1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
15 * Roll forward recovery scenarios.
17 * [Term] F: fsync_mark, D: dentry_mark
19 * 1. inode(x) | CP | inode(x) | dnode(F)
20 * -> Update the latest inode(x).
22 * 2. inode(x) | CP | inode(F) | dnode(F)
25 * 3. inode(x) | CP | dnode(F) | inode(x)
26 * -> Recover to the latest dnode(F), and drop the last inode(x)
28 * 4. inode(x) | CP | dnode(F) | inode(F)
31 * 5. CP | inode(x) | dnode(F)
32 * -> The inode(DF) was missing. Should drop this dnode(F).
34 * 6. CP | inode(DF) | dnode(F)
37 * 7. CP | dnode(F) | inode(DF)
38 * -> If f2fs_iget fails, then goto next to find inode(DF).
40 * 8. CP | dnode(F) | inode(x)
41 * -> If f2fs_iget fails, then goto next to find inode(DF).
42 * But it will fail due to no inode(DF).
45 static struct kmem_cache *fsync_entry_slab;
47 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
49 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
51 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
56 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
59 struct fsync_inode_entry *entry;
61 list_for_each_entry(entry, head, list)
62 if (entry->inode->i_ino == ino)
68 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
69 struct list_head *head, nid_t ino, bool quota_inode)
72 struct fsync_inode_entry *entry;
75 inode = f2fs_iget_retry(sbi->sb, ino);
77 return ERR_CAST(inode);
79 err = dquot_initialize(inode);
84 err = dquot_alloc_inode(inode);
89 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
91 list_add_tail(&entry->list, head);
99 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
102 /* inode should not be recovered, drop it */
103 f2fs_inode_synced(entry->inode);
106 list_del(&entry->list);
107 kmem_cache_free(fsync_entry_slab, entry);
110 static int recover_dentry(struct inode *inode, struct page *ipage,
111 struct list_head *dir_list)
113 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
114 nid_t pino = le32_to_cpu(raw_inode->i_pino);
115 struct f2fs_dir_entry *de;
116 struct fscrypt_name fname;
118 struct inode *dir, *einode;
119 struct fsync_inode_entry *entry;
123 entry = get_fsync_inode(dir_list, pino);
125 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
128 dir = ERR_CAST(entry);
129 err = PTR_ERR(entry);
136 memset(&fname, 0, sizeof(struct fscrypt_name));
137 fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen);
138 fname.disk_name.name = raw_inode->i_name;
140 if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
146 de = __f2fs_find_entry(dir, &fname, &page);
147 if (de && inode->i_ino == le32_to_cpu(de->ino))
151 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
152 if (IS_ERR(einode)) {
154 err = PTR_ERR(einode);
160 err = dquot_initialize(einode);
166 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
171 f2fs_delete_entry(de, page, dir, einode);
174 } else if (IS_ERR(page)) {
177 err = f2fs_add_dentry(dir, &fname, inode,
178 inode->i_ino, inode->i_mode);
185 f2fs_put_page(page, 0);
187 if (file_enc_name(inode))
188 name = "<encrypted>";
190 name = raw_inode->i_name;
191 f2fs_msg(inode->i_sb, KERN_NOTICE,
192 "%s: ino = %x, name = %s, dir = %lx, err = %d",
193 __func__, ino_of_node(ipage), name,
194 IS_ERR(dir) ? 0 : dir->i_ino, err);
198 static int recover_quota_data(struct inode *inode, struct page *page)
200 struct f2fs_inode *raw = F2FS_INODE(page);
202 uid_t i_uid = le32_to_cpu(raw->i_uid);
203 gid_t i_gid = le32_to_cpu(raw->i_gid);
206 memset(&attr, 0, sizeof(attr));
208 attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
209 attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
211 if (!uid_eq(attr.ia_uid, inode->i_uid))
212 attr.ia_valid |= ATTR_UID;
213 if (!gid_eq(attr.ia_gid, inode->i_gid))
214 attr.ia_valid |= ATTR_GID;
219 err = dquot_transfer(inode, &attr);
221 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
225 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
227 if (ri->i_inline & F2FS_PIN_FILE)
228 set_inode_flag(inode, FI_PIN_FILE);
230 clear_inode_flag(inode, FI_PIN_FILE);
231 if (ri->i_inline & F2FS_DATA_EXIST)
232 set_inode_flag(inode, FI_DATA_EXIST);
234 clear_inode_flag(inode, FI_DATA_EXIST);
237 static int recover_inode(struct inode *inode, struct page *page)
239 struct f2fs_inode *raw = F2FS_INODE(page);
243 inode->i_mode = le16_to_cpu(raw->i_mode);
245 err = recover_quota_data(inode, page);
249 i_uid_write(inode, le32_to_cpu(raw->i_uid));
250 i_gid_write(inode, le32_to_cpu(raw->i_gid));
252 if (raw->i_inline & F2FS_EXTRA_ATTR) {
253 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
254 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
259 i_projid = (projid_t)le32_to_cpu(raw->i_projid);
260 kprojid = make_kprojid(&init_user_ns, i_projid);
262 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
263 err = f2fs_transfer_project_quota(inode,
267 F2FS_I(inode)->i_projid = kprojid;
272 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
273 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
274 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
275 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
276 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
277 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
278 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
280 F2FS_I(inode)->i_advise = raw->i_advise;
281 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
282 f2fs_set_inode_flags(inode);
283 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
284 le16_to_cpu(raw->i_gc_failures);
286 recover_inline_flags(inode, raw);
288 f2fs_mark_inode_dirty_sync(inode, true);
290 if (file_enc_name(inode))
291 name = "<encrypted>";
293 name = F2FS_INODE(page)->i_name;
295 f2fs_msg(inode->i_sb, KERN_NOTICE,
296 "recover_inode: ino = %x, name = %s, inline = %x",
297 ino_of_node(page), name, raw->i_inline);
301 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
304 struct curseg_info *curseg;
305 struct page *page = NULL;
307 unsigned int loop_cnt = 0;
308 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
309 valid_user_blocks(sbi);
312 /* get node pages in the current segment */
313 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
314 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
317 struct fsync_inode_entry *entry;
319 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
322 page = f2fs_get_tmp_page(sbi, blkaddr);
328 if (!is_recoverable_dnode(page)) {
329 f2fs_put_page(page, 1);
333 if (!is_fsync_dnode(page))
336 entry = get_fsync_inode(head, ino_of_node(page));
338 bool quota_inode = false;
341 IS_INODE(page) && is_dent_dnode(page)) {
342 err = f2fs_recover_inode_page(sbi, page);
344 f2fs_put_page(page, 1);
351 * CP | dnode(F) | inode(DF)
352 * For this case, we should not give up now.
354 entry = add_fsync_inode(sbi, head, ino_of_node(page),
357 err = PTR_ERR(entry);
358 if (err == -ENOENT) {
362 f2fs_put_page(page, 1);
366 entry->blkaddr = blkaddr;
368 if (IS_INODE(page) && is_dent_dnode(page))
369 entry->last_dentry = blkaddr;
371 /* sanity check in order to detect looped node chain */
372 if (++loop_cnt >= free_blocks ||
373 blkaddr == next_blkaddr_of_node(page)) {
374 f2fs_msg(sbi->sb, KERN_NOTICE,
375 "%s: detect looped node chain, "
376 "blkaddr:%u, next:%u",
377 __func__, blkaddr, next_blkaddr_of_node(page));
378 f2fs_put_page(page, 1);
383 /* check next segment */
384 blkaddr = next_blkaddr_of_node(page);
385 f2fs_put_page(page, 1);
387 f2fs_ra_meta_pages_cond(sbi, blkaddr);
392 static void destroy_fsync_dnodes(struct list_head *head, int drop)
394 struct fsync_inode_entry *entry, *tmp;
396 list_for_each_entry_safe(entry, tmp, head, list)
397 del_fsync_inode(entry, drop);
400 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
401 block_t blkaddr, struct dnode_of_data *dn)
403 struct seg_entry *sentry;
404 unsigned int segno = GET_SEGNO(sbi, blkaddr);
405 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
406 struct f2fs_summary_block *sum_node;
407 struct f2fs_summary sum;
408 struct page *sum_page, *node_page;
409 struct dnode_of_data tdn = *dn;
416 sentry = get_seg_entry(sbi, segno);
417 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
420 /* Get the previous summary */
421 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
422 struct curseg_info *curseg = CURSEG_I(sbi, i);
423 if (curseg->segno == segno) {
424 sum = curseg->sum_blk->entries[blkoff];
429 sum_page = f2fs_get_sum_page(sbi, segno);
430 if (IS_ERR(sum_page))
431 return PTR_ERR(sum_page);
432 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
433 sum = sum_node->entries[blkoff];
434 f2fs_put_page(sum_page, 1);
436 /* Use the locked dnode page and inode */
437 nid = le32_to_cpu(sum.nid);
438 if (dn->inode->i_ino == nid) {
440 if (!dn->inode_page_locked)
441 lock_page(dn->inode_page);
442 tdn.node_page = dn->inode_page;
443 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
445 } else if (dn->nid == nid) {
446 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
450 /* Get the node page */
451 node_page = f2fs_get_node_page(sbi, nid);
452 if (IS_ERR(node_page))
453 return PTR_ERR(node_page);
455 offset = ofs_of_node(node_page);
456 ino = ino_of_node(node_page);
457 f2fs_put_page(node_page, 1);
459 if (ino != dn->inode->i_ino) {
462 /* Deallocate previous index in the node page */
463 inode = f2fs_iget_retry(sbi->sb, ino);
465 return PTR_ERR(inode);
467 ret = dquot_initialize(inode);
476 bidx = f2fs_start_bidx_of_node(offset, inode) +
477 le16_to_cpu(sum.ofs_in_node);
480 * if inode page is locked, unlock temporarily, but its reference
483 if (ino == dn->inode->i_ino && dn->inode_page_locked)
484 unlock_page(dn->inode_page);
486 set_new_dnode(&tdn, inode, NULL, NULL, 0);
487 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
490 if (tdn.data_blkaddr == blkaddr)
491 f2fs_truncate_data_blocks_range(&tdn, 1);
493 f2fs_put_dnode(&tdn);
495 if (ino != dn->inode->i_ino)
497 else if (dn->inode_page_locked)
498 lock_page(dn->inode_page);
502 if (datablock_addr(tdn.inode, tdn.node_page,
503 tdn.ofs_in_node) == blkaddr)
504 f2fs_truncate_data_blocks_range(&tdn, 1);
505 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
506 unlock_page(dn->inode_page);
510 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
513 struct dnode_of_data dn;
515 unsigned int start, end;
516 int err = 0, recovered = 0;
518 /* step 1: recover xattr */
519 if (IS_INODE(page)) {
520 f2fs_recover_inline_xattr(inode, page);
521 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
522 err = f2fs_recover_xattr_data(inode, page);
528 /* step 2: recover inline data */
529 if (f2fs_recover_inline_data(inode, page))
532 /* step 3: recover data indices */
533 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
534 end = start + ADDRS_PER_PAGE(page, inode);
536 set_new_dnode(&dn, inode, NULL, NULL, 0);
538 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
540 if (err == -ENOMEM) {
541 congestion_wait(BLK_RW_ASYNC, HZ/50);
547 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
549 err = f2fs_get_node_info(sbi, dn.nid, &ni);
553 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
555 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
556 f2fs_msg(sbi->sb, KERN_WARNING,
557 "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
558 inode->i_ino, ofs_of_node(dn.node_page),
564 for (; start < end; start++, dn.ofs_in_node++) {
567 src = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
568 dest = datablock_addr(dn.inode, page, dn.ofs_in_node);
570 if (__is_valid_data_blkaddr(src) &&
571 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
576 if (__is_valid_data_blkaddr(dest) &&
577 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
582 /* skip recovering if dest is the same as src */
586 /* dest is invalid, just invalidate src block */
587 if (dest == NULL_ADDR) {
588 f2fs_truncate_data_blocks_range(&dn, 1);
592 if (!file_keep_isize(inode) &&
593 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
594 f2fs_i_size_write(inode,
595 (loff_t)(start + 1) << PAGE_SHIFT);
598 * dest is reserved block, invalidate src block
599 * and then reserve one new block in dnode page.
601 if (dest == NEW_ADDR) {
602 f2fs_truncate_data_blocks_range(&dn, 1);
603 f2fs_reserve_new_block(&dn);
607 /* dest is valid block, try to recover from src to dest */
608 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
610 if (src == NULL_ADDR) {
611 err = f2fs_reserve_new_block(&dn);
613 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
614 err = f2fs_reserve_new_block(&dn);
615 /* We should not get -ENOSPC */
616 f2fs_bug_on(sbi, err);
621 /* Check the previous node page having this index */
622 err = check_index_in_prev_nodes(sbi, dest, &dn);
624 if (err == -ENOMEM) {
625 congestion_wait(BLK_RW_ASYNC, HZ/50);
631 /* write dummy data page */
632 f2fs_replace_block(sbi, &dn, src, dest,
633 ni.version, false, false);
638 copy_node_footer(dn.node_page, page);
639 fill_node_footer(dn.node_page, dn.nid, ni.ino,
640 ofs_of_node(page), false);
641 set_page_dirty(dn.node_page);
645 f2fs_msg(sbi->sb, KERN_NOTICE,
646 "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
648 file_keep_isize(inode) ? "keep" : "recover",
653 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
654 struct list_head *tmp_inode_list, struct list_head *dir_list)
656 struct curseg_info *curseg;
657 struct page *page = NULL;
661 /* get node pages in the current segment */
662 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
663 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
666 struct fsync_inode_entry *entry;
668 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
671 f2fs_ra_meta_pages_cond(sbi, blkaddr);
673 page = f2fs_get_tmp_page(sbi, blkaddr);
679 if (!is_recoverable_dnode(page)) {
680 f2fs_put_page(page, 1);
684 entry = get_fsync_inode(inode_list, ino_of_node(page));
688 * inode(x) | CP | inode(x) | dnode(F)
689 * In this case, we can lose the latest inode(x).
690 * So, call recover_inode for the inode update.
692 if (IS_INODE(page)) {
693 err = recover_inode(entry->inode, page);
695 f2fs_put_page(page, 1);
699 if (entry->last_dentry == blkaddr) {
700 err = recover_dentry(entry->inode, page, dir_list);
702 f2fs_put_page(page, 1);
706 err = do_recover_data(sbi, entry->inode, page);
708 f2fs_put_page(page, 1);
712 if (entry->blkaddr == blkaddr)
713 list_move_tail(&entry->list, tmp_inode_list);
715 /* check next segment */
716 blkaddr = next_blkaddr_of_node(page);
717 f2fs_put_page(page, 1);
720 f2fs_allocate_new_segments(sbi);
724 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
726 struct list_head inode_list, tmp_inode_list;
727 struct list_head dir_list;
730 unsigned long s_flags = sbi->sb->s_flags;
731 bool need_writecp = false;
736 if (s_flags & SB_RDONLY) {
737 f2fs_msg(sbi->sb, KERN_INFO,
738 "recover fsync data on readonly fs");
739 sbi->sb->s_flags &= ~SB_RDONLY;
743 /* Needed for iput() to work correctly and not trash data */
744 sbi->sb->s_flags |= SB_ACTIVE;
745 /* Turn on quotas so that they are updated correctly */
746 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
749 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
750 sizeof(struct fsync_inode_entry));
751 if (!fsync_entry_slab) {
756 INIT_LIST_HEAD(&inode_list);
757 INIT_LIST_HEAD(&tmp_inode_list);
758 INIT_LIST_HEAD(&dir_list);
760 /* prevent checkpoint */
761 mutex_lock(&sbi->cp_mutex);
763 /* step #1: find fsynced inode numbers */
764 err = find_fsync_dnodes(sbi, &inode_list, check_only);
765 if (err || list_empty(&inode_list))
775 /* step #2: recover data */
776 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
778 f2fs_bug_on(sbi, !list_empty(&inode_list));
780 /* restore s_flags to let iput() trash data */
781 sbi->sb->s_flags = s_flags;
784 destroy_fsync_dnodes(&inode_list, err);
785 destroy_fsync_dnodes(&tmp_inode_list, err);
787 /* truncate meta pages to be used by the recovery */
788 truncate_inode_pages_range(META_MAPPING(sbi),
789 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
792 truncate_inode_pages_final(NODE_MAPPING(sbi));
793 truncate_inode_pages_final(META_MAPPING(sbi));
795 clear_sbi_flag(sbi, SBI_POR_DOING);
797 mutex_unlock(&sbi->cp_mutex);
799 /* let's drop all the directory inodes for clean checkpoint */
800 destroy_fsync_dnodes(&dir_list, err);
803 set_sbi_flag(sbi, SBI_IS_RECOVERED);
806 struct cp_control cpc = {
807 .reason = CP_RECOVERY,
809 err = f2fs_write_checkpoint(sbi, &cpc);
813 kmem_cache_destroy(fsync_entry_slab);
816 /* Turn quotas off */
818 f2fs_quota_off_umount(sbi->sb);
820 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
822 return ret ? ret: err;