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 if (page->index == sbi->metapage_eio_ofs &&
102 sbi->metapage_eio_cnt++ == MAX_RETRY_META_PAGE_EIO) {
103 set_ckpt_flags(sbi, CP_ERROR_FLAG);
105 sbi->metapage_eio_ofs = page->index;
106 sbi->metapage_eio_cnt = 0;
108 f2fs_put_page(page, 1);
109 return ERR_PTR(-EIO);
115 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
117 return __get_meta_page(sbi, index, true);
120 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
126 page = __get_meta_page(sbi, index, true);
128 if (PTR_ERR(page) == -EIO &&
129 ++count <= DEFAULT_RETRY_IO_COUNT)
131 f2fs_stop_checkpoint(sbi, false);
137 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
139 return __get_meta_page(sbi, index, false);
142 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
145 struct seg_entry *se;
146 unsigned int segno, offset;
149 if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
152 segno = GET_SEGNO(sbi, blkaddr);
153 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
154 se = get_seg_entry(sbi, segno);
156 exist = f2fs_test_bit(offset, se->cur_valid_map);
157 if (!exist && type == DATA_GENERIC_ENHANCE) {
158 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
160 set_sbi_flag(sbi, SBI_NEED_FSCK);
166 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
167 block_t blkaddr, int type)
173 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
177 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
178 blkaddr < SM_I(sbi)->ssa_blkaddr))
182 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
183 blkaddr < __start_cp_addr(sbi)))
187 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
188 blkaddr < MAIN_BLKADDR(sbi)))
192 case DATA_GENERIC_ENHANCE:
193 case DATA_GENERIC_ENHANCE_READ:
194 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
195 blkaddr < MAIN_BLKADDR(sbi))) {
196 f2fs_warn(sbi, "access invalid blkaddr:%u",
198 set_sbi_flag(sbi, SBI_NEED_FSCK);
202 return __is_bitmap_valid(sbi, blkaddr, type);
206 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
207 blkaddr >= MAIN_BLKADDR(sbi)))
218 * Readahead CP/NAT/SIT/SSA/POR pages
220 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
224 block_t blkno = start;
225 struct f2fs_io_info fio = {
229 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
230 .encrypted_page = NULL,
232 .is_por = (type == META_POR),
234 struct blk_plug plug;
237 if (unlikely(type == META_POR))
238 fio.op_flags &= ~REQ_META;
240 blk_start_plug(&plug);
241 for (; nrpages-- > 0; blkno++) {
243 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
248 if (unlikely(blkno >=
249 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
251 /* get nat block addr */
252 fio.new_blkaddr = current_nat_addr(sbi,
253 blkno * NAT_ENTRY_PER_BLOCK);
256 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
258 /* get sit block addr */
259 fio.new_blkaddr = current_sit_addr(sbi,
260 blkno * SIT_ENTRY_PER_BLOCK);
265 fio.new_blkaddr = blkno;
271 page = f2fs_grab_cache_page(META_MAPPING(sbi),
272 fio.new_blkaddr, false);
275 if (PageUptodate(page)) {
276 f2fs_put_page(page, 1);
281 err = f2fs_submit_page_bio(&fio);
282 f2fs_put_page(page, err ? 1 : 0);
285 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
288 blk_finish_plug(&plug);
289 return blkno - start;
292 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
293 unsigned int ra_blocks)
296 bool readahead = false;
298 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
301 page = find_get_page(META_MAPPING(sbi), index);
302 if (!page || !PageUptodate(page))
304 f2fs_put_page(page, 0);
307 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
310 static int __f2fs_write_meta_page(struct page *page,
311 struct writeback_control *wbc,
312 enum iostat_type io_type)
314 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
316 trace_f2fs_writepage(page, META);
318 if (unlikely(f2fs_cp_error(sbi)))
320 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
322 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
325 f2fs_do_write_meta_page(sbi, page, io_type);
326 dec_page_count(sbi, F2FS_DIRTY_META);
328 if (wbc->for_reclaim)
329 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
333 if (unlikely(f2fs_cp_error(sbi)))
334 f2fs_submit_merged_write(sbi, META);
339 redirty_page_for_writepage(wbc, page);
340 return AOP_WRITEPAGE_ACTIVATE;
343 static int f2fs_write_meta_page(struct page *page,
344 struct writeback_control *wbc)
346 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
349 static int f2fs_write_meta_pages(struct address_space *mapping,
350 struct writeback_control *wbc)
352 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
355 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
358 /* collect a number of dirty meta pages and write together */
359 if (wbc->sync_mode != WB_SYNC_ALL &&
360 get_pages(sbi, F2FS_DIRTY_META) <
361 nr_pages_to_skip(sbi, META))
364 /* if locked failed, cp will flush dirty pages instead */
365 if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
368 trace_f2fs_writepages(mapping->host, wbc, META);
369 diff = nr_pages_to_write(sbi, META, wbc);
370 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
371 f2fs_up_write(&sbi->cp_global_sem);
372 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
376 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
377 trace_f2fs_writepages(mapping->host, wbc, META);
381 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
382 long nr_to_write, enum iostat_type io_type)
384 struct address_space *mapping = META_MAPPING(sbi);
385 pgoff_t index = 0, prev = ULONG_MAX;
389 struct writeback_control wbc = {
392 struct blk_plug plug;
396 blk_start_plug(&plug);
398 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
399 PAGECACHE_TAG_DIRTY))) {
402 for (i = 0; i < nr_pages; i++) {
403 struct page *page = pvec.pages[i];
405 if (prev == ULONG_MAX)
406 prev = page->index - 1;
407 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
408 pagevec_release(&pvec);
414 if (unlikely(page->mapping != mapping)) {
419 if (!PageDirty(page)) {
420 /* someone wrote it for us */
421 goto continue_unlock;
424 f2fs_wait_on_page_writeback(page, META, true, true);
426 if (!clear_page_dirty_for_io(page))
427 goto continue_unlock;
429 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
435 if (unlikely(nwritten >= nr_to_write))
438 pagevec_release(&pvec);
443 f2fs_submit_merged_write(sbi, type);
445 blk_finish_plug(&plug);
450 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
453 trace_f2fs_set_page_dirty(&folio->page, META);
455 if (!folio_test_uptodate(folio))
456 folio_mark_uptodate(folio);
457 if (!folio_test_dirty(folio)) {
458 filemap_dirty_folio(mapping, folio);
459 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
460 set_page_private_reference(&folio->page);
466 const struct address_space_operations f2fs_meta_aops = {
467 .writepage = f2fs_write_meta_page,
468 .writepages = f2fs_write_meta_pages,
469 .dirty_folio = f2fs_dirty_meta_folio,
470 .invalidate_folio = f2fs_invalidate_folio,
471 .releasepage = f2fs_release_page,
472 #ifdef CONFIG_MIGRATION
473 .migratepage = f2fs_migrate_page,
477 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
478 unsigned int devidx, int type)
480 struct inode_management *im = &sbi->im[type];
481 struct ino_entry *e = NULL, *new = NULL;
483 if (type == FLUSH_INO) {
485 e = radix_tree_lookup(&im->ino_root, ino);
491 new = f2fs_kmem_cache_alloc(ino_entry_slab,
492 GFP_NOFS, true, NULL);
494 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
496 spin_lock(&im->ino_lock);
497 e = radix_tree_lookup(&im->ino_root, ino);
500 spin_unlock(&im->ino_lock);
504 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
507 memset(e, 0, sizeof(struct ino_entry));
510 list_add_tail(&e->list, &im->ino_list);
511 if (type != ORPHAN_INO)
515 if (type == FLUSH_INO)
516 f2fs_set_bit(devidx, (char *)&e->dirty_device);
518 spin_unlock(&im->ino_lock);
519 radix_tree_preload_end();
522 kmem_cache_free(ino_entry_slab, new);
525 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
527 struct inode_management *im = &sbi->im[type];
530 spin_lock(&im->ino_lock);
531 e = radix_tree_lookup(&im->ino_root, ino);
534 radix_tree_delete(&im->ino_root, ino);
536 spin_unlock(&im->ino_lock);
537 kmem_cache_free(ino_entry_slab, e);
540 spin_unlock(&im->ino_lock);
543 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
545 /* add new dirty ino entry into list */
546 __add_ino_entry(sbi, ino, 0, type);
549 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
551 /* remove dirty ino entry from list */
552 __remove_ino_entry(sbi, ino, type);
555 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
556 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
558 struct inode_management *im = &sbi->im[mode];
561 spin_lock(&im->ino_lock);
562 e = radix_tree_lookup(&im->ino_root, ino);
563 spin_unlock(&im->ino_lock);
564 return e ? true : false;
567 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
569 struct ino_entry *e, *tmp;
572 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
573 struct inode_management *im = &sbi->im[i];
575 spin_lock(&im->ino_lock);
576 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
578 radix_tree_delete(&im->ino_root, e->ino);
579 kmem_cache_free(ino_entry_slab, e);
582 spin_unlock(&im->ino_lock);
586 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
587 unsigned int devidx, int type)
589 __add_ino_entry(sbi, ino, devidx, type);
592 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
593 unsigned int devidx, int type)
595 struct inode_management *im = &sbi->im[type];
597 bool is_dirty = false;
599 spin_lock(&im->ino_lock);
600 e = radix_tree_lookup(&im->ino_root, ino);
601 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
603 spin_unlock(&im->ino_lock);
607 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
609 struct inode_management *im = &sbi->im[ORPHAN_INO];
612 spin_lock(&im->ino_lock);
614 if (time_to_inject(sbi, FAULT_ORPHAN)) {
615 spin_unlock(&im->ino_lock);
616 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
620 if (unlikely(im->ino_num >= sbi->max_orphans))
624 spin_unlock(&im->ino_lock);
629 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
631 struct inode_management *im = &sbi->im[ORPHAN_INO];
633 spin_lock(&im->ino_lock);
634 f2fs_bug_on(sbi, im->ino_num == 0);
636 spin_unlock(&im->ino_lock);
639 void f2fs_add_orphan_inode(struct inode *inode)
641 /* add new orphan ino entry into list */
642 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
643 f2fs_update_inode_page(inode);
646 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
648 /* remove orphan entry from orphan list */
649 __remove_ino_entry(sbi, ino, ORPHAN_INO);
652 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
658 inode = f2fs_iget_retry(sbi->sb, ino);
661 * there should be a bug that we can't find the entry
664 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
665 return PTR_ERR(inode);
668 err = f2fs_dquot_initialize(inode);
676 /* truncate all the data during iput */
679 err = f2fs_get_node_info(sbi, ino, &ni, false);
683 /* ENOMEM was fully retried in f2fs_evict_inode. */
684 if (ni.blk_addr != NULL_ADDR) {
691 set_sbi_flag(sbi, SBI_NEED_FSCK);
692 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
697 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
699 block_t start_blk, orphan_blocks, i, j;
700 unsigned int s_flags = sbi->sb->s_flags;
706 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
709 if (bdev_read_only(sbi->sb->s_bdev)) {
710 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
714 if (s_flags & SB_RDONLY) {
715 f2fs_info(sbi, "orphan cleanup on readonly fs");
716 sbi->sb->s_flags &= ~SB_RDONLY;
721 * Turn on quotas which were not enabled for read-only mounts if
722 * filesystem has quota feature, so that they are updated correctly.
724 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
727 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
728 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
730 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
732 for (i = 0; i < orphan_blocks; i++) {
734 struct f2fs_orphan_block *orphan_blk;
736 page = f2fs_get_meta_page(sbi, start_blk + i);
742 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
743 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
744 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
746 err = recover_orphan_inode(sbi, ino);
748 f2fs_put_page(page, 1);
752 f2fs_put_page(page, 1);
754 /* clear Orphan Flag */
755 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
757 set_sbi_flag(sbi, SBI_IS_RECOVERED);
760 /* Turn quotas off */
762 f2fs_quota_off_umount(sbi->sb);
764 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
769 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
771 struct list_head *head;
772 struct f2fs_orphan_block *orphan_blk = NULL;
773 unsigned int nentries = 0;
774 unsigned short index = 1;
775 unsigned short orphan_blocks;
776 struct page *page = NULL;
777 struct ino_entry *orphan = NULL;
778 struct inode_management *im = &sbi->im[ORPHAN_INO];
780 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
783 * we don't need to do spin_lock(&im->ino_lock) here, since all the
784 * orphan inode operations are covered under f2fs_lock_op().
785 * And, spin_lock should be avoided due to page operations below.
787 head = &im->ino_list;
789 /* loop for each orphan inode entry and write them in Jornal block */
790 list_for_each_entry(orphan, head, list) {
792 page = f2fs_grab_meta_page(sbi, start_blk++);
794 (struct f2fs_orphan_block *)page_address(page);
795 memset(orphan_blk, 0, sizeof(*orphan_blk));
798 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
800 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
802 * an orphan block is full of 1020 entries,
803 * then we need to flush current orphan blocks
804 * and bring another one in memory
806 orphan_blk->blk_addr = cpu_to_le16(index);
807 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
808 orphan_blk->entry_count = cpu_to_le32(nentries);
809 set_page_dirty(page);
810 f2fs_put_page(page, 1);
818 orphan_blk->blk_addr = cpu_to_le16(index);
819 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
820 orphan_blk->entry_count = cpu_to_le32(nentries);
821 set_page_dirty(page);
822 f2fs_put_page(page, 1);
826 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
827 struct f2fs_checkpoint *ckpt)
829 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
832 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
833 if (chksum_ofs < CP_CHKSUM_OFFSET) {
834 chksum_ofs += sizeof(chksum);
835 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
836 F2FS_BLKSIZE - chksum_ofs);
841 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
842 struct f2fs_checkpoint **cp_block, struct page **cp_page,
843 unsigned long long *version)
845 size_t crc_offset = 0;
848 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
849 if (IS_ERR(*cp_page))
850 return PTR_ERR(*cp_page);
852 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
854 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
855 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
856 crc_offset > CP_CHKSUM_OFFSET) {
857 f2fs_put_page(*cp_page, 1);
858 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
862 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
863 if (crc != cur_cp_crc(*cp_block)) {
864 f2fs_put_page(*cp_page, 1);
865 f2fs_warn(sbi, "invalid crc value");
869 *version = cur_cp_version(*cp_block);
873 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
874 block_t cp_addr, unsigned long long *version)
876 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
877 struct f2fs_checkpoint *cp_block = NULL;
878 unsigned long long cur_version = 0, pre_version = 0;
879 unsigned int cp_blocks;
882 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
883 &cp_page_1, version);
887 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
889 if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
890 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
891 le32_to_cpu(cp_block->cp_pack_total_block_count));
894 pre_version = *version;
896 cp_addr += cp_blocks - 1;
897 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
898 &cp_page_2, version);
901 cur_version = *version;
903 if (cur_version == pre_version) {
904 *version = cur_version;
905 f2fs_put_page(cp_page_2, 1);
908 f2fs_put_page(cp_page_2, 1);
910 f2fs_put_page(cp_page_1, 1);
914 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
916 struct f2fs_checkpoint *cp_block;
917 struct f2fs_super_block *fsb = sbi->raw_super;
918 struct page *cp1, *cp2, *cur_page;
919 unsigned long blk_size = sbi->blocksize;
920 unsigned long long cp1_version = 0, cp2_version = 0;
921 unsigned long long cp_start_blk_no;
922 unsigned int cp_blks = 1 + __cp_payload(sbi);
927 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
932 * Finding out valid cp block involves read both
933 * sets( cp pack 1 and cp pack 2)
935 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
936 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
938 /* The second checkpoint pack should start at the next segment */
939 cp_start_blk_no += ((unsigned long long)1) <<
940 le32_to_cpu(fsb->log_blocks_per_seg);
941 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
944 if (ver_after(cp2_version, cp1_version))
957 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
958 memcpy(sbi->ckpt, cp_block, blk_size);
961 sbi->cur_cp_pack = 1;
963 sbi->cur_cp_pack = 2;
965 /* Sanity checking of checkpoint */
966 if (f2fs_sanity_check_ckpt(sbi)) {
968 goto free_fail_no_cp;
974 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
976 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
978 for (i = 1; i < cp_blks; i++) {
979 void *sit_bitmap_ptr;
980 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
982 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
983 if (IS_ERR(cur_page)) {
984 err = PTR_ERR(cur_page);
985 goto free_fail_no_cp;
987 sit_bitmap_ptr = page_address(cur_page);
988 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
989 f2fs_put_page(cur_page, 1);
992 f2fs_put_page(cp1, 1);
993 f2fs_put_page(cp2, 1);
997 f2fs_put_page(cp1, 1);
998 f2fs_put_page(cp2, 1);
1004 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1006 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1007 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1009 if (is_inode_flag_set(inode, flag))
1012 set_inode_flag(inode, flag);
1013 if (!f2fs_is_volatile_file(inode))
1014 list_add_tail(&F2FS_I(inode)->dirty_list,
1015 &sbi->inode_list[type]);
1016 stat_inc_dirty_inode(sbi, type);
1019 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1021 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1023 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1026 list_del_init(&F2FS_I(inode)->dirty_list);
1027 clear_inode_flag(inode, flag);
1028 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1031 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1033 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1034 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1036 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1037 !S_ISLNK(inode->i_mode))
1040 spin_lock(&sbi->inode_lock[type]);
1041 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1042 __add_dirty_inode(inode, type);
1043 inode_inc_dirty_pages(inode);
1044 spin_unlock(&sbi->inode_lock[type]);
1046 set_page_private_reference(&folio->page);
1049 void f2fs_remove_dirty_inode(struct inode *inode)
1051 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1052 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1054 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1055 !S_ISLNK(inode->i_mode))
1058 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1061 spin_lock(&sbi->inode_lock[type]);
1062 __remove_dirty_inode(inode, type);
1063 spin_unlock(&sbi->inode_lock[type]);
1066 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1068 struct list_head *head;
1069 struct inode *inode;
1070 struct f2fs_inode_info *fi;
1071 bool is_dir = (type == DIR_INODE);
1072 unsigned long ino = 0;
1074 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1075 get_pages(sbi, is_dir ?
1076 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1078 if (unlikely(f2fs_cp_error(sbi))) {
1079 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1080 get_pages(sbi, is_dir ?
1081 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085 spin_lock(&sbi->inode_lock[type]);
1087 head = &sbi->inode_list[type];
1088 if (list_empty(head)) {
1089 spin_unlock(&sbi->inode_lock[type]);
1090 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1091 get_pages(sbi, is_dir ?
1092 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1095 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1096 inode = igrab(&fi->vfs_inode);
1097 spin_unlock(&sbi->inode_lock[type]);
1099 unsigned long cur_ino = inode->i_ino;
1101 F2FS_I(inode)->cp_task = current;
1103 filemap_fdatawrite(inode->i_mapping);
1105 F2FS_I(inode)->cp_task = NULL;
1108 /* We need to give cpu to another writers. */
1115 * We should submit bio, since it exists several
1116 * wribacking dentry pages in the freeing inode.
1118 f2fs_submit_merged_write(sbi, DATA);
1124 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1126 struct list_head *head = &sbi->inode_list[DIRTY_META];
1127 struct inode *inode;
1128 struct f2fs_inode_info *fi;
1129 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1132 if (unlikely(f2fs_cp_error(sbi)))
1135 spin_lock(&sbi->inode_lock[DIRTY_META]);
1136 if (list_empty(head)) {
1137 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1140 fi = list_first_entry(head, struct f2fs_inode_info,
1142 inode = igrab(&fi->vfs_inode);
1143 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1145 sync_inode_metadata(inode, 0);
1147 /* it's on eviction */
1148 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1149 f2fs_update_inode_page(inode);
1156 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1158 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1159 struct f2fs_nm_info *nm_i = NM_I(sbi);
1160 nid_t last_nid = nm_i->next_scan_nid;
1162 next_free_nid(sbi, &last_nid);
1163 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1164 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1165 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1166 ckpt->next_free_nid = cpu_to_le32(last_nid);
1169 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1173 if (!is_journalled_quota(sbi))
1176 if (!f2fs_down_write_trylock(&sbi->quota_sem))
1178 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1180 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1182 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1183 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1185 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1188 f2fs_up_write(&sbi->quota_sem);
1193 * Freeze all the FS-operations for checkpoint.
1195 static int block_operations(struct f2fs_sb_info *sbi)
1197 struct writeback_control wbc = {
1198 .sync_mode = WB_SYNC_ALL,
1199 .nr_to_write = LONG_MAX,
1202 int err = 0, cnt = 0;
1205 * Let's flush inline_data in dirty node pages.
1207 f2fs_flush_inline_data(sbi);
1211 if (__need_flush_quota(sbi)) {
1214 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1215 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1216 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1217 goto retry_flush_dents;
1219 f2fs_unlock_all(sbi);
1221 /* only failed during mount/umount/freeze/quotactl */
1222 locked = down_read_trylock(&sbi->sb->s_umount);
1223 f2fs_quota_sync(sbi->sb, -1);
1225 up_read(&sbi->sb->s_umount);
1227 goto retry_flush_quotas;
1231 /* write all the dirty dentry pages */
1232 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1233 f2fs_unlock_all(sbi);
1234 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1238 goto retry_flush_quotas;
1242 * POR: we should ensure that there are no dirty node pages
1243 * until finishing nat/sit flush. inode->i_blocks can be updated.
1245 f2fs_down_write(&sbi->node_change);
1247 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1248 f2fs_up_write(&sbi->node_change);
1249 f2fs_unlock_all(sbi);
1250 err = f2fs_sync_inode_meta(sbi);
1254 goto retry_flush_quotas;
1258 f2fs_down_write(&sbi->node_write);
1260 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1261 f2fs_up_write(&sbi->node_write);
1262 atomic_inc(&sbi->wb_sync_req[NODE]);
1263 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1264 atomic_dec(&sbi->wb_sync_req[NODE]);
1266 f2fs_up_write(&sbi->node_change);
1267 f2fs_unlock_all(sbi);
1271 goto retry_flush_nodes;
1275 * sbi->node_change is used only for AIO write_begin path which produces
1276 * dirty node blocks and some checkpoint values by block allocation.
1278 __prepare_cp_block(sbi);
1279 f2fs_up_write(&sbi->node_change);
1283 static void unblock_operations(struct f2fs_sb_info *sbi)
1285 f2fs_up_write(&sbi->node_write);
1286 f2fs_unlock_all(sbi);
1289 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1294 if (!get_pages(sbi, type))
1297 if (unlikely(f2fs_cp_error(sbi)))
1300 if (type == F2FS_DIRTY_META)
1301 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1303 else if (type == F2FS_WB_CP_DATA)
1304 f2fs_submit_merged_write(sbi, DATA);
1306 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1307 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1309 finish_wait(&sbi->cp_wait, &wait);
1312 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1314 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1315 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1316 unsigned long flags;
1318 if (cpc->reason & CP_UMOUNT) {
1319 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1320 NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1321 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1322 f2fs_notice(sbi, "Disable nat_bits due to no space");
1323 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1324 f2fs_nat_bitmap_enabled(sbi)) {
1325 f2fs_enable_nat_bits(sbi);
1326 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1327 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1331 spin_lock_irqsave(&sbi->cp_lock, flags);
1333 if (cpc->reason & CP_TRIMMED)
1334 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1336 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1338 if (cpc->reason & CP_UMOUNT)
1339 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1341 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1343 if (cpc->reason & CP_FASTBOOT)
1344 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1346 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1349 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1351 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1353 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1354 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1356 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1357 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1359 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1361 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1362 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1364 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1366 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1367 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1369 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1371 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1372 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1374 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1376 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1377 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1379 /* set this flag to activate crc|cp_ver for recovery */
1380 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1381 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1383 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1386 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1387 void *src, block_t blk_addr)
1389 struct writeback_control wbc = {
1394 * pagevec_lookup_tag and lock_page again will take
1395 * some extra time. Therefore, f2fs_update_meta_pages and
1396 * f2fs_sync_meta_pages are combined in this function.
1398 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1401 f2fs_wait_on_page_writeback(page, META, true, true);
1403 memcpy(page_address(page), src, PAGE_SIZE);
1405 set_page_dirty(page);
1406 if (unlikely(!clear_page_dirty_for_io(page)))
1407 f2fs_bug_on(sbi, 1);
1409 /* writeout cp pack 2 page */
1410 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1411 if (unlikely(err && f2fs_cp_error(sbi))) {
1412 f2fs_put_page(page, 1);
1416 f2fs_bug_on(sbi, err);
1417 f2fs_put_page(page, 0);
1419 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1420 f2fs_submit_merged_write(sbi, META_FLUSH);
1423 static inline u64 get_sectors_written(struct block_device *bdev)
1425 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1428 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1430 if (f2fs_is_multi_device(sbi)) {
1434 for (i = 0; i < sbi->s_ndevs; i++)
1435 sectors += get_sectors_written(FDEV(i).bdev);
1440 return get_sectors_written(sbi->sb->s_bdev);
1443 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1445 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1446 struct f2fs_nm_info *nm_i = NM_I(sbi);
1447 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1449 unsigned int data_sum_blocks, orphan_blocks;
1452 int cp_payload_blks = __cp_payload(sbi);
1453 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1457 /* Flush all the NAT/SIT pages */
1458 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1460 /* start to update checkpoint, cp ver is already updated previously */
1461 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1462 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1463 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1464 ckpt->cur_node_segno[i] =
1465 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1466 ckpt->cur_node_blkoff[i] =
1467 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1468 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1469 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1471 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1472 ckpt->cur_data_segno[i] =
1473 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1474 ckpt->cur_data_blkoff[i] =
1475 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1476 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1477 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1480 /* 2 cp + n data seg summary + orphan inode blocks */
1481 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1482 spin_lock_irqsave(&sbi->cp_lock, flags);
1483 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1484 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1486 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1487 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1489 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1490 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1493 if (__remain_node_summaries(cpc->reason))
1494 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1495 cp_payload_blks + data_sum_blocks +
1496 orphan_blocks + NR_CURSEG_NODE_TYPE);
1498 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1499 cp_payload_blks + data_sum_blocks +
1502 /* update ckpt flag for checkpoint */
1503 update_ckpt_flags(sbi, cpc);
1505 /* update SIT/NAT bitmap */
1506 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1507 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1509 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1510 *((__le32 *)((unsigned char *)ckpt +
1511 le32_to_cpu(ckpt->checksum_offset)))
1512 = cpu_to_le32(crc32);
1514 start_blk = __start_cp_next_addr(sbi);
1516 /* write nat bits */
1517 if ((cpc->reason & CP_UMOUNT) &&
1518 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1519 __u64 cp_ver = cur_cp_version(ckpt);
1522 cp_ver |= ((__u64)crc32 << 32);
1523 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1525 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1526 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1527 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1528 (i << F2FS_BLKSIZE_BITS), blk + i);
1531 /* write out checkpoint buffer at block 0 */
1532 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1534 for (i = 1; i < 1 + cp_payload_blks; i++)
1535 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1539 write_orphan_inodes(sbi, start_blk);
1540 start_blk += orphan_blocks;
1543 f2fs_write_data_summaries(sbi, start_blk);
1544 start_blk += data_sum_blocks;
1546 /* Record write statistics in the hot node summary */
1547 kbytes_written = sbi->kbytes_written;
1548 kbytes_written += (f2fs_get_sectors_written(sbi) -
1549 sbi->sectors_written_start) >> 1;
1550 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1552 if (__remain_node_summaries(cpc->reason)) {
1553 f2fs_write_node_summaries(sbi, start_blk);
1554 start_blk += NR_CURSEG_NODE_TYPE;
1557 /* update user_block_counts */
1558 sbi->last_valid_block_count = sbi->total_valid_block_count;
1559 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1560 percpu_counter_set(&sbi->rf_node_block_count, 0);
1562 /* Here, we have one bio having CP pack except cp pack 2 page */
1563 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1564 /* Wait for all dirty meta pages to be submitted for IO */
1565 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1567 /* wait for previous submitted meta pages writeback */
1568 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1570 /* flush all device cache */
1571 err = f2fs_flush_device_cache(sbi);
1575 /* barrier and flush checkpoint cp pack 2 page if it can */
1576 commit_checkpoint(sbi, ckpt, start_blk);
1577 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1580 * invalidate intermediate page cache borrowed from meta inode which are
1581 * used for migration of encrypted, verity or compressed inode's blocks.
1583 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1584 f2fs_sb_has_compression(sbi))
1585 invalidate_mapping_pages(META_MAPPING(sbi),
1586 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1588 f2fs_release_ino_entry(sbi, false);
1590 f2fs_reset_fsync_node_info(sbi);
1592 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1593 clear_sbi_flag(sbi, SBI_NEED_CP);
1594 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1596 spin_lock(&sbi->stat_lock);
1597 sbi->unusable_block_count = 0;
1598 spin_unlock(&sbi->stat_lock);
1600 __set_cp_next_pack(sbi);
1603 * redirty superblock if metadata like node page or inode cache is
1604 * updated during writing checkpoint.
1606 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1607 get_pages(sbi, F2FS_DIRTY_IMETA))
1608 set_sbi_flag(sbi, SBI_IS_DIRTY);
1610 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1612 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1615 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1617 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1618 unsigned long long ckpt_ver;
1621 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1624 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1625 if (cpc->reason != CP_PAUSE)
1627 f2fs_warn(sbi, "Start checkpoint disabled!");
1629 if (cpc->reason != CP_RESIZE)
1630 f2fs_down_write(&sbi->cp_global_sem);
1632 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1633 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1634 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1636 if (unlikely(f2fs_cp_error(sbi))) {
1641 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1643 err = block_operations(sbi);
1647 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1649 f2fs_flush_merged_writes(sbi);
1651 /* this is the case of multiple fstrims without any changes */
1652 if (cpc->reason & CP_DISCARD) {
1653 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1654 unblock_operations(sbi);
1658 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1659 SIT_I(sbi)->dirty_sentries == 0 &&
1660 prefree_segments(sbi) == 0) {
1661 f2fs_flush_sit_entries(sbi, cpc);
1662 f2fs_clear_prefree_segments(sbi, cpc);
1663 unblock_operations(sbi);
1669 * update checkpoint pack index
1670 * Increase the version number so that
1671 * SIT entries and seg summaries are written at correct place
1673 ckpt_ver = cur_cp_version(ckpt);
1674 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1676 /* write cached NAT/SIT entries to NAT/SIT area */
1677 err = f2fs_flush_nat_entries(sbi, cpc);
1679 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1680 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1684 f2fs_flush_sit_entries(sbi, cpc);
1686 /* save inmem log status */
1687 f2fs_save_inmem_curseg(sbi);
1689 err = do_checkpoint(sbi, cpc);
1691 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1692 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1693 f2fs_release_discard_addrs(sbi);
1695 f2fs_clear_prefree_segments(sbi, cpc);
1698 f2fs_restore_inmem_curseg(sbi);
1700 unblock_operations(sbi);
1701 stat_inc_cp_count(sbi->stat_info);
1703 if (cpc->reason & CP_RECOVERY)
1704 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1706 /* update CP_TIME to trigger checkpoint periodically */
1707 f2fs_update_time(sbi, CP_TIME);
1708 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1710 if (cpc->reason != CP_RESIZE)
1711 f2fs_up_write(&sbi->cp_global_sem);
1715 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1719 for (i = 0; i < MAX_INO_ENTRY; i++) {
1720 struct inode_management *im = &sbi->im[i];
1722 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1723 spin_lock_init(&im->ino_lock);
1724 INIT_LIST_HEAD(&im->ino_list);
1728 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1729 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1730 F2FS_ORPHANS_PER_BLOCK;
1733 int __init f2fs_create_checkpoint_caches(void)
1735 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1736 sizeof(struct ino_entry));
1737 if (!ino_entry_slab)
1739 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1740 sizeof(struct inode_entry));
1741 if (!f2fs_inode_entry_slab) {
1742 kmem_cache_destroy(ino_entry_slab);
1748 void f2fs_destroy_checkpoint_caches(void)
1750 kmem_cache_destroy(ino_entry_slab);
1751 kmem_cache_destroy(f2fs_inode_entry_slab);
1754 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1756 struct cp_control cpc = { .reason = CP_SYNC, };
1759 f2fs_down_write(&sbi->gc_lock);
1760 err = f2fs_write_checkpoint(sbi, &cpc);
1761 f2fs_up_write(&sbi->gc_lock);
1766 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1768 struct ckpt_req_control *cprc = &sbi->cprc_info;
1769 struct ckpt_req *req, *next;
1770 struct llist_node *dispatch_list;
1771 u64 sum_diff = 0, diff, count = 0;
1774 dispatch_list = llist_del_all(&cprc->issue_list);
1777 dispatch_list = llist_reverse_order(dispatch_list);
1779 ret = __write_checkpoint_sync(sbi);
1780 atomic_inc(&cprc->issued_ckpt);
1782 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1783 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1785 complete(&req->wait);
1790 atomic_sub(count, &cprc->queued_ckpt);
1791 atomic_add(count, &cprc->total_ckpt);
1793 spin_lock(&cprc->stat_lock);
1794 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1795 if (cprc->peak_time < cprc->cur_time)
1796 cprc->peak_time = cprc->cur_time;
1797 spin_unlock(&cprc->stat_lock);
1800 static int issue_checkpoint_thread(void *data)
1802 struct f2fs_sb_info *sbi = data;
1803 struct ckpt_req_control *cprc = &sbi->cprc_info;
1804 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1806 if (kthread_should_stop())
1809 if (!llist_empty(&cprc->issue_list))
1810 __checkpoint_and_complete_reqs(sbi);
1812 wait_event_interruptible(*q,
1813 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1817 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1818 struct ckpt_req *wait_req)
1820 struct ckpt_req_control *cprc = &sbi->cprc_info;
1822 if (!llist_empty(&cprc->issue_list)) {
1823 __checkpoint_and_complete_reqs(sbi);
1825 /* already dispatched by issue_checkpoint_thread */
1827 wait_for_completion(&wait_req->wait);
1831 static void init_ckpt_req(struct ckpt_req *req)
1833 memset(req, 0, sizeof(struct ckpt_req));
1835 init_completion(&req->wait);
1836 req->queue_time = ktime_get();
1839 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1841 struct ckpt_req_control *cprc = &sbi->cprc_info;
1842 struct ckpt_req req;
1843 struct cp_control cpc;
1845 cpc.reason = __get_cp_reason(sbi);
1846 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1849 f2fs_down_write(&sbi->gc_lock);
1850 ret = f2fs_write_checkpoint(sbi, &cpc);
1851 f2fs_up_write(&sbi->gc_lock);
1856 if (!cprc->f2fs_issue_ckpt)
1857 return __write_checkpoint_sync(sbi);
1859 init_ckpt_req(&req);
1861 llist_add(&req.llnode, &cprc->issue_list);
1862 atomic_inc(&cprc->queued_ckpt);
1865 * update issue_list before we wake up issue_checkpoint thread,
1866 * this smp_mb() pairs with another barrier in ___wait_event(),
1867 * see more details in comments of waitqueue_active().
1871 if (waitqueue_active(&cprc->ckpt_wait_queue))
1872 wake_up(&cprc->ckpt_wait_queue);
1874 if (cprc->f2fs_issue_ckpt)
1875 wait_for_completion(&req.wait);
1877 flush_remained_ckpt_reqs(sbi, &req);
1882 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1884 dev_t dev = sbi->sb->s_bdev->bd_dev;
1885 struct ckpt_req_control *cprc = &sbi->cprc_info;
1887 if (cprc->f2fs_issue_ckpt)
1890 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1891 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1892 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1893 cprc->f2fs_issue_ckpt = NULL;
1897 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1902 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1904 struct ckpt_req_control *cprc = &sbi->cprc_info;
1906 if (cprc->f2fs_issue_ckpt) {
1907 struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt;
1909 cprc->f2fs_issue_ckpt = NULL;
1910 kthread_stop(ckpt_task);
1912 flush_remained_ckpt_reqs(sbi, NULL);
1916 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1918 struct ckpt_req_control *cprc = &sbi->cprc_info;
1920 atomic_set(&cprc->issued_ckpt, 0);
1921 atomic_set(&cprc->total_ckpt, 0);
1922 atomic_set(&cprc->queued_ckpt, 0);
1923 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1924 init_waitqueue_head(&cprc->ckpt_wait_queue);
1925 init_llist_head(&cprc->issue_list);
1926 spin_lock_init(&cprc->stat_lock);