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)
105 list_del(&entry->list);
106 kmem_cache_free(fsync_entry_slab, entry);
109 static int recover_dentry(struct inode *inode, struct page *ipage,
110 struct list_head *dir_list)
112 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
113 nid_t pino = le32_to_cpu(raw_inode->i_pino);
114 struct f2fs_dir_entry *de;
115 struct fscrypt_name fname;
117 struct inode *dir, *einode;
118 struct fsync_inode_entry *entry;
122 entry = get_fsync_inode(dir_list, pino);
124 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
127 dir = ERR_CAST(entry);
128 err = PTR_ERR(entry);
135 memset(&fname, 0, sizeof(struct fscrypt_name));
136 fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen);
137 fname.disk_name.name = raw_inode->i_name;
139 if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
145 de = __f2fs_find_entry(dir, &fname, &page);
146 if (de && inode->i_ino == le32_to_cpu(de->ino))
150 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
151 if (IS_ERR(einode)) {
153 err = PTR_ERR(einode);
159 err = dquot_initialize(einode);
165 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
170 f2fs_delete_entry(de, page, dir, einode);
173 } else if (IS_ERR(page)) {
176 err = f2fs_add_dentry(dir, &fname, inode,
177 inode->i_ino, inode->i_mode);
184 f2fs_put_page(page, 0);
186 if (file_enc_name(inode))
187 name = "<encrypted>";
189 name = raw_inode->i_name;
190 f2fs_msg(inode->i_sb, KERN_NOTICE,
191 "%s: ino = %x, name = %s, dir = %lx, err = %d",
192 __func__, ino_of_node(ipage), name,
193 IS_ERR(dir) ? 0 : dir->i_ino, err);
197 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
199 if (ri->i_inline & F2FS_PIN_FILE)
200 set_inode_flag(inode, FI_PIN_FILE);
202 clear_inode_flag(inode, FI_PIN_FILE);
203 if (ri->i_inline & F2FS_DATA_EXIST)
204 set_inode_flag(inode, FI_DATA_EXIST);
206 clear_inode_flag(inode, FI_DATA_EXIST);
209 static void recover_inode(struct inode *inode, struct page *page)
211 struct f2fs_inode *raw = F2FS_INODE(page);
214 inode->i_mode = le16_to_cpu(raw->i_mode);
215 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
216 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
217 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
218 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
219 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
220 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
221 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
223 F2FS_I(inode)->i_advise = raw->i_advise;
225 recover_inline_flags(inode, raw);
227 if (file_enc_name(inode))
228 name = "<encrypted>";
230 name = F2FS_INODE(page)->i_name;
232 f2fs_msg(inode->i_sb, KERN_NOTICE,
233 "recover_inode: ino = %x, name = %s, inline = %x",
234 ino_of_node(page), name, raw->i_inline);
237 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
240 struct curseg_info *curseg;
241 struct page *page = NULL;
243 unsigned int loop_cnt = 0;
244 unsigned int free_blocks = sbi->user_block_count -
245 valid_user_blocks(sbi);
248 /* get node pages in the current segment */
249 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
250 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
253 struct fsync_inode_entry *entry;
255 if (!f2fs_is_valid_meta_blkaddr(sbi, blkaddr, META_POR))
258 page = f2fs_get_tmp_page(sbi, blkaddr);
260 if (!is_recoverable_dnode(page))
263 if (!is_fsync_dnode(page))
266 entry = get_fsync_inode(head, ino_of_node(page));
268 bool quota_inode = false;
271 IS_INODE(page) && is_dent_dnode(page)) {
272 err = f2fs_recover_inode_page(sbi, page);
279 * CP | dnode(F) | inode(DF)
280 * For this case, we should not give up now.
282 entry = add_fsync_inode(sbi, head, ino_of_node(page),
285 err = PTR_ERR(entry);
286 if (err == -ENOENT) {
293 entry->blkaddr = blkaddr;
295 if (IS_INODE(page) && is_dent_dnode(page))
296 entry->last_dentry = blkaddr;
298 /* sanity check in order to detect looped node chain */
299 if (++loop_cnt >= free_blocks ||
300 blkaddr == next_blkaddr_of_node(page)) {
301 f2fs_msg(sbi->sb, KERN_NOTICE,
302 "%s: detect looped node chain, "
303 "blkaddr:%u, next:%u",
304 __func__, blkaddr, next_blkaddr_of_node(page));
309 /* check next segment */
310 blkaddr = next_blkaddr_of_node(page);
311 f2fs_put_page(page, 1);
313 f2fs_ra_meta_pages_cond(sbi, blkaddr);
315 f2fs_put_page(page, 1);
319 static void destroy_fsync_dnodes(struct list_head *head)
321 struct fsync_inode_entry *entry, *tmp;
323 list_for_each_entry_safe(entry, tmp, head, list)
324 del_fsync_inode(entry);
327 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
328 block_t blkaddr, struct dnode_of_data *dn)
330 struct seg_entry *sentry;
331 unsigned int segno = GET_SEGNO(sbi, blkaddr);
332 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
333 struct f2fs_summary_block *sum_node;
334 struct f2fs_summary sum;
335 struct page *sum_page, *node_page;
336 struct dnode_of_data tdn = *dn;
343 sentry = get_seg_entry(sbi, segno);
344 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
347 /* Get the previous summary */
348 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
349 struct curseg_info *curseg = CURSEG_I(sbi, i);
350 if (curseg->segno == segno) {
351 sum = curseg->sum_blk->entries[blkoff];
356 sum_page = f2fs_get_sum_page(sbi, segno);
357 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
358 sum = sum_node->entries[blkoff];
359 f2fs_put_page(sum_page, 1);
361 /* Use the locked dnode page and inode */
362 nid = le32_to_cpu(sum.nid);
363 if (dn->inode->i_ino == nid) {
365 if (!dn->inode_page_locked)
366 lock_page(dn->inode_page);
367 tdn.node_page = dn->inode_page;
368 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
370 } else if (dn->nid == nid) {
371 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
375 /* Get the node page */
376 node_page = f2fs_get_node_page(sbi, nid);
377 if (IS_ERR(node_page))
378 return PTR_ERR(node_page);
380 offset = ofs_of_node(node_page);
381 ino = ino_of_node(node_page);
382 f2fs_put_page(node_page, 1);
384 if (ino != dn->inode->i_ino) {
387 /* Deallocate previous index in the node page */
388 inode = f2fs_iget_retry(sbi->sb, ino);
390 return PTR_ERR(inode);
392 ret = dquot_initialize(inode);
401 bidx = f2fs_start_bidx_of_node(offset, inode) +
402 le16_to_cpu(sum.ofs_in_node);
405 * if inode page is locked, unlock temporarily, but its reference
408 if (ino == dn->inode->i_ino && dn->inode_page_locked)
409 unlock_page(dn->inode_page);
411 set_new_dnode(&tdn, inode, NULL, NULL, 0);
412 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
415 if (tdn.data_blkaddr == blkaddr)
416 f2fs_truncate_data_blocks_range(&tdn, 1);
418 f2fs_put_dnode(&tdn);
420 if (ino != dn->inode->i_ino)
422 else if (dn->inode_page_locked)
423 lock_page(dn->inode_page);
427 if (datablock_addr(tdn.inode, tdn.node_page,
428 tdn.ofs_in_node) == blkaddr)
429 f2fs_truncate_data_blocks_range(&tdn, 1);
430 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
431 unlock_page(dn->inode_page);
435 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
438 struct dnode_of_data dn;
440 unsigned int start, end;
441 int err = 0, recovered = 0;
443 /* step 1: recover xattr */
444 if (IS_INODE(page)) {
445 f2fs_recover_inline_xattr(inode, page);
446 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
447 err = f2fs_recover_xattr_data(inode, page);
453 /* step 2: recover inline data */
454 if (f2fs_recover_inline_data(inode, page))
457 /* step 3: recover data indices */
458 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
459 end = start + ADDRS_PER_PAGE(page, inode);
461 set_new_dnode(&dn, inode, NULL, NULL, 0);
463 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
465 if (err == -ENOMEM) {
466 congestion_wait(BLK_RW_ASYNC, HZ/50);
472 f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
474 f2fs_get_node_info(sbi, dn.nid, &ni);
475 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
476 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
478 for (; start < end; start++, dn.ofs_in_node++) {
481 src = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
482 dest = datablock_addr(dn.inode, page, dn.ofs_in_node);
484 /* skip recovering if dest is the same as src */
488 /* dest is invalid, just invalidate src block */
489 if (dest == NULL_ADDR) {
490 f2fs_truncate_data_blocks_range(&dn, 1);
494 if (!file_keep_isize(inode) &&
495 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
496 f2fs_i_size_write(inode,
497 (loff_t)(start + 1) << PAGE_SHIFT);
500 * dest is reserved block, invalidate src block
501 * and then reserve one new block in dnode page.
503 if (dest == NEW_ADDR) {
504 f2fs_truncate_data_blocks_range(&dn, 1);
505 f2fs_reserve_new_block(&dn);
509 /* dest is valid block, try to recover from src to dest */
510 if (f2fs_is_valid_meta_blkaddr(sbi, dest, META_POR)) {
512 if (src == NULL_ADDR) {
513 err = f2fs_reserve_new_block(&dn);
514 #ifdef CONFIG_F2FS_FAULT_INJECTION
516 err = f2fs_reserve_new_block(&dn);
518 /* We should not get -ENOSPC */
519 f2fs_bug_on(sbi, err);
524 /* Check the previous node page having this index */
525 err = check_index_in_prev_nodes(sbi, dest, &dn);
527 if (err == -ENOMEM) {
528 congestion_wait(BLK_RW_ASYNC, HZ/50);
534 /* write dummy data page */
535 f2fs_replace_block(sbi, &dn, src, dest,
536 ni.version, false, false);
541 copy_node_footer(dn.node_page, page);
542 fill_node_footer(dn.node_page, dn.nid, ni.ino,
543 ofs_of_node(page), false);
544 set_page_dirty(dn.node_page);
548 f2fs_msg(sbi->sb, KERN_NOTICE,
549 "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
551 file_keep_isize(inode) ? "keep" : "recover",
556 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
557 struct list_head *dir_list)
559 struct curseg_info *curseg;
560 struct page *page = NULL;
564 /* get node pages in the current segment */
565 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
566 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
569 struct fsync_inode_entry *entry;
571 if (!f2fs_is_valid_meta_blkaddr(sbi, blkaddr, META_POR))
574 f2fs_ra_meta_pages_cond(sbi, blkaddr);
576 page = f2fs_get_tmp_page(sbi, blkaddr);
578 if (!is_recoverable_dnode(page)) {
579 f2fs_put_page(page, 1);
583 entry = get_fsync_inode(inode_list, ino_of_node(page));
587 * inode(x) | CP | inode(x) | dnode(F)
588 * In this case, we can lose the latest inode(x).
589 * So, call recover_inode for the inode update.
592 recover_inode(entry->inode, page);
593 if (entry->last_dentry == blkaddr) {
594 err = recover_dentry(entry->inode, page, dir_list);
596 f2fs_put_page(page, 1);
600 err = do_recover_data(sbi, entry->inode, page);
602 f2fs_put_page(page, 1);
606 if (entry->blkaddr == blkaddr)
607 del_fsync_inode(entry);
609 /* check next segment */
610 blkaddr = next_blkaddr_of_node(page);
611 f2fs_put_page(page, 1);
614 f2fs_allocate_new_segments(sbi);
618 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
620 struct list_head inode_list;
621 struct list_head dir_list;
624 unsigned long s_flags = sbi->sb->s_flags;
625 bool need_writecp = false;
630 if (s_flags & SB_RDONLY) {
631 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
632 sbi->sb->s_flags &= ~SB_RDONLY;
636 /* Needed for iput() to work correctly and not trash data */
637 sbi->sb->s_flags |= SB_ACTIVE;
638 /* Turn on quotas so that they are updated correctly */
639 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
642 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
643 sizeof(struct fsync_inode_entry));
644 if (!fsync_entry_slab) {
649 INIT_LIST_HEAD(&inode_list);
650 INIT_LIST_HEAD(&dir_list);
652 /* prevent checkpoint */
653 mutex_lock(&sbi->cp_mutex);
655 /* step #1: find fsynced inode numbers */
656 err = find_fsync_dnodes(sbi, &inode_list, check_only);
657 if (err || list_empty(&inode_list))
667 /* step #2: recover data */
668 err = recover_data(sbi, &inode_list, &dir_list);
670 f2fs_bug_on(sbi, !list_empty(&inode_list));
672 destroy_fsync_dnodes(&inode_list);
674 /* truncate meta pages to be used by the recovery */
675 truncate_inode_pages_range(META_MAPPING(sbi),
676 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
679 truncate_inode_pages_final(NODE_MAPPING(sbi));
680 truncate_inode_pages_final(META_MAPPING(sbi));
683 clear_sbi_flag(sbi, SBI_POR_DOING);
684 mutex_unlock(&sbi->cp_mutex);
686 /* let's drop all the directory inodes for clean checkpoint */
687 destroy_fsync_dnodes(&dir_list);
689 if (!err && need_writecp) {
690 struct cp_control cpc = {
691 .reason = CP_RECOVERY,
693 err = f2fs_write_checkpoint(sbi, &cpc);
696 kmem_cache_destroy(fsync_entry_slab);
699 /* Turn quotas off */
701 f2fs_quota_off_umount(sbi->sb);
703 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
705 return ret ? ret: err;