1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
8 #include <asm/unaligned.h>
10 #include <linux/f2fs_fs.h>
11 #include <linux/sched/mm.h>
17 * Roll forward recovery scenarios.
19 * [Term] F: fsync_mark, D: dentry_mark
21 * 1. inode(x) | CP | inode(x) | dnode(F)
22 * -> Update the latest inode(x).
24 * 2. inode(x) | CP | inode(F) | dnode(F)
27 * 3. inode(x) | CP | dnode(F) | inode(x)
28 * -> Recover to the latest dnode(F), and drop the last inode(x)
30 * 4. inode(x) | CP | dnode(F) | inode(F)
33 * 5. CP | inode(x) | dnode(F)
34 * -> The inode(DF) was missing. Should drop this dnode(F).
36 * 6. CP | inode(DF) | dnode(F)
39 * 7. CP | dnode(F) | inode(DF)
40 * -> If f2fs_iget fails, then goto next to find inode(DF).
42 * 8. CP | dnode(F) | inode(x)
43 * -> If f2fs_iget fails, then goto next to find inode(DF).
44 * But it will fail due to no inode(DF).
47 static struct kmem_cache *fsync_entry_slab;
50 extern struct kmem_cache *f2fs_cf_name_slab;
53 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
55 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
57 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
62 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
65 struct fsync_inode_entry *entry;
67 list_for_each_entry(entry, head, list)
68 if (entry->inode->i_ino == ino)
74 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
75 struct list_head *head, nid_t ino, bool quota_inode)
78 struct fsync_inode_entry *entry;
81 inode = f2fs_iget_retry(sbi->sb, ino);
83 return ERR_CAST(inode);
85 err = f2fs_dquot_initialize(inode);
90 err = dquot_alloc_inode(inode);
95 entry = f2fs_kmem_cache_alloc(fsync_entry_slab,
96 GFP_F2FS_ZERO, true, NULL);
98 list_add_tail(&entry->list, head);
106 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
109 /* inode should not be recovered, drop it */
110 f2fs_inode_synced(entry->inode);
113 list_del(&entry->list);
114 kmem_cache_free(fsync_entry_slab, entry);
117 static int init_recovered_filename(const struct inode *dir,
118 struct f2fs_inode *raw_inode,
119 struct f2fs_filename *fname,
120 struct qstr *usr_fname)
124 memset(fname, 0, sizeof(*fname));
125 fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
126 fname->disk_name.name = raw_inode->i_name;
128 if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
129 return -ENAMETOOLONG;
131 if (!IS_ENCRYPTED(dir)) {
132 usr_fname->name = fname->disk_name.name;
133 usr_fname->len = fname->disk_name.len;
134 fname->usr_fname = usr_fname;
137 /* Compute the hash of the filename */
138 if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) {
140 * In this case the hash isn't computable without the key, so it
143 if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN)
145 fname->hash = get_unaligned((f2fs_hash_t *)
146 &raw_inode->i_name[fname->disk_name.len]);
147 } else if (IS_CASEFOLDED(dir)) {
148 err = f2fs_init_casefolded_name(dir, fname);
151 f2fs_hash_filename(dir, fname);
152 #ifdef CONFIG_UNICODE
153 /* Case-sensitive match is fine for recovery */
154 kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
155 fname->cf_name.name = NULL;
158 f2fs_hash_filename(dir, fname);
163 static int recover_dentry(struct inode *inode, struct page *ipage,
164 struct list_head *dir_list)
166 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
167 nid_t pino = le32_to_cpu(raw_inode->i_pino);
168 struct f2fs_dir_entry *de;
169 struct f2fs_filename fname;
170 struct qstr usr_fname;
172 struct inode *dir, *einode;
173 struct fsync_inode_entry *entry;
177 entry = get_fsync_inode(dir_list, pino);
179 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
182 dir = ERR_CAST(entry);
183 err = PTR_ERR(entry);
189 err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
193 de = __f2fs_find_entry(dir, &fname, &page);
194 if (de && inode->i_ino == le32_to_cpu(de->ino))
198 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
199 if (IS_ERR(einode)) {
201 err = PTR_ERR(einode);
207 err = f2fs_dquot_initialize(einode);
213 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
218 f2fs_delete_entry(de, page, dir, einode);
221 } else if (IS_ERR(page)) {
224 err = f2fs_add_dentry(dir, &fname, inode,
225 inode->i_ino, inode->i_mode);
232 f2fs_put_page(page, 0);
234 if (file_enc_name(inode))
235 name = "<encrypted>";
237 name = raw_inode->i_name;
238 f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
239 __func__, ino_of_node(ipage), name,
240 IS_ERR(dir) ? 0 : dir->i_ino, err);
244 static int recover_quota_data(struct inode *inode, struct page *page)
246 struct f2fs_inode *raw = F2FS_INODE(page);
248 uid_t i_uid = le32_to_cpu(raw->i_uid);
249 gid_t i_gid = le32_to_cpu(raw->i_gid);
252 memset(&attr, 0, sizeof(attr));
254 attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
255 attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
257 if (!uid_eq(attr.ia_uid, inode->i_uid))
258 attr.ia_valid |= ATTR_UID;
259 if (!gid_eq(attr.ia_gid, inode->i_gid))
260 attr.ia_valid |= ATTR_GID;
265 err = dquot_transfer(inode, &attr);
267 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
271 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
273 if (ri->i_inline & F2FS_PIN_FILE)
274 set_inode_flag(inode, FI_PIN_FILE);
276 clear_inode_flag(inode, FI_PIN_FILE);
277 if (ri->i_inline & F2FS_DATA_EXIST)
278 set_inode_flag(inode, FI_DATA_EXIST);
280 clear_inode_flag(inode, FI_DATA_EXIST);
283 static int recover_inode(struct inode *inode, struct page *page)
285 struct f2fs_inode *raw = F2FS_INODE(page);
289 inode->i_mode = le16_to_cpu(raw->i_mode);
291 err = recover_quota_data(inode, page);
295 i_uid_write(inode, le32_to_cpu(raw->i_uid));
296 i_gid_write(inode, le32_to_cpu(raw->i_gid));
298 if (raw->i_inline & F2FS_EXTRA_ATTR) {
299 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
300 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
305 i_projid = (projid_t)le32_to_cpu(raw->i_projid);
306 kprojid = make_kprojid(&init_user_ns, i_projid);
308 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
309 err = f2fs_transfer_project_quota(inode,
313 F2FS_I(inode)->i_projid = kprojid;
318 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
319 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
320 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
321 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
322 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
323 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
324 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
326 F2FS_I(inode)->i_advise = raw->i_advise;
327 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
328 f2fs_set_inode_flags(inode);
329 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
330 le16_to_cpu(raw->i_gc_failures);
332 recover_inline_flags(inode, raw);
334 f2fs_mark_inode_dirty_sync(inode, true);
336 if (file_enc_name(inode))
337 name = "<encrypted>";
339 name = F2FS_INODE(page)->i_name;
341 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
342 ino_of_node(page), name, raw->i_inline);
346 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
349 struct curseg_info *curseg;
350 struct page *page = NULL;
352 unsigned int loop_cnt = 0;
353 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
354 valid_user_blocks(sbi);
357 /* get node pages in the current segment */
358 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
359 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
362 struct fsync_inode_entry *entry;
364 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
367 page = f2fs_get_tmp_page(sbi, blkaddr);
373 if (!is_recoverable_dnode(page)) {
374 f2fs_put_page(page, 1);
378 if (!is_fsync_dnode(page))
381 entry = get_fsync_inode(head, ino_of_node(page));
383 bool quota_inode = false;
386 IS_INODE(page) && is_dent_dnode(page)) {
387 err = f2fs_recover_inode_page(sbi, page);
389 f2fs_put_page(page, 1);
396 * CP | dnode(F) | inode(DF)
397 * For this case, we should not give up now.
399 entry = add_fsync_inode(sbi, head, ino_of_node(page),
402 err = PTR_ERR(entry);
403 if (err == -ENOENT) {
407 f2fs_put_page(page, 1);
411 entry->blkaddr = blkaddr;
413 if (IS_INODE(page) && is_dent_dnode(page))
414 entry->last_dentry = blkaddr;
416 /* sanity check in order to detect looped node chain */
417 if (++loop_cnt >= free_blocks ||
418 blkaddr == next_blkaddr_of_node(page)) {
419 f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
421 next_blkaddr_of_node(page));
422 f2fs_put_page(page, 1);
427 /* check next segment */
428 blkaddr = next_blkaddr_of_node(page);
429 f2fs_put_page(page, 1);
431 f2fs_ra_meta_pages_cond(sbi, blkaddr);
436 static void destroy_fsync_dnodes(struct list_head *head, int drop)
438 struct fsync_inode_entry *entry, *tmp;
440 list_for_each_entry_safe(entry, tmp, head, list)
441 del_fsync_inode(entry, drop);
444 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
445 block_t blkaddr, struct dnode_of_data *dn)
447 struct seg_entry *sentry;
448 unsigned int segno = GET_SEGNO(sbi, blkaddr);
449 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
450 struct f2fs_summary_block *sum_node;
451 struct f2fs_summary sum;
452 struct page *sum_page, *node_page;
453 struct dnode_of_data tdn = *dn;
460 sentry = get_seg_entry(sbi, segno);
461 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
464 /* Get the previous summary */
465 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
466 struct curseg_info *curseg = CURSEG_I(sbi, i);
468 if (curseg->segno == segno) {
469 sum = curseg->sum_blk->entries[blkoff];
474 sum_page = f2fs_get_sum_page(sbi, segno);
475 if (IS_ERR(sum_page))
476 return PTR_ERR(sum_page);
477 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
478 sum = sum_node->entries[blkoff];
479 f2fs_put_page(sum_page, 1);
481 /* Use the locked dnode page and inode */
482 nid = le32_to_cpu(sum.nid);
483 if (dn->inode->i_ino == nid) {
485 if (!dn->inode_page_locked)
486 lock_page(dn->inode_page);
487 tdn.node_page = dn->inode_page;
488 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
490 } else if (dn->nid == nid) {
491 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
495 /* Get the node page */
496 node_page = f2fs_get_node_page(sbi, nid);
497 if (IS_ERR(node_page))
498 return PTR_ERR(node_page);
500 offset = ofs_of_node(node_page);
501 ino = ino_of_node(node_page);
502 f2fs_put_page(node_page, 1);
504 if (ino != dn->inode->i_ino) {
507 /* Deallocate previous index in the node page */
508 inode = f2fs_iget_retry(sbi->sb, ino);
510 return PTR_ERR(inode);
512 ret = f2fs_dquot_initialize(inode);
521 bidx = f2fs_start_bidx_of_node(offset, inode) +
522 le16_to_cpu(sum.ofs_in_node);
525 * if inode page is locked, unlock temporarily, but its reference
528 if (ino == dn->inode->i_ino && dn->inode_page_locked)
529 unlock_page(dn->inode_page);
531 set_new_dnode(&tdn, inode, NULL, NULL, 0);
532 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
535 if (tdn.data_blkaddr == blkaddr)
536 f2fs_truncate_data_blocks_range(&tdn, 1);
538 f2fs_put_dnode(&tdn);
540 if (ino != dn->inode->i_ino)
542 else if (dn->inode_page_locked)
543 lock_page(dn->inode_page);
547 if (f2fs_data_blkaddr(&tdn) == blkaddr)
548 f2fs_truncate_data_blocks_range(&tdn, 1);
549 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
550 unlock_page(dn->inode_page);
554 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
557 struct dnode_of_data dn;
559 unsigned int start, end;
560 int err = 0, recovered = 0;
562 /* step 1: recover xattr */
563 if (IS_INODE(page)) {
564 err = f2fs_recover_inline_xattr(inode, page);
567 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
568 err = f2fs_recover_xattr_data(inode, page);
574 /* step 2: recover inline data */
575 err = f2fs_recover_inline_data(inode, page);
582 /* step 3: recover data indices */
583 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
584 end = start + ADDRS_PER_PAGE(page, inode);
586 set_new_dnode(&dn, inode, NULL, NULL, 0);
588 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
590 if (err == -ENOMEM) {
591 memalloc_retry_wait(GFP_NOFS);
597 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
599 err = f2fs_get_node_info(sbi, dn.nid, &ni);
603 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
605 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
606 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
607 inode->i_ino, ofs_of_node(dn.node_page),
613 for (; start < end; start++, dn.ofs_in_node++) {
616 src = f2fs_data_blkaddr(&dn);
617 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
619 if (__is_valid_data_blkaddr(src) &&
620 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
625 if (__is_valid_data_blkaddr(dest) &&
626 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
631 /* skip recovering if dest is the same as src */
635 /* dest is invalid, just invalidate src block */
636 if (dest == NULL_ADDR) {
637 f2fs_truncate_data_blocks_range(&dn, 1);
641 if (!file_keep_isize(inode) &&
642 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
643 f2fs_i_size_write(inode,
644 (loff_t)(start + 1) << PAGE_SHIFT);
647 * dest is reserved block, invalidate src block
648 * and then reserve one new block in dnode page.
650 if (dest == NEW_ADDR) {
651 f2fs_truncate_data_blocks_range(&dn, 1);
652 f2fs_reserve_new_block(&dn);
656 /* dest is valid block, try to recover from src to dest */
657 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
659 if (src == NULL_ADDR) {
660 err = f2fs_reserve_new_block(&dn);
662 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
663 err = f2fs_reserve_new_block(&dn);
664 /* We should not get -ENOSPC */
665 f2fs_bug_on(sbi, err);
670 /* Check the previous node page having this index */
671 err = check_index_in_prev_nodes(sbi, dest, &dn);
673 if (err == -ENOMEM) {
674 memalloc_retry_wait(GFP_NOFS);
680 /* write dummy data page */
681 f2fs_replace_block(sbi, &dn, src, dest,
682 ni.version, false, false);
687 copy_node_footer(dn.node_page, page);
688 fill_node_footer(dn.node_page, dn.nid, ni.ino,
689 ofs_of_node(page), false);
690 set_page_dirty(dn.node_page);
694 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
695 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
700 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
701 struct list_head *tmp_inode_list, struct list_head *dir_list)
703 struct curseg_info *curseg;
704 struct page *page = NULL;
708 /* get node pages in the current segment */
709 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
710 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
713 struct fsync_inode_entry *entry;
715 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
718 f2fs_ra_meta_pages_cond(sbi, blkaddr);
720 page = f2fs_get_tmp_page(sbi, blkaddr);
726 if (!is_recoverable_dnode(page)) {
727 f2fs_put_page(page, 1);
731 entry = get_fsync_inode(inode_list, ino_of_node(page));
735 * inode(x) | CP | inode(x) | dnode(F)
736 * In this case, we can lose the latest inode(x).
737 * So, call recover_inode for the inode update.
739 if (IS_INODE(page)) {
740 err = recover_inode(entry->inode, page);
742 f2fs_put_page(page, 1);
746 if (entry->last_dentry == blkaddr) {
747 err = recover_dentry(entry->inode, page, dir_list);
749 f2fs_put_page(page, 1);
753 err = do_recover_data(sbi, entry->inode, page);
755 f2fs_put_page(page, 1);
759 if (entry->blkaddr == blkaddr)
760 list_move_tail(&entry->list, tmp_inode_list);
762 /* check next segment */
763 blkaddr = next_blkaddr_of_node(page);
764 f2fs_put_page(page, 1);
767 f2fs_allocate_new_segments(sbi);
771 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
773 struct list_head inode_list, tmp_inode_list;
774 struct list_head dir_list;
777 unsigned long s_flags = sbi->sb->s_flags;
778 bool need_writecp = false;
779 bool fix_curseg_write_pointer = false;
784 if (s_flags & SB_RDONLY) {
785 f2fs_info(sbi, "recover fsync data on readonly fs");
786 sbi->sb->s_flags &= ~SB_RDONLY;
790 /* Turn on quotas so that they are updated correctly */
791 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
794 INIT_LIST_HEAD(&inode_list);
795 INIT_LIST_HEAD(&tmp_inode_list);
796 INIT_LIST_HEAD(&dir_list);
798 /* prevent checkpoint */
799 down_write(&sbi->cp_global_sem);
801 /* step #1: find fsynced inode numbers */
802 err = find_fsync_dnodes(sbi, &inode_list, check_only);
803 if (err || list_empty(&inode_list))
813 /* step #2: recover data */
814 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
816 f2fs_bug_on(sbi, !list_empty(&inode_list));
818 f2fs_bug_on(sbi, sbi->sb->s_flags & SB_ACTIVE);
820 fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
822 destroy_fsync_dnodes(&inode_list, err);
823 destroy_fsync_dnodes(&tmp_inode_list, err);
825 /* truncate meta pages to be used by the recovery */
826 truncate_inode_pages_range(META_MAPPING(sbi),
827 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
830 truncate_inode_pages_final(NODE_MAPPING(sbi));
831 truncate_inode_pages_final(META_MAPPING(sbi));
835 * If fsync data succeeds or there is no fsync data to recover,
836 * and the f2fs is not read only, check and fix zoned block devices'
837 * write pointer consistency.
839 if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
840 f2fs_sb_has_blkzoned(sbi)) {
841 err = f2fs_fix_curseg_write_pointer(sbi);
846 clear_sbi_flag(sbi, SBI_POR_DOING);
848 up_write(&sbi->cp_global_sem);
850 /* let's drop all the directory inodes for clean checkpoint */
851 destroy_fsync_dnodes(&dir_list, err);
854 set_sbi_flag(sbi, SBI_IS_RECOVERED);
857 struct cp_control cpc = {
858 .reason = CP_RECOVERY,
860 err = f2fs_write_checkpoint(sbi, &cpc);
865 /* Turn quotas off */
867 f2fs_quota_off_umount(sbi->sb);
869 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
871 return ret ? ret : err;
874 int __init f2fs_create_recovery_cache(void)
876 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
877 sizeof(struct fsync_inode_entry));
878 if (!fsync_entry_slab)
883 void f2fs_destroy_recovery_cache(void)
885 kmem_cache_destroy(fsync_entry_slab);