1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
10 #include <linux/mpage.h>
11 #include <linux/writeback.h>
12 #include <linux/blkdev.h>
13 #include <linux/f2fs_fs.h>
14 #include <linux/pagevec.h>
15 #include <linux/swap.h>
16 #include <linux/kthread.h>
22 #include <trace/events/f2fs.h>
24 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *f2fs_inode_entry_slab;
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
31 f2fs_build_fault_attr(sbi, 0, 0);
32 set_ckpt_flags(sbi, CP_ERROR_FLAG);
34 f2fs_flush_merged_writes(sbi);
38 * We guarantee no failure on the returned page.
40 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
42 struct address_space *mapping = META_MAPPING(sbi);
45 page = f2fs_grab_cache_page(mapping, index, false);
50 f2fs_wait_on_page_writeback(page, META, true, true);
51 if (!PageUptodate(page))
52 SetPageUptodate(page);
56 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
59 struct address_space *mapping = META_MAPPING(sbi);
61 struct f2fs_io_info fio = {
65 .op_flags = REQ_META | REQ_PRIO,
68 .encrypted_page = NULL,
73 if (unlikely(!is_meta))
74 fio.op_flags &= ~REQ_META;
76 page = f2fs_grab_cache_page(mapping, index, false);
81 if (PageUptodate(page))
86 err = f2fs_submit_page_bio(&fio);
88 f2fs_put_page(page, 1);
92 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
95 if (unlikely(page->mapping != mapping)) {
96 f2fs_put_page(page, 1);
100 if (unlikely(!PageUptodate(page))) {
101 f2fs_put_page(page, 1);
102 return ERR_PTR(-EIO);
108 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
110 return __get_meta_page(sbi, index, true);
113 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
119 page = __get_meta_page(sbi, index, true);
121 if (PTR_ERR(page) == -EIO &&
122 ++count <= DEFAULT_RETRY_IO_COUNT)
124 f2fs_stop_checkpoint(sbi, false);
130 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
132 return __get_meta_page(sbi, index, false);
135 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
138 struct seg_entry *se;
139 unsigned int segno, offset;
142 if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
145 segno = GET_SEGNO(sbi, blkaddr);
146 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
147 se = get_seg_entry(sbi, segno);
149 exist = f2fs_test_bit(offset, se->cur_valid_map);
150 if (!exist && type == DATA_GENERIC_ENHANCE) {
151 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
153 set_sbi_flag(sbi, SBI_NEED_FSCK);
159 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
160 block_t blkaddr, int type)
166 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
170 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
171 blkaddr < SM_I(sbi)->ssa_blkaddr))
175 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
176 blkaddr < __start_cp_addr(sbi)))
180 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
181 blkaddr < MAIN_BLKADDR(sbi)))
185 case DATA_GENERIC_ENHANCE:
186 case DATA_GENERIC_ENHANCE_READ:
187 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
188 blkaddr < MAIN_BLKADDR(sbi))) {
189 f2fs_warn(sbi, "access invalid blkaddr:%u",
191 set_sbi_flag(sbi, SBI_NEED_FSCK);
195 return __is_bitmap_valid(sbi, blkaddr, type);
199 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
200 blkaddr >= MAIN_BLKADDR(sbi)))
211 * Readahead CP/NAT/SIT/SSA/POR pages
213 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
217 block_t blkno = start;
218 struct f2fs_io_info fio = {
222 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
223 .encrypted_page = NULL,
225 .is_por = (type == META_POR),
227 struct blk_plug plug;
230 if (unlikely(type == META_POR))
231 fio.op_flags &= ~REQ_META;
233 blk_start_plug(&plug);
234 for (; nrpages-- > 0; blkno++) {
236 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
241 if (unlikely(blkno >=
242 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
244 /* get nat block addr */
245 fio.new_blkaddr = current_nat_addr(sbi,
246 blkno * NAT_ENTRY_PER_BLOCK);
249 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
251 /* get sit block addr */
252 fio.new_blkaddr = current_sit_addr(sbi,
253 blkno * SIT_ENTRY_PER_BLOCK);
258 fio.new_blkaddr = blkno;
264 page = f2fs_grab_cache_page(META_MAPPING(sbi),
265 fio.new_blkaddr, false);
268 if (PageUptodate(page)) {
269 f2fs_put_page(page, 1);
274 err = f2fs_submit_page_bio(&fio);
275 f2fs_put_page(page, err ? 1 : 0);
278 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
281 blk_finish_plug(&plug);
282 return blkno - start;
285 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
288 bool readahead = false;
290 page = find_get_page(META_MAPPING(sbi), index);
291 if (!page || !PageUptodate(page))
293 f2fs_put_page(page, 0);
296 f2fs_ra_meta_pages(sbi, index, BIO_MAX_VECS, META_POR, true);
299 static int __f2fs_write_meta_page(struct page *page,
300 struct writeback_control *wbc,
301 enum iostat_type io_type)
303 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
305 trace_f2fs_writepage(page, META);
307 if (unlikely(f2fs_cp_error(sbi)))
309 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
311 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
314 f2fs_do_write_meta_page(sbi, page, io_type);
315 dec_page_count(sbi, F2FS_DIRTY_META);
317 if (wbc->for_reclaim)
318 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
322 if (unlikely(f2fs_cp_error(sbi)))
323 f2fs_submit_merged_write(sbi, META);
328 redirty_page_for_writepage(wbc, page);
329 return AOP_WRITEPAGE_ACTIVATE;
332 static int f2fs_write_meta_page(struct page *page,
333 struct writeback_control *wbc)
335 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
338 static int f2fs_write_meta_pages(struct address_space *mapping,
339 struct writeback_control *wbc)
341 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
344 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
347 /* collect a number of dirty meta pages and write together */
348 if (wbc->sync_mode != WB_SYNC_ALL &&
349 get_pages(sbi, F2FS_DIRTY_META) <
350 nr_pages_to_skip(sbi, META))
353 /* if locked failed, cp will flush dirty pages instead */
354 if (!down_write_trylock(&sbi->cp_global_sem))
357 trace_f2fs_writepages(mapping->host, wbc, META);
358 diff = nr_pages_to_write(sbi, META, wbc);
359 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
360 up_write(&sbi->cp_global_sem);
361 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
365 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
366 trace_f2fs_writepages(mapping->host, wbc, META);
370 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
371 long nr_to_write, enum iostat_type io_type)
373 struct address_space *mapping = META_MAPPING(sbi);
374 pgoff_t index = 0, prev = ULONG_MAX;
378 struct writeback_control wbc = {
381 struct blk_plug plug;
385 blk_start_plug(&plug);
387 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
388 PAGECACHE_TAG_DIRTY))) {
391 for (i = 0; i < nr_pages; i++) {
392 struct page *page = pvec.pages[i];
394 if (prev == ULONG_MAX)
395 prev = page->index - 1;
396 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
397 pagevec_release(&pvec);
403 if (unlikely(page->mapping != mapping)) {
408 if (!PageDirty(page)) {
409 /* someone wrote it for us */
410 goto continue_unlock;
413 f2fs_wait_on_page_writeback(page, META, true, true);
415 if (!clear_page_dirty_for_io(page))
416 goto continue_unlock;
418 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
424 if (unlikely(nwritten >= nr_to_write))
427 pagevec_release(&pvec);
432 f2fs_submit_merged_write(sbi, type);
434 blk_finish_plug(&plug);
439 static int f2fs_set_meta_page_dirty(struct page *page)
441 trace_f2fs_set_page_dirty(page, META);
443 if (!PageUptodate(page))
444 SetPageUptodate(page);
445 if (!PageDirty(page)) {
446 __set_page_dirty_nobuffers(page);
447 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
448 set_page_private_reference(page);
454 const struct address_space_operations f2fs_meta_aops = {
455 .writepage = f2fs_write_meta_page,
456 .writepages = f2fs_write_meta_pages,
457 .set_page_dirty = f2fs_set_meta_page_dirty,
458 .invalidatepage = f2fs_invalidate_page,
459 .releasepage = f2fs_release_page,
460 #ifdef CONFIG_MIGRATION
461 .migratepage = f2fs_migrate_page,
465 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
466 unsigned int devidx, int type)
468 struct inode_management *im = &sbi->im[type];
469 struct ino_entry *e = NULL, *new = NULL;
471 if (type == FLUSH_INO) {
473 e = radix_tree_lookup(&im->ino_root, ino);
479 new = f2fs_kmem_cache_alloc(ino_entry_slab,
480 GFP_NOFS, true, NULL);
482 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
484 spin_lock(&im->ino_lock);
485 e = radix_tree_lookup(&im->ino_root, ino);
488 spin_unlock(&im->ino_lock);
492 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
495 memset(e, 0, sizeof(struct ino_entry));
498 list_add_tail(&e->list, &im->ino_list);
499 if (type != ORPHAN_INO)
503 if (type == FLUSH_INO)
504 f2fs_set_bit(devidx, (char *)&e->dirty_device);
506 spin_unlock(&im->ino_lock);
507 radix_tree_preload_end();
510 kmem_cache_free(ino_entry_slab, new);
513 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
515 struct inode_management *im = &sbi->im[type];
518 spin_lock(&im->ino_lock);
519 e = radix_tree_lookup(&im->ino_root, ino);
522 radix_tree_delete(&im->ino_root, ino);
524 spin_unlock(&im->ino_lock);
525 kmem_cache_free(ino_entry_slab, e);
528 spin_unlock(&im->ino_lock);
531 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
533 /* add new dirty ino entry into list */
534 __add_ino_entry(sbi, ino, 0, type);
537 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
539 /* remove dirty ino entry from list */
540 __remove_ino_entry(sbi, ino, type);
543 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
544 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
546 struct inode_management *im = &sbi->im[mode];
549 spin_lock(&im->ino_lock);
550 e = radix_tree_lookup(&im->ino_root, ino);
551 spin_unlock(&im->ino_lock);
552 return e ? true : false;
555 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
557 struct ino_entry *e, *tmp;
560 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
561 struct inode_management *im = &sbi->im[i];
563 spin_lock(&im->ino_lock);
564 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
566 radix_tree_delete(&im->ino_root, e->ino);
567 kmem_cache_free(ino_entry_slab, e);
570 spin_unlock(&im->ino_lock);
574 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
575 unsigned int devidx, int type)
577 __add_ino_entry(sbi, ino, devidx, type);
580 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
581 unsigned int devidx, int type)
583 struct inode_management *im = &sbi->im[type];
585 bool is_dirty = false;
587 spin_lock(&im->ino_lock);
588 e = radix_tree_lookup(&im->ino_root, ino);
589 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
591 spin_unlock(&im->ino_lock);
595 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
597 struct inode_management *im = &sbi->im[ORPHAN_INO];
600 spin_lock(&im->ino_lock);
602 if (time_to_inject(sbi, FAULT_ORPHAN)) {
603 spin_unlock(&im->ino_lock);
604 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
608 if (unlikely(im->ino_num >= sbi->max_orphans))
612 spin_unlock(&im->ino_lock);
617 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
619 struct inode_management *im = &sbi->im[ORPHAN_INO];
621 spin_lock(&im->ino_lock);
622 f2fs_bug_on(sbi, im->ino_num == 0);
624 spin_unlock(&im->ino_lock);
627 void f2fs_add_orphan_inode(struct inode *inode)
629 /* add new orphan ino entry into list */
630 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
631 f2fs_update_inode_page(inode);
634 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
636 /* remove orphan entry from orphan list */
637 __remove_ino_entry(sbi, ino, ORPHAN_INO);
640 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
646 inode = f2fs_iget_retry(sbi->sb, ino);
649 * there should be a bug that we can't find the entry
652 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
653 return PTR_ERR(inode);
656 err = dquot_initialize(inode);
664 /* truncate all the data during iput */
667 err = f2fs_get_node_info(sbi, ino, &ni);
671 /* ENOMEM was fully retried in f2fs_evict_inode. */
672 if (ni.blk_addr != NULL_ADDR) {
679 set_sbi_flag(sbi, SBI_NEED_FSCK);
680 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
685 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
687 block_t start_blk, orphan_blocks, i, j;
688 unsigned int s_flags = sbi->sb->s_flags;
694 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
697 if (bdev_read_only(sbi->sb->s_bdev)) {
698 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
702 if (s_flags & SB_RDONLY) {
703 f2fs_info(sbi, "orphan cleanup on readonly fs");
704 sbi->sb->s_flags &= ~SB_RDONLY;
708 /* Needed for iput() to work correctly and not trash data */
709 sbi->sb->s_flags |= SB_ACTIVE;
712 * Turn on quotas which were not enabled for read-only mounts if
713 * filesystem has quota feature, so that they are updated correctly.
715 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
718 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
719 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
721 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
723 for (i = 0; i < orphan_blocks; i++) {
725 struct f2fs_orphan_block *orphan_blk;
727 page = f2fs_get_meta_page(sbi, start_blk + i);
733 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
734 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
735 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
737 err = recover_orphan_inode(sbi, ino);
739 f2fs_put_page(page, 1);
743 f2fs_put_page(page, 1);
745 /* clear Orphan Flag */
746 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
748 set_sbi_flag(sbi, SBI_IS_RECOVERED);
751 /* Turn quotas off */
753 f2fs_quota_off_umount(sbi->sb);
755 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
760 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
762 struct list_head *head;
763 struct f2fs_orphan_block *orphan_blk = NULL;
764 unsigned int nentries = 0;
765 unsigned short index = 1;
766 unsigned short orphan_blocks;
767 struct page *page = NULL;
768 struct ino_entry *orphan = NULL;
769 struct inode_management *im = &sbi->im[ORPHAN_INO];
771 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
774 * we don't need to do spin_lock(&im->ino_lock) here, since all the
775 * orphan inode operations are covered under f2fs_lock_op().
776 * And, spin_lock should be avoided due to page operations below.
778 head = &im->ino_list;
780 /* loop for each orphan inode entry and write them in Jornal block */
781 list_for_each_entry(orphan, head, list) {
783 page = f2fs_grab_meta_page(sbi, start_blk++);
785 (struct f2fs_orphan_block *)page_address(page);
786 memset(orphan_blk, 0, sizeof(*orphan_blk));
789 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
791 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
793 * an orphan block is full of 1020 entries,
794 * then we need to flush current orphan blocks
795 * and bring another one in memory
797 orphan_blk->blk_addr = cpu_to_le16(index);
798 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
799 orphan_blk->entry_count = cpu_to_le32(nentries);
800 set_page_dirty(page);
801 f2fs_put_page(page, 1);
809 orphan_blk->blk_addr = cpu_to_le16(index);
810 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
811 orphan_blk->entry_count = cpu_to_le32(nentries);
812 set_page_dirty(page);
813 f2fs_put_page(page, 1);
817 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
818 struct f2fs_checkpoint *ckpt)
820 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
823 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
824 if (chksum_ofs < CP_CHKSUM_OFFSET) {
825 chksum_ofs += sizeof(chksum);
826 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
827 F2FS_BLKSIZE - chksum_ofs);
832 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
833 struct f2fs_checkpoint **cp_block, struct page **cp_page,
834 unsigned long long *version)
836 size_t crc_offset = 0;
839 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
840 if (IS_ERR(*cp_page))
841 return PTR_ERR(*cp_page);
843 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
845 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
846 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
847 crc_offset > CP_CHKSUM_OFFSET) {
848 f2fs_put_page(*cp_page, 1);
849 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
853 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
854 if (crc != cur_cp_crc(*cp_block)) {
855 f2fs_put_page(*cp_page, 1);
856 f2fs_warn(sbi, "invalid crc value");
860 *version = cur_cp_version(*cp_block);
864 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
865 block_t cp_addr, unsigned long long *version)
867 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
868 struct f2fs_checkpoint *cp_block = NULL;
869 unsigned long long cur_version = 0, pre_version = 0;
872 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
873 &cp_page_1, version);
877 if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
878 sbi->blocks_per_seg) {
879 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
880 le32_to_cpu(cp_block->cp_pack_total_block_count));
883 pre_version = *version;
885 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
886 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
887 &cp_page_2, version);
890 cur_version = *version;
892 if (cur_version == pre_version) {
893 *version = cur_version;
894 f2fs_put_page(cp_page_2, 1);
897 f2fs_put_page(cp_page_2, 1);
899 f2fs_put_page(cp_page_1, 1);
903 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
905 struct f2fs_checkpoint *cp_block;
906 struct f2fs_super_block *fsb = sbi->raw_super;
907 struct page *cp1, *cp2, *cur_page;
908 unsigned long blk_size = sbi->blocksize;
909 unsigned long long cp1_version = 0, cp2_version = 0;
910 unsigned long long cp_start_blk_no;
911 unsigned int cp_blks = 1 + __cp_payload(sbi);
916 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
921 * Finding out valid cp block involves read both
922 * sets( cp pack 1 and cp pack 2)
924 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
925 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
927 /* The second checkpoint pack should start at the next segment */
928 cp_start_blk_no += ((unsigned long long)1) <<
929 le32_to_cpu(fsb->log_blocks_per_seg);
930 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
933 if (ver_after(cp2_version, cp1_version))
946 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
947 memcpy(sbi->ckpt, cp_block, blk_size);
950 sbi->cur_cp_pack = 1;
952 sbi->cur_cp_pack = 2;
954 /* Sanity checking of checkpoint */
955 if (f2fs_sanity_check_ckpt(sbi)) {
957 goto free_fail_no_cp;
963 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
965 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
967 for (i = 1; i < cp_blks; i++) {
968 void *sit_bitmap_ptr;
969 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
971 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
972 if (IS_ERR(cur_page)) {
973 err = PTR_ERR(cur_page);
974 goto free_fail_no_cp;
976 sit_bitmap_ptr = page_address(cur_page);
977 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
978 f2fs_put_page(cur_page, 1);
981 f2fs_put_page(cp1, 1);
982 f2fs_put_page(cp2, 1);
986 f2fs_put_page(cp1, 1);
987 f2fs_put_page(cp2, 1);
993 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
995 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
996 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
998 if (is_inode_flag_set(inode, flag))
1001 set_inode_flag(inode, flag);
1002 if (!f2fs_is_volatile_file(inode))
1003 list_add_tail(&F2FS_I(inode)->dirty_list,
1004 &sbi->inode_list[type]);
1005 stat_inc_dirty_inode(sbi, type);
1008 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1010 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1012 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1015 list_del_init(&F2FS_I(inode)->dirty_list);
1016 clear_inode_flag(inode, flag);
1017 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1020 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1022 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1023 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1025 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1026 !S_ISLNK(inode->i_mode))
1029 spin_lock(&sbi->inode_lock[type]);
1030 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1031 __add_dirty_inode(inode, type);
1032 inode_inc_dirty_pages(inode);
1033 spin_unlock(&sbi->inode_lock[type]);
1035 set_page_private_reference(page);
1038 void f2fs_remove_dirty_inode(struct inode *inode)
1040 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1041 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1043 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1044 !S_ISLNK(inode->i_mode))
1047 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1050 spin_lock(&sbi->inode_lock[type]);
1051 __remove_dirty_inode(inode, type);
1052 spin_unlock(&sbi->inode_lock[type]);
1055 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1057 struct list_head *head;
1058 struct inode *inode;
1059 struct f2fs_inode_info *fi;
1060 bool is_dir = (type == DIR_INODE);
1061 unsigned long ino = 0;
1063 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1064 get_pages(sbi, is_dir ?
1065 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1067 if (unlikely(f2fs_cp_error(sbi))) {
1068 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1069 get_pages(sbi, is_dir ?
1070 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1074 spin_lock(&sbi->inode_lock[type]);
1076 head = &sbi->inode_list[type];
1077 if (list_empty(head)) {
1078 spin_unlock(&sbi->inode_lock[type]);
1079 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1080 get_pages(sbi, is_dir ?
1081 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1084 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1085 inode = igrab(&fi->vfs_inode);
1086 spin_unlock(&sbi->inode_lock[type]);
1088 unsigned long cur_ino = inode->i_ino;
1090 F2FS_I(inode)->cp_task = current;
1092 filemap_fdatawrite(inode->i_mapping);
1094 F2FS_I(inode)->cp_task = NULL;
1097 /* We need to give cpu to another writers. */
1104 * We should submit bio, since it exists several
1105 * wribacking dentry pages in the freeing inode.
1107 f2fs_submit_merged_write(sbi, DATA);
1113 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1115 struct list_head *head = &sbi->inode_list[DIRTY_META];
1116 struct inode *inode;
1117 struct f2fs_inode_info *fi;
1118 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1121 if (unlikely(f2fs_cp_error(sbi)))
1124 spin_lock(&sbi->inode_lock[DIRTY_META]);
1125 if (list_empty(head)) {
1126 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1129 fi = list_first_entry(head, struct f2fs_inode_info,
1131 inode = igrab(&fi->vfs_inode);
1132 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1134 sync_inode_metadata(inode, 0);
1136 /* it's on eviction */
1137 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1138 f2fs_update_inode_page(inode);
1145 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1147 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1148 struct f2fs_nm_info *nm_i = NM_I(sbi);
1149 nid_t last_nid = nm_i->next_scan_nid;
1151 next_free_nid(sbi, &last_nid);
1152 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1153 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1154 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1155 ckpt->next_free_nid = cpu_to_le32(last_nid);
1158 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1162 if (!is_journalled_quota(sbi))
1165 down_write(&sbi->quota_sem);
1166 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1168 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1170 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1171 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1173 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1176 up_write(&sbi->quota_sem);
1181 * Freeze all the FS-operations for checkpoint.
1183 static int block_operations(struct f2fs_sb_info *sbi)
1185 struct writeback_control wbc = {
1186 .sync_mode = WB_SYNC_ALL,
1187 .nr_to_write = LONG_MAX,
1190 int err = 0, cnt = 0;
1193 * Let's flush inline_data in dirty node pages.
1195 f2fs_flush_inline_data(sbi);
1199 if (__need_flush_quota(sbi)) {
1202 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1203 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1204 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1205 goto retry_flush_dents;
1207 f2fs_unlock_all(sbi);
1209 /* only failed during mount/umount/freeze/quotactl */
1210 locked = down_read_trylock(&sbi->sb->s_umount);
1211 f2fs_quota_sync(sbi->sb, -1);
1213 up_read(&sbi->sb->s_umount);
1215 goto retry_flush_quotas;
1219 /* write all the dirty dentry pages */
1220 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1221 f2fs_unlock_all(sbi);
1222 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1226 goto retry_flush_quotas;
1230 * POR: we should ensure that there are no dirty node pages
1231 * until finishing nat/sit flush. inode->i_blocks can be updated.
1233 down_write(&sbi->node_change);
1235 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1236 up_write(&sbi->node_change);
1237 f2fs_unlock_all(sbi);
1238 err = f2fs_sync_inode_meta(sbi);
1242 goto retry_flush_quotas;
1246 down_write(&sbi->node_write);
1248 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1249 up_write(&sbi->node_write);
1250 atomic_inc(&sbi->wb_sync_req[NODE]);
1251 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1252 atomic_dec(&sbi->wb_sync_req[NODE]);
1254 up_write(&sbi->node_change);
1255 f2fs_unlock_all(sbi);
1259 goto retry_flush_nodes;
1263 * sbi->node_change is used only for AIO write_begin path which produces
1264 * dirty node blocks and some checkpoint values by block allocation.
1266 __prepare_cp_block(sbi);
1267 up_write(&sbi->node_change);
1271 static void unblock_operations(struct f2fs_sb_info *sbi)
1273 up_write(&sbi->node_write);
1274 f2fs_unlock_all(sbi);
1277 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1282 if (!get_pages(sbi, type))
1285 if (unlikely(f2fs_cp_error(sbi)))
1288 if (type == F2FS_DIRTY_META)
1289 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1291 else if (type == F2FS_WB_CP_DATA)
1292 f2fs_submit_merged_write(sbi, DATA);
1294 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1295 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1297 finish_wait(&sbi->cp_wait, &wait);
1300 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1302 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1303 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1304 unsigned long flags;
1306 if (cpc->reason & CP_UMOUNT) {
1307 if (le32_to_cpu(ckpt->cp_pack_total_block_count) >
1308 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks) {
1309 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1310 f2fs_notice(sbi, "Disable nat_bits due to no space");
1311 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1312 f2fs_nat_bitmap_enabled(sbi)) {
1313 f2fs_enable_nat_bits(sbi);
1314 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1315 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1319 spin_lock_irqsave(&sbi->cp_lock, flags);
1321 if (cpc->reason & CP_TRIMMED)
1322 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1324 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1326 if (cpc->reason & CP_UMOUNT)
1327 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1329 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1331 if (cpc->reason & CP_FASTBOOT)
1332 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1334 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1337 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1339 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1341 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1342 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1344 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1345 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1347 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1349 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1350 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1352 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1354 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1355 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1357 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1359 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1360 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1362 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1364 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1365 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1367 /* set this flag to activate crc|cp_ver for recovery */
1368 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1369 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1371 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1374 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1375 void *src, block_t blk_addr)
1377 struct writeback_control wbc = {
1382 * pagevec_lookup_tag and lock_page again will take
1383 * some extra time. Therefore, f2fs_update_meta_pages and
1384 * f2fs_sync_meta_pages are combined in this function.
1386 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1389 f2fs_wait_on_page_writeback(page, META, true, true);
1391 memcpy(page_address(page), src, PAGE_SIZE);
1393 set_page_dirty(page);
1394 if (unlikely(!clear_page_dirty_for_io(page)))
1395 f2fs_bug_on(sbi, 1);
1397 /* writeout cp pack 2 page */
1398 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1399 if (unlikely(err && f2fs_cp_error(sbi))) {
1400 f2fs_put_page(page, 1);
1404 f2fs_bug_on(sbi, err);
1405 f2fs_put_page(page, 0);
1407 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1408 f2fs_submit_merged_write(sbi, META_FLUSH);
1411 static inline u64 get_sectors_written(struct block_device *bdev)
1413 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1416 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1418 if (f2fs_is_multi_device(sbi)) {
1422 for (i = 0; i < sbi->s_ndevs; i++)
1423 sectors += get_sectors_written(FDEV(i).bdev);
1428 return get_sectors_written(sbi->sb->s_bdev);
1431 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1433 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1434 struct f2fs_nm_info *nm_i = NM_I(sbi);
1435 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1437 unsigned int data_sum_blocks, orphan_blocks;
1440 int cp_payload_blks = __cp_payload(sbi);
1441 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1445 /* Flush all the NAT/SIT pages */
1446 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1448 /* start to update checkpoint, cp ver is already updated previously */
1449 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1450 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1451 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1452 ckpt->cur_node_segno[i] =
1453 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1454 ckpt->cur_node_blkoff[i] =
1455 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1456 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1457 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1459 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1460 ckpt->cur_data_segno[i] =
1461 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1462 ckpt->cur_data_blkoff[i] =
1463 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1464 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1465 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1468 /* 2 cp + n data seg summary + orphan inode blocks */
1469 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1470 spin_lock_irqsave(&sbi->cp_lock, flags);
1471 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1472 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1474 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1475 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1477 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1478 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1481 if (__remain_node_summaries(cpc->reason))
1482 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1483 cp_payload_blks + data_sum_blocks +
1484 orphan_blocks + NR_CURSEG_NODE_TYPE);
1486 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1487 cp_payload_blks + data_sum_blocks +
1490 /* update ckpt flag for checkpoint */
1491 update_ckpt_flags(sbi, cpc);
1493 /* update SIT/NAT bitmap */
1494 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1495 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1497 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1498 *((__le32 *)((unsigned char *)ckpt +
1499 le32_to_cpu(ckpt->checksum_offset)))
1500 = cpu_to_le32(crc32);
1502 start_blk = __start_cp_next_addr(sbi);
1504 /* write nat bits */
1505 if ((cpc->reason & CP_UMOUNT) &&
1506 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1507 __u64 cp_ver = cur_cp_version(ckpt);
1510 cp_ver |= ((__u64)crc32 << 32);
1511 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1513 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1514 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1515 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1516 (i << F2FS_BLKSIZE_BITS), blk + i);
1519 /* write out checkpoint buffer at block 0 */
1520 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1522 for (i = 1; i < 1 + cp_payload_blks; i++)
1523 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1527 write_orphan_inodes(sbi, start_blk);
1528 start_blk += orphan_blocks;
1531 f2fs_write_data_summaries(sbi, start_blk);
1532 start_blk += data_sum_blocks;
1534 /* Record write statistics in the hot node summary */
1535 kbytes_written = sbi->kbytes_written;
1536 kbytes_written += (f2fs_get_sectors_written(sbi) -
1537 sbi->sectors_written_start) >> 1;
1538 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1540 if (__remain_node_summaries(cpc->reason)) {
1541 f2fs_write_node_summaries(sbi, start_blk);
1542 start_blk += NR_CURSEG_NODE_TYPE;
1545 /* update user_block_counts */
1546 sbi->last_valid_block_count = sbi->total_valid_block_count;
1547 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1549 /* Here, we have one bio having CP pack except cp pack 2 page */
1550 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1551 /* Wait for all dirty meta pages to be submitted for IO */
1552 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1554 /* wait for previous submitted meta pages writeback */
1555 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1557 /* flush all device cache */
1558 err = f2fs_flush_device_cache(sbi);
1562 /* barrier and flush checkpoint cp pack 2 page if it can */
1563 commit_checkpoint(sbi, ckpt, start_blk);
1564 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1567 * invalidate intermediate page cache borrowed from meta inode which are
1568 * used for migration of encrypted, verity or compressed inode's blocks.
1570 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1571 f2fs_sb_has_compression(sbi))
1572 invalidate_mapping_pages(META_MAPPING(sbi),
1573 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1575 f2fs_release_ino_entry(sbi, false);
1577 f2fs_reset_fsync_node_info(sbi);
1579 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1580 clear_sbi_flag(sbi, SBI_NEED_CP);
1581 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1583 spin_lock(&sbi->stat_lock);
1584 sbi->unusable_block_count = 0;
1585 spin_unlock(&sbi->stat_lock);
1587 __set_cp_next_pack(sbi);
1590 * redirty superblock if metadata like node page or inode cache is
1591 * updated during writing checkpoint.
1593 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1594 get_pages(sbi, F2FS_DIRTY_IMETA))
1595 set_sbi_flag(sbi, SBI_IS_DIRTY);
1597 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1599 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1602 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1604 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1605 unsigned long long ckpt_ver;
1608 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1611 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1612 if (cpc->reason != CP_PAUSE)
1614 f2fs_warn(sbi, "Start checkpoint disabled!");
1616 if (cpc->reason != CP_RESIZE)
1617 down_write(&sbi->cp_global_sem);
1619 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1620 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1621 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1623 if (unlikely(f2fs_cp_error(sbi))) {
1628 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1630 err = block_operations(sbi);
1634 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1636 f2fs_flush_merged_writes(sbi);
1638 /* this is the case of multiple fstrims without any changes */
1639 if (cpc->reason & CP_DISCARD) {
1640 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1641 unblock_operations(sbi);
1645 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1646 SIT_I(sbi)->dirty_sentries == 0 &&
1647 prefree_segments(sbi) == 0) {
1648 f2fs_flush_sit_entries(sbi, cpc);
1649 f2fs_clear_prefree_segments(sbi, cpc);
1650 unblock_operations(sbi);
1656 * update checkpoint pack index
1657 * Increase the version number so that
1658 * SIT entries and seg summaries are written at correct place
1660 ckpt_ver = cur_cp_version(ckpt);
1661 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1663 /* write cached NAT/SIT entries to NAT/SIT area */
1664 err = f2fs_flush_nat_entries(sbi, cpc);
1666 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1667 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1671 f2fs_flush_sit_entries(sbi, cpc);
1673 /* save inmem log status */
1674 f2fs_save_inmem_curseg(sbi);
1676 err = do_checkpoint(sbi, cpc);
1678 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1679 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1680 f2fs_release_discard_addrs(sbi);
1682 f2fs_clear_prefree_segments(sbi, cpc);
1685 f2fs_restore_inmem_curseg(sbi);
1687 unblock_operations(sbi);
1688 stat_inc_cp_count(sbi->stat_info);
1690 if (cpc->reason & CP_RECOVERY)
1691 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1693 /* update CP_TIME to trigger checkpoint periodically */
1694 f2fs_update_time(sbi, CP_TIME);
1695 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1697 if (cpc->reason != CP_RESIZE)
1698 up_write(&sbi->cp_global_sem);
1702 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1706 for (i = 0; i < MAX_INO_ENTRY; i++) {
1707 struct inode_management *im = &sbi->im[i];
1709 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1710 spin_lock_init(&im->ino_lock);
1711 INIT_LIST_HEAD(&im->ino_list);
1715 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1716 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1717 F2FS_ORPHANS_PER_BLOCK;
1720 int __init f2fs_create_checkpoint_caches(void)
1722 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1723 sizeof(struct ino_entry));
1724 if (!ino_entry_slab)
1726 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1727 sizeof(struct inode_entry));
1728 if (!f2fs_inode_entry_slab) {
1729 kmem_cache_destroy(ino_entry_slab);
1735 void f2fs_destroy_checkpoint_caches(void)
1737 kmem_cache_destroy(ino_entry_slab);
1738 kmem_cache_destroy(f2fs_inode_entry_slab);
1741 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1743 struct cp_control cpc = { .reason = CP_SYNC, };
1746 down_write(&sbi->gc_lock);
1747 err = f2fs_write_checkpoint(sbi, &cpc);
1748 up_write(&sbi->gc_lock);
1753 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1755 struct ckpt_req_control *cprc = &sbi->cprc_info;
1756 struct ckpt_req *req, *next;
1757 struct llist_node *dispatch_list;
1758 u64 sum_diff = 0, diff, count = 0;
1761 dispatch_list = llist_del_all(&cprc->issue_list);
1764 dispatch_list = llist_reverse_order(dispatch_list);
1766 ret = __write_checkpoint_sync(sbi);
1767 atomic_inc(&cprc->issued_ckpt);
1769 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1770 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1772 complete(&req->wait);
1777 atomic_sub(count, &cprc->queued_ckpt);
1778 atomic_add(count, &cprc->total_ckpt);
1780 spin_lock(&cprc->stat_lock);
1781 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1782 if (cprc->peak_time < cprc->cur_time)
1783 cprc->peak_time = cprc->cur_time;
1784 spin_unlock(&cprc->stat_lock);
1787 static int issue_checkpoint_thread(void *data)
1789 struct f2fs_sb_info *sbi = data;
1790 struct ckpt_req_control *cprc = &sbi->cprc_info;
1791 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1793 if (kthread_should_stop())
1796 if (!llist_empty(&cprc->issue_list))
1797 __checkpoint_and_complete_reqs(sbi);
1799 wait_event_interruptible(*q,
1800 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1804 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1805 struct ckpt_req *wait_req)
1807 struct ckpt_req_control *cprc = &sbi->cprc_info;
1809 if (!llist_empty(&cprc->issue_list)) {
1810 __checkpoint_and_complete_reqs(sbi);
1812 /* already dispatched by issue_checkpoint_thread */
1814 wait_for_completion(&wait_req->wait);
1818 static void init_ckpt_req(struct ckpt_req *req)
1820 memset(req, 0, sizeof(struct ckpt_req));
1822 init_completion(&req->wait);
1823 req->queue_time = ktime_get();
1826 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1828 struct ckpt_req_control *cprc = &sbi->cprc_info;
1829 struct ckpt_req req;
1830 struct cp_control cpc;
1832 cpc.reason = __get_cp_reason(sbi);
1833 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1836 down_write(&sbi->gc_lock);
1837 ret = f2fs_write_checkpoint(sbi, &cpc);
1838 up_write(&sbi->gc_lock);
1843 if (!cprc->f2fs_issue_ckpt)
1844 return __write_checkpoint_sync(sbi);
1846 init_ckpt_req(&req);
1848 llist_add(&req.llnode, &cprc->issue_list);
1849 atomic_inc(&cprc->queued_ckpt);
1852 * update issue_list before we wake up issue_checkpoint thread,
1853 * this smp_mb() pairs with another barrier in ___wait_event(),
1854 * see more details in comments of waitqueue_active().
1858 if (waitqueue_active(&cprc->ckpt_wait_queue))
1859 wake_up(&cprc->ckpt_wait_queue);
1861 if (cprc->f2fs_issue_ckpt)
1862 wait_for_completion(&req.wait);
1864 flush_remained_ckpt_reqs(sbi, &req);
1869 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1871 dev_t dev = sbi->sb->s_bdev->bd_dev;
1872 struct ckpt_req_control *cprc = &sbi->cprc_info;
1874 if (cprc->f2fs_issue_ckpt)
1877 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1878 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1879 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1880 cprc->f2fs_issue_ckpt = NULL;
1884 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1889 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1891 struct ckpt_req_control *cprc = &sbi->cprc_info;
1893 if (cprc->f2fs_issue_ckpt) {
1894 struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt;
1896 cprc->f2fs_issue_ckpt = NULL;
1897 kthread_stop(ckpt_task);
1899 flush_remained_ckpt_reqs(sbi, NULL);
1903 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1905 struct ckpt_req_control *cprc = &sbi->cprc_info;
1907 atomic_set(&cprc->issued_ckpt, 0);
1908 atomic_set(&cprc->total_ckpt, 0);
1909 atomic_set(&cprc->queued_ckpt, 0);
1910 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1911 init_waitqueue_head(&cprc->ckpt_wait_queue);
1912 init_llist_head(&cprc->issue_list);
1913 spin_lock_init(&cprc->stat_lock);