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>
16 * Roll forward recovery scenarios.
18 * [Term] F: fsync_mark, D: dentry_mark
20 * 1. inode(x) | CP | inode(x) | dnode(F)
21 * -> Update the latest inode(x).
23 * 2. inode(x) | CP | inode(F) | dnode(F)
26 * 3. inode(x) | CP | dnode(F) | inode(x)
27 * -> Recover to the latest dnode(F), and drop the last inode(x)
29 * 4. inode(x) | CP | dnode(F) | inode(F)
32 * 5. CP | inode(x) | dnode(F)
33 * -> The inode(DF) was missing. Should drop this dnode(F).
35 * 6. CP | inode(DF) | dnode(F)
38 * 7. CP | dnode(F) | inode(DF)
39 * -> If f2fs_iget fails, then goto next to find inode(DF).
41 * 8. CP | dnode(F) | inode(x)
42 * -> If f2fs_iget fails, then goto next to find inode(DF).
43 * But it will fail due to no inode(DF).
46 static struct kmem_cache *fsync_entry_slab;
48 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
50 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
52 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
57 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
60 struct fsync_inode_entry *entry;
62 list_for_each_entry(entry, head, list)
63 if (entry->inode->i_ino == ino)
69 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
70 struct list_head *head, nid_t ino, bool quota_inode)
73 struct fsync_inode_entry *entry;
76 inode = f2fs_iget_retry(sbi->sb, ino);
78 return ERR_CAST(inode);
80 err = dquot_initialize(inode);
85 err = dquot_alloc_inode(inode);
90 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
92 list_add_tail(&entry->list, head);
100 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
103 /* inode should not be recovered, drop it */
104 f2fs_inode_synced(entry->inode);
107 list_del(&entry->list);
108 kmem_cache_free(fsync_entry_slab, entry);
111 static int init_recovered_filename(const struct inode *dir,
112 struct f2fs_inode *raw_inode,
113 struct f2fs_filename *fname,
114 struct qstr *usr_fname)
118 memset(fname, 0, sizeof(*fname));
119 fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
120 fname->disk_name.name = raw_inode->i_name;
122 if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
123 return -ENAMETOOLONG;
125 if (!IS_ENCRYPTED(dir)) {
126 usr_fname->name = fname->disk_name.name;
127 usr_fname->len = fname->disk_name.len;
128 fname->usr_fname = usr_fname;
131 /* Compute the hash of the filename */
132 if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) {
134 * In this case the hash isn't computable without the key, so it
137 if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN)
139 fname->hash = get_unaligned((f2fs_hash_t *)
140 &raw_inode->i_name[fname->disk_name.len]);
141 } else if (IS_CASEFOLDED(dir)) {
142 err = f2fs_init_casefolded_name(dir, fname);
145 f2fs_hash_filename(dir, fname);
146 #ifdef CONFIG_UNICODE
147 /* Case-sensitive match is fine for recovery */
148 kfree(fname->cf_name.name);
149 fname->cf_name.name = NULL;
152 f2fs_hash_filename(dir, fname);
157 static int recover_dentry(struct inode *inode, struct page *ipage,
158 struct list_head *dir_list)
160 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
161 nid_t pino = le32_to_cpu(raw_inode->i_pino);
162 struct f2fs_dir_entry *de;
163 struct f2fs_filename fname;
164 struct qstr usr_fname;
166 struct inode *dir, *einode;
167 struct fsync_inode_entry *entry;
171 entry = get_fsync_inode(dir_list, pino);
173 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
176 dir = ERR_CAST(entry);
177 err = PTR_ERR(entry);
183 err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
187 de = __f2fs_find_entry(dir, &fname, &page);
188 if (de && inode->i_ino == le32_to_cpu(de->ino))
192 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
193 if (IS_ERR(einode)) {
195 err = PTR_ERR(einode);
201 err = dquot_initialize(einode);
207 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
212 f2fs_delete_entry(de, page, dir, einode);
215 } else if (IS_ERR(page)) {
218 err = f2fs_add_dentry(dir, &fname, inode,
219 inode->i_ino, inode->i_mode);
226 f2fs_put_page(page, 0);
228 if (file_enc_name(inode))
229 name = "<encrypted>";
231 name = raw_inode->i_name;
232 f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
233 __func__, ino_of_node(ipage), name,
234 IS_ERR(dir) ? 0 : dir->i_ino, err);
238 static int recover_quota_data(struct inode *inode, struct page *page)
240 struct f2fs_inode *raw = F2FS_INODE(page);
242 uid_t i_uid = le32_to_cpu(raw->i_uid);
243 gid_t i_gid = le32_to_cpu(raw->i_gid);
246 memset(&attr, 0, sizeof(attr));
248 attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
249 attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
251 if (!uid_eq(attr.ia_uid, inode->i_uid))
252 attr.ia_valid |= ATTR_UID;
253 if (!gid_eq(attr.ia_gid, inode->i_gid))
254 attr.ia_valid |= ATTR_GID;
259 err = dquot_transfer(inode, &attr);
261 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
265 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
267 if (ri->i_inline & F2FS_PIN_FILE)
268 set_inode_flag(inode, FI_PIN_FILE);
270 clear_inode_flag(inode, FI_PIN_FILE);
271 if (ri->i_inline & F2FS_DATA_EXIST)
272 set_inode_flag(inode, FI_DATA_EXIST);
274 clear_inode_flag(inode, FI_DATA_EXIST);
277 static int recover_inode(struct inode *inode, struct page *page)
279 struct f2fs_inode *raw = F2FS_INODE(page);
283 inode->i_mode = le16_to_cpu(raw->i_mode);
285 err = recover_quota_data(inode, page);
289 i_uid_write(inode, le32_to_cpu(raw->i_uid));
290 i_gid_write(inode, le32_to_cpu(raw->i_gid));
292 if (raw->i_inline & F2FS_EXTRA_ATTR) {
293 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
294 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
299 i_projid = (projid_t)le32_to_cpu(raw->i_projid);
300 kprojid = make_kprojid(&init_user_ns, i_projid);
302 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
303 err = f2fs_transfer_project_quota(inode,
307 F2FS_I(inode)->i_projid = kprojid;
312 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
313 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
314 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
315 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
316 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
317 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
318 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
320 F2FS_I(inode)->i_advise = raw->i_advise;
321 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
322 f2fs_set_inode_flags(inode);
323 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
324 le16_to_cpu(raw->i_gc_failures);
326 recover_inline_flags(inode, raw);
328 f2fs_mark_inode_dirty_sync(inode, true);
330 if (file_enc_name(inode))
331 name = "<encrypted>";
333 name = F2FS_INODE(page)->i_name;
335 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
336 ino_of_node(page), name, raw->i_inline);
340 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
343 struct curseg_info *curseg;
344 struct page *page = NULL;
346 unsigned int loop_cnt = 0;
347 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
348 valid_user_blocks(sbi);
351 /* get node pages in the current segment */
352 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
353 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
356 struct fsync_inode_entry *entry;
358 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
361 page = f2fs_get_tmp_page(sbi, blkaddr);
367 if (!is_recoverable_dnode(page)) {
368 f2fs_put_page(page, 1);
372 if (!is_fsync_dnode(page))
375 entry = get_fsync_inode(head, ino_of_node(page));
377 bool quota_inode = false;
380 IS_INODE(page) && is_dent_dnode(page)) {
381 err = f2fs_recover_inode_page(sbi, page);
383 f2fs_put_page(page, 1);
390 * CP | dnode(F) | inode(DF)
391 * For this case, we should not give up now.
393 entry = add_fsync_inode(sbi, head, ino_of_node(page),
396 err = PTR_ERR(entry);
397 if (err == -ENOENT) {
401 f2fs_put_page(page, 1);
405 entry->blkaddr = blkaddr;
407 if (IS_INODE(page) && is_dent_dnode(page))
408 entry->last_dentry = blkaddr;
410 /* sanity check in order to detect looped node chain */
411 if (++loop_cnt >= free_blocks ||
412 blkaddr == next_blkaddr_of_node(page)) {
413 f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
415 next_blkaddr_of_node(page));
416 f2fs_put_page(page, 1);
421 /* check next segment */
422 blkaddr = next_blkaddr_of_node(page);
423 f2fs_put_page(page, 1);
425 f2fs_ra_meta_pages_cond(sbi, blkaddr);
430 static void destroy_fsync_dnodes(struct list_head *head, int drop)
432 struct fsync_inode_entry *entry, *tmp;
434 list_for_each_entry_safe(entry, tmp, head, list)
435 del_fsync_inode(entry, drop);
438 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
439 block_t blkaddr, struct dnode_of_data *dn)
441 struct seg_entry *sentry;
442 unsigned int segno = GET_SEGNO(sbi, blkaddr);
443 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
444 struct f2fs_summary_block *sum_node;
445 struct f2fs_summary sum;
446 struct page *sum_page, *node_page;
447 struct dnode_of_data tdn = *dn;
454 sentry = get_seg_entry(sbi, segno);
455 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
458 /* Get the previous summary */
459 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
460 struct curseg_info *curseg = CURSEG_I(sbi, i);
462 if (curseg->segno == segno) {
463 sum = curseg->sum_blk->entries[blkoff];
468 sum_page = f2fs_get_sum_page(sbi, segno);
469 if (IS_ERR(sum_page))
470 return PTR_ERR(sum_page);
471 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
472 sum = sum_node->entries[blkoff];
473 f2fs_put_page(sum_page, 1);
475 /* Use the locked dnode page and inode */
476 nid = le32_to_cpu(sum.nid);
477 if (dn->inode->i_ino == nid) {
479 if (!dn->inode_page_locked)
480 lock_page(dn->inode_page);
481 tdn.node_page = dn->inode_page;
482 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
484 } else if (dn->nid == nid) {
485 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
489 /* Get the node page */
490 node_page = f2fs_get_node_page(sbi, nid);
491 if (IS_ERR(node_page))
492 return PTR_ERR(node_page);
494 offset = ofs_of_node(node_page);
495 ino = ino_of_node(node_page);
496 f2fs_put_page(node_page, 1);
498 if (ino != dn->inode->i_ino) {
501 /* Deallocate previous index in the node page */
502 inode = f2fs_iget_retry(sbi->sb, ino);
504 return PTR_ERR(inode);
506 ret = dquot_initialize(inode);
515 bidx = f2fs_start_bidx_of_node(offset, inode) +
516 le16_to_cpu(sum.ofs_in_node);
519 * if inode page is locked, unlock temporarily, but its reference
522 if (ino == dn->inode->i_ino && dn->inode_page_locked)
523 unlock_page(dn->inode_page);
525 set_new_dnode(&tdn, inode, NULL, NULL, 0);
526 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
529 if (tdn.data_blkaddr == blkaddr)
530 f2fs_truncate_data_blocks_range(&tdn, 1);
532 f2fs_put_dnode(&tdn);
534 if (ino != dn->inode->i_ino)
536 else if (dn->inode_page_locked)
537 lock_page(dn->inode_page);
541 if (f2fs_data_blkaddr(&tdn) == blkaddr)
542 f2fs_truncate_data_blocks_range(&tdn, 1);
543 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
544 unlock_page(dn->inode_page);
548 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
551 struct dnode_of_data dn;
553 unsigned int start, end;
554 int err = 0, recovered = 0;
556 /* step 1: recover xattr */
557 if (IS_INODE(page)) {
558 err = f2fs_recover_inline_xattr(inode, page);
561 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
562 err = f2fs_recover_xattr_data(inode, page);
568 /* step 2: recover inline data */
569 err = f2fs_recover_inline_data(inode, page);
576 /* step 3: recover data indices */
577 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
578 end = start + ADDRS_PER_PAGE(page, inode);
580 set_new_dnode(&dn, inode, NULL, NULL, 0);
582 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
584 if (err == -ENOMEM) {
585 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
591 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
593 err = f2fs_get_node_info(sbi, dn.nid, &ni);
597 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
599 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
600 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
601 inode->i_ino, ofs_of_node(dn.node_page),
607 for (; start < end; start++, dn.ofs_in_node++) {
610 src = f2fs_data_blkaddr(&dn);
611 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
613 if (__is_valid_data_blkaddr(src) &&
614 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
619 if (__is_valid_data_blkaddr(dest) &&
620 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
625 /* skip recovering if dest is the same as src */
629 /* dest is invalid, just invalidate src block */
630 if (dest == NULL_ADDR) {
631 f2fs_truncate_data_blocks_range(&dn, 1);
635 if (!file_keep_isize(inode) &&
636 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
637 f2fs_i_size_write(inode,
638 (loff_t)(start + 1) << PAGE_SHIFT);
641 * dest is reserved block, invalidate src block
642 * and then reserve one new block in dnode page.
644 if (dest == NEW_ADDR) {
645 f2fs_truncate_data_blocks_range(&dn, 1);
646 f2fs_reserve_new_block(&dn);
650 /* dest is valid block, try to recover from src to dest */
651 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
653 if (src == NULL_ADDR) {
654 err = f2fs_reserve_new_block(&dn);
656 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
657 err = f2fs_reserve_new_block(&dn);
658 /* We should not get -ENOSPC */
659 f2fs_bug_on(sbi, err);
664 /* Check the previous node page having this index */
665 err = check_index_in_prev_nodes(sbi, dest, &dn);
667 if (err == -ENOMEM) {
668 congestion_wait(BLK_RW_ASYNC,
675 /* write dummy data page */
676 f2fs_replace_block(sbi, &dn, src, dest,
677 ni.version, false, false);
682 copy_node_footer(dn.node_page, page);
683 fill_node_footer(dn.node_page, dn.nid, ni.ino,
684 ofs_of_node(page), false);
685 set_page_dirty(dn.node_page);
689 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
690 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
695 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
696 struct list_head *tmp_inode_list, struct list_head *dir_list)
698 struct curseg_info *curseg;
699 struct page *page = NULL;
703 /* get node pages in the current segment */
704 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
705 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
708 struct fsync_inode_entry *entry;
710 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
713 f2fs_ra_meta_pages_cond(sbi, blkaddr);
715 page = f2fs_get_tmp_page(sbi, blkaddr);
721 if (!is_recoverable_dnode(page)) {
722 f2fs_put_page(page, 1);
726 entry = get_fsync_inode(inode_list, ino_of_node(page));
730 * inode(x) | CP | inode(x) | dnode(F)
731 * In this case, we can lose the latest inode(x).
732 * So, call recover_inode for the inode update.
734 if (IS_INODE(page)) {
735 err = recover_inode(entry->inode, page);
737 f2fs_put_page(page, 1);
741 if (entry->last_dentry == blkaddr) {
742 err = recover_dentry(entry->inode, page, dir_list);
744 f2fs_put_page(page, 1);
748 err = do_recover_data(sbi, entry->inode, page);
750 f2fs_put_page(page, 1);
754 if (entry->blkaddr == blkaddr)
755 list_move_tail(&entry->list, tmp_inode_list);
757 /* check next segment */
758 blkaddr = next_blkaddr_of_node(page);
759 f2fs_put_page(page, 1);
762 f2fs_allocate_new_segments(sbi);
766 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
768 struct list_head inode_list, tmp_inode_list;
769 struct list_head dir_list;
772 unsigned long s_flags = sbi->sb->s_flags;
773 bool need_writecp = false;
774 bool fix_curseg_write_pointer = false;
779 if (s_flags & SB_RDONLY) {
780 f2fs_info(sbi, "recover fsync data on readonly fs");
781 sbi->sb->s_flags &= ~SB_RDONLY;
785 /* Needed for iput() to work correctly and not trash data */
786 sbi->sb->s_flags |= SB_ACTIVE;
787 /* Turn on quotas so that they are updated correctly */
788 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
791 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
792 sizeof(struct fsync_inode_entry));
793 if (!fsync_entry_slab) {
798 INIT_LIST_HEAD(&inode_list);
799 INIT_LIST_HEAD(&tmp_inode_list);
800 INIT_LIST_HEAD(&dir_list);
802 /* prevent checkpoint */
803 down_write(&sbi->cp_global_sem);
805 /* step #1: find fsynced inode numbers */
806 err = find_fsync_dnodes(sbi, &inode_list, check_only);
807 if (err || list_empty(&inode_list))
817 /* step #2: recover data */
818 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
820 f2fs_bug_on(sbi, !list_empty(&inode_list));
822 /* restore s_flags to let iput() trash data */
823 sbi->sb->s_flags = s_flags;
826 fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
828 destroy_fsync_dnodes(&inode_list, err);
829 destroy_fsync_dnodes(&tmp_inode_list, err);
831 /* truncate meta pages to be used by the recovery */
832 truncate_inode_pages_range(META_MAPPING(sbi),
833 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
836 truncate_inode_pages_final(NODE_MAPPING(sbi));
837 truncate_inode_pages_final(META_MAPPING(sbi));
841 * If fsync data succeeds or there is no fsync data to recover,
842 * and the f2fs is not read only, check and fix zoned block devices'
843 * write pointer consistency.
845 if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
846 f2fs_sb_has_blkzoned(sbi)) {
847 err = f2fs_fix_curseg_write_pointer(sbi);
852 clear_sbi_flag(sbi, SBI_POR_DOING);
854 up_write(&sbi->cp_global_sem);
856 /* let's drop all the directory inodes for clean checkpoint */
857 destroy_fsync_dnodes(&dir_list, err);
860 set_sbi_flag(sbi, SBI_IS_RECOVERED);
863 struct cp_control cpc = {
864 .reason = CP_RECOVERY,
866 err = f2fs_write_checkpoint(sbi, &cpc);
870 kmem_cache_destroy(fsync_entry_slab);
873 /* Turn quotas off */
875 f2fs_quota_off_umount(sbi->sb);
877 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
879 return ret ? ret : err;