4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
18 * Roll forward recovery scenarios.
20 * [Term] F: fsync_mark, D: dentry_mark
22 * 1. inode(x) | CP | inode(x) | dnode(F)
23 * -> Update the latest inode(x).
25 * 2. inode(x) | CP | inode(F) | dnode(F)
28 * 3. inode(x) | CP | dnode(F) | inode(x)
29 * -> Recover to the latest dnode(F), and drop the last inode(x)
31 * 4. inode(x) | CP | dnode(F) | inode(F)
34 * 5. CP | inode(x) | dnode(F)
35 * -> The inode(DF) was missing. Should drop this dnode(F).
37 * 6. CP | inode(DF) | dnode(F)
40 * 7. CP | dnode(F) | inode(DF)
41 * -> If f2fs_iget fails, then goto next to find inode(DF).
43 * 8. CP | dnode(F) | inode(x)
44 * -> If f2fs_iget fails, then goto next to find inode(DF).
45 * But it will fail due to no inode(DF).
48 static struct kmem_cache *fsync_entry_slab;
50 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
52 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
54 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
59 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
62 struct fsync_inode_entry *entry;
64 list_for_each_entry(entry, head, list)
65 if (entry->inode->i_ino == ino)
71 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
72 struct list_head *head, nid_t ino, bool quota_inode)
75 struct fsync_inode_entry *entry;
78 inode = f2fs_iget_retry(sbi->sb, ino);
80 return ERR_CAST(inode);
82 err = dquot_initialize(inode);
87 err = dquot_alloc_inode(inode);
92 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
94 list_add_tail(&entry->list, head);
102 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
105 /* inode should not be recovered, drop it */
106 f2fs_inode_synced(entry->inode);
109 list_del(&entry->list);
110 kmem_cache_free(fsync_entry_slab, entry);
113 static int recover_dentry(struct inode *inode, struct page *ipage,
114 struct list_head *dir_list)
116 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
117 nid_t pino = le32_to_cpu(raw_inode->i_pino);
118 struct f2fs_dir_entry *de;
119 struct fscrypt_name fname;
121 struct inode *dir, *einode;
122 struct fsync_inode_entry *entry;
126 entry = get_fsync_inode(dir_list, pino);
128 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
131 dir = ERR_CAST(entry);
132 err = PTR_ERR(entry);
139 memset(&fname, 0, sizeof(struct fscrypt_name));
140 fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen);
141 fname.disk_name.name = raw_inode->i_name;
143 if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
149 de = __f2fs_find_entry(dir, &fname, &page);
150 if (de && inode->i_ino == le32_to_cpu(de->ino))
154 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
155 if (IS_ERR(einode)) {
157 err = PTR_ERR(einode);
163 err = dquot_initialize(einode);
169 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
174 f2fs_delete_entry(de, page, dir, einode);
177 } else if (IS_ERR(page)) {
180 err = f2fs_add_dentry(dir, &fname, inode,
181 inode->i_ino, inode->i_mode);
188 f2fs_put_page(page, 0);
190 if (file_enc_name(inode))
191 name = "<encrypted>";
193 name = raw_inode->i_name;
194 f2fs_msg(inode->i_sb, KERN_NOTICE,
195 "%s: ino = %x, name = %s, dir = %lx, err = %d",
196 __func__, ino_of_node(ipage), name,
197 IS_ERR(dir) ? 0 : dir->i_ino, err);
201 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
203 if (ri->i_inline & F2FS_PIN_FILE)
204 set_inode_flag(inode, FI_PIN_FILE);
206 clear_inode_flag(inode, FI_PIN_FILE);
207 if (ri->i_inline & F2FS_DATA_EXIST)
208 set_inode_flag(inode, FI_DATA_EXIST);
210 clear_inode_flag(inode, FI_DATA_EXIST);
213 static void recover_inode(struct inode *inode, struct page *page)
215 struct f2fs_inode *raw = F2FS_INODE(page);
218 inode->i_mode = le16_to_cpu(raw->i_mode);
219 i_uid_write(inode, le32_to_cpu(raw->i_uid));
220 i_gid_write(inode, le32_to_cpu(raw->i_gid));
222 if (raw->i_inline & F2FS_EXTRA_ATTR) {
223 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)->sb) &&
224 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
228 i_projid = (projid_t)le32_to_cpu(raw->i_projid);
229 F2FS_I(inode)->i_projid =
230 make_kprojid(&init_user_ns, i_projid);
234 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
235 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
236 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
237 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
238 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
239 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
240 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
242 F2FS_I(inode)->i_advise = raw->i_advise;
243 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
244 f2fs_set_inode_flags(inode);
245 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
246 le16_to_cpu(raw->i_gc_failures);
248 recover_inline_flags(inode, raw);
250 f2fs_mark_inode_dirty_sync(inode, true);
252 if (file_enc_name(inode))
253 name = "<encrypted>";
255 name = F2FS_INODE(page)->i_name;
257 f2fs_msg(inode->i_sb, KERN_NOTICE,
258 "recover_inode: ino = %x, name = %s, inline = %x",
259 ino_of_node(page), name, raw->i_inline);
262 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
265 struct curseg_info *curseg;
266 struct page *page = NULL;
268 unsigned int loop_cnt = 0;
269 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
270 valid_user_blocks(sbi);
273 /* get node pages in the current segment */
274 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
275 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
278 struct fsync_inode_entry *entry;
280 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
283 page = f2fs_get_tmp_page(sbi, blkaddr);
289 if (!is_recoverable_dnode(page))
292 if (!is_fsync_dnode(page))
295 entry = get_fsync_inode(head, ino_of_node(page));
297 bool quota_inode = false;
300 IS_INODE(page) && is_dent_dnode(page)) {
301 err = f2fs_recover_inode_page(sbi, page);
308 * CP | dnode(F) | inode(DF)
309 * For this case, we should not give up now.
311 entry = add_fsync_inode(sbi, head, ino_of_node(page),
314 err = PTR_ERR(entry);
315 if (err == -ENOENT) {
322 entry->blkaddr = blkaddr;
324 if (IS_INODE(page) && is_dent_dnode(page))
325 entry->last_dentry = blkaddr;
327 /* sanity check in order to detect looped node chain */
328 if (++loop_cnt >= free_blocks ||
329 blkaddr == next_blkaddr_of_node(page)) {
330 f2fs_msg(sbi->sb, KERN_NOTICE,
331 "%s: detect looped node chain, "
332 "blkaddr:%u, next:%u",
333 __func__, blkaddr, next_blkaddr_of_node(page));
338 /* check next segment */
339 blkaddr = next_blkaddr_of_node(page);
340 f2fs_put_page(page, 1);
342 f2fs_ra_meta_pages_cond(sbi, blkaddr);
344 f2fs_put_page(page, 1);
348 static void destroy_fsync_dnodes(struct list_head *head, int drop)
350 struct fsync_inode_entry *entry, *tmp;
352 list_for_each_entry_safe(entry, tmp, head, list)
353 del_fsync_inode(entry, drop);
356 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
357 block_t blkaddr, struct dnode_of_data *dn)
359 struct seg_entry *sentry;
360 unsigned int segno = GET_SEGNO(sbi, blkaddr);
361 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
362 struct f2fs_summary_block *sum_node;
363 struct f2fs_summary sum;
364 struct page *sum_page, *node_page;
365 struct dnode_of_data tdn = *dn;
372 sentry = get_seg_entry(sbi, segno);
373 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
376 /* Get the previous summary */
377 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
378 struct curseg_info *curseg = CURSEG_I(sbi, i);
379 if (curseg->segno == segno) {
380 sum = curseg->sum_blk->entries[blkoff];
385 sum_page = f2fs_get_sum_page(sbi, segno);
386 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
387 sum = sum_node->entries[blkoff];
388 f2fs_put_page(sum_page, 1);
390 /* Use the locked dnode page and inode */
391 nid = le32_to_cpu(sum.nid);
392 if (dn->inode->i_ino == nid) {
394 if (!dn->inode_page_locked)
395 lock_page(dn->inode_page);
396 tdn.node_page = dn->inode_page;
397 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
399 } else if (dn->nid == nid) {
400 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
404 /* Get the node page */
405 node_page = f2fs_get_node_page(sbi, nid);
406 if (IS_ERR(node_page))
407 return PTR_ERR(node_page);
409 offset = ofs_of_node(node_page);
410 ino = ino_of_node(node_page);
411 f2fs_put_page(node_page, 1);
413 if (ino != dn->inode->i_ino) {
416 /* Deallocate previous index in the node page */
417 inode = f2fs_iget_retry(sbi->sb, ino);
419 return PTR_ERR(inode);
421 ret = dquot_initialize(inode);
430 bidx = f2fs_start_bidx_of_node(offset, inode) +
431 le16_to_cpu(sum.ofs_in_node);
434 * if inode page is locked, unlock temporarily, but its reference
437 if (ino == dn->inode->i_ino && dn->inode_page_locked)
438 unlock_page(dn->inode_page);
440 set_new_dnode(&tdn, inode, NULL, NULL, 0);
441 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
444 if (tdn.data_blkaddr == blkaddr)
445 f2fs_truncate_data_blocks_range(&tdn, 1);
447 f2fs_put_dnode(&tdn);
449 if (ino != dn->inode->i_ino)
451 else if (dn->inode_page_locked)
452 lock_page(dn->inode_page);
456 if (datablock_addr(tdn.inode, tdn.node_page,
457 tdn.ofs_in_node) == blkaddr)
458 f2fs_truncate_data_blocks_range(&tdn, 1);
459 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
460 unlock_page(dn->inode_page);
464 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
467 struct dnode_of_data dn;
469 unsigned int start, end;
470 int err = 0, recovered = 0;
472 /* step 1: recover xattr */
473 if (IS_INODE(page)) {
474 err = f2fs_recover_inline_xattr(inode, page);
477 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
478 err = f2fs_recover_xattr_data(inode, page);
484 /* step 2: recover inline data */
485 err = f2fs_recover_inline_data(inode, page);
492 /* step 3: recover data indices */
493 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
494 end = start + ADDRS_PER_PAGE(page, inode);
496 set_new_dnode(&dn, inode, NULL, NULL, 0);
498 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
500 if (err == -ENOMEM) {
501 congestion_wait(BLK_RW_ASYNC, HZ/50);
507 f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
509 err = f2fs_get_node_info(sbi, dn.nid, &ni);
513 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
515 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
516 f2fs_msg(sbi->sb, KERN_WARNING,
517 "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
518 inode->i_ino, ofs_of_node(dn.node_page),
524 for (; start < end; start++, dn.ofs_in_node++) {
527 src = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
528 dest = datablock_addr(dn.inode, page, dn.ofs_in_node);
530 /* skip recovering if dest is the same as src */
534 /* dest is invalid, just invalidate src block */
535 if (dest == NULL_ADDR) {
536 f2fs_truncate_data_blocks_range(&dn, 1);
540 if (!file_keep_isize(inode) &&
541 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
542 f2fs_i_size_write(inode,
543 (loff_t)(start + 1) << PAGE_SHIFT);
546 * dest is reserved block, invalidate src block
547 * and then reserve one new block in dnode page.
549 if (dest == NEW_ADDR) {
550 f2fs_truncate_data_blocks_range(&dn, 1);
551 f2fs_reserve_new_block(&dn);
555 /* dest is valid block, try to recover from src to dest */
556 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
558 if (src == NULL_ADDR) {
559 err = f2fs_reserve_new_block(&dn);
561 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
562 err = f2fs_reserve_new_block(&dn);
563 /* We should not get -ENOSPC */
564 f2fs_bug_on(sbi, err);
569 /* Check the previous node page having this index */
570 err = check_index_in_prev_nodes(sbi, dest, &dn);
572 if (err == -ENOMEM) {
573 congestion_wait(BLK_RW_ASYNC, HZ/50);
579 /* write dummy data page */
580 f2fs_replace_block(sbi, &dn, src, dest,
581 ni.version, false, false);
586 copy_node_footer(dn.node_page, page);
587 fill_node_footer(dn.node_page, dn.nid, ni.ino,
588 ofs_of_node(page), false);
589 set_page_dirty(dn.node_page);
593 f2fs_msg(sbi->sb, KERN_NOTICE,
594 "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
596 file_keep_isize(inode) ? "keep" : "recover",
601 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
602 struct list_head *tmp_inode_list, struct list_head *dir_list)
604 struct curseg_info *curseg;
605 struct page *page = NULL;
609 /* get node pages in the current segment */
610 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
611 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
614 struct fsync_inode_entry *entry;
616 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
619 f2fs_ra_meta_pages_cond(sbi, blkaddr);
621 page = f2fs_get_tmp_page(sbi, blkaddr);
627 if (!is_recoverable_dnode(page)) {
628 f2fs_put_page(page, 1);
632 entry = get_fsync_inode(inode_list, ino_of_node(page));
636 * inode(x) | CP | inode(x) | dnode(F)
637 * In this case, we can lose the latest inode(x).
638 * So, call recover_inode for the inode update.
641 recover_inode(entry->inode, page);
642 if (entry->last_dentry == blkaddr) {
643 err = recover_dentry(entry->inode, page, dir_list);
645 f2fs_put_page(page, 1);
649 err = do_recover_data(sbi, entry->inode, page);
651 f2fs_put_page(page, 1);
655 if (entry->blkaddr == blkaddr)
656 list_move_tail(&entry->list, tmp_inode_list);
658 /* check next segment */
659 blkaddr = next_blkaddr_of_node(page);
660 f2fs_put_page(page, 1);
663 f2fs_allocate_new_segments(sbi);
667 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
669 struct list_head inode_list, tmp_inode_list;
670 struct list_head dir_list;
673 unsigned long s_flags = sbi->sb->s_flags;
674 bool need_writecp = false;
679 if (s_flags & SB_RDONLY) {
680 f2fs_msg(sbi->sb, KERN_INFO,
681 "recover fsync data on readonly fs");
682 sbi->sb->s_flags &= ~SB_RDONLY;
686 /* Needed for iput() to work correctly and not trash data */
687 sbi->sb->s_flags |= SB_ACTIVE;
688 /* Turn on quotas so that they are updated correctly */
689 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
692 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
693 sizeof(struct fsync_inode_entry));
694 if (!fsync_entry_slab) {
699 INIT_LIST_HEAD(&inode_list);
700 INIT_LIST_HEAD(&tmp_inode_list);
701 INIT_LIST_HEAD(&dir_list);
703 /* prevent checkpoint */
704 mutex_lock(&sbi->cp_mutex);
706 /* step #1: find fsynced inode numbers */
707 err = find_fsync_dnodes(sbi, &inode_list, check_only);
708 if (err || list_empty(&inode_list))
718 /* step #2: recover data */
719 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
721 f2fs_bug_on(sbi, !list_empty(&inode_list));
723 /* restore s_flags to let iput() trash data */
724 sbi->sb->s_flags = s_flags;
727 destroy_fsync_dnodes(&inode_list, err);
728 destroy_fsync_dnodes(&tmp_inode_list, err);
730 /* truncate meta pages to be used by the recovery */
731 truncate_inode_pages_range(META_MAPPING(sbi),
732 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
735 truncate_inode_pages_final(NODE_MAPPING(sbi));
736 truncate_inode_pages_final(META_MAPPING(sbi));
738 clear_sbi_flag(sbi, SBI_POR_DOING);
740 mutex_unlock(&sbi->cp_mutex);
742 /* let's drop all the directory inodes for clean checkpoint */
743 destroy_fsync_dnodes(&dir_list, err);
746 set_sbi_flag(sbi, SBI_IS_RECOVERED);
749 struct cp_control cpc = {
750 .reason = CP_RECOVERY,
752 err = f2fs_write_checkpoint(sbi, &cpc);
756 kmem_cache_destroy(fsync_entry_slab);
759 /* Turn quotas off */
761 f2fs_quota_off_umount(sbi->sb);
763 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
765 return ret ? ret: err;