglist_lock);
928 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
929 if (unusable > ovp_holes)
930 return unusable - ovp_holes;
934 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
937 (overprovision_segments(sbi) - reserved_segments(sbi));
938 if (unusable > F2FS_OPTION(sbi).unusable_cap)
940 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
941 dirty_segments(sbi) > ovp_hole_segs)
946 /* This is only used by SBI_CP_DISABLED */
947 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
949 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
950 unsigned int segno = 0;
952 mutex_lock(&dirty_i->seglist_lock);
953 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
954 if (get_valid_blocks(sbi, segno, false))
956 if (get_ckpt_valid_blocks(sbi, segno))
958 mutex_unlock(&dirty_i->seglist_lock);
961 mutex_unlock(&dirty_i->seglist_lock);
965 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
966 struct block_device *bdev, block_t lstart,
967 block_t start, block_t len)
969 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
970 struct list_head *pend_list;
971 struct discard_cmd *dc;
973 f2fs_bug_on(sbi, !len);
975 pend_list = &dcc->pend_list[plist_idx(len)];
977 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
978 INIT_LIST_HEAD(&dc->list);
987 init_completion(&dc->wait);
988 list_add_tail(&dc->list, pend_list);
989 spin_lock_init(&dc->lock);
991 atomic_inc(&dcc->discard_cmd_cnt);
992 dcc->undiscard_blks += len;
997 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
998 struct block_device *bdev, block_t lstart,
999 block_t start, block_t len,
1000 struct rb_node *parent, struct rb_node **p,
1003 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1004 struct discard_cmd *dc;
1006 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1008 rb_link_node(&dc->rb_node, parent, p);
1009 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1014 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1015 struct discard_cmd *dc)
1017 if (dc->state == D_DONE)
1018 atomic_sub(dc->queued, &dcc->queued_discard);
1020 list_del(&dc->list);
1021 rb_erase_cached(&dc->rb_node, &dcc->root);
1022 dcc->undiscard_blks -= dc->len;
1024 kmem_cache_free(discard_cmd_slab, dc);
1026 atomic_dec(&dcc->discard_cmd_cnt);
1029 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1030 struct discard_cmd *dc)
1032 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1033 unsigned long flags;
1035 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1037 spin_lock_irqsave(&dc->lock, flags);
1039 spin_unlock_irqrestore(&dc->lock, flags);
1042 spin_unlock_irqrestore(&dc->lock, flags);
1044 f2fs_bug_on(sbi, dc->ref);
1046 if (dc->error == -EOPNOTSUPP)
1051 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1052 KERN_INFO, sbi->sb->s_id,
1053 dc->lstart, dc->start, dc->len, dc->error);
1054 __detach_discard_cmd(dcc, dc);
1057 static void f2fs_submit_discard_endio(struct bio *bio)
1059 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1060 unsigned long flags;
1062 spin_lock_irqsave(&dc->lock, flags);
1064 dc->error = blk_status_to_errno(bio->bi_status);
1066 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1068 complete_all(&dc->wait);
1070 spin_unlock_irqrestore(&dc->lock, flags);
1074 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1075 block_t start, block_t end)
1077 #ifdef CONFIG_F2FS_CHECK_FS
1078 struct seg_entry *sentry;
1080 block_t blk = start;
1081 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1085 segno = GET_SEGNO(sbi, blk);
1086 sentry = get_seg_entry(sbi, segno);
1087 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1089 if (end < START_BLOCK(sbi, segno + 1))
1090 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1093 map = (unsigned long *)(sentry->cur_valid_map);
1094 offset = __find_rev_next_bit(map, size, offset);
1095 f2fs_bug_on(sbi, offset != size);
1096 blk = START_BLOCK(sbi, segno + 1);
1101 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1102 struct discard_policy *dpolicy,
1103 int discard_type, unsigned int granularity)
1106 dpolicy->type = discard_type;
1107 dpolicy->sync = true;
1108 dpolicy->ordered = false;
1109 dpolicy->granularity = granularity;
1111 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1112 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1113 dpolicy->timeout = false;
1115 if (discard_type == DPOLICY_BG) {
1116 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1117 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1118 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1119 dpolicy->io_aware = true;
1120 dpolicy->sync = false;
1121 dpolicy->ordered = true;
1122 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1123 dpolicy->granularity = 1;
1124 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1126 } else if (discard_type == DPOLICY_FORCE) {
1127 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1128 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1129 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1130 dpolicy->io_aware = false;
1131 } else if (discard_type == DPOLICY_FSTRIM) {
1132 dpolicy->io_aware = false;
1133 } else if (discard_type == DPOLICY_UMOUNT) {
1134 dpolicy->io_aware = false;
1135 /* we need to issue all to keep CP_TRIMMED_FLAG */
1136 dpolicy->granularity = 1;
1137 dpolicy->timeout = true;
1141 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1142 struct block_device *bdev, block_t lstart,
1143 block_t start, block_t len);
1144 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1145 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1146 struct discard_policy *dpolicy,
1147 struct discard_cmd *dc,
1148 unsigned int *issued)
1150 struct block_device *bdev = dc->bdev;
1151 struct request_queue *q = bdev_get_queue(bdev);
1152 unsigned int max_discard_blocks =
1153 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1154 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1155 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1156 &(dcc->fstrim_list) : &(dcc->wait_list);
1157 int flag = dpolicy->sync ? REQ_SYNC : 0;
1158 block_t lstart, start, len, total_len;
1161 if (dc->state != D_PREP)
1164 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1167 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1169 lstart = dc->lstart;
1176 while (total_len && *issued < dpolicy->max_requests && !err) {
1177 struct bio *bio = NULL;
1178 unsigned long flags;
1181 if (len > max_discard_blocks) {
1182 len = max_discard_blocks;
1187 if (*issued == dpolicy->max_requests)
1192 if (time_to_inject(sbi, FAULT_DISCARD)) {
1193 f2fs_show_injection_info(sbi, FAULT_DISCARD);
1197 err = __blkdev_issue_discard(bdev,
1198 SECTOR_FROM_BLOCK(start),
1199 SECTOR_FROM_BLOCK(len),
1203 spin_lock_irqsave(&dc->lock, flags);
1204 if (dc->state == D_PARTIAL)
1205 dc->state = D_SUBMIT;
1206 spin_unlock_irqrestore(&dc->lock, flags);
1211 f2fs_bug_on(sbi, !bio);
1214 * should keep before submission to avoid D_DONE
1217 spin_lock_irqsave(&dc->lock, flags);
1219 dc->state = D_SUBMIT;
1221 dc->state = D_PARTIAL;
1223 spin_unlock_irqrestore(&dc->lock, flags);
1225 atomic_inc(&dcc->queued_discard);
1227 list_move_tail(&dc->list, wait_list);
1229 /* sanity check on discard range */
1230 __check_sit_bitmap(sbi, lstart, lstart + len);
1232 bio->bi_private = dc;
1233 bio->bi_end_io = f2fs_submit_discard_endio;
1234 bio->bi_opf |= flag;
1237 atomic_inc(&dcc->issued_discard);
1239 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1248 dcc->undiscard_blks -= len;
1249 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1254 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1255 struct block_device *bdev, block_t lstart,
1256 block_t start, block_t len,
1257 struct rb_node **insert_p,
1258 struct rb_node *insert_parent)
1260 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1262 struct rb_node *parent = NULL;
1263 bool leftmost = true;
1265 if (insert_p && insert_parent) {
1266 parent = insert_parent;
1271 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1274 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1278 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1279 struct discard_cmd *dc)
1281 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1284 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1285 struct discard_cmd *dc, block_t blkaddr)
1287 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288 struct discard_info di = dc->di;
1289 bool modified = false;
1291 if (dc->state == D_DONE || dc->len == 1) {
1292 __remove_discard_cmd(sbi, dc);
1296 dcc->undiscard_blks -= di.len;
1298 if (blkaddr > di.lstart) {
1299 dc->len = blkaddr - dc->lstart;
1300 dcc->undiscard_blks += dc->len;
1301 __relocate_discard_cmd(dcc, dc);
1305 if (blkaddr < di.lstart + di.len - 1) {
1307 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1308 di.start + blkaddr + 1 - di.lstart,
1309 di.lstart + di.len - 1 - blkaddr,
1315 dcc->undiscard_blks += dc->len;
1316 __relocate_discard_cmd(dcc, dc);
1321 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1322 struct block_device *bdev, block_t lstart,
1323 block_t start, block_t len)
1325 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1326 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1327 struct discard_cmd *dc;
1328 struct discard_info di = {0};
1329 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1330 struct request_queue *q = bdev_get_queue(bdev);
1331 unsigned int max_discard_blocks =
1332 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1333 block_t end = lstart + len;
1335 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1337 (struct rb_entry **)&prev_dc,
1338 (struct rb_entry **)&next_dc,
1339 &insert_p, &insert_parent, true, NULL);
1345 di.len = next_dc ? next_dc->lstart - lstart : len;
1346 di.len = min(di.len, len);
1351 struct rb_node *node;
1352 bool merged = false;
1353 struct discard_cmd *tdc = NULL;
1356 di.lstart = prev_dc->lstart + prev_dc->len;
1357 if (di.lstart < lstart)
1359 if (di.lstart >= end)
1362 if (!next_dc || next_dc->lstart > end)
1363 di.len = end - di.lstart;
1365 di.len = next_dc->lstart - di.lstart;
1366 di.start = start + di.lstart - lstart;
1372 if (prev_dc && prev_dc->state == D_PREP &&
1373 prev_dc->bdev == bdev &&
1374 __is_discard_back_mergeable(&di, &prev_dc->di,
1375 max_discard_blocks)) {
1376 prev_dc->di.len += di.len;
1377 dcc->undiscard_blks += di.len;
1378 __relocate_discard_cmd(dcc, prev_dc);
1384 if (next_dc && next_dc->state == D_PREP &&
1385 next_dc->bdev == bdev &&
1386 __is_discard_front_mergeable(&di, &next_dc->di,
1387 max_discard_blocks)) {
1388 next_dc->di.lstart = di.lstart;
1389 next_dc->di.len += di.len;
1390 next_dc->di.start = di.start;
1391 dcc->undiscard_blks += di.len;
1392 __relocate_discard_cmd(dcc, next_dc);
1394 __remove_discard_cmd(sbi, tdc);
1399 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1400 di.len, NULL, NULL);
1407 node = rb_next(&prev_dc->rb_node);
1408 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1412 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1413 struct block_device *bdev, block_t blkstart, block_t blklen)
1415 block_t lblkstart = blkstart;
1417 if (!f2fs_bdev_support_discard(bdev))
1420 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1422 if (f2fs_is_multi_device(sbi)) {
1423 int devi = f2fs_target_device_index(sbi, blkstart);
1425 blkstart -= FDEV(devi).start_blk;
1427 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1428 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1429 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1433 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1434 struct discard_policy *dpolicy)
1436 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1437 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1438 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1439 struct discard_cmd *dc;
1440 struct blk_plug plug;
1441 unsigned int pos = dcc->next_pos;
1442 unsigned int issued = 0;
1443 bool io_interrupted = false;
1445 mutex_lock(&dcc->cmd_lock);
1446 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1448 (struct rb_entry **)&prev_dc,
1449 (struct rb_entry **)&next_dc,
1450 &insert_p, &insert_parent, true, NULL);
1454 blk_start_plug(&plug);
1457 struct rb_node *node;
1460 if (dc->state != D_PREP)
1463 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1464 io_interrupted = true;
1468 dcc->next_pos = dc->lstart + dc->len;
1469 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1471 if (issued >= dpolicy->max_requests)
1474 node = rb_next(&dc->rb_node);
1476 __remove_discard_cmd(sbi, dc);
1477 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1480 blk_finish_plug(&plug);
1485 mutex_unlock(&dcc->cmd_lock);
1487 if (!issued && io_interrupted)
1492 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1493 struct discard_policy *dpolicy);
1495 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1496 struct discard_policy *dpolicy)
1498 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1499 struct list_head *pend_list;
1500 struct discard_cmd *dc, *tmp;
1501 struct blk_plug plug;
1503 bool io_interrupted = false;
1505 if (dpolicy->timeout)
1506 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1510 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1511 if (dpolicy->timeout &&
1512 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1515 if (i + 1 < dpolicy->granularity)
1518 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1519 return __issue_discard_cmd_orderly(sbi, dpolicy);
1521 pend_list = &dcc->pend_list[i];
1523 mutex_lock(&dcc->cmd_lock);
1524 if (list_empty(pend_list))
1526 if (unlikely(dcc->rbtree_check))
1527 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1529 blk_start_plug(&plug);
1530 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1531 f2fs_bug_on(sbi, dc->state != D_PREP);
1533 if (dpolicy->timeout &&
1534 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1537 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1538 !is_idle(sbi, DISCARD_TIME)) {
1539 io_interrupted = true;
1543 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1545 if (issued >= dpolicy->max_requests)
1548 blk_finish_plug(&plug);
1550 mutex_unlock(&dcc->cmd_lock);
1552 if (issued >= dpolicy->max_requests || io_interrupted)
1556 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1557 __wait_all_discard_cmd(sbi, dpolicy);
1561 if (!issued && io_interrupted)
1567 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1569 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1570 struct list_head *pend_list;
1571 struct discard_cmd *dc, *tmp;
1573 bool dropped = false;
1575 mutex_lock(&dcc->cmd_lock);
1576 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1577 pend_list = &dcc->pend_list[i];
1578 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1579 f2fs_bug_on(sbi, dc->state != D_PREP);
1580 __remove_discard_cmd(sbi, dc);
1584 mutex_unlock(&dcc->cmd_lock);
1589 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1591 __drop_discard_cmd(sbi);
1594 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1595 struct discard_cmd *dc)
1597 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1598 unsigned int len = 0;
1600 wait_for_completion_io(&dc->wait);
1601 mutex_lock(&dcc->cmd_lock);
1602 f2fs_bug_on(sbi, dc->state != D_DONE);
1607 __remove_discard_cmd(sbi, dc);
1609 mutex_unlock(&dcc->cmd_lock);
1614 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1615 struct discard_policy *dpolicy,
1616 block_t start, block_t end)
1618 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1619 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1620 &(dcc->fstrim_list) : &(dcc->wait_list);
1621 struct discard_cmd *dc, *tmp;
1623 unsigned int trimmed = 0;
1628 mutex_lock(&dcc->cmd_lock);
1629 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1630 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1632 if (dc->len < dpolicy->granularity)
1634 if (dc->state == D_DONE && !dc->ref) {
1635 wait_for_completion_io(&dc->wait);
1638 __remove_discard_cmd(sbi, dc);
1645 mutex_unlock(&dcc->cmd_lock);
1648 trimmed += __wait_one_discard_bio(sbi, dc);
1655 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1656 struct discard_policy *dpolicy)
1658 struct discard_policy dp;
1659 unsigned int discard_blks;
1662 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1665 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1666 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1667 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1668 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1670 return discard_blks;
1673 /* This should be covered by global mutex, &sit_i->sentry_lock */
1674 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1676 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1677 struct discard_cmd *dc;
1678 bool need_wait = false;
1680 mutex_lock(&dcc->cmd_lock);
1681 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1684 if (dc->state == D_PREP) {
1685 __punch_discard_cmd(sbi, dc, blkaddr);
1691 mutex_unlock(&dcc->cmd_lock);
1694 __wait_one_discard_bio(sbi, dc);
1697 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1699 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1701 if (dcc && dcc->f2fs_issue_discard) {
1702 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1704 dcc->f2fs_issue_discard = NULL;
1705 kthread_stop(discard_thread);
1709 /* This comes from f2fs_put_super */
1710 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1712 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1713 struct discard_policy dpolicy;
1716 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1717 dcc->discard_granularity);
1718 __issue_discard_cmd(sbi, &dpolicy);
1719 dropped = __drop_discard_cmd(sbi);
1721 /* just to make sure there is no pending discard commands */
1722 __wait_all_discard_cmd(sbi, NULL);
1724 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1728 static int issue_discard_thread(void *data)
1730 struct f2fs_sb_info *sbi = data;
1731 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1732 wait_queue_head_t *q = &dcc->discard_wait_queue;
1733 struct discard_policy dpolicy;
1734 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1740 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1741 dcc->discard_granularity);
1743 wait_event_interruptible_timeout(*q,
1744 kthread_should_stop() || freezing(current) ||
1746 msecs_to_jiffies(wait_ms));
1748 if (dcc->discard_wake)
1749 dcc->discard_wake = 0;
1751 /* clean up pending candidates before going to sleep */
1752 if (atomic_read(&dcc->queued_discard))
1753 __wait_all_discard_cmd(sbi, NULL);
1755 if (try_to_freeze())
1757 if (f2fs_readonly(sbi->sb))
1759 if (kthread_should_stop())
1761 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1762 wait_ms = dpolicy.max_interval;
1766 if (sbi->gc_mode == GC_URGENT_HIGH)
1767 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1769 sb_start_intwrite(sbi->sb);
1771 issued = __issue_discard_cmd(sbi, &dpolicy);
1773 __wait_all_discard_cmd(sbi, &dpolicy);
1774 wait_ms = dpolicy.min_interval;
1775 } else if (issued == -1){
1776 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1778 wait_ms = dpolicy.mid_interval;
1780 wait_ms = dpolicy.max_interval;
1783 sb_end_intwrite(sbi->sb);
1785 } while (!kthread_should_stop());
1789 #ifdef CONFIG_BLK_DEV_ZONED
1790 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1791 struct block_device *bdev, block_t blkstart, block_t blklen)
1793 sector_t sector, nr_sects;
1794 block_t lblkstart = blkstart;
1797 if (f2fs_is_multi_device(sbi)) {
1798 devi = f2fs_target_device_index(sbi, blkstart);
1799 if (blkstart < FDEV(devi).start_blk ||
1800 blkstart > FDEV(devi).end_blk) {
1801 f2fs_err(sbi, "Invalid block %x", blkstart);
1804 blkstart -= FDEV(devi).start_blk;
1807 /* For sequential zones, reset the zone write pointer */
1808 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1809 sector = SECTOR_FROM_BLOCK(blkstart);
1810 nr_sects = SECTOR_FROM_BLOCK(blklen);
1812 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1813 nr_sects != bdev_zone_sectors(bdev)) {
1814 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1815 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1819 trace_f2fs_issue_reset_zone(bdev, blkstart);
1820 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1821 sector, nr_sects, GFP_NOFS);
1824 /* For conventional zones, use regular discard if supported */
1825 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1829 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1830 struct block_device *bdev, block_t blkstart, block_t blklen)
1832 #ifdef CONFIG_BLK_DEV_ZONED
1833 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1834 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1836 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1839 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1840 block_t blkstart, block_t blklen)
1842 sector_t start = blkstart, len = 0;
1843 struct block_device *bdev;
1844 struct seg_entry *se;
1845 unsigned int offset;
1849 bdev = f2fs_target_device(sbi, blkstart, NULL);
1851 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1853 struct block_device *bdev2 =
1854 f2fs_target_device(sbi, i, NULL);
1856 if (bdev2 != bdev) {
1857 err = __issue_discard_async(sbi, bdev,
1867 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1868 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1870 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1871 sbi->discard_blks--;
1875 err = __issue_discard_async(sbi, bdev, start, len);
1879 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1882 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1883 int max_blocks = sbi->blocks_per_seg;
1884 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1885 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1886 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1887 unsigned long *discard_map = (unsigned long *)se->discard_map;
1888 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1889 unsigned int start = 0, end = -1;
1890 bool force = (cpc->reason & CP_DISCARD);
1891 struct discard_entry *de = NULL;
1892 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1895 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1899 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1900 SM_I(sbi)->dcc_info->nr_discards >=
1901 SM_I(sbi)->dcc_info->max_discards)
1905 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1906 for (i = 0; i < entries; i++)
1907 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1908 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1910 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1911 SM_I(sbi)->dcc_info->max_discards) {
1912 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1913 if (start >= max_blocks)
1916 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1917 if (force && start && end != max_blocks
1918 && (end - start) < cpc->trim_minlen)
1925 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1927 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1928 list_add_tail(&de->list, head);
1931 for (i = start; i < end; i++)
1932 __set_bit_le(i, (void *)de->discard_map);
1934 SM_I(sbi)->dcc_info->nr_discards += end - start;
1939 static void release_discard_addr(struct discard_entry *entry)
1941 list_del(&entry->list);
1942 kmem_cache_free(discard_entry_slab, entry);
1945 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1947 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1948 struct discard_entry *entry, *this;
1951 list_for_each_entry_safe(entry, this, head, list)
1952 release_discard_addr(entry);
1956 * Should call f2fs_clear_prefree_segments after checkpoint is done.
1958 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1960 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1963 mutex_lock(&dirty_i->seglist_lock);
1964 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1965 __set_test_and_free(sbi, segno, false);
1966 mutex_unlock(&dirty_i->seglist_lock);
1969 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1970 struct cp_control *cpc)
1972 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1973 struct list_head *head = &dcc->entry_list;
1974 struct discard_entry *entry, *this;
1975 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1976 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1977 unsigned int start = 0, end = -1;
1978 unsigned int secno, start_segno;
1979 bool force = (cpc->reason & CP_DISCARD);
1980 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
1982 mutex_lock(&dirty_i->seglist_lock);
1987 if (need_align && end != -1)
1989 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1990 if (start >= MAIN_SEGS(sbi))
1992 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1996 start = rounddown(start, sbi->segs_per_sec);
1997 end = roundup(end, sbi->segs_per_sec);
2000 for (i = start; i < end; i++) {
2001 if (test_and_clear_bit(i, prefree_map))
2002 dirty_i->nr_dirty[PRE]--;
2005 if (!f2fs_realtime_discard_enable(sbi))
2008 if (force && start >= cpc->trim_start &&
2009 (end - 1) <= cpc->trim_end)
2012 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2013 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2014 (end - start) << sbi->log_blocks_per_seg);
2018 secno = GET_SEC_FROM_SEG(sbi, start);
2019 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2020 if (!IS_CURSEC(sbi, secno) &&
2021 !get_valid_blocks(sbi, start, true))
2022 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2023 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2025 start = start_segno + sbi->segs_per_sec;
2031 mutex_unlock(&dirty_i->seglist_lock);
2033 /* send small discards */
2034 list_for_each_entry_safe(entry, this, head, list) {
2035 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2036 bool is_valid = test_bit_le(0, entry->discard_map);
2040 next_pos = find_next_zero_bit_le(entry->discard_map,
2041 sbi->blocks_per_seg, cur_pos);
2042 len = next_pos - cur_pos;
2044 if (f2fs_sb_has_blkzoned(sbi) ||
2045 (force && len < cpc->trim_minlen))
2048 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2052 next_pos = find_next_bit_le(entry->discard_map,
2053 sbi->blocks_per_seg, cur_pos);
2057 is_valid = !is_valid;
2059 if (cur_pos < sbi->blocks_per_seg)
2062 release_discard_addr(entry);
2063 dcc->nr_discards -= total_len;
2066 wake_up_discard_thread(sbi, false);
2069 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2071 dev_t dev = sbi->sb->s_bdev->bd_dev;
2072 struct discard_cmd_control *dcc;
2075 if (SM_I(sbi)->dcc_info) {
2076 dcc = SM_I(sbi)->dcc_info;
2080 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2084 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2085 INIT_LIST_HEAD(&dcc->entry_list);
2086 for (i = 0; i < MAX_PLIST_NUM; i++)
2087 INIT_LIST_HEAD(&dcc->pend_list[i]);
2088 INIT_LIST_HEAD(&dcc->wait_list);
2089 INIT_LIST_HEAD(&dcc->fstrim_list);
2090 mutex_init(&dcc->cmd_lock);
2091 atomic_set(&dcc->issued_discard, 0);
2092 atomic_set(&dcc->queued_discard, 0);
2093 atomic_set(&dcc->discard_cmd_cnt, 0);
2094 dcc->nr_discards = 0;
2095 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2096 dcc->undiscard_blks = 0;
2098 dcc->root = RB_ROOT_CACHED;
2099 dcc->rbtree_check = false;
2101 init_waitqueue_head(&dcc->discard_wait_queue);
2102 SM_I(sbi)->dcc_info = dcc;
2104 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2105 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2106 if (IS_ERR(dcc->f2fs_issue_discard)) {
2107 err = PTR_ERR(dcc->f2fs_issue_discard);
2109 SM_I(sbi)->dcc_info = NULL;
2116 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2118 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2123 f2fs_stop_discard_thread(sbi);
2126 * Recovery can cache discard commands, so in error path of
2127 * fill_super(), it needs to give a chance to handle them.
2129 if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2130 f2fs_issue_discard_timeout(sbi);
2133 SM_I(sbi)->dcc_info = NULL;
2136 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2138 struct sit_info *sit_i = SIT_I(sbi);
2140 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2141 sit_i->dirty_sentries++;
2148 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2149 unsigned int segno, int modified)
2151 struct seg_entry *se = get_seg_entry(sbi, segno);
2154 __mark_sit_entry_dirty(sbi, segno);
2157 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2160 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2162 if (segno == NULL_SEGNO)
2164 return get_seg_entry(sbi, segno)->mtime;
2167 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2168 unsigned long long old_mtime)
2170 struct seg_entry *se;
2171 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2172 unsigned long long ctime = get_mtime(sbi, false);
2173 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2175 if (segno == NULL_SEGNO)
2178 se = get_seg_entry(sbi, segno);
2183 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2184 se->valid_blocks + 1);
2186 if (ctime > SIT_I(sbi)->max_mtime)
2187 SIT_I(sbi)->max_mtime = ctime;
2190 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2192 struct seg_entry *se;
2193 unsigned int segno, offset;
2194 long int new_vblocks;
2196 #ifdef CONFIG_F2FS_CHECK_FS
2200 segno = GET_SEGNO(sbi, blkaddr);
2202 se = get_seg_entry(sbi, segno);
2203 new_vblocks = se->valid_blocks + del;
2204 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2206 f2fs_bug_on(sbi, (new_vblocks < 0 ||
2207 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2209 se->valid_blocks = new_vblocks;
2211 /* Update valid block bitmap */
2213 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2214 #ifdef CONFIG_F2FS_CHECK_FS
2215 mir_exist = f2fs_test_and_set_bit(offset,
2216 se->cur_valid_map_mir);
2217 if (unlikely(exist != mir_exist)) {
2218 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2220 f2fs_bug_on(sbi, 1);
2223 if (unlikely(exist)) {
2224 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2226 f2fs_bug_on(sbi, 1);
2231 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2232 sbi->discard_blks--;
2235 * SSR should never reuse block which is checkpointed
2236 * or newly invalidated.
2238 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2239 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2240 se->ckpt_valid_blocks++;
2243 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2244 #ifdef CONFIG_F2FS_CHECK_FS
2245 mir_exist = f2fs_test_and_clear_bit(offset,
2246 se->cur_valid_map_mir);
2247 if (unlikely(exist != mir_exist)) {
2248 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2250 f2fs_bug_on(sbi, 1);
2253 if (unlikely(!exist)) {
2254 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2256 f2fs_bug_on(sbi, 1);
2259 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2261 * If checkpoints are off, we must not reuse data that
2262 * was used in the previous checkpoint. If it was used
2263 * before, we must track that to know how much space we
2266 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2267 spin_lock(&sbi->stat_lock);
2268 sbi->unusable_block_count++;
2269 spin_unlock(&sbi->stat_lock);
2273 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2274 sbi->discard_blks++;
2276 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2277 se->ckpt_valid_blocks += del;
2279 __mark_sit_entry_dirty(sbi, segno);
2281 /* update total number of valid blocks to be written in ckpt area */
2282 SIT_I(sbi)->written_valid_blocks += del;
2284 if (__is_large_section(sbi))
2285 get_sec_entry(sbi, segno)->valid_blocks += del;
2288 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2290 unsigned int segno = GET_SEGNO(sbi, addr);
2291 struct sit_info *sit_i = SIT_I(sbi);
2293 f2fs_bug_on(sbi, addr == NULL_ADDR);
2294 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2297 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2299 /* add it into sit main buffer */
2300 down_write(&sit_i->sentry_lock);
2302 update_segment_mtime(sbi, addr, 0);
2303 update_sit_entry(sbi, addr, -1);
2305 /* add it into dirty seglist */
2306 locate_dirty_segment(sbi, segno);
2308 up_write(&sit_i->sentry_lock);
2311 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2313 struct sit_info *sit_i = SIT_I(sbi);
2314 unsigned int segno, offset;
2315 struct seg_entry *se;
2318 if (!__is_valid_data_blkaddr(blkaddr))
2321 down_read(&sit_i->sentry_lock);
2323 segno = GET_SEGNO(sbi, blkaddr);
2324 se = get_seg_entry(sbi, segno);
2325 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2327 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2330 up_read(&sit_i->sentry_lock);
2336 * This function should be resided under the curseg_mutex lock
2338 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2339 struct f2fs_summary *sum)
2341 struct curseg_info *curseg = CURSEG_I(sbi, type);
2342 void *addr = curseg->sum_blk;
2343 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2344 memcpy(addr, sum, sizeof(struct f2fs_summary));
2348 * Calculate the number of current summary pages for writing
2350 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2352 int valid_sum_count = 0;
2355 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2356 if (sbi->ckpt->alloc_type[i] == SSR)
2357 valid_sum_count += sbi->blocks_per_seg;
2360 valid_sum_count += le16_to_cpu(
2361 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2363 valid_sum_count += curseg_blkoff(sbi, i);
2367 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2368 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2369 if (valid_sum_count <= sum_in_page)
2371 else if ((valid_sum_count - sum_in_page) <=
2372 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2378 * Caller should put this summary page
2380 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2382 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2385 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2386 void *src, block_t blk_addr)
2388 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2390 memcpy(page_address(page), src, PAGE_SIZE);
2391 set_page_dirty(page);
2392 f2fs_put_page(page, 1);
2395 static void write_sum_page(struct f2fs_sb_info *sbi,
2396 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2398 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2401 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2402 int type, block_t blk_addr)
2404 struct curseg_info *curseg = CURSEG_I(sbi, type);
2405 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2406 struct f2fs_summary_block *src = curseg->sum_blk;
2407 struct f2fs_summary_block *dst;
2409 dst = (struct f2fs_summary_block *)page_address(page);
2410 memset(dst, 0, PAGE_SIZE);
2412 mutex_lock(&curseg->curseg_mutex);
2414 down_read(&curseg->journal_rwsem);
2415 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2416 up_read(&curseg->journal_rwsem);
2418 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2419 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2421 mutex_unlock(&curseg->curseg_mutex);
2423 set_page_dirty(page);
2424 f2fs_put_page(page, 1);
2427 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2429 struct curseg_info *curseg = CURSEG_I(sbi, type);
2430 unsigned int segno = curseg->segno + 1;
2431 struct free_segmap_info *free_i = FREE_I(sbi);
2433 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2434 return !test_bit(segno, free_i->free_segmap);
2439 * Find a new segment from the free segments bitmap to right order
2440 * This function should be returned with success, otherwise BUG
2442 static void get_new_segment(struct f2fs_sb_info *sbi,
2443 unsigned int *newseg, bool new_sec, int dir)
2445 struct free_segmap_info *free_i = FREE_I(sbi);
2446 unsigned int segno, secno, zoneno;
2447 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2448 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2449 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2450 unsigned int left_start = hint;
2455 spin_lock(&free_i->segmap_lock);
2457 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2458 segno = find_next_zero_bit(free_i->free_segmap,
2459 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2460 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2464 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2465 if (secno >= MAIN_SECS(sbi)) {
2466 if (dir == ALLOC_RIGHT) {
2467 secno = find_next_zero_bit(free_i->free_secmap,
2469 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2472 left_start = hint - 1;
2478 while (test_bit(left_start, free_i->free_secmap)) {
2479 if (left_start > 0) {
2483 left_start = find_next_zero_bit(free_i->free_secmap,
2485 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2490 segno = GET_SEG_FROM_SEC(sbi, secno);
2491 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2493 /* give up on finding another zone */
2496 if (sbi->secs_per_zone == 1)
2498 if (zoneno == old_zoneno)
2500 if (dir == ALLOC_LEFT) {
2501 if (!go_left && zoneno + 1 >= total_zones)
2503 if (go_left && zoneno == 0)
2506 for (i = 0; i < NR_CURSEG_TYPE; i++)
2507 if (CURSEG_I(sbi, i)->zone == zoneno)
2510 if (i < NR_CURSEG_TYPE) {
2511 /* zone is in user, try another */
2513 hint = zoneno * sbi->secs_per_zone - 1;
2514 else if (zoneno + 1 >= total_zones)
2517 hint = (zoneno + 1) * sbi->secs_per_zone;
2519 goto find_other_zone;
2522 /* set it as dirty segment in free segmap */
2523 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2524 __set_inuse(sbi, segno);
2526 spin_unlock(&free_i->segmap_lock);
2529 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2531 struct curseg_info *curseg = CURSEG_I(sbi, type);
2532 struct summary_footer *sum_footer;
2534 curseg->inited = true;
2535 curseg->segno = curseg->next_segno;
2536 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2537 curseg->next_blkoff = 0;
2538 curseg->next_segno = NULL_SEGNO;
2540 sum_footer = &(curseg->sum_blk->footer);
2541 memset(sum_footer, 0, sizeof(struct summary_footer));
2542 if (IS_DATASEG(curseg->seg_type))
2543 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2544 if (IS_NODESEG(curseg->seg_type))
2545 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2546 __set_sit_entry_type(sbi, curseg->seg_type, curseg->segno, modified);
2549 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2551 struct curseg_info *curseg = CURSEG_I(sbi, type);
2553 /* if segs_per_sec is large than 1, we need to keep original policy. */
2554 if (__is_large_section(sbi))
2555 return curseg->segno;
2557 /* inmem log may not locate on any segment after mount */
2558 if (!curseg->inited)
2561 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2564 if (test_opt(sbi, NOHEAP) &&
2565 (curseg->seg_type == CURSEG_HOT_DATA ||
2566 IS_NODESEG(curseg->seg_type)))
2569 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2570 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2572 /* find segments from 0 to reuse freed segments */
2573 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2576 return curseg->segno;
2580 * Allocate a current working segment.
2581 * This function always allocates a free segment in LFS manner.
2583 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2585 struct curseg_info *curseg = CURSEG_I(sbi, type);
2586 unsigned short seg_type = curseg->seg_type;
2587 unsigned int segno = curseg->segno;
2588 int dir = ALLOC_LEFT;
2591 write_sum_page(sbi, curseg->sum_blk,
2592 GET_SUM_BLOCK(sbi, segno));
2593 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2596 if (test_opt(sbi, NOHEAP))
2599 segno = __get_next_segno(sbi, type);
2600 get_new_segment(sbi, &segno, new_sec, dir);
2601 curseg->next_segno = segno;
2602 reset_curseg(sbi, type, 1);
2603 curseg->alloc_type = LFS;
2606 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2607 struct curseg_info *seg, block_t start)
2609 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2610 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2611 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2612 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2613 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2616 for (i = 0; i < entries; i++)
2617 target_map[i] = ckpt_map[i] | cur_map[i];
2619 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2621 seg->next_blkoff = pos;
2625 * If a segment is written by LFS manner, next block offset is just obtained
2626 * by increasing the current block offset. However, if a segment is written by
2627 * SSR manner, next block offset obtained by calling __next_free_blkoff
2629 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2630 struct curseg_info *seg)
2632 if (seg->alloc_type == SSR)
2633 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2639 * This function always allocates a used segment(from dirty seglist) by SSR
2640 * manner, so it should recover the existing segment information of valid blocks
2642 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2644 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2645 struct curseg_info *curseg = CURSEG_I(sbi, type);
2646 unsigned int new_segno = curseg->next_segno;
2647 struct f2fs_summary_block *sum_node;
2648 struct page *sum_page;
2650 write_sum_page(sbi, curseg->sum_blk,
2651 GET_SUM_BLOCK(sbi, curseg->segno));
2652 __set_test_and_inuse(sbi, new_segno);
2654 mutex_lock(&dirty_i->seglist_lock);
2655 __remove_dirty_segment(sbi, new_segno, PRE);
2656 __remove_dirty_segment(sbi, new_segno, DIRTY);
2657 mutex_unlock(&dirty_i->seglist_lock);
2659 reset_curseg(sbi, type, 1);
2660 curseg->alloc_type = SSR;
2661 __next_free_blkoff(sbi, curseg, 0);
2663 sum_page = f2fs_get_sum_page(sbi, new_segno);
2664 f2fs_bug_on(sbi, IS_ERR(sum_page));
2665 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2666 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2667 f2fs_put_page(sum_page, 1);
2670 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2672 struct curseg_info *curseg = CURSEG_I(sbi, type);
2674 mutex_lock(&curseg->curseg_mutex);
2675 if (!curseg->inited)
2678 if (get_valid_blocks(sbi, curseg->segno, false)) {
2679 write_sum_page(sbi, curseg->sum_blk,
2680 GET_SUM_BLOCK(sbi, curseg->segno));
2682 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2683 __set_test_and_free(sbi, curseg->segno, true);
2684 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2687 mutex_unlock(&curseg->curseg_mutex);
2690 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2692 struct curseg_info *curseg = CURSEG_I(sbi, type);
2694 mutex_lock(&curseg->curseg_mutex);
2695 if (!curseg->inited)
2697 if (get_valid_blocks(sbi, curseg->segno, false))
2700 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2701 __set_test_and_inuse(sbi, curseg->segno);
2702 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2704 mutex_unlock(&curseg->curseg_mutex);
2707 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2709 struct curseg_info *curseg = CURSEG_I(sbi, type);
2710 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2711 unsigned segno = NULL_SEGNO;
2713 bool reversed = false;
2715 /* f2fs_need_SSR() already forces to do this */
2716 if (!v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2717 curseg->next_segno = segno;
2721 /* For node segments, let's do SSR more intensively */
2722 if (IS_NODESEG(type)) {
2723 if (type >= CURSEG_WARM_NODE) {
2725 i = CURSEG_COLD_NODE;
2727 i = CURSEG_HOT_NODE;
2729 cnt = NR_CURSEG_NODE_TYPE;
2731 if (type >= CURSEG_WARM_DATA) {
2733 i = CURSEG_COLD_DATA;
2735 i = CURSEG_HOT_DATA;
2737 cnt = NR_CURSEG_DATA_TYPE;
2740 for (; cnt-- > 0; reversed ? i-- : i++) {
2743 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2744 curseg->next_segno = segno;
2749 /* find valid_blocks=0 in dirty list */
2750 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2751 segno = get_free_segment(sbi);
2752 if (segno != NULL_SEGNO) {
2753 curseg->next_segno = segno;
2761 * flush out current segment and replace it with new segment
2762 * This function should be returned with success, otherwise BUG
2764 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2765 int type, bool force)
2767 struct curseg_info *curseg = CURSEG_I(sbi, type);
2770 new_curseg(sbi, type, true);
2771 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2772 type == CURSEG_WARM_NODE)
2773 new_curseg(sbi, type, false);
2774 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2775 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2776 new_curseg(sbi, type, false);
2777 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2778 change_curseg(sbi, type);
2780 new_curseg(sbi, type, false);
2782 stat_inc_seg_type(sbi, curseg);
2785 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2786 unsigned int start, unsigned int end)
2788 struct curseg_info *curseg = CURSEG_I(sbi, type);
2791 down_read(&SM_I(sbi)->curseg_lock);
2792 mutex_lock(&curseg->curseg_mutex);
2793 down_write(&SIT_I(sbi)->sentry_lock);
2795 segno = CURSEG_I(sbi, type)->segno;
2796 if (segno < start || segno > end)
2799 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2800 change_curseg(sbi, type);
2802 new_curseg(sbi, type, true);
2804 stat_inc_seg_type(sbi, curseg);
2806 locate_dirty_segment(sbi, segno);
2808 up_write(&SIT_I(sbi)->sentry_lock);
2810 if (segno != curseg->segno)
2811 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2812 type, segno, curseg->segno);
2814 mutex_unlock(&curseg->curseg_mutex);
2815 up_read(&SM_I(sbi)->curseg_lock);
2818 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2820 struct curseg_info *curseg = CURSEG_I(sbi, type);
2821 unsigned int old_segno;
2823 if (!curseg->inited)
2826 if (!curseg->next_blkoff &&
2827 !get_valid_blocks(sbi, curseg->segno, false) &&
2828 !get_ckpt_valid_blocks(sbi, curseg->segno))
2832 old_segno = curseg->segno;
2833 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2834 locate_dirty_segment(sbi, old_segno);
2837 void f2fs_allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2839 down_write(&SIT_I(sbi)->sentry_lock);
2840 __allocate_new_segment(sbi, type);
2841 up_write(&SIT_I(sbi)->sentry_lock);
2844 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2848 down_write(&SIT_I(sbi)->sentry_lock);
2849 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2850 __allocate_new_segment(sbi, i);
2851 up_write(&SIT_I(sbi)->sentry_lock);
2854 static const struct segment_allocation default_salloc_ops = {
2855 .allocate_segment = allocate_segment_by_default,
2858 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2859 struct cp_control *cpc)
2861 __u64 trim_start = cpc->trim_start;
2862 bool has_candidate = false;
2864 down_write(&SIT_I(sbi)->sentry_lock);
2865 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2866 if (add_discard_addrs(sbi, cpc, true)) {
2867 has_candidate = true;
2871 up_write(&SIT_I(sbi)->sentry_lock);
2873 cpc->trim_start = trim_start;
2874 return has_candidate;
2877 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2878 struct discard_policy *dpolicy,
2879 unsigned int start, unsigned int end)
2881 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2882 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2883 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2884 struct discard_cmd *dc;
2885 struct blk_plug plug;
2887 unsigned int trimmed = 0;
2892 mutex_lock(&dcc->cmd_lock);
2893 if (unlikely(dcc->rbtree_check))
2894 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2897 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2899 (struct rb_entry **)&prev_dc,
2900 (struct rb_entry **)&next_dc,
2901 &insert_p, &insert_parent, true, NULL);
2905 blk_start_plug(&plug);
2907 while (dc && dc->lstart <= end) {
2908 struct rb_node *node;
2911 if (dc->len < dpolicy->granularity)
2914 if (dc->state != D_PREP) {
2915 list_move_tail(&dc->list, &dcc->fstrim_list);
2919 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2921 if (issued >= dpolicy->max_requests) {
2922 start = dc->lstart + dc->len;
2925 __remove_discard_cmd(sbi, dc);
2927 blk_finish_plug(&plug);
2928 mutex_unlock(&dcc->cmd_lock);
2929 trimmed += __wait_all_discard_cmd(sbi, NULL);
2930 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2934 node = rb_next(&dc->rb_node);
2936 __remove_discard_cmd(sbi, dc);
2937 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2939 if (fatal_signal_pending(current))
2943 blk_finish_plug(&plug);
2944 mutex_unlock(&dcc->cmd_lock);
2949 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2951 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2952 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2953 unsigned int start_segno, end_segno;
2954 block_t start_block, end_block;
2955 struct cp_control cpc;
2956 struct discard_policy dpolicy;
2957 unsigned long long trimmed = 0;
2959 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
2961 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2964 if (end < MAIN_BLKADDR(sbi))
2967 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2968 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2969 return -EFSCORRUPTED;
2972 /* start/end segment number in main_area */
2973 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2974 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2975 GET_SEGNO(sbi, end);
2977 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2978 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2981 cpc.reason = CP_DISCARD;
2982 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2983 cpc.trim_start = start_segno;
2984 cpc.trim_end = end_segno;
2986 if (sbi->discard_blks == 0)
2989 down_write(&sbi->gc_lock);
2990 err = f2fs_write_checkpoint(sbi, &cpc);
2991 up_write(&sbi->gc_lock);
2996 * We filed discard candidates, but actually we don't need to wait for
2997 * all of them, since they'll be issued in idle time along with runtime
2998 * discard option. User configuration looks like using runtime discard
2999 * or periodic fstrim instead of it.
3001 if (f2fs_realtime_discard_enable(sbi))
3004 start_block = START_BLOCK(sbi, start_segno);
3005 end_block = START_BLOCK(sbi, end_segno + 1);
3007 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3008 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3009 start_block, end_block);
3011 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3012 start_block, end_block);
3015 range->len = F2FS_BLK_TO_BYTES(trimmed);
3019 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
3021 struct curseg_info *curseg = CURSEG_I(sbi, type);
3023 return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3027 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3030 case WRITE_LIFE_SHORT:
3031 return CURSEG_HOT_DATA;
3032 case WRITE_LIFE_EXTREME:
3033 return CURSEG_COLD_DATA;
3035 return CURSEG_WARM_DATA;
3039 /* This returns write hints for each segment type. This hints will be
3040 * passed down to block layer. There are mapping tables which depend on
3041 * the mount option 'whint_mode'.
3043 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3045 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3049 * META WRITE_LIFE_NOT_SET
3053 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3054 * extension list " "
3057 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3058 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3059 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3060 * WRITE_LIFE_NONE " "
3061 * WRITE_LIFE_MEDIUM " "
3062 * WRITE_LIFE_LONG " "
3065 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3066 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3067 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3068 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3069 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3070 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
3072 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3076 * META WRITE_LIFE_MEDIUM;
3077 * HOT_NODE WRITE_LIFE_NOT_SET
3079 * COLD_NODE WRITE_LIFE_NONE
3080 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3081 * extension list " "
3084 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3085 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3086 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
3087 * WRITE_LIFE_NONE " "
3088 * WRITE_LIFE_MEDIUM " "
3089 * WRITE_LIFE_LONG " "
3092 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3093 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3094 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3095 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3096 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3097 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
3100 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3101 enum page_type type, enum temp_type temp)
3103 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3106 return WRITE_LIFE_NOT_SET;
3107 else if (temp == HOT)
3108 return WRITE_LIFE_SHORT;
3109 else if (temp == COLD)
3110 return WRITE_LIFE_EXTREME;
3112 return WRITE_LIFE_NOT_SET;
3114 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3117 return WRITE_LIFE_LONG;
3118 else if (temp == HOT)
3119 return WRITE_LIFE_SHORT;
3120 else if (temp == COLD)
3121 return WRITE_LIFE_EXTREME;
3122 } else if (type == NODE) {
3123 if (temp == WARM || temp == HOT)
3124 return WRITE_LIFE_NOT_SET;
3125 else if (temp == COLD)
3126 return WRITE_LIFE_NONE;
3127 } else if (type == META) {
3128 return WRITE_LIFE_MEDIUM;
3131 return WRITE_LIFE_NOT_SET;
3134 static int __get_segment_type_2(struct f2fs_io_info *fio)
3136 if (fio->type == DATA)
3137 return CURSEG_HOT_DATA;
3139 return CURSEG_HOT_NODE;
3142 static int __get_segment_type_4(struct f2fs_io_info *fio)
3144 if (fio->type == DATA) {
3145 struct inode *inode = fio->page->mapping->host;
3147 if (S_ISDIR(inode->i_mode))
3148 return CURSEG_HOT_DATA;
3150 return CURSEG_COLD_DATA;
3152 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3153 return CURSEG_WARM_NODE;
3155 return CURSEG_COLD_NODE;
3159 static int __get_segment_type_6(struct f2fs_io_info *fio)
3161 if (fio->type == DATA) {
3162 struct inode *inode = fio->page->mapping->host;
3164 if (is_cold_data(fio->page) || file_is_cold(inode) ||
3165 f2fs_compressed_file(inode))
3166 return CURSEG_COLD_DATA;
3167 if (file_is_hot(inode) ||
3168 is_inode_flag_set(inode, FI_HOT_DATA) ||
3169 f2fs_is_atomic_file(inode) ||
3170 f2fs_is_volatile_file(inode))
3171 return CURSEG_HOT_DATA;
3172 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3174 if (IS_DNODE(fio->page))
3175 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3177 return CURSEG_COLD_NODE;
3181 static int __get_segment_type(struct f2fs_io_info *fio)
3185 switch (F2FS_OPTION(fio->sbi).active_logs) {
3187 type = __get_segment_type_2(fio);
3190 type = __get_segment_type_4(fio);
3193 type = __get_segment_type_6(fio);
3196 f2fs_bug_on(fio->sbi, true);
3201 else if (IS_WARM(type))
3208 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3209 block_t old_blkaddr, block_t *new_blkaddr,
3210 struct f2fs_summary *sum, int type,
3211 struct f2fs_io_info *fio, bool from_gc)
3213 struct sit_info *sit_i = SIT_I(sbi);
3214 struct curseg_info *curseg = CURSEG_I(sbi, type);
3215 unsigned long long old_mtime;
3217 down_read(&SM_I(sbi)->curseg_lock);
3219 mutex_lock(&curseg->curseg_mutex);
3220 down_write(&sit_i->sentry_lock);
3222 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3224 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3227 * __add_sum_entry should be resided under the curseg_mutex
3228 * because, this function updates a summary entry in the
3229 * current summary block.
3231 __add_sum_entry(sbi, type, sum);
3233 __refresh_next_blkoff(sbi, curseg);
3235 stat_inc_block_count(sbi, curseg);
3238 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3240 update_segment_mtime(sbi, old_blkaddr, 0);
3243 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3246 * SIT information should be updated before segment allocation,
3247 * since SSR needs latest valid block information.
3249 update_sit_entry(sbi, *new_blkaddr, 1);
3250 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3251 update_sit_entry(sbi, old_blkaddr, -1);
3253 if (!__has_curseg_space(sbi, type))
3254 sit_i->s_ops->allocate_segment(sbi, type, false);
3257 * segment dirty status should be updated after segment allocation,
3258 * so we just need to update status only one time after previous
3259 * segment being closed.
3261 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3262 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3264 up_write(&sit_i->sentry_lock);
3266 if (page && IS_NODESEG(type)) {
3267 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3269 f2fs_inode_chksum_set(sbi, page);
3272 if (F2FS_IO_ALIGNED(sbi))
3276 struct f2fs_bio_info *io;
3278 INIT_LIST_HEAD(&fio->list);
3279 fio->in_list = true;
3280 io = sbi->write_io[fio->type] + fio->temp;
3281 spin_lock(&io->io_lock);
3282 list_add_tail(&fio->list, &io->io_list);
3283 spin_unlock(&io->io_lock);
3286 mutex_unlock(&curseg->curseg_mutex);
3288 up_read(&SM_I(sbi)->curseg_lock);
3291 static void update_device_state(struct f2fs_io_info *fio)
3293 struct f2fs_sb_info *sbi = fio->sbi;
3294 unsigned int devidx;
3296 if (!f2fs_is_multi_device(sbi))
3299 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3301 /* update device state for fsync */
3302 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3304 /* update device state for checkpoint */
3305 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3306 spin_lock(&sbi->dev_lock);
3307 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3308 spin_unlock(&sbi->dev_lock);
3312 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3314 int type = __get_segment_type(fio);
3315 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3318 down_read(&fio->sbi->io_order_lock);
3320 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3321 &fio->new_blkaddr, sum, type, fio,
3322 is_cold_data(fio->page));
3323 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3324 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3325 fio->old_blkaddr, fio->old_blkaddr);
3327 /* writeout dirty page into bdev */
3328 f2fs_submit_page_write(fio);
3330 fio->old_blkaddr = fio->new_blkaddr;
3334 update_device_state(fio);
3337 up_read(&fio->sbi->io_order_lock);
3340 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3341 enum iostat_type io_type)
3343 struct f2fs_io_info fio = {
3348 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3349 .old_blkaddr = page->index,
3350 .new_blkaddr = page->index,
3352 .encrypted_page = NULL,
3356 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3357 fio.op_flags &= ~REQ_META;
3359 set_page_writeback(page);
3360 ClearPageError(page);
3361 f2fs_submit_page_write(&fio);
3363 stat_inc_meta_count(sbi, page->index);
3364 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3367 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3369 struct f2fs_summary sum;
3371 set_summary(&sum, nid, 0, 0);
3372 do_write_page(&sum, fio);
3374 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3377 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3378 struct f2fs_io_info *fio)
3380 struct f2fs_sb_info *sbi = fio->sbi;
3381 struct f2fs_summary sum;
3383 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3384 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3385 do_write_page(&sum, fio);
3386 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3388 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3391 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3394 struct f2fs_sb_info *sbi = fio->sbi;
3397 fio->new_blkaddr = fio->old_blkaddr;
3398 /* i/o temperature is needed for passing down write hints */
3399 __get_segment_type(fio);
3401 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3403 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3404 set_sbi_flag(sbi, SBI_NEED_FSCK);
3405 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3407 return -EFSCORRUPTED;
3410 stat_inc_inplace_blocks(fio->sbi);
3412 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3413 err = f2fs_merge_page_bio(fio);
3415 err = f2fs_submit_page_bio(fio);
3417 update_device_state(fio);
3418 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3424 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3429 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3430 if (CURSEG_I(sbi, i)->segno == segno)
3436 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3437 block_t old_blkaddr, block_t new_blkaddr,
3438 bool recover_curseg, bool recover_newaddr,
3441 struct sit_info *sit_i = SIT_I(sbi);
3442 struct curseg_info *curseg;
3443 unsigned int segno, old_cursegno;
3444 struct seg_entry *se;
3446 unsigned short old_blkoff;
3448 segno = GET_SEGNO(sbi, new_blkaddr);
3449 se = get_seg_entry(sbi, segno);
3452 down_write(&SM_I(sbi)->curseg_lock);
3454 if (!recover_curseg) {
3455 /* for recovery flow */
3456 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3457 if (old_blkaddr == NULL_ADDR)
3458 type = CURSEG_COLD_DATA;
3460 type = CURSEG_WARM_DATA;
3463 if (IS_CURSEG(sbi, segno)) {
3464 /* se->type is volatile as SSR allocation */
3465 type = __f2fs_get_curseg(sbi, segno);
3466 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3468 type = CURSEG_WARM_DATA;
3472 f2fs_bug_on(sbi, !IS_DATASEG(type));
3473 curseg = CURSEG_I(sbi, type);
3475 mutex_lock(&curseg->curseg_mutex);
3476 down_write(&sit_i->sentry_lock);
3478 old_cursegno = curseg->segno;
3479 old_blkoff = curseg->next_blkoff;
3481 /* change the current segment */
3482 if (segno != curseg->segno) {
3483 curseg->next_segno = segno;
3484 change_curseg(sbi, type);
3487 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3488 __add_sum_entry(sbi, type, sum);
3490 if (!recover_curseg || recover_newaddr) {
3492 update_segment_mtime(sbi, new_blkaddr, 0);
3493 update_sit_entry(sbi, new_blkaddr, 1);
3495 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3496 invalidate_mapping_pages(META_MAPPING(sbi),
3497 old_blkaddr, old_blkaddr);
3499 update_segment_mtime(sbi, old_blkaddr, 0);
3500 update_sit_entry(sbi, old_blkaddr, -1);
3503 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3504 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3506 locate_dirty_segment(sbi, old_cursegno);
3508 if (recover_curseg) {
3509 if (old_cursegno != curseg->segno) {
3510 curseg->next_segno = old_cursegno;
3511 change_curseg(sbi, type);
3513 curseg->next_blkoff = old_blkoff;
3516 up_write(&sit_i->sentry_lock);
3517 mutex_unlock(&curseg->curseg_mutex);
3518 up_write(&SM_I(sbi)->curseg_lock);
3521 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3522 block_t old_addr, block_t new_addr,
3523 unsigned char version, bool recover_curseg,
3524 bool recover_newaddr)
3526 struct f2fs_summary sum;
3528 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3530 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3531 recover_curseg, recover_newaddr, false);
3533 f2fs_update_data_blkaddr(dn, new_addr);
3536 void f2fs_wait_on_page_writeback(struct page *page,
3537 enum page_type type, bool ordered, bool locked)
3539 if (PageWriteback(page)) {
3540 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3542 /* submit cached LFS IO */
3543 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3544 /* sbumit cached IPU IO */
3545 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3547 wait_on_page_writeback(page);
3548 f2fs_bug_on(sbi, locked && PageWriteback(page));
3550 wait_for_stable_page(page);
3555 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3557 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3560 if (!f2fs_post_read_required(inode))
3563 if (!__is_valid_data_blkaddr(blkaddr))
3566 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3568 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3569 f2fs_put_page(cpage, 1);
3573 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3578 for (i = 0; i < len; i++)
3579 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3582 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3584 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3585 struct curseg_info *seg_i;
3586 unsigned char *kaddr;
3591 start = start_sum_block(sbi);
3593 page = f2fs_get_meta_page(sbi, start++);
3595 return PTR_ERR(page);
3596 kaddr = (unsigned char *)page_address(page);
3598 /* Step 1: restore nat cache */
3599 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3600 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3602 /* Step 2: restore sit cache */
3603 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3604 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3605 offset = 2 * SUM_JOURNAL_SIZE;
3607 /* Step 3: restore summary entries */
3608 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3609 unsigned short blk_off;
3612 seg_i = CURSEG_I(sbi, i);
3613 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3614 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3615 seg_i->next_segno = segno;
3616 reset_curseg(sbi, i, 0);
3617 seg_i->alloc_type = ckpt->alloc_type[i];
3618 seg_i->next_blkoff = blk_off;
3620 if (seg_i->alloc_type == SSR)
3621 blk_off = sbi->blocks_per_seg;
3623 for (j = 0; j < blk_off; j++) {
3624 struct f2fs_summary *s;
3625 s = (struct f2fs_summary *)(kaddr + offset);
3626 seg_i->sum_blk->entries[j] = *s;
3627 offset += SUMMARY_SIZE;
3628 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3632 f2fs_put_page(page, 1);
3635 page = f2fs_get_meta_page(sbi, start++);
3637 return PTR_ERR(page);
3638 kaddr = (unsigned char *)page_address(page);
3642 f2fs_put_page(page, 1);
3646 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3648 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3649 struct f2fs_summary_block *sum;
3650 struct curseg_info *curseg;
3652 unsigned short blk_off;
3653 unsigned int segno = 0;
3654 block_t blk_addr = 0;
3657 /* get segment number and block addr */
3658 if (IS_DATASEG(type)) {
3659 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3660 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3662 if (__exist_node_summaries(sbi))
3663 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3665 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3667 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3669 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3671 if (__exist_node_summaries(sbi))
3672 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3673 type - CURSEG_HOT_NODE);
3675 blk_addr = GET_SUM_BLOCK(sbi, segno);
3678 new = f2fs_get_meta_page(sbi, blk_addr);
3680 return PTR_ERR(new);
3681 sum = (struct f2fs_summary_block *)page_address(new);
3683 if (IS_NODESEG(type)) {
3684 if (__exist_node_summaries(sbi)) {
3685 struct f2fs_summary *ns = &sum->entries[0];
3687 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3689 ns->ofs_in_node = 0;
3692 err = f2fs_restore_node_summary(sbi, segno, sum);
3698 /* set uncompleted segment to curseg */
3699 curseg = CURSEG_I(sbi, type);
3700 mutex_lock(&curseg->curseg_mutex);
3702 /* update journal info */
3703 down_write(&curseg->journal_rwsem);
3704 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3705 up_write(&curseg->journal_rwsem);
3707 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3708 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3709 curseg->next_segno = segno;
3710 reset_curseg(sbi, type, 0);
3711 curseg->alloc_type = ckpt->alloc_type[type];
3712 curseg->next_blkoff = blk_off;
3713 mutex_unlock(&curseg->curseg_mutex);
3715 f2fs_put_page(new, 1);
3719 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3721 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3722 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3723 int type = CURSEG_HOT_DATA;
3726 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3727 int npages = f2fs_npages_for_summary_flush(sbi, true);
3730 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3733 /* restore for compacted data summary */
3734 err = read_compacted_summaries(sbi);
3737 type = CURSEG_HOT_NODE;
3740 if (__exist_node_summaries(sbi))
3741 f2fs_ra_meta_pages(sbi,
3742 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3743 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3745 for (; type <= CURSEG_COLD_NODE; type++) {
3746 err = read_normal_summaries(sbi, type);
3751 /* sanity check for summary blocks */
3752 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3753 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3754 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3755 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3762 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3765 unsigned char *kaddr;
3766 struct f2fs_summary *summary;
3767 struct curseg_info *seg_i;
3768 int written_size = 0;
3771 page = f2fs_grab_meta_page(sbi, blkaddr++);
3772 kaddr = (unsigned char *)page_address(page);
3773 memset(kaddr, 0, PAGE_SIZE);
3775 /* Step 1: write nat cache */
3776 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3777 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3778 written_size += SUM_JOURNAL_SIZE;
3780 /* Step 2: write sit cache */
3781 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3782 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3783 written_size += SUM_JOURNAL_SIZE;
3785 /* Step 3: write summary entries */
3786 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3787 unsigned short blkoff;
3788 seg_i = CURSEG_I(sbi, i);
3789 if (sbi->ckpt->alloc_type[i] == SSR)
3790 blkoff = sbi->blocks_per_seg;
3792 blkoff = curseg_blkoff(sbi, i);
3794 for (j = 0; j < blkoff; j++) {
3796 page = f2fs_grab_meta_page(sbi, blkaddr++);
3797 kaddr = (unsigned char *)page_address(page);
3798 memset(kaddr, 0, PAGE_SIZE);
3801 summary = (struct f2fs_summary *)(kaddr + written_size);
3802 *summary = seg_i->sum_blk->entries[j];
3803 written_size += SUMMARY_SIZE;
3805 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3809 set_page_dirty(page);
3810 f2fs_put_page(page, 1);
3815 set_page_dirty(page);
3816 f2fs_put_page(page, 1);
3820 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3821 block_t blkaddr, int type)
3824 if (IS_DATASEG(type))
3825 end = type + NR_CURSEG_DATA_TYPE;
3827 end = type + NR_CURSEG_NODE_TYPE;
3829 for (i = type; i < end; i++)
3830 write_current_sum_page(sbi, i, blkaddr + (i - type));
3833 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3835 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3836 write_compacted_summaries(sbi, start_blk);
3838 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3841 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3843 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3846 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3847 unsigned int val, int alloc)
3851 if (type == NAT_JOURNAL) {
3852 for (i = 0; i < nats_in_cursum(journal); i++) {
3853 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3856 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3857 return update_nats_in_cursum(journal, 1);
3858 } else if (type == SIT_JOURNAL) {
3859 for (i = 0; i < sits_in_cursum(journal); i++)
3860 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3862 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3863 return update_sits_in_cursum(journal, 1);
3868 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3871 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3874 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3877 struct sit_info *sit_i = SIT_I(sbi);
3879 pgoff_t src_off, dst_off;
3881 src_off = current_sit_addr(sbi, start);
3882 dst_off = next_sit_addr(sbi, src_off);
3884 page = f2fs_grab_meta_page(sbi, dst_off);
3885 seg_info_to_sit_page(sbi, page, start);
3887 set_page_dirty(page);
3888 set_to_next_sit(sit_i, start);
3893 static struct sit_entry_set *grab_sit_entry_set(void)
3895 struct sit_entry_set *ses =
3896 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3899 INIT_LIST_HEAD(&ses->set_list);
3903 static void release_sit_entry_set(struct sit_entry_set *ses)
3905 list_del(&ses->set_list);
3906 kmem_cache_free(sit_entry_set_slab, ses);
3909 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3910 struct list_head *head)
3912 struct sit_entry_set *next = ses;
3914 if (list_is_last(&ses->set_list, head))
3917 list_for_each_entry_continue(next, head, set_list)
3918 if (ses->entry_cnt <= next->entry_cnt)
3921 list_move_tail(&ses->set_list, &next->set_list);
3924 static void add_sit_entry(unsigned int segno, struct list_head *head)
3926 struct sit_entry_set *ses;
3927 unsigned int start_segno = START_SEGNO(segno);
3929 list_for_each_entry(ses, head, set_list) {
3930 if (ses->start_segno == start_segno) {
3932 adjust_sit_entry_set(ses, head);
3937 ses = grab_sit_entry_set();
3939 ses->start_segno = start_segno;
3941 list_add(&ses->set_list, head);
3944 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3946 struct f2fs_sm_info *sm_info = SM_I(sbi);
3947 struct list_head *set_list = &sm_info->sit_entry_set;
3948 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3951 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3952 add_sit_entry(segno, set_list);
3955 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3957 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3958 struct f2fs_journal *journal = curseg->journal;
3961 down_write(&curseg->journal_rwsem);
3962 for (i = 0; i < sits_in_cursum(journal); i++) {
3966 segno = le32_to_cpu(segno_in_journal(journal, i));
3967 dirtied = __mark_sit_entry_dirty(sbi, segno);
3970 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3972 update_sits_in_cursum(journal, -i);
3973 up_write(&curseg->journal_rwsem);
3977 * CP calls this function, which flushes SIT entries including sit_journal,
3978 * and moves prefree segs to free segs.
3980 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3982 struct sit_info *sit_i = SIT_I(sbi);
3983 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3984 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3985 struct f2fs_journal *journal = curseg->journal;
3986 struct sit_entry_set *ses, *tmp;
3987 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3988 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3989 struct seg_entry *se;
3991 down_write(&sit_i->sentry_lock);
3993 if (!sit_i->dirty_sentries)
3997 * add and account sit entries of dirty bitmap in sit entry
4000 add_sits_in_set(sbi);
4003 * if there are no enough space in journal to store dirty sit
4004 * entries, remove all entries from journal and add and account
4005 * them in sit entry set.
4007 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4009 remove_sits_in_journal(sbi);
4012 * there are two steps to flush sit entries:
4013 * #1, flush sit entries to journal in current cold data summary block.
4014 * #2, flush sit entries to sit page.
4016 list_for_each_entry_safe(ses, tmp, head, set_list) {
4017 struct page *page = NULL;
4018 struct f2fs_sit_block *raw_sit = NULL;
4019 unsigned int start_segno = ses->start_segno;
4020 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4021 (unsigned long)MAIN_SEGS(sbi));
4022 unsigned int segno = start_segno;
4025 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4029 down_write(&curseg->journal_rwsem);
4031 page = get_next_sit_page(sbi, start_segno);
4032 raw_sit = page_address(page);
4035 /* flush dirty sit entries in region of current sit set */
4036 for_each_set_bit_from(segno, bitmap, end) {
4037 int offset, sit_offset;
4039 se = get_seg_entry(sbi, segno);
4040 #ifdef CONFIG_F2FS_CHECK_FS
4041 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4042 SIT_VBLOCK_MAP_SIZE))
4043 f2fs_bug_on(sbi, 1);
4046 /* add discard candidates */
4047 if (!(cpc->reason & CP_DISCARD)) {
4048 cpc->trim_start = segno;
4049 add_discard_addrs(sbi, cpc, false);
4053 offset = f2fs_lookup_journal_in_cursum(journal,
4054 SIT_JOURNAL, segno, 1);
4055 f2fs_bug_on(sbi, offset < 0);
4056 segno_in_journal(journal, offset) =
4058 seg_info_to_raw_sit(se,
4059 &sit_in_journal(journal, offset));
4060 check_block_count(sbi, segno,
4061 &sit_in_journal(journal, offset));
4063 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4064 seg_info_to_raw_sit(se,
4065 &raw_sit->entries[sit_offset]);
4066 check_block_count(sbi, segno,
4067 &raw_sit->entries[sit_offset]);
4070 __clear_bit(segno, bitmap);
4071 sit_i->dirty_sentries--;
4076 up_write(&curseg->journal_rwsem);
4078 f2fs_put_page(page, 1);
4080 f2fs_bug_on(sbi, ses->entry_cnt);
4081 release_sit_entry_set(ses);
4084 f2fs_bug_on(sbi, !list_empty(head));
4085 f2fs_bug_on(sbi, sit_i->dirty_sentries);
4087 if (cpc->reason & CP_DISCARD) {
4088 __u64 trim_start = cpc->trim_start;
4090 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4091 add_discard_addrs(sbi, cpc, false);
4093 cpc->trim_start = trim_start;
4095 up_write(&sit_i->sentry_lock);
4097 set_prefree_as_free_segments(sbi);
4100 static int build_sit_info(struct f2fs_sb_info *sbi)
4102 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4103 struct sit_info *sit_i;
4104 unsigned int sit_segs, start;
4105 char *src_bitmap, *bitmap;
4106 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4108 /* allocate memory for SIT information */
4109 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4113 SM_I(sbi)->sit_info = sit_i;
4116 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4119 if (!sit_i->sentries)
4122 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4123 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4125 if (!sit_i->dirty_sentries_bitmap)
4128 #ifdef CONFIG_F2FS_CHECK_FS
4129 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4131 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4133 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4137 bitmap = sit_i->bitmap;
4139 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4140 sit_i->sentries[start].cur_valid_map = bitmap;
4141 bitmap += SIT_VBLOCK_MAP_SIZE;
4143 sit_i->sentries[start].ckpt_valid_map = bitmap;
4144 bitmap += SIT_VBLOCK_MAP_SIZE;
4146 #ifdef CONFIG_F2FS_CHECK_FS
4147 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4148 bitmap += SIT_VBLOCK_MAP_SIZE;
4151 sit_i->sentries[start].discard_map = bitmap;
4152 bitmap += SIT_VBLOCK_MAP_SIZE;
4155 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4156 if (!sit_i->tmp_map)
4159 if (__is_large_section(sbi)) {
4160 sit_i->sec_entries =
4161 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4164 if (!sit_i->sec_entries)
4168 /* get information related with SIT */
4169 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4171 /* setup SIT bitmap from ckeckpoint pack */
4172 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4173 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4175 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4176 if (!sit_i->sit_bitmap)
4179 #ifdef CONFIG_F2FS_CHECK_FS
4180 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4181 sit_bitmap_size, GFP_KERNEL);
4182 if (!sit_i->sit_bitmap_mir)
4185 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4186 main_bitmap_size, GFP_KERNEL);
4187 if (!sit_i->invalid_segmap)
4191 /* init SIT information */
4192 sit_i->s_ops = &default_salloc_ops;
4194 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4195 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4196 sit_i->written_valid_blocks = 0;
4197 sit_i->bitmap_size = sit_bitmap_size;
4198 sit_i->dirty_sentries = 0;
4199 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4200 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4201 sit_i->mounted_time = ktime_get_boottime_seconds();
4202 init_rwsem(&sit_i->sentry_lock);
4206 static int build_free_segmap(struct f2fs_sb_info *sbi)
4208 struct free_segmap_info *free_i;
4209 unsigned int bitmap_size, sec_bitmap_size;
4211 /* allocate memory for free segmap information */
4212 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4216 SM_I(sbi)->free_info = free_i;
4218 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4219 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4220 if (!free_i->free_segmap)
4223 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4224 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4225 if (!free_i->free_secmap)
4228 /* set all segments as dirty temporarily */
4229 memset(free_i->free_segmap, 0xff, bitmap_size);
4230 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4232 /* init free segmap information */
4233 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4234 free_i->free_segments = 0;
4235 free_i->free_sections = 0;
4236 spin_lock_init(&free_i->segmap_lock);
4240 static int build_curseg(struct f2fs_sb_info *sbi)
4242 struct curseg_info *array;
4245 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4246 sizeof(*array)), GFP_KERNEL);
4250 SM_I(sbi)->curseg_array = array;
4252 for (i = 0; i < NO_CHECK_TYPE; i++) {
4253 mutex_init(&array[i].curseg_mutex);
4254 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4255 if (!array[i].sum_blk)
4257 init_rwsem(&array[i].journal_rwsem);
4258 array[i].journal = f2fs_kzalloc(sbi,
4259 sizeof(struct f2fs_journal), GFP_KERNEL);
4260 if (!array[i].journal)
4262 if (i < NR_PERSISTENT_LOG)
4263 array[i].seg_type = CURSEG_HOT_DATA + i;
4264 else if (i == CURSEG_COLD_DATA_PINNED)
4265 array[i].seg_type = CURSEG_COLD_DATA;
4266 array[i].segno = NULL_SEGNO;
4267 array[i].next_blkoff = 0;
4268 array[i].inited = false;
4270 return restore_curseg_summaries(sbi);
4273 static int build_sit_entries(struct f2fs_sb_info *sbi)
4275 struct sit_info *sit_i = SIT_I(sbi);
4276 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4277 struct f2fs_journal *journal = curseg->journal;
4278 struct seg_entry *se;
4279 struct f2fs_sit_entry sit;
4280 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4281 unsigned int i, start, end;
4282 unsigned int readed, start_blk = 0;
4284 block_t total_node_blocks = 0;
4287 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4290 start = start_blk * sit_i->sents_per_block;
4291 end = (start_blk + readed) * sit_i->sents_per_block;
4293 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4294 struct f2fs_sit_block *sit_blk;
4297 se = &sit_i->sentries[start];
4298 page = get_current_sit_page(sbi, start);
4300 return PTR_ERR(page);
4301 sit_blk = (struct f2fs_sit_block *)page_address(page);
4302 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4303 f2fs_put_page(page, 1);
4305 err = check_block_count(sbi, start, &sit);
4308 seg_info_from_raw_sit(se, &sit);
4309 if (IS_NODESEG(se->type))
4310 total_node_blocks += se->valid_blocks;
4312 /* build discard map only one time */
4313 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4314 memset(se->discard_map, 0xff,
4315 SIT_VBLOCK_MAP_SIZE);
4317 memcpy(se->discard_map,
4319 SIT_VBLOCK_MAP_SIZE);
4320 sbi->discard_blks +=
4321 sbi->blocks_per_seg -
4325 if (__is_large_section(sbi))
4326 get_sec_entry(sbi, start)->valid_blocks +=
4329 start_blk += readed;
4330 } while (start_blk < sit_blk_cnt);
4332 down_read(&curseg->journal_rwsem);
4333 for (i = 0; i < sits_in_cursum(journal); i++) {
4334 unsigned int old_valid_blocks;
4336 start = le32_to_cpu(segno_in_journal(journal, i));
4337 if (start >= MAIN_SEGS(sbi)) {
4338 f2fs_err(sbi, "Wrong journal entry on segno %u",
4340 err = -EFSCORRUPTED;
4344 se = &sit_i->sentries[start];
4345 sit = sit_in_journal(journal, i);
4347 old_valid_blocks = se->valid_blocks;
4348 if (IS_NODESEG(se->type))
4349 total_node_blocks -= old_valid_blocks;
4351 err = check_block_count(sbi, start, &sit);
4354 seg_info_from_raw_sit(se, &sit);
4355 if (IS_NODESEG(se->type))
4356 total_node_blocks += se->valid_blocks;
4358 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4359 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4361 memcpy(se->discard_map, se->cur_valid_map,
4362 SIT_VBLOCK_MAP_SIZE);
4363 sbi->discard_blks += old_valid_blocks;
4364 sbi->discard_blks -= se->valid_blocks;
4367 if (__is_large_section(sbi)) {
4368 get_sec_entry(sbi, start)->valid_blocks +=
4370 get_sec_entry(sbi, start)->valid_blocks -=
4374 up_read(&curseg->journal_rwsem);
4376 if (!err && total_node_blocks != valid_node_count(sbi)) {
4377 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4378 total_node_blocks, valid_node_count(sbi));
4379 err = -EFSCORRUPTED;
4385 static void init_free_segmap(struct f2fs_sb_info *sbi)
4389 struct seg_entry *sentry;
4391 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4392 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4394 sentry = get_seg_entry(sbi, start);
4395 if (!sentry->valid_blocks)
4396 __set_free(sbi, start);
4398 SIT_I(sbi)->written_valid_blocks +=
4399 sentry->valid_blocks;
4402 /* set use the current segments */
4403 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4404 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4405 __set_test_and_inuse(sbi, curseg_t->segno);
4409 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4411 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4412 struct free_segmap_info *free_i = FREE_I(sbi);
4413 unsigned int segno = 0, offset = 0, secno;
4414 block_t valid_blocks, usable_blks_in_seg;
4415 block_t blks_per_sec = BLKS_PER_SEC(sbi);
4418 /* find dirty segment based on free segmap */
4419 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4420 if (segno >= MAIN_SEGS(sbi))
4423 valid_blocks = get_valid_blocks(sbi, segno, false);
4424 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4425 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4427 if (valid_blocks > usable_blks_in_seg) {
4428 f2fs_bug_on(sbi, 1);
4431 mutex_lock(&dirty_i->seglist_lock);
4432 __locate_dirty_segment(sbi, segno, DIRTY);
4433 mutex_unlock(&dirty_i->seglist_lock);
4436 if (!__is_large_section(sbi))
4439 mutex_lock(&dirty_i->seglist_lock);
4440 for (segno = 0; segno < MAIN_SECS(sbi); segno += blks_per_sec) {
4441 valid_blocks = get_valid_blocks(sbi, segno, true);
4442 secno = GET_SEC_FROM_SEG(sbi, segno);
4444 if (!valid_blocks || valid_blocks == blks_per_sec)
4446 if (IS_CURSEC(sbi, secno))
4448 set_bit(secno, dirty_i->dirty_secmap);
4450 mutex_unlock(&dirty_i->seglist_lock);
4453 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4455 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4456 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4458 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4459 if (!dirty_i->victim_secmap)
4464 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4466 struct dirty_seglist_info *dirty_i;
4467 unsigned int bitmap_size, i;
4469 /* allocate memory for dirty segments list information */
4470 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4475 SM_I(sbi)->dirty_info = dirty_i;
4476 mutex_init(&dirty_i->seglist_lock);
4478 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4480 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4481 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4483 if (!dirty_i->dirty_segmap[i])
4487 if (__is_large_section(sbi)) {
4488 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4489 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4490 bitmap_size, GFP_KERNEL);
4491 if (!dirty_i->dirty_secmap)
4495 init_dirty_segmap(sbi);
4496 return init_victim_secmap(sbi);
4499 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4504 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4505 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4507 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4508 struct curseg_info *curseg = CURSEG_I(sbi, i);
4509 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4510 unsigned int blkofs = curseg->next_blkoff;
4512 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4515 if (curseg->alloc_type == SSR)
4518 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4519 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4523 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4524 i, curseg->segno, curseg->alloc_type,
4525 curseg->next_blkoff, blkofs);
4526 return -EFSCORRUPTED;
4532 #ifdef CONFIG_BLK_DEV_ZONED
4534 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4535 struct f2fs_dev_info *fdev,
4536 struct blk_zone *zone)
4538 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4539 block_t zone_block, wp_block, last_valid_block;
4540 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4542 struct seg_entry *se;
4544 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4547 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4548 wp_segno = GET_SEGNO(sbi, wp_block);
4549 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4550 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4551 zone_segno = GET_SEGNO(sbi, zone_block);
4552 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4554 if (zone_segno >= MAIN_SEGS(sbi))
4558 * Skip check of zones cursegs point to, since
4559 * fix_curseg_write_pointer() checks them.
4561 for (i = 0; i < NO_CHECK_TYPE; i++)
4562 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4563 CURSEG_I(sbi, i)->segno))
4567 * Get last valid block of the zone.
4569 last_valid_block = zone_block - 1;
4570 for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4571 segno = zone_segno + s;
4572 se = get_seg_entry(sbi, segno);
4573 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4574 if (f2fs_test_bit(b, se->cur_valid_map)) {
4575 last_valid_block = START_BLOCK(sbi, segno) + b;
4578 if (last_valid_block >= zone_block)
4583 * If last valid block is beyond the write pointer, report the
4584 * inconsistency. This inconsistency does not cause write error
4585 * because the zone will not be selected for write operation until
4586 * it get discarded. Just report it.
4588 if (last_valid_block >= wp_block) {
4589 f2fs_notice(sbi, "Valid block beyond write pointer: "
4590 "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4591 GET_SEGNO(sbi, last_valid_block),
4592 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4593 wp_segno, wp_blkoff);
4598 * If there is no valid block in the zone and if write pointer is
4599 * not at zone start, reset the write pointer.
4601 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4603 "Zone without valid block has non-zero write "
4604 "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4605 wp_segno, wp_blkoff);
4606 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4607 zone->len >> log_sectors_per_block);
4609 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4618 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4619 block_t zone_blkaddr)
4623 for (i = 0; i < sbi->s_ndevs; i++) {
4624 if (!bdev_is_zoned(FDEV(i).bdev))
4626 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4627 zone_blkaddr <= FDEV(i).end_blk))
4634 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4636 memcpy(data, zone, sizeof(struct blk_zone));
4640 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4642 struct curseg_info *cs = CURSEG_I(sbi, type);
4643 struct f2fs_dev_info *zbd;
4644 struct blk_zone zone;
4645 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4646 block_t cs_zone_block, wp_block;
4647 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4648 sector_t zone_sector;
4651 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4652 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4654 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4658 /* report zone for the sector the curseg points to */
4659 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4660 << log_sectors_per_block;
4661 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4662 report_one_zone_cb, &zone);
4664 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4669 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4672 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4673 wp_segno = GET_SEGNO(sbi, wp_block);
4674 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4675 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4677 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4681 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4682 "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4683 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4685 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4686 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4687 allocate_segment_by_default(sbi, type, true);
4689 /* check consistency of the zone curseg pointed to */
4690 if (check_zone_write_pointer(sbi, zbd, &zone))
4693 /* check newly assigned zone */
4694 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4695 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4697 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4701 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4702 << log_sectors_per_block;
4703 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4704 report_one_zone_cb, &zone);
4706 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4711 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4714 if (zone.wp != zone.start) {
4716 "New zone for curseg[%d] is not yet discarded. "
4717 "Reset the zone: curseg[0x%x,0x%x]",
4718 type, cs->segno, cs->next_blkoff);
4719 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4720 zone_sector >> log_sectors_per_block,
4721 zone.len >> log_sectors_per_block);
4723 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4732 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4736 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4737 ret = fix_curseg_write_pointer(sbi, i);
4745 struct check_zone_write_pointer_args {
4746 struct f2fs_sb_info *sbi;
4747 struct f2fs_dev_info *fdev;
4750 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4752 struct check_zone_write_pointer_args *args;
4753 args = (struct check_zone_write_pointer_args *)data;
4755 return check_zone_write_pointer(args->sbi, args->fdev, zone);
4758 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4761 struct check_zone_write_pointer_args args;
4763 for (i = 0; i < sbi->s_ndevs; i++) {
4764 if (!bdev_is_zoned(FDEV(i).bdev))
4768 args.fdev = &FDEV(i);
4769 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4770 check_zone_write_pointer_cb, &args);
4778 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4779 unsigned int dev_idx)
4781 if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4783 return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4786 /* Return the zone index in the given device */
4787 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4790 block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4792 return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4793 sbi->log_blocks_per_blkz;
4797 * Return the usable segments in a section based on the zone's
4798 * corresponding zone capacity. Zone is equal to a section.
4800 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4801 struct f2fs_sb_info *sbi, unsigned int segno)
4803 unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4805 dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4806 zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4808 /* Conventional zone's capacity is always equal to zone size */
4809 if (is_conv_zone(sbi, zone_idx, dev_idx))
4810 return sbi->segs_per_sec;
4813 * If the zone_capacity_blocks array is NULL, then zone capacity
4814 * is equal to the zone size for all zones
4816 if (!FDEV(dev_idx).zone_capacity_blocks)
4817 return sbi->segs_per_sec;
4819 /* Get the segment count beyond zone capacity block */
4820 unusable_segs_in_sec = (sbi->blocks_per_blkz -
4821 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
4822 sbi->log_blocks_per_seg;
4823 return sbi->segs_per_sec - unusable_segs_in_sec;
4827 * Return the number of usable blocks in a segment. The number of blocks
4828 * returned is always equal to the number of blocks in a segment for
4829 * segments fully contained within a sequential zone capacity or a
4830 * conventional zone. For segments partially contained in a sequential
4831 * zone capacity, the number of usable blocks up to the zone capacity
4832 * is returned. 0 is returned in all other cases.
4834 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4835 struct f2fs_sb_info *sbi, unsigned int segno)
4837 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
4838 unsigned int zone_idx, dev_idx, secno;
4840 secno = GET_SEC_FROM_SEG(sbi, segno);
4841 seg_start = START_BLOCK(sbi, segno);
4842 dev_idx = f2fs_target_device_index(sbi, seg_start);
4843 zone_idx = get_zone_idx(sbi, secno, dev_idx);
4846 * Conventional zone's capacity is always equal to zone size,
4847 * so, blocks per segment is unchanged.
4849 if (is_conv_zone(sbi, zone_idx, dev_idx))
4850 return sbi->blocks_per_seg;
4852 if (!FDEV(dev_idx).zone_capacity_blocks)
4853 return sbi->blocks_per_seg;
4855 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4856 sec_cap_blkaddr = sec_start_blkaddr +
4857 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
4860 * If segment starts before zone capacity and spans beyond
4861 * zone capacity, then usable blocks are from seg start to
4862 * zone capacity. If the segment starts after the zone capacity,
4863 * then there are no usable blocks.
4865 if (seg_start >= sec_cap_blkaddr)
4867 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
4868 return sec_cap_blkaddr - seg_start;
4870 return sbi->blocks_per_seg;
4873 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4878 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4883 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
4889 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
4895 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
4898 if (f2fs_sb_has_blkzoned(sbi))
4899 return f2fs_usable_zone_blks_in_seg(sbi, segno);
4901 return sbi->blocks_per_seg;
4904 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
4907 if (f2fs_sb_has_blkzoned(sbi))
4908 return f2fs_usable_zone_segs_in_sec(sbi, segno);
4910 return sbi->segs_per_sec;
4914 * Update min, max modified time for cost-benefit GC algorithm
4916 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4918 struct sit_info *sit_i = SIT_I(sbi);
4921 down_write(&sit_i->sentry_lock);
4923 sit_i->min_mtime = ULLONG_MAX;
4925 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4927 unsigned long long mtime = 0;
4929 for (i = 0; i < sbi->segs_per_sec; i++)
4930 mtime += get_seg_entry(sbi, segno + i)->mtime;
4932 mtime = div_u64(mtime, sbi->segs_per_sec);
4934 if (sit_i->min_mtime > mtime)
4935 sit_i->min_mtime = mtime;
4937 sit_i->max_mtime = get_mtime(sbi, false);
4938 up_write(&sit_i->sentry_lock);
4941 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4943 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4944 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4945 struct f2fs_sm_info *sm_info;
4948 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4953 sbi->sm_info = sm_info;
4954 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4955 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4956 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4957 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4958 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4959 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4960 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4961 sm_info->rec_prefree_segments = sm_info->main_segments *
4962 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4963 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4964 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4966 if (!f2fs_lfs_mode(sbi))
4967 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4968 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4969 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4970 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4971 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4972 sm_info->min_ssr_sections = reserved_sections(sbi);
4974 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4976 init_rwsem(&sm_info->curseg_lock);
4978 if (!f2fs_readonly(sbi->sb)) {
4979 err = f2fs_create_flush_cmd_control(sbi);
4984 err = create_discard_cmd_control(sbi);
4988 err = build_sit_info(sbi);
4991 err = build_free_segmap(sbi);
4994 err = build_curseg(sbi);
4998 /* reinit free segmap based on SIT */
4999 err = build_sit_entries(sbi);
5003 init_free_segmap(sbi);
5004 err = build_dirty_segmap(sbi);
5008 err = sanity_check_curseg(sbi);
5012 init_min_max_mtime(sbi);
5016 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5017 enum dirty_type dirty_type)
5019 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5021 mutex_lock(&dirty_i->seglist_lock);
5022 kvfree(dirty_i->dirty_segmap[dirty_type]);
5023 dirty_i->nr_dirty[dirty_type] = 0;
5024 mutex_unlock(&dirty_i->seglist_lock);
5027 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5029 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5030 kvfree(dirty_i->victim_secmap);
5033 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5035 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5041 /* discard pre-free/dirty segments list */
5042 for (i = 0; i < NR_DIRTY_TYPE; i++)
5043 discard_dirty_segmap(sbi, i);
5045 if (__is_large_section(sbi)) {
5046 mutex_lock(&dirty_i->seglist_lock);
5047 kvfree(dirty_i->dirty_secmap);
5048 mutex_unlock(&dirty_i->seglist_lock);
5051 destroy_victim_secmap(sbi);
5052 SM_I(sbi)->dirty_info = NULL;
5056 static void destroy_curseg(struct f2fs_sb_info *sbi)
5058 struct curseg_info *array = SM_I(sbi)->curseg_array;
5063 SM_I(sbi)->curseg_array = NULL;
5064 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5065 kvfree(array[i].sum_blk);
5066 kvfree(array[i].journal);
5071 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5073 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5076 SM_I(sbi)->free_info = NULL;
5077 kvfree(free_i->free_segmap);
5078 kvfree(free_i->free_secmap);
5082 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5084 struct sit_info *sit_i = SIT_I(sbi);
5089 if (sit_i->sentries)
5090 kvfree(sit_i->bitmap);
5091 kvfree(sit_i->tmp_map);
5093 kvfree(sit_i->sentries);
5094 kvfree(sit_i->sec_entries);
5095 kvfree(sit_i->dirty_sentries_bitmap);
5097 SM_I(sbi)->sit_info = NULL;
5098 kvfree(sit_i->sit_bitmap);
5099 #ifdef CONFIG_F2FS_CHECK_FS
5100 kvfree(sit_i->sit_bitmap_mir);
5101 kvfree(sit_i->invalid_segmap);
5106 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5108 struct f2fs_sm_info *sm_info = SM_I(sbi);
5112 f2fs_destroy_flush_cmd_control(sbi, true);
5113 destroy_discard_cmd_control(sbi);
5114 destroy_dirty_segmap(sbi);
5115 destroy_curseg(sbi);
5116 destroy_free_segmap(sbi);
5117 destroy_sit_info(sbi);
5118 sbi->sm_info = NULL;
5122 int __init f2fs_create_segment_manager_caches(void)
5124 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5125 sizeof(struct discard_entry));
5126 if (!discard_entry_slab)
5129 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5130 sizeof(struct discard_cmd));
5131 if (!discard_cmd_slab)
5132 goto destroy_discard_entry;
5134 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5135 sizeof(struct sit_entry_set));
5136 if (!sit_entry_set_slab)
5137 goto destroy_discard_cmd;
5139 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5140 sizeof(struct inmem_pages));
5141 if (!inmem_entry_slab)
5142 goto destroy_sit_entry_set;
5145 destroy_sit_entry_set:
5146 kmem_cache_destroy(sit_entry_set_slab);
5147 destroy_discard_cmd:
5148 kmem_cache_destroy(discard_cmd_slab);
5149 destroy_discard_entry:
5150 kmem_cache_destroy(discard_entry_slab);
5155 void f2fs_destroy_segment_manager_caches(void)
5157 kmem_cache_destroy(sit_entry_set_slab);
5158 kmem_cache_destroy(discard_cmd_slab);
5159 kmem_cache_destroy(discard_entry_slab);
5160 kmem_cache_destroy(inmem_entry_slab);