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,
32 f2fs_build_fault_attr(sbi, 0, 0);
33 set_ckpt_flags(sbi, CP_ERROR_FLAG);
35 f2fs_flush_merged_writes(sbi);
37 f2fs_handle_stop(sbi, reason);
42 * We guarantee no failure on the returned page.
44 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
46 struct address_space *mapping = META_MAPPING(sbi);
49 page = f2fs_grab_cache_page(mapping, index, false);
54 f2fs_wait_on_page_writeback(page, META, true, true);
55 if (!PageUptodate(page))
56 SetPageUptodate(page);
60 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
63 struct address_space *mapping = META_MAPPING(sbi);
65 struct f2fs_io_info fio = {
69 .op_flags = REQ_META | REQ_PRIO,
72 .encrypted_page = NULL,
77 if (unlikely(!is_meta))
78 fio.op_flags &= ~REQ_META;
80 page = f2fs_grab_cache_page(mapping, index, false);
85 if (PageUptodate(page))
90 err = f2fs_submit_page_bio(&fio);
92 f2fs_put_page(page, 1);
96 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
99 if (unlikely(page->mapping != mapping)) {
100 f2fs_put_page(page, 1);
104 if (unlikely(!PageUptodate(page))) {
105 f2fs_handle_page_eio(sbi, page->index, META);
106 f2fs_put_page(page, 1);
107 return ERR_PTR(-EIO);
113 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
115 return __get_meta_page(sbi, index, true);
118 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
124 page = __get_meta_page(sbi, index, true);
126 if (PTR_ERR(page) == -EIO &&
127 ++count <= DEFAULT_RETRY_IO_COUNT)
129 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
135 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
137 return __get_meta_page(sbi, index, false);
140 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
143 struct seg_entry *se;
144 unsigned int segno, offset;
147 if (type == DATA_GENERIC)
150 segno = GET_SEGNO(sbi, blkaddr);
151 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
152 se = get_seg_entry(sbi, segno);
154 exist = f2fs_test_bit(offset, se->cur_valid_map);
155 if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
156 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
158 set_sbi_flag(sbi, SBI_NEED_FSCK);
162 if (!exist && type == DATA_GENERIC_ENHANCE) {
163 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
165 set_sbi_flag(sbi, SBI_NEED_FSCK);
171 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
172 block_t blkaddr, int type)
178 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
182 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
183 blkaddr < SM_I(sbi)->ssa_blkaddr))
187 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
188 blkaddr < __start_cp_addr(sbi)))
192 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
193 blkaddr < MAIN_BLKADDR(sbi)))
197 case DATA_GENERIC_ENHANCE:
198 case DATA_GENERIC_ENHANCE_READ:
199 case DATA_GENERIC_ENHANCE_UPDATE:
200 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
201 blkaddr < MAIN_BLKADDR(sbi))) {
202 f2fs_warn(sbi, "access invalid blkaddr:%u",
204 set_sbi_flag(sbi, SBI_NEED_FSCK);
208 return __is_bitmap_valid(sbi, blkaddr, type);
212 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
213 blkaddr >= MAIN_BLKADDR(sbi)))
224 * Readahead CP/NAT/SIT/SSA/POR pages
226 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
230 block_t blkno = start;
231 struct f2fs_io_info fio = {
235 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
236 .encrypted_page = NULL,
238 .is_por = (type == META_POR),
240 struct blk_plug plug;
243 if (unlikely(type == META_POR))
244 fio.op_flags &= ~REQ_META;
246 blk_start_plug(&plug);
247 for (; nrpages-- > 0; blkno++) {
249 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
254 if (unlikely(blkno >=
255 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
257 /* get nat block addr */
258 fio.new_blkaddr = current_nat_addr(sbi,
259 blkno * NAT_ENTRY_PER_BLOCK);
262 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
264 /* get sit block addr */
265 fio.new_blkaddr = current_sit_addr(sbi,
266 blkno * SIT_ENTRY_PER_BLOCK);
271 fio.new_blkaddr = blkno;
277 page = f2fs_grab_cache_page(META_MAPPING(sbi),
278 fio.new_blkaddr, false);
281 if (PageUptodate(page)) {
282 f2fs_put_page(page, 1);
287 err = f2fs_submit_page_bio(&fio);
288 f2fs_put_page(page, err ? 1 : 0);
291 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
295 blk_finish_plug(&plug);
296 return blkno - start;
299 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
300 unsigned int ra_blocks)
303 bool readahead = false;
305 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
308 page = find_get_page(META_MAPPING(sbi), index);
309 if (!page || !PageUptodate(page))
311 f2fs_put_page(page, 0);
314 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
317 static int __f2fs_write_meta_page(struct page *page,
318 struct writeback_control *wbc,
319 enum iostat_type io_type)
321 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
323 trace_f2fs_writepage(page, META);
325 if (unlikely(f2fs_cp_error(sbi)))
327 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
329 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
332 f2fs_do_write_meta_page(sbi, page, io_type);
333 dec_page_count(sbi, F2FS_DIRTY_META);
335 if (wbc->for_reclaim)
336 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
340 if (unlikely(f2fs_cp_error(sbi)))
341 f2fs_submit_merged_write(sbi, META);
346 redirty_page_for_writepage(wbc, page);
347 return AOP_WRITEPAGE_ACTIVATE;
350 static int f2fs_write_meta_page(struct page *page,
351 struct writeback_control *wbc)
353 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
356 static int f2fs_write_meta_pages(struct address_space *mapping,
357 struct writeback_control *wbc)
359 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
362 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
365 /* collect a number of dirty meta pages and write together */
366 if (wbc->sync_mode != WB_SYNC_ALL &&
367 get_pages(sbi, F2FS_DIRTY_META) <
368 nr_pages_to_skip(sbi, META))
371 /* if locked failed, cp will flush dirty pages instead */
372 if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
375 trace_f2fs_writepages(mapping->host, wbc, META);
376 diff = nr_pages_to_write(sbi, META, wbc);
377 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
378 f2fs_up_write(&sbi->cp_global_sem);
379 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
383 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
384 trace_f2fs_writepages(mapping->host, wbc, META);
388 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
389 long nr_to_write, enum iostat_type io_type)
391 struct address_space *mapping = META_MAPPING(sbi);
392 pgoff_t index = 0, prev = ULONG_MAX;
396 struct writeback_control wbc = {
399 struct blk_plug plug;
403 blk_start_plug(&plug);
405 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
406 PAGECACHE_TAG_DIRTY))) {
409 for (i = 0; i < nr_pages; i++) {
410 struct page *page = pvec.pages[i];
412 if (prev == ULONG_MAX)
413 prev = page->index - 1;
414 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
415 pagevec_release(&pvec);
421 if (unlikely(page->mapping != mapping)) {
426 if (!PageDirty(page)) {
427 /* someone wrote it for us */
428 goto continue_unlock;
431 f2fs_wait_on_page_writeback(page, META, true, true);
433 if (!clear_page_dirty_for_io(page))
434 goto continue_unlock;
436 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
442 if (unlikely(nwritten >= nr_to_write))
445 pagevec_release(&pvec);
450 f2fs_submit_merged_write(sbi, type);
452 blk_finish_plug(&plug);
457 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
460 trace_f2fs_set_page_dirty(&folio->page, META);
462 if (!folio_test_uptodate(folio))
463 folio_mark_uptodate(folio);
464 if (filemap_dirty_folio(mapping, folio)) {
465 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
466 set_page_private_reference(&folio->page);
472 const struct address_space_operations f2fs_meta_aops = {
473 .writepage = f2fs_write_meta_page,
474 .writepages = f2fs_write_meta_pages,
475 .dirty_folio = f2fs_dirty_meta_folio,
476 .invalidate_folio = f2fs_invalidate_folio,
477 .release_folio = f2fs_release_folio,
478 .migrate_folio = filemap_migrate_folio,
481 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
482 unsigned int devidx, int type)
484 struct inode_management *im = &sbi->im[type];
485 struct ino_entry *e = NULL, *new = NULL;
487 if (type == FLUSH_INO) {
489 e = radix_tree_lookup(&im->ino_root, ino);
495 new = f2fs_kmem_cache_alloc(ino_entry_slab,
496 GFP_NOFS, true, NULL);
498 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
500 spin_lock(&im->ino_lock);
501 e = radix_tree_lookup(&im->ino_root, ino);
504 spin_unlock(&im->ino_lock);
508 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
511 memset(e, 0, sizeof(struct ino_entry));
514 list_add_tail(&e->list, &im->ino_list);
515 if (type != ORPHAN_INO)
519 if (type == FLUSH_INO)
520 f2fs_set_bit(devidx, (char *)&e->dirty_device);
522 spin_unlock(&im->ino_lock);
523 radix_tree_preload_end();
526 kmem_cache_free(ino_entry_slab, new);
529 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
531 struct inode_management *im = &sbi->im[type];
534 spin_lock(&im->ino_lock);
535 e = radix_tree_lookup(&im->ino_root, ino);
538 radix_tree_delete(&im->ino_root, ino);
540 spin_unlock(&im->ino_lock);
541 kmem_cache_free(ino_entry_slab, e);
544 spin_unlock(&im->ino_lock);
547 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
549 /* add new dirty ino entry into list */
550 __add_ino_entry(sbi, ino, 0, type);
553 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
555 /* remove dirty ino entry from list */
556 __remove_ino_entry(sbi, ino, type);
559 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
560 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
562 struct inode_management *im = &sbi->im[mode];
565 spin_lock(&im->ino_lock);
566 e = radix_tree_lookup(&im->ino_root, ino);
567 spin_unlock(&im->ino_lock);
568 return e ? true : false;
571 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
573 struct ino_entry *e, *tmp;
576 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
577 struct inode_management *im = &sbi->im[i];
579 spin_lock(&im->ino_lock);
580 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
582 radix_tree_delete(&im->ino_root, e->ino);
583 kmem_cache_free(ino_entry_slab, e);
586 spin_unlock(&im->ino_lock);
590 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
591 unsigned int devidx, int type)
593 __add_ino_entry(sbi, ino, devidx, type);
596 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
597 unsigned int devidx, int type)
599 struct inode_management *im = &sbi->im[type];
601 bool is_dirty = false;
603 spin_lock(&im->ino_lock);
604 e = radix_tree_lookup(&im->ino_root, ino);
605 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
607 spin_unlock(&im->ino_lock);
611 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
613 struct inode_management *im = &sbi->im[ORPHAN_INO];
616 spin_lock(&im->ino_lock);
618 if (time_to_inject(sbi, FAULT_ORPHAN)) {
619 spin_unlock(&im->ino_lock);
620 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
624 if (unlikely(im->ino_num >= sbi->max_orphans))
628 spin_unlock(&im->ino_lock);
633 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
635 struct inode_management *im = &sbi->im[ORPHAN_INO];
637 spin_lock(&im->ino_lock);
638 f2fs_bug_on(sbi, im->ino_num == 0);
640 spin_unlock(&im->ino_lock);
643 void f2fs_add_orphan_inode(struct inode *inode)
645 /* add new orphan ino entry into list */
646 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
647 f2fs_update_inode_page(inode);
650 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
652 /* remove orphan entry from orphan list */
653 __remove_ino_entry(sbi, ino, ORPHAN_INO);
656 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
662 inode = f2fs_iget_retry(sbi->sb, ino);
665 * there should be a bug that we can't find the entry
668 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
669 return PTR_ERR(inode);
672 err = f2fs_dquot_initialize(inode);
680 /* truncate all the data during iput */
683 err = f2fs_get_node_info(sbi, ino, &ni, false);
687 /* ENOMEM was fully retried in f2fs_evict_inode. */
688 if (ni.blk_addr != NULL_ADDR) {
695 set_sbi_flag(sbi, SBI_NEED_FSCK);
696 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
701 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
703 block_t start_blk, orphan_blocks, i, j;
704 unsigned int s_flags = sbi->sb->s_flags;
710 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
713 if (bdev_read_only(sbi->sb->s_bdev)) {
714 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
718 if (s_flags & SB_RDONLY) {
719 f2fs_info(sbi, "orphan cleanup on readonly fs");
720 sbi->sb->s_flags &= ~SB_RDONLY;
725 * Turn on quotas which were not enabled for read-only mounts if
726 * filesystem has quota feature, so that they are updated correctly.
728 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
731 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
732 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
734 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
736 for (i = 0; i < orphan_blocks; i++) {
738 struct f2fs_orphan_block *orphan_blk;
740 page = f2fs_get_meta_page(sbi, start_blk + i);
746 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
747 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
748 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
750 err = recover_orphan_inode(sbi, ino);
752 f2fs_put_page(page, 1);
756 f2fs_put_page(page, 1);
758 /* clear Orphan Flag */
759 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
761 set_sbi_flag(sbi, SBI_IS_RECOVERED);
764 /* Turn quotas off */
766 f2fs_quota_off_umount(sbi->sb);
768 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
773 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
775 struct list_head *head;
776 struct f2fs_orphan_block *orphan_blk = NULL;
777 unsigned int nentries = 0;
778 unsigned short index = 1;
779 unsigned short orphan_blocks;
780 struct page *page = NULL;
781 struct ino_entry *orphan = NULL;
782 struct inode_management *im = &sbi->im[ORPHAN_INO];
784 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
787 * we don't need to do spin_lock(&im->ino_lock) here, since all the
788 * orphan inode operations are covered under f2fs_lock_op().
789 * And, spin_lock should be avoided due to page operations below.
791 head = &im->ino_list;
793 /* loop for each orphan inode entry and write them in Jornal block */
794 list_for_each_entry(orphan, head, list) {
796 page = f2fs_grab_meta_page(sbi, start_blk++);
798 (struct f2fs_orphan_block *)page_address(page);
799 memset(orphan_blk, 0, sizeof(*orphan_blk));
802 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
804 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
806 * an orphan block is full of 1020 entries,
807 * then we need to flush current orphan blocks
808 * and bring another one in memory
810 orphan_blk->blk_addr = cpu_to_le16(index);
811 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
812 orphan_blk->entry_count = cpu_to_le32(nentries);
813 set_page_dirty(page);
814 f2fs_put_page(page, 1);
822 orphan_blk->blk_addr = cpu_to_le16(index);
823 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
824 orphan_blk->entry_count = cpu_to_le32(nentries);
825 set_page_dirty(page);
826 f2fs_put_page(page, 1);
830 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
831 struct f2fs_checkpoint *ckpt)
833 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
836 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
837 if (chksum_ofs < CP_CHKSUM_OFFSET) {
838 chksum_ofs += sizeof(chksum);
839 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
840 F2FS_BLKSIZE - chksum_ofs);
845 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
846 struct f2fs_checkpoint **cp_block, struct page **cp_page,
847 unsigned long long *version)
849 size_t crc_offset = 0;
852 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
853 if (IS_ERR(*cp_page))
854 return PTR_ERR(*cp_page);
856 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
858 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
859 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
860 crc_offset > CP_CHKSUM_OFFSET) {
861 f2fs_put_page(*cp_page, 1);
862 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
866 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
867 if (crc != cur_cp_crc(*cp_block)) {
868 f2fs_put_page(*cp_page, 1);
869 f2fs_warn(sbi, "invalid crc value");
873 *version = cur_cp_version(*cp_block);
877 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
878 block_t cp_addr, unsigned long long *version)
880 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
881 struct f2fs_checkpoint *cp_block = NULL;
882 unsigned long long cur_version = 0, pre_version = 0;
883 unsigned int cp_blocks;
886 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
887 &cp_page_1, version);
891 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
893 if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
894 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
895 le32_to_cpu(cp_block->cp_pack_total_block_count));
898 pre_version = *version;
900 cp_addr += cp_blocks - 1;
901 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
902 &cp_page_2, version);
905 cur_version = *version;
907 if (cur_version == pre_version) {
908 *version = cur_version;
909 f2fs_put_page(cp_page_2, 1);
912 f2fs_put_page(cp_page_2, 1);
914 f2fs_put_page(cp_page_1, 1);
918 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
920 struct f2fs_checkpoint *cp_block;
921 struct f2fs_super_block *fsb = sbi->raw_super;
922 struct page *cp1, *cp2, *cur_page;
923 unsigned long blk_size = sbi->blocksize;
924 unsigned long long cp1_version = 0, cp2_version = 0;
925 unsigned long long cp_start_blk_no;
926 unsigned int cp_blks = 1 + __cp_payload(sbi);
931 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
936 * Finding out valid cp block involves read both
937 * sets( cp pack 1 and cp pack 2)
939 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
940 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
942 /* The second checkpoint pack should start at the next segment */
943 cp_start_blk_no += ((unsigned long long)1) <<
944 le32_to_cpu(fsb->log_blocks_per_seg);
945 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
948 if (ver_after(cp2_version, cp1_version))
961 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
962 memcpy(sbi->ckpt, cp_block, blk_size);
965 sbi->cur_cp_pack = 1;
967 sbi->cur_cp_pack = 2;
969 /* Sanity checking of checkpoint */
970 if (f2fs_sanity_check_ckpt(sbi)) {
972 goto free_fail_no_cp;
978 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
980 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
982 for (i = 1; i < cp_blks; i++) {
983 void *sit_bitmap_ptr;
984 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
986 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
987 if (IS_ERR(cur_page)) {
988 err = PTR_ERR(cur_page);
989 goto free_fail_no_cp;
991 sit_bitmap_ptr = page_address(cur_page);
992 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
993 f2fs_put_page(cur_page, 1);
996 f2fs_put_page(cp1, 1);
997 f2fs_put_page(cp2, 1);
1001 f2fs_put_page(cp1, 1);
1002 f2fs_put_page(cp2, 1);
1008 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1010 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1011 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1013 if (is_inode_flag_set(inode, flag))
1016 set_inode_flag(inode, flag);
1017 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1018 stat_inc_dirty_inode(sbi, type);
1021 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1023 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1025 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1028 list_del_init(&F2FS_I(inode)->dirty_list);
1029 clear_inode_flag(inode, flag);
1030 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1033 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1035 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1036 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1038 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1039 !S_ISLNK(inode->i_mode))
1042 spin_lock(&sbi->inode_lock[type]);
1043 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1044 __add_dirty_inode(inode, type);
1045 inode_inc_dirty_pages(inode);
1046 spin_unlock(&sbi->inode_lock[type]);
1048 set_page_private_reference(&folio->page);
1051 void f2fs_remove_dirty_inode(struct inode *inode)
1053 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1054 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1056 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1057 !S_ISLNK(inode->i_mode))
1060 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1063 spin_lock(&sbi->inode_lock[type]);
1064 __remove_dirty_inode(inode, type);
1065 spin_unlock(&sbi->inode_lock[type]);
1068 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1071 struct list_head *head;
1072 struct inode *inode;
1073 struct f2fs_inode_info *fi;
1074 bool is_dir = (type == DIR_INODE);
1075 unsigned long ino = 0;
1077 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1078 get_pages(sbi, is_dir ?
1079 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1081 if (unlikely(f2fs_cp_error(sbi))) {
1082 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1083 get_pages(sbi, is_dir ?
1084 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1088 spin_lock(&sbi->inode_lock[type]);
1090 head = &sbi->inode_list[type];
1091 if (list_empty(head)) {
1092 spin_unlock(&sbi->inode_lock[type]);
1093 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1094 get_pages(sbi, is_dir ?
1095 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1098 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1099 inode = igrab(&fi->vfs_inode);
1100 spin_unlock(&sbi->inode_lock[type]);
1102 unsigned long cur_ino = inode->i_ino;
1105 F2FS_I(inode)->cp_task = current;
1106 F2FS_I(inode)->wb_task = current;
1108 filemap_fdatawrite(inode->i_mapping);
1110 F2FS_I(inode)->wb_task = NULL;
1112 F2FS_I(inode)->cp_task = NULL;
1115 /* We need to give cpu to another writers. */
1122 * We should submit bio, since it exists several
1123 * wribacking dentry pages in the freeing inode.
1125 f2fs_submit_merged_write(sbi, DATA);
1131 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1133 struct list_head *head = &sbi->inode_list[DIRTY_META];
1134 struct inode *inode;
1135 struct f2fs_inode_info *fi;
1136 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1139 if (unlikely(f2fs_cp_error(sbi)))
1142 spin_lock(&sbi->inode_lock[DIRTY_META]);
1143 if (list_empty(head)) {
1144 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1147 fi = list_first_entry(head, struct f2fs_inode_info,
1149 inode = igrab(&fi->vfs_inode);
1150 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1152 sync_inode_metadata(inode, 0);
1154 /* it's on eviction */
1155 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1156 f2fs_update_inode_page(inode);
1163 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1165 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1166 struct f2fs_nm_info *nm_i = NM_I(sbi);
1167 nid_t last_nid = nm_i->next_scan_nid;
1169 next_free_nid(sbi, &last_nid);
1170 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1171 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1172 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1173 ckpt->next_free_nid = cpu_to_le32(last_nid);
1176 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1180 if (!is_journalled_quota(sbi))
1183 if (!f2fs_down_write_trylock(&sbi->quota_sem))
1185 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1187 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1189 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1190 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1192 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1195 f2fs_up_write(&sbi->quota_sem);
1200 * Freeze all the FS-operations for checkpoint.
1202 static int block_operations(struct f2fs_sb_info *sbi)
1204 struct writeback_control wbc = {
1205 .sync_mode = WB_SYNC_ALL,
1206 .nr_to_write = LONG_MAX,
1209 int err = 0, cnt = 0;
1212 * Let's flush inline_data in dirty node pages.
1214 f2fs_flush_inline_data(sbi);
1218 if (__need_flush_quota(sbi)) {
1221 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1222 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1223 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1224 goto retry_flush_dents;
1226 f2fs_unlock_all(sbi);
1228 /* only failed during mount/umount/freeze/quotactl */
1229 locked = down_read_trylock(&sbi->sb->s_umount);
1230 f2fs_quota_sync(sbi->sb, -1);
1232 up_read(&sbi->sb->s_umount);
1234 goto retry_flush_quotas;
1238 /* write all the dirty dentry pages */
1239 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1240 f2fs_unlock_all(sbi);
1241 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1245 goto retry_flush_quotas;
1249 * POR: we should ensure that there are no dirty node pages
1250 * until finishing nat/sit flush. inode->i_blocks can be updated.
1252 f2fs_down_write(&sbi->node_change);
1254 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1255 f2fs_up_write(&sbi->node_change);
1256 f2fs_unlock_all(sbi);
1257 err = f2fs_sync_inode_meta(sbi);
1261 goto retry_flush_quotas;
1265 f2fs_down_write(&sbi->node_write);
1267 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1268 f2fs_up_write(&sbi->node_write);
1269 atomic_inc(&sbi->wb_sync_req[NODE]);
1270 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1271 atomic_dec(&sbi->wb_sync_req[NODE]);
1273 f2fs_up_write(&sbi->node_change);
1274 f2fs_unlock_all(sbi);
1278 goto retry_flush_nodes;
1282 * sbi->node_change is used only for AIO write_begin path which produces
1283 * dirty node blocks and some checkpoint values by block allocation.
1285 __prepare_cp_block(sbi);
1286 f2fs_up_write(&sbi->node_change);
1290 static void unblock_operations(struct f2fs_sb_info *sbi)
1292 f2fs_up_write(&sbi->node_write);
1293 f2fs_unlock_all(sbi);
1296 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1301 if (!get_pages(sbi, type))
1304 if (unlikely(f2fs_cp_error(sbi)))
1307 if (type == F2FS_DIRTY_META)
1308 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1310 else if (type == F2FS_WB_CP_DATA)
1311 f2fs_submit_merged_write(sbi, DATA);
1313 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1314 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1316 finish_wait(&sbi->cp_wait, &wait);
1319 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1321 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1322 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1323 unsigned long flags;
1325 if (cpc->reason & CP_UMOUNT) {
1326 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1327 NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1328 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1329 f2fs_notice(sbi, "Disable nat_bits due to no space");
1330 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1331 f2fs_nat_bitmap_enabled(sbi)) {
1332 f2fs_enable_nat_bits(sbi);
1333 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1334 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1338 spin_lock_irqsave(&sbi->cp_lock, flags);
1340 if (cpc->reason & CP_TRIMMED)
1341 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1343 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1345 if (cpc->reason & CP_UMOUNT)
1346 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1348 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1350 if (cpc->reason & CP_FASTBOOT)
1351 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1353 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1356 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1358 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1360 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1361 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1363 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1364 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1366 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1368 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1369 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1371 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1373 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1374 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1376 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1378 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1379 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1381 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1383 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1384 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1386 /* set this flag to activate crc|cp_ver for recovery */
1387 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1388 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1390 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1393 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1394 void *src, block_t blk_addr)
1396 struct writeback_control wbc = {
1401 * pagevec_lookup_tag and lock_page again will take
1402 * some extra time. Therefore, f2fs_update_meta_pages and
1403 * f2fs_sync_meta_pages are combined in this function.
1405 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1408 f2fs_wait_on_page_writeback(page, META, true, true);
1410 memcpy(page_address(page), src, PAGE_SIZE);
1412 set_page_dirty(page);
1413 if (unlikely(!clear_page_dirty_for_io(page)))
1414 f2fs_bug_on(sbi, 1);
1416 /* writeout cp pack 2 page */
1417 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1418 if (unlikely(err && f2fs_cp_error(sbi))) {
1419 f2fs_put_page(page, 1);
1423 f2fs_bug_on(sbi, err);
1424 f2fs_put_page(page, 0);
1426 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1427 f2fs_submit_merged_write(sbi, META_FLUSH);
1430 static inline u64 get_sectors_written(struct block_device *bdev)
1432 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1435 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1437 if (f2fs_is_multi_device(sbi)) {
1441 for (i = 0; i < sbi->s_ndevs; i++)
1442 sectors += get_sectors_written(FDEV(i).bdev);
1447 return get_sectors_written(sbi->sb->s_bdev);
1450 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1452 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1453 struct f2fs_nm_info *nm_i = NM_I(sbi);
1454 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1456 unsigned int data_sum_blocks, orphan_blocks;
1459 int cp_payload_blks = __cp_payload(sbi);
1460 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1464 /* Flush all the NAT/SIT pages */
1465 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1467 /* start to update checkpoint, cp ver is already updated previously */
1468 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1469 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1470 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1471 ckpt->cur_node_segno[i] =
1472 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1473 ckpt->cur_node_blkoff[i] =
1474 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1475 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1476 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1478 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1479 ckpt->cur_data_segno[i] =
1480 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1481 ckpt->cur_data_blkoff[i] =
1482 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1483 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1484 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1487 /* 2 cp + n data seg summary + orphan inode blocks */
1488 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1489 spin_lock_irqsave(&sbi->cp_lock, flags);
1490 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1491 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1493 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1494 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1496 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1497 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1500 if (__remain_node_summaries(cpc->reason))
1501 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1502 cp_payload_blks + data_sum_blocks +
1503 orphan_blocks + NR_CURSEG_NODE_TYPE);
1505 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1506 cp_payload_blks + data_sum_blocks +
1509 /* update ckpt flag for checkpoint */
1510 update_ckpt_flags(sbi, cpc);
1512 /* update SIT/NAT bitmap */
1513 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1514 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1516 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1517 *((__le32 *)((unsigned char *)ckpt +
1518 le32_to_cpu(ckpt->checksum_offset)))
1519 = cpu_to_le32(crc32);
1521 start_blk = __start_cp_next_addr(sbi);
1523 /* write nat bits */
1524 if ((cpc->reason & CP_UMOUNT) &&
1525 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1526 __u64 cp_ver = cur_cp_version(ckpt);
1529 cp_ver |= ((__u64)crc32 << 32);
1530 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1532 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1533 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1534 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1535 (i << F2FS_BLKSIZE_BITS), blk + i);
1538 /* write out checkpoint buffer at block 0 */
1539 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1541 for (i = 1; i < 1 + cp_payload_blks; i++)
1542 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1546 write_orphan_inodes(sbi, start_blk);
1547 start_blk += orphan_blocks;
1550 f2fs_write_data_summaries(sbi, start_blk);
1551 start_blk += data_sum_blocks;
1553 /* Record write statistics in the hot node summary */
1554 kbytes_written = sbi->kbytes_written;
1555 kbytes_written += (f2fs_get_sectors_written(sbi) -
1556 sbi->sectors_written_start) >> 1;
1557 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1559 if (__remain_node_summaries(cpc->reason)) {
1560 f2fs_write_node_summaries(sbi, start_blk);
1561 start_blk += NR_CURSEG_NODE_TYPE;
1564 /* update user_block_counts */
1565 sbi->last_valid_block_count = sbi->total_valid_block_count;
1566 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1567 percpu_counter_set(&sbi->rf_node_block_count, 0);
1569 /* Here, we have one bio having CP pack except cp pack 2 page */
1570 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1571 /* Wait for all dirty meta pages to be submitted for IO */
1572 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1574 /* wait for previous submitted meta pages writeback */
1575 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1577 /* flush all device cache */
1578 err = f2fs_flush_device_cache(sbi);
1582 /* barrier and flush checkpoint cp pack 2 page if it can */
1583 commit_checkpoint(sbi, ckpt, start_blk);
1584 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1587 * invalidate intermediate page cache borrowed from meta inode which are
1588 * used for migration of encrypted, verity or compressed inode's blocks.
1590 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1591 f2fs_sb_has_compression(sbi))
1592 invalidate_mapping_pages(META_MAPPING(sbi),
1593 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1595 f2fs_release_ino_entry(sbi, false);
1597 f2fs_reset_fsync_node_info(sbi);
1599 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1600 clear_sbi_flag(sbi, SBI_NEED_CP);
1601 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1603 spin_lock(&sbi->stat_lock);
1604 sbi->unusable_block_count = 0;
1605 spin_unlock(&sbi->stat_lock);
1607 __set_cp_next_pack(sbi);
1610 * redirty superblock if metadata like node page or inode cache is
1611 * updated during writing checkpoint.
1613 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1614 get_pages(sbi, F2FS_DIRTY_IMETA))
1615 set_sbi_flag(sbi, SBI_IS_DIRTY);
1617 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1619 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1622 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1624 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1625 unsigned long long ckpt_ver;
1628 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1631 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1632 if (cpc->reason != CP_PAUSE)
1634 f2fs_warn(sbi, "Start checkpoint disabled!");
1636 if (cpc->reason != CP_RESIZE)
1637 f2fs_down_write(&sbi->cp_global_sem);
1639 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1640 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1641 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1643 if (unlikely(f2fs_cp_error(sbi))) {
1648 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1650 err = block_operations(sbi);
1654 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1656 f2fs_flush_merged_writes(sbi);
1658 /* this is the case of multiple fstrims without any changes */
1659 if (cpc->reason & CP_DISCARD) {
1660 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1661 unblock_operations(sbi);
1665 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1666 SIT_I(sbi)->dirty_sentries == 0 &&
1667 prefree_segments(sbi) == 0) {
1668 f2fs_flush_sit_entries(sbi, cpc);
1669 f2fs_clear_prefree_segments(sbi, cpc);
1670 unblock_operations(sbi);
1676 * update checkpoint pack index
1677 * Increase the version number so that
1678 * SIT entries and seg summaries are written at correct place
1680 ckpt_ver = cur_cp_version(ckpt);
1681 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1683 /* write cached NAT/SIT entries to NAT/SIT area */
1684 err = f2fs_flush_nat_entries(sbi, cpc);
1686 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1687 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1691 f2fs_flush_sit_entries(sbi, cpc);
1693 /* save inmem log status */
1694 f2fs_save_inmem_curseg(sbi);
1696 err = do_checkpoint(sbi, cpc);
1698 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1699 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1700 f2fs_release_discard_addrs(sbi);
1702 f2fs_clear_prefree_segments(sbi, cpc);
1705 f2fs_restore_inmem_curseg(sbi);
1707 unblock_operations(sbi);
1708 stat_inc_cp_count(sbi->stat_info);
1710 if (cpc->reason & CP_RECOVERY)
1711 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1713 /* update CP_TIME to trigger checkpoint periodically */
1714 f2fs_update_time(sbi, CP_TIME);
1715 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1717 if (cpc->reason != CP_RESIZE)
1718 f2fs_up_write(&sbi->cp_global_sem);
1722 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1726 for (i = 0; i < MAX_INO_ENTRY; i++) {
1727 struct inode_management *im = &sbi->im[i];
1729 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1730 spin_lock_init(&im->ino_lock);
1731 INIT_LIST_HEAD(&im->ino_list);
1735 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1736 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1737 F2FS_ORPHANS_PER_BLOCK;
1740 int __init f2fs_create_checkpoint_caches(void)
1742 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1743 sizeof(struct ino_entry));
1744 if (!ino_entry_slab)
1746 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1747 sizeof(struct inode_entry));
1748 if (!f2fs_inode_entry_slab) {
1749 kmem_cache_destroy(ino_entry_slab);
1755 void f2fs_destroy_checkpoint_caches(void)
1757 kmem_cache_destroy(ino_entry_slab);
1758 kmem_cache_destroy(f2fs_inode_entry_slab);
1761 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1763 struct cp_control cpc = { .reason = CP_SYNC, };
1766 f2fs_down_write(&sbi->gc_lock);
1767 err = f2fs_write_checkpoint(sbi, &cpc);
1768 f2fs_up_write(&sbi->gc_lock);
1773 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1775 struct ckpt_req_control *cprc = &sbi->cprc_info;
1776 struct ckpt_req *req, *next;
1777 struct llist_node *dispatch_list;
1778 u64 sum_diff = 0, diff, count = 0;
1781 dispatch_list = llist_del_all(&cprc->issue_list);
1784 dispatch_list = llist_reverse_order(dispatch_list);
1786 ret = __write_checkpoint_sync(sbi);
1787 atomic_inc(&cprc->issued_ckpt);
1789 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1790 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1792 complete(&req->wait);
1797 atomic_sub(count, &cprc->queued_ckpt);
1798 atomic_add(count, &cprc->total_ckpt);
1800 spin_lock(&cprc->stat_lock);
1801 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1802 if (cprc->peak_time < cprc->cur_time)
1803 cprc->peak_time = cprc->cur_time;
1804 spin_unlock(&cprc->stat_lock);
1807 static int issue_checkpoint_thread(void *data)
1809 struct f2fs_sb_info *sbi = data;
1810 struct ckpt_req_control *cprc = &sbi->cprc_info;
1811 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1813 if (kthread_should_stop())
1816 if (!llist_empty(&cprc->issue_list))
1817 __checkpoint_and_complete_reqs(sbi);
1819 wait_event_interruptible(*q,
1820 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1824 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1825 struct ckpt_req *wait_req)
1827 struct ckpt_req_control *cprc = &sbi->cprc_info;
1829 if (!llist_empty(&cprc->issue_list)) {
1830 __checkpoint_and_complete_reqs(sbi);
1832 /* already dispatched by issue_checkpoint_thread */
1834 wait_for_completion(&wait_req->wait);
1838 static void init_ckpt_req(struct ckpt_req *req)
1840 memset(req, 0, sizeof(struct ckpt_req));
1842 init_completion(&req->wait);
1843 req->queue_time = ktime_get();
1846 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1848 struct ckpt_req_control *cprc = &sbi->cprc_info;
1849 struct ckpt_req req;
1850 struct cp_control cpc;
1852 cpc.reason = __get_cp_reason(sbi);
1853 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1856 f2fs_down_write(&sbi->gc_lock);
1857 ret = f2fs_write_checkpoint(sbi, &cpc);
1858 f2fs_up_write(&sbi->gc_lock);
1863 if (!cprc->f2fs_issue_ckpt)
1864 return __write_checkpoint_sync(sbi);
1866 init_ckpt_req(&req);
1868 llist_add(&req.llnode, &cprc->issue_list);
1869 atomic_inc(&cprc->queued_ckpt);
1872 * update issue_list before we wake up issue_checkpoint thread,
1873 * this smp_mb() pairs with another barrier in ___wait_event(),
1874 * see more details in comments of waitqueue_active().
1878 if (waitqueue_active(&cprc->ckpt_wait_queue))
1879 wake_up(&cprc->ckpt_wait_queue);
1881 if (cprc->f2fs_issue_ckpt)
1882 wait_for_completion(&req.wait);
1884 flush_remained_ckpt_reqs(sbi, &req);
1889 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1891 dev_t dev = sbi->sb->s_bdev->bd_dev;
1892 struct ckpt_req_control *cprc = &sbi->cprc_info;
1894 if (cprc->f2fs_issue_ckpt)
1897 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1898 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1899 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1900 cprc->f2fs_issue_ckpt = NULL;
1904 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1909 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1911 struct ckpt_req_control *cprc = &sbi->cprc_info;
1912 struct task_struct *ckpt_task;
1914 if (!cprc->f2fs_issue_ckpt)
1917 ckpt_task = cprc->f2fs_issue_ckpt;
1918 cprc->f2fs_issue_ckpt = NULL;
1919 kthread_stop(ckpt_task);
1921 f2fs_flush_ckpt_thread(sbi);
1924 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1926 struct ckpt_req_control *cprc = &sbi->cprc_info;
1928 flush_remained_ckpt_reqs(sbi, NULL);
1930 /* Let's wait for the previous dispatched checkpoint. */
1931 while (atomic_read(&cprc->queued_ckpt))
1932 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1935 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1937 struct ckpt_req_control *cprc = &sbi->cprc_info;
1939 atomic_set(&cprc->issued_ckpt, 0);
1940 atomic_set(&cprc->total_ckpt, 0);
1941 atomic_set(&cprc->queued_ckpt, 0);
1942 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1943 init_waitqueue_head(&cprc->ckpt_wait_queue);
1944 init_llist_head(&cprc->issue_list);
1945 spin_lock_init(&cprc->stat_lock);