1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/sched/mm.h>
13 #include <linux/prefetch.h>
14 #include <linux/kthread.h>
15 #include <linux/swap.h>
16 #include <linux/timer.h>
17 #include <linux/freezer.h>
18 #include <linux/sched/signal.h>
19 #include <linux/random.h>
26 #include <trace/events/f2fs.h>
28 #define __reverse_ffz(x) __reverse_ffs(~(x))
30 static struct kmem_cache *discard_entry_slab;
31 static struct kmem_cache *discard_cmd_slab;
32 static struct kmem_cache *sit_entry_set_slab;
33 static struct kmem_cache *inmem_entry_slab;
35 static unsigned long __reverse_ulong(unsigned char *str)
37 unsigned long tmp = 0;
38 int shift = 24, idx = 0;
40 #if BITS_PER_LONG == 64
44 tmp |= (unsigned long)str[idx++] << shift;
45 shift -= BITS_PER_BYTE;
51 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
52 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 static inline unsigned long __reverse_ffs(unsigned long word)
58 #if BITS_PER_LONG == 64
59 if ((word & 0xffffffff00000000UL) == 0)
64 if ((word & 0xffff0000) == 0)
69 if ((word & 0xff00) == 0)
74 if ((word & 0xf0) == 0)
79 if ((word & 0xc) == 0)
84 if ((word & 0x2) == 0)
90 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
91 * f2fs_set_bit makes MSB and LSB reversed in a byte.
92 * @size must be integral times of unsigned long.
95 * f2fs_set_bit(0, bitmap) => 1000 0000
96 * f2fs_set_bit(7, bitmap) => 0000 0001
98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
99 unsigned long size, unsigned long offset)
101 const unsigned long *p = addr + BIT_WORD(offset);
102 unsigned long result = size;
108 size -= (offset & ~(BITS_PER_LONG - 1));
109 offset %= BITS_PER_LONG;
115 tmp = __reverse_ulong((unsigned char *)p);
117 tmp &= ~0UL >> offset;
118 if (size < BITS_PER_LONG)
119 tmp &= (~0UL << (BITS_PER_LONG - size));
123 if (size <= BITS_PER_LONG)
125 size -= BITS_PER_LONG;
131 return result - size + __reverse_ffs(tmp);
134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 unsigned long size, unsigned long offset)
137 const unsigned long *p = addr + BIT_WORD(offset);
138 unsigned long result = size;
144 size -= (offset & ~(BITS_PER_LONG - 1));
145 offset %= BITS_PER_LONG;
151 tmp = __reverse_ulong((unsigned char *)p);
154 tmp |= ~0UL << (BITS_PER_LONG - offset);
155 if (size < BITS_PER_LONG)
160 if (size <= BITS_PER_LONG)
162 size -= BITS_PER_LONG;
168 return result - size + __reverse_ffz(tmp);
171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
173 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
174 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
175 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177 if (f2fs_lfs_mode(sbi))
179 if (sbi->gc_mode == GC_URGENT_HIGH)
181 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
184 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
185 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
188 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
190 struct inmem_pages *new;
192 set_page_private_atomic(page);
194 new = f2fs_kmem_cache_alloc(inmem_entry_slab,
195 GFP_NOFS, true, NULL);
197 /* add atomic page indices to the list */
199 INIT_LIST_HEAD(&new->list);
201 /* increase reference count with clean state */
203 mutex_lock(&F2FS_I(inode)->inmem_lock);
204 list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
205 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
206 mutex_unlock(&F2FS_I(inode)->inmem_lock);
208 trace_f2fs_register_inmem_page(page, INMEM);
211 static int __revoke_inmem_pages(struct inode *inode,
212 struct list_head *head, bool drop, bool recover,
215 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
216 struct inmem_pages *cur, *tmp;
219 list_for_each_entry_safe(cur, tmp, head, list) {
220 struct page *page = cur->page;
223 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
227 * to avoid deadlock in between page lock and
230 if (!trylock_page(page))
236 f2fs_wait_on_page_writeback(page, DATA, true, true);
239 struct dnode_of_data dn;
242 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
244 set_new_dnode(&dn, inode, NULL, NULL, 0);
245 err = f2fs_get_dnode_of_data(&dn, page->index,
248 if (err == -ENOMEM) {
249 memalloc_retry_wait(GFP_NOFS);
256 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
262 if (cur->old_addr == NEW_ADDR) {
263 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
264 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
266 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
267 cur->old_addr, ni.version, true, true);
271 /* we don't need to invalidate this in the sccessful status */
272 if (drop || recover) {
273 ClearPageUptodate(page);
274 clear_page_private_gcing(page);
276 detach_page_private(page);
277 set_page_private(page, 0);
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);
331 mutex_lock(&fi->inmem_lock);
332 if (list_empty(&fi->inmem_pages)) {
333 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
335 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
336 if (!list_empty(&fi->inmem_ilist))
337 list_del_init(&fi->inmem_ilist);
338 if (f2fs_is_atomic_file(inode)) {
339 clear_inode_flag(inode, FI_ATOMIC_FILE);
342 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
344 mutex_unlock(&fi->inmem_lock);
347 __revoke_inmem_pages(inode, &fi->inmem_pages,
349 mutex_unlock(&fi->inmem_lock);
353 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
355 struct f2fs_inode_info *fi = F2FS_I(inode);
356 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
357 struct list_head *head = &fi->inmem_pages;
358 struct inmem_pages *cur = NULL;
360 f2fs_bug_on(sbi, !page_private_atomic(page));
362 mutex_lock(&fi->inmem_lock);
363 list_for_each_entry(cur, head, list) {
364 if (cur->page == page)
368 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
369 list_del(&cur->list);
370 mutex_unlock(&fi->inmem_lock);
372 dec_page_count(sbi, F2FS_INMEM_PAGES);
373 kmem_cache_free(inmem_entry_slab, cur);
375 ClearPageUptodate(page);
376 clear_page_private_atomic(page);
377 f2fs_put_page(page, 0);
379 detach_page_private(page);
380 set_page_private(page, 0);
382 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
385 static int __f2fs_commit_inmem_pages(struct inode *inode)
387 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
388 struct f2fs_inode_info *fi = F2FS_I(inode);
389 struct inmem_pages *cur, *tmp;
390 struct f2fs_io_info fio = {
395 .op_flags = REQ_SYNC | REQ_PRIO,
396 .io_type = FS_DATA_IO,
398 struct list_head revoke_list;
399 bool submit_bio = false;
402 INIT_LIST_HEAD(&revoke_list);
404 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
405 struct page *page = cur->page;
408 if (page->mapping == inode->i_mapping) {
409 trace_f2fs_commit_inmem_page(page, INMEM);
411 f2fs_wait_on_page_writeback(page, DATA, true, true);
413 set_page_dirty(page);
414 if (clear_page_dirty_for_io(page)) {
415 inode_dec_dirty_pages(inode);
416 f2fs_remove_dirty_inode(inode);
420 fio.old_blkaddr = NULL_ADDR;
421 fio.encrypted_page = NULL;
422 fio.need_lock = LOCK_DONE;
423 err = f2fs_do_write_data_page(&fio);
425 if (err == -ENOMEM) {
426 memalloc_retry_wait(GFP_NOFS);
432 /* record old blkaddr for revoking */
433 cur->old_addr = fio.old_blkaddr;
437 list_move_tail(&cur->list, &revoke_list);
441 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
445 * try to revoke all committed pages, but still we could fail
446 * due to no memory or other reason, if that happened, EAGAIN
447 * will be returned, which means in such case, transaction is
448 * already not integrity, caller should use journal to do the
449 * recovery or rewrite & commit last transaction. For other
450 * error number, revoking was done by filesystem itself.
452 err = __revoke_inmem_pages(inode, &revoke_list,
455 /* drop all uncommitted pages */
456 __revoke_inmem_pages(inode, &fi->inmem_pages,
459 __revoke_inmem_pages(inode, &revoke_list,
460 false, false, false);
466 int f2fs_commit_inmem_pages(struct inode *inode)
468 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
469 struct f2fs_inode_info *fi = F2FS_I(inode);
472 f2fs_balance_fs(sbi, true);
474 f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
477 set_inode_flag(inode, FI_ATOMIC_COMMIT);
479 mutex_lock(&fi->inmem_lock);
480 err = __f2fs_commit_inmem_pages(inode);
481 mutex_unlock(&fi->inmem_lock);
483 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
486 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
492 * This function balances dirty node and dentry pages.
493 * In addition, it controls garbage collection.
495 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
497 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
498 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
499 f2fs_stop_checkpoint(sbi, false);
502 /* balance_fs_bg is able to be pending */
503 if (need && excess_cached_nats(sbi))
504 f2fs_balance_fs_bg(sbi, false);
506 if (!f2fs_is_checkpoint_ready(sbi))
510 * We should do GC or end up with checkpoint, if there are so many dirty
511 * dir/node pages without enough free segments.
513 if (has_not_enough_free_secs(sbi, 0, 0)) {
514 if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
515 sbi->gc_thread->f2fs_gc_task) {
518 prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
519 TASK_UNINTERRUPTIBLE);
520 wake_up(&sbi->gc_thread->gc_wait_queue_head);
522 finish_wait(&sbi->gc_thread->fggc_wq, &wait);
524 f2fs_down_write(&sbi->gc_lock);
525 f2fs_gc(sbi, false, false, false, NULL_SEGNO);
530 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
532 int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
533 unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
534 unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
535 unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
536 unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
537 unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
538 unsigned int threshold = sbi->blocks_per_seg * factor *
539 DEFAULT_DIRTY_THRESHOLD;
540 unsigned int global_threshold = threshold * 3 / 2;
542 if (dents >= threshold || qdata >= threshold ||
543 nodes >= threshold || meta >= threshold ||
546 return dents + qdata + nodes + meta + imeta > global_threshold;
549 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
551 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
554 /* try to shrink extent cache when there is no enough memory */
555 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
556 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
558 /* check the # of cached NAT entries */
559 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
560 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
562 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
563 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
565 f2fs_build_free_nids(sbi, false, false);
567 if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
568 excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
571 /* there is background inflight IO or foreground operation recently */
572 if (is_inflight_io(sbi, REQ_TIME) ||
573 (!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
576 /* exceed periodical checkpoint timeout threshold */
577 if (f2fs_time_over(sbi, CP_TIME))
580 /* checkpoint is the only way to shrink partial cached entries */
581 if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
582 f2fs_available_free_memory(sbi, INO_ENTRIES))
586 if (test_opt(sbi, DATA_FLUSH) && from_bg) {
587 struct blk_plug plug;
589 mutex_lock(&sbi->flush_lock);
591 blk_start_plug(&plug);
592 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
593 blk_finish_plug(&plug);
595 mutex_unlock(&sbi->flush_lock);
597 f2fs_sync_fs(sbi->sb, true);
598 stat_inc_bg_cp_count(sbi->stat_info);
601 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
602 struct block_device *bdev)
604 int ret = blkdev_issue_flush(bdev);
606 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
607 test_opt(sbi, FLUSH_MERGE), ret);
611 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
616 if (!f2fs_is_multi_device(sbi))
617 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
619 for (i = 0; i < sbi->s_ndevs; i++) {
620 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
622 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
629 static int issue_flush_thread(void *data)
631 struct f2fs_sb_info *sbi = data;
632 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
633 wait_queue_head_t *q = &fcc->flush_wait_queue;
635 if (kthread_should_stop())
638 if (!llist_empty(&fcc->issue_list)) {
639 struct flush_cmd *cmd, *next;
642 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
643 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
645 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
647 ret = submit_flush_wait(sbi, cmd->ino);
648 atomic_inc(&fcc->issued_flush);
650 llist_for_each_entry_safe(cmd, next,
651 fcc->dispatch_list, llnode) {
653 complete(&cmd->wait);
655 fcc->dispatch_list = NULL;
658 wait_event_interruptible(*q,
659 kthread_should_stop() || !llist_empty(&fcc->issue_list));
663 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
665 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
666 struct flush_cmd cmd;
669 if (test_opt(sbi, NOBARRIER))
672 if (!test_opt(sbi, FLUSH_MERGE)) {
673 atomic_inc(&fcc->queued_flush);
674 ret = submit_flush_wait(sbi, ino);
675 atomic_dec(&fcc->queued_flush);
676 atomic_inc(&fcc->issued_flush);
680 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
681 f2fs_is_multi_device(sbi)) {
682 ret = submit_flush_wait(sbi, ino);
683 atomic_dec(&fcc->queued_flush);
685 atomic_inc(&fcc->issued_flush);
690 init_completion(&cmd.wait);
692 llist_add(&cmd.llnode, &fcc->issue_list);
695 * update issue_list before we wake up issue_flush thread, this
696 * smp_mb() pairs with another barrier in ___wait_event(), see
697 * more details in comments of waitqueue_active().
701 if (waitqueue_active(&fcc->flush_wait_queue))
702 wake_up(&fcc->flush_wait_queue);
704 if (fcc->f2fs_issue_flush) {
705 wait_for_completion(&cmd.wait);
706 atomic_dec(&fcc->queued_flush);
708 struct llist_node *list;
710 list = llist_del_all(&fcc->issue_list);
712 wait_for_completion(&cmd.wait);
713 atomic_dec(&fcc->queued_flush);
715 struct flush_cmd *tmp, *next;
717 ret = submit_flush_wait(sbi, ino);
719 llist_for_each_entry_safe(tmp, next, list, llnode) {
722 atomic_dec(&fcc->queued_flush);
726 complete(&tmp->wait);
734 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
736 dev_t dev = sbi->sb->s_bdev->bd_dev;
737 struct flush_cmd_control *fcc;
740 if (SM_I(sbi)->fcc_info) {
741 fcc = SM_I(sbi)->fcc_info;
742 if (fcc->f2fs_issue_flush)
747 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
750 atomic_set(&fcc->issued_flush, 0);
751 atomic_set(&fcc->queued_flush, 0);
752 init_waitqueue_head(&fcc->flush_wait_queue);
753 init_llist_head(&fcc->issue_list);
754 SM_I(sbi)->fcc_info = fcc;
755 if (!test_opt(sbi, FLUSH_MERGE))
759 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
760 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
761 if (IS_ERR(fcc->f2fs_issue_flush)) {
762 err = PTR_ERR(fcc->f2fs_issue_flush);
764 SM_I(sbi)->fcc_info = NULL;
771 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
773 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
775 if (fcc && fcc->f2fs_issue_flush) {
776 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
778 fcc->f2fs_issue_flush = NULL;
779 kthread_stop(flush_thread);
783 SM_I(sbi)->fcc_info = NULL;
787 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
791 if (!f2fs_is_multi_device(sbi))
794 if (test_opt(sbi, NOBARRIER))
797 for (i = 1; i < sbi->s_ndevs; i++) {
798 int count = DEFAULT_RETRY_IO_COUNT;
800 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
804 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
806 congestion_wait(BLK_RW_ASYNC,
808 } while (ret && --count);
811 f2fs_stop_checkpoint(sbi, false);
815 spin_lock(&sbi->dev_lock);
816 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
817 spin_unlock(&sbi->dev_lock);
823 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
824 enum dirty_type dirty_type)
826 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
828 /* need not be added */
829 if (IS_CURSEG(sbi, segno))
832 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
833 dirty_i->nr_dirty[dirty_type]++;
835 if (dirty_type == DIRTY) {
836 struct seg_entry *sentry = get_seg_entry(sbi, segno);
837 enum dirty_type t = sentry->type;
839 if (unlikely(t >= DIRTY)) {
843 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
844 dirty_i->nr_dirty[t]++;
846 if (__is_large_section(sbi)) {
847 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
848 block_t valid_blocks =
849 get_valid_blocks(sbi, segno, true);
851 f2fs_bug_on(sbi, unlikely(!valid_blocks ||
852 valid_blocks == BLKS_PER_SEC(sbi)));
854 if (!IS_CURSEC(sbi, secno))
855 set_bit(secno, dirty_i->dirty_secmap);
860 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
861 enum dirty_type dirty_type)
863 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
864 block_t valid_blocks;
866 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
867 dirty_i->nr_dirty[dirty_type]--;
869 if (dirty_type == DIRTY) {
870 struct seg_entry *sentry = get_seg_entry(sbi, segno);
871 enum dirty_type t = sentry->type;
873 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
874 dirty_i->nr_dirty[t]--;
876 valid_blocks = get_valid_blocks(sbi, segno, true);
877 if (valid_blocks == 0) {
878 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
879 dirty_i->victim_secmap);
880 #ifdef CONFIG_F2FS_CHECK_FS
881 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
884 if (__is_large_section(sbi)) {
885 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
888 valid_blocks == BLKS_PER_SEC(sbi)) {
889 clear_bit(secno, dirty_i->dirty_secmap);
893 if (!IS_CURSEC(sbi, secno))
894 set_bit(secno, dirty_i->dirty_secmap);
900 * Should not occur error such as -ENOMEM.
901 * Adding dirty entry into seglist is not critical operation.
902 * If a given segment is one of current working segments, it won't be added.
904 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
906 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
907 unsigned short valid_blocks, ckpt_valid_blocks;
908 unsigned int usable_blocks;
910 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
913 usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
914 mutex_lock(&dirty_i->seglist_lock);
916 valid_blocks = get_valid_blocks(sbi, segno, false);
917 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
919 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
920 ckpt_valid_blocks == usable_blocks)) {
921 __locate_dirty_segment(sbi, segno, PRE);
922 __remove_dirty_segment(sbi, segno, DIRTY);
923 } else if (valid_blocks < usable_blocks) {
924 __locate_dirty_segment(sbi, segno, DIRTY);
926 /* Recovery routine with SSR needs this */
927 __remove_dirty_segment(sbi, segno, DIRTY);
930 mutex_unlock(&dirty_i->seglist_lock);
933 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
934 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
936 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
939 mutex_lock(&dirty_i->seglist_lock);
940 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
941 if (get_valid_blocks(sbi, segno, false))
943 if (IS_CURSEG(sbi, segno))
945 __locate_dirty_segment(sbi, segno, PRE);
946 __remove_dirty_segment(sbi, segno, DIRTY);
948 mutex_unlock(&dirty_i->seglist_lock);
951 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
954 (overprovision_segments(sbi) - reserved_segments(sbi));
955 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
956 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
957 block_t holes[2] = {0, 0}; /* DATA and NODE */
959 struct seg_entry *se;
962 mutex_lock(&dirty_i->seglist_lock);
963 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
964 se = get_seg_entry(sbi, segno);
965 if (IS_NODESEG(se->type))
966 holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
969 holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
972 mutex_unlock(&dirty_i->seglist_lock);
974 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
975 if (unusable > ovp_holes)
976 return unusable - ovp_holes;
980 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
983 (overprovision_segments(sbi) - reserved_segments(sbi));
984 if (unusable > F2FS_OPTION(sbi).unusable_cap)
986 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
987 dirty_segments(sbi) > ovp_hole_segs)
992 /* This is only used by SBI_CP_DISABLED */
993 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
995 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
996 unsigned int segno = 0;
998 mutex_lock(&dirty_i->seglist_lock);
999 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
1000 if (get_valid_blocks(sbi, segno, false))
1002 if (get_ckpt_valid_blocks(sbi, segno, false))
1004 mutex_unlock(&dirty_i->seglist_lock);
1007 mutex_unlock(&dirty_i->seglist_lock);
1011 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
1012 struct block_device *bdev, block_t lstart,
1013 block_t start, block_t len)
1015 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1016 struct list_head *pend_list;
1017 struct discard_cmd *dc;
1019 f2fs_bug_on(sbi, !len);
1021 pend_list = &dcc->pend_list[plist_idx(len)];
1023 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
1024 INIT_LIST_HEAD(&dc->list);
1026 dc->lstart = lstart;
1033 init_completion(&dc->wait);
1034 list_add_tail(&dc->list, pend_list);
1035 spin_lock_init(&dc->lock);
1037 atomic_inc(&dcc->discard_cmd_cnt);
1038 dcc->undiscard_blks += len;
1043 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
1044 struct block_device *bdev, block_t lstart,
1045 block_t start, block_t len,
1046 struct rb_node *parent, struct rb_node **p,
1049 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1050 struct discard_cmd *dc;
1052 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1054 rb_link_node(&dc->rb_node, parent, p);
1055 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1060 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1061 struct discard_cmd *dc)
1063 if (dc->state == D_DONE)
1064 atomic_sub(dc->queued, &dcc->queued_discard);
1066 list_del(&dc->list);
1067 rb_erase_cached(&dc->rb_node, &dcc->root);
1068 dcc->undiscard_blks -= dc->len;
1070 kmem_cache_free(discard_cmd_slab, dc);
1072 atomic_dec(&dcc->discard_cmd_cnt);
1075 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1076 struct discard_cmd *dc)
1078 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1079 unsigned long flags;
1081 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1083 spin_lock_irqsave(&dc->lock, flags);
1085 spin_unlock_irqrestore(&dc->lock, flags);
1088 spin_unlock_irqrestore(&dc->lock, flags);
1090 f2fs_bug_on(sbi, dc->ref);
1092 if (dc->error == -EOPNOTSUPP)
1097 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1098 KERN_INFO, sbi->sb->s_id,
1099 dc->lstart, dc->start, dc->len, dc->error);
1100 __detach_discard_cmd(dcc, dc);
1103 static void f2fs_submit_discard_endio(struct bio *bio)
1105 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1106 unsigned long flags;
1108 spin_lock_irqsave(&dc->lock, flags);
1110 dc->error = blk_status_to_errno(bio->bi_status);
1112 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1114 complete_all(&dc->wait);
1116 spin_unlock_irqrestore(&dc->lock, flags);
1120 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1121 block_t start, block_t end)
1123 #ifdef CONFIG_F2FS_CHECK_FS
1124 struct seg_entry *sentry;
1126 block_t blk = start;
1127 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1131 segno = GET_SEGNO(sbi, blk);
1132 sentry = get_seg_entry(sbi, segno);
1133 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1135 if (end < START_BLOCK(sbi, segno + 1))
1136 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1139 map = (unsigned long *)(sentry->cur_valid_map);
1140 offset = __find_rev_next_bit(map, size, offset);
1141 f2fs_bug_on(sbi, offset != size);
1142 blk = START_BLOCK(sbi, segno + 1);
1147 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1148 struct discard_policy *dpolicy,
1149 int discard_type, unsigned int granularity)
1151 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1154 dpolicy->type = discard_type;
1155 dpolicy->sync = true;
1156 dpolicy->ordered = false;
1157 dpolicy->granularity = granularity;
1159 dpolicy->max_requests = dcc->max_discard_request;
1160 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1161 dpolicy->timeout = false;
1163 if (discard_type == DPOLICY_BG) {
1164 dpolicy->min_interval = dcc->min_discard_issue_time;
1165 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1166 dpolicy->max_interval = dcc->max_discard_issue_time;
1167 dpolicy->io_aware = true;
1168 dpolicy->sync = false;
1169 dpolicy->ordered = true;
1170 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1171 dpolicy->granularity = 1;
1172 if (atomic_read(&dcc->discard_cmd_cnt))
1173 dpolicy->max_interval =
1174 dcc->min_discard_issue_time;
1176 } else if (discard_type == DPOLICY_FORCE) {
1177 dpolicy->min_interval = dcc->min_discard_issue_time;
1178 dpolicy->mid_interval = dcc->mid_discard_issue_time;
1179 dpolicy->max_interval = dcc->max_discard_issue_time;
1180 dpolicy->io_aware = false;
1181 } else if (discard_type == DPOLICY_FSTRIM) {
1182 dpolicy->io_aware = false;
1183 } else if (discard_type == DPOLICY_UMOUNT) {
1184 dpolicy->io_aware = false;
1185 /* we need to issue all to keep CP_TRIMMED_FLAG */
1186 dpolicy->granularity = 1;
1187 dpolicy->timeout = true;
1191 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1192 struct block_device *bdev, block_t lstart,
1193 block_t start, block_t len);
1194 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1195 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1196 struct discard_policy *dpolicy,
1197 struct discard_cmd *dc,
1198 unsigned int *issued)
1200 struct block_device *bdev = dc->bdev;
1201 struct request_queue *q = bdev_get_queue(bdev);
1202 unsigned int max_discard_blocks =
1203 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1204 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1205 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1206 &(dcc->fstrim_list) : &(dcc->wait_list);
1207 int flag = dpolicy->sync ? REQ_SYNC : 0;
1208 block_t lstart, start, len, total_len;
1211 if (dc->state != D_PREP)
1214 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1217 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1219 lstart = dc->lstart;
1226 while (total_len && *issued < dpolicy->max_requests && !err) {
1227 struct bio *bio = NULL;
1228 unsigned long flags;
1231 if (len > max_discard_blocks) {
1232 len = max_discard_blocks;
1237 if (*issued == dpolicy->max_requests)
1242 if (time_to_inject(sbi, FAULT_DISCARD)) {
1243 f2fs_show_injection_info(sbi, FAULT_DISCARD);
1247 err = __blkdev_issue_discard(bdev,
1248 SECTOR_FROM_BLOCK(start),
1249 SECTOR_FROM_BLOCK(len),
1253 spin_lock_irqsave(&dc->lock, flags);
1254 if (dc->state == D_PARTIAL)
1255 dc->state = D_SUBMIT;
1256 spin_unlock_irqrestore(&dc->lock, flags);
1261 f2fs_bug_on(sbi, !bio);
1264 * should keep before submission to avoid D_DONE
1267 spin_lock_irqsave(&dc->lock, flags);
1269 dc->state = D_SUBMIT;
1271 dc->state = D_PARTIAL;
1273 spin_unlock_irqrestore(&dc->lock, flags);
1275 atomic_inc(&dcc->queued_discard);
1277 list_move_tail(&dc->list, wait_list);
1279 /* sanity check on discard range */
1280 __check_sit_bitmap(sbi, lstart, lstart + len);
1282 bio->bi_private = dc;
1283 bio->bi_end_io = f2fs_submit_discard_endio;
1284 bio->bi_opf |= flag;
1287 atomic_inc(&dcc->issued_discard);
1289 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1298 dcc->undiscard_blks -= len;
1299 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1304 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1305 struct block_device *bdev, block_t lstart,
1306 block_t start, block_t len,
1307 struct rb_node **insert_p,
1308 struct rb_node *insert_parent)
1310 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1312 struct rb_node *parent = NULL;
1313 bool leftmost = true;
1315 if (insert_p && insert_parent) {
1316 parent = insert_parent;
1321 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1324 __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1328 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1329 struct discard_cmd *dc)
1331 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1334 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1335 struct discard_cmd *dc, block_t blkaddr)
1337 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1338 struct discard_info di = dc->di;
1339 bool modified = false;
1341 if (dc->state == D_DONE || dc->len == 1) {
1342 __remove_discard_cmd(sbi, dc);
1346 dcc->undiscard_blks -= di.len;
1348 if (blkaddr > di.lstart) {
1349 dc->len = blkaddr - dc->lstart;
1350 dcc->undiscard_blks += dc->len;
1351 __relocate_discard_cmd(dcc, dc);
1355 if (blkaddr < di.lstart + di.len - 1) {
1357 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1358 di.start + blkaddr + 1 - di.lstart,
1359 di.lstart + di.len - 1 - blkaddr,
1365 dcc->undiscard_blks += dc->len;
1366 __relocate_discard_cmd(dcc, dc);
1371 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1372 struct block_device *bdev, block_t lstart,
1373 block_t start, block_t len)
1375 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1376 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1377 struct discard_cmd *dc;
1378 struct discard_info di = {0};
1379 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1380 struct request_queue *q = bdev_get_queue(bdev);
1381 unsigned int max_discard_blocks =
1382 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1383 block_t end = lstart + len;
1385 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1387 (struct rb_entry **)&prev_dc,
1388 (struct rb_entry **)&next_dc,
1389 &insert_p, &insert_parent, true, NULL);
1395 di.len = next_dc ? next_dc->lstart - lstart : len;
1396 di.len = min(di.len, len);
1401 struct rb_node *node;
1402 bool merged = false;
1403 struct discard_cmd *tdc = NULL;
1406 di.lstart = prev_dc->lstart + prev_dc->len;
1407 if (di.lstart < lstart)
1409 if (di.lstart >= end)
1412 if (!next_dc || next_dc->lstart > end)
1413 di.len = end - di.lstart;
1415 di.len = next_dc->lstart - di.lstart;
1416 di.start = start + di.lstart - lstart;
1422 if (prev_dc && prev_dc->state == D_PREP &&
1423 prev_dc->bdev == bdev &&
1424 __is_discard_back_mergeable(&di, &prev_dc->di,
1425 max_discard_blocks)) {
1426 prev_dc->di.len += di.len;
1427 dcc->undiscard_blks += di.len;
1428 __relocate_discard_cmd(dcc, prev_dc);
1434 if (next_dc && next_dc->state == D_PREP &&
1435 next_dc->bdev == bdev &&
1436 __is_discard_front_mergeable(&di, &next_dc->di,
1437 max_discard_blocks)) {
1438 next_dc->di.lstart = di.lstart;
1439 next_dc->di.len += di.len;
1440 next_dc->di.start = di.start;
1441 dcc->undiscard_blks += di.len;
1442 __relocate_discard_cmd(dcc, next_dc);
1444 __remove_discard_cmd(sbi, tdc);
1449 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1450 di.len, NULL, NULL);
1457 node = rb_next(&prev_dc->rb_node);
1458 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1462 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1463 struct block_device *bdev, block_t blkstart, block_t blklen)
1465 block_t lblkstart = blkstart;
1467 if (!f2fs_bdev_support_discard(bdev))
1470 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1472 if (f2fs_is_multi_device(sbi)) {
1473 int devi = f2fs_target_device_index(sbi, blkstart);
1475 blkstart -= FDEV(devi).start_blk;
1477 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1478 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1479 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1483 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1484 struct discard_policy *dpolicy)
1486 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1487 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1488 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1489 struct discard_cmd *dc;
1490 struct blk_plug plug;
1491 unsigned int pos = dcc->next_pos;
1492 unsigned int issued = 0;
1493 bool io_interrupted = false;
1495 mutex_lock(&dcc->cmd_lock);
1496 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1498 (struct rb_entry **)&prev_dc,
1499 (struct rb_entry **)&next_dc,
1500 &insert_p, &insert_parent, true, NULL);
1504 blk_start_plug(&plug);
1507 struct rb_node *node;
1510 if (dc->state != D_PREP)
1513 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1514 io_interrupted = true;
1518 dcc->next_pos = dc->lstart + dc->len;
1519 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1521 if (issued >= dpolicy->max_requests)
1524 node = rb_next(&dc->rb_node);
1526 __remove_discard_cmd(sbi, dc);
1527 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1530 blk_finish_plug(&plug);
1535 mutex_unlock(&dcc->cmd_lock);
1537 if (!issued && io_interrupted)
1542 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1543 struct discard_policy *dpolicy);
1545 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1546 struct discard_policy *dpolicy)
1548 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1549 struct list_head *pend_list;
1550 struct discard_cmd *dc, *tmp;
1551 struct blk_plug plug;
1553 bool io_interrupted = false;
1555 if (dpolicy->timeout)
1556 f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1560 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1561 if (dpolicy->timeout &&
1562 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1565 if (i + 1 < dpolicy->granularity)
1568 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1569 return __issue_discard_cmd_orderly(sbi, dpolicy);
1571 pend_list = &dcc->pend_list[i];
1573 mutex_lock(&dcc->cmd_lock);
1574 if (list_empty(pend_list))
1576 if (unlikely(dcc->rbtree_check))
1577 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1578 &dcc->root, false));
1579 blk_start_plug(&plug);
1580 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1581 f2fs_bug_on(sbi, dc->state != D_PREP);
1583 if (dpolicy->timeout &&
1584 f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1587 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1588 !is_idle(sbi, DISCARD_TIME)) {
1589 io_interrupted = true;
1593 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1595 if (issued >= dpolicy->max_requests)
1598 blk_finish_plug(&plug);
1600 mutex_unlock(&dcc->cmd_lock);
1602 if (issued >= dpolicy->max_requests || io_interrupted)
1606 if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1607 __wait_all_discard_cmd(sbi, dpolicy);
1611 if (!issued && io_interrupted)
1617 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1619 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1620 struct list_head *pend_list;
1621 struct discard_cmd *dc, *tmp;
1623 bool dropped = false;
1625 mutex_lock(&dcc->cmd_lock);
1626 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1627 pend_list = &dcc->pend_list[i];
1628 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1629 f2fs_bug_on(sbi, dc->state != D_PREP);
1630 __remove_discard_cmd(sbi, dc);
1634 mutex_unlock(&dcc->cmd_lock);
1639 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1641 __drop_discard_cmd(sbi);
1644 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1645 struct discard_cmd *dc)
1647 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1648 unsigned int len = 0;
1650 wait_for_completion_io(&dc->wait);
1651 mutex_lock(&dcc->cmd_lock);
1652 f2fs_bug_on(sbi, dc->state != D_DONE);
1657 __remove_discard_cmd(sbi, dc);
1659 mutex_unlock(&dcc->cmd_lock);
1664 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1665 struct discard_policy *dpolicy,
1666 block_t start, block_t end)
1668 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1669 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1670 &(dcc->fstrim_list) : &(dcc->wait_list);
1671 struct discard_cmd *dc, *tmp;
1673 unsigned int trimmed = 0;
1678 mutex_lock(&dcc->cmd_lock);
1679 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1680 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1682 if (dc->len < dpolicy->granularity)
1684 if (dc->state == D_DONE && !dc->ref) {
1685 wait_for_completion_io(&dc->wait);
1688 __remove_discard_cmd(sbi, dc);
1695 mutex_unlock(&dcc->cmd_lock);
1698 trimmed += __wait_one_discard_bio(sbi, dc);
1705 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1706 struct discard_policy *dpolicy)
1708 struct discard_policy dp;
1709 unsigned int discard_blks;
1712 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1715 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1716 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1717 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1718 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1720 return discard_blks;
1723 /* This should be covered by global mutex, &sit_i->sentry_lock */
1724 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1726 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1727 struct discard_cmd *dc;
1728 bool need_wait = false;
1730 mutex_lock(&dcc->cmd_lock);
1731 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1734 if (dc->state == D_PREP) {
1735 __punch_discard_cmd(sbi, dc, blkaddr);
1741 mutex_unlock(&dcc->cmd_lock);
1744 __wait_one_discard_bio(sbi, dc);
1747 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1749 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1751 if (dcc && dcc->f2fs_issue_discard) {
1752 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1754 dcc->f2fs_issue_discard = NULL;
1755 kthread_stop(discard_thread);
1759 /* This comes from f2fs_put_super */
1760 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1762 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1763 struct discard_policy dpolicy;
1766 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1767 dcc->discard_granularity);
1768 __issue_discard_cmd(sbi, &dpolicy);
1769 dropped = __drop_discard_cmd(sbi);
1771 /* just to make sure there is no pending discard commands */
1772 __wait_all_discard_cmd(sbi, NULL);
1774 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1778 static int issue_discard_thread(void *data)
1780 struct f2fs_sb_info *sbi = data;
1781 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1782 wait_queue_head_t *q = &dcc->discard_wait_queue;
1783 struct discard_policy dpolicy;
1784 unsigned int wait_ms = dcc->min_discard_issue_time;
1790 if (sbi->gc_mode == GC_URGENT_HIGH ||
1791 !f2fs_available_free_memory(sbi, DISCARD_CACHE))
1792 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1794 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1795 dcc->discard_granularity);
1797 if (!atomic_read(&dcc->discard_cmd_cnt))
1798 wait_ms = dpolicy.max_interval;
1800 wait_event_interruptible_timeout(*q,
1801 kthread_should_stop() || freezing(current) ||
1803 msecs_to_jiffies(wait_ms));
1805 if (dcc->discard_wake)
1806 dcc->discard_wake = 0;
1808 /* clean up pending candidates before going to sleep */
1809 if (atomic_read(&dcc->queued_discard))
1810 __wait_all_discard_cmd(sbi, NULL);
1812 if (try_to_freeze())
1814 if (f2fs_readonly(sbi->sb))
1816 if (kthread_should_stop())
1818 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1819 wait_ms = dpolicy.max_interval;
1822 if (!atomic_read(&dcc->discard_cmd_cnt))
1825 sb_start_intwrite(sbi->sb);
1827 issued = __issue_discard_cmd(sbi, &dpolicy);
1829 __wait_all_discard_cmd(sbi, &dpolicy);
1830 wait_ms = dpolicy.min_interval;
1831 } else if (issued == -1) {
1832 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1834 wait_ms = dpolicy.mid_interval;
1836 wait_ms = dpolicy.max_interval;
1839 sb_end_intwrite(sbi->sb);
1841 } while (!kthread_should_stop());
1845 #ifdef CONFIG_BLK_DEV_ZONED
1846 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1847 struct block_device *bdev, block_t blkstart, block_t blklen)
1849 sector_t sector, nr_sects;
1850 block_t lblkstart = blkstart;
1853 if (f2fs_is_multi_device(sbi)) {
1854 devi = f2fs_target_device_index(sbi, blkstart);
1855 if (blkstart < FDEV(devi).start_blk ||
1856 blkstart > FDEV(devi).end_blk) {
1857 f2fs_err(sbi, "Invalid block %x", blkstart);
1860 blkstart -= FDEV(devi).start_blk;
1863 /* For sequential zones, reset the zone write pointer */
1864 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1865 sector = SECTOR_FROM_BLOCK(blkstart);
1866 nr_sects = SECTOR_FROM_BLOCK(blklen);
1868 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1869 nr_sects != bdev_zone_sectors(bdev)) {
1870 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1871 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1875 trace_f2fs_issue_reset_zone(bdev, blkstart);
1876 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1877 sector, nr_sects, GFP_NOFS);
1880 /* For conventional zones, use regular discard if supported */
1881 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1885 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1886 struct block_device *bdev, block_t blkstart, block_t blklen)
1888 #ifdef CONFIG_BLK_DEV_ZONED
1889 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1890 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1892 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1895 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1896 block_t blkstart, block_t blklen)
1898 sector_t start = blkstart, len = 0;
1899 struct block_device *bdev;
1900 struct seg_entry *se;
1901 unsigned int offset;
1905 bdev = f2fs_target_device(sbi, blkstart, NULL);
1907 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1909 struct block_device *bdev2 =
1910 f2fs_target_device(sbi, i, NULL);
1912 if (bdev2 != bdev) {
1913 err = __issue_discard_async(sbi, bdev,
1923 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1924 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1926 if (f2fs_block_unit_discard(sbi) &&
1927 !f2fs_test_and_set_bit(offset, se->discard_map))
1928 sbi->discard_blks--;
1932 err = __issue_discard_async(sbi, bdev, start, len);
1936 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1939 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1940 int max_blocks = sbi->blocks_per_seg;
1941 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1942 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1943 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1944 unsigned long *discard_map = (unsigned long *)se->discard_map;
1945 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1946 unsigned int start = 0, end = -1;
1947 bool force = (cpc->reason & CP_DISCARD);
1948 struct discard_entry *de = NULL;
1949 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1952 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
1953 !f2fs_block_unit_discard(sbi))
1957 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1958 SM_I(sbi)->dcc_info->nr_discards >=
1959 SM_I(sbi)->dcc_info->max_discards)
1963 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1964 for (i = 0; i < entries; i++)
1965 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1966 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1968 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1969 SM_I(sbi)->dcc_info->max_discards) {
1970 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1971 if (start >= max_blocks)
1974 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1975 if (force && start && end != max_blocks
1976 && (end - start) < cpc->trim_minlen)
1983 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1984 GFP_F2FS_ZERO, true, NULL);
1985 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1986 list_add_tail(&de->list, head);
1989 for (i = start; i < end; i++)
1990 __set_bit_le(i, (void *)de->discard_map);
1992 SM_I(sbi)->dcc_info->nr_discards += end - start;
1997 static void release_discard_addr(struct discard_entry *entry)
1999 list_del(&entry->list);
2000 kmem_cache_free(discard_entry_slab, entry);
2003 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2005 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2006 struct discard_entry *entry, *this;
2009 list_for_each_entry_safe(entry, this, head, list)
2010 release_discard_addr(entry);
2014 * Should call f2fs_clear_prefree_segments after checkpoint is done.
2016 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2018 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2021 mutex_lock(&dirty_i->seglist_lock);
2022 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2023 __set_test_and_free(sbi, segno, false);
2024 mutex_unlock(&dirty_i->seglist_lock);
2027 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2028 struct cp_control *cpc)
2030 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2031 struct list_head *head = &dcc->entry_list;
2032 struct discard_entry *entry, *this;
2033 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2034 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2035 unsigned int start = 0, end = -1;
2036 unsigned int secno, start_segno;
2037 bool force = (cpc->reason & CP_DISCARD);
2038 bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2039 DISCARD_UNIT_SECTION;
2041 if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2042 section_alignment = true;
2044 mutex_lock(&dirty_i->seglist_lock);
2049 if (section_alignment && end != -1)
2051 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2052 if (start >= MAIN_SEGS(sbi))
2054 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2057 if (section_alignment) {
2058 start = rounddown(start, sbi->segs_per_sec);
2059 end = roundup(end, sbi->segs_per_sec);
2062 for (i = start; i < end; i++) {
2063 if (test_and_clear_bit(i, prefree_map))
2064 dirty_i->nr_dirty[PRE]--;
2067 if (!f2fs_realtime_discard_enable(sbi))
2070 if (force && start >= cpc->trim_start &&
2071 (end - 1) <= cpc->trim_end)
2074 if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2075 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2076 (end - start) << sbi->log_blocks_per_seg);
2080 secno = GET_SEC_FROM_SEG(sbi, start);
2081 start_segno = GET_SEG_FROM_SEC(sbi, secno);
2082 if (!IS_CURSEC(sbi, secno) &&
2083 !get_valid_blocks(sbi, start, true))
2084 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2085 sbi->segs_per_sec << sbi->log_blocks_per_seg);
2087 start = start_segno + sbi->segs_per_sec;
2093 mutex_unlock(&dirty_i->seglist_lock);
2095 if (!f2fs_block_unit_discard(sbi))
2098 /* send small discards */
2099 list_for_each_entry_safe(entry, this, head, list) {
2100 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2101 bool is_valid = test_bit_le(0, entry->discard_map);
2105 next_pos = find_next_zero_bit_le(entry->discard_map,
2106 sbi->blocks_per_seg, cur_pos);
2107 len = next_pos - cur_pos;
2109 if (f2fs_sb_has_blkzoned(sbi) ||
2110 (force && len < cpc->trim_minlen))
2113 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2117 next_pos = find_next_bit_le(entry->discard_map,
2118 sbi->blocks_per_seg, cur_pos);
2122 is_valid = !is_valid;
2124 if (cur_pos < sbi->blocks_per_seg)
2127 release_discard_addr(entry);
2128 dcc->nr_discards -= total_len;
2132 wake_up_discard_thread(sbi, false);
2135 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2137 dev_t dev = sbi->sb->s_bdev->bd_dev;
2138 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2141 if (!f2fs_realtime_discard_enable(sbi))
2144 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2145 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2146 if (IS_ERR(dcc->f2fs_issue_discard))
2147 err = PTR_ERR(dcc->f2fs_issue_discard);
2152 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2154 struct discard_cmd_control *dcc;
2157 if (SM_I(sbi)->dcc_info) {
2158 dcc = SM_I(sbi)->dcc_info;
2162 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2166 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2167 if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2168 dcc->discard_granularity = sbi->blocks_per_seg;
2169 else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2170 dcc->discard_granularity = BLKS_PER_SEC(sbi);
2172 INIT_LIST_HEAD(&dcc->entry_list);
2173 for (i = 0; i < MAX_PLIST_NUM; i++)
2174 INIT_LIST_HEAD(&dcc->pend_list[i]);
2175 INIT_LIST_HEAD(&dcc->wait_list);
2176 INIT_LIST_HEAD(&dcc->fstrim_list);
2177 mutex_init(&dcc->cmd_lock);
2178 atomic_set(&dcc->issued_discard, 0);
2179 atomic_set(&dcc->queued_discard, 0);
2180 atomic_set(&dcc->discard_cmd_cnt, 0);
2181 dcc->nr_discards = 0;
2182 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2183 dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2184 dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2185 dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2186 dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2187 dcc->undiscard_blks = 0;
2189 dcc->root = RB_ROOT_CACHED;
2190 dcc->rbtree_check = false;
2192 init_waitqueue_head(&dcc->discard_wait_queue);
2193 SM_I(sbi)->dcc_info = dcc;
2195 err = f2fs_start_discard_thread(sbi);
2198 SM_I(sbi)->dcc_info = NULL;
2204 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2206 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2211 f2fs_stop_discard_thread(sbi);
2214 * Recovery can cache discard commands, so in error path of
2215 * fill_super(), it needs to give a chance to handle them.
2217 if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2218 f2fs_issue_discard_timeout(sbi);
2221 SM_I(sbi)->dcc_info = NULL;
2224 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2226 struct sit_info *sit_i = SIT_I(sbi);
2228 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2229 sit_i->dirty_sentries++;
2236 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2237 unsigned int segno, int modified)
2239 struct seg_entry *se = get_seg_entry(sbi, segno);
2243 __mark_sit_entry_dirty(sbi, segno);
2246 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2249 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2251 if (segno == NULL_SEGNO)
2253 return get_seg_entry(sbi, segno)->mtime;
2256 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2257 unsigned long long old_mtime)
2259 struct seg_entry *se;
2260 unsigned int segno = GET_SEGNO(sbi, blkaddr);
2261 unsigned long long ctime = get_mtime(sbi, false);
2262 unsigned long long mtime = old_mtime ? old_mtime : ctime;
2264 if (segno == NULL_SEGNO)
2267 se = get_seg_entry(sbi, segno);
2272 se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2273 se->valid_blocks + 1);
2275 if (ctime > SIT_I(sbi)->max_mtime)
2276 SIT_I(sbi)->max_mtime = ctime;
2279 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2281 struct seg_entry *se;
2282 unsigned int segno, offset;
2283 long int new_vblocks;
2285 #ifdef CONFIG_F2FS_CHECK_FS
2289 segno = GET_SEGNO(sbi, blkaddr);
2291 se = get_seg_entry(sbi, segno);
2292 new_vblocks = se->valid_blocks + del;
2293 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2295 f2fs_bug_on(sbi, (new_vblocks < 0 ||
2296 (new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2298 se->valid_blocks = new_vblocks;
2300 /* Update valid block bitmap */
2302 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2303 #ifdef CONFIG_F2FS_CHECK_FS
2304 mir_exist = f2fs_test_and_set_bit(offset,
2305 se->cur_valid_map_mir);
2306 if (unlikely(exist != mir_exist)) {
2307 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2309 f2fs_bug_on(sbi, 1);
2312 if (unlikely(exist)) {
2313 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2315 f2fs_bug_on(sbi, 1);
2320 if (f2fs_block_unit_discard(sbi) &&
2321 !f2fs_test_and_set_bit(offset, se->discard_map))
2322 sbi->discard_blks--;
2325 * SSR should never reuse block which is checkpointed
2326 * or newly invalidated.
2328 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2329 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2330 se->ckpt_valid_blocks++;
2333 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2334 #ifdef CONFIG_F2FS_CHECK_FS
2335 mir_exist = f2fs_test_and_clear_bit(offset,
2336 se->cur_valid_map_mir);
2337 if (unlikely(exist != mir_exist)) {
2338 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2340 f2fs_bug_on(sbi, 1);
2343 if (unlikely(!exist)) {
2344 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2346 f2fs_bug_on(sbi, 1);
2349 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2351 * If checkpoints are off, we must not reuse data that
2352 * was used in the previous checkpoint. If it was used
2353 * before, we must track that to know how much space we
2356 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2357 spin_lock(&sbi->stat_lock);
2358 sbi->unusable_block_count++;
2359 spin_unlock(&sbi->stat_lock);
2363 if (f2fs_block_unit_discard(sbi) &&
2364 f2fs_test_and_clear_bit(offset, se->discard_map))
2365 sbi->discard_blks++;
2367 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2368 se->ckpt_valid_blocks += del;
2370 __mark_sit_entry_dirty(sbi, segno);
2372 /* update total number of valid blocks to be written in ckpt area */
2373 SIT_I(sbi)->written_valid_blocks += del;
2375 if (__is_large_section(sbi))
2376 get_sec_entry(sbi, segno)->valid_blocks += del;
2379 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2381 unsigned int segno = GET_SEGNO(sbi, addr);
2382 struct sit_info *sit_i = SIT_I(sbi);
2384 f2fs_bug_on(sbi, addr == NULL_ADDR);
2385 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2388 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2389 f2fs_invalidate_compress_page(sbi, addr);
2391 /* add it into sit main buffer */
2392 down_write(&sit_i->sentry_lock);
2394 update_segment_mtime(sbi, addr, 0);
2395 update_sit_entry(sbi, addr, -1);
2397 /* add it into dirty seglist */
2398 locate_dirty_segment(sbi, segno);
2400 up_write(&sit_i->sentry_lock);
2403 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2405 struct sit_info *sit_i = SIT_I(sbi);
2406 unsigned int segno, offset;
2407 struct seg_entry *se;
2410 if (!__is_valid_data_blkaddr(blkaddr))
2413 down_read(&sit_i->sentry_lock);
2415 segno = GET_SEGNO(sbi, blkaddr);
2416 se = get_seg_entry(sbi, segno);
2417 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2419 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2422 up_read(&sit_i->sentry_lock);
2428 * This function should be resided under the curseg_mutex lock
2430 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2431 struct f2fs_summary *sum)
2433 struct curseg_info *curseg = CURSEG_I(sbi, type);
2434 void *addr = curseg->sum_blk;
2436 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2437 memcpy(addr, sum, sizeof(struct f2fs_summary));
2441 * Calculate the number of current summary pages for writing
2443 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2445 int valid_sum_count = 0;
2448 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2449 if (sbi->ckpt->alloc_type[i] == SSR)
2450 valid_sum_count += sbi->blocks_per_seg;
2453 valid_sum_count += le16_to_cpu(
2454 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2456 valid_sum_count += curseg_blkoff(sbi, i);
2460 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2461 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2462 if (valid_sum_count <= sum_in_page)
2464 else if ((valid_sum_count - sum_in_page) <=
2465 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2471 * Caller should put this summary page
2473 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2475 if (unlikely(f2fs_cp_error(sbi)))
2476 return ERR_PTR(-EIO);
2477 return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2480 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2481 void *src, block_t blk_addr)
2483 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2485 memcpy(page_address(page), src, PAGE_SIZE);
2486 set_page_dirty(page);
2487 f2fs_put_page(page, 1);
2490 static void write_sum_page(struct f2fs_sb_info *sbi,
2491 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2493 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2496 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2497 int type, block_t blk_addr)
2499 struct curseg_info *curseg = CURSEG_I(sbi, type);
2500 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2501 struct f2fs_summary_block *src = curseg->sum_blk;
2502 struct f2fs_summary_block *dst;
2504 dst = (struct f2fs_summary_block *)page_address(page);
2505 memset(dst, 0, PAGE_SIZE);
2507 mutex_lock(&curseg->curseg_mutex);
2509 down_read(&curseg->journal_rwsem);
2510 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2511 up_read(&curseg->journal_rwsem);
2513 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2514 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2516 mutex_unlock(&curseg->curseg_mutex);
2518 set_page_dirty(page);
2519 f2fs_put_page(page, 1);
2522 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2523 struct curseg_info *curseg, int type)
2525 unsigned int segno = curseg->segno + 1;
2526 struct free_segmap_info *free_i = FREE_I(sbi);
2528 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2529 return !test_bit(segno, free_i->free_segmap);
2534 * Find a new segment from the free segments bitmap to right order
2535 * This function should be returned with success, otherwise BUG
2537 static void get_new_segment(struct f2fs_sb_info *sbi,
2538 unsigned int *newseg, bool new_sec, int dir)
2540 struct free_segmap_info *free_i = FREE_I(sbi);
2541 unsigned int segno, secno, zoneno;
2542 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2543 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2544 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2545 unsigned int left_start = hint;
2550 spin_lock(&free_i->segmap_lock);
2552 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2553 segno = find_next_zero_bit(free_i->free_segmap,
2554 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2555 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2559 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2560 if (secno >= MAIN_SECS(sbi)) {
2561 if (dir == ALLOC_RIGHT) {
2562 secno = find_first_zero_bit(free_i->free_secmap,
2564 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2567 left_start = hint - 1;
2573 while (test_bit(left_start, free_i->free_secmap)) {
2574 if (left_start > 0) {
2578 left_start = find_first_zero_bit(free_i->free_secmap,
2580 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2585 segno = GET_SEG_FROM_SEC(sbi, secno);
2586 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2588 /* give up on finding another zone */
2591 if (sbi->secs_per_zone == 1)
2593 if (zoneno == old_zoneno)
2595 if (dir == ALLOC_LEFT) {
2596 if (!go_left && zoneno + 1 >= total_zones)
2598 if (go_left && zoneno == 0)
2601 for (i = 0; i < NR_CURSEG_TYPE; i++)
2602 if (CURSEG_I(sbi, i)->zone == zoneno)
2605 if (i < NR_CURSEG_TYPE) {
2606 /* zone is in user, try another */
2608 hint = zoneno * sbi->secs_per_zone - 1;
2609 else if (zoneno + 1 >= total_zones)
2612 hint = (zoneno + 1) * sbi->secs_per_zone;
2614 goto find_other_zone;
2617 /* set it as dirty segment in free segmap */
2618 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2619 __set_inuse(sbi, segno);
2621 spin_unlock(&free_i->segmap_lock);
2624 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2626 struct curseg_info *curseg = CURSEG_I(sbi, type);
2627 struct summary_footer *sum_footer;
2628 unsigned short seg_type = curseg->seg_type;
2630 curseg->inited = true;
2631 curseg->segno = curseg->next_segno;
2632 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2633 curseg->next_blkoff = 0;
2634 curseg->next_segno = NULL_SEGNO;
2636 sum_footer = &(curseg->sum_blk->footer);
2637 memset(sum_footer, 0, sizeof(struct summary_footer));
2639 sanity_check_seg_type(sbi, seg_type);
2641 if (IS_DATASEG(seg_type))
2642 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2643 if (IS_NODESEG(seg_type))
2644 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2645 __set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2648 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2650 struct curseg_info *curseg = CURSEG_I(sbi, type);
2651 unsigned short seg_type = curseg->seg_type;
2653 sanity_check_seg_type(sbi, seg_type);
2654 if (f2fs_need_rand_seg(sbi))
2655 return prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
2657 /* if segs_per_sec is large than 1, we need to keep original policy. */
2658 if (__is_large_section(sbi))
2659 return curseg->segno;
2661 /* inmem log may not locate on any segment after mount */
2662 if (!curseg->inited)
2665 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2668 if (test_opt(sbi, NOHEAP) &&
2669 (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2672 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2673 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2675 /* find segments from 0 to reuse freed segments */
2676 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2679 return curseg->segno;
2683 * Allocate a current working segment.
2684 * This function always allocates a free segment in LFS manner.
2686 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2688 struct curseg_info *curseg = CURSEG_I(sbi, type);
2689 unsigned short seg_type = curseg->seg_type;
2690 unsigned int segno = curseg->segno;
2691 int dir = ALLOC_LEFT;
2694 write_sum_page(sbi, curseg->sum_blk,
2695 GET_SUM_BLOCK(sbi, segno));
2696 if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2699 if (test_opt(sbi, NOHEAP))
2702 segno = __get_next_segno(sbi, type);
2703 get_new_segment(sbi, &segno, new_sec, dir);
2704 curseg->next_segno = segno;
2705 reset_curseg(sbi, type, 1);
2706 curseg->alloc_type = LFS;
2707 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2708 curseg->fragment_remained_chunk =
2709 prandom_u32() % sbi->max_fragment_chunk + 1;
2712 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2713 int segno, block_t start)
2715 struct seg_entry *se = get_seg_entry(sbi, segno);
2716 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2717 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2718 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2719 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2722 for (i = 0; i < entries; i++)
2723 target_map[i] = ckpt_map[i] | cur_map[i];
2725 return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2729 * If a segment is written by LFS manner, next block offset is just obtained
2730 * by increasing the current block offset. However, if a segment is written by
2731 * SSR manner, next block offset obtained by calling __next_free_blkoff
2733 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2734 struct curseg_info *seg)
2736 if (seg->alloc_type == SSR) {
2738 __next_free_blkoff(sbi, seg->segno,
2739 seg->next_blkoff + 1);
2742 if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) {
2743 /* To allocate block chunks in different sizes, use random number */
2744 if (--seg->fragment_remained_chunk <= 0) {
2745 seg->fragment_remained_chunk =
2746 prandom_u32() % sbi->max_fragment_chunk + 1;
2748 prandom_u32() % sbi->max_fragment_hole + 1;
2754 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2756 return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2760 * This function always allocates a used segment(from dirty seglist) by SSR
2761 * manner, so it should recover the existing segment information of valid blocks
2763 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2765 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2766 struct curseg_info *curseg = CURSEG_I(sbi, type);
2767 unsigned int new_segno = curseg->next_segno;
2768 struct f2fs_summary_block *sum_node;
2769 struct page *sum_page;
2772 write_sum_page(sbi, curseg->sum_blk,
2773 GET_SUM_BLOCK(sbi, curseg->segno));
2775 __set_test_and_inuse(sbi, new_segno);
2777 mutex_lock(&dirty_i->seglist_lock);
2778 __remove_dirty_segment(sbi, new_segno, PRE);
2779 __remove_dirty_segment(sbi, new_segno, DIRTY);
2780 mutex_unlock(&dirty_i->seglist_lock);
2782 reset_curseg(sbi, type, 1);
2783 curseg->alloc_type = SSR;
2784 curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2786 sum_page = f2fs_get_sum_page(sbi, new_segno);
2787 if (IS_ERR(sum_page)) {
2788 /* GC won't be able to use stale summary pages by cp_error */
2789 memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2792 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2793 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2794 f2fs_put_page(sum_page, 1);
2797 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2798 int alloc_mode, unsigned long long age);
2800 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2801 int target_type, int alloc_mode,
2802 unsigned long long age)
2804 struct curseg_info *curseg = CURSEG_I(sbi, type);
2806 curseg->seg_type = target_type;
2808 if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2809 struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2811 curseg->seg_type = se->type;
2812 change_curseg(sbi, type, true);
2814 /* allocate cold segment by default */
2815 curseg->seg_type = CURSEG_COLD_DATA;
2816 new_curseg(sbi, type, true);
2818 stat_inc_seg_type(sbi, curseg);
2821 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2823 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2825 if (!sbi->am.atgc_enabled)
2828 f2fs_down_read(&SM_I(sbi)->curseg_lock);
2830 mutex_lock(&curseg->curseg_mutex);
2831 down_write(&SIT_I(sbi)->sentry_lock);
2833 get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2835 up_write(&SIT_I(sbi)->sentry_lock);
2836 mutex_unlock(&curseg->curseg_mutex);
2838 f2fs_up_read(&SM_I(sbi)->curseg_lock);
2841 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2843 __f2fs_init_atgc_curseg(sbi);
2846 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2848 struct curseg_info *curseg = CURSEG_I(sbi, type);
2850 mutex_lock(&curseg->curseg_mutex);
2851 if (!curseg->inited)
2854 if (get_valid_blocks(sbi, curseg->segno, false)) {
2855 write_sum_page(sbi, curseg->sum_blk,
2856 GET_SUM_BLOCK(sbi, curseg->segno));
2858 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2859 __set_test_and_free(sbi, curseg->segno, true);
2860 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2863 mutex_unlock(&curseg->curseg_mutex);
2866 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2868 __f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2870 if (sbi->am.atgc_enabled)
2871 __f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2874 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2876 struct curseg_info *curseg = CURSEG_I(sbi, type);
2878 mutex_lock(&curseg->curseg_mutex);
2879 if (!curseg->inited)
2881 if (get_valid_blocks(sbi, curseg->segno, false))
2884 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2885 __set_test_and_inuse(sbi, curseg->segno);
2886 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2888 mutex_unlock(&curseg->curseg_mutex);
2891 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2893 __f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2895 if (sbi->am.atgc_enabled)
2896 __f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2899 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2900 int alloc_mode, unsigned long long age)
2902 struct curseg_info *curseg = CURSEG_I(sbi, type);
2903 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2904 unsigned segno = NULL_SEGNO;
2905 unsigned short seg_type = curseg->seg_type;
2907 bool reversed = false;
2909 sanity_check_seg_type(sbi, seg_type);
2911 /* f2fs_need_SSR() already forces to do this */
2912 if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2913 curseg->next_segno = segno;
2917 /* For node segments, let's do SSR more intensively */
2918 if (IS_NODESEG(seg_type)) {
2919 if (seg_type >= CURSEG_WARM_NODE) {
2921 i = CURSEG_COLD_NODE;
2923 i = CURSEG_HOT_NODE;
2925 cnt = NR_CURSEG_NODE_TYPE;
2927 if (seg_type >= CURSEG_WARM_DATA) {
2929 i = CURSEG_COLD_DATA;
2931 i = CURSEG_HOT_DATA;
2933 cnt = NR_CURSEG_DATA_TYPE;
2936 for (; cnt-- > 0; reversed ? i-- : i++) {
2939 if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2940 curseg->next_segno = segno;
2945 /* find valid_blocks=0 in dirty list */
2946 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2947 segno = get_free_segment(sbi);
2948 if (segno != NULL_SEGNO) {
2949 curseg->next_segno = segno;
2957 * flush out current segment and replace it with new segment
2958 * This function should be returned with success, otherwise BUG
2960 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2961 int type, bool force)
2963 struct curseg_info *curseg = CURSEG_I(sbi, type);
2966 new_curseg(sbi, type, true);
2967 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2968 curseg->seg_type == CURSEG_WARM_NODE)
2969 new_curseg(sbi, type, false);
2970 else if (curseg->alloc_type == LFS &&
2971 is_next_segment_free(sbi, curseg, type) &&
2972 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2973 new_curseg(sbi, type, false);
2974 else if (f2fs_need_SSR(sbi) &&
2975 get_ssr_segment(sbi, type, SSR, 0))
2976 change_curseg(sbi, type, true);
2978 new_curseg(sbi, type, false);
2980 stat_inc_seg_type(sbi, curseg);
2983 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2984 unsigned int start, unsigned int end)
2986 struct curseg_info *curseg = CURSEG_I(sbi, type);
2989 f2fs_down_read(&SM_I(sbi)->curseg_lock);
2990 mutex_lock(&curseg->curseg_mutex);
2991 down_write(&SIT_I(sbi)->sentry_lock);
2993 segno = CURSEG_I(sbi, type)->segno;
2994 if (segno < start || segno > end)
2997 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2998 change_curseg(sbi, type, true);
3000 new_curseg(sbi, type, true);
3002 stat_inc_seg_type(sbi, curseg);
3004 locate_dirty_segment(sbi, segno);
3006 up_write(&SIT_I(sbi)->sentry_lock);
3008 if (segno != curseg->segno)
3009 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3010 type, segno, curseg->segno);
3012 mutex_unlock(&curseg->curseg_mutex);
3013 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3016 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3017 bool new_sec, bool force)
3019 struct curseg_info *curseg = CURSEG_I(sbi, type);
3020 unsigned int old_segno;
3022 if (!curseg->inited)
3025 if (force || curseg->next_blkoff ||
3026 get_valid_blocks(sbi, curseg->segno, new_sec))
3029 if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3032 old_segno = curseg->segno;
3033 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
3034 locate_dirty_segment(sbi, old_segno);
3037 static void __allocate_new_section(struct f2fs_sb_info *sbi,
3038 int type, bool force)
3040 __allocate_new_segment(sbi, type, true, force);
3043 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3045 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3046 down_write(&SIT_I(sbi)->sentry_lock);
3047 __allocate_new_section(sbi, type, force);
3048 up_write(&SIT_I(sbi)->sentry_lock);
3049 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3052 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3056 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3057 down_write(&SIT_I(sbi)->sentry_lock);
3058 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3059 __allocate_new_segment(sbi, i, false, false);
3060 up_write(&SIT_I(sbi)->sentry_lock);
3061 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3064 static const struct segment_allocation default_salloc_ops = {
3065 .allocate_segment = allocate_segment_by_default,
3068 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3069 struct cp_control *cpc)
3071 __u64 trim_start = cpc->trim_start;
3072 bool has_candidate = false;
3074 down_write(&SIT_I(sbi)->sentry_lock);
3075 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3076 if (add_discard_addrs(sbi, cpc, true)) {
3077 has_candidate = true;
3081 up_write(&SIT_I(sbi)->sentry_lock);
3083 cpc->trim_start = trim_start;
3084 return has_candidate;
3087 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3088 struct discard_policy *dpolicy,
3089 unsigned int start, unsigned int end)
3091 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3092 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3093 struct rb_node **insert_p = NULL, *insert_parent = NULL;
3094 struct discard_cmd *dc;
3095 struct blk_plug plug;
3097 unsigned int trimmed = 0;
3102 mutex_lock(&dcc->cmd_lock);
3103 if (unlikely(dcc->rbtree_check))
3104 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
3105 &dcc->root, false));
3107 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
3109 (struct rb_entry **)&prev_dc,
3110 (struct rb_entry **)&next_dc,
3111 &insert_p, &insert_parent, true, NULL);
3115 blk_start_plug(&plug);
3117 while (dc && dc->lstart <= end) {
3118 struct rb_node *node;
3121 if (dc->len < dpolicy->granularity)
3124 if (dc->state != D_PREP) {
3125 list_move_tail(&dc->list, &dcc->fstrim_list);
3129 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3131 if (issued >= dpolicy->max_requests) {
3132 start = dc->lstart + dc->len;
3135 __remove_discard_cmd(sbi, dc);
3137 blk_finish_plug(&plug);
3138 mutex_unlock(&dcc->cmd_lock);
3139 trimmed += __wait_all_discard_cmd(sbi, NULL);
3140 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
3144 node = rb_next(&dc->rb_node);
3146 __remove_discard_cmd(sbi, dc);
3147 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3149 if (fatal_signal_pending(current))
3153 blk_finish_plug(&plug);
3154 mutex_unlock(&dcc->cmd_lock);
3159 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3161 __u64 start = F2FS_BYTES_TO_BLK(range->start);
3162 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3163 unsigned int start_segno, end_segno;
3164 block_t start_block, end_block;
3165 struct cp_control cpc;
3166 struct discard_policy dpolicy;
3167 unsigned long long trimmed = 0;
3169 bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3171 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3174 if (end < MAIN_BLKADDR(sbi))
3177 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3178 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3179 return -EFSCORRUPTED;
3182 /* start/end segment number in main_area */
3183 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3184 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3185 GET_SEGNO(sbi, end);
3187 start_segno = rounddown(start_segno, sbi->segs_per_sec);
3188 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3191 cpc.reason = CP_DISCARD;
3192 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3193 cpc.trim_start = start_segno;
3194 cpc.trim_end = end_segno;
3196 if (sbi->discard_blks == 0)
3199 f2fs_down_write(&sbi->gc_lock);
3200 err = f2fs_write_checkpoint(sbi, &cpc);
3201 f2fs_up_write(&sbi->gc_lock);
3206 * We filed discard candidates, but actually we don't need to wait for
3207 * all of them, since they'll be issued in idle time along with runtime
3208 * discard option. User configuration looks like using runtime discard
3209 * or periodic fstrim instead of it.
3211 if (f2fs_realtime_discard_enable(sbi))
3214 start_block = START_BLOCK(sbi, start_segno);
3215 end_block = START_BLOCK(sbi, end_segno + 1);
3217 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3218 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3219 start_block, end_block);
3221 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3222 start_block, end_block);
3225 range->len = F2FS_BLK_TO_BYTES(trimmed);
3229 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3230 struct curseg_info *curseg)
3232 return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3236 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3239 case WRITE_LIFE_SHORT:
3240 return CURSEG_HOT_DATA;
3241 case WRITE_LIFE_EXTREME:
3242 return CURSEG_COLD_DATA;
3244 return CURSEG_WARM_DATA;
3248 /* This returns write hints for each segment type. This hints will be
3249 * passed down to block layer. There are mapping tables which depend on
3250 * the mount option 'whint_mode'.
3252 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3254 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3258 * META WRITE_LIFE_NOT_SET
3262 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3263 * extension list " "
3266 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3267 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3268 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3269 * WRITE_LIFE_NONE " "
3270 * WRITE_LIFE_MEDIUM " "
3271 * WRITE_LIFE_LONG " "
3274 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3275 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3276 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3277 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3278 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3279 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
3281 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3285 * META WRITE_LIFE_MEDIUM;
3286 * HOT_NODE WRITE_LIFE_NOT_SET
3288 * COLD_NODE WRITE_LIFE_NONE
3289 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
3290 * extension list " "
3293 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3294 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3295 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
3296 * WRITE_LIFE_NONE " "
3297 * WRITE_LIFE_MEDIUM " "
3298 * WRITE_LIFE_LONG " "
3301 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
3302 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
3303 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
3304 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
3305 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
3306 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
3309 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3310 enum page_type type, enum temp_type temp)
3312 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3315 return WRITE_LIFE_NOT_SET;
3316 else if (temp == HOT)
3317 return WRITE_LIFE_SHORT;
3318 else if (temp == COLD)
3319 return WRITE_LIFE_EXTREME;
3321 return WRITE_LIFE_NOT_SET;
3323 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3326 return WRITE_LIFE_LONG;
3327 else if (temp == HOT)
3328 return WRITE_LIFE_SHORT;
3329 else if (temp == COLD)
3330 return WRITE_LIFE_EXTREME;
3331 } else if (type == NODE) {
3332 if (temp == WARM || temp == HOT)
3333 return WRITE_LIFE_NOT_SET;
3334 else if (temp == COLD)
3335 return WRITE_LIFE_NONE;
3336 } else if (type == META) {
3337 return WRITE_LIFE_MEDIUM;
3340 return WRITE_LIFE_NOT_SET;
3343 static int __get_segment_type_2(struct f2fs_io_info *fio)
3345 if (fio->type == DATA)
3346 return CURSEG_HOT_DATA;
3348 return CURSEG_HOT_NODE;
3351 static int __get_segment_type_4(struct f2fs_io_info *fio)
3353 if (fio->type == DATA) {
3354 struct inode *inode = fio->page->mapping->host;
3356 if (S_ISDIR(inode->i_mode))
3357 return CURSEG_HOT_DATA;
3359 return CURSEG_COLD_DATA;
3361 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3362 return CURSEG_WARM_NODE;
3364 return CURSEG_COLD_NODE;
3368 static int __get_segment_type_6(struct f2fs_io_info *fio)
3370 if (fio->type == DATA) {
3371 struct inode *inode = fio->page->mapping->host;
3373 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3374 return CURSEG_COLD_DATA_PINNED;
3376 if (page_private_gcing(fio->page)) {
3377 if (fio->sbi->am.atgc_enabled &&
3378 (fio->io_type == FS_DATA_IO) &&
3379 (fio->sbi->gc_mode != GC_URGENT_HIGH))
3380 return CURSEG_ALL_DATA_ATGC;
3382 return CURSEG_COLD_DATA;
3384 if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3385 return CURSEG_COLD_DATA;
3386 if (file_is_hot(inode) ||
3387 is_inode_flag_set(inode, FI_HOT_DATA) ||
3388 f2fs_is_atomic_file(inode) ||
3389 f2fs_is_volatile_file(inode))
3390 return CURSEG_HOT_DATA;
3391 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3393 if (IS_DNODE(fio->page))
3394 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3396 return CURSEG_COLD_NODE;
3400 static int __get_segment_type(struct f2fs_io_info *fio)
3404 switch (F2FS_OPTION(fio->sbi).active_logs) {
3406 type = __get_segment_type_2(fio);
3409 type = __get_segment_type_4(fio);
3412 type = __get_segment_type_6(fio);
3415 f2fs_bug_on(fio->sbi, true);
3420 else if (IS_WARM(type))
3427 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3428 block_t old_blkaddr, block_t *new_blkaddr,
3429 struct f2fs_summary *sum, int type,
3430 struct f2fs_io_info *fio)
3432 struct sit_info *sit_i = SIT_I(sbi);
3433 struct curseg_info *curseg = CURSEG_I(sbi, type);
3434 unsigned long long old_mtime;
3435 bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3436 struct seg_entry *se = NULL;
3438 f2fs_down_read(&SM_I(sbi)->curseg_lock);
3440 mutex_lock(&curseg->curseg_mutex);
3441 down_write(&sit_i->sentry_lock);
3444 f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3445 se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3446 sanity_check_seg_type(sbi, se->type);
3447 f2fs_bug_on(sbi, IS_NODESEG(se->type));
3449 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3451 f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3453 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3456 * __add_sum_entry should be resided under the curseg_mutex
3457 * because, this function updates a summary entry in the
3458 * current summary block.
3460 __add_sum_entry(sbi, type, sum);
3462 __refresh_next_blkoff(sbi, curseg);
3464 stat_inc_block_count(sbi, curseg);
3467 old_mtime = get_segment_mtime(sbi, old_blkaddr);
3469 update_segment_mtime(sbi, old_blkaddr, 0);
3472 update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3475 * SIT information should be updated before segment allocation,
3476 * since SSR needs latest valid block information.
3478 update_sit_entry(sbi, *new_blkaddr, 1);
3479 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3480 update_sit_entry(sbi, old_blkaddr, -1);
3482 if (!__has_curseg_space(sbi, curseg)) {
3484 get_atssr_segment(sbi, type, se->type,
3487 sit_i->s_ops->allocate_segment(sbi, type, false);
3490 * segment dirty status should be updated after segment allocation,
3491 * so we just need to update status only one time after previous
3492 * segment being closed.
3494 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3495 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3497 up_write(&sit_i->sentry_lock);
3499 if (page && IS_NODESEG(type)) {
3500 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3502 f2fs_inode_chksum_set(sbi, page);
3506 struct f2fs_bio_info *io;
3508 if (F2FS_IO_ALIGNED(sbi))
3511 INIT_LIST_HEAD(&fio->list);
3512 fio->in_list = true;
3513 io = sbi->write_io[fio->type] + fio->temp;
3514 spin_lock(&io->io_lock);
3515 list_add_tail(&fio->list, &io->io_list);
3516 spin_unlock(&io->io_lock);
3519 mutex_unlock(&curseg->curseg_mutex);
3521 f2fs_up_read(&SM_I(sbi)->curseg_lock);
3524 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3525 block_t blkaddr, unsigned int blkcnt)
3527 if (!f2fs_is_multi_device(sbi))
3531 unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3532 unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3534 /* update device state for fsync */
3535 f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3537 /* update device state for checkpoint */
3538 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3539 spin_lock(&sbi->dev_lock);
3540 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3541 spin_unlock(&sbi->dev_lock);
3551 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3553 int type = __get_segment_type(fio);
3554 bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3557 f2fs_down_read(&fio->sbi->io_order_lock);
3559 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3560 &fio->new_blkaddr, sum, type, fio);
3561 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3562 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3563 fio->old_blkaddr, fio->old_blkaddr);
3564 f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3567 /* writeout dirty page into bdev */
3568 f2fs_submit_page_write(fio);
3570 fio->old_blkaddr = fio->new_blkaddr;
3574 f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3577 f2fs_up_read(&fio->sbi->io_order_lock);
3580 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3581 enum iostat_type io_type)
3583 struct f2fs_io_info fio = {
3588 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3589 .old_blkaddr = page->index,
3590 .new_blkaddr = page->index,
3592 .encrypted_page = NULL,
3596 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3597 fio.op_flags &= ~REQ_META;
3599 set_page_writeback(page);
3600 ClearPageError(page);
3601 f2fs_submit_page_write(&fio);
3603 stat_inc_meta_count(sbi, page->index);
3604 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3607 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3609 struct f2fs_summary sum;
3611 set_summary(&sum, nid, 0, 0);
3612 do_write_page(&sum, fio);
3614 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3617 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3618 struct f2fs_io_info *fio)
3620 struct f2fs_sb_info *sbi = fio->sbi;
3621 struct f2fs_summary sum;
3623 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3624 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3625 do_write_page(&sum, fio);
3626 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3628 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3631 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3634 struct f2fs_sb_info *sbi = fio->sbi;
3637 fio->new_blkaddr = fio->old_blkaddr;
3638 /* i/o temperature is needed for passing down write hints */
3639 __get_segment_type(fio);
3641 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3643 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3644 set_sbi_flag(sbi, SBI_NEED_FSCK);
3645 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3647 err = -EFSCORRUPTED;
3651 if (f2fs_cp_error(sbi)) {
3656 invalidate_mapping_pages(META_MAPPING(sbi),
3657 fio->new_blkaddr, fio->new_blkaddr);
3659 stat_inc_inplace_blocks(fio->sbi);
3661 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3662 err = f2fs_merge_page_bio(fio);
3664 err = f2fs_submit_page_bio(fio);
3666 f2fs_update_device_state(fio->sbi, fio->ino,
3667 fio->new_blkaddr, 1);
3668 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3673 if (fio->bio && *(fio->bio)) {
3674 struct bio *bio = *(fio->bio);
3676 bio->bi_status = BLK_STS_IOERR;
3683 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3688 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3689 if (CURSEG_I(sbi, i)->segno == segno)
3695 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3696 block_t old_blkaddr, block_t new_blkaddr,
3697 bool recover_curseg, bool recover_newaddr,
3700 struct sit_info *sit_i = SIT_I(sbi);
3701 struct curseg_info *curseg;
3702 unsigned int segno, old_cursegno;
3703 struct seg_entry *se;
3705 unsigned short old_blkoff;
3706 unsigned char old_alloc_type;
3708 segno = GET_SEGNO(sbi, new_blkaddr);
3709 se = get_seg_entry(sbi, segno);
3712 f2fs_down_write(&SM_I(sbi)->curseg_lock);
3714 if (!recover_curseg) {
3715 /* for recovery flow */
3716 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3717 if (old_blkaddr == NULL_ADDR)
3718 type = CURSEG_COLD_DATA;
3720 type = CURSEG_WARM_DATA;
3723 if (IS_CURSEG(sbi, segno)) {
3724 /* se->type is volatile as SSR allocation */
3725 type = __f2fs_get_curseg(sbi, segno);
3726 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3728 type = CURSEG_WARM_DATA;
3732 f2fs_bug_on(sbi, !IS_DATASEG(type));
3733 curseg = CURSEG_I(sbi, type);
3735 mutex_lock(&curseg->curseg_mutex);
3736 down_write(&sit_i->sentry_lock);
3738 old_cursegno = curseg->segno;
3739 old_blkoff = curseg->next_blkoff;
3740 old_alloc_type = curseg->alloc_type;
3742 /* change the current segment */
3743 if (segno != curseg->segno) {
3744 curseg->next_segno = segno;
3745 change_curseg(sbi, type, true);
3748 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3749 __add_sum_entry(sbi, type, sum);
3751 if (!recover_curseg || recover_newaddr) {
3753 update_segment_mtime(sbi, new_blkaddr, 0);
3754 update_sit_entry(sbi, new_blkaddr, 1);
3756 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3757 invalidate_mapping_pages(META_MAPPING(sbi),
3758 old_blkaddr, old_blkaddr);
3759 f2fs_invalidate_compress_page(sbi, old_blkaddr);
3761 update_segment_mtime(sbi, old_blkaddr, 0);
3762 update_sit_entry(sbi, old_blkaddr, -1);
3765 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3766 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3768 locate_dirty_segment(sbi, old_cursegno);
3770 if (recover_curseg) {
3771 if (old_cursegno != curseg->segno) {
3772 curseg->next_segno = old_cursegno;
3773 change_curseg(sbi, type, true);
3775 curseg->next_blkoff = old_blkoff;
3776 curseg->alloc_type = old_alloc_type;
3779 up_write(&sit_i->sentry_lock);
3780 mutex_unlock(&curseg->curseg_mutex);
3781 f2fs_up_write(&SM_I(sbi)->curseg_lock);
3784 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3785 block_t old_addr, block_t new_addr,
3786 unsigned char version, bool recover_curseg,
3787 bool recover_newaddr)
3789 struct f2fs_summary sum;
3791 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3793 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3794 recover_curseg, recover_newaddr, false);
3796 f2fs_update_data_blkaddr(dn, new_addr);
3799 void f2fs_wait_on_page_writeback(struct page *page,
3800 enum page_type type, bool ordered, bool locked)
3802 if (PageWriteback(page)) {
3803 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3805 /* submit cached LFS IO */
3806 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3807 /* sbumit cached IPU IO */
3808 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3810 wait_on_page_writeback(page);
3811 f2fs_bug_on(sbi, locked && PageWriteback(page));
3813 wait_for_stable_page(page);
3818 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3820 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3823 if (!f2fs_post_read_required(inode))
3826 if (!__is_valid_data_blkaddr(blkaddr))
3829 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3831 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3832 f2fs_put_page(cpage, 1);
3836 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3841 for (i = 0; i < len; i++)
3842 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3845 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3847 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3848 struct curseg_info *seg_i;
3849 unsigned char *kaddr;
3854 start = start_sum_block(sbi);
3856 page = f2fs_get_meta_page(sbi, start++);
3858 return PTR_ERR(page);
3859 kaddr = (unsigned char *)page_address(page);
3861 /* Step 1: restore nat cache */
3862 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3863 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3865 /* Step 2: restore sit cache */
3866 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3867 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3868 offset = 2 * SUM_JOURNAL_SIZE;
3870 /* Step 3: restore summary entries */
3871 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3872 unsigned short blk_off;
3875 seg_i = CURSEG_I(sbi, i);
3876 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3877 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3878 seg_i->next_segno = segno;
3879 reset_curseg(sbi, i, 0);
3880 seg_i->alloc_type = ckpt->alloc_type[i];
3881 seg_i->next_blkoff = blk_off;
3883 if (seg_i->alloc_type == SSR)
3884 blk_off = sbi->blocks_per_seg;
3886 for (j = 0; j < blk_off; j++) {
3887 struct f2fs_summary *s;
3889 s = (struct f2fs_summary *)(kaddr + offset);
3890 seg_i->sum_blk->entries[j] = *s;
3891 offset += SUMMARY_SIZE;
3892 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3896 f2fs_put_page(page, 1);
3899 page = f2fs_get_meta_page(sbi, start++);
3901 return PTR_ERR(page);
3902 kaddr = (unsigned char *)page_address(page);
3906 f2fs_put_page(page, 1);
3910 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3912 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3913 struct f2fs_summary_block *sum;
3914 struct curseg_info *curseg;
3916 unsigned short blk_off;
3917 unsigned int segno = 0;
3918 block_t blk_addr = 0;
3921 /* get segment number and block addr */
3922 if (IS_DATASEG(type)) {
3923 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3924 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3926 if (__exist_node_summaries(sbi))
3927 blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3929 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3931 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3933 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3935 if (__exist_node_summaries(sbi))
3936 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3937 type - CURSEG_HOT_NODE);
3939 blk_addr = GET_SUM_BLOCK(sbi, segno);
3942 new = f2fs_get_meta_page(sbi, blk_addr);
3944 return PTR_ERR(new);
3945 sum = (struct f2fs_summary_block *)page_address(new);
3947 if (IS_NODESEG(type)) {
3948 if (__exist_node_summaries(sbi)) {
3949 struct f2fs_summary *ns = &sum->entries[0];
3952 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3954 ns->ofs_in_node = 0;
3957 err = f2fs_restore_node_summary(sbi, segno, sum);
3963 /* set uncompleted segment to curseg */
3964 curseg = CURSEG_I(sbi, type);
3965 mutex_lock(&curseg->curseg_mutex);
3967 /* update journal info */
3968 down_write(&curseg->journal_rwsem);
3969 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3970 up_write(&curseg->journal_rwsem);
3972 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3973 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3974 curseg->next_segno = segno;
3975 reset_curseg(sbi, type, 0);
3976 curseg->alloc_type = ckpt->alloc_type[type];
3977 curseg->next_blkoff = blk_off;
3978 mutex_unlock(&curseg->curseg_mutex);
3980 f2fs_put_page(new, 1);
3984 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3986 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3987 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3988 int type = CURSEG_HOT_DATA;
3991 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3992 int npages = f2fs_npages_for_summary_flush(sbi, true);
3995 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3998 /* restore for compacted data summary */
3999 err = read_compacted_summaries(sbi);
4002 type = CURSEG_HOT_NODE;
4005 if (__exist_node_summaries(sbi))
4006 f2fs_ra_meta_pages(sbi,
4007 sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4008 NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4010 for (; type <= CURSEG_COLD_NODE; type++) {
4011 err = read_normal_summaries(sbi, type);
4016 /* sanity check for summary blocks */
4017 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4018 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4019 f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4020 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4027 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4030 unsigned char *kaddr;
4031 struct f2fs_summary *summary;
4032 struct curseg_info *seg_i;
4033 int written_size = 0;
4036 page = f2fs_grab_meta_page(sbi, blkaddr++);
4037 kaddr = (unsigned char *)page_address(page);
4038 memset(kaddr, 0, PAGE_SIZE);
4040 /* Step 1: write nat cache */
4041 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4042 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4043 written_size += SUM_JOURNAL_SIZE;
4045 /* Step 2: write sit cache */
4046 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4047 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4048 written_size += SUM_JOURNAL_SIZE;
4050 /* Step 3: write summary entries */
4051 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4052 unsigned short blkoff;
4054 seg_i = CURSEG_I(sbi, i);
4055 if (sbi->ckpt->alloc_type[i] == SSR)
4056 blkoff = sbi->blocks_per_seg;
4058 blkoff = curseg_blkoff(sbi, i);
4060 for (j = 0; j < blkoff; j++) {
4062 page = f2fs_grab_meta_page(sbi, blkaddr++);
4063 kaddr = (unsigned char *)page_address(page);
4064 memset(kaddr, 0, PAGE_SIZE);
4067 summary = (struct f2fs_summary *)(kaddr + written_size);
4068 *summary = seg_i->sum_blk->entries[j];
4069 written_size += SUMMARY_SIZE;
4071 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4075 set_page_dirty(page);
4076 f2fs_put_page(page, 1);
4081 set_page_dirty(page);
4082 f2fs_put_page(page, 1);
4086 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4087 block_t blkaddr, int type)
4091 if (IS_DATASEG(type))
4092 end = type + NR_CURSEG_DATA_TYPE;
4094 end = type + NR_CURSEG_NODE_TYPE;
4096 for (i = type; i < end; i++)
4097 write_current_sum_page(sbi, i, blkaddr + (i - type));
4100 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4102 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4103 write_compacted_summaries(sbi, start_blk);
4105 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4108 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4110 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4113 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4114 unsigned int val, int alloc)
4118 if (type == NAT_JOURNAL) {
4119 for (i = 0; i < nats_in_cursum(journal); i++) {
4120 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4123 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4124 return update_nats_in_cursum(journal, 1);
4125 } else if (type == SIT_JOURNAL) {
4126 for (i = 0; i < sits_in_cursum(journal); i++)
4127 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4129 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4130 return update_sits_in_cursum(journal, 1);
4135 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4138 return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4141 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4144 struct sit_info *sit_i = SIT_I(sbi);
4146 pgoff_t src_off, dst_off;
4148 src_off = current_sit_addr(sbi, start);
4149 dst_off = next_sit_addr(sbi, src_off);
4151 page = f2fs_grab_meta_page(sbi, dst_off);
4152 seg_info_to_sit_page(sbi, page, start);
4154 set_page_dirty(page);
4155 set_to_next_sit(sit_i, start);
4160 static struct sit_entry_set *grab_sit_entry_set(void)
4162 struct sit_entry_set *ses =
4163 f2fs_kmem_cache_alloc(sit_entry_set_slab,
4164 GFP_NOFS, true, NULL);
4167 INIT_LIST_HEAD(&ses->set_list);
4171 static void release_sit_entry_set(struct sit_entry_set *ses)
4173 list_del(&ses->set_list);
4174 kmem_cache_free(sit_entry_set_slab, ses);
4177 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4178 struct list_head *head)
4180 struct sit_entry_set *next = ses;
4182 if (list_is_last(&ses->set_list, head))
4185 list_for_each_entry_continue(next, head, set_list)
4186 if (ses->entry_cnt <= next->entry_cnt)
4189 list_move_tail(&ses->set_list, &next->set_list);
4192 static void add_sit_entry(unsigned int segno, struct list_head *head)
4194 struct sit_entry_set *ses;
4195 unsigned int start_segno = START_SEGNO(segno);
4197 list_for_each_entry(ses, head, set_list) {
4198 if (ses->start_segno == start_segno) {
4200 adjust_sit_entry_set(ses, head);
4205 ses = grab_sit_entry_set();
4207 ses->start_segno = start_segno;
4209 list_add(&ses->set_list, head);
4212 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4214 struct f2fs_sm_info *sm_info = SM_I(sbi);
4215 struct list_head *set_list = &sm_info->sit_entry_set;
4216 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4219 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4220 add_sit_entry(segno, set_list);
4223 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4225 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4226 struct f2fs_journal *journal = curseg->journal;
4229 down_write(&curseg->journal_rwsem);
4230 for (i = 0; i < sits_in_cursum(journal); i++) {
4234 segno = le32_to_cpu(segno_in_journal(journal, i));
4235 dirtied = __mark_sit_entry_dirty(sbi, segno);
4238 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4240 update_sits_in_cursum(journal, -i);
4241 up_write(&curseg->journal_rwsem);
4245 * CP calls this function, which flushes SIT entries including sit_journal,
4246 * and moves prefree segs to free segs.
4248 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4250 struct sit_info *sit_i = SIT_I(sbi);
4251 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4252 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4253 struct f2fs_journal *journal = curseg->journal;
4254 struct sit_entry_set *ses, *tmp;
4255 struct list_head *head = &SM_I(sbi)->sit_entry_set;
4256 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4257 struct seg_entry *se;
4259 down_write(&sit_i->sentry_lock);
4261 if (!sit_i->dirty_sentries)
4265 * add and account sit entries of dirty bitmap in sit entry
4268 add_sits_in_set(sbi);
4271 * if there are no enough space in journal to store dirty sit
4272 * entries, remove all entries from journal and add and account
4273 * them in sit entry set.
4275 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4277 remove_sits_in_journal(sbi);
4280 * there are two steps to flush sit entries:
4281 * #1, flush sit entries to journal in current cold data summary block.
4282 * #2, flush sit entries to sit page.
4284 list_for_each_entry_safe(ses, tmp, head, set_list) {
4285 struct page *page = NULL;
4286 struct f2fs_sit_block *raw_sit = NULL;
4287 unsigned int start_segno = ses->start_segno;
4288 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4289 (unsigned long)MAIN_SEGS(sbi));
4290 unsigned int segno = start_segno;
4293 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4297 down_write(&curseg->journal_rwsem);
4299 page = get_next_sit_page(sbi, start_segno);
4300 raw_sit = page_address(page);
4303 /* flush dirty sit entries in region of current sit set */
4304 for_each_set_bit_from(segno, bitmap, end) {
4305 int offset, sit_offset;
4307 se = get_seg_entry(sbi, segno);
4308 #ifdef CONFIG_F2FS_CHECK_FS
4309 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4310 SIT_VBLOCK_MAP_SIZE))
4311 f2fs_bug_on(sbi, 1);
4314 /* add discard candidates */
4315 if (!(cpc->reason & CP_DISCARD)) {
4316 cpc->trim_start = segno;
4317 add_discard_addrs(sbi, cpc, false);
4321 offset = f2fs_lookup_journal_in_cursum(journal,
4322 SIT_JOURNAL, segno, 1);
4323 f2fs_bug_on(sbi, offset < 0);
4324 segno_in_journal(journal, offset) =
4326 seg_info_to_raw_sit(se,
4327 &sit_in_journal(journal, offset));
4328 check_block_count(sbi, segno,
4329 &sit_in_journal(journal, offset));
4331 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4332 seg_info_to_raw_sit(se,
4333 &raw_sit->entries[sit_offset]);
4334 check_block_count(sbi, segno,
4335 &raw_sit->entries[sit_offset]);
4338 __clear_bit(segno, bitmap);
4339 sit_i->dirty_sentries--;
4344 up_write(&curseg->journal_rwsem);
4346 f2fs_put_page(page, 1);
4348 f2fs_bug_on(sbi, ses->entry_cnt);
4349 release_sit_entry_set(ses);
4352 f2fs_bug_on(sbi, !list_empty(head));
4353 f2fs_bug_on(sbi, sit_i->dirty_sentries);
4355 if (cpc->reason & CP_DISCARD) {
4356 __u64 trim_start = cpc->trim_start;
4358 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4359 add_discard_addrs(sbi, cpc, false);
4361 cpc->trim_start = trim_start;
4363 up_write(&sit_i->sentry_lock);
4365 set_prefree_as_free_segments(sbi);
4368 static int build_sit_info(struct f2fs_sb_info *sbi)
4370 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4371 struct sit_info *sit_i;
4372 unsigned int sit_segs, start;
4373 char *src_bitmap, *bitmap;
4374 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4375 unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4377 /* allocate memory for SIT information */
4378 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4382 SM_I(sbi)->sit_info = sit_i;
4385 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4388 if (!sit_i->sentries)
4391 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4392 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4394 if (!sit_i->dirty_sentries_bitmap)
4397 #ifdef CONFIG_F2FS_CHECK_FS
4398 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4400 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4402 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4406 bitmap = sit_i->bitmap;
4408 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4409 sit_i->sentries[start].cur_valid_map = bitmap;
4410 bitmap += SIT_VBLOCK_MAP_SIZE;
4412 sit_i->sentries[start].ckpt_valid_map = bitmap;
4413 bitmap += SIT_VBLOCK_MAP_SIZE;
4415 #ifdef CONFIG_F2FS_CHECK_FS
4416 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4417 bitmap += SIT_VBLOCK_MAP_SIZE;
4421 sit_i->sentries[start].discard_map = bitmap;
4422 bitmap += SIT_VBLOCK_MAP_SIZE;
4426 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4427 if (!sit_i->tmp_map)
4430 if (__is_large_section(sbi)) {
4431 sit_i->sec_entries =
4432 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4435 if (!sit_i->sec_entries)
4439 /* get information related with SIT */
4440 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4442 /* setup SIT bitmap from ckeckpoint pack */
4443 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4444 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4446 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4447 if (!sit_i->sit_bitmap)
4450 #ifdef CONFIG_F2FS_CHECK_FS
4451 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4452 sit_bitmap_size, GFP_KERNEL);
4453 if (!sit_i->sit_bitmap_mir)
4456 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4457 main_bitmap_size, GFP_KERNEL);
4458 if (!sit_i->invalid_segmap)
4462 /* init SIT information */
4463 sit_i->s_ops = &default_salloc_ops;
4465 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4466 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4467 sit_i->written_valid_blocks = 0;
4468 sit_i->bitmap_size = sit_bitmap_size;
4469 sit_i->dirty_sentries = 0;
4470 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4471 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4472 sit_i->mounted_time = ktime_get_boottime_seconds();
4473 init_rwsem(&sit_i->sentry_lock);
4477 static int build_free_segmap(struct f2fs_sb_info *sbi)
4479 struct free_segmap_info *free_i;
4480 unsigned int bitmap_size, sec_bitmap_size;
4482 /* allocate memory for free segmap information */
4483 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4487 SM_I(sbi)->free_info = free_i;
4489 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4490 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4491 if (!free_i->free_segmap)
4494 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4495 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4496 if (!free_i->free_secmap)
4499 /* set all segments as dirty temporarily */
4500 memset(free_i->free_segmap, 0xff, bitmap_size);
4501 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4503 /* init free segmap information */
4504 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4505 free_i->free_segments = 0;
4506 free_i->free_sections = 0;
4507 spin_lock_init(&free_i->segmap_lock);
4511 static int build_curseg(struct f2fs_sb_info *sbi)
4513 struct curseg_info *array;
4516 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4517 sizeof(*array)), GFP_KERNEL);
4521 SM_I(sbi)->curseg_array = array;
4523 for (i = 0; i < NO_CHECK_TYPE; i++) {
4524 mutex_init(&array[i].curseg_mutex);
4525 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4526 if (!array[i].sum_blk)
4528 init_rwsem(&array[i].journal_rwsem);
4529 array[i].journal = f2fs_kzalloc(sbi,
4530 sizeof(struct f2fs_journal), GFP_KERNEL);
4531 if (!array[i].journal)
4533 if (i < NR_PERSISTENT_LOG)
4534 array[i].seg_type = CURSEG_HOT_DATA + i;
4535 else if (i == CURSEG_COLD_DATA_PINNED)
4536 array[i].seg_type = CURSEG_COLD_DATA;
4537 else if (i == CURSEG_ALL_DATA_ATGC)
4538 array[i].seg_type = CURSEG_COLD_DATA;
4539 array[i].segno = NULL_SEGNO;
4540 array[i].next_blkoff = 0;
4541 array[i].inited = false;
4543 return restore_curseg_summaries(sbi);
4546 static int build_sit_entries(struct f2fs_sb_info *sbi)
4548 struct sit_info *sit_i = SIT_I(sbi);
4549 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4550 struct f2fs_journal *journal = curseg->journal;
4551 struct seg_entry *se;
4552 struct f2fs_sit_entry sit;
4553 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4554 unsigned int i, start, end;
4555 unsigned int readed, start_blk = 0;
4557 block_t total_node_blocks = 0;
4560 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4563 start = start_blk * sit_i->sents_per_block;
4564 end = (start_blk + readed) * sit_i->sents_per_block;
4566 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4567 struct f2fs_sit_block *sit_blk;
4570 se = &sit_i->sentries[start];
4571 page = get_current_sit_page(sbi, start);
4573 return PTR_ERR(page);
4574 sit_blk = (struct f2fs_sit_block *)page_address(page);
4575 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4576 f2fs_put_page(page, 1);
4578 err = check_block_count(sbi, start, &sit);
4581 seg_info_from_raw_sit(se, &sit);
4582 if (IS_NODESEG(se->type))
4583 total_node_blocks += se->valid_blocks;
4585 if (f2fs_block_unit_discard(sbi)) {
4586 /* build discard map only one time */
4587 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4588 memset(se->discard_map, 0xff,
4589 SIT_VBLOCK_MAP_SIZE);
4591 memcpy(se->discard_map,
4593 SIT_VBLOCK_MAP_SIZE);
4594 sbi->discard_blks +=
4595 sbi->blocks_per_seg -
4600 if (__is_large_section(sbi))
4601 get_sec_entry(sbi, start)->valid_blocks +=
4604 start_blk += readed;
4605 } while (start_blk < sit_blk_cnt);
4607 down_read(&curseg->journal_rwsem);
4608 for (i = 0; i < sits_in_cursum(journal); i++) {
4609 unsigned int old_valid_blocks;
4611 start = le32_to_cpu(segno_in_journal(journal, i));
4612 if (start >= MAIN_SEGS(sbi)) {
4613 f2fs_err(sbi, "Wrong journal entry on segno %u",
4615 err = -EFSCORRUPTED;
4619 se = &sit_i->sentries[start];
4620 sit = sit_in_journal(journal, i);
4622 old_valid_blocks = se->valid_blocks;
4623 if (IS_NODESEG(se->type))
4624 total_node_blocks -= old_valid_blocks;
4626 err = check_block_count(sbi, start, &sit);
4629 seg_info_from_raw_sit(se, &sit);
4630 if (IS_NODESEG(se->type))
4631 total_node_blocks += se->valid_blocks;
4633 if (f2fs_block_unit_discard(sbi)) {
4634 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4635 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4637 memcpy(se->discard_map, se->cur_valid_map,
4638 SIT_VBLOCK_MAP_SIZE);
4639 sbi->discard_blks += old_valid_blocks;
4640 sbi->discard_blks -= se->valid_blocks;
4644 if (__is_large_section(sbi)) {
4645 get_sec_entry(sbi, start)->valid_blocks +=
4647 get_sec_entry(sbi, start)->valid_blocks -=
4651 up_read(&curseg->journal_rwsem);
4653 if (!err && total_node_blocks != valid_node_count(sbi)) {
4654 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4655 total_node_blocks, valid_node_count(sbi));
4656 err = -EFSCORRUPTED;
4662 static void init_free_segmap(struct f2fs_sb_info *sbi)
4666 struct seg_entry *sentry;
4668 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4669 if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4671 sentry = get_seg_entry(sbi, start);
4672 if (!sentry->valid_blocks)
4673 __set_free(sbi, start);
4675 SIT_I(sbi)->written_valid_blocks +=
4676 sentry->valid_blocks;
4679 /* set use the current segments */
4680 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4681 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4683 __set_test_and_inuse(sbi, curseg_t->segno);
4687 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4689 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4690 struct free_segmap_info *free_i = FREE_I(sbi);
4691 unsigned int segno = 0, offset = 0, secno;
4692 block_t valid_blocks, usable_blks_in_seg;
4693 block_t blks_per_sec = BLKS_PER_SEC(sbi);
4696 /* find dirty segment based on free segmap */
4697 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4698 if (segno >= MAIN_SEGS(sbi))
4701 valid_blocks = get_valid_blocks(sbi, segno, false);
4702 usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4703 if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4705 if (valid_blocks > usable_blks_in_seg) {
4706 f2fs_bug_on(sbi, 1);
4709 mutex_lock(&dirty_i->seglist_lock);
4710 __locate_dirty_segment(sbi, segno, DIRTY);
4711 mutex_unlock(&dirty_i->seglist_lock);
4714 if (!__is_large_section(sbi))
4717 mutex_lock(&dirty_i->seglist_lock);
4718 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4719 valid_blocks = get_valid_blocks(sbi, segno, true);
4720 secno = GET_SEC_FROM_SEG(sbi, segno);
4722 if (!valid_blocks || valid_blocks == blks_per_sec)
4724 if (IS_CURSEC(sbi, secno))
4726 set_bit(secno, dirty_i->dirty_secmap);
4728 mutex_unlock(&dirty_i->seglist_lock);
4731 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4733 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4734 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4736 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4737 if (!dirty_i->victim_secmap)
4742 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4744 struct dirty_seglist_info *dirty_i;
4745 unsigned int bitmap_size, i;
4747 /* allocate memory for dirty segments list information */
4748 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4753 SM_I(sbi)->dirty_info = dirty_i;
4754 mutex_init(&dirty_i->seglist_lock);
4756 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4758 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4759 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4761 if (!dirty_i->dirty_segmap[i])
4765 if (__is_large_section(sbi)) {
4766 bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4767 dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4768 bitmap_size, GFP_KERNEL);
4769 if (!dirty_i->dirty_secmap)
4773 init_dirty_segmap(sbi);
4774 return init_victim_secmap(sbi);
4777 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4782 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4783 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4785 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4786 struct curseg_info *curseg = CURSEG_I(sbi, i);
4787 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4788 unsigned int blkofs = curseg->next_blkoff;
4790 if (f2fs_sb_has_readonly(sbi) &&
4791 i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4794 sanity_check_seg_type(sbi, curseg->seg_type);
4796 if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4798 "Current segment has invalid alloc_type:%d",
4799 curseg->alloc_type);
4800 return -EFSCORRUPTED;
4803 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4806 if (curseg->alloc_type == SSR)
4809 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4810 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4814 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4815 i, curseg->segno, curseg->alloc_type,
4816 curseg->next_blkoff, blkofs);
4817 return -EFSCORRUPTED;
4823 #ifdef CONFIG_BLK_DEV_ZONED
4825 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4826 struct f2fs_dev_info *fdev,
4827 struct blk_zone *zone)
4829 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4830 block_t zone_block, wp_block, last_valid_block;
4831 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4833 struct seg_entry *se;
4835 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4838 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4839 wp_segno = GET_SEGNO(sbi, wp_block);
4840 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4841 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4842 zone_segno = GET_SEGNO(sbi, zone_block);
4843 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4845 if (zone_segno >= MAIN_SEGS(sbi))
4849 * Skip check of zones cursegs point to, since
4850 * fix_curseg_write_pointer() checks them.
4852 for (i = 0; i < NO_CHECK_TYPE; i++)
4853 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4854 CURSEG_I(sbi, i)->segno))
4858 * Get last valid block of the zone.
4860 last_valid_block = zone_block - 1;
4861 for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4862 segno = zone_segno + s;
4863 se = get_seg_entry(sbi, segno);
4864 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4865 if (f2fs_test_bit(b, se->cur_valid_map)) {
4866 last_valid_block = START_BLOCK(sbi, segno) + b;
4869 if (last_valid_block >= zone_block)
4874 * If last valid block is beyond the write pointer, report the
4875 * inconsistency. This inconsistency does not cause write error
4876 * because the zone will not be selected for write operation until
4877 * it get discarded. Just report it.
4879 if (last_valid_block >= wp_block) {
4880 f2fs_notice(sbi, "Valid block beyond write pointer: "
4881 "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4882 GET_SEGNO(sbi, last_valid_block),
4883 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4884 wp_segno, wp_blkoff);
4889 * If there is no valid block in the zone and if write pointer is
4890 * not at zone start, reset the write pointer.
4892 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4894 "Zone without valid block has non-zero write "
4895 "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4896 wp_segno, wp_blkoff);
4897 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4898 zone->len >> log_sectors_per_block);
4900 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4909 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4910 block_t zone_blkaddr)
4914 for (i = 0; i < sbi->s_ndevs; i++) {
4915 if (!bdev_is_zoned(FDEV(i).bdev))
4917 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4918 zone_blkaddr <= FDEV(i).end_blk))
4925 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4928 memcpy(data, zone, sizeof(struct blk_zone));
4932 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4934 struct curseg_info *cs = CURSEG_I(sbi, type);
4935 struct f2fs_dev_info *zbd;
4936 struct blk_zone zone;
4937 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4938 block_t cs_zone_block, wp_block;
4939 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4940 sector_t zone_sector;
4943 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4944 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4946 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4950 /* report zone for the sector the curseg points to */
4951 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4952 << log_sectors_per_block;
4953 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4954 report_one_zone_cb, &zone);
4956 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4961 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4964 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4965 wp_segno = GET_SEGNO(sbi, wp_block);
4966 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4967 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4969 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4973 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4974 "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4975 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4977 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4978 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4980 f2fs_allocate_new_section(sbi, type, true);
4982 /* check consistency of the zone curseg pointed to */
4983 if (check_zone_write_pointer(sbi, zbd, &zone))
4986 /* check newly assigned zone */
4987 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4988 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4990 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4994 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4995 << log_sectors_per_block;
4996 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4997 report_one_zone_cb, &zone);
4999 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5004 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5007 if (zone.wp != zone.start) {
5009 "New zone for curseg[%d] is not yet discarded. "
5010 "Reset the zone: curseg[0x%x,0x%x]",
5011 type, cs->segno, cs->next_blkoff);
5012 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
5013 zone_sector >> log_sectors_per_block,
5014 zone.len >> log_sectors_per_block);
5016 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5025 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5029 for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5030 ret = fix_curseg_write_pointer(sbi, i);
5038 struct check_zone_write_pointer_args {
5039 struct f2fs_sb_info *sbi;
5040 struct f2fs_dev_info *fdev;
5043 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5046 struct check_zone_write_pointer_args *args;
5048 args = (struct check_zone_write_pointer_args *)data;
5050 return check_zone_write_pointer(args->sbi, args->fdev, zone);
5053 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5056 struct check_zone_write_pointer_args args;
5058 for (i = 0; i < sbi->s_ndevs; i++) {
5059 if (!bdev_is_zoned(FDEV(i).bdev))
5063 args.fdev = &FDEV(i);
5064 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5065 check_zone_write_pointer_cb, &args);
5073 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
5074 unsigned int dev_idx)
5076 if (!bdev_is_zoned(FDEV(dev_idx).bdev))
5078 return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
5081 /* Return the zone index in the given device */
5082 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
5085 block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5087 return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
5088 sbi->log_blocks_per_blkz;
5092 * Return the usable segments in a section based on the zone's
5093 * corresponding zone capacity. Zone is equal to a section.
5095 static inline unsigned int f2fs_usable_zone_segs_in_sec(
5096 struct f2fs_sb_info *sbi, unsigned int segno)
5098 unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
5100 dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
5101 zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
5103 /* Conventional zone's capacity is always equal to zone size */
5104 if (is_conv_zone(sbi, zone_idx, dev_idx))
5105 return sbi->segs_per_sec;
5108 * If the zone_capacity_blocks array is NULL, then zone capacity
5109 * is equal to the zone size for all zones
5111 if (!FDEV(dev_idx).zone_capacity_blocks)
5112 return sbi->segs_per_sec;
5114 /* Get the segment count beyond zone capacity block */
5115 unusable_segs_in_sec = (sbi->blocks_per_blkz -
5116 FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
5117 sbi->log_blocks_per_seg;
5118 return sbi->segs_per_sec - unusable_segs_in_sec;
5122 * Return the number of usable blocks in a segment. The number of blocks
5123 * returned is always equal to the number of blocks in a segment for
5124 * segments fully contained within a sequential zone capacity or a
5125 * conventional zone. For segments partially contained in a sequential
5126 * zone capacity, the number of usable blocks up to the zone capacity
5127 * is returned. 0 is returned in all other cases.
5129 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5130 struct f2fs_sb_info *sbi, unsigned int segno)
5132 block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5133 unsigned int zone_idx, dev_idx, secno;
5135 secno = GET_SEC_FROM_SEG(sbi, segno);
5136 seg_start = START_BLOCK(sbi, segno);
5137 dev_idx = f2fs_target_device_index(sbi, seg_start);
5138 zone_idx = get_zone_idx(sbi, secno, dev_idx);
5141 * Conventional zone's capacity is always equal to zone size,
5142 * so, blocks per segment is unchanged.
5144 if (is_conv_zone(sbi, zone_idx, dev_idx))
5145 return sbi->blocks_per_seg;
5147 if (!FDEV(dev_idx).zone_capacity_blocks)
5148 return sbi->blocks_per_seg;
5150 sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5151 sec_cap_blkaddr = sec_start_blkaddr +
5152 FDEV(dev_idx).zone_capacity_blocks[zone_idx];
5155 * If segment starts before zone capacity and spans beyond
5156 * zone capacity, then usable blocks are from seg start to
5157 * zone capacity. If the segment starts after the zone capacity,
5158 * then there are no usable blocks.
5160 if (seg_start >= sec_cap_blkaddr)
5162 if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5163 return sec_cap_blkaddr - seg_start;
5165 return sbi->blocks_per_seg;
5168 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5173 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5178 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5184 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5190 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5193 if (f2fs_sb_has_blkzoned(sbi))
5194 return f2fs_usable_zone_blks_in_seg(sbi, segno);
5196 return sbi->blocks_per_seg;
5199 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5202 if (f2fs_sb_has_blkzoned(sbi))
5203 return f2fs_usable_zone_segs_in_sec(sbi, segno);
5205 return sbi->segs_per_sec;
5209 * Update min, max modified time for cost-benefit GC algorithm
5211 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5213 struct sit_info *sit_i = SIT_I(sbi);
5216 down_write(&sit_i->sentry_lock);
5218 sit_i->min_mtime = ULLONG_MAX;
5220 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5222 unsigned long long mtime = 0;
5224 for (i = 0; i < sbi->segs_per_sec; i++)
5225 mtime += get_seg_entry(sbi, segno + i)->mtime;
5227 mtime = div_u64(mtime, sbi->segs_per_sec);
5229 if (sit_i->min_mtime > mtime)
5230 sit_i->min_mtime = mtime;
5232 sit_i->max_mtime = get_mtime(sbi, false);
5233 sit_i->dirty_max_mtime = 0;
5234 up_write(&sit_i->sentry_lock);
5237 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5239 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5240 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5241 struct f2fs_sm_info *sm_info;
5244 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5249 sbi->sm_info = sm_info;
5250 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5251 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5252 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5253 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5254 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5255 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5256 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5257 sm_info->rec_prefree_segments = sm_info->main_segments *
5258 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5259 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5260 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5262 if (!f2fs_lfs_mode(sbi))
5263 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5264 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5265 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5266 sm_info->min_seq_blocks = sbi->blocks_per_seg;
5267 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5268 sm_info->min_ssr_sections = reserved_sections(sbi);
5270 INIT_LIST_HEAD(&sm_info->sit_entry_set);
5272 init_f2fs_rwsem(&sm_info->curseg_lock);
5274 if (!f2fs_readonly(sbi->sb)) {
5275 err = f2fs_create_flush_cmd_control(sbi);
5280 err = create_discard_cmd_control(sbi);
5284 err = build_sit_info(sbi);
5287 err = build_free_segmap(sbi);
5290 err = build_curseg(sbi);
5294 /* reinit free segmap based on SIT */
5295 err = build_sit_entries(sbi);
5299 init_free_segmap(sbi);
5300 err = build_dirty_segmap(sbi);
5304 err = sanity_check_curseg(sbi);
5308 init_min_max_mtime(sbi);
5312 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5313 enum dirty_type dirty_type)
5315 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5317 mutex_lock(&dirty_i->seglist_lock);
5318 kvfree(dirty_i->dirty_segmap[dirty_type]);
5319 dirty_i->nr_dirty[dirty_type] = 0;
5320 mutex_unlock(&dirty_i->seglist_lock);
5323 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5325 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5327 kvfree(dirty_i->victim_secmap);
5330 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5332 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5338 /* discard pre-free/dirty segments list */
5339 for (i = 0; i < NR_DIRTY_TYPE; i++)
5340 discard_dirty_segmap(sbi, i);
5342 if (__is_large_section(sbi)) {
5343 mutex_lock(&dirty_i->seglist_lock);
5344 kvfree(dirty_i->dirty_secmap);
5345 mutex_unlock(&dirty_i->seglist_lock);
5348 destroy_victim_secmap(sbi);
5349 SM_I(sbi)->dirty_info = NULL;
5353 static void destroy_curseg(struct f2fs_sb_info *sbi)
5355 struct curseg_info *array = SM_I(sbi)->curseg_array;
5360 SM_I(sbi)->curseg_array = NULL;
5361 for (i = 0; i < NR_CURSEG_TYPE; i++) {
5362 kfree(array[i].sum_blk);
5363 kfree(array[i].journal);
5368 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5370 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5374 SM_I(sbi)->free_info = NULL;
5375 kvfree(free_i->free_segmap);
5376 kvfree(free_i->free_secmap);
5380 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5382 struct sit_info *sit_i = SIT_I(sbi);
5387 if (sit_i->sentries)
5388 kvfree(sit_i->bitmap);
5389 kfree(sit_i->tmp_map);
5391 kvfree(sit_i->sentries);
5392 kvfree(sit_i->sec_entries);
5393 kvfree(sit_i->dirty_sentries_bitmap);
5395 SM_I(sbi)->sit_info = NULL;
5396 kvfree(sit_i->sit_bitmap);
5397 #ifdef CONFIG_F2FS_CHECK_FS
5398 kvfree(sit_i->sit_bitmap_mir);
5399 kvfree(sit_i->invalid_segmap);
5404 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5406 struct f2fs_sm_info *sm_info = SM_I(sbi);
5410 f2fs_destroy_flush_cmd_control(sbi, true);
5411 destroy_discard_cmd_control(sbi);
5412 destroy_dirty_segmap(sbi);
5413 destroy_curseg(sbi);
5414 destroy_free_segmap(sbi);
5415 destroy_sit_info(sbi);
5416 sbi->sm_info = NULL;
5420 int __init f2fs_create_segment_manager_caches(void)
5422 discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5423 sizeof(struct discard_entry));
5424 if (!discard_entry_slab)
5427 discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5428 sizeof(struct discard_cmd));
5429 if (!discard_cmd_slab)
5430 goto destroy_discard_entry;
5432 sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5433 sizeof(struct sit_entry_set));
5434 if (!sit_entry_set_slab)
5435 goto destroy_discard_cmd;
5437 inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5438 sizeof(struct inmem_pages));
5439 if (!inmem_entry_slab)
5440 goto destroy_sit_entry_set;
5443 destroy_sit_entry_set:
5444 kmem_cache_destroy(sit_entry_set_slab);
5445 destroy_discard_cmd:
5446 kmem_cache_destroy(discard_cmd_slab);
5447 destroy_discard_entry:
5448 kmem_cache_destroy(discard_entry_slab);
5453 void f2fs_destroy_segment_manager_caches(void)
5455 kmem_cache_destroy(sit_entry_set_slab);
5456 kmem_cache_destroy(discard_cmd_slab);
5457 kmem_cache_destroy(discard_entry_slab);
5458 kmem_cache_destroy(inmem_entry_slab);